1 /* Driver of optimization process 2 Copyright (C) 2003-2020 Free Software Foundation, Inc. 3 Contributed by Jan Hubicka 4 5 This file is part of GCC. 6 7 GCC is free software; you can redistribute it and/or modify it under 8 the terms of the GNU General Public License as published by the Free 9 Software Foundation; either version 3, or (at your option) any later 10 version. 11 12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY 13 WARRANTY; without even the implied warranty of MERCHANTABILITY or 14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 15 for more details. 16 17 You should have received a copy of the GNU General Public License 18 along with GCC; see the file COPYING3. If not see 19 <http://www.gnu.org/licenses/>. */ 20 21 /* This module implements main driver of compilation process. 22 23 The main scope of this file is to act as an interface in between 24 tree based frontends and the backend. 25 26 The front-end is supposed to use following functionality: 27 28 - finalize_function 29 30 This function is called once front-end has parsed whole body of function 31 and it is certain that the function body nor the declaration will change. 32 33 (There is one exception needed for implementing GCC extern inline 34 function.) 35 36 - varpool_finalize_decl 37 38 This function has same behavior as the above but is used for static 39 variables. 40 41 - add_asm_node 42 43 Insert new toplevel ASM statement 44 45 - finalize_compilation_unit 46 47 This function is called once (source level) compilation unit is finalized 48 and it will no longer change. 49 50 The symbol table is constructed starting from the trivially needed 51 symbols finalized by the frontend. Functions are lowered into 52 GIMPLE representation and callgraph/reference lists are constructed. 53 Those are used to discover other necessary functions and variables. 54 55 At the end the bodies of unreachable functions are removed. 56 57 The function can be called multiple times when multiple source level 58 compilation units are combined. 59 60 - compile 61 62 This passes control to the back-end. Optimizations are performed and 63 final assembler is generated. This is done in the following way. Note 64 that with link time optimization the process is split into three 65 stages (compile time, linktime analysis and parallel linktime as 66 indicated bellow). 67 68 Compile time: 69 70 1) Inter-procedural optimization. 71 (ipa_passes) 72 73 This part is further split into: 74 75 a) early optimizations. These are local passes executed in 76 the topological order on the callgraph. 77 78 The purpose of early optimizations is to optimize away simple 79 things that may otherwise confuse IP analysis. Very simple 80 propagation across the callgraph is done i.e. to discover 81 functions without side effects and simple inlining is performed. 82 83 b) early small interprocedural passes. 84 85 Those are interprocedural passes executed only at compilation 86 time. These include, for example, transactional memory lowering, 87 unreachable code removal and other simple transformations. 88 89 c) IP analysis stage. All interprocedural passes do their 90 analysis. 91 92 Interprocedural passes differ from small interprocedural 93 passes by their ability to operate across whole program 94 at linktime. Their analysis stage is performed early to 95 both reduce linking times and linktime memory usage by 96 not having to represent whole program in memory. 97 98 d) LTO streaming. When doing LTO, everything important gets 99 streamed into the object file. 100 101 Compile time and or linktime analysis stage (WPA): 102 103 At linktime units gets streamed back and symbol table is 104 merged. Function bodies are not streamed in and not 105 available. 106 e) IP propagation stage. All IP passes execute their 107 IP propagation. This is done based on the earlier analysis 108 without having function bodies at hand. 109 f) Ltrans streaming. When doing WHOPR LTO, the program 110 is partitioned and streamed into multiple object files. 111 112 Compile time and/or parallel linktime stage (ltrans) 113 114 Each of the object files is streamed back and compiled 115 separately. Now the function bodies becomes available 116 again. 