xref: /netbsd-src/external/gpl3/gcc/dist/NEWS (revision 4d6fc14bc9b0c5bf3e30be318c143ee82cadd108)
1This file contains information about GCC releases which has been generated
2automatically from the online release notes.  It covers releases of GCC
3(and the former EGCS project) since EGCS 1.0, on the line of development
4that led to GCC 3. For information on GCC 2.8.1 and older releases of GCC 2,
5see ONEWS.
6
7======================================================================
8http://gcc.gnu.org/gcc-10/index.html
9
10                             GCC 10 Release Series
11
12   April 8, 2021
13
14   The [1]GNU project and the GCC developers are pleased to announce the
15   release of GCC 10.3.
16
17   This release is a bug-fix release, containing fixes for regressions in
18   GCC 10.2 relative to previous releases of GCC.
19
20Release History
21
22   GCC 10.3
23          April 8, 2021 ([2]changes, [3]documentation)
24
25   GCC 10.2
26          July 23, 2020 ([4]changes, [5]documentation)
27
28   GCC 10.1
29          May 7, 2020 ([6]changes, [7]documentation)
30
31References and Acknowledgements
32
33   GCC used to stand for the GNU C Compiler, but since the compiler
34   supports several other languages aside from C, it now stands for the
35   GNU Compiler Collection.
36
37   A list of [8]successful builds is updated as new information becomes
38   available.
39
40   The GCC developers would like to thank the numerous people that have
41   contributed new features, improvements, bug fixes, and other changes as
42   well as test results to GCC. This [9]amazing group of volunteers is
43   what makes GCC successful.
44
45   For additional information about GCC please refer to the [10]GCC
46   project web site or contact the [11]GCC development mailing list.
47
48   To obtain GCC please use [12]our mirror sites or [13]our version
49   control system.
50
51
52    For questions related to the use of GCC, please consult these web
53    pages and the [14]GCC manuals. If that fails, the
54    [15]gcc-help@gcc.gnu.org mailing list might help. Comments on these
55    web pages and the development of GCC are welcome on our developer
56    list at [16]gcc@gcc.gnu.org. All of [17]our lists have public
57    archives.
58
59   Copyright (C) [18]Free Software Foundation, Inc. Verbatim copying and
60   distribution of this entire article is permitted in any medium,
61   provided this notice is preserved.
62
63   These pages are [19]maintained by the GCC team. Last modified
64   2021-04-08[20].
65
66References
67
68   1. http://www.gnu.org/
69   2. http://gcc.gnu.org/gcc-10/changes.html
70   3. http://gcc.gnu.org/onlinedocs/10.3.0/
71   4. http://gcc.gnu.org/gcc-10/changes.html
72   5. http://gcc.gnu.org/onlinedocs/10.2.0/
73   6. http://gcc.gnu.org/gcc-10/changes.html
74   7. http://gcc.gnu.org/onlinedocs/10.1.0/
75   8. http://gcc.gnu.org/gcc-10/buildstat.html
76   9. http://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Contributors.html
77  10. http://gcc.gnu.org/index.html
78  11. mailto:gcc@gcc.gnu.org
79  12. http://gcc.gnu.org/mirrors.html
80  13. http://gcc.gnu.org/git.html
81  14. https://gcc.gnu.org/onlinedocs/
82  15. mailto:gcc-help@gcc.gnu.org
83  16. mailto:gcc@gcc.gnu.org
84  17. https://gcc.gnu.org/lists.html
85  18. https://www.fsf.org/
86  19. https://gcc.gnu.org/about.html
87  20. http://validator.w3.org/check/referer
88======================================================================
89http://gcc.gnu.org/gcc-10/changes.html
90
91                             GCC 10 Release Series
92                        Changes, New Features, and Fixes
93
94   This page is a "brief" summary of some of the huge number of
95   improvements in GCC 10. You may also want to check out our [1]Porting
96   to GCC 10 page and the [2]full GCC documentation.
97
98Caveats
99
100     * An ABI incompatibility between C++14 and C++17 has been fixed. On
101       some targets a class with a zero-sized subobject would be passed
102       incorrectly when compiled as C++17 or C++20. See the [3]C++ notes
103       below for more details.
104     * The deprecated Profile Mode and array_allocator extensions have
105       been removed from libstdc++.
106     * The non-standard std::__is_nullptr_t type trait is deprecated and
107       will be removed from libstdc++ in a future release. The standard
108       trait std::is_null_pointer should be instead.
109     * The minimum version of the [4]MPFR library required for building
110       GCC has been increased to version 3.1.0 (released 2011-10-03).
111     * The automatic template instantiation at link time (-frepo) has been
112       removed.
113     * The --param allow-store-data-races internal parameter has been
114       removed in favor of a new official option -fallow-store-data-races.
115       While default behavior is unchanged and the new option allows to
116       correctly maintain a per compilation unit setting across link-time
117       optimization, alteration of the default via --param
118       allow-store-data-races will now be diagnosed and build systems have
119       to be adjusted accordingly.
120     * Offloading to Heterogeneous System Architecture Intermediate
121       Language (HSAIL) has been deprecated and will likely be removed in
122       a future release.
123     * The type of the std::iterator base class of
124       std::istreambuf_iterator was changed in C++98 mode to be consistent
125       with C++11 and later standards. See the [5]libstdc++ notes below
126       for more details.
127
128General Improvements
129
130     * New built-in functions:
131          + The [6]__has_builtin built-in preprocessor operator can be
132            used to query support for built-in functions provided by GCC
133            and other compilers that support it.
134          + __builtin_roundeven for the corresponding function from
135            ISO/IEC TS 18661.
136     * New command-line options:
137          + [7]-fallocation-dce removes unneeded pairs of new and delete
138            operators.
139          + [8]-fprofile-partial-training can now be used to inform the
140            compiler that code paths not covered by the training run
141            should not be optimized for size.
142          + [9]-fprofile-reproducible controls level of reproducibility of
143            profile gathered by [10]-fprofile-generate. This makes it
144            possible to rebuild program with same outcome which is useful,
145            for example, for distribution packages.
146          + [11]-fprofile-prefix-path can be used in combination with
147            -fprofile-generate=profile_dir and -fprofile-use=profile_dir
148            to inform GCC where the base directory of build source tree is
149            in case it differs between instrumentation and optimized
150            builds.
151          + [12]-fanalyzer enables a new static analysis pass and
152            associated warnings. This pass performs a time-consuming
153            exploration of paths through the code in the hope of detecting
154            various common errors, such as double-free bugs. This option
155            should be regarded as experimental in this release. In
156            particular, analysis of non-C code is unlikely to work.
157     * Inter-procedural optimization improvements:
158          + The inter-procedural scalar replacement of aggregates
159            (IPA-SRA) pass was re-implemented to work at link-time and can
160            now also remove computing and returning unused return values.
161          + [13]-finline-functions is now enabled at -O2 and was retuned
162            for better code size versus runtime performance trade-offs.
163            Inliner heuristics was also significantly sped up to avoid
164            negative impact to -flto -O2 compile times.
165          + Inliner heuristics and function cloning can now use
166            value-range information to predict effectivity of individual
167            transformations.
168          + During link-time optimization the C++ One Definition Rule is
169            used to increase precision of type based alias analysis.
170     * Link-time optimization improvements:
171          + A new binary [14]lto-dump has been added. It dumps various
172            information about LTO bytecode object files.
173          + The parallel phase of the LTO can automatically detect a
174            running make's jobserver or fall back to number of available
175            cores.
176          + The LTO bytecode can be compressed with the [15]zstd
177            algorithm. The configure script automatically detects zstd
178            support.
179          + Most --param values can now be specified at translation unit
180            granularity. This includes all parameters controlling the
181            inliner and other inter-procedural optimizations. Unlike
182            earlier releases, GCC 10 will ignore parameters controlling
183            optimizations specified at link-time and apply parameters
184            specified at compile-time in the same manner as done for
185            optimization flags.
186     * Profile driven optimization improvements:
187          + Profile maintenance during compilation and hot/cold code
188            partitioning have been improved.
189          + Using [16]-fprofile-values, an instrumented binary can track
190            multiple values (up to 4) for e.g. indirect calls and provide
191            more precise profile information.
192
193New Languages and Language-Specific Improvements
194
195     * Version 2.6 of the [17]OpenACC specification is now supported in
196       the C, C++ and Fortran compilers. See the [18]implementation status
197       section on the OpenACC wiki page and the [19]run-time library
198       documentation for further information.
199     * GCC 10 adds a number of newly implemented [20]OpenMP 5.0 features
200       on top of the GCC 9 release such as conditional lastprivate clause,
201       scan and loop directives, order(concurrent) and use_device_addr
202       clauses support, if clause on simd construct or partial support for
203       the declare variant directive, getting closer to full support of
204       the OpenMP 5.0 standard.
205     * OpenMP and OpenACC now support [21]offloading to AMD Radeon (GCN)
206       GPUs; supported are the third-generation Fiji (fiji) and the
207       fifth-generation VEGA 10/VEGA 20 (gfx900 or gfx906).
208
209  C family
210
211     * New attributes:
212          + The access function and type attribute has been added to
213            describe how a function accesses objects passed to it by
214            pointer or reference, and to associate such arguments with
215            integer arguments denoting the objects' sizes. The attribute
216            is used to enable the detection of invalid accesses by
217            user-defined functions, such as those diagnosed by
218            -Wstringop-overflow.
219          + The symver attribute can be used to bind symbols to specific
220            version nodes on ELF platforms. This is preferred to using
221            inline assembly with GNU as symver directive because the
222            latter is not compatible with link-time optimizations.
223     * New warnings:
224          + [22]-Wstring-compare, enabled by -Wextra, warns about equality
225            and inequality expressions between zero and the result of a
226            call to either strcmp and strncmp that evaluate to a constant
227            as a result of the length of one argument being greater than
228            the size of the array pointed to by the other.
229          + [23]-Wzero-length-bounds, enabled by -Warray-bounds, warns
230            about accesses to elements of zero-length arrays that might
231            overlap other members of the same object.
232     * Enhancements to existing warnings:
233          + [24]-Warray-bounds detects more out-of-bounds accesses to
234            member arrays as well as accesses to elements of zero-length
235            arrays.
236          + [25]-Wformat-overflow makes full use of string length
237            information computed by the strlen optimization pass.
238          + [26]-Wrestrict detects overlapping accesses to dynamically
239            allocated objects.
240          + [27]-Wreturn-local-addr diagnoses more instances of return
241            statements returning addresses of automatic variables.
242          + [28]-Wstringop-overflow detects more out-of-bounds stores to
243            member arrays including zero-length arrays, dynamically
244            allocated objects and variable length arrays, as well as more
245            instances of reads of unterminated character arrays by string
246            built-in functions. The warning also detects out-of-bounds
247            accesses by calls to user-defined functions declared with the
248            new attribute access.
249          + [29]-Warith-conversion re-enables warnings from -Wconversion,
250            -Wfloat-conversion, and -Wsign-conversion that are now off by
251            default for an expression where the result of an arithmetic
252            operation will not fit in the target type due to promotion,
253            but the operands of the expression do fit in the target type.
254     * Extended characters in identifiers may now be specified directly in
255       the input encoding (UTF-8, by default), in addition to the UCN
256       syntax (\uNNNN or \UNNNNNNNN) that is already supported:
257
258static const int p = 3;
259int get_na�ve_pi() {
260  return p;
261}
262
263  C
264
265     * Several new features from the upcoming C2X revision of the ISO C
266       standard are supported with -std=c2x and -std=gnu2x. Some of these
267       features are also supported as extensions when compiling for older
268       language versions. In addition to the features listed, some
269       features previously supported as extensions and now added to the C
270       standard are enabled by default in C2X mode and not diagnosed with
271       -std=c2x -Wpedantic.
272          + The [[]] attribute syntax is supported, as in C++. Existing
273            attributes can be used with this syntax in forms such as
274            [[gnu::const]]. The standard attributes [[deprecated]],
275            [[fallthrough]] and [[maybe_unused]] are supported.
276          + UTF-8 character constants using the u8'' syntax are supported.
277          + <float.h> defines macros FLT_NORM_MAX, DBL_NORM_MAX and
278            LDBL_NORM_MAX.
279          + When decimal floating-point arithmetic is supported, <float.h>
280            defines macros DEC32_TRUE_MIN, DEC64_TRUE_MIN and
281            DEC128_TRUE_MIN, in addition to the macros that were
282            previously only defined if __STDC_WANT_DEC_FP__ was defined
283            before including <float.h>.
284          + In C2X mode, empty parentheses in a function definition give
285            that function a type with a prototype for subsequent calls;
286            other old-style function definitions are diagnosed by default
287            in C2X mode.
288          + The strftime format checking supports the %OB and %Ob formats.
289          + In C2X mode, -fno-fp-int-builtin-inexact is enabled by
290            default.
291     * GCC now defaults to -fno-common. As a result, global variable
292       accesses are more efficient on various targets. In C, global
293       variables with multiple tentative definitions now result in linker
294       errors. With -fcommon such definitions are silently merged during
295       linking.
296
297  C++
298
299     * Several C++20 features have been implemented:
300          + Concepts, including P0734R0, P0857R0, P1084R2, P1141R2,
301            P0848R3, P1616R1, P1452R2
302          + P1668R1, Permitting Unevaluated inline-assembly in constexpr
303            Functions
304          + P1161R3, Deprecate a[b,c]
305          + P0848R3, Conditionally Trivial Special Member Functions
306          + P1091R3, Extending structured bindings
307          + P1143R2, Adding the constinit keyword
308          + P1152R4, Deprecating volatile
309          + P0388R4, Permit conversions to arrays of unknown bound
310          + P0784R7, constexpr new
311          + P1301R4, [[nodiscard("with reason")]]
312          + P1814R0, class template argument deduction for alias templates
313          + P1816R0, class template argument deduction for aggregates
314          + P0960R3, Parenthesized initialization of aggregates
315          + P1331R2, Allow trivial default initialization in constexpr
316            contexts
317          + P1327R1, Allowing dynamic_cast and polymorphic typeid in
318            constexpr contexts
319          + P0912R5, Coroutines (requires -fcoroutines)
320     * Several C++ Defect Reports have been resolved, e.g.:
321          + DR 1560, lvalue-to-rvalue conversion in ?:
322          + DR 1813, __is_standard_layout for a class with repeated bases
323          + DR 2094, volatile scalars are trivially copyable,
324          + DR 2096, constraints on literal unions
325          + DR 2413, typename in conversion-function-ids
326          + DR 2352, Similar types and reference binding
327          + DR 1601, Promotion of enumeration with fixed underlying type
328          + DR 330, Qualification conversions and pointers to arrays of
329            pointers
330          + DR 1307, Overload resolution based on size of array
331            initializer-list
332          + DR 1710, Missing template keyword in class-or-decltype
333     * New warnings:
334          + [30]-Wmismatched-tags, disabled by default, warns about
335            declarations of structs, classes, and class templates and
336            their specializations with a class-key that does not match
337            either the definition or the first declaration if no
338            definition is provided. The option is provided to ease
339            portability to Windows-based compilers.
340          + [31]-Wredundant-tags, disabled by default, warns about
341            redundant class-key and enum-key in contexts where the key can
342            be eliminated without causing an syntactic ambiguity.
343     * G++ can now detect modifying constant objects in constexpr
344       evaluation (which is undefined behavior).
345     * G++ no longer emits bogus -Wsign-conversion warnings with explicit
346       casts.
347     * Narrowing is now detected in more contexts (e.g., case values).
348     * Memory consumption of the compiler has been reduced in constexpr
349       evaluation.
350     * The noexcept-specifier is now properly treated as a complete-class
351       context as per [class.mem].
352     * The attribute deprecated can now be used on namespaces too.
353     * The ABI of passing and returning certain C++ classes by value
354       changed on several targets in GCC 10, including [32]AArch64,
355       [33]ARM, [34]PowerPC ELFv2, [35]S/390 and [36]Itanium. These
356       changes affect classes with a zero-sized subobject (an empty base
357       class, or data member with the [[no_unique_address]] attribute)
358       where all other non-static data members have the same type (this is
359       called a "homogeneous aggregate" in some ABI specifications, or if
360       there is only one such member, a "single element"). In -std=c++17
361       and -std=c++20 modes, classes with an empty base class were not
362       considered to have a single element or to be a homogeneous
363       aggregate, and so could be passed differently (in the wrong
364       registers or at the wrong stack address). This could make code
365       compiled with -std=c++17 and -std=c++14 ABI incompatible. This has
366       been corrected and the empty bases are ignored in those ABI
367       decisions, so functions compiled with -std=c++14 and -std=c++17 are
368       now ABI compatible again. Example: struct empty {}; struct S :
369       empty { float f; }; void f(S);. Similarly, in classes containing
370       non-static data members with empty class types using the C++20
371       [[no_unique_address]] attribute, those members weren't ignored in
372       the ABI argument passing decisions as they should be. Both of these
373       ABI changes are now diagnosed with -Wpsabi.
374
375    Runtime Library (libstdc++)
376
377     * Improved experimental C++2a support, including:
378          + Library concepts in <concepts> and <iterator>.
379          + Constrained algorithms in <ranges>, <algorithm>, and <memory>
380            (thanks to Patrick Palka).
381          + New algorithms shift_left and shift_right (thanks to Patrick
382            Palka).
383          + std::span (thanks to JeanHeyd Meneide).
384          + Three-way comparisons in <compare> and throughout the library.
385          + Constexpr support in <algorithm> and elsewhere (thanks to
386            Edward Smith-Rowland).
387          + <stop_token> and std::jthread (thanks to Thomas Rodgers).
388          + std::atomic_ref and std::atomic<floating point>.
389          + Integer comparison functions (cmp_equal, cmp_less etc.).
390          + std::ssize, std::to_array.
391          + std::construct_at, std::destroy, constexpr std::allocator.
392          + Mathematical constants in <numbers>.
393     * Support for RDSEED in std::random_device.
394     * Reduced header dependencies, leading to faster compilation for some
395       code.
396     * The std::iterator base class of std::istreambuf_iterator was
397       changed in C++98 mode to be consistent with C++11 and later
398       standards. This is expected to have no noticeable effect except in
399       the unlikely case of a class which has potentially overlapping
400       subobjects of type std::istreambuf_iterator<C> and another iterator
401       type with a std::iterator<input_iterator_tag, C, ...> base class.
402       The layout of such a type might change when compiled as C++98.
403       [37]Bug 92285 has more details and concrete examples.
404
405  D
406
407     * Support for static foreach has been implemented.
408     * Aliases can now be created directly from any __traits that return
409       symbols or tuples. Previously, an AliasSeq was necessary in order
410       to alias their return.
411     * It is now possible to detect the language ABI specified for a
412       struct, class, or interface using __traits(getLinkage, ...).
413     * Support for core.math.toPrec intrinsics has been added. These
414       intrinsics guarantee the rounding to specific floating-point
415       precisions at specified points in the code.
416     * Support for pragma(inline) has been implemented. Previously the
417       pragma was recognized, but had no effect on the compilation.
418     * Optional parentheses in asm operands are deprecated and will be
419       removed in a future release.
420     * All content imported files are now included in the make dependency
421       list when compiling with -M.
422     * Compiler recognized attributes provided by the gcc.attribute module
423       will now take effect when applied to function prototypes as well as
424       when applied to full function declarations.
425     * Added a --enable-libphobos-checking configure option to control
426       whether run-time checks are compiled into the D runtime library.
427     * Added a --with-libphobos-druntime-only configure option to indicate
428       whether to build only the core D runtime library, or both the core
429       and standard libraries into libphobos.
430
431  Fortran
432
433     * use_device_addr of version 5.0 of the [38]OpenMP specification is
434       now supported. Note that otherwise OpenMP 4.5 is partially
435       supported in the Fortran compiler; the largest missing item is
436       structure element mapping.
437     * The default buffer size for I/O using unformatted files has been
438       increased to 1048576. The buffer size for can now be set at runtime
439       via the environment variables GFORTRAN_FORMATTED_BUFFER_SIZE and
440       GFORTRAN_UNFORMATTED_BUFFER_SIZE for formatted and unformatted
441       files, respectively.
442     * Mismatches between actual and dummy argument lists in a single file
443       are now rejected with an error. Use the new option
444       -fallow-argument-mismatch to turn these errors into warnings; this
445       option is implied with -std=legacy. -Wargument-mismatch has been
446       removed.
447     * The handling of a BOZ literal constant has been reworked to provide
448       better conformance to the Fortran 2008 and 2018 standards. In these
449       Fortran standards, a BOZ literal constant is a typeless and
450       kindless entity. As a part of the rework, documented and
451       undocumented extensions to the Fortran standard now emit errors
452       during compilation. Some of these extensions are permitted with the
453       -fallow-invalid-boz option, which degrades the error to a warning
454       and the code is compiled as with older gfortran.
455     * At any optimization level except-Os, gfortran now uses inline
456       packing for arguments instead of calling a library routine. If the
457       source contains a large number of arguments that need to be
458       repacked, code size or time for compilation can become excessive.
459       If that is the case, -fno-inline-arg-packing can be used to disable
460       inline argument packing.
461     * Legacy extensions:
462          + For formatted input/output, if the explicit widths after the
463            data-edit descriptors I, F and G have been omitted, default
464            widths are used.
465          + A blank format item at the end of a format specification, i.e.
466            nothing following the final comma, is allowed. Use the option
467            -fdec-blank-format-item; this option is implied with -fdec.
468          + The existing support for AUTOMATIC and STATIC attributes has
469            been extended to allow variables with the AUTOMATIC attribute
470            to be used in EQUIVALENCE statements. Use -fdec-static; this
471            option is implied by -fdec.
472          + Allow character literals in assignments and DATA statements
473            for numeric (INTEGER, REAL, or COMPLEX) or LOGICAL variables.
474            Use the option -fdec-char-conversions; this option is implied
475            with -fdec.
476          + DEC comparisons, i.e. allow Hollerith constants to be used in
477            comparisons with INTEGER, REAL, COMPLEX and CHARACTER
478            expressions. Use the option -fdec.
479     * Character type names in errors and warnings now include len in
480       addition to kind; * is used for assumed length. The kind is omitted
481       if it is the default kind. Examples: CHARACTER(12), CHARACTER(6,4).
482     * CO_BROADCAST now supports derived type variables including objects
483       with allocatable components. In this case, the optional arguments
484       STAT= and ERRMSG= are currently ignored.
485     * The handling of module and submodule names has been reworked to
486       allow the full 63-character length mandated by the standard.
487       Previously symbol names were truncated if the combined length of
488       module, submodule, and function name exceeded 126 characters. This
489       change therefore breaks the ABI, but only for cases where this 126
490       character limit was exceeded.
491
492libgccjit
493
494     * The libgccjit API gained four new entry points:
495          + [39]gcc_jit_version_major, [40]gcc_jit_version_minor, and
496            [41]gcc_jit_version_patchlevel for programmatically checking
497            the libgccjit version from client code, and
498          + [42]gcc_jit_context_new_bitfield
499
500New Targets and Target Specific Improvements
501
502  AArch64 & arm
503
504     * The AArch64 and arm ports now support condition flag output
505       constraints in inline assembly, as indicated by the
506       __GCC_ASM_FLAG_OUTPUTS__. On arm this feature is only available for
507       A32 and T32 targets. Please refer to the documentation for more
508       details.
509
510  AArch64
511
512     * There have been several improvements related to the Scalable Vector
513       Extension (SVE):
514          + The SVE ACLE types and intrinsics are now supported. They can
515            be accessed using the header file arm_sve.h.
516          + It is now possible to create fixed-length SVE types using the
517            arm_sve_vector_bits attribute. For example:
518#if __ARM_FEATURE_SVE_BITS==512
519typedef svint32_t vec512 __attribute__((arm_sve_vector_bits(512)));
520typedef svbool_t pred512 __attribute__((arm_sve_vector_bits(512)));
521#endif
522          + -mlow-precision-div, -mlow-precision-sqrt and
523            -mlow-precision-recip-sqrt now work for SVE.
524          + -msve-vector-bits=128 now generates vector-length-specific
525            code for little-endian targets. It continues to generate
526            vector-length-agnostic code for big-endian targets, just as
527            previous releases did for all targets.
528          + The vectorizer is now able to use extending loads and
529            truncating stores, including gather loads and scatter stores.
530          + The vectorizer now compares the cost of vectorizing with SVE
531            and vectorizing with Advanced SIMD and tries to pick the best
532            one. Previously it would always use SVE if possible.
533          + If a vector loop uses Advanced SIMD rather than SVE, the
534            vectorizer now considers using SVE to vectorize the left-over
535            elements (the "scalar tail" or "epilog").
536          + Besides these specific points, there have been many general
537            improvements to the way that the vectorizer uses SVE.
538     * The -mbranch-protection=pac-ret option now accepts the optional
539       argument +b-key extension to perform return address signing with
540       the B-key instead of the A-key.
541     * The option -moutline-atomics has been added to aid deployment of
542       the Large System Extensions (LSE) on GNU/Linux systems built with a
543       baseline architecture targeting Armv8-A. When the option is
544       specified code is emitted to detect the presence of LSE
545       instructions at runtime and use them for standard atomic
546       operations. For more information please refer to the documentation.
547     * The Transactional Memory Extension is now supported through ACLE
548       intrinsics. It can be enabled through the +tme option extension
549       (for example, -march=armv8.5-a+tme).
550     * A number of features from Armv8.5-A are now supported through ACLE
551       intrinsics. These include:
552          + The random number instructions that can be enabled through the
553            (already present in GCC 9.1) +rng option extension.
554          + Floating-point intrinsics to round to integer instructions
555            from Armv8.5-A when targeting -march=armv8.5-a or later.
556          + Memory Tagging Extension intrinsics enabled through the
557            +memtag option extension.
558     * Similarly, the following Armv8.6-A features are now supported
559       through ACLE intrinsics:
560          + The bfloat16 extension. This extension is enabled
561            automatically when Armv8.6-A is selected (such as by
562            -march=armv8.6-a). It can also be enabled for Armv8.2-A and
563            later using the +bf16 option extension.
564          + The Matrix Multiply extension. This extension is split into
565            three parts, one for each supported data type:
566               o Support for 8-bit integer matrix multiply instructions.
567                 This extension is enabled automatically when Armv8.6-A is
568                 selected. It can also be enabled for Armv8.2-A and later
569                 using the +i8mm option extension.
570               o Support for 32-bit floating-point matrix multiply
571                 instructions. This extension can be enabled using the
572                 +f32mm option extension, which also has the effect of
573                 enabling SVE.
574               o Support for 64-bit floating-point matrix multiply
575                 instructions. This extension can be enabled using the
576                 +f64mm option extension, which likewise has the effect of
577                 enabling SVE.
578     * SVE2 is now supported through ACLE intrinsics and (to a limited
579       extent) through autovectorization. It can be enabled through the
580       +sve2 option extension (for example, -march=armv8.5-a+sve2).
581       Additional extensions can be enabled through +sve2-sm4, +sve2-aes,
582       +sve2-sha3 and +sve2-bitperm.
583     * Support has been added for the following processors (GCC
584       identifiers in parentheses):
585          + Arm Cortex-A77 (cortex-a77).
586          + Arm Cortex-A76AE (cortex-a76ae).
587          + Arm Cortex-A65 (cortex-a65).
588          + Arm Cortex-A65AE (cortex-a65ae).
589          + Arm Cortex-A34 (cortex-a34).
590          + Marvell ThunderX3 (thunderx3t110).
591       The GCC identifiers can be used as arguments to the -mcpu or -mtune
592       options, for example: -mcpu=cortex-a77 or -mtune=cortex-a65ae or as
593       arguments to the equivalent target attributes and pragmas.
594
595  arm
596
597     * Support for the FDPIC ABI has been added. It uses 64-bit function
598       descriptors to represent pointers to functions, and enables code
599       sharing on MMU-less systems. The corresponding target triple is
600       arm-uclinuxfdpiceabi, and the C library is uclibc-ng.
601     * Support has been added for the Arm EABI on NetBSD through the
602       arm*-*-netbsdelf-*eabi* triplet.
603     * The handling of 64-bit integer operations has been significantly
604       reworked and improved leading to improved performance and reduced
605       stack usage when using 64-bit integral data types. The option
606       -mneon-for-64bits is now deprecated and will be removed in a future
607       release.
608     * Support has been added for the following processors (GCC
609       identifiers in parentheses):
610          + Arm Cortex-A77 (cortex-a77).
611          + Arm Cortex-A76AE (cortex-a76ae).
612          + Arm Cortex-M35P (cortex-m35p).
613          + Arm Cortex-M55 (cortex-m55).
614       The GCC identifiers can be used as arguments to the -mcpu or -mtune
615       options, for example: -mcpu=cortex-a77 or -mtune=cortex-m35p.
616     * Support has been extended for the ACLE [43]data-processing
617       intrinsics to include 32-bit SIMD, saturating arithmetic, 16-bit
618       multiplication and other related intrinsics aimed at DSP algorithm
619       optimization.
620     * Support for -mpure-code in Thumb-1 (v6m) has been added: this
621       M-profile feature is no longer restricted to targets with MOVT. For
622       example, -mcpu=cortex-m0 now supports this option.
623     * Support for the [44]Armv8.1-M Mainline Architecture has been added.
624          + Armv8.1-M Mainline can be enabled by using the
625            -march=armv8.1-m.main command-line option.
626     * Support for the [45]MVE beta ACLE intrinsics has been added. These
627       intrinsics can be enabled by including the arm_mve.h header file
628       and passing the +mve or +mve.fp option extensions (for example:
629       -march=armv8.1-m.main+mve).
630     * Support for the Custom Datapath Extension beta ACLE [46]intrinsics
631       has been added.
632     * Support for Armv8.1-M Mainline Security Extensions architecture has
633       been added. The -mcmse option, when used in combination with an
634       Armv8.1-M Mainline architecture (for example: -march=armv8.1-m.main
635       -mcmse), now leads to the generation of improved code sequences
636       when changing security states.
637
638  AMD Radeon (GCN)
639
640     * The code generation and in particular the vectorization support has
641       been much improved.
642
643  ARC
644
645     * The interrupt service routine functions save all used registers,
646       including extension registers and auxiliary registers used by Zero
647       Overhead Loops.
648     * Improve code size by using multiple short instructions instead of a
649       single long mov or ior instruction when its long immediate constant
650       is known.
651     * Fix usage of the accumulator register for ARC600.
652     * Fix issues with uncached attribute.
653     * Remove -mq-class option.
654     * Improve 64-bit integer addition and subtraction operations.
655
656  AVR
657
658     * Support for the XMEGA-like devices
659
660     ATtiny202, ATtiny204, ATtiny402, ATtiny404, ATtiny406, ATtiny804,
661     ATtiny806, ATtiny807, ATtiny1604, ATtiny1606, ATtiny1607, ATmega808,
662     ATmega809, ATmega1608, ATmega1609, ATmega3208, ATmega3209,
663     ATmega4808, ATmega4809
664       has been added.
665     * A new command-line option -nodevicespecs has been added. It allows
666       to provide a custom device-specs file by means of
667
668     avr-gcc -nodevicespecs -specs=my-spec-file <options>
669       and without the need to provide options -B and -mmcu=. See [47]AVR
670       command-line options for details. This feature is also available in
671       GCC 9.3+ and GCC 8.4+.
672     * New command-line options -mdouble=[32,64] and -mlong-double=[32,64]
673       have been added. They allow to choose the size (in bits) of the
674       double and long double types, respectively. Whether or not the
675       mentioned layouts are available, whether the options act as a
676       multilib option, and the default for either option are controlled
677       by the new [48]AVR configure options --with-double= and
678       --with-long-double=.
679     * A new configure option --with-libf7= has been added. It controls to
680       which level avr-libgcc provides 64-bit floating point support by
681       means of [49]Libf7.
682     * A new configure option --with-double-comparison= has been added.
683       It's unlikely you need to set this option by hand.
684
685  IA-32/x86-64
686
687     * Support to expand __builtin_roundeven into the appropriate SSE 4.1
688       instruction has been added.
689     * New ISA extension support for Intel ENQCMD was added to GCC. ENQCMD
690       intrinsics are available via the -menqcmd compiler switch.
691     * GCC now supports the Intel CPU named Cooperlake through
692       -march=cooperlake. The switch enables the AVX512BF16 ISA
693       extensions.
694     * GCC now supports the Intel CPU named Tigerlake through
695       -march=tigerlake. The switch enables the MOVDIRI MOVDIR64B
696       AVX512VP2INTERSECT ISA extensions.
697
698  MIPS
699
700     * The mips*-*-linux* targets now mark object files with appropriate
701       GNU-stack note, facilitating use of non-executable stack hardening
702       on GNU/Linux. The soft-float targets have this feature enabled by
703       default, while for hard-float targets it is required for GCC to be
704       configured with --with-glibc-version=2.31 against glibc 2.31 or
705       later.
706
707  PowerPC / PowerPC64 / RS6000
708
709     * Many vector builtins have been listed as deprecated in the
710       [50]64-Bit ELF V2 ABI Specification for quite a number of years.
711       The vector builtins listed in Tables A.8 through A.10 are now
712       deprecated for GCC 10, and will likely be removed from support in
713       GCC 11. Note that this does not result in any loss of function.
714       These deprecated builtins generally provide somewhat nonsensical
715       argument lists (for example, mixing signed, unsigned, and bool
716       vector arguments arbitrarily), or are duplicate builtins that are
717       inconsistent with the expected naming scheme. We expect that this
718       will be unlikely to affect much if any code, and any required code
719       changes will be trivial.
720
721  PRU
722
723     * A new back end targeting TI PRU I/O processors has been contributed
724       to GCC.
725
726  RISC-V
727
728     * The riscv*-*-* targets now require GNU binutils version 2.30 or
729       later, to support new assembly instructions produced by GCC.
730
731  V850
732
733     * The ABI for V850 nested functions has been changed. Previously the
734       V850 port used %r20 for the static chain pointer, now the port uses
735       %r19. This corrects a long standing latent bug in the v850 port
736       where a call to a nested function would unexpectedly change the
737       value in %r20.
738
739Operating Systems
740
741Improvements for plugin authors
742
743     * GCC diagnostics can now have a chain of events associated with
744       them, describing a path through the code that triggers the problem.
745       These can be printed by the diagnostics subsystem in various ways,
746       controlled by the [51]-fdiagnostics-path-format option, or captured
747       in JSON form via [52]-fdiagnostics-format=json.
748     * GCC diagnostics can now be associated with [53]CWE weakness
749       identifiers, which will appear on the standard error stream, and in
750       the JSON output from [54]-fdiagnostics-format=json.
751
752Other significant improvements
753
754     * To allow inline expansion of both memcpy and memmove, the existing
755       movmem instruction patterns used for non-overlapping memory copies
756       have been renamed to cpymem. The movmem name is now used for
757       overlapping memory moves, consistent with the library functions
758       memcpy and memmove.
759     * For many releases, when GCC emits a warning it prints the option
760       controlling that warning. As of GCC 10, that option text is now a
761       clickable hyperlink for the documentation of that option (assuming
762       a [55]sufficiently capable terminal). This behavior can be
763       controlled via a new [56]-fdiagnostics-urls option (along with
764       various environment variables and heuristics documented with that
765       option).
766
767GCC 10.1
768
769   This is the [57]list of problem reports (PRs) from GCC's bug tracking
770   system that are known to be fixed in the 10.1 release. This list might
771   not be complete (that is, it is possible that some PRs that have been
772   fixed are not listed here).
773
774GCC 10.2
775
776   This is the [58]list of problem reports (PRs) from GCC's bug tracking
777   system that are known to be fixed in the 10.2 release. This list might
778   not be complete (that is, it is possible that some PRs that have been
779   fixed are not listed here).
780
781GCC 10.3
782
783   This is the [59]list of problem reports (PRs) from GCC's bug tracking
784   system that are known to be fixed in the 10.3 release. This list might
785   not be complete (that is, it is possible that some PRs that have been
786   fixed are not listed here).
787
788  Target Specific Changes
789
790    AArch64
791
792     * A bug with the Random Number intrinsics in the arm_acle.h header
793       that resulted in an incorrect status result being returned has been
794       fixed.
795     * GCC now supports the Fujitsu A64FX. The associated -mcpu and -mtune
796       options are -mcpu=a64fx and -mtune=a64fx respectively. In
797       particular, -mcpu=a64fx generates code for Armv8.2-A with SVE and
798       tunes the code for the A64FX. This includes tuning the SVE code,
799       although by default the code is still length-agnostic and so works
800       for all SVE implementations. Adding -msve-vector-bits=512 makes the
801       code specific to 512-bit SVE.
802
803
804    For questions related to the use of GCC, please consult these web
805    pages and the [60]GCC manuals. If that fails, the
806    [61]gcc-help@gcc.gnu.org mailing list might help. Comments on these
807    web pages and the development of GCC are welcome on our developer
808    list at [62]gcc@gcc.gnu.org. All of [63]our lists have public
809    archives.
810
811   Copyright (C) [64]Free Software Foundation, Inc. Verbatim copying and
812   distribution of this entire article is permitted in any medium,
813   provided this notice is preserved.
814
815   These pages are [65]maintained by the GCC team. Last modified
816   2021-04-08[66].
817
818References
819
820   1. http://gcc.gnu.org/gcc-10/porting_to.html
821   2. http://gcc.gnu.org/onlinedocs/index.html#current
822   3. http://gcc.gnu.org/gcc-10/changes.html#empty_base
823   4. https://www.mpfr.org/
824   5. http://gcc.gnu.org/gcc-10/changes.html#iterator_base
825   6. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/cpp/_005f_005fhas_005fbuiltin.html#g_t_005f_005fhas_005fbuiltin
826   7. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Optimize-Options.html#index-fno-allocation-dce
827   8. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Optimize-Options.html#index-fprofile-partial-training
828   9. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Instrumentation-Options.html#index-fprofile-reproducible
829  10. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Instrumentation-Options.html#index-fprofile-generate
830  11. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Instrumentation-Options.html#index-fprofile-prefix-path
831  12. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Static-Analyzer-Options.html
832  13. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Optimize-Options.html#index-finline-functions
833  14. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/lto-dump.html
834  15. https://facebook.github.io/zstd/
835  16. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Optimize-Options.html#index-fprofile-values
836  17. https://www.openacc.org/
837  18. https://gcc.gnu.org/wiki/OpenACC#status-10
838  19. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/libgomp/#toc-Enabling-OpenACC-1
839  20. https://www.openmp.org/specifications/
840  21. https://gcc.gnu.org/wiki/Offloading
841  22. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Warning-Options.html#index-Wstring-compare
842  23. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Warning-Options.html#index-Wzero-length-bounds
843  24. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Warning-Options.html#index-Warray-bounds
844  25. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Warning-Options.html#index-Wformat-overflow
845  26. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Warning-Options.html#index-Wrestrict
846  27. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Warning-Options.html#index-Wreturn-local-addr
847  28. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Warning-Options.html#index-Wstringop-overflow
848  29. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Warning-Options.html#index-Warith-conversion
849  30. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/C_002b_002b-Dialect-Options.html#index-Wmismatched-tags
850  31. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/C_002b_002b-Dialect-Options.html#index-Wredundant-tags
851  32. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94383
852  33. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94711
853  34. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94707
854  35. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94704
855  36. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94706
856  37. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92285
857  38. https://www.openmp.org/specifications/
858  39. https://gcc.gnu.org/onlinedocs/jit/topics/compatibility.html#c.gcc_jit_version_major
859  40. https://gcc.gnu.org/onlinedocs/jit/topics/compatibility.html#c.gcc_jit_version_minor
860  41. https://gcc.gnu.org/onlinedocs/jit/topics/compatibility.html#c.gcc_jit_version_patchlevel
861  42. https://gcc.gnu.org/onlinedocs/jit/topics/types.html#c.gcc_jit_context_new_bitfield
862  43. https://developer.arm.com/documentation/101028/0009/Data-processing-intrinsics
863  44. https://developer.arm.com/architectures/cpu-architecture/m-profile
864  45. https://developer.arm.com/architectures/instruction-sets/simd-isas/helium/helium-intrinsics
865  46. https://developer.arm.com/documentation/101028/0010/Custom-Datapath-Extension
866  47. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/AVR-Options.html#index-nodevicespecs
867  48. https://gcc.gnu.org/install/configure.html#avr
868  49. https://gcc.gnu.org/wiki/avr-gcc#Libf7
869  50. https://openpowerfoundation.org/?resource_lib=64-bit-elf-v2-abi-specification-power-architecture
870  51. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Diagnostic-Message-Formatting-Options.html#index-fdiagnostics-path-format
871  52. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Diagnostic-Message-Formatting-Options.html#index-fdiagnostics-format
872  53. https://cwe.mitre.org/
873  54. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Diagnostic-Message-Formatting-Options.html#index-fdiagnostics-format
874  55. https://gist.github.com/egmontkob/eb114294efbcd5adb1944c9f3cb5feda
875  56. https://gcc.gnu.org/onlinedocs/gcc-10.1.0/gcc/Diagnostic-Message-Formatting-Options.html#index-fdiagnostics-urls
876  57. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=10.0
877  58. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=10.2
878  59. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=10.3
879  60. https://gcc.gnu.org/onlinedocs/
880  61. mailto:gcc-help@gcc.gnu.org
881  62. mailto:gcc@gcc.gnu.org
882  63. https://gcc.gnu.org/lists.html
883  64. https://www.fsf.org/
884  65. https://gcc.gnu.org/about.html
885  66. http://validator.w3.org/check/referer
886======================================================================
887http://gcc.gnu.org/gcc-9/index.html
888
889                              GCC 9 Release Series
890
891   Mar 12, 2020
892
893   The [1]GNU project and the GCC developers are pleased to announce the
894   release of GCC 9.3.
895
896   This release is a bug-fix release, containing fixes for regressions in
897   GCC 9.2 relative to previous releases of GCC.
898
899Release History
900
901   GCC 9.3
902          Mar 12, 2020 ([2]changes, [3]documentation)
903
904   GCC 9.2
905          Aug 12, 2019 ([4]changes, [5]documentation)
906
907   GCC 9.1
908          May 3, 2019 ([6]changes, [7]documentation)
909
910References and Acknowledgements
911
912   GCC used to stand for the GNU C Compiler, but since the compiler
913   supports several other languages aside from C, it now stands for the
914   GNU Compiler Collection.
915
916   A list of [8]successful builds is updated as new information becomes
917   available.
918
919   The GCC developers would like to thank the numerous people that have
920   contributed new features, improvements, bug fixes, and other changes as
921   well as test results to GCC. This [9]amazing group of volunteers is
922   what makes GCC successful.
923
924   For additional information about GCC please refer to the [10]GCC
925   project web site or contact the [11]GCC development mailing list.
926
927   To obtain GCC please use [12]our mirror sites or [13]our version
928   control system.
929
930
931    For questions related to the use of GCC, please consult these web
932    pages and the [14]GCC manuals. If that fails, the
933    [15]gcc-help@gcc.gnu.org mailing list might help. Comments on these
934    web pages and the development of GCC are welcome on our developer
935    list at [16]gcc@gcc.gnu.org. All of [17]our lists have public
936    archives.
937
938   Copyright (C) [18]Free Software Foundation, Inc. Verbatim copying and
939   distribution of this entire article is permitted in any medium,
940   provided this notice is preserved.
941
942   These pages are [19]maintained by the GCC team. Last modified
943   2020-03-12[20].
944
945References
946
947   1. http://www.gnu.org/
948   2. http://gcc.gnu.org/gcc-9/changes.html
949   3. http://gcc.gnu.org/onlinedocs/9.3.0/
950   4. http://gcc.gnu.org/gcc-9/changes.html
951   5. http://gcc.gnu.org/onlinedocs/9.2.0/
952   6. http://gcc.gnu.org/gcc-9/changes.html
953   7. http://gcc.gnu.org/onlinedocs/9.1.0/
954   8. http://gcc.gnu.org/gcc-9/buildstat.html
955   9. http://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Contributors.html
956  10. http://gcc.gnu.org/index.html
957  11. mailto:gcc@gcc.gnu.org
958  12. http://gcc.gnu.org/mirrors.html
959  13. http://gcc.gnu.org/git.html
960  14. https://gcc.gnu.org/onlinedocs/
961  15. mailto:gcc-help@gcc.gnu.org
962  16. mailto:gcc@gcc.gnu.org
963  17. https://gcc.gnu.org/lists.html
964  18. https://www.fsf.org/
965  19. https://gcc.gnu.org/about.html
966  20. http://validator.w3.org/check/referer
967======================================================================
968http://gcc.gnu.org/gcc-9/changes.html
969
970                              GCC 9 Release Series
971                        Changes, New Features, and Fixes
972
973   This page is a "brief" summary of some of the huge number of
974   improvements in GCC 9. You may also want to check out our [1]Porting to
975   GCC 9 page and the [2]full GCC documentation.
976
977Caveats
978
979     * On Arm targets (arm*-*-*), [3]a bug in the implementation of the
980       procedure call standard (AAPCS) in the GCC 6, 7 and 8 releases has
981       been fixed: a structure containing a bit-field based on a 64-bit
982       integral type and where no other element in a structure required
983       64-bit alignment could be passed incorrectly to functions. This is
984       an ABI change. If the option -Wpsabi is enabled (on by default) the
985       compiler will emit a diagnostic note for code that might be
986       affected.
987     * Support for a number of older systems and recently unmaintained or
988       untested target ports of GCC has been declared obsolete in GCC 9.
989       Unless there is activity to revive them, the next release of GCC
990       will have their sources permanently removed.
991       The following ports for individual systems on particular
992       architectures have been obsoleted:
993          + Solaris 10 (*-*-solaris2.10). Details can be found in the
994            [4]announcement.
995          + Cell Broadband Engine SPU (spu*-*-*). Details can be found in
996            the [5]announcement.
997     * A change to the C++ std::rotate algorithm in GCC 9.1.0 can cause
998       ABI incompatibilities with object files compiled with other
999       versions of GCC. If the std::rotate algorithm is called with an
1000       empty range then it might cause a divide-by-zero error (as a SIGFPE
1001       signal) and crash. The change has been reverted for GCC 9.2.0 and
1002       future releases. For more details see [6]Bug 90920. The problem can
1003       be avoided by recompiling any objects that might call std::rotate
1004       with an empty range, so that the GCC 9.1.0 definition of
1005       std::rotate is not used.
1006     * The automatic template instantiation at link time ([7]-frepo) has
1007       been deprecated and will be removed in a future release.
1008     * The --with-default-libstdcxx-abi=gcc4-compatible configure option
1009       is broken in the 9.1 and 9.2 releases, producing a shared library
1010       with missing symbols (see [8]Bug 90361). As a workaround, configure
1011       without that option and build GCC as normal, then edit the
1012       installed <bits/c++config.h> headers to define the
1013       _GLIBCXX_USE_CXX11_ABI macro to 0.
1014
1015General Improvements
1016
1017   The following GCC command line options have been introduced or
1018   improved.
1019     * All command line options that take a byte-size argument accept
1020       64-bit integers as well as standard SI and IEC suffixes such as kb
1021       and KiB, MB and MiB, or GB and GiB denoting the corresponding
1022       multiples of bytes. See [9]Invoking GCC for more.
1023     * A new option,
1024       [10]-flive-patching=[inline-only-static|inline-clone], has been
1025       introduced to provide a safe compilation for live-patching. At the
1026       same time, provides multiple-level control on the enabled IPA
1027       optimizations. See the user guide for more details about the
1028       option.
1029     * A new option, --completion, has been added to provide more fine
1030       option completion in a shell. It is intended to be used by
1031       Bash-completion.
1032     * GCC's diagnostics now print source code with a left margin showing
1033       line numbers, configurable with
1034       [11]-fno-diagnostics-show-line-numbers.
1035       GCC's diagnostics can also now label regions of the source code to
1036       show pertinent information, such as the types within an expression.
1037$ g++ t.cc
1038t.cc: In function 'int test(const shape&, const shape&)':
1039t.cc:15:4: error: no match for 'operator+' (operand types are 'boxed_value<doubl
1040e>' and 'boxed_value<double>')
1041   14 |   return (width(s1) * height(s1)
1042      |           ~~~~~~~~~~~~~~~~~~~~~~
1043      |                     |
1044      |                     boxed_value<[...]>
1045   15 |    + width(s2) * height(s2));
1046      |    ^ ~~~~~~~~~~~~~~~~~~~~~~
1047      |                |
1048      |                boxed_value<[...]>
1049
1050       These labels can be disabled via [12]-fno-diagnostics-show-labels.
1051     * A new option [13]-fdiagnostics-format=json has been introduced for
1052       emitting diagnostics in a machine-readable format.
1053     * The alignment-related options [14]-falign-functions,
1054       [15]-falign-labels, [16]-falign-loops, and [17]-falign-jumps
1055       received support for a secondary alignment (e.g.
1056       -falign-loops=n:m:n2:m2).
1057     * New pair of profiling options ([18]-fprofile-filter-files and
1058       [19]-fprofile-exclude-files) has been added. The options help to
1059       filter which source files are instrumented.
1060     * AddressSanitizer generates more compact redzones for automatic
1061       variables. That helps to reduce memory footprint of a sanitized
1062       binary.
1063     * Numerous improvements have been made to the output of
1064       [20]-fopt-info.
1065       Messages are now prefixed with optimized, missed, or note, rather
1066       than the old behavior of all being prefixed with note.
1067       The output from -fopt-info can now contain information on inlining
1068       decisions:
1069$ g++ -c inline.cc -O2 -fopt-info-inline-all
1070inline.cc:24:11: note: Considering inline candidate void foreach(T, T, void (*)(
1071E)) [with T = char**; E = char*]/2.
1072inline.cc:24:11: optimized:  Inlining void foreach(T, T, void (*)(E)) [with T =
1073char**; E = char*]/2 into int main(int, char**)/1.
1074inline.cc:19:12: missed:   not inlinable: void inline_me(char*)/0 -> int std::pu
1075ts(const char*)/3, function body not available
1076inline.cc:13:8: optimized:  Inlined void inline_me(char*)/4 into int main(int, c
1077har**)/1 which now has time 127.363637 and size 11, net change of +0.
1078Unit growth for small function inlining: 16->16 (0%)
1079
1080Inlined 2 calls, eliminated 1 functions
1081
1082
1083       The output from the vectorizer has been rationalized so that failed
1084       attempts to vectorize a loop are displayed in the form
1085    [LOOP-LOCATION]: couldn't vectorize this loop
1086    [PROBLEM-LOCATION]: because of [REASON]
1087
1088       rather than an exhaustive log of all decisions made by the
1089       vectorizer. For example:
1090$ gcc -c v.c -O3 -fopt-info-all-vec
1091v.c:7:3: missed: couldn't vectorize loop
1092v.c:10:7: missed: statement clobbers memory: __asm__ __volatile__("" :  :  : "me
1093mory");
1094v.c:3:6: note: vectorized 0 loops in function.
1095v.c:10:7: missed: statement clobbers memory: __asm__ __volatile__("" :  :  : "me
1096mory");
1097
1098       The old behavior can be obtained via a new -internals suboption of
1099       -fopt-info.
1100     * A new option, [21]-fsave-optimization-record has been added, which
1101       writes a SRCFILE.opt-record.json.gz file describing the
1102       optimization decisions made by GCC. This is similar to the output
1103       of -fopt-info, but with additional metadata such as the inlining
1104       chain, and profile information (if available).
1105     * Inter-procedural propagation of stack alignment can now be
1106       controlled by [22]-fipa-stack-alignment.
1107     * Propagation of addressability, readonly and writeonly flags on
1108       static variables can now be controlled by
1109       [23]-fipa-reference-addressable.
1110
1111   The following built-in functions have been introduced.
1112     * [24]__builtin_expect_with_probability to provide branch prediction
1113       probability hints to the optimizer.
1114     * [25]__builtin_has_attribute determines whether a function, type, or
1115       variable has been declared with some attribute.
1116     * [26]__builtin_speculation_safe_value can be used to help mitigate
1117       against unsafe speculative execution.
1118
1119   The following attributes have been introduced.
1120     * The [27]copy function attribute has been added. The attribute can
1121       also be applied to type definitions and to variable declarations.
1122
1123   A large number of improvements to code generation have been made,
1124   including but not limited to the following.
1125     * Switch expansion has been improved by using a different strategy
1126       (jump table, bit test, decision tree) for a subset of switch cases.
1127     * A linear function expression defined as a switch statement can be
1128       transformed by [28]-ftree-switch-conversion. For example:
1129
1130int
1131foo (int how)
1132{
1133  switch (how) {
1134    case 2: how = 205; break;
1135    case 3: how = 305; break;
1136    case 4: how = 405; break;
1137    case 5: how = 505; break;
1138    case 6: how = 605; break;
1139  }
1140  return how;
1141}
1142
1143       can be transformed into 100 * how + 5 (for values defined in the
1144       switch statement).
1145     * Inter-procedural optimization improvements:
1146          + Inliner defaults was tuned to better suits modern C++
1147            codebases especially when built with link time optimizations.
1148            New parameters max-inline-insns-small, max-inline-insns-size,
1149            uninlined-function-insns, uninlined-function-time,
1150            uninlined-thunk-insns, and uninlined-thunk-time were added.
1151          + Hot/cold partitioning is now more precise and aggressive.
1152          + Improved scalability for very large translation units
1153            (especially when link-time optimizing large programs).
1154     * Profile driven optimization improvements:
1155          + [29]-fprofile-use now enables [30]-fversion-loops-for-strides,
1156            [31]-floop-interchange, [32]-floop-unroll-and-jam,
1157            [33]-ftree-loop-distribution.
1158          + Streaming of counter histograms was removed. This reduces the
1159            size of profile files. Histogram is computed on the fly with
1160            link-time optimization. Parameter hot-bb-count-ws-permille was
1161            reduced from 999 to 990 to account for more precise
1162            histograms.
1163     * Link-time optimization improvements:
1164          + Types are now simplified prior streaming resulting in
1165            significant reductions of the LTO object files, link-time
1166            memory use, and improvements of link-time parallelism.
1167          + Default number of partitions (--param lto-partitions) was
1168            increased from 32 to 128 enabling effective use of CPUs with
1169            more than 32 hyperthreads. --param
1170            lto-max-streaming-parallelism can now be used to control
1171            number of streaming processes.
1172          + Warnings on C++ One Decl Rule violations (-Wodr) are now more
1173            informative and produce fewer redundant results.
1174       Overall compile time of Firefox 66 and LibreOffice 6.2.3 on an
1175       8-core machine was reduced by about 5% compared to GCC 8.3, and the
1176       size of LTO object files by 7%. LTO link-time improves by 11% on an
1177       8-core machine and scales significantly better for more parallel
1178       build environments. The serial stage of the link-time optimization
1179       is 28% faster consuming 20% less memory. The parallel stage now
1180       scales to up to 128 partitions rather than 32 and reduces memory
1181       use for every worker by 30%.
1182
1183   The following improvements to the gcov command-line utility have been
1184   made.
1185     * The gcov tool received a new option [34]--use-hotness-colors (-q)
1186       that can provide perf-like coloring of hot functions.
1187     * The gcov tool has changed its intermediate format to a new JSON
1188       format.
1189
1190New Languages and Language specific improvements
1191
1192   [35]OpenACC support in C, C++, and Fortran continues to be maintained
1193   and improved. Most of the OpenACC 2.5 specification is implemented. See
1194   the [36]implementation status section on the OpenACC wiki page for
1195   further information.
1196
1197  C family
1198
1199     * Version 5.0 of the [37]OpenMP specification is now partially
1200       supported in the C and C++ compilers. For details which features of
1201       OpenMP 5.0 are and which are not supported in the GCC 9 release see
1202       [38]this mail.
1203     * New extensions:
1204          + [39]__builtin_convertvector built-in for vector conversions
1205            has been added.
1206     * New warnings:
1207          + [40]-Waddress-of-packed-member, enabled by default, warns
1208            about an unaligned pointer value from the address of a packed
1209            member of a struct or union.
1210     * Enhancements to existing warnings:
1211          + [41]-Warray-bounds detects more instances of out-of-bounds
1212            indices.
1213          + [42]-Wattribute-alias also detects attribute mismatches
1214            between alias declarations and their targets, in addition to
1215            mismatches between their types.
1216          + [43]-Wformat-overflow and [44]-Wformat-truncation have been
1217            extended to all formatted input/output functions (where
1218            applicable) and enhanced to detect a subset of instances of
1219            reading past the end of unterminated constant character arrays
1220            in %s directives.
1221          + [45]-Wmissing-attributes detects instances of missing function
1222            attributes on declarations of aliases and weak references.
1223          + [46]-Wstringop-truncation also detects a subset of instances
1224            of reading past the end of unterminated constant character
1225            arrays,
1226     * If a macro is used with the wrong argument count, the C and C++
1227       front ends now show the definition of that macro via a note.
1228     * The spelling corrector now considers transposed letters, and the
1229       threshold for similarity has been tightened, to avoid nonsensical
1230       suggestions.
1231
1232  C
1233
1234     * There is now experimental support for -std=c2x, to select support
1235       for the upcoming C2X revision of the ISO C standard. This standard
1236       is in the early stages of development and the only feature
1237       supported in GCC 9 is _Static_assert with a single argument
1238       (support for _Static_assert with two arguments was added in C11 and
1239       GCC 4.6). There are also new options -std=gnu2x, for C2X with GNU
1240       extensions, and -Wc11-c2x-compat, to warn for uses of features
1241       added in C2X (such warnings are also enabled by use of -Wpedantic
1242       if not using -std=c2x or -std=gnu2x).
1243     * New warnings:
1244          + [47]-Wabsolute-value warns for calls to standard functions
1245            that compute the absolute value of an argument when a more
1246            appropriate standard function is available. For example,
1247            calling abs(3.14) triggers the warning because the appropriate
1248            function to call to compute the absolute value of a double
1249            argument is fabs. The option also triggers warnings when the
1250            argument in a call to such a function has an unsigned type.
1251            This warning can be suppressed with an explicit type cast and
1252            it is also enabled by -Wextra.
1253
1254  C++
1255
1256     * New warnings:
1257          + [48]-Wdeprecated-copy, implied by -Wextra, warns about the
1258            C++11 deprecation of implicitly declared copy constructor and
1259            assignment operator if one of them is user-provided.
1260            -Wdeprecated-copy-dtor also warns if the destructor is
1261            user-provided, as specified in C++11.
1262          + [49]-Winit-list-lifetime, on by default, warns about uses of
1263            std::initializer_list that are likely to result in a dangling
1264            pointer, such as returning or assigning from a temporary list.
1265          + [50]-Wredundant-move, implied by -Wextra, warns about
1266            redundant calls to std::move.
1267          + [51]-Wpessimizing-move, implied by -Wall, warns when a call to
1268            std::move prevents copy elision.
1269          + [52]-Wclass-conversion, on by default, warns when a conversion
1270            function will never be called due to the type it converts to.
1271     * The C++ front end has experimental support for some of the upcoming
1272       C++2a draft features with the -std=c++2a or -std=gnu++2a flags,
1273       including range-based for statements with initializer, default
1274       constructible and assignable stateless lambdas, lambdas in
1275       unevaluated contexts, language support for empty data members,
1276       allowing pack expansion in lambda init-capture, likely and unlikely
1277       attributes, class types in non-type template parameters, allowing
1278       virtual function calls in constant expressions, explicit(bool),
1279       std::is_constant_evaluated, nested inline namespaces, etc. For a
1280       full list of new features, see [53]the C++ status page.
1281     * The C++ front end now preserves source locations for literals,
1282       id-expression, and mem-initializer for longer. For example it is
1283       now able to pin-point the pertinent locations for bad
1284       initializations such as these
1285$ g++ -c bad-inits.cc
1286bad-inits.cc:10:14: error: cannot convert 'json' to 'int' in initialization
1287   10 |   { 3, json::object },
1288      |        ~~~~~~^~~~~~
1289      |              |
1290      |              json
1291bad-inits.cc:14:31: error: initializer-string for array of chars is too long [-f
1292permissive]
1293   14 | char buffers[3][5] = { "red", "green", "blue" };
1294      |                               ^~~~~~~
1295bad-inits.cc: In constructor 'X::X()':
1296bad-inits.cc:17:13: error: invalid conversion from 'int' to 'void*' [-fpermissiv
1297e]
1298   17 |   X() : one(42), two(42), three(42)
1299      |             ^~
1300      |             |
1301      |             int
1302
1303       rather than emitting the error at the final closing parenthesis or
1304       brace.
1305     * Error-reporting of overload resolution has been special-cased to
1306       make the case of a single failed candidate easier to read. For
1307       example:
1308$ g++ param-type-mismatch.cc
1309param-type-mismatch.cc: In function 'int test(int, const char*, float)':
1310param-type-mismatch.cc:8:32: error: cannot convert 'const char*' to 'const char*
1311*'
1312    8 |   return foo::member_1 (first, second, third);
1313      |                                ^~~~~~
1314      |                                |
1315      |                                const char*
1316param-type-mismatch.cc:3:46: note:   initializing argument 2 of 'static int foo:
1317:member_1(int, const char**, float)'
1318    3 |   static int member_1 (int one, const char **two, float three);
1319      |                                 ~~~~~~~~~~~~~^~~
1320
1321       highlights both the problematic argument, and the parameter that it
1322       can't be converted to.
1323     * Diagnostics involving binary operators now use color to distinguish
1324       the two operands, and label them separately (as per the example of
1325       source labelling above).
1326     * Diagnostics involving function calls now highlight the pertinent
1327       parameter of the declaration in more places.
1328$ g++ bad-conversion.cc
1329bad-conversion.cc: In function 'void caller()':
1330bad-conversion.cc:9:14: error: cannot convert 'bool' to 'void*'
1331    9 |   callee (0, false, 2);
1332      |              ^~~~~
1333      |              |
1334      |              bool
1335bad-conversion.cc:3:19: note:   initializing argument 2 of 'void callee(int, voi
1336d*, int)'
1337    3 | void callee (int, void *, int)
1338      |                   ^~~~~~
1339
1340     * The C++ front end's implementation of [54]-Wformat now shows
1341       precise locations within string literals, and underlines the
1342       pertinent arguments at bogus call sites (the C front end has been
1343       doing this since GCC 7). For example:
1344$ g++ -c bad-printf.cc -Wall
1345bad-printf.cc: In function 'void print_field(const char*, float, long int, long
1346int)':
1347bad-printf.cc:6:17: warning: field width specifier '*' expects argument of type
1348'int', but argument 3 has type 'long int' [-Wformat=]
1349    6 |   printf ("%s: %*ld ", fieldname, column - width, value);
1350      |                ~^~~               ~~~~~~~~~~~~~~
1351      |                 |                        |
1352      |                 int                      long int
1353bad-printf.cc:6:19: warning: format '%ld' expects argument of type 'long int', b
1354ut argument 4 has type 'double' [-Wformat=]
1355    6 |   printf ("%s: %*ld ", fieldname, column - width, value);
1356      |                ~~~^                               ~~~~~
1357      |                   |                               |
1358      |                   long int                        double
1359      |                %*f
1360
1361     * The C++ front end has gained new fix-it hints for forgetting the
1362       return *this; needed by various C++ operators:
1363$ g++ -c operator.cc
1364operator.cc: In member function 'boxed_ptr& boxed_ptr::operator=(const boxed_ptr
1365&)':
1366operator.cc:7:3: warning: no return statement in function returning non-void [-W
1367return-type]
1368    6 |     m_ptr = other.m_ptr;
1369  +++ |+    return *this;
1370    7 |   }
1371      |   ^
1372
1373       for when the compiler needs a typename:
1374$ g++ -c template.cc
1375template.cc:3:3: error: need 'typename' before 'Traits::type' because 'Traits' i
1376s a dependent scope
1377    3 |   Traits::type type;
1378      |   ^~~~~~
1379      |   typename
1380
1381       when trying to use an accessor member as if it were a data member:
1382$ g++ -c fncall.cc
1383fncall.cc: In function 'void hangman(const mystring&)':
1384fncall.cc:12:11: error: invalid use of member function 'int mystring::get_length
1385() const' (did you forget the '()' ?)
1386   12 |   if (str.get_length > 0)
1387      |       ~~~~^~~~~~~~~~
1388      |                     ()
1389
1390       for C++11's scoped enums:
1391$ g++ -c enums.cc
1392enums.cc: In function 'void json::test(const json::value&)':
1393enums.cc:12:26: error: 'STRING' was not declared in this scope; did you mean 'js
1394on::kind::STRING'?
1395   12 |     if (v.get_kind () == STRING)
1396      |                          ^~~~~~
1397      |                          json::kind::STRING
1398enums.cc:3:44: note: 'json::kind::STRING' declared here
1399    3 |   enum class kind { OBJECT, ARRAY, NUMBER, STRING, TRUE, FALSE, NULL_ };
1400      |                                            ^~~~~~
1401
1402       and a tweak to integrate the suggestions about misspelled members
1403       with that for accessors:
1404$ g++ -c accessor-fixit.cc
1405accessor-fixit.cc: In function 'int test(t*)':
1406accessor-fixit.cc:17:15: error: 'class t' has no member named 'ratio'; did you m
1407ean 'int t::m_ratio'? (accessible via 'int t::get_ratio() const')
1408   17 |   return ptr->ratio;
1409      |               ^~~~~
1410      |               get_ratio()
1411
1412       In addition, various diagnostics in the C++ front-end have been
1413       streamlined by consolidating the suggestion into the initial error,
1414       rather than emitting a follow-up note:
1415$ g++ typo.cc
1416typo.cc:5:13: error: 'BUFSIZE' was not declared in this scope; did you mean 'BUF
1417_SIZE'?
1418    5 | uint8_t buf[BUFSIZE];
1419      |             ^~~~~~~
1420      |             BUF_SIZE
1421
1422    Runtime Library (libstdc++)
1423
1424     * Improved support for C++17, including:
1425          + The C++17 implementation is no longer experimental.
1426          + Parallel algorithms and <execution> (requires [55]Thread
1427            Building Blocks 2018 or newer).
1428          + <memory_resource>.
1429          + Using the types and functions in <filesystem> does not require
1430            linking with -lstdc++fs now.
1431     * Improved experimental support for C++2a, including:
1432          + Type traits std::remove_cvref, std::unwrap_reference,
1433            std::unwrap_decay_ref, std::is_nothrow_convertible, and
1434            std::type_identity.
1435          + Headers <bit> and <version>.
1436          + Uniform container erasure (std::erase_if).
1437          + contains member of maps and sets.
1438          + String prefix and suffix checking (starts_with, ends_with).
1439          + Functions std::midpoint and std::lerp for interpolation.
1440          + std::bind_front.
1441          + std::visit<R>.
1442          + std::assume_aligned.
1443          + Uses-allocator construction utilities.
1444          + std::pmr::polymorphic_allocator<std::byte>.
1445          + Library support for char8_t type.
1446          + Destroying delete.
1447          + std::is_constant_evaluated() function.
1448     * Support for opening file streams with wide character paths on
1449       Windows
1450     * Incomplete support for the C++17 Filesystem library and the
1451       Filesystem TS on Windows.
1452     * Incomplete, experimental support for the Networking TS.
1453
1454  D
1455
1456     * Support for the D programming language has been added to GCC,
1457       implementing version 2.076 of the language and run-time library.
1458
1459  Fortran
1460
1461     * Asynchronous I/O is now fully supported. The program needs to be
1462       linked against the pthreads library to use it, otherwise the I/O is
1463       done synchronously. For systems which do not support POSIX
1464       condition variables, such as AIX, all I/O is still done
1465       synchronously.
1466     * The BACK argument for MINLOC and MAXLOC has been implemented.
1467     * The FINDLOC intrinsic function has been implemented.
1468     * The IS_CONTIGUOUS intrinsic function has been implemented.
1469     * Direct access to the real and imaginary parts of a complex variable
1470       via c%re and c%im has been implemented.
1471     * Type parameter inquiry via str%len and a%kind has been implemented.
1472     * C descriptors and the ISO_Fortran_binding.h source file have been
1473       implemented.
1474     * The MAX and MIN intrinsics are no longer guaranteed to return any
1475       particular value in case one of the arguments is a NaN. Note that
1476       this conforms to the Fortran standard and to what other Fortran
1477       compilers do. If there is a need to handle that case in some
1478       specific way, one needs to explicitly check for NaN's before
1479       calling MAX or MIN, e.g. by using the IEEE_IS_NAN function from the
1480       intrinsic module IEEE_ARITHMETIC.
1481     * A new command-line option [56]-fdec-include, set also by the
1482       [57]-fdec option, has been added to increase compatibility with
1483       legacy code. With this option, an INCLUDE directive is also parsed
1484       as a statement, which allows the directive to be spread across
1485       multiple source lines with line continuations.
1486     * A new [58]BUILTIN directive, has been added. The purpose of the
1487       directive is to provide an API between the GCC compiler and the GNU
1488       C Library which would define vector implementations of math
1489       routines.
1490
1491  Go
1492
1493     * GCC 9 provides a complete implementation of the Go 1.12.2 user
1494       packages.
1495
1496libgccjit
1497
1498     * The libgccjit API gained a new entry point:
1499       [59]gcc_jit_context_add_driver_option.
1500
1501New Targets and Target Specific Improvements
1502
1503  AArch64 & Arm
1504
1505     * Support has been added for the following processors (GCC
1506       identifiers in parentheses):
1507          + Arm Cortex-A76 (cortex-a76).
1508          + Arm Cortex-A55/Cortex-A76 DynamIQ big.LITTLE
1509            (cortex-a76.cortex-a55).
1510          + Arm Neoverse N1 (neoverse-n1).
1511       The GCC identifiers can be used as arguments to the -mcpu or -mtune
1512       options, for example: -mcpu=cortex-a76 or
1513       -mtune=cortex-a76.cortex-a55 or as arguments to the equivalent
1514       target attributes and pragmas.
1515     * The Armv8.3-A complex number instructions are now supported via
1516       intrinsics when the option -march=armv8.3-a or equivalent is
1517       specified. For the half-precision floating-point variants of these
1518       instructions use the architecture extension flag +fp16, e.g.
1519       -march=armv8.3-a+fp16.
1520       The intrinsics are defined by the ACLE specification.
1521     * The Armv8.5-A architecture is now supported through the
1522       -march=armv8.5-a option.
1523     * The Armv8.5-A architecture also adds some security features that
1524       are optional to all older architecture versions. These are now
1525       supported and only affect the assembler.
1526          + Speculation Barrier instruction through the -march=armv8-a+sb
1527            option.
1528          + Execution and Data Prediction Restriction instructions through
1529            the -march=armv8-a+predres option.
1530          + Speculative Store Bypass Safe instruction through the
1531            -march=armv8-a+ssbs option. This does not require a compiler
1532            option for Arm and thus -march=armv8-a+ssbs is an
1533            AArch64-specific option.
1534
1535      AArch64 specific
1536
1537     * Support has been added for the Arm Neoverse E1 processor
1538       (-mcpu=neoverse-e1).
1539     * The AArch64 port now has support for stack clash protection using
1540       the [60]-fstack-clash-protection option. The probing interval/guard
1541       size can be set by using --param
1542       stack-clash-protection-guard-size=12|16. The value of this
1543       parameter must be in bytes represented as a power of two. The two
1544       supported values for this parameter are 12 (for a 4KiB size, 2^12)
1545       and 16 (for a 64KiB size, 2^16). The default value is 16 (64Kb) and
1546       can be changed at configure time using the flag
1547       --with-stack-clash-protection-guard-size=12|16.
1548     * The option -msign-return-address= has been deprecated. This has
1549       been replaced by the new -mbranch-protection= option. This new
1550       option can now be used to enable the return address signing as well
1551       as the new Branch Target Identification feature of Armv8.5-A
1552       architecture. For more information on the arguments accepted by
1553       this option, please refer to [61]AArch64-Options.
1554     * The following optional extensions to Armv8.5-A architecture are now
1555       supported and only affect the assembler.
1556          + Random Number Generation instructions through the
1557            -march=armv8.5-a+rng option.
1558          + Memory Tagging Extension through the -march=armv8.5-a+memtag
1559            option.
1560
1561      Arm specific
1562
1563     * Support for the deprecated Armv2 and Armv3 architectures and their
1564       variants has been removed. Their corresponding -march values and
1565       the -mcpu options that used these architectures have been removed.
1566     * Support for the Armv5 and Armv5E architectures (which have no known
1567       implementations) has been removed. Note that Armv5T, Armv5TE and
1568       Armv5TEJ architectures remain supported.
1569     * Corrected FPU configurations for Cortex-R7 and Cortex-R8 when using
1570       their respective -mcpu options.
1571
1572  AMD GCN
1573
1574     * A new back end targeting AMD GCN GPUs has been contributed to GCC.
1575       The implementation is currently limited to compiling
1576       single-threaded, stand-alone programs. Future versions will add
1577       support for offloading multi-threaded kernels via OpenMP and
1578       OpenACC. The following devices are supported (GCC identifiers in
1579       parentheses):
1580          + Fiji (fiji).
1581          + Vega 10 (gfx900).
1582
1583  ARC
1584
1585     * LRA is now on by default for the ARC target. This can be controlled
1586       by -mlra.
1587     * Add support for frame code-density and branch-and-index
1588       instructions.
1589
1590  C-SKY
1591
1592     * A new back end targeting C-SKY V2 processors has been contributed
1593       to GCC.
1594
1595  IA-32/x86-64
1596
1597     * Support of Intel MPX (Memory Protection Extensions) has been
1598       removed.
1599     * New ISA extension support for Intel PTWRITE was added to GCC.
1600       PTWRITE intrinsics are available via the -mptwrite compiler switch.
1601     * GCC now supports the Intel CPU named Cascade Lake with AVX512
1602       extensions through -march=cascadelake. The switch enables the
1603       following ISA extensions: AVX512F, AVX512VL, AVX512CD, AVX512BW,
1604       AVX512DQ, AVX512VNNI.
1605
1606  OpenRISC
1607
1608     * A new back end targeting OpenRISC processors has been contributed
1609       to GCC.
1610
1611  S/390, System z, IBM z Systems
1612
1613     * Support for the arch13 architecture has been added. When using the
1614       -march=arch13 option, the compiler will generate code making use of
1615       the new instructions introduced with the vector enhancement
1616       facility 2 and the miscellaneous instruction extension facility 2.
1617       The -mtune=arch13 option enables arch13 specific instruction
1618       scheduling without making use of new instructions.
1619     * Builtins for the new vector instructions have been added and can be
1620       enabled using the -mzvector option.
1621     * Support for ESA architecture machines g5 and g6 is deprecated since
1622       GCC 6.1.0 and has been removed now.
1623     * When compiling with -march=z14 or higher GCC emits alignments hints
1624       on the vector load/store instructions (8 or 16 byte).
1625     * Functions now have a default alignment of 16 bytes. This helps with
1626       branch prediction effects.
1627     * -mfentry is now supported. As well as the mcount mechanism the
1628       __fentry__ is called before the function prologue. However, since
1629       just a single instruction is required to call __fentry__ the call
1630       sequence imposes a smaller overhead than mcount (4 instructions).
1631       The produced code is compatible only with newer glibc versions,
1632       which provide the __fentry__ symbol and do not clobber r0 when
1633       resolving lazily bound functions. -mfentry is only supported when
1634       generating 64 bit code and does not work with nested C functions.
1635     * The -mnop-mcount option can be used to emit NOP instructions
1636       instead of an mcount or fentry call stub.
1637     * With the -mrecord-mcount option a __mcount_loc section is generated
1638       containing pointers to each profiling call stub. This is useful for
1639       automatically patching in and out calls.
1640
1641Operating Systems
1642
1643  Solaris
1644
1645     * g++ now unconditionally enables large file support when compiling
1646       32-bit code.
1647     * Support for the AddressSanitizer and UndefinedBehaviorSanitizer has
1648       been merged from LLVM. For the moment, this only works for 32-bit
1649       code on both SPARC and x86.
1650     * An initial port of the D runtime library has been completed on
1651       Solaris 11/x86. It requires the use of GNU as. Solaris 11/SPARC
1652       support is still work-in-progress.
1653
1654  Windows
1655
1656     * A C++ Microsoft ABI bitfield layout bug, [62]PR87137 has been
1657       fixed. A non-field declaration could cause the current bitfield
1658       allocation unit to be completed, incorrectly placing a following
1659       bitfield into a new allocation unit. The Microsoft ABI is selected
1660       for:
1661          + Mingw targets
1662          + PowerPC, IA-32 or x86-64 targets when the -mms-bitfields
1663            option is specified, or __attribute__((ms_struct)) is used
1664          + SuperH targets when the -mhitachi option is specified, or
1665            __attribute__((renesas)) is used
1666
1667Improvements for plugin authors
1668
1669     * GCC's diagnostic subsystem now has a way to logically group
1670       together related diagnostics, auto_diagnostic_group. Such
1671       diagnostics will be nested by the output of
1672       [63]-fdiagnostics-format=json.
1673     * GCC now has a set of [64]user experience guidelines for GCC, with
1674       information and advice on implementing new diagnostics.
1675
1676Other significant improvements
1677
1678     * GCC's internal "selftest" suite now runs for C++ as well as C (in
1679       debug builds of the compiler).
1680
1681GCC 9.1
1682
1683   This is the [65]list of problem reports (PRs) from GCC's bug tracking
1684   system that are known to be fixed in the 9.1 release. This list might
1685   not be complete (that is, it is possible that some PRs that have been
1686   fixed are not listed here).
1687
1688GCC 9.2
1689
1690   This is the [66]list of problem reports (PRs) from GCC's bug tracking
1691   system that are known to be fixed in the 9.2 release. This list might
1692   not be complete (that is, it is possible that some PRs that have been
1693   fixed are not listed here).
1694
1695GCC 9.3
1696
1697   This is the [67]list of problem reports (PRs) from GCC's bug tracking
1698   system that are known to be fixed in the 9.3 release. This list might
1699   not be complete (that is, it is possible that some PRs that have been
1700   fixed are not listed here).
1701
1702GCC 9.4
1703
1704  Target Specific Changes
1705
1706    AArch64
1707
1708     * The option -moutline-atomics has been added to aid deployment of
1709       the Large System Extensions (LSE) on GNU/Linux systems built with a
1710       baseline architecture targeting Armv8-A. When the option is
1711       specified code is emitted to detect the presence of LSE
1712       instructions at run time and use them for standard atomic
1713       operations. For more information please refer to the documentation.
1714     * GCC now supports the Fujitsu A64FX. The associated -mcpu and -mtune
1715       options are -mcpu=a64fx and -mtune=a64fx respectively. In
1716       particular, -mcpu=a64fx generates code for Armv8.2-A with SVE and
1717       tunes the code for the A64FX. This includes tuning the SVE code,
1718       although by default the code is still length-agnostic and so works
1719       for all SVE implementations. Adding -msve-vector-bits=512 makes the
1720       code specific to 512-bit SVE.
1721
1722
1723    For questions related to the use of GCC, please consult these web
1724    pages and the [68]GCC manuals. If that fails, the
1725    [69]gcc-help@gcc.gnu.org mailing list might help. Comments on these
1726    web pages and the development of GCC are welcome on our developer
1727    list at [70]gcc@gcc.gnu.org. All of [71]our lists have public
1728    archives.
1729
1730   Copyright (C) [72]Free Software Foundation, Inc. Verbatim copying and
1731   distribution of this entire article is permitted in any medium,
1732   provided this notice is preserved.
1733
1734   These pages are [73]maintained by the GCC team. Last modified
1735   2021-03-28[74].
1736
1737References
1738
1739   1. http://gcc.gnu.org/gcc-9/porting_to.html
1740   2. http://gcc.gnu.org/onlinedocs/index.html#current
1741   3. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88469
1742   4. https://gcc.gnu.org/ml/gcc/2018-10/msg00139.html
1743   5. https://gcc.gnu.org/ml/gcc/2019-04/msg00023.html
1744   6. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90920
1745   7. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/C_002b_002b-Dialect-Options.html#index-frepo
1746   8. https://gcc.gnu.org/PR90361
1747   9. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Invoking-GCC.html#Invoking-GCC
1748  10. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-flive-patching
1749  11. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Diagnostic-Message-Formatting-Options.html#index-fno-diagnostics-show-line-numbers
1750  12. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Diagnostic-Message-Formatting-Options.html#index-fno-diagnostics-show-labels
1751  13. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Diagnostic-Message-Formatting-Options.html#index-fdiagnostics-format
1752  14. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-falign-functions
1753  15. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-falign-labels
1754  16. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-falign-loops
1755  17. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-falign-jumps
1756  18. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Instrumentation-Options.html#index-fprofile-filter-files
1757  19. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Instrumentation-Options.html#index-fprofile-exclude-files
1758  20. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Developer-Options.html#index-fopt-info
1759  21. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Developer-Options.html#index-fsave-optimization-record
1760  22. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-fipa-stack-alignment
1761  23. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-fipa-reference-addressable
1762  24. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Other-Builtins.html#index-_005f_005fbuiltin_005fexpect_005fwith_005fprobability
1763  25. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Other-Builtins.html#index-_005f_005fbuiltin_005fhas_005fattribute-1
1764  26. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Other-Builtins.html#index-_005f_005fbuiltin_005fspeculation_005fsafe_005fvalue-1
1765  27. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Common-Function-Attributes.html#index-copy-function-attribute
1766  28. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-ftree-switch-conversion
1767  29. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-fprofile-use
1768  30. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-fversion-loops-for-strides
1769  31. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-floop-interchange
1770  32. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-floop-unroll-and-jam
1771  33. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Optimize-Options.html#index-ftree-loop-distribution
1772  34. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Invoking-Gcov.html#Invoking-Gcov
1773  35. https://www.openacc.org/
1774  36. https://gcc.gnu.org/wiki/OpenACC#status-9
1775  37. https://www.openmp.org/specifications/
1776  38. https://gcc.gnu.org/ml/gcc-patches/2018-11/msg00628.html
1777  39. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Vector-Extensions.html#index-_005f_005fbuiltin_005fconvertvector
1778  40. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Warning-Options.html#index-Waddress-of-packed-member
1779  41. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Warning-Options.html#index-Warray-bounds
1780  42. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Warning-Options.html#index-Wattribute-alias
1781  43. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Warning-Options.html#index-Wformat-overflow
1782  44. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Warning-Options.html#index-Wformat-truncation
1783  45. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Warning-Options.html#index-Wmissing-attributes
1784  46. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Warning-Options.html#index-Wstringop-truncation
1785  47. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Warning-Options.html#index-Wabsolute-value
1786  48. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/C_002b_002b-Dialect-Options.html#index-Wdeprecated-copy
1787  49. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/C_002b_002b-Dialect-Options.html#index-Winit-list-lifetime
1788  50. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/C_002b_002b-Dialect-Options.html#index-Wredundant-move
1789  51. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/C_002b_002b-Dialect-Options.html#index-Wpessimizing-move
1790  52. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/C_002b_002b-Dialect-Options.html#index-Wclass-conversion
1791  53. http://gcc.gnu.org/projects/cxx-status.html#cxx2a
1792  54. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Warning-Options.html#index-Wformat
1793  55. https://github.com/oneapi-src/oneTBB
1794  56. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gfortran/Fortran-Dialect-Options.html#index-fdec-include
1795  57. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gfortran/Fortran-Dialect-Options.html#index-fdec
1796  58. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gfortran/BUILTIN-directive.html#BUILTIN-directive
1797  59. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/jit/topics/contexts.html#gcc_jit_context_add_driver_option
1798  60. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Instrumentation-Options.html#index-fstack-protector
1799  61. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/AArch64-Options.html#AArch64-Options
1800  62. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87137
1801  63. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Diagnostic-Message-Formatting-Options.html#index-fdiagnostics-format
1802  64. https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gccint/User-Experience-Guidelines.html
1803  65. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=9.0
1804  66. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=9.2
1805  67. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=9.3
1806  68. https://gcc.gnu.org/onlinedocs/
1807  69. mailto:gcc-help@gcc.gnu.org
1808  70. mailto:gcc@gcc.gnu.org
1809  71. https://gcc.gnu.org/lists.html
1810  72. https://www.fsf.org/
1811  73. https://gcc.gnu.org/about.html
1812  74. http://validator.w3.org/check/referer
1813======================================================================
1814http://gcc.gnu.org/gcc-8/index.html
1815
1816                              GCC 8 Release Series
1817
1818   Mar 4, 2020
1819
1820   The [1]GNU project and the GCC developers are pleased to announce the
1821   release of GCC 8.4.
1822
1823   This release is a bug-fix release, containing fixes for regressions in
1824   GCC 8.3 relative to previous releases of GCC.
1825
1826Release History
1827
1828   GCC 8.4
1829          Mar 4, 2020 ([2]changes, [3]documentation)
1830
1831   GCC 8.3
1832          Feb 22, 2019 ([4]changes, [5]documentation)
1833
1834   GCC 8.2
1835          Jul 26, 2018 ([6]changes, [7]documentation)
1836
1837   GCC 8.1
1838          May 2, 2018 ([8]changes, [9]documentation)
1839
1840References and Acknowledgements
1841
1842   GCC used to stand for the GNU C Compiler, but since the compiler
1843   supports several other languages aside from C, it now stands for the
1844   GNU Compiler Collection.
1845
1846   A list of [10]successful builds is updated as new information becomes
1847   available.
1848
1849   The GCC developers would like to thank the numerous people that have
1850   contributed new features, improvements, bug fixes, and other changes as
1851   well as test results to GCC. This [11]amazing group of volunteers is
1852   what makes GCC successful.
1853
1854   For additional information about GCC please refer to the [12]GCC
1855   project web site or contact the [13]GCC development mailing list.
1856
1857   To obtain GCC please use [14]our mirror sites or [15]our version
1858   control system.
1859
1860
1861    For questions related to the use of GCC, please consult these web
1862    pages and the [16]GCC manuals. If that fails, the
1863    [17]gcc-help@gcc.gnu.org mailing list might help. Comments on these
1864    web pages and the development of GCC are welcome on our developer
1865    list at [18]gcc@gcc.gnu.org. All of [19]our lists have public
1866    archives.
1867
1868   Copyright (C) [20]Free Software Foundation, Inc. Verbatim copying and
1869   distribution of this entire article is permitted in any medium,
1870   provided this notice is preserved.
1871
1872   These pages are [21]maintained by the GCC team. Last modified
1873   2020-03-04[22].
1874
1875References
1876
1877   1. http://www.gnu.org/
1878   2. http://gcc.gnu.org/gcc-8/changes.html
1879   3. http://gcc.gnu.org/onlinedocs/8.4.0/
1880   4. http://gcc.gnu.org/gcc-8/changes.html
1881   5. http://gcc.gnu.org/onlinedocs/8.3.0/
1882   6. http://gcc.gnu.org/gcc-8/changes.html
1883   7. http://gcc.gnu.org/onlinedocs/8.2.0/
1884   8. http://gcc.gnu.org/gcc-8/changes.html
1885   9. http://gcc.gnu.org/onlinedocs/8.1.0/
1886  10. http://gcc.gnu.org/gcc-8/buildstat.html
1887  11. http://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Contributors.html
1888  12. http://gcc.gnu.org/index.html
1889  13. mailto:gcc@gcc.gnu.org
1890  14. http://gcc.gnu.org/mirrors.html
1891  15. http://gcc.gnu.org/git.html
1892  16. https://gcc.gnu.org/onlinedocs/
1893  17. mailto:gcc-help@gcc.gnu.org
1894  18. mailto:gcc@gcc.gnu.org
1895  19. https://gcc.gnu.org/lists.html
1896  20. https://www.fsf.org/
1897  21. https://gcc.gnu.org/about.html
1898  22. http://validator.w3.org/check/referer
1899======================================================================
1900http://gcc.gnu.org/gcc-8/changes.html
1901
1902                              GCC 8 Release Series
1903                        Changes, New Features, and Fixes
1904
1905   This page is a "brief" summary of some of the huge number of
1906   improvements in GCC 8. You may also want to check out our [1]Porting to
1907   GCC 8 page and the [2]full GCC documentation.
1908
1909Caveats
1910
1911     * Support for the obsolete SDB/coff debug info format has been
1912       removed. The option -gcoff no longer does anything.
1913     * The Cilk+ extensions to the C and C++ languages have been removed.
1914     * The MPX extensions to the C and C++ languages have been deprecated
1915       and will be removed in a future release.
1916     * The extension allowing arithmetic on std::atomic<void*> and types
1917       like std::atomic<R(*)()> has been deprecated.
1918     * The non-standard C++0x std::copy_exception function was removed.
1919       std::make_exception_ptr should be used instead.
1920     * Support for the powerpc*-*-*spe* target ports which have been
1921       recently unmaintained and untested in GCC has been declared
1922       obsolete in GCC 8 as [3]announced. Unless there is activity to
1923       revive them, the next release of GCC will have their sources
1924       permanently removed.
1925
1926General Improvements
1927
1928     * Inter-procedural optimization improvements:
1929          + Reworked run-time estimation metrics leading to more realistic
1930            guesses driving inliner and cloning heuristics.
1931          + The ipa-pure-const pass is extended to propagate the malloc
1932            attribute, and the corresponding warning option
1933            -Wsuggest-attribute=malloc emits a diagnostic for functions
1934            which can be annotated with the malloc attribute.
1935     * Profile driven optimization improvements:
1936          + New infrastructure for representing profiles (both statically
1937            guessed and profile feedback) which allows propagation of
1938            additional information about the reliability of the profile.
1939          + A number of improvements in the profile updating code solving
1940            problems found by new verification code.
1941          + Static detection of code which is not executed in a valid run
1942            of the program. This includes paths which trigger undefined
1943            behavior as well as calls to functions declared with the cold
1944            attribute. Newly the noreturn attribute does not imply all
1945            effects of cold to differentiate between exit (which is
1946            noreturn) and abort (which is in addition not executed in
1947            valid runs).
1948          + -freorder-blocks-and-partition, a pass splitting function
1949            bodies into hot and cold regions, is now enabled by default at
1950            -O2 and higher for x86 and x86-64.
1951     * Link-time optimization improvements:
1952          + We have significantly improved debug information on ELF
1953            targets using DWARF by properly preserving language-specific
1954            information. This allows for example the libstdc++
1955            pretty-printers to work with LTO optimized executables.
1956     * A new option -fcf-protection=[full|branch|return|none] is
1957       introduced to perform code instrumentation to increase program
1958       security by checking that target addresses of control-flow transfer
1959       instructions (such as indirect function call, function return,
1960       indirect jump) are valid. Currently the instrumentation is
1961       supported on x86 GNU/Linux targets only. See the user guide for
1962       further information about the option syntax and section "New
1963       Targets and Target Specific Improvements" for IA-32/x86-64 for more
1964       details.
1965     * The -gcolumn-info option is now enabled by default. It includes
1966       column information in addition to just filenames and line numbers
1967       in DWARF debugging information.
1968     * The polyhedral-based loop nest optimization pass
1969       -floop-nest-optimize has been overhauled. It's still considered
1970       experimental and may not result in any runtime improvements.
1971     * Two new classical loop nest optimization passes have been added.
1972       -floop-unroll-and-jam performs outer loop unrolling and fusing of
1973       the inner loop copies. -floop-interchange exchanges loops in a loop
1974       nest to improve data locality. Both passes are enabled by default
1975       at -O3 and above.
1976     * The classic loop nest optimization pass -ftree-loop-distribution
1977       has been improved and enabled by default at -O3 and above. It
1978       supports loop nest distribution in some restricted scenarios; it
1979       also supports cancellable innermost loop distribution with loop
1980       versioning under run-time alias checks.
1981     * The new option -fstack-clash-protection causes the compiler to
1982       insert probes whenever stack space is allocated statically or
1983       dynamically to reliably detect stack overflows and thus mitigate
1984       the attack vector that relies on jumping over a stack guard page as
1985       provided by the operating system.
1986     * A new pragma GCC unroll has been implemented in the C family of
1987       languages, as well as Fortran and Ada, so as to make it possible
1988       for the user to have a finer-grained control over the loop
1989       unrolling optimization.
1990     * GCC has been enhanced to detect more instances of meaningless or
1991       mutually exclusive attribute specifications and handle such
1992       conflicts more consistently. Mutually exclusive attribute
1993       specifications are ignored with a warning regardless of whether
1994       they appear on the same declaration or on distinct declarations of
1995       the same entity. For example, because the noreturn attribute on the
1996       second declaration below is mutually exclusive with the malloc
1997       attribute on the first, it is ignored and a warning is issued.
1998>
1999      void* __attribute__ ((malloc)) f (unsigned);
2000      void* __attribute__ ((noreturn)) f (unsigned);
2001
2002      warning: ignoring attribute 'noreturn' because it conflicts with attribute
2003 'malloc' [-Wattributes]
2004     * The gcov tool can distinguish functions that begin on a same line
2005       in a source file. This can be a different template instantiation or
2006       a class constructor:
2007
2008File 'ins.C'
2009Lines executed:100.00% of 8
2010Creating 'ins.C.gcov'
2011
2012        -:    0:Source:ins.C
2013        -:    0:Graph:ins.gcno
2014        -:    0:Data:ins.gcda
2015        -:    0:Runs:1
2016        -:    0:Programs:1
2017        -:    1:template<class T>
2018        -:    2:class Foo
2019        -:    3:{
2020        -:    4: public:
2021        2:    5:   Foo(): b (1000) {}
2022------------------
2023Foo<char>::Foo():
2024        1:    5:   Foo(): b (1000) {}
2025------------------
2026Foo<int>::Foo():
2027        1:    5:   Foo(): b (1000) {}
2028------------------
2029        2:    6:   void inc () { b++; }
2030------------------
2031Foo<char>::inc():
2032        1:    6:   void inc () { b++; }
2033------------------
2034Foo<int>::inc():
2035        1:    6:   void inc () { b++; }
2036------------------
2037        -:    7:
2038        -:    8:  private:
2039        -:    9:   int b;
2040        -:   10:};
2041        -:   11:
2042        1:   12:int main(int argc, char **argv)
2043        -:   13:{
2044        1:   14:  Foo<int> a;
2045        1:   15:  Foo<char> b;
2046        -:   16:
2047        1:   17:  a.inc ();
2048        1:   18:  b.inc ();
2049        1:   19:}
2050
2051     * The gcov tool has more accurate numbers for execution of lines in a
2052       source file.
2053     * The gcov tool can use TERM colors to provide more readable output.
2054     * AddressSanitizer gained a new pair of sanitization options,
2055       -fsanitize=pointer-compare and -fsanitize=pointer-subtract, which
2056       warn about subtraction (or comparison) of pointers that point to a
2057       different memory object:
2058
2059int
2060main ()
2061{
2062  /* Heap allocated memory.  */
2063  char *heap1 = (char *)__builtin_malloc (42);
2064  char *heap2 = (char *)__builtin_malloc (42);
2065  if (heap1 > heap2)
2066      return 1;
2067
2068  return 0;
2069}
2070
2071==17465==ERROR: AddressSanitizer: invalid-pointer-pair: 0x604000000010 0x6040000
207200050
2073    #0 0x40070f in main /tmp/pointer-compare.c:7
2074    #1 0x7ffff6a72a86 in __libc_start_main (/lib64/libc.so.6+0x21a86)
2075    #2 0x400629 in _start (/tmp/a.out+0x400629)
2076
20770x604000000010 is located 0 bytes inside of 42-byte region [0x604000000010,0x604
207800000003a)
2079allocated by thread T0 here:
2080    #0 0x7ffff6efb390 in __interceptor_malloc ../../../../libsanitizer/asan/asan
2081_malloc_linux.cc:86
2082    #1 0x4006ea in main /tmp/pointer-compare.c:5
2083    #2 0x7ffff6a72a86 in __libc_start_main (/lib64/libc.so.6+0x21a86)
2084
20850x604000000050 is located 0 bytes inside of 42-byte region [0x604000000050,0x604
208600000007a)
2087allocated by thread T0 here:
2088    #0 0x7ffff6efb390 in __interceptor_malloc ../../../../libsanitizer/asan/asan
2089_malloc_linux.cc:86
2090    #1 0x4006f8 in main /tmp/pointer-compare.c:6
2091    #2 0x7ffff6a72a86 in __libc_start_main (/lib64/libc.so.6+0x21a86)
2092
2093SUMMARY: AddressSanitizer: invalid-pointer-pair /tmp/pointer-compare.c:7 in main
2094
2095     * The store merging pass has been enhanced to handle bit-fields and
2096       not just constant stores, but also data copying from adjacent
2097       memory locations into other adjacent memory locations, including
2098       bitwise logical operations on the data. The pass can also handle
2099       byte swapping into memory locations.
2100     * The undefined behavior sanitizer gained two new options included in
2101       -fsanitize=undefined: -fsanitize=builtin which diagnoses at run
2102       time invalid arguments to __builtin_clz or __builtin_ctz prefixed
2103       builtins, and -fsanitize=pointer-overflow which performs cheap run
2104       time tests for pointer wrapping.
2105     * A new attribute no_sanitize can be applied to functions to instruct
2106       the compiler not to do sanitization of the options provided as
2107       arguments to the attribute. Acceptable values for no_sanitize match
2108       those acceptable by the -fsanitize command-line option.
2109
2110void __attribute__ ((no_sanitize ("alignment", "object-size")))
2111f () { /* Do something. */; }
2112
2113New Languages and Language specific improvements
2114
2115  Ada
2116
2117     * For its internal exception handling used on the host for error
2118       recovery in the front-end, the compiler now relies on the native
2119       exception handling mechanism of the host platform, which should be
2120       more efficient than the former mechanism.
2121
2122  BRIG (HSAIL)
2123
2124   In this release cycle, the focus for the BRIGFE was on stabilization
2125   and performance improvements. Also a couple of completely new features
2126   were added.
2127     * Improved support for function and module scope group segment
2128       variables. PRM specs define function and module scope group segment
2129       variables as an experimental feature. However, PRM test suite uses
2130       them. Now group segment is handled by separate book keeping of
2131       module scope and function (kernel) offsets. Each function has a
2132       "frame" in the group segment offset to which is given as an
2133       argument, similar to traditional call stack frame handling.
2134     * Reduce the number of type conversions due to the untyped HSAIL
2135       registers. Instead of always representing the HSAIL's untyped
2136       registers as unsigned int, the gccbrig now pre-analyzes the BRIG
2137       code and builds the register variables as a type used the most when
2138       storing or reading data to/from each register. This reduces the
2139       number of total casts which cannot be always optimized away.
2140     * Support for BRIG_KIND_NONE directives.
2141     * Made -O3 the default optimization level for BRIGFE.
2142     * Fixed illegal addresses generated from address expressions which
2143       refer only to offset 0.
2144     * Fixed a bug with reg+offset addressing on 32b segments. In 'large'
2145       mode, the offset is treated as 32bits unless it's in global,
2146       read-only or kernarg address space.
2147     * Fixed a crash caused sometimes by calls with more than 4 arguments.
2148     * Fixed a mis-execution issue with kernels that have both unexpanded
2149       ID functions and calls to subfunctions.
2150     * Treat HSAIL barrier builtins as setjmp/longjump style functions to
2151       avoid illegal optimizations.
2152     * Ensure per WI copies of private variables are aligned correctly.
2153     * libhsail-rt: Assume the host runtime allocates the work group
2154       memory.
2155
2156  C family
2157
2158     * New command-line options have been added for the C and C++
2159       compilers:
2160          + [4]-Wmultistatement-macros warns about unsafe macros expanding
2161            to multiple statements used as a body of a statement such as
2162            if, else, while, switch, or for.
2163          + [5]-Wstringop-truncation warns for calls to bounded string
2164            manipulation functions such as strncat, strncpy, and stpncpy
2165            that might either truncate the copied string or leave the
2166            destination unchanged. For example, the following call to
2167            strncat is diagnosed because it appends just three of the four
2168            characters from the source string.
2169void append (char *buf, size_t bufsize)
2170{
2171    strncat (buf, ".txt", 3);
2172}
2173warning: 'strncat' output truncated copying 3 bytes from a string of length 4 [-
2174Wstringop-truncation]
2175            Similarly, in the following example, the call to strncpy
2176            specifies the size of the destination buffer as the bound. If
2177            the length of the source string is equal to or greater than
2178            this size the result of the copy will not be NUL-terminated.
2179            Therefore, the call is also diagnosed. To avoid the warning,
2180            specify sizeof buf - 1 as the bound and set the last element
2181            of the buffer to NUL.
2182void copy (const char *s)
2183{
2184    char buf[80];
2185    strncpy (buf, s, sizeof buf);
2186    ...
2187}
2188warning: 'strncpy' specified bound 80 equals destination size [-Wstringop-trunca
2189tion]
2190            The -Wstringop-truncation option is included in -Wall.
2191            Note that due to GCC bug [6]82944, defining strncat, strncpy,
2192            or stpncpy as a macro in a system header as some
2193            implementations do, suppresses the warning.
2194          + [7]-Wif-not-aligned controls warnings issued in response to
2195            invalid uses of objects declared with attribute
2196            [8]warn_if_not_aligned.
2197            The -Wif-not-aligned option is included in -Wall.
2198          + [9]-Wmissing-attributes warns when a declaration of a function
2199            is missing one or more attributes that a related function is
2200            declared with and whose absence may adversely affect the
2201            correctness or efficiency of generated code. For example, in
2202            C++, the warning is issued when an explicit specialization of
2203            a primary template declared with attribute alloc_align,
2204            alloc_size, assume_aligned, format, format_arg, malloc, or
2205            nonnull is declared without it. Attributes deprecated, error,
2206            and warning suppress the warning.
2207            The -Wmissing-attributes option is included in -Wall.
2208          + [10]-Wpacked-not-aligned warns when a struct or union declared
2209            with attribute packed defines a member with an explicitly
2210            specified alignment greater than 1. Such a member will wind up
2211            under-aligned. For example, a warning will be issued for the
2212            definition of struct A in the following:
2213struct __attribute__ ((aligned (8)))
2214S8 { char a[8]; };
2215
2216struct __attribute__ ((packed)) A
2217{
2218    struct S8 s8;
2219};
2220warning: alignment 1 of 'struct S' is less than 8 [-Wpacked-not-aligned]
2221            The -Wpacked-not-aligned option is included in -Wall.
2222          + -Wcast-function-type warns when a function pointer is cast to
2223            an incompatible function pointer. This warning is enabled by
2224            -Wextra.
2225          + -Wsizeof-pointer-div warns for suspicious divisions of the
2226            size of a pointer by the size of the elements it points to,
2227            which looks like the usual way to compute the array size but
2228            won't work out correctly with pointers. This warning is
2229            enabled by -Wall.
2230          + -Wcast-align=strict warns whenever a pointer is cast such that
2231            the required alignment of the target is increased. For
2232            example, warn if a char * is cast to an int * regardless of
2233            the target machine.
2234          + -fprofile-abs-path creates absolute path names in the .gcno
2235            files. This allows gcov to find the correct sources in
2236            projects where compilations occur with different working
2237            directories.
2238     * -fno-strict-overflow is now mapped to -fwrapv -fwrapv-pointer and
2239       signed integer overflow is now undefined by default at all
2240       optimization levels. Using -fsanitize=signed-integer-overflow is
2241       now the preferred way to audit code, -Wstrict-overflow is
2242       deprecated.
2243     * The [11]-Warray-bounds option has been improved to detect more
2244       instances of out-of-bounds array indices and pointer offsets. For
2245       example, negative or excessive indices into flexible array members
2246       and string literals are detected.
2247     * The [12]-Wrestrict option introduced in GCC 7 has been enhanced to
2248       detect many more instances of overlapping accesses to objects via
2249       restrict-qualified arguments to standard memory and string
2250       manipulation functions such as memcpy and strcpy. For example, the
2251       strcpy call in the function below attempts to truncate the string
2252       by replacing its initial characters with the last four. However,
2253       because the function writes the terminating NUL into a[4], the
2254       copies overlap and the call is diagnosed.
2255void f (void)
2256{
2257    char a[] = "abcd1234";
2258    strcpy (a, a + 4);
2259    ...
2260}
2261warning: 'strcpy' accessing 5 bytes at offsets 0 and 4 overlaps 1 byte at offset
2262 4 [-Wrestrict]
2263       The -Wrestrict option is included in -Wall.
2264     * Several optimizer enhancements have enabled improvements to the
2265       [13]-Wformat-overflow and [14]-Wformat-truncation options. The
2266       warnings detect more instances of buffer overflow and truncation
2267       than in GCC 7 and are better at avoiding certain kinds of false
2268       positives.
2269     * When reporting mismatching argument types at a function call, the C
2270       and C++ compilers now underline both the argument and the pertinent
2271       parameter in the declaration.
2272$ gcc arg-type-mismatch.cc
2273arg-type-mismatch.cc: In function 'int caller(int, int, float)':
2274arg-type-mismatch.cc:5:24: error: invalid conversion from 'int' to 'const char*'
2275 [-fpermissive]
2276   return callee(first, second, third);
2277                        ^~~~~~
2278arg-type-mismatch.cc:1:40: note:   initializing argument 2 of 'int callee(int, c
2279onst char*, float)'
2280 extern int callee(int one, const char *two, float three);
2281                            ~~~~~~~~~~~~^~~
2282
2283     * When reporting on unrecognized identifiers, the C and C++ compilers
2284       will now emit fix-it hints suggesting #include directives for
2285       various headers in the C and C++ standard libraries.
2286$ gcc incomplete.c
2287incomplete.c: In function 'test':
2288incomplete.c:3:10: error: 'NULL' undeclared (first use in this function)
2289   return NULL;
2290          ^~~~
2291incomplete.c:3:10: note: 'NULL' is defined in header '<stddef.h>'; did you forge
2292t to '#include <stddef.h>'?
2293incomplete.c:1:1:
2294+#include <stddef.h>
2295 const char *test(void)
2296incomplete.c:3:10:
2297   return NULL;
2298          ^~~~
2299incomplete.c:3:10: note: each undeclared identifier is reported only once for ea
2300ch function it appears in
2301
2302$ gcc incomplete.cc
2303incomplete.cc:1:6: error: 'string' in namespace 'std' does not name a type
2304 std::string s("hello world");
2305      ^~~~~~
2306incomplete.cc:1:1: note: 'std::string' is defined in header '<string>'; did you
2307forget to '#include <string>'?
2308+#include <string>
2309 std::string s("hello world");
2310 ^~~
2311
2312     * The C and C++ compilers now use more intuitive locations when
2313       reporting on missing semicolons, and offer fix-it hints:
2314$ gcc t.c
2315t.c: In function 'test':
2316t.c:3:12: error: expected ';' before '}' token
2317   return 42
2318            ^
2319            ;
2320 }
2321 ~
2322
2323     * When reporting on missing '}' and ')' tokens, the C and C++
2324       compilers will now highlight the corresponding '{' and '(' token,
2325       issuing a 'note' if it's on a separate line:
2326$ gcc unclosed.c
2327unclosed.c: In function 'log_when_out_of_range':
2328unclosed.c:12:50: error: expected ')' before '{' token
2329       && (temperature < MIN || temperature > MAX) {
2330                                                  ^~
2331                                                  )
2332unclosed.c:11:6: note: to match this '('
2333   if (logging_enabled && check_range ()
2334      ^
2335
2336       or highlighting it directly if it's on the same line:
2337$ gcc unclosed-2.c
2338unclosed-2.c: In function 'test':
2339unclosed-2.c:8:45: error: expected ')' before '{' token
2340   if (temperature < MIN || temperature > MAX {
2341      ~                                      ^~
2342                                             )
2343
2344       They will also emit fix-it hints.
2345
2346  C++
2347
2348     * GCC 8 (-fabi-version=12) has a couple of corrections to the calling
2349       convention, which changes the ABI for some uncommon code:
2350          + Passing an empty class as an argument now takes up no space on
2351            x86_64, as required by the psABI.
2352          + Passing or returning a class with only deleted copy and move
2353            constructors now uses the same calling convention as a class
2354            with a non-trivial copy or move constructor. This only affects
2355            C++17 mode, as in earlier standards passing or returning such
2356            a class was impossible.
2357          + WARNING: In GCC 8.1 the second change mistakenly also affects
2358            classes with a deleted copy constructor and defaulted trivial
2359            move constructor (bug [15]c++/86094). This issue is fixed in
2360            GCC 8.2 (-fabi-version=13).
2361       You can test whether these changes affect your code with -Wabi=11
2362       (or -Wabi=12 in GCC 8.2 for the third issue); if these changes are
2363       problematic for your project, the GCC 7 ABI can be selected with
2364       -fabi-version=11.
2365     * The value of the C++11 alignof operator has been corrected to match
2366       C _Alignof (minimum alignment) rather than GNU __alignof__
2367       (preferred alignment); on ia32 targets this means that
2368       alignof(double) is now 4 rather than 8. Code that wants the
2369       preferred alignment should use __alignof__ instead.
2370     * New command-line options have been added for the C++ compiler to
2371       control warnings:
2372          + [16]-Wclass-memaccess warns when objects of non-trivial class
2373            types are manipulated in potentially unsafe ways by raw memory
2374            functions such as memcpy, or realloc. The warning helps detect
2375            calls that bypass user-defined constructors or copy-assignment
2376            operators, corrupt virtual table pointers, data members of
2377            const-qualified types or references, or member pointers. The
2378            warning also detects calls that would bypass access controls
2379            to data members. For example, a call such as:
2380        memcpy (&std::cout, &std::cerr, sizeof std::cout);
2381            results in
2382        warning: 'void* memcpy(void*, const void*, long unsigned int)' writing t
2383o an object of type 'std::ostream' {aka 'class std::basic_ostream<char>'} with n
2384o trivial copy-assignment [-Wclass-memaccess]
2385            The -Wclass-memaccess option is included in -Wall.
2386     * The C++ front end has experimental support for some of the upcoming
2387       C++2a draft features with the -std=c++2a or -std=gnu++2a flags,
2388       including designated initializers, default member initializers for
2389       bit-fields, __VA_OPT__ (except that #__VA_OPT__ is unsupported),
2390       lambda [=, this] captures, etc. For a full list of new features,
2391       see [17]the C++ status page.
2392     * When reporting on attempts to access private fields of a class or
2393       struct, the C++ compiler will now offer fix-it hints showing how to
2394       use an accessor function to get at the field in question, if one
2395       exists.
2396$ gcc accessor.cc
2397accessor.cc: In function 'void test(foo*)':
2398accessor.cc:12:12: error: 'double foo::m_ratio' is private within this context
2399   if (ptr->m_ratio >= 0.5)
2400            ^~~~~~~
2401accessor.cc:7:10: note: declared private here
2402   double m_ratio;
2403          ^~~~~~~
2404accessor.cc:12:12: note: field 'double foo::m_ratio' can be accessed via 'double
2405 foo::get_ratio() const'
2406   if (ptr->m_ratio >= 0.5)
2407            ^~~~~~~
2408            get_ratio()
2409
2410     * The C++ compiler can now give you a hint if you use a macro before
2411       it was defined (e.g. if you mess up the order of your #include
2412       directives):
2413$ gcc ordering.cc
2414ordering.cc:2:24: error: expected ';' at end of member declaration
2415   virtual void clone() const OVERRIDE { }
2416                        ^~~~~
2417                             ;
2418ordering.cc:2:30: error: 'OVERRIDE' does not name a type
2419   virtual void clone() const OVERRIDE { }
2420                              ^~~~~~~~
2421ordering.cc:2:30: note: the macro 'OVERRIDE' had not yet been defined
2422In file included from ordering.cc:5:
2423c++11-compat.h:2: note: it was later defined here
2424 #define OVERRIDE override
2425
2426
2427     * The -Wold-style-cast diagnostic can now emit fix-it hints telling
2428       you when you can use a static_cast, const_cast, or
2429       reinterpret_cast.
2430$ gcc -c old-style-cast-fixits.cc -Wold-style-cast
2431old-style-cast-fixits.cc: In function 'void test(void*)':
2432old-style-cast-fixits.cc:5:19: warning: use of old-style cast to 'struct foo*' [
2433-Wold-style-cast]
2434   foo *f = (foo *)ptr;
2435                   ^~~
2436            ----------
2437            static_cast<foo *> (ptr)
2438
2439     * When reporting on problems within extern "C" linkage
2440       specifications, the C++ compiler will now display the location of
2441       the start of the extern "C".
2442$ gcc -c extern-c.cc
2443extern-c.cc:3:1: error: template with C linkage
2444 template <typename T> void test (void);
2445 ^~~~~~~~
2446In file included from extern-c.cc:1:
2447unclosed.h:1:1: note: 'extern "C"' linkage started here
2448 extern "C" {
2449 ^~~~~~~~~~
2450extern-c.cc:3:39: error: expected '}' at end of input
2451 template <typename T> void test (void);
2452                                       ^
2453In file included from extern-c.cc:1:
2454unclosed.h:1:12: note: to match this '{'
2455 extern "C" {
2456            ^
2457
2458     * When reporting on mismatching template types, the C++ compiler will
2459       now use color to highlight the mismatching parts of the template,
2460       and will elide the parameters that are common between two
2461       mismatching templates, printing [...] instead:
2462$ gcc templates.cc
2463templates.cc: In function 'void test()':
2464templates.cc:9:8: error: could not convert 'vector<double>()' from 'vector<doubl
2465e>' to 'vector<int>'
2466   fn_1(vector<double> ());
2467        ^~~~~~~~~~~~~~~~~
2468templates.cc:10:8: error: could not convert 'map<int, double>()' from 'map<[...]
2469,double>' to 'map<[...],int>'
2470   fn_2(map<int, double>());
2471        ^~~~~~~~~~~~~~~~~~
2472
2473       Those [...] elided parameters can be seen using -fno-elide-type:
2474$ gcc templates.cc -fno-elide-type
2475templates.cc: In function 'void test()':
2476templates.cc:9:8: error: could not convert 'vector<double>()' from 'vector<doubl
2477e>' to 'vector<int>'
2478   fn_1(vector<double> ());
2479        ^~~~~~~~~~~~~~~~~
2480templates.cc:10:8: error: could not convert 'map<int, double>()' from 'map<int,d
2481ouble>' to 'map<int,int>'
2482   fn_2(map<int, double>());
2483        ^~~~~~~~~~~~~~~~~~
2484
2485       The C++ compiler has also gained an option
2486       -fdiagnostics-show-template-tree which visualizes such mismatching
2487       templates in a hierarchical form:
2488$ gcc templates-2.cc -fdiagnostics-show-template-tree
2489templates-2.cc: In function 'void test()':
2490templates-2.cc:9:8: error: could not convert 'vector<double>()' from 'vector<dou
2491ble>' to 'vector<int>'
2492  vector<
2493    [double != int]>
2494   fn_1(vector<double> ());
2495        ^~~~~~~~~~~~~~~~~
2496templates-2.cc:10:8: error: could not convert 'map<map<int, vector<double> >, ve
2497ctor<double> >()' from 'map<map<[...],vector<double>>,vector<double>>' to 'map<m
2498ap<[...],vector<float>>,vector<float>>'
2499  map<
2500    map<
2501      [...],
2502      vector<
2503        [double != float]>>,
2504    vector<
2505      [double != float]>>
2506   fn_2(map<map<int, vector<double>>, vector<double>> ());
2507        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2508
2509       which again works with -fno-elide-type:
2510$ gcc templates-2.cc -fdiagnostics-show-template-tree -fno-elide-type
2511templates-2.cc: In function 'void test()':
2512templates-2.cc:9:8: error: could not convert 'vector<double>()' from 'vector<dou
2513ble>' to 'vector<int>'
2514  vector<
2515    [double != int]>
2516   fn_1(vector<double> ());
2517        ^~~~~~~~~~~~~~~~~
2518templates-2.cc:10:8: error: could not convert 'map<map<int, vector<double> >, ve
2519ctor<double> >()' from 'map<map<int,vector<double>>,vector<double>>' to 'map<map
2520<int,vector<float>>,vector<float>>'
2521  map<
2522    map<
2523      int,
2524      vector<
2525        [double != float]>>,
2526    vector<
2527      [double != float]>>
2528   fn_2(map<map<int, vector<double>>, vector<double>> ());
2529        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2530
2531     * Flowing off the end of a non-void function is considered
2532       unreachable and may be subject to optimization on that basis. As a
2533       result of this change, -Wreturn-type warnings are enabled by
2534       default for C++.
2535
2536    Runtime Library (libstdc++)
2537
2538     * Improved experimental support for C++17, including the following
2539       features:
2540          + Deduction guides to support class template argument deduction.
2541          + std::filesystem implementation.
2542          + std::char_traits<char> and std::char_traits<wchar_t> are
2543            usable in constant expressions.
2544          + std::to_chars and std::from_chars (for integers only, not for
2545            floating point types).
2546     * Experimental support for C++2a: std::to_address (thanks to Glen
2547       Fernandes) and std::endian.
2548     * On GNU/Linux, std::random_device::entropy() accesses the kernel's
2549       entropy count for the random device, if known (thanks to Xi
2550       Ruoyao).
2551     * Support for std::experimental::source_location.
2552     * AddressSanitizer integration for std::vector, detecting
2553       out-of-range accesses to the unused capacity of a vector.
2554     * Extensions __gnu_cxx::airy_ai and __gnu_cxx::airy_bi added to the
2555       Mathematical Special Functions.
2556
2557  Fortran
2558
2559     * The main version of libfortran has been changed to 5.
2560     * Parameterized derived types, a major feature of Fortran 2003, have
2561       been implemented.
2562     * Partial support is provided for Fortran 2018 teams, which are
2563       hierarchical subsets of images that execute independently of other
2564       image subsets.
2565     * The maximum rank for arrays has been increased to 15, conforming to
2566       the Fortran 2008 standard.
2567     * Transformational intrinsics are now fully supported in
2568       initialization expressions.
2569     * New flag -fc-prototypes to write C prototypes for BIND(C)
2570       procedures and variables.
2571     * If -fmax-stack-var-size is honored if given together with -Ofast,
2572       -fstack-arrays is no longer set in that case.
2573     * New options -fdefault-real-16 and -fdefault-real-10 to control the
2574       default kind of REAL variables.
2575     * A warning is now issued if an array subscript inside a DO loop
2576       could lead to an out-of-bounds-access. The new option
2577       -Wdo-subscript, enabled by -Wextra, warns about this even if the
2578       compiler can not prove that the code will be executed.
2579     * The Fortran front end now attempts to interchange loops if it is
2580       deemed profitable. So far, this is restricted to FORALL and DO
2581       CONCURRENT statements with multiple indices. This behavior be
2582       controlled with the new flag -ffrontend-loop-interchange, which is
2583       enabled with optimization by default. The
2584       -Wfrontend-loop-interchange option warns about such occurrences.
2585     * When an actual argument contains too few elements for a dummy
2586       argument, an error is now issued. The -std=legacy option can be
2587       used to still compile such code.
2588     * The RECL= argument to OPEN and INQUIRE statements now allows 64-bit
2589       integers, making records larger than 2GiB possible.
2590     * The GFORTRAN_DEFAULT_RECL environment variable no longer has any
2591       effect. The record length for preconnected units is now larger than
2592       any practical limit, same as for sequential access units opened
2593       without an explicit RECL= specifier.
2594     * Character variables longer than HUGE(0) elements are now possible
2595       on 64-bit targets. Note that this changes the procedure call ABI
2596       for all procedures with character arguments on 64-bit targets, as
2597       the type of the hidden character length argument has changed. The
2598       hidden character length argument is now of type INTEGER(C_SIZE_T).
2599     * Partial support is provided for Fortran 2018 teams, which are
2600       hierarchical subsets of images that execute independently of other
2601       image subsets.
2602
2603  Go
2604
2605     * GCC 8 provides a complete implementation of the Go 1.10.1 user
2606       packages.
2607     * The garbage collector is now fully concurrent. As before, values
2608       stored on the stack are scanned conservatively, but value stored in
2609       the heap are scanned precisely.
2610     * Escape analysis is fully implemented and enabled by default in the
2611       Go front end. This significantly reduces the number of heap
2612       allocations by allocating values on the stack instead.
2613
2614libgccjit
2615
2616   The libgccjit API gained four new entry points:
2617     * [18]gcc_jit_type_get_vector and
2618     * [19]gcc_jit_context_new_rvalue_from_vector for working with
2619       vectors,
2620     * [20]gcc_jit_type_get_aligned
2621     * [21]gcc_jit_function_get_address
2622
2623   The C code generated by [22]gcc_jit_context_dump_reproducer_to_file is
2624   now easier-to-read.
