utah rcsid 1.4 86/08/03 01:03:22: Add 'configuration' flags which are
really 'assembler style' flags, clean up makefile. Add macros to replace
some trivial functions. Change FIX macros to punt if e

utah rcsid 1.4 86/08/03 01:03:22: Add 'configuration' flags which are
really 'assembler style' flags, clean up makefile. Add macros to replace
some trivial functions. Change FIX macros to punt if errors have been
seen.

show more ...

utah rcsid 1.33 87/11/28 02:59:18: Be cooperative with SCCON and SSCON by
masking and sign-extending ASG OPSIMP constant operands. Added a few
comments in the tortuous AND code too.

utah rcsid 1.32 87/09/24 22:24:49: Missing case for FORARG in sconv() --
floating to short integral conversions.

utah rcsid 1.31 87/09/08 19:12:14: Fix bug where stack push of 0.0 only
pushes one word of zeroes... Ack.

utah rcsid 1.30 87/09/07 20:41:00: Another minor battle in the struggle
between makearg() and float arguments. This time we kluge the issue and
avoid szty() (which predicts that float values are onl

utah rcsid 1.30 87/09/07 20:41:00: Another minor battle in the struggle
between makearg() and float arguments. This time we kluge the issue and
avoid szty() (which predicts that float values are only 4 bytes when SPRECC
is used).

show more ...

utah rcsid 1.29 87/08/25 00:05:43: Took care of some bugs in the extension
of sconv() to handle FORARG evaluations. Conversion of UCHAR/USHORT to
floating is now handled in sconv() so that ZG can ta

utah rcsid 1.29 87/08/25 00:05:43: Took care of some bugs in the extension
of sconv() to handle FORARG evaluations. Conversion of UCHAR/USHORT to
floating is now handled in sconv() so that ZG can take advantage of it;
this cures a bug with assignment ops with UCHAR/USHORT lhs and floating rhs
which caused the lhs to be treated as signed. Conversions to FLOAT are now
handled in sconv() too.

show more ...

conversions to insert an INT conversion and thus avoid the expensive and
redundant UNSIGNED to FLOAT/DOUBLE overhead. (7) Delete INT to UNSIGNED
and reverse conversions on function calls. (8) Delet

conversions to insert an INT conversion and thus avoid the expensive and
redundant UNSIGNED to FLOAT/DOUBLE overhead. (7) Delete INT to UNSIGNED
and reverse conversions on function calls. (8) Delete FLOAT to DOUBLE
conversions below logical operators.

show more ...

utah rcsid 1.28 87/08/24 01:55:20: (1) Changed code for ZZ so that we can
generate optimal AND tests against constants. (2) Fixed botch in my
makearg() routine. (3) Generate 'pushl' instead of 'mov

utah rcsid 1.28 87/08/24 01:55:20: (1) Changed code for ZZ so that we can
generate optimal AND tests against constants. (2) Fixed botch in my
makearg() routine. (3) Generate 'pushl' instead of 'movl' when working on
FORARG in sconv(). (4) Permitted NAME nodes as possible indexing bases,
using pc relative deferred addressing. (5) Produce positive bicl masks if
possible, so we can use literals. (6) Modify UCHAR/USHORT to FLOAT/DOUBLE

show more ...

utah rcsid 1.27 87/07/17 17:09:50: Cosmetic change to tlen() -- use
#define'd constants. Also fix use of local machine 'sizeof (int)' in
zzzcode to #define'd target machine constants.

shifts. (6) Add ZV entry in zzzcode() to handle FORARG goals for
conversions and leaf nodes. (7) Fix minor bug with ZB entry and unsigned
char/short.

utah rcsid 1.26 87/07/17 15:31:14: Some changes to reduce differences with
the tahoe compiler. (1) #include <ctype.h> not "ctype.h". (2) Simplify
eobl2(), hopcode() and hoptab[], flshape(), acon(),

utah rcsid 1.26 87/07/17 15:31:14: Some changes to reduce differences with
the tahoe compiler. (1) #include <ctype.h> not "ctype.h". (2) Simplify
eobl2(), hopcode() and hoptab[], flshape(), acon(), and cbgen() and
ccbranches[]. (3) Farm out conversion and structure assignment code in
zzzcode() to sconv() and stasg() functions. (4) Delete lots of
commented-out old code. (5) Fix bug in optim2() code for degenerate left

show more ...