2625
2626New Targets and Target Specific Improvements
2627
2628  AArch64
2629
2630     * The Armv8.4-A architecture is now supported. It can be used by
2631       specifying the -march=armv8.4-a option.
2632     * The Dot Product instructions are now supported as an optional
2633       extension to the Armv8.2-A architecture and newer and are mandatory
2634       on Armv8.4-A. The extension can be used by specifying the +dotprod
2635       architecture extension. E.g. -march=armv8.2-a+dotprod.
2636     * The Armv8-A +crypto extension has now been split into two
2637       extensions for finer grained control:
2638          + +aes which contains the Armv8-A AES crytographic instructions.
2639          + +sha2 which contains the Armv8-A SHA2 and SHA1 cryptographic
2640            instructions.
2641       Using +crypto will now enable these two extensions.
2642     * New Armv8.4-A FP16 Floating Point Multiplication Variant
2643       instructions have been added. These instructions are mandatory in
2644       Armv8.4-A but available as an optional extension to Armv8.2-A and
2645       Armv8.3-A. The new extension can be used by specifying the +fp16fml
2646       architectural extension on Armv8.2-A and Armv8.3-A. On Armv8.4-A
2647       the instructions can be enabled by specifying +fp16.
2648     * New cryptographic instructions have been added as optional
2649       extensions to Armv8.2-A and newer. These instructions can be
2650       enabled with:
2651          + +sha3 New SHA3 and SHA2 instructions from Armv8.4-A. This
2652            implies +sha2.
2653          + +sm4 New SM3 and SM4 instructions from Armv8.4-A.
2654     * The Scalable Vector Extension (SVE) is now supported as an optional
2655       extension to the Armv8.2-A architecture and newer. This support
2656       includes automatic vectorization with SVE instructions, but it does
2657       not yet include the SVE Arm C Language Extensions (ACLE). It can be
2658       enabled by specifying the +sve architecture extension (for example,
2659       -march=armv8.2-a+sve). By default, the generated code works with
2660       all vector lengths, but it can be made specific to N-bit vectors
2661       using -msve-vector-bits=N.
2662     * Support has been added for the following processors (GCC
2663       identifiers in parentheses):
2664          + Arm Cortex-A75 (cortex-a75).
2665          + Arm Cortex-A55 (cortex-a55).
2666          + Arm Cortex-A55/Cortex-A75 DynamIQ big.LITTLE
2667            (cortex-a75.cortex-a55).
2668       The GCC identifiers can be used as arguments to the -mcpu or -mtune
2669       options, for example: -mcpu=cortex-a75 or -mtune=cortex-a75 or as
2670       arguments to the equivalent target attributes and pragmas.
2671
2672  ARC
2673
2674     * Added support for:
2675          + Fast interrupts.
2676          + Naked functions.
2677          + aux variable attributes.
2678          + uncached type qualifier.
2679          + Secure functions via sjli instruction.
2680     * New exception handling implementation.
2681     * Revamped trampoline implementation.
2682     * Refactored small data feature implementation, controlled via -G
2683       command line option.
2684     * New support for reduced register set ARC architecture
2685       configurations, controlled via -mrf16 command line option.
2686     * Refurbished and improved support for zero overhead loops.
2687       Introduced -mlpc-width command line option to control the width of
2688       lp_count register.
2689
2690  ARM
2691
2692     * The -mfpu option now takes a new option setting of -mfpu=auto. When
2693       set to this the floating-point and SIMD settings are derived from
2694       the settings of the -mcpu or -march options. The internal CPU
2695       configurations have been updated with information about the
2696       permitted floating-point configurations supported. See the user
2697       guide for further information about the extended option syntax for
2698       controlling architectural extensions via the -march option.
2699       -mfpu=auto is now the default setting unless the compiler has been
2700       configured with an explicit --with-fpu option.
2701     * The -march and -mcpu options now accept optional extensions to the
2702       architecture or CPU option, allowing the user to enable or disable
2703       any such extensions supported by that architecture or CPU such as
2704       (but not limited to) floating-point and AdvancedSIMD. For example:
2705       the option -mcpu=cortex-a53+nofp will generate code for the
2706       Cortex-A53 processor with no floating-point support. This, in
2707       combination with the new -mfpu=auto option, provides a
2708       straightforward way of specifying a valid build target through a
2709       single -mcpu or -march option. The -mtune option accepts the same
2710       arguments as -mcpu but only the CPU name has an effect on tuning.
2711       The architecture extensions do not have any effect. For details of
2712       what extensions a particular architecture or CPU option supports
2713       please refer to the [23]documentation.
2714     * The -mstructure-size-boundary option has been deprecated and will
2715       be removed in a future release.
2716     * The default link behavior for Armv6 and Armv7-R targets has been
2717       changed to produce BE8 format when generating big-endian images. A
2718       new flag -mbe32 can be used to force the linker to produce legacy
2719       BE32 format images. There is no change of behavior for Armv6-M and
2720       other Armv7 or later targets: these already defaulted to BE8
2721       format. This change brings GCC into alignment with other compilers
2722       for the ARM architecture.
2723     * The Armv8-R architecture is now supported. It can be used by
2724       specifying the -march=armv8-r option.
2725     * The Armv8.3-A architecture is now supported. It can be used by
2726       specifying the -march=armv8.3-a option.
2727     * The Armv8.4-A architecture is now supported. It can be used by
2728       specifying the -march=armv8.4-a option.
2729     * The Dot Product instructions are now supported as an optional
2730       extension to the Armv8.2-A architecture and newer and are mandatory
2731       on Armv8.4-A. The extension can be used by specifying the +dotprod
2732       architecture extension. E.g. -march=armv8.2-a+dotprod.
2733     * Support for setting extensions and architectures using the GCC
2734       target pragma and attribute has been added. It can be used by
2735       specifying #pragma GCC target ("arch=..."), #pragma GCC target
2736       ("+extension"), __attribute__((target("arch=..."))) or
2737       __attribute__((target("+extension"))).
2738     * New Armv8.4-A FP16 Floating Point Multiplication Variant
2739       instructions have been added. These instructions are mandatory in
2740       Armv8.4-A but available as an optional extension to Armv8.2-A and
2741       Armv8.3-A. The new extension can be used by specifying the +fp16fml
2742       architectural extension on Armv8.2-A and Armv8.3-A. On Armv8.4-A
2743       the instructions can be enabled by specifying +fp16.
2744     * Support has been added for the following processors (GCC
2745       identifiers in parentheses):
2746          + Arm Cortex-A75 (cortex-a75).
2747          + Arm Cortex-A55 (cortex-a55).
2748          + Arm Cortex-A55/Cortex-A75 DynamIQ big.LITTLE
2749            (cortex-a75.cortex-a55).
2750          + Arm Cortex-R52 for Armv8-R (cortex-r52).
2751       The GCC identifiers can be used as arguments to the -mcpu or -mtune
2752       options, for example: -mcpu=cortex-a75 or -mtune=cortex-r52 or as
2753       arguments to the equivalent target attributes and pragmas.
2754
2755  AVR
2756
2757     * The AVR port now supports the following XMEGA-like devices:
2758
2759     ATtiny212, ATtiny214, ATtiny412, ATtiny414, ATtiny416, ATtiny417,
2760     ATtiny814, ATtiny816, ATtiny817, ATtiny1614, ATtiny1616, ATtiny1617,
2761     ATtiny3214, ATtiny3216, ATtiny3217
2762       The new devices are listed under [24]-mmcu=avrxmega3.
2763          + These devices see flash memory in the RAM address space, so
2764            that features like PROGMEM and __flash are not needed any more
2765            (as opposed to other AVR families for which read-only data
2766            will be located in RAM except special, non-standard features
2767            are used to locate and access such data). This requires that
2768            the compiler is used with Binutils 2.29 or newer so that
2769            [25]read-only data will be located in flash memory.
2770          + A new command-line option -mshort-calls is supported. This
2771            option is used internally for multilib selection of the
2772            avrxmega3 variants. It is not an optimization option. Do not
2773            set it by hand.
2774     * The compiler now generates [26]efficient interrupt service routine
2775       (ISR) prologues and epilogues. This is achieved by using the new
2776       [27]AVR pseudo instruction __gcc_isr which is supported and
2777       resolved by the GNU assembler.
2778          + As the __gcc_isr pseudo-instruction will be resolved by the
2779            assembler, inline assembly is transparent to the process. This
2780            means that when inline assembly uses an instruction like INC
2781            that clobbers the condition code, then the assembler will
2782            detect this and generate an appropriate ISR prologue /
2783            epilogue chunk to save / restore SREG as needed.
2784          + A new command-line option -mno-gas-isr-prologues disables the
2785            generation of the __gcc_isr pseudo instruction. Any non-naked
2786            ISR will save and restore SREG, tmp_reg and zero_reg, no
2787            matter whether the respective register is clobbered or used.
2788          + The feature is turned on per default for all optimization
2789            levels except for -O0 and -Og. It is explicitly enabled by
2790            means of option -mgas-isr-prologues.
2791          + Support has been added for a new [28]AVR function attribute
2792            no_gccisr. It can be used to disable __gcc_isr pseudo
2793            instruction generation for individual ISRs.
2794          + This optimization is only available if GCC is configured with
2795            GNU Binutils 2.29 or newer; or at least with a version of
2796            Binutils that implements feature [29]PR21683.
2797     * The compiler no more saves / restores registers in main; the effect
2798       is the same as if attribute OS_task was specified for main. This
2799       optimization can be switched off by the new command-line option
2800       -mno-main-is-OS_task.
2801
2802  IA-32/x86-64
2803
2804     * The x86 port now supports the naked function attribute.
2805     * Better tuning for znver1 and Intel Core based CPUs.
2806     * Vectorization cost metrics has been reworked leading to significant
2807       improvements on some benchmarks.
2808     * GCC now supports the Intel CPU named Cannonlake through
2809       -march=cannonlake. The switch enables the AVX512VBMI, AVX512IFMA
2810       and SHA ISA extensions.
2811     * GCC now supports the Intel CPU named Icelake through
2812       -march=icelake. The switch enables the AVX512VNNI, GFNI, VAES,
2813       AVX512VBMI2, VPCLMULQDQ, AVX512BITALG, RDPID and AVX512VPOPCNTDQ
2814       ISA extensions.
2815     * GCC now supports the Intel Control-flow Enforcement Technology
2816       (CET) extension through -fcf-protection option.
2817
2818  NDS32
2819
2820     * New command-line options -mext-perf, -mext-perf2, and -mext-string
2821       have been added for performance extension instructions.
2822
2823  Nios II
2824
2825     * The Nios II back end has been improved to generate better-optimized
2826       code. Changes include switching to LRA, more accurate cost models,
2827       and more compact code for addressing static variables.
2828     * New command-line options -mgprel-sec= and -mr0rel-sec= have been
2829       added.
2830     * The stack-smashing protection options are now enabled on Nios II.
2831
2832  PA-RISC
2833
2834     * The default call ABI on 32-bit linux has been changed from callee
2835       copies to caller copies. This affects objects larger than eight
2836       bytes passed by value. The goal is to improve compatibility with
2837       x86 and resolve issues with OpenMP.
2838     * Other PA-RISC targets are unchanged.
2839
2840  PowerPC / PowerPC64 / RS6000
2841
2842     * The PowerPC SPE support is split off to a separate powerpcspe port.
2843       The separate port is deprecated and might be removed in a future
2844       release.
2845     * The Paired Single support (as used on some PPC750 CPUs, -mpaired,
2846       powerpc*-*-linux*paired*) is deprecated and will be removed in a
2847       future release.
2848     * The Xilinx floating point support (-mxilinx-fpu,
2849       powerpc-xilinx-eabi*) is deprecated and will be removed in a future
2850       release.
2851     * Support for using big-endian AltiVec intrinsics on a little-endian
2852       target (-maltivec=be) is deprecated and will be removed in a future
2853       release.
2854
2855  Tile
2856
2857     * The TILE-Gx port is deprecated and will be removed in a future
2858       release.
2859
2860Operating Systems
2861
2862  Windows
2863
2864     * GCC on Microsoft Windows can now be configured via
2865       --enable-mingw-wildcard or --disable-mingw-wildcard to force a
2866       specific behavior for GCC itself with regards to supporting the
2867       wildcard character. Prior versions of GCC would follow the
2868       configuration of the MinGW runtime. This behavior can still be
2869       obtained by not using the above options or by using
2870       --enable-mingw-wildcard=platform.
2871
2872Improvements for plugin authors
2873
2874     * Plugins can now register a callback hook for when comments are
2875       encountered by the C and C++ compilers, e.g. allowing for plugins
2876       to handle documentation markup in code comments.
2877     * The gdbinit support script for debugging GCC now has a
2878       break-on-diagnostic command, providing an easy way to trigger a
2879       breakpoint whenever a diagnostic is emitted.
2880     * The API for creating fix-it hints now supports newlines, and for
2881       emitting mutually incompatible fix-it hints for one diagnostic.
2882
2883GCC 8.1
2884
2885   This is the [30]list of problem reports (PRs) from GCC's bug tracking
2886   system that are known to be fixed in the 8.1 release. This list might
2887   not be complete (that is, it is possible that some PRs that have been
2888   fixed are not listed here).
2889
2890GCC 8.2
2891
2892   This is the [31]list of problem reports (PRs) from GCC's bug tracking
2893   system that are known to be fixed in the 8.2 release. This list might
2894   not be complete (that is, it is possible that some PRs that have been
2895   fixed are not listed here).
2896
2897  General Improvements
2898
2899     * Fixed LTO link-time performance problems caused by an overflow in
2900       the partitioning algorithm while building large binaries.
2901
2902  Language Specific Changes
2903
2904    C++
2905
2906   GCC 8.2 fixed a bug introduced in GCC 8.1 affecting passing or
2907   returning of classes with a deleted copy constructor and defaulted
2908   trivial move constructor (bug [32]c++/86094). GCC 8.2 introduces
2909   -fabi-version=13 and makes it the default, ABI incompatibilities
2910   between GCC 8.1 and 8.2 can be reported with -Wabi=12. See [33]C++
2911   changes for more details.
2912
2913  Target Specific Changes
2914
2915    IA-32/x86-64
2916
2917     * -mtune=native performance regression [34]PR84413 on Intel Skylake
2918       processors has been fixed.
2919
2920GCC 8.3
2921
2922   This is the [35]list of problem reports (PRs) from GCC's bug tracking
2923   system that are known to be fixed in the 8.3 release. This list might
2924   not be complete (that is, it is possible that some PRs that have been
2925   fixed are not listed here).
2926
2927  Windows
2928
2929     * A C++ Microsoft ABI bitfield layout bug, [36]PR87137 has been
2930       fixed. A non-field declaration could cause the current bitfield
2931       allocation unit to be completed, incorrectly placing a following
2932       bitfield into a new allocation unit. The Microsoft ABI is selected
2933       for:
2934          + Mingw targets
2935          + PowerPC, IA-32 or x86-64 targets when the -mms-bitfields
2936            option is specified, or __attribute__((ms_struct)) is used
2937          + SuperH targets when the -mhitachi option is specified, or
2938            __attribute__((renesas)) is used
2939       GCC 8 introduced additional cases of this defect, but rather than
2940       resolve only those regressions, we decided to resolve all the cases
2941       of this defect in single change.
2942
2943GCC 8.4
2944
2945   This is the [37]list of problem reports (PRs) from GCC's bug tracking
2946   system that are known to be fixed in the 8.4 release. This list might
2947   not be complete (that is, it is possible that some PRs that have been
2948   fixed are not listed here).
2949
2950GCC 8.5
2951
2952  Target Specific Changes
2953
2954    AArch64
2955
2956     * The option -moutline-atomics has been added to aid deployment of
2957       the Large System Extensions (LSE) on GNU/Linux systems built with a
2958       baseline architecture targeting Armv8-A. When the option is
2959       specified code is emitted to detect the presence of LSE
2960       instructions at run time and use them for standard atomic
2961       operations. For more information please refer to the documentation.
2962
2963
2964    For questions related to the use of GCC, please consult these web
2965    pages and the [38]GCC manuals. If that fails, the
2966    [39]gcc-help@gcc.gnu.org mailing list might help. Comments on these
2967    web pages and the development of GCC are welcome on our developer
2968    list at [40]gcc@gcc.gnu.org. All of [41]our lists have public
2969    archives.
2970
2971   Copyright (C) [42]Free Software Foundation, Inc. Verbatim copying and
2972   distribution of this entire article is permitted in any medium,
2973   provided this notice is preserved.
2974
2975   These pages are [43]maintained by the GCC team. Last modified
2976   2021-03-28[44].
2977
2978References
2979
2980   1. http://gcc.gnu.org/gcc-8/porting_to.html
2981   2. http://gcc.gnu.org/onlinedocs/index.html#current
2982   3. https://gcc.gnu.org/ml/gcc/2018-04/msg00102.html
2983   4. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Warning-Options.html#index-Wmultistatement-macros
2984   5. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Warning-Options.html#index-Wstringop-truncation
2985   6. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82944
2986   7. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Warning-Options.html#index-Wif-not-aligned
2987   8. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Common-Variable-Attributes.html#index-warn_005fif_005fnot_005faligned-variable-attribute
2988   9. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Warning-Options.html#index-Wmissing-attributes
2989  10. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Warning-Options.html#index-Wpacked-not-aligned
2990  11. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Warning-Options.html#index-Warray-bounds
2991  12. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Warning-Options.html#index-Wrestrict
2992  13. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Warning-Options.html#index-Wformat-overflow
2993  14. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/Warning-Options.html#index-Wformat-truncation
2994  15. https://gcc.gnu.org/PR86094
2995  16. https://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Dialect-Options.html#index-Wclass-memaccess
2996  17. http://gcc.gnu.org/projects/cxx-status.html#cxx2a
2997  18. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/jit/topics/types.html#gcc_jit_type_get_vector
2998  19. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/jit/topics/expressions.html#gcc_jit_context_new_rvalue_from_vector
2999  20. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/jit/topics/types.html#gcc_jit_type_get_aligned
3000  21. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/jit/topics/function-pointers.html#gcc_jit_function_get_address
3001  22. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/jit/topics/contexts.html#gcc_jit_context_dump_reproducer_to_file
3002  23. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/ARM-Options.html#ARM-Options
3003  24. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/AVR-Options.html
3004  25. https://sourceware.org/PR21472
3005  26. https://gcc.gnu.org/PR20296
3006  27. https://sourceware.org/binutils/docs-2.29/as/AVR-Pseudo-Instructions.html
3007  28. https://gcc.gnu.org/onlinedocs/gcc-8.1.0/gcc/AVR-Function-Attributes.html
3008  29. https://sourceware.org/PR21683
3009  30. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=8.0
3010  31. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=8.2
3011  32. https://gcc.gnu.org/PR86094
3012  33. http://gcc.gnu.org/gcc-8/changes.html#cxx
3013  34. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84413
3014  35. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=8.3
3015  36. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87137
3016  37. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=8.4
3017  38. https://gcc.gnu.org/onlinedocs/
3018  39. mailto:gcc-help@gcc.gnu.org
3019  40. mailto:gcc@gcc.gnu.org
3020  41. https://gcc.gnu.org/lists.html
3021  42. https://www.fsf.org/
3022  43. https://gcc.gnu.org/about.html
3023  44. http://validator.w3.org/check/referer
3024======================================================================
3025http://gcc.gnu.org/gcc-7/index.html
3026
3027                              GCC 7 Release Series
3028
3029   Nov 14, 2019
3030
3031   The [1]GNU project and the GCC developers are pleased to announce the
3032   release of GCC 7.5.
3033
3034   This release is a bug-fix release, containing fixes for regressions in
3035   GCC 7.4 relative to previous releases of GCC.
3036
3037Release History
3038
3039   GCC 7.5
3040          Nov 14, 2019 ([2]changes, [3]documentation)
3041
3042   GCC 7.4
3043          Dec 6, 2018 ([4]changes, [5]documentation)
3044
3045   GCC 7.3
3046          Jan 25, 2018 ([6]changes, [7]documentation)
3047
3048   GCC 7.2
3049          Aug 14, 2017 ([8]changes, [9]documentation)
3050
3051   GCC 7.1
3052          May 2, 2017 ([10]changes, [11]documentation)
3053
3054References and Acknowledgements
3055
3056   GCC used to stand for the GNU C Compiler, but since the compiler
3057   supports several other languages aside from C, it now stands for the
3058   GNU Compiler Collection.
3059
3060   A list of [12]successful builds is updated as new information becomes
3061   available.
3062
3063   The GCC developers would like to thank the numerous people that have
3064   contributed new features, improvements, bug fixes, and other changes as
3065   well as test results to GCC. This [13]amazing group of volunteers is
3066   what makes GCC successful.
3067
3068   For additional information about GCC please refer to the [14]GCC
3069   project web site or contact the [15]GCC development mailing list.
3070
3071   To obtain GCC please use [16]our mirror sites or [17]our version
3072   control system.
3073
3074
3075    For questions related to the use of GCC, please consult these web
3076    pages and the [18]GCC manuals. If that fails, the
3077    [19]gcc-help@gcc.gnu.org mailing list might help. Comments on these
3078    web pages and the development of GCC are welcome on our developer
3079    list at [20]gcc@gcc.gnu.org. All of [21]our lists have public
3080    archives.
3081
3082   Copyright (C) [22]Free Software Foundation, Inc. Verbatim copying and
3083   distribution of this entire article is permitted in any medium,
3084   provided this notice is preserved.
3085
3086   These pages are [23]maintained by the GCC team. Last modified
3087   2020-01-14[24].
3088
3089References
3090
3091   1. http://www.gnu.org/
3092   2. http://gcc.gnu.org/gcc-7/changes.html
3093   3. http://gcc.gnu.org/onlinedocs/7.5.0/
3094   4. http://gcc.gnu.org/gcc-7/changes.html
3095   5. http://gcc.gnu.org/onlinedocs/7.4.0/
3096   6. http://gcc.gnu.org/gcc-7/changes.html
3097   7. http://gcc.gnu.org/onlinedocs/7.3.0/
3098   8. http://gcc.gnu.org/gcc-7/changes.html
3099   9. http://gcc.gnu.org/onlinedocs/7.2.0/
3100  10. http://gcc.gnu.org/gcc-7/changes.html
3101  11. http://gcc.gnu.org/onlinedocs/7.1.0/
3102  12. http://gcc.gnu.org/gcc-7/buildstat.html
3103  13. http://gcc.gnu.org/onlinedocs/gcc/Contributors.html
3104  14. http://gcc.gnu.org/index.html
3105  15. mailto:gcc@gcc.gnu.org
3106  16. http://gcc.gnu.org/mirrors.html
3107  17. http://gcc.gnu.org/git.html
3108  18. https://gcc.gnu.org/onlinedocs/
3109  19. mailto:gcc-help@gcc.gnu.org
3110  20. mailto:gcc@gcc.gnu.org
3111  21. https://gcc.gnu.org/lists.html
3112  22. https://www.fsf.org/
3113  23. https://gcc.gnu.org/about.html
3114  24. http://validator.w3.org/check/referer
3115======================================================================
3116http://gcc.gnu.org/gcc-7/changes.html
3117
3118                              GCC 7 Release Series
3119                        Changes, New Features, and Fixes
3120
3121   This page is a brief summary of some of the huge number of improvements
3122   in GCC 7. For more information, see the [1]Porting to GCC 7 page and
3123   the [2]full GCC documentation.
3124
3125Caveats
3126
3127     * GCC now uses [3]LRA (a new local register allocator) by default for
3128       new targets.
3129     * The non-standard C++0x type traits has_trivial_default_constructor,
3130       has_trivial_copy_constructor and has_trivial_copy_assign have been
3131       removed.
3132     * The libstdc++ [4]Profile Mode has been deprecated and will be
3133       removed in a future version.
3134     * The Cilk+ extensions to the C and C++ languages have been
3135       deprecated.
3136     * On ARM targets (arm*-*-*), [5]a bug introduced in GCC 5 that
3137       affects conformance to the procedure call standard (AAPCS) has been
3138       fixed. The bug affects some C++ code where class objects are passed
3139       by value to functions and could result in incorrect or inconsistent
3140       code being generated. This is an ABI change. If the option -Wpsabi
3141       is enabled (on by default) the compiler will emit a diagnostic note
3142       for code that might be affected.
3143
3144General Optimizer Improvements
3145
3146     * GCC 7 can determine the return value or range of return values of
3147       some calls to the sprintf family of functions and make it available
3148       to other optimization passes. Some calls to the snprintf function
3149       with a zero size argument can be folded into constants. This
3150       optimization is included in -O1 and can be selectively controlled
3151       by the -fprintf-return-value option.
3152     * A new store merging pass has been added. It merges constant stores
3153       to adjacent memory locations into fewer, wider, stores. It is
3154       enabled by the -fstore-merging option and at the -O2 optimization
3155       level or higher (and -Os).
3156     * A new code hoisting optimization has been added to the partial
3157       redundancy elimination pass. It attempts to move evaluation of
3158       expressions executed on all paths to the function exit as early as
3159       possible. This primarily helps improve code size, but can improve
3160       the speed of the generated code as well. It is enabled by the
3161       -fcode-hoisting option and at the -O2 optimization level or higher
3162       (and -Os).
3163     * A new interprocedural bitwise constant propagation optimization has
3164       been added, which propagates knowledge about which bits of
3165       variables are known to be zero (including pointer alignment
3166       information) across the call graph. It is enabled by the
3167       -fipa-bit-cp option if -fipa-cp is enabled as well, and is enabled
3168       at the -O2 optimization level and higher (and -Os). This
3169       optimization supersedes interprocedural alignment propagation of
3170       GCC 6, and therefore the option -fipa-cp-alignment is now
3171       deprecated and ignored.
3172     * A new interprocedural value range propagation optimization has been
3173       added, which propagates integral range information across the call
3174       graph when variable values can be proven to be within those ranges.
3175       It is enabled by the -fipa-vrp option and at the -O2 optimization
3176       level and higher (and -Os).
3177     * A new loop splitting optimization pass has been added. Certain
3178       loops which contain a condition that is always true on one side of
3179       the iteration space and always false on the other are split into
3180       two loops, such that each of the two new loops iterates on just one
3181       side of the iteration space and the condition does not need to be
3182       checked inside of the loop. It is enabled by the -fsplit-loops
3183       option and at the -O3 optimization level or higher.
3184     * The shrink-wrapping optimization can now separate portions of
3185       prologues and epilogues to improve performance if some of the work
3186       done traditionally by prologues and epilogues is not needed on
3187       certain paths. This is controlled by the -fshrink-wrap-separate
3188       option, enabled by default. It requires target support, which is
3189       currently only implemented in the PowerPC and AArch64 ports.
3190     * AddressSanitizer gained a new sanitization option,
3191       -fsanitize-address-use-after-scope, which enables sanitization of
3192       variables whose address is taken and used after a scope where the
3193       variable is defined:
3194
3195int
3196main (int argc, char **argv)
3197{
3198  char *ptr;
3199    {
3200      char my_char;
3201      ptr = &my_char;
3202    }
3203
3204  *ptr = 123;
3205  return *ptr;
3206}
3207
3208==28882==ERROR: AddressSanitizer: stack-use-after-scope on address 0x7fffb8dba99
32090 at pc 0x0000004006d5 bp 0x7fffb8dba960 sp 0x7fffb8dba958
3210WRITE of size 1 at 0x7fffb8dba990 thread T0
3211    #0 0x4006d4 in main /tmp/use-after-scope-1.c:10
3212    #1 0x7f9c71943290 in __libc_start_main (/lib64/libc.so.6+0x20290)
3213    #2 0x400739 in _start (/tmp/a.out+0x400739)
3214
3215Address 0x7fffb8dba990 is located in stack of thread T0 at offset 32 in frame
3216    #0 0x40067f in main /tmp/use-after-scope-1.c:3
3217
3218  This frame has 1 object(s):
3219    [32, 33) 'my_char' <== Memory access at offset 32 is inside this variable
3220
3221       The option is enabled by default with -fsanitize=address and
3222       disabled by default with -fsanitize=kernel-address. Compared to the
3223       LLVM compiler, where the option already exists, the implementation
3224       in the GCC compiler has some improvements and advantages:
3225          + Complex uses of gotos and case labels are properly handled and
3226            should not report any false positive or false negatives.
3227          + C++ temporaries are sanitized.
3228          + Sanitization can handle invalid memory stores that are
3229            optimized out by the LLVM compiler when optimization is
3230            enabled.
3231     * The -fsanitize=signed-integer-overflow suboption of the
3232       UndefinedBehavior Sanitizer now diagnoses arithmetic overflows even
3233       on arithmetic operations with generic vectors.
3234     * Version 5 of the [6]DWARF debugging information standard is
3235       supported through the -gdwarf-5 option. The DWARF version 4
3236       debugging information remains the default until consumers of
3237       debugging information are adjusted.
3238
3239New Languages and Language specific improvements
3240
3241   OpenACC support in C, C++, and Fortran continues to be maintained and
3242   improved. See the [7]OpenACC and [8]Offloading wiki pages for further
3243   information.
3244
3245  Ada
3246
3247     * On mainstream native platforms, Ada programs no longer require the
3248       stack to be made executable in order to run properly.
3249
3250  BRIG (HSAIL)
3251
3252   Support for processing BRIG 1.0 files was added in this release. BRIG
3253   is a binary format for HSAIL (Heterogeneous System Architecture
3254   Intermediate Language). The BRIG front end can be used for implementing
3255   HSAIL "finalizers" (compilation of HSAIL to a native ISA) for
3256   GCC-supported targets. An implementation of an HSAIL runtime library,
3257   libhsail-rt is also included.
3258
3259  C family
3260
3261     * New command-line options have been added for the C and C++
3262       compilers:
3263          + -Wimplicit-fallthrough warns when a switch case falls through.
3264            This warning has five different levels. The compiler is able
3265            to parse a wide range of fallthrough comments, depending on
3266            the level. It also handles control-flow statements, such as
3267            ifs. It's possible to suppress the warning by either adding a
3268            fallthrough comment, or by using a null statement:
3269            __attribute__ ((fallthrough)); (C, C++), or [[fallthrough]];
3270            (C++17), or [[gnu::fallthrough]]; (C++11/C++14). This warning
3271            is enabled by -Wextra.
3272          + -Wpointer-compare warns when a pointer is compared with a zero
3273            character constant. Such code is now invalid in C++11 and GCC
3274            rejects it. This warning is enabled by default.
3275          + -Wduplicated-branches warns when an if-else has identical
3276            branches.
3277          + -Wrestrict warns when an argument passed to a
3278            restrict-qualified parameter aliases with another argument.
3279          + -Wmemset-elt-size warns for memset calls, when the first
3280            argument references an array, and the third argument is a
3281            number equal to the number of elements of the array, but not
3282            the size of the array. This warning is enabled by -Wall.
3283          + -Wint-in-bool-context warns about suspicious uses of integer
3284            values where boolean values are expected. This warning is
3285            enabled by -Wall.
3286          + -Wswitch-unreachable warns when a switch statement has
3287            statements between the controlling expression and the first
3288            case label which will never be executed. This warning is
3289            enabled by default.
3290          + -Wexpansion-to-defined warns when defined is used outside #if.
3291            This warning is enabled by -Wextra or -Wpedantic.
3292          + -Wregister warns about uses of the register storage specifier.
3293            In C++17 this keyword has been removed and for C++17 this is a
3294            pedantic warning enabled by default. The warning is not
3295            emitted for the GNU Explicit Register Variables extension.
3296          + -Wvla-larger-than=N warns about unbounded uses of
3297            variable-length arrays, and about bounded uses of
3298            variable-length arrays whose bound can be larger than N bytes.
3299          + -Wduplicate-decl-specifier warns when a declaration has
3300            duplicate const, volatile, restrict or _Atomic specifier. This
3301            warning is enabled by -Wall.
3302     * GCC 6's C and C++ front ends were able to offer suggestions for
3303       misspelled field names:
3304
3305spellcheck-fields.cc:52:13: error: 'struct s' has no member named 'colour'; did
3306you mean 'color'?
3307   return ptr->colour;
3308               ^~~~~~
3309
3310       GCC 7 greatly expands the scope of these suggestions. Firstly, it
3311       adds fix-it hints to such suggestions:
3312
3313spellcheck-fields.cc:52:13: error: 'struct s' has no member named 'colour'; did
3314you mean 'color'?
3315   return ptr->colour;
3316               ^~~~~~
3317               color
3318
3319       The suggestions now cover many other things, such as misspelled
3320       function names:
3321
3322spellcheck-identifiers.c:11:3: warning: implicit declaration of function 'gtk_wi
3323dget_showall'; did you mean 'gtk_widget_show_all'? [-Wimplicit-function-declarat
3324ion]
3325   gtk_widget_showall (w);
3326   ^~~~~~~~~~~~~~~~~~
3327   gtk_widget_show_all
3328
3329       misspelled macro names and enum values:
3330
3331spellcheck-identifiers.cc:85:11: error: 'MAX_ITEM' undeclared here (not in a fun
3332ction); did you mean 'MAX_ITEMS'?
3333 int array[MAX_ITEM];
3334           ^~~~~~~~
3335           MAX_ITEMS
3336
3337       misspelled type names:
3338
3339spellcheck-typenames.c:7:14: error: unknown type name 'singed'; did you mean 'si
3340gned'?
3341 void test (singed char e);
3342            ^~~~~~
3343            signed
3344
3345       and, in the C front end, named initializers:
3346
3347test.c:7:20: error: 'struct s' has no member named 'colour'; did you mean 'color
3348'?
3349 struct s test = { .colour = 3 };
3350                    ^~~~~~
3351                    color
3352
3353     * The preprocessor can now offer suggestions for misspelled
3354       directives, e.g.:
3355
3356test.c:5:2: error:invalid preprocessing directive #endfi; did you mean #endif?
3357 #endfi
3358  ^~~~~
3359  endif
3360
3361     * Warnings about format strings now underline the pertinent part of
3362       the string, and can offer suggested fixes. In some cases, the
3363       pertinent argument is underlined.
3364
3365test.c:51:29: warning: format '%s' expects argument of type 'char *', but argume
3366nt 3 has type 'int' [-Wformat=]
3367   printf ("foo: %d  bar: %s baz: %d", 100, i + j, 102);
3368                          ~^                ~~~~~
3369                          %d
3370
3371     * The new -Wdangling-else command-line option has been split out of
3372       -Wparentheses and warns about dangling else.
3373     * The -Wshadow warning has been split into three variants.
3374       -Wshadow=global warns for any shadowing. This is the default when
3375       using -Wshadow without any argument. -Wshadow=local only warns for
3376       a local variable shadowing another local variable or parameter.
3377       -Wshadow=compatible-local only warns for a local variable shadowing
3378       another local variable or parameter whose type is compatible (in
3379       C++ compatible means that the type of the shadowing variable can be
3380       converted to that of the shadowed variable).
3381       The following example shows the different kinds of shadow warnings:
3382
3383enum operation { add, count };
3384struct container { int nr; };
3385
3386int
3387container_count (struct container c, int count)
3388{
3389  int r = 0;
3390  for (int count = 0; count > 0; count--)
3391    {
3392      struct container count = c;
3393      r += count.nr;
3394    }
3395  return r;
3396}
3397
3398       -Wshadow=compatible-local will warn for the parameter being
3399       shadowed with the same type:
3400
3401warn-test.c:8:12: warning: declaration of 'count' shadows a parameter [-Wshadow=
3402compatible-local]
3403   for (int count = 0; count > 0; count--)
3404            ^~~~~
3405warn-test.c:5:42: note: shadowed declaration is here
3406 container_count (struct container c, int count)
3407                                          ^~~~~
3408
3409       -Wshadow=local will warn for the above and for the shadowed
3410       declaration with incompatible type:
3411
3412warn-test.c:10:24: warning: declaration of 'count' shadows a previous local [-Ws
3413hadow=local]
3414       struct container count = c;
3415                        ^~~~~
3416warn-test.c:8:12: note: shadowed declaration is here
3417   for (int count = 0; count > 0; count--)
3418            ^~~~~
3419
3420       -Wshadow=global will warn for all of the above and the shadowing of
3421       the global declaration:
3422
3423warn-test.c:5:42: warning: declaration of 'count' shadows a global declaration [
3424-Wshadow]
3425 container_count (struct container c, int count)
3426                                          ^~~~~
3427warn-test.c:1:23: note: shadowed declaration is here
3428 enum operation { add, count };
3429                       ^~~~~
3430
3431     * GCC 7 contains a number of enhancements that help detect buffer
3432       overflow and other forms of invalid memory accesses.
3433          + The -Walloc-size-larger-than=size option detects calls to
3434            standard and user-defined memory allocation functions
3435            decorated with attribute alloc_size whose argument exceeds the
3436            specified size (PTRDIFF_MAX by default). The option also
3437            detects arithmetic overflow in the computation of the size in
3438            two-argument allocation functions like calloc where the total
3439            size is the product of the two arguments. Since calls with an
3440            excessive size cannot succeed they are typically the result of
3441            programming errors. Such bugs have been known to be the source
3442            of security vulnerabilities and a target of exploits.
3443            -Walloc-size-larger-than=PTRDIFF_MAX is included in -Wall.
3444            For example, the following call to malloc incorrectly tries to
3445            avoid passing a negative argument to the function and instead
3446            ends up unconditionally invoking it with an argument less than
3447            or equal to zero. Since after conversion to the type of the
3448            argument of the function (size_t) a negative argument results
3449            in a value in excess of the maximum PTRDIFF_MAX the call is
3450            diagnosed.
3451
3452void* f (int n)
3453{
3454  return malloc (n > 0 ? 0 : n);
3455}
3456
3457warning: argument 1 range [2147483648, 4294967295] exceeds maximum object size 2
3458147483647 [-Walloc-size-larger-than=]
3459
3460          + The -Walloc-zero option detects calls to standard and
3461            user-defined memory allocation functions decorated with
3462            attribute alloc_size with a zero argument. -Walloc-zero is not
3463            included in either -Wall or -Wextra and must be explicitly
3464            enabled.
3465          + The -Walloca option detects all calls to the alloca function
3466            in the program. -Walloca is not included in either -Wall or
3467            -Wextra and must be explicitly enabled.
3468          + The -Walloca-larger-than=size option detects calls to the
3469            alloca function whose argument either may exceed the specified
3470            size, or that is not known to be sufficiently constrained to
3471            avoid exceeding it. -Walloca-larger-than is not included in
3472            either -Wall or -Wextra and must be explicitly enabled.
3473            For example, compiling the following snippet with
3474            -Walloca-larger-than=1024 results in a warning because even
3475            though the code appears to call alloca only with sizes of 1kb
3476            and less, since n is signed, a negative value would result in
3477            a call to the function well in excess of the limit.
3478
3479void f (int n)
3480{
3481  char *d;
3482  if (n < 1025)
3483    d = alloca (n);
3484  else
3485    d = malloc (n);
3486  ...
3487}
3488
3489warning: argument to 'alloca may be too large due to conversion from 'int' to 'l
3490ong unsigned int' [-Walloca-larger-than=]
3491
3492            In contrast, a call to alloca that isn't bounded at all such
3493            as in the following function will elicit the warning below
3494            regardless of the size argument to the option.
3495
3496void f (size_t n)
3497{
3498  char *d = alloca (n);
3499  ...
3500}
3501
3502warning: unbounded use of 'alloca' [-Walloca-larger-than=]
3503
3504          + The -Wformat-overflow=level option detects certain and likely
3505            buffer overflow in calls to the sprintf family of formatted
3506            output functions. Although the option is enabled even without
3507            optimization it works best with -O2 and higher.
3508            For example, in the following snippet the call to sprintf is
3509            diagnosed because even though its output has been constrained
3510            using the modulo operation it could result in as many as three
3511            bytes if mday were negative. The solution is to either
3512            allocate a larger buffer or make sure the argument is not
3513            negative, for example by changing mday's type to unsigned or
3514            by making the type of the second operand of the modulo
3515            expression unsigned: 100U.
3516
3517void* f (int mday)
3518{
3519  char *buf = malloc (3);
3520  sprintf (buf, "%02i", mday % 100);
3521  return buf;
3522}
3523
3524warning: 'sprintf may write a terminating nul past the end of the destination [-
3525Wformat-overflow=]
3526note: 'sprintf' output between 3 and 4 bytes into a destination of size 3
3527
3528          + The -Wformat-truncation=level option detects certain and
3529            likely output truncation in calls to the snprintf family of
3530            formatted output functions. -Wformat-truncation=1 is included
3531            in -Wall and enabled without optimization but works best with
3532            -O2 and higher.
3533            For example, the following function attempts to format an
3534            integer between 0 and 255 in hexadecimal, including the 0x
3535            prefix, into a buffer of four characters. But since the
3536            function must always terminate output by the null character
3537            ('\0') such a buffer is only big enough to fit just one digit
3538            plus the prefix. Therefore the snprintf call is diagnosed. To
3539            avoid the warning either use a bigger buffer or handle the
3540            function's return value which indicates whether or not its
3541            output has been truncated.
3542
3543void f (unsigned x)
3544{
3545  char d[4];
3546  snprintf (d, sizeof d, "%#02x", x & 0xff);
3547  ...
3548}
3549
3550warning: 'snprintf' output may be truncated before the last format character [-W
3551format-truncation=]
3552note: 'snprintf' output between 3 and 5 bytes into a destination of size 4
3553
3554          + The -Wnonnull option has been enhanced to detect a broader set
3555            of cases of passing null pointers to functions that expect a
3556            non-null argument (those decorated with attribute nonnull). By
3557            taking advantage of optimizations the option can detect many
3558            more cases of the problem than in prior GCC versions.
3559          + The -Wstringop-overflow=type option detects buffer overflow in
3560            calls to string handling functions like memcpy and strcpy. The
3561            option relies on [9]Object Size Checking and has an effect
3562            similar to defining the _FORTIFY_SOURCE macro.
3563            -Wstringop-overflow=2 is enabled by default.
3564            For example, in the following snippet, because the call to
3565            strncat specifies a maximum that allows the function to write
3566            past the end of the destination, it is diagnosed. To correct
3567            the problem and avoid the overflow the function should be
3568            called with a size of at most sizeof d - strlen(d) - 1.
3569
3570void f (const char *fname)
3571{
3572  char d[8];
3573  strncpy (d, "/tmp/", sizeof d);
3574  strncat (d, fname, sizeof d);
3575  ...
3576}
3577
3578warning: specified bound 8 equals the size of the destination [-Wstringop-overfl
3579ow=]
3580
3581     * The <limits.h> header provided by GCC defines macros such as
3582       INT_WIDTH for the width in bits of integer types, if
3583       __STDC_WANT_IEC_60559_BFP_EXT__ is defined before the header is
3584       included. The <stdint.h> header defines such macros as SIZE_WIDTH
3585       and INTMAX_WIDTH for the width of some standard typedef names for
3586       integer types, again if __STDC_WANT_IEC_60559_BFP_EXT__ is defined
3587       before the header is included; note that GCC's implementation of
3588       this header is only used for freestanding compilations, not hosted
3589       compilations, on most systems. These macros come from ISO/IEC TS
3590       18661-1:2014.
3591     * The <float.h> header provided by GCC defines the macro
3592       CR_DECIMAL_DIG, from ISO/IEC TS 18661-1:2014, if
3593       __STDC_WANT_IEC_60559_BFP_EXT__ is defined before the header is
3594       included. This represents the number of decimal digits for which
3595       conversions between decimal character strings and binary formats,
3596       in both directions, are correctly rounded, and currently has the
3597       value of UINTMAX_MAX on all systems, reflecting that GCC's
3598       compile-time conversions are correctly rounded for any number of
3599       digits.
3600     * New __builtin_add_overflow_p, __builtin_sub_overflow_p,
3601       __builtin_mul_overflow_p built-in functions have been added. These
3602       work similarly to their siblings without the _p suffix, but do not
3603       actually store the result of the arithmetics anywhere, just return
3604       whether the operation would overflow. Calls to these built-ins with
3605       integer constant arguments evaluate to integer constants
3606       expressions.
3607       For example, in the following, c is assigned the result of a * b
3608       only if the multiplication does not overflow, otherwise it is
3609       assigned the value zero. The multiplication is performed at
3610       compile-time and without triggering a -Woverflow warning.
3611
3612enum {
3613  a = 12345678,
3614  b = 87654321,
3615  c = __builtin_mul_overflow_p (a, b, a) ? 0 : a * b
3616};
3617
3618  C
3619
3620     * The C front end now supports type names _FloatN for floating-point
3621       types with IEEE interchange formats and _FloatNx for floating-point
3622       types with IEEE extended formats. These type names come from
3623       ISO/IEC TS 18661-3:2015.
3624       The set of types supported depends on the target for which GCC is
3625       configured. Most targets support _Float32, _Float32x and _Float64.
3626       _Float128 is supported on targets where IEEE binary128 encoding was
3627       already supported as long double or __float128. _Float64x is
3628       supported on targets where a type with either binary128 or Intel
3629       extended precision format is available.
3630       Constants with these types are supported using suffixes fN, FN, fNx
3631       and FNx (e.g., 1.2f128 or 2.3F64x). Macros such as FLT128_MAX are
3632       defined in <float.h> if __STDC_WANT_IEC_60559_TYPES_EXT__ is
3633       defined before it is included.
3634       These new types are always distinct from each other and from float,
3635       double and long double, even if they have the same encoding.
3636       Complex types such as _Complex _Float128 are also supported.
3637       Type-generic built-in functions such as __builtin_isinf support the
3638       new types, and the following type-specific built-in functions have
3639       versions (suffixed fN or fNx) for the new types:
3640       __builtin_copysign, __builtin_fabs, __builtin_huge_val,
3641       __builtin_inf, __builtin_nan, __builtin_nans.
3642     * Compilation with -fopenmp is now compatible with the C11 _Atomic
3643       keyword.
3644
3645  C++
3646
3647     * The C++ front end has experimental support for all of the current
3648       C++17 draft with the -std=c++1z or -std=gnu++1z flags, including if
3649       constexpr, class template argument deduction, auto template
3650       parameters, and structured bindings. For a full list of new
3651       features, see [10]the C++ status page.
3652     * C++17 support for new of over-aligned types can be enabled in other
3653       modes with the -faligned-new flag.
3654     * The C++17 evaluation order requirements can be selected in other
3655       modes with the -fstrong-eval-order flag, or disabled in C++17 mode
3656       with -fno-strong-eval-order.
3657     * The default semantics of inherited constructors has changed in all
3658       modes, following [11]P0136. Essentially, overload resolution
3659       happens as if calling the inherited constructor directly, and the
3660       compiler fills in construction of the other bases and members as
3661       needed. Most uses should not need any changes. The old behavior can
3662       be restored with -fno-new-inheriting-ctors, or -fabi-version less
3663       than 11.
3664     * The resolution of DR 150 on matching of template template
3665       parameters, allowing default template arguments to make a template
3666       match a parameter, is currently enabled by default in C++17 mode
3667       only. The default can be overridden with -f{no-,}new-ttp-matching.
3668     * The C++ front end will now provide fix-it hints for some missing
3669       semicolons, allowing for automatic fixes by IDEs:
3670
3671test.cc:4:11: error: expected ';' after class definition
3672 class a {}
3673           ^
3674           ;
3675
3676     * -Waligned-new has been added to the C++ front end. It warns about
3677       new of type with extended alignment without -faligned-new.
3678
3679    Runtime Library (libstdc++)
3680
3681     * The type of exception thrown by iostreams, std::ios_base::failure,
3682       now uses the [12]cxx11 ABI.
3683     * Experimental support for C++17, including the following new
3684       features:
3685          + std::string_view;
3686          + std::any, std::optional, and std::variant;
3687          + std::invoke, std::is_invocable, std::is_nothrow_invocable, and
3688            invoke_result;
3689          + std::is_swappable, and std::is_nothrow_swappable;
3690          + std::apply, and std::make_from_tuple;
3691          + std::void_t, std::bool_constant, std::conjunction,
3692            std::disjunction, and std::negation;
3693          + Variable templates for type traits;
3694          + Mathematical Special Functions;
3695          + std::chrono::floor, std::chrono::ceil, std::chrono::round, and
3696            std::chrono::abs;
3697          + std::clamp, std::gcd, std::lcm, 3-dimensional std::hypot;
3698          + std::scoped_lock, std::shared_mutex,
3699            std::atomic<T>::is_always_lock_free;
3700          + std::sample, std::default_searcher, std::boyer_moore_searcher
3701            and std::boyer_moore_horspool_searcher;
3702          + Extraction and re-insertion of map and set nodes, try_emplace
3703            members for maps, and functions for accessing containers
3704            std::size, std::empty, and std::data;
3705          + std::shared_ptr support for arrays,
3706            std::shared_ptr<T>::weak_type,
3707            std::enable_shared_from_this<T>::weak_from_this(), and
3708            std::owner_less<void>;
3709          + std::byte;
3710          + std::as_const, std::not_fn,
3711            std::has_unique_object_representations, constexpr
3712            std::addressof.
3713       Thanks to Daniel Kr�gler, Tim Shen, Edward Smith-Rowland, and Ville
3714       Voutilainen for work on the C++17 support.
3715     * A new power-of-two rehashing policy for use with the _Hashtable
3716       internals, thanks to Fran�ois Dumont.
3717
3718  Fortran
3719
3720     * Support for a number of extensions for compatibility with legacy
3721       code with new flags:
3722          + -fdec-structure Support for DEC STRUCTURE and UNION
3723          + -fdec-intrinsic-ints Support for new integer intrinsics with
3724            B/I/J/K prefixes such as BABS, JIAND...
3725          + -fdec-math Support for additional math intrinsics, including
3726            COTAN and degree-valued trigonometric functions such as TAND,
3727            ASIND...
3728          + -fdec Enable the -fdec-* family of extensions.
3729     * New flag -finit-derived to allow default initialization of
3730       derived-type variables.
3731     * Improved DO loops with step equal to 1 or -1, generates faster code
3732       without a loop preheader. A new warning, -Wundefined-do-loop, warns
3733       when a loop iterates either to HUGE(i) (with step equal to 1), or
3734       to -HUGE(i) (with step equal to -1). Invalid behavior can be caught
3735       at run time with -fcheck=do enabled:
3736
3737program test
3738  implicit none
3739  integer(1) :: i
3740  do i = -HUGE(i)+10, -HUGE(i)-1, -1
3741    print *, i
3742  end do
3743end program test
3744
3745At line 8 of file do_check_12.f90
3746Fortran runtime error: Loop iterates infinitely
3747
3748     * Version 4.5 of the [13]OpenMP specification is now partially
3749       supported in the Fortran compiler; the largest missing item is
3750       structure element mapping.
3751     * User-defined derived-type input/output (UDTIO) is added.
3752     * Derived type coarrays with allocatable and pointer components are
3753       partially supported.
3754     * Non-constant stop codes and error stop codes (Fortran 2015
3755       feature).
3756     * Derived types with allocatable components of recursive type.
3757     * Intrinsic assignment to polymorphic variables.
3758     * Improved submodule support.
3759     * Improved diagnostics (polymorphic results in pure functions).
3760     * Coarray: Support for failed images (Fortan 2015 feature).
3761
3762  Go
3763
3764     * GCC 7 provides a complete implementation of the Go 1.8.1 user
3765       packages.
3766     * Compared to the Go 1.8.1 toolchain, the garbage collector is more
3767       conservative and less concurrent.
3768     * Escape analysis is available for experimental use via the
3769       -fgo-optimize-allocs option. The -fgo-debug-escape prints
3770       information useful for debugging escape analysis choices.
3771
3772  Java (GCJ)
3773
3774   The GCC Java front end and associated libjava runtime library have been
3775   removed from GCC.
3776
3777libgccjit
3778
3779   The libgccjit API gained support for marking calls as requiring
3780   tail-call optimization via a new entry point:
3781   [14]gcc_jit_rvalue_set_bool_require_tail_call.
3782
3783   libgccjit performs numerous checks at the API boundary, but if these
3784   succeed, it previously ignored errors and other diagnostics emitted
3785   within the core of GCC, and treated the compile of a gcc_jit_context as
3786   having succeeded. As of GCC 7 it now ensures that if any diagnostics
3787   are emitted, they are visible from the libgccjit API, and that the the
3788   context is flagged as having failed.
3789
3790New Targets and Target Specific Improvements
3791
3792  AArch64
3793
3794     * GCC has been updated to the latest revision of the procedure call
3795       standard (AAPCS64) to provide support for parameter passing when
3796       data types have been over-aligned.
3797     * The ARMv8.3-A architecture is now supported. It can be used by
3798       specifying the -march=armv8.3-a option.
3799     * The option -msign-return-address= is supported to enable return
3800       address protection using ARMv8.3-A Pointer Authentication
3801       Extensions. For more information on the arguments accepted by this
3802       option, please refer to [15]AArch64-Options.
3803     * The ARMv8.2-A architecture and the ARMv8.2-A 16-bit Floating-Point
3804       Extensions are now supported. They can be used by specifying the
3805       -march=armv8.2-a or -march=armv8.2-a+fp16 options. The 16-bit
3806       Floating-Point Extensions introduce new half-precision data
3807       processing floating-point instructions.
3808     * Support has been added for the following processors (GCC
3809       identifiers in parentheses): ARM Cortex-A73 (cortex-a73), Broadcom
3810       Vulcan (vulcan), Cavium ThunderX CN81xx (thunderxt81), Cavium
3811       ThunderX CN83xx (thunderxt83), Cavium ThunderX CN88xx
3812       (thunderxt88), Cavium ThunderX CN88xx pass 1.x (thunderxt88p1),
3813       Cavium ThunderX 2 CN99xx (thunderx2t99), Qualcomm Falkor (falkor).
3814       The GCC identifiers can be used as arguments to the -mcpu or -mtune
3815       options, for example: -mcpu=cortex-a73 or -mtune=vulcan or as
3816       arguments to the equivalent target attributes and pragmas.
3817
3818  ARC
3819
3820     * Added support for ARC HS and ARC EM processors.
3821     * Added support for ARC EM variation found in Intel QuarkSE SoCs.
3822     * Added support for NPS400 ARC700 based CPUs.
3823     * Thread Local Storage is now supported by ARC CPUs.
3824     * Fixed errors for ARC600 when using 32x16 multiplier option.
3825     * Fixed PIE for ARC CPUs.
3826     * New CPU templates are supported via multilib.
3827
3828  ARM
3829
3830     * Support for the ARMv5 and ARMv5E architectures has been deprecated
3831       (which have no known implementations) and will be removed in a
3832       future GCC release. Note that ARMv5T, ARMv5TE and ARMv5TEJ
3833       architectures remain supported. The values armv5 and armv5e of
3834       -march are thus deprecated.
3835     * The ARMv8.2-A architecture and the ARMv8.2-A 16-bit Floating-Point
3836       Extensions are now supported. They can be used by specifying the
3837       -march=armv8.2-a or -march=armv8.2-a+fp16 options. The 16-bit
3838       Floating-Point Extensions introduce new half-precision data
3839       processing floating-point instructions.
3840     * The ARMv8-M architecture is now supported in its two architecture
3841       profiles: ARMv8-M Baseline and ARMv8-M Mainline with its DSP and
3842       Floating-Point Extensions. They can be used by specifying the
3843       -march=armv8-m.base, armv8-m.main or armv8-m.main+dsp options.
3844     * Support has been added for the following processors (GCC
3845       identifiers in parentheses): ARM Cortex-A73 (cortex-a73), ARM
3846       Cortex-M23 (cortex-m23) and ARM Cortex-M33 (cortex-m33). The GCC
3847       identifiers can be used as arguments to the -mcpu or -mtune
3848       options, for example: -mcpu=cortex-a73 or -mtune=cortex-m33.
3849     * A new command-line option -mpure-code has been added. It does not
3850       allow constant data to be placed in code sections. This option is
3851       only available when generating non-PIC code for ARMv7-M targets.
3852     * Support for the ACLE Coprocessor Intrinsics has been added. This
3853       enables the generation of coprocessor instructions through the use
3854       of intrinsics such as cdp, ldc, and others.
3855     * The configure option --with-multilib-list now accepts the value
3856       rmprofile to build multilib libraries for a range of embedded
3857       targets. See our [16]installation instructions for details.
3858
3859  AVR
3860
3861     * On the reduced Tiny cores, the progmem [17]variable attribute is
3862       now properly supported. Respective read-only variables are located
3863       in flash memory in section .progmem.data. No special code is needed
3864       to access such variables; the compiler automatically adds an offset
3865       of 0x4000 to all addresses, which is needed to access variables in
3866       flash memory. As opposed to ordinary cores where it is sufficient
3867       to specify the progmem attribute with definitions, on the reduced
3868       Tiny cores the attribute also has to be specified with (external)
3869       declarations:
3870
3871extern const int array[] __attribute__((__progmem__));
3872
3873int get_value2 (void)
3874{
3875  /* Access via addresses array + 0x4004 and array + 0x4005. */
3876  return array[2];
3877}
3878
3879const int* get_address (unsigned idx)
3880{
3881  /* Returns array + 0x4000 + 2 * idx. */
3882  return &array[idx];
3883}
3884
3885     * A new command-line option -Wmisspelled-isr has been added. It turns
3886       off -- or turns into errors -- warnings that are reported for
3887       interrupt service routines (ISRs) which don't follow AVR-LibC's
3888       naming convention of prefixing ISR names with __vector.
3889     * __builtin_avr_nops(n) is a new [18]built-in function that inserts n
3890       NOP instructions into the instruction stream. n must be a value
3891       known at compile time.
3892
3893  IA-32/x86-64
3894
3895     * Support for the AVX-512 Fused Multiply Accumulation Packed Single
3896       precision (4FMAPS), AVX-512 Vector Neural Network Instructions Word
3897       variable precision (4VNNIW), AVX-512 Vector Population Count
3898       (VPOPCNTDQ) and Software Guard Extensions (SGX) ISA extensions has
3899       been added.
3900
3901  NVPTX
3902
3903     * OpenMP target regions can now be offloaded to NVidia PTX GPGPUs.
3904       See the [19]Offloading Wiki on how to configure it.
3905
3906  PowerPC / PowerPC64 / RS6000
3907
3908     * The PowerPC port now uses LRA by default.
3909     * GCC now diagnoses inline assembly that clobbers register r2. This
3910       has always been invalid code, and is no longer quietly tolerated.
3911     * The PowerPC port's support for ISA 3.0 (-mcpu=power9) has been
3912       enhanced to generate more of the new instructions by default, and
3913       to provide more built-in functions to generate code for other new
3914       instructions.
3915     * The configuration option --enable-gnu-indirect-function is now
3916       enabled by default on PowerPC GNU/Linux builds.
3917     * The PowerPC port will now allow 64-bit and 32-bit integer types to
3918       be allocated to the VSX vector registers (ISA 2.06 and above). In
3919       addition, on ISA 3.0, 16-bit and 8-bit integer types can be
3920       allocated in the vector registers. Previously, only 64-bit integer
3921       types were allowed in the traditional floating point registers.
3922     * New options -mstack-protector-guard=global,
3923       -mstack-protector-guard=tls, -mstack-protector-guard-reg=, and
3924       -mstack-protector-guard-offset= change how the stack protector gets
3925       the value to use as canary.
3926
3927  S/390, System z, IBM z Systems, IBM Z
3928
3929     * Support for the IBM z14 processor has been added. When using the
3930       -march=z14 option, the compiler will generate code making use of
3931       the new instructions introduced with the vector enhancement
3932       facility and the miscellaneous instruction extension facility 2.
3933       The -mtune=z14 option enables z14 specific instruction scheduling
3934       without making use of new instructions.
3935     * Builtins for the new vector instructions have been added and can be
3936       enabled using the -mzvector option.
3937
3938  RISC-V
3939
3940     * Support for the RISC-V instruction set has been added.
3941
3942  RX
3943
3944   Basic support for atomic built-in function has been added. It is
3945   currently implemented by flipping interrupts off and on as needed.
3946
3947  SH
3948
3949     * Support for SH5/SH64 has been removed.
3950     * Improved utilization of delay slots on SH2A.
3951     * Improved utilization of zero-displacement conditional branches.
3952     * The following deprecated options have been removed
3953          + -mcbranchdi
3954          + -mcmpeqdi
3955          + -minvalid-symbols
3956          + -msoft-atomic
3957          + -mspace
3958          + -madjust-unroll
3959     * Support for the following SH2A instructions has been added
3960          + mov.b @-Rm,R0
3961          + mov.w @-Rm,R0
3962          + mov.l @-Rm,R0
3963          + mov.b R0,@Rn+
3964          + mov.w R0,@Rn+
3965          + mov.l R0,@Rn+
3966
3967  SPARC
3968
3969     * The SPARC port now uses LRA by default.
3970     * Support for the new Subtract-Extended-with-Carry instruction
3971       available in SPARC M7 (Niagara 7) has been added.
3972
3973Operating Systems
3974
3975  AIX
3976
3977     * Visibility support has been enabled for AIX 7.1 and above.
3978
3979  Fuchsia
3980
3981     * Support has been added for the [20]Fuchsia OS.
3982
3983  RTEMS
3984
3985     * The ABI changes on ARM so that no short enums are used by default.
3986
3987Other significant improvements
3988
3989     * -fverbose-asm previously emitted information on the meanings of
3990       assembly expressions. This has been extended so that it now also
3991       prints comments showing the source lines that correspond to the
3992       assembly, making it easier to read the generated assembly
3993       (especially with larger functions). For example, given this C
3994       source file:
3995
3996int test (int n)
3997{
3998  int i;
3999  int total = 0;
4000
4001  for (i = 0; i < n; i++)
4002    total += i * i;
4003  return total;
4004}
4005
4006       -fverbose-asm now gives output similar to this for the function
4007       body (when compiling for x86_64, with -Os):
4008
4009       .text
4010       .globl  test
4011       .type   test, @@function
4012test:
4013.LFB0:
4014       .cfi_startproc
4015# example.c:4:   int total = 0;
4016       xorl    %eax, %eax      # <retval>
4017# example.c:6:   for (i = 0; i < n; i++)
4018       xorl    %edx, %edx      # i
4019.L2:
4020# example.c:6:   for (i = 0; i < n; i++)
4021       cmpl    %edi, %edx      # n, i
4022       jge     .L5     #,
4023# example.c:7:     total += i * i;
4024       movl    %edx, %ecx      # i, tmp92
4025       imull   %edx, %ecx      # i, tmp92
4026# example.c:6:   for (i = 0; i < n; i++)
4027       incl    %edx    # i
4028# example.c:7:     total += i * i;
4029       addl    %ecx, %eax      # tmp92, <retval>
4030       jmp     .L2     #
4031.L5:
4032# example.c:10: }
4033       ret
4034       .cfi_endproc
4035
4036     * Two new options have been added for printing fix-it hints:
4037          + -fdiagnostics-parseable-fixits allows for fix-it hints to be
4038            emitted in a machine-readable form, suitable for consumption
4039            by IDEs. For example, given:
4040
4041spellcheck-fields.cc:52:13: error: 'struct s' has no member named 'colour'; did
4042you mean 'color'?
4043   return ptr->colour;
4044               ^~~~~~
4045               color
4046
4047            it will emit:
4048
4049fix-it:"spellcheck-fields.cc":{52:13-52:19}:"color"
4050
4051          + -fdiagnostics-generate-patch will print a patch in "unified"
4052            format after any diagnostics are printed, showing the result
4053            of applying all fix-it hints. For the above example it would
4054            emit:
4055
4056--- spellcheck-fields.cc
4057+++ spellcheck-fields.cc
4058@@ -49,5 +49,5 @@
4059
4060 color get_color(struct s *ptr)
4061 {
4062-  return ptr->colour;
4063+  return ptr->color;
4064 }
4065
4066     * The gcc and g++ driver programs will now provide suggestions for
4067       misspelled arguments to command-line options.
4068
4069$ gcc -c test.c -ftls-model=global-dinamic
4070gcc: error: unknown TLS model 'global-dinamic'
4071gcc: note: valid arguments to '-ftls-model=' are: global-dynamic initial-exec lo
4072cal-dynamic local-exec; did you mean 'global-dynamic'?
4073
4074     * The compiler will now provide suggestions for misspelled
4075       parameters.
4076
4077$ gcc -c test.c --param max-early-inliner-iteration=3
4078cc1: error: invalid --param name 'max-early-inliner-iteration'; did you mean 'ma
4079x-early-inliner-iterations'?
4080
4081     * Profile-guided optimization (PGO) instrumentation, as well as test
4082       coverage (GCOV), can newly instrument constructors (functions marks
4083       with __attribute__((constructor))), destructors and C++
4084       constructors (and destructors) of classes that are used as the type
4085       of a global variable.
4086     * A new option -fprofile-update=atomic prevents creation of corrupted
4087       profiles created during an instrumentation run (-fprofile=generate)
4088       of an application. The downside of the option is a speed penalty.
4089       Providing -pthread on the command line selects atomic profile
4090       updating (when supported by the target).
4091     * GCC's already extensive testsuite has gained some new capabilities,
4092       to further improve the reliability of the compiler:
4093          + GCC now has an internal unit-testing API and a suite of tests
4094            for programmatic self-testing of subsystems.
4095          + GCC's C front end has been extended so that it can parse dumps
4096            of GCC's internal representations, allowing for DejaGnu tests
4097            that more directly exercise specific optimization passes. This
4098            covers both the [21]GIMPLE representation (for testing
4099            higher-level optimizations) and the [22]RTL representation,
4100            allowing for more direct testing of lower-level details, such
4101            as register allocation and instruction selection.
4102
4103GCC 7.1
4104
4105   This is the [23]list of problem reports (PRs) from GCC's bug tracking
4106   system that are known to be fixed in the 7.1 release. This list might
4107   not be complete (that is, it is possible that some PRs that have been
4108   fixed are not listed here).
4109
4110GCC 7.2
4111
4112   This is the [24]list of problem reports (PRs) from GCC's bug tracking
4113   system that are known to be fixed in the 7.2 release. This list might
4114   not be complete (that is, it is possible that some PRs that have been
4115   fixed are not listed here).
4116
4117  Target Specific Changes
4118
4119    SPARC
4120
4121     * Support for the SPARC M8 processor has been added.
4122     * The switches -mfix-ut700 and -mfix-gr712rc have been added to work
4123       around an erratum in LEON3FT processors.
4124     * Use of the Floating-point Multiply Single to Double (FsMULd)
4125       instruction can now be controlled by the -mfsmuld and -fno-fsmuld
4126       options.
4127
4128  Operating Systems
4129
4130    RTEMS
4131
4132     * The Ada run-time support uses now thread-local storage (TLS).
4133     * Support for RISC-V has been added.
4134     * Support for 64-bit PowerPC using the ELFv2 ABI with 64-bit long
4135       double has been added.
4136
4137GCC 7.3
4138
4139   This is the [25]list of problem reports (PRs) from GCC's bug tracking
4140   system that are known to be fixed in the 7.3 release. This list might
4141   not be complete (that is, it is possible that some PRs that have been
4142   fixed are not listed here).
4143
4144  Target Specific Changes
4145
4146    SPARC
4147
4148     * Workarounds for the four [26]LEON3FT errata GRLIB-TN-0010..0013
4149       have been added. Relevant errata are activated by the target
4150       specific -mfix-ut699, -mfix-ut700 and -mfix-gr712rc switches.
4151
4152  Operating Systems
4153
4154    RTEMS
4155
4156     * Support has been added for Epiphany target.
4157
4158GCC 7.4
4159
4160   This is the [27]list of problem reports (PRs) from GCC's bug tracking
4161   system that are known to be fixed in the 7.4 release. This list might
4162   not be complete (that is, it is possible that some PRs that have been
4163   fixed are not listed here).
4164
4165GCC 7.5
4166
4167   This is the [28]list of problem reports (PRs) from GCC's bug tracking
4168   system that are known to be fixed in the 7.5 release. This list might
4169   not be complete (that is, it is possible that some PRs that have been
4170   fixed are not listed here).
4171
4172
4173    For questions related to the use of GCC, please consult these web
4174    pages and the [29]GCC manuals. If that fails, the
4175    [30]gcc-help@gcc.gnu.org mailing list might help. Comments on these
4176    web pages and the development of GCC are welcome on our developer
4177    list at [31]gcc@gcc.gnu.org. All of [32]our lists have public
4178    archives.
4179
4180   Copyright (C) [33]Free Software Foundation, Inc. Verbatim copying and
4181   distribution of this entire article is permitted in any medium,
4182   provided this notice is preserved.
4183
4184   These pages are [34]maintained by the GCC team. Last modified
4185   2020-04-07[35].
4186
4187References
4188
4189   1. http://gcc.gnu.org/gcc-7/porting_to.html
4190   2. http://gcc.gnu.org/onlinedocs/index.html#current
4191   3. https://gcc.gnu.org/wiki/LRAIsDefault
4192   4. https://gcc.gnu.org/onlinedocs/gcc-7.1.0/libstdc++/manual/manual/profile_mode.html
4193   5. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77728
4194   6. http://www.dwarfstd.org/Download.php
4195   7. https://gcc.gnu.org/wiki/OpenACC
4196   8. https://gcc.gnu.org/wiki/Offloading
4197   9. https://gcc.gnu.org/onlinedocs/gcc-7.1.0/gcc/Object-Size-Checking.html
4198  10. https://gcc.gnu.org/projects/cxx-status.html#cxx1z
4199  11. https://wg21.link/p0136
4200  12. https://gcc.gnu.org/onlinedocs/gcc-7.1.0/libstdc++/manual/using_dual_abi.html
4201  13. https://www.openmp.org/specifications/
4202  14. https://gcc.gnu.org/onlinedocs/gcc-7.1.0/jit/topics/expressions.html#gcc_jit_rvalue_set_bool_require_tail_call
4203  15. https://gcc.gnu.org/onlinedocs/gcc-7.1.0/gcc/AArch64-Options.html#AArch64-Options
4204  16. https://gcc.gnu.org/install/configure.html
4205  17. https://gcc.gnu.org/onlinedocs/gcc-7.1.0/gcc/AVR-Variable-Attributes.html
4206  18. https://gcc.gnu.org/onlinedocs/gcc-7.1.0/gcc/AVR-Built-in-Functions.html
4207  19. https://gcc.gnu.org/wiki/Offloading
4208  20. https://fuchsia.googlesource.com/
4209  21. https://gcc.gnu.org/onlinedocs/gcc-7.1.0/gccint/GIMPLE-Tests.html
4210  22. https://gcc.gnu.org/onlinedocs/gcc-7.1.0/gccint/RTL-Tests.html
4211  23. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=7.0
4212  24. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=7.2
4213  25. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=7.3
4214  26. https://www.gaisler.com/index.php/information/app-tech-notes
4215  27. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=7.4
4216  28. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=7.5
4217  29. https://gcc.gnu.org/onlinedocs/
4218  30. mailto:gcc-help@gcc.gnu.org
4219  31. mailto:gcc@gcc.gnu.org
4220  32. https://gcc.gnu.org/lists.html
4221  33. https://www.fsf.org/
4222  34. https://gcc.gnu.org/about.html
4223  35. http://validator.w3.org/check/referer
4224======================================================================
4225http://gcc.gnu.org/gcc-6/index.html
4226
4227                              GCC 6 Release Series
4228
4229   (This release series is no longer supported.)
4230
4231   October 26, 2018
4232
4233   The [1]GNU project and the GCC developers are pleased to announce the
4234   release of GCC 6.5.
4235
4236   This release is a bug-fix release, containing fixes for regressions in
4237   GCC 6.4 relative to previous releases of GCC.
4238
4239Release History
4240
4241   GCC 6.5
4242          October 26, 2018 ([2]changes, [3]documentation)
4243
4244   GCC 6.4
4245          July 4, 2017 ([4]changes, [5]documentation)
4246
4247   GCC 6.3
4248          December 21, 2016 ([6]changes, [7]documentation)
4249
4250   GCC 6.2
4251          August 22, 2016 ([8]changes, [9]documentation)
4252
4253   GCC 6.1
4254          April 27, 2016 ([10]changes, [11]documentation)
4255
4256References and Acknowledgements
4257
4258   GCC used to stand for the GNU C Compiler, but since the compiler
4259   supports several other languages aside from C, it now stands for the
4260   GNU Compiler Collection.
4261
4262   A list of [12]successful builds is updated as new information becomes
4263   available.
4264
4265   The GCC developers would like to thank the numerous people that have
4266   contributed new features, improvements, bug fixes, and other changes as
4267   well as test results to GCC. This [13]amazing group of volunteers is
4268   what makes GCC successful.
4269
4270   For additional information about GCC please refer to the [14]GCC
4271   project web site or contact the [15]GCC development mailing list.
4272
4273   To obtain GCC please use [16]our mirror sites or [17]our version
4274   control system.
4275
4276
4277    For questions related to the use of GCC, please consult these web
4278    pages and the [18]GCC manuals. If that fails, the
4279    [19]gcc-help@gcc.gnu.org mailing list might help. Comments on these
4280    web pages and the development of GCC are welcome on our developer
4281    list at [20]gcc@gcc.gnu.org. All of [21]our lists have public
4282    archives.
4283
4284   Copyright (C) [22]Free Software Foundation, Inc. Verbatim copying and
4285   distribution of this entire article is permitted in any medium,
4286   provided this notice is preserved.
4287
4288   These pages are [23]maintained by the GCC team. Last modified
4289   2020-01-14[24].
4290
4291References
4292
4293   1. http://www.gnu.org/
4294   2. http://gcc.gnu.org/gcc-6/changes.html
4295   3. http://gcc.gnu.org/onlinedocs/6.5.0/
4296   4. http://gcc.gnu.org/gcc-6/changes.html
4297   5. http://gcc.gnu.org/onlinedocs/6.4.0/
4298   6. http://gcc.gnu.org/gcc-6/changes.html
4299   7. http://gcc.gnu.org/onlinedocs/6.3.0/
4300   8. http://gcc.gnu.org/gcc-6/changes.html
4301   9. http://gcc.gnu.org/onlinedocs/6.2.0/
4302  10. http://gcc.gnu.org/gcc-6/changes.html
4303  11. http://gcc.gnu.org/onlinedocs/6.1.0/
4304  12. http://gcc.gnu.org/gcc-6/buildstat.html
4305  13. http://gcc.gnu.org/onlinedocs/gcc/Contributors.html
4306  14. http://gcc.gnu.org/index.html
4307  15. mailto:gcc@gcc.gnu.org
4308  16. http://gcc.gnu.org/mirrors.html
4309  17. http://gcc.gnu.org/git.html
4310  18. https://gcc.gnu.org/onlinedocs/
4311  19. mailto:gcc-help@gcc.gnu.org
4312  20. mailto:gcc@gcc.gnu.org
4313  21. https://gcc.gnu.org/lists.html
4314  22. https://www.fsf.org/
4315  23. https://gcc.gnu.org/about.html
4316  24. http://validator.w3.org/check/referer
4317======================================================================
4318http://gcc.gnu.org/gcc-6/changes.html
4319
4320                              GCC 6 Release Series
4321                        Changes, New Features, and Fixes
4322
4323   This page is a brief summary of some of the huge number of improvements
4324   in GCC 6. For more information, see the [1]Porting to GCC 6 page and
4325   the [2]full GCC documentation.
4326
4327Caveats
4328
4329     * The default mode for C++ is now -std=gnu++14 instead of
4330       -std=gnu++98.
4331     * Support for a number of older systems and recently unmaintained or
4332       untested target ports of GCC has been declared obsolete in GCC 6.
4333       Unless there is activity to revive them, the next release of GCC
4334       will have their sources permanently removed.
4335       The following ports for individual systems on particular
4336       architectures have been obsoleted:
4337          + SH5 / SH64 (sh64-*-*) as announced [3]here.
4338     * The AVR port requires binutils version 2.26.1 or later for the fix
4339       for [4]PR71151 to work.
4340     * The GCC 6.5 release has an accidental ABI incompatibility for
4341       nested std::pair objects, for more details see [5]PR 87822. The bug
4342       causes a layout change for pairs where the first member is also a
4343       pair, e.g. std::pair<std::pair<X, Y>, Z>. The GCC 6 release series
4344       is closed so the bug in GCC 6.5 will not be fixed upstream, but
4345       there is a patch in the bug report to allow it to be fixed by
4346       anybody packaging GCC 6.5 or installing it themselves.
4347
4348General Optimizer Improvements
4349
4350     * UndefinedBehaviorSanitizer gained a new sanitization option,
4351       -fsanitize=bounds-strict, which enables strict checking of array
4352       bounds. In particular, it enables -fsanitize=bounds as well as
4353       instrumentation of flexible array member-like arrays.
4354     * Type-based alias analysis now disambiguates accesses to different
4355       pointers. This improves precision of the alias oracle by about
4356       20-30% on higher-level C++ programs. Programs doing invalid type
4357       punning of pointer types may now need -fno-strict-aliasing to work
4358       correctly.
4359     * Alias analysis now correctly supports the weakref and alias
4360       attributes. This allows accessing both a variable and its alias in
4361       one translation unit which is common with link-time optimization.
4362     * Value range propagation now assumes that the this pointer in C++
4363       member functions is non-null. This eliminates common null pointer
4364       checks but also breaks some non-conforming code-bases (such as
4365       Qt-5, Chromium, KDevelop). As a temporary work-around
4366       -fno-delete-null-pointer-checks can be used. Wrong code can be
4367       identified by using -fsanitize=undefined.
4368     * Link-time optimization improvements:
4369          + warning and error attributes are now correctly preserved by
4370            declaration linking and thus -D_FORTIFY_SOURCE=2 is now
4371            supported with -flto.
4372          + Type merging was fixed to handle C and Fortran
4373            interoperability rules as defined by the Fortran 2008 language
4374            standard.
4375            As an exception, CHARACTER(KIND=C_CHAR) is not inter-operable
4376            with char in all cases because it is an array while char is
4377            scalar. INTEGER(KIND=C_SIGNED_CHAR) should be used instead. In
4378            general, this inter-operability cannot be implemented, for
4379            example on targets where the argument passing convention for
4380            arrays differs from scalars.
4381          + More type information is now preserved at link time, reducing
4382            the loss of accuracy of the type-based alias analysis compared
4383            to builds without link-time optimization.
4384          + Invalid type punning on global variables and declarations is
4385            now reported with -Wodr-type-mismatch.
4386          + The size of LTO object files was reduced by about 11%
4387            (measured by compiling Firefox 46.0).
4388          + Link-time parallelization (enabled using -flto=n) was
4389            significantly improved by decreasing the size of streamed data
4390            when partitioning programs. The size of streamed IL while
4391            compiling Firefox 46.0 was reduced by 66%.
4392          + The linker plugin was extended to pass information about the
4393            type of binary produced to the GCC back end. (That can also be
4394            controlled manually by -flinker-output.) This makes it
4395            possible to properly configure the code generator and support
4396            incremental linking. Incremental linking of LTO objects by gcc
4397            -r is now supported for plugin-enabled setups.
4398            There are two ways to perform incremental linking:
4399              1. Linking by ld -r will result in an object file with all
4400                 sections from individual object files mechanically
4401                 merged. This delays the actual link-time optimization to
4402                 the final linking step and thus permits whole program
4403                 optimization. Linking the final binary with such object
4404                 files is however slower.
4405              2. Linking by gcc -r will lead to link-time optimization and
4406                 emit the final binary into the object file. Linking such
4407                 an object file is fast but avoids any benefits from whole
4408                 program optimization.
4409            GCC 7 will support incremental link-time optimization with gcc
4410            -r.
4411     * Inter-procedural optimization improvements:
4412          + Basic jump threading is now performed before profile
4413            construction and inline analysis, resulting in more realistic
4414            size and time estimates that drive the heuristics of the
4415            inliner and function cloning passes.
4416          + Function cloning now more aggressively eliminates unused
4417            function parameters.
4418
4419New Languages and Language specific improvements
4420
4421   Compared to GCC 5, the GCC 6 release series includes a much improved
4422   implementation of the [6]OpenACC 2.0a specification. Highlights are:
4423     * In addition to single-threaded host-fallback execution, offloading
4424       is supported for nvptx (Nvidia GPUs) on x86_64 and PowerPC 64-bit
4425       little-endian GNU/Linux host systems. For nvptx offloading, with
4426       the OpenACC parallel construct, the execution model allows for an
4427       arbitrary number of gangs, up to 32 workers, and 32 vectors.
4428     * Initial support for parallelized execution of OpenACC kernels
4429       constructs:
4430          + Parallelization of a kernels region is switched on by
4431            -fopenacc combined with -O2 or higher.
4432          + Code is offloaded onto multiple gangs, but executes with just
4433            one worker, and a vector length of 1.
4434          + Directives inside a kernels region are not supported.
4435          + Loops with reductions can be parallelized.
4436          + Only kernels regions with one loop nest are parallelized.
4437          + Only the outer-most loop of a loop nest can be parallelized.
4438          + Loop nests containing sibling loops are not parallelized.
4439       Typically, using the OpenACC parallel construct gives much better
4440       performance, compared to the initial support of the OpenACC kernels
4441       construct.
4442     * The device_type clause is not supported. The bind and nohost
4443       clauses are not supported. The host_data directive is not supported
4444       in Fortran.
4445     * Nested parallelism (cf. CUDA dynamic parallelism) is not supported.
4446     * Usage of OpenACC constructs inside multithreaded contexts (such as
4447       created by OpenMP, or pthread programming) is not supported.
4448     * If a call to the acc_on_device function has a compile-time constant
4449       argument, the function call evaluates to a compile-time constant
4450       value only for C and C++ but not for Fortran.
4451
4452   See the [7]OpenACC and [8]Offloading wiki pages for further
4453   information.
4454
4455  C family
4456
4457     * Version 4.5 of the [9]OpenMP specification is now supported in the
4458       C and C++ compilers.
4459     * The C and C++ compilers now support attributes on enumerators. For
4460       instance, it is now possible to mark enumerators as deprecated:
4461
4462enum {
4463  newval,
4464  oldval __attribute__ ((deprecated ("too old")))
4465};
4466
4467     * Source locations for the C and C++ compilers are now tracked as
4468       ranges, rather than just points, making it easier to identify the
4469       subexpression of interest within a complicated expression. For
4470       example:
4471
4472test.cc: In function 'int test(int, int, foo, int, int)':
4473test.cc:5:16: error: no match for 'operator*' (operand types are 'int' and 'foo'
4474)
4475   return p + q * r * s + t;
4476              ~~^~~
4477
4478       In addition, there is now initial support for precise diagnostic
4479       locations within strings:
4480
4481format-strings.c:3:14: warning: field width specifier '*' expects a matching 'in
4482t' argument [-Wformat=]
4483   printf("%*d");
4484            ^
4485
4486     * Diagnostics can now contain "fix-it hints", which are displayed in
4487       context underneath the relevant source code. For example:
4488
4489fixits.c: In function 'bad_deref':
4490fixits.c:11:13: error: 'ptr' is a pointer; did you mean to use '->'?
4491   return ptr.x;
4492             ^
4493             ->
4494
4495     * The C and C++ compilers now offer suggestions for misspelled field
4496       names:
4497
4498spellcheck-fields.cc:52:13: error: 'struct s' has no member named 'colour'; did
4499you mean 'color'?
4500   return ptr->colour;
4501               ^~~~~~
4502
4503     * New command-line options have been added for the C and C++
4504       compilers:
4505          + -Wshift-negative-value warns about left shifting a negative
4506            value.
4507          + -Wshift-overflow warns about left shift overflows. This
4508            warning is enabled by default. -Wshift-overflow=2 also warns
4509            about left-shifting 1 into the sign bit.
4510          + -Wtautological-compare warns if a self-comparison always
4511            evaluates to true or false. This warning is enabled by -Wall.
4512          + -Wnull-dereference warns if the compiler detects paths that
4513            trigger erroneous or undefined behavior due to dereferencing a
4514            null pointer. This option is only active when
4515            -fdelete-null-pointer-checks is active, which is enabled by
4516            optimizations in most targets. The precision of the warnings
4517            depends on the optimization options used.
4518          + -Wduplicated-cond warns about duplicated conditions in an
4519            if-else-if chain.
4520          + -Wmisleading-indentation warns about places where the
4521            indentation of the code gives a misleading idea of the block
4522            structure of the code to a human reader. For example, given
4523            [10]CVE-2014-1266:
4524
4525sslKeyExchange.c: In function 'SSLVerifySignedServerKeyExchange':
4526sslKeyExchange.c:629:3: warning: this 'if' clause does not guard... [-Wmisleadin
4527g-indentation]
4528    if ((err = SSLHashSHA1.update(&hashCtx, &signedParams)) != 0)
4529    ^~
4530sslKeyExchange.c:631:5: note: ...this statement, but the latter is misleadingly
4531indented as if it is guarded by the 'if'
4532        goto fail;
4533        ^~~~
4534
4535            This warning is enabled by -Wall.
4536     * The C and C++ compilers now emit saner error messages if
4537       merge-conflict markers are present in a source file.
4538
4539test.c:3:1: error: version control conflict marker in file
4540 <<<<<<< HEAD
4541 ^~~~~~~
4542
4543  C
4544
4545     * It is possible to disable warnings when an initialized field of a
4546       structure or a union with side effects is being overridden when
4547       using designated initializers via a new warning option
4548       -Woverride-init-side-effects.
4549     * A new type attribute scalar_storage_order applying to structures
4550       and unions has been introduced. It specifies the storage order (aka
4551       endianness) in memory of scalar fields in structures or unions.
4552
4553  C++
4554
4555     * The default mode has been changed to -std=gnu++14.
4556     * [11]C++ Concepts are now supported when compiling with -fconcepts.
4557     * -flifetime-dse is more aggressive in dead-store elimination in
4558       situations where a memory store to a location precedes a
4559       constructor to that memory location.
4560     * G++ now supports [12]C++17 fold expressions, u8 character literals,
4561       extended static_assert, and nested namespace definitions.
4562     * G++ now allows constant evaluation for all non-type template
4563       arguments.
4564     * G++ now supports C++ Transactional Memory when compiling with
4565       -fgnu-tm.
4566
4567    Runtime Library (libstdc++)
4568
4569     * Extensions to the C++ Library to support mathematical special
4570       functions (ISO/IEC 29124:2010), thanks to Edward Smith-Rowland.
4571     * Experimental support for C++17, including the following new
4572       features:
4573          + std::uncaught_exceptions function (this is also available for
4574            -std=gnu++NN modes);
4575          + new member functions try_emplace and insert_or_assign for
4576            unique_key maps;
4577          + non-member functions std::size, std::empty, and std::data for
4578            accessing containers and arrays;
4579          + std::invoke;
4580          + std::shared_mutex;
4581          + std::void_t and std::bool_constant metaprogramming utilities.
4582       Thanks to Ville Voutilainen for contributing many of the C++17
4583       features.
4584     * An experimental implementation of the File System TS.
4585     * Experimental support for most features of the second version of the
4586       Library Fundamentals TS. This includes polymorphic memory resources
4587       and array support in shared_ptr, thanks to Fan You.
4588     * Some assertions checked by Debug Mode can now also be enabled by
4589       _GLIBCXX_ASSERTIONS. The subset of checks enabled by the new macro
4590       have less run-time overhead than the full _GLIBCXX_DEBUG checks and
4591       don't affect the library ABI, so can be enabled per-translation
4592       unit.
4593     * Timed mutex types are supported on more targets, including Darwin.
4594     * Improved std::locale support for DragonFly and FreeBSD, thanks to
4595       John Marino and Andreas Tobler.
4596
4597  Fortran
4598
4599     * Fortran 2008 SUBMODULE support.
4600     * Fortran 2015 EVENT_TYPE, EVENT_POST, EVENT_WAIT, and EVENT_QUERY
4601       support.
4602     * Improved support for Fortran 2003 deferred-length character
4603       variables.
4604     * Improved support for OpenMP and OpenACC.
4605     * The MATMUL intrinsic is now inlined for straightforward cases if
4606       front-end optimization is active. The maximum size for inlining can
4607       be set to n with the -finline-matmul-limit=n option and turned off
4608       with -finline-matmul-limit=0.
4609     * The -Wconversion-extra option will warn about REAL constants which
4610       have excess precision for their kind.
4611     * The -Winteger-division option has been added, which warns about
4612       divisions of integer constants which are truncated. This option is
4613       included in -Wall by default.
4614
4615libgccjit
4616
4617     * The driver code is now run in-process within libgccjit, providing a
4618       small speed-up of the compilation process.
4619     * The API has gained entrypoints for
4620          + [13]timing how long was spent in different parts of code,
4621          + [14]creating switch statements,
4622          + [15]allowing unreachable basic blocks in a function, and
4623          + [16]adding arbitrary command-line options to a compilation.
4624
4625New Targets and Target Specific Improvements
4626
4627  AArch64
4628
4629     * A number of AArch64-specific options have been added. The most
4630       important ones are summarised in this section; for more detailed
4631       information please refer to the documentation.
4632     * The command-line options -march=native, -mcpu=native and
4633       -mtune=native are now available on native AArch64 GNU/Linux
4634       systems. Specifying these options causes GCC to auto-detect the
4635       host CPU and choose the optimal setting for that system.
4636     * -fpic is now supported when generating code for the small code
4637       model (-mcmodel=small). The size of the global offset table (GOT)
4638       is limited to 28KiB under the LP64 SysV ABI, and 15KiB under the
4639       ILP32 SysV ABI.
4640     * The AArch64 port now supports target attributes and pragmas. Please
4641       refer to the [17]documentation for details of available attributes
4642       and pragmas as well as usage instructions.
4643     * Link-time optimization across translation units with different
4644       target-specific options is now supported.
4645     * The option -mtls-size= is now supported. It can be used to specify
4646       the bit size of TLS offsets, allowing GCC to generate better TLS
4647       instruction sequences.
4648     * The option -fno-plt is now fully functional.
4649     * The ARMv8.1-A architecture and the Large System Extensions are now
4650       supported. They can be used by specifying the -march=armv8.1-a
4651       option. Additionally, the +lse option extension can be used in a
4652       similar fashion to other option extensions. The Large System
4653       Extensions introduce new instructions that are used in the
4654       implementation of atomic operations.
4655     * The ACLE half-precision floating-point type __fp16 is now supported
4656       in the C and C++ languages.
4657     * The ARM Cortex-A35 processor is now supported via the
4658       -mcpu=cortex-a35 and -mtune=cortex-a35 options as well as the
4659       equivalent target attributes and pragmas.
4660     * The Qualcomm QDF24xx processor is now supported via the
4661       -mcpu=qdf24xx and -mtune=qdf24xx options as well as the equivalent
4662       target attributes and pragmas.
4663     * Code generation for the ARM Cortex-A57 processor is improved. Among
4664       general code generation improvements, a better algorithm is added
4665       for allocating registers to floating-point multiply-accumulate
4666       instructions offering increased performance when compiling with
4667       -mcpu=cortex-a57 or -mtune=cortex-a57.
4668     * Code generation for the ARM Cortex-A53 processor is improved. A
4669       more accurate instruction scheduling model for the processor is now
4670       used, and a number of compiler tuning parameters have been set to
4671       offer increased performance when compiling with -mcpu=cortex-a53 or
4672       -mtune=cortex-a53.
4673     * Code generation for the Samsung Exynos M1 processor is improved. A
4674       more accurate instruction scheduling model for the processor is now
4675       used, and a number of compiler tuning parameters have been set to
4676       offer increased performance when compiling with -mcpu=exynos-m1 or
4677       -mtune=exynos-m1.
4678     * Improvements in the generation of conditional branches and literal
4679       pools allow the compiler to compile functions of a large size.
4680       Constant pools are now placed into separate rodata sections. The
4681       new option -mpc-relative-literal-loads generates per-function
4682       literal pools, limiting the maximum size of functions to 1MiB.
4683     * Several correctness issues generating Advanced SIMD instructions
4684       for big-endian targets have been fixed resulting in improved code
4685       generation for ACLE intrinsics with -mbig-endian.
4686
4687  ARM
4688
4689     * Support for revisions of the ARM architecture prior to ARMv4t has
4690       been deprecated and will be removed in a future GCC release. The
4691       -mcpu and -mtune values that are deprecated are: arm2, arm250,
4692       arm3, arm6, arm60, arm600, arm610, arm620, arm7, arm7d, arm7di,
4693       arm70, arm700, arm700i, arm710, arm720, arm710c, arm7100, arm7500,
4694       arm7500fe, arm7m, arm7dm, arm7dmi, arm8, arm810, strongarm,
4695       strongarm110, strongarm1100, strongarm1110, fa526, fa626. The value
4696       arm7tdmi is still supported. The values of -march that are
4697       deprecated are: armv2,armv2a,armv3,armv3m,armv4.
4698     * The ARM port now supports target attributes and pragmas. Please
4699       refer to the [18]documentation for details of available attributes
4700       and pragmas as well as usage instructions.
4701     * Support has been added for the following processors (GCC
4702       identifiers in parentheses): ARM Cortex-A32 (cortex-a32), ARM
4703       Cortex-A35 (cortex-a35) and ARM Cortex-R8 (cortex-r8). The GCC
4704       identifiers can be used as arguments to the -mcpu or -mtune
4705       options, for example: -mcpu=cortex-a32 or -mtune=cortex-a35.
4706
4707  Heterogeneous Systems Architecture
4708
4709     * GCC can now generate HSAIL (Heterogeneous System Architecture
4710       Intermediate Language) for simple OpenMP device constructs if
4711       configured with --enable-offload-targets=hsa. A new libgomp plugin
4712       then runs the HSA GPU kernels implementing these constructs on HSA
4713       capable GPUs via a standard HSA run time.
4714       If the HSA compilation back end determines it cannot output HSAIL
4715       for a particular input, it gives a warning by default. These
4716       warnings can be suppressed with -Wno-hsa. To give a few examples,
4717       the HSA back end does not implement compilation of code using
4718       function pointers, automatic allocation of variable sized arrays,
4719       functions with variadic arguments as well as a number of other less
4720       common programming constructs.
4721       When compilation for HSA is enabled, the compiler attempts to
4722       compile composite OpenMP constructs
4723
4724#pragma omp target teams distribute parallel for
4725
4726       into parallel HSA GPU kernels.
4727
4728  IA-32/x86-64
4729
4730     * GCC now supports the Intel CPU named Skylake with AVX-512
4731       extensions through -march=skylake-avx512. The switch enables the
4732       following ISA extensions: AVX-512F, AVX512VL, AVX-512CD, AVX-512BW,
4733       AVX-512DQ.
4734     * Support for new AMD instructions monitorx and mwaitx has been
4735       added. This includes new intrinsic and built-in support. It is
4736       enabled through option -mmwaitx. The instructions monitorx and
4737       mwaitx implement the same functionality as the old monitor and
4738       mwait instructions. In addition mwaitx adds a configurable timer.
4739       The timer value is received as third argument and stored in
4740       register %ebx.
4741     * x86-64 targets now allow stack realignment from a word-aligned
4742       stack pointer using the command-line option -mstackrealign or
4743       __attribute__ ((force_align_arg_pointer)). This allows functions
4744       compiled with a vector-aligned stack to be invoked from objects
4745       that keep only word-alignment.
4746     * Support for address spaces __seg_fs, __seg_gs, and __seg_tls. These
4747       can be used to access data via the %fs and %gs segments without
4748       having to resort to inline assembly. Please refer to the
4749       [19]documentation for usage instructions.
4750     * Support for AMD Zen (family 17h) processors is now available
4751       through the -march=znver1 and -mtune=znver1 options.
4752
4753  MeP
4754
4755     * Support for the MeP (mep-elf) architecture has been deprecated and
4756       will be removed in a future GCC release.
4757
4758  MSP430
4759
4760     * The MSP430 compiler now has the ability to automatically distribute
4761       code and data between low memory (addresses below 64K) and high
4762       memory. This only applies to parts that actually have both memory
4763       regions and only if the linker script for the part has been
4764       specifically set up to support this feature.
4765       A new attribute of either can be applied to both functions and
4766       data, and this tells the compiler to place the object into low
4767       memory if there is room and into high memory otherwise. Two other
4768       new attributes - lower and upper - can be used to explicitly state
4769       that an object should be placed in the specified memory region. If
4770       there is not enough left in that region the compilation will fail.
4771       Two new command-line options - -mcode-region=[lower|upper|either]
4772       and -mdata-region=[lower|upper|either] - can be used to tell the
4773       compiler what to do with objects that do not have one of these new
4774       attributes.
4775
4776  PowerPC / PowerPC64 / RS6000
4777
4778     * PowerPC64 now supports IEEE 128-bit floating-point using the
4779       __float128 data type. In GCC 6, this is not enabled by default, but
4780       you can enable it with -mfloat128. The IEEE 128-bit floating-point
4781       support requires the use of the VSX instruction set. IEEE 128-bit
4782       floating-point values are passed and returned as a single vector
4783       value. The software emulator for IEEE 128-bit floating-point
4784       support is only built on PowerPC GNU/Linux systems where the
4785       default CPU is at least power7. On future ISA 3.0 systems (POWER 9
4786       and later), you will be able to use the -mfloat128-hardware option
4787       to use the ISA 3.0 instructions that support IEEE 128-bit
4788       floating-point. An additional type (__ibm128) has been added to
4789       refer to the IBM extended double type that normally implements long
4790       double. This will allow for a future transition to implementing
4791       long double with IEEE 128-bit floating-point.
4792     * Basic support has been added for POWER9 hardware that will use the
4793       recently published OpenPOWER ISA 3.0 instructions. The following
4794       new switches are available:
4795          + -mcpu=power9: Implement all of the ISA 3.0 instructions
4796            supported by the compiler.
4797          + -mtune=power9: In the future, apply tuning for POWER9 systems.
4798            Currently, POWER8 tunings are used.
4799          + -mmodulo: Generate code using the ISA 3.0 integer instructions
4800            (modulus, count trailing zeros, array index support, integer
4801            multiply/add).
4802          + -mpower9-fusion: Generate code to suitably fuse instruction
4803            sequences for a POWER9 system.
4804          + -mpower9-dform: Generate code to use the new D-form
4805            (register+offset) memory instructions for the vector
4806            registers.
4807          + -mpower9-vector: Generate code using the new ISA 3.0 vector
4808            (VSX or Altivec) instructions.
4809          + -mpower9-minmax: Reserved for future development.
4810          + -mtoc-fusion: Keep TOC entries together to provide more fusion
4811            opportunities.
4812     * New constraints have been added to support IEEE 128-bit
4813       floating-point and ISA 3.0 instructions:
4814          + wb: Altivec register if -mpower9-dform is enabled.
4815          + we: VSX register if -mpower9-vector is enabled for 64-bit code
4816            generation.
4817          + wo: VSX register if -mpower9-vector is enabled.
4818          + wp: Reserved for future use if long double is implemented with
4819            IEEE 128-bit floating-point instead of IBM extended double.
4820          + wq: VSX register if -mfloat128 is enabled.
4821          + wF: Memory operand suitable for POWER9 fusion load/store.
4822          + wG: Memory operand suitable for TOC fusion memory references.
4823          + wL: Integer constant identifying the element number mfvsrld
4824            accesses within a vector.
4825     * Support has been added for __builtin_cpu_is() and
4826       __builtin_cpu_supports(), allowing for very fast access to
4827       AT_PLATFORM, AT_HWCAP, and AT_HWCAP2 values. This requires use of
4828       glibc 2.23 or later.
4829     * All hardware transactional memory builtins now correctly behave as
4830       memory barriers. Programmers can use #ifdef __TM_FENCE__ to
4831       determine whether their "old" compiler treats the builtins as
4832       barriers.
4833     * Split-stack support has been added for gccgo on PowerPC64 for both
4834       big- and little-endian (but not for 32-bit). The gold linker from
4835       at least binutils 2.25.1 must be available in the PATH when
4836       configuring and building gccgo to enable split stack. (The
4837       requirement for binutils 2.25.1 applies to PowerPC64 only.) The
4838       split-stack feature allows a small initial stack size to be
4839       allocated for each goroutine, which increases as needed.
4840     * GCC on PowerPC now supports the standard lround function.
4841     * A new configuration option ---with-advance-toolchain=at was added
4842       for PowerPC 64-bit GNU/Linux systems to use the header files,
4843       library files, and the dynamic linker from a specific Advance
4844       Toolchain release instead of the default versions that are provided
4845       by the GNU/Linux distribution. In general, this option is intended
4846       for the developers of GCC, and it is not intended for general use.
4847     * The "q", "S", "T", and "t" asm-constraints have been removed.
4848     * The "b", "B", "m", "M", and "W" format modifiers have been removed.
4849
4850  S/390, System z, IBM z Systems
4851
4852     * Support for the IBM z13 processor has been added. When using the
4853       -march=z13 option, the compiler will generate code making use of
4854       the new instructions and registers introduced with the vector
4855       extension facility. The -mtune=z13 option enables z13 specific
4856       instruction scheduling without making use of new instructions.
4857       Compiling code with -march=z13 reduces the default alignment of
4858       vector types bigger than 8 bytes to 8. This is an ABI change and
4859       care must be taken when linking modules compiled with different
4860       arch levels which interchange variables containing vector type
4861       values. For newly compiled code the GNU linker will emit a warning.
4862     * The -mzvector option enables a C/C++ language extension. This
4863       extension provides a new keyword vector which can be used to define
4864       vector type variables. (Note: This is not available when enforcing
4865       strict standard compliance e.g. with -std=c99. Either enable GNU
4866       extensions with e.g. -std=gnu99 or use __vector instead of vector.)
4867       Additionally a set of overloaded builtins is provided which is
4868       partially compatible to the PowerPC Altivec builtins. In order to
4869       make use of these builtins the vecintrin.h header file needs to be
4870       included.
4871     * The new command-line options -march=native, and -mtune=native are
4872       now available on native IBM z Systems. Specifying these options
4873       causes GCC to auto-detect the host CPU and choose the optimal
4874       setting for that system. If GCC is unable to detect the host CPU
4875       these options have no effect.
4876     * The IBM z Systems port now supports target attributes and pragmas.
4877       Please refer to the [20]documentation for details of available
4878       attributes and pragmas as well as usage instructions.
4879     * -fsplit-stack is now supported as part of the IBM z Systems port.
4880       This feature requires a recent gold linker to be used.
4881     * Support for the g5 and g6 -march=/-mtune= CPU level switches has
4882       been deprecated and will be removed in a future GCC release. -m31
4883       from now on defaults to -march=z900 if not specified otherwise.
4884       -march=native on a g5/g6 machine will default to -march=z900.
4885
4886  SH
4887
4888     * Support for SH5 / SH64 has been declared obsolete and will be
4889       removed in future releases.
4890     * Support for the FDPIC ABI has been added. It can be enabled using
4891       the new -mfdpic target option and --enable-fdpic configure option.
4892
4893  SPARC
4894
4895     * An ABI bug has been fixed in 64-bit mode. Unfortunately, this
4896       change will break binary compatibility with earlier releases for
4897       code it affects, but this should be pretty rare in practice. The
4898       conditions are: a 16-byte structure containing a double or a 8-byte
4899       vector in the second half is passed to a subprogram in slot #15,
4900       for example as 16th parameter if the first 15 ones have at most 8
4901       bytes. The double or vector was wrongly passed in floating-point
4902       register %d32 in lieu of on the stack as per the SPARC calling
4903       conventions.
4904
4905Operating Systems
4906
4907  AIX
4908
4909     * DWARF debugging support for AIX 7.1 has been enabled as an optional
4910       debugging format. A more recent Technology Level (TL) and GCC built
4911       with that level are required for full exploitation of DWARF
4912       debugging capabilities.
4913
4914  Linux
4915
4916     * Support for the [21]musl C library was added for the AArch64, ARM,
4917       MicroBlaze, MIPS, MIPS64, PowerPC, PowerPC64, SH, i386, x32 and
4918       x86_64 targets. It can be selected using the new -mmusl option in
4919       case musl is not the default libc. GCC defaults to musl libc if it
4920       is built with a target triplet matching the *-linux-musl* pattern.
4921
4922  RTEMS
4923
4924     * The RTEMS thread model implementation changed. Mutexes now use
4925       self-contained objects defined in Newlib <sys/lock.h> instead of
4926       Classic API semaphores. The keys for thread specific data and the
4927       once function are directly defined via <pthread.h>. Self-contained
4928       condition variables are provided via Newlib <sys/lock.h>. The RTEMS
4929       thread model also supports C++11 threads.
4930     * OpenMP support now uses self-contained objects provided by Newlib
4931       <sys/lock.h> and offers a significantly better performance compared
4932       to the POSIX configuration of libgomp. It is possible to configure
4933       thread pools for each scheduler instance via the environment
4934       variable GOMP_RTEMS_THREAD_POOLS.
4935
4936  Solaris
4937
4938     * Solaris 12 is now fully supported. Minimal support had already been
4939       present in GCC 5.3.
4940     * Solaris 12 provides a full set of startup files (crt1.o, crti.o,
4941       crtn.o), which GCC now prefers over its own ones.
4942     * Position independent executables (PIE) are now supported on Solaris
4943       12.
4944     * Constructor priority is now supported on Solaris 12 with the system
4945       linker.
4946     * libvtv has been ported to Solaris 11 and up.
4947
4948  Windows
4949
4950     * The option -mstackrealign is now automatically activated in 32-bit
4951       mode whenever the use of SSE instructions is requested.
4952
4953Other significant improvements
4954
4955     * The gcc and g++ driver programs will now provide suggestions for
4956       misspelled command-line options.
4957
4958$ gcc -static-libfortran test.f95
4959gcc: error: unrecognized command line option '-static-libfortran'; did you mean
4960'-static-libgfortran'?
4961
4962     * The --enable-default-pie configure option enables generation of PIE
4963       by default.
4964
4965                                    GCC 6.2
4966
4967   This is the [22]list of problem reports (PRs) from GCC's bug tracking
4968   system that are known to be fixed in the 6.2 release. This list might
4969   not be complete (that is, it is possible that some PRs that have been
4970   fixed are not listed here).
4971
4972Target Specific Changes
4973
4974  SPARC
4975
4976     * Support for --with-cpu-32 and --with-cpu-64 configure options has
4977       been added on bi-architecture platforms.
4978     * Support for the SPARC M7 (Niagara 7) processor has been added.
4979     * Support for the VIS 4.0 instruction set has been added.
4980
4981                                    GCC 6.3
4982
4983   This is the [23]list of problem reports (PRs) from GCC's bug tracking
4984   system that are known to be fixed in the 6.3 release. This list might
4985   not be complete (that is, it is possible that some PRs that have been
4986   fixed are not listed here).
4987
4988Target Specific Changes
4989
4990  IA-32/x86-64
4991
4992     * Support for the [24]deprecated pcommit instruction has been
4993       removed.
4994
4995                                    GCC 6.4
4996
4997   This is the [25]list of problem reports (PRs) from GCC's bug tracking
4998   system that are known to be fixed in the 6.4 release. This list might
4999   not be complete (that is, it is possible that some PRs that have been
5000   fixed are not listed here).
5001
5002Operating Systems
5003
5004  RTEMS
5005
5006     * The ABI changes on ARM so that no short enums are used by default.
5007
5008                                    GCC 6.5
5009
5010   This is the [26]list of problem reports (PRs) from GCC's bug tracking
5011   system that are known to be fixed in the 6.5 release. This list might
5012   not be complete (that is, it is possible that some PRs that have been
5013   fixed are not listed here).
5014
5015
5016    For questions related to the use of GCC, please consult these web
5017    pages and the [27]GCC manuals. If that fails, the
5018    [28]gcc-help@gcc.gnu.org mailing list might help. Comments on these
5019    web pages and the development of GCC are welcome on our developer
5020    list at [29]gcc@gcc.gnu.org. All of [30]our lists have public
5021    archives.
5022
5023   Copyright (C) [31]Free Software Foundation, Inc. Verbatim copying and
5024   distribution of this entire article is permitted in any medium,
5025   provided this notice is preserved.
5026
5027   These pages are [32]maintained by the GCC team. Last modified
5028   2020-08-01[33].
5029
5030References
5031
5032   1. http://gcc.gnu.org/gcc-6/porting_to.html
5033   2. http://gcc.gnu.org/onlinedocs/index.html#current
5034   3. https://gcc.gnu.org/ml/gcc/2015-08/msg00101.html
5035   4. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71151
5036   5. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87822
5037   6. https://www.openacc.org/
5038   7. https://gcc.gnu.org/wiki/OpenACC
5039   8. https://gcc.gnu.org/wiki/Offloading
5040   9. https://www.openmp.org/specifications/
5041  10. https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2014-1266
5042  11. http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2015/n4377.pdf
5043  12. https://gcc.gnu.org/projects/cxx-status.html#cxx1z
5044  13. https://gcc.gnu.org/onlinedocs/gcc-6.1.0/jit/topics/performance.html
5045  14. https://gcc.gnu.org/onlinedocs/gcc-6.1.0/jit/topics/functions.html#gcc_jit_block_end_with_switch
5046  15. https://gcc.gnu.org/onlinedocs/gcc-6.1.0/jit/topics/contexts.html#gcc_jit_context_set_bool_allow_unreachable_blocks
5047  16. https://gcc.gnu.org/onlinedocs/gcc-6.1.0/jit/topics/contexts.html#gcc_jit_context_add_command_line_option
5048  17. https://gcc.gnu.org/onlinedocs/gcc-6.1.0/gcc/AArch64-Function-Attributes.html#AArch64-Function-Attributes
5049  18. https://gcc.gnu.org/onlinedocs/gcc-6.1.0/gcc/ARM-Function-Attributes.html#ARM-Function-Attributes
5050  19. https://gcc.gnu.org/onlinedocs/gcc-6.1.0/gcc/Named-Address-Spaces.html#Named-Address-Spaces
5051  20. https://gcc.gnu.org/onlinedocs/gcc-6.1.0/gcc/S_002f390-Function-Attributes.html#S_002f390-Function-Attributes
5052  21. http://www.musl-libc.org/
5053  22. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=6.2
5054  23. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=6.3
5055  24. https://software.intel.com/content/www/us/en/develop/blogs/deprecate-pcommit-instruction.html
5056  25. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=6.4
5057  26. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=6.5
5058  27. https://gcc.gnu.org/onlinedocs/
5059  28. mailto:gcc-help@gcc.gnu.org
5060  29. mailto:gcc@gcc.gnu.org
5061  30. https://gcc.gnu.org/lists.html
5062  31. https://www.fsf.org/
5063  32. https://gcc.gnu.org/about.html
5064  33. http://validator.w3.org/check/referer
5065======================================================================
5066http://gcc.gnu.org/gcc-5/index.html
5067
5068                              GCC 5 Release Series
5069
5070   (This release series is no longer supported.)
5071
5072   October 10, 2017
5073
5074   The [1]GNU project and the GCC developers are pleased to announce the
5075   release of GCC 5.5.
5076
5077   This release is a bug-fix release, containing fixes for regressions in
5078   GCC 5.4 relative to previous releases of GCC.
5079
5080Release History
5081
5082   GCC 5.5
5083          October 10, 2017 ([2]changes, [3]documentation)
5084
5085   GCC 5.4
5086          June 3, 2016 ([4]changes, [5]documentation)
5087
5088   GCC 5.3
5089          December 4, 2015 ([6]changes, [7]documentation)
5090
5091   GCC 5.2
5092          July 16, 2015 ([8]changes, [9]documentation)
5093
5094   GCC 5.1
5095          April 22, 2015 ([10]changes, [11]documentation)
5096
5097References and Acknowledgements
5098
5099   GCC used to stand for the GNU C Compiler, but since the compiler
5100   supports several other languages aside from C, it now stands for the
5101   GNU Compiler Collection.
5102
5103   A list of [12]successful builds is updated as new information becomes
5104   available.
5105
5106   The GCC developers would like to thank the numerous people that have
5107   contributed new features, improvements, bug fixes, and other changes as
5108   well as test results to GCC. This [13]amazing group of volunteers is
5109   what makes GCC successful.
5110
5111   For additional information about GCC please refer to the [14]GCC
5112   project web site or contact the [15]GCC development mailing list.
5113
5114   To obtain GCC please use [16]our mirror sites or [17]our version
5115   control system.
5116
5117
5118    For questions related to the use of GCC, please consult these web
5119    pages and the [18]GCC manuals. If that fails, the
5120    [19]gcc-help@gcc.gnu.org mailing list might help. Comments on these
5121    web pages and the development of GCC are welcome on our developer
5122    list at [20]gcc@gcc.gnu.org. All of [21]our lists have public
5123    archives.
5124
5125   Copyright (C) [22]Free Software Foundation, Inc. Verbatim copying and
5126   distribution of this entire article is permitted in any medium,
5127   provided this notice is preserved.
5128
5129   These pages are [23]maintained by the GCC team. Last modified
5130   2020-01-14[24].
5131
5132References
5133
5134   1. http://www.gnu.org/
5135   2. http://gcc.gnu.org/gcc-5/changes.html
5136   3. http://gcc.gnu.org/onlinedocs/5.5.0/
5137   4. http://gcc.gnu.org/gcc-5/changes.html
5138   5. http://gcc.gnu.org/onlinedocs/5.4.0/
5139   6. http://gcc.gnu.org/gcc-5/changes.html
5140   7. http://gcc.gnu.org/onlinedocs/5.3.0/
5141   8. http://gcc.gnu.org/gcc-5/changes.html
5142   9. http://gcc.gnu.org/onlinedocs/5.2.0/
5143  10. http://gcc.gnu.org/gcc-5/changes.html
5144  11. http://gcc.gnu.org/onlinedocs/5.1.0/
5145  12. http://gcc.gnu.org/gcc-5/buildstat.html
5146  13. http://gcc.gnu.org/onlinedocs/gcc/Contributors.html
5147  14. http://gcc.gnu.org/index.html
5148  15. mailto:gcc@gcc.gnu.org
5149  16. http://gcc.gnu.org/mirrors.html
5150  17. http://gcc.gnu.org/git.html
5151  18. https://gcc.gnu.org/onlinedocs/
5152  19. mailto:gcc-help@gcc.gnu.org
5153  20. mailto:gcc@gcc.gnu.org
5154  21. https://gcc.gnu.org/lists.html
5155  22. https://www.fsf.org/
5156  23. https://gcc.gnu.org/about.html
5157  24. http://validator.w3.org/check/referer
5158======================================================================
5159http://gcc.gnu.org/gcc-5/changes.html
5160
5161                              GCC 5 Release Series
5162                        Changes, New Features, and Fixes
5163
5164Caveats
5165
5166     * The default mode for C is now -std=gnu11 instead of -std=gnu89.
5167     * The C++ runtime library (libstdc++) uses a new ABI by default (see
5168       [1]below).
5169     * The Graphite framework for loop optimizations no longer requires
5170       the CLooG library, only ISL version 0.14 (recommended) or 0.12.2.
5171       The installation manual contains more information about
5172       requirements to build GCC.
5173     * The non-standard C++0x type traits has_trivial_default_constructor,
5174       has_trivial_copy_constructor and has_trivial_copy_assign have been
5175       deprecated and will be removed in a future version. The standard
5176       C++11 traits is_trivially_default_constructible,
5177       is_trivially_copy_constructible and is_trivially_copy_assignable
5178       should be used instead.
5179     * On AVR, support has been added for the devices
5180       ATtiny4/5/9/10/20/40. This requires Binutils 2.25 or newer.
5181     * The AVR port uses a new scheme to describe supported devices: For
5182       each supported device the compiler provides a device-specific
5183       [2]spec file. If the compiler is used together with AVR-LibC, this
5184       requires at least GCC 5.2 and a version of AVR-LibC which
5185       implements [3]feature #44574.
5186
5187General Optimizer Improvements
5188
5189     * Inter-procedural optimization improvements:
5190          + An Identical Code Folding (ICF) pass (controlled via
5191            -fipa-icf) has been added. Compared to the identical code
5192            folding performed by the Gold linker this pass does not
5193            require function sections. It also performs merging before
5194            inlining, so inter-procedural optimizations are aware of the
5195            code re-use. On the other hand not all unifications performed
5196            by a linker are doable by GCC which must honor aliasing
5197            information. During link-time optimization of Firefox, this
5198            pass unifies about 31000 functions, that is 14% overall.
5199          + The devirtualization pass was significantly improved by adding
5200            better support for speculative devirtualization and dynamic
5201            type detection. About 50% of virtual calls in Firefox are now
5202            speculatively devirtualized during link-time optimization.
5203          + A new comdat localization pass allows the linker to eliminate
5204            more dead code in presence of C++ inline functions.
5205          + Virtual tables are now optimized. Local aliases are used to
5206            reduce dynamic linking time of C++ virtual tables on ELF
5207            targets and data alignment has been reduced to limit data
5208            segment bloat.
5209          + A new -fno-semantic-interposition option can be used to
5210            improve code quality of shared libraries where interposition
5211            of exported symbols is not allowed.
5212          + Write-only variables are now detected and optimized out.
5213          + With profile feedback the function inliner can now bypass
5214            --param inline-insns-auto and --param inline-insns-single
5215            limits for hot calls.
5216          + The IPA reference pass was significantly sped up making it
5217            feasible to enable -fipa-reference with -fprofile-generate.
5218            This also solves a bottleneck seen when building Chromium with
5219            link-time optimization.
5220          + The symbol table and call-graph API was reworked to C++ and
5221            simplified.
5222          + The interprocedural propagation of constants now also
5223            propagates alignments of pointer parameters. This for example
5224            means that the vectorizer often does not need to generate loop
5225            prologues and epilogues to make up for potential
5226            misalignments.
5227     * Link-time optimization improvements:
5228          + One Definition Rule based merging of C++ types has been
5229            implemented. Type merging enables better devirtualization and
5230            alias analysis. Streaming extra information needed to merge
5231            types adds about 2-6% of memory size and object size increase.
5232            This can be controlled by -flto-odr-type-merging.
5233          + Command-line optimization and target options are now streamed
5234            on a per-function basis and honored by the link-time
5235            optimizer. This change makes link-time optimization a more
5236            transparent replacement of per-file optimizations. It is now
5237            possible to build projects that require different optimization
5238            settings for different translation units (such as -ffast-math,
5239            -mavx, or -finline). Contrary to earlier GCC releases, the
5240            optimization and target options passed on the link command
5241            line are ignored.
5242            Note that this applies only to those command-line options that
5243            can be passed to optimize and target attributes. Command-line
5244            options affecting global code generation (such as -fpic),
5245            warnings (such as -Wodr), optimizations affecting the way
5246            static variables are optimized (such as -fcommon), debug
5247            output (such as -g), and --param parameters can be applied
5248            only to the whole link-time optimization unit. In these cases,
5249            it is recommended to consistently use the same options at both
5250            compile time and link time.
5251          + GCC bootstrap now uses slim LTO object files.
5252          + Memory usage and link times were improved. Tree merging was
5253            sped up, memory usage of GIMPLE declarations and types was
5254            reduced, and, support for on-demand streaming of variable
5255            constructors was added.
5256     * Feedback directed optimization improvements:
5257          + A new auto-FDO mode uses profiles collected by low overhead
5258            profiling tools (perf) instead of more expensive program
5259            instrumentation (via -fprofile-generate). SPEC2006 benchmarks
5260            on x86-64 improve by 4.7% with auto-FDO and by 7.3% with
5261            traditional feedback directed optimization.
5262          + Profile precision was improved in presence of C++ inline and
5263            extern inline functions.
5264          + The new gcov-tool utility allows manipulating profiles.
5265          + Profiles are now more tolerant to source file changes (this
5266            can be controlled by --param profile-func-internal-id).
5267     * Register allocation improvements:
5268          + A new local register allocator (LRA) sub-pass, controlled by
5269            -flra-remat, implements control-flow sensitive global register
5270            rematerialization. Instead of spilling and restoring a
5271            register value, it is recalculated if it is profitable. The
5272            sub-pass improved SPEC2000 generated code by 1% and 0.5%
5273            correspondingly on ARM and x86-64.
5274          + Reuse of the PIC hard register, instead of using a fixed
5275            register, was implemented on x86/x86-64 targets. This improves
5276            generated PIC code performance as more hard registers can be
5277            used. Shared libraries can significantly benefit from this
5278            optimization. Currently it is switched on only for x86/x86-64
5279            targets. As RA infrastructure is already implemented for PIC
5280            register reuse, other targets might follow this in the future.
5281          + A simple form of inter-procedural RA was implemented. When it
5282            is known that a called function does not use caller-saved
5283            registers, save/restore code is not generated around the call
5284            for such registers. This optimization can be controlled by
5285            -fipa-ra
5286          + LRA is now much more effective at generating spills of general
5287            registers into vector registers instead of memory on
5288            architectures (e.g., modern Intel processors) where this is
5289            profitable.
5290     * UndefinedBehaviorSanitizer gained a few new sanitization options:
5291          + -fsanitize=float-divide-by-zero: detect floating-point
5292            division by zero;
5293          + -fsanitize=float-cast-overflow: check that the result of
5294            floating-point type to integer conversions do not overflow;
5295          + -fsanitize=bounds: enable instrumentation of array bounds and
5296            detect out-of-bounds accesses;
5297          + -fsanitize=alignment: enable alignment checking, detect
5298            various misaligned objects;
5299          + -fsanitize=object-size: enable object size checking, detect
5300            various out-of-bounds accesses.
5301          + -fsanitize=vptr: enable checking of C++ member function calls,
5302            member accesses and some conversions between pointers to base
5303            and derived classes, detect if the referenced object does not
5304            have the correct dynamic type.
5305     * Pointer Bounds Checker, a bounds violation detector, has been added
5306       and can be enabled via -fcheck-pointer-bounds. Memory accesses are
5307       instrumented with run-time checks of used pointers against their
5308       bounds to detect pointer bounds violations (overflows). The Pointer
5309       Bounds Checker is available on x86/x86-64 GNU/Linux targets with a
5310       new ISA extension Intel MPX support. See the Pointer Bounds Checker
5311       [4]Wiki page for more details.
5312
5313New Languages and Language specific improvements
5314
5315     * [5]OpenMP 4.0 specification offloading features are now supported
5316       by the C, C++, and Fortran compilers. Generic changes:
5317          + Infrastructure (suitable for any vendor).
5318          + Testsuite which covers offloading from the [6]OpenMP 4.0
5319            Examples document.
5320       Specific for upcoming Intel Xeon Phi products:
5321          + Run-time library.
5322          + Card emulator.
5323     * GCC 5 includes a preliminary implementation of the OpenACC 2.0a
5324       specification. OpenACC is intended for programming accelerator
5325       devices such as GPUs. See [7]the OpenACC wiki page for more
5326       information.
5327
5328  C family
5329
5330     * The default setting of the -fdiagnostics-color= command-line option
5331       is now [8]configurable when building GCC using configuration option
5332       --with-diagnostics-color=. The possible values are: never, always,
5333       auto and auto-if-env. The new default auto uses color only when the
5334       standard error is a terminal. The default in GCC 4.9 was
5335       auto-if-env, which is equivalent to auto if there is a non-empty
5336       GCC_COLORS environment variable, and never otherwise. As in GCC
5337       4.9, an empty GCC_COLORS variable in the environment will always
5338       disable colors, no matter what the default is or what command-line
5339       options are used.
5340     * A new command-line option -Wswitch-bool has been added for the C
5341       and C++ compilers, which warns whenever a switch statement has an
5342       index of boolean type.
5343     * A new command-line option -Wlogical-not-parentheses has been added
5344       for the C and C++ compilers, which warns about "logical not" used
5345       on the left hand side operand of a comparison.
5346     * A new command-line option -Wsizeof-array-argument has been added
5347       for the C and C++ compilers, which warns when the sizeof operator
5348       is applied to a parameter that has been declared as an array in a
5349       function definition.
5350     * A new command-line option -Wbool-compare has been added for the C
5351       and C++ compilers, which warns about boolean expressions compared
5352       with an integer value different from true/false.
5353     * Full support for [9]Cilk Plus has been added to the GCC compiler.
5354       Cilk Plus is an extension to the C and C++ languages to support
5355       data and task parallelism.
5356     * A new attribute no_reorder prevents reordering of selected symbols
5357       against other such symbols or inline assembler. This enables to
5358       link-time optimize the Linux kernel without having to resort to
5359       -fno-toplevel-reorder that disables several optimizations.
5360     * New preprocessor constructs, __has_include and __has_include_next,
5361       to test the availability of headers have been added.
5362       This demonstrates a way to include the header <optional> only if it
5363       is available:
5364
5365#ifdef __has_include
5366#  if __has_include(<optional>)
5367#    include <optional>
5368#    define have_optional 1
5369#  elif __has_include(<experimental/optional>)
5370#    include <experimental/optional>
5371#    define have_optional 1
5372#    define experimental_optional
5373#  else
5374#    define have_optional 0
5375#  endif
5376#endif
5377
5378       The header search paths for __has_include and __has_include_next
5379       are equivalent to those of the standard directive #include and the
5380       extension #include_next respectively.
5381     * A new built-in function-like macro to determine the existence of an
5382       attribute, __has_attribute, has been added. The equivalent built-in
5383       macro __has_cpp_attribute was added to C++ to support
5384       [10]Feature-testing recommendations for C++. The macro
5385       __has_attribute is added to all C-like languages as an extension:
5386
5387int
5388#ifdef __has_attribute
5389#  if __has_attribute(__noinline__)
5390  __attribute__((__noinline__))
5391#  endif
5392#endif
5393foo(int x);
5394
5395       If an attribute exists, a nonzero constant integer is returned. For
5396       standardized C++ attributes a date is returned, otherwise the
5397       constant returned is 1. Both __has_attribute and
5398       __has_cpp_attribute will add underscores to an attribute name if
5399       necessary to resolve the name. For C++11 and onwards the attribute
5400       may be scoped.
5401     * A new set of built-in functions for arithmetics with overflow
5402       checking has been added: __builtin_add_overflow,
5403       __builtin_sub_overflow and __builtin_mul_overflow and for
5404       compatibility with clang also other variants. These builtins have
5405       two integral arguments (which don't need to have the same type),
5406       the arguments are extended to infinite precision signed type, +, -
5407       or * is performed on those, and the result is stored in an integer
5408       variable pointed to by the last argument. If the stored value is
5409       equal to the infinite precision result, the built-in functions
5410       return false, otherwise true. The type of the integer variable that
5411       will hold the result can be different from the types of the first
5412       two arguments. The following snippet demonstrates how this can be
5413       used in computing the size for the calloc function:
5414
5415void *
5416calloc (size_t x, size_t y)
5417{
5418  size_t sz;
5419  if (__builtin_mul_overflow (x, y, &sz))
5420    return NULL;
5421  void *ret = malloc (sz);
5422  if (ret) memset (res, 0, sz);
5423  return ret;
5424}
5425
5426       On e.g. i?86 or x86-64 the above will result in a mul instruction
5427       followed by a jump on overflow.
5428     * The option -fextended-identifiers is now enabled by default for
5429       C++, and for C99 and later C versions. Various bugs in the
5430       implementation of extended identifiers have been fixed.
5431
5432  C
5433
5434     * The default mode has been changed to -std=gnu11.
5435     * A new command-line option -Wc90-c99-compat has been added to warn
5436       about features not present in ISO C90, but present in ISO C99.
5437     * A new command-line option -Wc99-c11-compat has been added to warn
5438       about features not present in ISO C99, but present in ISO C11.
5439     * It is possible to disable warnings about conversions between
5440       pointers that have incompatible types via a new warning option
5441       -Wno-incompatible-pointer-types; warnings about implicit
5442       incompatible integer to pointer and pointer to integer conversions
5443       via a new warning option -Wno-int-conversion; and warnings about
5444       qualifiers on pointers being discarded via a new warning option
5445       -Wno-discarded-qualifiers.
5446     * To allow proper use of const qualifiers with multidimensional
5447       arrays, GCC will not warn about incompatible pointer types anymore
5448       for conversions between pointers to arrays with and without const
5449       qualifier (except when using -pedantic). Instead, a new warning is
5450       emitted only if the const qualifier is lost. This can be controlled
5451       with a new warning option -Wno-discarded-array-qualifiers.
5452     * The C front end now generates more precise caret diagnostics.
5453     * The -pg command-line option now only affects the current file in an
5454       LTO build.
5455
5456  C++
5457
5458     * G++ now supports [11]C++14 variable templates.
5459     * -Wnon-virtual-dtor doesn't warn anymore for final classes.
5460     * Excessive template instantiation depth is now a fatal error. This
5461       prevents excessive diagnostics that usually do not help to identify
5462       the problem.
5463     * G++ and libstdc++ now implement the feature-testing macros from
5464       [12]Feature-testing recommendations for C++.
5465     * G++ now allows typename in a template template parameter.
5466
5467template<template<typename> typename X> struct D; // OK
5468
5469     * G++ now supports [13]C++14 aggregates with non-static data member
5470       initializers.
5471
5472struct A { int i, j = i; };
5473A a = { 42 }; // a.j is also 42
5474
5475     * G++ now supports [14]C++14 extended constexpr.
5476
5477constexpr int f (int i)
5478{
5479  int j = 0;
5480  for (; i > 0; --i)
5481    ++j;
5482  return j;
5483}
5484
5485constexpr int i = f(42); // i is 42
5486
5487     * G++ now supports the [15]C++14 sized deallocation functions.
5488
5489void operator delete (void *, std::size_t) noexcept;
5490void operator delete[] (void *, std::size_t) noexcept;
5491
5492     * A new One Definition Rule violation warning (controlled by -Wodr)
5493       detects mismatches in type definitions and virtual table contents
5494       during link-time optimization.
5495     * New warnings -Wsuggest-final-types and -Wsuggest-final-methods help
5496       developers to annotate programs with final specifiers (or anonymous
5497       namespaces) to improve code generation. These warnings can be used
5498       at compile time, but they are more useful in combination with
5499       link-time optimization.
5500     * G++ no longer supports [16]N3639 variable length arrays, as they
5501       were removed from the C++14 working paper prior to ratification.
5502       GNU VLAs are still supported, so VLA support is now the same in
5503       C++14 mode as in C++98 and C++11 modes.
5504     * G++ now allows passing a non-trivially-copyable class via C
5505       varargs, which is conditionally-supported with
5506       implementation-defined semantics in the standard. This uses the
5507       same calling convention as a normal value parameter.
5508     * G++ now defaults to -fabi-version=9 and -fabi-compat-version=2. So
5509       various mangling bugs are fixed, but G++ will still emit aliases
5510       with the old, wrong mangling where feasible. -Wabi=2 will warn
5511       about differences between ABI version 2 and the current setting.
5512     * G++ 5.2 fixes the alignment of std::nullptr_t. Most code is likely
5513       to be unaffected, but -Wabi=8 will warn about a non-static data
5514       member with type std::nullptr_t which changes position due to this
5515       change.
5516
5517    Runtime Library (libstdc++)
5518
5519     * A [17]Dual ABI is provided by the library. A new ABI is enabled by
5520       default. The old ABI is still supported and can be used by defining
5521       the macro _GLIBCXX_USE_CXX11_ABI to 0 before including any C++
5522       standard library headers.
5523     * A new implementation of std::string is enabled by default, using
5524       the small string optimization instead of copy-on-write reference
5525       counting.
5526     * A new implementation of std::list is enabled by default, with an
5527       O(1) size() function;
5528     * [18]Full support for C++11, including the following new features:
5529          + std::deque and std::vector<bool> meet the allocator-aware
5530            container requirements;
5531          + movable and swappable iostream classes;
5532          + support for std::align and std::aligned_union;
5533          + type traits std::is_trivially_copyable,
5534            std::is_trivially_constructible, std::is_trivially_assignable
5535            etc.;
5536          + I/O manipulators std::put_time, std::get_time, std::hexfloat
5537            and std::defaultfloat;
5538          + generic locale-aware std::isblank;
5539          + locale facets for Unicode conversion;
5540          + atomic operations for std::shared_ptr;
5541          + std::notify_all_at_thread_exit() and functions for making
5542            futures ready at thread exit.
5543     * Support for the C++11 hexfloat manipulator changes how the num_put
5544       facet formats floating point types when
5545       ios_base::fixed|ios_base::scientific is set in a stream's fmtflags.
5546       This change affects all language modes, even though the C++98
5547       standard gave no special meaning to that combination of flags. To
5548       prevent the use of hexadecimal notation for floating point types
5549       use str.unsetf(std::ios_base::floatfield) to clear the relevant
5550       bits in str.flags().
5551     * [19]Full experimental support for C++14, including the following
5552       new features:
5553          + std::is_final type trait;
5554          + heterogeneous comparison lookup in associative containers.
5555          + global functions cbegin, cend, rbegin, rend, crbegin, and
5556            crend for range access to containers, arrays and initializer
5557            lists.
5558     * [20]Improved experimental support for the Library Fundamentals TS,
5559       including:
5560          + class std::experimental::any;
5561          + function template std::experimental::apply;
5562          + function template std::experimental::sample;
5563          + function template std::experimental::search and related
5564            searcher types;
5565          + variable templates for type traits;
5566          + function template std::experimental::not_fn.
5567     * New random number distributions logistic_distribution and
5568       uniform_on_sphere_distribution as extensions.
5569     * [21]GDB Xmethods for containers and std::unique_ptr.
5570
5571  Fortran
5572
5573     * Compatibility notice:
5574          + The version of the module files (.mod) has been incremented.
5575          + For free-form source files [22]-Werror=line-truncation is now
5576            enabled by default. Note that comments exceeding the line
5577            length are not diagnosed. (For fixed-form source code, the
5578            same warning is available but turned off by default, such that
5579            excess characters are ignored. -ffree-line-length-n and
5580            -ffixed-line-length-n can be used to modify the default line
5581            lengths of 132 and 72 columns, respectively.)
5582          + The -Wtabs option is now more sensible: with -Wtabs the
5583            compiler warns if it encounters tabs and with -Wno-tabs this
5584            warning is turned off. Before, -Wno-tabs warned and -Wtabs
5585            disabled the warning. As before, this warning is also enabled
5586            by -Wall, -pedantic and the f95, f2003, f2008 and f2008ts
5587            options of -std=.
5588     * Incomplete support for colorizing diagnostics emitted by gfortran
5589       has been added. The option [23]-fdiagnostics-color controls when
5590       color is used in diagnostics. The default value of this option can
5591       be [24]configured when building GCC. The GCC_COLORS environment
5592       variable can be used to customize the colors or disable coloring
5593       completely. Sample diagnostics output:
5594      $ gfortran -fdiagnostics-color=always -Wuse-without-only test.f90
5595      test.f90:6:1:
5596
5597       0 continue
5598       1
5599      Error: Zero is not a valid statement label at (1)
5600      test.f90:9:6:
5601
5602         USE foo
5603            1
5604      Warning: USE statement at (1) has no ONLY qualifier [-Wuse-without-only]
5605
5606     * The -Wuse-without-only option has been added to warn when a USE
5607       statement has no ONLY qualifier and thus implicitly imports all
5608       public entities of the used module.
5609     * Formatted READ and WRITE statements now work correctly in
5610       locale-aware programs. For more information and potential caveats,
5611       see [25]Section 5.3 Thread-safety of the runtime library in the
5612       manual.
5613     * [26]Fortran 2003:
5614          + The intrinsic IEEE modules (IEEE_FEATURES, IEEE_EXCEPTIONS and
5615            IEEE_ARITHMETIC) are now supported.
5616     * [27]Fortran 2008:
5617          + [28]Coarrays: Full experimental support of Fortran 2008's
5618            coarrays with -fcoarray=lib except for allocatable/pointer
5619            components of derived-type coarrays. GCC currently only ships
5620            with a single-image library (libcaf_single), but multi-image
5621            support based on MPI and GASNet is provided by the libraries
5622            of the [29]OpenCoarrays project.
5623     * TS18508 Additional Parallel Features in Fortran:
5624          + Support for the collective intrinsic subroutines CO_MAX,
5625            CO_MIN, CO_SUM, CO_BROADCAST and CO_REDUCE has been added,
5626            including -fcoarray=lib support.
5627          + Support for the new atomic intrinsics has been added,
5628            including -fcoarray=lib support.
5629     * Fortran 2015:
5630          + Support for IMPLICIT NONE (external, type).
5631          + ERROR STOP is now permitted in pure procedures.
5632
5633  Go
5634
5635     * GCC 5 provides a complete implementation of the Go 1.4.2 release.
5636     * Building GCC 5 with Go enabled will install two new programs:
5637       [30]go and [31]gofmt.
5638
5639libgccjit
5640
5641   New in GCC 5 is the ability to build GCC as a shared library for
5642   embedding in other processes (such as interpreters), suitable for
5643   Just-In-Time compilation to machine code.
5644
5645   The shared library has a [32]C API and a [33]C++ wrapper API providing
5646   some "syntactic sugar". There are also bindings available from 3rd
5647   parties for [34]Python and for [35]D.
5648
5649   For example, this library can be used by interpreters for [36]compiling
5650   functions from bytecode to machine code.
5651
5652   The library can also be used for ahead-of-time compilation, enabling
5653   GCC to be plugged into a pre-existing front end. An example of using
5654   this to build a compiler for an esoteric language we'll refer to as
5655   "brainf" can be seen [37]here.
5656
5657   libgccjit is licensed under the GPLv3 (or at your option, any later
5658   version)
5659
5660   It should be regarded as experimental at this time.
5661
5662New Targets and Target Specific Improvements
5663
5664  Reporting stack usage
5665
5666     * The BFIN, FT32, H8300, IQ2000 and M32C targets now support the
5667       -fstack-usage option.
5668
5669  AArch64
5670
5671     * Code generation for the ARM Cortex-A57 processor has been improved.
5672       A more accurate instruction scheduling model for the processor is
5673       now used, and a number of compiler tuning parameters have been set
5674       to offer increased performance when compiling with -mcpu=cortex-a57
5675       or -mtune=cortex-a57.
5676     * A workaround for the ARM Cortex-A53 erratum 835769 has been added
5677       and can be enabled by giving the -mfix-cortex-a53-835769 option.
5678       Alternatively it can be enabled by default by configuring GCC with
5679       the --enable-fix-cortex-a53-835769 option.
5680     * The optional cryptographic extensions to the ARMv8-A architecture
5681       are no longer enabled by default when specifying the
5682       -mcpu=cortex-a53, -mcpu=cortex-a57 or -mcpu=cortex-a57.cortex-a53
5683       options. To enable these extensions add +crypto to the value of
5684       -mcpu or -march e.g. -mcpu=cortex-a53+crypto.
5685     * Support has been added for the following processors (GCC
5686       identifiers in parentheses): ARM Cortex-A72 (cortex-a72) and
5687       initial support for its big.LITTLE combination with the ARM
5688       Cortex-A53 (cortex-a72.cortex-a53), Cavium ThunderX (thunderx),
5689       Applied Micro X-Gene 1 (xgene1), and Samsung Exynos M1 (exynos-m1).
5690       The GCC identifiers can be used as arguments to the -mcpu or -mtune
5691       options, for example: -mcpu=xgene1 or -mtune=cortex-a72.cortex-a53.
5692       Using -mcpu=cortex-a72 requires a version of GNU binutils that has
5693       support for the Cortex-A72.
5694     * The transitional options -mlra and -mno-lra have been removed. The
5695       AArch64 backend now uses the local register allocator (LRA) only.
5696
5697  ARM
5698
5699     * Thumb-1 assembly code is now generated in unified syntax. The new
5700       option -masm-syntax-unified specifies whether inline assembly code
5701       is using unified syntax. By default the option is off which means
5702       non-unified syntax is used. However this is subject to change in
5703       future releases. Eventually the non-unified syntax will be
5704       deprecated.
5705     * It is now a configure-time error to use the --with-cpu configure
5706       option with either of --with-tune or --with-arch.
5707     * Code generation for the ARM Cortex-A57 processor has been improved.
5708       A more accurate instruction scheduling model for the processor is
5709       now used, and a number of compiler tuning parameters have been set
5710       to offer increased performance when compiling with -mcpu=cortex-a57
5711       or -mtune=cortex-a57.
5712     * Support has been added for the following processors (GCC
5713       identifiers in parentheses): ARM Cortex-A17 (cortex-a17) and
5714       initial support for its big.LITTLE combination with the ARM
5715       Cortex-A7 (cortex-a17.cortex-a7), ARM Cortex-A72 (cortex-a72) and
5716       initial support for its big.LITTLE combination with the ARM
5717       Cortex-A53 (cortex-a72.cortex-a53), ARM Cortex-M7 (cortex-m7),
5718       Applied Micro X-Gene 1 (xgene1), and Samsung Exynos M1 (exynos-m1).
5719       The GCC identifiers can be used as arguments to the -mcpu or -mtune
5720       options, for example: -mcpu=xgene1 or -mtune=cortex-a72.cortex-a53.
5721       Using -mcpu=cortex-a72 requires a version of GNU binutils that has
5722       support for the Cortex-A72.
5723     * The deprecated option -mwords-little-endian has been removed.
5724     * The options -mapcs, -mapcs-frame, -mtpcs-frame and
5725       -mtpcs-leaf-frame which are only applicable to the old ABI have
5726       been deprecated.
5727     * The transitional options -mlra and -mno-lra have been removed. The
5728       ARM backend now uses the local register allocator (LRA) only.
5729
5730  AVR
5731
5732     * The compiler no more supports individual devices like ATmega8.
5733       Specifying, say, -mmcu=atmega8 triggers the usage of the
5734       device-specific [38]spec file specs-atmega8 which is part of the
5735       installation and describes options for the sub-processes like
5736       compiler proper, assembler and linker. You can add support for a
5737       new device -mmcu=mydevice as follows:
5738         1. In an empty directory /someplace, create a new directory
5739            device-specs.
5740         2. Copy a device spec file from the installed device-specs
5741            folder, follow the comments in that file and then save it as
5742            /someplace/device-specs/specs-mydevice.
5743         3. Add -B /someplace -mmcu=mydevice to the compiler's
5744            command-line options. Notice that /someplace must specify an
5745            absolute path and that mydevice must not start with "avr".
5746         4. Provided you have a device-specific library libmydevice.a
5747            available, you can put it at /someplace, dito for a
5748            device-specific startup file crtmydevice.o.
5749       The contents of the device spec files depend on the compiler's
5750       configuration, in particular on --with-avrlibc=no and whether or
5751       not it is configured for RTEMS.
5752     * A new command-line option -nodevicelib has been added. It prevents
5753       the compiler from linking against AVR-LibC's device-specific
5754       library libdevice.a.
5755     * The following three command-line options have been added:
5756
5757        -mrmw
5758                Set if the device supports the read-modify-write
5759                instructions LAC, LAS, LAT and XCH.
5760
5761        -mn-flash=size
5762                Specify the flash size of the device in units of 64 KiB,
5763                rounded up to the next integer as needed. This option
5764                affects the availability of the [39]AVR address-spaces.
5765
5766        -mskip-bug
5767                Set if the device is affected by the respective silicon
5768                bug.
5769
5770       In general, you don't need to set these options by hand. The new
5771       device-specific spec file will set them as needed.
5772
5773  IA-32/x86-64
5774
5775     * New [40]ISA extensions support AVX-512{BW,DQ,VL,IFMA,VBMI} of
5776       Intel's CPU codenamed Skylake Server was added to GCC. That
5777       includes inline assembly support, new intrinsics, and basic
5778       autovectorization. These new AVX-512 extensions are available via
5779       the following GCC switches: AVX-512 Vector Length EVEX feature:
5780       -mavx512vl, AVX-512 Byte and Word instructions: -mavx512bw, AVX-512
5781       Dword and Qword instructions: -mavx512dq, AVX-512 FMA-52
5782       instructions: -mavx512ifma and for AVX-512 Vector Bit Manipulation
5783       Instructions: -mavx512vbmi.
5784     * New ISA extensions support Intel MPX was added to GCC. This new
5785       extension is available via the -mmpx compiler switch. Intel MPX is
5786       a set of processor features which, with compiler, run-time library
5787       and OS support, brings increased robustness to software by run-time
5788       checking pointer references against their bounds. In GCC Intel MPX
5789       is supported by Pointer Bounds Checker and libmpx run-time
5790       libraries.
5791     * The new -mrecord-mcount option for -pg generates a Linux kernel
5792       style table of pointers to mcount or __fentry__ calls at the
5793       beginning of functions. The new -mnop-mcount option in addition
5794       also generates nops in place of the __fentry__ or mcount call, so
5795       that a call per function can be later patched in. This can be used
5796       for low overhead tracing or hot code patching.
5797     * The new -malign-data option controls how GCC aligns variables.
5798       -malign-data=compat uses increased alignment compatible with GCC
5799       4.8 and earlier, -malign-data=abi uses alignment as specified by
5800       the psABI, and -malign-data=cacheline uses increased alignment to
5801       match the cache line size. -malign-data=compat is the default.
5802     * The new -mskip-rax-setup option skips setting up the RAX register
5803       when SSE is disabled and there are no variable arguments passed in
5804       vector registers. This can be used to optimize the Linux kernel.
5805
5806  MIPS
5807
5808     * MIPS Releases 3 and 5 are now directly supported. Use the
5809       command-line options -mips32r3, -mips64r3, -mips32r5 and -mips64r5
5810       to enable code-generation for these processors.
5811     * The Imagination P5600 processor is now supported using the
5812       -march=p5600 command-line option.
5813     * The Cavium Octeon3 processor is now supported using the
5814       -march=octeon3 command-line option.
5815     * MIPS Release 6 is now supported using the -mips32r6 and -mips64r6
5816       command-line options.
5817     * The o32 ABI has been modified and extended. The o32 64-bit
5818       floating-point register support is now obsolete and has been
5819       removed. It has been replaced by three ABI extensions FPXX, FP64A,
5820       and FP64. The meaning of the -mfp64 command-line option has
5821       changed. It is now used to enable the FP64A and FP64 ABI
5822       extensions.
5823          + The FPXX extension requires that code generated to access
5824            double-precision values use even-numbered registers. Code that
5825            adheres to this extension is link-compatible with all other
5826            o32 double-precision ABI variants and will execute correctly
5827            in all hardware FPU modes. The command-line options -mabi=32
5828            -mfpxx can be used to enable this extension. MIPS II is the
5829            minimum processor required.
5830          + The o32 FP64A extension requires that floating-point registers
5831            be 64-bit and odd-numbered single-precision registers are not
5832            allowed. Code that adheres to the o32 FP64A variant is
5833            link-compatible with all other o32 double-precision ABI
5834            variants. The command-line options -mabi=32 -mfp64
5835            -mno-odd-spreg can be used to enable this extension. MIPS32R2
5836            is the minimum processor required.
5837          + The o32 FP64 extension also requires that floating-point
5838            registers be 64-bit, but permits the use of single-precision
5839            registers. Code that adheres to the o32 FP64 variant is
5840            link-compatible with o32 FPXX and o32 FP64A variants only,
5841            i.e. it is not compatible with the original o32
5842            double-precision ABI. The command-line options -mabi=32 -mfp64
5843            -modd-spreg can be used to enable this extension. MIPS32R2 is
5844            the minimum processor required.
5845       The new ABI variants can be enabled by default using the configure
5846       time options --with-fp-32=[32|xx|64] and --with(out)-odd-sp-reg-32.
5847       It is strongly recommended that all vendors begin to set o32 FPXX
5848       as the default ABI. This will be required to run the generated code
5849       on MIPSR5 cores in conjunction with future MIPS SIMD (MSA) code and
5850       MIPSR6 cores.
5851     * GCC will now pass all floating-point options to the assembler if
5852       GNU binutils 2.25 is used. As a result, any inline assembly code
5853       that uses hard-float instructions should be amended to include a
5854       .set directive to override the global assembler options when
5855       compiling for soft-float targets.
5856
5857  NDS32
5858
5859     * The variadic function ABI implementation is now compatible with
5860       past Andes toolchains where the caller uses registers to pass
5861       arguments and the callee is in charge of pushing them on stack.
5862     * The options -mforce-fp-as-gp, -mforbid-fp-as-gp, and -mex9 have
5863       been removed since they are not yet available in the nds32 port of
5864       GNU binutils.
5865     * A new option -mcmodel=[small|medium|large] supports varied code
5866       models on code generation. The -mgp-direct option became
5867       meaningless and can be discarded.
5868
5869  RX
5870
5871     * A new command line option -mno-allow-string-insns can be used to
5872       disable the generation of the SCMPU, SMOVU, SMOVB, SMOVF, SUNTIL,
5873       SWHILE and RMPA instructions. An erratum released by Renesas shows
5874       that it is unsafe to use these instructions on addresses within the
5875       I/O space of the processor. The new option can be used when the
5876       programmer is concerned that the I/O space might be accessed. The
5877       default is still to enable these instructions.
5878
5879  SH
5880
5881     * The compiler will now pass the appropriate --isa= option to the
5882       assembler.
5883     * The default handling for the GBR has been changed from call
5884       clobbered to call preserved. The old behavior can be reinstated by
5885       specifying the option -fcall-used-gbr.
5886     * Support for the SH4A fpchg instruction has been added which will be
5887       utilized when switching between single and double precision FPU
5888       modes.
5889     * The compiler no longer uses the __fpscr_values array for switching
5890       between single and double FPU precision modes on non-SH4A targets.
5891       Instead mode switching will now be performed by storing, modifying
5892       and reloading the FPSCR, so that other FPSCR bits are preserved
5893       across mode switches. The __fpscr_values array that is defined in
5894       libgcc is still present for backwards compatibility, but it will
5895       not be referenced by compiler generated code anymore.
5896     * New builtin functions __builtin_sh_get_fpscr and
5897       __builtin_sh_set_fpscr have been added. The __builtin_sh_set_fpscr
5898       function will mask the specified bits in such a way that the SZ, PR
5899       and FR mode bits will be preserved, while changing the other bits.
5900       These new functions do not reference the __fpscr_values array. The
5901       old functions __set_fpscr and __get_fpscr in libgcc which access
5902       the __fpscr_values array are still present for backwards
5903       compatibility, but their usage is highly discouraged.
5904     * Some improvements to code generated for __atomic built-in
5905       functions.
5906     * When compiling for SH2E the compiler will no longer force the usage
5907       of delay slots for conditional branch instructions bt and bf. The
5908       old behavior can be reinstated (e.g. to work around a hardware bug
5909       in the original SH7055) by specifying the new option
5910       -mcbranch-force-delay-slot.
5911
5912Operating Systems
5913
5914  AIX
5915
5916     * GCC now supports stabs debugging continuation lines to allow long
5917       stabs debug information without overflow that generates AIX linker
5918       errors.
5919
5920  DragonFly BSD
5921
5922     * GCC now supports the DragonFly BSD operating system.
5923
5924  FreeBSD
5925
5926     * GCC now supports the FreeBSD operating system for the arm port
5927       through the arm*-*-freebsd* target triplets.
5928
5929  VxWorks MILS
5930
5931     * GCC now supports the MILS (Multiple Independent Levels of Security)
5932       variant of WindRiver's VxWorks operating system for PowerPC
5933       targets.
5934
5935Other significant improvements
5936
5937     * The gcc-ar, gcc-nm, gcc-ranlib wrappers now understand a -B option
5938       to set the compiler to use.
5939
5940     * When the new command-line option -freport-bug is used, GCC
5941       automatically generates a developer-friendly reproducer whenever an
5942       internal compiler error is encountered.
5943
5944                                    GCC 5.2
5945
5946   This is the [41]list of problem reports (PRs) from GCC's bug tracking
5947   system that are known to be fixed in the 5.2 release. This list might
5948   not be complete (that is, it is possible that some PRs that have been
5949   fixed are not listed here).
5950
5951Target Specific Changes
5952
5953  IA-32/x86-64
5954
5955     * Support for new AMD instructions monitorx and mwaitx has been
5956       added. This includes new intrinsic and built-in support. It is
5957       enabled through option -mmwaitx. The instructions monitorx and
5958       mwaitx implement the same functionality as the old monitor and
5959       mwait instructions. In addition mwaitx adds a configurable timer.
5960       The timer value is received as third argument and stored in
5961       register %ebx.
5962
5963  S/390, System z, IBM z Systems
5964
5965     * Support for the IBM z13 processor has been added. When using the
5966       -march=z13 option, the compiler will generate code making use of
5967       the new instructions and registers introduced with the vector
5968       extension facility. The -mtune=z13 option enables z13 specific
5969       instruction scheduling without making use of new instructions.
5970       Compiling code with -march=z13 reduces the default alignment of
5971       vector types bigger than 8 bytes to 8. This is an ABI change and
5972       care must be taken when linking modules compiled with different
5973       arch levels which interchange variables containing vector type
5974       values. For newly compiled code the GNU linker will emit a warning.
5975     * The -mzvector option enables a C/C++ language extension. This
5976       extension provides a new keyword vector which can be used to define
5977       vector type variables. (Note: This is not available when enforcing
5978       strict standard compliance e.g. with -std=c99. Either enable GNU
5979       extensions with e.g. -std=gnu99 or use __vector instead of vector.)
5980       Additionally a set of overloaded builtins is provided which is
5981       partially compatible to the PowerPC Altivec builtins. In order to
5982       make use of these builtins the vecintrin.h header file needs to be
5983       included.
5984
5985                                    GCC 5.3
5986
5987   This is the [42]list of problem reports (PRs) from GCC's bug tracking
5988   system that are known to be fixed in the 5.3 release. This list might
5989   not be complete (that is, it is possible that some PRs that have been
5990   fixed are not listed here).
5991
5992Target Specific Changes
5993
5994  IA-32/x86-64
5995
5996     * GCC now supports the Intel CPU named Skylake with AVX-512
5997       extensions through -march=skylake-avx512. The switch enables the
5998       following ISA extensions: AVX-512F, AVX512VL, AVX-512CD, AVX-512BW,
5999       AVX-512DQ.
6000
6001  S/390, System z, IBM z Systems
6002
6003     * With this version of GCC IBM z Systems support has been added to
6004       the GO runtime environment. GCC 5.3 has proven to be able to
6005       compile larger GO applications on IBM z Systems.
6006
6007                                    GCC 5.4
6008
6009   This is the [43]list of problem reports (PRs) from GCC's bug tracking
6010   system that are known to be fixed in the 5.4 release. This list might
6011   not be complete (that is, it is possible that some PRs that have been
6012   fixed are not listed here).
6013
6014                                    GCC 5.5
6015
6016   This is the [44]list of problem reports (PRs) from GCC's bug tracking
6017   system that are known to be fixed in the 5.5 release. This list might
6018   not be complete (that is, it is possible that some PRs that have been
6019   fixed are not listed here).
6020
6021Target Specific Changes
6022
6023  IA-32/x86-64
6024
6025     * Support for the [45]deprecated pcommit instruction has been
6026       removed.
6027
6028
6029    For questions related to the use of GCC, please consult these web
6030    pages and the [46]GCC manuals. If that fails, the
6031    [47]gcc-help@gcc.gnu.org mailing list might help. Comments on these
6032    web pages and the development of GCC are welcome on our developer
6033    list at [48]gcc@gcc.gnu.org. All of [49]our lists have public
6034    archives.
6035
6036   Copyright (C) [50]Free Software Foundation, Inc. Verbatim copying and
6037   distribution of this entire article is permitted in any medium,
6038   provided this notice is preserved.
6039
6040   These pages are [51]maintained by the GCC team. Last modified
6041   2020-05-31[52].
6042
6043References
6044
6045   1. http://gcc.gnu.org/gcc-5/changes.html#libstdcxx
6046   2. https://gcc.gnu.org/onlinedocs/gcc/Spec-Files.html
6047   3. https://savannah.nongnu.org/bugs/?44574
6048   4. https://gcc.gnu.org/wiki/Intel%20MPX%20support%20in%20the%20GCC%20compiler
6049   5. https://www.openmp.org/wp-content/uploads/OpenMP4.0.0.pdf
6050   6. https://www.openmp.org/wp-content/uploads/OpenMP4.0.0.Examples.pdf
6051   7. https://gcc.gnu.org/wiki/OpenACC
6052   8. https://gcc.gnu.org/install/configure.html
6053   9. https://www.cilkplus.org/
6054  10. https://isocpp.org/std/standing-documents/sd-6-sg10-feature-test-recommendations
6055  11. http://gcc.gnu.org/projects/cxx1y.html
6056  12. https://isocpp.org/std/standing-documents/sd-6-sg10-feature-test-recommendations
6057  13. http://gcc.gnu.org/projects/cxx1y.html
6058  14. http://gcc.gnu.org/projects/cxx1y.html
6059  15. http://gcc.gnu.org/projects/cxx1y.html
6060  16. http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2013/n3639.html
6061  17. http://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html
6062  18. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/libstdc++/manual/manual/status.html#status.iso.2011
6063  19. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/libstdc++/manual/manual/status.html#status.iso.2014
6064  20. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/libstdc++/manual/manual/status.html#status.iso.2014
6065  21. https://sourceware.org/gdb/current/onlinedocs/gdb/Xmethods-In-Python.html
6066  22. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/gfortran/Error-and-Warning-Options.html
6067  23. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/gcc/Language-Independent-Options.html
6068  24. https://gcc.gnu.org/install/configure.html
6069  25. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/gfortran/Thread-safety-of-the-runtime-library.html
6070  26. https://gcc.gnu.org/wiki/Fortran2003Status
6071  27. https://gcc.gnu.org/wiki/Fortran2008Status
6072  28. https://gcc.gnu.org/wiki/Coarray
6073  29. http://www.opencoarrays.org/
6074  30. https://golang.org/cmd/go/
6075  31. https://golang.org/cmd/gofmt/
6076  32. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/jit/intro/index.html
6077  33. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/jit/cp/index.html
6078  34. https://github.com/davidmalcolm/pygccjit
6079  35. https://github.com/ibuclaw/gccjitd
6080  36. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/jit/intro/tutorial04.html
6081  37. https://gcc.gnu.org/onlinedocs/gcc-5.1.0/jit/intro/tutorial05.html
6082  38. https://gcc.gnu.org/onlinedocs/gcc/Spec-Files.html
6083  39. https://gcc.gnu.org/onlinedocs/gcc/Named-Address-Spaces.html
6084  40. https://software.intel.com/sites/default/files/managed/39/c5/325462-sdm-vol-1-2abcd-3abcd.pdf
6085  41. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=5.2
6086  42. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=5.3
6087  43. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=5.4
6088  44. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=5.5
6089  45. https://software.intel.com/content/www/us/en/develop/blogs/deprecate-pcommit-instruction.html
6090  46. https://gcc.gnu.org/onlinedocs/
6091  47. mailto:gcc-help@gcc.gnu.org
6092  48. mailto:gcc@gcc.gnu.org
6093  49. https://gcc.gnu.org/lists.html
6094  50. https://www.fsf.org/
6095  51. https://gcc.gnu.org/about.html
6096  52. http://validator.w3.org/check/referer
6097======================================================================
6098http://gcc.gnu.org/gcc-4.9/index.html
6099
6100                             GCC 4.9 Release Series
6101
6102   (This release series is no longer supported.)
6103
6104   Aug 3, 2016
6105
6106   The [1]GNU project and the GCC developers are pleased to announce the
6107   release of GCC 4.9.4.
6108
6109   This release is a bug-fix release, containing fixes for regressions in
6110   GCC 4.9.3 relative to previous releases of GCC.
6111
6112Release History
6113
6114   GCC 4.9.4
6115          Aug 3, 2016 ([2]changes, [3]documentation)
6116
6117   GCC 4.9.3
6118          June 26, 2015 ([4]changes, [5]documentation)
6119
6120   GCC 4.9.2
6121          October 30, 2014 ([6]changes, [7]documentation)
6122
6123   GCC 4.9.1
6124          July 16, 2014 ([8]changes, [9]documentation)
6125
6126   GCC 4.9.0
6127          April 22, 2014 ([10]changes, [11]documentation)
6128
6129References and Acknowledgements
6130
6131   GCC used to stand for the GNU C Compiler, but since the compiler
6132   supports several other languages aside from C, it now stands for the
6133   GNU Compiler Collection.
6134
6135   A list of [12]successful builds is updated as new information becomes
6136   available.
6137
6138   The GCC developers would like to thank the numerous people that have
6139   contributed new features, improvements, bug fixes, and other changes as
6140   well as test results to GCC. This [13]amazing group of volunteers is
6141   what makes GCC successful.
6142
6143   For additional information about GCC please refer to the [14]GCC
6144   project web site or contact the [15]GCC development mailing list.
6145
6146   To obtain GCC please use [16]our mirror sites or [17]our version
6147   control system.
6148
6149
6150    For questions related to the use of GCC, please consult these web
6151    pages and the [18]GCC manuals. If that fails, the
6152    [19]gcc-help@gcc.gnu.org mailing list might help. Comments on these
6153    web pages and the development of GCC are welcome on our developer
6154    list at [20]gcc@gcc.gnu.org. All of [21]our lists have public
6155    archives.
6156
6157   Copyright (C) [22]Free Software Foundation, Inc. Verbatim copying and
6158   distribution of this entire article is permitted in any medium,
6159   provided this notice is preserved.
6160
6161   These pages are [23]maintained by the GCC team. Last modified
6162   2020-01-14[24].
6163
6164References
6165
6166   1. http://www.gnu.org/
6167   2. http://gcc.gnu.org/gcc-4.9/changes.html
6168   3. http://gcc.gnu.org/onlinedocs/4.9.4/
6169   4. http://gcc.gnu.org/gcc-4.9/changes.html
6170   5. http://gcc.gnu.org/onlinedocs/4.9.3/
6171   6. http://gcc.gnu.org/gcc-4.9/changes.html
6172   7. http://gcc.gnu.org/onlinedocs/4.9.2/
6173   8. http://gcc.gnu.org/gcc-4.9/changes.html
6174   9. http://gcc.gnu.org/onlinedocs/4.9.1/
6175  10. http://gcc.gnu.org/gcc-4.9/changes.html
6176  11. https://gcc.gnu.org/onlinedocs/4.9.0/
6177  12. http://gcc.gnu.org/gcc-4.9/buildstat.html
6178  13. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
6179  14. http://gcc.gnu.org/index.html
6180  15. mailto:gcc@gcc.gnu.org
6181  16. http://gcc.gnu.org/mirrors.html
6182  17. http://gcc.gnu.org/git.html
6183  18. https://gcc.gnu.org/onlinedocs/
6184  19. mailto:gcc-help@gcc.gnu.org
6185  20. mailto:gcc@gcc.gnu.org
6186  21. https://gcc.gnu.org/lists.html
6187  22. https://www.fsf.org/
6188  23. https://gcc.gnu.org/about.html
6189  24. http://validator.w3.org/check/referer
6190======================================================================
6191http://gcc.gnu.org/gcc-4.9/changes.html
6192
6193                             GCC 4.9 Release Series
6194                        Changes, New Features, and Fixes
6195
6196Caveats
6197
6198     * The mudflap run time checker has been removed. The mudflap options
6199       remain, but do nothing.
6200     * Support for a number of older systems and recently unmaintained or
6201       untested target ports of GCC has been declared obsolete in GCC 4.9.
6202       Unless there is activity to revive them, the next release of GCC
6203       will have their sources permanently removed.
6204       The following ports for individual systems on particular
6205       architectures have been obsoleted:
6206          + Solaris 9 (*-*-solaris2.9). Details can be found in the
6207            [1]announcement.
6208     * On AArch64, the singleton vector types int64x1_t, uint64x1_t and
6209       float64x1_t exported by arm_neon.h are defined to be the same as
6210       their base types. This results in incorrect application of
6211       parameter passing rules to arguments of types int64x1_t and
6212       uint64x1_t, with respect to the AAPCS64 ABI specification. In
6213       addition, names of C++ functions with parameters of these types
6214       (including float64x1_t) are not mangled correctly. The current
6215       typedef declarations also unintentionally allow implicit casting
6216       between singleton vector types and their base types. These issues
6217       will be resolved in a near future release. See [2]PR60825 for more
6218       information.
6219
6220   More information on porting to GCC 4.9 from previous versions of GCC
6221   can be found in the [3]porting guide for this release.
6222
6223General Optimizer Improvements
6224
6225     * AddressSanitizer, a fast memory error detector, is now available on
6226       ARM.
6227     * UndefinedBehaviorSanitizer (ubsan), a fast undefined behavior
6228       detector, has been added and can be enabled via
6229       -fsanitize=undefined. Various computations will be instrumented to
6230       detect undefined behavior at runtime. UndefinedBehaviorSanitizer is
6231       currently available for the C and C++ languages.
6232     * Link-time optimization (LTO) improvements:
6233          + Type merging was rewritten. The new implementation is
6234            significantly faster and uses less memory.
6235          + Better partitioning algorithm resulting in less streaming
6236            during link time.
6237          + Early removal of virtual methods reduces the size of object
6238            files and improves link-time memory usage and compile time.
6239          + Function bodies are now loaded on-demand and released early
6240            improving overall memory usage at link time.
6241          + C++ hidden keyed methods can now be optimized out.
6242          + When using a linker plugin, compiling with the -flto option
6243            now generates slim object files (.o) which only contain
6244            intermediate language representation for LTO. Use
6245            -ffat-lto-objects to create files which contain additionally
6246            the object code. To generate static libraries suitable for LTO
6247            processing, use gcc-ar and gcc-ranlib; to list symbols from a
6248            slim object file use gcc-nm. (This requires that ar, ranlib
6249            and nm have been compiled with plugin support.)
6250       Memory usage building Firefox with debug enabled was reduced from
6251       15GB to 3.5GB; link time from 1700 seconds to 350 seconds.
6252     * Inter-procedural optimization improvements:
6253          + New type inheritance analysis module improving
6254            devirtualization. Devirtualization now takes into account
6255            anonymous name-spaces and the C++11 final keyword.
6256          + New speculative devirtualization pass (controlled by
6257            -fdevirtualize-speculatively.
6258          + Calls that were speculatively made direct are turned back to
6259            indirect where direct call is not cheaper.
6260          + Local aliases are introduced for symbols that are known to be
6261            semantically equivalent across shared libraries improving
6262            dynamic linking times.
6263     * Feedback directed optimization improvements:
6264          + Profiling of programs using C++ inline functions is now more
6265            reliable.
6266          + New time profiling determines typical order in which functions
6267            are executed.
6268          + A new function reordering pass (controlled by
6269            -freorder-functions) significantly reduces startup time of
6270            large applications. Until binutils support is completed, it is
6271            effective only with link-time optimization.
6272          + Feedback driven indirect call removal and devirtualization now
6273            handle cross-module calls when link-time optimization is
6274            enabled.
6275
6276New Languages and Language specific improvements
6277
6278     * Version 4.0 of the [4]OpenMP specification is now supported in the
6279       C and C++ compilers and starting with the 4.9.1 release also in the
6280       Fortran compiler. The new -fopenmp-simd option can be used to
6281       enable OpenMP's SIMD directives while ignoring other OpenMP
6282       directives. The new [5]-fsimd-cost-model= option permits to tune
6283       the vectorization cost model for loops annotated with OpenMP and
6284       Cilk Plus simd directives. -Wopenmp-simd warns when the current
6285       cost model overrides simd directives set by the user.
6286     * The -Wdate-time option has been added for the C, C++ and Fortran
6287       compilers, which warns when the __DATE__, __TIME__ or __TIMESTAMP__
6288       macros are used. Those macros might prevent bit-wise-identical
6289       reproducible compilations.
6290
6291  Ada
6292
6293     * GNAT switched to Ada 2012 instead of Ada 2005 by default.
6294
6295  C family
6296
6297     * Support for colorizing diagnostics emitted by GCC has been added.
6298       The [6]-fdiagnostics-color=auto will enable it when outputting to
6299       terminals, -fdiagnostics-color=always unconditionally. The
6300       GCC_COLORS environment variable can be used to customize the colors
6301       or disable coloring. If GCC_COLORS variable is present in the
6302       environment, the default is -fdiagnostics-color=auto, otherwise
6303       -fdiagnostics-color=never.
6304       Sample diagnostics output:
6305    $ g++ -fdiagnostics-color=always -S -Wall test.C
6306    test.C: In function `int foo()':
6307    test.C:1:14: warning: no return statement in function returning non-void [-W
6308return-type]
6309     int foo () { }
6310                  ^
6311    test.C:2:46: error: template instantiation depth exceeds maximum of 900 (use
6312 -ftemplate-depth= to increase the maximum) instantiating `struct X<100>'
6313     template <int N> struct X { static const int value = X<N-1>::value; }; temp
6314late struct X<1000>;
6315                                                  ^
6316    test.C:2:46:   recursively required from `const int X<999>::value'
6317    test.C:2:46:   required from `const int X<1000>::value'
6318    test.C:2:88:   required from here
6319
6320    test.C:2:46: error: incomplete type `X<100>' used in nested name specifier
6321
6322     * With the new [7]#pragma GCC ivdep, the user can assert that there
6323       are no loop-carried dependencies which would prevent concurrent
6324       execution of consecutive iterations using SIMD (single instruction
6325       multiple data) instructions.
6326     * Support for Cilk Plus has been added and can be enabled with the
6327       -fcilkplus option. Cilk Plus is an extension to the C and C++
6328       languages to support data and task parallelism. The present
6329       implementation follows ABI version 1.2; all features but _Cilk_for
6330       have been implemented.
6331
6332  C
6333
6334     * ISO C11 atomics (the _Atomic type specifier and qualifier and the
6335       <stdatomic.h> header) are now supported.
6336     * ISO C11 generic selections (_Generic keyword) are now supported.
6337     * ISO C11 thread-local storage (_Thread_local, similar to GNU C
6338       __thread) is now supported.
6339     * ISO C11 support is now at a similar level of completeness to ISO
6340       C99 support: substantially complete modulo bugs, extended
6341       identifiers (supported except for corner cases when
6342       -fextended-identifiers is used), floating-point issues (mainly but
6343       not entirely relating to optional C99 features from Annexes F and
6344       G) and the optional Annexes K (Bounds-checking interfaces) and L
6345       (Analyzability).
6346     * A new C extension __auto_type provides a subset of the
6347       functionality of C++11 auto in GNU C.
6348
6349  C++
6350
6351     * The G++ implementation of [8]C++1y return type deduction for normal
6352       functions has been updated to conform to [9]N3638, the proposal
6353       accepted into the working paper. Most notably, it adds
6354       decltype(auto) for getting decltype semantics rather than the
6355       template argument deduction semantics of plain auto:
6356
6357int& f();
6358         auto  i1 = f(); // int
6359decltype(auto) i2 = f(); // int&
6360
6361     * G++ supports [10]C++1y lambda capture initializers:
6362
6363[x = 42]{ ... };
6364
6365       Actually, they have been accepted since GCC 4.5, but now the
6366       compiler doesn't warn about them with -std=c++1y, and supports
6367       parenthesized and brace-enclosed initializers as well.
6368     * G++ supports [11]C++1y variable length arrays. G++ has supported
6369       GNU/C99-style VLAs for a long time, but now additionally supports
6370       initializers and lambda capture by reference. In C++1y mode G++
6371       will complain about VLA uses that are not permitted by the draft
6372       standard, such as forming a pointer to VLA type or applying sizeof
6373       to a VLA variable. Note that it now appears that VLAs will not be
6374       part of C++14, but will be part of a separate document and then
6375       perhaps C++17.
6376
6377void f(int n) {
6378  int a[n] = { 1, 2, 3 }; // throws std::bad_array_length if n < 3
6379  [&a]{ for (int i : a) { cout << i << endl; } }();
6380  &a; // error, taking address of VLA
6381}
6382
6383     * G++ supports the [12]C++1y [[deprecated]] attribute modulo bugs in
6384       the underlying [[gnu::deprecated]] attribute. Classes and functions
6385       can be marked deprecated and a diagnostic message added:
6386
6387class A;
6388int bar(int n);
6389#if __cplusplus > 201103
6390class [[deprecated("A is deprecated in C++14; Use B instead")]] A;
6391[[deprecated("bar is unsafe; use foo() instead")]]
6392int bar(int n);
6393
6394int foo(int n);
6395class B;
6396#endif
6397A aa; // warning: 'A' is deprecated : A is deprecated in C++14; Use B instead
6398int j = bar(2); // warning: 'int bar(int)' is deprecated : bar is unsafe; use fo
6399o() instead
6400
6401     * G++ supports [13]C++1y digit separators. Long numeric literals can
6402       be subdivided with a single quote ' to enhance readability:
6403
6404int i = 1048576;
6405int j = 1'048'576;
6406int k = 0x10'0000;
6407int m = 0'004'000'000;
6408int n = 0b0001'0000'0000'0000'0000'0000;
6409
6410double x = 1.602'176'565e-19;
6411double y = 1.602'176'565e-1'9;
6412
6413     * G++ supports [14]C++1y generic (polymorphic) lambdas.
6414
6415// a functional object that will increment any type
6416auto incr = [](auto x) { return x++; };
6417
6418     * As a GNU extension, G++ supports explicit template parameter syntax
6419       for generic lambdas. This can be combined in the expected way with
6420       the standard auto syntax.
6421
6422// a functional object that will add two like-type objects
6423auto add = [] <typename T> (T a, T b) { return a + b; };
6424
6425     * G++ supports unconstrained generic functions as specified by �4.1.2
6426       and �5.1.1 of [15]N3889: Concepts Lite Specification. Briefly, auto
6427       may be used as a type-specifier in a parameter declaration of any
6428       function declarator in order to introduce an implicit function
6429       template parameter, akin to generic lambdas.
6430
6431// the following two function declarations are equivalent
6432auto incr(auto x) { return x++; }
6433template <typename T>
6434auto incr(T x) { return x++; }
6435
6436    Runtime Library (libstdc++)
6437
6438     * [16]Improved support for C++11, including:
6439          + support for <regex>;
6440          + The associative containers in <map> and <set> and the
6441            unordered associative containers in <unordered_map> and
6442            <unordered_set> meet the allocator-aware container
6443            requirements;
6444     * [17]Improved experimental support for the upcoming ISO C++
6445       standard, C++14, including:
6446          + fixing constexpr member functions without const;
6447          + implementation of the std::exchange() utility function;
6448          + addressing tuples by type;
6449          + implemention of std::make_unique;
6450          + implemention of std::shared_lock;
6451          + making std::result_of SFINAE-friendly;
6452          + adding operator() to std::integral_constant;
6453          + adding user-defined literals for standard library types
6454            std::basic_string, std::chrono::duration, and std::complex;
6455          + adding two range overloads to non-modifying sequence oprations
6456            std::equal and std::mismatch;
6457          + adding IO manipulators for quoted strings;
6458          + adding constexpr members to <utility>, <complex>, <chrono>,
6459            and some containers;
6460          + adding compile-time std::integer_sequence;
6461          + adding cleaner transformation traits;
6462          + making <functional>s operator functors easier to use and more
6463            generic;
6464     * An implementation of std::experimental::optional.
6465     * An implementation of std::experimental::string_view.
6466     * The non-standard function std::copy_exception has been deprecated
6467       and will be removed in a future version. std::make_exception_ptr
6468       should be used instead.
6469
6470  Fortran
6471
6472     * Compatibility notice:
6473          + Module files: The version of the module files (.mod) has been
6474            incremented; additionally, module files are now compressed.
6475            Fortran MODULEs compiled by earlier GCC versions have to be
6476            recompiled, when they are USEd by files compiled with GCC 4.9.
6477            GCC 4.9 is not able to read .mod files of earlier GCC
6478            versions; attempting to do so gives an error message. Note:
6479            The ABI of the produced assembler data itself has not changed:
6480            object files and libraries are fully compatible with older
6481            versions (except as stated below).
6482          + ABI changes:
6483               o The [18]argument passing ABI has changed for scalar dummy
6484                 arguments of type INTEGER, REAL, COMPLEX and LOGICAL,
6485                 which have both the VALUE and the OPTIONAL attributes.
6486               o To support finalization the virtual table associated with
6487                 polymorphic variables has changed. Code containing CLASS
6488                 should be recompiled, including all files which define
6489                 derived types involved in the type definition used by
6490                 polymorphic variables. (Note: Due to the incremented
6491                 module version, trying to mix old code with new code will
6492                 usually give an error message.)
6493          + GNU Fortran no longer deallocates allocatable variables or
6494            allocatable components of variables declared in the main
6495            program. Since Fortran 2008, the standard explicitly states
6496            that variables declared in the Fortran main program
6497            automatically have the SAVE attribute.
6498          + When opening files, the close-on-exec flag is set if the
6499            system supports such a feature. This is generally considered
6500            good practice these days, but if there is a need to pass file
6501            descriptors to child processes the parent process must now
6502            remember to clear the close-on-exec flag by calling fcntl(),
6503            e.g. via ISO_C_BINDING, before executing the child process.
6504     * The deprecated command-line option -fno-whole-file has been
6505       removed. (-fwhole-file is the default since GCC 4.6.)
6506       -fwhole-file/-fno-whole-file continue to be accepted but do not
6507       influence the code generation.
6508     * The compiler no longer unconditionally warns about DO loops with
6509       zero iterations. This warning is now controlled by the -Wzerotrip
6510       option, which is implied by -Wall.
6511     * The new NO_ARG_CHECK attribute of the [19]!GCC$ directive can be
6512       used to disable the type-kind-rank (TKR) argument check for a dummy
6513       argument. The feature is similar to ISO/IEC TS 29133:2012's
6514       TYPE(*), except that it additionally also disables the rank check.
6515       Variables with NO_ARG_CHECK have to be dummy arguments and may only
6516       be used as argument to ISO_C_BINDING's C_LOC and as actual argument
6517       to another NO_ARG_CHECK dummy argument; also the other constraints
6518       of TYPE(*) apply. The dummy arguments should be declared as scalar
6519       or assumed-size variable of type type(*) (recommended) - or of type
6520       integer, real, complex or logical. With NO_ARG_CHECK, a pointer to
6521       the data without further type or shape information is passed,
6522       similar to C's void*. Note that also TS 29113's
6523       type(*),dimension(..) accepts arguments of any type and rank;
6524       contrary to NO_ARG_CHECK assumed-rank arguments pass an array
6525       descriptor which contains the array shape and stride of the
6526       argument.
6527     * [20]Fortran 2003:
6528          + Finalization is now supported. It is currently only done for a
6529            subset of those situations in which it should occur.
6530          + Experimental support for scalar character components with
6531            deferred length (i.e. allocatable string length) in derived
6532            types has been added. (Deferred-length character variables are
6533            supported since GCC 4.6.)
6534     * [21]Fortran 2008:
6535          + When STOP or ERROR STOP are used to terminate the execution
6536            and any exception (but inexact) is signaling, a warning is
6537            printed to ERROR_UNIT, indicating which exceptions are
6538            signaling. The [22]-ffpe-summary= command-line option can be
6539            used to fine-tune for which exceptions the warning should be
6540            shown.
6541          + Rounding on input (READ) is now handled on systems where
6542            strtod honours the rounding mode. (For output, rounding is
6543            supported since GCC 4.5.) Note that for input, the compatible
6544            rounding mode is handled as nearest (i.e., rounding to an even
6545            least significant [cf. IEC 60559:1989] for a tie, while
6546            compatible rounds away from zero in that case).
6547
6548  Go
6549
6550     * GCC 4.9 provides a complete implementation of the Go 1.2.1 release.
6551
6552New Targets and Target Specific Improvements
6553
6554  AArch64
6555
6556     * The ARMv8-A crypto and CRC instructions are now supported through
6557       intrinsics. These are enabled when the architecture supports these
6558       and are available through the -march=armv8-a+crc and
6559       -march=armv8-a+crypto options.
6560     * Initial support for ILP32 has now been added to the compiler. This
6561       is now available through the command-line option -mabi=ilp32.
6562       Support for ILP32 is considered experimental as the ABI
6563       specification is still beta.
6564     * Coverage of more of the ISA including the SIMD extensions has been
6565       added. The Advanced SIMD intrinsics have also been improved.
6566     * The new local register allocator (LRA) is now on by default for the
6567       AArch64 backend.
6568     * The REE (Redundant extension elimination) pass has now been enabled
6569       by default for the AArch64 backend.
6570     * Tuning for the Cortex-A53 and Cortex-A57 has been improved.
6571     * Initial big.LITTLE tuning support for the combination of Cortex-A57
6572       and Cortex-A53 was added through the -mcpu=cortex-a57.cortex-a53
6573       option.
6574     * A number of structural changes have been made to both the ARM and
6575       AArch64 backends to facilitate improved code-generation.
6576     * As of GCC 4.9.2 a workaround for the ARM Cortex-A53 erratum 835769
6577       has been added and can be enabled by giving the
6578       -mfix-cortex-a53-835769 option. Alternatively it can be enabled by
6579       default by configuring GCC with the --enable-fix-cortex-a53-835769
6580       option.
6581
6582  ARC
6583
6584     * A port for Synopsys Designware ARC has been contributed by Embecosm
6585       and Synopsys Inc.
6586
6587  ARM
6588
6589     * Use of Advanced SIMD (Neon) for 64-bit scalar computations has been
6590       disabled by default. This was found to generate better code in only
6591       a small number of cases. It can be turned back on with the
6592       -mneon-for-64bits option.
6593     * Further support for the ARMv8-A architecture, notably implementing
6594       the restriction around IT blocks in the Thumb32 instruction set has
6595       been added. The -mrestrict-it option can be used with
6596       -march=armv7-a or the -march=armv7ve options to make code
6597       generation fully compatible with the deprecated instructions in
6598       ARMv8-A.
6599     * Support has now been added for the ARMv7ve variant of the
6600       architecture. This can be used by the -march=armv7ve option.
6601     * The ARMv8-A crypto and CRC instructions are now supported through
6602       intrinsics and are available through the -march=armv8-a+crc and
6603       mfpu=crypto-neon-fp-armv8 options.
6604     * LRA is now on by default for the ARM target. This can be turned off
6605       using the -mno-lra option. This option is a purely transitionary
6606       command-line option and will be removed in a future release. We are
6607       interested in any bug reports regarding functional and performance
6608       regressions with LRA.
6609     * A new option -mslow-flash-data to improve performance of programs
6610       fetching data on slow flash memory has now been introduced for the
6611       ARMv7-M profile cores.
6612     * A new option -mpic-data-is-text-relative for targets that allows
6613       data segments to be relative to text segments has been added. This
6614       is on by default for all targets except VxWorks RTP.
6615     * A number of infrastructural changes have been made to both the ARM
6616       and AArch64 backends to facilitate improved code-generation.
6617     * GCC now supports Cortex-A12 and the Cortex-R7 through the
6618       -mcpu=cortex-a12 and -mcpu=cortex-r7 options.
6619     * GCC now has tuning for the Cortex-A57 and Cortex-A53 through the
6620       -mcpu=cortex-a57 and -mcpu=cortex-a53 options.
6621     * Initial big.LITTLE tuning support for the combination of Cortex-A57
6622       and Cortex-A53 was added through the -mcpu=cortex-a57.cortex-a53
6623       option. Similar support was added for the combination of Cortex-A15
6624       and Cortex-A7 through the -mcpu=cortex-a15.cortex-a7 option.
6625     * Further performance optimizations for the Cortex-A15 and the
6626       Cortex-M4 have been added.
6627     * A number of code generation improvements for Thumb2 to reduce code
6628       size when compiling for the M-profile processors.
6629
6630  AVR
6631
6632     * A new command-line option -mfract-convert-truncate has been added.
6633       It allows compiler to use truncation instead of rounding towards
6634       zero for fractional fixed-point types.
6635
6636  IA-32/x86-64
6637
6638     * -mfpmath=sse is now implied by -ffast-math on all targets where
6639       SSE2 is supported.
6640     * Intel AVX-512 support was added to GCC. That includes inline
6641       assembly support, new registers and extending existing ones, new
6642       intrinsics (covered by corresponding testsuite), and basic
6643       autovectorization. AVX-512 instructions are available via the
6644       following GCC switches: AVX-512 foundation instructions: -mavx512f,
6645       AVX-512 prefetch instructions: -mavx512pf, AVX-512 exponential and
6646       reciprocal instructions: -mavx512er, AVX-512 conflict detection
6647       instructions: -mavx512cd.
6648     * It is now possible to call x86 intrinsics from select functions in
6649       a file that are tagged with the corresponding target attribute
6650       without having to compile the entire file with the -mxxx option.
6651       This improves the usability of x86 intrinsics and is particularly
6652       useful when doing [23]Function Multiversioning.
6653     * GCC now supports the new Intel microarchitecture named Silvermont
6654       through -march=silvermont.
6655     * GCC now supports the new Intel microarchitecture named Broadwell
6656       through -march=broadwell.
6657     * Optimizing for other Intel microarchitectures have been renamed to
6658       -march=nehalem, westmere, sandybridge, ivybridge, haswell, bonnell.
6659     * -march=generic has been retuned for better support of Intel core
6660       and AMD Bulldozer architectures. Performance of AMD K7, K8, Intel
6661       Pentium-M, and Pentium4 based CPUs is no longer considered
6662       important for generic.
6663     * -mtune=intel can now be used to generate code running well on the
6664       most current Intel processors, which are Haswell and Silvermont for
6665       GCC 4.9.
6666     * Support to encode 32-bit assembly instructions in 16-bit format is
6667       now available through the -m16 command-line option.
6668     * Better inlining of memcpy and memset that is aware of value ranges
6669       and produces shorter alignment prologues.
6670     * -mno-accumulate-outgoing-args is now honored when unwind
6671       information is output. Argument accumulation is also now turned off
6672       for portions of programs optimized for size.
6673     * Support for new AMD family 15h processors (Excavator core) is now
6674       available through the -march=bdver4 and -mtune=bdver4 options.
6675
6676  MSP430
6677
6678     * A new command-line option -mcpu= has been added to the MSP430
6679       backend. This option is used to specify the ISA to be used.
6680       Accepted values are msp430 (the default), msp430x and msp430xv2.
6681       The ISA is no longer deduced from the -mmcu= option as there are
6682       far too many different MCU names. The -mmcu= option is still
6683       supported, and this is still used to select linker scripts and
6684       generate a C preprocessor symbol that will be recognised by the
6685       msp430.h header file.
6686
6687  NDS32
6688
6689     * A new nds32 port supports the 32-bit architecture from Andes
6690       Technology Corporation.
6691     * The port provides initial support for the V2, V3, V3m instruction
6692       set architectures.
6693
6694  Nios II
6695
6696     * A port for the Altera Nios II has been contributed by Mentor
6697       Graphics.
6698
6699  PowerPC / PowerPC64 / RS6000
6700
6701     * GCC now supports Power ISA 2.07, which includes support for
6702       Hardware Transactional Memory (HTM), Quadword atomics and several
6703       VMX and VSX additions, including Crypto, 64-bit integer, 128-bit
6704       integer and decimal integer operations.
6705     * Support for the POWER8 processor is now available through the
6706       -mcpu=power8 and -mtune=power8 options.
6707     * The libitm library has been modified to add a HTM fastpath that
6708       automatically uses POWER's HTM hardware instructions when it is
6709       executing on a HTM enabled processor.
6710     * Support for the new powerpc64le-linux platform has been added. It
6711       defaults to generating code that conforms to the ELFV2 ABI.
6712
6713  S/390, System z
6714
6715     * Support for the Transactional Execution Facility included with the
6716       IBM zEnterprise zEC12 processor has been added. A set of GCC style
6717       builtins as well as XLC style builtins are provided. The builtins
6718       are enabled by default when using the -march=zEC12 option but can
6719       explicitly be disabled with -mno-htm. Using the GCC builtins also
6720       libitm supports hardware transactions on S/390.
6721     * The hotpatch features allows to prepare functions for hotpatching.
6722       A certain amount of bytes is reserved before the function entry
6723       label plus a NOP is inserted at its very beginning to implement a
6724       backward jump when applying a patch. The feature can either be
6725       enabled per compilation unit via the command-line option -mhotpatch
6726       or per function using the hotpatch attribute.
6727     * The shrink wrap optimization is now supported on S/390 and enabled
6728       by default.
6729     * A major rework of the routines to determine which registers need to
6730       be saved and restored in function prologue/epilogue now allow to
6731       use floating point registers as save slots. This will happen for
6732       certain leaf function with -march=z10 or higher.
6733     * The LRA rtl pass replaces reload by default on S/390.
6734
6735  RX
6736
6737     * The port now allows to specify the RX100, RX200, and RX600
6738       processors with the command-line options -mcpu=rx100, -mcpu=rx200
6739       and -mcpu=rx600.
6740
6741  SH
6742
6743     * Minor improvements to code generated for integer arithmetic and
6744       code that involves the T bit.
6745     * Added support for the SH2A clips and clipu instructions. The
6746       compiler will now try to utilize them for min/max expressions such
6747       as max (-128, min (127, x)).
6748     * Added support for the cmp/str instruction through built-in
6749       functions such as __builtin_strlen. When not optimizing for size,
6750       the compiler will now expand calls to e.g. strlen as an inlined
6751       sequences which utilize the cmp/str instruction.
6752     * Improved code generated around volatile memory loads and stores.
6753     * The option -mcbranchdi has been deprecated. Specifying it will
6754       result in a warning and will not influence code generation.
6755     * The option -mcmpeqdi has been deprecated. Specifying it will result
6756       in a warning and will not influence code generation.
6757
6758GCC 4.9.1
6759
6760   This is the [24]list of problem reports (PRs) from GCC's bug tracking
6761   system that are known to be fixed in the 4.9.1 release. This list might
6762   not be complete (that is, it is possible that some PRs that have been
6763   fixed are not listed here).
6764
6765   Version 4.0 of the OpenMP specification is supported even in Fortran,
6766   not just C and C++.
6767
6768GCC 4.9.2
6769
6770   This is the [25]list of problem reports (PRs) from GCC's bug tracking
6771   system that are known to be fixed in the 4.9.2 release. This list might
6772   not be complete (that is, it is possible that some PRs that have been
6773   fixed are not listed here).
6774
6775GCC 4.9.3
6776
6777   This is the [26]list of problem reports (PRs) from GCC's bug tracking
6778   system that are known to be fixed in the 4.9.3 release. This list might
6779   not be complete (that is, it is possible that some PRs that have been
6780   fixed are not listed here).
6781
6782GCC 4.9.4
6783
6784   This is the [27]list of problem reports (PRs) from GCC's bug tracking
6785   system that are known to be fixed in the 4.9.4 release. This list might
6786   not be complete (that is, it is possible that some PRs that have been
6787   fixed are not listed here).
6788
6789
6790    For questions related to the use of GCC, please consult these web
6791    pages and the [28]GCC manuals. If that fails, the
6792    [29]gcc-help@gcc.gnu.org mailing list might help. Comments on these
6793    web pages and the development of GCC are welcome on our developer
6794    list at [30]gcc@gcc.gnu.org. All of [31]our lists have public
6795    archives.
6796
6797   Copyright (C) [32]Free Software Foundation, Inc. Verbatim copying and
6798   distribution of this entire article is permitted in any medium,
6799   provided this notice is preserved.
6800
6801   These pages are [33]maintained by the GCC team. Last modified
6802   2021-03-31[34].
6803
6804References
6805
6806   1. https://gcc.gnu.org/ml/gcc-patches/2013-05/msg00728.html
6807   2. https://gcc.gnu.org/PR60825
6808   3. https://gcc.gnu.org/gcc-4.9/porting_to.html
6809   4. https://www.openmp.org/specifications/
6810   5. https://gcc.gnu.org/onlinedocs/gcc-4.9.2/gcc/Optimize-Options.html#index-fsimd-cost-model-908
6811   6. https://gcc.gnu.org/onlinedocs/gcc-4.9.2/gcc/Language-Independent-Options.html#index-fdiagnostics-color-252
6812   7. https://gcc.gnu.org/onlinedocs/gcc-4.9.2/gcc/Loop-Specific-Pragmas.html
6813   8. http://gcc.gnu.org/projects/cxx1y.html
6814   9. http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2013/n3638.html
6815  10. http://gcc.gnu.org/projects/cxx1y.html
6816  11. http://gcc.gnu.org/projects/cxx1y.html
6817  12. http://gcc.gnu.org/projects/cxx1y.html
6818  13. http://gcc.gnu.org/projects/cxx1y.html
6819  14. http://gcc.gnu.org/projects/cxx1y.html
6820  15. http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n3889.pdf
6821  16. https://gcc.gnu.org/onlinedocs/gcc-4.9.2/libstdc++/manual/manual/status.html#status.iso.2011
6822  17. https://gcc.gnu.org/onlinedocs/libstdc++/manual/status.html#status.iso.2014
6823  18. https://gcc.gnu.org/onlinedocs/gcc-4.9.2/gfortran/Argument-passing-conventions.html
6824  19. https://gcc.gnu.org/onlinedocs/gcc-4.9.2/gfortran/GNU-Fortran-Compiler-Directives.html
6825  20. https://gcc.gnu.org/wiki/Fortran2003Status
6826  21. https://gcc.gnu.org/wiki/Fortran2008Status
6827  22. https://gcc.gnu.org/onlinedocs/gcc-4.9.2/gfortran/Debugging-Options.html
6828  23. https://gcc.gnu.org/onlinedocs/gcc-4.9.2/gcc/Function-Multiversioning.html
6829  24. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.9.1
6830  25. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.9.2
6831  26. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.9.3
6832  27. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.9.4
6833  28. https://gcc.gnu.org/onlinedocs/
6834  29. mailto:gcc-help@gcc.gnu.org
6835  30. mailto:gcc@gcc.gnu.org
6836  31. https://gcc.gnu.org/lists.html
6837  32. https://www.fsf.org/
6838  33. https://gcc.gnu.org/about.html
6839  34. http://validator.w3.org/check/referer
6840======================================================================
6841http://gcc.gnu.org/gcc-4.8/index.html
6842
6843                             GCC 4.8 Release Series
6844
6845   (This release series is no longer supported.)
6846
6847   June 23, 2015
6848
6849   The [1]GNU project and the GCC developers are pleased to announce the
6850   release of GCC 4.8.5.
6851
6852   This release is a bug-fix release, containing fixes for regressions in
6853   GCC 4.8.4 relative to previous releases of GCC.
6854
6855Release History
6856
6857   GCC 4.8.5
6858          June 23, 2015 ([2]changes, [3]documentation)
6859
6860   GCC 4.8.4
6861          December 19, 2014 ([4]changes, [5]documentation)
6862
6863   GCC 4.8.3
6864          May 22, 2014 ([6]changes, [7]documentation)
6865
6866   GCC 4.8.2
6867          October 16, 2013 ([8]changes, [9]documentation)
6868
6869   GCC 4.8.1
6870          May 31, 2013 ([10]changes, [11]documentation)
6871
6872   GCC 4.8.0
6873          March 22, 2013 ([12]changes, [13]documentation)
6874
6875References and Acknowledgements
6876
6877   GCC used to stand for the GNU C Compiler, but since the compiler
6878   supports several other languages aside from C, it now stands for the
6879   GNU Compiler Collection.
6880
6881   A list of [14]successful builds is updated as new information becomes
6882   available.
6883
6884   The GCC developers would like to thank the numerous people that have
6885   contributed new features, improvements, bug fixes, and other changes as
6886   well as test results to GCC. This [15]amazing group of volunteers is
6887   what makes GCC successful.
6888
6889   For additional information about GCC please refer to the [16]GCC
6890   project web site or contact the [17]GCC development mailing list.
6891
6892   To obtain GCC please use [18]our mirror sites or [19]our version
6893   control system.
6894
6895
6896    For questions related to the use of GCC, please consult these web
6897    pages and the [20]GCC manuals. If that fails, the
6898    [21]gcc-help@gcc.gnu.org mailing list might help. Comments on these
6899    web pages and the development of GCC are welcome on our developer
6900    list at [22]gcc@gcc.gnu.org. All of [23]our lists have public
6901    archives.
6902
6903   Copyright (C) [24]Free Software Foundation, Inc. Verbatim copying and
6904   distribution of this entire article is permitted in any medium,
6905   provided this notice is preserved.
6906
6907   These pages are [25]maintained by the GCC team. Last modified
6908   2020-01-14[26].
6909
6910References
6911
6912   1. http://www.gnu.org/
6913   2. http://gcc.gnu.org/gcc-4.8/changes.html
6914   3. https://gcc.gnu.org/onlinedocs/4.8.5/
6915   4. http://gcc.gnu.org/gcc-4.8/changes.html
6916   5. https://gcc.gnu.org/onlinedocs/4.8.4/
6917   6. http://gcc.gnu.org/gcc-4.8/changes.html
6918   7. https://gcc.gnu.org/onlinedocs/4.8.3/
6919   8. http://gcc.gnu.org/gcc-4.8/changes.html
6920   9. https://gcc.gnu.org/onlinedocs/4.8.2/
6921  10. http://gcc.gnu.org/gcc-4.8/changes.html
6922  11. https://gcc.gnu.org/onlinedocs/4.8.1/
6923  12. http://gcc.gnu.org/gcc-4.8/changes.html
6924  13. https://gcc.gnu.org/onlinedocs/4.8.0/
6925  14. http://gcc.gnu.org/gcc-4.8/buildstat.html
6926  15. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
6927  16. http://gcc.gnu.org/index.html
6928  17. mailto:gcc@gcc.gnu.org
6929  18. http://gcc.gnu.org/mirrors.html
6930  19. http://gcc.gnu.org/git.html
6931  20. https://gcc.gnu.org/onlinedocs/
6932  21. mailto:gcc-help@gcc.gnu.org
6933  22. mailto:gcc@gcc.gnu.org
6934  23. https://gcc.gnu.org/lists.html
6935  24. https://www.fsf.org/
6936  25. https://gcc.gnu.org/about.html
6937  26. http://validator.w3.org/check/referer
6938======================================================================
6939http://gcc.gnu.org/gcc-4.8/changes.html
6940
6941                             GCC 4.8 Release Series
6942                        Changes, New Features, and Fixes
6943
6944Caveats
6945
6946   GCC now uses C++ as its implementation language. This means that to
6947   build GCC from sources, you will need a C++ compiler that understands
6948   C++ 2003. For more details on the rationale and specific changes,
6949   please refer to the [1]C++ conversion page.
6950
6951   To enable the Graphite framework for loop optimizations you now need
6952   CLooG version 0.18.0 and ISL version 0.11.1. Both can be obtained from
6953   the [2]GCC infrastructure directory. The installation manual contains
6954   more information about requirements to build GCC.
6955
6956   GCC now uses a more aggressive analysis to derive an upper bound for
6957   the number of iterations of loops using constraints imposed by language
6958   standards. This may cause non-conforming programs to no longer work as
6959   expected, such as SPEC CPU 2006 464.h264ref and 416.gamess. A new
6960   option, -fno-aggressive-loop-optimizations, was added to disable this
6961   aggressive analysis. In some loops that have known constant number of
6962   iterations, but undefined behavior is known to occur in the loop before
6963   reaching or during the last iteration, GCC will warn about the
6964   undefined behavior in the loop instead of deriving lower upper bound of
6965   the number of iterations for the loop. The warning can be disabled with
6966   -Wno-aggressive-loop-optimizations.
6967
6968   On ARM, a bug has been fixed in GCC's implementation of the AAPCS rules
6969   for the layout of vectors that could lead to wrong code being
6970   generated. Vectors larger than 8 bytes in size are now by default
6971   aligned to an 8-byte boundary. This is an ABI change: code that makes
6972   explicit use of vector types may be incompatible with binary objects
6973   built with older versions of GCC. Auto-vectorized code is not affected
6974   by this change.
6975
6976   On AVR, support has been removed for the command-line option
6977   -mshort-calls deprecated in GCC 4.7.
6978
6979   On AVR, the configure option --with-avrlibc supported since GCC 4.7.2
6980   is turned on per default for all non-RTEMS configurations. This option
6981   arranges for a better integration of [3]AVR Libc with avr-gcc. For
6982   technical details, see [4]PR54461. To turn off the option in non-RTEMS
6983   configurations, use --with-avrlibc=no. If the compiler is configured
6984   for RTEMS, the option is always turned off.
6985
6986   More information on porting to GCC 4.8 from previous versions of GCC
6987   can be found in the [5]porting guide for this release.
6988
6989General Optimizer Improvements (and Changes)
6990
6991     * DWARF4 is now the default when generating DWARF debug information.
6992       When -g is used on a platform that uses DWARF debugging
6993       information, GCC will now default to -gdwarf-4
6994       -fno-debug-types-section.
6995       GDB 7.5, Valgrind 3.8.0 and elfutils 0.154 debug information
6996       consumers support DWARF4 by default. Before GCC 4.8 the default
6997       version used was DWARF2. To make GCC 4.8 generate an older DWARF
6998       version use -g together with -gdwarf-2 or -gdwarf-3. The default
6999       for Darwin and VxWorks is still -gdwarf-2 -gstrict-dwarf.
7000     * A new general optimization level, -Og, has been introduced. It
7001       addresses the need for fast compilation and a superior debugging
7002       experience while providing a reasonable level of run-time
7003       performance. Overall experience for development should be better
7004       than the default optimization level -O0.
7005     * A new option -ftree-partial-pre was added to control the partial
7006       redundancy elimination (PRE) optimization. This option is enabled
7007       by default at the -O3 optimization level, and it makes PRE more
7008       aggressive.
7009     * The option -fconserve-space has been removed; it was no longer
7010       useful on most targets since GCC supports putting variables into
7011       BSS without making them common.
7012     * The struct reorg and matrix reorg optimizations (command-line
7013       options -fipa-struct-reorg and -fipa-matrix-reorg) have been
7014       removed. They did not always work correctly, nor did they work with
7015       link-time optimization (LTO), hence were only applicable to
7016       programs consisting of a single translation unit.
7017     * Several scalability bottle-necks have been removed from GCC's
7018       optimization passes. Compilation of extremely large functions, e.g.
7019       due to the use of the flatten attribute in the "Eigen" C++ linear
7020       algebra templates library, is significantly faster than previous
7021       releases of GCC.
7022     * Link-time optimization (LTO) improvements:
7023          + LTO partitioning has been rewritten for better reliability and
7024            maintanibility. Several important bugs leading to link
7025            failures have been fixed.
7026     * Interprocedural optimization improvements:
7027          + A new symbol table has been implemented. It builds on existing
7028            callgraph and varpool modules and provide a new API. Unusual
7029            symbol visibilities and aliases are handled more consistently
7030            leading to, for example, more aggressive unreachable code
7031            removal with LTO.
7032          + The inline heuristic can now bypass limits on the size of of
7033            inlined functions when the inlining is particularly
7034            profitable. This happens, for example, when loop bounds or
7035            array strides get propagated.
7036          + Values passed through aggregates (either by value or
7037            reference) are now propagated at the inter-procedural level
7038            leading to better inlining decisions (for example in the case
7039            of Fortran array descriptors) and devirtualization.
7040     * [6]AddressSanitizer , a fast memory error detector, has been added
7041       and can be enabled via -fsanitize=address. Memory access
7042       instructions will be instrumented to detect heap-, stack-, and
7043       global-buffer overflow as well as use-after-free bugs. To get nicer
7044       stacktraces, use -fno-omit-frame-pointer. The AddressSanitizer is
7045       available on IA-32/x86-64/x32/PowerPC/PowerPC64 GNU/Linux and on
7046       x86-64 Darwin.
7047     * [7]ThreadSanitizer has been added and can be enabled via
7048       -fsanitize=thread. Instructions will be instrumented to detect data
7049       races. The ThreadSanitizer is available on x86-64 GNU/Linux.
7050     * A new local register allocator (LRA) has been implemented, which
7051       replaces the 26 year old reload pass and improves generated code
7052       quality. For now it is active on the IA-32 and x86-64 targets.
7053     * Support for transactional memory has been implemented on the
7054       following architectures: IA-32/x86-64, ARM, PowerPC, SH, SPARC, and
7055       Alpha.
7056
7057New Languages and Language specific improvements
7058
7059  C family
7060
7061     * Each diagnostic emitted now includes the original source line and a
7062       caret '^' indicating the column. The option
7063       -fno-diagnostics-show-caret suppresses this information.
7064     * The option -ftrack-macro-expansion=2 is now enabled by default.
7065       This allows the compiler to display the macro expansion stack in
7066       diagnostics. Combined with the caret information, an example
7067       diagnostic showing these two features is:
7068
7069t.c:1:94: error: invalid operands to binary < (have `struct mystruct' and `float
7070')
7071 #define MYMAX(A,B)    __extension__ ({ __typeof__(A) __a = (A); __typeof__(B) _
7072_b = (B); __a < __b ? __b : __a; })
7073
7074              ^
7075t.c:7:7: note: in expansion of macro 'MYMAX'
7076   X = MYMAX(P, F);
7077       ^
7078
7079     * A new -Wsizeof-pointer-memaccess warning has been added (also
7080       enabled by -Wall) to warn about suspicious length parameters to
7081       certain string and memory built-in functions if the argument uses
7082       sizeof. This warning warns e.g. about memset (ptr, 0, sizeof
7083       (ptr)); if ptr is not an array, but a pointer, and suggests a
7084       possible fix, or about memcpy (&foo, ptr, sizeof (&foo));.
7085     * The new option -Wpedantic is an alias for -pedantic, which is now
7086       deprecated. The forms -Wno-pedantic, -Werror=pedantic, and
7087       -Wno-error=pedantic work in the same way as for any other -W
7088       option. One caveat is that -Werror=pedantic is not equivalent to
7089       -pedantic-errors, since the latter makes into errors some warnings
7090       that are not controlled by -Wpedantic, and the former only affects
7091       diagnostics that are disabled when using -Wno-pedantic.
7092     * The option -Wshadow no longer warns if a declaration shadows a
7093       function declaration, unless the former declares a function or
7094       pointer to function, because this is [8]a common and valid case in
7095       real-world code.
7096
7097  C++
7098
7099     * G++ now implements the [9]C++11 thread_local keyword; this differs
7100       from the GNU __thread keyword primarily in that it allows dynamic
7101       initialization and destruction semantics. Unfortunately, this
7102       support requires a run-time penalty for references to
7103       non-function-local thread_local variables defined in a different
7104       translation unit even if they don't need dynamic initialization, so
7105       users may want to continue to use __thread for TLS variables with
7106       static initialization semantics.
7107       If the programmer can be sure that no use of the variable in a
7108       non-defining TU needs to trigger dynamic initialization (either
7109       because the variable is statically initialized, or a use of the
7110       variable in the defining TU will be executed before any uses in
7111       another TU), they can avoid this overhead with the
7112       -fno-extern-tls-init option.
7113       OpenMP threadprivate variables now also support dynamic
7114       initialization and destruction by the same mechanism.
7115     * G++ now implements the [10]C++11 attribute syntax, e.g.
7116
7117[[noreturn]] void f();
7118
7119       and also the alignment specifier, e.g.
7120
7121alignas(double) int i;
7122
7123     * G++ now implements [11]C++11 inheriting constructors, e.g.
7124
7125struct A { A(int); };
7126struct B: A { using A::A; }; // defines B::B(int)
7127B b(42); // OK
7128
7129     * As of GCC 4.8.1, G++ implements the change to decltype semantics
7130       from [12]N3276.
7131
7132struct A f();
7133decltype(f()) g();    // OK, return type of f() is not required to be complete.
7134
7135     * As of GCC 4.8.1, G++ implements [13]C++11 ref-qualifiers, e.g.
7136
7137struct A { int f() &; };
7138int i = A().f();  // error, f() requires an lvalue object
7139
7140     * G++ now supports a -std=c++1y option for experimentation with
7141       features proposed for the next revision of the standard, expected
7142       around 2014. Currently the only difference from -std=c++11 is
7143       support for return type deduction in normal functions, as proposed
7144       in [14]N3386. Status of C++1y features in GCC 4.8 can be found
7145       [15]here.
7146     * The G++ namespace association extension, __attribute ((strong)),
7147       has been deprecated. Inline namespaces should be used instead.
7148     * G++ now supports a -fext-numeric-literal option to control whether
7149       GNU numeric literal suffixes are accepted as extensions or
7150       processed as C++11 user-defined numeric literal suffixes. The flag
7151       is on (use suffixes for GNU literals) by default for -std=gnu++*,
7152       and -std=c++98. The flag is off (use suffixes for user-defined
7153       literals) by default for -std=c++11 and later.
7154
7155    Runtime Library (libstdc++)
7156
7157     * [16]Improved experimental support for the new ISO C++ standard,
7158       C++11, including:
7159          + forward_list meets the allocator-aware container requirements;
7160          + this_thread::sleep_for(), this_thread::sleep_until() and
7161            this_thread::yield() are defined without requiring the
7162            configure option --enable-libstdcxx-time;
7163     * Improvements to <random>:
7164          + SSE optimized normal_distribution.
7165          + Use of hardware RNG instruction for random_device on new x86
7166            processors (requires the assembler to support the
7167            instruction.)
7168       and <ext/random>:
7169          + New random number engine simd_fast_mersenne_twister_engine
7170            with an optimized SSE implementation.
7171          + New random number distributions beta_distribution,
7172            normal_mv_distribution, rice_distribution,
7173            nakagami_distribution, pareto_distribution, k_distribution,
7174            arcsine_distribution, hoyt_distribution.
7175     * Added --disable-libstdcxx-verbose configure option to disable
7176       diagnostic messages issued when a process terminates abnormally.
7177       This may be useful for embedded systems to reduce the size of
7178       executables that link statically to the library.
7179
7180  Fortran
7181
7182     * Compatibility notice:
7183          + Module files: The version of module files (.mod) has been
7184            incremented. Fortran MODULEs compiled by earlier GCC versions
7185            have to be recompiled, when they are USEd by files compiled
7186            with GCC 4.8. GCC 4.8 is not able to read .mod files created
7187            by earlier versions; attempting to do so gives an error
7188            message.
7189            Note: The ABI of the produced assembler data itself has not
7190            changed; object files and libraries are fully compatible with
7191            older versions except as noted below.
7192          + ABI: Some internal names (used in the assembler/object file)
7193            have changed for symbols declared in the specification part of
7194            a module. If an affected module - or a file using it via use
7195            association - is recompiled, the module and all files which
7196            directly use such symbols have to be recompiled as well. This
7197            change only affects the following kind of module symbols:
7198               o Procedure pointers. Note: C-interoperable function
7199                 pointers (type(c_funptr)) are not affected nor are
7200                 procedure-pointer components.
7201               o Deferred-length character strings.
7202     * The [17]BACKTRACE intrinsic subroutine has been added. It shows a
7203       backtrace at an arbitrary place in user code; program execution
7204       continues normally afterwards.
7205     * The [18]-Wc-binding-type warning option has been added (disabled by
7206       default). It warns if the a variable might not be C interoperable;
7207       in particular, if the variable has been declared using an intrinsic
7208       type with default kind instead of using a kind parameter defined
7209       for C interoperability in the intrinsic ISO_C_Binding module.
7210       Before, this warning was always printed. The -Wc-binding-type
7211       option is enabled by -Wall.
7212     * The [19]-Wrealloc-lhs and -Wrealloc-lhs-all warning command-line
7213       options have been added, which diagnose when code is inserted for
7214       automatic (re)allocation of a variable during assignment. This
7215       option can be used to decide whether it is safe to use
7216       [20]-fno-realloc-lhs. Additionally, it can be used to find
7217       automatic (re)allocation in hot loops. (For arrays, replacing
7218       "var=" by "var(:)=" disables the automatic reallocation.)
7219     * The [21]-Wcompare-reals command-line option has been added. When
7220       this is set, warnings are issued when comparing REAL or COMPLEX
7221       types for equality and inequality; consider replacing a == b by
7222       abs(a-b) < eps with a suitable eps. -Wcompare-reals is enabled by
7223       -Wextra.
7224     * The [22]-Wtarget-lifetime command-line option has been added
7225       (enabled with -Wall), which warns if the pointer in a pointer
7226       assignment might outlive its target.
7227     * Reading floating point numbers which use "q" for the exponential
7228       (such as 4.0q0) is now supported as vendor extension for better
7229       compatibility with old data files. It is strongly recommended to
7230       use for I/O the equivalent but standard conforming "e" (such as
7231       4.0e0).
7232       (For Fortran source code, consider replacing the "q" in
7233       floating-point literals by a kind parameter (e.g. 4.0e0_qp with a
7234       suitable qp). Note that - in Fortran source code - replacing "q" by
7235       a simple "e" is not equivalent.)
7236     * The GFORTRAN_TMPDIR environment variable for specifying a
7237       non-default directory for files opened with STATUS="SCRATCH", is
7238       not used anymore. Instead gfortran checks the POSIX/GNU standard
7239       TMPDIR environment variable. If TMPDIR is not defined, gfortran
7240       falls back to other methods to determine the directory for
7241       temporary files as documented in the [23]user manual.
7242     * [24]Fortran 2003:
7243          + Support for unlimited polymorphic variables (CLASS(*)) has
7244            been added. Nonconstant character lengths are not yet
7245            supported.
7246     * [25]TS 29113:
7247          + Assumed types (TYPE(*)) are now supported.
7248          + Experimental support for assumed-rank arrays (dimension(..))
7249            has been added. Note that currently gfortran's own array
7250            descriptor is used, which is different from the one defined in
7251            TS29113, see [26]gfortran's header file or use the [27]Chasm
7252            Language Interoperability Tools.
7253
7254  Go
7255
7256     * GCC 4.8.2 provides a complete implementation of the Go 1.1.2
7257       release.
7258     * GCC 4.8.0 and 4.8.1 implement a preliminary version of the Go 1.1
7259       release. The library support is not quite complete.
7260     * Go has been tested on GNU/Linux and Solaris platforms for various
7261       processors including x86, x86_64, PowerPC, SPARC, and Alpha. It may
7262       work on other platforms as well.
7263
7264New Targets and Target Specific Improvements
7265
7266  AArch64
7267
7268     * A new port has been added to support AArch64, the new 64-bit
7269       architecture from ARM. Note that this is a separate port from the
7270       existing 32-bit ARM port.
7271     * The port provides initial support for the Cortex-A53 and the
7272       Cortex-A57 processors with the command line options
7273       -mcpu=cortex-a53 and -mcpu=cortex-a57.
7274     * As of GCC 4.8.4 a workaround for the ARM Cortex-A53 erratum 835769
7275       has been added and can be enabled by giving the
7276       -mfix-cortex-a53-835769 option. Alternatively it can be enabled by
7277       default by configuring GCC with the --enable-fix-cortex-a53-835769
7278       option.
7279
7280  ARM
7281
7282     * Initial support has been added for the AArch32 extensions defined
7283       in the ARMv8 architecture.
7284     * Code generation improvements for the Cortex-A7 and Cortex-A15 CPUs.
7285     * A new option, -mcpu=marvell-pj4, has been added to generate code
7286       for the Marvell PJ4 processor.
7287     * The compiler can now automatically generate the VFMA, VFMS, REVSH
7288       and REV16 instructions.
7289     * A new vectorizer cost model for Advanced SIMD configurations to
7290       improve the auto-vectorization strategies used.
7291     * The scheduler now takes into account the number of live registers
7292       to reduce the amount of spilling that can occur. This should
7293       improve code performance in large functions. The limit can be
7294       removed by using the option -fno-sched-pressure.
7295     * Improvements have been made to the Marvell iWMMX code generation
7296       and support for the iWMMX2 SIMD unit has been added. The option
7297       -mcpu=iwmmxt2 can be used to enable code generation for the latter.
7298     * A number of code generation improvements for Thumb2 to reduce code
7299       size when compiling for the M-profile processors.
7300     * The RTEMS (arm-rtems) port has been updated to use the EABI.
7301     * Code generation support for the old FPA and Maverick floating-point
7302       architectures has been removed. Ports that previously relied on
7303       these features have also been removed. This includes the targets:
7304          + arm*-*-linux-gnu (use arm*-*-linux-gnueabi)
7305          + arm*-*-elf (use arm*-*-eabi)
7306          + arm*-*-uclinux* (use arm*-*-uclinux*eabi)
7307          + arm*-*-ecos-elf (no alternative)
7308          + arm*-*-freebsd (no alternative)
7309          + arm*-wince-pe* (no alternative).
7310
7311  AVR
7312
7313     * Support for the "Embedded C" fixed-point has been added. For
7314       details, see the [28]GCC wiki and the [29]user manual. The support
7315       is not complete.
7316     * A new print modifier %r for register operands in inline assembler
7317       is supported. It will print the raw register number without the
7318       register prefix 'r':
7319    /* Return the most significant byte of 'val', a 64-bit value.  */
7320
7321    unsigned char msb (long long val)
7322    {
7323      unsigned char c;
7324      __asm__ ("mov %0, %r1+7" : "=r" (c) : "r" (val));
7325      return c;
7326    }
7327       The inline assembler in this example will generate code like
7328    mov r24, 8+7
7329       provided c is allocated to R24 and val is allocated to R8...R15.
7330       This works because the GNU assembler accepts plain register numbers
7331       without register prefix.
7332     * Static initializers with 3-byte symbols are supported now:
7333    extern const __memx char foo;
7334    const __memx void *pfoo = &foo;
7335       This requires at least Binutils 2.23.
7336
7337  IA-32/x86-64
7338
7339     * Allow -mpreferred-stack-boundary=3 for the x86-64 architecture with
7340       SSE extensions disabled. Since the x86-64 ABI requires 16 byte
7341       stack alignment, this is ABI incompatible and intended to be used
7342       in controlled environments where stack space is an important
7343       limitation. This option will lead to wrong code when functions
7344       compiled with 16 byte stack alignment (such as functions from a
7345       standard library) are called with misaligned stack. In this case,
7346       SSE instructions may lead to misaligned memory access traps. In
7347       addition, variable arguments will be handled incorrectly for 16
7348       byte aligned objects (including x87 long double and __int128),
7349       leading to wrong results. You must build all modules with
7350       -mpreferred-stack-boundary=3, including any libraries. This
7351       includes the system libraries and startup modules.
7352     * Support for the new Intel processor codename Broadwell with RDSEED,
7353       ADCX, ADOX, PREFETCHW is available through -madx, -mprfchw,
7354       -mrdseed command-line options.
7355     * Support for the Intel RTM and HLE intrinsics, built-in functions
7356       and code generation is available via -mrtm and -mhle.
7357     * Support for the Intel FXSR, XSAVE and XSAVEOPT instruction sets.
7358       Intrinsics and built-in functions are available via -mfxsr, -mxsave
7359       and -mxsaveopt respectively.
7360     * New -maddress-mode=[short|long] options for x32.
7361       -maddress-mode=short overrides default 64-bit addresses to 32-bit
7362       by emitting the 0x67 address-size override prefix. This is the
7363       default address mode for x32.
7364     * New built-in functions to detect run-time CPU type and ISA:
7365          + A built-in function __builtin_cpu_is has been added to detect
7366            if the run-time CPU is of a particular type. It returns a
7367            positive integer on a match and zero otherwise. It accepts one
7368            string literal argument, the CPU name. For example,
7369            __builtin_cpu_is("westmere") returns a positive integer if the
7370            run-time CPU is an Intel Core i7 Westmere processor. Please
7371            refer to the [30]user manual for the list of valid CPU names
7372            recognized.
7373          + A built-in function __builtin_cpu_supports has been added to
7374            detect if the run-time CPU supports a particular ISA feature.
7375            It returns a positive integer on a match and zero otherwise.
7376            It accepts one string literal argument, the ISA feature. For
7377            example, __builtin_cpu_supports("ssse3") returns a positive
7378            integer if the run-time CPU supports SSSE3 instructions.
7379            Please refer to the [31]user manual for the list of valid ISA
7380            names recognized.
7381       Caveat: If these built-in functions are called before any static
7382       constructors are invoked, like during IFUNC initialization, then
7383       the CPU detection initialization must be explicitly run using this
7384       newly provided built-in function, __builtin_cpu_init. The
7385       initialization needs to be done only once. For example, this is how
7386       the invocation would look like inside an IFUNC initializer:
7387    static void (*some_ifunc_resolver(void))(void)
7388    {
7389      __builtin_cpu_init();
7390      if (__builtin_cpu_is("amdfam10h") ...
7391      if (__builtin_cpu_supports("popcnt") ...
7392    }
7393
7394     * Function Multiversioning Support with G++:
7395       It is now possible to create multiple function versions each
7396       targeting a specific processor and/or ISA. Function versions have
7397       the same signature but different target attributes. For example,
7398       here is a program with function versions:
7399    __attribute__ ((target ("default")))
7400    int foo(void)
7401    {
7402      return 1;
7403    }
7404
7405    __attribute__ ((target ("sse4.2")))
7406    int foo(void)
7407    {
7408      return 2;
7409    }
7410
7411    int main (void)
7412    {
7413      int (*p) = &foo;
7414      assert ((*p)() == foo());
7415      return 0;
7416    }
7417
7418       Please refer to this [32]wiki for more information.
7419     * The x86 back end has been improved to allow option -fschedule-insns
7420       to work reliably. This option can be used to schedule instructions
7421       better and leads to improved performace in certain cases.
7422     * Windows MinGW-w64 targets (*-w64-mingw*) require at least r5437
7423       from the Mingw-w64 trunk.
7424     * Support for new AMD family 15h processors (Steamroller core) is now
7425       available through the -march=bdver3 and -mtune=bdver3 options.
7426     * Support for new AMD family 16h processors (Jaguar core) is now
7427       available through the -march=btver2 and -mtune=btver2 options.
7428
7429  FRV
7430
7431     * This target now supports the -fstack-usage command-line option.
7432
7433  MIPS
7434
7435     * GCC can now generate code specifically for the R4700, Broadcom XLP
7436       and MIPS 34kn processors. The associated -march options are
7437       -march=r4700, -march=xlp and -march=34kn respectively.
7438     * GCC now generates better DSP code for MIPS 74k cores thanks to
7439       further scheduling optimizations.
7440     * The MIPS port now supports the -fstack-check option.
7441     * GCC now passes the -mmcu and -mno-mcu options to the assembler.
7442     * Previous versions of GCC would silently accept -fpic and -fPIC for
7443       -mno-abicalls targets like mips*-elf. This combination was not
7444       intended or supported, and did not generate position-independent
7445       code. GCC 4.8 now reports an error when this combination is used.
7446
7447  PowerPC / PowerPC64 / RS6000
7448
7449     * SVR4 configurations (GNU/Linux, FreeBSD, NetBSD) no longer save,
7450       restore or update the VRSAVE register by default. The respective
7451       operating systems manage the VRSAVE register directly.
7452     * Large TOC support has been added for AIX through the command line
7453       option -mcmodel=large.
7454     * Native Thread-Local Storage support has been added for AIX.
7455     * VMX (Altivec) and VSX instruction sets now are enabled implicitly
7456       when targetting processors that support those hardware features on
7457       AIX 6.1 and above.
7458
7459  RX
7460
7461     * This target will now issue a warning message whenever multiple fast
7462       interrupt handlers are found in the same compilation unit. This
7463       feature can be turned off by the new
7464       -mno-warn-multiple-fast-interrupts command-line option.
7465
7466  S/390, System z
7467
7468     * Support for the IBM zEnterprise zEC12 processor has been added.
7469       When using the -march=zEC12 option, the compiler will generate code
7470       making use of the following new instructions:
7471          + load and trap instructions
7472          + 2 new compare and trap instructions
7473          + rotate and insert selected bits - without CC clobber
7474       The -mtune=zEC12 option enables zEC12 specific instruction
7475       scheduling without making use of new instructions.
7476     * Register pressure sensitive instruction scheduling is enabled by
7477       default.
7478     * The ifunc function attribute is enabled by default.
7479     * memcpy and memcmp invokations on big memory chunks or with run time
7480       lengths are not generated inline anymore when tuning for z10 or
7481       higher. The purpose is to make use of the IFUNC optimized versions
7482       in Glibc.
7483
7484  SH
7485
7486     * The default alignment settings have been reduced to be less
7487       aggressive. This results in more compact code for optimization
7488       levels other than -Os.
7489     * Improved support for the __atomic built-in functions:
7490          + A new option -matomic-model=model selects the model for the
7491            generated atomic sequences. The following models are
7492            supported:
7493
7494              soft-gusa
7495                      Software gUSA sequences (SH3* and SH4* only). On
7496                      SH4A targets this will now also partially utilize
7497                      the movco.l and movli.l instructions. This is the
7498                      default when the target is sh3*-*-linux* or
7499                      sh4*-*-linux*.
7500
7501              hard-llcs
7502                      Hardware movco.l / movli.l sequences (SH4A only).
7503
7504              soft-tcb
7505                      Software thread control block sequences.
7506
7507              soft-imask
7508                      Software interrupt flipping sequences (privileged
7509                      mode only). This is the default when the target is
7510                      sh1*-*-linux* or sh2*-*-linux*.
7511
7512              none
7513                      Generates function calls to the respective __atomic
7514                      built-in functions. This is the default for SH64
7515                      targets or when the target is not sh*-*-linux*.
7516
7517          + The option -msoft-atomic has been deprecated. It is now an
7518            alias for -matomic-model=soft-gusa.
7519          + A new option -mtas makes the compiler generate the tas.b
7520            instruction for the __atomic_test_and_set built-in function
7521            regardless of the selected atomic model.
7522          + The __sync functions in libgcc now reflect the selected atomic
7523            model when building the toolchain.
7524     * Added support for the mov.b and mov.w instructions with
7525       displacement addressing.
7526     * Added support for the SH2A instructions movu.b and movu.w.
7527     * Various improvements to code generated for integer arithmetic.
7528     * Improvements to conditional branches and code that involves the T
7529       bit. A new option -mzdcbranch tells the compiler to favor
7530       zero-displacement branches. This is enabled by default for SH4*
7531       targets.
7532     * The pref instruction will now be emitted by the __builtin_prefetch
7533       built-in function for SH3* targets.
7534     * The fmac instruction will now be emitted by the fmaf standard
7535       function and the __builtin_fmaf built-in function.
7536     * The -mfused-madd option has been deprecated in favor of the
7537       machine-independent -ffp-contract option. Notice that the fmac
7538       instruction will now be generated by default for expressions like a
7539       * b + c. This is due to the compiler default setting
7540       -ffp-contract=fast.
7541     * Added new options -mfsrra and -mfsca to allow the compiler using
7542       the fsrra and fsca instructions on targets other than SH4A (where
7543       they are already enabled by default).
7544     * Added support for the __builtin_bswap32 built-in function. It is
7545       now expanded as a sequence of swap.b and swap.w instructions
7546       instead of a library function call.
7547     * The behavior of the -mieee option has been fixed and the negative
7548       form -mno-ieee has been added to control the IEEE conformance of
7549       floating point comparisons. By default -mieee is now enabled and
7550       the option -ffinite-math-only implicitly sets -mno-ieee.
7551     * Added support for the built-in functions __builtin_thread_pointer
7552       and __builtin_set_thread_pointer. This assumes that GBR is used to
7553       hold the thread pointer of the current thread. Memory loads and
7554       stores relative to the address returned by __builtin_thread_pointer
7555       will now also utilize GBR based displacement address modes.
7556     * The -mdiv= option for targets other than SHmedia has been fixed and
7557       documented.
7558
7559  SPARC
7560
7561     * Added optimized instruction scheduling for Niagara4.
7562
7563  TILE-Gx
7564
7565     * Added support for the -mcmodel=MODEL command-line option. The
7566       models supported are small and large.
7567
7568  V850
7569
7570     * This target now supports the E3V5 architecture via the use of the
7571       new -mv850e3v5 command-line option. It also has experimental
7572       support for the e3v5 LOOP instruction which can be enabled via the
7573       new -mloop command-line option.
7574
7575  XStormy16
7576
7577     * This target now supports the -fstack-usage command-line option.
7578
7579Operating Systems
7580
7581  OpenBSD
7582
7583     * Support for OpenBSD/amd64 (x86_64-*-openbsd*) has been added and
7584       support for OpenBSD/i386 (i386-*-openbsd*) has been rejuvenated.
7585
7586  Windows (Cygwin)
7587
7588     * Executables are now linked against shared libgcc by default. The
7589       previous default was to link statically, which can still be done by
7590       explicitly specifying -static or static-libgcc on the command line.
7591       However it is strongly advised against, as it will cause problems
7592       for any application that makes use of DLLs compiled by GCC. It
7593       should be alright for a monolithic stand-alone application that
7594       only links against the Windows DLLs, but offers little or no
7595       benefit.
7596
7597GCC 4.8.1
7598
7599   This is the [33]list of problem reports (PRs) from GCC's bug tracking
7600   system that are known to be fixed in the 4.8.1 release. This list might
7601   not be complete (that is, it is possible that some PRs that have been
7602   fixed are not listed here).
7603
7604   The C++11 <chrono> std::chrono::system_clock and
7605   std::chrono::steady_clock classes have changed ABI in GCC 4.8.1, they
7606   both are now separate (never typedefs of each other), both use
7607   std::chrono::nanoseconds resolution, on most GNU/Linux configurations
7608   std::chrono::steady_clock is now finally monotonic, and both classes
7609   are mangled differently than in the previous GCC releases.
7610   std::chrono::system_clock::now() with std::chrono::microseconds resp.
7611   std::chrono::seconds resolution is still exported for backwards
7612   compatibility with default configured libstdc++. Note that libstdc++
7613   configured with --enable-libstdcxx-time= used to be ABI incompatible
7614   with default configured libstdc++ for those two classes and no ABI
7615   compatibility can be offered for those configurations, so any C++11
7616   code that uses those classes and has been compiled and linked against
7617   libstdc++ configured with the non-default --enable-libstdcxx-time=
7618   configuration option needs to be recompiled.
7619
7620GCC 4.8.2
7621
7622   This is the [34]list of problem reports (PRs) from GCC's bug tracking
7623   system that are known to be fixed in the 4.8.2 release. This list might
7624   not be complete (that is, it is possible that some PRs that have been
7625   fixed are not listed here).
7626
7627GCC 4.8.3
7628
7629   This is the [35]list of problem reports (PRs) from GCC's bug tracking
7630   system that are known to be fixed in the 4.8.3 release. This list might
7631   not be complete (that is, it is possible that some PRs that have been
7632   fixed are not listed here).
7633
7634   Support for the new powerpc64le-linux platform has been added. It
7635   defaults to generating code that conforms to the ELFV2 ABI.
7636
7637GCC 4.8.4
7638
7639   This is the [36]list of problem reports (PRs) from GCC's bug tracking
7640   system that are known to be fixed in the 4.8.4 release. This list might
7641   not be complete (that is, it is possible that some PRs that have been
7642   fixed are not listed here).
7643
7644GCC 4.8.5
7645
7646   This is the [37]list of problem reports (PRs) from GCC's bug tracking
7647   system that are known to be fixed in the 4.8.5 release. This list might
7648   not be complete (that is, it is possible that some PRs that have been
7649   fixed are not listed here).
7650
7651
7652    For questions related to the use of GCC, please consult these web
7653    pages and the [38]GCC manuals. If that fails, the
7654    [39]gcc-help@gcc.gnu.org mailing list might help. Comments on these
7655    web pages and the development of GCC are welcome on our developer
7656    list at [40]gcc@gcc.gnu.org. All of [41]our lists have public
7657    archives.
7658
7659   Copyright (C) [42]Free Software Foundation, Inc. Verbatim copying and
7660   distribution of this entire article is permitted in any medium,
7661   provided this notice is preserved.
7662
7663   These pages are [43]maintained by the GCC team. Last modified
7664   2020-03-26[44].
7665
7666References
7667
7668   1. https://gcc.gnu.org/wiki/cxx-conversion
7669   2. ftp://gcc.gnu.org/pub/gcc/infrastructure/
7670   3. http://www.nongnu.org/avr-libc/
7671   4. https://gcc.gnu.org/PR54461
7672   5. https://gcc.gnu.org/gcc-4.8/porting_to.html
7673   6. https://github.com/google/sanitizers
7674   7. https://code.google.com/archive/p/data-race-test/wikis/ThreadSanitizer.wiki
7675   8. https://lkml.org/lkml/2006/11/28/239
7676   9. http://gcc.gnu.org/gcc-4.8/cxx0x_status.html
7677  10. http://gcc.gnu.org/gcc-4.8/cxx0x_status.html
7678  11. http://gcc.gnu.org/gcc-4.8/cxx0x_status.html
7679  12. http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2011/n3276.pdf
7680  13. http://gcc.gnu.org/gcc-4.8/cxx0x_status.html
7681  14. http://www.open-std.org/JTC1/SC22/WG21/docs/papers/2012/n3386.html
7682  15. http://gcc.gnu.org/projects/cxx1y.html
7683  16. https://gcc.gnu.org/onlinedocs/gcc-4.8.4/libstdc++/manual/manual/status.html#status.iso.2011
7684  17. https://gcc.gnu.org/onlinedocs/gfortran/BACKTRACE.html
7685  18. https://gcc.gnu.org/onlinedocs/gfortran/Error-and-Warning-Options.html
7686  19. https://gcc.gnu.org/onlinedocs/gfortran/Error-and-Warning-Options.html
7687  20. https://gcc.gnu.org/onlinedocs/gfortran/Code-Gen-Options.html
7688  21. https://gcc.gnu.org/onlinedocs/gfortran/Error-and-Warning-Options.html
7689  22. https://gcc.gnu.org/onlinedocs/gfortran/Error-and-Warning-Options.html
7690  23. https://gcc.gnu.org/onlinedocs/gfortran/TMPDIR.html
7691  24. https://gcc.gnu.org/wiki/Fortran2003Status
7692  25. https://gcc.gnu.org/wiki/TS29113Status
7693  26. https://gcc.gnu.org/git/?p=gcc.git;a=blob;f=libgfortran/libgfortran.h
7694  27. http://chasm-interop.sourceforge.net/
7695  28. https://gcc.gnu.org/wiki/avr-gcc#Fixed-Point_Support
7696  29. https://gcc.gnu.org/onlinedocs/gcc/Fixed-Point.html
7697  30. https://gcc.gnu.org/onlinedocs/gcc/x86-Built-in-Functions.html
7698  31. https://gcc.gnu.org/onlinedocs/gcc/x86-Built-in-Functions.html
7699  32. https://gcc.gnu.org/wiki/FunctionMultiVersioning
7700  33. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.8.1
7701  34. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.8.2
7702  35. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.8.3
7703  36. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.8.4
7704  37. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.8.5
7705  38. https://gcc.gnu.org/onlinedocs/
7706  39. mailto:gcc-help@gcc.gnu.org
7707  40. mailto:gcc@gcc.gnu.org
7708  41. https://gcc.gnu.org/lists.html
7709  42. https://www.fsf.org/
7710  43. https://gcc.gnu.org/about.html
7711  44. http://validator.w3.org/check/referer
7712======================================================================
7713http://gcc.gnu.org/gcc-4.7/index.html
7714
7715                             GCC 4.7 Release Series
7716
7717   (This release series is no longer supported.)
7718
7719   June 12, 2014
7720
7721   The [1]GNU project and the GCC developers are pleased to announce the
7722   release of GCC 4.7.4.
7723
7724   This release is a bug-fix release, containing fixes for regressions in
7725   GCC 4.7.3 relative to previous releases of GCC.
7726
7727Release History
7728
7729   GCC 4.7.4
7730          June 12, 2014 ([2]changes, [3]documentation)
7731
7732   GCC 4.7.3
7733          April 11, 2013 ([4]changes, [5]documentation)
7734
7735   GCC 4.7.2
7736          September 20, 2012 ([6]changes, [7]documentation)
7737
7738   GCC 4.7.1
7739          June 14, 2012 ([8]changes, [9]documentation)
7740
7741   GCC 4.7.0
7742          March 22, 2012 ([10]changes, [11]documentation)
7743
7744References and Acknowledgements
7745
7746   GCC used to stand for the GNU C Compiler, but since the compiler
7747   supports several other languages aside from C, it now stands for the
7748   GNU Compiler Collection.
7749
7750   A list of [12]successful builds is updated as new information becomes
7751   available.
7752
7753   The GCC developers would like to thank the numerous people that have
7754   contributed new features, improvements, bug fixes, and other changes as
7755   well as test results to GCC. This [13]amazing group of volunteers is
7756   what makes GCC successful.
7757
7758   For additional information about GCC please refer to the [14]GCC
7759   project web site or contact the [15]GCC development mailing list.
7760
7761   To obtain GCC please use [16]our mirror sites or [17]our version
7762   control system.
7763
7764
7765    For questions related to the use of GCC, please consult these web
7766    pages and the [18]GCC manuals. If that fails, the
7767    [19]gcc-help@gcc.gnu.org mailing list might help. Comments on these
7768    web pages and the development of GCC are welcome on our developer
7769    list at [20]gcc@gcc.gnu.org. All of [21]our lists have public
7770    archives.
7771
7772   Copyright (C) [22]Free Software Foundation, Inc. Verbatim copying and
7773   distribution of this entire article is permitted in any medium,
7774   provided this notice is preserved.
7775
7776   These pages are [23]maintained by the GCC team. Last modified
7777   2020-01-14[24].
7778
7779References
7780
7781   1. http://www.gnu.org/
7782   2. http://gcc.gnu.org/gcc-4.7/changes.html
7783   3. https://gcc.gnu.org/onlinedocs/4.7.4/
7784   4. http://gcc.gnu.org/gcc-4.7/changes.html
7785   5. https://gcc.gnu.org/onlinedocs/4.7.3/
7786   6. http://gcc.gnu.org/gcc-4.7/changes.html
7787   7. https://gcc.gnu.org/onlinedocs/4.7.2/
7788   8. http://gcc.gnu.org/gcc-4.7/changes.html
7789   9. https://gcc.gnu.org/onlinedocs/4.7.1/
7790  10. http://gcc.gnu.org/gcc-4.7/changes.html
7791  11. https://gcc.gnu.org/onlinedocs/4.7.0/
7792  12. http://gcc.gnu.org/gcc-4.7/buildstat.html
7793  13. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
7794  14. http://gcc.gnu.org/index.html
7795  15. mailto:gcc@gcc.gnu.org
7796  16. http://gcc.gnu.org/mirrors.html
7797  17. http://gcc.gnu.org/git.html
7798  18. https://gcc.gnu.org/onlinedocs/
7799  19. mailto:gcc-help@gcc.gnu.org
7800  20. mailto:gcc@gcc.gnu.org
7801  21. https://gcc.gnu.org/lists.html
7802  22. https://www.fsf.org/
7803  23. https://gcc.gnu.org/about.html
7804  24. http://validator.w3.org/check/referer
7805======================================================================
7806http://gcc.gnu.org/gcc-4.7/changes.html
7807
7808                             GCC 4.7 Release Series
7809                        Changes, New Features, and Fixes
7810
7811Caveats
7812
7813     * The -fconserve-space flag has been deprecated. The flag had no
7814       effect for most targets: only targets without a global .bss section
7815       and without support for switchable sections. Furthermore, the flag
7816       only had an effect for G++, where it could result in wrong
7817       semantics (please refer to the GCC manual for further details). The
7818       flag will be removed in GCC 4.8
7819     * Support for a number of older systems and recently unmaintained or
7820       untested target ports of GCC has been declared obsolete in GCC 4.7.
7821       Unless there is activity to revive them, the next release of GCC
7822       will have their sources permanently removed.
7823       All GCC ports for the following processor architectures have been
7824       declared obsolete:
7825          + picoChip (picochip-*)
7826       The following ports for individual systems on particular
7827       architectures have been obsoleted:
7828          + IRIX 6.5 (mips-sgi-irix6.5)
7829          + MIPS OpenBSD (mips*-*-openbsd*)
7830          + Solaris 8 (*-*-solaris2.8). Details can be found in the
7831            [1]announcement.
7832          + Tru64 UNIX V5.1 (alpha*-dec-osf5.1*)
7833     * On ARM, when compiling for ARMv6 (but not ARMv6-M), ARMv7-A,
7834       ARMv7-R, or ARMv7-M, the new option -munaligned-access is active by
7835       default, which for some sources generates code that accesses memory
7836       on unaligned addresses. This requires the kernel of those systems
7837       to enable such accesses (controlled by CP15 register c1, refer to
7838       ARM documentation). Alternatively, or for compatibility with
7839       kernels where unaligned accesses are not supported, all code has to
7840       be compiled with -mno-unaligned-access. Upstream Linux kernel
7841       releases have automatically and unconditionally supported unaligned
7842       accesses as emitted by GCC due to this option being active since
7843       version 2.6.28.
7844     * Support on ARM for the legacy floating-point accelerator (FPA) and
7845       the mixed-endian floating-point format that it used has been
7846       obsoleted. The ports that still use this format have been obsoleted
7847       as well. Many legacy ARM ports already provide an alternative that
7848       uses the VFP floating-point format. The obsolete ports will be
7849       deleted in the next release.
7850       The obsolete ports with alternatives are:
7851          + arm*-*-rtems (use arm*-*-rtemseabi)
7852          + arm*-*-linux-gnu (use arm*-*-linux-gnueabi)
7853          + arm*-*-elf (use arm*-*-eabi)
7854          + arm*-*-uclinux* (use arm*-*-uclinux*eabi)
7855       Note, however, that these alternatives are not binary compatible
7856       with their legacy counterparts (although some can support running
7857       legacy applications).