utah rcsid 1.25 87/07/07 20:30:32: Permit the ZE code for INCR/DECR in
zzzcode() to take a NAME as its second operand: this permits FLOAT and
DOUBLE operands to be handled, since floating constants a

utah rcsid 1.25 87/07/07 20:30:32: Permit the ZE code for INCR/DECR in
zzzcode() to take a NAME as its second operand: this permits FLOAT and
DOUBLE operands to be handled, since floating constants are represented by
NAMEs.

show more ...

utah rcsid 1.24 87/07/07 19:37:31: Prevent core dumps in shumul() when
using -x debugging...

utah rcsid 1.23 87/05/31 20:46:39: More work on degenerate operations in
optim2() and degenerate().

utah rcsid 1.22 87/05/15 17:03:03: First pass at handling degenerate
operations in optim2(). The idea is mainly that we can avoid some table
complications by assuming that no degenerate operations d

utah rcsid 1.22 87/05/15 17:03:03: First pass at handling degenerate
operations in optim2(). The idea is mainly that we can avoid some table
complications by assuming that no degenerate operations dribble down there.
Since I did some degenerate operations, I decided to do a few others so
that some operations aren't outstandingly slow.

show more ...

on a certain amount of tree canonicalization there.

utah rcsid 1.21 87/04/24 19:05:37: Optimize comparisons with constants
which are out of range of short integral types. The purpose of this was
mainly to handle a problem in the code table -- it's ha

utah rcsid 1.21 87/04/24 19:05:37: Optimize comparisons with constants
which are out of range of short integral types. The purpose of this was
mainly to handle a problem in the code table -- it's hard to generate the
best code for short comparisons unless you know that the ridiculous cases
have been taken care of. The optimization could have gone in either
clocal() or optim2(); the latter seemed more reasonable since I could count

show more ...

utah rcsid 1.20 87/04/19 23:47:13: Add support for handling unsigned div
and rem inline when the divisor is a constant: (1) ZJ template escape for
stuff that's too complex for the table; (2) a real v

utah rcsid 1.20 87/04/19 23:47:13: Add support for handling unsigned div
and rem inline when the divisor is a constant: (1) ZJ template escape for
stuff that's too complex for the table; (2) a real version of upput() which
lets the table access both words of the 64-bit dividend operand of the ediv
instruction; (3) a hack to hardops() to detect constant divisors -- it
handles the easiest case and passes everything else to the table.

show more ...

utah rcsid 1.19 87/03/18 00:02:55: The earlier rewrite of shtemp() left the
code for STARNM forms commented out for some reason...

utah rcsid 1.18 87/02/08 22:50:12: Replaced shtemp() routine -- shtemp() is
supposed to find leaf nodes which don't use any temporary registers, and
the old version didn't do that well.

utah rcsid 1.17 87/01/16 02:45:01: Permit nonpositive offsets from the
argument pointer for excessively clever program which want to recover funny
stuff on the stack.

utah rcsid 1.16 87/01/15 01:36:58: Changed optim2()'s SCONV optimization
code to avoid screwing up function types. Also some minor lint fixes.

utah rcsid 1.15 86/11/27 20:26:18: update to ucb sid 1.15 -- upput calling
sequence changes (sam). Added a couple /*ARGSUSED*/ lint directives;
probably a lost cause.

utah rcsid 1.14 86/06/18 14:35:25: The idea of converting ASSIGN+SCONV into
ASSIGN with modified types was cute, but unfortunately certain varieties of
ASSIGN are not supported -- in particular doubl

utah rcsid 1.14 86/06/18 14:35:25: The idea of converting ASSIGN+SCONV into
ASSIGN with modified types was cute, but unfortunately certain varieties of
ASSIGN are not supported -- in particular double => bit field. We chicken
out and leave bitfields alone.

show more ...

utah rcsid 1.10 87/11/26 16:26:36: Delete isitfloat() -- make float
constants predictable and ANSI compatible.