7858       The obsolete ports that currently lack a modern alternative are:
7859          + arm*-*-ecos-elf
7860          + arm*-*-freebsd
7861          + arm*-wince-pe*
7862       New ports that support more recent versions of the architecture are
7863       welcome.
7864     * Support for the Maverick co-processor on ARM has been obsoleted.
7865       Code to support it will be deleted in the next release.
7866     * Support has been removed for Unix International threads on Solaris
7867       2, so the --enable-threads=solaris configure option and the
7868       -threads compiler option don't work any longer.
7869     * Support has been removed for the Solaris BSD Compatibility Package,
7870       which lives in /usr/ucbinclude and /usr/ucblib. It has been removed
7871       from Solaris 11, and was only intended as a migration aid from
7872       SunOS 4 to SunOS 5. The -compat-bsd compiler option is not
7873       recognized any longer.
7874     * The AVR port's libgcc has been improved and its multilib structure
7875       has been enhanced. As a result, all objects contributing to an
7876       application must either be compiled with GCC versions up to 4.6.x
7877       or with GCC versions 4.7.1 or later. If the compiler is used with
7878       AVR Libc, you need a version that supports the new layout, i.e.
7879       implements [2]#35407.
7880     * The AVR port's -mshort-calls command-line option has been
7881       deprecated. It will be removed in the GCC 4.8 release. See -mrelax
7882       for a replacement.
7883     * The AVR port only references startup code that clears .bss and the
7884       common section resp. initializes the .data and .rodata section
7885       provided respective sections (or subsections thereof) are not
7886       empty, see [3]PR18145. Applications that put all static storage
7887       objects into non-standard sections and / or define all static
7888       storage objects in assembler modules, must reference __do_clear_bss
7889       resp. __do_copy_data by hand or undefine the symbol(s) by means of
7890       -Wl,-u,__do_clear_bss resp. -Wl,-u,__do_copy_data.
7891     * The ARM port's -mwords-little-endian option has been deprecated. It
7892       will be removed in a future release.
7893     * Support has been removed for the NetWare x86 configuration
7894       obsoleted in GCC 4.6.
7895     * It is no longer possible to use the "l" constraint in MIPS16 asm
7896       statements.
7897     * GCC versions 4.7.0 and 4.7.1 had changes to the C++ standard
7898       library which affected the ABI in C++11 mode: a data member was
7899       added to std::list changing its size and altering the definitions
7900       of some member functions, and std::pair's move constructor was
7901       non-trivial which altered the calling convention for functions with
7902       std::pair arguments or return types. The ABI incompatibilities have
7903       been fixed for GCC version 4.7.2 but as a result C++11 code
7904       compiled with GCC 4.7.0 or 4.7.1 may be incompatible with C++11
7905       code compiled with different GCC versions and with C++98/C++03 code
7906       compiled with any version.
7907     * On ARM, a bug has been fixed in GCC's implementation of the AAPCS
7908       rules for the layout of vectors that could lead to wrong code being
7909       generated. Vectors larger than 8 bytes in size are now by default
7910       aligned to an 8-byte boundary. This is an ABI change: code that
7911       makes explicit use of vector types may be incompatible with binary
7912       objects built with older versions of GCC. Auto-vectorized code is
7913       not affected by this change. (This change affects GCC versions
7914       4.7.2 and later.)
7915     * More information on porting to GCC 4.7 from previous versions of
7916       GCC can be found in the [4]porting guide for this release.
7917
7918General Optimizer Improvements
7919
7920     * Support for a new parameter --param case-values-threshold=n was
7921       added to allow users to control the cutoff between doing switch
7922       statements as a series of if statements and using a jump table.
7923     * Link-time optimization (LTO) improvements:
7924          + Improved scalability and reduced memory usage. Link time
7925            optimization of Firefox now requires 3GB of RAM on a 64-bit
7926            system, while over 8GB was needed previously. Linking time has
7927            been improved, too. The serial stage of linking Firefox has
7928            been sped up by about a factor of 10.
7929          + Reduced size of object files and temporary storage used during
7930            linking.
7931          + Streaming performance (both outbound and inbound) has been
7932            improved.
7933          + ld -r is now supported with LTO.
7934          + Several bug fixes, especially in symbol table handling and
7935            merging.
7936     * Interprocedural optimization improvements:
7937          + Heuristics now take into account that after inlining code will
7938            be optimized out because of known values (or properties) of
7939            function parameters. For example:
7940void foo(int a)
7941{
7942  if (a > 10)
7943    ... huge code ...
7944}
7945void bar (void)
7946{
7947  foo (0);
7948}
7949
7950            The call of foo will be inlined into bar even when optimizing
7951            for code size. Constructs based on __builtin_constant_p are
7952            now understood by the inliner and code size estimates are
7953            evaluated a lot more realistically.
7954          + The representation of C++ virtual thunks and aliases (both
7955            implicit and defined via the alias attribute) has been
7956            re-engineered. Aliases no longer pose optimization barriers
7957            and calls to an alias can be inlined and otherwise optimized.
7958          + The inter-procedural constant propagation pass has been
7959            rewritten. It now performs generic function specialization.
7960            For example when compiling the following:
7961void foo(bool flag)
7962{
7963  if (flag)
7964    ... do something ...
7965  else
7966    ... do something else ...
7967}
7968void bar (void)
7969{
7970  foo (false);
7971  foo (true);
7972  foo (false);
7973  foo (true);
7974  foo (false);
7975  foo (true);
7976}
7977
7978            GCC will now produce two copies of foo. One with flag being
7979            true, while other with flag being false. This leads to
7980            performance improvements previously possible only by inlining
7981            all calls. Cloning causes a lot less code size growth.
7982     * A string length optimization pass has been added. It attempts to
7983       track string lengths and optimize various standard C string
7984       functions like strlen, strchr, strcpy, strcat, stpcpy and their
7985       _FORTIFY_SOURCE counterparts into faster alternatives. This pass is
7986       enabled by default at -O2 or above, unless optimizing for size, and
7987       can be disabled by the -fno-optimize-strlen option. The pass can
7988       e.g. optimize
7989char *bar (const char *a)
7990{
7991  size_t l = strlen (a) + 2;
7992  char *p = malloc (l); if (p == NULL) return p;
7993  strcpy (p, a); strcat (p, "/"); return p;
7994}
7995
7996       into:
7997char *bar (const char *a)
7998{
7999  size_t tmp = strlen (a);
8000  char *p = malloc (tmp + 2); if (p == NULL) return p;
8001  memcpy (p, a, tmp); memcpy (p + tmp, "/", 2); return p;
8002}
8003
8004       or for hosted compilations where stpcpy is available in the runtime
8005       and headers provide its prototype, e.g.
8006void foo (char *a, const char *b, const char *c, const char *d)
8007{
8008  strcpy (a, b); strcat (a, c); strcat (a, d);
8009}
8010
8011       can be optimized into:
8012void foo (char *a, const char *b, const char *c, const char *d)
8013{
8014  strcpy (stpcpy (stpcpy (a, b), c), d);
8015}
8016
8017New Languages and Language specific improvements
8018
8019     * Version 3.1 of the OpenMP specification is now supported for the C,
8020       C++, and Fortran compilers.
8021
8022  Ada
8023
8024     * The command-line option -feliminate-unused-debug-types has been
8025       re-enabled by default, as it is for the other languages, leading to
8026       a reduction in debug info size of 12.5% and more for relevant
8027       cases, as well as to a small compilation speedup.
8028
8029  C family
8030
8031     * A new built-in, __builtin_assume_aligned, has been added, through
8032       which the compiler can be hinted about pointer alignment and can
8033       use it to improve generated code.
8034     * A new warning option -Wunused-local-typedefs was added for C, C++,
8035       Objective-C and Objective-C++. This warning diagnoses typedefs
8036       locally defined in a function, and otherwise not used.
8037     * A new experimental command-line option -ftrack-macro-expansion was
8038       added for C, C++, Objective-C, Objective-C++ and Fortran. It allows
8039       the compiler to emit diagnostic about the current macro expansion
8040       stack when a compilation error occurs in a macro expansion.
8041     * Experimental support for transactional memory has been added. It
8042       includes support in the compiler, as well as a supporting runtime
8043       library called libitm. To compile code with transactional memory
8044       constructs, use the -fgnu-tm option.
8045       Support is currently available for Alpha, ARM, PowerPC, SH, SPARC,
8046       and 32-bit/64-bit x86 platforms.
8047       For more details on transactional memory see [5]the GCC WiKi.
8048     * Support for atomic operations specifying the C++11/C11 memory model
8049       has been added. These new __atomic routines replace the existing
8050       __sync built-in routines.
8051       Atomic support is also available for memory blocks. Lock-free
8052       instructions will be used if a memory block is the same size and
8053       alignment as a supported integer type. Atomic operations which do
8054       not have lock-free support are left as function calls. A set of
8055       library functions is available on the GCC atomic wiki in the
8056       "External Atomics Library" section.
8057       For more details on the memory models and features, see the
8058       [6]atomic wiki.
8059     * When a binary operation is performed on vector types and one of the
8060       operands is a uniform vector, it is possible to replace the vector
8061       with the generating element. For example:
8062typedef int v4si __attribute__ ((vector_size (16)));
8063v4si res, a = {1,2,3,4};
8064int x;
8065
8066res = 2 + a;  /* means {2,2,2,2} + a  */
8067res = a - x;  /* means a - {x,x,x,x}  */
8068
8069  C
8070
8071     * There is support for some more features from the C11 revision of
8072       the ISO C standard. GCC now accepts the options -std=c11 and
8073       -std=gnu11, in addition to the previous -std=c1x and -std=gnu1x.
8074          + Unicode strings (previously supported only with options such
8075            as -std=gnu11, now supported with -std=c11), and the
8076            predefined macros __STDC_UTF_16__ and __STDC_UTF_32__.
8077          + Nonreturning functions (_Noreturn and <stdnoreturn.h>).
8078          + Alignment support (_Alignas, _Alignof, max_align_t,
8079            <stdalign.h>).
8080          + A built-in function __builtin_complex is provided to support C
8081            library implementation of the CMPLX family of macros.
8082
8083  C++
8084
8085     * G++ now accepts the -std=c++11, -std=gnu++11, and -Wc++11-compat
8086       options, which are equivalent to -std=c++0x, -std=gnu++0x, and
8087       -Wc++0x-compat, respectively.
8088     * G++ now implements [7]C++11 extended friend syntax:
8089
8090template<class W>
8091class Q
8092{
8093  static const int I = 2;
8094public:
8095  friend W;
8096};
8097
8098struct B
8099{
8100  int ar[Q<B>::I];
8101};
8102
8103     * Thanks to Ville Voutilainen, G++ now implements [8]C++11 explicit
8104       override control.
8105
8106struct B {
8107  virtual void f() const final;
8108  virtual void f(int);
8109};
8110
8111struct D : B {
8112  void f() const;            // error: D::f attempts to override final B::f
8113  void f(long) override;     // error: doesn't override anything
8114  void f(int) override;      // ok
8115};
8116
8117struct E final { };
8118struct F: E { }; // error: deriving from final class
8119
8120     * G++ now implements [9]C++11 non-static data member initializers.
8121
8122struct A {
8123  int i = 42;
8124} a; // initializes a.i to 42
8125
8126     * Thanks to Ed Smith-Rowland, G++ now implements [10]C++11
8127       user-defined literals.
8128
8129// Not actually a good approximation.  :)
8130constexpr long double operator"" _degrees (long double d) { return d * 0.0175; }
8131long double pi = 180.0_degrees;
8132
8133     * G++ now implements [11]C++11 alias-declarations.
8134
8135template <class T> using Ptr = T*;
8136Ptr<int> ip;  // decltype(ip) is int*
8137
8138     * Thanks to Ville Voutilainen and Pedro Lamar�o, G++ now implements
8139       [12]C++11 delegating constructors.
8140
8141struct A {
8142  A(int);
8143  A(): A(42) { } // delegate to the A(int) constructor
8144};
8145
8146     * G++ now fully implements C++11 atomic classes rather than just
8147       integer derived classes.
8148
8149class POD {
8150  int a;
8151  int b;
8152};
8153std::atomic<POD> my_atomic_POD;
8154
8155     * G++ now sets the predefined macro __cplusplus to the correct value,
8156       199711L for C++98/03, and 201103L for C++11.
8157     * G++ now correctly implements the two-phase lookup rules such that
8158       an unqualified name used in a template must have an appropriate
8159       declaration found either in scope at the point of definition of the
8160       template or by argument-dependent lookup at the point of
8161       instantiation. As a result, code that relies on a second
8162       unqualified lookup at the point of instantiation to find functions
8163       declared after the template or in dependent bases will be rejected.
8164       The compiler will suggest ways to fix affected code, and using the
8165       -fpermissive compiler flag will allow the code to compile with a
8166       warning.
8167
8168template <class T>
8169void f() { g(T()); } // error, g(int) not found by argument-dependent lookup
8170void g(int) { } // fix by moving this declaration before the declaration of f
8171
8172template <class T>
8173struct A: T {
8174  // error, B::g(B) not found by argument-dependent lookup
8175  void f() { g(T()); } // fix by using this->g or A::g
8176};
8177
8178struct B { void g(B); };
8179
8180int main()
8181{
8182  f<int>();
8183  A<B>().f();
8184}
8185
8186     * G++ now properly re-uses stack space allocated for temporary
8187       objects when their lifetime ends, which can significantly lower
8188       stack consumption for some C++ functions. As a result of this, some
8189       code with undefined behavior will now break:
8190
8191const int &f(const int &i) { return i; }
8192....
8193const int &x = f(1);
8194const int &y = f(2);
8195
8196       Here, x refers to the temporary allocated to hold the 1 argument,
8197       which only lives until the end of the initialization; it
8198       immediately becomes a dangling reference. So the next statement
8199       re-uses the stack slot to hold the 2 argument, and users of x get
8200       that value instead.
8201       Note that this should not cause any change of behavior for
8202       temporaries of types with non-trivial destructors, as they are
8203       already destroyed at end of full-expression; the change is that now
8204       the storage is released as well.
8205     * A new command-line option -Wdelete-non-virtual-dtor has been added
8206       to warn when delete is used to destroy an instance of a class which
8207       has virtual functions and non-virtual destructor. It is unsafe to
8208       delete an instance of a derived class through a pointer to a base
8209       class if the base class does not have a virtual destructor. This
8210       warning is enabled by -Wall.
8211     * A new command-line option -Wzero-as-null-pointer-constant has been
8212       added to warn when a literal '0' is used as null pointer constant.
8213       It can be useful to facilitate the conversion to nullptr in C++11.
8214     * As per C++98, access-declarations are now deprecated by G++.
8215       Using-declarations are to be used instead. Furthermore, some
8216       efforts have been made to improve the support of class scope
8217       using-declarations. In particular, using-declarations referring to
8218       a dependent type now work as expected ([13]bug c++/14258).
8219     * The ELF symbol visibility of a template instantiation is now
8220       properly constrained by the visibility of its template arguments
8221       ([14]bug c++/35688).
8222
8223    Runtime Library (libstdc++)
8224
8225     * [15]Improved experimental support for the new ISO C++ standard,
8226       C++11, including:
8227          + using noexcept in most of the library;
8228          + implementations of pointer_traits, allocator_traits and
8229            scoped_allocator_adaptor;
8230          + uses-allocator construction for tuple;
8231          + vector meets the allocator-aware container requirements;
8232          + replacing monotonic_clock with steady_clock;
8233          + enabling the thread support library on most POSIX targets;
8234          + many small improvements to conform to the FDIS.
8235     * Added --enable-clocale=newlib configure option.
8236     * Debug Mode iterators for unordered associative containers.
8237     * Avoid polluting the global namespace and do not include <unistd.h>.
8238
8239  Fortran
8240
8241     * The compile flag [16]-fstack-arrays has been added, which causes
8242       all local arrays to be put on stack memory. For some programs this
8243       will improve the performance significantly. If your program uses
8244       very large local arrays, it is possible that you will have to
8245       extend your runtime limits for stack memory.
8246     * The [17]-Ofast flag now also implies [18]-fno-protect-parens and
8247       [19]-fstack-arrays.
8248     * Front-end optimizations can now be selected by the
8249       [20]-ffrontend-optimize option and deselected by the
8250       -fno-frontend-optimize option.
8251     * When front-end optimization removes a function call,
8252       [21]-Wfunction-elimination warns about that.
8253     * When performing front-end-optimization, the
8254       [22]-faggressive-function-elimination option allows the removal of
8255       duplicate function calls even for impure functions.
8256     * The flag [23]-Wreal-q-constant has been added, which warns if
8257       floating-point literals have been specified using q (such as
8258       1.0q0); the q marker is now supported as a vendor extension to
8259       denote quad precision (REAL(16) or, if not available, REAL(10)).
8260       Consider using a kind parameter (such as in 1.0_qp) instead, which
8261       can be obtained via [24]SELECTED_REAL_KIND.
8262     * The GFORTRAN_USE_STDERR environment variable has been removed. GNU
8263       Fortran now always prints error messages to standard error. If you
8264       wish to redirect standard error, please consult the manual for your
8265       OS, shell, batch environment etc. as appropriate.
8266     * The -fdump-core option and GFORTRAN_ERROR_DUMPCORE environment
8267       variable have been removed. When encountering a serious error,
8268       gfortran will now always abort the program. Whether a core dump is
8269       generated depends on the user environment settings; see the ulimit
8270       -c setting for POSIX shells, limit coredumpsize for C shells, and
8271       the [25]WER user-mode dumps settings on Windows.
8272     * The [26]-fbacktrace option is now enabled by default. When
8273       encountering a fatal error, gfortran will attempt to print a
8274       backtrace to standard error before aborting. It can be disabled
8275       with -fno-backtrace. Note: On POSIX targets with the addr2line
8276       utility from GNU binutils, GNU Fortran can print a backtrace with
8277       function name, file name, line number information in addition to
8278       the addresses; otherwise only the addresses are printed.
8279     * [27]Fortran 2003:
8280          + Generic interface names which have the same name as derived
8281            types are now supported, which allows to write constructor
8282            functions. Note that Fortran does not support static
8283            constructor functions; only default initialization or an
8284            explicit structure-constructor initialization are available.
8285          + [28]Polymorphic (class) arrays are now supported.
8286     * [29]Fortran 2008:
8287          + Support for the DO CONCURRENT construct has been added, which
8288            allows the user to specify that individual loop iterations
8289            have no interdependencies.
8290          + [30]Coarrays: Full single-image support except for polymorphic
8291            coarrays. Additionally, preliminary support for multiple
8292            images via an MPI-based [31]coarray communication library has
8293            been added. Note: The library version is not yet usable as
8294            remote coarray access is not yet possible.
8295     * [32]TS 29113:
8296          + New flag [33]-std=f2008ts permits programs that are expected
8297            to conform to the Fortran 2008 standard and the draft
8298            Technical Specification (TS) 29113 on Further Interoperability
8299            of Fortran with C.
8300          + The OPTIONAL attribute is now allowed for dummy arguments of
8301            BIND(C) procedures.
8302          + The RANK intrinsic has been added.
8303          + The implementation of the ASYNCHRONOUS attribute in GCC is
8304            compatible with the candidate draft of TS 29113 (since GCC
8305            4.6).
8306
8307  Go
8308
8309     * GCC 4.7 implements the [34]Go 1 language standard. The library
8310       support in 4.7.0 is not quite complete, due to release timing.
8311       Release 4.7.1 includes complete support for Go 1. The Go library is
8312       from the Go 1.0.1 release.
8313     * Go has been tested on GNU/Linux and Solaris platforms. It may work
8314       on other platforms as well.
8315
8316New Targets and Target Specific Improvements
8317
8318  ARM
8319
8320     * GCC now supports the Cortex-A7 processor implementing the v7-a
8321       version of the architecture using the option -mcpu=cortex-a7.
8322     * The default vector size in auto-vectorization for NEON is now 128
8323       bits. If vectorization fails thusly, the vectorizer tries again
8324       with 64-bit vectors.
8325     * A new option -mvectorize-with-neon-double was added to allow users
8326       to change the vector size to 64 bits.
8327
8328  AVR
8329
8330     * GCC now supports the XMEGA architecture. This requires GNU binutils
8331       2.22 or later.
8332     * Support for the [35]named address spaces __flash, __flash1, ...,
8333       __flash5 and __memx has been added. These address spaces locate
8334       read-only data in flash memory and allow reading from flash memory
8335       by means of ordinary C code, i.e. without the need of (inline)
8336       assembler code:
8337
8338const __flash int values[] = { 42, 31 };
8339
8340int add_values (const __flash int *p, int i)
8341{
8342    return values[i] + *p;
8343}
8344
8345     * Support has been added for the AVR-specific configure option
8346       --with-avrlibc=yes in order to arrange for better integration of
8347       [36]AVR-Libc. This configure option is supported in avr-gcc 4.7.2
8348       and newer and will only take effect in non-RTEMS configurations. If
8349       avr-gcc is configured for RTEMS, the option will be ignored which
8350       is the same as specifying --with-avrlibc=no. See [37]PR54461 for
8351       more technical details.
8352     * Support for AVR-specific [38]built-in functions has been added.
8353     * Support has been added for the signed and unsigned 24-bit scalar
8354       integer types __int24 and __uint24.
8355     * New command-line options -maccumulate-args, -mbranch-cost=cost and
8356       -mstrict-X were added to allow better fine-tuning of code
8357       optimization.
8358     * The command-line option -fdata-sections now also takes affect on
8359       the section names of variables with the progmem attribute.
8360     * A new inline assembler print modifier %i to print a RAM address as
8361       I/O address has been added:
8362
8363#include <avr/io.h> /* Port Definitions from AVR-LibC */
8364
8365void set_portb (uint8_t value)
8366{
8367    asm volatile ("out %i0, %1" :: "n" (&PORTB), "r" (value) : "memory");
8368}
8369
8370       The offset between an I/O address and the RAM address for that I/O
8371       location is device-specific. This offset is taken into account when
8372       printing a RAM address with the %i modifier so that the address is
8373       suitable to be used as operand in an I/O command. The address must
8374       be a constant integer known at compile time.
8375     * The inline assembler constraint "R" to represent integers in the
8376       range -6 ... 5 has been removed without replacement.
8377     * Many optimizations to:
8378          + 64-bit integer arithmetic
8379          + Widening multiplication
8380          + Integer division by a constant
8381          + Avoid constant reloading in multi-byte instructions.
8382          + Micro-optimizations for special instruction sequences.
8383          + Generic built-in functions like __builtin_ffs*,
8384            __builtin_clz*, etc.
8385          + If-else decision trees generated by switch instructions
8386          + Merging of data located in flash memory
8387          + New libgcc variants for devices with 8-bit wide stack pointer
8388          + ...
8389     * Better documentation:
8390          + Handling of EIND and indirect jumps on devices with more than
8391            128 KiB of program memory.
8392          + Handling of the RAMPD, RAMPX, RAMPY and RAMPZ special function
8393            registers.
8394          + Function attributes OS_main and OS_task.
8395          + AVR-specific built-in macros.
8396
8397  C6X
8398
8399     * Support has been added for the Texas Instruments C6X family of
8400       processors.
8401
8402  CR16
8403
8404     * Support has been added for National Semiconductor's CR16
8405       architecture.
8406
8407  Epiphany
8408
8409     * Support has been added for Adapteva's Epiphany architecture.
8410
8411  IA-32/x86-64
8412
8413     * Support for Intel AVX2 intrinsics, built-in functions and code
8414       generation is available via -mavx2.
8415     * Support for Intel BMI2 intrinsics, built-in functions and code
8416       generation is available via -mbmi2.
8417     * Implementation and automatic generation of __builtin_clz* using the
8418       lzcnt instruction is available via -mlzcnt.
8419     * Support for Intel FMA3 intrinsics and code generation is available
8420       via -mfma.
8421     * A new -mfsgsbase command-line option is available that makes GCC
8422       generate new segment register read/write instructions through
8423       dedicated built-ins.
8424     * Support for the new Intel rdrnd instruction is available via
8425       -mrdrnd.
8426     * Two additional AVX vector conversion instructions are available via
8427       -mf16c.
8428     * Support for new Intel processor codename IvyBridge with RDRND,
8429       FSGSBASE and F16C is available through -march=core-avx-i.
8430     * Support for the new Intel processor codename Haswell with AVX2,
8431       FMA, BMI, BMI2, LZCNT is available through -march=core-avx2.
8432     * Support for new AMD family 15h processors (Piledriver core) is now
8433       available through -march=bdver2 and -mtune=bdver2 options.
8434     * Support for [39]the x32 psABI is now available through the -mx32
8435       option.
8436     * Windows mingw targets are using the -mms-bitfields option by
8437       default.
8438     * Windows x86 targets are using the __thiscall calling convention for
8439       C++ class-member functions.
8440     * Support for the configure option --with-threads=posix for Windows
8441       mingw targets.
8442
8443  MIPS
8444
8445     * GCC now supports thread-local storage (TLS) for MIPS16. This
8446       requires GNU binutils 2.22 or later.
8447     * GCC can now generate code specifically for the Cavium Octeon+ and
8448       Octeon2 processors. The associated command-line options are
8449       -march=octeon+ and -march=octeon2 respectively. Both options
8450       require GNU binutils 2.22 or later.
8451     * GCC can now work around certain 24k errata, under the control of
8452       the command-line option -mfix-24k. These workarounds require GNU
8453       binutils 2.20 or later.
8454     * 32-bit MIPS GNU/Linux targets such as mips-linux-gnu can now build
8455       n32 and n64 multilibs. The result is effectively a 64-bit GNU/Linux
8456       toolchain that generates 32-bit code by default. Use the
8457       configure-time option --enable-targets=all to select these extra
8458       multilibs.
8459     * Passing -fno-delayed-branch now also stops the assembler from
8460       automatically filling delay slots.
8461
8462  PowerPC/PowerPC64
8463
8464     * Vectors of type vector long long or vector long are passed and
8465       returned using the same method as other vectors with the VSX
8466       instruction set. Previously GCC did not adhere to the ABI for
8467       128-bit vectors with 64-bit integer base types (PR 48857). This
8468       will also be fixed in the GCC 4.6.1 and 4.5.4 releases.
8469     * A new option -mno-pointers-to-nested-functions was added to allow
8470       AIX 32-bit/64-bit and GNU/Linux 64-bit PowerPC users to specify
8471       that the compiler should not load up the chain register (r11)
8472       before calling a function through a pointer. If you use this
8473       option, you cannot call nested functions through a pointer, or call
8474       other languages that might use the static chain.
8475     * A new option msave-toc-indirect was added to allow AIX
8476       32-bit/64-bit and GNU/Linux 64-bit PowerPC users control whether we
8477       save the TOC in the prologue for indirect calls or generate the
8478       save inline. This can speed up some programs that call through a
8479       function pointer a lot, but it can slow down other functions that
8480       only call through a function pointer in exceptional cases.
8481     * The PowerPC port will now enable machine-specific built-in
8482       functions when the user switches the target machine using the
8483       #pragma GCC target or __attribute__ ((__target__ ("target"))) code
8484       sequences. In addition, the target macros are updated. However, due
8485       to the way the -save-temps switch is implemented, you won't see the
8486       effect of these additional macros being defined in preprocessor
8487       output.
8488
8489  SH
8490
8491     * A new option -msoft-atomic has been added. When it is specified,
8492       GCC will generate GNU/Linux-compatible gUSA atomic sequences for
8493       the new __atomic routines.
8494     * Since it is neither supported by GAS nor officially documented,
8495       code generation for little endian SH2A has been disabled.
8496       Specifying -ml with -m2a* will now result in a compiler error.
8497     * The defunct -mbranch-cost option has been fixed.
8498     * Some improvements to the generated code of:
8499          + Utilization of the tst #imm,R0 instruction.
8500          + Dynamic shift instructions on SH2A.
8501          + Integer absolute value calculations.
8502     * The -mdiv= option for targets other than SHmedia has been fixed and
8503       documented.
8504
8505  SPARC
8506
8507     * The option -mflat has been reinstated. When it is specified, the
8508       compiler will generate code for a single register window model.
8509       This is essentially a new implementation and the corresponding
8510       debugger support has been added to GDB 7.4.
8511     * Support for the options -mtune=native and -mcpu=native has been
8512       added on selected native platforms (GNU/Linux and Solaris).
8513     * Support for the SPARC T3 (Niagara 3) processor has been added.
8514     * VIS:
8515          + An intrinsics header visintrin.h has been added.
8516          + Builtin intrinsics for the VIS 1.0 edge handling and pixel
8517            compare instructions have been added.
8518          + The little-endian version of alignaddr is now supported.
8519          + When possible, VIS builtins are marked const, which should
8520            increase the compiler's ability to optimize VIS operations.
8521          + The compiler now properly tracks the %gsr register and how it
8522            behaves as an input for various VIS instructions.
8523          + Akin to fzero, the compiler can now generate fone instructions
8524            in order to set all of the bits of a floating-point register
8525            to 1.
8526          + The documentation for the VIS intrinsics in the GCC manual has
8527            been brought up to date and many inaccuracies were fixed.
8528          + Intrinsics for the VIS 2.0 bmask, bshuffle, and
8529            non-condition-code setting edge instructions have been added.
8530            Their availability is controlled by the new -mvis2 and
8531            -mno-vis2 options. They are enabled by default on
8532            UltraSPARC-III and later CPUs.
8533     * Support for UltraSPARC Fused Multiply-Add floating-point extensions
8534       has been added. These instructions are enabled by default on SPARC
8535       T3 (Niagara 3) and later CPUs.
8536
8537  TILE-Gx/TILEPro
8538
8539     * Support has been added for the Tilera TILE-Gx and TILEPro families
8540       of processors.
8541
8542Other significant improvements
8543
8544     * A new option (-grecord-gcc-switches) was added that appends
8545       compiler command-line options that might affect code generation to
8546       the DW_AT_producer attribute string in the DWARF debugging
8547       information.
8548     * GCC now supports various new GNU extensions to the DWARF debugging
8549       information format, like [40]entry value and [41]call site
8550       information, [42]typed DWARF stack or [43]a more compact macro
8551       representation. Support for these extensions has been added to GDB
8552       7.4. They can be disabled through the -gstrict-dwarf command-line
8553       option.
8554
8555GCC 4.7.1
8556
8557   This is the [44]list of problem reports (PRs) from GCC's bug tracking
8558   system that are known to be fixed in the 4.7.1 release. This list might
8559   not be complete (that is, it is possible that some PRs that have been
8560   fixed are not listed here).
8561
8562   The Go front end in the 4.7.1 release fully supports the [45]Go 1
8563   language standard.
8564
8565GCC 4.7.2
8566
8567   This is the [46]list of problem reports (PRs) from GCC's bug tracking
8568   system that are known to be fixed in the 4.7.2 release. This list might
8569   not be complete (that is, it is possible that some PRs that have been
8570   fixed are not listed here).
8571
8572GCC 4.7.3
8573
8574   This is the [47]list of problem reports (PRs) from GCC's bug tracking
8575   system that are known to be fixed in the 4.7.3 release. This list might
8576   not be complete (that is, it is possible that some PRs that have been
8577   fixed are not listed here).
8578
8579GCC 4.7.4
8580
8581   This is the [48]list of problem reports (PRs) from GCC's bug tracking
8582   system that are known to be fixed in the 4.7.4 release. This list might
8583   not be complete (that is, it is possible that some PRs that have been
8584   fixed are not listed here).
8585
8586
8587    For questions related to the use of GCC, please consult these web
8588    pages and the [49]GCC manuals. If that fails, the
8589    [50]gcc-help@gcc.gnu.org mailing list might help. Comments on these
8590    web pages and the development of GCC are welcome on our developer
8591    list at [51]gcc@gcc.gnu.org. All of [52]our lists have public
8592    archives.
8593
8594   Copyright (C) [53]Free Software Foundation, Inc. Verbatim copying and
8595   distribution of this entire article is permitted in any medium,
8596   provided this notice is preserved.
8597
8598   These pages are [54]maintained by the GCC team. Last modified
8599   2019-11-28[55].
8600
8601References
8602
8603   1. https://gcc.gnu.org/ml/gcc-patches/2011-03/msg01263.html
8604   2. http://savannah.nongnu.org/bugs/?35407
8605   3. https://gcc.gnu.org/PR18145
8606   4. https://gcc.gnu.org/gcc-4.7/porting_to.html
8607   5. https://gcc.gnu.org/wiki/TransactionalMemory
8608   6. https://gcc.gnu.org/wiki/Atomic/GCCMM
8609   7. http://gcc.gnu.org/gcc-4.7/cxx0x_status.html
8610   8. http://gcc.gnu.org/gcc-4.7/cxx0x_status.html
8611   9. http://gcc.gnu.org/gcc-4.7/cxx0x_status.html
8612  10. http://gcc.gnu.org/gcc-4.7/cxx0x_status.html
8613  11. http://gcc.gnu.org/gcc-4.7/cxx0x_status.html
8614  12. http://gcc.gnu.org/gcc-4.7/cxx0x_status.html
8615  13. https://gcc.gnu.org/PR14258
8616  14. https://gcc.gnu.org/PR35688
8617  15. https://gcc.gnu.org/onlinedocs/gcc-4.7.4/libstdc++/manual/manual/status.html#status.iso.2011
8618  16. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/Code-Gen-Options.html#index-g_t_0040code_007bfstack-arrays_007d-254
8619  17. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gcc/Optimize-Options.html#index-Ofast-689
8620  18. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/Code-Gen-Options.html#index-g_t_0040code_007bfno-protect-parens_007d-270
8621  19. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/Code-Gen-Options.html#index-g_t_0040code_007bfstack-arrays_007d-254
8622  20. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/Code-Gen-Options.html#index-g_t_0040code_007bfrontend-optimize_007d-275
8623  21. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/Error-and-Warning-Options.html#index-g_t_0040code_007bWfunction-elimination_007d-170
8624  22. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/Code-Gen-Options.html#index-g_t_0040code_007bfaggressive-function-elimination_007d-270
8625  23. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/Error-and-Warning-Options.html#index-g_t_0040code_007bWreal-q-constant_007d-149
8626  24. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/SELECTED_005fREAL_005fKIND.html
8627  25. https://docs.microsoft.com/en-us/windows/win32/wer/collecting-user-mode-dumps
8628  26. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/Debugging-Options.html#index-g_t_0040code_007bfno-backtrace_007d-183
8629  27. https://gcc.gnu.org/wiki/Fortran2003Status
8630  28. https://gcc.gnu.org/wiki/OOP
8631  29. https://gcc.gnu.org/wiki/Fortran2008Status
8632  30. https://gcc.gnu.org/wiki/Coarray
8633  31. https://gcc.gnu.org/wiki/CoarrayLib
8634  32. https://gcc.gnu.org/wiki/TS29113Status
8635  33. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gfortran/Fortran-Dialect-Options.html#index-g_t_0040code_007bstd_003d_007d_0040var_007bstd_007d-option-53
8636  34. https://golang.org/doc/go1
8637  35. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gcc/Named-Address-Spaces.html
8638  36. http://nongnu.org/avr-libc/
8639  37. https://gcc.gnu.org/PR54461
8640  38. https://gcc.gnu.org/onlinedocs/gcc-4.7.1/gcc/AVR-Built%5f002din-Functions.html
8641  39. https://sites.google.com/site/x32abi/
8642  40. http://www.dwarfstd.org/ShowIssue.php?issue=100909.1
8643  41. http://www.dwarfstd.org/ShowIssue.php?issue=100909.2
8644  42. http://www.dwarfstd.org/ShowIssue.php?issue=140425.1
8645  43. http://www.dwarfstd.org/ShowIssue.php?issue=110722.1
8646  44. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.7.1
8647  45. https://golang.org/doc/go1
8648  46. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.7.2
8649  47. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.7.3
8650  48. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.7.4
8651  49. https://gcc.gnu.org/onlinedocs/
8652  50. mailto:gcc-help@gcc.gnu.org
8653  51. mailto:gcc@gcc.gnu.org
8654  52. https://gcc.gnu.org/lists.html
8655  53. https://www.fsf.org/
8656  54. https://gcc.gnu.org/about.html
8657  55. http://validator.w3.org/check/referer
8658======================================================================
8659http://gcc.gnu.org/gcc-4.6/index.html
8660
8661                             GCC 4.6 Release Series
8662
8663   (This release series is no longer supported.)
8664
8665   April 12, 2013
8666
8667   The [1]GNU project and the GCC developers are pleased to announce the
8668   release of GCC 4.6.4.
8669
8670   This release is a bug-fix release, containing fixes for regressions in
8671   GCC 4.6.3 relative to previous releases of GCC.
8672
8673Release History
8674
8675   GCC 4.6.4
8676          April 12, 2013 ([2]changes, [3]documentation)
8677
8678   GCC 4.6.3
8679          March 1, 2012 ([4]changes, [5]documentation)
8680
8681   GCC 4.6.2
8682          October 26, 2011 ([6]changes, [7]documentation)
8683
8684   GCC 4.6.1
8685          June 27, 2011 ([8]changes, [9]documentation)
8686
8687   GCC 4.6.0
8688          March 25, 2011 ([10]changes, [11]documentation)
8689
8690References and Acknowledgements
8691
8692   GCC used to stand for the GNU C Compiler, but since the compiler
8693   supports several other languages aside from C, it now stands for the
8694   GNU Compiler Collection.
8695
8696   A list of [12]successful builds is updated as new information becomes
8697   available.
8698
8699   The GCC developers would like to thank the numerous people that have
8700   contributed new features, improvements, bug fixes, and other changes as
8701   well as test results to GCC. This [13]amazing group of volunteers is
8702   what makes GCC successful.
8703
8704   For additional information about GCC please refer to the [14]GCC
8705   project web site or contact the [15]GCC development mailing list.
8706
8707   To obtain GCC please use [16]our mirror sites or [17]our version
8708   control system.
8709
8710
8711    For questions related to the use of GCC, please consult these web
8712    pages and the [18]GCC manuals. If that fails, the
8713    [19]gcc-help@gcc.gnu.org mailing list might help. Comments on these
8714    web pages and the development of GCC are welcome on our developer
8715    list at [20]gcc@gcc.gnu.org. All of [21]our lists have public
8716    archives.
8717
8718   Copyright (C) [22]Free Software Foundation, Inc. Verbatim copying and
8719   distribution of this entire article is permitted in any medium,
8720   provided this notice is preserved.
8721
8722   These pages are [23]maintained by the GCC team. Last modified
8723   2020-01-14[24].
8724
8725References
8726
8727   1. http://www.gnu.org/
8728   2. http://gcc.gnu.org/gcc-4.6/changes.html
8729   3. https://gcc.gnu.org/onlinedocs/4.6.4/
8730   4. http://gcc.gnu.org/gcc-4.6/changes.html
8731   5. https://gcc.gnu.org/onlinedocs/4.6.3/
8732   6. http://gcc.gnu.org/gcc-4.6/changes.html
8733   7. https://gcc.gnu.org/onlinedocs/4.6.2/
8734   8. http://gcc.gnu.org/gcc-4.6/changes.html
8735   9. https://gcc.gnu.org/onlinedocs/4.6.1/
8736  10. http://gcc.gnu.org/gcc-4.6/changes.html
8737  11. https://gcc.gnu.org/onlinedocs/4.6.0/
8738  12. http://gcc.gnu.org/gcc-4.6/buildstat.html
8739  13. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
8740  14. http://gcc.gnu.org/index.html
8741  15. mailto:gcc@gcc.gnu.org
8742  16. http://gcc.gnu.org/mirrors.html
8743  17. http://gcc.gnu.org/git.html
8744  18. https://gcc.gnu.org/onlinedocs/
8745  19. mailto:gcc-help@gcc.gnu.org
8746  20. mailto:gcc@gcc.gnu.org
8747  21. https://gcc.gnu.org/lists.html
8748  22. https://www.fsf.org/
8749  23. https://gcc.gnu.org/about.html
8750  24. http://validator.w3.org/check/referer
8751======================================================================
8752http://gcc.gnu.org/gcc-4.6/changes.html
8753
8754                             GCC 4.6 Release Series
8755                        Changes, New Features, and Fixes
8756
8757Caveats
8758
8759     * The options -b <machine> and -V <version> have been removed because
8760       they were unreliable. Instead, users should directly run
8761       <machine>-gcc when cross-compiling, or <machine>-gcc-<version> to
8762       run a different version of gcc.
8763     * GCC now has stricter checks for invalid command-line options. In
8764       particular, when gcc was called to link object files rather than
8765       compile source code, it would previously accept and ignore all
8766       options starting with --, including linker options such as
8767       --as-needed and --export-dynamic, although such options would
8768       result in errors if any source code was compiled. Such options, if
8769       unknown to the compiler, are now rejected in all cases; if the
8770       intent was to pass them to the linker, options such as
8771       -Wl,--as-needed should be used.
8772     * Versions of the GNU C library up to and including 2.11.1 included
8773       an [1]incorrect implementation of the cproj function. GCC optimizes
8774       its builtin cproj according to the behavior specified and allowed
8775       by the ISO C99 standard. If you want to avoid discrepancies between
8776       the C library and GCC's builtin transformations when using cproj in
8777       your code, use GLIBC 2.12 or later. If you are using an older GLIBC
8778       and actually rely on the incorrect behavior of cproj, then you can
8779       disable GCC's transformations using -fno-builtin-cproj.
8780     * The C-only intermodule optimization framework (IMA, enabled by
8781       -combine) has been removed in favor of the new generic link-time
8782       optimization framework (LTO) introduced in [2]GCC 4.5.0.
8783     * GCC now ships with the LGPL-licensed libquadmath library, which
8784       provides quad-precision mathematical functions for targets with a
8785       __float128 datatype. __float128 is available for targets on 32-bit
8786       x86, x86-64 and Itanium architectures. The libquadmath library is
8787       automatically built on such targets when building the Fortran
8788       compiler.
8789     * New -Wunused-but-set-variable and -Wunused-but-set-parameter
8790       warnings were added for C, C++, Objective-C and Objective-C++.
8791       These warnings diagnose variables respective parameters which are
8792       only set in the code and never otherwise used. Usually such
8793       variables are useless and often even the value assigned to them is
8794       computed needlessly, sometimes expensively. The
8795       -Wunused-but-set-variable warning is enabled by default by -Wall
8796       flag and -Wunused-but-set-parameter by -Wall -Wextra flags.
8797     * On ARM, a bug has been fixed in GCC's implementation of the AAPCS
8798       rules for the layout of vectors that could lead to wrong code being
8799       generated. Vectors larger than 8 bytes in size are now by default
8800       aligned to an 8-byte boundary. This is an ABI change: code that
8801       makes explicit use of vector types may be incompatible with binary
8802       objects built with older versions of GCC. Auto-vectorized code is
8803       not affected by this change. (This change affects GCC versions
8804       4.6.4 and later, with the exception of versions 4.7.0 and 4.7.1.)
8805     * On AVR, variables with the progmem attribute to locate data in
8806       flash memory must be qualified as const.
8807     * Support for a number of older systems and recently unmaintained or
8808       untested target ports of GCC has been declared obsolete in GCC 4.6.
8809       Unless there is activity to revive them, the next release of GCC
8810       will have their sources permanently removed.
8811       All GCC ports for the following processor architectures have been
8812       declared obsolete:
8813          + Argonaut ARC (arc-*)
8814          + National Semiconductor CRX (crx-*)
8815          + Motorola 68HC11 and 68HC12 (m68hc11-*-*, m6811-*-*,
8816            m68hc12-*-*, m6812-*-*)
8817          + Sunplus S+core (score-*)
8818       The following ports for individual systems on particular
8819       architectures have been obsoleted:
8820          + Interix (i[34567]86-*-interix3*)
8821          + NetWare x86 (i[3456x]86-*-netware*)
8822          + Generic ARM PE (arm-*-pe* other than arm*-wince-pe*)
8823          + MCore PE (mcore-*-pe*)
8824          + SH SymbianOS (sh*-*-symbianelf*)
8825          + GNU Hurd on Alpha and PowerPC (alpha*-*-gnu*, powerpc*-*-gnu*)
8826          + M68K uClinux old ABI (m68k-*-uclinuxoldabi*)
8827          + a.out NetBSD (arm*-*-netbsd*, i[34567]86-*-netbsd*,
8828            vax-*-netbsd*, but not *-*-netbsdelf*)
8829       The i[34567]86-*-pe alias for Cygwin targets has also been
8830       obsoleted; users should configure for i[34567]86-*-cygwin* instead.
8831       Certain configure options to control the set of libraries built
8832       with GCC on some targets have been obsoleted. On ARM targets, the
8833       options --disable-fpu, --disable-26bit, --disable-underscore,
8834       --disable-interwork, --disable-biendian and --disable-nofmult have
8835       been obsoleted. On MIPS targets, the options
8836       --disable-single-float, --disable-biendian and --disable-softfloat
8837       have been obsoleted.
8838     * Support has been removed for all the [3]configurations obsoleted in
8839       GCC 4.5.
8840     * More information on porting to GCC 4.6 from previous versions of
8841       GCC can be found in the [4]porting guide for this release.
8842
8843General Optimizer Improvements
8844
8845     * A new general optimization level, -Ofast, has been introduced. It
8846       combines the existing optimization level -O3 with options that can
8847       affect standards compliance but result in better optimized code.
8848       For example, -Ofast enables -ffast-math.
8849     * Link-time optimization improvements:
8850          + The [5]Scalable Whole Program Optimizer (WHOPR) project has
8851            stabilized to the point of being usable. It has become the
8852            default mode when using the LTO optimization model. Link time
8853            optimization can now split itself into multiple parallel
8854            compilations. Parallelism is controlled with -flto=n (where n
8855            specifies the number of compilations to execute in parallel).
8856            GCC can also cooperate with a GNU make job server by
8857            specifying the -flto=jobserver option and adding + to the
8858            beginning of the Makefile rule executing the linker.
8859            Classical LTO mode can be enforced by -flto-partition=none.
8860            This may result in small code quality improvements.
8861          + A large number of bugs were fixed. GCC itself, Mozilla Firefox
8862            and other large applications can be built with LTO enabled.
8863          + The linker plugin support improvements
8864               o Linker plugin is now enabled by default when the linker
8865                 is detected to have plugin support. This is the case for
8866                 GNU ld 2.21.51 or newer (on ELF and Cygwin targets) and
8867                 the Gold linker on ELF targets. Plugin support of the
8868                 Apple linker on Darwin is not compatible with GCC. The
8869                 linker plugin can also be controlled by the
8870                 -fuse-linker-plugin command-line option.
8871               o Resolution information from the linker plugin is used to
8872                 drive whole program assumptions. Use of the linker plugin
8873                 results in more aggressive optimization on binaries and
8874                 on shared libraries that use the hidden visibility
8875                 attribute. Consequently the use of -fwhole-program is not
8876                 necessary in addition to LTO.
8877          + Hidden symbols used from non-LTO objects now have to be
8878            explicitly annotated with externally_visible when the linker
8879            plugin is not used.
8880          + C++ inline functions and virtual tables are now privatized
8881            more aggressively, leading to better inter-procedural
8882            optimization and faster dynamic linking.
8883          + Memory usage and intermediate language streaming performance
8884            have been improved.
8885          + Static constructors and destructors from individual units are
8886            inlined into a single function. This can significantly improve
8887            startup times of large C++ applications where static
8888            constructors are very common. For example, static constructors
8889            are used when including the iostream header.
8890          + Support for the Ada language has been added.
8891     * Interprocedural optimization improvements
8892          + The interprocedural framework was re-tuned for link time
8893            optimization. Several scalability issues were resolved.
8894          + Improved auto-detection of const and pure functions. Newly,
8895            noreturn functions are auto-detected.
8896            The [6]-Wsuggest-attribute=[const|pure|noreturn] flag is
8897            available that informs users when adding attributes to headers
8898            might improve code generation.
8899          + A number of inlining heuristic improvements. In particular:
8900               o Partial inlining is now supported and enabled by default
8901                 at -O2 and greater. The feature can be controlled via
8902                 -fpartial-inlining.
8903                 Partial inlining splits functions with short hot path to
8904                 return. This allows more aggressive inlining of the hot
8905                 path leading to better performance and often to code size
8906                 reductions (because cold parts of functions are not
8907                 duplicated).
8908               o Scalability for large compilation units was improved
8909                 significantly.
8910               o Inlining of callbacks is now more aggressive.
8911               o Virtual methods are considered for inlining when the
8912                 caller is inlined and devirtualization is then possible.
8913               o Inlining when optimizing for size (either in cold regions
8914                 of a program or when compiling with -Os) was improved to
8915                 better handle C++ programs with larger abstraction
8916                 penalty, leading to smaller and faster code.
8917          + The IPA reference optimization pass detecting global variables
8918            used or modified by functions was strengthened and sped up.
8919          + Functions whose address was taken are now optimized out when
8920            all references to them are dead.
8921          + A new inter-procedural static profile estimation pass detects
8922            functions that are executed once or unlikely to be executed.
8923            Unlikely executed functions are optimized for size. Functions
8924            executed once are optimized for size except for the inner
8925            loops.
8926          + On most targets with named section support, functions used
8927            only at startup (static constructors and main), functions used
8928            only at exit and functions detected to be cold are placed into
8929            separate text segment subsections. This extends the
8930            -freorder-functions feature and is controlled by the same
8931            switch. The goal is to improve the startup time of large C++
8932            programs.
8933            Proper function placement requires linker support. GNU ld
8934            2.21.51 on ELF targets was updated to place those functions
8935            together within the text section leading to better code
8936            locality and faster startup times of large C++ programs. The
8937            feature is also supported in the Apple linker. Support in the
8938            gold linker is planned.
8939     * A new switch -fstack-usage has been added. It makes the compiler
8940       output stack usage information for the program, on a per-function
8941       basis, in an auxiliary file.
8942     * A new switch -fcombine-stack-adjustments has been added. It can be
8943       used to enable or disable the compiler's stack-slot combining pass
8944       which before was enabled automatically at -O1 and above, but could
8945       not be controlled on its own.
8946     * A new switch -fstrict-volatile-bitfields has been added. Using it
8947       indicates that accesses to volatile bitfields should use a single
8948       access of the width of the field's type. This option can be useful
8949       for precisely defining and accessing memory-mapped peripheral
8950       registers from C or C++.
8951
8952Compile time and memory usage improvements
8953
8954     * Datastructures used by the dataflow framework in GCC were
8955       reorganized for better memory usage and more cache locality.
8956       Compile time is improved especially on units with large functions
8957       (possibly resulting from a lot of inlining) not fitting into the
8958       processor cache. The compile time of the GCC C compiler binary with
8959       link-time optimization went down by over 10% (benchmarked on x86-64
8960       target).
8961
8962New Languages and Language specific improvements
8963
8964  Ada
8965
8966     * Stack checking has been improved on selected architectures (Alpha,
8967       IA-32/x86-64, RS/6000 and SPARC): it now will detect stack
8968       overflows in all cases on these architectures.
8969     * Initial support for Ada 2012 has been added.
8970
8971  C family
8972
8973     * A new warning, enabled by -Wdouble-promotion, has been added that
8974       warns about cases where a value of type float is implicitly
8975       promoted to double. This is especially helpful for CPUs that handle
8976       the former in hardware, but emulate the latter in software.
8977     * A new function attribute leaf was introduced. This attribute allows
8978       better inter-procedural optimization across calls to functions that
8979       return to the current unit only via returning or exception
8980       handling. This is the case for most library functions that have no
8981       callbacks.
8982     * Support for a new data type __int128 for targets having wide enough
8983       machine-mode support.
8984     * The new function attribute callee_pop_aggregate allows to specify
8985       if the caller or callee is responsible for popping the aggregate
8986       return pointer value from the stack.
8987     * Support for selectively enabling and disabling warnings via #pragma
8988       GCC diagnostic has been added. For instance:
8989#pragma GCC diagnostic error "-Wuninitialized"
8990  foo(a);                       /* error is given for this one */
8991#pragma GCC diagnostic push
8992#pragma GCC diagnostic ignored "-Wuninitialized"
8993  foo(b);                       /* no diagnostic for this one */
8994#pragma GCC diagnostic pop
8995  foo(c);                       /* error is given for this one */
8996#pragma GCC diagnostic pop
8997  foo(d);                       /* depends on command-line options */
8998
8999     * The -fmax-errors=N option is now supported. Using this option
9000       causes the compiler to exit after N errors have been issued.
9001
9002  C
9003
9004     * There is now experimental support for some features from the
9005       upcoming C1X revision of the ISO C standard. This support may be
9006       selected with -std=c1x, or -std=gnu1x for C1X with GNU extensions.
9007       Note that this support is experimental and may change incompatibly
9008       in future releases for consistency with changes to the C1X standard
9009       draft. The following features are newly supported as described in
9010       the N1539 draft of C1X (with changes agreed at the March 2011 WG14
9011       meeting); some other features were already supported with no
9012       compiler changes being needed, or have some support but not in full
9013       accord with N1539 (as amended).
9014          + Static assertions (_Static_assert keyword)
9015          + Typedef redefinition
9016          + New macros in <float.h>
9017          + Anonymous structures and unions
9018     * The new -fplan9-extensions option directs the compiler to support
9019       some extensions for anonymous struct fields which are implemented
9020       by the Plan 9 compiler. A pointer to a struct may be automatically
9021       converted to a pointer to an anonymous field when calling a
9022       function, in order to make the types match. An anonymous struct
9023       field whose type is a typedef name may be referred to using the
9024       typedef name.
9025
9026  C++
9027
9028     * Improved [7]experimental support for the upcoming C++0x ISO C++
9029       standard, including support for constexpr (thanks to Gabriel Dos
9030       Reis and Jason Merrill), nullptr (thanks to Magnus Fromreide),
9031       noexcept, unrestricted unions, range-based for loops (thanks to
9032       Rodrigo Rivas Costa), opaque enum declarations (thanks also to
9033       Rodrigo), implicitly deleted functions and implicit move
9034       constructors.
9035     * When an extern declaration within a function does not match a
9036       declaration in the enclosing context, G++ now properly declares the
9037       name within the namespace of the function rather than the namespace
9038       which was open just before the function definition ([8]c++/43145).
9039     * GCC now warns by default when casting integers to larger pointer
9040       types. These warnings can be disabled with the option
9041       -Wno-int-to-pointer-cast, which is now also available in C++.
9042     * G++ no longer optimizes using the assumption that a value of
9043       enumeration type will fall within the range specified by the
9044       standard, since that assumption is easily violated with a
9045       conversion from integer type ([9]c++/43680). The old behavior can
9046       be restored with -fstrict-enums.
9047     * The new -fnothrow-opt flag changes the semantics of a throw()
9048       exception specification to match the proposed semantics of the
9049       noexcept specification: just call terminate if an exception tries
9050       to propagate out of a function with such an exception
9051       specification. This dramatically reduces or eliminates the code
9052       size overhead from adding the exception specification.
9053     * The new -Wnoexcept flag will suggest adding a noexcept qualifier to
9054       a function that the compiler can tell doesn't throw if it would
9055       change the value of a noexcept expression.
9056     * The -Wshadow option now warns if a local variable or type
9057       declaration shadows another type in C++. Note that the compiler
9058       will not warn if a local variable shadows a struct/class/enum, but
9059       will warn if it shadows an explicit typedef.
9060     * When an identifier is not found in the current scope, G++ now
9061       offers suggestions about which identifier might have been intended.
9062     * G++ now issues clearer diagnostics for missing semicolons after
9063       class, struct, and union definitions.
9064     * G++ now issues clearer diagnostics for missing semicolons after
9065       class member declarations.
9066     * G++ now issues clearer diagnostics when a colon is used in a place
9067       where a double-colon was intended.
9068     * G++ no longer accepts mutable on reference members ([10]c++/33558).
9069       Use -fpermissive to allow the old, non-conforming behaviour.
9070     * A few mangling fixes have been made, to attribute const/volatile on
9071       function pointer types, decltype of a plain decl, and use of a
9072       function parameter in the declaration of another parameter. By
9073       default the compiler still uses the old mangling, but emits aliases
9074       with the new mangling on targets that support strong aliases. Users
9075       can switch over entirely to the new mangling with -fabi-version=5
9076       or -fabi-version=0. -Wabi will now warn about code that uses the
9077       old mangling.
9078     * In 4.6.0 and 4.6.1 G++ no longer allows objects of const-qualified
9079       type to be default initialized unless the type has a user-declared
9080       default constructor. In 4.6.2 G++ implements the proposed
9081       resolution of [11]DR 253, so default initialization is allowed if
9082       it initializes all subobjects. Code that fails to compile can be
9083       fixed by providing an initializer e.g.
9084    struct A { A(); };
9085    struct B : A { int i; };
9086    const B b = B();
9087       Use -fpermissive to allow the old, non-conforming behaviour.
9088
9089    Runtime Library (libstdc++)
9090
9091     * [12]Improved experimental support for the upcoming ISO C++
9092       standard, C++0x, including using constexpr and nullptr.
9093     * Performance improvements to the [13]Debug Mode, thanks to Fran�ois
9094       Dumont.
9095     * Atomic operations used for reference-counting are annotated so that
9096       they can be understood by race detectors such as Helgrind, see
9097       [14]Data Race Hunting.
9098     * Most libstdc++ standard headers have been changed to no longer
9099       include the cstddef header as an implementation detail. Code that
9100       relied on that header being included as side-effect of including
9101       other standard headers will need to include cstddef explicitly.
9102
9103  Fortran
9104
9105     * On systems supporting the libquadmath library, GNU Fortran now also
9106       supports a quad-precision, kind=16 floating-point data type
9107       (REAL(16), COMPLEX(16)). As the data type is not fully supported in
9108       hardware, calculations might be one to two orders of magnitude
9109       slower than with the 4, 8 or 10 bytes floating-point data types.
9110       This change does not affect systems which support REAL(16) in
9111       hardware nor those which do not support libquadmath.
9112     * Much improved compile time for large array constructors.
9113     * In order to reduce execution time and memory consumption, use of
9114       temporary arrays in assignment expressions is avoided for many
9115       cases. The compiler now reverses loops in order to avoid generating
9116       a temporary array where possible.
9117     * Improved diagnostics, especially with -fwhole-file.
9118     * The -fwhole-file flag is now enabled by default. This improves code
9119       generation and diagnostics. It can be disabled using the deprecated
9120       -fno-whole-file flag.
9121     * Support the generation of Makefile dependencies via the [15]-M...
9122       flags of GCC; you may need to specify the -cpp option in addition.
9123       The dependencies take modules, Fortran's include, and CPP's
9124       #include into account. Note: Using -M for the module path is no
9125       longer supported, use -J instead.
9126     * The flag -Wconversion has been modified to only issue warnings
9127       where a conversion leads to information loss. This drastically
9128       reduces the number of warnings; -Wconversion is thus now enabled
9129       with -Wall. The flag -Wconversion-extra has been added and also
9130       warns about other conversions; -Wconversion-extra typically issues
9131       a huge number of warnings, most of which can be ignored.
9132     * A new command-line option -Wunused-dummy-argument warns about
9133       unused dummy arguments and is included in -Wall. Before,
9134       -Wunused-variable also warned about unused dummy arguments.
9135     * Fortran 2003 support has been extended:
9136          + Improved support for polymorphism between libraries and
9137            programs and for complicated inheritance patterns (cf.
9138            [16]object-oriented programming).
9139          + Experimental support of the ASSOCIATE construct.
9140          + In pointer assignments it is now possible to specify the lower
9141            bounds of the pointer and, for a rank-1 or a simply contiguous
9142            data-target, to remap the bounds.
9143          + Automatic (re)allocation: In intrinsic assignments to
9144            allocatable variables the left-hand side will be automatically
9145            allocated (if unallocated) or reallocated (if the shape or
9146            type parameter is different). To avoid the small performance
9147            penalty, you can use a(:) = ... instead of a = ... for arrays
9148            and character strings - or disable the feature using -std=f95
9149            or -fno-realloc-lhs.
9150          + Deferred type parameter: For scalar allocatable and pointer
9151            variables the character length can be deferred.
9152          + Namelist variables with allocatable and pointer attribute and
9153            nonconstant length type parameter are supported.
9154     * Fortran 2008 support has been extended:
9155          + Experimental [17]coarray support (for one image only, i.e.
9156            num_images() == 1); use the [18]-fcoarray=single flag to
9157            enable it.
9158          + The STOP and the new ERROR STOP statements now support all
9159            constant expressions.
9160          + Support for the CONTIGUOUS attribute.
9161          + Support for ALLOCATE with MOLD.
9162          + Support for the STORAGE_SIZE intrinsic inquiry function.
9163          + Support of the NORM2 and PARITY intrinsic functions.
9164          + The following bit intrinsics were added: POPCNT and POPPAR for
9165            counting the number of 1 bits and returning the parity; BGE,
9166            BGT, BLE, and BLT for bitwise comparisons; DSHIFTL and DSHIFTR
9167            for combined left and right shifts, MASKL and MASKR for simple
9168            left and right justified masks, MERGE_BITS for a bitwise merge
9169            using a mask, SHIFTA, SHIFTL and SHIFTR for shift operations,
9170            and the transformational bit intrinsics IALL, IANY and
9171            IPARITY.
9172          + Support of the EXECUTE_COMMAND_LINE intrinsic subroutine.
9173          + Support for the IMPURE attribute for procedures, which allows
9174            for ELEMENTAL procedures without the restrictions of PURE.
9175          + Null pointers (including NULL()) and not allocated variables
9176            can be used as actual argument to optional non-pointer,
9177            non-allocatable dummy arguments, denoting an absent argument.
9178          + Non-pointer variables with TARGET attribute can be used as
9179            actual argument to POINTER dummies with INTENT(IN)
9180          + Pointers including procedure pointers and those in a derived
9181            type (pointer components) can now be initialized by a target
9182            instead of only by NULL.
9183          + The EXIT statement (with construct-name) can now be used to
9184            leave not only the DO but also the ASSOCIATE, BLOCK, IF,
9185            SELECT CASE and SELECT TYPE constructs.
9186          + Internal procedures can now be used as actual argument.
9187          + The named constants INTEGER_KINDS, LOGICAL_KINDS, REAL_KINDS
9188            and CHARACTER_KINDS of the intrinsic module ISO_FORTRAN_ENV
9189            have been added; these arrays contain the supported kind
9190            values for the respective types.
9191          + The module procedures C_SIZEOF of the intrinsic module
9192            ISO_C_BINDINGS and COMPILER_VERSION and COMPILER_OPTIONS of
9193            ISO_FORTRAN_ENV have been implemented.
9194          + Minor changes: obsolescence diagnostics for ENTRY was added
9195            for -std=f2008; a line may start with a semicolon; for
9196            internal and module procedures END can be used instead of END
9197            SUBROUTINE and END FUNCTION; SELECTED_REAL_KIND now also takes
9198            a RADIX argument; intrinsic types are supported for
9199            TYPE(intrinsic-type-spec); multiple type-bound procedures can
9200            be declared in a single PROCEDURE statement; implied-shape
9201            arrays are supported for named constants (PARAMETER). The
9202            transformational, three argument versions of BESSEL_JN and
9203            BESSEL_YN were added - the elemental, two-argument version had
9204            been added in GCC 4.4; note that the transformational
9205            functions use a recurrence algorithm.
9206
9207  Go
9208
9209   Support for the [19]Go programming language has been added to GCC. It
9210   is not enabled by default when you build GCC; use the
9211   --enable-languages configure option to build it. The driver program for
9212   compiling Go code is gccgo.
9213
9214   Go is currently known to work on GNU/Linux and RTEMS. Solaris support
9215   is in progress. It may or may not work on other platforms.
9216
9217  Objective-C and Objective-C++
9218
9219     * The -fobjc-exceptions flag is now required to enable Objective-C
9220       exception and synchronization syntax (introduced by the keywords
9221       @try, @catch, @finally and @synchronized).
9222     * A number of Objective-C 2.0 features and extensions are now
9223       supported by GCC. These features are enabled by default; you can
9224       disable them by using the new -fobjc-std=objc1 command-line option.
9225     * The Objective-C 2.0 dot-syntax is now supported. It is an
9226       alternative syntax for using getters and setters; object.count is
9227       automatically converted into [object count] or [object setCount:
9228       ...] depending on context; for example if (object.count > 0) is
9229       automatically compiled into the equivalent of if ([object count] >
9230       0) while object.count = 0; is automatically compiled into the
9231       equivalent ot [object setCount: 0];. The dot-syntax can be used
9232       with instance and class objects and with any setters or getters, no
9233       matter if they are part of a declared property or not.
9234     * Objective-C 2.0 declared properties are now supported. They are
9235       declared using the new @property keyword, and are most commonly
9236       used in conjunction with the new Objective-C 2.0 dot-syntax. The
9237       nonatomic, readonly, readwrite, assign, retain, copy, setter and
9238       getter attributes are all supported. Marking declared properties
9239       with __attribute__ ((deprecated)) is supported too.
9240     * The Objective-C 2.0 @synthesize and @dynamic keywords are
9241       supported. @synthesize causes the compiler to automatically
9242       synthesize a declared property, while @dynamic is used to disable
9243       all warnings for a declared property for which no implementation is
9244       provided at compile time. Synthesizing declared properties requires
9245       runtime support in most useful cases; to be able to use it with the
9246       GNU runtime, appropriate helper functions have been added to the
9247       GNU Objective-C runtime ABI, and are implemented by the GNU
9248       Objective-C runtime library shipped with GCC.
9249     * The Objective-C 2.0 fast enumeration syntax is supported in
9250       Objective-C. This is currently not yet available in Objective-C++.
9251       Fast enumeration requires support in the runtime, and such support
9252       has been added to the GNU Objective-C runtime library (shipped with
9253       GCC).
9254     * The Objective-C 2.0 @optional keyword is supported. It allows you
9255       to mark methods or properties in a protocol as optional as opposed
9256       to required.
9257     * The Objective-C 2.0 @package keyword is supported. It has currently
9258       the same effect as the @public keyword.
9259     * Objective-C 2.0 method attributes are supported. Currently the
9260       supported attributes are deprecated, sentinel, noreturn and format.
9261     * Objective-C 2.0 method argument attributes are supported. The most
9262       widely used attribute is unused, to mark an argument as unused in
9263       the implementation.
9264     * Objective-C 2.0 class and protocol attributes are supported.
9265       Currently the only supported attribute is deprecated.
9266     * Objective-C 2.0 class extensions are supported. A class extension
9267       has the same syntax as a category declaration with no category
9268       name, and the methods and properties declared in it are added
9269       directly to the main class. It is mostly used as an alternative to
9270       a category to add methods to a class without advertising them in
9271       the public headers, with the advantage that for class extensions
9272       the compiler checks that all the privately declared methods are
9273       actually implemented.
9274     * As a result of these enhancements, GCC can now be used to build
9275       Objective-C and Objective-C++ software that uses Foundation and
9276       other important system frameworks with the NeXT runtime on Darwin 9
9277       and Darwin 10 (OSX 10.5 and 10.6).
9278     * Many bugs in the compiler have been fixed in this release; in
9279       particular, LTO can now be used when compiling Objective-C and
9280       Objective-C++ and the parser is much more robust in dealing with
9281       invalid code.
9282
9283    Runtime Library (libobjc)
9284
9285     * The GNU Objective-C runtime library now defines the macro
9286       __GNU_LIBOBJC__ (with a value that is increased at every release
9287       where there is any change to the API) in objc/objc.h, making it
9288       easy to determine if the GNU Objective-C runtime library is being
9289       used, and if so, which version. Previous versions of the GNU
9290       Objective-C runtime library (and other Objective-C runtime
9291       libraries such as the Apple one) do not define this macro.
9292     * A new Objective-C 2.0 API, almost identical to the one implemented
9293       by the Apple Objective-C runtime, has been implemented in the GNU
9294       Objective-C runtime library. The new API hides the internals of
9295       most runtime structures but provides a more extensive set of
9296       functions to operate on them. It is much easier, for example, to
9297       create or modify classes at runtime. The new API also makes it
9298       easier to port software from Apple to GNU as almost no changes
9299       should be required. The old API is still supported for backwards
9300       compatibility; including the old objc/objc-api.h header file
9301       automatically selects the old API, while including the new
9302       objc/runtime.h header file automatically selects the new API.
9303       Support for the old API is being phased out and upgrading the
9304       software to use the new API is strongly recommended. To check for
9305       the availability of the new API, the __GNU_LIBOBJC__ macro can be
9306       used as older versions of the GNU Objective-C runtime library,
9307       which do not support the new API, do not define such a macro.
9308     * Runtime support for @synchronized has been added.
9309     * Runtime support for Objective-C 2.0 synthesized property accessors
9310       has been added.
9311     * Runtime support for Objective-C 2.0 fast enumeration has been
9312       added.
9313
9314New Targets and Target Specific Improvements
9315
9316  ARM
9317
9318     * GCC now supports the Cortex-M4 processor implementing the v7-em
9319       version of the architecture using the option -mcpu=cortex-m4.
9320     * Scheduling descriptions for the Cortex-M4, the Neon and the
9321       floating point units of the Cortex-A9 and a pipeline description
9322       for the Cortex-A5 have been added.
9323     * Synchronization primitives such as __sync_fetch_and_add and friends
9324       are now inlined for supported architectures rather than calling
9325       into a kernel helper function.
9326     * SSA loop prefetching is enabled by default for the Cortex-A9 at
9327       -O3.
9328     * Several improvements were committed to improve code generation for
9329       the ARM architecture including a rewritten implementation for load
9330       and store multiples.
9331     * Several enhancements were committed to improve SIMD code generation
9332       for NEON by adding support for widening instructions, misaligned
9333       loads and stores, vector conditionals and support for 64 bit
9334       arithmetic.
9335     * Support was added for the Faraday cores fa526, fa606te, fa626te,
9336       fmp626te, fmp626 and fa726te and can be used with the respective
9337       names as parameters to the -mcpu= option.
9338     * Basic support was added for Cortex-A15 and is available through
9339       -mcpu=cortex-a15.
9340     * GCC for AAPCS configurations now more closely adheres to the AAPCS
9341       specification by enabling -fstrict-volatile-bitfields by default.
9342
9343  IA-32/x86-64
9344
9345     * The new -fsplit-stack option permits programs to use a
9346       discontiguous stack. This is useful for threaded programs, in that
9347       it is no longer necessary to specify the maximum stack size when
9348       creating a thread. This feature is currently only implemented for
9349       32-bit and 64-bit x86 GNU/Linux targets.
9350     * Support for emitting profiler counter calls before function
9351       prologues. This is enabled via a new command-line option -mfentry.
9352     * Optimization for the Intel Core 2 processors is now available
9353       through the -march=core2 and -mtune=core2 options.
9354     * Support for Intel Core i3/i5/i7 processors is now available through
9355       the -march=corei7 and -mtune=corei7 options.
9356     * Support for Intel Core i3/i5/i7 processors with AVX is now
9357       available through the -march=corei7-avx and -mtune=corei7-avx
9358       options.
9359     * Support for AMD Bobcat (family 14) processors is now available
9360       through the -march=btver1 and -mtune=btver1 options.
9361     * Support for AMD Bulldozer (family 15) processors is now available
9362       through the -march=bdver1 and -mtune=bdver1 options.
9363     * The default setting (when not optimizing for size) for 32-bit
9364       GNU/Linux and Darwin x86 targets has been changed to
9365       -fomit-frame-pointer. The default can be reverted to
9366       -fno-omit-frame-pointer by configuring GCC with the
9367       --enable-frame-pointer configure option.
9368     * Darwin, FreeBSD, Solaris 2, MinGW and Cygwin now all support
9369       __float128 on 32-bit and 64-bit x86 targets.
9370     * AVX floating-point arithmetic can now be enabled by default at
9371       configure time with the new --with-fpmath=avx option.
9372     * The SSA loop prefetching pass is enabled when using -O3 when
9373       optimizing for CPUs where prefetching is beneficial (AMD CPUs newer
9374       than K6).
9375     * Support for TBM (Trailing Bit Manipulation) built-in functions and
9376       code generation is available via -mtbm.
9377     * Support for AMD's BMI (Bit Manipulation) built-in functions and
9378       code generation is available via -mbmi.
9379
9380  MicroBlaze
9381
9382     * Support has been added for the Xilinx MicroBlaze softcore processor
9383       (microblaze-elf) embedded target. This configurable processor is
9384       supported on several Xilinx Spartan and Virtex FPGAs.
9385
9386  MIPS
9387
9388     * GCC now supports the Loongson 3A processor. Its canonical -march=
9389       and -mtune= name is loongson3a.
9390
9391  MN10300 / AM33
9392
9393     * The inline assembly register constraint "A" has been renamed "c".
9394       This constraint is used to select a floating-point register that
9395       can be used as the destination of a multiply-accumulate
9396       instruction.
9397     * New inline assembly register constraints "A" and "D" have been
9398       added. These constraint letters resolve to all general registers
9399       when compiling for AM33, and resolve to address registers only or
9400       data registers only when compiling for MN10300.
9401     * The MDR register is represented in the compiler. One can access the
9402       register via the "z" constraint in inline assembly. It can be
9403       marked as clobbered or used as a local register variable via the
9404       "mdr" name. The compiler uses the RETF instruction if the function
9405       does not modify the MDR register, so it is important that inline
9406       assembly properly annotate any usage of the register.
9407
9408  PowerPC/PowerPC64
9409
9410     * GCC now supports the Applied Micro Titan processor with
9411       -mcpu=titan.
9412     * The -mrecip option has been added, which indicates whether the
9413       reciprocal and reciprocal square root instructions should be used.
9414     * The -mveclibabi=mass option can be used to enable the compiler to
9415       autovectorize mathematical functions using the Mathematical
9416       Acceleration Subsystem library.
9417     * The -msingle-pic-base option has been added, which instructs the
9418       compiler to avoid loading the PIC base register in function
9419       prologues. The PIC base register must be initialized by the runtime
9420       system.
9421     * The -mblock-move-inline-limit option has been added, which enables
9422       the user to control the maximum size of inlined memcpy calls and
9423       similar.
9424     * PowerPC64 GNU/Linux support for applications requiring a large TOC
9425       section has been improved. A new command-line option,
9426       -mcmodel=MODEL, controls this feature; valid values for MODEL are
9427       small, medium, or large.
9428     * The Altivec builtin functions vec_ld and vec_st have been modified
9429       to generate the Altivec memory instructions LVX and STVX, even if
9430       the -mvsx option is used. In the initial GCC 4.5 release, these
9431       builtin functions were changed to generate VSX memory reference
9432       instructions instead of Altivec memory instructions, but there are
9433       differences between the two instructions. If the VSX instruction
9434       set is available, you can now use the new builtin functions
9435       vec_vsx_ld and vec_vsx_st which always generates the VSX memory
9436       instructions.
9437     * The GCC compiler on AIX now defaults to a process layout with a
9438       larger data space allowing larger programs to be compiled.
9439     * The GCC long double type on AIX 6.1 and above has reverted to 64
9440       bit double precision, matching the AIX XL compiler default, because
9441       of missing C99 symbols required by the GCC runtime.
9442     * The default processor scheduling model and tuning for PowerPC64
9443       GNU/Linux and for AIX 6.1 and above now is POWER7.
9444     * Starting with GCC 4.6.1, vectors of type vector long long or vector
9445       long are passed and returned in the same method as other vectors
9446       with the VSX instruction set. Previously the GCC compiler did not
9447       adhere to the ABI for 128-bit vectors with 64-bit integer base
9448       types (PR 48857). This is also fixed in the GCC 4.5.4 release.
9449
9450  S/390, zSeries and System z9/z10, IBM zEnterprise z196
9451
9452     * Support for the zEnterprise z196 processor has been added. When
9453       using the -march=z196 option, the compiler will generate code
9454       making use of the following instruction facilities:
9455          + Conditional load/store
9456          + Distinct-operands
9457          + Floating-point-extension
9458          + Interlocked-access
9459          + Population-count
9460       The -mtune=z196 option avoids the compare and branch instructions
9461       as well as the load address instruction with an index register as
9462       much as possible and performs instruction scheduling appropriate
9463       for the new out-of-order pipeline architecture.
9464     * When using the -m31 -mzarch options the generated code still
9465       conforms to the 32-bit ABI but uses the general purpose registers
9466       as 64-bit registers internally. This requires a Linux kernel saving
9467       the whole 64-bit registers when doing a context switch. Kernels
9468       providing that feature indicate that by the 'highgprs' string in
9469       /proc/cpuinfo.
9470     * The SSA loop prefetching pass is enabled when using -O3.
9471
9472  SPARC
9473
9474     * GCC now supports the LEON series of SPARC V8 processors. The code
9475       generated by the compiler can either be tuned to it by means of the
9476       --with-tune=leon configure option and -mtune=leon compilation
9477       option, or the compiler can be built for the sparc-leon-{elf,linux}
9478       and sparc-leon3-{elf,linux} targets directly.
9479     * GCC has stopped sign/zero-extending parameter registers in the
9480       callee for functions taking parameters with sub-word size in 32-bit
9481       mode, since this is redundant with the specification of the ABI.
9482       GCC has never done so in 64-bit mode since this is also redundant.
9483     * The command-line option -mfix-at697f has been added to enable the
9484       documented workaround for the single erratum of the Atmel AT697F
9485       processor.
9486
9487Operating Systems
9488
9489  Android
9490
9491     * GCC now supports the Bionic C library and provides a convenient way
9492       of building native libraries and applications for the Android
9493       platform. Refer to the documentation of the -mandroid and -mbionic
9494       options for details on building native code. At the moment, Android
9495       support is enabled only for ARM.
9496
9497  Darwin/Mac OS X
9498
9499     * General
9500          + Initial support for CFString types has been added.
9501            This allows GCC to build projects including the system Core
9502            Foundation frameworks. The GCC Objective-C family supports
9503            CFString "toll-free bridged" as per the Mac OS X system tools.
9504            CFString is also recognized in the context of format
9505            attributes and arguments (see the documentation for format
9506            attributes for limitations). At present, 8-bit character types
9507            are supported.
9508          + Object file size reduction.
9509            The Darwin zeroed memory allocators have been re-written to
9510            make more use of .zerofill sections. For non-debug code, this
9511            can reduce object file size significantly.
9512          + Objective-C family 64-bit support (NeXT ABI 2).
9513            Initial support has been added to support 64-bit Objective-C
9514            code using the Darwin/OS X native (NeXT) runtime. ABI version
9515            2 will be selected automatically when 64-bit code is built.
9516          + Objective-C family 32-bit ABI 1.
9517            For 32-bit code ABI 1 is also now also allowed. At present it
9518            must be selected manually using -fobjc-abi-version=1 where
9519            applicable - i.e. on Darwin 9/10 (OS X 10.5/10.6).
9520     * x86 Architecture
9521          + The -mdynamic-no-pic option has been enabled.
9522            Code supporting -mdynamic-no-pic optimization has been added
9523            and is applicable to -m32 builds. The compiler bootstrap uses
9524            the option where appropriate.
9525          + The default value for -mtune= has been changed.
9526            Since Darwin systems are primarily Xeon, Core-2 or similar the
9527            default tuning has been changed to -mtune=core2.
9528          + Enable 128-bit long double (__float128) support on Darwin.
9529     * PPC Architecture
9530          + Darwin64 ABI.
9531            Several significant bugs have been fixed, such that GCC now
9532            produces code compatible with the Darwin64 PowerPC ABI.
9533          + libffi and boehm-gc.
9534            The Darwin ports of the libffi and boehm-gc libraries have
9535            been upgraded to include a Darwin64 implementation. This means
9536            that powerpc*-*-darwin9 platforms may now, for example, build
9537            Java applications with -m64 enabled.
9538          + Plug-in support has been enabled.
9539          + The -fsection-anchors option is now available although,
9540            presently, not heavily tested.
9541
9542  Solaris 2
9543
9544    New Features
9545
9546     * Support symbol versioning with the Sun linker.
9547     * Allow libstdc++ to leverage full ISO C99 support on Solaris 10+.
9548     * Support thread-local storage (TLS) with the Sun assembler on
9549       Solaris 2/x86.
9550     * Support TLS on Solaris 8/9 if prerequisites are met.
9551     * Support COMDAT group with the GNU assembler and recent Sun linker.
9552     * Support the Sun assembler visibility syntax.
9553     * Default Solaris 2/x86 to -march=pentium4 (Solaris 10+) resp.
9554       -march=pentiumpro (Solaris 8/9).
9555     * Don't use SSE on Solaris 8/9 x86 by default.
9556     * Enable 128-bit long double (__float128) support on Solaris 2/x86.
9557
9558    ABI Change
9559
9560     * Change the ABI for returning 8-byte vectors like __m64 in MMX
9561       registers on Solaris 10+/x86 to match the Sun Studio 12.1+
9562       compilers. This is an incompatible change. If you use such types,
9563       you must either recompile all your code with the new compiler or
9564       use the new -mvect8-ret-in-mem option to remain compatible with
9565       previous versions of GCC and Sun Studio.
9566
9567  Windows x86/x86_64
9568
9569     * Initial support for decimal floating point.
9570     * Support for the __thiscall calling-convention.
9571     * Support for hot-patchable function prologues via the
9572       ms_hook_prologue attribute for x86_64 in addition to 32-bit x86.
9573     * Improvements of stack-probing and stack-allocation mechanisms.
9574     * Support of push/pop-macro pragma as preprocessor command.
9575       With #pragma push_macro("macro-name") the current definition of
9576       macro-name is saved and can be restored with #pragma
9577       pop_macro("macro-name") to its saved definition.
9578     * Enable 128-bit long double (__float128) support on MinGW and
9579       Cygwin.
9580
9581Other significant improvements
9582
9583  Installation changes
9584
9585     * An install-strip make target is provided that installs stripped
9586       executables, and may install libraries with unneeded or debugging
9587       sections stripped.
9588     * On Power7 systems, there is a potential problem if you build the
9589       GCC compiler with a host compiler using options that enable the VSX
9590       instruction set generation. If the host compiler has been patched
9591       so that the vec_ld and vec_st builtin functions generate Altivec
9592       memory instructions instead of VSX memory instructions, then you
9593       should be able to build the compiler with VSX instruction
9594       generation.
9595
9596Changes for GCC Developers
9597
9598   Note: these changes concern developers that develop GCC itself or
9599   software that integrates with GCC, such as plugins, and not the general
9600   GCC users.
9601     * The gengtype utility, which previously was internal to the GCC
9602       build process, has been enchanced to provide GC root information
9603       for plugins as necessary.
9604     * The old GC allocation interface of ggc_alloc and friends was
9605       replaced with a type-safe alternative.
9606
9607GCC 4.6.1
9608
9609   This is the [20]list of problem reports (PRs) from GCC's bug tracking
9610   system that are known to be fixed in the 4.6.1 release. This list might
9611   not be complete (that is, it is possible that some PRs that have been
9612   fixed are not listed here).
9613
9614GCC 4.6.2
9615
9616   This is the [21]list of problem reports (PRs) from GCC's bug tracking
9617   system that are known to be fixed in the 4.6.2 release. This list might
9618   not be complete (that is, it is possible that some PRs that have been
9619   fixed are not listed here).
9620
9621GCC 4.6.3
9622
9623   This is the [22]list of problem reports (PRs) from GCC's bug tracking
9624   system that are known to be fixed in the 4.6.3 release. This list might
9625   not be complete (that is, it is possible that some PRs that have been
9626   fixed are not listed here).
9627
9628GCC 4.6.4
9629
9630   This is the [23]list of problem reports (PRs) from GCC's bug tracking
9631   system that are known to be fixed in the 4.6.4 release. This list might
9632   not be complete (that is, it is possible that some PRs that have been
9633   fixed are not listed here).
9634
9635
9636    For questions related to the use of GCC, please consult these web
9637    pages and the [24]GCC manuals. If that fails, the
9638    [25]gcc-help@gcc.gnu.org mailing list might help. Comments on these
9639    web pages and the development of GCC are welcome on our developer
9640    list at [26]gcc@gcc.gnu.org. All of [27]our lists have public
9641    archives.
9642
9643   Copyright (C) [28]Free Software Foundation, Inc. Verbatim copying and
9644   distribution of this entire article is permitted in any medium,
9645   provided this notice is preserved.
9646
9647   These pages are [29]maintained by the GCC team. Last modified
9648   2019-11-28[30].
9649
9650References
9651
9652   1. https://sourceware.org/bugzilla/show_bug.cgi?id=10401
9653   2. http://gcc.gnu.org/gcc-4.5/changes.html
9654   3. http://gcc.gnu.org/gcc-4.5/changes.html#obsoleted
9655   4. http://gcc.gnu.org/gcc-4.6/porting_to.html
9656   5. http://gcc.gnu.org/projects/lto/whopr.pdf
9657   6. https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html#Warning-Options
9658   7. http://gcc.gnu.org/gcc-4.6/cxx0x_status.html
9659   8. https://gcc.gnu.org/PR43145
9660   9. https://gcc.gnu.org/PR43680
9661  10. https://gcc.gnu.org/PR33558
9662  11. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_active.html#253
9663  12. https://gcc.gnu.org/onlinedocs/gcc-4.6.4/libstdc++/manual/manual/status.html#status.iso.200x
9664  13. https://gcc.gnu.org/onlinedocs/libstdc++/manual/debug_mode.html
9665  14. https://gcc.gnu.org/onlinedocs/libstdc++/manual/debug.html#debug.races
9666  15. https://gcc.gnu.org/onlinedocs/gcc/Preprocessor-Options.html
9667  16. https://gcc.gnu.org/wiki/OOP
9668  17. https://gcc.gnu.org/wiki/Coarray
9669  18. https://gcc.gnu.org/onlinedocs/gfortran/Code-Gen-Options.html#index-g_t_0040code_007bfcoarray_007d-233
9670  19. https://golang.org/
9671  20. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.6.1
9672  21. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.6.2
9673  22. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.6.3
9674  23. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.6.4
9675  24. https://gcc.gnu.org/onlinedocs/
9676  25. mailto:gcc-help@gcc.gnu.org
9677  26. mailto:gcc@gcc.gnu.org
9678  27. https://gcc.gnu.org/lists.html
9679  28. https://www.fsf.org/
9680  29. https://gcc.gnu.org/about.html
9681  30. http://validator.w3.org/check/referer
9682======================================================================
9683http://gcc.gnu.org/gcc-4.5/index.html
9684
9685                             GCC 4.5 Release Series
9686
9687   (This release series is no longer supported.)
9688
9689   Jul 2, 2012
9690
9691   The [1]GNU project and the GCC developers are pleased to announce the
9692   release of GCC 4.5.4.
9693
9694   This release is a bug-fix release, containing fixes for regressions in
9695   GCC 4.5.3 relative to previous releases of GCC.
9696
9697Release History
9698
9699   GCC 4.5.4
9700          Jul 2, 2012 ([2]changes)
9701
9702   GCC 4.5.3
9703          Apr 28, 2011 ([3]changes)
9704
9705   GCC 4.5.2
9706          Dec 16, 2010 ([4]changes)
9707
9708   GCC 4.5.1
9709          Jul 31, 2010 ([5]changes)
9710
9711   GCC 4.5.0
9712          April 14, 2010 ([6]changes)
9713
9714References and Acknowledgements
9715
9716   GCC used to stand for the GNU C Compiler, but since the compiler
9717   supports several other languages aside from C, it now stands for the
9718   GNU Compiler Collection.
9719
9720   A list of [7]successful builds is updated as new information becomes
9721   available.
9722
9723   The GCC developers would like to thank the numerous people that have
9724   contributed new features, improvements, bug fixes, and other changes as
9725   well as test results to GCC. This [8]amazing group of volunteers is
9726   what makes GCC successful.
9727
9728   For additional information about GCC please refer to the [9]GCC project
9729   web site or contact the [10]GCC development mailing list.
9730
9731   To obtain GCC please use [11]our mirror sites or [12]our version
9732   control system.
9733
9734
9735    For questions related to the use of GCC, please consult these web
9736    pages and the [13]GCC manuals. If that fails, the
9737    [14]gcc-help@gcc.gnu.org mailing list might help. Comments on these
9738    web pages and the development of GCC are welcome on our developer
9739    list at [15]gcc@gcc.gnu.org. All of [16]our lists have public
9740    archives.
9741
9742   Copyright (C) [17]Free Software Foundation, Inc. Verbatim copying and
9743   distribution of this entire article is permitted in any medium,
9744   provided this notice is preserved.
9745
9746   These pages are [18]maintained by the GCC team. Last modified
9747   2020-01-14[19].
9748
9749References
9750
9751   1. http://www.gnu.org/
9752   2. http://gcc.gnu.org/gcc-4.5/changes.html
9753   3. http://gcc.gnu.org/gcc-4.5/changes.html
9754   4. http://gcc.gnu.org/gcc-4.5/changes.html
9755   5. http://gcc.gnu.org/gcc-4.5/changes.html
9756   6. http://gcc.gnu.org/gcc-4.5/changes.html
9757   7. http://gcc.gnu.org/gcc-4.5/buildstat.html
9758   8. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
9759   9. http://gcc.gnu.org/index.html
9760  10. mailto:gcc@gcc.gnu.org
9761  11. http://gcc.gnu.org/mirrors.html
9762  12. http://gcc.gnu.org/git.html
9763  13. https://gcc.gnu.org/onlinedocs/
9764  14. mailto:gcc-help@gcc.gnu.org
9765  15. mailto:gcc@gcc.gnu.org
9766  16. https://gcc.gnu.org/lists.html
9767  17. https://www.fsf.org/
9768  18. https://gcc.gnu.org/about.html
9769  19. http://validator.w3.org/check/referer
9770======================================================================
9771http://gcc.gnu.org/gcc-4.5/changes.html
9772
9773                             GCC 4.5 Release Series
9774                        Changes, New Features, and Fixes
9775
9776Caveats
9777
9778     * GCC now requires the [1]MPC library in order to build. See the
9779       [2]prerequisites page for version requirements.
9780     * Support for a number of older systems and recently unmaintained or
9781       untested target ports of GCC has been declared obsolete in GCC 4.5.
9782       Unless there is activity to revive them, the next release of GCC
9783       will have their sources permanently removed.
9784       The following ports for individual systems on particular
9785       architectures have been obsoleted:
9786          + IRIX releases before 6.5 (mips-sgi-irix5*,
9787            mips-sgi-irix6.[0-4])
9788          + Solaris 7 (*-*-solaris2.7)
9789          + Tru64 UNIX releases before V5.1 (alpha*-dec-osf4*,
9790            alpha-dec-osf5.0*)
9791          + Details for the IRIX, Solaris 7, and Tru64 UNIX obsoletions
9792            can be found in the [3]announcement.
9793       Support for the classic POWER architecture implemented in the
9794       original RIOS and RIOS2 processors of the old IBM RS/6000 product
9795       line has been obsoleted in the rs6000 port. This does not affect
9796       the new generation Power and PowerPC architectures.
9797     * Support has been removed for all the [4]configurations obsoleted in
9798       GCC 4.4.
9799     * Support has been removed for the protoize and unprotoize utilities,
9800       obsoleted in GCC 4.4.
9801     * Support has been removed for tuning for Itanium1 (Merced) variants.
9802       Note that code tuned for Itanium2 should also run correctly on
9803       Itanium1.
9804     * GCC now generates unwind info also for epilogues. DWARF debuginfo
9805       generated by GCC now uses more features of DWARF3 than before, and
9806       also some DWARF4 features. GDB older than 7.0 is not able to handle
9807       either of these, so to debug GCC 4.5 generated binaries or
9808       libraries GDB 7.0 or later is needed. You can disable use of DWARF4
9809       features with the -gdwarf-3 -gstrict-dwarf options, or use
9810       -gdwarf-2 -gstrict-dwarf to restrict GCC to just DWARF2, but
9811       epilogue unwind info is emitted unconditionally whenever unwind
9812       info is emitted.
9813     * On x86 targets, code containing floating-point calculations may run
9814       significantly more slowly when compiled with GCC 4.5 in strict C99
9815       conformance mode than they did with earlier GCC versions. This is
9816       due to stricter standard conformance of the compiler and can be
9817       avoided by using the option -fexcess-precision=fast; also see
9818       [5]below.
9819     * The function attribute noinline no longer prevents GCC from cloning
9820       the function. A new attribute noclone has been introduced for this
9821       purpose. Cloning a function means that it is duplicated and the new
9822       copy is specialized for certain contexts (for example when a
9823       parameter is a known constant).
9824
9825General Optimizer Improvements
9826
9827     * The -save-temps now takes an optional argument. The -save-temps and
9828       -save-temps=cwd switches write the temporary files in the current
9829       working directory based on the original source file. The
9830       -save-temps=obj switch will write files into the directory
9831       specified with the -o option, and the intermediate filenames are
9832       based on the output file. This will allow the user to get the
9833       compiler intermediate files when doing parallel builds without two
9834       builds of the same filename located in different directories from
9835       interfering with each other.
9836     * Debugging dumps are now created in the same directory as the object
9837       file rather than in the current working directory. This allows the
9838       user to get debugging dumps when doing parallel builds without two
9839       builds of the same filename interfering with each other.
9840     * GCC has been integrated with the MPC library. This allows GCC to
9841       evaluate complex arithmetic at compile time [6]more accurately. It
9842       also allows GCC to evaluate calls to complex built-in math
9843       functions having constant arguments and replace them at compile
9844       time with their mathematically equivalent results. In doing so, GCC
9845       can generate correct results regardless of the math library
9846       implementation or floating point precision of the host platform.
9847       This also allows GCC to generate identical results regardless of
9848       whether one compiles in native or cross-compile configurations to a
9849       particular target. The following built-in functions take advantage
9850       of this new capability: cacos, cacosh, casin, casinh, catan,
9851       catanh, ccos, ccosh, cexp, clog, cpow, csin, csinh, csqrt, ctan,
9852       and ctanh. The float and long double variants of these functions
9853       (e.g. csinf and csinl) are also handled.
9854     * A new link-time optimizer has been added ([7]-flto). When this
9855       option is used, GCC generates a bytecode representation of each
9856       input file and writes it to specially-named sections in each object
9857       file. When the object files are linked together, all the function
9858       bodies are read from these named sections and instantiated as if
9859       they had been part of the same translation unit. This enables
9860       interprocedural optimizations to work across different files (and
9861       even different languages), potentially improving the performance of
9862       the generated code. To use the link-timer optimizer, -flto needs to
9863       be specified at compile time and during the final link. If the
9864       program does not require any symbols to be exported, it is possible
9865       to combine -flto and the experimental [8]-fwhopr with
9866       [9]-fwhole-program to allow the interprocedural optimizers to use
9867       more aggressive assumptions.
9868     * The automatic parallelization pass was enhanced to support
9869       parallelization of outer loops.
9870     * Automatic parallelization can be enabled as part of Graphite. In
9871       addition to -ftree-parallelize-loops=, specify
9872       -floop-parallelize-all to enable the Graphite-based optimization.
9873     * The infrastructure for optimizing based on [10]restrict qualified
9874       pointers has been rewritten and should result in code generation
9875       improvements. Optimizations based on restrict qualified pointers
9876       are now also available when using -fno-strict-aliasing.
9877     * There is a new optimization pass that attempts to change prototype
9878       of functions to avoid unused parameters, pass only relevant parts
9879       of structures and turn arguments passed by reference to arguments
9880       passed by value when possible. It is enabled by -O2 and above as
9881       well as -Os and can be manually invoked using the new command-line
9882       switch -fipa-sra.
9883     * GCC now optimize exception handling code. In particular cleanup
9884       regions that are proved to not have any effect are optimized out.
9885
9886New Languages and Language specific improvements
9887
9888  All languages
9889
9890     * The -fshow-column option is now on by default. This means error
9891       messages now have a column associated with them.
9892
9893  Ada
9894
9895     * Compilation of programs heavily using discriminated record types
9896       with variant parts has been sped up and generates more compact
9897       code.
9898     * Stack checking now works reasonably well on most plaforms. In some
9899       specific cases, stack overflows may still fail to be detected, but
9900       a compile-time warning will be issued for these cases.
9901
9902  C family
9903
9904     * If a header named in a #include directive is not found, the
9905       compiler exits immediately. This avoids a cascade of errors arising
9906       from declarations expected to be found in that header being
9907       missing.
9908     * A new built-in function __builtin_unreachable() has been added that
9909       tells the compiler that control will never reach that point. It may
9910       be used after asm statements that terminate by transferring control
9911       elsewhere, and in other places that are known to be unreachable.
9912     * The -Wlogical-op option now warns for logical expressions such as
9913       (c == 1 && c == 2) and (c != 1 || c != 2), which are likely to be
9914       mistakes. This option is disabled by default.
9915     * An asm goto feature has been added to allow asm statements that
9916       jump to C labels.
9917     * C++0x raw strings are supported for C++ and for C with -std=gnu99.
9918     * The deprecated attribute now takes an optional string argument, for
9919       example, __attribute__((deprecated("text string"))), that will be
9920       printed together with the deprecation warning.
9921
9922  C
9923
9924     * The -Wenum-compare option, which warns when comparing values of
9925       different enum types, now works for C. It formerly only worked for
9926       C++. This warning is enabled by -Wall. It may be avoided by using a
9927       type cast.
9928     * The -Wcast-qual option now warns about casts which are unsafe in
9929       that they permit const-correctness to be violated without further
9930       warnings. Specifically, it warns about cases where a qualifier is
9931       added when all the lower types are not const. For example, it warns
9932       about a cast from char ** to const char **.
9933     * The -Wc++-compat option is significantly improved. It issues new
9934       warnings for:
9935          + Using C++ reserved operator names as identifiers.
9936          + Conversions to enum types without explicit casts.
9937          + Using va_arg with an enum type.
9938          + Using different enum types in the two branches of ?:.
9939          + Using ++ or -- on a variable of enum type.
9940          + Using the same name as both a struct, union or enum tag and a
9941            typedef, unless the typedef refers to the tagged type itself.
9942          + Using a struct, union, or enum which is defined within another
9943            struct or union.
9944          + A struct field defined using a typedef if there is a field in
9945            the struct, or an enclosing struct, whose name is the typedef
9946            name.
9947          + Duplicate definitions at file scope.
9948          + Uninitialized const variables.
9949          + A global variable with an anonymous struct, union, or enum
9950            type.
9951          + Using a string constant to initialize a char array whose size
9952            is the length of the string.
9953     * The new -Wjump-misses-init option warns about cases where a goto or
9954       switch skips the initialization of a variable. This sort of branch
9955       is an error in C++ but not in C. This warning is enabled by
9956       -Wc++-compat.
9957     * GCC now ensures that a C99-conforming <stdint.h> is present on most
9958       targets, and uses information about the types in this header to
9959       implement the Fortran bindings to those types. GCC does not ensure
9960       the presence of such a header, and does not implement the Fortran
9961       bindings, on the following targets: NetBSD, VxWorks, VMS,
9962       SymbianOS, WinCE, LynxOS, Netware, QNX, Interix, TPF.
9963     * GCC now implements C90- and C99-conforming rules for constant
9964       expressions. This may cause warnings or errors for some code using
9965       expressions that can be folded to a constant but are not constant
9966       expressions as defined by ISO C.
9967     * All known target-independent C90 and C90 Amendment 1 conformance
9968       bugs, and all known target-independent C99 conformance bugs not
9969       related to floating point or extended identifiers, have been fixed.
9970     * The C decimal floating point support now includes support for the
9971       FLOAT_CONST_DECIMAL64 pragma.
9972     * The named address space feature from ISO/IEC TR 18037 is now
9973       supported. This is currently only implemented for the SPU
9974       processor.
9975
9976  C++
9977
9978     * Improved [11]experimental support for the upcoming C++0x ISO C++
9979       standard, including support for raw strings, lambda expressions and
9980       explicit type conversion operators.
9981     * When printing the name of a class template specialization, G++ will
9982       now omit any template arguments which come from default template
9983       arguments. This behavior (and the pretty-printing of function
9984       template specializations as template signature and arguments) can
9985       be disabled with the -fno-pretty-templates option.
9986     * Access control is now applied to typedef names used in a template,
9987       which may cause G++ to reject some ill-formed code that was
9988       accepted by earlier releases. The -fno-access-control option can be
9989       used as a temporary workaround until the code is corrected.
9990     * Compilation time for code that uses templates should now scale
9991       linearly with the number of instantiations rather than
9992       quadratically, as template instantiations are now looked up using
9993       hash tables.
9994     * Declarations of functions that look like builtin declarations of
9995       library functions are only considered to be redeclarations if they
9996       are declared with extern "C". This may cause problems with code
9997       that omits extern "C" on hand-written declarations of C library
9998       functions such as abort or memcpy. Such code is ill-formed, but was
9999       accepted by earlier releases.
10000     * Diagnostics that used to complain about passing non-POD types to
10001       ... or jumping past the declaration of a non-POD variable now check
10002       for triviality rather than PODness, as per C++0x.
10003     * In C++0x mode local and anonymous classes are now allowed as
10004       template arguments, and in declarations of variables and functions
10005       with linkage, so long as any such declaration that is used is also
10006       defined ([12]DR 757).
10007     * Labels may now have attributes, as has been permitted for a while
10008       in C. This is only permitted when the label definition and the
10009       attribute specifier is followed by a semicolon--i.e., the label
10010       applies to an empty statement. The only useful attribute for a
10011       label is unused.
10012     * G++ now implements [13]DR 176. Previously G++ did not support using
10013       the injected-class-name of a template base class as a type name,
10014       and lookup of the name found the declaration of the template in the
10015       enclosing scope. Now lookup of the name finds the
10016       injected-class-name, which can be used either as a type or as a
10017       template, depending on whether or not the name is followed by a
10018       template argument list. As a result of this change, some code that
10019       was previously accepted may be ill-formed because
10020         1. The injected-class-name is not accessible because it's from a
10021            private base, or
10022         2. The injected-class-name cannot be used as an argument for a
10023            template template parameter.
10024       In either of these cases, the code can be fixed by adding a
10025       nested-name-specifier to explicitly name the template. The first
10026       can be worked around with -fno-access-control; the second is only
10027       rejected with -pedantic.
10028     * A new standard mangling for SIMD vector types has been added, to
10029       avoid name clashes on systems with vectors of varying length. By
10030       default the compiler still uses the old mangling, but emits aliases
10031       with the new mangling on targets that support strong aliases. Users
10032       can switch over entirely to the new mangling with -fabi-version=4
10033       or -fabi-version=0. -Wabi will now warn about code that uses the
10034       old mangling.
10035     * The command-line option -ftemplate-depth-N is now written as
10036       -ftemplate-depth=N and the old form is deprecated.
10037     * Conversions between NULL and non-pointer types are now warned by
10038       default. The new option -Wno-conversion-null disables these
10039       warnings. Previously these warnings were only available when using
10040       -Wconversion explicitly.
10041
10042    Runtime Library (libstdc++)
10043
10044     * Improved experimental support for the upcoming ISO C++ standard,
10045       C++0x, including:
10046          + Support for <future>, <functional>, and <random>.
10047          + Existing facilities now exploit explicit operators and the
10048            newly implemented core C++0x features.
10049          + The header <cstdatomic> has been renamed to <atomic>.
10050     * An experimental [14]profile mode has been added. This is an
10051       implementation of many C++ standard library constructs with an
10052       additional analysis layer that gives performance improvement advice
10053       based on recognition of suboptimal usage patterns. For example,
10054#include <vector>
10055int main()
10056{
10057  std::vector<int> v;
10058  for (int k = 0; k < 1024; ++k)
10059    v.insert(v.begin(), k);
10060}
10061
10062       When instrumented via the profile mode, can return suggestions
10063       about the initial size and choice of the container used as follows:
10064vector-to-list: improvement = 5: call stack = 0x804842c ...
10065    : advice = change std::vector to std::list
10066vector-size: improvement = 3: call stack = 0x804842c ...
10067    : advice = change initial container size from 0 to 1024
10068
10069       These constructs can be substituted for the normal libstdc++
10070       constructs on a piecemeal basis, or all existing components can be
10071       transformed via the -D_GLIBCXX_PROFILE macro.
10072     * [15]Support for decimal floating-point arithmetic (aka ISO C++ TR
10073       24733) has been added. This support is in header file
10074       <decimal/decimal>, uses namespace std::decimal, and includes
10075       classes decimal32, decimal64, and decimal128.
10076     * Sources have been audited for application of function attributes
10077       nothrow, const, pure, and noreturn.
10078     * Python pretty-printers have been added for many standard library
10079       components that simplify the internal representation and present a
10080       more intuitive view of components when used with
10081       appropriately-advanced versions of GDB. For more information,
10082       please consult the more [16]detailed description.
10083     * The default behavior for comparing typeinfo names has changed, so
10084       in <typeinfo>, __GXX_MERGED_TYPEINFO_NAMES now defaults to zero.
10085     * The new -static-libstdc++ option directs g++ to link the C++
10086       library statically, even if the default would normally be to link
10087       it dynamically.
10088
10089  Fortran
10090
10091     * The COMMON default padding has been changed - instead of adding the
10092       padding before a variable it is now added afterwards, which
10093       increases the compatibility with other vendors and helps to obtain
10094       the correct output in some cases. Cf. also the -falign-commons
10095       option ([17]added in 4.4).
10096     * The -finit-real= option now also supports the value snan for
10097       signaling not-a-number; to be effective, one additionally needs to
10098       enable trapping (e.g. via -ffpe-trap=). Note: Compile-time
10099       optimizations can turn a signaling NaN into a quiet one.
10100     * The new option -fcheck= has been added with the options bounds,
10101       array-temps, do, pointer, and recursive. The bounds and array-temps
10102       options are equivalent to -fbounds-check and
10103       -fcheck-array-temporaries. The do option checks for invalid
10104       modification of loop iteration variables, and the recursive option
10105       tests for recursive calls to subroutines/functions which are not
10106       marked as recursive. With pointer pointer association checks in
10107       calls are performed; however, neither undefined pointers nor
10108       pointers in expressions are handled. Using -fcheck=all enables all
10109       these run-time checks.
10110     * The run-time checking -fcheck=bounds now warns about invalid string
10111       lengths of character dummy arguments. Additionally, more
10112       compile-time checks have been added.
10113     * The new option [18]-fno-protect-parens has been added; if set, the
10114       compiler may reorder REAL and COMPLEX expressions without regard to
10115       parentheses.
10116     * GNU Fortran no longer links against libgfortranbegin. As before,
10117       MAIN__ (assembler symbol name) is the actual Fortran main program,
10118       which is invoked by the main function. However, main is now
10119       generated and put in the same object file as MAIN__. For the time
10120       being, libgfortranbegin still exists for backward compatibility.
10121       For details see the new [19]Mixed-Language Programming chapter in
10122       the manual.
10123     * The I/O library was restructured for performance and cleaner code.
10124     * Array assignments and WHERE are now run in parallel when OpenMP's
10125       WORKSHARE is used.
10126     * The experimental option -fwhole-file was added. The option allows
10127       whole-file checking of procedure arguments and allows for better
10128       optimizations. It can also be used with -fwhole-program, which is
10129       now also supported in gfortran.
10130     * More Fortran 2003 and Fortran 2008 mathematical functions can now
10131       be used as initialization expressions.
10132     * Some extended attributes such as STDCALL are now supported via the
10133       [20]GCC$ compiler directive.
10134     * For Fortran 77 compatibility: If -fno-sign-zero is used, the SIGN
10135       intrinsic behaves now as if zero were always positive.
10136     * For legacy compatibiliy: On Cygwin and MinGW, the special files
10137       CONOUT$ and CONIN$ (and CONERR$ which maps to CONOUT$) are now
10138       supported.
10139     * Fortran 2003 support has been extended:
10140          + Procedure-pointer function results and procedure-pointer
10141            components (including PASS),
10142          + allocatable scalars (experimental),
10143          + DEFERRED type-bound procedures,
10144          + the ERRMSG= argument of the ALLOCATE and DEALLOCATE statements
10145            have been implemented.
10146          + The ALLOCATE statement supports type-specs and the SOURCE=
10147            argument.
10148          + OPERATOR(*) and ASSIGNMENT(=) are now allowed as GENERIC
10149            type-bound procedure (i.e. as type-bound operators).
10150          + Rounding (ROUND=, RZ, ...) for output is now supported.
10151          + The INT_FAST{8,16,32,64,128}_T kind type parameters of the
10152            intrinsic module ISO_C_BINDING are now supported, except for
10153            the targets listed above as ones where GCC does not have
10154            <stdint.h> type information.
10155          + Extensible derived types with type-bound procedure or
10156            procedure pointer with PASS attribute now have to use CLASS in
10157            line with the Fortran 2003 standard; the workaround to use
10158            TYPE is no longer supported.
10159          + [21]Experimental, incomplete support for polymorphism,
10160            including CLASS, SELECT TYPE and dynamic dispatch of
10161            type-bound procedure calls. Some features do not work yet such
10162            as unlimited polymorphism (CLASS(*)).
10163     * Fortran 2008 support has been extended:
10164          + The OPEN statement now supports the NEWUNIT= option, which
10165            returns a unique file unit, thus preventing inadvertent use of
10166            the same unit in different parts of the program.
10167          + Support for unlimited format items has been added.
10168          + The INT{8,16,32} and REAL{32,64,128} kind type parameters of
10169            the intrinsic module ISO_FORTRAN_ENV are now supported.
10170          + Using complex arguments with TAN, SINH, COSH, TANH, ASIN,
10171            ACOS, and ATAN is now possible; the functions ASINH, ACOSH,
10172            and ATANH have been added (for real and complex arguments) and
10173            ATAN(Y,X) is now an alias for ATAN2(Y,X).
10174          + The BLOCK construct has been implemented.
10175
10176New Targets and Target Specific Improvements
10177
10178  AIX
10179
10180     * Full cross-toolchain support now available with GNU Binutils
10181
10182  ARM
10183
10184     * GCC now supports the Cortex-M0 and Cortex-A5 processors.
10185     * GCC now supports the ARM v7E-M architecture.
10186     * GCC now supports VFPv4-based FPUs and FPUs with
10187       single-precision-only VFP.
10188     * GCC has many improvements to optimization for other ARM processors,
10189       including scheduling support for the integer pipeline on Cortex-A9.
10190     * GCC now supports the IEEE 754-2008 half-precision floating-point
10191       type, and a variant ARM-specific half-precision type. This type is
10192       specified using __fp16, with the layout determined by
10193       -mfp16-format. With appropriate -mfpu options, the Cortex-A9 and
10194       VFPv4 half-precision instructions will be used.
10195     * GCC now supports the variant of AAPCS that uses VFP registers for
10196       parameter passing and return values.
10197
10198  AVR
10199
10200     * The -mno-tablejump option has been removed because it has the same
10201       effect as the -fno-jump-tables option.
10202     * Added support for these new AVR devices:
10203          + ATmega8U2
10204          + ATmega16U2
10205          + ATmega32U2
10206
10207  IA-32/x86-64
10208
10209     * GCC now will set the default for -march= based on the configure
10210       target.
10211     * GCC now supports handling floating-point excess precision arising
10212       from use of the x87 floating-point unit in a way that conforms to
10213       ISO C99. This is enabled with -fexcess-precision=standard and with
10214       standards conformance options such as -std=c99, and may be disabled
10215       using -fexcess-precision=fast.
10216     * Support for the Intel Atom processor is now available through the
10217       -march=atom and -mtune=atom options.
10218     * A new -mcrc32 option is now available to enable crc32 intrinsics.
10219     * A new -mmovbe option is now available to enable GCC to use the
10220       movbe instruction to implement __builtin_bswap32 and
10221       __builtin_bswap64.
10222     * SSE math now can be enabled by default at configure time with the
10223       new --with-fpmath=sse option.
10224     * There is a new intrinsic header file, <x86intrin.h>. It should be
10225       included before using any IA-32/x86-64 intrinsics.
10226     * Support for the XOP, FMA4, and LWP instruction sets for the AMD
10227       Orochi processors are now available with the -mxop, -mfma4, and
10228       -mlwp options.
10229     * The -mabm option enables GCC to use the popcnt and lzcnt
10230       instructions on AMD processors.
10231     * The -mpopcnt option enables GCC to use the popcnt instructions on
10232       both AMD and Intel processors.
10233
10234  M68K/ColdFire
10235
10236     * GCC now supports ColdFire 51xx, 5221x, 5225x, 52274, 52277, 5301x
10237       and 5441x devices.
10238     * GCC now supports thread-local storage (TLS) on M68K and ColdFire
10239       processors.
10240
10241  MeP
10242
10243   Support has been added for the Toshiba Media embedded Processor (MeP,
10244   or mep-elf) embedded target.
10245
10246  MIPS
10247
10248     * GCC now supports MIPS 1004K processors.
10249     * GCC can now be configured with options --with-arch-32,
10250       --with-arch-64, --with-tune-32 and --with-tune-64 to control the
10251       default optimization separately for 32-bit and 64-bit modes.
10252     * MIPS targets now support an alternative _mcount interface, in which
10253       register $12 points to the function's save slot for register $31.
10254       This interface is selected by the -mcount-ra-address option; see
10255       the documentation for more details.
10256     * GNU/Linux targets can now generate read-only .eh_frame sections.
10257       This optimization requires GNU binutils 2.20 or above, and is only
10258       available if GCC is configured with a suitable version of binutils.
10259     * GNU/Linux targets can now attach special relocations to indirect
10260       calls, so that the linker can turn them into direct jumps or
10261       branches. This optimization requires GNU binutils 2.20 or later,
10262       and is automatically selected if GCC is configured with an
10263       appropriate version of binutils. It can be explicitly enabled or
10264       disabled using the -mrelax-pic-calls command-line option.
10265     * GCC now generates more heavily-optimized atomic operations on
10266       Octeon processors.
10267     * MIPS targets now support the -fstack-protector option.
10268     * GCC now supports an -msynci option, which specifies that synci is
10269       enough to flush the instruction cache, without help from the
10270       operating system. GCC uses this information to optimize
10271       automatically-generated cache flush operations, such as those used
10272       for nested functions in C. There is also a --with-synci
10273       configure-time option, which makes -msynci the default.
10274     * GCC supports four new function attributes for interrupt handlers:
10275       interrupt, use_shadow_register_set, keep_interrupts_masked and
10276       use_debug_exception_return. See the documentation for more details
10277       about these attributes.
10278
10279  RS/6000 (POWER/PowerPC)
10280
10281     * GCC now supports the Power ISA 2.06, which includes the VSX
10282       instructions that add vector 64-bit floating point support, new
10283       population count instructions, and conversions between floating
10284       point and unsigned types.
10285     * Support for the power7 processor is now available through the
10286       -mcpu=power7 and -mtune=power7.
10287     * GCC will now vectorize loops that contain simple math functions
10288       like copysign when generating code for altivec or VSX targets.
10289     * Support for the A2 processor is now available through the -mcpu=a2
10290       and -mtune=a2 options.
10291     * Support for the 476 processor is now available through the
10292       -mcpu={476,476fp} and -mtune={476,476fp} options.
10293     * Support for the e500mc64 processor is now available through the
10294       -mcpu=e500mc64 and -mtune=e500mc64 options.
10295     * GCC can now be configured with options --with-cpu-32,
10296       --with-cpu-64, --with-tune-32 and --with-tune-64 to control the
10297       default optimization separately for 32-bit and 64-bit modes.
10298     * Starting with GCC 4.5.4, vectors of type vector long long or vector
10299       long are passed and returned in the same method as other vectors
10300       with the VSX instruction set. Previously the GCC compiler did not
10301       adhere to the ABI for 128-bit vectors with 64-bit integer base
10302       types (PR 48857). This is also fixed in the GCC 4.6.1 release.
10303
10304  RX
10305
10306   Support has been added for the Renesas RX Processor (rx-elf) target.
10307
10308Operating Systems
10309
10310  Windows (Cygwin and MinGW)
10311
10312     * GCC now installs all the major language runtime libraries as DLLs
10313       when configured with the --enable-shared option.
10314     * GCC now makes use of the new support for aligned common variables
10315       in versions of binutils >= 2.20 to fix bugs in the support for SSE
10316       data types.
10317     * Improvements to the libffi support library increase the reliability
10318       of code generated by GCJ on all Windows platforms. Libgcj is
10319       enabled by default for the first time.
10320     * Libtool improvements simplify installation by placing the generated
10321       DLLs in the correct binaries directory.
10322     * Numerous other minor bugfixes and improvements, and substantial
10323       enhancements to the Fortran language support library.
10324
10325   >
10326
10327Other significant improvements
10328
10329  Plugins
10330
10331     * It is now possible to extend the compiler without having to modify
10332       its source code. A new option -fplugin=file.so tells GCC to load
10333       the shared object file.so and execute it as part of the compiler.
10334       The internal documentation describes the details on how plugins can
10335       interact with the compiler.
10336
10337  Installation changes
10338
10339     * The move to newer autotools changed default installation
10340       directories and switches to control them: The --with-datarootdir,
10341       --with-docdir, --with-pdfdir, and --with-htmldir switches are not
10342       used any more. Instead, you can now use --datarootdir, --docdir,
10343       --htmldir, and --pdfdir. The default installation directories have
10344       changed as follows according to the GNU Coding Standards:
10345
10346       datarootdir read-only architecture-independent data root [PREFIX/share]
10347       localedir   locale-specific message catalogs [DATAROOTDIR/locale]
10348       docdir      documentation root [DATAROOTDIR/doc/PACKAGE]
10349       htmldir     html documentation [DOCDIR]
10350       dvidir      dvi documentation [DOCDIR]
10351       pdfdir      pdf documentation [DOCDIR]
10352       psdir       ps documentation [DOCDIR]
10353       The following variables have new default values:
10354
10355       datadir read-only architecture-independent data [DATAROOTDIR]
10356       infodir info documentation [DATAROOTDIR/info]
10357       mandir  man documentation [DATAROOTDIR/man]
10358
10359GCC 4.5.1
10360
10361   This is the [22]list of problem reports (PRs) from GCC's bug tracking
10362   system that are known to be fixed in the 4.5.1 release. This list might
10363   not be complete (that is, it is possible that some PRs that have been
10364   fixed are not listed here).
10365
10366  All languages
10367
10368     * GCC's new link-time optimizer ([23]-flto) now also works on a few
10369       non-ELF targets:
10370          + Cygwin (*-cygwin*)
10371          + MinGW (*-mingw*)
10372          + Darwin on x86-64 (x86_64-apple-darwin*)
10373       LTO is not enabled by default for these targets. To enable LTO, you
10374       should configure with the --enable-lto option.
10375
10376GCC 4.5.2
10377
10378   This is the [24]list of problem reports (PRs) from GCC's bug tracking
10379   system that are known to be fixed in the 4.5.2 release. This list might
10380   not be complete (that is, it is possible that some PRs that have been
10381   fixed are not listed here).
10382
10383GCC 4.5.3
10384
10385   This is the [25]list of problem reports (PRs) from GCC's bug tracking
10386   system that are known to be fixed in the 4.5.3 release. This list might
10387   not be complete (that is, it is possible that some PRs that have been
10388   fixed are not listed here).
10389
10390   On the PowerPC compiler, the Altivec builtin functions vec_ld and
10391   vec_st have been modified to generate the Altivec memory instructions
10392   LVX and STVX, even if the -mvsx option is used. In the initial GCC 4.5
10393   release, these builtin functions were changed to generate VSX memory
10394   reference instructions instead of Altivec memory instructions, but
10395   there are differences between the two instructions. If the VSX
10396   instruction set is available, you can now use the new builtin functions
10397   vec_vsx_ld and vec_vsx_st which always generates the VSX memory
10398   instructions.
10399
10400GCC 4.5.4
10401
10402   This is the [26]list of problem reports (PRs) from GCC's bug tracking
10403   system that are known to be fixed in the 4.5.4 release. This list might
10404   not be complete (that is, it is possible that some PRs that have been
10405   fixed are not listed here).
10406
10407
10408    For questions related to the use of GCC, please consult these web
10409    pages and the [27]GCC manuals. If that fails, the
10410    [28]gcc-help@gcc.gnu.org mailing list might help. Comments on these
10411    web pages and the development of GCC are welcome on our developer
10412    list at [29]gcc@gcc.gnu.org. All of [30]our lists have public
10413    archives.
10414
10415   Copyright (C) [31]Free Software Foundation, Inc. Verbatim copying and
10416   distribution of this entire article is permitted in any medium,
10417   provided this notice is preserved.
10418
10419   These pages are [32]maintained by the GCC team. Last modified
10420   2019-11-28[33].
10421
10422References
10423
10424   1. http://www.multiprecision.org/mpc/
10425   2. https://gcc.gnu.org/install/prerequisites.html
10426   3. https://gcc.gnu.org/ml/gcc/2010-01/msg00510.html
10427   4. http://gcc.gnu.org/gcc-4.4/changes.html#obsoleted
10428   5. http://gcc.gnu.org/gcc-4.5/changes.html#x86
10429   6. https://gcc.gnu.org/PR30789
10430   7. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-flto-801
10431   8. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-fwhopr-802
10432   9. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-fwhole-program-800
10433  10. https://gcc.gnu.org/onlinedocs/gcc/Restricted-Pointers.html
10434  11. http://gcc.gnu.org/gcc-4.5/cxx0x_status.html
10435  12. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#757
10436  13. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#176
10437  14. https://gcc.gnu.org/onlinedocs/libstdc++/manual/profile_mode.html
10438  15. https://gcc.gnu.org/onlinedocs/libstdc++/manual/status.html#status.iso.tr24733
10439  16. https://sourceware.org/gdb/wiki/STLSupport
10440  17. http://gcc.gnu.org/gcc-4.4/changes.html
10441  18. https://gcc.gnu.org/onlinedocs/gfortran/Code-Gen-Options.html
10442  19. https://gcc.gnu.org/onlinedocs/gfortran/Mixed-Language-Programming.html
10443  20. https://gcc.gnu.org/onlinedocs/gfortran/GNU-Fortran-Compiler-Directives.html
10444  21. https://gcc.gnu.org/wiki/OOP
10445  22. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.5.1
10446  23. https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-flto-801
10447  24. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.5.2
10448  25. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.5.3
10449  26. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.5.4
10450  27. https://gcc.gnu.org/onlinedocs/
10451  28. mailto:gcc-help@gcc.gnu.org
10452  29. mailto:gcc@gcc.gnu.org
10453  30. https://gcc.gnu.org/lists.html
10454  31. https://www.fsf.org/
10455  32. https://gcc.gnu.org/about.html
10456  33. http://validator.w3.org/check/referer
10457======================================================================
10458http://gcc.gnu.org/gcc-4.4/index.html
10459
10460                             GCC 4.4 Release Series
10461
10462   This release series is no longer maintained.
10463
10464   March 13, 2012
10465
10466   The [1]GNU project and the GCC developers are pleased to announce the
10467   release of GCC 4.4.7.
10468
10469   This release is a bug-fix release, containing fixes for regressions in
10470   GCC 4.4.6 relative to previous releases of GCC.
10471
10472Release History
10473
10474   GCC 4.4.7
10475          March 13, 2012 ([2]changes)
10476
10477   GCC 4.4.6
10478          April 16, 2011 ([3]changes)
10479
10480   GCC 4.4.5
10481          October 1, 2010 ([4]changes)
10482
10483   GCC 4.4.4
10484          April 29, 2010 ([5]changes)
10485
10486   GCC 4.4.3
10487          January 21, 2010 ([6]changes)
10488
10489   GCC 4.4.2
10490          October 15, 2009 ([7]changes)
10491
10492   GCC 4.4.1
10493          July 22, 2009 ([8]changes)
10494
10495   GCC 4.4.0
10496          April 21, 2009 ([9]changes)
10497
10498References and Acknowledgements
10499
10500   GCC used to stand for the GNU C Compiler, but since the compiler
10501   supports several other languages aside from C, it now stands for the
10502   GNU Compiler Collection.
10503
10504   A list of [10]successful builds is updated as new information becomes
10505   available.
10506
10507   The GCC developers would like to thank the numerous people that have
10508   contributed new features, improvements, bug fixes, and other changes as
10509   well as test results to GCC. This [11]amazing group of volunteers is
10510   what makes GCC successful.
10511
10512   For additional information about GCC please refer to the [12]GCC
10513   project web site or contact the [13]GCC development mailing list.
10514
10515   To obtain GCC please use [14]our mirror sites or [15]our version
10516   control system.
10517
10518
10519    For questions related to the use of GCC, please consult these web
10520    pages and the [16]GCC manuals. If that fails, the
10521    [17]gcc-help@gcc.gnu.org mailing list might help. Comments on these
10522    web pages and the development of GCC are welcome on our developer
10523    list at [18]gcc@gcc.gnu.org. All of [19]our lists have public
10524    archives.
10525
10526   Copyright (C) [20]Free Software Foundation, Inc. Verbatim copying and
10527   distribution of this entire article is permitted in any medium,
10528   provided this notice is preserved.
10529
10530   These pages are [21]maintained by the GCC team. Last modified
10531   2020-01-14[22].
10532
10533References
10534
10535   1. http://www.gnu.org/
10536   2. http://gcc.gnu.org/gcc-4.4/changes.html
10537   3. http://gcc.gnu.org/gcc-4.4/changes.html
10538   4. http://gcc.gnu.org/gcc-4.4/changes.html
10539   5. http://gcc.gnu.org/gcc-4.4/changes.html
10540   6. http://gcc.gnu.org/gcc-4.4/changes.html
10541   7. http://gcc.gnu.org/gcc-4.4/changes.html
10542   8. http://gcc.gnu.org/gcc-4.4/changes.html
10543   9. http://gcc.gnu.org/gcc-4.4/changes.html
10544  10. http://gcc.gnu.org/gcc-4.4/buildstat.html
10545  11. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
10546  12. http://gcc.gnu.org/index.html
10547  13. mailto:gcc@gcc.gnu.org
10548  14. http://gcc.gnu.org/mirrors.html
10549  15. http://gcc.gnu.org/git.html
10550  16. https://gcc.gnu.org/onlinedocs/
10551  17. mailto:gcc-help@gcc.gnu.org
10552  18. mailto:gcc@gcc.gnu.org
10553  19. https://gcc.gnu.org/lists.html
10554  20. https://www.fsf.org/
10555  21. https://gcc.gnu.org/about.html
10556  22. http://validator.w3.org/check/referer
10557======================================================================
10558http://gcc.gnu.org/gcc-4.4/changes.html
10559
10560                             GCC 4.4 Release Series
10561                        Changes, New Features, and Fixes
10562
10563   The latest release in the 4.4 release series is [1]GCC 4.4.7.
10564
10565Caveats
10566
10567     * __builtin_stdarg_start has been completely removed from GCC.
10568       Support for <varargs.h> had been deprecated since GCC 4.0. Use
10569       __builtin_va_start as a replacement.
10570     * Some of the errors issued by the C++ front end that could be
10571       downgraded to warnings in previous releases by using -fpermissive
10572       are now warnings by default. They can be converted into errors by
10573       using -pedantic-errors.
10574     * Use of the cpp assertion extension will now emit a warning when
10575       -Wdeprecated or -pedantic is used. This extension has been
10576       deprecated for many years, but never warned about.
10577     * Packed bit-fields of type char were not properly bit-packed on many
10578       targets prior to GCC 4.4. On these targets, the fix in GCC 4.4
10579       causes an ABI change. For example there is no longer a 4-bit
10580       padding between field a and b in this structure:
10581    struct foo
10582    {
10583      char a:4;
10584      char b:8;
10585    } __attribute__ ((packed));
10586       There is a new warning to help identify fields that are affected:
10587    foo.c:5: note: Offset of packed bit-field 'b' has changed in GCC 4.4
10588       The warning can be disabled with -Wno-packed-bitfield-compat.
10589     * On ARM EABI targets, the C++ mangling of the va_list type has been
10590       changed to conform to the current revision of the EABI. This does
10591       not affect the libstdc++ library included with GCC.
10592     * The SCOUNT and POS bits of the MIPS DSP control register are now
10593       treated as global. Previous versions of GCC treated these fields as
10594       call-clobbered instead.
10595     * The MIPS port no longer recognizes the h asm constraint. It was
10596       necessary to remove this constraint in order to avoid generating
10597       unpredictable code sequences.
10598       One of the main uses of the h constraint was to extract the high
10599       part of a multiplication on 64-bit targets. For example:
10600    asm ("dmultu\t%1,%2" : "=h" (result) : "r" (x), "r" (y));
10601       You can now achieve the same effect using 128-bit types:
10602    typedef unsigned int uint128_t __attribute__((mode(TI)));
10603    result = ((uint128_t) x * y) >> 64;
10604       The second sequence is better in many ways. For example, if x and y
10605       are constants, the compiler can perform the multiplication at
10606       compile time. If x and y are not constants, the compiler can
10607       schedule the runtime multiplication better than it can schedule an
10608       asm statement.
10609     * Support for a number of older systems and recently unmaintained or
10610       untested target ports of GCC has been declared obsolete in GCC 4.4.
10611       Unless there is activity to revive them, the next release of GCC
10612       will have their sources permanently removed.
10613       The following ports for individual systems on particular
10614       architectures have been obsoleted:
10615          + Generic a.out on IA32 and m68k (i[34567]86-*-aout*,
10616            m68k-*-aout*)
10617          + Generic COFF on ARM, H8300, IA32, m68k and SH (arm-*-coff*,
10618            armel-*-coff*, h8300-*-*, i[34567]86-*-coff*, m68k-*-coff*,
10619            sh-*-*). This does not affect other more specific targets
10620            using the COFF object format on those architectures, or the
10621            more specific H8300 and SH targets (h8300-*-rtems*,
10622            h8300-*-elf*, sh-*-elf*, sh-*-symbianelf*, sh-*-linux*,
10623            sh-*-netbsdelf*, sh-*-rtems*, sh-wrs-vxworks).
10624          + 2BSD on PDP-11 (pdp11-*-bsd)
10625          + AIX 4.1 and 4.2 on PowerPC (rs6000-ibm-aix4.[12]*,
10626            powerpc-ibm-aix4.[12]*)
10627          + Tuning support for Itanium1 (Merced) variants. Note that code
10628            tuned for Itanium2 should also run correctly on Itanium1.
10629     * The protoize and unprotoize utilities have been obsoleted and will
10630       be removed in GCC 4.5. These utilities have not been installed by
10631       default since GCC 3.0.
10632     * Support has been removed for all the [2]configurations obsoleted in
10633       GCC 4.3.
10634     * Unknown -Wno-* options are now silently ignored by GCC if no other
10635       diagnostics are issued. If other diagnostics are issued, then GCC
10636       warns about the unknown options.
10637     * More information on porting to GCC 4.4 from previous versions of
10638       GCC can be found in the [3]porting guide for this release.
10639
10640General Optimizer Improvements
10641
10642     * A new command-line switch -findirect-inlining has been added. When
10643       turned on it allows the inliner to also inline indirect calls that
10644       are discovered to have known targets at compile time thanks to
10645       previous inlining.
10646     * A new command-line switch -ftree-switch-conversion has been added.
10647       This new pass turns simple initializations of scalar variables in
10648       switch statements into initializations from a static array, given
10649       that all the values are known at compile time and the ratio between
10650       the new array size and the original switch branches does not exceed
10651       the parameter --param switch-conversion-max-branch-ratio (default
10652       is eight).
10653     * A new command-line switch -ftree-builtin-call-dce has been added.
10654       This optimization eliminates unnecessary calls to certain builtin
10655       functions when the return value is not used, in cases where the
10656       calls can not be eliminated entirely because the function may set
10657       errno. This optimization is on by default at -O2 and above.
10658     * A new command-line switch -fconserve-stack directs the compiler to
10659       minimize stack usage even if it makes the generated code slower.
10660       This affects inlining decisions.
10661     * When the assembler supports it, the compiler will now emit unwind
10662       information using assembler .cfi directives. This makes it possible
10663       to use such directives in inline assembler code. The new option
10664       -fno-dwarf2-cfi-asm directs the compiler to not use .cfi
10665       directives.
10666     * The [4]Graphite branch has been merged. This merge has brought in a
10667       new framework for loop optimizations based on a polyhedral
10668       intermediate representation. These optimizations apply to all the
10669       languages supported by GCC. The following new code transformations
10670       are available in GCC 4.4:
10671          + -floop-interchange performs loop interchange transformations
10672            on loops. Interchanging two nested loops switches the inner
10673            and outer loops. For example, given a loop like:
10674          DO J = 1, M
10675            DO I = 1, N
10676              A(J, I) = A(J, I) * C
10677            ENDDO
10678          ENDDO
10679
10680            loop interchange will transform the loop as if the user had
10681            written:
10682          DO I = 1, N
10683            DO J = 1, M
10684              A(J, I) = A(J, I) * C
10685            ENDDO
10686          ENDDO
10687
10688            which can be beneficial when N is larger than the caches,
10689            because in Fortran, the elements of an array are stored in
10690            memory contiguously by column, and the original loop iterates
10691            over rows, potentially creating at each access a cache miss.
10692          + -floop-strip-mine performs loop strip mining transformations
10693            on loops. Strip mining splits a loop into two nested loops.
10694            The outer loop has strides equal to the strip size and the
10695            inner loop has strides of the original loop within a strip.
10696            For example, given a loop like:
10697          DO I = 1, N
10698            A(I) = A(I) + C
10699          ENDDO
10700
10701            loop strip mining will transform the loop as if the user had
10702            written:
10703          DO II = 1, N, 4
10704            DO I = II, min (II + 3, N)
10705              A(I) = A(I) + C
10706            ENDDO
10707          ENDDO
10708
10709          + -floop-block performs loop blocking transformations on loops.
10710            Blocking strip mines each loop in the loop nest such that the
10711            memory accesses of the element loops fit inside caches. For
10712            example, given a loop like:
10713          DO I = 1, N
10714            DO J = 1, M
10715              A(J, I) = B(I) + C(J)
10716            ENDDO
10717          ENDDO
10718
10719            loop blocking will transform the loop as if the user had
10720            written:
10721          DO II = 1, N, 64
10722            DO JJ = 1, M, 64
10723              DO I = II, min (II + 63, N)
10724                DO J = JJ, min (JJ + 63, M)
10725                  A(J, I) = B(I) + C(J)
10726                ENDDO
10727              ENDDO
10728            ENDDO
10729          ENDDO
10730
10731            which can be beneficial when M is larger than the caches,
10732            because the innermost loop will iterate over a smaller amount
10733            of data that can be kept in the caches.
10734     * A new register allocator has replaced the old one. It is called
10735       integrated register allocator (IRA) because coalescing, register
10736       live range splitting, and hard register preferencing are done
10737       on-the-fly during coloring. It also has better integration with the
10738       reload pass. IRA is a regional register allocator which uses modern
10739       Chaitin-Briggs coloring instead of Chow's priority coloring used in
10740       the old register allocator. More info about IRA internals and
10741       options can be found in the GCC manuals.
10742     * A new instruction scheduler and software pipeliner, based on the
10743       selective scheduling approach, has been added. The new pass
10744       performs instruction unification, register renaming, substitution
10745       through register copies, and speculation during scheduling. The
10746       software pipeliner is able to pipeline non-countable loops. The new
10747       pass is targeted at scheduling-eager in-order platforms. In GCC 4.4
10748       it is available for the Intel Itanium platform working by default
10749       as the second scheduling pass (after register allocation) at the
10750       -O3 optimization level.
10751     * When using -fprofile-generate with a multi-threaded program, the
10752       profile counts may be slightly wrong due to race conditions. The
10753       new -fprofile-correction option directs the compiler to apply
10754       heuristics to smooth out the inconsistencies. By default the
10755       compiler will give an error message when it finds an inconsistent
10756       profile.
10757     * The new -fprofile-dir=PATH option permits setting the directory
10758       where profile data files are stored when using -fprofile-generate
10759       and friends, and the directory used when reading profile data files
10760       using -fprofile-use and friends.
10761
10762New warning options
10763
10764     * The new -Wframe-larger-than=NUMBER option directs GCC to emit a
10765       warning if any stack frame is larger than NUMBER bytes. This may be
10766       used to help ensure that code fits within a limited amount of stack
10767       space.
10768     * The command-line option -Wlarger-than-N is now written as
10769       -Wlarger-than=N and the old form is deprecated.
10770     * The new -Wno-mudflap option disables warnings about constructs
10771       which can not be instrumented when using -fmudflap.
10772
10773New Languages and Language specific improvements
10774
10775     * Version 3.0 of the OpenMP specification is now supported for the C,
10776       C++, and Fortran compilers.
10777     * New character data types, per [5]TR 19769: New character types in
10778       C, are now supported for the C compiler in -std=gnu99 mode, as
10779       __CHAR16_TYPE__ and __CHAR32_TYPE__, and for the C++ compiler in
10780       -std=c++0x and -std=gnu++0x modes, as char16_t and char32_t too.
10781
10782  C family
10783
10784     * A new optimize attribute was added to allow programmers to change
10785       the optimization level and particular optimization options for an
10786       individual function. You can also change the optimization options
10787       via the GCC optimize pragma for functions defined after the pragma.
10788       The GCC push_options pragma and the GCC pop_options pragma allow
10789       you temporarily save and restore the options used. The GCC
10790       reset_options pragma restores the options to what was specified on
10791       the command line.
10792     * Uninitialized warnings do not require enabling optimization
10793       anymore, that is, -Wuninitialized can be used together with -O0.
10794       Nonetheless, the warnings given by -Wuninitialized will probably be
10795       more accurate if optimization is enabled.
10796     * -Wparentheses now warns about expressions such as (!x | y) and (!x
10797       & y). Using explicit parentheses, such as in ((!x) | y), silences
10798       this warning.
10799     * -Wsequence-point now warns within if, while,do while and for
10800       conditions, and within for begin/end expressions.
10801     * A new option -dU is available to dump definitions of preprocessor
10802       macros that are tested or expanded.
10803
10804  C++
10805
10806     * [6]Improved experimental support for the upcoming ISO C++ standard,
10807       C++0x. Including support for auto, inline namespaces, generalized
10808       initializer lists, defaulted and deleted functions, new character
10809       types, and scoped enums.
10810     * Those errors that may be downgraded to warnings to build legacy
10811       code now mention -fpermissive when -fdiagnostics-show-option is
10812       enabled.
10813     * -Wconversion now warns if the result of a static_cast to enumeral
10814       type is unspecified because the value is outside the range of the
10815       enumeral type.
10816     * -Wuninitialized now warns if a non-static reference or non-static
10817       const member appears in a class without constructors.
10818     * G++ now properly implements value-initialization, so objects with
10819       an initializer of () and an implicitly defined default constructor
10820       will be zero-initialized before the default constructor is called.
10821
10822    Runtime Library (libstdc++)
10823
10824     * Improved experimental support for the upcoming ISO C++ standard,
10825       C++0x, including:
10826          + Support for <chrono>, <condition_variable>, <cstdatomic>,
10827            <forward_list>, <initializer_list>, <mutex>, <ratio>,
10828            <system_error>, and <thread>.
10829          + unique_ptr, <algorithm> additions, exception propagation, and
10830            support for the new character types in <string> and <limits>.
10831          + Existing facilities now exploit initializer lists, defaulted
10832            and deleted functions, and the newly implemented core C++0x
10833            features.
10834          + Some standard containers are more efficient together with
10835            stateful allocators, i.e., no allocator is constructed on the
10836            fly at element construction time.
10837     * Experimental support for non-standard pointer types in containers.
10838     * The long standing libstdc++/30928 has been fixed for targets
10839       running glibc 2.10 or later.
10840     * As usual, many small and larger bug fixes, in particular quite a
10841       few corner cases in <locale>.
10842
10843  Fortran
10844
10845     * GNU Fortran now employs libcpp directly instead of using cc1 as an
10846       external preprocessor. The [7]-cpp option was added to allow manual
10847       invocation of the preprocessor without relying on filename
10848       extensions.
10849     * The [8]-Warray-temporaries option warns about array temporaries
10850       generated by the compiler, as an aid to optimization.
10851     * The [9]-fcheck-array-temporaries option has been added, printing a
10852       notification at run time, when an array temporary had to be created
10853       for an function argument. Contrary to -Warray-temporaries the
10854       warning is only printed if the array is noncontiguous.
10855     * Improved generation of DWARF debugging symbols
10856     * If using an intrinsic not part of the selected standard (via -std=
10857       and -fall-intrinsics) gfortran will now treat it as if this
10858       procedure were declared EXTERNAL and try to link to a user-supplied
10859       procedure. -Wintrinsics-std will warn whenever this happens. The
10860       now-useless option -Wnonstd-intrinsic was removed.
10861     * The flag -falign-commons has been added to control the alignment of
10862       variables in COMMON blocks, which is enabled by default in line
10863       with previous GCC version. Using -fno-align-commons one can force
10864       commons to be contiguous in memory as required by the Fortran
10865       standard, however, this slows down the memory access. The option
10866       -Walign-commons, which is enabled by default, warns when padding
10867       bytes were added for alignment. The proper solution is to sort the
10868       common objects by decreasing storage size, which avoids the
10869       alignment problems.
10870     * Fortran 2003 support has been extended:
10871          + Wide characters (ISO 10646, UCS-4, kind=4) and UTF-8 I/O is
10872            now supported (except internal reads from/writes to wide
10873            strings). [10]-fbackslash now supports also \unnnn and
10874            \Unnnnnnnn to enter Unicode characters.
10875          + Asynchronous I/O (implemented as synchronous I/O) and the
10876            decimal=, size=, sign=, pad=, blank=, and delim= specifiers
10877            are now supported in I/O statements.
10878          + Support for Fortran 2003 structure constructors and for array
10879            constructor with typespec has been added.
10880          + Procedure Pointers (but not yet as component in derived types
10881            and as function results) are now supported.
10882          + Abstract types, type extension, and type-bound procedures
10883            (both PROCEDURE and GENERIC but not as operators). Note: As
10884            CLASS/polymorphyic types are not implemented, type-bound
10885            procedures with PASS accept as non-standard extension TYPE
10886            arguments.
10887     * Fortran 2008 support has been added:
10888          + The -std=f2008 option and support for the file extensions
10889            .f2008 and .F2008 has been added.
10890          + The g0 format descriptor is now supported.
10891          + The Fortran 2008 mathematical intrinsics ASINH, ACOSH, ATANH,
10892            ERF, ERFC, GAMMA, LOG_GAMMA, BESSEL_*, HYPOT, and ERFC_SCALED
10893            are now available (some of them existed as GNU extension
10894            before). Note: The hyperbolic functions are not yet supporting
10895            complex arguments and the three- argument version of BESSEL_*N
10896            is not available.
10897          + The bit intrinsics LEADZ and TRAILZ have been added.
10898
10899  Java (GCJ)
10900
10901  Ada
10902
10903     * The Ada runtime now supports multilibs on many platforms including
10904       x86_64, SPARC and PowerPC. Their build is enabled by default.
10905
10906New Targets and Target Specific Improvements
10907
10908  ARM
10909
10910     * GCC now supports optimizing for the Cortex-A9, Cortex-R4 and
10911       Cortex-R4F processors and has many other improvements to
10912       optimization for ARM processors.
10913     * GCC now supports the VFPv3 variant with 16 double-precision
10914       registers with -mfpu=vfpv3-d16. The option -mfpu=vfp3 has been
10915       renamed to -mfpu=vfpv3.
10916     * GCC now supports the -mfix-cortex-m3-ldrd option to work around an
10917       erratum on Cortex-M3 processors.
10918     * GCC now supports the __sync_* atomic operations for ARM EABI
10919       GNU/Linux.
10920     * The section anchors optimization is now enabled by default when
10921       optimizing for ARM.
10922     * GCC now uses a new EABI-compatible profiling interface for EABI
10923       targets. This requires a function __gnu_mcount_nc, which is
10924       provided by GNU libc versions 2.8 and later.
10925
10926  AVR
10927
10928     * The -mno-tablejump option has been deprecated because it has the
10929       same effect as the -fno-jump-tables option.
10930     * Added support for these new AVR devices:
10931          + ATA6289
10932          + ATtiny13A
10933          + ATtiny87
10934          + ATtiny167
10935          + ATtiny327
10936          + ATmega8C1
10937          + ATmega16C1
10938          + ATmega32C1
10939          + ATmega8M1
10940          + ATmega16M1
10941          + ATmega32M1
10942          + ATmega32U4
10943          + ATmega16HVB
10944          + ATmega4HVD
10945          + ATmega8HVD
10946          + ATmega64C1
10947          + ATmega64M1
10948          + ATmega16U4
10949          + ATmega32U6
10950          + ATmega128RFA1
10951          + AT90PWM81
10952          + AT90SCR100
10953          + M3000F
10954          + M3000S
10955          + M3001B
10956
10957  IA-32/x86-64
10958
10959     * Support for Intel AES built-in functions and code generation is
10960       available via -maes.
10961     * Support for Intel PCLMUL built-in function and code generation is
10962       available via -mpclmul.
10963     * Support for Intel AVX built-in functions and code generation is
10964       available via -mavx.
10965     * Automatically align the stack for local variables with alignment
10966       requirement.
10967     * GCC can now utilize the SVML library for vectorizing calls to a set
10968       of C99 functions if -mveclibabi=svml is specified and you link to
10969       an SVML ABI compatible library.
10970     * On x86-64, the ABI has been changed in the following cases to
10971       conform to the x86-64 ABI:
10972          + Passing/returning structures with flexible array member:
10973  struct foo
10974    {
10975      int i;
10976      int flex[];
10977    };
10978          + Passing/returning structures with complex float member:
10979  struct foo
10980    {
10981      int i;
10982      __complex__ float f;
10983    };
10984          + Passing/returning unions with long double member:
10985  union foo
10986    {
10987      int x;
10988      long double ld;
10989    };
10990       Code built with previous versions of GCC that uses any of these is
10991       not compatible with code built with GCC 4.4.0 or later.
10992     * A new target attribute was added to allow programmers to change the
10993       target options like -msse2 or -march=k8 for an individual function.
10994       You can also change the target options via the GCC target pragma
10995       for functions defined after the pragma.
10996     * GCC can now be configured with options --with-arch-32,
10997       --with-arch-64, --with-cpu-32, --with-cpu-64, --with-tune-32 and
10998       --with-tune-64 to control the default optimization separately for
10999       32-bit and 64-bit modes.
11000
11001  IA-32/IA64
11002
11003     * Support for __float128 (TFmode) IEEE quad type and corresponding
11004       TCmode IEEE complex quad type is available via the soft-fp library
11005       on IA-32/IA64 targets. This includes basic arithmetic operations
11006       (addition, subtraction, negation, multiplication and division) on
11007       __float128 real and TCmode complex values, the full set of IEEE
11008       comparisons between __float128 values, conversions to and from
11009       float, double and long double floating point types, as well as
11010       conversions to and from signed or unsigned integer, signed or
11011       unsigned long integer and signed or unsigned quad (TImode, IA64
11012       only) integer types. Additionally, all operations generate the full
11013       set of IEEE exceptions and support the full set of IEEE rounding
11014       modes.
11015
11016  M68K/ColdFire
11017
11018     * GCC now supports instruction scheduling for ColdFire V1, V3 and V4
11019       processors. (Scheduling support for ColdFire V2 processors was
11020       added in GCC 4.3.)
11021     * GCC now supports the -mxgot option to support programs requiring
11022       many GOT entries on ColdFire.
11023     * The m68k-*-linux-gnu target now builds multilibs by default.
11024
11025  MIPS
11026
11027     * MIPS Technologies have extended the original MIPS SVR4 ABI to
11028       include support for procedure linkage tables (PLTs) and copy
11029       relocations. These extensions allow GNU/Linux executables to use a
11030       significantly more efficient code model than the one defined by the
11031       original ABI.
11032       GCC support for this code model is available via a new command-line
11033       option, -mplt. There is also a new configure-time option,
11034       --with-mips-plt, to make -mplt the default.
11035       The new code model requires support from the assembler, the linker,
11036       and the runtime C library. This support is available in binutils
11037       2.19 and GLIBC 2.9.
11038     * GCC can now generate MIPS16 code for 32-bit GNU/Linux executables
11039       and 32-bit GNU/Linux shared libraries. This feature requires GNU
11040       binutils 2.19 or above.
11041     * Support for RMI's XLR processor is now available through the
11042       -march=xlr and -mtune=xlr options.
11043     * 64-bit targets can now perform 128-bit multiplications inline,
11044       instead of relying on a libgcc function.
11045     * Native GNU/Linux toolchains now support -march=native and
11046       -mtune=native, which select the host processor.
11047     * GCC now supports the R10K, R12K, R14K and R16K processors. The
11048       canonical -march= and -mtune= names for these processors are
11049       r10000, r12000, r14000 and r16000 respectively.
11050     * GCC can now work around the side effects of speculative execution
11051       on R10K processors. Please see the documentation of the
11052       -mr10k-cache-barrier option for details.
11053     * Support for the MIPS64 Release 2 instruction set has been added.
11054       The option -march=mips64r2 enables generation of these
11055       instructions.
11056     * GCC now supports Cavium Networks' Octeon processor. This support is
11057       available through the -march=octeon and -mtune=octeon options.
11058     * GCC now supports STMicroelectronics' Loongson 2E/2F processors. The
11059       canonical -march= and -mtune= names for these processors are
11060       loongson2e and loongson2f.
11061
11062  picochip
11063
11064   Picochip is a 16-bit processor. A typical picoChip contains over 250
11065   small cores, each with small amounts of memory. There are three
11066   processor variants (STAN, MEM and CTRL) with different instruction sets
11067   and memory configurations and they can be chosen using the -mae option.
11068
11069   This port is intended to be a "C" only port.
11070
11071  Power Architecture and PowerPC
11072
11073     * GCC now supports the e300c2, e300c3 and e500mc processors.
11074     * GCC now supports Xilinx processors with a single-precision FPU.
11075     * Decimal floating point is now supported for e500 processors.
11076
11077  S/390, zSeries and System z9/z10
11078
11079     * Support for the IBM System z10 EC/BC processor has been added. When
11080       using the -march=z10 option, the compiler will generate code making
11081       use of instructions provided by the General-Instruction-Extension
11082       Facility and the Execute-Extension Facility.
11083
11084  VxWorks
11085
11086     * GCC now supports the thread-local storage mechanism used on
11087       VxWorks.
11088
11089  Xtensa
11090
11091     * GCC now supports thread-local storage (TLS) for Xtensa processor
11092       configurations that include the Thread Pointer option. TLS also
11093       requires support from the assembler and linker; this support is
11094       provided in the GNU binutils beginning with version 2.19.
11095
11096Documentation improvements
11097
11098Other significant improvements
11099
11100GCC 4.4.1
11101
11102   This is the [11]list of problem reports (PRs) from GCC's bug tracking
11103   system that are known to be fixed in the 4.4.1 release. This list might
11104   not be complete (that is, it is possible that some PRs that have been
11105   fixed are not listed here).
11106
11107GCC 4.4.2
11108
11109   This is the [12]list of problem reports (PRs) from GCC's bug tracking
11110   system that are known to be fixed in the 4.4.2 release. This list might
11111   not be complete (that is, it is possible that some PRs that have been
11112   fixed are not listed here).
11113
11114GCC 4.4.3
11115
11116   This is the [13]list of problem reports (PRs) from GCC's bug tracking
11117   system that are known to be fixed in the 4.4.3 release. This list might
11118   not be complete (that is, it is possible that some PRs that have been
11119   fixed are not listed here).
11120
11121GCC 4.4.4
11122
11123   This is the [14]list of problem reports (PRs) from GCC's bug tracking
11124   system that are known to be fixed in the 4.4.4 release. This list might
11125   not be complete (that is, it is possible that some PRs that have been
11126   fixed are not listed here).
11127
11128GCC 4.4.5
11129
11130   This is the [15]list of problem reports (PRs) from GCC's bug tracking
11131   system that are known to be fixed in the 4.4.5 release. This list might
11132   not be complete (that is, it is possible that some PRs that have been
11133   fixed are not listed here).
11134
11135GCC 4.4.6
11136
11137   This is the [16]list of problem reports (PRs) from GCC's bug tracking
11138   system that are known to be fixed in the 4.4.6 release. This list might
11139   not be complete (that is, it is possible that some PRs that have been
11140   fixed are not listed here).
11141
11142GCC 4.4.7
11143
11144   This is the [17]list of problem reports (PRs) from GCC's bug tracking
11145   system that are known to be fixed in the 4.4.7 release. This list might
11146   not be complete (that is, it is possible that some PRs that have been
11147   fixed are not listed here).
11148
11149
11150    For questions related to the use of GCC, please consult these web
11151    pages and the [18]GCC manuals. If that fails, the
11152    [19]gcc-help@gcc.gnu.org mailing list might help. Comments on these
11153    web pages and the development of GCC are welcome on our developer
11154    list at [20]gcc@gcc.gnu.org. All of [21]our lists have public
11155    archives.
11156
11157   Copyright (C) [22]Free Software Foundation, Inc. Verbatim copying and
11158   distribution of this entire article is permitted in any medium,
11159   provided this notice is preserved.
11160
11161   These pages are [23]maintained by the GCC team. Last modified
11162   2019-11-28[24].
11163
11164References
11165
11166   1. http://gcc.gnu.org/gcc-4.4/changes.html#4.4.7
11167   2. http://gcc.gnu.org/gcc-4.3/changes.html#obsoleted
11168   3. http://gcc.gnu.org/gcc-4.4/porting_to.html
11169   4. https://gcc.gnu.org/wiki/Graphite
11170   5. http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1040.pdf
11171   6. http://gcc.gnu.org/gcc-4.4/cxx0x_status.html
11172   7. https://gcc.gnu.org/onlinedocs/gfortran/Preprocessing-Options.html
11173   8. https://gcc.gnu.org/onlinedocs/gfortran/Error-and-Warning-Options.html#index-g_t_0040code_007bWarray-temporaries_007d-125
11174   9. https://gcc.gnu.org/onlinedocs/gfortran/Code-Gen-Options.html#index-g_t_0040code_007bfcheck-array-temporaries_007d-221
11175  10. https://gcc.gnu.org/onlinedocs/gfortran/Fortran-Dialect-Options.html#index-g_t_0040code_007bbackslash_007d-34
11176  11. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.4.1
11177  12. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.4.2
11178  13. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.4.3
11179  14. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.4.4
11180  15. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.4.5
11181  16. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.4.6
11182  17. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.4.7
11183  18. https://gcc.gnu.org/onlinedocs/
11184  19. mailto:gcc-help@gcc.gnu.org
11185  20. mailto:gcc@gcc.gnu.org
11186  21. https://gcc.gnu.org/lists.html
11187  22. https://www.fsf.org/
11188  23. https://gcc.gnu.org/about.html
11189  24. http://validator.w3.org/check/referer
11190======================================================================
11191http://gcc.gnu.org/gcc-4.3/index.html
11192
11193                             GCC 4.3 Release Series
11194
11195   (This release series is no longer supported.)
11196
11197   Jun 27, 2011
11198
11199   The [1]GNU project and the GCC developers are pleased to announce the
11200   release of GCC 4.3.6.
11201
11202   This release is a bug-fix release, containing fixes for regressions in
11203   GCC 4.3.5 relative to previous releases of GCC.
11204
11205Release History
11206
11207   GCC 4.3.6
11208          Jun 27, 2011 ([2]changes)
11209
11210   GCC 4.3.5
11211          May 22, 2010 ([3]changes)
11212
11213   GCC 4.3.4
11214          August 4, 2009 ([4]changes)
11215
11216   GCC 4.3.3
11217          January 24, 2009 ([5]changes)
11218
11219   GCC 4.3.2
11220          August 27, 2008 ([6]changes)
11221
11222   GCC 4.3.1
11223          June 6, 2008 ([7]changes)
11224
11225   GCC 4.3.0
11226          March 5, 2008 ([8]changes)
11227
11228References and Acknowledgements
11229
11230   GCC used to stand for the GNU C Compiler, but since the compiler
11231   supports several other languages aside from C, it now stands for the
11232   GNU Compiler Collection.
11233
11234   A list of [9]successful builds is updated as new information becomes
11235   available.
11236
11237   The GCC developers would like to thank the numerous people that have
11238   contributed new features, improvements, bug fixes, and other changes as
11239   well as test results to GCC. This [10]amazing group of volunteers is
11240   what makes GCC successful.
11241
11242   For additional information about GCC please refer to the [11]GCC
11243   project web site or contact the [12]GCC development mailing list.
11244
11245   To obtain GCC please use [13]our mirror sites or [14]our version
11246   control system.
11247
11248
11249    For questions related to the use of GCC, please consult these web
11250    pages and the [15]GCC manuals. If that fails, the
11251    [16]gcc-help@gcc.gnu.org mailing list might help. Comments on these
11252    web pages and the development of GCC are welcome on our developer
11253    list at [17]gcc@gcc.gnu.org. All of [18]our lists have public
11254    archives.
11255
11256   Copyright (C) [19]Free Software Foundation, Inc. Verbatim copying and
11257   distribution of this entire article is permitted in any medium,
11258   provided this notice is preserved.
11259
11260   These pages are [20]maintained by the GCC team. Last modified
11261   2020-01-14[21].
11262
11263References
11264
11265   1. http://www.gnu.org/
11266   2. http://gcc.gnu.org/gcc-4.3/changes.html
11267   3. http://gcc.gnu.org/gcc-4.3/changes.html
11268   4. http://gcc.gnu.org/gcc-4.3/changes.html
11269   5. http://gcc.gnu.org/gcc-4.3/changes.html
11270   6. http://gcc.gnu.org/gcc-4.3/changes.html
11271   7. http://gcc.gnu.org/gcc-4.3/changes.html
11272   8. http://gcc.gnu.org/gcc-4.3/changes.html
11273   9. http://gcc.gnu.org/gcc-4.3/buildstat.html
11274  10. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
11275  11. http://gcc.gnu.org/index.html
11276  12. mailto:gcc@gcc.gnu.org
11277  13. http://gcc.gnu.org/mirrors.html
11278  14. http://gcc.gnu.org/git.html
11279  15. https://gcc.gnu.org/onlinedocs/
11280  16. mailto:gcc-help@gcc.gnu.org
11281  17. mailto:gcc@gcc.gnu.org
11282  18. https://gcc.gnu.org/lists.html
11283  19. https://www.fsf.org/
11284  20. https://gcc.gnu.org/about.html
11285  21. http://validator.w3.org/check/referer
11286======================================================================
11287http://gcc.gnu.org/gcc-4.3/changes.html
11288
11289                             GCC 4.3 Release Series
11290                        Changes, New Features, and Fixes
11291
11292   The latest release in the 4.3 release series is [1]GCC 4.3.5.
11293
11294Caveats
11295
11296     * GCC requires the [2]GMP and [3]MPFR libraries for building all the
11297       various front-end languages it supports. See the [4]prerequisites
11298       page for version requirements.
11299     * ColdFire targets now treat long double as having the same format as
11300       double. In earlier versions of GCC, they used the 68881 long double
11301       format instead.
11302     * The m68k-uclinux target now uses the same calling conventions as
11303       m68k-linux-gnu. You can select the original calling conventions by
11304       configuring for m68k-uclinuxoldabi instead. Note that
11305       m68k-uclinuxoldabi also retains the original 80-bit long double on
11306       ColdFire targets.
11307     * The -fforce-mem option has been removed because it has had no
11308       effect in the last few GCC releases.
11309     * The i386 -msvr3-shlib option has been removed since it is no longer
11310       used.
11311     * Fastcall for i386 has been changed not to pass aggregate arguments
11312       in registers, following Microsoft compilers.
11313     * Support for the AOF assembler has been removed from the ARM back
11314       end; this affects only the targets arm-semi-aof and armel-semi-aof,
11315       which are no longer recognized. We removed these targets without a
11316       deprecation period because we discovered that they have been
11317       unusable since GCC 4.0.0.
11318     * Support for the TMS320C3x/C4x processor (targets c4x-* and tic4x-*)
11319       has been removed. This support had been deprecated since GCC 4.0.0.
11320     * Support for a number of older systems and recently unmaintained or
11321       untested target ports of GCC has been declared obsolete in GCC 4.3.
11322       Unless there is activity to revive them, the next release of GCC
11323       will have their sources permanently removed.
11324       All GCC ports for the following processor architectures have been
11325       declared obsolete:
11326          + Morpho MT (mt-*)
11327       The following aliases for processor architectures have been
11328       declared obsolete. Users should use the indicated generic target
11329       names instead, with compile-time options such as -mcpu or
11330       configure-time options such as --with-cpu to control the
11331       configuration more precisely.
11332          + strongarm*-*-*, ep9312*-*-*, xscale*-*-* (use arm*-*-*
11333            instead).
11334          + parisc*-*-* (use hppa*-*-* instead).
11335          + m680[012]0-*-* (use m68k-*-* instead).
11336       All GCC ports for the following operating systems have been
11337       declared obsolete:
11338          + BeOS (*-*-beos*)
11339          + kaOS (*-*-kaos*)
11340          + GNU/Linux using the a.out object format (*-*-linux*aout*)
11341          + GNU/Linux using version 1 of the GNU C Library
11342            (*-*-linux*libc1*)
11343          + Solaris versions before Solaris 7 (*-*-solaris2.[0-6],
11344            *-*-solaris2.[0-6].*)
11345          + Miscellaneous System V (*-*-sysv*)
11346          + WindISS (*-*-windiss*)
11347       Also, those for some individual systems on particular architectures
11348       have been obsoleted:
11349          + UNICOS/mk on DEC Alpha (alpha*-*-unicosmk*)
11350          + CRIS with a.out object format (cris-*-aout)
11351          + BSD 4.3 on PA-RISC (hppa1.1-*-bsd*)
11352          + OSF/1 on PA-RISC (hppa1.1-*-osf*)
11353          + PRO on PA-RISC (hppa1.1-*-pro*)
11354          + Sequent PTX on IA32 (i[34567]86-sequent-ptx4*,
11355            i[34567]86-sequent-sysv4*)
11356          + SCO Open Server 5 on IA32 (i[34567]86-*-sco3.2v5*)
11357          + UWIN on IA32 (i[34567]86-*-uwin*) (support for UWIN as a host
11358            was previously [5]removed in 2001, leaving only the support
11359            for UWIN as a target now being deprecated)
11360          + ChorusOS on PowerPC (powerpc-*-chorusos*)
11361          + All VAX configurations apart from NetBSD and OpenBSD
11362            (vax-*-bsd*, vax-*-sysv*, vax-*-ultrix*)
11363     * The [6]-Wconversion option has been modified. Its purpose now is to
11364       warn for implicit conversions that may alter a value. This new
11365       behavior is available for both C and C++. Warnings about
11366       conversions between signed and unsigned integers can be disabled by
11367       using -Wno-sign-conversion. In C++, they are disabled by default
11368       unless -Wsign-conversion is explicitly requested. The old behavior
11369       of -Wconversion, that is, warn for prototypes causing a type
11370       conversion that is different from what would happen to the same
11371       argument in the absence of a prototype, has been moved to a new
11372       option -Wtraditional-conversion, which is only available for C.
11373     * The -m386, -m486, -mpentium and -mpentiumpro tuning options have
11374       been removed because they were deprecated for more than 3 GCC major
11375       releases. Use -mtune=i386, -mtune=i486, -mtune=pentium or
11376       -mtune=pentiumpro as a replacement.
11377     * The -funsafe-math-optimizations option now automatically turns on
11378       -fno-trapping-math in addition to -fno-signed-zeros, as it enables
11379       reassociation and thus may introduce or remove traps.
11380     * The -ftree-vectorize option is now on by default under -O3. In
11381       order to generate code for a SIMD extension, it has to be enabled
11382       as well: use -maltivec for PowerPC platforms and -msse/-msse2 for
11383       i?86 and x86_64.
11384     * More information on porting to GCC 4.3 from previous versions of
11385       GCC can be found in the [7]porting guide for this release.
11386
11387General Optimizer Improvements
11388
11389     * The GCC middle-end has been integrated with the MPFR library. This
11390       allows GCC to evaluate and replace at compile-time calls to
11391       built-in math functions having constant arguments with their
11392       mathematically equivalent results. In making use of MPFR, GCC can
11393       generate correct results regardless of the math library
11394       implementation or floating point precision of the host platform.
11395       This also allows GCC to generate identical results regardless of
11396       whether one compiles in native or cross-compile configurations to a
11397       particular target. The following built-in functions take advantage
11398       of this new capability: acos, acosh, asin, asinh, atan2, atan,
11399       atanh, cbrt, cos, cosh, drem, erf, erfc, exp10, exp2, exp, expm1,
11400       fdim, fma, fmax, fmin, gamma_r, hypot, j0, j1, jn, lgamma_r, log10,
11401       log1p, log2, log, pow10, pow, remainder, remquo, sin, sincos, sinh,
11402       tan, tanh, tgamma, y0, y1 and yn. The float and long double
11403       variants of these functions (e.g. sinf and sinl) are also handled.
11404       The sqrt and cabs functions with constant arguments were already
11405       optimized in prior GCC releases. Now they also use MPFR.
11406     * A new forward propagation pass on RTL was added. The new pass
11407       replaces several slower transformations, resulting in compile-time
11408       improvements as well as better code generation in some cases.
11409     * A new command-line switch -frecord-gcc-switches has been added to
11410       GCC, although it is only enabled for some targets. The switch
11411       causes the command line that was used to invoke the compiler to be
11412       recorded into the object file that is being created. The exact
11413       format of this recording is target and binary file format
11414       dependent, but it usually takes the form of a note section
11415       containing ASCII text. The switch is related to the -fverbose-asm
11416       switch, but that one only records the information in the assembler
11417       output file as comments, so the information never reaches the
11418       object file.
11419     * The inliner heuristic is now aware of stack frame consumption. New
11420       command-line parameters --param large-stack-frame and --param
11421       large-stack-frame-growth can be used to limit stack frame size
11422       growth caused by inlining.
11423     * During feedback directed optimizations, the expected block size the
11424       memcpy, memset and bzero functions operate on is discovered and for
11425       cases of commonly used small sizes, specialized inline code is
11426       generated.
11427     * __builtin_expect no longer requires its argument to be a compile
11428       time constant.
11429     * Interprocedural optimization was reorganized to work on functions
11430       in SSA form. This enables more precise and cheaper dataflow
11431       analysis and makes writing interprocedural optimizations easier.
11432       The following improvements have been implemented on top of this
11433       framework:
11434          + Pre-inline optimization: Selected local optimization passes
11435            are run before the inliner (and other interprocedural passes)
11436            are executed. This significantly improves the accuracy of code
11437            growth estimates used by the inliner and reduces the overall
11438            memory footprint for large compilation units.
11439          + Early inlining (a simple bottom-up inliner pass inlining only
11440            functions whose body is smaller than the expected call
11441            overhead) is now executed with the early optimization passes,
11442            thus inlining already optimized function bodies into an
11443            unoptimized function that is subsequently optimized by early
11444            optimizers. This enables the compiler to quickly eliminate
11445            abstraction penalty in C++ programs.
11446          + Interprocedural constant propagation now operate on SSA form
11447            increasing accuracy of the analysis.
11448     * A new internal representation for GIMPLE statements has been
11449       contributed, resulting in compile-time memory savings.
11450     * The vectorizer was enhanced to support vectorization of outer
11451       loops, intra-iteration parallelism (loop-aware SLP), vectorization
11452       of strided accesses and loops with multiple data-types. Run-time
11453       dependency testing using loop versioning was added. The cost model,
11454       turned on by -fvect-cost-model, was developed.
11455
11456New Languages and Language specific improvements
11457
11458     * We have added new command-line options
11459       -finstrument-functions-exclude-function-list and
11460       -finstrument-functions-exclude-file-list. They provide more control
11461       over which functions are annotated by the -finstrument-functions
11462       option.
11463
11464  C family
11465
11466     * Implicit conversions between generic vector types are now only
11467       permitted when the two vectors in question have the same number of
11468       elements and compatible element types. (Note that the restriction
11469       involves compatible element types, not implicitly-convertible
11470       element types: thus, a vector type with element type int may not be
11471       implicitly converted to a vector type with element type unsigned
11472       int.) This restriction, which is in line with specifications for
11473       SIMD architectures such as AltiVec, may be relaxed using the flag
11474       -flax-vector-conversions. This flag is intended only as a
11475       compatibility measure and should not be used for new code.
11476     * -Warray-bounds has been added and is now enabled by default for
11477       -Wall . It produces warnings for array subscripts that can be
11478       determined at compile time to be always out of bounds.
11479       -Wno-array-bounds will disable the warning.
11480     * The constructor and destructor function attributes now accept
11481       optional priority arguments which control the order in which the
11482       constructor and destructor functions are run.
11483     * New [8]command-line options -Wtype-limits, -Wold-style-declaration,
11484       -Wmissing-parameter-type, -Wempty-body, -Wclobbered and
11485       -Wignored-qualifiers have been added for finer control of the
11486       diverse warnings enabled by -Wextra.
11487     * A new function attribute alloc_size has been added to mark up
11488       malloc style functions. For constant sized allocations this can be
11489       used to find out the size of the returned pointer using the
11490       __builtin_object_size() function for buffer overflow checking and
11491       similar. This supplements the already built-in malloc and calloc
11492       constant size handling.
11493     * Integer constants written in binary are now supported as a GCC
11494       extension. They consist of a prefix 0b or 0B, followed by a
11495       sequence of 0 and 1 digits.
11496     * A new predefined macro __COUNTER__ has been added. It expands to
11497       sequential integral values starting from 0. In conjunction with the
11498       ## operator, this provides a convenient means to generate unique
11499       identifiers.
11500     * A new command-line option -fdirectives-only has been added. It
11501       enables a special preprocessing mode which improves the performance
11502       of applications like distcc and ccache.
11503     * Fixed-point data types and operators have been added. They are
11504       based on Chapter 4 of the Embedded-C specification (n1169.pdf).
11505       Currently, only MIPS targets are supported.
11506     * Decimal floating-point arithmetic based on draft ISO/IEC TR 24732,
11507       N1241, is now supported as a GCC extension to C for targets
11508       i[34567]86-*-linux-gnu, powerpc*-*-linux-gnu, s390*-ibm-linux-gnu,
11509       and x86_64-*-linux-gnu. The feature introduces new data types
11510       _Decimal32, _Decimal64, and _Decimal128 with constant suffixes DF,
11511       DD, and DL.
11512
11513  C++
11514
11515     * [9]Experimental support for the upcoming ISO C++ standard, C++0x.
11516     * -Wc++0x-compat has been added and is now enabled by default for
11517       -Wall. It produces warnings for constructs whose meaning differs
11518       between ISO C++ 1998 and C++0x.
11519     * The -Wparentheses option now works for C++ as it does for C. It
11520       warns if parentheses are omitted when operators with confusing
11521       precedence are nested. It also warns about ambiguous else
11522       statements. Since -Wparentheses is enabled by -Wall, this may cause
11523       additional warnings with existing C++ code which uses -Wall. These
11524       new warnings may be disabled by using -Wall -Wno-parentheses.
11525     * The -Wmissing-declarations now works for C++ as it does for C.
11526     * The -fvisibility-ms-compat flag was added, to make it easier to
11527       port larger projects using shared libraries from Microsoft's Visual
11528       Studio to ELF and Mach-O systems.
11529     * C++ attribute handling has been overhauled for template arguments
11530       (ie dependent types). In particular, __attribute__((aligned(T)));
11531       works for C++ types.
11532
11533    Runtime Library (libstdc++)
11534
11535     * [10]Experimental support for the upcoming ISO C++ standard, C++0x.
11536     * Support for TR1 mathematical special functions and regular
11537       expressions.
11538     * Default what implementations give more elaborate exception strings
11539       for bad_cast, bad_typeid, bad_exception, and bad_alloc.
11540     * Header dependencies have been streamlined, reducing unnecessary
11541       includes and pre-processed bloat.
11542     * Variadic template implementations of items in <tuple> and
11543       <functional>.
11544     * An experimental [11]parallel mode has been added. This is a
11545       parallel implementation of many C++ Standard library algorithms,
11546       like std::accumulate, std::for_each, std::transform, or std::sort,
11547       to give but four examples. These algorithms can be substituted for
11548       the normal (sequential) libstdc++ algorithms on a piecemeal basis,
11549       or all existing algorithms can be transformed via the
11550       -D_GLIBCXX_PARALLEL macro.
11551     * Debug mode versions of classes in <unordered_set> and
11552       <unordered_map>.
11553     * Formal deprecation of <ext/hash_set> and <ext/hash_map>, which are
11554       now <backward/hash_set> and <backward/hash_map>. This code:
11555    #include <ext/hash_set>
11556    __gnu_cxx::hash_set<int> s;
11557
11558       Can be transformed (in order of preference) to:
11559    #include <tr1/unordered_set>
11560    std::tr1::unordered_set<int> s;
11561
11562       or
11563    #include <backward/hash_set>
11564    __gnu_cxx::hash_set<int> s;
11565
11566       Similar transformations apply to __gnu_cxx::hash_map,
11567       __gnu_cxx::hash_multimap, __gnu_cxx::hash_set,
11568       __gnu_cxx::hash_multiset.
11569
11570  Fortran
11571
11572     * Due to the fact that the GMP and MPFR libraries are required for
11573       all languages, Fortran is no longer special in this regard and is
11574       available by default.
11575     * The [12]-fexternal-blas option has been added, which generates
11576       calls to BLAS routines for intrinsic matrix operations such as
11577       matmul rather than using the built-in algorithms.
11578     * Support to give a backtrace (compiler flag -fbacktrace or
11579       environment variable GFORTRAN_ERROR_BACKTRACE; on glibc systems
11580       only) or a core dump (-fdump-core, GFORTRAN_ERROR_DUMPCORE) when a
11581       run-time error occured.
11582     * GNU Fortran now defines __GFORTRAN__ when it runs the C
11583       preprocessor (CPP).
11584     * The [13]-finit-local-zero, -finit-real, -finit-integer,
11585       -finit-character, and -finit-logical options have been added, which
11586       can be used to initialize local variables.
11587     * The intrinsic procedures [14]GAMMA and [15]LGAMMA have been added,
11588       which calculate the Gamma function and its logarithm. Use EXTERNAL
11589       gamma if you want to use your own gamma function.
11590     * GNU Fortran now regards the backslash character as literal (as
11591       required by the Fortran 2003 standard); using [16]-fbackslash GNU
11592       Fortran interprets backslashes as C-style escape characters.
11593     * The [17]interpretation of binary, octal and hexadecimal (BOZ)
11594       literal constants has been changed. Before they were always
11595       interpreted as integer; now they are bit-wise transferred as
11596       argument of INT, REAL, DBLE and CMPLX as required by the Fortran
11597       2003 standard, and for real and complex variables in DATA
11598       statements or when directly assigned to real and complex variables.
11599       Everywhere else and especially in expressions they are still
11600       regarded as integer constants.
11601     * Fortran 2003 support has been extended:
11602          + Intrinsic statements IMPORT, PROTECTED, VALUE and VOLATILE
11603          + Pointer intent
11604          + Intrinsic module ISO_ENV_FORTRAN
11605          + Interoperability with C (ISO C Bindings)
11606          + ABSTRACT INTERFACES and PROCEDURE statements (without POINTER
11607            attribute)
11608          + Fortran 2003 BOZ
11609
11610  Java (GCJ)
11611
11612     * GCJ now uses the Eclipse Java compiler for its Java parsing needs.
11613       This enables the use of all 1.5 language features, and fixes most
11614       existing front end bugs.
11615     * libgcj now supports all 1.5 language features which require runtime
11616       support: foreach, enum, annotations, generics, and auto-boxing.
11617     * We've made many changes to the tools shipped with gcj.
11618          + The old jv-scan tool has been removed. This tool never really
11619            worked properly. There is no replacement.
11620          + gcjh has been rewritten. Some of its more obscure options no
11621            longer work, but are still recognized in an attempt at
11622            compatibility. gjavah is a new program with similar
11623            functionality but different command-line options.
11624          + grmic and grmiregistry have been rewritten. grmid has been
11625            added.
11626          + gjar replaces the old fastjar.
11627          + gjarsigner (used for signing jars), gkeytool (used for key
11628            management), gorbd (for CORBA), gserialver (computes
11629            serialization UIDs), and gtnameserv (also for CORBA) are now
11630            installed.
11631     * The ability to dump the contents of the java run time heap to a
11632       file for off-line analysis has been added. The heap dumps may be
11633       analyzed with the new gc-analyze tool. They may be generated on
11634       out-of-memory conditions or on demand and are controlled by the new
11635       run time class gnu.gcj.util.GCInfo.
11636     * java.util.TimeZone can now read files from /usr/share/zoneinfo to
11637       provide correct, updated, timezone information. This means that
11638       packagers no longer have to update libgcj when a time zone change
11639       is published.
11640
11641New Targets and Target Specific Improvements
11642
11643  IA-32/x86-64
11644
11645     * Tuning for Intel Core 2 processors is available via -mtune=core2
11646       and -march=core2.
11647     * Tuning for AMD Geode processors is available via -mtune=geode and
11648       -march=geode.
11649     * Code generation of block move (memcpy) and block set (memset) was
11650       rewritten. GCC can now pick the best algorithm (loop, unrolled
11651       loop, instruction with rep prefix or a library call) based on the
11652       size of the block being copied and the CPU being optimized for. A
11653       new option -minline-stringops-dynamically has been added. With this
11654       option string operations of unknown size are expanded such that
11655       small blocks are copied by in-line code, while for large blocks a
11656       library call is used. This results in faster code than
11657       -minline-all-stringops when the library implementation is capable
11658       of using cache hierarchy hints. The heuristic choosing the
11659       particular algorithm can be overwritten via -mstringop-strategy.
11660       Newly also memset of values different from 0 is inlined.
11661     * GCC no longer places the cld instruction before string operations.
11662       Both i386 and x86-64 ABI documents mandate the direction flag to be
11663       clear at the entry of a function. It is now invalid to set the flag
11664       in asm statement without reseting it afterward.
11665     * Support for SSSE3 built-in functions and code generation are
11666       available via -mssse3.
11667     * Support for SSE4.1 built-in functions and code generation are
11668       available via -msse4.1.
11669     * Support for SSE4.2 built-in functions and code generation are
11670       available via -msse4.2.
11671     * Both SSE4.1 and SSE4.2 support can be enabled via -msse4.
11672     * A new set of options -mpc32, -mpc64 and -mpc80 have been added to
11673       allow explicit control of x87 floating point precision.
11674     * Support for __float128 (TFmode) IEEE quad type and corresponding
11675       TCmode IEEE complex quad type is available via the soft-fp library
11676       on x86_64 targets. This includes basic arithmetic operations
11677       (addition, subtraction, negation, multiplication and division) on
11678       __float128 real and TCmode complex values, the full set of IEEE
11679       comparisons between __float128 values, conversions to and from
11680       float, double and long double floating point types, as well as
11681       conversions to and from signed or unsigned integer, signed or
11682       unsigned long integer and signed or unsigned quad (TImode) integer
11683       types. Additionally, all operations generate the full set of IEEE
11684       exceptions and support the full set of IEEE rounding modes.
11685     * GCC can now utilize the ACML library for vectorizing calls to a set
11686       of C99 functions on x86_64 if -mveclibabi=acml is specified and you
11687       link to an ACML ABI compatible library.
11688
11689  ARM
11690
11691     * Compiler and Library support for Thumb-2 and the ARMv7 architecture
11692       has been added.
11693
11694  CRIS
11695
11696    New features
11697
11698     * Compiler and Library support for the CRIS v32 architecture, as
11699       found in Axis Communications ETRAX FS and ARTPEC-3 chips, has been
11700       added.
11701
11702    Configuration changes
11703
11704     * The cris-*-elf target now includes support for CRIS v32, including
11705       libraries, through the -march=v32 option.
11706     * A new crisv32-*-elf target defaults to generate code for CRIS v32.
11707     * A new crisv32-*-linux* target defaults to generate code for CRIS
11708       v32.
11709     * The cris-*-aout target has been obsoleted.
11710
11711    Improved support for built-in functions
11712
11713     * GCC can now use the lz and swapwbr instructions to implement the
11714       __builtin_clz, __builtin_ctz and __builtin_ffs family of functions.
11715     * __builtin_bswap32 is now implemented using the swapwb instruction,
11716       when available.
11717
11718  m68k and ColdFire
11719
11720    New features
11721
11722     * Support for several new ColdFire processors has been added. You can
11723       generate code for them using the new -mcpu option.
11724     * All targets now support ColdFire processors.
11725     * m68k-uclinux targets have improved support for C++ constructors and
11726       destructors, and for shared libraries.
11727     * It is now possible to set breakpoints on the first or last line of
11728       a function, even if there are no statements on that line.
11729
11730    Optimizations
11731
11732     * Support for sibling calls has been added.
11733     * More use is now made of the ColdFire mov3q instruction.
11734     * __builtin_clz is now implemented using the ff1 ColdFire
11735       instruction, when available.
11736     * GCC now honors the -m68010 option. 68010 code now uses clr rather
11737       than move to zero volatile memory.
11738     * 68020 targets and above can now use symbol(index.size*scale)
11739       addresses for indexed array accesses. Earlier compilers would
11740       always load the symbol into a base register first.
11741
11742    Configuration changes
11743
11744     * All m68k and ColdFire targets now allow the default processor to be
11745       set at configure time using --with-cpu.
11746     * A --with-arch configuration option has been added. This option
11747       allows you to restrict a target to ColdFire or non-ColdFire
11748       processors.
11749
11750    Preprocessor macros
11751
11752     * An __mcfv*__ macro is now defined for all ColdFire targets.
11753       (Earlier versions of GCC only defined __mcfv4e__.)
11754     * __mcf_cpu_*, __mcf_family_* and __mcffpu__ macros have been added.
11755     * All targets now define __mc68010 and __mc68010__ when generating
11756       68010 code.
11757
11758    Command-line changes
11759
11760     * New command-line options -march, -mcpu, -mtune and -mhard-float
11761       have been added. These options apply to both m68k and ColdFire
11762       targets.
11763     * -mno-short, -mno-bitfield and -mno-rtd are now accepted as negative
11764       versions of -mshort, etc.
11765     * -fforce-addr has been removed. It is now ignored by the compiler.
11766
11767    Other improvements
11768
11769     * ColdFire targets now try to maintain a 4-byte-aligned stack where
11770       possible.
11771     * m68k-uclinux targets now try to avoid situations that lead to the
11772       load-time error: BINFMT_FLAT: reloc outside program.
11773
11774  MIPS
11775
11776    Changes to existing configurations
11777
11778     * libffi and libjava now support all three GNU/Linux ABIs: o32, n32
11779       and n64. Every GNU/Linux configuration now builds these libraries
11780       by default.
11781     * GNU/Linux configurations now generate -mno-shared code unless
11782       overridden by -fpic, -fPIC, -fpie or -fPIE.
11783     * mipsisa32*-linux-gnu configurations now generate hard-float code by
11784       default, just like other mipsisa32* and mips*-linux-gnu
11785       configurations. You can build a soft-float version of any
11786       mips*-linux-gnu configuration by passing --with-float=soft to
11787       configure.
11788     * mips-wrs-vxworks now supports run-time processes (RTPs).
11789
11790    Changes to existing command-line options
11791
11792     * The -march and -mtune options no longer accept 24k as a processor
11793       name. Please use 24kc, 24kf2_1 or 24kf1_1 instead.
11794     * The -march and -mtune options now accept 24kf2_1, 24kef2_1 and
11795       34kf2_1 as synonyms for 24kf, 24kef and 34kf respectively. The
11796       options also accept 24kf1_1, 24kef1_1 and 34kf1_1 as synonyms for
11797       24kx, 24kex and 34kx.
11798
11799    New configurations
11800
11801   GCC now supports the following configurations:
11802     * mipsisa32r2*-linux-gnu*, which generates MIPS32 revision 2 code by
11803       default. Earlier releases also recognized this configuration, but
11804       they treated it in the same way as mipsisa32*-linux-gnu*. Note that
11805       you can customize any mips*-linux-gnu* configuration to a
11806       particular ISA or processor by passing an appropriate --with-arch
11807       option to configure.
11808     * mipsisa*-sde-elf*, which provides compatibility with MIPS
11809       Technologies' SDE toolchains. The configuration uses the SDE
11810       libraries by default, but you can use it like other newlib-based
11811       ELF configurations by passing --with-newlib to configure. It is the
11812       only configuration besides mips64vr*-elf* to build MIPS16 as well
11813       as non-MIPS16 libraries.
11814     * mipsisa*-elfoabi*, which is similar to the general mipsisa*-elf*
11815       configuration, but uses the o32 and o64 ABIs instead of the 32-bit
11816       and 64-bit forms of the EABI.
11817
11818    New processors and application-specific extensions
11819
11820     * Support for the SmartMIPS ASE is available through the new
11821       -msmartmips option.
11822     * Support for revision 2 of the DSP ASE is available through the new
11823       -mdspr2 option. A new preprocessor macro called __mips_dsp_rev
11824       indicates the revision of the ASE in use.
11825     * Support for the 4KS and 74K families of processors is available
11826       through the -march and -mtune options.
11827
11828    Improved support for built-in functions
11829
11830     * GCC can now use load-linked, store-conditional and sync
11831       instructions to implement atomic built-in functions such as
11832       __sync_fetch_and_add. The memory reference must be 4 bytes wide for
11833       32-bit targets and either 4 or 8 bytes wide for 64-bit targets.
11834     * GCC can now use the clz and dclz instructions to implement the
11835       __builtin_ctz and __builtin_ffs families of functions.
11836     * There is a new __builtin___clear_cache function for flushing the
11837       instruction cache. GCC expands this function inline on MIPS32
11838       revision 2 targets, otherwise it calls the function specified by
11839       -mcache-flush-func.
11840
11841    MIPS16 improvements
11842
11843     * GCC can now compile objects that contain a mixture of MIPS16 and
11844       non-MIPS16 code. There are two new attributes, mips16 and nomips16,
11845       for specifying which mode a function should use.
11846     * A new option called -minterlink-mips16 makes non-MIPS16 code
11847       link-compatible with MIPS16 code.
11848     * After many bug fixes, the long-standing MIPS16 -mhard-float support
11849       should now work fairly reliably.
11850     * GCC can now use the MIPS16e save and restore instructions.
11851     * -fsection-anchors now works in MIPS16 mode. MIPS16 code compiled
11852       with -G0 -fsection-anchors is often smaller than code compiled with
11853       -G8. However, please note that you must usually compile all objects
11854       in your application with the same -G option; see the documentation
11855       of -G for details.
11856     * A new option called-mcode-readable specifies which instructions are
11857       allowed to load from the code segment. -mcode-readable=yes is the
11858       default and says that any instruction may load from the code
11859       segment. The other alternatives are -mcode-readable=pcrel, which
11860       says that only PC-relative MIPS16 instructions may load from the
11861       code segment, and -mcode-readable=no, which says that no
11862       instruction may do so. Please see the documentation for more
11863       details, including example uses.
11864
11865    Small-data improvements
11866
11867   There are three new options for controlling small data:
11868     * -mno-extern-sdata, which disables small-data accesses for
11869       externally-defined variables. Code compiled with -Gn
11870       -mno-extern-sdata will be link-compatible with any -G setting
11871       between -G0 and -Gn inclusive.
11872     * -mno-local-sdata, which disables the use of small-data sections for
11873       data that is not externally visible. This option can be a useful
11874       way of reducing small-data usage in less performance-critical parts
11875       of an application.
11876     * -mno-gpopt, which disables the use of the $gp register while still
11877       honoring the -G limit when placing externally-visible data. This
11878       option implies -mno-extern-sdata and -mno-local-sdata and it can be
11879       useful in situations where $gp does not necessarily hold the
11880       expected value.
11881
11882    Miscellaneous improvements
11883
11884     * There is a new option called -mbranch-cost for tweaking the
11885       perceived cost of branches.
11886     * If GCC is configured to use a version of GAS that supports the
11887       .gnu_attribute directive, it will use that directive to record
11888       certain properties of the output code. .gnu_attribute is new to GAS
11889       2.18.
11890     * There are two new function attributes, near and far, for overriding
11891       the command-line setting of -mlong-calls on a function-by-function
11892       basis.
11893     * -mfp64, which previously required a 64-bit target, now works with
11894       MIPS32 revision 2 targets as well. The mipsisa*-elfoabi* and
11895       mipsisa*-sde-elf* configurations provide suitable library support.
11896     * GCC now recognizes the -mdmx and -mmt options and passes them down
11897       to the assembler. It does nothing else with the options at present.
11898
11899  SPU (Synergistic Processor Unit) of the Cell Broadband Engine Architecture
11900  (BEA)
11901
11902     * Support has been added for this new architecture.
11903
11904  RS6000 (POWER/PowerPC)
11905
11906     * Support for the PowerPC 750CL paired-single instructions has been
11907       added with a new powerpc-*-linux*paired* target configuration. It
11908       is enabled by an associated -mpaired option and can be accessed
11909       using new built-in functions.
11910     * Support for auto-detecting architecture and system configuration to
11911       auto-select processor optimization tuning.
11912     * Support for VMX on AIX 5.3 has been added.
11913     * Support for AIX Version 6.1 has been added.
11914
11915  S/390, zSeries and System z9
11916
11917     * Support for the IBM System z9 EC/BC processor (z9 GA3) has been
11918       added. When using the -march=z9-ec option, the compiler will
11919       generate code making use of instructions provided by the decimal
11920       floating point facility and the floating point conversion facility
11921       (pfpo). Besides the instructions used to implement decimal floating
11922       point operations these facilities also contain instructions to move
11923       between general purpose and floating point registers and to modify
11924       and copy the sign-bit of floating point values.
11925     * When the -march=z9-ec option is used the new
11926       -mhard-dfp/-mno-hard-dfp options can be used to specify whether the
11927       decimal floating point hardware instructions will be used or not.
11928       If none of them is given the hardware support is enabled by
11929       default.
11930     * The -mstack-guard option can now be omitted when using stack
11931       checking via -mstack-size in order to let GCC choose a sensible
11932       stack guard value according to the frame size of each function.
11933     * Various changes to improve performance of generated code have been
11934       implemented, including:
11935          + The condition code set by an add logical with carry
11936            instruction is now available for overflow checks like: a + b +
11937            carry < b.
11938          + The test data class instruction is now used to implement
11939            sign-bit and infinity checks of binary and decimal floating
11940            point numbers.
11941
11942  SPARC
11943
11944     * Support for the Sun UltraSPARC T2 (Niagara 2) processor has been
11945       added.
11946
11947  Xtensa
11948
11949     * Stack unwinding for exception handling now uses by default a
11950       specialized version of DWARF unwinding. This is not
11951       binary-compatible with the setjmp/longjmp (sjlj) unwinding used for
11952       Xtensa with previous versions of GCC.
11953     * For Xtensa processors that include the Conditional Store option,
11954       the built-in functions for atomic memory access are now implemented
11955       using S32C1I instructions.
11956     * If the Xtensa NSA option is available, GCC will use it to implement
11957       the __builtin_ctz and __builtin_clz functions.
11958
11959Documentation improvements
11960
11961     * Existing libstdc++ documentation has been edited and restructured
11962       into a single DocBook XML manual. The results can be viewed online
11963       [18]here.
11964
11965Other significant improvements
11966
11967     * The compiler's --help command-line option has been extended so that
11968       it now takes an optional set of arguments. These arguments restrict
11969       the information displayed to specific classes of command-line
11970       options, and possibly only a subset of those options. It is also
11971       now possible to replace the descriptive text associated with each
11972       displayed option with an indication of its current value, or for
11973       binary options, whether it has been enabled or disabled.
11974       Here are some examples. The following will display all the options
11975       controlling warning messages:
11976      --help=warnings
11977
11978       Whereas this will display all the undocumented, target specific
11979       options:
11980      --help=target,undocumented
11981
11982       This sequence of commands will display the binary optimizations
11983       that are enabled by -O3:
11984      gcc -c -Q -O3 --help=optimizers > /tmp/O3-opts
11985      gcc -c -Q -O2 --help=optimizers > /tmp/O2-opts
11986      diff /tmp/O2-opts /tmp/O3-opts | grep enabled
11987
11988     * The configure options --with-pkgversion and --with-bugurl have been
11989       added. These allow distributors of GCC to include a
11990       distributor-specific string in manuals and --version output and to
11991       specify the URL for reporting bugs in their versions of GCC.
11992
11993GCC 4.3.1
11994
11995   This is the [19]list of problem reports (PRs) from GCC's bug tracking
11996   system that are known to be fixed in the 4.3.1 release. This list might
11997   not be complete (that is, it is possible that some PRs that have been
11998   fixed are not listed here).
11999
12000Target Specific Changes
12001
12002  IA-32/x86-64
12003
12004    ABI changes
12005
12006     * Starting with GCC 4.3.1, decimal floating point variables are
12007       aligned to their natural boundaries when they are passed on the
12008       stack for i386.
12009
12010    Command-line changes
12011
12012     * Starting with GCC 4.3.1, the -mcld option has been added to
12013       automatically generate a cld instruction in the prologue of
12014       functions that use string instructions. This option is used for
12015       backward compatibility on some operating systems and can be enabled
12016       by default for 32-bit x86 targets by configuring GCC with the
12017       --enable-cld configure option.
12018
12019GCC 4.3.2
12020
12021   This is the [20]list of problem reports (PRs) from GCC's bug tracking
12022   system that are known to be fixed in the 4.3.2 release. This list might
12023   not be complete (that is, it is possible that some PRs that have been
12024   fixed are not listed here).
12025
12026GCC 4.3.3
12027
12028   This is the [21]list of problem reports (PRs) from GCC's bug tracking
12029   system that are known to be fixed in the 4.3.3 release. This list might
12030   not be complete (that is, it is possible that some PRs that have been
12031   fixed are not listed here).
12032
12033GCC 4.3.4
12034
12035   This is the [22]list of problem reports (PRs) from GCC's bug tracking
12036   system that are known to be fixed in the 4.3.4 release. This list might
12037   not be complete (that is, it is possible that some PRs that have been
12038   fixed are not listed here).
12039
12040GCC 4.3.5
12041
12042   This is the [23]list of problem reports (PRs) from GCC's bug tracking
12043   system that are known to be fixed in the 4.3.5 release. This list might
12044   not be complete (that is, it is possible that some PRs that have been
12045   fixed are not listed here).
12046
12047GCC 4.3.6
12048
12049   This is the [24]list of problem reports (PRs) from GCC's bug tracking
12050   system that are known to be fixed in the 4.3.6 release. This list might
12051   not be complete (that is, it is possible that some PRs that have been
12052   fixed are not listed here).
12053
12054
12055    For questions related to the use of GCC, please consult these web
12056    pages and the [25]GCC manuals. If that fails, the
12057    [26]gcc-help@gcc.gnu.org mailing list might help. Comments on these
12058    web pages and the development of GCC are welcome on our developer
12059    list at [27]gcc@gcc.gnu.org. All of [28]our lists have public
12060    archives.
12061
12062   Copyright (C) [29]Free Software Foundation, Inc. Verbatim copying and
12063   distribution of this entire article is permitted in any medium,
12064   provided this notice is preserved.
12065
12066   These pages are [30]maintained by the GCC team. Last modified
12067   2019-11-28[31].
12068
12069References
12070
12071   1. http://gcc.gnu.org/gcc-4.3/changes.html#4.3.5
12072   2. https://gmplib.org/
12073   3. https://www.mpfr.org/
12074   4. https://gcc.gnu.org/install/prerequisites.html
12075   5. https://gcc.gnu.org/ml/gcc-announce/2001/msg00000.html
12076   6. https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html#Warning-Options
12077   7. http://gcc.gnu.org/gcc-4.3/porting_to.html
12078   8. https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html
12079   9. http://gcc.gnu.org/gcc-4.3/cxx0x_status.html
12080  10. http://gcc.gnu.org/gcc-4.3/cxx0x_status.html
12081  11. https://gcc.gnu.org/onlinedocs/libstdc++/manual/parallel_mode.html
12082  12. https://gcc.gnu.org/onlinedocs/gfortran/Code-Gen-Options.html#Code-Gen-Options
12083  13. https://gcc.gnu.org/onlinedocs/gfortran/Code-Gen-Options.html#index-g_t_0040code_007bfinit-local-zero_007d-167
12084  14. https://gcc.gnu.org/onlinedocs/gcc-4.3.0/gfortran/GAMMA.html
12085  15. https://gcc.gnu.org/onlinedocs/gcc-4.3.0/gfortran/LGAMMA.html
12086  16. https://gcc.gnu.org/onlinedocs/gfortran/Fortran-Dialect-Options.html
12087  17. https://gcc.gnu.org/onlinedocs/gfortran/BOZ-literal-constants.html
12088  18. https://gcc.gnu.org/onlinedocs/libstdc++/
12089  19. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.3.1
12090  20. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.3.2
12091  21. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.3.3
12092  22. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.3.4
12093  23. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.3.5
12094  24. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.3.6
12095  25. https://gcc.gnu.org/onlinedocs/
12096  26. mailto:gcc-help@gcc.gnu.org
12097  27. mailto:gcc@gcc.gnu.org
12098  28. https://gcc.gnu.org/lists.html
12099  29. https://www.fsf.org/
12100  30. https://gcc.gnu.org/about.html
12101  31. http://validator.w3.org/check/referer
12102======================================================================
12103http://gcc.gnu.org/gcc-4.2/index.html
12104
12105                             GCC 4.2 Release Series
12106
12107   (This release series is no longer supported.)
12108
12109   May 19, 2008
12110
12111   The [1]GNU project and the GCC developers are pleased to announce the
12112   release of GCC 4.2.4.
12113
12114   This release is a bug-fix release, containing fixes for regressions in
12115   GCC 4.2.3 relative to previous releases of GCC.
12116
12117Release History
12118
12119   GCC 4.2.4
12120          May 19, 2008 ([2]changes)
12121
12122   GCC 4.2.3
12123          February 1, 2008 ([3]changes)
12124
12125   GCC 4.2.2
12126          October 7, 2007 ([4]changes)
12127
12128   GCC 4.2.1
12129          July 18, 2007 ([5]changes)
12130
12131   GCC 4.2.0
12132          May 13, 2007 ([6]changes)
12133
12134References and Acknowledgements
12135
12136   GCC used to stand for the GNU C Compiler, but since the compiler
12137   supports several other languages aside from C, it now stands for the
12138   GNU Compiler Collection.
12139
12140   A list of [7]successful builds is updated as new information becomes
12141   available.
12142
12143   The GCC developers would like to thank the numerous people that have
12144   contributed new features, improvements, bug fixes, and other changes as
12145   well as test results to GCC. This [8]amazing group of volunteers is
12146   what makes GCC successful.
12147
12148   For additional information about GCC please refer to the [9]GCC project
12149   web site or contact the [10]GCC development mailing list.
12150
12151   To obtain GCC please use [11]our mirror sites or [12]our version
12152   control system.
12153
12154
12155    For questions related to the use of GCC, please consult these web
12156    pages and the [13]GCC manuals. If that fails, the
12157    [14]gcc-help@gcc.gnu.org mailing list might help. Comments on these
12158    web pages and the development of GCC are welcome on our developer
12159    list at [15]gcc@gcc.gnu.org. All of [16]our lists have public
12160    archives.
12161
12162   Copyright (C) [17]Free Software Foundation, Inc. Verbatim copying and
12163   distribution of this entire article is permitted in any medium,
12164   provided this notice is preserved.
12165
12166   These pages are [18]maintained by the GCC team. Last modified
12167   2020-01-14[19].
12168
12169References
12170
12171   1. http://www.gnu.org/
12172   2. http://gcc.gnu.org/gcc-4.2/changes.html
12173   3. http://gcc.gnu.org/gcc-4.2/changes.html
12174   4. http://gcc.gnu.org/gcc-4.2/changes.html
12175   5. http://gcc.gnu.org/gcc-4.2/changes.html
12176   6. http://gcc.gnu.org/gcc-4.2/changes.html
12177   7. http://gcc.gnu.org/gcc-4.2/buildstat.html
12178   8. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
12179   9. http://gcc.gnu.org/index.html
12180  10. mailto:gcc@gcc.gnu.org
12181  11. http://gcc.gnu.org/mirrors.html
12182  12. http://gcc.gnu.org/git.html
12183  13. https://gcc.gnu.org/onlinedocs/
12184  14. mailto:gcc-help@gcc.gnu.org
12185  15. mailto:gcc@gcc.gnu.org
12186  16. https://gcc.gnu.org/lists.html
12187  17. https://www.fsf.org/
12188  18. https://gcc.gnu.org/about.html
12189  19. http://validator.w3.org/check/referer
12190======================================================================
12191http://gcc.gnu.org/gcc-4.2/changes.html
12192
12193                             GCC 4.2 Release Series
12194                        Changes, New Features, and Fixes
12195
12196Caveats
12197
12198     * GCC no longer accepts the -fshared-data option. This option has had
12199       no effect in any GCC 4 release; the targets to which the option
12200       used to apply had been removed before GCC 4.0.
12201
12202General Optimizer Improvements
12203
12204     * New command-line options specify the possible relationships among
12205       parameters and between parameters and global data. For example,
12206       -fargument-noalias-anything specifies that arguments do not alias
12207       any other storage.
12208       Each language will automatically use whatever option is required by
12209       the language standard. You should not need to use these options
12210       yourself.
12211
12212New Languages and Language specific improvements
12213
12214     * [1]OpenMP is now supported for the C, C++ and Fortran compilers.
12215     * New command-line options -fstrict-overflow and -Wstrict-overflow
12216       have been added. -fstrict-overflow tells the compiler that it may
12217       assume that the program follows the strict signed overflow
12218       semantics permitted for the language: for C and C++ this means that
12219       the compiler may assume that signed overflow does not occur. For
12220       example, a loop like
12221      for (i = 1; i > 0; i *= 2)
12222
12223       is presumably intended to continue looping until i overflows. With
12224       -fstrict-overflow, the compiler may assume that signed overflow
12225       will not occur, and transform this into an infinite loop.
12226       -fstrict-overflow is turned on by default at -O2, and may be
12227       disabled via -fno-strict-overflow. The -Wstrict-overflow option may
12228       be used to warn about cases where the compiler assumes that signed
12229       overflow will not occur. It takes five different levels:
12230       -Wstrict-overflow=1 to 5. See the [2]documentation for details.
12231       -Wstrict-overflow=1 is enabled by -Wall.
12232     * The new command-line option -fno-toplevel-reorder directs GCC to
12233       emit top-level functions, variables, and asm statements in the same
12234       order that they appear in the input file. This is intended to
12235       support existing code which relies on a particular ordering (for
12236       example, code which uses top-level asm statements to switch
12237       sections). For new code, it is generally better to use function and
12238       variable attributes. The -fno-toplevel-reorder option may be used
12239       for most cases which currently use -fno-unit-at-a-time. The
12240       -fno-unit-at-a-time option will be removed in some future version
12241       of GCC. If you know of a case which requires -fno-unit-at-a-time
12242       which is not fixed by -fno-toplevel-reorder, please open a bug
12243       report.
12244
12245  C family
12246
12247     * The pragma redefine_extname will now macro expand its tokens for
12248       compatibility with SunPRO.
12249     * In the next release of GCC, 4.3, -std=c99 or -std=gnu99 will direct
12250       GCC to handle inline functions as specified in the C99 standard. In
12251       preparation for this, GCC 4.2 will warn about any use of non-static
12252       inline functions in gnu99 or c99 mode. This new warning may be
12253       disabled with the new gnu_inline function attribute or the new
12254       -fgnu89-inline command-line option. Also, GCC 4.2 and later will
12255       define one of the preprocessor macros __GNUC_GNU_INLINE__ or
12256       __GNUC_STDC_INLINE__ to indicate the semantics of inline functions
12257       in the current compilation.
12258     * A new command-line option -Waddress has been added to warn about
12259       suspicious uses of memory addresses as, for example, using the
12260       address of a function in a conditional expression, and comparisons
12261       against the memory address of a string literal. This warning is
12262       enabled by -Wall.
12263
12264  C++
12265
12266     * C++ visibility handling has been overhauled.
12267       Restricted visiblity is propagated from classes to members, from
12268       functions to local statics, and from templates and template
12269       arguments to instantiations, unless the latter has explicitly
12270       declared visibility.
12271       The visibility attribute for a class must come between the
12272       class-key and the name, not after the closing brace.
12273       Attributes are now allowed for enums and elaborated-type-specifiers
12274       that only declare a type.
12275       Members of the anonymous namespace are now local to a particular
12276       translation unit, along with any other declarations which use them,
12277       though they are still treated as having external linkage for
12278       language semantics.
12279     * The (undocumented) extension which permitted templates with default
12280       arguments to be bound to template template parameters with fewer
12281       parameters has been removed. For example:
12282        template <template <typename> class C>
12283        void f(C<double>) {}
12284
12285        template <typename T, typename U = int>
12286        struct S {};
12287
12288        template void f(S<double>);
12289
12290       is no longer accepted by G++. The reason this code is not accepted
12291       is that S is a template with two parameters; therefore, it cannot
12292       be bound to C which has only one parameter.
12293     * The <?, >?, <?=, and >?= operators, deprecated in previous GCC
12294       releases, have been removed.
12295     * The command-line option -fconst-strings, deprecated in previous GCC
12296       releases, has been removed.
12297     * The configure variable enable-__cxa_atexit is now enabled by
12298       default for more targets. Enabling this variable is necessary in
12299       order for static destructors to be executed in the correct order,
12300       but it depends upon the presence of a non-standard C library in the
12301       target library in order to work. The variable is now enabled for
12302       more targets which are known to have suitable C libraries.
12303     * -Wextra will produce warnings for if statements with a semicolon as
12304       the only body, to catch code like:
12305         if (a);
12306            return 1;
12307         return 0;
12308
12309       To suppress the warning in valid cases, use { } instead.
12310     * The C++ front end now also produces strict aliasing warnings when
12311       -fstrict-aliasing -Wstrict-aliasing is in effect.
12312
12313    Runtime Library (libstdc++)
12314
12315     * Added support for TR1 <random>, <complex>, and C compatibility
12316       headers. In addition, a lock-free version of shared_ptr was
12317       contributed as part of Phillip Jordan's Google Summer of Code
12318       project on lock-free containers.
12319     * In association with the Summer of Code work on lock-free
12320       containers, the interface for atomic builtins was adjusted,
12321       creating simpler alternatives for non-threaded code paths. Also,
12322       usage was consolidated and all elements were moved from namespace
12323       std to namespace__gnu_cxx. Affected interfaces are the functions
12324       __exchange_and_add, __atomic_add, and the objects __mutex,
12325       __recursive_mutex, and __scoped_lock.
12326     * Support for versioning weak symbol names via namespace association
12327       was added. However, as this changes the names of exported symbols,
12328       this is turned off by default in the current ABI. Intrepid users
12329       can enable this feature by using
12330       --enable-symvers=gnu-versioned-namespace during configuration.
12331     * Revised, simplified, and expanded policy-based associative
12332       containers, including data types for tree and trie forms
12333       (basic_tree, tree, trie), lists (list_update), and both
12334       collision-chaining and probing hash-based containers
12335       (basic_hash_table, cc_hash_table, gp_hash_table). More details per
12336       the [3]documentation.
12337     * The implementation of the debug mode was modified, whereby the
12338       debug namespaces were nested inside of namespace std and namespace
12339       __gnu_cxx in order to resolve some long standing corner cases
12340       involving name lookup. Debug functionality from the policy-based
12341       data structures was consolidated and enabled with the single macro,
12342       _GLIBCXX_DEBUG. See PR 26142 for more information.
12343     * Added extensions for type traits: __conditional_type,
12344       __numeric_traits, __add_unsigned, __removed_unsigned, __enable_if.
12345     * Added a typelist implementation for compile-time meta-programming.
12346       Elements for typelist construction and operation can be found
12347       within namespace __gnu_cxx::typelist.
12348     * Added a new allocator, __gnu_cxx::throw_allocator, for testing
12349       exception-safety.
12350     * Enabled library-wide visibility control, allowing -fvisibility to
12351       be used.
12352     * Consolidated all nested namespaces and the conversion of
12353       __gnu_internal implementation-private details to anonymous
12354       namespaces whenever possible.
12355     * Implemented LWG resolutions DR 431 and DR 538.
12356
12357  Fortran
12358
12359     * Support for allocatable components has been added (TR 15581 and
12360       Fortran 2003).
12361     * Support for the Fortran 2003 streaming IO extension has been added.
12362     * The GNU Fortran compiler now uses 4-byte record markers by default
12363       for unformatted files to be compatible with g77 and most other
12364       compilers. The implementation allows for records greater than 2 GB
12365       and is compatible with several other compilers. Older versions of
12366       gfortran used 8-byte record markers by default (on most systems).
12367       In order to change the length of the record markers, e.g. to read
12368       unformatted files created by older gfortran versions, the
12369       [4]-frecord-marker=8 option can be used.
12370
12371  Java (GCJ)
12372
12373     * A new command-line option -static-libgcj has been added for targets
12374       that use a linker compatible with GNU Binutils. As its name
12375       implies, this causes libgcj to be linked statically. In some cases
12376       this causes the resulting executable to start faster and use less
12377       memory than if the shared version of libgcj were used. However
12378       caution should be used as it can also cause essential parts of the
12379       library to be omitted. Some of these issues are discussed in:
12380       [5]https://gcc.gnu.org/wiki/Statically_linking_libgcj
12381     * fastjar is no longer bundled with GCC. To build libgcj, you will
12382       need either InfoZIP (both zip and unzip) or an external jar
12383       program. In the former case, the GCC build will install a jar shell
12384       script that is based on InfoZIP and provides the same functionality
12385       as fastjar.
12386
12387New Targets and Target Specific Improvements
12388
12389  IA-32/x86-64
12390
12391     * -mtune=generic can now be used to generate code running well on
12392       common x86 chips. This includes AMD Athlon, AMD Opteron, Intel
12393       Pentium-M, Intel Pentium 4 and Intel Core 2.
12394     * -mtune=native and -march=native will produce code optimized for the
12395       host architecture as detected using the cpuid instruction.
12396     * Added a new command-line option -fstackrealign and and
12397       __attribute__ ((force_align_arg_pointer)) to realign the stack at
12398       runtime. This allows functions compiled with a vector-aligned stack
12399       to be invoked from legacy objects that keep only word-alignment.
12400
12401  SPARC
12402
12403     * The default CPU setting has been changed from V7 to V9 in 32-bit
12404       mode on Solaris 7 and above. This is already the case in 64-bit
12405       mode. It can be overridden by specifying --with-cpu at configure
12406       time.
12407     * Back-end support of built-in functions for atomic memory access has
12408       been implemented.
12409     * Support for the Sun UltraSPARC T1 (Niagara) processor has been
12410       added.
12411
12412  M32C
12413
12414     * Various bug fixes have made some functions (notably, functions
12415       returning structures) incompatible with previous releases.
12416       Recompiling all libraries is recommended. Note that code quality
12417       has considerably improved since 4.1, making a recompile even more
12418       beneficial.
12419
12420  MIPS
12421
12422     * Added support for the Broadcom SB-1A core.
12423
12424  IA-64
12425
12426     * Added support for IA-64 data and control speculation. By default
12427       speculation is enabled only during second scheduler pass. A number
12428       of machine flags was introduced to control the usage of speculation
12429       for both scheduler passes.
12430
12431  HPPA
12432
12433     * Added Java language support (libffi and libjava) for 32-bit HP-UX
12434       11 target.
12435
12436Obsolete Systems
12437
12438Documentation improvements
12439
12440  PDF Documentation
12441
12442     * A make pdf target has been added to the top-level makefile,
12443       enabling automated production of PDF documentation files.
12444       (Front-ends external to GCC should modify their Make-lang.in file
12445       to add a lang.pdf: target.)
12446
12447Other significant improvements
12448
12449  Build system improvements
12450
12451     * All the components of the compiler are now bootstrapped by default.
12452       This improves the resilience to bugs in the system compiler or
12453       binary compatibility problems, as well as providing better testing
12454       of GCC 4.2 itself. In addition, if you build the compiler from a
12455       combined tree, the assembler, linker, etc. will also be
12456       bootstrapped (i.e. built with themselves).
12457       You can disable this behavior, and go back to the pre-GCC 4.2 set
12458       up, by configuring GCC with --disable-bootstrap.
12459     * The rules that configure follows to find target tools resemble more
12460       closely the locations that the built compiler will search. In
12461       addition, you can use the new configure option --with-target-tools
12462       to specify where to find the target tools used during the build,
12463       without affecting what the built compiler will use.
12464       This can be especially useful when building packages of GCC. For
12465       example, you may want to build GCC with GNU as or ld, even if the
12466       resulting compiler to work with the native assembler and linker. To
12467       do so, you can use --with-target-tools to point to the native
12468       tools.
12469
12470  Incompatible changes to the build system
12471
12472     * Front-ends external to GCC should modify their Make-lang.in file to
12473       replace double-colon rules (e.g. dvi::) with normal rules (like
12474       lang.dvi:). Front-end makefile hooks do not use double-colon rules
12475       anymore.
12476     * Up to GCC 4.1, a popular way to specify the target tools used
12477       during the build was to create directories named gas, binutils,
12478       etc. in the build tree, and create links to the tools from there.
12479       This does not work any more when the compiler is bootstrapped. The
12480       new configure option --with-target-tools provides a better way to
12481       achieve the same effect, and works for all native and cross
12482       settings.
12483
12484
12485    For questions related to the use of GCC, please consult these web
12486    pages and the [6]GCC manuals. If that fails, the
12487    [7]gcc-help@gcc.gnu.org mailing list might help. Comments on these
12488    web pages and the development of GCC are welcome on our developer
12489    list at [8]gcc@gcc.gnu.org. All of [9]our lists have public archives.
12490
12491   Copyright (C) [10]Free Software Foundation, Inc. Verbatim copying and
12492   distribution of this entire article is permitted in any medium,
12493   provided this notice is preserved.
12494
12495   These pages are [11]maintained by the GCC team. Last modified
12496   2019-11-28[12].
12497
12498References
12499
12500   1. http://gcc.gnu.org/projects/gomp/
12501   2. https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html
12502   3. https://gcc.gnu.org/onlinedocs/libstdc++/ext/pb_ds/index.html
12503   4. https://gcc.gnu.org/onlinedocs/gfortran/Runtime-Options.html
12504   5. https://gcc.gnu.org/wiki/Statically_linking_libgcj
12505   6. https://gcc.gnu.org/onlinedocs/
12506   7. mailto:gcc-help@gcc.gnu.org
12507   8. mailto:gcc@gcc.gnu.org
12508   9. https://gcc.gnu.org/lists.html
12509  10. https://www.fsf.org/
12510  11. https://gcc.gnu.org/about.html
12511  12. http://validator.w3.org/check/referer
12512======================================================================
12513http://gcc.gnu.org/gcc-4.1/index.html
12514
12515                             GCC 4.1 Release Series
12516
12517   (This release series is no longer supported.)
12518
12519   February 13, 2007
12520
12521   The [1]GNU project and the GCC developers are pleased to announce the
12522   release of GCC 4.1.2.
12523
12524   This release is a bug-fix release, containing fixes for regressions in
12525   GCC 4.1.1 relative to previous releases of GCC.
12526
12527Release History
12528
12529   GCC 4.1.2
12530          February 13, 2007 ([2]changes)
12531
12532   GCC 4.1.1
12533          May 24, 2006 ([3]changes)
12534
12535   GCC 4.1.0
12536          February 28, 2006 ([4]changes)
12537
12538References and Acknowledgements
12539
12540   GCC used to stand for the GNU C Compiler, but since the compiler
12541   supports several other languages aside from C, it now stands for the
12542   GNU Compiler Collection.
12543
12544   A list of [5]successful builds is updated as new information becomes
12545   available.
12546
12547   The GCC developers would like to thank the numerous people that have
12548   contributed new features, improvements, bug fixes, and other changes as
12549   well as test results to GCC. This [6]amazing group of volunteers is
12550   what makes GCC successful.
12551
12552   For additional information about GCC please refer to the [7]GCC project
12553   web site or contact the [8]GCC development mailing list.
12554
12555   To obtain GCC please use [9]our mirror sites or [10]our version control
12556   system.
12557
12558
12559    For questions related to the use of GCC, please consult these web
12560    pages and the [11]GCC manuals. If that fails, the
12561    [12]gcc-help@gcc.gnu.org mailing list might help. Comments on these
12562    web pages and the development of GCC are welcome on our developer
12563    list at [13]gcc@gcc.gnu.org. All of [14]our lists have public
12564    archives.
12565
12566   Copyright (C) [15]Free Software Foundation, Inc. Verbatim copying and
12567   distribution of this entire article is permitted in any medium,
12568   provided this notice is preserved.
12569
12570   These pages are [16]maintained by the GCC team. Last modified
12571   2020-01-14[17].
12572
12573References
12574
12575   1. http://www.gnu.org/
12576   2. http://gcc.gnu.org/gcc-4.1/changes.html#4.1.2
12577   3. http://gcc.gnu.org/gcc-4.1/changes.html
12578   4. http://gcc.gnu.org/gcc-4.1/changes.html
12579   5. http://gcc.gnu.org/gcc-4.1/buildstat.html
12580   6. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
12581   7. http://gcc.gnu.org/index.html
12582   8. mailto:gcc@gcc.gnu.org
12583   9. http://gcc.gnu.org/mirrors.html
12584  10. http://gcc.gnu.org/git.html
12585  11. https://gcc.gnu.org/onlinedocs/
12586  12. mailto:gcc-help@gcc.gnu.org
12587  13. mailto:gcc@gcc.gnu.org
12588  14. https://gcc.gnu.org/lists.html
12589  15. https://www.fsf.org/
12590  16. https://gcc.gnu.org/about.html
12591  17. http://validator.w3.org/check/referer
12592======================================================================
12593http://gcc.gnu.org/gcc-4.1/changes.html
12594
12595                             GCC 4.1 Release Series
12596                        Changes, New Features, and Fixes
12597
12598   The latest release in the 4.1 release series is [1]GCC 4.1.2.
12599
12600Caveats
12601
12602General Optimizer Improvements
12603
12604     * GCC now has infrastructure for inter-procedural optimizations and
12605       the following inter-procedural optimizations are implemented:
12606          + Profile guided inlining. When doing profile feedback guided
12607            optimization, GCC can now use the profile to make better
12608            informed decisions on whether inlining of a function is
12609            profitable or not. This means that GCC will no longer inline
12610            functions at call sites that are not executed very often, and
12611            that functions at hot call sites are more likely to be
12612            inlined.
12613            A new parameter min-inline-recursive-probability is also now
12614            available to throttle recursive inlining of functions with
12615            small average recursive depths.
12616          + Discovery of pure and const functions, a form of side-effects
12617            analysis. While older GCC releases could also discover such
12618            special functions, the new IPA-based pass runs earlier so that
12619            the results are available to more optimizers. The pass is also
12620            simply more powerful than the old one.
12621          + Analysis of references to static variables and type escape
12622            analysis, also forms of side-effects analysis. The results of
12623            these passes allow the compiler to be less conservative about
12624            call-clobbered variables and references. This results in more
12625            redundant loads being eliminated and in making static
12626            variables candidates for register promotion.
12627          + Improvement of RTL-based alias analysis. The results of type
12628            escape analysis are fed to the RTL type-based alias analyzer,
12629            allowing it to disambiguate more memory references.
12630          + Interprocedural constant propagation and function versioning.
12631            This pass looks for functions that are always called with the
12632            same constant value for one or more of the function arguments,
12633            and propagates those constants into those functions.
12634          + GCC will now eliminate static variables whose usage was
12635            optimized out.
12636          + -fwhole-program --combine can now be used to make all
12637            functions in program static allowing whole program
12638            optimization. As an exception, the main function and all
12639            functions marked with the new externally_visible attribute are
12640            kept global so that programs can link with runtime libraries.
12641     * GCC can now do a form of partial dead code elimination (PDCE) that
12642       allows code motion of expressions to the paths where the result of
12643       the expression is actually needed. This is not always a win, so the
12644       pass has been limited to only consider profitable cases. Here is an
12645       example:
12646    int foo (int *, int *);
12647    int
12648    bar (int d)
12649    {
12650      int a, b, c;
12651      b = d + 1;
12652      c = d + 2;
12653      a = b + c;
12654      if (d)
12655        {
12656          foo (&b, &c);
12657          a = b + c;
12658        }
12659      printf ("%d\n", a);
12660    }
12661
12662       The a = b + c can be sunk to right before the printf. Normal code
12663       sinking will not do this, it will sink the first one above into the
12664       else-branch of the conditional jump, which still gives you two
12665       copies of the code.
12666     * GCC now has a value range propagation pass. This allows the
12667       compiler to eliminate bounds checks and branches. The results of
12668       the pass can also be used to accurately compute branch
12669       probabilities.
12670     * The pass to convert PHI nodes to straight-line code (a form of
12671       if-conversion for GIMPLE) has been improved significantly. The two
12672       most significant improvements are an improved algorithm to
12673       determine the order in which the PHI nodes are considered, and an
12674       improvement that allow the pass to consider if-conversions of basic
12675       blocks with more than two predecessors.
12676     * Alias analysis improvements. GCC can now differentiate between
12677       different fields of structures in Tree-SSA's virtual operands form.
12678       This lets stores/loads from non-overlapping structure fields not
12679       conflict. A new algorithm to compute points-to sets was contributed
12680       that can allows GCC to see now that p->a and p->b, where p is a
12681       pointer to a structure, can never point to the same field.
12682     * Various enhancements to auto-vectorization:
12683          + Incrementally preserve SSA form when vectorizing.
12684          + Incrementally preserve loop-closed form when vectorizing.
12685          + Improvements to peeling for alignment: generate better code
12686            when the misalignment of an access is known at compile time,
12687            or when different accesses are known to have the same
12688            misalignment, even if the misalignment amount itself is
12689            unknown.
12690          + Consider dependence distance in the vectorizer.
12691          + Externalize generic parts of data reference analysis to make
12692            this analysis available to other passes.
12693          + Vectorization of conditional code.
12694          + Reduction support.
12695     * GCC can now partition functions in sections of hot and cold code.
12696       This can significantly improve performance due to better
12697       instruction cache locality. This feature works best together with
12698       profile feedback driven optimization.
12699     * A new pass to avoid saving of unneeded arguments to the stack in
12700       vararg functions if the compiler can prove that they will not be
12701       needed.
12702     * Transition of basic block profiling to tree level implementation
12703       has been completed. The new implementation should be considerably
12704       more reliable (hopefully avoiding profile mismatch errors when
12705       using -fprofile-use or -fbranch-probabilities) and can be used to
12706       drive higher level optimizations, such as inlining.
12707       The -ftree-based-profiling command-line option was removed and
12708       -fprofile-use now implies disabling old RTL level loop optimizer
12709       (-fno-loop-optimize). Speculative prefetching optimization
12710       (originally enabled by -fspeculative-prefetching) was removed.
12711
12712New Languages and Language specific improvements
12713
12714  C and Objective-C
12715
12716     * The old Bison-based C and Objective-C parser has been replaced by a
12717       new, faster hand-written recursive-descent parser.
12718
12719  Ada
12720
12721     * The build infrastructure for the Ada runtime library and tools has
12722       been changed to be better integrated with the rest of the build
12723       infrastructure of GCC. This should make doing cross builds of Ada a
12724       bit easier.
12725
12726  C++
12727
12728     * ARM-style name-injection of friend declarations is no longer the
12729       default. For example:
12730          struct S {
12731            friend void f();
12732          };
12733
12734          void g() { f(); }
12735       will not be accepted; instead a declaration of f will need to be
12736       present outside of the scope of S. The new -ffriend-injection
12737       option will enable the old behavior.
12738     * The (undocumented) extension which permitted templates with default
12739       arguments to be bound to template template parameters with fewer
12740       parameters has been deprecated, and will be removed in the next
12741       major release of G++. For example:
12742       template <template <typename> class C>
12743       void f(C<double>) {}
12744
12745       template <typename T, typename U = int>
12746       struct S {};
12747
12748       template void f(S<double>);
12749
12750       makes use of the deprecated extension. The reason this code is not
12751       valid ISO C++ is that S is a template with two parameters;
12752       therefore, it cannot be bound to C which has only one parameter.
12753
12754    Runtime Library (libstdc++)
12755
12756     * Optimization work:
12757          + A new implementation of std::search_n is provided, better
12758            performing in case of random access iterators.
12759          + Added further efficient specializations of istream functions,
12760            i.e., character array and string extractors.
12761          + Other smaller improvements throughout.
12762     * Policy-based associative containers, designed for high-performance,
12763       flexibility and semantic safety are delivered in ext/pb_assoc.
12764     * A versatile string class, __gnu_cxx::__versa_string, providing
12765       facilities conforming to the standard requirements for
12766       basic_string, is delivered in <ext/vstring.h>. In particular:
12767          + Two base classes are provided: the default one avoids
12768            reference counting and is optimized for short strings; the
12769            alternate one, still uses it while improving in a few low
12770            level areas (e.g., alignment). See vstring_fwd.h for some
12771            useful typedefs.
12772          + Various algorithms have been rewritten (e.g., replace), the
12773            code streamlined and simple optimizations added.
12774          + Option 3 of DR 431 is implemented for both available bases,
12775            thus improving the support for stateful allocators.
12776     * As usual, many bugs have been fixed (e.g., libstdc++/13583,
12777       libstdc++/23953) and LWG resolutions put into effect for the first
12778       time (e.g., DR 280, DR 464, N1780 recommendations for DR 233, TR1
12779       Issue 6.19). The implementation status of TR1 is now tracked in the
12780       docs in tr1.html.
12781
12782  Objective-C++
12783
12784     * A new language front end for Objective-C++ has been added. This
12785       language allows users to mix the object oriented features of
12786       Objective-C with those of C++.
12787
12788  Java (GCJ)
12789
12790     * Core library (libgcj) updates based on GNU Classpath 0.15 - 0.19
12791       features (plus some 0.20 bug-fixes)
12792          + Networking
12793               o The java.net.HttpURLConnection implementation no longer
12794                 buffers the entire response body in memory. This means
12795                 that response bodies larger than available memory can now
12796                 be handled.
12797          + (N)IO
12798               o NIO FileChannel.map implementation, fast bulk put
12799                 implementation for DirectByteBuffer (speeds up this
12800                 method 10x).
12801               o FileChannel.lock() and FileChannel.force() implemented.
12802          + XML
12803               o gnu.xml fix for nodes created outside a namespace
12804                 context.
12805               o Add support for output indenting and
12806                 cdata-section-elements output instruction in
12807                 xml.transform.
12808               o xml.xpath corrections for cases where elements/attributes
12809                 might have been created in non-namespace-aware mode.
12810                 Corrections to handling of XSL variables and minor
12811                 conformance updates.
12812          + AWT
12813               o GNU JAWT implementation, the AWT Native Interface, which
12814                 allows direct access to native screen resources from
12815                 within a Canvas's paint method. GNU Classpath Examples
12816                 comes with a Demo, see libjava/classpath/examples/README.
12817               o awt.datatransfer updated to 1.5 with support for
12818                 FlavorEvents. The gtk+ awt peers now allow copy/paste of
12819                 text, images, URIs/files and serialized objects with
12820                 other applications and tracking clipboard change events
12821                 with gtk+ 2.6 (for gtk+ 2.4 only text and serialized
12822                 objects are supported). A GNU Classpath Examples
12823                 datatransfer Demo was added to show the new
12824                 functionality.
12825               o Split gtk+ awt peers event handling in two threads and
12826                 improve gdk lock handling (solves several awt lock ups).
12827               o Speed up awt Image loading.
12828               o Better gtk+ scrollbar peer implementation when using gtk+
12829                 >= 2.6.
12830               o Handle image loading errors correctly for gdkpixbuf and
12831                 MediaTracker.
12832               o Better handle GDK lock. Properly prefix gtkpeer native
12833                 functions (cp_gtk).
12834               o GdkGraphics2D has been updated to use Cairo 0.5.x or
12835                 higher.
12836               o BufferedImage and GtkImage rewrites. All image drawing
12837                 operations should now work correctly (flipping requires
12838                 gtk+ >= 2.6)
12839               o Future Graphics2D, image and text work is documented at:
12840                 [2]http://developer.classpath.org/mediation/ClasspathGrap
12841                 hicsImagesText
12842               o When gtk+ 2.6 or higher is installed the default log
12843                 handler will produce stack traces whenever a WARNING,
12844                 CRITICAL or ERROR message is produced.
12845          + Free Swing
12846               o The RepaintManager has been reworked for more efficient
12847                 painting, especially for large GUIs.
12848               o The layout manager OverlayLayout has been implemented,
12849                 the BoxLayout has been rewritten to make use of the
12850                 SizeRequirements utility class and caching for more
12851                 efficient layout.
12852               o Improved accessibility support.
12853               o Significant progress has been made in the implementation
12854                 of the javax.swing.plaf.metal package, with most UI
12855                 delegates in a working state now. Please test this with
12856                 your own applications and provide feedback that will help
12857                 us to improve this package.
12858               o The GUI demo (gnu.classpath.examples.swing.Demo) has been
12859                 extended to highlight various features in our Free Swing
12860                 implementation. And it includes a look and feel switcher
12861                 for Metal (default), Ocean and GNU themes.
12862               o The javax.swing.plaf.multi package is now implemented.
12863               o Editing and several key actions for JTree and JTable were
12864                 implemented.
12865               o Lots of icons and look and feel improvements for Free
12866                 Swing basic and metal themes were added. Try running the
12867                 GNU Classpath Swing Demo in examples
12868                 (gnu.classpath.examples.swing.Demo) with:
12869                 -Dswing.defaultlaf=javax.swing.plaf.basic.BasicLookAndFee
12870                 l or
12871                 -Dswing.defaultlaf=javax.swing.plaf.metal.MetalLookAndFee
12872                 l
12873               o Start of styled text capabilites for java.swing.text.
12874               o DefaultMutableTreeNode pre-order, post-order, depth-first
12875                 and breadth-first traversal enumerations implemented.
12876               o JInternalFrame colors and titlebar draw properly.
12877               o JTree is working up to par (icons, selection and keyboard
12878                 traversal).
12879               o JMenus were made more compatible in visual and
12880                 programmatic behavior.
12881               o JTable changeSelection and multiple selections
12882                 implemented.
12883               o JButton and JToggleButton change states work properly
12884                 now.
12885               o JFileChooser fixes.
12886               o revalidate() and repaint() fixes which make Free Swing
12887                 much more responsive.
12888               o MetalIconFactory implemented.
12889               o Free Swing Top-Level Compatibility. JFrame, JDialog,
12890                 JApplet, JInternalFrame, and JWindow are now 1.5
12891                 compatible in the sense that you can call add() and
12892                 setLayout() directly on them, which will have the same
12893                 effect as calling getContentPane().add() and
12894                 getContentPane().setLayout().
12895               o The JTree interface has been completed. JTrees now
12896                 recognizes mouse clicks and selections work.
12897               o BoxLayout works properly now.
12898               o Fixed GrayFilter to actually work.
12899               o Metal SplitPane implemented.
12900               o Lots of Free Swing text and editor stuff work now.
12901          + Free RMI and Corba
12902               o Andrew Watson, Vice President and Technical Director of
12903                 the Object Management Group, has officially assigned us
12904                 20 bit Vendor Minor Code Id: 0x47430 ("GC") that will
12905                 mark remote classpath-specific system exceptions.
12906                 Obtaining the VMCID means that GNU Classpath now is a
12907                 recogniseable type of node in a highly interoperable
12908                 CORBA world.
12909               o GNU Classpath now includes the first working draft to
12910                 support the RMI over IIOP protocol. The current
12911                 implementation is capable of remote invocations,
12912                 transferring various Serializables and Externalizables
12913                 via RMI-IIOP protocol. It can flatten graphs and, at
12914                 least for the simple cases, is interoperable with 1.5
12915                 JDKs.
12916               o org.omg.PortableInterceptor and related functionality in
12917                 other packages is now implemented:
12918                    # The sever and client interceptors work as required
12919                      since 1.4.
12920                    # The IOR interceptor works as needed for 1.5.
12921               o The org.omg.DynamicAny package is completed and passes
12922                 the prepared tests.
12923               o The Portable Object Adapter should now support the output
12924                 of the recent IDL to java compilers. These compilers now
12925                 generate servants and not CORBA objects as before, making
12926                 the output depend on the existing POA implementation.
12927                 Completing POA means that such code can already be tried
12928                 to run on Classpath. Our POA is tested for the following
12929                 usager scenarios:
12930                    # POA converts servant to the CORBA object.
12931                    # Servant provides to the CORBA object.
12932                    # POA activates new CORBA object with the given Object
12933                      Id (byte array) that is later accessible for the
12934                      servant.
12935                    # During the first call, the ServantActivator provides
12936                      servant for this and all subsequent calls on the
12937                      current object.
12938                    # During each call, the ServantLocator provides
12939                      servant for this call only.
12940                    # ServantLocator or ServantActivator forwards call to
12941                      another server.
12942                    # POA has a single servant, responsible for all
12943                      objects.
12944                    # POA has a default servant, but some objects are
12945                      explicitly connected to they specific servants.
12946                 The POA is verified using tests from the former
12947                 cost.omg.org.
12948               o The CORBA implementation is now a working prototype that
12949                 should support features up to 1.3 inclusive. We invite
12950                 groups writing CORBA dependent applications to try
12951                 Classpath implementation, reporting any possible bugs.
12952                 The CORBA prototype is interoperable with Sun's
12953                 implementation v 1.4, transferring object references,
12954                 primitive types, narrow and wide strings, arrays,
12955                 structures, trees, abstract interfaces and value types
12956                 (feature of CORBA 2.3) between these two platforms.
12957                 Remote exceptions are transferred and handled correctly.
12958                 The stringified object references (IORs) from various
12959                 sources are parsed as required. The transient (for
12960                 current session) and permanent (till jre restart)
12961                 redirections work. Both Little and Big Endian encoded
12962                 messages are accepted. The implementation is verified
12963                 using tests from the former cost.omg.org. The current
12964                 release includes working examples (see the examples
12965                 directory), demonstrating the client-server
12966                 communication, using either CORBA Request or IDL-based
12967                 stub (usually generated by a IDL to java compiler). These
12968                 examples also show how to use the Classpath CORBA naming
12969                 service. The IDL to java compiler is not yet written, but
12970                 as our library must be compatible, it naturally accepts
12971                 the output of other idlj implementations.
12972          + Misc
12973               o Updated TimeZone data against Olson tzdata2005l.
12974               o Make zip and jar packages UTF-8 clean.
12975               o "native" code builds and compiles (warning free) on
12976                 Darwin and Solaris.
12977               o java.util.logging.FileHandler now rotates files.
12978               o Start of a generic JDWP framework in gnu/classpath/jdwp.
12979                 This is unfinished, but feedback (at classpath@gnu.org)
12980                 from runtime hackers is greatly appreciated. Although
12981                 most of the work is currently being done around gcj/gij
12982                 we want this framework to be as VM neutral as possible.
12983                 Early design is described in:
12984                 [3]https://gcc.gnu.org/ml/java/2005-05/msg00260.html
12985               o QT4 AWT peers, enable by giving configure
12986                 --enable-qt-peer. Included, but not ready for production
12987                 yet. They are explicitly disabled and not supported. But
12988                 if you want to help with the development of these new
12989                 features we are interested in feedback. You will have to
12990                 explicitly enable them to try them out (and they will
12991                 most likely contain bugs).
12992               o Documentation fixes all over the place. See
12993                 [4]http://developer.classpath.org/doc/
12994
12995New Targets and Target Specific Improvements
12996
12997  IA-32/x86-64
12998
12999     * The x86-64 medium model (that allows building applications whose
13000       data segment exceeds 4GB) was redesigned to match latest ABI draft.
13001       New implementation split large datastructures into separate segment
13002       improving performance of accesses to small datastructures and also
13003       allows linking of small model libraries into medium model programs
13004       as long as the libraries are not accessing the large datastructures
13005       directly. Medium model is also supported in position independent
13006       code now.
13007       The ABI change results in partial incompatibility among medium
13008       model objects. Linking medium model libraries (or objects) compiled
13009       with new compiler into medium model program compiled with older
13010       will likely result in exceeding ranges of relocations.
13011       Binutils 2.16.91 or newer are required for compiling medium model
13012       now.
13013
13014  RS6000 (POWER/PowerPC)
13015
13016     * The AltiVec vector primitives in <altivec.h> are now implemented in
13017       a way that puts a smaller burden on the preprocessor, instead
13018       processing the "overloading" in the front ends. This should benefit
13019       compilation speed on AltiVec vector code.
13020     * AltiVec initializers now are generated more efficiently.
13021     * The popcountb instruction available on POWER5 now is generated.
13022     * The floating point round to integer instructions available on
13023       POWER5+ now is generated.
13024     * Floating point divides can be synthesized using the floating point
13025       reciprocal estimate instructions.
13026     * Double precision floating point constants are initialized as single
13027       precision values if they can be represented exactly.
13028
13029  S/390, zSeries and System z9
13030
13031     * Support for the IBM System z9 109 processor has been added. When
13032       using the -march=z9-109 option, the compiler will generate code
13033       making use of instructions provided by the extended immediate
13034       facility.
13035     * Support for 128-bit IEEE floating point has been added. When using
13036       the -mlong-double-128 option, the compiler will map the long double
13037       data type to 128-bit IEEE floating point. Using this option
13038       constitutes an ABI change, and requires glibc support.
13039     * Various changes to improve performance of generated code have been
13040       implemented, including:
13041          + In functions that do not require a literal pool, register %r13
13042            (which is traditionally reserved as literal pool pointer), can
13043            now be freely used for other purposes by the compiler.
13044          + More precise tracking of register use allows the compiler to
13045            generate more efficient function prolog and epilog code in
13046            certain cases.
13047          + The SEARCH STRING, COMPARE LOGICAL STRING, and MOVE STRING
13048            instructions are now used to implement C string functions.
13049          + The MOVE CHARACTER instruction with single byte overlap is now
13050            used to implement the memset function with non-zero fill byte.
13051          + The LOAD ZERO instructions are now used where appropriate.
13052          + The INSERT CHARACTERS UNDER MASK, STORE CHARACTERS UNDER MASK,
13053            and INSERT IMMEDIATE instructions are now used more frequently
13054            to optimize bitfield operations.
13055          + The BRANCH ON COUNT instruction is now used more frequently.
13056            In particular, the fact that a loop contains a subroutine call
13057            no longer prevents the compiler from using this instruction.
13058          + The compiler is now aware that all shift and rotate
13059            instructions implicitly truncate the shift count to six bits.
13060     * Back-end support for the following generic features has been
13061       implemented:
13062          + The full set of [5]built-in functions for atomic memory
13063            access.
13064          + The -fstack-protector feature.
13065          + The optimization pass avoiding unnecessary stores of incoming
13066            argument registers in functions with variable argument list.
13067
13068  SPARC
13069
13070     * The default code model in 64-bit mode has been changed from
13071       Medium/Anywhere to Medium/Middle on Solaris.
13072     * TLS support is disabled by default on Solaris prior to release 10.
13073       It can be enabled on TLS-capable Solaris 9 versions (4/04 release
13074       and later) by specifying --enable-tls at configure time.
13075
13076  MorphoSys
13077
13078     * Support has been added for this new architecture.
13079
13080Obsolete Systems
13081
13082Documentation improvements
13083
13084Other significant improvements
13085
13086     * GCC can now emit code for protecting applications from
13087       stack-smashing attacks. The protection is realized by buffer
13088       overflow detection and reordering of stack variables to avoid
13089       pointer corruption.
13090     * Some built-in functions have been fortified to protect them against
13091       various buffer overflow (and format string) vulnerabilities.
13092       Compared to the mudflap bounds checking feature, the safe builtins
13093       have far smaller overhead. This means that programs built using
13094       safe builtins should not experience any measurable slowdown.
13095
13096GCC 4.1.2
13097
13098   This is the [6]list of problem reports (PRs) from GCC's bug tracking
13099   system that are known to be fixed in the 4.1.2 release. This list might
13100   not be complete (that is, it is possible that some PRs that have been
13101   fixed are not listed here).
13102
13103   When generating code for a shared library, GCC now recognizes that
13104   global functions may be replaced when the program runs. Therefore, it
13105   is now more conservative in deducing information from the bodies of
13106   functions. For example, in this example:
13107    void f() {}
13108    void g() {
13109     try { f(); }
13110     catch (...) {
13111       cout << "Exception";
13112     }
13113    }
13114
13115   G++ would previously have optimized away the catch clause, since it
13116   would have concluded that f cannot throw exceptions. Because users may
13117   replace f with another function in the main body of the program, this
13118   optimization is unsafe, and is no longer performed. If you wish G++ to
13119   continue to optimize as before, you must add a throw() clause to the
13120   declaration of f to make clear that it does not throw exceptions.
13121
13122
13123    For questions related to the use of GCC, please consult these web
13124    pages and the [7]GCC manuals. If that fails, the
13125    [8]gcc-help@gcc.gnu.org mailing list might help. Comments on these
13126    web pages and the development of GCC are welcome on our developer
13127    list at [9]gcc@gcc.gnu.org. All of [10]our lists have public
13128    archives.
13129
13130   Copyright (C) [11]Free Software Foundation, Inc. Verbatim copying and
13131   distribution of this entire article is permitted in any medium,
13132   provided this notice is preserved.
13133
13134   These pages are [12]maintained by the GCC team. Last modified
13135   2019-11-28[13].
13136
13137References
13138
13139   1. http://gcc.gnu.org/gcc-4.1/changes.html#4.1.2
13140   2. http://developer.classpath.org/mediation/ClasspathGraphicsImagesText
13141   3. https://gcc.gnu.org/ml/java/2005-05/msg00260.html
13142   4. http://developer.classpath.org/doc/
13143   5. https://gcc.gnu.org/onlinedocs/gcc-4.1.0/gcc/Atomic-Builtins.html
13144   6. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.1.2
13145   7. https://gcc.gnu.org/onlinedocs/
13146   8. mailto:gcc-help@gcc.gnu.org
13147   9. mailto:gcc@gcc.gnu.org
13148  10. https://gcc.gnu.org/lists.html
13149  11. https://www.fsf.org/
13150  12. https://gcc.gnu.org/about.html
13151  13. http://validator.w3.org/check/referer
13152======================================================================
13153http://gcc.gnu.org/gcc-4.0/index.html
13154
13155                             GCC 4.0 Release Series
13156
13157   (This release series is no longer supported.)
13158
13159   January 31, 2007
13160
13161   The [1]GNU project and the GCC developers are pleased to announce the
13162   release of GCC 4.0.4.
13163
13164   This release is a bug-fix release, containing fixes for regressions in
13165   GCC 4.0.3 relative to previous releases of GCC.
13166
13167Release History
13168
13169   GCC 4.0.4
13170          January 31, 2007 ([2]changes)
13171
13172   GCC 4.0.3
13173          March 10, 2006 ([3]changes)
13174
13175   GCC 4.0.2
13176          September 28, 2005 ([4]changes)
13177
13178   GCC 4.0.1
13179          July 7, 2005 ([5]changes)
13180
13181   GCC 4.0.0
13182          April 20, 2005 ([6]changes)
13183
13184References and Acknowledgements
13185
13186   GCC used to stand for the GNU C Compiler, but since the compiler
13187   supports several other languages aside from C, it now stands for the
13188   GNU Compiler Collection.
13189
13190   A list of [7]successful builds is updated as new information becomes
13191   available.
13192
13193   The GCC developers would like to thank the numerous people that have
13194   contributed new features, improvements, bug fixes, and other changes as
13195   well as test results to GCC. This [8]amazing group of volunteers is
13196   what makes GCC successful.
13197
13198   For additional information about GCC please refer to the [9]GCC project
13199   web site or contact the [10]GCC development mailing list.
13200
13201   To obtain GCC please use [11]our mirror sites, or [12]our version
13202   control system.
13203
13204
13205    For questions related to the use of GCC, please consult these web
13206    pages and the [13]GCC manuals. If that fails, the
13207    [14]gcc-help@gcc.gnu.org mailing list might help. Comments on these
13208    web pages and the development of GCC are welcome on our developer
13209    list at [15]gcc@gcc.gnu.org. All of [16]our lists have public
13210    archives.
13211
13212   Copyright (C) [17]Free Software Foundation, Inc. Verbatim copying and
13213   distribution of this entire article is permitted in any medium,
13214   provided this notice is preserved.
13215
13216   These pages are [18]maintained by the GCC team. Last modified
13217   2020-01-14[19].
13218
13219References
13220
13221   1. http://www.gnu.org/
13222   2. http://gcc.gnu.org/gcc-4.0/changes.html#4.0.4
13223   3. http://gcc.gnu.org/gcc-4.0/changes.html#4.0.3
13224   4. http://gcc.gnu.org/gcc-4.0/changes.html#4.0.2
13225   5. http://gcc.gnu.org/gcc-4.0/changes.html#4.0.1
13226   6. http://gcc.gnu.org/gcc-4.0/changes.html
13227   7. http://gcc.gnu.org/gcc-4.0/buildstat.html
13228   8. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
13229   9. http://gcc.gnu.org/index.html
13230  10. mailto:gcc@gcc.gnu.org
13231  11. http://gcc.gnu.org/mirrors.html
13232  12. http://gcc.gnu.org/git.html
13233  13. https://gcc.gnu.org/onlinedocs/
13234  14. mailto:gcc-help@gcc.gnu.org
13235  15. mailto:gcc@gcc.gnu.org
13236  16. https://gcc.gnu.org/lists.html
13237  17. https://www.fsf.org/
13238  18. https://gcc.gnu.org/about.html
13239  19. http://validator.w3.org/check/referer
13240======================================================================
13241http://gcc.gnu.org/gcc-4.0/changes.html
13242
13243                             GCC 4.0 Release Series
13244                        Changes, New Features, and Fixes
13245
13246   The latest release in the 4.0 release series is [1]GCC 4.0.4.
13247
13248Caveats
13249
13250     * GCC now generates location lists by default when compiling with
13251       debug info and optimization.
13252          + GDB 6.0 and older crashes when it sees location lists. GDB 6.1
13253            or later is needed to debug binaries containing location
13254            lists.
13255          + When you are trying to view a value of a variable in a part of
13256            a function where it has no location (for example when the
13257            variable is no longer used and thus its location was used for
13258            something else) GDB will say that it is not available.
13259       You can disable generating location lists by -fno-var-tracking.
13260     * GCC no longer accepts the -fwritable-strings option. Use named
13261       character arrays when you need a writable string.
13262     * The options -freduce-all-givs and -fmove-all-movables have been
13263       discontinued. They were used to circumvent a shortcoming in the
13264       heuristics of the old loop optimization code with respect to common
13265       Fortran constructs. The new (tree) loop optimizer works differently
13266       and doesn't need those work-arounds.
13267     * The graph-coloring register allocator, formerly enabled by the
13268       option -fnew-ra, has been discontinued.
13269     * -I- has been deprecated. -iquote is meant to replace the need for
13270       this option.
13271     * The MIPS -membedded-pic and -mrnames options have been removed.
13272     * All MIPS targets now require the GNU assembler. In particular, IRIX
13273       configurations can no longer use the MIPSpro assemblers, although
13274       they do still support the MIPSpro linkers.
13275     * The SPARC option -mflat has been removed.
13276     * English-language diagnostic messages will now use Unicode quotation
13277       marks in UTF-8 locales. (Non-English messages already used the
13278       quotes appropriate for the language in previous releases.) If your
13279       terminal does not support UTF-8 but you are using a UTF-8 locale
13280       (such locales are the default on many GNU/Linux systems) then you
13281       should set LC_CTYPE=C in the environment to disable that locale.
13282       Programs that parse diagnostics and expect plain ASCII
13283       English-language messages should set LC_ALL=C. See [2]Markus Kuhn's
13284       explanation of Unicode quotation marks for more information.
13285     * The specs file is no longer installed on most platforms. Most users
13286       will be totally unaffected. However, if you are accustomed to
13287       editing the specs file yourself, you will now have to use the
13288       -dumpspecs option to generate the specs file, and then edit the
13289       resulting file.
13290
13291General Optimizer Improvements
13292
13293     * The [3]tree ssa branch has been merged. This merge has brought in a
13294       completely new optimization framework based on a higher level
13295       intermediate representation than the existing RTL representation.
13296       Numerous new code transformations based on the new framework are
13297       available in GCC 4.0, including:
13298          + Scalar replacement of aggregates
13299          + Constant propagation
13300          + Value range propagation
13301          + Partial redundancy elimination
13302          + Load and store motion
13303          + Strength reduction
13304          + Dead store elimination
13305          + Dead and unreachable code elimination
13306          + [4]Autovectorization
13307          + Loop interchange
13308          + Tail recursion by accumulation
13309       Many of these passes outperform their counterparts from previous
13310       GCC releases.
13311     * [5]Swing Modulo Scheduling (SMS). An RTL level instruction
13312       scheduling optimization intended for loops that perform heavy
13313       computations.
13314
13315New Languages and Language specific improvements
13316
13317  C family
13318
13319     * The sentinel attribute has been added to GCC. This function
13320       attribute allows GCC to warn when variadic functions such as execl
13321       are not NULL terminated. See the GCC manual for a complete
13322       description of its behavior.
13323     * Given __attribute__((alias("target"))) it is now an error if target
13324       is not a symbol, defined in the same translation unit. This also
13325       applies to aliases created by #pragma weak alias=target. This is
13326       because it's meaningless to define an alias to an undefined symbol.
13327       On Solaris, the native assembler would have caught this error, but
13328       GNU as does not.
13329
13330  C and Objective-C
13331
13332     * The -Wstrict-aliasing=2 option has been added. This warning catches
13333       all unsafe cases, but it may also give a warning for some cases
13334       that are safe.
13335     * The cast-as-lvalue, conditional-expression-as-lvalue and
13336       compound-expression-as-lvalue extensions, which were deprecated in
13337       3.3.4 and 3.4, have been removed.
13338     * The -fwritable-strings option, which was deprecated in 3.4, has
13339       been removed.
13340     * #pragma pack() semantics have been brought closer to those used by
13341       other compilers. This also applies to C++.
13342     * Taking the address of a variable with register storage is invalid
13343       in C. GCC now issues an error instead of a warning.
13344     * Arrays of incomplete element type are invalid in C. GCC now issues
13345       an error for such arrays. Declarations such as extern struct s x[];
13346       (where struct s has not been defined) can be moved after the
13347       definition of struct s. Function parameters declared as arrays of
13348       incomplete type can instead be declared as pointers.
13349
13350  C++
13351
13352     * When compiling without optimizations (-O0), the C++ front end is
13353       much faster than in any previous versions of GCC. Independent
13354       testers have measured speed-ups up to 25% in real-world production
13355       code, compared to the 3.4 family (which was already the fastest
13356       version to date). Upgrading from older versions might show even
13357       bigger improvements.
13358     * ELF visibility attributes can now be applied to a class type, so
13359       that it affects every member function of a class at once, without
13360       having to specify each individually:
13361class __attribute__ ((visibility("hidden"))) Foo
13362{
13363   int foo1();
13364   void foo2();
13365};
13366       The syntax is deliberately similar to the __declspec() system used
13367       by Microsoft Windows based compilers, allowing cross-platform
13368       projects to easily reuse their existing macro system for denoting
13369       exports and imports. By explicitly marking internal classes never
13370       used outside a binary as hidden, one can completely avoid PLT
13371       indirection overheads during their usage by the compiler. You can
13372       find out more about the advantages of this at
13373       [6]https://www.akkadia.org/drepper/dsohowto.pdf
13374     * The -fvisibility-inlines-hidden option has been added which marks
13375       all inlineable functions as having hidden ELF visibility, thus
13376       removing their symbol and typeinfo from the exported symbol table
13377       of the output ELF binary. Using this option can reduce the exported
13378       symbol count of template-heavy code by up to 40% with no code
13379       change at all, thus notably improving link and load times for the
13380       binary as well as a reduction in size of up to 10%. Also, check the
13381       new [7]-fvisibility option.
13382     * The compiler now uses the library interface specified by the [8]C++
13383       ABI for thread-safe initialization of function-scope static
13384       variables. Most users should leave this alone, but embedded
13385       programmers may want to disable this by specifying
13386       -fno-threadsafe-statics for a small savings in code size.
13387     * Taking the address of an explicit register variable is no longer
13388       supported. Note that C++ allows taking the address of variables
13389       with register storage so this will continue to compile with a
13390       warning. For example, assuming that r0 is a machine register:
13391register int foo asm ("r0");
13392register int bar;
13393&foo; // error, no longer accepted
13394&bar; // OK, with a warning
13395     * G++ has an undocumented extension to virtual function covariancy
13396       rules that allowed the overrider to return a type that was
13397       implicitly convertable to the overridden function's return type.
13398       For instance a function returning void * could be overridden by a
13399       function returning T *. This is now deprecated and will be removed
13400       in a future release.
13401     * The G++ minimum and maximum operators (<? and >?) and their
13402       compound forms (<?=) and >?=) have been deprecated and will be
13403       removed in a future version. Code using these operators should be
13404       modified to use std::min and std::max instead.
13405     * Declaration of nested classes of class templates as friends are
13406       supported:
13407template <typename T> struct A {
13408  class B {};
13409};
13410class C {
13411  template <typename T> friend class A<T>::B;
13412};
13413       This complements the feature member functions of class templates as
13414       friends introduced in GCC 3.4.0.
13415     * When declaring a friend class using an unqualified name, classes
13416       outside the innermost non-class scope are not searched:
13417class A;
13418namespace N {
13419  class B {
13420    friend class A;   // Refer to N::A which has not been declared yet
13421                      // because name outside namespace N are not searched
13422    friend class ::A; // Refer to ::A
13423  };
13424}
13425       Hiding the friend name until declaration is still not implemented.
13426     * Friends of classes defined outside their namespace are correctly
13427       handled:
13428namespace N {
13429  class A;
13430}
13431class N::A {
13432  friend class B; // Refer to N::B in GCC 4.0.0
13433                  // but ::B in earlier versions of GCC
13434};
13435
13436    Runtime Library (libstdc++)
13437
13438     * Optimization work:
13439          + Added efficient specializations of istream functions for char
13440            and wchar_t.
13441          + Further performance tuning of strings, in particular wrt
13442            single-char append and getline.
13443          + iter_swap - and therefore most of the mutating algorithms -
13444            now makes an unqualified call to swap when the value_type of
13445            the two iterators is the same.
13446     * A large subset of the features in Technical Report 1 (TR1 for
13447       short) is experimentally delivered (i.e., no guarantees about the
13448       implementation are provided. In particular it is not promised that
13449       the library will remain link-compatible when code using TR1 is
13450       used):
13451          + General utilities such as reference_wrapper and shared_ptr.
13452          + Function objects, i.e., result_of, mem_fn, bind, function.
13453          + Support for metaprogramming.
13454          + New containers such as tuple, array, unordered_set,
13455            unordered_map, unordered_multiset, unordered_multimap.
13456     * As usual, many bugs have been fixed and LWG resolutions implemented
13457       for the first time (e.g., DR 409).
13458
13459  Java
13460
13461     * In order to prevent naming conflicts with other implementations of
13462       these tools, some GCJ binaries have been renamed:
13463          + rmic is now grmic,
13464          + rmiregistry is now grmiregistry, and
13465          + jar is now fastjar.
13466       In particular, these names were problematic for the jpackage.org
13467       packaging conventions which install symlinks in /usr/bin that point
13468       to the preferred versions of these tools.
13469     * The -findirect-dispatch argument to the compiler now works and
13470       generates code following a new "binary compatibility" ABI. Code
13471       compiled this way follows the binary compatibility rules of the
13472       Java Language Specification.
13473     * libgcj now has support for using GCJ as a JIT, using the
13474       gnu.gcj.jit family of system properties.
13475     * libgcj can now find a shared library corresponding to the bytecode
13476       representation of a class. See the documentation for the new
13477       gcj-dbtool program, and the new gnu.gcj.precompiled.db.path system
13478       property.
13479     * There have been many improvements to the class library. Here are
13480       some highlights:
13481          + Much more of AWT and Swing exist.
13482          + Many new packages and classes were added, including
13483            java.util.regex, java.net.URI, javax.crypto,
13484            javax.crypto.interfaces, javax.crypto.spec, javax.net,
13485            javax.net.ssl, javax.security.auth,
13486            javax.security.auth.callback, javax.security.auth.login,
13487            javax.security.auth.x500, javax.security.sasl, org.ietf.jgss,
13488            javax.imageio, javax.imageio.event, javax.imageio.spi,
13489            javax.print, javax.print.attribute,
13490            javax.print.attribute.standard, javax.print.event, and
13491            javax.xml
13492          + Updated SAX and DOM, and imported GNU JAXP
13493
13494  Fortran
13495
13496     * A new [9]Fortran front end has replaced the aging GNU Fortran 77
13497       front end. The new front end supports Fortran 90 and Fortran 95. It
13498       may not yet be as stable as the old Fortran front end.
13499
13500  Ada
13501
13502     * Ada (with tasking and Zero Cost Exceptions) is now available on
13503       many more targets, including but not limited to: alpha-linux,
13504       hppa-hpux, hppa-linux, powerpc-darwin, powerpc-linux, s390-linux,
13505       s390x-linux, sparc-linux.
13506     * Some of the new Ada 2005 features are now implemented like
13507       Wide_Wide_Character and Ada.Containers.
13508     * Many bugs have been fixed, tools and documentation improved.
13509     * To compile Ada from the sources, install an older working Ada
13510       compiler and then use --enable-languages=ada at configuration time,
13511       since the Ada front end is not currently activated by default. See
13512       the [10]Installing GCC for details.
13513
13514New Targets and Target Specific Improvements
13515
13516  H8/300
13517
13518     * The frame layout has changed. In the new layout, the prologue of a
13519       function first saves registers and then allocate space for locals,
13520       resulting in an 1% improvement on code size.
13521
13522  IA-32/x86-64 (AMD64)
13523
13524     * The acos, asin, drem, exp10, exp2, expm1, fmod, ilogb, log10,
13525       log1p, log2, logb and tan mathematical builtins (and their float
13526       and long double variants) are now implemented as inline x87
13527       intrinsics when using -ffast-math.
13528     * The ceil, floor, nearbyint, rint and trunc mathematical builtins
13529       (and their float and long double variants) are now implemented as
13530       inline x87 intrinsics when using -ffast-math.
13531     * The x87's fsincos instruction is now used automatically with
13532       -ffast-math when calculating both the sin and cos of the same
13533       argument.
13534     * Instruction selection for multiplication and division by constants
13535       has been improved.
13536
13537  IA-64
13538
13539     * Floating point division, integer division and sqrt are now inlined,
13540       resulting in significant performance improvements on some codes.
13541
13542  MIPS
13543
13544     * Division by zero checks now use conditional traps if the target
13545       processor supports them. This decreases code size by one word per
13546       division operation. The old behavior (branch and break) can be
13547       obtained either at configure time by passing --with-divide=breaks
13548       to configure or at runtime by passing -mdivide-breaks to GCC.
13549     * Support for MIPS64 paired-single instructions has been added. It is
13550       enabled by -mpaired-single and can be accessed using both the
13551       target-independent vector extensions and new MIPS-specific built-in
13552       functions.
13553     * Support for the MIPS-3D ASE has been added. It is enabled by
13554       -mips3d and provides new MIPS-3D-specific built-in functions.
13555     * The -mexplicit-relocs option now supports static n64 code (as is
13556       used, for example, in 64-bit linux kernels). -mexplicit-relocs
13557       should now be feature-complete and is enabled by default when GCC
13558       is configured to use a compatible assembler.
13559     * Support for the NEC VR4130 series has been added. This support
13560       includes the use of VR-specific instructions and a new VR4130
13561       scheduler. Full VR4130 support can be selected with -march=vr4130
13562       while code for any ISA can be tuned for the VR4130 using
13563       -mtune=vr4130. There is also a new -mvr4130-align option that
13564       produces better schedules at the cost of increased code size.
13565     * Support for the Broadcom SB-1 has been extended. There is now an
13566       SB-1 scheduler as well as support for the SB-1-specific
13567       paired-single instructions. Full SB-1 support can be selected with
13568       -march=sb1 while code for any ISA can be optimized for the SB-1
13569       using -mtune=sb1.
13570     * The compiler can now work around errata in R4000, R4400, VR4120 and
13571       VR4130 processors. These workarounds are enabled by -mfix-r4000,
13572       -mfix-r4400, -mfix-vr4120 and -mfix-vr4130 respectively. The VR4120
13573       and VR4130 workarounds need binutils 2.16 or above.
13574     * IRIX shared libraries are now installed into the standard library
13575       directories: o32 libraries go into lib/, n32 libraries go into
13576       lib32/ and n64 libraries go into lib64/.
13577     * The compiler supports a new -msym32 option. It can be used to
13578       optimize n64 code in which all symbols are known to have 32-bit
13579       values.
13580
13581  S/390 and zSeries
13582
13583     * New command-line options help to generate code intended to run in
13584       an environment where stack space is restricted, e.g. Linux kernel
13585       code:
13586          + -mwarn-framesize and -mwarn-dynamicstack trigger compile-time
13587            warnings for single functions that require large or dynamic
13588            stack frames.
13589          + -mstack-size and -mstack-guard generate code that checks for
13590            stack overflow at run time.
13591          + -mpacked-stack generates code that reduces the stack frame
13592            size of many functions by reusing unneeded parts of the stack
13593            bias area.
13594     * The -msoft-float option now ensures that generated code never
13595       accesses floating point registers.
13596     * The s390x-ibm-tpf target now fully supports C++, including
13597       exceptions and threads.
13598     * Various changes to improve performance of the generated code have
13599       been implemented, including:
13600          + GCC now uses sibling calls where possible.
13601          + Condition code handling has been optimized, allowing GCC to
13602            omit redundant comparisons in certain cases.
13603          + The cost function guiding many optimizations has been refined
13604            to more accurately represent the z900 and z990 processors.
13605          + The ADD LOGICAL WITH CARRY and SUBTRACT LOGICAL WITH BORROW
13606            instructions are now used to avoid conditional branches in
13607            certain cases.
13608          + The back end now uses the LEGITIMIZE_RELOAD_ADDRESS feature to
13609            optimize address arithmetic required to access large stack
13610            frames.
13611          + GCC now makes more efficient use of memory-to-memory type
13612            instructions (MVC, CLC, ...).
13613          + More precise tracking of special register use allows better
13614            instruction scheduling, in particular of the function prologue
13615            and epilogue sequences.
13616          + The Java front end now generates inline code to implement
13617            integer division, instead of calling library routines.
13618
13619  SPARC
13620
13621     * The options -mv8, -msparclite, -mcypress, -msupersparc, -mf930 and
13622       -mf934 have been removed. They have been replaced with -mcpu=xxx.
13623     * The internal model used to estimate the relative cost of each
13624       instruction has been updated. It is expected to give better results
13625       on recent UltraSPARC processors.
13626     * Code generation for function prologues and epilogues has been
13627       improved, resulting in better scheduling and allowing multiple exit
13628       points in functions.
13629     * Support for Sun's Visual Instruction Set (VIS) has been enhanced.
13630       It is enabled by -mvis and provides new built-in functions for VIS
13631       instructions on UltraSPARC processors.
13632     * The option -mapp-regs has been turned on by default on Solaris too.
13633
13634  NetWare
13635
13636     * Novell NetWare (on ix86, no other hardware platform was ever really
13637       supported by this OS) has been re-enabled and the ABI supported by
13638       GCC has been brought into sync with that of MetroWerks CodeWarrior
13639       (the ABI previously supported was that of some Unix systems, which
13640       NetWare never tried to support).
13641
13642Obsolete Systems
13643
13644   Support for a number of older systems has been declared obsolete in GCC
13645   4.0. Unless there is activity to revive them, the next release of GCC
13646   will have their sources permanently removed.
13647
13648   All GCC ports for the following processor architectures have been
13649   declared obsolete:
13650     * Intel i860
13651     * Ubicom IP2022
13652     * National Semiconductor NS32K (ns32k)
13653     * Texas Instruments TMS320C[34]x
13654
13655   Also, those for some individual systems have been obsoleted:
13656     * SPARC family
13657          + SPARClite-based systems (sparclite-*-coff, sparclite-*-elf,
13658            sparc86x-*-elf)
13659          + OpenBSD 32-bit (sparc-*-openbsd*)
13660
13661Documentation improvements
13662
13663Other significant improvements
13664
13665     * Location lists are now generated by default when compiling with
13666       debug info and optimization. Location lists provide more accurate
13667       debug info about locations of variables and they allow debugging
13668       code compiled with -fomit-frame-pointer.
13669     * The -fvisibility option has been added which allows the default ELF
13670       visibility of all symbols to be set per compilation and the new
13671       #pragma GCC visibility preprocessor command allows the setting of
13672       default ELF visibility for a region of code. Using
13673       -fvisibility=hidden especially in combination with the new
13674       -fvisibility-inlines-hidden can yield substantial improvements in
13675       output binary quality including avoiding PLT indirection overheads,
13676       reduction of the exported symbol count by up to 60% (with resultant
13677       improvements to link and load times), better scope for the
13678       optimizer to improve code and up to a 20% reduction in binary size.
13679       Using these options correctly yields a binary with a similar symbol
13680       count to a Windows DLL.
13681       Perhaps more importantly, this new feature finally allows (with
13682       careful planning) complete avoidance of symbol clashes when
13683       manually loading shared objects with RTLD_GLOBAL, thus finally
13684       solving problems many projects such as python were forced to use
13685       RTLD_LOCAL for (with its resulting issues for C++ correctness). You
13686       can find more information about using these options at
13687       [11]https://gcc.gnu.org/wiki/Visibility.
13688     __________________________________________________________________
13689
13690GCC 4.0.1
13691
13692   This is the [12]list of problem reports (PRs) from GCC's bug tracking
13693   system that are known to be fixed in the 4.0.1 release. This list might
13694   not be complete (that is, it is possible that some PRs that have been
13695   fixed are not listed here).
13696
13697GCC 4.0.2
13698
13699   This is the [13]list of problem reports (PRs) from GCC's bug tracking
13700   system that are known to be fixed in the 4.0.2 release. This list might
13701   not be complete (that is, it is possible that some PRs that have been
13702   fixed are not listed here).
13703
13704   Unfortunately, due to a release engineering failure, this release has a
13705   regression on Solaris that will affect some C++ programs. We suggest
13706   that Solaris users apply a [14]patch that corrects the problem. Users
13707   who do not wish to apply the patch should explicitly link C++ programs
13708   with the -pthreads option, even if they do not use threads. This
13709   problem has been corrected in the current 4.0 branch sources and will
13710   not be present in GCC 4.0.3.
13711
13712GCC 4.0.3
13713
13714   Starting with this release, the function getcontext is recognized by
13715   the compiler as having the same semantics as the setjmp function. In
13716   particular, the compiler will ensure that all registers are dead before
13717   calling such a function and will emit a warning about the variables
13718   that may be clobbered after the second return from the function.
13719
13720GCC 4.0.4
13721
13722   This is the [15]list of problem reports (PRs) from GCC's bug tracking
13723   system that are known to be fixed in the 4.0.4 release. This list might
13724   not be complete (that is, it is possible that some PRs that have been
13725   fixed are not listed here).
13726
13727   The 4.0.4 release is provided for those that require a high degree of
13728   binary compatibility with previous 4.0.x releases. For most users, the
13729   GCC team recommends that version 4.1.1 or later be used instead."
13730
13731
13732    For questions related to the use of GCC, please consult these web
13733    pages and the [16]GCC manuals. If that fails, the
13734    [17]gcc-help@gcc.gnu.org mailing list might help. Comments on these
13735    web pages and the development of GCC are welcome on our developer
13736    list at [18]gcc@gcc.gnu.org. All of [19]our lists have public
13737    archives.
13738
13739   Copyright (C) [20]Free Software Foundation, Inc. Verbatim copying and
13740   distribution of this entire article is permitted in any medium,
13741   provided this notice is preserved.
13742
13743   These pages are [21]maintained by the GCC team. Last modified
13744   2019-11-28[22].
13745
13746References
13747
13748   1. http://gcc.gnu.org/gcc-4.0/changes.html#4.0.4
13749   2. https://www.cl.cam.ac.uk/~mgk25/ucs/quotes.html
13750   3. http://gcc.gnu.org/projects/tree-ssa/
13751   4. http://gcc.gnu.org/projects/tree-ssa/vectorization.html
13752   5. http://gcc.gnu.org/news/sms.html
13753   6. https://www.akkadia.org/drepper/dsohowto.pdf
13754   7. http://gcc.gnu.org/gcc-4.0/changes.html#visibility
13755   8. https://itanium-cxx-abi.github.io/cxx-abi/
13756   9. http://gcc.gnu.org/fortran/
13757  10. https://gcc.gnu.org/install/
13758  11. https://gcc.gnu.org/wiki/Visibility
13759  12. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.0.1
13760  13. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.0.2
13761  14. https://gcc.gnu.org/ml/gcc-cvs/2005-09/msg00984.html
13762  15. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=4.0.4
13763  16. https://gcc.gnu.org/onlinedocs/
13764  17. mailto:gcc-help@gcc.gnu.org
13765  18. mailto:gcc@gcc.gnu.org
13766  19. https://gcc.gnu.org/lists.html
13767  20. https://www.fsf.org/
13768  21. https://gcc.gnu.org/about.html
13769  22. http://validator.w3.org/check/referer
13770======================================================================
13771http://gcc.gnu.org/gcc-3.4/index.html
13772
13773                             GCC 3.4 Release Series
13774
13775   (This release series is no longer supported.)
13776
13777   May 26, 2006
13778
13779   The [1]GNU project and the GCC developers are pleased to announce the
13780   release of GCC 3.4.6.
13781
13782   This release is a bug-fix release, containing fixes for regressions in
13783   GCC 3.4.4 relative to previous releases of GCC. This is the last of the
13784   3.4.x series.
13785
13786   The GCC 3.4 release series includes numerous [2]new features,
13787   improvements, bug fixes, and other changes, thanks to an [3]amazing
13788   group of volunteers.
13789
13790Release History
13791
13792   GCC 3.4.6
13793          March 6, 2006 ([4]changes)
13794
13795   GCC 3.4.5
13796          November 30, 2005 ([5]changes)
13797
13798   GCC 3.4.4
13799          May 18, 2005 ([6]changes)
13800
13801   GCC 3.4.3
13802          November 4, 2004 ([7]changes)
13803
13804   GCC 3.4.2
13805          September 6, 2004 ([8]changes)
13806
13807   GCC 3.4.1
13808          July 1, 2004 ([9]changes)
13809
13810   GCC 3.4.0
13811          April 18, 2004 ([10]changes)
13812
13813References and Acknowledgements
13814
13815   GCC used to stand for the GNU C Compiler, but since the compiler
13816   supports several other languages aside from C, it now stands for the
13817   GNU Compiler Collection.
13818
13819   A list of [11]successful builds is updated as new information becomes
13820   available.
13821
13822   The GCC developers would like to thank the numerous people that have
13823   contributed new features, improvements, bug fixes, and other changes as
13824   well as test results to GCC. This [12]amazing group of volunteers is
13825   what makes GCC successful.
13826
13827   For additional information about GCC please refer to the [13]GCC
13828   project web site or contact the [14]GCC development mailing list.
13829
13830   To obtain GCC please use [15]our mirror sites, or [16]our version
13831   control system.
13832
13833
13834    For questions related to the use of GCC, please consult these web
13835    pages and the [17]GCC manuals. If that fails, the
13836    [18]gcc-help@gcc.gnu.org mailing list might help. Comments on these
13837    web pages and the development of GCC are welcome on our developer
13838    list at [19]gcc@gcc.gnu.org. All of [20]our lists have public
13839    archives.
13840
13841   Copyright (C) [21]Free Software Foundation, Inc. Verbatim copying and
13842   distribution of this entire article is permitted in any medium,
13843   provided this notice is preserved.
13844
13845   These pages are [22]maintained by the GCC team. Last modified
13846   2020-01-14[23].
13847
13848References
13849
13850   1. http://www.gnu.org/
13851   2. http://gcc.gnu.org/gcc-3.4/changes.html
13852   3. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
13853   4. http://gcc.gnu.org/gcc-3.4/changes.html#3.4.6
13854   5. http://gcc.gnu.org/gcc-3.4/changes.html#3.4.5
13855   6. http://gcc.gnu.org/gcc-3.4/changes.html#3.4.4
13856   7. http://gcc.gnu.org/gcc-3.4/changes.html#3.4.3
13857   8. http://gcc.gnu.org/gcc-3.4/changes.html#3.4.2
13858   9. http://gcc.gnu.org/gcc-3.4/changes.html#3.4.1
13859  10. http://gcc.gnu.org/gcc-3.4/changes.html
13860  11. http://gcc.gnu.org/gcc-3.4/buildstat.html
13861  12. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
13862  13. http://gcc.gnu.org/index.html
13863  14. mailto:gcc@gcc.gnu.org
13864  15. http://gcc.gnu.org/mirrors.html
13865  16. http://gcc.gnu.org/git.html
13866  17. https://gcc.gnu.org/onlinedocs/
13867  18. mailto:gcc-help@gcc.gnu.org
13868  19. mailto:gcc@gcc.gnu.org
13869  20. https://gcc.gnu.org/lists.html
13870  21. https://www.fsf.org/
13871  22. https://gcc.gnu.org/about.html
13872  23. http://validator.w3.org/check/referer
13873======================================================================
13874http://gcc.gnu.org/gcc-3.4/changes.html
13875
13876                             GCC 3.4 Release Series
13877                        Changes, New Features, and Fixes
13878
13879   The final release in the 3.4 release series is [1]GCC 3.4.6. The series
13880   is now closed.
13881
13882   GCC 3.4 has [2]many improvements in the C++ front end. Before reporting
13883   a bug, please make sure it's really GCC, and not your code, that is
13884   broken.
13885
13886Caveats
13887
13888     * GNU Make is now required to build GCC.
13889     * With -nostdinc the preprocessor used to ignore both standard
13890       include paths and include paths contained in environment variables.
13891       It was neither documented nor intended that environment variable
13892       paths be ignored, so this has been corrected.
13893     * GCC no longer accepts the options -fvolatile, -fvolatile-global and
13894       -fvolatile-static. It is unlikely that they worked correctly in any
13895       3.x release.
13896     * GCC no longer ships <varargs.h>. Use <stdarg.h> instead.
13897     * Support for all the systems [3]obsoleted in GCC 3.3 has been
13898       removed from GCC 3.4. See below for a [4]list of systems which are
13899       obsoleted in this release.
13900     * GCC now requires an ISO C90 (ANSI C89) C compiler to build. K&R C
13901       compilers will not work.
13902     * The implementation of the [5]MIPS ABIs has changed. As a result,
13903       the code generated for certain MIPS targets will not be binary
13904       compatible with earlier releases.
13905     * In previous releases, the MIPS port had a fake "hilo" register with
13906       the user-visible name accum. This register has been removed.
13907     * The implementation of the [6]SPARC ABIs has changed. As a result,
13908       the code generated will not be binary compatible with earlier
13909       releases in certain cases.
13910     * The configure option --enable-threads=pthreads has been removed;
13911       use --enable-threads=posix instead, which should have the same
13912       effect.
13913     * Code size estimates used by inlining heuristics for C, Objective-C,
13914       C++ and Java have been redesigned significantly. As a result the
13915       parameters of -finline-insns, --param max-inline-insns-single and
13916       --param max-inline-insns-auto need to be reconsidered.
13917     * --param max-inline-slope and --param min-inline-insns have been
13918       removed; they are not needed for the new bottom-up inlining
13919       heuristics.
13920     * The new unit-at-a-time compilation scheme has several compatibility
13921       issues:
13922          + The order in which functions, variables, and top-level asm
13923            statements are emitted may have changed. Code relying on some
13924            particular ordering needs to be updated. The majority of such
13925            top-level asm statements can be replaced by section
13926            attributes.
13927          + Unreferenced static variables and functions are removed. This
13928            may result in undefined references when an asm statement
13929            refers to the variable/function directly. In that case either
13930            the variable/function shall be listed in asm statement operand
13931            or in the case of top-level asm statements the attribute used
13932            shall be used to force function/variable to be always output
13933            and considered as a possibly used by unknown code.
13934            For variables the attribute is accepted only by GCC 3.4 and
13935            newer, while for earlier versions it is sufficient to use
13936            unused to silence warnings about the variables not being
13937            referenced. To keep code portable across different GCC
13938            versions, you can use appropriate preprocessor conditionals.
13939          + Static functions now can use non-standard passing conventions
13940            that may break asm statements calling functions directly.
13941            Again the attribute used shall be used to prevent this
13942            behavior.
13943       As a temporary workaround, -fno-unit-at-a-time can be used, but
13944       this scheme may not be supported by future releases of GCC.
13945     * GCC 3.4 automatically places zero-initialized variables in the .bss
13946       section on some operating systems. Versions of GNU Emacs up to (and
13947       including) 21.3 will not work correctly when using this
13948       optimization; you can use -fno-zero-initialized-in-bss to disable
13949       it.
13950     * If GCC 3.4 is configured with --enable-threads=posix (the default
13951       on most targets that support pthreads) then _REENTRANT will be
13952       defined unconditionally by some libstdc++ headers. C++ code which
13953       relies on that macro to detect whether multi-threaded code is being
13954       compiled might change in meaning, possibly resulting in linker
13955       errors for single-threaded programs. Affected users of [7]Boost
13956       should compile single-threaded code with -DBOOST_DISABLE_THREADS.
13957       See Bugzilla for [8]more information.
13958
13959General Optimizer Improvements
13960
13961     * Usability of the profile feedback and coverage testing has been
13962       improved.
13963          + Performance of profiled programs has been improved by faster
13964            profile merging code.
13965          + Better use of the profile feedback for optimization (loop
13966            unrolling and loop peeling).
13967          + File locking support allowing fork() calls and parallel runs
13968            of profiled programs.
13969          + Coverage file format has been redesigned.
13970          + gcov coverage tool has been improved.
13971          + make profiledbootstrap available to build a faster compiler.
13972            Experiments made on i386 hardware showed an 11% speedup on -O0
13973            and a 7.5% speedup on -O2 compilation of a [9]large C++
13974            testcase.
13975          + New value profiling pass enabled via -fprofile-values
13976          + New value profile transformations pass enabled via -fvpt aims
13977            to optimize some code sequences by exploiting knowledge about
13978            value ranges or other properties of the operands. At the
13979            moment a conversion of expensive divisions into cheaper
13980            operations has been implemented.
13981          + New -fprofile-generate and -fprofile-use command-line options
13982            to simplify the use of profile feedback.
13983     * A new unit-at-a-time compilation scheme for C, Objective-C, C++ and
13984       Java which is enabled via -funit-at-a-time (and implied by -O2). In
13985       this scheme a whole file is parsed first and optimized later. The
13986       following basic inter-procedural optimizations are implemented:
13987          + Removal of unreachable functions and variables
13988          + Discovery of local functions (functions with static linkage
13989            whose address is never taken)
13990          + On i386, these local functions use register parameter passing
13991            conventions.
13992          + Reordering of functions in topological order of the call graph
13993            to enable better propagation of optimizing hints (such as the
13994            stack alignments needed by functions) in the back end.
13995          + Call graph based out-of-order inlining heuristics which allows
13996            to limit overall compilation unit growth (--param
13997            inline-unit-growth).
13998       Overall, the unit-at-a-time scheme produces a 1.3% improvement for
13999       the SPECint2000 benchmark on the i386 architecture (AMD Athlon
14000       CPU).
14001     * More realistic code size estimates used by inlining for C,
14002       Objective-C, C++ and Java. The growth of large functions can now be
14003       limited via --param large-function-insns and --param
14004       large-function-growth.
14005     * A new cfg-level loop optimizer pass replaces the old loop unrolling
14006       pass and adds two other loop transformations -- loop peeling and
14007       loop unswitching -- and also uses the profile feedback to limit
14008       code growth. (The three optimizations are enabled by
14009       -funroll-loops, -fpeel-loops and -funswitch-loops flags,
14010       respectively).
14011       The old loop unroller still can be enabled by -fold-unroll-loops
14012       and may produce better code in some cases, especially when the
14013       webizer optimization pass is not run.
14014     * A new web construction pass enabled via -fweb (and implied by -O3)
14015       improves the quality of register allocation, CSE, first scheduling
14016       pass and some other optimization passes by avoiding re-use of
14017       pseudo registers with non-overlapping live ranges. The pass almost
14018       always improves code quality but does make debugging difficult and
14019       thus is not enabled by default by -O2
14020       The pass is especially effective as cleanup after code duplication
14021       passes, such as the loop unroller or the tracer.
14022     * Experimental implementations of superblock or trace scheduling in
14023       the second scheduling pass can be enabled via
14024       -fsched2-use-superblocks and -fsched2-use-traces, respectively.
14025
14026New Languages and Language specific improvements
14027
14028  Ada
14029
14030     * The Ada front end has been updated to include numerous bug fixes
14031       and enhancements. These include:
14032          + Improved project file support
14033          + Additional set of warnings about potential wrong code
14034          + Improved error messages
14035          + Improved code generation
14036          + Improved cross reference information
14037          + Improved inlining
14038          + Better run-time check elimination
14039          + Better error recovery
14040          + More efficient implementation of unbounded strings
14041          + Added features in GNAT.Sockets, GNAT.OS_Lib, GNAT.Debug_Pools,
14042            ...
14043          + New GNAT.xxxx packages (e.g. GNAT.Strings,
14044            GNAT.Exception_Action)
14045          + New pragmas
14046          + New -gnatS switch replacing gnatpsta
14047          + Implementation of new Ada features (in particular limited
14048            with, limited aggregates)
14049
14050  C/Objective-C/C++
14051
14052     * Precompiled headers are now supported. Precompiled headers can
14053       dramatically speed up compilation of some projects. There are some
14054       known defects in the current precompiled header implementation that
14055       will result in compiler crashes in relatively rare situations.
14056       Therefore, precompiled headers should be considered a "technology
14057       preview" in this release. Read the manual for details about how to
14058       use precompiled headers.
14059     * File handling in the preprocessor has been rewritten. GCC no longer
14060       gets confused by symlinks and hardlinks, and now has a correct
14061       implementation of #import and #pragma once. These two directives
14062       have therefore been un-deprecated.
14063     * The undocumented extension that allowed C programs to have a label
14064       at the end of a compound statement, which has been deprecated since
14065       GCC 3.0, has been removed.
14066     * The cast-as-lvalue extension has been removed for C++ and
14067       deprecated for C and Objective-C. In particular, code like this:
14068        int i;
14069        (char) i = 5;
14070
14071       or this:
14072        char *p;
14073        ((int *) p)++;
14074
14075       is no longer accepted for C++ and will not be accepted for C and
14076       Objective-C in a future version.
14077     * The conditional-expression-as-lvalue extension has been deprecated
14078       for C and Objective-C. In particular, code like this:
14079        int a, b, c;
14080        (a ? b : c) = 2;
14081
14082       will not be accepted for C and Objective-C in a future version.
14083     * The compound-expression-as-lvalue extension has been deprecated for
14084       C and Objective-C. In particular, code like this:
14085        int a, b;
14086        (a, b) = 2;
14087
14088       will not be accepted for C and Objective-C in a future version. A
14089       possible non-intrusive workaround is the following:
14090        (*(a, &b)) = 2;
14091
14092     * Several [10]built-in functions such as __builtin_popcount for
14093       counting bits, finding the highest and lowest bit in a word, and
14094       parity have been added.
14095     * The -fwritable-strings option has been deprecated and will be
14096       removed.
14097     * Many C math library functions are now recognized as built-ins and
14098       optimized.
14099     * The C, C++, and Objective-C compilers can now handle source files
14100       written in any character encoding supported by the host C library.
14101       The default input character set is taken from the current locale,
14102       and may be overridden with the -finput-charset command line option.
14103       In the future we will add support for inline encoding markers.
14104
14105  C++
14106
14107     * G++ is now much closer to full conformance to the ISO/ANSI C++
14108       standard. This means, among other things, that a lot of invalid
14109       constructs which used to be accepted in previous versions will now
14110       be rejected. It is very likely that existing C++ code will need to
14111       be fixed. This document lists some of the most common issues.
14112     * A hand-written recursive-descent C++ parser has replaced the
14113       YACC-derived C++ parser from previous GCC releases. The new parser
14114       contains much improved infrastructure needed for better parsing of
14115       C++ source codes, handling of extensions, and clean separation
14116       (where possible) between proper semantics analysis and parsing. The
14117       new parser fixes many bugs that were found in the old parser.
14118     * You must now use the typename and template keywords to disambiguate
14119       dependent names, as required by the C++ standard.
14120        struct K {
14121          typedef int mytype_t;
14122        };
14123
14124        template <class T1> struct A {
14125          template <class T2> struct B {
14126              void callme(void);
14127            };
14128
14129          template <int N> void bar(void)
14130          {
14131            // Use 'typename' to tell the parser that T1::mytype_t names
14132            //  a type. This is needed because the name is dependent (in
14133            //  this case, on template parameter T1).
14134            typename T1::mytype_t x;
14135            x = 0;
14136          }
14137        };
14138
14139        template <class T> void template_func(void)
14140        {
14141          // Use 'template' to prefix member templates within
14142          //  dependent types (a has type A<T>, which depends on
14143          //  the template parameter T).
14144          A<T> a;
14145          a.template bar<0>();
14146
14147          // Use 'template' to tell the parser that B is a nested
14148          //  template class (dependent on template parameter T), and
14149          //  'typename' because the whole A<T>::B<int> is
14150          //  the name of a type (again, dependent).
14151          typename A<T>::template B<int> b;
14152          b.callme();
14153        }
14154
14155        void non_template_func(void)
14156        {
14157          // Outside of any template class or function, no names can be
14158          //  dependent, so the use of the keyword 'typename' and 'template'
14159          //  is not needed (and actually forbidden).
14160          A<K> a;
14161          a.bar<0>();
14162          A<K>::B<float> b;
14163          b.callme();
14164        }
14165     * In a template definition, unqualified names will no longer find
14166       members of a dependent base (as specified by [temp.dep]/3 in the
14167       C++ standard). For example,
14168        template <typename T> struct B {
14169          int m;
14170          int n;
14171          int f ();
14172          int g ();
14173        };
14174        int n;
14175        int g ();
14176        template <typename T> struct C : B<T> {
14177          void h ()
14178          {
14179            m = 0; // error
14180            f ();  // error
14181            n = 0; // ::n is modified
14182            g ();  // ::g is called
14183          }
14184        };
14185       You must make the names dependent, e.g. by prefixing them with
14186       this->. Here is the corrected definition of C<T>::h,
14187        template <typename T> void C<T>::h ()
14188        {
14189          this->m = 0;
14190          this->f ();
14191          this->n = 0
14192          this->g ();
14193        }
14194       As an alternative solution (unfortunately not backwards compatible
14195       with GCC 3.3), you may use using declarations instead of this->:
14196        template <typename T> struct C : B<T> {
14197          using B<T>::m;
14198          using B<T>::f;
14199          using B<T>::n;
14200          using B<T>::g;
14201          void h ()
14202          {
14203            m = 0;
14204            f ();
14205            n = 0;
14206            g ();
14207          }
14208        };
14209     * In templates, all non-dependent names are now looked up and bound
14210       at definition time (while parsing the code), instead of later when
14211       the template is instantiated. For instance:
14212        void foo(int);
14213
14214        template <int> struct A {
14215          static void bar(void){
14216            foo('a');
14217          }
14218        };
14219
14220        void foo(char);
14221
14222        int main()
14223        {
14224          A<0>::bar();    // Calls foo(int), used to call foo(char).
14225        }
14226
14227     * In an explicit instantiation of a class template, you must use
14228       class or struct before the template-id:
14229        template <int N>
14230        class A {};
14231
14232        template A<0>;         // error, not accepted anymore
14233        template class A<0>;   // OK
14234     * The "named return value" and "implicit typename" extensions have
14235       been removed.
14236     * Default arguments in function types have been deprecated and will
14237       be removed.
14238     * ARM-style name-injection of friend declarations has been deprecated
14239       and will be removed. For example: struct S { friend void f(); };
14240       void g() { f(); } will not be accepted by future versions of G++;
14241       instead a declaration of "f" will need to be present outside of the
14242       scope of "S".
14243     * Covariant returns are implemented for all but varadic functions
14244       that require an adjustment.
14245     * When -pedantic is used, G++ now issues errors about spurious
14246       semicolons. For example,
14247        namespace N {}; // Invalid semicolon.
14248        void f() {}; // Invalid semicolon.
14249     * G++ no longer accepts attributes for a declarator after the
14250       initializer associated with that declarator. For example,
14251        X x(1) __attribute__((...));
14252       is no longer accepted. Instead, use:
14253        X x __attribute__((...)) (1);
14254     * Inside the scope of a template class, the name of the class itself
14255       can be treated as either a class or a template. So GCC used to
14256       accept the class name as argument of type template, and template
14257       template parameter. However this is not C++ standard compliant. Now
14258       the name is not treated as a valid template template argument
14259       unless you qualify the name by its scope. For example, the code
14260       below no longer compiles.
14261        template <template <class> class TT> class X {};
14262        template <class T> class Y {
14263          X<Y> x; // Invalid, Y is always a type template parameter.
14264        };
14265       The valid code for the above example is
14266          X< ::Y> x; // Valid.
14267       (Notice the space between < and : to prevent GCC to interpret this
14268       as a digraph for [.)
14269     * Friend declarations that refer to template specializations are
14270       rejected if the template has not already been declared. For
14271       example,
14272        template <typename T>
14273        class C {
14274          friend void f<> (C&);
14275        };
14276       is rejected. You must first declare f as a template,
14277        template <typename T>
14278        void f(T);
14279     * In case of friend declarations, every name used in the friend
14280       declaration must be accessible at the point of that declaration.
14281       Previous versions of G++ used to be less strict about this and
14282       allowed friend declarations for private class members, for example.
14283       See the ISO C++ Standard Committee's [11]defect report #209 for
14284       details.
14285     * Declaration of member functions of class templates as friends are
14286       supported. For example,
14287        template <typename T> struct A {
14288          void f();
14289        };
14290        class C {
14291          template <typename T> friend void A<T>::f();
14292        };
14293     * You must use template <> to introduce template specializations, as
14294       required by the standard. For example,
14295        template <typename T>
14296        struct S;
14297
14298        struct S<int> { };
14299       is rejected. You must write,
14300        template <> struct S<int> {};
14301     * G++ used to accept code like this,
14302        struct S {
14303          int h();
14304          void f(int i = g());
14305          int g(int i = h());
14306        };
14307       This behavior is not mandated by the standard. Now G++ issues an
14308       error about this code. To avoid the error, you must move the
14309       declaration of g before the declaration of f. The default arguments
14310       for g must be visible at the point where it is called.
14311     * The C++ ABI Section 3.3.3 specifications for the array construction
14312       routines __cxa_vec_new2 and __cxa_vec_new3 were changed to return
14313       NULL when the allocator argument returns NULL. These changes are
14314       incorporated into the libstdc++ runtime library.
14315     * Using a name introduced by a typedef in a friend declaration or in
14316       an explicit instantiation is now rejected, as specified by the ISO
14317       C++ standard.
14318        class A;
14319        typedef A B;
14320        class C {
14321          friend class B;      // error, no typedef name here
14322          friend B;            // error, friend always needs class/struct/enum
14323          friend class A;      // OK
14324        };
14325
14326        template <int> class Q {};
14327        typedef Q<0> R;
14328        template class R;      // error, no typedef name here
14329        template class Q<0>;   // OK
14330     * When allocating an array with a new expression, GCC used to allow
14331       parentheses around the type name. This is actually ill-formed and
14332       it is now rejected:
14333        int* a = new (int)[10];    // error, not accepted anymore
14334        int* a = new int[10];      // OK
14335     * When binding an rvalue of class type to a reference, the copy
14336       constructor of the class must be accessible. For instance, consider
14337       the following code:
14338        class A
14339        {
14340        public:
14341          A();
14342
14343        private:
14344          A(const A&);   // private copy ctor
14345        };
14346
14347        A makeA(void);
14348        void foo(const A&);
14349
14350        void bar(void)
14351        {
14352          foo(A());       // error, copy ctor is not accessible
14353          foo(makeA());   // error, copy ctor is not accessible
14354
14355          A a1;
14356          foo(a1);        // OK, a1 is a lvalue
14357        }
14358       This might be surprising at first sight, especially since most
14359       popular compilers do not correctly implement this rule ([12]further
14360       details).
14361     * When forming a pointer to member or a pointer to member function,
14362       access checks for class visibility (public, protected, private) are
14363       now performed using the qualifying scope of the name itself. This
14364       is better explained with an example:
14365        class A
14366        {
14367        public:
14368          void pub_func();
14369        protected:
14370          void prot_func();
14371        private:
14372          void priv_func();
14373        };
14374
14375        class B : public A
14376        {
14377        public:
14378          void foo()
14379          {
14380            &A::pub_func;   // OK, pub_func is accessible through A
14381            &A::prot_func;  // error, cannot access prot_func through A
14382            &A::priv_func;  // error, cannot access priv_func through A
14383
14384            &B::pub_func;   // OK, pub_func is accessible through B
14385            &B::prot_func;  // OK, can access prot_func through B (within B)
14386            &B::priv_func;  // error, cannot access priv_func through B
14387          }
14388        };
14389
14390    Runtime Library (libstdc++)
14391
14392     * Optimization work:
14393          + Streamlined streambuf, filebuf, separate synched with C
14394            Standard I/O streambuf.
14395          + All formatted I/O now uses cached locale information.
14396          + STL optimizations (memory/speed for list, red-black trees as
14397            used by sets and maps).
14398          + More use of GCC builtins.
14399          + String optimizations (avoid contention on
14400            increment/decrement-and-test of the reference count in the
14401            empty-string object, constructor from input_iterators
14402            speedup).
14403     * Static linkage size reductions.
14404     * Large File Support (files larger than 2 GB on 32-bit systems).
14405     * Wide character and variable encoding filebuf work (UTF-8, Unicode).
14406     * Generic character traits.
14407     * Also support wchar_t specializations on Mac OS 10.3.x, FreeBSD 5.x,
14408       Solaris 2.7 and above, AIX 5.x, Irix 6.5.
14409     * The allocator class is now standard-conformant, and two additional
14410       extension allocators have been added, mt_alloc and
14411       bitmap_allocator.
14412     * PCH support: -include bits/stdc++.h (2x compile speedup).
14413     * Rewrote __cxa_demangle with support for C++ style allocators.
14414     * New debug modes for STL containers and iterators.
14415     * Testsuite rewrite: five times as many tests, plus increasingly
14416       sophisticated tests, including I/O, MT, multi-locale, wide and
14417       narrow characters.
14418     * Use current versions of GNU "autotools" for build/configuration.
14419
14420  Objective-C
14421
14422     * The Objective-C front end has been updated to include the numerous
14423       bug fixes and enhancements previously available only in Apple's
14424       version of GCC. These include:
14425          + Structured exception (@try... @catch... @finally, @throw) and
14426            synchronization (@synchronized) support. These are accessible
14427            via the -fobjc-exceptions switch; as of this writing, they may
14428            only be used in conjunction with -fnext-runtime on Mac OS X
14429            10.3 and later. See [13]Options Controlling Objective-C
14430            Dialect for more information.
14431          + An overhaul of @encode logic. The C99 _Bool and C++ bool type
14432            may now be encoded as 'B'. In addition, the back-end/codegen
14433            dependencies have been removed.
14434          + An overhaul of message dispatch construction, ensuring that
14435            the various receiver types (and casts thereof) are handled
14436            properly, and that correct diagnostics are issued.
14437          + Support for "Zero-Link" (-fzero-link) and "Fix-and-Continue"
14438            (-freplace-objc-classes) debugging modes, currently available
14439            on Mac OS X 10.3 and later. See [14]Options Controlling
14440            Objective-C Dialect for more information.
14441          + Access to optimized runtime entry points (-fno-nil-receivers )
14442            on the assumption that message receivers are never nil. This
14443            is currently available on Mac OS X 10.3 and later. See
14444            [15]Options Controlling Objective-C Dialect for more
14445            information.
14446
14447  Java
14448
14449     * Compiling a .jar file will now cause non-.class entries to be
14450       automatically compiled as resources.
14451     * libgcj has been ported to Darwin.
14452     * Jeff Sturm has adapted Jan Hubicka's call graph optimization code
14453       to gcj.
14454     * libgcj has a new gcjlib URL type; this lets URLClassLoader load
14455       code from shared libraries.
14456     * libgcj has been much more completely merged with [16]GNU Classpath.
14457     * Class loading is now much more correct; in particular the caller's
14458       class loader is now used when that is required.
14459     * [17]Eclipse 2.x will run out of the box using gij.
14460     * Parts of java.nio have been implemented. Direct and indirect
14461       buffers work, as do fundamental file and socket operations.
14462     * java.awt has been improved, though it is still not ready for
14463       general use.
14464     * The HTTP protocol handler now uses HTTP/1.1 and can handle the POST
14465       method.
14466     * The MinGW port has matured. Enhancements include socket timeout
14467       support, thread interruption, improved Runtime.exec() handling and
14468       support for accented characters in filenames.
14469
14470  Fortran
14471
14472     * Fortran improvements are listed in the [18]Fortran documentation.
14473
14474New Targets and Target Specific Improvements
14475
14476  Alpha
14477
14478     * Several [19]built-in functions have been added such as
14479       __builtin_alpha_zap to allow utilizing the more obscure
14480       instructions of the CPU.
14481     * Parameter passing of complex arguments has changed to match the
14482       ABI. This change is incompatible with previous GCC versions, but
14483       does fix compatibility with the Tru64 compiler and several corner
14484       cases where GCC was incompatible with itself.
14485
14486  ARM
14487
14488     * Nicolas Pitre has contributed his hand-coded floating-point support
14489       code for ARM. It is both significantly smaller and faster than the
14490       existing C-based implementation, even when building applications
14491       for Thumb. The arm-elf configuration has been converted to use the
14492       new code.
14493     * Support for the Intel's iWMMXt architecture, a second generation
14494       XScale processor, has been added. Enabled at run time with the
14495       -mcpu=iwmmxt command line switch.
14496     * A new ARM target has been added: arm-wince-pe. This is similar to
14497       the arm-pe target, but it defaults to using the APCS32 ABI.
14498     * The existing ARM pipeline description has been converted to the use
14499       the [20]DFA processor pipeline model. There is not much change in
14500       code performance, but the description is now [21]easier to
14501       understand.
14502     * Support for the Cirrus EP9312 Maverick floating point co-processor
14503       added. Enabled at run time with the -mcpu=ep9312 command line
14504       switch. Note however that the multilibs to support this chip are
14505       currently disabled in gcc/config/arm/t-arm-elf, so if you want to
14506       enable their production you will have to uncomment the entries in
14507       that file.
14508
14509  H8/300
14510
14511     * Support for long long has been added.
14512     * Support for saveall attribute has been added.
14513     * Pavel Pisa contributed hand-written 32-bit-by-32-bit division code
14514       for H8/300H and H8S, which is much faster than the previous
14515       implementation.
14516     * A lot of small performance improvements.
14517
14518  IA-32/AMD64 (x86-64)
14519
14520     * Tuning for K8 (AMD Opteron/Athlon64) core is available via
14521       -march=k8 and -mcpu=k8.
14522     * Scalar SSE code generation carefully avoids reformatting penalties,
14523       hidden dependencies and minimizes the number of uops generated on
14524       both Intel and AMD CPUs.
14525     * Vector MMX and SSE operands are now passed in registers to improve
14526       performance and match the argument passing convention used by the
14527       Intel C++ Compiler. As a result it is not possible to call
14528       functions accepting vector arguments compiled by older GCC version.
14529     * Conditional jump elimination is now more aggressive on modern CPUs.
14530     * The Athlon ports has been converted to use the DFA processor
14531       pipeline description.
14532     * Optimization of indirect tail calls is now possible in a similar
14533       fashion as direct sibcall optimization.
14534     * Further small performance improvements.
14535     * -m128bit-long-double is now less buggy.
14536     * __float128 support in 64-bit compilation.
14537     * Support for data structures exceeding 2GB in 64-bit mode.
14538     * -mcpu has been renamed to -mtune.
14539
14540  IA-64
14541
14542     * Tuning code for the Itanium 2 processor has been added. The
14543       generation of code tuned for Itanium 2 (option -mtune=itanium2) is
14544       enabled by default now. To generate code tuned for Itanium 1 the
14545       option -mtune=itanium1 should be used.
14546     * [22]DFA processor pipeline descriptions for the IA-64 processors
14547       have been added. This resulted in about 3% improvement on the
14548       SPECInt2000 benchmark for Itanium 2.
14549     * Instruction bundling for the IA-64 processors has been rewritten
14550       using the DFA pipeline hazard recognizer. It resulted in about 60%
14551       compiler speedup on the SPECInt2000 C programs.
14552
14553  M32R
14554
14555     * Support for the M32R/2 processor has been added by Renesas.
14556     * Support for an M32R GNU/Linux target and PIC code generation has
14557       been added by Renesas.
14558
14559  M68000
14560
14561     * Bernardo Innocenti (Develer S.r.l.) has contributed the
14562       m68k-uclinux target, based on former work done by Paul Dale
14563       (SnapGear Inc.). Code generation for the ColdFire processors family
14564       has been enhanced and extended to support the MCF 53xx and MCF 54xx
14565       cores, integrating former work done by Peter Barada (Motorola).
14566
14567  MIPS
14568
14569    Processor-specific changes
14570
14571     * Support for the RM7000 and RM9000 processors has been added. It can
14572       be selected using the -march compiler option and should work with
14573       any MIPS I (mips-*) or MIPS III (mips64-*) configuration.
14574     * Support for revision 2 of the MIPS32 ISA has been added. It can be
14575       selected with the command-line option -march=mips32r2.
14576     * There is a new option, -mfix-sb1, to work around certain SB-1
14577       errata.
14578
14579    Configuration
14580
14581     * It is possible to customize GCC using the following configure-time
14582       options:
14583          + --with-arch, which specifies the default value of the -march
14584            option.
14585          + --with-tune, which specifies the default value of the -mtune
14586            option.
14587          + --with-abi, which specifies the default ABI.
14588          + --with-float=soft, which tells GCC to use software floating
14589            point by default.
14590          + --with-float=hard, which tells GCC to use hardware floating
14591            point by default.
14592     * A 64-bit GNU/Linux port has been added. The associated
14593       configurations are mips64-linux-gnu and mips64el-linux-gnu.
14594     * The 32-bit GNU/Linux port now supports Java.
14595     * The IRIX 6 configuration now supports the o32 ABI and will build
14596       o32 multilibs by default. This support is compatible with both
14597       binutils and the SGI tools, but note that several features,
14598       including debugging information and DWARF2 exception handling, are
14599       only available when using the GNU assembler. Use of the GNU
14600       assembler and linker (version 2.15 or above) is strongly
14601       recommended.
14602     * The IRIX 6 configuration now supports 128-bit long doubles.
14603     * There are two new RTEMS-specific configurations, mips-rtems and
14604       mipsel-rtems.
14605     * There are two new *-elf configurations, mipsisa32r2-elf and
14606       mipsisa32r2el-elf.
14607
14608    General
14609
14610     * Several [23]ABI bugs have been fixed. Unfortunately, these changes
14611       will break binary compatibility with earlier releases.
14612     * GCC can now use explicit relocation operators when generating
14613       -mabicalls code. This behavior is controlled by -mexplicit-relocs
14614       and can have several performance benefits. For example:
14615          + It allows for more optimization of GOT accesses, including
14616            better scheduling and redundancy elimination.
14617          + It allows sibling calls to be implemented as jumps.
14618          + n32 and n64 leaf functions can use a call-clobbered global
14619            pointer instead of $28.
14620          + The code to set up $gp can be removed from functions that
14621            don't need it.
14622     * A new option, -mxgot, allows the GOT to be bigger than 64k. This
14623       option is equivalent to the assembler's -xgot option and should be
14624       used instead of -Wa,-xgot.
14625     * Frame pointer elimination is now supported when generating 64-bit
14626       MIPS16 code.
14627     * Inline block moves have been optimized to take more account of
14628       alignment information.
14629     * Many internal changes have been made to the MIPS port, mostly aimed
14630       at reducing the reliance on assembler macros.
14631
14632  PowerPC
14633
14634     * GCC 3.4 releases have a number of fixes for PowerPC and PowerPC64
14635       [24]ABI incompatibilities regarding the way parameters are passed
14636       during functions calls. These changes may result in incompatibility
14637       between code compiled with GCC 3.3 and GCC 3.4.
14638
14639    PowerPC Darwin
14640
14641     * Support for shared/dylib gcc libraries has been added. It is
14642       enabled by default on powerpc-apple-darwin7.0.0 and up.
14643     * Libgcj is enabled by default. On systems older than
14644       powerpc-apple-darwin7.0.0 you need to install dlcompat.
14645     * 128-bit IBM extended precision format support added for long
14646       double.
14647
14648    PowerPC64 GNU/Linux
14649
14650     * By default, PowerPC64 GNU/Linux now uses natural alignment of
14651       structure elements. The old four byte alignment for double, with
14652       special rules for a struct starting with a double, can be chosen
14653       with -malign-power. This change may result in incompatibility
14654       between code compiled with GCC 3.3 and GCC 3.4.
14655     * -mabi=altivec is now the default rather than -mabi=no-altivec.
14656     * 128-bit IBM extended precision format support added for long
14657       double.
14658
14659  S/390 and zSeries
14660
14661     * New command-line options allow to specify the intended execution
14662       environment for generated code:
14663          + -mesa/-mzarch allows to specify whether to generate code
14664            running in ESA/390 mode or in z/Architecture mode (this is
14665            applicable to 31-bit code only).
14666          + -march allows to specify a minimum processor architecture
14667            level (g5, g6, z900, or z990).
14668          + -mtune allows to specify which processor to tune for.
14669     * It is possible to customize GCC using the following configure-time
14670       options:
14671          + --with-mode, which specifies whether to default to assuming
14672            ESA/390 or z/Architecture mode.
14673          + --with-arch, which specifies the default value of the -march
14674            option.
14675          + --with-tune, which specifies the default value of the -mtune
14676            option.
14677     * Support for the z990 processor has been added, and can be selected
14678       using -march=z990 or -mtune=z990. This includes instruction
14679       scheduling tuned for the superscalar instruction pipeline of the
14680       z990 processor as well as support for all new instructions provided
14681       by the long-displacement facility.
14682     * Support to generate 31-bit code optimized for zSeries processors
14683       (running in ESA/390 or in z/Architecture mode) has been added. This
14684       can be selected using -march=z900 and -mzarch respectively.
14685     * Instruction scheduling for the z900 and z990 processors now uses
14686       the DFA pipeline hazard recognizer.
14687     * GCC no longer generates code to maintain a stack backchain,
14688       previously used to generate stack backtraces for debugging
14689       purposes. As replacement that does not incur runtime overhead,
14690       DWARF-2 call frame information is provided by GCC; this is
14691       supported by GDB 6.1. The old behavior can be restored using the
14692       -mbackchain option.
14693     * The stack frame size of functions may now exceed 2 GB in 64-bit
14694       code.
14695     * A port for the 64-bit IBM TPF operating system has been added; the
14696       configuration is s390x-ibm-tpf. This configuration is supported as
14697       cross-compilation target only.
14698     * Various changes to improve the generated code have been
14699       implemented, including:
14700          + GCC now uses the MULTIPLY AND ADD and MULTIPLY AND SUBTRACT
14701            instructions to significantly speed up many floating-point
14702            applications.
14703          + GCC now uses the ADD LOGICAL WITH CARRY and SUBTRACT LOGICAL
14704            WITH BORROW instructions to speed up long long arithmetic.
14705          + GCC now uses the SEARCH STRING instruction to implement
14706            strlen().
14707          + In many cases, function call overhead for 31-bit code has been
14708            reduced by placing the literal pool after the function code
14709            instead of after the function prolog.
14710          + Register 14 is no longer reserved in 64-bit code.
14711          + Handling of global register variables has been improved.
14712
14713  SPARC
14714
14715     * The option -mflat is deprecated.
14716     * Support for large (> 2GB) frames has been added to the 64-bit port.
14717     * Several [25]ABI bugs have been fixed. Unfortunately, these changes
14718       will break binary compatibility with earlier releases.
14719     * The default debugging format has been switched from STABS to
14720       DWARF-2 for 32-bit code on Solaris 7 and later. DWARF-2 is already
14721       the default debugging format for 64-bit code on Solaris.
14722
14723  SuperH
14724
14725     * Support for the SH2E processor has been added. Enabled at run time
14726       with the -m2e command line switch, or at configure time by
14727       specifying sh2e as the machine part of the target triple.
14728
14729  V850
14730
14731     * Support for the Mitsubishi V850E1 processor has been added. This is
14732       a variant of the V850E processor with some additional debugging
14733       instructions.
14734
14735  Xtensa
14736
14737     * Several ABI bugs have been fixed. Unfortunately, these changes
14738       break binary compatibility with earlier releases.
14739          + For big-endian processors, the padding of aggregate return
14740            values larger than a word has changed. If the size of an
14741            aggregate return value is not a multiple of 32 bits, previous
14742            versions of GCC inserted padding in the most-significant bytes
14743            of the first return value register. Aggregates larger than a
14744            word are now padded in the least-significant bytes of the last
14745            return value register used. Aggregates smaller than a word are
14746            still padded in the most-significant bytes. The return value
14747            padding has not changed for little-endian processors.
14748          + Function arguments with 16-byte alignment are now properly
14749            aligned.
14750          + The implementation of the va_list type has changed. A va_list
14751            value created by va_start from a previous release cannot be
14752            used with va_arg from this release, or vice versa.
14753     * More processor configuration options for Xtensa processors are
14754       supported:
14755          + the ABS instruction is now optional;
14756          + the ADDX* and SUBX* instructions are now optional;
14757          + an experimental CONST16 instruction can be used to synthesize
14758            constants instead of loading them from constant pools.
14759       These and other Xtensa processor configuration options can no
14760       longer be enabled or disabled by command-line options; the
14761       processor configuration must be specified by the xtensa-config.h
14762       header file when building GCC. Additionally, the
14763       -mno-serialize-volatile option is no longer supported.
14764
14765Obsolete Systems
14766
14767   Support for a number of older systems has been declared obsolete in GCC
14768   3.4. Unless there is activity to revive them, the next release of GCC
14769   will have their sources permanently removed.
14770
14771   All configurations of the following processor architectures have been
14772   declared obsolete:
14773     * Mitsubishi D30V, d30v-*
14774     * AT&T DSP1600 and DSP1610, dsp16xx-*
14775     * Intel 80960, i960
14776
14777   Also, some individual systems have been obsoleted:
14778     * ARM Family
14779          + Support for generating code for operation in APCS/26 mode
14780            (-mapcs-26).
14781     * IBM ESA/390
14782          + "Bigfoot" port, i370-*. (The other port, s390-*, is actively
14783            maintained and supported.)
14784     * Intel 386 family
14785          + MOSS, i?86-moss-msdos and i?86-*-moss*
14786          + NCR 3000 running System V r.4, i?86-ncr-sysv4*
14787          + FreeBSD with a.out object format, i?86-*-freebsd*aout* and
14788            i?86-*-freebsd2*
14789          + GNU/Linux with a.out object format, i?86-linux*aout*
14790          + GNU/Linux with libc5, a.k.a. glibc1, i?86-linux*libc1*
14791          + Interix versions before Interix 3, i?86-*-interix
14792          + Mach microkernel, i?86-mach*
14793          + SCO UnixWare with UDK, i?86-*-udk*
14794          + Generic System V releases 1, 2, and 3, i?86-*-sysv[123]*
14795          + VSTa microkernel, i386-*-vsta
14796     * Motorola M68000 family
14797          + HPUX, m68k-hp-hpux* and m68000-hp-hpux*
14798          + NetBSD with a.out object format (before NetBSD 1.4),
14799            m68k-*-*-netbsd* except m68k-*-*-netbsdelf*
14800          + Generic System V r.4, m68k-*-sysv4*
14801     * VAX
14802          + Generic VAX, vax-*-* (This is generic VAX only; we have not
14803            obsoleted any VAX triples for specific operating systems.)
14804
14805Documentation improvements
14806
14807Other significant improvements
14808
14809     * The build system has undergone several significant cleanups.
14810       Subdirectories will only be configured if they are being built, and
14811       all subdirectory configures are run from the make command. The top
14812       level has been autoconfiscated.
14813     * Building GCC no longer writes to its source directory. This should
14814       help those wishing to share a read-only source directory over NFS
14815       or build from a CD. The exceptions to this feature are if you
14816       configure with either --enable-maintainer-mode or
14817       --enable-generated-files-in-srcdir.
14818     * The -W warning option has been renamed to -Wextra, which is more
14819       easily understood. The older spelling will be retained for
14820       backwards compatibility.
14821     * Substantial improvements in compile time have been made,
14822       particularly for non-optimizing compilations.
14823     __________________________________________________________________
14824
14825GCC 3.4.0
14826
14827  Bug Fixes
14828
14829   A vast number of bugs have been fixed in 3.4.0, too many to publish a
14830   complete list here. [26]Follow this link to query the Bugzilla database
14831   for the list of over 900 bugs fixed in 3.4.0. This is the list of all
14832   bugs marked as resolved and fixed in 3.4.0 that are not flagged as 3.4
14833   regressions.
14834     __________________________________________________________________
14835
14836GCC 3.4.1
14837
14838  Bug Fixes
14839
14840   This section lists the problem reports (PRs) from GCC's bug tracking
14841   system that are known to be fixed in the 3.4.1 release. This list might
14842   not be complete (that is, it is possible that some PRs that have been
14843   fixed are not listed here).
14844
14845    Bootstrap failures
14846
14847     * [27]10129 Ada bootstrap fails on PPC-Darwin - invalid assembler
14848       emitted - PIC related
14849     * [28]14576 [ARM] ICE in libiberty when building gcc-3.4 for arm-elf
14850     * [29]14760 A bug in configure.in prevents using both
14851       --program-suffix and --program-prefix
14852     * [30]14671 [hppa64] bootstrap fails: ICE in
14853       save_call_clobbered_regs, in caller_save.c
14854     * [31]15093 [alpha][Java] make bootstrap fails to configure libffi on
14855       Alpha
14856     * [32]15178 Solaris 9/x86 fails linking after stage 3
14857
14858    Multi-platform internal compiler errors (ICEs)
14859
14860     * [33]12753 (preprocessor) Memory corruption in preprocessor on bad
14861       input
14862     * [34]13985 ICE in gcc.c-torture/compile/930621-1.c
14863     * [35]14810 (c++) tree check failures with invalid code involving
14864       templates
14865     * [36]14883 (c++) ICE on invalid code, in cp_parser_lookup_name, in
14866       cp/parser.c
14867     * [37]15044 (c++) ICE on syntax error, template header
14868     * [38]15057 (c++) Compiling of conditional value throw constructs
14869       cause a segmentation violation
14870     * [39]15064 (c++) typeid of template parameter gives ICE
14871     * [40]15142 (c++) ICE when passing a string where a char* is expected
14872       in a throw statement
14873     * [41]15159 ICE in rtl_verify_flow_info_1
14874     * [42]15165 (c++) ICE in instantiate_template
14875     * [43]15193 Unary minus using pointer to V4SF vector causes
14876       -fforce-mem to exhaust all memory
14877     * [44]15209 (c++) Runs out of memory with packed structs
14878     * [45]15227 (c++) Trouble with invalid function definition
14879     * [46]15285 (c++) instantiate_type ICE when forming pointer to
14880       template function
14881     * [47]15299 (c++) ICE in resolve_overloaded_unification
14882     * [48]15329 (c++) ICE on constructor of member template
14883     * [49]15550 ICE in extract_insn, in recog.c
14884     * [50]15554 (c++) ICE in tsubst_copy, in cp/pt.c
14885     * [51]15640 (c++) ICE on invalid code in arg_assoc, in
14886       cp/name-lookup.c
14887     * [52]15666 [unit-at-a-time] Gcc abort on valid code
14888     * [53]15696 (c++) ICE with bad pointer-to-member code
14889     * [54]15701 (c++) ICE with friends and template template parameter
14890     * [55]15761 ICE in do_SUBST, in combine.c
14891     * [56]15829 (c++) ICE on Botan-1.3.13 due to -funroll-loops
14892
14893    Ada
14894
14895     * [57]14538 All RTEMS targets broken for gnat
14896
14897    C front end
14898
14899     * [58]12391 missing warning about assigning to an incomplete type
14900     * [59]14649 atan(1.0) should not be a constant expression
14901     * [60]15004 [unit-at-a-time] no warning for unused paramater in
14902       static function
14903     * [61]15749 --pedantic-errors behaves differently from --pedantic
14904       with C-compiler on GNU/Linux
14905
14906    C++ compiler and library
14907
14908     * [62]10646 non-const reference is incorrectly matched in a "const T"
14909       partial specialization
14910     * [63]12077 wcin.rdbuf()->in_avail() return value too high
14911     * [64]13598 enc_filebuf doesn't work
14912     * [65]14211 const_cast returns lvalue but should be rvalue
14913     * [66]14220 num_put::do_put() undesired float/double behavior
14914     * [67]14245 problem with user-defined allocators in std::basic_string
14915     * [68]14340 libstdc++ Debug mode: failure to convert iterator to
14916       const_iterator
14917     * [69]14600 __gnu_cxx::stdio_sync_filebuf should expose internal
14918       FILE*
14919     * [70]14668 no warning anymore for reevaluation of declaration
14920     * [71]14775 LFS (large file support) tests missing
14921     * [72]14821 Duplicate namespace alias declaration should not conflict
14922     * [73]14930 Friend declaration ignored
14923     * [74]14932 cannot use offsetof to get offsets of array elements in
14924       g++ 3.4.0
14925     * [75]14950 [non unit-at-a-time] always_inline does not mix with
14926       templates and -O0
14927     * [76]14962 g++ ignores #pragma redefine_extname
14928     * [77]14975 Segfault on low-level write error during imbue
14929     * [78]15002 Linewise stream input is unusably slow (std::string slow)
14930     * [79]15025 compiler accepts redeclaration of template as
14931       non-template
14932     * [80]15046 [arm] Math functions misdetected by cross configuration
14933     * [81]15069 a bit test on a variable of enum type is miscompiled
14934     * [82]15074 g++ -lsupc++ still links against libstdc++
14935     * [83]15083 spurious "statement has no effect" warning
14936     * [84]15096 parse error with templates and pointer to const member
14937     * [85]15287 combination of operator[] and operator .* fails in
14938       templates
14939     * [86]15317 __attribute__ unused in first parameter of constructor
14940       gives error
14941     * [87]15337 sizeof on incomplete type diagnostic
14942     * [88]15361 bitset<>::_Find_next fails
14943     * [89]15412 _GLIBCXX_ symbols symbols defined and used in different
14944       namespaces
14945     * [90]15427 valid code results in incomplete type error
14946     * [91]15471 Incorrect member pointer offsets in anonymous
14947       structs/unions
14948     * [92]15503 nested template problem
14949     * [93]15507 compiler hangs while laying out union
14950     * [94]15542 operator & and template definitions
14951     * [95]15565 SLES9: leading + sign for unsigned int with showpos
14952     * [96]15625 friend defined inside a template fails to find static
14953       function
14954     * [97]15629 Function templates, overloads, and friend name injection
14955     * [98]15742 'noreturn' attribute ignored in method of template
14956       functions.
14957     * [99]15775 Allocator::pointer consistently ignored
14958     * [100]15821 Duplicate namespace alias within namespace rejected
14959     * [101]15862 'enum yn' fails (confict with undeclared builtin)
14960     * [102]15875 rejects pointer to member in template
14961     * [103]15877 valid code using templates and anonymous enums is
14962       rejected
14963     * [104]15947 Puzzling error message for wrong destructor declaration
14964       in template class
14965     * [105]16020 cannot copy __gnu_debug::bitset
14966     * [106]16154 input iterator concept too restrictive
14967     * [107]16174 deducing top-level consts
14968
14969    Java
14970
14971     * [108]14315 Java compiler is not parallel make safe
14972
14973    Fortran
14974
14975     * [109]15151 [g77] incorrect logical i/o in 64-bit mode
14976
14977    Objective-C
14978
14979     * [110]7993 private variables cannot be shadowed in subclasses
14980
14981    Optimization bugs
14982
14983     * [111]15228 useless copies of floating point operands
14984     * [112]15345 [non-unit-at-a-time] unreferenced nested inline
14985       functions not optimized away
14986     * [113]15945 Incorrect floating point optimization
14987     * [114]15526 ftrapv aborts on 0 * (-1)
14988     * [115]14690 Miscompiled POOMA tests
14989     * [116]15112 GCC generates code to write to unchanging memory
14990
14991    Preprocessor
14992
14993     * [117]15067 Minor glitch in the source of cpp
14994
14995    Main driver program bugs
14996
14997     * [118]1963 collect2 interprets -oldstyle_liblookup as -o
14998       ldstyle_liblookup
14999
15000    x86-specific (Intel/AMD)
15001
15002     * [119]15717 Error: can't resolve `L0' {*ABS* section} - `xx' {*UND*
15003       section}
15004
15005    HPPA-specific
15006
15007     * [120]14782 GCC produces an unaligned data access at -O2
15008     * [121]14828 FAIL: gcc.c-torture/execute/20030408-1.c execution, -O2
15009     * [122]15202 ICE in reload_cse_simplify_operands, in postreload.c
15010
15011    IA64-specific
15012
15013     * [123]14610 __float80 constants incorrectly emitted
15014     * [124]14813 init_array sections are initialized in the wrong order
15015     * [125]14857 GCC segfault on duplicated asm statement
15016     * [126]15598 Gcc 3.4 ICE on valid code
15017     * [127]15653 Gcc 3.4 ICE on valid code
15018
15019    MIPS-specific
15020
15021     * [128]15189 wrong filling of delay slot with -march=mips1 -G0
15022       -mno-split-addresses -mno-explicit-relocs
15023     * [129]15331 Assembler error building gnatlib on IRIX 6.5 with GNU as
15024       2.14.91
15025     * [130]16144 Bogus reference to __divdf3 when -O1
15026     * [131]16176 Miscompilation of unaligned data in MIPS backend
15027
15028    PowerPC-specific
15029
15030     * [132]11591 ICE in gcc.dg/altivec-5.c
15031     * [133]12028 powerpc-eabispe produces bad sCOND operation
15032     * [134]14478 rs6000 geu/ltu patterns generate incorrect code
15033     * [135]14567 long double and va_arg complex args
15034     * [136]14715 Altivec stack layout may overlap gpr save with stack
15035       temps
15036     * [137]14902 (libstdc++) Stream checking functions fail when -pthread
15037       option is used.
15038     * [138]14924 Compiler ICE on valid code
15039     * [139]14960 -maltivec affects vector return with -mabi=no-altivec
15040     * [140]15106 vector varargs failure passing from altivec to
15041       non-altivec code for -m32
15042     * [141]16026 ICE in function.c:4804, assign_parms, when -mpowerpc64 &
15043       half-word operation
15044     * [142]15191 -maltivec -mabi=no-altivec results in mis-aligned lvx
15045       and stvx
15046     * [143]15662 Segmentation fault when an exception is thrown - even if
15047       try and catch are specified
15048
15049    s390-specific
15050
15051     * [144]15054 Bad code due to overlapping stack temporaries
15052
15053    SPARC-specific
15054
15055     * [145]15783 ICE with union assignment in 64-bit mode
15056     * [146]15626 GCC 3.4 emits "ld: warning: relocation error:
15057       R_SPARC_UA32"
15058
15059    x86-64-specific
15060
15061     * [147]14326 boehm-gc hardcodes to 3DNow! prefetch for x86_64
15062     * [148]14723 Backported -march=nocona from mainline
15063     * [149]15290 __float128 failed to pass to function properly
15064
15065    Cygwin/Mingw32-specific
15066
15067     * [150]15250 Option -mms-bitfields support on GCC 3.4 is not
15068       conformant to MS layout
15069     * [151]15551 -mtune=pentium4 -O2 with sjlj EH breaks stack probe
15070       worker on windows32 targets
15071
15072    Bugs specific to embedded processors
15073
15074     * [152]8309 [m68k] -m5200 produces erroneous SImode set of short
15075       varaible on stack
15076     * [153]13250 [SH] Gcc code for rotation clobbers the register, but
15077       gcc continues to use the register as if it was not clobbered
15078     * [154]13803 [coldfire] movqi operand constraints too restrictivefor
15079       TARGET_COLDFIRE
15080     * [155]14093 [SH] ICE for code when using -mhitachi option in SH
15081     * [156]14457 [m6811hc] ICE with simple c++ source
15082     * [157]14542 [m6811hc] ICE on simple source
15083     * [158]15100 [SH] cc1plus got hang-up on
15084       libstdc++-v3/testsuite/abi_check.cc
15085     * [159]15296 [CRIS] Delayed branch scheduling causing invalid code on
15086       cris-*
15087     * [160]15396 [SH] ICE with -O2 -fPIC
15088     * [161]15782 [coldfire] m68k_output_mi_thunk emits wrong code for
15089       ColdFire
15090
15091    Testsuite problems (compiler not affected)
15092
15093     * [162]11610 libstdc++ testcases 27_io/* don't work properly remotely
15094     * [163]15488 (libstdc++) possibly insufficient file permissions for
15095       executing test suite
15096     * [164]15489 (libstdc++) testsuite_files determined incorrectly
15097
15098    Documentation bugs
15099
15100     * [165]13928 (libstdc++) no whatis info in some man pages generated
15101       by doxygen
15102     * [166]14150 Ada documentation out of date
15103     * [167]14949 (c++) Need to document method visibility changes
15104     * [168]15123 libstdc++-doc: Allocators.3 manpage is empty
15105     __________________________________________________________________
15106
15107GCC 3.4.2
15108
15109  Bug Fixes
15110
15111   This section lists the problem reports (PRs) from GCC's bug tracking
15112   system that are known to be fixed in the 3.4.2 release. This list might
15113   not be complete (that is, it is possible that some PRs that have been
15114   fixed are not listed here).
15115
15116    Bootstrap failures and issues
15117
15118     * [169]16469 [mips-sgi-irix5.3] bootstrap fails in
15119       libstdc++-v3/testsuite
15120     * [170]16344 [hppa-linux-gnu] libstdc++'s PCH built by
15121       profiledbootstrap does not work with the built compiler
15122     * [171]16842 [Solaris/x86] mkheaders can not find mkheaders.conf
15123
15124    Multi-platform internal compiler errors (ICEs)
15125
15126     * [172]12608 (c++) ICE: expected class 't', have 'x' (error_mark) in
15127       cp_parser_class_specifier, in cp/parser.c
15128     * [173]14492 ICE in loc_descriptor_from_tree, in dwarf2out.c
15129     * [174]15461 (c++) ICE due to NRV and inlining
15130     * [175]15890 (c++) ICE in c_expand_expr, in c-common.c
15131     * [176]16180 ICE: segmentation fault in RTL optimization
15132     * [177]16224 (c++) ICE in write_unscoped_name (template/namespace)
15133     * [178]16408 ICE: in delete_insn, in cfgrtl.c
15134     * [179]16529 (c++) ICE for: namespace-alias shall not be declared as
15135       the name of any other entity
15136     * [180]16698 (c++) ICE with exceptions and declaration of __cxa_throw
15137     * [181]16706 (c++) ICE in finish_member_declaration, in
15138       cp/semantics.c
15139     * [182]16810 (c++) Legal C++ program with cast gives ICE in
15140       build_ptrmemfunc
15141     * [183]16851 (c++) ICE when throwing a comma expression
15142     * [184]16870 (c++) Boost.Spirit causes ICE in tsubst, in cp/pt.c
15143     * [185]16904 (c++) ICE in finish_class_member_access_expr, in
15144       cp/typeck.c
15145     * [186]16905 (c++) ICE (segfault) with exceptions
15146     * [187]16964 (c++) ICE in cp_parser_class_specifier due to
15147       redefinition
15148     * [188]17068 (c++) ICE: tree check: expected class 'd', have 'x'
15149       (identifier_node) in dependent_template_p, in cp/pt.c
15150
15151    Preprocessor bugs
15152
15153     * [189]16366 Preprocessor option -remap causes memory corruption
15154
15155    Optimization
15156
15157     * [190]15345 unreferenced nested inline functions not optimized away
15158     * [191]16590 Incorrect execution when compiling with -O2
15159     * [192]16693 Bitwise AND is lost when used within a cast to an enum
15160       of the same precision
15161     * [193]17078 Jump into if(0) substatement fails
15162
15163    Problems in generated debug information
15164
15165     * [194]13956 incorrect stabs for nested local variables
15166
15167    C front end bugs
15168
15169     * [195]16684 GCC should not warn about redundant redeclarations of
15170       built-ins
15171
15172    C++ compiler and library
15173
15174     * [196]12658 Thread safety problems in locale::global() and
15175       locale::locale()
15176     * [197]13092 g++ accepts invalid pointer-to-member conversion
15177     * [198]15320 Excessive memory consumption
15178     * [199]16246 Incorrect template argument deduction
15179     * [200]16273 Memory exhausted when using nested classes and virtual
15180       functions
15181     * [201]16401 ostringstream in gcc 3.4.x very slow for big data
15182     * [202]16411 undefined reference to
15183       __gnu_cxx::stdio_sync_filebuf<char, std::char_traits<char>
15184       >::file()
15185     * [203]16489 G++ incorrectly rejects use of a null constant integral
15186       expression as a null constant pointer
15187     * [204]16618 offsetof fails with constant member
15188     * [205]16637 syntax error reported for valid input code
15189     * [206]16717 __attribute__((constructor)) broken in C++
15190     * [207]16813 compiler error in DEBUG version of range insertion
15191       std::map::insert
15192     * [208]16853 pointer-to-member initialization from incompatible one
15193       accepted
15194     * [209]16889 ambiguity is not detected
15195     * [210]16959 Segmentation fault in ios_base::sync_with_stdio
15196
15197    Java compiler and library
15198
15199     * [211]7587 direct threaded interpreter not thread-safe
15200     * [212]16473 ServerSocket accept() leaks file descriptors
15201     * [213]16478 Hash synchronization deadlock with finalizers
15202
15203    Alpha-specific
15204
15205     * [214]10695 ICE in dwarf2out_frame_debug_expr, in dwarf2out.c
15206     * [215]16974 could not split insn (ice in final_scan_insn, in
15207       final.c)
15208
15209    x86-specific
15210
15211     * [216]16298 ICE in output_operand
15212     * [217]17113 ICE with SSE2 intrinsics
15213
15214    x86-64 specific
15215
15216     * [218]14697 libstdc++ couldn't find 32bit libgcc_s
15217
15218    MIPS-specific
15219
15220     * [219]15869 [mips64] No NOP after LW (with -mips1 -O0)
15221     * [220]16325 [mips64] value profiling clobbers gp on mips
15222     * [221]16357 [mipsisa64-elf] ICE copying 7 bytes between extern
15223       char[]s
15224     * [222]16380 [mips64] Use of uninitialised register after dbra
15225       conversion
15226     * [223]16407 [mips64] Unaligned access to local variables
15227     * [224]16643 [mips64] verify_local_live_at_start ICE after
15228       crossjumping & cfgcleanup
15229
15230    ARM-specific
15231
15232     * [225]15927 THUMB -O2: strength-reduced iteration variable ends up
15233       off by 1
15234     * [226]15948 THUMB: ICE with non-commutative cbranch
15235     * [227]17019 THUMB: bad switch statement in md code for
15236       addsi3_cbranch_scratch
15237
15238    IA64-specific
15239
15240     * [228]16130 ICE on valid code: in bundling, in config/ia64/ia64.c
15241       (-mtune=merced)
15242     * [229]16142 ICE on valid code: in bundling, in config/ia64/ia64.c
15243       (-mtune=itanium)
15244     * [230]16278 Gcc failed to build Linux kernel with -mtune=merced
15245     * [231]16414 ICE on valid code: typo in comparison of asm_noperands
15246       result
15247     * [232]16445 ICE on valid code: don't count ignored insns
15248     * [233]16490 ICE (segfault) while compiling with -fprofile-use
15249     * [234]16683 ia64 does not honor SUBTARGET_EXTRA_SPECS
15250
15251    PowerPC-specific
15252
15253     * [235]16195 (ppc64): Miscompilation of GCC 3.3.x by 3.4.x
15254     * [236]16239 ICE on ppc64 (mozilla 1.7 compile, -O1 -fno-exceptions
15255       issue)
15256
15257    SPARC-specific
15258
15259     * [237]16199 ICE while compiling apache 2.0.49
15260     * [238]16416 -m64 doesn't imply -mcpu=v9 anymore
15261     * [239]16430 ICE when returning non-C aggregates larger than 16 bytes
15262
15263    Bugs specific to embedded processors
15264
15265     * [240]16379 [m32r] can't output large model function call of memcpy
15266     * [241]17093 [m32r] ICE with -msdata=use -O0
15267     * [242]17119 [m32r] ICE at switch case 0x8000
15268
15269    DJGPP-specific
15270
15271     * [243]15928 libstdc++ in 3.4.x doesn't cross-compile for djgpp
15272
15273    Alpha Tru64-specific
15274
15275     * [244]16210 libstdc++ gratuitously omits "long long" I/O
15276
15277    Testsuite, documentation issues (compiler is not affected):
15278
15279     * [245]15488 (libstdc++) possibly insufficient file permissions for
15280       executing test suite
15281     * [246]16250 ada/doctools runs makeinfo even in release tarball
15282     __________________________________________________________________
15283
15284GCC 3.4.3
15285
15286   This is the [247]list of problem reports (PRs) from GCC's bug tracking
15287   system that are known to be fixed in the 3.4.3 release. This list might
15288   not be complete (that is, it is possible that some PRs that have been
15289   fixed are not listed here).
15290
15291    Bootstrap failures
15292
15293     * [248]17369 [ia64] Bootstrap failure with binutils-2.15.90.0.1.1
15294     * [249]17850 [arm-elf] bootstrap failure - libstdc++ uses strtold
15295       when undeclared
15296
15297    Internal compiler errors (ICEs) affecting multiple platforms
15298
15299     * [250]13948 (java) GCJ segmentation fault while compiling GL4Java
15300       .class files
15301     * [251]14492 ICE in loc_descriptor_from_tree, in dwarf2out.c
15302     * [252]16301 (c++) ICE when "strong" attribute is attached to a using
15303       directive
15304     * [253]16566 ICE with flexible arrays
15305     * [254]17023 ICE with nested functions in parameter declaration
15306     * [255]17027 ICE with noreturn function in loop at -O2
15307     * [256]17524 ICE in grokdeclarator, in cp/decl.c
15308     * [257]17826 (c++) ICE in cp_tree_equal
15309
15310    C and optimization bugs
15311
15312     * [258]15526 -ftrapv aborts on 0 * (-1)
15313     * [259]16999 #ident stopped working
15314     * [260]17503 quadratic behaviour in invalid_mode_change_p
15315     * [261]17581 Long long arithmetic fails inside a switch/case
15316       statement when compiled with -O2
15317     * [262]18129 -fwritable-strings doesn't work
15318
15319    C++ compiler and library bugs
15320
15321     * [263]10975 incorrect initial ostringstream::tellp()
15322     * [264]11722 Unbuffered filebuf::sgetn is slow
15323     * [265]14534 Unrecognizing static function as a template parameter
15324       when its return value is also templated
15325     * [266]15172 Copy constructor optimization in aggregate
15326       initialization
15327     * [267]15786 Bad error message for frequently occuring error.
15328     * [268]16162 Rejects valid member-template-definition
15329     * [269]16612 empty basic_strings can't live in shared memory
15330     * [270]16715 std::basic_iostream is instantiated when used, even
15331       though instantiations are already contained in libstdc++
15332     * [271]16848 code in /ext/demangle.h appears broken
15333     * [272]17132 GCC fails to eliminate function template specialization
15334       when argument deduction fails
15335     * [273]17259 One more _S_leaf incorrectly qualified with _RopeRep::
15336       in ropeimpl.h
15337     * [274]17327 use of `enumeral_type' in template type unification
15338     * [275]17393 "unused variable '._0'" warning with -Wall
15339     * [276]17501 Confusion with member templates
15340     * [277]17537 g++ not passing -lstdc++ to linker when all command line
15341       arguments are libraries
15342     * [278]17585 usage of unqualified name of static member from within
15343       class not allowed
15344     * [279]17821 Poor diagnostic for using "." instead of "->"
15345     * [280]17829 wrong error: call of overloaded function is ambiguous
15346     * [281]17851 Misleading diagnostic for invalid function declarations
15347       with undeclared types
15348     * [282]17976 Destructor is called twice
15349     * [283]18020 rejects valid definition of enum value in template
15350     * [284]18093 bogus conflict in namespace aliasing
15351     * [285]18140 C++ parser bug when using >> in templates
15352
15353    Fortran
15354
15355     * [286]17541 data statements with double precision constants fail
15356
15357    x86-specific
15358
15359     * [287]17853 -O2 ICE for MMX testcase
15360
15361    SPARC-specific
15362
15363     * [288]17245 ICE compiling gsl-1.5 statistics/lag1.c
15364
15365    Darwin-specific
15366
15367     * [289]17167 FATAL:Symbol L_foo$stub already defined.
15368
15369    AIX-specific
15370
15371     * [290]17277 could not catch an exception when specified -maix64
15372
15373    Solaris-specific
15374
15375     * [291]17505 <cmath> calls acosf(), ceilf(), and other functions
15376       missing from system libraries
15377
15378    HP/UX specific:
15379
15380     * [292]17684 /usr/ccs/bin/ld: Can't create libgcc_s.sl
15381
15382    ARM-specific
15383
15384     * [293]17384 ICE with mode attribute on structures
15385
15386    MIPS-specific
15387
15388     * [294]17770 No NOP after LWL with -mips1
15389
15390    Other embedded target specific
15391
15392     * [295]11476 [arc-elf] gcc ICE on newlib's vfprintf.c
15393     * [296]14064 [avr-elf] -fdata-sections triggers ICE
15394     * [297]14678 [m68hc11-elf] gcc ICE
15395     * [298]15583 [powerpc-rtems] powerpc-rtems lacks __USE_INIT_FINI__
15396     * [299]15790 [i686-coff] Alignment error building gcc with i686-coff
15397       target
15398     * [300]15886 [SH] Miscompilation with -O2 -fPIC
15399     * [301]16884 [avr-elf] [fweb related] bug while initializing
15400       variables
15401
15402    Bugs relating to debugger support
15403
15404     * [302]13841 missing debug info for _Complex function arguments
15405     * [303]15860 [big-endian targets] No DW_AT_location debug info is
15406       emitted for formal arguments to a function that uses "register"
15407       qualifiers
15408
15409    Testsuite issues (compiler not affected)
15410
15411     * [304]17465 Testsuite in libffi overrides LD_LIBRARY_PATH
15412     * [305]17469 Testsuite in libstdc++ overrides LD_LIBRARY_PATH
15413     * [306]18138 [mips-sgi-irix6.5] libgcc_s.so.1 not found by 64-bit
15414       testsuite
15415
15416    Documentation
15417
15418     * [307]15498 typo in gcc manual: non-existing locale example en_UK,
15419       should be en_GB
15420     * [308]15747 [mips-sgi-irix5.3] /bin/sh hangs during bootstrap:
15421       document broken shell
15422     * [309]16406 USE_LD_AS_NEEDED undocumented
15423     __________________________________________________________________
15424
15425GCC 3.4.4
15426
15427   This is the [310]list of problem reports (PRs) from GCC's bug tracking
15428   system that are known to be fixed in the 3.4.4 release. This list might
15429   not be complete (that is, it is possible that some PRs that have been
15430   fixed are not listed here).
15431     __________________________________________________________________
15432
15433GCC 3.4.5
15434
15435   This is the [311]list of problem reports (PRs) from GCC's bug tracking
15436   system that are known to be fixed in the 3.4.5 release. This list might
15437   not be complete (that is, it is possible that some PRs that have been
15438   fixed are not listed here).
15439
15440    Bootstrap issues
15441
15442     * [312]24688 sco_math fixincl breaks math.h
15443
15444    C compiler bugs
15445
15446     * [313]17188 struct Foo { } redefinition
15447     * [314]20187 wrong code for ((unsigned char)(unsigned long
15448       long)((a?a:1)&(a*b)))?0:1)
15449     * [315]21873 infinite warning loop on bad array initializer
15450     * [316]21899 enum definition accepts values to be overriden
15451     * [317]22061 ICE in find_function_data, in function.c
15452     * [318]22308 Failure to diagnose violation of constraint 6.516p2
15453     * [319]22458 ICE on missing brace
15454     * [320]22589 ICE casting to long long
15455     * [321]24101 Segfault with preprocessed source
15456
15457    C++ compiler and library bugs
15458
15459     * [322]10611 operations on vector mode not recognized in C++
15460     * [323]13377 unexpected behavior of namespace usage directive
15461     * [324]16002 Strange error message with new parser
15462     * [325]17413 local classes as template argument
15463     * [326]17609 spurious error message after using keyword
15464     * [327]17618 ICE in cp_convert_to_pointer, in cp/cvt.c
15465     * [328]18124 ICE with invalid template template parameter
15466     * [329]18155 typedef in template declaration not rejected
15467     * [330]18177 ICE with const_cast for undeclared variable
15468     * [331]18368 C++ error message regression
15469     * [332]16378 ICE when returning a copy of a packed member
15470     * [333]18466 int ::i; accepted
15471     * [334]18512 ICE on invalid usage of template base class
15472     * [335]18454 ICE when returning undefined type
15473     * [336]18738 typename not allowed with non-dependent qualified name
15474     * [337]18803 rejects access to operator() in template
15475     * [338]19004 ICE in uses_template_parms, in cp/pt.c
15476     * [339]19208 Spurious error about variably modified type
15477     * [340]18253 bad error message / ICE for invalid template parameter
15478     * [341]19608 ICE after friend function definition in local class
15479     * [342]19884 ICE on explicit instantiation of a non-template
15480       constructor
15481     * [343]20153 ICE when C++ template function contains anonymous union
15482     * [344]20563 Infinite loop in diagnostic (and ice after error
15483       message)
15484     * [345]20789 ICE with incomplete type in template
15485     * [346]21336 Internal compiler error when using custom new operators
15486     * [347]21768 ICE in error message due to violation of coding
15487       conventions
15488     * [348]21853 constness of pointer to data member ignored
15489     * [349]21903 Default argument of template function causes a
15490       compile-time error
15491     * [350]21983 multiple diagnostics
15492     * [351]21987 New testsuite failure
15493       g++.dg/warn/conversion-function-1.C
15494     * [352]22153 ICE on invalid template specialization
15495     * [353]22172 Internal compiler error, seg fault.
15496     * [354]21286 filebuf::xsgetn vs pipes
15497     * [355]22233 ICE with wrong number of template parameters
15498     * [356]22508 ICE after invalid operator new
15499     * [357]22545 ICE with pointer to class member & user defined
15500       conversion operator
15501     * [358]23528 Wrong default allocator in ext/hash_map
15502     * [359]23550 char_traits requirements/1.cc test bad math
15503     * [360]23586 Bad diagnostic for invalid namespace-name
15504     * [361]23624 ICE in invert_truthvalue, in fold-const.c
15505     * [362]23639 Bad error message: not a member of '<declaration error>'
15506     * [363]23797 ICE on typename outside template
15507     * [364]23965 Bogus error message: no matching function for call to
15508       'foo(<type error>)'
15509     * [365]24052 &#`label_decl' not supported by dump_expr#<expression
15510       error>
15511     * [366]24580 virtual base class cause exception not to be caught
15512
15513    Problems in generated debug information
15514
15515     * [367]24267 Bad DWARF for altivec vectors
15516
15517    Optimizations issues
15518
15519     * [368]17810 ICE in verify_local_live_at_start
15520     * [369]17860 Wrong generated code for loop with varying bound
15521     * [370]21709 ICE on compile-time complex NaN
15522     * [371]21964 broken tail call at -O2 or more
15523     * [372]22167 Strange optimization bug when using -Os
15524     * [373]22619 Compilation failure for real_const_1.f and
15525       real_const_2.f90
15526     * [374]23241 Invalid code generated for comparison of uchar to 255
15527     * [375]23478 Miscompilation due to reloading of a var that is also
15528       used in EH pad
15529     * [376]24470 segmentation fault in cc1plus when compiling with -O
15530     * [377]24950 ICE in operand_subword_force
15531
15532    Precompiled headers problems
15533
15534     * [378]14400 Cannot compile qt-x11-free-3.3.0
15535     * [379]14940 PCH largefile test fails on various platforms
15536
15537    Preprocessor bugs
15538
15539     * [380]20239 ICE on empty preprocessed input
15540     * [381]15220 "gcc -E -MM -MG" reports missing system headers in
15541       source directory
15542
15543    Testsuite issues
15544
15545     * [382]19275 gcc.dg/20020919-1.c fails with -fpic/-fPIC on
15546       i686-pc-linux-gnu
15547
15548    Alpha specific
15549
15550     * [383]21888 bootstrap failure with linker relaxation enabled
15551
15552    ARM specific
15553
15554     * [384]15342 [arm-linux]: ICE in verify_local_live_at_start
15555     * [385]23985 Memory aliasing information incorrect in inlined memcpy
15556
15557    ColdFile specific
15558
15559     * [386]16719 Illegal move of byte into address register causes
15560       compiler to ICE
15561
15562    HPPA specific
15563
15564     * [387]21723 ICE while building libgfortran
15565     * [388]21841 -mhp-ld/-mgnu-ld documentation
15566
15567    IA-64 specific
15568
15569     * [389]23644 IA-64 hardware models and configuration options
15570       documentation error
15571     * [390]24718 Shared libgcc not used for linking by default
15572
15573    M68000 specific
15574
15575     * [391]18421 ICE in reload_cse_simplify_operands, in postreload.c
15576
15577    MIPS specific
15578
15579     * [392]20621 ICE in change_address_1, in emit-rtl.c
15580
15581    PowerPC and PowerPC64 specific
15582
15583     * [393]18583 error on valid code: const
15584       __attribute__((altivec(vector__))) doesn't work in arrays
15585     * [394]20191 ICE in reload_cse_simplify_operands
15586     * [395]22083 AIX: TARGET_C99_FUNCTIONS is wrongly defined
15587     * [396]23070 CALL_V4_CLEAR_FP_ARGS flag not properly set
15588     * [397]23404 gij trashes args of functions with more than 8 fp args
15589     * [398]23539 C & C++ compiler generating misaligned references
15590       regardless of compiler flags
15591     * [399]24102 floatdisf2_internal2 broken
15592     * [400]24465 -mminimal-toc miscompilation of __thread vars
15593
15594    Solaris specific
15595
15596     * [401]19933 Problem with define of HUGE_VAL in math_c99
15597     * [402]21889 Native Solaris assembler cannot grok DTP-relative debug
15598       symbols
15599
15600    SPARC specific
15601
15602     * [403]19300 PCH failures on sparc-linux
15603     * [404]20301 Assembler labels have a leading "-"
15604     * [405]20673 C PCH testsuite assembly comparison failure
15605
15606    x86 and x86_64 specific
15607
15608     * [406]18582 ICE with arrays of type V2DF
15609     * [407]19340 Compilation SEGFAULTs with -O1 -fschedule-insns2
15610       -fsched2-use-traces
15611     * [408]21716 ICE in reg-stack.c's swap_rtx_condition
15612     * [409]24315 amd64 fails -fpeephole2
15613     __________________________________________________________________
15614
15615GCC 3.4.6
15616
15617   This is the [410]list of problem reports (PRs) from GCC's bug tracking
15618   system that are known to be fixed in the 3.4.6 release. This list might
15619   not be complete (that is, it is possible that some PRs that have been
15620   fixed are not listed here).
15621
15622
15623    For questions related to the use of GCC, please consult these web
15624    pages and the [411]GCC manuals. If that fails, the
15625    [412]gcc-help@gcc.gnu.org mailing list might help. Comments on these
15626    web pages and the development of GCC are welcome on our developer
15627    list at [413]gcc@gcc.gnu.org. All of [414]our lists have public
15628    archives.
15629
15630   Copyright (C) [415]Free Software Foundation, Inc. Verbatim copying and
15631   distribution of this entire article is permitted in any medium,
15632   provided this notice is preserved.
15633
15634   These pages are [416]maintained by the GCC team. Last modified
15635   2019-11-28[417].
15636
15637References
15638
15639   1. http://gcc.gnu.org/gcc-3.4/changes.html#3.4.6
15640   2. http://gcc.gnu.org/gcc-3.4/changes.html#cplusplus
15641   3. http://gcc.gnu.org/gcc-3.3/changes.html#obsolete_systems
15642   4. http://gcc.gnu.org/gcc-3.4/changes.html#obsolete_systems
15643   5. http://gcc.gnu.org/gcc-3.4/mips-abi.html
15644   6. http://gcc.gnu.org/gcc-3.4/sparc-abi.html
15645   7. https://www.boost.org/
15646   8. https://gcc.gnu.org/PR11953
15647   9. https://gcc.gnu.org/PR8361
15648  10. https://gcc.gnu.org/onlinedocs/gcc-3.4.3/gcc/Other-Builtins.html#Other%20Builtins
15649  11. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_closed.html#209
15650  12. http://gcc.gnu.org/bugs/#cxx_rvalbind
15651  13. https://gcc.gnu.org/onlinedocs/gcc-3.4.3/gcc/Objective-C-Dialect-Options.html
15652  14. https://gcc.gnu.org/onlinedocs/gcc-3.4.3/gcc/Objective-C-Dialect-Options.html
15653  15. https://gcc.gnu.org/onlinedocs/gcc-3.4.3/gcc/Objective-C-Dialect-Options.html
15654  16. http://www.gnu.org/software/classpath/
15655  17. http://www.eclipse.org/
15656  18. https://gcc.gnu.org/onlinedocs/gcc-3.4.3/g77/News.html
15657  19. https://gcc.gnu.org/onlinedocs/gcc-3.4.3/gcc/Alpha-Built-in-Functions.html
15658  20. https://gcc.gnu.org/onlinedocs/gcc-3.4.3/gccint/Processor-pipeline-description.html
15659  21. https://gcc.gnu.org/onlinedocs/gcc-3.4.3/gccint/Comparison-of-the-two-descriptions.html
15660  22. https://gcc.gnu.org/onlinedocs/gcc-3.4.3/gccint/Processor-pipeline-description.html
15661  23. http://gcc.gnu.org/gcc-3.4/mips-abi.html
15662  24. http://gcc.gnu.org/gcc-3.4/powerpc-abi.html
15663  25. http://gcc.gnu.org/gcc-3.4/sparc-abi.html
15664  26. https://gcc.gnu.org/bugzilla/buglist.cgi?short_desc_type=notregexp&short_desc=%5C%5B3%5C.4.*%5BRr%5Degression&target_milestone=3.4.0&bug_status=RESOLVED&resolution=FIXED
15665  27. https://gcc.gnu.org/PR10129
15666  28. https://gcc.gnu.org/PR14576
15667  29. https://gcc.gnu.org/PR14760
15668  30. https://gcc.gnu.org/PR14671
15669  31. https://gcc.gnu.org/PR15093
15670  32. https://gcc.gnu.org/PR15178
15671  33. https://gcc.gnu.org/PR12753
15672  34. https://gcc.gnu.org/PR13985
15673  35. https://gcc.gnu.org/PR14810
15674  36. https://gcc.gnu.org/PR14883
15675  37. https://gcc.gnu.org/PR15044
15676  38. https://gcc.gnu.org/PR15057
15677  39. https://gcc.gnu.org/PR15064
15678  40. https://gcc.gnu.org/PR15142
15679  41. https://gcc.gnu.org/PR15159
15680  42. https://gcc.gnu.org/PR15165
15681  43. https://gcc.gnu.org/PR15193
15682  44. https://gcc.gnu.org/PR15209
15683  45. https://gcc.gnu.org/PR15227
15684  46. https://gcc.gnu.org/PR15285
15685  47. https://gcc.gnu.org/PR15299
15686  48. https://gcc.gnu.org/PR15329
15687  49. https://gcc.gnu.org/PR15550
15688  50. https://gcc.gnu.org/PR15554
15689  51. https://gcc.gnu.org/PR15640
15690  52. https://gcc.gnu.org/PR15666
15691  53. https://gcc.gnu.org/PR15696
15692  54. https://gcc.gnu.org/PR15701
15693  55. https://gcc.gnu.org/PR15761
15694  56. https://gcc.gnu.org/PR15829
15695  57. https://gcc.gnu.org/PR14538
15696  58. https://gcc.gnu.org/PR12391
15697  59. https://gcc.gnu.org/PR14649
15698  60. https://gcc.gnu.org/PR15004
15699  61. https://gcc.gnu.org/PR15749
15700  62. https://gcc.gnu.org/PR10646
15701  63. https://gcc.gnu.org/PR12077
15702  64. https://gcc.gnu.org/PR13598
15703  65. https://gcc.gnu.org/PR14211
15704  66. https://gcc.gnu.org/PR14220
15705  67. https://gcc.gnu.org/PR14245
15706  68. https://gcc.gnu.org/PR14340
15707  69. https://gcc.gnu.org/PR14600
15708  70. https://gcc.gnu.org/PR14668
15709  71. https://gcc.gnu.org/PR14775
15710  72. https://gcc.gnu.org/PR14821
15711  73. https://gcc.gnu.org/PR14930
15712  74. https://gcc.gnu.org/PR14932
15713  75. https://gcc.gnu.org/PR14950
15714  76. https://gcc.gnu.org/PR14962
15715  77. https://gcc.gnu.org/PR14975
15716  78. https://gcc.gnu.org/PR15002
15717  79. https://gcc.gnu.org/PR15025
15718  80. https://gcc.gnu.org/PR15046
15719  81. https://gcc.gnu.org/PR15069
15720  82. https://gcc.gnu.org/PR15074
15721  83. https://gcc.gnu.org/PR15083
15722  84. https://gcc.gnu.org/PR15096
15723  85. https://gcc.gnu.org/PR15287
15724  86. https://gcc.gnu.org/PR15317
15725  87. https://gcc.gnu.org/PR15337
15726  88. https://gcc.gnu.org/PR15361
15727  89. https://gcc.gnu.org/PR15412
15728  90. https://gcc.gnu.org/PR15427
15729  91. https://gcc.gnu.org/PR15471
15730  92. https://gcc.gnu.org/PR15503
15731  93. https://gcc.gnu.org/PR15507
15732  94. https://gcc.gnu.org/PR15542
15733  95. https://gcc.gnu.org/PR15565
15734  96. https://gcc.gnu.org/PR15625
15735  97. https://gcc.gnu.org/PR15629
15736  98. https://gcc.gnu.org/PR15742
15737  99. https://gcc.gnu.org/PR15775
15738 100. https://gcc.gnu.org/PR15821
15739 101. https://gcc.gnu.org/PR15862
15740 102. https://gcc.gnu.org/PR15875
15741 103. https://gcc.gnu.org/PR15877
15742 104. https://gcc.gnu.org/PR15947
15743 105. https://gcc.gnu.org/PR16020
15744 106. https://gcc.gnu.org/PR16154
15745 107. https://gcc.gnu.org/PR16174
15746 108. https://gcc.gnu.org/PR14315
15747 109. https://gcc.gnu.org/PR15151
15748 110. https://gcc.gnu.org/PR7993
15749 111. https://gcc.gnu.org/PR15228
15750 112. https://gcc.gnu.org/PR15345
15751 113. https://gcc.gnu.org/PR15945
15752 114. https://gcc.gnu.org/PR15526
15753 115. https://gcc.gnu.org/PR14690
15754 116. https://gcc.gnu.org/PR15112
15755 117. https://gcc.gnu.org/PR15067
15756 118. https://gcc.gnu.org/PR1963
15757 119. https://gcc.gnu.org/PR15717
15758 120. https://gcc.gnu.org/PR14782
15759 121. https://gcc.gnu.org/PR14828
15760 122. https://gcc.gnu.org/PR15202
15761 123. https://gcc.gnu.org/PR14610
15762 124. https://gcc.gnu.org/PR14813
15763 125. https://gcc.gnu.org/PR14857
15764 126. https://gcc.gnu.org/PR15598
15765 127. https://gcc.gnu.org/PR15653
15766 128. https://gcc.gnu.org/PR15189
15767 129. https://gcc.gnu.org/PR15331
15768 130. https://gcc.gnu.org/PR16144
15769 131. https://gcc.gnu.org/PR16176
15770 132. https://gcc.gnu.org/PR11591
15771 133. https://gcc.gnu.org/PR12028
15772 134. https://gcc.gnu.org/PR14478
15773 135. https://gcc.gnu.org/PR14567
15774 136. https://gcc.gnu.org/PR14715
15775 137. https://gcc.gnu.org/PR14902
15776 138. https://gcc.gnu.org/PR14924
15777 139. https://gcc.gnu.org/PR14960
15778 140. https://gcc.gnu.org/PR15106
15779 141. https://gcc.gnu.org/PR16026
15780 142. https://gcc.gnu.org/PR15191
15781 143. https://gcc.gnu.org/PR15662
15782 144. https://gcc.gnu.org/PR15054
15783 145. https://gcc.gnu.org/PR15783
15784 146. https://gcc.gnu.org/PR15626
15785 147. https://gcc.gnu.org/PR14326
15786 148. https://gcc.gnu.org/PR14723
15787 149. https://gcc.gnu.org/PR15290
15788 150. https://gcc.gnu.org/PR15250
15789 151. https://gcc.gnu.org/PR15551
15790 152. https://gcc.gnu.org/PR8309
15791 153. https://gcc.gnu.org/PR13250
15792 154. https://gcc.gnu.org/PR13803
15793 155. https://gcc.gnu.org/PR14093
15794 156. https://gcc.gnu.org/PR14457
15795 157. https://gcc.gnu.org/PR14542
15796 158. https://gcc.gnu.org/PR15100
15797 159. https://gcc.gnu.org/PR15296
15798 160. https://gcc.gnu.org/PR15396
15799 161. https://gcc.gnu.org/PR15782
15800 162. https://gcc.gnu.org/PR11610
15801 163. https://gcc.gnu.org/PR15488
15802 164. https://gcc.gnu.org/PR15489
15803 165. https://gcc.gnu.org/PR13928
15804 166. https://gcc.gnu.org/PR14150
15805 167. https://gcc.gnu.org/PR14949
15806 168. https://gcc.gnu.org/PR15123
15807 169. https://gcc.gnu.org/PR16469
15808 170. https://gcc.gnu.org/PR16344
15809 171. https://gcc.gnu.org/PR16842
15810 172. https://gcc.gnu.org/PR12608
15811 173. https://gcc.gnu.org/PR14492
15812 174. https://gcc.gnu.org/PR15461
15813 175. https://gcc.gnu.org/PR15890
15814 176. https://gcc.gnu.org/PR16180
15815 177. https://gcc.gnu.org/PR16224
15816 178. https://gcc.gnu.org/PR16408
15817 179. https://gcc.gnu.org/PR16529
15818 180. https://gcc.gnu.org/PR16698
15819 181. https://gcc.gnu.org/PR16706
15820 182. https://gcc.gnu.org/PR16810
15821 183. https://gcc.gnu.org/PR16851
15822 184. https://gcc.gnu.org/PR16870
15823 185. https://gcc.gnu.org/PR16904
15824 186. https://gcc.gnu.org/PR16905
15825 187. https://gcc.gnu.org/PR16964
15826 188. https://gcc.gnu.org/PR17068
15827 189. https://gcc.gnu.org/PR16366
15828 190. https://gcc.gnu.org/PR15345
15829 191. https://gcc.gnu.org/PR16590
15830 192. https://gcc.gnu.org/PR16693
15831 193. https://gcc.gnu.org/PR17078
15832 194. https://gcc.gnu.org/PR13956
15833 195. https://gcc.gnu.org/PR16684
15834 196. https://gcc.gnu.org/PR12658
15835 197. https://gcc.gnu.org/PR13092
15836 198. https://gcc.gnu.org/PR15320
15837 199. https://gcc.gnu.org/PR16246
15838 200. https://gcc.gnu.org/PR16273
15839 201. https://gcc.gnu.org/PR16401
15840 202. https://gcc.gnu.org/PR16411
15841 203. https://gcc.gnu.org/PR16489
15842 204. https://gcc.gnu.org/PR16618
15843 205. https://gcc.gnu.org/PR16637
15844 206. https://gcc.gnu.org/PR16717
15845 207. https://gcc.gnu.org/PR16813
15846 208. https://gcc.gnu.org/PR16853
15847 209. https://gcc.gnu.org/PR16889
15848 210. https://gcc.gnu.org/PR16959
15849 211. https://gcc.gnu.org/PR7587
15850 212. https://gcc.gnu.org/PR16473
15851 213. https://gcc.gnu.org/PR16478
15852 214. https://gcc.gnu.org/PR10695
15853 215. https://gcc.gnu.org/PR16974
15854 216. https://gcc.gnu.org/PR16298
15855 217. https://gcc.gnu.org/PR17113
15856 218. https://gcc.gnu.org/PR14697
15857 219. https://gcc.gnu.org/PR15869
15858 220. https://gcc.gnu.org/PR16325
15859 221. https://gcc.gnu.org/PR16357
15860 222. https://gcc.gnu.org/PR16380
15861 223. https://gcc.gnu.org/PR16407
15862 224. https://gcc.gnu.org/PR16643
15863 225. https://gcc.gnu.org/PR15927
15864 226. https://gcc.gnu.org/PR15948
15865 227. https://gcc.gnu.org/PR17019
15866 228. https://gcc.gnu.org/PR16130
15867 229. https://gcc.gnu.org/PR16142
15868 230. https://gcc.gnu.org/PR16278
15869 231. https://gcc.gnu.org/PR16414
15870 232. https://gcc.gnu.org/PR16445
15871 233. https://gcc.gnu.org/PR16490
15872 234. https://gcc.gnu.org/PR16683
15873 235. https://gcc.gnu.org/PR16195
15874 236. https://gcc.gnu.org/PR16239
15875 237. https://gcc.gnu.org/PR16199
15876 238. https://gcc.gnu.org/PR16416
15877 239. https://gcc.gnu.org/PR16430
15878 240. https://gcc.gnu.org/PR16379
15879 241. https://gcc.gnu.org/PR17093
15880 242. https://gcc.gnu.org/PR17119
15881 243. https://gcc.gnu.org/PR15928
15882 244. https://gcc.gnu.org/PR16210
15883 245. https://gcc.gnu.org/PR15488
15884 246. https://gcc.gnu.org/PR16250
15885 247. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=3.4.3
15886 248. https://gcc.gnu.org/PR17369
15887 249. https://gcc.gnu.org/PR17850
15888 250. https://gcc.gnu.org/PR13948
15889 251. https://gcc.gnu.org/PR14492
15890 252. https://gcc.gnu.org/PR16301
15891 253. https://gcc.gnu.org/PR16566
15892 254. https://gcc.gnu.org/PR17023
15893 255. https://gcc.gnu.org/PR17027
15894 256. https://gcc.gnu.org/PR17524
15895 257. https://gcc.gnu.org/PR17826
15896 258. https://gcc.gnu.org/PR15526
15897 259. https://gcc.gnu.org/PR16999
15898 260. https://gcc.gnu.org/PR17503
15899 261. https://gcc.gnu.org/PR17581
15900 262. https://gcc.gnu.org/PR18129
15901 263. https://gcc.gnu.org/PR10975
15902 264. https://gcc.gnu.org/PR11722
15903 265. https://gcc.gnu.org/PR14534
15904 266. https://gcc.gnu.org/PR15172
15905 267. https://gcc.gnu.org/PR15786
15906 268. https://gcc.gnu.org/PR16162
15907 269. https://gcc.gnu.org/PR16612
15908 270. https://gcc.gnu.org/PR16715
15909 271. https://gcc.gnu.org/PR16848
15910 272. https://gcc.gnu.org/PR17132
15911 273. https://gcc.gnu.org/PR17259
15912 274. https://gcc.gnu.org/PR17327
15913 275. https://gcc.gnu.org/PR17393
15914 276. https://gcc.gnu.org/PR17501
15915 277. https://gcc.gnu.org/PR17537
15916 278. https://gcc.gnu.org/PR17585
15917 279. https://gcc.gnu.org/PR17821
15918 280. https://gcc.gnu.org/PR17829
15919 281. https://gcc.gnu.org/PR17851
15920 282. https://gcc.gnu.org/PR17976
15921 283. https://gcc.gnu.org/PR18020
15922 284. https://gcc.gnu.org/PR18093
15923 285. https://gcc.gnu.org/PR18140
15924 286. https://gcc.gnu.org/PR17541
15925 287. https://gcc.gnu.org/PR17853
15926 288. https://gcc.gnu.org/PR17245
15927 289. https://gcc.gnu.org/PR17167
15928 290. https://gcc.gnu.org/PR17277
15929 291. https://gcc.gnu.org/PR17505
15930 292. https://gcc.gnu.org/PR17684
15931 293. https://gcc.gnu.org/PR17384
15932 294. https://gcc.gnu.org/PR17770
15933 295. https://gcc.gnu.org/PR11476
15934 296. https://gcc.gnu.org/PR14064
15935 297. https://gcc.gnu.org/PR14678
15936 298. https://gcc.gnu.org/PR15583
15937 299. https://gcc.gnu.org/PR15790
15938 300. https://gcc.gnu.org/PR15886
15939 301. https://gcc.gnu.org/PR16884
15940 302. https://gcc.gnu.org/PR13841
15941 303. https://gcc.gnu.org/PR15860
15942 304. https://gcc.gnu.org/PR17465
15943 305. https://gcc.gnu.org/PR17469
15944 306. https://gcc.gnu.org/PR18138
15945 307. https://gcc.gnu.org/PR15498
15946 308. https://gcc.gnu.org/PR15747
15947 309. https://gcc.gnu.org/PR16406
15948 310. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=3.4.4
15949 311. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=3.4.5
15950 312. https://gcc.gnu.org/PR24688
15951 313. https://gcc.gnu.org/PR17188
15952 314. https://gcc.gnu.org/PR20187
15953 315. https://gcc.gnu.org/PR21873
15954 316. https://gcc.gnu.org/PR21899
15955 317. https://gcc.gnu.org/PR22061
15956 318. https://gcc.gnu.org/PR22208
15957 319. https://gcc.gnu.org/PR22458
15958 320. https://gcc.gnu.org/PR22589
15959 321. https://gcc.gnu.org/PR24101
15960 322. https://gcc.gnu.org/PR10611
15961 323. https://gcc.gnu.org/PR13377
15962 324. https://gcc.gnu.org/PR16002
15963 325. https://gcc.gnu.org/PR17413
15964 326. https://gcc.gnu.org/PR17609
15965 327. https://gcc.gnu.org/PR17618
15966 328. https://gcc.gnu.org/PR18124
15967 329. https://gcc.gnu.org/PR18155
15968 330. https://gcc.gnu.org/PR18177
15969 331. https://gcc.gnu.org/PR18368
15970 332. https://gcc.gnu.org/PR18378
15971 333. https://gcc.gnu.org/PR18466
15972 334. https://gcc.gnu.org/PR18512
15973 335. https://gcc.gnu.org/PR18545
15974 336. https://gcc.gnu.org/PR18738
15975 337. https://gcc.gnu.org/PR18803
15976 338. https://gcc.gnu.org/PR19004
15977 339. https://gcc.gnu.org/PR19208
15978 340. https://gcc.gnu.org/PR19253
15979 341. https://gcc.gnu.org/PR19608
15980 342. https://gcc.gnu.org/PR19884
15981 343. https://gcc.gnu.org/PR20153
15982 344. https://gcc.gnu.org/PR20563
15983 345. https://gcc.gnu.org/PR20789
15984 346. https://gcc.gnu.org/PR21336
15985 347. https://gcc.gnu.org/PR21768
15986 348. https://gcc.gnu.org/PR21853
15987 349. https://gcc.gnu.org/PR21903
15988 350. https://gcc.gnu.org/PR21983
15989 351. https://gcc.gnu.org/PR21987
15990 352. https://gcc.gnu.org/PR22153
15991 353. https://gcc.gnu.org/PR22172
15992 354. https://gcc.gnu.org/PR21286
15993 355. https://gcc.gnu.org/PR22233
15994 356. https://gcc.gnu.org/PR22508
15995 357. https://gcc.gnu.org/PR22545
15996 358. https://gcc.gnu.org/PR23528
15997 359. https://gcc.gnu.org/PR23550
15998 360. https://gcc.gnu.org/PR23586
15999 361. https://gcc.gnu.org/PR23624
16000 362. https://gcc.gnu.org/PR23639
16001 363. https://gcc.gnu.org/PR23797
16002 364. https://gcc.gnu.org/PR23965
16003 365. https://gcc.gnu.org/PR24052
16004 366. https://gcc.gnu.org/PR24580
16005 367. https://gcc.gnu.org/PR24267
16006 368. https://gcc.gnu.org/PR17810
16007 369. https://gcc.gnu.org/PR17860
16008 370. https://gcc.gnu.org/PR21709
16009 371. https://gcc.gnu.org/PR21964
16010 372. https://gcc.gnu.org/PR22167
16011 373. https://gcc.gnu.org/PR22619
16012 374. https://gcc.gnu.org/PR23241
16013 375. https://gcc.gnu.org/PR23478
16014 376. https://gcc.gnu.org/PR24470
16015 377. https://gcc.gnu.org/PR24950
16016 378. https://gcc.gnu.org/PR14400
16017 379. https://gcc.gnu.org/PR14940
16018 380. https://gcc.gnu.org/PR20239
16019 381. https://gcc.gnu.org/PR15220
16020 382. https://gcc.gnu.org/PR19275
16021 383. https://gcc.gnu.org/PR21888
16022 384. https://gcc.gnu.org/PR15342
16023 385. https://gcc.gnu.org/PR23985
16024 386. https://gcc.gnu.org/PR16719
16025 387. https://gcc.gnu.org/PR21723
16026 388. https://gcc.gnu.org/PR21841
16027 389. https://gcc.gnu.org/PR23644
16028 390. https://gcc.gnu.org/PR24718
16029 391. https://gcc.gnu.org/PR18421
16030 392. https://gcc.gnu.org/PR20621
16031 393. https://gcc.gnu.org/PR18583
16032 394. https://gcc.gnu.org/PR20191
16033 395. https://gcc.gnu.org/PR22083
16034 396. https://gcc.gnu.org/PR23070
16035 397. https://gcc.gnu.org/PR23404
16036 398. https://gcc.gnu.org/PR23539
16037 399. https://gcc.gnu.org/PR24102
16038 400. https://gcc.gnu.org/PR24465
16039 401. https://gcc.gnu.org/PR19933
16040 402. https://gcc.gnu.org/PR21889
16041 403. https://gcc.gnu.org/PR19300
16042 404. https://gcc.gnu.org/PR20301
16043 405. https://gcc.gnu.org/PR20673
16044 406. https://gcc.gnu.org/PR18582
16045 407. https://gcc.gnu.org/PR19340
16046 408. https://gcc.gnu.org/PR21716
16047 409. https://gcc.gnu.org/PR24315
16048 410. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=3.4.6
16049 411. https://gcc.gnu.org/onlinedocs/
16050 412. mailto:gcc-help@gcc.gnu.org
16051 413. mailto:gcc@gcc.gnu.org
16052 414. https://gcc.gnu.org/lists.html
16053 415. https://www.fsf.org/
16054 416. https://gcc.gnu.org/about.html
16055 417. http://validator.w3.org/check/referer
16056======================================================================
16057http://gcc.gnu.org/gcc-3.3/index.html
16058
16059                             GCC 3.3 Release Series
16060
16061   (This release series is no longer supported.)
16062
16063   May 03, 2005
16064
16065   The [1]GNU project and the GCC developers are pleased to announce the
16066   release of GCC 3.3.6.
16067
16068   This release is a bug-fix release, containing fixes for regressions in
16069   GCC 3.3.5 relative to previous releases of GCC.
16070
16071   This release is the last of the series 3.3.x.
16072
16073   The GCC 3.3 release series includes numerous [2]new features,
16074   improvements, bug fixes, and other changes, thanks to an [3]amazing
16075   group of volunteers.
16076
16077Release History
16078
16079   GCC 3.3.6
16080          May 3, 2005 ([4]changes)
16081
16082   GCC 3.3.5
16083          September 30, 2004 ([5]changes)
16084
16085   GCC 3.3.4
16086          May 31, 2004 ([6]changes)
16087
16088   GCC 3.3.3
16089          February 14, 2004 ([7]changes)
16090
16091   GCC 3.3.2
16092          October 16, 2003 ([8]changes)
16093
16094   GCC 3.3.1
16095          August 8, 2003 ([9]changes)
16096
16097   GCC 3.3
16098          May 14, 2003 ([10]changes)
16099
16100References and Acknowledgements
16101
16102   GCC used to stand for the GNU C Compiler, but since the compiler
16103   supports several other languages aside from C, it now stands for the
16104   GNU Compiler Collection.
16105
16106   A list of [11]successful builds is updated as new information becomes
16107   available.
16108
16109   The GCC developers would like to thank the numerous people that have
16110   contributed new features, improvements, bug fixes, and other changes as
16111   well as test results to GCC. This [12]amazing group of volunteers is
16112   what makes GCC successful.
16113
16114   For additional information about GCC please refer to the [13]GCC
16115   project web site or contact the [14]GCC development mailing list.
16116
16117   To obtain GCC please use [15]our mirror sites, or our CVS server.
16118
16119
16120    For questions related to the use of GCC, please consult these web
16121    pages and the [16]GCC manuals. If that fails, the
16122    [17]gcc-help@gcc.gnu.org mailing list might help. Comments on these
16123    web pages and the development of GCC are welcome on our developer
16124    list at [18]gcc@gcc.gnu.org. All of [19]our lists have public
16125    archives.
16126
16127   Copyright (C) [20]Free Software Foundation, Inc. Verbatim copying and
16128   distribution of this entire article is permitted in any medium,
16129   provided this notice is preserved.
16130
16131   These pages are [21]maintained by the GCC team. Last modified
16132   2019-11-28[22].
16133
16134References
16135
16136   1. http://www.gnu.org/
16137   2. http://gcc.gnu.org/gcc-3.3/changes.html
16138   3. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
16139   4. http://gcc.gnu.org/gcc-3.3/changes.html#3.3.6
16140   5. http://gcc.gnu.org/gcc-3.3/changes.html#3.3.5
16141   6. http://gcc.gnu.org/gcc-3.3/changes.html#3.3.4
16142   7. http://gcc.gnu.org/gcc-3.3/changes.html#3.3.3
16143   8. http://gcc.gnu.org/gcc-3.3/changes.html#3.3.2
16144   9. http://gcc.gnu.org/gcc-3.3/changes.html#3.3.1
16145  10. http://gcc.gnu.org/gcc-3.3/changes.html
16146  11. http://gcc.gnu.org/gcc-3.3/buildstat.html
16147  12. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
16148  13. http://gcc.gnu.org/index.html
16149  14. mailto:gcc@gcc.gnu.org
16150  15. http://gcc.gnu.org/mirrors.html
16151  16. https://gcc.gnu.org/onlinedocs/
16152  17. mailto:gcc-help@gcc.gnu.org
16153  18. mailto:gcc@gcc.gnu.org
16154  19. https://gcc.gnu.org/lists.html
16155  20. https://www.fsf.org/
16156  21. https://gcc.gnu.org/about.html
16157  22. http://validator.w3.org/check/referer
16158======================================================================
16159http://gcc.gnu.org/gcc-3.3/changes.html
16160
16161                             GCC 3.3 Release Series
16162                        Changes, New Features, and Fixes
16163
16164   The latest release in the 3.3 release series is [1]GCC 3.3.6.
16165
16166Caveats
16167
16168     * The preprocessor no longer accepts multi-line string literals. They
16169       were deprecated in 3.0, 3.1, and 3.2.
16170     * The preprocessor no longer supports the -A- switch when appearing
16171       alone. -A- followed by an assertion is still supported.
16172     * Support for all the systems [2]obsoleted in GCC 3.1 has been
16173       removed from GCC 3.3. See below for a [3]list of systems which are
16174       obsoleted in this release.
16175     * Checking for null format arguments has been decoupled from the rest
16176       of the format checking mechanism. Programs which use the format
16177       attribute may regain this functionality by using the new [4]nonnull
16178       function attribute. Note that all functions for which GCC has a
16179       built-in format attribute, an appropriate built-in nonnull
16180       attribute is also applied.
16181     * The DWARF (version 1) debugging format has been deprecated and will
16182       be removed in a future version of GCC. Version 2 of the DWARF
16183       debugging format will continue to be supported for the foreseeable
16184       future.
16185     * The C and Objective-C compilers no longer accept the "Naming Types"
16186       extension (typedef foo = bar); it was already unavailable in C++.
16187       Code which uses it will need to be changed to use the "typeof"
16188       extension instead: typedef typeof(bar) foo. (We have removed this
16189       extension without a period of deprecation because it has caused the
16190       compiler to crash since version 3.0 and no one noticed until very
16191       recently. Thus we conclude it is not in widespread use.)
16192     * The -traditional C compiler option has been removed. It was
16193       deprecated in 3.1 and 3.2. (Traditional preprocessing remains
16194       available.) The <varargs.h> header, used for writing variadic
16195       functions in traditional C, still exists but will produce an error
16196       message if used.
16197     * GCC 3.3.1 automatically places zero-initialized variables in the
16198       .bss section on some operating systems. Versions of GNU Emacs up to
16199       (and including) 21.3 will not work correctly when using this
16200       optimization; you can use -fno-zero-initialized-in-bss to disable
16201       it.
16202
16203General Optimizer Improvements
16204
16205     * A new scheme for accurately describing processor pipelines, the
16206       [5]DFA scheduler, has been added.
16207     * Pavel Nejedly, Charles University Prague, has contributed new file
16208       format used by the edge coverage profiler (-fprofile-arcs).
16209       The new format is robust and diagnoses common mistakes where
16210       profiles from different versions (or compilations) of the program
16211       are combined resulting in nonsensical profiles and slow code to
16212       produced with profile feedback. Additionally this format allows
16213       extra data to be gathered. Currently, overall statistics are
16214       produced helping optimizers to identify hot spots of a program
16215       globally replacing the old intra-procedural scheme and resulting in
16216       better code. Note that the gcov tool from older GCC versions will
16217       not be able to parse the profiles generated by GCC 3.3 and vice
16218       versa.
16219     * Jan Hubicka, SuSE Labs, has contributed a new superblock formation
16220       pass enabled using -ftracer. This pass simplifies the control flow
16221       of functions allowing other optimizations to do better job.
16222       He also contributed the function reordering pass
16223       (-freorder-functions) to optimize function placement using profile
16224       feedback.
16225
16226New Languages and Language specific improvements
16227
16228  C/ObjC/C++
16229
16230     * The preprocessor now accepts directives within macro arguments. It
16231       processes them just as if they had not been within macro arguments.
16232     * The separate ISO and traditional preprocessors have been completely
16233       removed. The front end handles either type of preprocessed output
16234       if necessary.
16235     * In C99 mode preprocessor arithmetic is done in the precision of the
16236       target's intmax_t, as required by that standard.
16237     * The preprocessor can now copy comments inside macros to the output
16238       file when the macro is expanded. This feature, enabled using the
16239       -CC option, is intended for use by applications which place
16240       metadata or directives inside comments, such as lint.
16241     * The method of constructing the list of directories to be searched
16242       for header files has been revised. If a directory named by a -I
16243       option is a standard system include directory, the option is
16244       ignored to ensure that the default search order for system
16245       directories and the special treatment of system header files are
16246       not defeated.
16247     * A few more [6]ISO C99 features now work correctly.
16248     * A new function attribute, nonnull, has been added which allows
16249       pointer arguments to functions to be specified as requiring a
16250       non-null value. The compiler currently uses this information to
16251       issue a warning when it detects a null value passed in such an
16252       argument slot.
16253     * A new type attribute, may_alias, has been added. Accesses to
16254       objects with types with this attribute are not subjected to
16255       type-based alias analysis, but are instead assumed to be able to
16256       alias any other type of objects, just like the char type.
16257
16258  C++
16259
16260     * Type based alias analysis has been implemented for C++ aggregate
16261       types.
16262
16263  Objective-C
16264
16265     * Generate an error if Objective-C objects are passed by value in
16266       function and method calls.
16267     * When -Wselector is used, check the whole list of selectors at the
16268       end of compilation, and emit a warning if a @selector() is not
16269       known.
16270     * Define __NEXT_RUNTIME__ when compiling for the NeXT runtime.
16271     * No longer need to include objc/objc-class.h to compile self calls
16272       in class methods (NeXT runtime only).
16273     * New -Wundeclared-selector option.
16274     * Removed selector bloating which was causing object files to be 10%
16275       bigger on average (GNU runtime only).
16276     * Using at run time @protocol() objects has been fixed in certain
16277       situations (GNU runtime only).
16278     * Type checking has been fixed and improved in many situations
16279       involving protocols.
16280
16281  Java
16282
16283     * The java.sql and javax.sql packages now implement the JDBC 3.0 (JDK
16284       1.4) API.
16285     * The JDK 1.4 assert facility has been implemented.
16286     * The bytecode interpreter is now direct threaded and thus faster.
16287
16288  Fortran
16289
16290     * Fortran improvements are listed in [7]the Fortran documentation.
16291
16292  Ada
16293
16294     * Ada tasking now works with glibc 2.3.x threading libraries.
16295
16296New Targets and Target Specific Improvements
16297
16298     * The following changes have been made to the HP-PA port:
16299          + The port now defaults to scheduling for the PA8000 series of
16300            processors.
16301          + Scheduling support for the PA7300 processor has been added.
16302          + The 32-bit port now supports weak symbols under HP-UX 11.
16303          + The handling of initializers and finalizers has been improved
16304            under HP-UX 11. The 64-bit port no longer uses collect2.
16305          + Dwarf2 EH support has been added to the 32-bit GNU/Linux port.
16306          + ABI fixes to correct the passing of small structures by value.
16307     * The SPARC, HP-PA, SH4, and x86/pentium ports have been converted to
16308       use the DFA processor pipeline description.
16309     * The following NetBSD configurations for the SuperH processor family
16310       have been added:
16311          + SH3, big-endian, sh-*-netbsdelf*
16312          + SH3, little-endian, shle-*-netbsdelf*
16313          + SH5, SHmedia, big-endian, 32-bit default, sh5-*-netbsd*
16314          + SH5, SHmedia, little-endian, 32-bit default, sh5le-*-netbsd*
16315          + SH5, SHmedia, big-endian, 64-bit default, sh64-*-netbsd*
16316          + SH5, SHmedia, little-endian, 64-bit default, sh64le-*-netbsd*
16317     * The following changes have been made to the IA-32/x86-64 port:
16318          + SSE2 and 3dNOW! intrinsics are now supported.
16319          + Support for thread local storage has been added to the IA-32
16320            and x86-64 ports.
16321          + The x86-64 port has been significantly improved.
16322     * The following changes have been made to the MIPS port:
16323          + All configurations now accept the -mabi switch. Note that you
16324            will need appropriate multilibs for this option to work
16325            properly.
16326          + ELF configurations will always pass an ABI flag to the
16327            assembler, except when the MIPS EABI is selected.
16328          + -mabi=64 no longer selects MIPS IV code.
16329          + The -mcpu option, which was deprecated in 3.1 and 3.2, has
16330            been removed from this release.
16331          + -march now changes the core ISA level. In previous releases,
16332            it would change the use of processor-specific extensions, but
16333            would leave the core ISA unchanged. For example, mips64-elf
16334            -march=r8000 will now generate MIPS IV code.
16335          + Under most configurations, -mipsN now acts as a synonym for
16336            -march.
16337          + There are some new preprocessor macros to describe the -march
16338            and -mtune settings. See the documentation of those options
16339            for details.
16340          + Support for the NEC VR-Series processors has been added. This
16341            includes the 54xx, 5500, and 41xx series.
16342          + Support for the Sandcraft sr71k processor has been added.
16343     * The following changes have been made to the S/390 port:
16344          + Support to build the Java runtime libraries has been added.
16345            Java is now enabled by default on s390-*-linux* and
16346            s390x-*-linux* targets.
16347          + Multilib support for the s390x-*-linux* target has been added;
16348            this allows to build 31-bit binaries using the -m31 option.
16349          + Support for thread local storage has been added.
16350          + Inline assembler code may now use the 'Q' constraint to
16351            specify memory operands without index register.
16352          + Various platform-specific performance improvements have been
16353            implemented; in particular, the compiler now uses the BRANCH
16354            ON COUNT family of instructions and makes more frequent use of
16355            the TEST UNDER MASK family of instructions.
16356     * The following changes have been made to the PowerPC port:
16357          + Support for IBM Power4 processor added.
16358          + Support for Motorola e500 SPE added.
16359          + Support for AIX 5.2 added.
16360          + Function and Data sections now supported on AIX.
16361          + Sibcall optimizations added.
16362     * The support for H8 Tiny is added to the H8/300 port with -mn.
16363
16364Obsolete Systems
16365
16366   Support for a number of older systems has been declared obsolete in GCC
16367   3.3. Unless there is activity to revive them, the next release of GCC
16368   will have their sources permanently removed.
16369
16370   All configurations of the following processor architectures have been
16371   declared obsolete:
16372     * Matsushita MN10200, mn10200-*-*
16373     * Motorola 88000, m88k-*-*
16374     * IBM ROMP, romp-*-*
16375
16376   Also, some individual systems have been obsoleted:
16377     * Alpha
16378          + Interix, alpha*-*-interix*
16379          + Linux libc1, alpha*-*-linux*libc1*
16380          + Linux ECOFF, alpha*-*-linux*ecoff*
16381     * ARM
16382          + Generic a.out, arm*-*-aout*
16383          + Conix, arm*-*-conix*
16384          + "Old ABI," arm*-*-oabi
16385          + StrongARM/COFF, strongarm-*-coff*
16386     * HPPA (PA-RISC)
16387          + Generic OSF, hppa1.0-*-osf*
16388          + Generic BSD, hppa1.0-*-bsd*
16389          + HP/UX versions 7, 8, and 9, hppa1.[01]-*-hpux[789]*
16390          + HiUX, hppa*-*-hiux*
16391          + Mach Lites, hppa*-*-lites*
16392     * Intel 386 family
16393          + Windows NT 3.x, i?86-*-win32
16394     * MC68000 family
16395          + HP systems, m68000-hp-bsd* and m68k-hp-bsd*
16396          + Sun systems, m68000-sun-sunos*, m68k-sun-sunos*, and
16397            m68k-sun-mach*
16398          + AT&T systems, m68000-att-sysv*
16399          + Atari systems, m68k-atari-sysv*
16400          + Motorola systems, m68k-motorola-sysv*
16401          + NCR systems, m68k-ncr-sysv*
16402          + Plexus systems, m68k-plexus-sysv*
16403          + Commodore systems, m68k-cbm-sysv*
16404          + Citicorp TTI, m68k-tti-*
16405          + Unos, m68k-crds-unos*
16406          + Concurrent RTU, m68k-ccur-rtu*
16407          + Linux a.out, m68k-*-linux*aout*
16408          + Linux libc1, m68k-*-linux*libc1*
16409          + pSOS, m68k-*-psos*
16410     * MIPS
16411          + Generic ECOFF, mips*-*-ecoff*
16412          + SINIX, mips-sni-sysv4
16413          + Orion RTEMS, mips64orion-*-rtems*
16414     * National Semiconductor 32000
16415          + OpenBSD, ns32k-*-openbsd*
16416     * POWER (aka RS/6000) and PowerPC
16417          + AIX versions 1, 2, and 3, rs6000-ibm-aix[123]*
16418          + Bull BOSX, rs6000-bull-bosx
16419          + Generic Mach, rs6000-*-mach*
16420          + Generic SysV, powerpc*-*-sysv*
16421          + Linux libc1, powerpc*-*-linux*libc1*
16422     * Sun SPARC
16423          + Generic a.out, sparc-*-aout*, sparclet-*-aout*,
16424            sparclite-*-aout*, and sparc86x-*-aout*
16425          + NetBSD a.out, sparc-*-netbsd*aout*
16426          + Generic BSD, sparc-*-bsd*
16427          + ChorusOS, sparc-*-chorusos*
16428          + Linux a.out, sparc-*-linux*aout*
16429          + Linux libc1, sparc-*-linux*libc1*
16430          + LynxOS, sparc-*-lynxos*
16431          + Solaris on HAL hardware, sparc-hal-solaris2*
16432          + SunOS versions 3 and 4, sparc-*-sunos[34]*
16433     * NEC V850
16434          + RTEMS, v850-*-rtems*
16435     * VAX
16436          + VMS, vax-*-vms*
16437
16438Documentation improvements
16439
16440Other significant improvements
16441
16442     * Almost all front-end dependencies in the compiler have been
16443       separated out into a set of language hooks. This should make adding
16444       a new front end clearer and easier.
16445     * One effect of removing the separate preprocessor is a small
16446       increase in the robustness of the compiler in general, and the
16447       maintainability of target descriptions. Previously target-specific
16448       built-in macros and others, such as __FAST_MATH__, had to be
16449       handled with so-called specs that were hard to maintain. Often they
16450       would fail to behave properly when conflicting options were
16451       supplied on the command line, and define macros in the user's
16452       namespace even when strict ISO compliance was requested.
16453       Integrating the preprocessor has cleanly solved these issues.
16454     * The Makefile suite now supports redirection of make install by
16455       means of the variable DESTDIR.
16456     __________________________________________________________________
16457
16458GCC 3.3
16459
16460   Detailed release notes for the GCC 3.3 release follow.
16461
16462  Bug Fixes
16463
16464    bootstrap failures
16465
16466     * [8]10140 cross compiler build failures: missing __mempcpy (DUP:
16467       [9]10198,[10]10338)
16468
16469    Internal compiler errors (multi-platform)
16470
16471     * [11]3581 large string causes segmentation fault in cc1
16472     * [12]4382 __builtin_{set,long}jmp with -O3 can crash the compiler
16473     * [13]5533 (c++) ICE when processing std::accumulate(begin, end,
16474       init, invalid_op)
16475     * [14]6387 -fpic -gdwarf-2 -g1 combination gives ICE in dwarf2out
16476     * [15]6412 (c++) ICE in retrieve_specialization
16477     * [16]6620 (c++) partial template specialization causes an ICE
16478       (segmentation fault)
16479     * [17]6663 (c++) ICE with attribute aligned
16480     * [18]7068 ICE with incomplete types
16481     * [19]7083 (c++) ICE using -gstabs with dodgy class derivation
16482     * [20]7647 (c++) ICE when data member has the name of the enclosing
16483       class
16484     * [21]7675 ICE in fixup_var_refs_1
16485     * [22]7718 'complex' template instantiation causes ICE
16486     * [23]8116 (c++) ICE in member template function
16487     * [24]8358 (ada) Ada compiler accesses freed memory, crashes
16488     * [25]8511 (c++) ICE: (hopefully) reproducible cc1plus segmentation
16489       fault
16490     * [26]8564 (c++) ICE in find_function_data, in function.c
16491     * [27]8660 (c++) template overloading ICE in tsubst_expr, in cp/pt.c
16492     * [28]8766 (c++) ICE after failed initialization of static template
16493       variable
16494     * [29]8803 ICE in instantiate_virtual_regs_1, in function.c
16495     * [30]8846 (c++) ICE after diagnostic if fr_FR@euro locale is set
16496     * [31]8906 (c++) ICE (Segmentation fault) when parsing nested-class
16497       definition
16498     * [32]9216 (c++) ICE on missing template parameter
16499     * [33]9261 (c++) ICE in arg_assoc, in cp/decl2.c
16500     * [34]9263 (fortran) ICE caused by invalid PARAMETER in implied DO
16501       loop
16502     * [35]9429 (c++) ICE in template instantiation with a pointered new
16503       operator
16504     * [36]9516 Internal error when using a big array
16505     * [37]9600 (c++) ICE with typedefs in template class
16506     * [38]9629 (c++) virtual inheritance segfault
16507     * [39]9672 (c++) ICE: Error reporting routines re-entered
16508     * [40]9749 (c++) ICE in write_expression on invalid function
16509       prototype
16510     * [41]9794 (fortran) ICE: floating point exception during constant
16511       folding
16512     * [42]9829 (c++) Missing colon in nested namespace usage causes ICE
16513     * [43]9916 (c++) ICE with noreturn function in ?: statement
16514     * [44]9936 ICE with local function and variable-length 2d array
16515     * [45]10262 (c++) cc1plus crashes with large generated code
16516     * [46]10278 (c++) ICE in parser for invalid code
16517     * [47]10446 (c++) ICE on definition of nonexistent member function of
16518       nested class in a class template
16519     * [48]10451 (c++) ICE in grokdeclarator on spurious mutable
16520       declaration
16521     * [49]10506 (c++) ICE in build_new at cp/init.c with
16522       -fkeep-inline-functions and multiple inheritance
16523     * [50]10549 (c++) ICE in store_bit_field on bitfields that exceed the
16524       precision of the declared type
16525
16526    Optimization bugs
16527
16528     * [51]2001 Inordinately long compile times in reload CSE regs
16529     * [52]2391 Exponential compilation time explosion in combine
16530     * [53]2960 Duplicate loop conditions even with -Os
16531     * [54]4046 redundant conditional branch
16532     * [55]6405 Loop-unrolling related performance regressions
16533     * [56]6798 very long compile time with large case-statement
16534     * [57]6871 const objects shouldn't be moved to .bss
16535     * [58]6909 problem w/ -Os on modified loop-2c.c test case
16536     * [59]7189 gcc -O2 -Wall does not print ``control reaches end of
16537       non-void function'' warning
16538     * [60]7642 optimization problem with signbit()
16539     * [61]8634 incorrect code for inlining of memcpy under -O2
16540     * [62]8750 Cygwin prolog generation erroneously emitting __alloca as
16541       regular function call
16542
16543    C front end
16544
16545     * [63]2161 long if-else cascade overflows parser stack
16546     * [64]4319 short accepted on typedef'd char
16547     * [65]8602 incorrect line numbers in warning messages when using
16548       inline functions
16549     * [66]9177 -fdump-translation-unit: C front end deletes function_decl
16550       AST nodes and breaks debugging dumps
16551     * [67]9853 miscompilation of non-constant structure initializer
16552
16553    c++ compiler and library
16554
16555     * [68]45 legal template specialization code is rejected (DUP:
16556       [69]3784)
16557     * [70]764 lookup failure: friend operator and dereferencing a pointer
16558       and templates (DUP: [71]5116)
16559     * [72]2862 gcc accepts invalid explicit instantiation syntax (DUP:
16560       2863)
16561     * [73]3663 G++ doesn't check access control during template
16562       instantiation
16563     * [74]3797 gcc fails to emit explicit specialization of a template
16564       member
16565     * [75]3948 Two destructors are called when no copy destructor is
16566       defined (ABI change)
16567     * [76]4137 Conversion operator within template is not accepted
16568     * [77]4361 bogus ambiguity taking the address of a member template
16569     * [78]4802 g++ accepts illegal template code (access to private
16570       member; DUP: [79]5837)
16571     * [80]4803 inline function is used but never defined, and g++ does
16572       not object
16573     * [81]5094 Partial specialization cannot be friend?
16574     * [82]5730 complex<double>::norm() -- huge slowdown from egcs-2.91.66
16575     * [83]6713 Regression wrt 3.0.4: g++ -O2 leads to seg fault at run
16576       time
16577     * [84]7015 certain __asm__ constructs rejected
16578     * [85]7086 compile time regression (quadratic behavior in
16579       fixup_var_refs)
16580     * [86]7099 G++ doesn't set the noreturn attribute on std::exit and
16581       std::abort
16582     * [87]7247 copy constructor missing when inlining enabled (invalid
16583       optimization?)
16584     * [88]7441 string array initialization compilation time regression
16585       from seconds to minutes
16586     * [89]7768 __PRETTY_FUNCTION__ for template destructor is wrong
16587     * [90]7804 bad printing of floating point constant in warning message
16588     * [91]8099 Friend classes and template specializations
16589     * [92]8117 member function pointers and multiple inheritance
16590     * [93]8205 using declaration and multiple inheritance
16591     * [94]8645 unnecessary non-zero checks in stl_tree.h
16592     * [95]8724 explicit destructor call for incomplete class allowed
16593     * [96]8805 compile time regression with many member variables
16594     * [97]8691 -O3 and -fno-implicit-templates are incompatible
16595     * [98]8700 unhelpful error message for binding temp to reference
16596     * [99]8724 explicit destructor call for incomplete class allowed
16597     * [100]8949 numeric_limits<>::denorm_min() and is_iec559 problems
16598     * [101]9016 Failure to consistently constant fold "constant" C++
16599       objects
16600     * [102]9053 g++ confused about ambiguity of overloaded function
16601       templates
16602     * [103]9152 undefined virtual thunks
16603     * [104]9182 basic_filebuf<> does not report errors in codecvt<>::out
16604     * [105]9297 data corruption due to codegen bug (when copying.)
16605     * [106]9318 i/ostream::operator>>/<<(streambuf*) broken
16606     * [107]9320 Incorrect usage of traits_type::int_type in stdio_filebuf
16607     * [108]9400 bogus -Wshadow warning: shadowed declaration of this in
16608       local classes
16609     * [109]9424 i/ostream::operator>>/<<(streambuf*) drops characters
16610     * [110]9425 filebuf::pbackfail broken (DUP: [111]9439)
16611     * [112]9474 GCC freezes in compiling a weird code mixing <iostream>
16612       and <iostream.h>
16613     * [113]9548 Incorrect results from setf(ios::fixed) and precision(-1)
16614       [114][DR 231]
16615     * [115]9555 ostream inserters fail to set badbit on exception
16616     * [116]9561 ostream inserters rethrow exception of wrong type
16617     * [117]9563 ostream::sentry returns true after a failed preparation
16618     * [118]9582 one-definition rule violation in std::allocator
16619     * [119]9622 __PRETTY_FUNCTION__ incorrect in template destructors
16620     * [120]9683 bug in initialization chains for static const variables
16621       from template classes
16622     * [121]9791 -Woverloaded-virtual reports hiding of destructor
16623     * [122]9817 collate::compare doesn't handle nul characters
16624     * [123]9825 filebuf::sputbackc breaks sbumpc
16625     * [124]9826 operator>>(basic_istream, basic_string) fails to compile
16626       with custom traits
16627     * [125]9924 Multiple using statements for builtin functions not
16628       allowed
16629     * [126]9946 destructor is not called for temporary object
16630     * [127]9964 filebuf::close() sometimes fails to close file
16631     * [128]9988 filebuf::overflow writes EOF to file
16632     * [129]10033 optimization breaks polymorphic references w/ typeid
16633       operator
16634     * [130]10097 filebuf::underflow drops characters
16635     * [131]10132 filebuf destructor can throw exceptions
16636     * [132]10180 gcc fails to warn about non-inlined function
16637     * [133]10199 method parametrized by template does not work everywhere
16638     * [134]10300 use of array-new (nothrow) in segfaults on NULL return
16639     * [135]10427 Stack corruption with variable-length automatic arrays
16640       and virtual destructors
16641     * [136]10503 Compilation never stops in fixed_type_or_null
16642
16643    Objective-C
16644
16645     * [137]5956 selectors aren't matched properly when added to the
16646       selector table
16647
16648    Fortran compiler and library
16649
16650     * [138]1832 list directed i/o overflow hangs, -fbounds-check doesn't
16651       detect
16652     * [139]3924 g77 generates code that is rejected by GAS if COFF debug
16653       info requested
16654     * [140]5634 doc: explain that configure --prefix=~/... does not work
16655     * [141]6367 multiple repeat counts confuse namelist read into array
16656     * [142]6491 Logical operations error on logicals when using
16657       -fugly-logint
16658     * [143]6742 Generation of C++ Prototype for FORTRAN and extern "C"
16659     * [144]7113 Failure of g77.f-torture/execute/f90-intrinsic-bit.f -Os
16660       on irix6.5
16661     * [145]7236 OPEN(...,RECL=nnn,...) without ACCESS='DIRECT' should
16662       assume a direct access file
16663     * [146]7278 g77 "bug"; the executable misbehaves (with -O2
16664       -fno-automatic)
16665     * [147]7384 DATE_AND_TIME milliseconds field inactive on Windows
16666     * [148]7388 Incorrect output with 0-based array of characters
16667     * [149]8587 Double complex zero ** double precision number -> NaN
16668       instead of zero
16669     * [150]9038 -ffixed-line-length-none -x f77-cpp-input gives: Warning:
16670       unknown register name line-length-none
16671     * [151]10197 Direct access files not unformatted by default
16672
16673    Java compiler and library
16674
16675     * [152]6005 gcj fails to build rhug on alpha
16676     * [153]6389 System.getProperty("") should always throw an
16677       IllegalArgumentException
16678     * [154]6576 java.util.ResourceBundle.getResource ignores locale
16679     * [155]6652 new java.io.File("").getCanonicalFile() throws exception
16680     * [156]7060 getMethod() doesn't search super interface
16681     * [157]7073 bytecode interpreter gives wrong answer for interface
16682       getSuperclass()
16683     * [158]7180 possible bug in
16684       javax.naming.spi.NamingManager.getPlusPath()
16685     * [159]7416 java.security startup refs "GNU libgcj.security"
16686     * [160]7570 Runtime.exec with null envp: child doesn't inherit parent
16687       env (DUP: [161]7578)
16688     * [162]7611 Internal error while compiling libjava with -O
16689     * [163]7709 NullPointerException in _Jv_ResolvePoolEntry
16690     * [164]7766 ZipInputStream.available returns 0 immediately after
16691       construction
16692     * [165]7785 Calendar.getTimeInMillis/setTimeInMillis should be public
16693     * [166]7786 TimeZone.getDSTSavings() from JDK1.4 not implemented
16694     * [167]8142 '$' in class names vs. dlopen 'dynamic string tokens'
16695     * [168]8234 ZipInputStream chokes when InputStream.read() returns
16696       small chunks
16697     * [169]8415 reflection bug: exception info for Method
16698     * [170]8481 java.Random.nextInt(int) may return negative
16699     * [171]8593 Error reading GZIPped files with BufferedReader
16700     * [172]8759 java.beans.Introspector has no flushCaches() or
16701       flushFromCaches() methods
16702     * [173]8997 spin() calls Thread.sleep
16703     * [174]9253 on win32, java.io.File.listFiles("C:\\") returns pwd
16704       instead of the root content of C:
16705     * [175]9254 java::lang::Object::wait(), threads-win32.cc returns
16706       wrong return codes
16707     * [176]9271 Severe bias in java.security.SecureRandom
16708
16709    Ada compiler and library
16710
16711     * [177]6767 make gnatlib-shared fails on -laddr2line
16712     * [178]9911 gnatmake fails to link when GCC configured with
16713       --with-sjlj-exceptions=yes
16714     * [179]10020 Can't bootstrap gcc on AIX with Ada enabled
16715     * [180]10546 Ada tasking not working on Red Hat 9
16716
16717    preprocessor
16718
16719     * [181]7029 preprocessor should ignore #warning with -M
16720
16721    ARM-specific
16722
16723     * [182]2903 [arm] Optimization bug with long long arithmetic
16724     * [183]7873 arm-linux-gcc fails when assigning address to a bit field
16725
16726    FreeBSD-specific
16727
16728     * [184]7680 float functions undefined in math.h/cmath with #define
16729       _XOPEN_SOURCE
16730
16731    HP-UX or HP-PA-specific
16732
16733     * [185]8705 [HP-PA] ICE in emit_move_insn_1, in expr.c
16734     * [186]9986 [HP-UX] Incorrect transformation of fputs_unlocked to
16735       fputc_unlocked
16736     * [187]10056 [HP-PA] ICE at -O2 when building c++ code from doxygen
16737
16738    m68hc11-specific
16739
16740     * [188]6744 Bad assembler code generated: reference to pseudo
16741       register z
16742     * [189]7361 Internal compiler error in reload_cse_simplify_operands,
16743       in reload1.c
16744
16745    MIPS-specific
16746
16747     * [190]9496 [mips-linux] bug in optimizer?
16748
16749    PowerPC-specific
16750
16751     * [191]7067 -Os with -mcpu=powerpc optimizes for speed (?) instead of
16752       space
16753     * [192]8480 reload ICEs for LAPACK code on powerpc64-linux
16754     * [193]8784 [AIX] Internal compiler error in simplify_gen_subreg
16755     * [194]10315 [powerpc] ICE: in extract_insn, in recog.c
16756
16757    SPARC-specific
16758
16759     * [195]10267 (documentation) Wrong build instructions for
16760       *-*-solaris2*
16761
16762    x86-specific (Intel/AMD)
16763
16764     * [196]7916 ICE in instantiate_virtual_register_1
16765     * [197]7926 (c++) i486 instructions in header files make c++ programs
16766       crash on i386
16767     * [198]8555 ICE in gen_split_1231
16768     * [199]8994 ICE with -O -march=pentium4
16769     * [200]9426 ICE with -fssa -funroll-loops -fprofile-arcs
16770     * [201]9806 ICE in inline assembly with -fPIC flag
16771     * [202]10077 gcc -msse2 generates movd to move dwords between xmm
16772       regs
16773     * [203]10233 64-bit comparison only comparing bottom 32-bits
16774     * [204]10286 type-punning doesn't work with __m64 and -O
16775     * [205]10308 [x86] ICE with -O -fgcse or -O2
16776     __________________________________________________________________
16777
16778GCC 3.3.1
16779
16780  Bug Fixes
16781
16782   This section lists the problem reports (PRs) from GCC's bug tracking
16783   system that are known to be fixed in the 3.3.1 release. This list might
16784   not be complete (that is, it is possible that some PRs that have been
16785   fixed are not listed here).
16786
16787    Bootstrap failures
16788
16789     * [206]11272 [Solaris] make bootstrap fails while building libstdc++
16790
16791    Internal compiler errors (multi-platform)
16792
16793     * [207]5754 ICE on invalid nested template class
16794     * [208]6597 ICE in set_mem_alias_set compiling Qt with -O2 on ia64
16795       and --enable-checking
16796     * [209]6949 (c++) ICE in tsubst_decl, in cp/pt.c
16797     * [210]7053 (c++) ICE when declaring a function already defined as a
16798       friend method of a template class
16799     * [211]8164 (c++) ICE when using different const expressions as
16800       template parameter
16801     * [212]8384 (c++) ICE in is_base_type, in dwarf2out.c
16802     * [213]9559 (c++) ICE with invalid initialization of a static const
16803     * [214]9649 (c++) ICE in finish_member_declaration, in cp/semantics.c
16804       when redeclaring a static member variable
16805     * [215]9864 (fortran) ICE in add_abstract_origin_attribute, in
16806       dwarfout.c with -g -O -finline-functions
16807     * [216]10432 (c++) ICE in poplevel, in cp/decl.c
16808     * [217]10475 ICE in subreg_highpart_offset for code with long long
16809     * [218]10635 (c++) ICE when dereferencing an incomplete type casted
16810       from a void pointer
16811     * [219]10661 (c++) ICE in instantiate_decl, in cp/pt.c while
16812       instantiating static member variables
16813     * [220]10700 ICE in copy_to_mode_reg on 64-bit targets
16814     * [221]10712 (c++) ICE in constructor_name_full, in cp/decl2.c
16815     * [222]10796 (c++) ICE when defining an enum with two values: -1 and
16816       MAX_INT_64BIT
16817     * [223]10890 ICE in merge_assigned_reloads building Linux 2.4.2x
16818       sched.c
16819     * [224]10939 (c++) ICE with template code
16820     * [225]10956 (c++) ICE when specializing a template member function
16821       of a template class, in tsubst, in cp/pt.c
16822     * [226]11041 (c++) ICE: const myclass &x = *x; (when operator*()
16823       defined)
16824     * [227]11059 (c++) ICE with empty union
16825     * [228]11083 (c++) ICE in commit_one_edge_insertion, in cfgrtl.c with
16826       -O2 -fnon-call-exceptions
16827     * [229]11105 (c++) ICE in mangle_conv_op_name_for_type
16828     * [230]11149 (c++) ICE on error when instantiation with call function
16829       of a base type
16830     * [231]11228 (c++) ICE on new-expression using array operator new and
16831       default-initialization
16832     * [232]11282 (c++) Infinite memory usage after syntax error
16833     * [233]11301 (fortran) ICE with -fno-globals
16834     * [234]11308 (c++) ICE when using an enum type name as if it were a
16835       class or namespace
16836     * [235]11473 (c++) ICE with -gstabs when empty struct inherits from
16837       an empty struct
16838     * [236]11503 (c++) ICE when instantiating template with ADDR_EXPR
16839     * [237]11513 (c++) ICE in push_template_decl_real, in cp/pt.c:
16840       template member functions
16841
16842    Optimization bugs
16843
16844     * [238]11198 -O2 -frename-registers generates wrong code (aliasing
16845       problem)
16846     * [239]11304 Wrong code production with -fomit-frame-pointer
16847     * [240]11381 volatile memory access optimized away
16848     * [241]11536 [strength-reduce] -O2 optimization produces wrong code
16849     * [242]11557 constant folding bug generates wrong code
16850
16851    C front end
16852
16853     * [243]5897 No warning for statement after return
16854     * [244]11279 DWARF-2 output mishandles large enums
16855
16856    Preprocessor bugs
16857
16858     * [245]11022 no warning for non-compatible macro redefinition
16859
16860    C++ compiler and library
16861
16862     * [246]2330 static_cast<>() to a private base is allowed
16863     * [247]5388 Incorrect message "operands to ?: have different types"
16864     * [248]5390 Libiberty fails to demangle multi-digit template
16865       parameters
16866     * [249]7877 Incorrect parameter passing to specializations of member
16867       function templates
16868     * [250]9393 Anonymous namespaces and compiling the same file twice
16869     * [251]10032 -pedantic converts some errors to warnings
16870     * [252]10468 const typeof(x) is non-const, but only in templates
16871     * [253]10527 confused error message with "new int()" parameter
16872       initializer
16873     * [254]10679 parameter MIN_INLINE_INSNS is not honored
16874     * [255]10682 gcc chokes on a typedef for an enum inside a class
16875       template
16876     * [256]10689 pow(std::complex(0),1/3) returns (nan, nan) instead of
16877       0.
16878     * [257]10845 template member function (with nested template as
16879       parameter) cannot be called anymore if another unrelated template
16880       member function is defined
16881     * [258]10849 Cannot define an out-of-class specialization of a
16882       private nested template class
16883     * [259]10888 Suppress -Winline warnings for system headers
16884     * [260]10929 -Winline warns about functions for which no definition
16885       is visible
16886     * [261]10931 valid conversion static_cast<const unsigned
16887       int&>(lvalue-of-type-int) is rejected
16888     * [262]10940 Bad code with explicit specialization
16889     * [263]10968 If member function implicitly instantiated, explicit
16890       instantiation of class fails to instantiate it
16891     * [264]10990 Cannot convert with dynamic_cast<> to a private base
16892       class from within a member function
16893     * [265]11039 Bad interaction between implicit typename deprecation
16894       and friendship
16895     * [266]11062 (libstdc++) avoid __attribute__ ((unused)); say
16896       "__unused__" instead
16897     * [267]11095 C++ iostream manipulator causes segfault when called
16898       with negative argument
16899     * [268]11098 g++ doesn't emit complete debugging information for
16900       local variables in destructors
16901     * [269]11137 GNU/Linux shared library constructors not called unless
16902       there's one global object
16903     * [270]11154 spurious ambiguity report for template class
16904       specialization
16905     * [271]11329 Compiler cannot find user defined implicit typecast
16906     * [272]11332 Spurious error with casts in ?: expression
16907     * [273]11431 static_cast behavior with subclasses when default
16908       constructor available
16909     * [274]11528 money_get facet does not accept "$.00" as valid
16910     * [275]11546 Type lookup problems in out-of-line definition of a
16911       class doubly nested from a template class
16912     * [276]11567 C++ code containing templated member function with same
16913       name as pure virtual member function results in linking failure
16914     * [277]11645 Failure to deal with using and private inheritance
16915
16916    Java compiler and library
16917
16918     * [278]5179 Qualified static field access doesn't initialize its
16919       class
16920     * [279]8204 gcj -O2 to native reorders certain instructions
16921       improperly
16922     * [280]10838 java.io.ObjectInputStream syntax error
16923     * [281]10886 The RMI registry that comes with GCJ does not work
16924       correctly
16925     * [282]11349 JNDI URL context factories not located correctly
16926
16927    x86-specific (Intel/AMD)
16928
16929     * [283]4823 ICE on inline assembly code
16930     * [284]8878 miscompilation with -O and SSE
16931     * [285]9815 (c++ library) atomicity.h - fails to compile with -O3
16932       -masm=intel
16933     * [286]10402 (inline assembly) [x86] ICE in merge_assigned_reloads,
16934       in reload1.c
16935     * [287]10504 ICE with SSE2 code and -O3 -mcpu=pentium4 -msse2
16936     * [288]10673 ICE for x86-64 on freebsd libc vfprintf.c source
16937     * [289]11044 [x86] out of range loop instructions for FP code on K6
16938     * [290]11089 ICE: instantiate_virtual_regs_lossage while using SSE
16939       built-ins
16940     * [291]11420 [x86_64] gcc generates invalid asm code when "-O -fPIC"
16941       is used
16942
16943    SPARC- or Solaris- specific
16944
16945     * [292]9362 solaris 'as' dies when fed .s and "-gstabs"
16946     * [293]10142 [SPARC64] gcc produces wrong code when passing
16947       structures by value
16948     * [294]10663 New configure check aborts with Sun tools.
16949     * [295]10835 combinatorial explosion in scheduler on HyperSPARC
16950     * [296]10876 ICE in calculate_giv_inc when building KDE
16951     * [297]10955 wrong code at -O3 for structure argument in context of
16952       structure return
16953     * [298]11018 -mcpu=ultrasparc busts tar-1.13.25
16954     * [299]11556 [sparc64] ICE in gen_reg_rtx() while compiling 2.6.x
16955       Linux kernel
16956
16957    ia64 specific
16958
16959     * [300]10907 gcc violates the ia64 ABI (GP must be preserved)
16960     * [301]11320 scheduler bug (in machine depended reorganization pass)
16961     * [302]11599 bug with conditional and __builtin_prefetch
16962
16963    PowerPC specific
16964
16965     * [303]9745 [powerpc] gcc mis-compiles libmcrypt (alias problem
16966       during loop)
16967     * [304]10871 error in rs6000_stack_info save_size computation
16968     * [305]11440 gcc mis-compiles c++ code (libkhtml) with -O2, -fno-gcse
16969       cures it
16970
16971    m68k-specific
16972
16973     * [306]7594 [m68k] ICE on legal code associated with simplify-rtx
16974     * [307]10557 [m68k] ICE in subreg_offset_representable_p
16975     * [308]11054 [m68k] ICE in reg_overlap_mentioned_p
16976
16977    ARM-specific
16978
16979     * [309]10834 [arm] GCC 3.3 still generates incorrect instructions for
16980       functions with __attribute__ ((interrupt ("IRQ")))
16981     * [310]10842 [arm] Clobbered link register is copied to pc under
16982       certain circumstances
16983     * [311]11052 [arm] noce_process_if_block() can lose REG_INC notes
16984     * [312]11183 [arm] ICE in change_address_1 (3.3) / subreg_hard_regno
16985       (3.4)
16986
16987    MIPS-specific
16988
16989     * [313]11084 ICE in propagate_one_insn, in flow.c
16990
16991    SH-specific
16992
16993     * [314]10331 can't compile c++ part of gcc cross compiler for sh-elf
16994     * [315]10413 [SH] ICE in reload_cse_simplify_operands, in reload1.c
16995     * [316]11096 i686-linux to sh-linux cross compiler fails to compile
16996       C++ files
16997
16998    GNU/Linux (or Hurd?) specific
16999
17000     * [317]2873 Bogus fixinclude of stdio.h from glibc 2.2.3
17001
17002    UnixWare specific
17003
17004     * [318]3163 configure bug: gcc/aclocal.m4 mmap test fails on UnixWare
17005       7.1.1
17006
17007    Cygwin (or mingw) specific
17008
17009     * [319]5287 ICE with dllimport attribute
17010     * [320]10148 [MingW/CygWin] Compiler dumps core
17011
17012    DJGPP specific
17013
17014     * [321]8787 GCC fails to emit .intel_syntax when invoked with
17015       -masm=intel on DJGPP
17016
17017    Darwin (and MacOS X) specific
17018
17019     * [322]10900 trampolines crash
17020
17021    Documentation
17022
17023     * [323]1607 (c++) Format attributes on methods undocumented
17024     * [324]4252 Invalid option `-fdump-translation-unit'
17025     * [325]4490 Clarify restrictions on -m96bit-long-double,
17026       -m128bit-long-double
17027     * [326]10355 document an issue with regparm attribute on some systems
17028       (e.g. Solaris)
17029     * [327]10726 (fortran) Documentation for function "IDate Intrinsic
17030       (Unix)" is wrong
17031     * [328]10805 document bug in old version of Sun assembler
17032     * [329]10815 warn against GNU binutils on AIX
17033     * [330]10877 document need for newer binutils on i?86-*-linux-gnu
17034     * [331]11280 Manual incorrect with respect to -freorder-blocks
17035     * [332]11466 Document -mlittle-endian and its restrictions for the
17036       sparc64 port
17037
17038    Testsuite bugs (compiler itself is not affected)
17039
17040     * [333]10737 newer bison causes g++.dg/parse/crash2.C to incorrectly
17041       report failure
17042     * [334]10810 gcc-3.3 fails make check: buffer overrun in
17043       test_demangle.c
17044     __________________________________________________________________
17045
17046GCC 3.3.2
17047
17048  Bug Fixes
17049
17050   This section lists the problem reports (PRs) from GCC's bug tracker
17051   that are known to be fixed in the 3.3.2 release. This list might not be
17052   complete (that is, it is possible that some PRs that have been fixed
17053   are not listed here).
17054
17055    Bootstrap failures and problems
17056
17057     * [335]8336 [SCO5] bootstrap config still tries to use COFF options
17058     * [336]9330 [alpha-osf] Bootstrap failure on Compaq Tru64 with
17059       --enable-threads=posix
17060     * [337]9631 [hppa64-linux] gcc-3.3 fails to bootstrap
17061     * [338]9877 fixincludes makes a bad sys/byteorder.h on svr5 (UnixWare
17062       7.1.1)
17063     * [339]11687 xstormy16-elf build fails in libf2c
17064     * [340]12263 [SGI IRIX] bootstrap fails during compile of
17065       libf2c/libI77/backspace.c
17066     * [341]12490 buffer overflow in scan-decls.c (during Solaris 9
17067       fix-header processing)
17068
17069    Internal compiler errors (multi-platform)
17070
17071     * [342]7277 Casting integers to vector types causes ICE
17072     * [343]7939 (c++) ICE on invalid function template specialization
17073     * [344]11063 (c++) ICE on parsing initialization list of const array
17074       member
17075     * [345]11207 ICE with negative index in array element designator
17076     * [346]11522 (fortran) g77 dwarf-2 ICE in
17077       add_abstract_origin_attribute
17078     * [347]11595 (c++) ICE on duplicate label definition
17079     * [348]11646 (c++) ICE in commit_one_edge_insertion with
17080       -fnon-call-exceptions -fgcse -O
17081     * [349]11665 ICE in struct initializer when taking address
17082     * [350]11852 (c++) ICE with bad struct initializer.
17083     * [351]11878 (c++) ICE in cp_expr_size
17084     * [352]11883 ICE with any -O on mercury-generated C code
17085     * [353]11991 (c++) ICE in cxx_incomplete_type_diagnostic, in
17086       cp/typeck2.c when applying typeid operator to template template
17087       parameter
17088     * [354]12146 ICE in lookup_template_function, in cp/pt.c
17089     * [355]12215 ICE in make_label_edge with -fnon-call-exceptions
17090       -fno-gcse -O2
17091     * [356]12369 (c++) ICE with templates and friends
17092     * [357]12446 ICE in emit_move_insn on complicated array reference
17093     * [358]12510 ICE in final_scan_insn
17094     * [359]12544 ICE with large parameters used in nested functions
17095
17096    C and optimization bugs
17097
17098     * [360]9862 spurious warnings with -W -finline-functions
17099     * [361]10962 lookup_field is a linear search on a linked list (can be
17100       slow if large struct)
17101     * [362]11370 -Wunreachable-code gives false complaints
17102     * [363]11637 invalid assembly with -fnon-call-exceptions
17103     * [364]11885 Problem with bitfields in packed structs
17104     * [365]12082 Inappropriate unreachable code warnings
17105     * [366]12180 Inline optimization fails for variadic function
17106     * [367]12340 loop unroller + gcse produces wrong code
17107
17108    C++ compiler and library
17109
17110     * [368]3907 nested template parameter collides with member name
17111     * [369]5293 confusing message when binding a temporary to a reference
17112     * [370]5296 [DR115] Pointers to functions and to template functions
17113       behave differently in deduction
17114     * [371]7939 ICE on function template specialization
17115     * [372]8656 Unable to assign function with __attribute__ and pointer
17116       return type to an appropriate variable
17117     * [373]10147 Confusing error message for invalid template function
17118       argument
17119     * [374]11400 std::search_n() makes assumptions about Size parameter
17120     * [375]11409 issues with using declarations, overloading, and
17121       built-in functions
17122     * [376]11740 ctype<wchar_t>::do_is(mask, wchar_t) doesn't handle
17123       multiple bits in mask
17124     * [377]11786 operator() call on variable in other namespace not
17125       recognized
17126     * [378]11867 static_cast ignores ambiguity
17127     * [379]11928 bug with conversion operators that are typedefs
17128     * [380]12114 Uninitialized memory accessed in dtor
17129     * [381]12163 static_cast + explicit constructor regression
17130     * [382]12181 Wrong code with comma operator and c++
17131     * [383]12236 regparm and fastcall messes up parameters
17132     * [384]12266 incorrect instantiation of unneeded template during
17133       overload resolution
17134     * [385]12296 istream::peek() doesn't set eofbit
17135     * [386]12298 [sjlj exceptions] Stack unwind destroys
17136       not-yet-constructed object
17137     * [387]12369 ICE with templates and friends
17138     * [388]12337 apparently infinite loop in g++
17139     * [389]12344 stdcall attribute ignored if function returns a pointer
17140     * [390]12451 missing(late) class forward declaration in cxxabi.h
17141     * [391]12486 g++ accepts invalid use of a qualified name
17142
17143    x86 specific (Intel/AMD)
17144
17145     * [392]8869 [x86 MMX] ICE with const variable optimization and MMX
17146       builtins
17147     * [393]9786 ICE in fixup_abnormal_edges with -fnon-call-exceptions
17148       -O2
17149     * [394]11689 g++3.3 emits un-assembleable code for k6 architecture
17150     * [395]12116 [k6] Invalid assembly output values with X-MAME code
17151     * [396]12070 ICE converting between double and long double with
17152       -msoft-float
17153
17154    ia64-specific
17155
17156     * [397]11184 [ia64 hpux] ICE on __builtin_apply building libobjc
17157     * [398]11535 __builtin_return_address may not work on ia64
17158     * [399]11693 [ia64] ICE in gen_nop_type
17159     * [400]12224 [ia64] Thread-local storage doesn't work
17160
17161    PowerPC-specific
17162
17163     * [401]11087 [powerpc64-linux] GCC miscompiles raid1.c from linux
17164       kernel
17165     * [402]11319 loop miscompiled on ppc32
17166     * [403]11949 ICE Compiler segfault with ffmpeg -maltivec code
17167
17168    SPARC-specific
17169
17170     * [404]11662 wrong code for expr. with cast to long long and
17171       exclusive or
17172     * [405]11965 invalid assembler code for a shift < 32 operation
17173     * [406]12301 (c++) stack corruption when a returned expression throws
17174       an exception
17175
17176    Alpha-specific
17177
17178     * [407]11717 [alpha-linux] unrecognizable insn compiling for.c of
17179       kernel 2.4.22-pre8
17180
17181    HPUX-specific
17182
17183     * [408]11313 problem with #pragma weak and static inline functions
17184     * [409]11712 __STDC_EXT__ not defined for C++ by default anymore?
17185
17186    Solaris specific
17187
17188     * [410]12166 Profiled programs crash if PROFDIR is set
17189
17190    Solaris-x86 specific
17191
17192     * [411]12101 i386 Solaris no longer works with GNU as?
17193
17194    Miscellaneous embedded target-specific bugs
17195
17196     * [412]10988 [m32r-elf] wrong blockmove code with -O3
17197     * [413]11805 [h8300-unknown-coff] [H8300] ICE for simple code with
17198       -O2
17199     * [414]11902 [sh4] spec file improperly inserts rpath even when none
17200       needed
17201     * [415]11903 [sh4] -pthread fails to link due to error in spec file
17202       on sh4
17203     __________________________________________________________________
17204
17205GCC 3.3.3
17206
17207  Minor features
17208
17209   In addition to the bug fixes documented below, this release contains
17210   few minor features such as:
17211     * Support for --with-sysroot
17212     * Support for automatic detection of executable stacks
17213     * Support for SSE3 instructions
17214     * Support for thread local storage debugging under GDB on S390
17215
17216  Bug Fixes
17217
17218   This section lists the problem reports (PRs) from GCC's bug tracker
17219   that are known to be fixed in the 3.3.3 release. This list might not be
17220   complete (that is, it is possible that some PRs that have been fixed
17221   are not listed here).
17222
17223    Bootstrap failures and issues
17224
17225     * [416]11890 Building cross gcc-3.3.1 for sparc-sun-solaris2.6 fails
17226     * [417]12399 boehm-gc fails (when building a cross compiler): libtool
17227       unable to infer tagged configuration
17228     * [418]13068 mklibgcc.in doesn't handle multi-level multilib
17229       subdirectories properly
17230
17231    Internal compiler errors (multi-platform)
17232
17233     * [419]10060 ICE (stack overflow) on huge file (300k lines) due to
17234       recursive behaviour of copy_rtx_if_shared, in emit_rtl.c
17235     * [420]10555 (c++) ICE on undefined template argument
17236     * [421]10706 (c++) ICE in mangle_class_name_for_template
17237     * [422]11496 (fortran) error in flow_loops_find when -funroll-loops
17238       active
17239     * [423]11741 ICE in pre_insert_copy_insn, in gcse.c
17240     * [424]12440 GCC crashes during compilation of quicktime4linux 2.0.0
17241     * [425]12632 (fortran) -fbounds-check ICE
17242     * [426]12712 (c++) ICE on short legit C++ code fragment with gcc
17243       3.3.2
17244     * [427]12726 (c++) ICE (segfault) on trivial code
17245     * [428]12890 (c++) ICE on compilation of class with throwing method
17246     * [429]12900 (c++) ICE in rtl_verify_flow_info_1
17247     * [430]13060 (fortran) ICE in fixup_var_refs_1, in function.c on
17248       correct code with -O2 -fno-force-mem
17249     * [431]13289 (c++) ICE in regenerate_decl_from_template on recursive
17250       template
17251     * [432]13318 ICE: floating point exception in the loop optimizer
17252     * [433]13392 (c++) ICE in convert_from_eh_region_ranges_1, in
17253       except.c
17254     * [434]13574 (c++) invalid array default initializer in class lets
17255       gcc consume all memory and die
17256     * [435]13475 ICE on SIMD variables with partial value initialization
17257     * [436]13797 (c++) ICE on invalid template parameter
17258     * [437]13824 (java) gcj SEGV with simple .java program
17259
17260    C and optimization bugs
17261
17262     * [438]8776 loop invariants are not removed (most likely)
17263     * [439]10339 [sparc,ppc,ppc64] Invalid optimization: replacing
17264       strncmp by memcmp
17265     * [440]11350 undefined labels with -Os -fPIC
17266     * [441]12826 Optimizer removes reference through volatile pointer
17267     * [442]12500 stabs debug info: void no longer a predefined / builtin
17268       type
17269     * [443]12941 builtin-bitops-1.c miscompilation (latent bug)
17270     * [444]12953 tree inliner bug (in inline_forbidden_p) and fix
17271     * [445]13041 linux-2.6/sound/core/oss/rate.c miscompiled
17272     * [446]13507 spurious printf format warning
17273     * [447]13382 Type information for const pointer disappears during
17274       optimization.
17275     * [448]13394 noreturn attribute ignored on recursive invokation
17276     * [449]13400 Compiled code crashes storing to read-only location
17277     * [450]13521 Endless loop in calculate_global_regs_live
17278
17279    C++ compiler and library
17280
17281   Some of the bug fixes in this list were made to implement decisions
17282   that the ISO C++ standards committee has made concerning several defect
17283   reports (DRs). Links in the list below point to detailed discussion of
17284   the relevant defect report.
17285     * [451]2094 unimplemented: use of `ptrmem_cst' in template type
17286       unification
17287     * [452]2294 using declaration confusion
17288     * [453]5050 template instantiation depth exceeds limit: recursion
17289       problem?
17290     * [454]9371 Bad exception handling in
17291       i/ostream::operator>>/<<(streambuf*)
17292     * [455]9546 bad exception handling in ostream members
17293     * [456]10081 basic_ios::_M_cache_locale leaves NULL members in the
17294       face of unknown locales
17295     * [457]10093 [458][DR 61] Setting failbit in exceptions doesn't work
17296     * [459]10095 istream::operator>>(int&) sets ios::badbit when
17297       ios::failbit is set.
17298     * [460]11554 Warning about reordering of initializers doesn't mention
17299       location of constructor
17300     * [461]12297 istream::sentry::sentry() handles eof() incorrectly.
17301     * [462]12352 Exception safety problems in src/localename.cc
17302     * [463]12438 Memory leak in locale::combine()
17303     * [464]12540 Memory leak in locale::locale(const char*)
17304     * [465]12594 DRs [466]60 [TC] and [467]63 [TC] not implemented
17305     * [468]12657 Resolution of [469]DR 292 (WP) still unimplemented
17306     * [470]12696 memory eating infinite loop in diagnostics (error
17307       recovery problem)
17308     * [471]12815 Code compiled with optimization behaves unexpectedly
17309     * [472]12862 Conflicts between typedefs/enums and namespace member
17310       declarations
17311     * [473]12926 Wrong value after assignment in initialize list using
17312       bit-fields
17313     * [474]12967 Resolution of [475]DR 300 [WP] still unimplemented
17314     * [476]12971 Resolution of [477]DR 328 [WP] still unimplemented
17315     * [478]13007 basic_streambuf::pubimbue, imbue wrong
17316     * [479]13009 Implicitly-defined assignment operator writes to wrong
17317       memory
17318     * [480]13057 regparm attribute not applied to destructor
17319     * [481]13070 -Wformat option ignored in g++
17320     * [482]13081 forward template declarations in <complex> let inlining
17321       fail
17322     * [483]13239 Assertion does not seem to work correctly anymore
17323     * [484]13262 "xxx is private within this context" when initializing a
17324       self-contained template class
17325     * [485]13290 simple typo in concept checking for std::generate_n
17326     * [486]13323 Template code does not compile in presence of typedef
17327     * [487]13369 __verify_grouping (and __add_grouping?) not correct
17328     * [488]13371 infinite loop with packed struct and inlining
17329     * [489]13445 Template argument replacement "dereferences" a typedef
17330     * [490]13461 Fails to access protected-ctor from public constant
17331     * [491]13462 Non-standard-conforming type set::pointer
17332     * [492]13478 gcc uses wrong constructor to initialize a const
17333       reference
17334     * [493]13544 "conflicting types" for enums in different scopes
17335     * [494]13650 string::compare should not (always) use
17336       traits_type::length()
17337     * [495]13683 bogus warning about passing non-PODs through ellipsis
17338     * [496]13688 Derived class is denied access to protected base class
17339       member class
17340     * [497]13774 Member variable cleared in virtual multiple inheritance
17341       class
17342     * [498]13884 Protect sstream.tcc from extern template use
17343
17344    Java compiler and library
17345
17346     * [499]10746 [win32] garbage collection crash in GCJ
17347
17348    Objective-C compiler and library
17349
17350     * [500]11433 Crash due to dereferencing null pointer when querying
17351       protocol
17352
17353    Fortran compiler and library
17354
17355     * [501]12633 logical expression gives incorrect result with
17356       -fugly-logint option
17357     * [502]13037 [gcse-lm] g77 generates incorrect code
17358     * [503]13213 Hex constant problem when compiling with -fugly-logint
17359       and -ftypeless-boz
17360
17361    x86-specific (Intel/AMD)
17362
17363     * [504]4490 ICE with -m128bit-long-double
17364     * [505]12292 [x86_64] ICE: RTL check: expected code `const_int', have
17365       `reg' in make_field_assignment, in combine.c
17366     * [506]12441 ICE: can't find a register to spill
17367     * [507]12943 array static-init failure under -fpic, -fPIC
17368     * [508]13608 Incorrect code with -O3 -ffast-math
17369
17370    PowerPC-specific
17371
17372     * [509]11598 testcase gcc.dg/20020118-1.c fails runtime check of
17373       __attribute__((aligned(16)))
17374     * [510]11793 ICE in extract_insn, in recog.c (const_vector's)
17375     * [511]12467 vmsumubm emitted when vmsummbm appropriate (typo in
17376       altivec.md)
17377     * [512]12537 g++ generates writeable text sections
17378
17379    SPARC-specific
17380
17381     * [513]12496 wrong result for __atomic_add(&value, -1) when using -O0
17382       -m64
17383     * [514]12865 mprotect call to make trampoline executable may fail
17384     * [515]13354 ICE in sparc_emit_set_const32
17385
17386    ARM-specific
17387
17388     * [516]10467 [arm] ICE in pre_insert_copy_insn,
17389
17390    ia64-specific
17391
17392     * [517]11226 ICE passing struct arg with two floats
17393     * [518]11227 ICE for _Complex float, _Complex long double args
17394     * [519]12644 GCC 3.3.2 fails to compile glibc on ia64
17395     * [520]13149 build gcc-3.3.2 1305 error:unrecognizable insn
17396     * Various fixes for libunwind
17397
17398    Alpha-specific
17399
17400     * [521]12654 Incorrect comparison code generated for Alpha
17401     * [522]12965 SEGV+ICE in cc1plus on alpha-linux with -O2
17402     * [523]13031 ICE (unrecognizable insn) when building gnome-libs-1.4.2
17403
17404    HPPA-specific
17405
17406     * [524]11634 [hppa] ICE in verify_local_live_at_start, in flow.c
17407     * [525]12158 [hppa] compilation does not terminate at -O1
17408
17409    S390-specific
17410
17411     * [526]11992 Wrong built-in code for memcmp with length 1<<24: only
17412       (1<<24)-1 possible for CLCL-Instruction
17413
17414    SH-specific
17415
17416     * [527]9365 segfault in gen_far_branch (config/sh/sh.c)
17417     * [528]10392 optimizer generates faulty array indexing
17418     * [529]11322 SH profiler outputs multiple definitions of symbol
17419     * [530]13069 gcc/config/sh/rtems.h broken
17420     * [531]13302 Putting a va_list in a struct causes seg fault
17421     * [532]13585 Incorrect optimization of call to sfunc
17422     * Fix inappropriately exported libgcc functions from the shared
17423       library
17424
17425    Other embedded target specific
17426
17427     * [533]8916 [mcore] unsigned char assign gets hosed.
17428     * [534]11576 [h8300] ICE in change_address_1, in emit-rtl.c
17429     * [535]13122 [h8300] local variable gets corrupted by function call
17430       when -fomit-frame-pointer is given
17431     * [536]13256 [cris] strict_low_part mistreated in delay slots
17432     * [537]13373 [mcore] optimization with -frerun-cse-after-loop
17433       -fexpensive-optimizations produces wrong code on mcore
17434
17435    GNU HURD-specific
17436
17437     * [538]12561 gcc/config/t-gnu needs updating to work with
17438       --with-sysroot
17439
17440    Tru64 Unix specific
17441
17442     * [539]6243 testsuite fails almost all tests due to no libintl in
17443       LD_LIBRARY_PATH during test.
17444     * [540]11397 weak aliases broken on Tru64 UNIX
17445
17446    AIX-specific
17447
17448     * [541]12505 build failure due to defines of uchar in cpphash.h and
17449       sys/types.h
17450     * [542]13150 WEAK symbols not exported by collect2
17451
17452    IRIX-specific
17453
17454     * [543]12666 fixincludes problem on IRIX 6.5.19m
17455
17456    Solaris-specific
17457
17458     * [544]12969 Including sys/byteorder.h breaks configure checks
17459
17460    Testsuite problems (compiler is not affected)
17461
17462     * [545]10819 testsuite creates CR+LF on compiler version lines in
17463       test summary files
17464     * [546]11612 abi_check not finding correct libgcc_s.so.1
17465
17466    Miscellaneous
17467
17468     * [547]13211 using -###, incorrect warnings about unused linker file
17469       are produced
17470     __________________________________________________________________
17471
17472GCC 3.3.4
17473
17474   This is the [548]list of problem reports (PRs) from GCC's bug tracking
17475   system that are known to be fixed in the 3.3.4 release. This list might
17476   not be complete (that is, it is possible that some PRs that have been
17477   fixed are not listed here).
17478     __________________________________________________________________
17479
17480GCC 3.3.5
17481
17482   This is the [549]list of problem reports (PRs) from GCC's bug tracking
17483   system that are known to be fixed in the 3.3.5 release. This list might
17484   not be complete (that is, it is possible that some PRs that have been
17485   fixed are not listed here).
17486     __________________________________________________________________
17487
17488GCC 3.3.6
17489
17490   This is the [550]list of problem reports (PRs) from GCC's bug tracking
17491   system that are known to be fixed in the 3.3.6 release. This list might
17492   not be complete (that is, it is possible that some PRs that have been
17493   fixed are not listed here).
17494
17495
17496    For questions related to the use of GCC, please consult these web
17497    pages and the [551]GCC manuals. If that fails, the
17498    [552]gcc-help@gcc.gnu.org mailing list might help. Comments on these
17499    web pages and the development of GCC are welcome on our developer
17500    list at [553]gcc@gcc.gnu.org. All of [554]our lists have public
17501    archives.
17502
17503   Copyright (C) [555]Free Software Foundation, Inc. Verbatim copying and
17504   distribution of this entire article is permitted in any medium,
17505   provided this notice is preserved.
17506
17507   These pages are [556]maintained by the GCC team. Last modified
17508   2019-11-28[557].
17509
17510References
17511
17512   1. http://gcc.gnu.org/gcc-3.3/changes.html#3.3.6
17513   2. http://gcc.gnu.org/gcc-3.1/changes.html#obsolete_systems
17514   3. http://gcc.gnu.org/gcc-3.3/changes.html#obsolete_systems
17515   4. http://gcc.gnu.org/gcc-3.3/changes.html#nonnull_attribute
17516   5. http://gcc.gnu.org/news/dfa.html
17517   6. http://gcc.gnu.org/c99status.html
17518   7. https://gcc.gnu.org/onlinedocs/gcc-3.3.6/g77/News.html
17519   8. https://gcc.gnu.org/PR10140
17520   9. https://gcc.gnu.org/PR10198
17521  10. https://gcc.gnu.org/PR10338
17522  11. https://gcc.gnu.org/PR3581
17523  12. https://gcc.gnu.org/PR4382
17524  13. https://gcc.gnu.org/PR5533
17525  14. https://gcc.gnu.org/PR6387
17526  15. https://gcc.gnu.org/PR6412
17527  16. https://gcc.gnu.org/PR6620
17528  17. https://gcc.gnu.org/PR6663
17529  18. https://gcc.gnu.org/PR7068
17530  19. https://gcc.gnu.org/PR7083
17531  20. https://gcc.gnu.org/PR7647
17532  21. https://gcc.gnu.org/PR7675
17533  22. https://gcc.gnu.org/PR7718
17534  23. https://gcc.gnu.org/PR8116
17535  24. https://gcc.gnu.org/PR8358
17536  25. https://gcc.gnu.org/PR8511
17537  26. https://gcc.gnu.org/PR8564
17538  27. https://gcc.gnu.org/PR8660
17539  28. https://gcc.gnu.org/PR8766
17540  29. https://gcc.gnu.org/PR8803
17541  30. https://gcc.gnu.org/PR8846
17542  31. https://gcc.gnu.org/PR8906
17543  32. https://gcc.gnu.org/PR9216
17544  33. https://gcc.gnu.org/PR9261
17545  34. https://gcc.gnu.org/PR9263
17546  35. https://gcc.gnu.org/PR9429
17547  36. https://gcc.gnu.org/PR9516
17548  37. https://gcc.gnu.org/PR9600
17549  38. https://gcc.gnu.org/PR9629
17550  39. https://gcc.gnu.org/PR9672
17551  40. https://gcc.gnu.org/PR9749
17552  41. https://gcc.gnu.org/PR9794
17553  42. https://gcc.gnu.org/PR9829
17554  43. https://gcc.gnu.org/PR9916
17555  44. https://gcc.gnu.org/PR9936
17556  45. https://gcc.gnu.org/PR10262
17557  46. https://gcc.gnu.org/PR10278
17558  47. https://gcc.gnu.org/PR10446
17559  48. https://gcc.gnu.org/PR10451
17560  49. https://gcc.gnu.org/PR10506
17561  50. https://gcc.gnu.org/PR10549
17562  51. https://gcc.gnu.org/PR2001
17563  52. https://gcc.gnu.org/PR2391
17564  53. https://gcc.gnu.org/PR2960
17565  54. https://gcc.gnu.org/PR4046
17566  55. https://gcc.gnu.org/PR6405
17567  56. https://gcc.gnu.org/PR6798
17568  57. https://gcc.gnu.org/PR6871
17569  58. https://gcc.gnu.org/PR6909
17570  59. https://gcc.gnu.org/PR7189
17571  60. https://gcc.gnu.org/PR7642
17572  61. https://gcc.gnu.org/PR8634
17573  62. https://gcc.gnu.org/PR8750
17574  63. https://gcc.gnu.org/PR2161
17575  64. https://gcc.gnu.org/PR4319
17576  65. https://gcc.gnu.org/PR8602
17577  66. https://gcc.gnu.org/PR9177
17578  67. https://gcc.gnu.org/PR9853
17579  68. https://gcc.gnu.org/PR45
17580  69. https://gcc.gnu.org/PR3784
17581  70. https://gcc.gnu.org/PR764
17582  71. https://gcc.gnu.org/PR5116
17583  72. https://gcc.gnu.org/PR2862
17584  73. https://gcc.gnu.org/PR3663
17585  74. https://gcc.gnu.org/PR3797
17586  75. https://gcc.gnu.org/PR3948
17587  76. https://gcc.gnu.org/PR4137
17588  77. https://gcc.gnu.org/PR4361
17589  78. https://gcc.gnu.org/PR4802
17590  79. https://gcc.gnu.org/PR5837
17591  80. https://gcc.gnu.org/PR4803
17592  81. https://gcc.gnu.org/PR5094
17593  82. https://gcc.gnu.org/PR5730
17594  83. https://gcc.gnu.org/PR6713
17595  84. https://gcc.gnu.org/PR7015
17596  85. https://gcc.gnu.org/PR7086
17597  86. https://gcc.gnu.org/PR7099
17598  87. https://gcc.gnu.org/PR7247
17599  88. https://gcc.gnu.org/PR7441
17600  89. https://gcc.gnu.org/PR7768
17601  90. https://gcc.gnu.org/PR7804
17602  91. https://gcc.gnu.org/PR8099
17603  92. https://gcc.gnu.org/PR8117
17604  93. https://gcc.gnu.org/PR8205
17605  94. https://gcc.gnu.org/PR8645
17606  95. https://gcc.gnu.org/PR8724
17607  96. https://gcc.gnu.org/PR8805
17608  97. https://gcc.gnu.org/PR8691
17609  98. https://gcc.gnu.org/PR8700
17610  99. https://gcc.gnu.org/PR8724
17611 100. https://gcc.gnu.org/PR8949
17612 101. https://gcc.gnu.org/PR9016
17613 102. https://gcc.gnu.org/PR9053
17614 103. https://gcc.gnu.org/PR9152
17615 104. https://gcc.gnu.org/PR9182
17616 105. https://gcc.gnu.org/PR9297
17617 106. https://gcc.gnu.org/PR9318
17618 107. https://gcc.gnu.org/PR9320
17619 108. https://gcc.gnu.org/PR9400
17620 109. https://gcc.gnu.org/PR9424
17621 110. https://gcc.gnu.org/PR9425
17622 111. https://gcc.gnu.org/PR9439
17623 112. https://gcc.gnu.org/PR9474
17624 113. https://gcc.gnu.org/PR9548
17625 114. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#231
17626 115. https://gcc.gnu.org/PR9555
17627 116. https://gcc.gnu.org/PR9561
17628 117. https://gcc.gnu.org/PR9563
17629 118. https://gcc.gnu.org/PR9582
17630 119. https://gcc.gnu.org/PR9622
17631 120. https://gcc.gnu.org/PR9683
17632 121. https://gcc.gnu.org/PR9791
17633 122. https://gcc.gnu.org/PR9817
17634 123. https://gcc.gnu.org/PR9825
17635 124. https://gcc.gnu.org/PR9826
17636 125. https://gcc.gnu.org/PR9924
17637 126. https://gcc.gnu.org/PR9946
17638 127. https://gcc.gnu.org/PR9964
17639 128. https://gcc.gnu.org/PR9988
17640 129. https://gcc.gnu.org/PR10033
17641 130. https://gcc.gnu.org/PR10097
17642 131. https://gcc.gnu.org/PR10132
17643 132. https://gcc.gnu.org/PR10180
17644 133. https://gcc.gnu.org/PR10199
17645 134. https://gcc.gnu.org/PR10300
17646 135. https://gcc.gnu.org/PR10427
17647 136. https://gcc.gnu.org/PR10503
17648 137. https://gcc.gnu.org/PR5956
17649 138. https://gcc.gnu.org/PR1832
17650 139. https://gcc.gnu.org/PR3924
17651 140. https://gcc.gnu.org/PR5634
17652 141. https://gcc.gnu.org/PR6367
17653 142. https://gcc.gnu.org/PR6491
17654 143. https://gcc.gnu.org/PR6742
17655 144. https://gcc.gnu.org/PR7113
17656 145. https://gcc.gnu.org/PR7236
17657 146. https://gcc.gnu.org/PR7278
17658 147. https://gcc.gnu.org/PR7384
17659 148. https://gcc.gnu.org/PR7388
17660 149. https://gcc.gnu.org/PR8587
17661 150. https://gcc.gnu.org/PR9038
17662 151. https://gcc.gnu.org/PR10197
17663 152. https://gcc.gnu.org/PR6005
17664 153. https://gcc.gnu.org/PR6389
17665 154. https://gcc.gnu.org/PR6576
17666 155. https://gcc.gnu.org/PR6652
17667 156. https://gcc.gnu.org/PR7060
17668 157. https://gcc.gnu.org/PR7073
17669 158. https://gcc.gnu.org/PR7180
17670 159. https://gcc.gnu.org/PR7416
17671 160. https://gcc.gnu.org/PR7570
17672 161. https://gcc.gnu.org/PR7578
17673 162. https://gcc.gnu.org/PR7611
17674 163. https://gcc.gnu.org/PR7709
17675 164. https://gcc.gnu.org/PR7766
17676 165. https://gcc.gnu.org/PR7785
17677 166. https://gcc.gnu.org/PR7786
17678 167. https://gcc.gnu.org/PR8142
17679 168. https://gcc.gnu.org/PR8234
17680 169. https://gcc.gnu.org/PR8415
17681 170. https://gcc.gnu.org/PR8481
17682 171. https://gcc.gnu.org/PR8593
17683 172. https://gcc.gnu.org/PR8759
17684 173. https://gcc.gnu.org/PR8997
17685 174. https://gcc.gnu.org/PR9253
17686 175. https://gcc.gnu.org/PR9254
17687 176. https://gcc.gnu.org/PR9271
17688 177. https://gcc.gnu.org/PR6767
17689 178. https://gcc.gnu.org/PR9911
17690 179. https://gcc.gnu.org/PR10020
17691 180. https://gcc.gnu.org/PR10546
17692 181. https://gcc.gnu.org/PR7029
17693 182. https://gcc.gnu.org/PR2903
17694 183. https://gcc.gnu.org/PR7873
17695 184. https://gcc.gnu.org/PR7680
17696 185. https://gcc.gnu.org/PR8705
17697 186. https://gcc.gnu.org/PR9986
17698 187. https://gcc.gnu.org/PR10056
17699 188. https://gcc.gnu.org/PR6744
17700 189. https://gcc.gnu.org/PR7361
17701 190. https://gcc.gnu.org/PR9496
17702 191. https://gcc.gnu.org/PR7067
17703 192. https://gcc.gnu.org/PR8480
17704 193. https://gcc.gnu.org/PR8784
17705 194. https://gcc.gnu.org/PR10315
17706 195. https://gcc.gnu.org/PR10267
17707 196. https://gcc.gnu.org/PR7916
17708 197. https://gcc.gnu.org/PR7926
17709 198. https://gcc.gnu.org/PR8555
17710 199. https://gcc.gnu.org/PR8994
17711 200. https://gcc.gnu.org/PR9426
17712 201. https://gcc.gnu.org/PR9806
17713 202. https://gcc.gnu.org/PR10077
17714 203. https://gcc.gnu.org/PR10233
17715 204. https://gcc.gnu.org/PR10286
17716 205. https://gcc.gnu.org/PR10308
17717 206. https://gcc.gnu.org/PR11272
17718 207. https://gcc.gnu.org/PR5754
17719 208. https://gcc.gnu.org/PR6597
17720 209. https://gcc.gnu.org/PR6949
17721 210. https://gcc.gnu.org/PR7053
17722 211. https://gcc.gnu.org/PR8164
17723 212. https://gcc.gnu.org/PR8384
17724 213. https://gcc.gnu.org/PR9559
17725 214. https://gcc.gnu.org/PR9649
17726 215. https://gcc.gnu.org/PR9864
17727 216. https://gcc.gnu.org/PR10432
17728 217. https://gcc.gnu.org/PR10475
17729 218. https://gcc.gnu.org/PR10635
17730 219. https://gcc.gnu.org/PR10661
17731 220. https://gcc.gnu.org/PR10700
17732 221. https://gcc.gnu.org/PR10712
17733 222. https://gcc.gnu.org/PR10796
17734 223. https://gcc.gnu.org/PR10890
17735 224. https://gcc.gnu.org/PR10939
17736 225. https://gcc.gnu.org/PR10956
17737 226. https://gcc.gnu.org/PR11041
17738 227. https://gcc.gnu.org/PR11059
17739 228. https://gcc.gnu.org/PR11083
17740 229. https://gcc.gnu.org/PR11105
17741 230. https://gcc.gnu.org/PR11149
17742 231. https://gcc.gnu.org/PR11228
17743 232. https://gcc.gnu.org/PR11282
17744 233. https://gcc.gnu.org/PR11301
17745 234. https://gcc.gnu.org/PR11308
17746 235. https://gcc.gnu.org/PR11473
17747 236. https://gcc.gnu.org/PR11503
17748 237. https://gcc.gnu.org/PR11513
17749 238. https://gcc.gnu.org/PR11198
17750 239. https://gcc.gnu.org/PR11304
17751 240. https://gcc.gnu.org/PR11381
17752 241. https://gcc.gnu.org/PR11536
17753 242. https://gcc.gnu.org/PR11557
17754 243. https://gcc.gnu.org/PR5897
17755 244. https://gcc.gnu.org/PR11279
17756 245. https://gcc.gnu.org/PR11022
17757 246. https://gcc.gnu.org/PR2330
17758 247. https://gcc.gnu.org/PR5388
17759 248. https://gcc.gnu.org/PR5390
17760 249. https://gcc.gnu.org/PR7877
17761 250. https://gcc.gnu.org/PR9393
17762 251. https://gcc.gnu.org/PR10032
17763 252. https://gcc.gnu.org/PR10468
17764 253. https://gcc.gnu.org/PR10527
17765 254. https://gcc.gnu.org/PR10679
17766 255. https://gcc.gnu.org/PR10682
17767 256. https://gcc.gnu.org/PR10689
17768 257. https://gcc.gnu.org/PR10845
17769 258. https://gcc.gnu.org/PR10849
17770 259. https://gcc.gnu.org/PR10888
17771 260. https://gcc.gnu.org/PR10929
17772 261. https://gcc.gnu.org/PR10931
17773 262. https://gcc.gnu.org/PR10940
17774 263. https://gcc.gnu.org/PR10968
17775 264. https://gcc.gnu.org/PR10990
17776 265. https://gcc.gnu.org/PR11039
17777 266. https://gcc.gnu.org/PR11062
17778 267. https://gcc.gnu.org/PR11095
17779 268. https://gcc.gnu.org/PR11098
17780 269. https://gcc.gnu.org/PR11137
17781 270. https://gcc.gnu.org/PR11154
17782 271. https://gcc.gnu.org/PR11329
17783 272. https://gcc.gnu.org/PR11332
17784 273. https://gcc.gnu.org/PR11431
17785 274. https://gcc.gnu.org/PR11528
17786 275. https://gcc.gnu.org/PR11546
17787 276. https://gcc.gnu.org/PR11567
17788 277. https://gcc.gnu.org/PR11645
17789 278. https://gcc.gnu.org/PR5179
17790 279. https://gcc.gnu.org/PR8204
17791 280. https://gcc.gnu.org/PR10838
17792 281. https://gcc.gnu.org/PR10886
17793 282. https://gcc.gnu.org/PR11349
17794 283. https://gcc.gnu.org/PR4823
17795 284. https://gcc.gnu.org/PR8878
17796 285. https://gcc.gnu.org/PR9815
17797 286. https://gcc.gnu.org/PR10402
17798 287. https://gcc.gnu.org/PR10504
17799 288. https://gcc.gnu.org/PR10673
17800 289. https://gcc.gnu.org/PR11044
17801 290. https://gcc.gnu.org/PR11089
17802 291. https://gcc.gnu.org/PR11420
17803 292. https://gcc.gnu.org/PR9362
17804 293. https://gcc.gnu.org/PR10142
17805 294. https://gcc.gnu.org/PR10663
17806 295. https://gcc.gnu.org/PR10835
17807 296. https://gcc.gnu.org/PR10876
17808 297. https://gcc.gnu.org/PR10955
17809 298. https://gcc.gnu.org/PR11018
17810 299. https://gcc.gnu.org/PR11556
17811 300. https://gcc.gnu.org/PR10907
17812 301. https://gcc.gnu.org/PR11320
17813 302. https://gcc.gnu.org/PR11599
17814 303. https://gcc.gnu.org/PR9745
17815 304. https://gcc.gnu.org/PR10871
17816 305. https://gcc.gnu.org/PR11440
17817 306. https://gcc.gnu.org/PR7594
17818 307. https://gcc.gnu.org/PR10557
17819 308. https://gcc.gnu.org/PR11054
17820 309. https://gcc.gnu.org/PR10834
17821 310. https://gcc.gnu.org/PR10842
17822 311. https://gcc.gnu.org/PR11052
17823 312. https://gcc.gnu.org/PR11183
17824 313. https://gcc.gnu.org/PR11084
17825 314. https://gcc.gnu.org/PR10331
17826 315. https://gcc.gnu.org/PR10413
17827 316. https://gcc.gnu.org/PR11096
17828 317. https://gcc.gnu.org/PR2873
17829 318. https://gcc.gnu.org/PR3163
17830 319. https://gcc.gnu.org/PR5287
17831 320. https://gcc.gnu.org/PR10148
17832 321. https://gcc.gnu.org/PR8787
17833 322. https://gcc.gnu.org/PR10900
17834 323. https://gcc.gnu.org/PR1607
17835 324. https://gcc.gnu.org/PR4252
17836 325. https://gcc.gnu.org/PR4490
17837 326. https://gcc.gnu.org/PR10355
17838 327. https://gcc.gnu.org/PR10726
17839 328. https://gcc.gnu.org/PR10805
17840 329. https://gcc.gnu.org/PR10815
17841 330. https://gcc.gnu.org/PR10877
17842 331. https://gcc.gnu.org/PR11280
17843 332. https://gcc.gnu.org/PR11466
17844 333. https://gcc.gnu.org/PR10737
17845 334. https://gcc.gnu.org/PR10810
17846 335. https://gcc.gnu.org/PR8336
17847 336. https://gcc.gnu.org/PR9330
17848 337. https://gcc.gnu.org/PR9631
17849 338. https://gcc.gnu.org/PR9877
17850 339. https://gcc.gnu.org/PR11687
17851 340. https://gcc.gnu.org/PR12263
17852 341. https://gcc.gnu.org/PR12490
17853 342. https://gcc.gnu.org/PR7277
17854 343. https://gcc.gnu.org/PR7939
17855 344. https://gcc.gnu.org/PR11063
17856 345. https://gcc.gnu.org/PR11207
17857 346. https://gcc.gnu.org/PR11522
17858 347. https://gcc.gnu.org/PR11595
17859 348. https://gcc.gnu.org/PR11646
17860 349. https://gcc.gnu.org/PR11665
17861 350. https://gcc.gnu.org/PR11852
17862 351. https://gcc.gnu.org/PR11878
17863 352. https://gcc.gnu.org/PR11883
17864 353. https://gcc.gnu.org/PR11991
17865 354. https://gcc.gnu.org/PR12146
17866 355. https://gcc.gnu.org/PR12215
17867 356. https://gcc.gnu.org/PR12369
17868 357. https://gcc.gnu.org/PR12446
17869 358. https://gcc.gnu.org/PR12510
17870 359. https://gcc.gnu.org/PR12544
17871 360. https://gcc.gnu.org/PR9862
17872 361. https://gcc.gnu.org/PR10962
17873 362. https://gcc.gnu.org/PR11370
17874 363. https://gcc.gnu.org/PR11637
17875 364. https://gcc.gnu.org/PR11885
17876 365. https://gcc.gnu.org/PR12082
17877 366. https://gcc.gnu.org/PR12180
17878 367. https://gcc.gnu.org/PR12340
17879 368. https://gcc.gnu.org/PR3907
17880 369. https://gcc.gnu.org/PR5293
17881 370. https://gcc.gnu.org/PR5296
17882 371. https://gcc.gnu.org/PR7939
17883 372. https://gcc.gnu.org/PR8656
17884 373. https://gcc.gnu.org/PR10147
17885 374. https://gcc.gnu.org/PR11400
17886 375. https://gcc.gnu.org/PR11409
17887 376. https://gcc.gnu.org/PR11740
17888 377. https://gcc.gnu.org/PR11786
17889 378. https://gcc.gnu.org/PR11867
17890 379. https://gcc.gnu.org/PR11928
17891 380. https://gcc.gnu.org/PR12114
17892 381. https://gcc.gnu.org/PR12163
17893 382. https://gcc.gnu.org/PR12181
17894 383. https://gcc.gnu.org/PR12236
17895 384. https://gcc.gnu.org/PR12266
17896 385. https://gcc.gnu.org/PR12296
17897 386. https://gcc.gnu.org/PR12298
17898 387. https://gcc.gnu.org/PR12369
17899 388. https://gcc.gnu.org/PR12337
17900 389. https://gcc.gnu.org/PR12344
17901 390. https://gcc.gnu.org/PR12451
17902 391. https://gcc.gnu.org/PR12486
17903 392. https://gcc.gnu.org/PR8869
17904 393. https://gcc.gnu.org/PR9786
17905 394. https://gcc.gnu.org/PR11689
17906 395. https://gcc.gnu.org/PR12116
17907 396. https://gcc.gnu.org/PR12070
17908 397. https://gcc.gnu.org/PR11184
17909 398. https://gcc.gnu.org/PR11535
17910 399. https://gcc.gnu.org/PR11693
17911 400. https://gcc.gnu.org/PR12224
17912 401. https://gcc.gnu.org/PR11087
17913 402. https://gcc.gnu.org/PR11319
17914 403. https://gcc.gnu.org/PR11949
17915 404. https://gcc.gnu.org/PR11662
17916 405. https://gcc.gnu.org/PR11965
17917 406. https://gcc.gnu.org/PR12301
17918 407. https://gcc.gnu.org/PR11717
17919 408. https://gcc.gnu.org/PR11313
17920 409. https://gcc.gnu.org/PR11712
17921 410. https://gcc.gnu.org/PR12166
17922 411. https://gcc.gnu.org/PR12101
17923 412. https://gcc.gnu.org/PR10988
17924 413. https://gcc.gnu.org/PR11805
17925 414. https://gcc.gnu.org/PR11902
17926 415. https://gcc.gnu.org/PR11903
17927 416. https://gcc.gnu.org/PR11890
17928 417. https://gcc.gnu.org/PR12399
17929 418. https://gcc.gnu.org/PR13068
17930 419. https://gcc.gnu.org/PR10060
17931 420. https://gcc.gnu.org/PR10555
17932 421. https://gcc.gnu.org/PR10706
17933 422. https://gcc.gnu.org/PR11496
17934 423. https://gcc.gnu.org/PR11741
17935 424. https://gcc.gnu.org/PR12440
17936 425. https://gcc.gnu.org/PR12632
17937 426. https://gcc.gnu.org/PR12712
17938 427. https://gcc.gnu.org/PR12726
17939 428. https://gcc.gnu.org/PR12890
17940 429. https://gcc.gnu.org/PR12900
17941 430. https://gcc.gnu.org/PR13060
17942 431. https://gcc.gnu.org/PR13289
17943 432. https://gcc.gnu.org/PR13318
17944 433. https://gcc.gnu.org/PR13392
17945 434. https://gcc.gnu.org/PR13574
17946 435. https://gcc.gnu.org/PR13475
17947 436. https://gcc.gnu.org/PR13797
17948 437. https://gcc.gnu.org/PR13824
17949 438. https://gcc.gnu.org/PR8776
17950 439. https://gcc.gnu.org/PR10339
17951 440. https://gcc.gnu.org/PR11350
17952 441. https://gcc.gnu.org/PR12826
17953 442. https://gcc.gnu.org/PR12500
17954 443. https://gcc.gnu.org/PR12941
17955 444. https://gcc.gnu.org/PR12953
17956 445. https://gcc.gnu.org/PR13041
17957 446. https://gcc.gnu.org/PR13507
17958 447. https://gcc.gnu.org/PR13382
17959 448. https://gcc.gnu.org/PR13394
17960 449. https://gcc.gnu.org/PR13400
17961 450. https://gcc.gnu.org/PR13521
17962 451. https://gcc.gnu.org/PR2094
17963 452. https://gcc.gnu.org/PR2294
17964 453. https://gcc.gnu.org/PR5050
17965 454. https://gcc.gnu.org/PR9371
17966 455. https://gcc.gnu.org/PR9546
17967 456. https://gcc.gnu.org/PR10081
17968 457. https://gcc.gnu.org/PR10093
17969 458. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#61
17970 459. https://gcc.gnu.org/PR10095
17971 460. https://gcc.gnu.org/PR11554
17972 461. https://gcc.gnu.org/PR12297
17973 462. https://gcc.gnu.org/PR12352
17974 463. https://gcc.gnu.org/PR12438
17975 464. https://gcc.gnu.org/PR12540
17976 465. https://gcc.gnu.org/PR12594
17977 466. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#60
17978 467. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#63
17979 468. https://gcc.gnu.org/PR12657
17980 469. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#292
17981 470. https://gcc.gnu.org/PR12696
17982 471. https://gcc.gnu.org/PR12815
17983 472. https://gcc.gnu.org/PR12862
17984 473. https://gcc.gnu.org/PR12926
17985 474. https://gcc.gnu.org/PR12967
17986 475. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html
17987 476. https://gcc.gnu.org/PR12971
17988 477. http://www.open-std.org/jtc1/sc22/wg21/docs/cwg_defects.html#328
17989 478. https://gcc.gnu.org/PR13007
17990 479. https://gcc.gnu.org/PR13009
17991 480. https://gcc.gnu.org/PR13057
17992 481. https://gcc.gnu.org/PR13070
17993 482. https://gcc.gnu.org/PR13081
17994 483. https://gcc.gnu.org/PR13239
17995 484. https://gcc.gnu.org/PR13262
17996 485. https://gcc.gnu.org/PR13290
17997 486. https://gcc.gnu.org/PR13323
17998 487. https://gcc.gnu.org/PR13369
17999 488. https://gcc.gnu.org/PR13371
18000 489. https://gcc.gnu.org/PR13445
18001 490. https://gcc.gnu.org/PR13461
18002 491. https://gcc.gnu.org/PR13462
18003 492. https://gcc.gnu.org/PR13478
18004 493. https://gcc.gnu.org/PR13544
18005 494. https://gcc.gnu.org/PR13650
18006 495. https://gcc.gnu.org/PR13683
18007 496. https://gcc.gnu.org/PR13688
18008 497. https://gcc.gnu.org/PR13774
18009 498. https://gcc.gnu.org/PR13884
18010 499. https://gcc.gnu.org/PR10746
18011 500. https://gcc.gnu.org/PR11433
18012 501. https://gcc.gnu.org/PR12633
18013 502. https://gcc.gnu.org/PR13037
18014 503. https://gcc.gnu.org/PR13213
18015 504. https://gcc.gnu.org/PR4490
18016 505. https://gcc.gnu.org/PR12292
18017 506. https://gcc.gnu.org/PR12441
18018 507. https://gcc.gnu.org/PR12943
18019 508. https://gcc.gnu.org/PR13608
18020 509. https://gcc.gnu.org/PR11598
18021 510. https://gcc.gnu.org/PR11793
18022 511. https://gcc.gnu.org/PR12467
18023 512. https://gcc.gnu.org/PR12537
18024 513. https://gcc.gnu.org/PR12496
18025 514. https://gcc.gnu.org/PR12865
18026 515. https://gcc.gnu.org/PR13354
18027 516. https://gcc.gnu.org/PR10467
18028 517. https://gcc.gnu.org/PR11226
18029 518. https://gcc.gnu.org/PR11227
18030 519. https://gcc.gnu.org/PR12644
18031 520. https://gcc.gnu.org/PR13149
18032 521. https://gcc.gnu.org/PR12654
18033 522. https://gcc.gnu.org/PR12965
18034 523. https://gcc.gnu.org/PR13031
18035 524. https://gcc.gnu.org/PR11634
18036 525. https://gcc.gnu.org/PR12158
18037 526. https://gcc.gnu.org/PR11992
18038 527. https://gcc.gnu.org/PR9365
18039 528. https://gcc.gnu.org/PR10392
18040 529. https://gcc.gnu.org/PR11322
18041 530. https://gcc.gnu.org/PR13069
18042 531. https://gcc.gnu.org/PR13302
18043 532. https://gcc.gnu.org/PR13585
18044 533. https://gcc.gnu.org/PR8916
18045 534. https://gcc.gnu.org/PR11576
18046 535. https://gcc.gnu.org/PR13122
18047 536. https://gcc.gnu.org/PR13256
18048 537. https://gcc.gnu.org/PR13373
18049 538. https://gcc.gnu.org/PR12561
18050 539. https://gcc.gnu.org/PR6243
18051 540. https://gcc.gnu.org/PR11397
18052 541. https://gcc.gnu.org/PR12505
18053 542. https://gcc.gnu.org/PR13150
18054 543. https://gcc.gnu.org/PR12666
18055 544. https://gcc.gnu.org/PR12969
18056 545. https://gcc.gnu.org/PR10819
18057 546. https://gcc.gnu.org/PR11612
18058 547. https://gcc.gnu.org/PR13211
18059 548. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=3.3.4
18060 549. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=3.3.5
18061 550. https://gcc.gnu.org/bugzilla/buglist.cgi?bug_status=RESOLVED&resolution=FIXED&target_milestone=3.3.6
18062 551. https://gcc.gnu.org/onlinedocs/
18063 552. mailto:gcc-help@gcc.gnu.org
18064 553. mailto:gcc@gcc.gnu.org
18065 554. https://gcc.gnu.org/lists.html
18066 555. https://www.fsf.org/
18067 556. https://gcc.gnu.org/about.html
18068 557. http://validator.w3.org/check/referer
18069======================================================================
18070http://gcc.gnu.org/gcc-3.2/index.html
18071
18072                             GCC 3.2 Release Series
18073
18074   (This release series is no longer supported.)
18075
18076   April 25, 2003
18077
18078   The [1]GNU project and the GCC developers are pleased to announce the
18079   release of GCC 3.2.3.
18080
18081   The purpose of the GCC 3.2 release series is to provide a stable
18082   platform for OS distributors to use building their next releases. A
18083   primary objective was to stabilize the C++ ABI; we believe that the
18084   interface to the compiler and the C++ standard library are now
18085   relatively stable.
18086
18087   Be aware that C++ code compiled by GCC 3.2.x will (in general) not
18088   interoperate with code compiled by GCC 3.1.1 or earlier.
18089
18090   Please refer to our [2]detailed list of news, caveats, and bug-fixes
18091   for further information.
18092
18093Release History
18094
18095   GCC 3.2.3
18096          April 25, 2003 ([3]changes)
18097
18098   GCC 3.2.2
18099          February 5, 2003 ([4]changes)
18100
18101   GCC 3.2.1
18102          November 19, 2002 ([5]changes)
18103
18104   GCC 3.2
18105          August 14, 2002 ([6]changes)
18106
18107References and Acknowledgements
18108
18109   GCC used to stand for the GNU C Compiler, but since the compiler
18110   supports several other languages aside from C, it now stands for the
18111   GNU Compiler Collection.
18112
18113   A list of [7]successful builds is updated as new information becomes
18114   available.
18115
18116   The GCC developers would like to thank the numerous people that have
18117   contributed new features, improvements, bug fixes, and other changes as
18118   well as test results to GCC. This [8]amazing group of volunteers is
18119   what makes GCC successful.
18120
18121   For additional information about GCC please refer to the [9]GCC project
18122   web site or contact the [10]GCC development mailing list.
18123
18124   To obtain GCC please use [11]our mirror sites, or our CVS server.
18125
18126
18127    For questions related to the use of GCC, please consult these web
18128    pages and the [12]GCC manuals. If that fails, the
18129    [13]gcc-help@gcc.gnu.org mailing list might help. Comments on these
18130    web pages and the development of GCC are welcome on our developer
18131    list at [14]gcc@gcc.gnu.org. All of [15]our lists have public
18132    archives.
18133
18134   Copyright (C) [16]Free Software Foundation, Inc. Verbatim copying and
18135   distribution of this entire article is permitted in any medium,
18136   provided this notice is preserved.
18137
18138   These pages are [17]maintained by the GCC team. Last modified
18139   2019-11-28[18].
18140
18141References
18142
18143   1. http://www.gnu.org/
18144   2. http://gcc.gnu.org/gcc-3.2/changes.html
18145   3. http://gcc.gnu.org/gcc-3.2/changes.html#3.2.3
18146   4. http://gcc.gnu.org/gcc-3.2/changes.html#3.2.2
18147   5. http://gcc.gnu.org/gcc-3.2/changes.html#3.2.1
18148   6. http://gcc.gnu.org/gcc-3.2/changes.html#3.2
18149   7. http://gcc.gnu.org/gcc-3.2/buildstat.html
18150   8. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
18151   9. http://gcc.gnu.org/index.html
18152  10. mailto:gcc@gcc.gnu.org
18153  11. http://gcc.gnu.org/mirrors.html
18154  12. https://gcc.gnu.org/onlinedocs/
18155  13. mailto:gcc-help@gcc.gnu.org
18156  14. mailto:gcc@gcc.gnu.org
18157  15. https://gcc.gnu.org/lists.html
18158  16. https://www.fsf.org/
18159  17. https://gcc.gnu.org/about.html
18160  18. http://validator.w3.org/check/referer
18161======================================================================
18162http://gcc.gnu.org/gcc-3.2/changes.html
18163
18164                             GCC 3.2 Release Series
18165                        Changes, New Features, and Fixes
18166
18167   The latest release in the 3.2 release series is [1]GCC 3.2.3.
18168
18169Caveats and New Features
18170
18171  Caveats
18172
18173     * The C++ compiler does not correctly zero-initialize
18174       pointers-to-data members. You must explicitly initialize them. For
18175       example: int S::*m(0); will work, but depending on
18176       default-initialization to zero will not work. This bug cannot be
18177       fixed in GCC 3.2 without inducing unacceptable risks. It will be
18178       fixed in GCC 3.3.
18179     * This GCC release is based on the GCC 3.1 sourcebase, and thus has
18180       all the [2]changes in the GCC 3.1 series. In addition, GCC 3.2 has
18181       a number of C++ ABI fixes which make its C++ compiler generate
18182       binary code which is incompatible with the C++ compilers found in
18183       earlier GCC releases, including GCC 3.1 and GCC 3.1.1.
18184
18185  Frontend Enhancements
18186
18187    C/C++/Objective-C
18188
18189     * The method of constructing the list of directories to be searched
18190       for header files has been revised. If a directory named by a -I
18191       option is a standard system include directory, the option is
18192       ignored to ensure that the default search order for system
18193       directories and the special treatment of system header files are
18194       not defeated.
18195     * The C and Objective-C compilers no longer accept the "Naming Types"
18196       extension (typedef foo = bar); it was already unavailable in C++.
18197       Code which uses it will need to be changed to use the "typeof"
18198       extension instead: typedef typeof(bar) foo. (We have removed this
18199       extension without a period of deprecation because it has caused the
18200       compiler to crash since version 3.0 and no one noticed until very
18201       recently. Thus we conclude it is not in widespread use.)
18202
18203    C++
18204
18205     * GCC 3.2 fixed serveral differences between the C++ ABI implemented
18206       in GCC and the multi-vendor standard, but more have been found
18207       since the release. 3.2.1 adds a new warning, -Wabi, to warn about
18208       code which is affected by these bugs. We will fix these bugs in
18209       some future release, once we are confident that all have been
18210       found; until then, it is our intention to make changes to the ABI
18211       only if they are necessary for correct compilation of C++, as
18212       opposed to conformance to the ABI documents.
18213     * For details on how to build an ABI compliant compiler for GNU/Linux
18214       systems, check the [3]common C++ ABI page.
18215
18216  New Targets and Target Specific Improvements
18217
18218    IA-32
18219
18220     * Fixed a number of bugs in SSE and MMX intrinsics.
18221     * Fixed common compiler crashes with SSE instruction set enabled
18222       (implied by -march=pentium3, pentium4, athlon-xp)
18223     * __m128 and __m128i is not 128bit aligned when used in structures.
18224
18225    x86-64
18226
18227     * A bug whereby the compiler could generate bad code for bzero has
18228       been fixed.
18229     * ABI fixes (implying ABI incompatibilities with previous version in
18230       some corner cases)
18231     * Fixed prefetch code generation
18232     __________________________________________________________________
18233
18234GCC 3.2.3
18235
18236   3.2.3 is a bug fix release only; there are no new features that were
18237   not present in GCC 3.2.2.
18238
18239  Bug Fixes
18240
18241   This section lists the problem reports (PRs) from GCC's bug tracking
18242   system that are known to be fixed in the 3.2.3 release. This list might
18243   not be complete (that is, it is possible that some PRs that have been
18244   fixed are not listed here), and some of the titles have been changed to
18245   make them more clear.
18246
18247    Internal Compiler Errors (multi-platform)
18248
18249     * [4]3782: (c++) -quiet -fstats produces a segmentation fault in
18250       cc1plus
18251     * [5]6440: (c++) template specializations cause ICE
18252     * [6]7050: (c++) ICE on: (i ? get_string() : throw)
18253     * [7]7741: ICE on conflicting types (make_decl_rtl in varasm.c)
18254     * [8]7982: (c++) ICE due to infinite recursion (using STL set)
18255     * [9]8068: exceedingly high (infinite) memory usage
18256     * [10]8178: ICE with __builtin_ffs
18257     * [11]8396: ICE in copy_to_mode_reg, in explow.c
18258     * [12]8674: (c++) ICE in cp_expr_size, in cp/cp-lang.c
18259     * [13]9768: ICE when optimizing inline code at -O2
18260     * [14]9798: (c++) Infinite recursion (segfault) in
18261       cp/decl.c:push_using_directive with recursive using directives
18262     * [15]9799: mismatching structure initializer with nested flexible
18263       array member: ICE
18264     * [16]9928: ICE on duplicate enum declaration
18265     * [17]10114: ICE in mem_loc_descriptor, in dwarf2out.c (affects
18266       sparc, alpha)
18267     * [18]10352: ICE in find_reloads_toplev
18268     * [19]10336: ICE with -Wunreachable-code
18269
18270    C/optimizer bugs:
18271
18272     * [20]8224: Incorrect joining of signed and unsigned division
18273     * [21]8613: -O2 produces wrong code with builtin strlen and
18274       postincrements
18275     * [22]8828: gcc reports some code is unreachable when it is not
18276     * [23]9226: GCSE breaking argument passing
18277     * [24]9853: miscompilation of non-constant structure initializer
18278     * [25]9797: C99-style struct initializers are miscompiled
18279     * [26]9967: Some standard C function calls should not be replaced
18280       when optimizing for size
18281     * [27]10116: ce2: invalid merge of join_bb in the context of switch
18282       statements
18283     * [28]10171: wrong code for inlined function
18284     * [29]10175: -Wunreachable-code doesn't work for single lines
18285
18286    C++ compiler and library:
18287
18288     * [30]8316: Confusing diagnostic for code that misuses conversion
18289       operators
18290     * [31]9169: filebuf output fails if codecvt<>::out returns noconv
18291     * [32]9420: incomplete type incorrectly reported
18292     * [33]9459: typeof in return type specification of template not
18293       supported
18294     * [34]9507: filebuf::open handles ios_base::ate incorrectly
18295     * [35]9538: Out-of-bounds memory access in streambuf::sputbackc
18296     * [36]9602: Total confusion about template/friend/virtual/abstract
18297     * [37]9993: destructor not called for local object created within and
18298       returned from infinite loop
18299     * [38]10167: ieee_1003.1-2001 locale specialisations on a glibc-2.3.2
18300       system
18301
18302    Java compiler and library:
18303
18304     * [39]9652: libgcj build fails on irix6.5.1[78]
18305     * [40]10144: gas on solaris complains about bad .stabs lines for
18306       java, native as unaffected
18307
18308    x86-specific (Intel/AMD):
18309
18310     * [41]8746: gcc miscompiles Linux kernel ppa driver on x86
18311     * [42]9888: -mcpu=k6 -Os produces out of range loop instructions
18312     * [43]9638: Cross-build for target i386-elf and i586-pc-linux-gnu
18313       failed
18314     * [44]9954: Cross-build for target i586-pc-linux-gnu (--with-newlib)
18315       failed
18316
18317    SPARC-specific:
18318
18319     * [45]7784: [Sparc] ICE in extract_insn, in recog.c
18320     * [46]7796: sparc extra failure with -m64 on execute/930921-1.c in
18321       unroll.c
18322     * [47]8281: ICE when compiling with -O2 -fPIC for Ultrasparc
18323     * [48]8366: [Sparc] C testsuite failure with -m64 -fpic -O in
18324       execute/loop-2d.c
18325     * [49]8726: gcc -O2 miscompiles Samba 2.2.7 on 32-bit sparc
18326     * [50]9414: Scheduling bug on Ultrasparc
18327     * [51]10067: GCC-3.2.2 outputs invalid asm on sparc64
18328
18329    m68k-specific:
18330
18331     * [52]7248: broken "inclusive or" code
18332     * [53]8343: m68k-elf/rtems ICE at instantiate_virtual_regs_1
18333
18334    PowerPC-specific:
18335
18336     * [54]9732: Wrong code with -O2 -fPIC
18337     * [55]10073: ICE: powerpc cannot split insn
18338
18339    Alpha-specific:
18340
18341     * [56]7702: optimization problem on a DEC alpha under OSF1
18342     * [57]9671: gcc.3.2.2 does not build on a HP Tru64 Unix v5.1B system
18343
18344    HP-specific:
18345
18346     * [58]8694: <string> breaks <ctype.h> on HP-UX 10.20 (DUP: 9275)
18347     * [59]9953: (ada) gcc 3.2.x can't build 3.3-branch ada on HP-UX 10
18348       (missing symbol)
18349     * [60]10271: Floating point args don't get reloaded across function
18350       calls with -O2
18351
18352    MIPS specific:
18353
18354     * [61]6362: mips-irix6 gcc-3.1 C testsuite failure with -mips4 in
18355       compile/920501-4.c
18356
18357    CRIS specific:
18358
18359     * [62]10377: gcc-3.2.2 creates bad assembler code for cris
18360
18361    Miscellaneous and minor bugs:
18362
18363     * [63]6955: collect2 says "core dumped" when there is no core
18364     __________________________________________________________________
18365
18366GCC 3.2.2
18367
18368   Beginning with 3.2.2, GCC's Makefile suite supports redirection of make
18369   install by means of the DESTDIR variable. Parts of the GCC tree have
18370   featured that support long before, but now it is available even from
18371   the top level.
18372
18373   Other than that, GCC 3.2.2 is a bug fix release only; there are no new
18374   features that were not present in GCC 3.2.1.
18375
18376  Bug Fixes
18377
18378   On the following i386-based systems GCC 3.2.1 broke the C ABI wrt.
18379   functions returning structures: Cygwin, FreeBSD (GCC 3.2.1 as shipped
18380   with FreeBSD 5.0 does not have this problem), Interix, a.out-based
18381   GNU/Linux and NetBSD, OpenBSD, and Darwin. GCC 3.2.2 reverts this ABI
18382   change, and thus restores ABI-compatibility with previous releases
18383   (except GCC 3.2.1) on these platforms.
18384
18385   This section lists the problem reports (PRs) from GCC's bug tracking
18386   system that are known to be fixed in the 3.2.2 release. This list might
18387   not be complete (that is, it is possible that some PRs that have been
18388   fixed are not listed here) and some of the titles have been changed to
18389   make them more clear.
18390
18391    Internal Compiler Errors (multi-platform)
18392
18393     * [64]5919: (c++) ICE when passing variable array to template
18394       function
18395     * [65]7129: (c++) ICE with min/max assignment operators (<?= and >?=)
18396     * [66]7507: ICE with -O2 when address of called function is a
18397       complicated expression
18398     * [67]7622: ICE with nested inline functions if function's address is
18399       taken
18400     * [68]7681: (fortran) ICE in compensate_edge, in reg-stack.c (also PR
18401       [69]9258)
18402     * [70]8031: (c++) ICE in code comparing typeids and casting from
18403       virtual base
18404     * [71]8275: ICE in simplify_subreg
18405     * [72]8332: (c++) builtin strlen/template interaction causes ICE
18406     * [73]8372: (c++) ICE on explicit call of destructor
18407     * [74]8439: (c, not c++) empty struct causes ICE
18408     * [75]8442: (c++) ICE with nested template classes
18409     * [76]8518: ICE when compiling mplayer ("extern inline" issue)
18410     * [77]8615: (c++) ICE with out-of-range character constant template
18411       argument
18412     * [78]8663: (c++) ICE in cp_expr_size, at cp-lang.c:307
18413     * [79]8799: (c++) ICE: error reporting routines re-entered
18414     * [80]9328: (c++) ICE with typeof(X) for overloaded X
18415     * [81]9465: (preprocessor) cpp -traditional ICE on null bytes
18416
18417    C++ (compiler and library) bugs
18418
18419     * [82]47: scoping in nested classes is broken
18420     * [83]6745: problems with iostream rdbuf() member function
18421     * [84]8214: conversion from const char* const to char* sometimes
18422       accepted illegally
18423     * [85]8493: builtin strlen and overload resolution (same bug as
18424       [86]8332)
18425     * [87]8503: strange behaviour of function types
18426     * [88]8727: compiler confused by inheritance from an anonymous struct
18427     * [89]7445: poor performance of std::locale::classic() in
18428       multi-threaded applications
18429     * [90]8230: mishandling of overflow in vector<T>::resize
18430     * [91]8399: sync_with_stdio(false) breaks unformatted input
18431     * [92]8662: illegal access of private member of unnamed class is
18432       accepted
18433     * [93]8707: "make distclean" fails in libstdc++-v3 directory
18434     * [94]8708: __USE_MALLOC doesn't work
18435     * [95]8790: Use of non-thread-safe strtok in src/localename.cc
18436     * [96]8887: Bug in date formats with --enable-clocale=generic
18437     * [97]9076: Call Frame Instructions are not handled correctly during
18438       unwind operation
18439     * [98]9151: std::setprecision limited to 16 digits when outputting a
18440       double to a stream
18441     * [99]9168: codecvt<char, char, mbstate_t> overwrites output buffers
18442     * [100]9269: libstdc++ headers: explicit specialization of function
18443       must precede its first use
18444     * [101]9322: return value of basic_streambuf<>::getloc affected by
18445       locale::global
18446     * [102]9433: segfault in runtime support for dynamic_cast
18447
18448    C and optimizer bugs
18449
18450     * [103]8032: GCC incorrectly initializes static structs that have
18451       flexible arrays
18452     * [104]8639: simple arithmetic expression broken
18453     * [105]8794: optimization improperly eliminates certain expressions
18454     * [106]8832: traditional "asm volatile" code is illegally optimized
18455     * [107]8988: loop optimizer bug: with -O2, code is generated that
18456       segfaults (found on i386, bug present for all platforms)
18457     * [108]9492: structure copy clobbers subsequent stores to structure
18458
18459    Objective-C bugs
18460
18461     * [109]9267: Objective-C parser won't build with newer bison versions
18462       (e.g. 1.875)
18463
18464    Ada bugs
18465
18466     * [110]8344: Ada build problem due to conflict between gcc/final.o,
18467       gcc/ada/final.o
18468
18469    Preprocessor bugs
18470
18471     * [111]8524: _Pragma within macros is improperly expanded
18472     * [112]8880: __WCHAR_TYPE__ macro incorrectly set to "long int" with
18473       -fshort-wchar
18474
18475    ARM-specific
18476
18477     * [113]9090: arm ICE with >= -O2; regression from gcc-2.95
18478
18479    x86-specific (Intel/AMD)
18480
18481     * [114]8588: ICE in extract_insn, at recog.c:NNNN (shift instruction)
18482     * [115]8599: loop unroll bug with -march=k6-3
18483     * [116]9506: ABI breakage in structure return (affects BSD and
18484       Cygwin, but not GNU/Linux)
18485
18486    FreeBSD 5.0 specific
18487
18488     * [117]9484: GCC 3.2.1 Bootstrap failure on FreeBSD 5.0
18489
18490    RTEMS-specific
18491
18492     * [118]9292: hppa1.1-rtems configurery problems
18493     * [119]9293: [m68k-elf/rtems] config/m68k/t-crtstuff bug
18494     * [120]9295: [mips-rtems] config/mips/rtems.h init/fini issue
18495     * [121]9296: gthr-rtems regression
18496     * [122]9316: powerpc-rtems: extending multilibs
18497
18498    HP-PA specific
18499
18500     * [123]9493: ICE with -O2 when building a simple function
18501
18502    Documentation
18503
18504     * [124]7341: hyperlink to gcov in GCC documentation doesn't work
18505     * [125]8947: Please add a warning about "-malign-double" in docs
18506     * [126]7448, [127]8882: typo cleanups
18507     __________________________________________________________________
18508
18509GCC 3.2.1
18510
18511   3.2.1 adds a new warning, -Wabi. This option warns when GNU C++
18512   generates code that is known not to be binary-compatible with the
18513   vendor-neutral ia32/ia64 ABI. Please consult the GCC manual, included
18514   in the distribution, for details.
18515
18516   This release also removes an old GCC extension, "naming types", and the
18517   documentation now directs users to use a different GCC extension,
18518   __typeof__, instead. The feature had evidently been broken for a while.
18519
18520   Otherwise, 3.2.1 is a bug fix release only; other than bug fixes and
18521   the new warning there are no new features that were not present in GCC
18522   3.2.
18523
18524   In addition, the previous fix for [128]PR 7445 (poor performance of
18525   std::locale::classic() in multi-threaded applications) was reverted
18526   ("unfixed"), because the "fix" was not thread-safe.
18527
18528  Bug Fixes
18529
18530   This section lists the problem reports (PRs) from GCC's bug tracking
18531   system that are known to be fixed in the 3.2.1 release. This list might
18532   not be complete (that is, it is possible that some PRs that have been
18533   fixed are not listed here). As you can see, the number of bug fixes is
18534   quite large, so it is strongly recommended that users of earlier GCC
18535   3.x releases upgrade to GCC 3.2.1.
18536
18537    Internal Compiler Errors (multi-platform)
18538
18539     * [129]2521: (c++) ICE in build_ptrmemfunc, in cp/typeck.c
18540     * [130]5661: (c++) ICE instantiating template on array of unknown
18541       size (bad code)
18542     * [131]6419: (c++) ICE in make_decl_rtl for "longest" attribute on
18543       64-bit platforms
18544     * [132]6994: (c++) ICE in find_function_data
18545     * [133]7150: preprocessor: GCC -dM -E gives an ICE
18546     * [134]7160: ICE when optimizing branches without a return value
18547     * [135]7228: (c++) ICE when using member template and template
18548       function
18549     * [136]7266: (c++) ICE with -pedantic on missing typename
18550     * [137]7353: ICE from use of "Naming Types" extension, see above
18551     * [138]7411: ICE in instantiate_virtual_regs_1, in function.c
18552     * [139]7478: (c++) ICE on static_cast inside template
18553     * [140]7526: preprocessor core dump when _Pragma implies #pragma
18554       dependency
18555     * [141]7721: (c++) ICE on simple (but incorrect) template ([142]7803
18556       is a duplicate)
18557     * [143]7754: (c++) ICE on union with template parameter
18558     * [144]7788: (c++) redeclaring a definition as an incomplete class
18559       causes ICE
18560     * [145]8031: (c++) ICE in comptypes, in cp/typeck.c
18561     * [146]8055: preprocessor dies with SIG11 when building FreeBSD
18562       kernel
18563     * [147]8067: (c++) ICE due to mishandling of __FUNCTION__ and related
18564       variables
18565     * [148]8134: (c++) ICE in force_store_init_value on legal code
18566     * [149]8149: (c++) ICE on incomplete type
18567     * [150]8160: (c++) ICE in build_modify_expr, in cp/typeck.c: array
18568       initialization
18569
18570    C++ (compiler and library) bugs
18571
18572     * [151]5607: No pointer adjustment in covariant return types
18573     * [152]6579: Infinite loop with statement expressions in member
18574       initialization
18575     * [153]6803: Default copy constructor bug in GCC 3.1
18576     * [154]7176: g++ confused by friend and static member with same name
18577     * [155]7188: Segfault with template class and recursive (incorrect)
18578       initializer list
18579     * [156]7306: Regression: GCC 3.x fails to compile code with virtual
18580       inheritance if a method has a variable number of arguments
18581     * [157]7461: ctype<char>::classic_table() returns offset array on
18582       Cygwin
18583     * [158]7524: f(const float arg[3]) fails
18584     * [159]7584: Erroneous ambiguous base error on using declaration
18585     * [160]7676: Member template overloading problem
18586     * [161]7679: infinite loop when a right parenthesis is missing
18587     * [162]7811: default locale not taken from environment
18588     * [163]7961: compare( char *) implemented incorrectly in
18589       basic_string<>
18590     * [164]8071: basic_ostream::operator<<(streambuf*) loops forever if
18591       streambuf::underflow() leaves gptr() NULL (dups: [165]8127,
18592       [166]6745)
18593     * [167]8096: deque::at() throws std::range_error instead of
18594       std::out_of_range
18595     * [168]8127: cout << cin.rdbuf() infinite loop
18596     * [169]8218: Excessively large memory consumed for classes with large
18597       array members
18598     * [170]8287: GCC 3.2: Destructor called for non-constructed local
18599       object
18600     * [171]8347: empty vector range used in string construction causes
18601       core dump
18602     * [172]8348: fail() flag is set in istringstream when eof() flag is
18603       set
18604     * [173]8391: regression: infinite loop in cp/decl2.c(finish_file)
18605
18606    C and optimizer bugs
18607
18608     * [174]6627: -fno-align-functions doesn't seem to disable function
18609       alignment
18610     * [175]6631: life_analysis misoptimizes code to initialize fields of
18611       a structure
18612     * [176]7102: unsigned char division results in floating exception
18613     * [177]7120: Run once loop should *always* be unrolled
18614       (pessimization)
18615     * [178]7209: Bug involving array referencing and ?: operator
18616     * [179]7515: invalid inlining of global function with -O3
18617     * [180]7814: incorrect scheduling for glibc-2.2.92 strcpy test
18618     * [181]8467: bug in sibling call optimization
18619
18620    Preprocessor bugs
18621
18622     * [182]4890: incorrect line markers from the traditional preprocessor
18623     * [183]7357: -M option omits system headers files (making it the same
18624       as -MM)
18625     * [184]7358: Changes to Sun's make Dependencies
18626     * [185]7602: C++ header files found in CPLUS_INCLUDE_PATH treated as
18627       C headers
18628     * [186]7862: Interrupting GCC -MD removes .d file but not .o
18629     * [187]8190: Failed compilation deletes -MD dependency file
18630     * [188]8524: _Pragma within macro is improperly expanded
18631
18632    x86 specific (Intel/AMD)
18633
18634     * [189]5351: (i686-only) function pass-by-value structure copy
18635       corrupts stack ([190]7591 is a duplicate)
18636     * [191]6845, [192]7034, [193]7124, [194]7174: ICE's with
18637       -march=pentium3/pentium2/athlon (these are all the same underlying
18638       bug, in MMX register use)
18639     * [195]7134, [196]7375, [197]7390: ICE with -march=athlon (maybe same
18640       as above?)
18641     * [198]6890: xmmintrin.h, _MM_TRANSPOSE4_PS is broken
18642     * [199]6981: wrong code in 64-bit manipulation on x86
18643     * [200]7242: GCC -mcpu=pentium[23] doesn't define __tune_pentiumpro__
18644       macro
18645     * [201]7396: ix86: cmpgt_ss, cmpge_ss, cmpngt_ss, and cmpnge_ss SSE
18646       intrinsics are broken
18647     * [202]7630: GCC 3.2 breaks on Mozilla 1.0's JS sources with
18648       -march=pentium4
18649     * [203]7693: Typo in i386 mmintrin.h header
18650     * [204]7723: ICE - Pentium3 sse - GCC 3.2
18651     * [205]7951: ICE on -march=pentium4 -O2 -mfpmath=sse
18652     * [206]8146: (i686 only) gcc 3.2 miscompiles gcc 2.95.3
18653
18654    PowerPC specific
18655
18656     * [207]5967: GCC bug when profiling nested functions on powerpc
18657     * [208]6984: wrong code generated with -O2, -O3, -Os for do-while
18658       loop on PowerPC
18659     * [209]7114: PowerPC: ICE building strcoll.op from glibc-2.2.5
18660     * [210]7130: miscompiled code for GCC-3.1 on
18661       powerpc-unknown-linux-gnu with -funroll-all-loops
18662     * [211]7133: PowerPC ICE: unrecognizable insn
18663     * [212]7380: ICE in extract_insn, at recog.c:2148
18664     * [213]8252: ICE on Altivec code with optimization turned on
18665     * [214]8451: Altivec ICE in GCC 3.2
18666
18667    HP/PA specific
18668
18669     * [215]7250: __ashrdi3 returns wrong value on 32 bit hppa
18670
18671    SPARC specific
18672
18673     * [216]6668: when using --disable-multilib, libgcc_s.so is installed
18674       in the wrong place on sparc-solaris
18675     * [217]7151: ICE when compiling for UltraSPARC
18676     * [218]7335: SPARC: ICE in verify_wide_reg (flow.c:557) with long
18677       double and -O1
18678     * [219]7842: [REGRESSION] SPARC code gen bug
18679
18680    ARM specific
18681
18682     * [220]7856: [arm] invalid offset in constant pool reference
18683     * [221]7967: optimization produces wrong code (ARM)
18684
18685    Alpha specific
18686
18687     * [222]7374: __builtin_fabsl broken on alpha
18688
18689    IBM s390 specific
18690
18691     * [223]7370: ICE in fixup_var_refs_1 on s390x
18692     * [224]7409: loop optimization bug on s390x-linux-gnu
18693     * [225]8232: s390x: ICE when using bcmp with int length argument
18694
18695    SCO specific
18696
18697     * [226]7623: SCO OpenServer build fails with machmode.def: undefined
18698       symbol: BITS_PER_UNIT
18699
18700    m68k/Coldfire specific
18701
18702     * [227]8314: crtbegin, crtend need to be multilib'ed for this
18703       platform
18704
18705    Documentation
18706
18707     * [228]761: Document some undocumented options
18708     * [229]5610: Fix documentation about invoking SSE instructions
18709       (-mfpmath=sse)
18710     * [230]7484: List -Wmissing-declarations as C-only option
18711     * [231]7531: -mcmodel not documented for x86-64
18712     * [232]8120: Update documentation of bad use of ##
18713     __________________________________________________________________
18714
18715GCC 3.2
18716
18717   3.2 is a small bug fix release, but there is a change to the
18718   application binary interface (ABI), hence the change to the second part
18719   of the version number.
18720
18721   The main purpose of the 3.2 release is to correct a couple of problems
18722   in the C++ ABI, with the intention of providing a stable interface
18723   going forward.  Accordingly, 3.2 is only a small change to 3.1.1.
18724
18725  Bug Fixes
18726
18727    C++
18728
18729     * [233]7320: g++ 3.2 relocation problem
18730     * [234]7470: vtable: virtual function pointers not in declaration
18731       order
18732
18733    libstdc++
18734
18735     * [235]6410: Trouble with non-ASCII monetary symbols and wchar_t
18736     * [236]6503, [237]6642, [238]7186: Problems with comparing or
18737       subtracting various types of const and non-const iterators
18738     * [239]7216: ambiguity with basic_iostream::traits_type
18739     * [240]7220: problem with basic_istream::ignore(0,delimiter)
18740     * [241]7222: locale::operator==() doesn't work on std::locale("")
18741     * [242]7286: placement operator delete issue
18742     * [243]7442: cxxabi.h does not match the C++ ABI
18743     * [244]7445: poor performance of std::locale::classic() in
18744       multi-threaded applications
18745
18746    x86-64 specific
18747
18748     * [245]7291: off-by-one in generated inline bzero code for x86-64
18749
18750
18751    For questions related to the use of GCC, please consult these web
18752    pages and the [246]GCC manuals. If that fails, the
18753    [247]gcc-help@gcc.gnu.org mailing list might help. Comments on these
18754    web pages and the development of GCC are welcome on our developer
18755    list at [248]gcc@gcc.gnu.org. All of [249]our lists have public
18756    archives.
18757
18758   Copyright (C) [250]Free Software Foundation, Inc. Verbatim copying and
18759   distribution of this entire article is permitted in any medium,
18760   provided this notice is preserved.
18761
18762   These pages are [251]maintained by the GCC team. Last modified
18763   2019-11-28[252].
18764
18765References
18766
18767   1. http://gcc.gnu.org/gcc-3.2/changes.html#3.2.3
18768   2. http://gcc.gnu.org/gcc-3.1/changes.html
18769   3. http://gcc.gnu.org/gcc-3.2/c++-abi.html
18770   4. https://gcc.gnu.org/PR3782
18771   5. https://gcc.gnu.org/PR6440
18772   6. https://gcc.gnu.org/PR7050
18773   7. https://gcc.gnu.org/PR7741
18774   8. https://gcc.gnu.org/PR7982
18775   9. https://gcc.gnu.org/PR8068
18776  10. https://gcc.gnu.org/PR8178
18777  11. https://gcc.gnu.org/PR8396
18778  12. https://gcc.gnu.org/PR8674
18779  13. https://gcc.gnu.org/PR9768
18780  14. https://gcc.gnu.org/PR9798
18781  15. https://gcc.gnu.org/PR9799
18782  16. https://gcc.gnu.org/PR9928
18783  17. https://gcc.gnu.org/PR10114
18784  18. https://gcc.gnu.org/PR10352
18785  19. https://gcc.gnu.org/PR10336
18786  20. https://gcc.gnu.org/PR8224
18787  21. https://gcc.gnu.org/PR8613
18788  22. https://gcc.gnu.org/PR8828
18789  23. https://gcc.gnu.org/PR9226
18790  24. https://gcc.gnu.org/PR9853
18791  25. https://gcc.gnu.org/PR9797
18792  26. https://gcc.gnu.org/PR9967
18793  27. https://gcc.gnu.org/PR10116
18794  28. https://gcc.gnu.org/PR10171
18795  29. https://gcc.gnu.org/PR10175
18796  30. https://gcc.gnu.org/PR8316
18797  31. https://gcc.gnu.org/PR9169
18798  32. https://gcc.gnu.org/PR9420
18799  33. https://gcc.gnu.org/PR9459
18800  34. https://gcc.gnu.org/PR9507
18801  35. https://gcc.gnu.org/PR9538
18802  36. https://gcc.gnu.org/PR9602
18803  37. https://gcc.gnu.org/PR9993
18804  38. https://gcc.gnu.org/PR10167
18805  39. https://gcc.gnu.org/PR9652
18806  40. https://gcc.gnu.org/PR10144
18807  41. https://gcc.gnu.org/PR8746
18808  42. https://gcc.gnu.org/PR9888
18809  43. https://gcc.gnu.org/PR9638
18810  44. https://gcc.gnu.org/PR9954
18811  45. https://gcc.gnu.org/PR7784
18812  46. https://gcc.gnu.org/PR7796
18813  47. https://gcc.gnu.org/PR8281
18814  48. https://gcc.gnu.org/PR8366
18815  49. https://gcc.gnu.org/PR8726
18816  50. https://gcc.gnu.org/PR9414
18817  51. https://gcc.gnu.org/PR10067
18818  52. https://gcc.gnu.org/PR7248
18819  53. https://gcc.gnu.org/PR8343
18820  54. https://gcc.gnu.org/PR9732
18821  55. https://gcc.gnu.org/PR10073
18822  56. https://gcc.gnu.org/PR7702
18823  57. https://gcc.gnu.org/PR9671
18824  58. https://gcc.gnu.org/PR8694
18825  59. https://gcc.gnu.org/PR9953
18826  60. https://gcc.gnu.org/PR10271
18827  61. https://gcc.gnu.org/PR6362
18828  62. https://gcc.gnu.org/PR10377
18829  63. https://gcc.gnu.org/PR6955
18830  64. https://gcc.gnu.org/PR5919
18831  65. https://gcc.gnu.org/PR7129
18832  66. https://gcc.gnu.org/PR7507
18833  67. https://gcc.gnu.org/PR7622
18834  68. https://gcc.gnu.org/PR7681
18835  69. https://gcc.gnu.org/PR9528
18836  70. https://gcc.gnu.org/PR8031
18837  71. https://gcc.gnu.org/PR8275
18838  72. https://gcc.gnu.org/PR8332
18839  73. https://gcc.gnu.org/PR8372
18840  74. https://gcc.gnu.org/PR8439
18841  75. https://gcc.gnu.org/PR8442
18842  76. https://gcc.gnu.org/PR8518
18843  77. https://gcc.gnu.org/PR8615
18844  78. https://gcc.gnu.org/PR8663
18845  79. https://gcc.gnu.org/PR8799
18846  80. https://gcc.gnu.org/PR9328
18847  81. https://gcc.gnu.org/PR9465
18848  82. https://gcc.gnu.org/PR47
18849  83. https://gcc.gnu.org/PR6745
18850  84. https://gcc.gnu.org/PR8214
18851  85. https://gcc.gnu.org/PR8493
18852  86. https://gcc.gnu.org/PR8332
18853  87. https://gcc.gnu.org/PR8503
18854  88. https://gcc.gnu.org/PR8727
18855  89. https://gcc.gnu.org/PR7445
18856  90. https://gcc.gnu.org/PR8230
18857  91. https://gcc.gnu.org/PR8399
18858  92. https://gcc.gnu.org/PR8662
18859  93. https://gcc.gnu.org/PR8707
18860  94. https://gcc.gnu.org/PR8708
18861  95. https://gcc.gnu.org/PR8790
18862  96. https://gcc.gnu.org/PR8887
18863  97. https://gcc.gnu.org/PR9076
18864  98. https://gcc.gnu.org/PR9151
18865  99. https://gcc.gnu.org/PR9168
18866 100. https://gcc.gnu.org/PR9269
18867 101. https://gcc.gnu.org/PR9322
18868 102. https://gcc.gnu.org/PR9433
18869 103. https://gcc.gnu.org/PR8032
18870 104. https://gcc.gnu.org/PR8639
18871 105. https://gcc.gnu.org/PR8794
18872 106. https://gcc.gnu.org/PR8832
18873 107. https://gcc.gnu.org/PR8988
18874 108. https://gcc.gnu.org/PR9492
18875 109. https://gcc.gnu.org/PR9267
18876 110. https://gcc.gnu.org/PR8344
18877 111. https://gcc.gnu.org/PR8524
18878 112. https://gcc.gnu.org/PR8880
18879 113. https://gcc.gnu.org/PR9090
18880 114. https://gcc.gnu.org/PR8588
18881 115. https://gcc.gnu.org/PR8599
18882 116. https://gcc.gnu.org/PR9506
18883 117. https://gcc.gnu.org/PR9484
18884 118. https://gcc.gnu.org/PR9292
18885 119. https://gcc.gnu.org/PR9293
18886 120. https://gcc.gnu.org/PR9295
18887 121. https://gcc.gnu.org/PR9296
18888 122. https://gcc.gnu.org/PR9316
18889 123. https://gcc.gnu.org/PR9493
18890 124. https://gcc.gnu.org/PR7341
18891 125. https://gcc.gnu.org/PR8947
18892 126. https://gcc.gnu.org/PR7448
18893 127. https://gcc.gnu.org/PR8882
18894 128. https://gcc.gnu.org/PR7445
18895 129. https://gcc.gnu.org/PR2521
18896 130. https://gcc.gnu.org/PR5661
18897 131. https://gcc.gnu.org/PR6419
18898 132. https://gcc.gnu.org/PR6994
18899 133. https://gcc.gnu.org/PR7150
18900 134. https://gcc.gnu.org/PR7160
18901 135. https://gcc.gnu.org/PR7228
18902 136. https://gcc.gnu.org/PR7266
18903 137. https://gcc.gnu.org/PR7353
18904 138. https://gcc.gnu.org/PR7411
18905 139. https://gcc.gnu.org/PR7478
18906 140. https://gcc.gnu.org/PR7526
18907 141. https://gcc.gnu.org/PR7721
18908 142. https://gcc.gnu.org/PR7803
18909 143. https://gcc.gnu.org/PR7754
18910 144. https://gcc.gnu.org/PR7788
18911 145. https://gcc.gnu.org/PR8031
18912 146. https://gcc.gnu.org/PR8055
18913 147. https://gcc.gnu.org/PR8067
18914 148. https://gcc.gnu.org/PR8134
18915 149. https://gcc.gnu.org/PR8149
18916 150. https://gcc.gnu.org/PR8160
18917 151. https://gcc.gnu.org/PR5607
18918 152. https://gcc.gnu.org/PR6579
18919 153. https://gcc.gnu.org/PR6803
18920 154. https://gcc.gnu.org/PR7176
18921 155. https://gcc.gnu.org/PR7188
18922 156. https://gcc.gnu.org/PR7306
18923 157. https://gcc.gnu.org/PR7461
18924 158. https://gcc.gnu.org/PR7524
18925 159. https://gcc.gnu.org/PR7584
18926 160. https://gcc.gnu.org/PR7676
18927 161. https://gcc.gnu.org/PR7679
18928 162. https://gcc.gnu.org/PR7811
18929 163. https://gcc.gnu.org/PR7961
18930 164. https://gcc.gnu.org/PR8071
18931 165. https://gcc.gnu.org/PR8127
18932 166. https://gcc.gnu.org/PR6745
18933 167. https://gcc.gnu.org/PR8096
18934 168. https://gcc.gnu.org/PR8127
18935 169. https://gcc.gnu.org/PR8218
18936 170. https://gcc.gnu.org/PR8287
18937 171. https://gcc.gnu.org/PR8347
18938 172. https://gcc.gnu.org/PR8348
18939 173. https://gcc.gnu.org/PR8391
18940 174. https://gcc.gnu.org/PR6627
18941 175. https://gcc.gnu.org/PR6631
18942 176. https://gcc.gnu.org/PR7102
18943 177. https://gcc.gnu.org/PR7120
18944 178. https://gcc.gnu.org/PR7209
18945 179. https://gcc.gnu.org/PR7515
18946 180. https://gcc.gnu.org/PR7814
18947 181. https://gcc.gnu.org/PR8467
18948 182. https://gcc.gnu.org/PR4890
18949 183. https://gcc.gnu.org/PR7357
18950 184. https://gcc.gnu.org/PR7358
18951 185. https://gcc.gnu.org/PR7602
18952 186. https://gcc.gnu.org/PR7862
18953 187. https://gcc.gnu.org/PR8190
18954 188. https://gcc.gnu.org/PR8524
18955 189. https://gcc.gnu.org/PR5351
18956 190. https://gcc.gnu.org/PR7591
18957 191. https://gcc.gnu.org/PR6845
18958 192. https://gcc.gnu.org/PR7034
18959 193. https://gcc.gnu.org/PR7124
18960 194. https://gcc.gnu.org/PR7174
18961 195. https://gcc.gnu.org/PR7134
18962 196. https://gcc.gnu.org/PR7375
18963 197. https://gcc.gnu.org/PR7390
18964 198. https://gcc.gnu.org/PR6890
18965 199. https://gcc.gnu.org/PR6981
18966 200. https://gcc.gnu.org/PR7242
18967 201. https://gcc.gnu.org/PR7396
18968 202. https://gcc.gnu.org/PR7630
18969 203. https://gcc.gnu.org/PR7693
18970 204. https://gcc.gnu.org/PR7723
18971 205. https://gcc.gnu.org/PR7951
18972 206. https://gcc.gnu.org/PR8146
18973 207. https://gcc.gnu.org/PR5967
18974 208. https://gcc.gnu.org/PR6984
18975 209. https://gcc.gnu.org/PR7114
18976 210. https://gcc.gnu.org/PR7130
18977 211. https://gcc.gnu.org/PR7133
18978 212. https://gcc.gnu.org/PR7380
18979 213. https://gcc.gnu.org/PR8252
18980 214. https://gcc.gnu.org/PR8451
18981 215. https://gcc.gnu.org/PR7250
18982 216. https://gcc.gnu.org/PR6668
18983 217. https://gcc.gnu.org/PR7151
18984 218. https://gcc.gnu.org/PR7335
18985 219. https://gcc.gnu.org/PR7842
18986 220. https://gcc.gnu.org/PR7856
18987 221. https://gcc.gnu.org/PR7967
18988 222. https://gcc.gnu.org/PR7374
18989 223. https://gcc.gnu.org/PR7370
18990 224. https://gcc.gnu.org/PR7409
18991 225. https://gcc.gnu.org/PR8232
18992 226. https://gcc.gnu.org/PR7623
18993 227. https://gcc.gnu.org/PR8314
18994 228. https://gcc.gnu.org/PR761
18995 229. https://gcc.gnu.org/PR5610
18996 230. https://gcc.gnu.org/PR7484
18997 231. https://gcc.gnu.org/PR7531
18998 232. https://gcc.gnu.org/PR8120
18999 233. https://gcc.gnu.org/PR7320
19000 234. https://gcc.gnu.org/PR7470
19001 235. https://gcc.gnu.org/PR6410
19002 236. https://gcc.gnu.org/PR6503
19003 237. https://gcc.gnu.org/PR6642
19004 238. https://gcc.gnu.org/PR7186
19005 239. https://gcc.gnu.org/PR7216
19006 240. https://gcc.gnu.org/PR7220
19007 241. https://gcc.gnu.org/PR7222
19008 242. https://gcc.gnu.org/PR7286
19009 243. https://gcc.gnu.org/PR7442
19010 244. https://gcc.gnu.org/PR7445
19011 245. https://gcc.gnu.org/PR7291
19012 246. https://gcc.gnu.org/onlinedocs/
19013 247. mailto:gcc-help@gcc.gnu.org
19014 248. mailto:gcc@gcc.gnu.org
19015 249. https://gcc.gnu.org/lists.html
19016 250. https://www.fsf.org/
19017 251. https://gcc.gnu.org/about.html
19018 252. http://validator.w3.org/check/referer
19019======================================================================
19020http://gcc.gnu.org/gcc-3.1/index.html
19021
19022                                    GCC 3.1
19023
19024   (This release series is no longer supported.)
19025
19026   July 27, 2002
19027
19028   The [1]GNU project and the GCC developers are pleased to announce the
19029   release of GCC 3.1.1.
19030
19031   The links below still apply to GCC 3.1.1.
19032
19033   May 15, 2002
19034
19035   The [2]GNU project and the GCC developers are pleased to announce the
19036   release of GCC 3.1.
19037
19038   GCC used to stand for the GNU C Compiler, but since the compiler
19039   supports several other languages aside from C, it now stands for the
19040   GNU Compiler Collection.
19041
19042   A list of [3]successful builds is updated as new information becomes
19043   available.
19044
19045   The GCC developers would like to thank the numerous people that have
19046   contributed [4]new features, improvements, bug fixes, and other changes
19047   as well as test results to GCC. This [5]amazing group of volunteers is
19048   what makes GCC successful.
19049
19050   For additional information about GCC please refer to the [6]GCC project
19051   web site or contact the [7]GCC development mailing list.
19052
19053   To obtain GCC please use [8]our mirror sites, or our CVS server.
19054     __________________________________________________________________
19055
19056
19057    For questions related to the use of GCC, please consult these web
19058    pages and the [9]GCC manuals. If that fails, the
19059    [10]gcc-help@gcc.gnu.org mailing list might help. Comments on these
19060    web pages and the development of GCC are welcome on our developer
19061    list at [11]gcc@gcc.gnu.org. All of [12]our lists have public
19062    archives.
19063
19064   Copyright (C) [13]Free Software Foundation, Inc. Verbatim copying and
19065   distribution of this entire article is permitted in any medium,
19066   provided this notice is preserved.
19067
19068   These pages are [14]maintained by the GCC team. Last modified
19069   2019-11-28[15].
19070
19071References
19072
19073   1. http://www.gnu.org/
19074   2. http://www.gnu.org/
19075   3. http://gcc.gnu.org/gcc-3.1/buildstat.html
19076   4. http://gcc.gnu.org/gcc-3.1/changes.html
19077   5. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
19078   6. http://gcc.gnu.org/index.html
19079   7. mailto:gcc@gcc.gnu.org
19080   8. http://gcc.gnu.org/mirrors.html
19081   9. https://gcc.gnu.org/onlinedocs/
19082  10. mailto:gcc-help@gcc.gnu.org
19083  11. mailto:gcc@gcc.gnu.org
19084  12. https://gcc.gnu.org/lists.html
19085  13. https://www.fsf.org/
19086  14. https://gcc.gnu.org/about.html
19087  15. http://validator.w3.org/check/referer
19088======================================================================
19089http://gcc.gnu.org/gcc-3.1/changes.html
19090
19091                             GCC 3.1 Release Series
19092                        Changes, New Features, and Fixes
19093
19094Additional changes in GCC 3.1.1
19095
19096     * A bug related to how structures and unions are returned has been
19097       fixed for powerpc-*-netbsd*.
19098     * An important bug in the implementation of -fprefetch-loop-arrays
19099       has been fixed. Previously the optimization prefetched random
19100       blocks of memory for most targets except for i386.
19101     * The Java compiler now compiles Java programs much faster and also
19102       works with parallel make.
19103     * Nested functions have been fixed for mips*-*-netbsd*.
19104     * Some missing floating point support routines have beed added for
19105       mips*-*-netbsd*.
19106     * This [1]message gives additional information about the bugs fixed
19107       in this release.
19108
19109Caveats
19110
19111     * The -traditional C compiler option has been deprecated and will be
19112       removed in GCC 3.3. (It remains possible to preprocess non-C code
19113       with the traditional preprocessor.)
19114     * The default debugging format for most ELF platforms (including
19115       GNU/Linux and FreeBSD; notable exception is Solaris) has changed
19116       from stabs to DWARF2. This requires GDB 5.1.1 or later.
19117
19118General Optimizer Improvements
19119
19120     * Jan Hubicka, SuSE Labs, together with Richard Henderson, Red Hat,
19121       and Andreas Jaeger, SuSE Labs, has contributed [2]infrastructure
19122       for profile driven optimizations.
19123       Options -fprofile-arcs and -fbranch-probabilities can now be used
19124       to improve speed of the generated code by profiling the actual
19125       program behaviour on typical runs. In the absence of profile info
19126       the compiler attempts to guess the profile statically.
19127     * [3]SPEC2000 and SPEC95 benchmark suites are now used daily to
19128       monitor performance of the generated code.
19129       According to the SPECInt2000 results on an AMD Athlon CPU, the code
19130       generated by GCC 3.1 is 6% faster on the average (8.2% faster with
19131       profile feedback) compared to GCC 3.0. The code produced by GCC 3.0
19132       is about 2.1% faster compared to 2.95.3. Tests were done using the
19133       -O2 -march=athlon command-line options.
19134     * Alexandre Oliva, of Red Hat, has generalized the tree inlining
19135       infrastructure developed by CodeSourcery, LLC for the C++ front
19136       end, so that it is now used in the C front end too. Inlining
19137       functions as trees exposes them earlier to the compiler, giving it
19138       more opportunities for optimization.
19139     * Support for data prefetching instructions has been added to the GCC
19140       back end and several targets. A new __builtin_prefetch intrinsic is
19141       available to explicitly insert prefetch instructions and
19142       experimental support for loop array prefetching has been added (see
19143       -fprefetch-loop-array documentation).
19144     * Support for emitting debugging information for macros has been
19145       added for DWARF2. It is activated using -g3.
19146
19147New Languages and Language specific improvements
19148
19149  C/C++
19150
19151     * A few more [4]ISO C99 features.
19152     * The preprocessor is 10-50% faster than the preprocessor in GCC 3.0.
19153     * The preprocessor's symbol table has been merged with the symbol
19154       table of the C, C++ and Objective-C front ends.
19155     * The preprocessor consumes less memory than the preprocessor in GCC
19156       3.0, often significantly so. On normal input files, it typically
19157       consumes less memory than pre-3.0 cccp-based GCC, too.
19158
19159  C++
19160
19161     * -fhonor-std and -fno-honor-std have been removed. -fno-honor-std
19162       was a workaround to allow std compliant code to work with the
19163       non-std compliant libstdc++-v2. libstdc++-v3 is std compliant.
19164     * The C++ ABI has been fixed so that void (A::*)() const is mangled
19165       as "M1AKFvvE", rather than "MK1AFvvE" as before. This change only
19166       affects pointer to cv-qualified member function types.
19167     * The C++ ABI has been changed to correctly handle this code:
19168    struct A {
19169      void operator delete[] (void *, size_t);
19170    };
19171
19172    struct B : public A {
19173    };
19174
19175    new B[10];
19176
19177       The amount of storage allocated for the array will be greater than
19178       it was in 3.0, in order to store the number of elements in the
19179       array, so that the correct size can be passed to operator delete[]
19180       when the array is deleted. Previously, the value passed to operator
19181       delete[] was unpredictable.
19182       This change will only affect code that declares a two-argument
19183       operator delete[] with a second parameter of type size_t in a base
19184       class, and does not override that definition in a derived class.
19185     * The C++ ABI has been changed so that:
19186    struct A {
19187      void operator delete[] (void *, size_t);
19188      void operator delete[] (void *);
19189    };
19190
19191       does not cause unnecessary storage to be allocated when an array of
19192       A objects is allocated.
19193       This change will only affect code that declares both of these forms
19194       of operator delete[], and declared the two-argument form before the
19195       one-argument form.
19196     * The C++ ABI has been changed so that when a parameter is passed by
19197       value, any cleanup for that parameter is performed in the caller,
19198       as specified by the ia64 C++ ABI, rather than the called function
19199       as before. As a result, classes with a non-trivial destructor but a
19200       trivial copy constructor will be passed and returned by invisible
19201       reference, rather than by bitwise copy as before.
19202     * G++ now supports the "named return value optimization": for code
19203       like
19204    A f () {
19205      A a;
19206      ...
19207      return a;
19208    }
19209
19210       G++ will allocate a in the return value slot, so that the return
19211       becomes a no-op. For this to work, all return statements in the
19212       function must return the same variable.
19213     * Improvements to the C++ library are listed in [5]the libstdc++-v3
19214       FAQ.
19215
19216  Objective-C
19217
19218     * Annoying linker warnings (due to incorrect code being generated)
19219       have been fixed.
19220     * If a class method cannot be found, the compiler no longer issues a
19221       warning if a corresponding instance method exists in the root
19222       class.
19223     * Forward @protocol declarations have been fixed.
19224     * Loading of categories has been fixed in certain situations (GNU run
19225       time only).
19226     * The class lookup in the run-time library has been rewritten so that
19227       class method dispatch is more than twice as fast as it used to be
19228       (GNU run time only).
19229
19230  Java
19231
19232     * libgcj now includes RMI, java.lang.ref.*, javax.naming, and
19233       javax.transaction.
19234     * Property files and other system resources can be compiled into
19235       executables which use libgcj using the new gcj --resource feature.
19236     * libgcj has been ported to more platforms. In particular there is
19237       now a mostly-functional mingw32 (Windows) target port.
19238     * JNI and CNI invocation interfaces were implemented, so gcj-compiled
19239       Java code can now be called from a C/C++ application.
19240     * gcj can now use builtin functions for certain known methods, for
19241       instance Math.cos.
19242     * gcj can now automatically remove redundant array-store checks in
19243       some common cases.
19244     * The --no-store-checks optimization option was added. This can be
19245       used to omit runtime store checks for code which is known not to
19246       throw ArrayStoreException
19247     * The following third party interface standards were added to libgcj:
19248       org.w3c.dom and org.xml.sax.
19249     * java.security has been merged with GNU Classpath. The new package
19250       is now JDK 1.2 compliant, and much more complete.
19251     * A bytecode verifier was added to the libgcj interpreter.
19252     * java.lang.Character was rewritten to comply with the Unicode 3.0
19253       standard, and improve performance.
19254     * Partial support for many more locales was added to libgcj.
19255     * Socket timeouts have been implemented.
19256     * libgcj has been merged into a single shared library. There are no
19257       longer separate shared libraries for the garbage collector and
19258       zlib.
19259     * Several performance improvements were made to gcj and libgcj:
19260          + Hash synchronization (thin locks)
19261          + A special allocation path for finalizer-free objects
19262          + Thread-local allocation
19263          + Parallel GC, and other GC tweaks
19264
19265  Fortran
19266
19267   Fortran improvements are listed in [6]the Fortran documentation.
19268
19269  Ada
19270
19271   [7]AdaCore, has contributed its GNAT Ada 95 front end and associated
19272   tools. The GNAT compiler fully implements the Ada language as defined
19273   by the ISO/IEC 8652 standard.
19274
19275   Please note that the integration of the Ada front end is still work in
19276   progress.
19277
19278New Targets and Target Specific Improvements
19279
19280     * Hans-Peter Nilsson has contributed a port to MMIX, the CPU
19281       architecture used in new editions of Donald E. Knuth's The Art of
19282       Computer Programming.
19283     * Axis Communications has contributed its port to the CRIS CPU
19284       architecture, used in the ETRAX system-on-a-chip series.
19285     * Alexandre Oliva, of Red Hat, has contributed a port to the SuperH
19286       SH5 64-bit RISC microprocessor architecture, extending the existing
19287       SH port.
19288     * UltraSPARC is fully supported in 64-bit mode. The option -m64
19289       enables it.
19290     * For compatibility with the Sun compiler #pragma redefine_extname
19291       has been implemented on Solaris.
19292     * The x86 back end has had some noticeable work done to it.
19293          + SuSE Labs developers Jan Hubicka, Bo Thorsen and Andreas
19294            Jaeger have contributed a port to the AMD x86-64 architecture.
19295            For more information on x86-64 see http://www.x86-64.org.
19296          + The compiler now supports MMX, 3DNow!, SSE, and SSE2
19297            instructions. Options -mmmx, -m3dnow, -msse, and -msse2 will
19298            enable the respective instruction sets. Intel C++ compatible
19299            MMX/3DNow!/SSE intrinsics are implemented. SSE2 intrinsics
19300            will be added in next major release.
19301          + Following those improvements, targets for Pentium MMX, K6-2,
19302            K6-3, Pentium III, Pentium 4, and Athlon 4 Mobile/XP/MP were
19303            added. Refer to the documentation on -march= and -mcpu=
19304            options for details.
19305          + For those targets that support it, -mfpmath=sse will cause the
19306            compiler to generate SSE/SSE2 instructions for floating point
19307            math instead of x87 instructions. Usually, this will lead to
19308            quicker code -- especially on the Pentium 4. Note that only
19309            scalar floating point instructions are used and GCC does not
19310            exploit SIMD features yet.
19311          + Prefetch support has been added to the Pentium III, Pentium 4,
19312            K6-2, K6-3, and Athlon series.
19313          + Code generated for floating point to integer conversions has
19314            been improved leading to better performance of many 3D
19315            applications.
19316     * The PowerPC back end has added 64-bit PowerPC GNU/Linux support.
19317     * C++ support for AIX has been improved.
19318     * Aldy Hernandez, of Red Hat, Inc, has contributed extensions to the
19319       PowerPC port supporting the AltiVec programming model (SIMD). The
19320       support, though presently useful, is experimental and is expected
19321       to stabilize for 3.2. The support is written to conform to
19322       Motorola's AltiVec specs. See -maltivec.
19323
19324Obsolete Systems
19325
19326   Support for a number of older systems has been declared obsolete in GCC
19327   3.1. Unless there is activity to revive them, the next release of GCC
19328   will have their sources permanently removed.
19329
19330   All configurations of the following processor architectures have been
19331   declared obsolete:
19332     * MIL-STD-1750A, 1750a-*-*
19333     * AMD A29k, a29k-*-*
19334     * Convex, c*-convex-*
19335     * Clipper, clipper-*-*
19336     * Elxsi, elxsi-*-*
19337     * Intel i860, i860-*-*
19338     * Sun picoJava, pj-*-* and pjl-*-*
19339     * Western Electric 32000, we32k-*-*
19340
19341   Most configurations of the following processor architectures have been
19342   declared obsolete, but we are preserving a few systems which may have
19343   active developers. It is unlikely that the remaining systems will
19344   survive much longer unless we see definite signs of port activity.
19345     * Motorola 88000 except
19346          + Generic a.out, m88k-*-aout*
19347          + Generic SVR4, m88k-*-sysv4
19348          + OpenBSD, m88k-*-openbsd*
19349     * NS32k except
19350          + NetBSD, ns32k-*-netbsd*
19351          + OpenBSD, ns32k-*-openbsd*.
19352     * ROMP except
19353          + OpenBSD, romp-*-openbsd*.
19354
19355   Finally, only some configurations of these processor architectures are
19356   being obsoleted.
19357     * Alpha:
19358          + OSF/1, alpha*-*-osf[123]*. (Digital Unix and Tru64 Unix, aka
19359            alpha*-*-osf[45], are still supported.)
19360     * ARM:
19361          + RISCiX, arm-*-riscix*.
19362     * i386:
19363          + 386BSD, i?86-*-bsd*
19364          + Chorus, i?86-*-chorusos*
19365          + DG/UX, i?86-*-dgux*
19366          + FreeBSD 1.x, i?86-*-freebsd1.*
19367          + IBM AIX, i?86-*-aix*
19368          + ISC UNIX, i?86-*-isc*
19369          + GNU/Linux with pre-BFD linker, i?86-*-linux*oldld*
19370          + NEXTstep, i?86-next-*
19371          + OSF UNIX, i?86-*-osf1* and i?86-*-osfrose*
19372          + RTEMS/coff, i?86-*-rtemscoff*
19373          + RTEMS/go32, i?86-go32-rtems*
19374          + Sequent/BSD, i?86-sequent-bsd*
19375          + Sequent/ptx before version 3, i?86-sequent-ptx[12]* and
19376            i?86-sequent-sysv3*
19377          + SunOS, i?86-*-sunos*
19378     * Motorola 68000:
19379          + Altos, m68[k0]*-altos-*
19380          + Apollo, m68[k0]*-apollo-*
19381          + Apple A/UX, m68[k0]*-apple-*
19382          + Bull, m68[k0]*-bull-*
19383          + Convergent, m68[k0]*-convergent-*
19384          + Generic SVR3, m68[k0]*-*-sysv3*
19385          + ISI, m68[k0]*-isi-*
19386          + LynxOS, m68[k0]*-*-lynxos*
19387          + NEXT, m68[k0]*-next-*
19388          + RTEMS/coff, m68[k0]*-*-rtemscoff*
19389          + Sony, m68[k0]*-sony-*
19390     * MIPS:
19391          + DEC Ultrix, mips-*-ultrix* and mips-dec-*
19392          + Generic BSD, mips-*-bsd*
19393          + Generic System V, mips-*-sysv*
19394          + IRIX before version 5, mips-sgi-irix[1234]*
19395          + RiscOS, mips-*-riscos*
19396          + Sony, mips-sony-*
19397          + Tandem, mips-tandem-*
19398     * SPARC:
19399          + RTEMS/a.out, sparc-*-rtemsaout*.
19400
19401Documentation improvements
19402
19403     * The old manual ("Using and Porting the GNU Compiler Collection")
19404       has been replaced by a users manual ("Using the GNU Compiler
19405       Collection") and a separate internals reference manual ("GNU
19406       Compiler Collection Internals").
19407     * More complete and much improved documentation about GCC's internal
19408       representation used by the C and C++ front ends.
19409     * Many cleanups and improvements in general.
19410
19411
19412    For questions related to the use of GCC, please consult these web
19413    pages and the [8]GCC manuals. If that fails, the
19414    [9]gcc-help@gcc.gnu.org mailing list might help. Comments on these
19415    web pages and the development of GCC are welcome on our developer
19416    list at [10]gcc@gcc.gnu.org. All of [11]our lists have public
19417    archives.
19418
19419   Copyright (C) [12]Free Software Foundation, Inc. Verbatim copying and
19420   distribution of this entire article is permitted in any medium,
19421   provided this notice is preserved.
19422
19423   These pages are [13]maintained by the GCC team. Last modified
19424   2019-11-28[14].
19425
19426References
19427
19428   1. https://gcc.gnu.org/ml/gcc/2002-07/msg01208.html
19429   2. http://gcc.gnu.org/news/profiledriven.html
19430   3. http://gcc.gnu.org/benchmarks/
19431   4. http://gcc.gnu.org/c99status.html
19432   5. https://gcc.gnu.org/onlinedocs/libstdc++/faq.html
19433   6. https://gcc.gnu.org/onlinedocs/gcc-3.1.1/g77/News.html
19434   7. https://www.adacore.com/
19435   8. https://gcc.gnu.org/onlinedocs/
19436   9. mailto:gcc-help@gcc.gnu.org
19437  10. mailto:gcc@gcc.gnu.org
19438  11. https://gcc.gnu.org/lists.html
19439  12. https://www.fsf.org/
19440  13. https://gcc.gnu.org/about.html
19441  14. http://validator.w3.org/check/referer
19442======================================================================
19443http://gcc.gnu.org/gcc-3.0/index.html
19444
19445                                   GCC 3.0.4
19446
19447   (This release series is no longer supported.)
19448
19449   February 20, 2002
19450
19451   The [1]GNU project and the GCC developers are pleased to announce the
19452   release of GCC 3.0.4, which is a bug-fix release for the GCC 3.0
19453   series.
19454
19455   GCC used to stand for the GNU C Compiler, but since the compiler
19456   supports several other languages aside from C, it now stands for the
19457   GNU Compiler Collection.
19458
19459   GCC 3.0.x has several new optimizations, new targets, new languages and
19460   many other new features, relative to GCC 2.95.x. See the [2]new
19461   features page for a more complete list.
19462
19463   A list of [3]successful builds is updated as new information becomes
19464   available.
19465
19466   The GCC developers would like to thank the numerous people that have
19467   contributed new features, test results, bug fixes, etc to GCC. This
19468   [4]amazing group of volunteers is what makes GCC successful.
19469
19470   And finally, we can't in good conscience fail to mention some
19471   [5]caveats to using GCC 3.0.x.
19472
19473   For additional information about GCC please refer to the [6]GCC project
19474   web site or contact the [7]GCC development mailing list.
19475
19476   To obtain GCC please use [8]our mirror sites, or our CVS server.
19477     __________________________________________________________________
19478
19479Previous 3.0.x Releases
19480
19481   December 20, 2001: GCC 3.0.3 has been released.
19482   October 25, 2001: GCC 3.0.2 has been released.
19483   August 20, 2001: GCC 3.0.1 has been released.
19484   June 18, 2001: GCC 3.0 has been released.
19485
19486
19487    For questions related to the use of GCC, please consult these web
19488    pages and the [9]GCC manuals. If that fails, the
19489    [10]gcc-help@gcc.gnu.org mailing list might help. Comments on these
19490    web pages and the development of GCC are welcome on our developer
19491    list at [11]gcc@gcc.gnu.org. All of [12]our lists have public
19492    archives.
19493
19494   Copyright (C) [13]Free Software Foundation, Inc. Verbatim copying and
19495   distribution of this entire article is permitted in any medium,
19496   provided this notice is preserved.
19497
19498   These pages are [14]maintained by the GCC team. Last modified
19499   2019-11-28[15].
19500
19501References
19502
19503   1. http://www.gnu.org/
19504   2. http://gcc.gnu.org/gcc-3.0/features.html
19505   3. http://gcc.gnu.org/gcc-3.0/buildstat.html
19506   4. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
19507   5. http://gcc.gnu.org/gcc-3.0/caveats.html
19508   6. http://gcc.gnu.org/index.html
19509   7. mailto:gcc@gcc.gnu.org
19510   8. http://gcc.gnu.org/mirrors.html
19511   9. https://gcc.gnu.org/onlinedocs/
19512  10. mailto:gcc-help@gcc.gnu.org
19513  11. mailto:gcc@gcc.gnu.org
19514  12. https://gcc.gnu.org/lists.html
19515  13. https://www.fsf.org/
19516  14. https://gcc.gnu.org/about.html
19517  15. http://validator.w3.org/check/referer
19518======================================================================
19519http://gcc.gnu.org/gcc-3.0/features.html
19520
19521                              GCC 3.0 New Features
19522
19523Additional changes in GCC 3.0.4
19524
19525     * GCC 3.0 now supports newer versions of the [1]NetBSD operating
19526       system, which use the ELF object file format, on x86 processors.
19527     * Correct debugging information is generated from functions that have
19528       lines from multiple files (e.g. yacc output).
19529     * A fix for whitespace handling in the -traditional preprocessor,
19530       which can affect Fortran.
19531     * Fixes to the exception handling runtime.
19532     * More fixes for bad code generation in C++.
19533     * A fix for shared library generation under AIX 4.3.
19534     * Documentation updates.
19535     * Port of GCC to Tensilica's Xtensa processor contributed.
19536     * A fix for compiling the PPC Linux kernel (FAT fs wouldn't link).
19537
19538Additional changes in GCC 3.0.3
19539
19540     * A fix to correct an accidental change to the PowerPC ABI.
19541     * Fixes for bad code generation on a variety of architectures.
19542     * Improvements to the debugging information generated for C++
19543       classes.
19544     * Fixes for bad code generation in C++.
19545     * A fix to avoid crashes in the C++ demangler.
19546     * A fix to the C++ standard library to avoid buffer overflows.
19547     * Miscellaneous improvements for a variety of architectures.
19548
19549Additional changes in GCC 3.0.2
19550
19551     * Fixes for bad code generation during loop unrolling.
19552     * Fixes for bad code generation by the sibling call optimization.
19553     * Minor improvements to x86 code generation.
19554     * Implementation of function descriptors in C++ vtables for IA64.
19555     * Numerous minor bug-fixes.
19556
19557Additional changes in GCC 3.0.1
19558
19559     * C++ fixes for incorrect code-generation.
19560     * Improved cross-compiling support for the C++ standard library.
19561     * Fixes for some embedded targets that worked in GCC 2.95.3, but not
19562       in GCC 3.0.
19563     * Fixes for various exception-handling bugs.
19564     * A port to the S/390 architecture.
19565
19566General Optimizer Improvements
19567
19568     * [2]Basic block reordering pass.
19569     * New if-conversion pass with support for conditional (predicated)
19570       execution.
19571     * New tail call and sibling call elimination optimizations.
19572     * New register renaming pass.
19573     * New (experimental) [3]static single assignment (SSA) representation
19574       support.
19575     * New dead-code elimination pass implemented using the SSA
19576       representation.
19577     * [4]Global null pointer test elimination.
19578     * [5]Global code hoisting/unification.
19579     * More builtins and optimizations for stdio.h, string.h and old BSD
19580       functions, as well as for ISO C99 functions.
19581     * New builtin __builtin_expect for giving hints to the branch
19582       predictor.
19583
19584New Languages and Language specific improvements
19585
19586     * The GNU Compiler for the Java(TM) language (GCJ) is now integrated
19587       and supported, including the run-time library containing most
19588       common non-GUI Java classes, a bytecode interpreter, and the Boehm
19589       conservative garbage collector. Many bugs have been fixed. GCJ can
19590       compile Java source or Java bytecodes to either native code or Java
19591       class files, and supports native methods written in either the
19592       standard JNI or the more efficient and convenient CNI.
19593     * Here is a [6]partial list of C++ improvements, both new features
19594       and those no longer supported.
19595     * New C++ ABI. On the IA-64 platform GCC is capable of
19596       inter-operating with other IA-64 compilers.
19597     * The new ABI also significantly reduces the size of symbol and debug
19598       information.
19599     * New C++ support library and many C++ bug fixes, vastly improving
19600       our conformance to the ISO C++ standard.
19601     * New [7]inliner for C++.
19602     * Rewritten C preprocessor, integrated into the C, C++ and Objective
19603       C compilers, with very many improvements including ISO C99 support
19604       and [8]improvements to dependency generation.
19605     * Support for more [9]ISO C99 features.
19606     * Many improvements to support for checking calls to format functions
19607       such as printf and scanf, including support for ISO C99 format
19608       features, extensions from the Single Unix Specification and GNU
19609       libc 2.2, checking of strfmon formats and features to assist in
19610       auditing for format string security bugs.
19611     * New warnings for C code that may have undefined semantics because
19612       of violations of sequence point rules in the C standard (such as a
19613       = a++;, a[n] = b[n++]; and a[i++] = i;), included in -Wall.
19614     * Additional warning option -Wfloat-equal.
19615     * Improvements to -Wtraditional.
19616     * Fortran improvements are listed in [10]the Fortran documentation.
19617
19618New Targets and Target Specific Improvements
19619
19620     * New x86 back end, generating much improved code.
19621     * Support for a generic i386-elf target contributed.
19622     * New option to emit x86 assembly code using Intel style syntax
19623       (-mintel-syntax).
19624     * HPUX 11 support contributed.
19625     * Improved PowerPC code generation, including scheduled prologue and
19626       epilogue.
19627     * Port of GCC to Intel's IA-64 processor contributed.
19628     * Port of GCC to Motorola's MCore 210 and 340 contributed.
19629     * New unified back-end for Arm, Thumb and StrongArm contributed.
19630     * Port of GCC to Intel's XScale processor contributed.
19631     * Port of GCC to Atmel's AVR microcontrollers contributed.
19632     * Port of GCC to Mitsubishi's D30V processor contributed.
19633     * Port of GCC to Matsushita's AM33 processor (a member of the MN10300
19634       processor family) contributed.
19635     * Port of GCC to Fujitsu's FR30 processor contributed.
19636     * Port of GCC to Motorola's 68HC11 and 68HC12 processors contributed.
19637     * Port of GCC to Sun's picoJava processor core contributed.
19638
19639Documentation improvements
19640
19641     * Substantially rewritten and improved C preprocessor manual.
19642     * Many improvements to other documentation.
19643     * Manpages for gcc, cpp and gcov are now generated automatically from
19644       the master Texinfo manual, eliminating the problem of manpages
19645       being out of date. (The generated manpages are only extracts from
19646       the full manual, which is provided in Texinfo form, from which
19647       info, HTML, other formats and a printed manual can be generated.)
19648     * Generated info files are included in the release tarballs alongside
19649       their Texinfo sources, avoiding problems on some platforms with
19650       building makeinfo as part of the GCC distribution.
19651
19652Other significant improvements
19653
19654     * Garbage collection used internally by the compiler for most memory
19655       allocation instead of obstacks.
19656     * Lengauer and Tarjan algorithm used for computing dominators in the
19657       CFG. This algorithm can be significantly faster and more space
19658       efficient than our older algorithm.
19659     * gccbug script provided to assist in submitting bug reports to our
19660       bug tracking system. (Bug reports previously submitted directly to
19661       our mailing lists, for which you received no bug tracking number,
19662       should be submitted again using gccbug if you can reproduce the
19663       problem with GCC 3.0.)
19664     * The internal libgcc library is [11]built as a shared library on
19665       systems that support it.
19666     * Extensive testsuite included with GCC, with many new tests. In
19667       addition to tests for GCC bugs that have been fixed, many tests
19668       have been added for language features, compiler warnings and
19669       builtin functions.
19670     * Additional language-independent warning options -Wpacked, -Wpadded,
19671       -Wunreachable-code and -Wdisabled-optimization.
19672     * Target-independent options -falign-functions, -falign-loops and
19673       -falign-jumps.
19674
19675   Plus a great many bug fixes and almost all the [12]features found in
19676   GCC 2.95.
19677
19678
19679    For questions related to the use of GCC, please consult these web
19680    pages and the [13]GCC manuals. If that fails, the
19681    [14]gcc-help@gcc.gnu.org mailing list might help. Comments on these
19682    web pages and the development of GCC are welcome on our developer
19683    list at [15]gcc@gcc.gnu.org. All of [16]our lists have public
19684    archives.
19685
19686   Copyright (C) [17]Free Software Foundation, Inc. Verbatim copying and
19687   distribution of this entire article is permitted in any medium,
19688   provided this notice is preserved.
19689
19690   These pages are [18]maintained by the GCC team. Last modified
19691   2019-11-28[19].
19692
19693References
19694
19695   1. http://www.netbsd.org/
19696   2. http://gcc.gnu.org/news/reorder.html
19697   3. http://gcc.gnu.org/news/ssa.html
19698   4. http://gcc.gnu.org/news/null.html
19699   5. http://gcc.gnu.org/news/unify.html
19700   6. http://gcc.gnu.org/gcc-3.0/c++features.html
19701   7. http://gcc.gnu.org/news/inlining.html
19702   8. http://gcc.gnu.org/news/dependencies.html
19703   9. http://gcc.gnu.org/c99status.html
19704  10. https://gcc.gnu.org/onlinedocs/gcc-3.4.6/g77/News.html
19705  11. http://gcc.gnu.org/gcc-3.0/libgcc.html
19706  12. http://gcc.gnu.org/gcc-2.95/features.html
19707  13. https://gcc.gnu.org/onlinedocs/
19708  14. mailto:gcc-help@gcc.gnu.org
19709  15. mailto:gcc@gcc.gnu.org
19710  16. https://gcc.gnu.org/lists.html
19711  17. https://www.fsf.org/
19712  18. https://gcc.gnu.org/about.html
19713  19. http://validator.w3.org/check/referer
19714======================================================================
19715http://gcc.gnu.org/gcc-3.0/caveats.html
19716
19717                                GCC 3.0 Caveats
19718
19719     * -fstrict-aliasing is now part of -O2 and higher optimization
19720       levels. This allows the compiler to assume the strictest aliasing
19721       rules applicable to the language being compiled. For C and C++,
19722       this activates optimizations based on the type of expressions. This
19723       optimization may thus break old, non-compliant code.
19724     * Enumerations are now properly promoted to int in function
19725       parameters and function returns. Normally this change is not
19726       visible, but when using -fshort-enums this is an ABI change.
19727     * The undocumented extension that allowed C programs to have a label
19728       at the end of a compound statement has been deprecated and may be
19729       removed in a future version. Programs that now generate a warning
19730       about this may be fixed by adding a null statement (a single
19731       semicolon) after the label.
19732     * The poorly documented extension that allowed string constants in C,
19733       C++ and Objective C to contain unescaped newlines has been
19734       deprecated and may be removed in a future version. Programs using
19735       this extension may be fixed in several ways: the bare newline may
19736       be replaced by \n, or preceded by \n\, or string concatenation may
19737       be used with the bare newline preceded by \n" and " placed at the
19738       start of the next line.
19739     * The Chill compiler is not included in GCC 3.0, because of the lack
19740       of a volunteer to convert it to use garbage collection.
19741     * Certain non-standard iostream methods from earlier versions of
19742       libstdc++ are not included in libstdc++ v3, i.e. filebuf::attach,
19743       ostream::form, and istream::gets.
19744     * The new C++ ABI is not yet fully supported by current (as of
19745       2001-07-01) releases and development versions of GDB, or any
19746       earlier versions. There is a problem setting breakpoints by line
19747       number, and other related issues that have been fixed in GCC 3.0
19748       but not yet handled in GDB:
19749       [1]https://gcc.gnu.org/ml/gcc-bugs/2001-06/msg00421.html
19750
19751
19752    For questions related to the use of GCC, please consult these web
19753    pages and the [2]GCC manuals. If that fails, the
19754    [3]gcc-help@gcc.gnu.org mailing list might help. Comments on these
19755    web pages and the development of GCC are welcome on our developer
19756    list at [4]gcc@gcc.gnu.org. All of [5]our lists have public archives.
19757
19758   Copyright (C) [6]Free Software Foundation, Inc. Verbatim copying and
19759   distribution of this entire article is permitted in any medium,
19760   provided this notice is preserved.
19761
19762   These pages are [7]maintained by the GCC team. Last modified
19763   2019-11-28[8].
19764
19765References
19766
19767   1. https://gcc.gnu.org/ml/gcc-bugs/2001-06/msg00421.html
19768   2. https://gcc.gnu.org/onlinedocs/
19769   3. mailto:gcc-help@gcc.gnu.org
19770   4. mailto:gcc@gcc.gnu.org
19771   5. https://gcc.gnu.org/lists.html
19772   6. https://www.fsf.org/
19773   7. https://gcc.gnu.org/about.html
19774   8. http://validator.w3.org/check/referer
19775======================================================================
19776http://gcc.gnu.org/gcc-2.95/index.html
19777
19778                                    GCC 2.95
19779
19780   (This release series is no longer supported.)
19781
19782   March 16, 2001: The GNU project and the GCC developers are pleased to
19783   announce the release of GCC version 2.95.3.
19784
19785Release History
19786
19787   GCC 2.95.3
19788          March 16, 2001
19789
19790   GCC 2.95.2
19791          October 27, 1999
19792
19793   GCC 2.95.1
19794          August 19, 1999
19795
19796   GCC 2.95
19797          July 31, 1999. This is the first release of GCC since the April
19798          1999 GCC/EGCS reunification and includes nearly a year's worth
19799          of new development and bugfixes.
19800
19801References and Acknowledgements
19802
19803   GCC used to stand for the GNU C Compiler, but since the compiler
19804   supports several other languages aside from C, it now stands for the
19805   GNU Compiler Collection.
19806
19807   The whole suite has been extensively [1]regression tested and
19808   [2]package tested. It should be reliable and suitable for widespread
19809   use.
19810
19811   The compiler has several new optimizations, new targets, new languages
19812   and other new features. See the [3]new features page for a more
19813   complete list of new features found in the GCC 2.95 releases.
19814
19815   The sources include installation instructions in both HTML and
19816   plaintext forms in the install directory in the distribution. However,
19817   the most up to date installation instructions and [4]build/test status
19818   are on the web pages. We will update those pages as new information
19819   becomes available.
19820
19821   The GCC developers would like to thank the numerous people that have
19822   contributed new features, test results, bugfixes, etc to GCC. This
19823   [5]amazing group of volunteers is what makes GCC successful.
19824
19825   And finally, we can't in good conscience fail to mention some
19826   [6]caveats to using GCC 2.95.
19827
19828   Download GCC 2.95 from one of our many [7]mirror sites.
19829
19830   For additional information about GCC please see the [8]GCC project web
19831   server or contact the [9]GCC development mailing list.
19832
19833
19834    For questions related to the use of GCC, please consult these web
19835    pages and the [10]GCC manuals. If that fails, the
19836    [11]gcc-help@gcc.gnu.org mailing list might help. Comments on these
19837    web pages and the development of GCC are welcome on our developer
19838    list at [12]gcc@gcc.gnu.org. All of [13]our lists have public
19839    archives.
19840
19841   Copyright (C) [14]Free Software Foundation, Inc. Verbatim copying and
19842   distribution of this entire article is permitted in any medium,
19843   provided this notice is preserved.
19844
19845   These pages are [15]maintained by the GCC team. Last modified
19846   2019-11-28[16].
19847
19848References
19849
19850   1. http://gcc.gnu.org/gcc-2.95/regress.html
19851   2. http://gcc.gnu.org/gcc-2.95/othertest.html
19852   3. http://gcc.gnu.org/gcc-2.95/features.html
19853   4. http://gcc.gnu.org/gcc-2.95/buildstat.html
19854   5. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
19855   6. http://gcc.gnu.org/gcc-2.95/caveats.html
19856   7. http://gcc.gnu.org/mirrors.html
19857   8. http://gcc.gnu.org/index.html
19858   9. mailto:gcc@gcc.gnu.org
19859  10. https://gcc.gnu.org/onlinedocs/
19860  11. mailto:gcc-help@gcc.gnu.org
19861  12. mailto:gcc@gcc.gnu.org
19862  13. https://gcc.gnu.org/lists.html
19863  14. https://www.fsf.org/
19864  15. https://gcc.gnu.org/about.html
19865  16. http://validator.w3.org/check/referer
19866======================================================================
19867http://gcc.gnu.org/gcc-2.95/features.html
19868
19869                             GCC 2.95 New Features
19870
19871     * General Optimizer Improvements:
19872          + [1]Localized register spilling to improve speed and code
19873            density especially on small register class machines.
19874          + [2]Global CSE using lazy code motion algorithms.
19875          + [3]Improved global constant/copy propagation.
19876          + [4]Improved control flow graph analysis and manipulation.
19877          + [5]Local dead store elimination.
19878          + [6]Memory Load hoisting/store sinking in loops.
19879          + [7]Type based alias analysis is enabled by default. Note this
19880            feature will expose bugs in the Linux kernel. Please refer to
19881            the FAQ (as shipped with GCC 2.95) for additional information
19882            on this issue.
19883          + Major revamp of GIV detection, combination and simplification
19884            to improve loop performance.
19885          + Major improvements to register allocation and reloading.
19886     * New Languages and Language specific improvements
19887          + [8]Many C++ improvements.
19888          + [9]Many Fortran improvements.
19889          + [10]Java front-end has been integrated. A [11]runtime library
19890            is available separately.
19891          + [12]ISO C99 support
19892          + [13]Chill front-end and runtime has been integrated.
19893          + Boehm garbage collector support in libobjc.
19894          + More support for various pragmas which appear in vendor
19895            include files
19896     * New Targets and Target Specific Improvements
19897          + [14]SPARC backend rewrite.
19898          + -mschedule=8000 will optimize code for PA8000 class
19899            processors; -mpa-risc-2-0 will generate code for PA2.0
19900            processors
19901          + Various micro-optimizations for the ia32 port. K6
19902            optimizations
19903          + Compiler will attempt to align doubles in the stack on the
19904            ia32 port
19905          + Alpha EV6 support
19906          + PowerPC 750
19907          + RS6000/PowerPC: -mcpu=401 was added as an alias for -mcpu=403.
19908            -mcpu=e603e was added to do -mcpu=603e and -msoft-float.
19909          + c3x, c4x
19910          + HyperSPARC
19911          + SparcLite86x
19912          + sh4
19913          + Support for new systems (OpenBSD, FreeBSD, UWIN, Interix,
19914            arm-linux)
19915          + vxWorks targets include support for vxWorks threads
19916          + StrongARM 110 and ARM9 support added. ARM Scheduling
19917            parameters rewritten.
19918          + Various changes to the MIPS port to avoid assembler macros,
19919            which in turn improves performance
19920          + Various performance improvements to the i960 port.
19921          + Major rewrite of ns32k port
19922     * Other significant improvements
19923          + [15]Ability to dump cfg information and display it using vcg.
19924          + The new faster scheme for fixing vendor header files is
19925            enabled by default.
19926          + Experimental internationalization support.
19927          + multibyte character support
19928          + Some compile-time speedups for pathological problems
19929          + Better support for complex types
19930     * Plus the usual mountain of bugfixes
19931     * Core compiler is based on the gcc2 development tree from Sept 30,
19932       1998, so we have all of the [16]features found in GCC 2.8.
19933
19934Additional Changes in GCC 2.95.1
19935
19936     * Generic bugfixes and improvements
19937          + Various documentation fixes related to the GCC/EGCS merger.
19938          + Fix memory management bug which could lead to spurious aborts,
19939            core dumps or random parsing errors in the compiler.
19940          + Fix a couple bugs in the dwarf1 and dwarf2 debug record
19941            support.
19942          + Fix infinite loop in the CSE optimizer.
19943          + Avoid undefined behavior in compiler FP emulation code
19944          + Fix install problem when prefix is overridden on the make
19945            install command.
19946          + Fix problem with unwanted installation of assert.h on some
19947            systems.
19948          + Fix problem with finding the wrong assembler in a single tree
19949            build.
19950          + Avoid increasing the known alignment of a register that is
19951            already known to be a pointer.
19952     * Platform specific bugfixes and improvements
19953          + Codegen bugfix for prologue/epilogue for cpu32 target.
19954          + Fix long long code generation bug for the Coldfire target.
19955          + Fix various aborts in the SH compiler.
19956          + Fix bugs in libgcc support library for the SH.
19957          + Fix alpha ev6 code generation bug.
19958          + Fix problems with EXIT_SUCCESS/EXIT_FAILURE redefinitions on
19959            AIX platforms.
19960          + Fix -fpic code generation bug for rs6000/ppc svr4 targets.
19961          + Fix varargs/stdarg code generation bug for rs6000/ppc svr4
19962            targets.
19963          + Fix weak symbol handling for rs6000/ppc svr4 targets.
19964          + Fix various problems with 64bit code generation for the
19965            rs6000/ppc port.
19966          + Fix codegen bug which caused tetex to be mis-compiled on the
19967            x86.
19968          + Fix compiler abort in new cfg code exposed by x86 port.
19969          + Fix out of range array reference in code convert flat
19970            registers to the x87 stacked FP register file.
19971          + Fix minor vxworks configuration bug.
19972          + Fix return type of bsearch for SunOS 4.x.
19973     * Language & Runtime specific fixes.
19974          + The G++ signature extension has been deprecated. It will be
19975            removed in the next major release of G++. Use of signatures
19976            will result in a warning from the compiler.
19977          + Several bugs relating to templates and namespaces were fixed.
19978          + A bug that caused crashes when combining templates with -g on
19979            DWARF1 platforms was fixed.
19980          + Pointers-to-members, virtual functions, and multiple
19981            inheritance should now work together correctly.
19982          + Some code-generation bugs relating to function try blocks were
19983            fixed.
19984          + G++ is a little bit more lenient with certain archaic
19985            constructs than in GCC 2.95.
19986          + Fix to prevent shared library version #s from bring truncated
19987            to 1 digit
19988          + Fix missing std:: in the libstdc++ library.
19989          + Fix stream locking problems in libio.
19990          + Fix problem in java compiler driver.
19991
19992Additional Changes in GCC 2.95.2
19993
19994   The -fstrict-aliasing is not enabled by default for GCC 2.95.2. While
19995   the optimizations performed by -fstrict-aliasing are valid according to
19996   the C and C++ standards, the optimization have caused some problems,
19997   particularly with old non-conforming code.
19998
19999   The GCC developers are experimenting with ways to warn users about code
20000   which violates the C/C++ standards, but those warnings are not ready
20001   for widespread use at this time. Rather than wait for those warnings
20002   the GCC developers have chosen to disable -fstrict-aliasing by default
20003   for the GCC 2.95.2 release.
20004
20005   We strongly encourage developers to find and fix code which violates
20006   the C/C++ standards as -fstrict-aliasing may be enabled by default in
20007   future releases. Use the option -fstrict-aliasing to re-enable these
20008   optimizations.
20009     * Generic bugfixes and improvements
20010          + Fix incorrectly optimized memory reference in global common
20011            subexpression elimination (GCSE) optimization pass.
20012          + Fix code generation bug in regmove.c in which it could
20013            incorrectly change a "const" value.
20014          + Fix bug in optimization of conditionals involving volatile
20015            memory references.
20016          + Avoid over-allocation of stack space for some procedures.
20017          + Fixed bug in the compiler which caused incorrect optimization
20018            of an obscure series of bit manipulations, shifts and
20019            arithmetic.
20020          + Fixed register allocator bug which caused teTeX to be
20021            mis-compiled on SPARC targets.
20022          + Avoid incorrect optimization of degenerate case statements for
20023            certain targets such as the ARM.
20024          + Fix out of range memory reference in the jump optimizer.
20025          + Avoid dereferencing null pointer in fix-header.
20026          + Fix test for GCC specific features so that it is possible to
20027            bootstrap with gcc-2.6.2 and older versions of GCC.
20028          + Fix typo in scheduler which could potentially cause out of
20029            range memory accesses.
20030          + Avoid incorrect loop reversal which caused incorrect code for
20031            certain loops on PowerPC targets.
20032          + Avoid incorrect optimization of switch statements on certain
20033            targets (for example the ARM).
20034     * Platform specific bugfixes and improvements
20035          + Work around bug in Sun V5.0 compilers which caused bootstrap
20036            comparison failures on SPARC targets.
20037          + Fix SPARC backend bug which caused aborts in final.c.
20038          + Fix sparc-hal-solaris2* configuration fragments.
20039          + Fix bug in sparc block profiling.
20040          + Fix obscure code generation bug for the PARISC targets.
20041          + Define __STDC_EXT__ for HPUX configurations.
20042          + Various POWERPC64 code generation bugfixes.
20043          + Fix abort for PPC targets using ELF (ex GNU/Linux).
20044          + Fix collect2 problems for AIX targets.
20045          + Correct handling of .file directive for PPC targets.
20046          + Fix bug in fix_trunc x86 patterns.
20047          + Fix x86 port to correctly pop the FP stack for functions that
20048            return structures in memory.
20049          + Fix minor bug in strlen x86 pattern.
20050          + Use stabs debugging instead of dwarf1 for x86-solaris targets.
20051          + Fix template repository code to handle leading underscore in
20052            mangled names.
20053          + Fix weak/weak alias support for OpenBSD.
20054          + GNU/Linux for the ARM has C++ compatible include files.
20055     * Language & Runtime specific fixes.
20056          + Fix handling of constructor attribute in the C front-end which
20057            caused problems building the Chill runtime library on some
20058            targets.
20059          + Fix minor problem merging type qualifiers in the C front-end.
20060          + Fix aliasing bug for pointers and references (C/C++).
20061          + Fix incorrect "non-constant initializer bug" when -traditional
20062            or -fwritable-strings is enabled.
20063          + Fix build error for Chill front-end on SunOS.
20064          + Do not complain about duplicate instantiations when using
20065            -frepo (C++).
20066          + Fix array bounds handling in C++ front-end which caused
20067            problems with dwarf debugging information in some
20068            circumstances.
20069          + Fix minor namespace problem.
20070          + Fix problem linking java programs.
20071
20072Additional Changes in GCC 2.95.3
20073
20074     * Generic bugfixes and improvements
20075          + Fix numerous problems that caused incorrect optimization in
20076            the register reloading code.
20077          + Fix numerous problems that caused incorrect optimization in
20078            the loop optimizer.
20079          + Fix aborts in the functions build_insn_chain and scan_loops
20080            under some circumstances.
20081          + Fix an alias analysis bug.
20082          + Fix an infinite compilation bug in the combiner.
20083          + A few problems with complex number support have been fixed.
20084          + It is no longer possible for gcc to act as a fork bomb when
20085            installed incorrectly.
20086          + The -fpack-struct option should be recognized now.
20087          + Fixed a bug that caused incorrect code to be generated due to
20088            a lost stack adjustment.
20089     * Platform specific bugfixes and improvements
20090          + Support building ARM toolchains hosted on Windows.
20091          + Fix attribute calculations in ARM toolchains.
20092          + arm-linux support has been improved.
20093          + Fix a PIC failure on sparc targets.
20094          + On ix86 targets, the regparm attribute should now work
20095            reliably.
20096          + Several updates for the h8300 port.
20097          + Fix problem building libio with glibc 2.2.
20098
20099
20100    For questions related to the use of GCC, please consult these web
20101    pages and the [17]GCC manuals. If that fails, the
20102    [18]gcc-help@gcc.gnu.org mailing list might help. Comments on these
20103    web pages and the development of GCC are welcome on our developer
20104    list at [19]gcc@gcc.gnu.org. All of [20]our lists have public
20105    archives.
20106
20107   Copyright (C) [21]Free Software Foundation, Inc. Verbatim copying and
20108   distribution of this entire article is permitted in any medium,
20109   provided this notice is preserved.
20110
20111   These pages are [22]maintained by the GCC team. Last modified
20112   2019-11-28[23].
20113
20114References
20115
20116   1. http://gcc.gnu.org/news/spill.html
20117   2. http://gcc.gnu.org/news/lcm.html
20118   3. http://gcc.gnu.org/news/cprop.html
20119   4. http://gcc.gnu.org/news/cfg.html
20120   5. http://gcc.gnu.org/news/dse.html
20121   6. http://gcc.gnu.org/news/hoist.html
20122   7. http://gcc.gnu.org/news/alias.html
20123   8. http://gcc.gnu.org/gcc-2.95/c++features.html
20124   9. https://gcc.gnu.org/onlinedocs/gcc-3.4.6/g77/News.html
20125  10. http://gcc.gnu.org/news/gcj-announce.txt
20126  11. http://gcc.gnu.org/news/javaannounce.html
20127  12. http://gcc.gnu.org/c99status.html
20128  13. http://gcc.gnu.org/news/chill.html
20129  14. http://gcc.gnu.org/news/sparc.html
20130  15. http://gcc.gnu.org/news/egcs-vcg.html
20131  16. http://gcc.gnu.org/egcs-1.0/features-2.8.html
20132  17. https://gcc.gnu.org/onlinedocs/
20133  18. mailto:gcc-help@gcc.gnu.org
20134  19. mailto:gcc@gcc.gnu.org
20135  20. https://gcc.gnu.org/lists.html
20136  21. https://www.fsf.org/
20137  22. https://gcc.gnu.org/about.html
20138  23. http://validator.w3.org/check/referer
20139======================================================================
20140http://gcc.gnu.org/gcc-2.95/caveats.html
20141
20142                                GCC 2.95 Caveats
20143
20144     * GCC 2.95 will issue an error for invalid asm statements that had
20145       been silently accepted by earlier versions of the compiler. This is
20146       particularly noticeable when compiling older versions of the Linux
20147       kernel (2.0.xx). Please refer to the FAQ (as shipped with GCC 2.95)
20148       for more information on this issue.
20149     * GCC 2.95 implements type based alias analysis to disambiguate
20150       memory references. Some programs, particularly the Linux kernel
20151       violate ANSI/ISO aliasing rules and therefore may not operate
20152       correctly when compiled with GCC 2.95. Please refer to the FAQ (as
20153       shipped with GCC 2.95) for more information on this issue.
20154     * GCC 2.95 has a known bug in its handling of complex variables for
20155       64bit targets. Instead of silently generating incorrect code, GCC
20156       2.95 will issue a fatal error for situations it can not handle.
20157       This primarily affects the Fortran community as Fortran makes more
20158       use of complex variables than C or C++.
20159     * GCC 2.95 has an integrated libstdc++, but does not have an
20160       integrated libg++. Furthermore old libg++ releases will not work
20161       with GCC 2.95. You can retrieve a recent copy of libg++ from the
20162       [1]GCC ftp server.
20163       Note most C++ programs only need libstdc++.
20164     * Exception handling may not work with shared libraries, particularly
20165       on alphas, hppas, rs6000/powerpc and mips based platforms.
20166       Exception handling is known to work on x86 GNU/Linux platforms with
20167       shared libraries.
20168     * In general, GCC 2.95 is more rigorous about rejecting invalid C++
20169       code or deprecated C++ constructs than G++ 2.7, G++ 2.8, EGCS 1.0,
20170       or EGCS 1.1. As a result it may be necessary to fix C++ code before
20171       it will compile with GCC 2.95.
20172     * G++ is also converting toward the ISO C++ standard; as a result
20173       code which was previously valid (and thus accepted by other
20174       compilers and older versions of g++) may no longer be accepted. The
20175       flag -fpermissive may allow some non-conforming code to compile
20176       with GCC 2.95.
20177     * GCC 2.95 compiled C++ code is not binary compatible with EGCS
20178       1.1.x, EGCS 1.0.x or GCC 2.8.x.
20179     * GCC 2.95 does not have changes from the GCC 2.8 tree that were made
20180       between Sept 30, 1998 and April 30, 1999 (the official end of the
20181       GCC 2.8 project). Future GCC releases will include all the changes
20182       from the defunct GCC 2.8 sources.
20183
20184
20185    For questions related to the use of GCC, please consult these web
20186    pages and the [2]GCC manuals. If that fails, the
20187    [3]gcc-help@gcc.gnu.org mailing list might help. Comments on these
20188    web pages and the development of GCC are welcome on our developer
20189    list at [4]gcc@gcc.gnu.org. All of [5]our lists have public archives.
20190
20191   Copyright (C) [6]Free Software Foundation, Inc. Verbatim copying and
20192   distribution of this entire article is permitted in any medium,
20193   provided this notice is preserved.
20194
20195   These pages are [7]maintained by the GCC team. Last modified
20196   2019-11-28[8].
20197
20198References
20199
20200   1. ftp://gcc.gnu.org/pub/gcc/infrastructure/libg++-2.8.1.3.tar.gz
20201   2. https://gcc.gnu.org/onlinedocs/
20202   3. mailto:gcc-help@gcc.gnu.org
20203   4. mailto:gcc@gcc.gnu.org
20204   5. https://gcc.gnu.org/lists.html
20205   6. https://www.fsf.org/
20206   7. https://gcc.gnu.org/about.html
20207   8. http://validator.w3.org/check/referer
20208======================================================================
20209http://gcc.gnu.org/egcs-1.1/index.html
20210
20211                                    EGCS 1.1
20212
20213   September 3, 1998: We are pleased to announce the release of EGCS 1.1.
20214   December 1, 1998: We are pleased to announce the release of EGCS 1.1.1.
20215   March 15, 1999: We are pleased to announce the release of EGCS 1.1.2.
20216
20217   EGCS is a free software project to further the development of the GNU
20218   compilers using an open development environment.
20219
20220   EGCS 1.1 is a major new release of the EGCS compiler system. It has
20221   been [1]extensively tested and is believed to be stable and suitable
20222   for widespread use.
20223
20224   EGCS 1.1 is based on an June 6, 1998 snapshot of the GCC 2.8
20225   development sources; it contains all of the new features found in GCC
20226   2.8.1 as well as all new development from GCC up to June 6, 1998.
20227
20228   EGCS 1.1 also contains many improvements and features not found in GCC
20229   or in older versions of EGCS:
20230     * Global common subexpression elimination and global constant/copy
20231       propagation (aka [2]gcse)
20232     * Ongoing improvements to the [3]alias analysis support to allow for
20233       better optimizations throughout the compiler.
20234     * Vastly improved [4]C++ compiler and integrated C++ runtime
20235       libraries.
20236     * Fixes for the /tmp symlink race security problems.
20237     * New targets including mips16, arm-thumb and 64 bit PowerPC.
20238     * Improvements to GNU Fortran (g77) compiler and runtime library made
20239       since g77 version 0.5.23.
20240
20241   See the [5]new features page for a more complete list of new features
20242   found in EGCS 1.1 releases.
20243
20244   EGCS 1.1.1 is a minor update to fix several serious problems in EGCS
20245   1.1:
20246     * General improvements and fixes
20247          + Avoid some stack overflows when compiling large functions.
20248          + Avoid incorrect loop invariant code motions.
20249          + Fix some core dumps on Linux kernel code.
20250          + Bring back the imake -Di386 and friends fix from EGCS 1.0.2.
20251          + Fix code generation problem in gcse.
20252          + Various documentation related fixes.
20253     * g++/libstdc++ improvements and fixes
20254          + MT safe EH fix for setjmp/longjmp based exception handling.
20255          + Fix a few bad interactions between optimization and exception
20256            handling.
20257          + Fixes for demangling of template names starting with "__".
20258          + Fix a bug that would fail to run destructors in some cases
20259            with -O2.
20260          + Fix 'new' of classes with virtual bases.
20261          + Fix crash building Qt on the Alpha.
20262          + Fix failure compiling WIFEXITED macro on GNU/Linux.
20263          + Fix some -frepo failures.
20264     * g77 and libf2c improvements and fixes
20265          + Various documentation fixes.
20266          + Avoid compiler crash on RAND intrinsic.
20267          + Fix minor bugs in makefiles exposed by BSD make programs.
20268          + Define _XOPEN_SOURCE for libI77 build to avoid potential
20269            problems on some 64-bit systems.
20270          + Fix problem with implicit endfile on rewind.
20271          + Fix spurious recursive I/O errors.
20272     * platform specific improvements and fixes
20273          + Match all versions of UnixWare7.
20274          + Do not assume x86 SVR4 or UnixWare targets can handle stabs.
20275          + Fix PPC/RS6000 LEGITIMIZE_ADDRESS macro and bug in conversion
20276            from unsigned ints to double precision floats.
20277          + Fix ARM ABI issue with NetBSD.
20278          + Fix a few arm code generation bugs.
20279          + Fixincludes will fix additional broken SCO OpenServer header
20280            files.
20281          + Fix a m68k backend bug which caused invalid offsets in reg+d
20282            addresses.
20283          + Fix problems with 64bit AIX 4.3 support.
20284          + Fix handling of long longs for varargs/stdarg functions on the
20285            ppc.
20286          + Minor fixes to CPP predefines for Windows.
20287          + Fix code generation problems with gpr<->fpr copies for 64bit
20288            ppc.
20289          + Fix a few coldfire code generation bugs.
20290          + Fix some more header file problems on SunOS 4.x.
20291          + Fix assert.h handling for RTEMS.
20292          + Fix Windows handling of TREE_SYMBOL_REFERENCED.
20293          + Fix x86 compiler abort in reg-stack pass.
20294          + Fix cygwin/windows problem with section attributes.
20295          + Fix Alpha code generation problem exposed by SMP Linux
20296            kernels.
20297          + Fix typo in m68k 32->64bit integer conversion.
20298          + Make sure target libraries build with -fPIC for PPC & Alpha
20299            targets.
20300
20301   EGCS 1.1.2 is a minor update to fix several serious problems in EGCS
20302   1.1.1:
20303     * General improvements and fixes
20304          + Fix bug in loop optimizer which caused the SPARC (and
20305            potentially other) ports to segfault.
20306          + Fix infinite recursion in alias analysis and combiner code.
20307          + Fix bug in regclass preferencing.
20308          + Fix incorrect loop reversal which caused incorrect code to be
20309            generated for several targets.
20310          + Fix return value for builtin memcpy.
20311          + Reduce compile time for certain loops which exposed quadratic
20312            behavior in the loop optimizer.
20313          + Fix bug which caused volatile memory to be written multiple
20314            times when only one write was needed/desired.
20315          + Fix compiler abort in caller-save.c
20316          + Fix combiner bug which caused incorrect code generation for
20317            certain division by constant operations.
20318          + Fix incorrect code generation due to a bug in range check
20319            optimizations.
20320          + Fix incorrect code generation due to mis-handling of clobbered
20321            values in CSE.
20322          + Fix compiler abort/segfault due to incorrect register
20323            splitting when unrolling loops.
20324          + Fix code generation involving autoincremented addresses with
20325            ternary operators.
20326          + Work around bug in the scheduler which caused qt to be
20327            mis-compiled on some platforms.
20328          + Fix code generation problems with -fshort-enums.
20329          + Tighten security for temporary files.
20330          + Improve compile time for codes which make heavy use of
20331            overloaded functions.
20332          + Fix multiply defined constructor/destructor symbol problems.
20333          + Avoid setting bogus RPATH environment variable during
20334            bootstrap.
20335          + Avoid GNU-make dependencies in the texinfo subdir.
20336          + Install CPP wrapper script in $(prefix)/bin if --enable-cpp.
20337            --enable-cpp=<dirname> can be used to specify an additional
20338            install directory for the cpp wrapper script.
20339          + Fix CSE bug which caused incorrect label-label refs to appear
20340            on some platforms.
20341          + Avoid linking in EH routines from libgcc if they are not
20342            needed.
20343          + Avoid obscure bug in aliasing code.
20344          + Fix bug in weak symbol handling.
20345     * Platform-specific improvements and fixes
20346          + Fix detection of PPro/PII on Unixware 7.
20347          + Fix compiler segfault when building spec99 and other programs
20348            for SPARC targets.
20349          + Fix code-generation bugs for integer and floating point
20350            conditional move instructions on the PPro/PII.
20351          + Use fixincludes to fix byteorder problems on i?86-*-sysv.
20352          + Fix build failure for the arc port.
20353          + Fix floating point format configuration for i?86-gnu port.
20354          + Fix problems with hppa1.0-hp-hpux10.20 configuration when
20355            threads are enabled.
20356          + Fix coldfire code generation bugs.
20357          + Fix "unrecognized insn" problems for Alpha and PPC ports.
20358          + Fix h8/300 code generation problem with floating point values
20359            in memory.
20360          + Fix unrecognized insn problems for the m68k port.
20361          + Fix namespace-pollution problem for the x86 port.
20362          + Fix problems with old assembler on x86 NeXT systems.
20363          + Fix PIC code-generation problems for the SPARC port.
20364          + Fix minor bug with LONG_CALLS in PowerPC SVR4 support.
20365          + Fix minor ISO namespace violation in Alpha varargs/stdarg
20366            support.
20367          + Fix incorrect "braf" instruction usage for the SH port.
20368          + Fix minor bug in va-sh which prevented its use with -ansi.
20369          + Fix problems recognizing and supporting FreeBSD.
20370          + Handle OpenBSD systems correctly.
20371          + Minor fixincludes fix for Digital UNIX 4.0B.
20372          + Fix problems with ctors/dtors in SCO shared libraries.
20373          + Abort instead of generating incorrect code for PPro/PII
20374            floating point conditional moves.
20375          + Avoid multiply defined symbols on GNU/Linux systems using
20376            libc-5.4.xx.
20377          + Fix abort in alpha compiler.
20378     * Fortran-specific fixes
20379          + Fix the IDate intrinsic (VXT) (in libg2c) so the returned year
20380            is in the documented, non-Y2K-compliant range of 0-99, instead
20381            of being returned as 100 in the year 2000.
20382          + Fix the `Date_and_Time' intrinsic (in libg2c) to return the
20383            milliseconds value properly in Values(8).
20384          + Fix the `LStat' intrinsic (in libg2c) to return device-ID
20385            information properly in SArray(7).
20386
20387   Each release includes installation instructions in both HTML and
20388   plaintext forms (see the INSTALL directory in the toplevel directory of
20389   the distribution). However, we also keep the most up to date
20390   installation instructions and [6]build/test status on our web page. We
20391   will update those pages as new information becomes available.
20392
20393   The EGCS project would like to thank the numerous people that have
20394   contributed new features, test results, bugfixes, etc. This [7]amazing
20395   group of volunteers is what makes EGCS successful.
20396
20397   And finally, we can't in good conscience fail to mention some
20398   [8]caveats to using EGCS 1.1.
20399
20400   Download EGCS from egcs.cygnus.com (USA California).
20401
20402   The EGCS 1.1 release is also available on many mirror sites.
20403   [9]Goto mirror list to find a closer site.
20404
20405
20406    For questions related to the use of GCC, please consult these web
20407    pages and the [10]GCC manuals. If that fails, the
20408    [11]gcc-help@gcc.gnu.org mailing list might help. Comments on these
20409    web pages and the development of GCC are welcome on our developer
20410    list at [12]gcc@gcc.gnu.org. All of [13]our lists have public
20411    archives.
20412
20413   Copyright (C) [14]Free Software Foundation, Inc. Verbatim copying and
20414   distribution of this entire article is permitted in any medium,
20415   provided this notice is preserved.
20416
20417   These pages are [15]maintained by the GCC team. Last modified
20418   2019-11-28[16].
20419
20420References
20421
20422   1. http://gcc.gnu.org/egcs-1.1/egcs-1.1-test.html
20423   2. http://gcc.gnu.org/news/gcse.html
20424   3. http://gcc.gnu.org/news/alias.html
20425   4. http://gcc.gnu.org/egcs-1.1/c++features.html
20426   5. http://gcc.gnu.org/egcs-1.1/features.html
20427   6. http://gcc.gnu.org/egcs-1.1/buildstat.html
20428   7. https://gcc.gnu.org/onlinedocs/gcc/Contributors.html
20429   8. http://gcc.gnu.org/egcs-1.1/caveats.html
20430   9. http://gcc.gnu.org/mirrors.html
20431  10. https://gcc.gnu.org/onlinedocs/
20432  11. mailto:gcc-help@gcc.gnu.org
20433  12. mailto:gcc@gcc.gnu.org
20434  13. https://gcc.gnu.org/lists.html
20435  14. https://www.fsf.org/
20436  15. https://gcc.gnu.org/about.html
20437  16. http://validator.w3.org/check/referer
20438======================================================================
20439http://gcc.gnu.org/egcs-1.1/features.html
20440
20441                             EGCS 1.1 new features
20442
20443     * Integrated GNU Fortran (g77) compiler and runtime library with
20444       improvements, based on g77 version 0.5.23.
20445     * Vast improvements in the C++ compiler; so many they have [1]page of
20446       their own!
20447     * Compiler implements [2]global common subexpression elimination and
20448       global copy/constant propagation.
20449     * More major improvements in the [3]alias analysis code.
20450     * More major improvements in the exception handling code to improve
20451       performance, lower static overhead and provide the infrastructure
20452       for future improvements.
20453     * The infamous /tmp symlink race security problems have been fixed.
20454     * The regmove optimization pass has been nearly completely rewritten
20455       to improve performance of generated code.
20456     * The compiler now recomputes register usage information before local
20457       register allocation. By providing more accurate information to the
20458       priority based allocator, we get better register allocation.
20459     * The register reloading phase of the compiler optimizes spill code
20460       much better than in previous releases.
20461     * Some bad interactions between the register allocator and
20462       instruction scheduler have been fixed, resulting in much better
20463       code for certain programs. Additionally, we have tuned the
20464       scheduler in various ways to improve performance of generated code
20465       for some architectures.
20466     * The compiler's branch shortening algorithms have been significantly
20467       improved to work better on targets which align jump targets.
20468     * The compiler now supports -Os to prefer optimizing for code space
20469       over optimizing for code speed.
20470     * The compiler will now totally eliminate library calls which compute
20471       constant values. This primarily helps targets with no integer
20472       div/mul support and targets without floating point support.
20473     * The compiler now supports an extensive "--help" option.
20474     * cpplib has been greatly improved and may be suitable for limited
20475       use.
20476     * Memory footprint for the compiler has been significantly reduced
20477       for some pathological cases.
20478     * The time to build EGCS has been improved for certain targets
20479       (particularly the alpha and mips platforms).
20480     * Many infrastructure improvements throughout the compiler, plus the
20481       usual mountain of bugfixes and minor improvements.
20482     * Target dependent improvements:
20483          + SPARC port now includes V8 plus and V9 support as well as
20484            performance tuning for Ultra class machines. The SPARC port
20485            now uses the Haifa scheduler.
20486          + Alpha port has been tuned for the EV6 processor and has an
20487            optimized expansion of memcpy/bzero. The Alpha port now uses
20488            the Haifa scheduler.
20489          + RS6000/PowerPC: support for the Power64 architecture and AIX
20490            4.3. The RS6000/PowerPC port now uses the Haifa scheduler.
20491          + x86: Alignment of static store data and jump targets is per
20492            Intel recommendations now. Various improvements throughout the
20493            x86 port to improve performance on Pentium processors
20494            (including improved epilogue sequences for Pentium chips and
20495            backend improvements which should help register allocation on
20496            all x86 variants. Conditional move support has been fixed and
20497            enabled for PPro processors. The x86 port also better supports
20498            64bit operations now. Unixware 7, a System V Release 5 target,
20499            is now supported and SCO OpenServer targets can support GAS.
20500          + MIPS has improved multiply/multiply-add support and now
20501            includes mips16 ISA support.
20502          + M68k has many micro-optimizations and Coldfire fixes.
20503     * Core compiler is based on the GCC development tree from June 9,
20504       1998, so we have all of the [4]features found in GCC 2.8.
20505
20506
20507    For questions related to the use of GCC, please consult these web
20508    pages and the [5]GCC manuals. If that fails, the
20509    [6]gcc-help@gcc.gnu.org mailing list might help. Comments on these
20510    web pages and the development of GCC are welcome on our developer
20511    list at [7]gcc@gcc.gnu.org. All of [8]our lists have public archives.
20512
20513   Copyright (C) [9]Free Software Foundation, Inc. Verbatim copying and
20514   distribution of this entire article is permitted in any medium,
20515   provided this notice is preserved.
20516
20517   These pages are [10]maintained by the GCC team. Last modified
20518   2019-11-28[11].
20519
20520References
20521
20522   1. http://gcc.gnu.org/egcs-1.1/c++features.html
20523   2. http://gcc.gnu.org/news/gcse.html
20524   3. http://gcc.gnu.org/news/alias.html
20525   4. http://gcc.gnu.org/egcs-1.0/features-2.8.html
20526   5. https://gcc.gnu.org/onlinedocs/
20527   6. mailto:gcc-help@gcc.gnu.org
20528   7. mailto:gcc@gcc.gnu.org
20529   8. https://gcc.gnu.org/lists.html
20530   9. https://www.fsf.org/
20531  10. https://gcc.gnu.org/about.html
20532  11. http://validator.w3.org/check/referer
20533======================================================================
20534http://gcc.gnu.org/egcs-1.1/caveats.html
20535
20536                                EGCS 1.1 Caveats
20537
20538     * EGCS has an integrated libstdc++, but does not have an integrated
20539       libg++. Furthermore old libg++ releases will not work with EGCS; HJ
20540       Lu has made a libg++-2.8.1.2 snapshot available which may work with
20541       EGCS.
20542       Note most C++ programs only need libstdc++.
20543     * Exception handling may not work with shared libraries, particularly
20544       on alphas, hppas, rs6000/powerpc and mips based platforms.
20545       Exception handling is known to work on x86-linux platforms with
20546       shared libraries.
20547     * Some versions of the Linux kernel have bugs which prevent them from
20548       being compiled or from running when compiled by EGCS. See the FAQ
20549       (as shipped with EGCS 1.1) for additional information.
20550     * In general, EGCS is more rigorous about rejecting invalid C++ code
20551       or deprecated C++ constructs than g++-2.7, g++-2.8 or EGCS 1.0. As
20552       a result it may be necessary to fix C++ code before it will compile
20553       with EGCS.
20554     * G++ is also converting toward the ISO C++ standard; as a result
20555       code which was previously valid (and thus accepted by other
20556       compilers and older versions of g++) may no longer be accepted.
20557     * EGCS 1.1 compiled C++ code is not binary compatible with EGCS 1.0.x
20558       or GCC 2.8.x due to changes necessary to support thread safe
20559       exception handling.
20560
20561
20562    For questions related to the use of GCC, please consult these web
20563    pages and the [1]GCC manuals. If that fails, the
20564    [2]gcc-help@gcc.gnu.org mailing list might help. Comments on these
20565    web pages and the development of GCC are welcome on our developer
20566    list at [3]gcc@gcc.gnu.org. All of [4]our lists have public archives.
20567
20568   Copyright (C) [5]Free Software Foundation, Inc. Verbatim copying and
20569   distribution of this entire article is permitted in any medium,
20570   provided this notice is preserved.
20571
20572   These pages are [6]maintained by the GCC team. Last modified
20573   2019-11-28[7].
20574
20575References
20576
20577   1. https://gcc.gnu.org/onlinedocs/
20578   2. mailto:gcc-help@gcc.gnu.org
20579   3. mailto:gcc@gcc.gnu.org
20580   4. https://gcc.gnu.org/lists.html
20581   5. https://www.fsf.org/
20582   6. https://gcc.gnu.org/about.html
20583   7. http://validator.w3.org/check/referer
20584======================================================================
20585http://gcc.gnu.org/egcs-1.0/index.html
20586
20587                                    EGCS 1.0
20588
20589   December 3, 1997: We are pleased to announce the release of EGCS 1.0.
20590   January 6, 1998: We are pleased to announce the release of EGCS 1.0.1.
20591   March 16, 1998: We are pleased to announce the release of EGCS 1.0.2.
20592   May 15, 1998 We are pleased to announce the release of EGCS 1.0.3.
20593
20594   EGCS is a collaborative effort involving several groups of hackers
20595   using an open development model to accelerate development and testing
20596   of GNU compilers and runtime libraries.
20597
20598   An important goal of EGCS is to allow wide scale testing of
20599   experimental features and optimizations; therefore, EGCS contains some
20600   features and optimizations which are still under development. However,
20601   EGCS has been carefully tested and should be comparable in quality to
20602   most GCC releases.
20603
20604   EGCS 1.0 is based on an August 2, 1997 snapshot of the GCC 2.8
20605   development sources; it contains nearly all of the new features found
20606   in GCC 2.8.
20607
20608   EGCS 1.0 also contains many improvements and features not found in GCC
20609   2.7 and even the GCC 2.8 series (which was released after the original
20610   EGCS 1.0 release).
20611     * Integrated C++ runtime libraries, including support for most major
20612       GNU/Linux systems!
20613     * The integrated libstdc++ library includes a verbatim copy of SGI's
20614       STL release.
20615     * Integrated GNU Fortran compiler.
20616     * New instruction scheduler.
20617     * New alias analysis code.
20618
20619   See the [1]new features page for a more complete list of new features.
20620
20621   EGCS 1.0.1 is a minor update to the EGCS 1.0 compiler to fix a few
20622   critical bugs and add support for Red Hat 5.0 Linux. Changes since the
20623   EGCS 1.0 release:
20624     * Add support for Red Hat 5.0 Linux and better support for Linux
20625       systems using glibc2.
20626       Many programs failed to link when compiled with EGCS 1.0 on Red Hat
20627       5.0 or on systems with newer versions of glibc2. EGCS 1.0.1 should
20628       fix these problems.
20629     * Compatibility with both EGCS 1.0 and GCC 2.8 libgcc exception
20630       handling interfaces.
20631       To avoid future compatibility problems, we strongly urge anyone who
20632       is planning on distributing shared libraries that contain C++ code
20633       to upgrade to EGCS 1.0.1 first.
20634       Soon after EGCS 1.0 was released, the GCC developers made some
20635       incompatible changes in libgcc's exception handling interfaces.
20636       These changes were needed to solve problems on some platforms. This
20637       means that GCC 2.8.0, when released, will not be seamlessly
20638       compatible with shared libraries built by EGCS 1.0. The reason is
20639       that the libgcc.a in GCC 2.8.0 will not contain a function needed
20640       by the old interface.
20641       The result of this is that there may be compatibility problems with
20642       shared libraries built by EGCS 1.0 when used with GCC 2.8.0.
20643       With EGCS 1.0.1, generated code uses the new (GCC 2.8.0) interface,
20644       and libgcc.a has the support routines for both the old and the new
20645       interfaces (so EGCS 1.0.1 and EGCS 1.0 code can be freely mixed,
20646       and EGCS 1.0.1 and GCC 2.8.0 code can be freely mixed).
20647       The maintainers of GCC 2.x have decided against including seamless
20648       support for the old interface in 2.8.0, since it was never
20649       "official", so to avoid future compatibility problems we recommend
20650       against distributing any shared libraries built by EGCS 1.0 that
20651       contain C++ code (upgrade to 1.0.1 and use that).
20652     * Various bugfixes in the x86, hppa, mips, and rs6000/ppc back ends.
20653       The x86 changes fix code generation errors exposed when building
20654       glibc2 and the usual GNU/Linux dynamic linker (ld.so).
20655       The hppa change fixes a compiler abort when configured for use with
20656       RTEMS.
20657       The MIPS changes fix problems with the definition of LONG_MAX on
20658       newer systems, allow for command line selection of the target ABI,
20659       and fix one code generation problem.
20660       The rs6000/ppc change fixes some problems with passing structures
20661       to varargs/stdarg functions.
20662     * A few machine independent bugfixes, mostly to fix code generation
20663       errors when building Linux kernels or glibc.
20664     * Fix a few critical exception handling and template bugs in the C++
20665       compiler.
20666     * Fix Fortran namelist bug on alphas.
20667     * Fix build problems on x86-solaris systems.
20668
20669   EGCS 1.0.2 is a minor update to the EGCS 1.0.1 compiler to fix several
20670   serious problems in EGCS 1.0.1.
20671     * General improvements and fixes
20672          + Memory consumption significantly reduced, especially for
20673            templates and inline functions.
20674          + Fix various problems with glibc2.1.
20675          + Fix loop optimization bug exposed by rs6000/ppc port.
20676          + Fix to avoid potential code generation problems in jump.c.
20677          + Fix some undefined symbol problems in dwarf1 debug support.
20678     * g++/libstdc++ improvements and fixes
20679          + libstdc++ in the EGCS release has been updated and should be
20680            link compatible with libstdc++-2.8.
20681          + Various fixes in libio/libstdc++ to work better on GNU/Linux
20682            systems.
20683          + Fix problems with duplicate symbols on systems that do not
20684            support weak symbols.
20685          + Memory corruption bug and undefined symbols in bastring have
20686            been fixed.
20687          + Various exception handling fixes.
20688          + Fix compiler abort for very long thunk names.
20689     * g77 improvements and fixes
20690          + Fix compiler crash for omitted bound in Fortran CASE
20691            statement.
20692          + Add missing entries to g77 lang-options.
20693          + Fix problem with -fpedantic in the g77 compiler.
20694          + Fix "backspace" problem with g77 on alphas.
20695          + Fix x86 backend problem with Fortran literals and -fpic.
20696          + Fix some of the problems with negative subscripts for g77 on
20697            alphas.
20698          + Fixes for Fortran builds on cygwin32/mingw32.
20699     * platform specific improvements and fixes
20700          + Fix long double problems on x86 (exposed by glibc).
20701          + x86 ports define i386 again to keep imake happy.
20702          + Fix exception handling support on NetBSD ports.
20703          + Several changes to collect2 to fix many problems with AIX.
20704          + Define __ELF__ for GNU/Linux on rs6000.
20705          + Fix -mcall-linux problem on GNU/Linux on rs6000.
20706          + Fix stdarg/vararg problem for GNU/Linux on rs6000.
20707          + Allow autoconf to select a proper install problem on AIX 3.1.
20708          + m68k port support includes -mcpu32 option as well as cpu32
20709            multilibs.
20710          + Fix stdarg bug for irix6.
20711          + Allow EGCS to build on irix5 without the gnu assembler.
20712          + Fix problem with static linking on sco5.
20713          + Fix bootstrap on sco5 with native compiler.
20714          + Fix for abort building newlib on H8 target.
20715          + Fix fixincludes handling of math.h on SunOS.
20716          + Minor fix for Motorola 3300 m68k systems.
20717
20718   EGCS 1.0.3 is a minor update to the EGCS 1.0.2 compiler to fix a few
20719   problems reported by Red Hat for builds of Red Hat 5.1.
20720     * Generic bugfixes:
20721          + Fix a typo in the libio library which resulted in incorrect
20722            behavior of istream::get.
20723          + Fix the Fortran negative array index problem.
20724          + Fix a major problem with the ObjC runtime thread support
20725            exposed by glibc2.
20726          + Reduce memory consumption of the Haifa scheduler.
20727     * Target specific bugfixes:
20728          + Fix one x86 floating point code generation bug exposed by
20729            glibc2 builds.
20730          + Fix one x86 internal compiler error exposed by glibc2 builds.
20731          + Fix profiling bugs on the Alpha.
20732          + Fix ImageMagick & emacs 20.2 build problems on the Alpha.
20733          + Fix rs6000/ppc bug when converting values from integer types
20734            to floating point types.
20735
20736   The EGCS 1.0 releases include installation instructions in both HTML
20737   and plaintext forms (see the INSTALL directory in the toplevel
20738   directory of the distribution). However, we also keep the most up to
20739   date installation instructions and [2]build/test status on our web
20740   page. We will update those pages as new information becomes available.
20741
20742   And, we can't in good conscience fail to mention some [3]caveats to
20743   using EGCS.
20744
20745   Update: Big thanks to Stanford for providing a high speed link for
20746   downloading EGCS (go.cygnus.com)!
20747
20748   Download EGCS from ftp.cygnus.com (USA California) or go.cygnus.com
20749   (USA California -- High speed link provided by Stanford).
20750
20751   The EGCS 1.0 release is also available many mirror sites.
20752   [4]Goto mirror list to find a closer site
20753
20754   We'd like to thank the numerous people that have contributed new
20755   features, test results, bugfixes, etc. Unfortunately, they're far too
20756   numerous to mention by name.
20757
20758
20759    For questions related to the use of GCC, please consult these web
20760    pages and the [5]GCC manuals. If that fails, the
20761    [6]gcc-help@gcc.gnu.org mailing list might help. Comments on these
20762    web pages and the development of GCC are welcome on our developer
20763    list at [7]gcc@gcc.gnu.org. All of [8]our lists have public archives.
20764
20765   Copyright (C) [9]Free Software Foundation, Inc. Verbatim copying and
20766   distribution of this entire article is permitted in any medium,
20767   provided this notice is preserved.
20768
20769   These pages are [10]maintained by the GCC team. Last modified
20770   2019-11-28[11].
20771
20772References
20773
20774   1. http://gcc.gnu.org/egcs-1.0/features.html
20775   2. http://gcc.gnu.org/egcs-1.0/buildstat.html
20776   3. http://gcc.gnu.org/egcs-1.0/caveats.html
20777   4. http://gcc.gnu.org/mirrors.html
20778   5. https://gcc.gnu.org/onlinedocs/
20779   6. mailto:gcc-help@gcc.gnu.org
20780   7. mailto:gcc@gcc.gnu.org
20781   8. https://gcc.gnu.org/lists.html
20782   9. https://www.fsf.org/
20783  10. https://gcc.gnu.org/about.html
20784  11. http://validator.w3.org/check/referer
20785======================================================================
20786http://gcc.gnu.org/egcs-1.0/features.html
20787
20788                               EGCS 1.0 features
20789
20790     * Core compiler is based on the gcc2 development tree from Aug 2,
20791       1997, so we have most of the [1]features found in GCC 2.8.
20792     * Integrated GNU Fortran compiler based on g77-0.5.22-19970929.
20793     * Vast improvements in the C++ compiler; so many they have [2]page of
20794       their own!
20795     * Integrated C++ runtime libraries, including support for most major
20796       GNU/Linux systems!
20797     * New instruction scheduler from IBM Haifa which includes support for
20798       function wide instruction scheduling as well as superscalar
20799       scheduling.
20800     * Significantly improved alias analysis code.
20801     * Improved register allocation for two address machines.
20802     * Significant code generation improvements for Fortran code on
20803       Alphas.
20804     * Various optimizations from the g77 project as well as improved loop
20805       optimizations.
20806     * Dwarf2 debug format support for some targets.
20807     * egcs libstdc++ includes the SGI STL implementation without changes.
20808     * As a result of these and other changes, egcs libstc++ is not binary
20809       compatible with previous releases of libstdc++.
20810     * Various new ports -- UltraSPARC, Irix6.2 & Irix6.3 support, The SCO
20811       Openserver 5 family (5.0.{0,2,4} and Internet FastStart 1.0 and
20812       1.1), Support for RTEMS on several embedded targets, Support for
20813       arm-linux, Mitsubishi M32R, Hitachi H8/S, Matsushita MN102 and
20814       MN103, NEC V850, Sparclet, Solaris & GNU/Linux on PowerPCs, etc.
20815     * Integrated testsuites for gcc, g++, g77, libstdc++ and libio.
20816     * RS6000/PowerPC ports generate code which can run on all
20817       RS6000/PowerPC variants by default.
20818     * -mcpu= and -march= switches for the x86 port to allow better
20819       control over how the x86 port generates code.
20820     * Includes the template repository patch (aka repo patch); note the
20821       new template code makes repo obsolete for ELF systems using gnu-ld
20822       such as GNU/Linux.
20823     * Plus the usual assortment of bugfixes and improvements.
20824
20825
20826    For questions related to the use of GCC, please consult these web
20827    pages and the [3]GCC manuals. If that fails, the
20828    [4]gcc-help@gcc.gnu.org mailing list might help. Comments on these
20829    web pages and the development of GCC are welcome on our developer
20830    list at [5]gcc@gcc.gnu.org. All of [6]our lists have public archives.
20831
20832   Copyright (C) [7]Free Software Foundation, Inc. Verbatim copying and
20833   distribution of this entire article is permitted in any medium,
20834   provided this notice is preserved.
20835
20836   These pages are [8]maintained by the GCC team. Last modified
20837   2019-11-28[9].
20838
20839References
20840
20841   1. http://gcc.gnu.org/egcs-1.0/features-2.8.html
20842   2. http://gcc.gnu.org/egcs-1.0/c++features.html
20843   3. https://gcc.gnu.org/onlinedocs/
20844   4. mailto:gcc-help@gcc.gnu.org
20845   5. mailto:gcc@gcc.gnu.org
20846   6. https://gcc.gnu.org/lists.html
20847   7. https://www.fsf.org/
20848   8. https://gcc.gnu.org/about.html
20849   9. http://validator.w3.org/check/referer
20850======================================================================
20851http://gcc.gnu.org/egcs-1.0/caveats.html
20852
20853                                EGCS 1.0 Caveats
20854
20855     * EGCS has an integrated libstdc++, but does not have an integrated
20856       libg++. Furthermore old libg++ releases will not work with egc; HJ
20857       Lu has made a libg++-2.8.1.2 available which may work with EGCS.
20858       Note most C++ programs only need libstdc++.
20859     * Note that using -pedantic or -Wreturn-type can cause an explosion
20860       in the amount of memory needed for template-heavy C++ code, such as
20861       code that uses STL. Also note that -Wall includes -Wreturn-type, so
20862       if you use -Wall you will need to specify -Wno-return-type to turn
20863       it off.
20864     * Exception handling may not work with shared libraries, particularly
20865       on alphas, hppas, and mips based platforms. Exception handling is
20866       known to work on x86-linux platforms with shared libraries.
20867     * Some versions of the Linux kernel have bugs which prevent them from
20868       being compiled or from running when compiled by EGCS. See the FAQ
20869       (as shipped with EGCS 1.0) for additional information.
20870     * In general, EGCS is more rigorous about rejecting invalid C++ code
20871       or deprecated C++ constructs than G++ 2.7. As a result it may be
20872       necessary to fix C++ code before it will compile with EGCS.
20873     * G++ is also aggressively tracking the C++ standard; as a result
20874       code which was previously valid (and thus accepted by other
20875       compilers and older versions of G++) may no longer be accepted.
20876     * EGCS 1.0 may not work with Red Hat Linux 5.0 on all targets. EGCS
20877       1.0.x and later releases should work with Red Hat Linux 5.0.
20878
20879
20880    For questions related to the use of GCC, please consult these web
20881    pages and the [1]GCC manuals. If that fails, the
20882    [2]gcc-help@gcc.gnu.org mailing list might help. Comments on these
20883    web pages and the development of GCC are welcome on our developer
20884    list at [3]gcc@gcc.gnu.org. All of [4]our lists have public archives.
20885
20886   Copyright (C) [5]Free Software Foundation, Inc. Verbatim copying and
20887   distribution of this entire article is permitted in any medium,
20888   provided this notice is preserved.
20889
20890   These pages are [6]maintained by the GCC team. Last modified
20891   2019-11-28[7].
20892
20893References
20894
20895   1. https://gcc.gnu.org/onlinedocs/
20896   2. mailto:gcc-help@gcc.gnu.org
20897   3. mailto:gcc@gcc.gnu.org
20898   4. https://gcc.gnu.org/lists.html
20899   5. https://www.fsf.org/
20900   6. https://gcc.gnu.org/about.html
20901   7. http://validator.w3.org/check/referer
20902======================================================================
20903