117 118 2) Virtual clone materialization 119 (cgraph_materialize_clone) 120 121 IP passes can produce copies of existing functions (such 122 as versioned clones or inline clones) without actually 123 manipulating their bodies by creating virtual clones in 124 the callgraph. At this time the virtual clones are 125 turned into real functions 126 3) IP transformation 127 128 All IP passes transform function bodies based on earlier 129 decision of the IP propagation. 130 131 4) late small IP passes 132 133 Simple IP passes working within single program partition. 134 135 5) Expansion 136 (expand_all_functions) 137 138 At this stage functions that needs to be output into 139 assembler are identified and compiled in topological order 140 6) Output of variables and aliases 141 Now it is known what variable references was not optimized 142 out and thus all variables are output to the file. 143 144 Note that with -fno-toplevel-reorder passes 5 and 6 145 are combined together in cgraph_output_in_order. 146 147 Finally there are functions to manipulate the callgraph from 148 backend. 149 - cgraph_add_new_function is used to add backend produced 150 functions introduced after the unit is finalized. 151 The functions are enqueue for later processing and inserted 152 into callgraph with cgraph_process_new_functions. 153 154 - cgraph_function_versioning 155 156 produces a copy of function into new one (a version) 157 and apply simple transformations 158 */ 159 160 #include "config.h" 161 #include "system.h" 162 #include "coretypes.h" 163 #include "backend.h" 164 #include "target.h" 165 #include "rtl.h" 166 #include "tree.h" 167 #include "gimple.h" 168 #include "cfghooks.h" 169 #include "regset.h" /* FIXME: For reg_obstack. */ 170 #include "alloc-pool.h" 171 #include "tree-pass.h" 172 #include "stringpool.h" 173 #include "gimple-ssa.h" 174 #include "cgraph.h" 175 #include "coverage.h" 176 #include "lto-streamer.h" 177 #include "fold-const.h" 178 #include "varasm.h" 179 #include "stor-layout.h" 180 #include "output.h" 181 #include "cfgcleanup.h" 182 #include "gimple-fold.h" 183 #include "gimplify.h" 184 #include "gimple-iterator.h" 185 #include "gimplify-me.h" 186 #include "tree-cfg.h" 187 #include "tree-into-ssa.h" 188 #include "tree-ssa.h" 189 #include "langhooks.h" 190 #include "toplev.h" 191 #include "debug.h" 192 #include "symbol-summary.h" 193 #include "tree-vrp.h" 194 #include "ipa-prop.h" 195 #include "gimple-pretty-print.h" 196 #include "plugin.h" 197 #include "ipa-fnsummary.h" 198 #include "ipa-utils.h" 199 #include "except.h" 200 #include "cfgloop.h" 201 #include "context.h" 202 #include "pass_manager.h" 203 #include "tree-nested.h" 204 #include "dbgcnt.h" 205 #include "lto-section-names.h" 206 #include "stringpool.h" 207 #include "attribs.h" 208 #include "ipa-inline.h" 209 210 /* Queue of cgraph nodes scheduled to be added into cgraph. This is a 211 secondary queue used during optimization to accommodate passes that 212 may generate new functions that need to be optimized and expanded. */ 213 vec<cgraph_node *> cgraph_new_nodes; 214 215 static void expand_all_functions (void); 216 static void mark_functions_to_output (void); 217 static void handle_alias_pairs (void); 218 219 /* Used for vtable lookup in thunk adjusting. */ 220 static GTY (()) tree vtable_entry_type; 221 222 /* Return true if this symbol is a function from the C frontend specified 223 directly in RTL form (with "__RTL"). */ 224 225 bool 226 symtab_node::native_rtl_p () const 227 { 228 if (TREE_CODE (decl) != FUNCTION_DECL) 229 return false; 230 if (!DECL_STRUCT_FUNCTION (decl)) 231 return false; 232 return DECL_STRUCT_FUNCTION (decl)->curr_properties & PROP_rtl; 233 } 234 235 /* Determine if symbol declaration is needed. That is, visible to something 236 either outside this translation unit, something magic in the system 237 configury */ 238 bool 239 symtab_node::needed_p (void) 240 { 241 /* Double check that no one output the function into assembly file 242 early. */ 243 if (!native_rtl_p ()) 244 gcc_checking_assert 245 (!DECL_ASSEMBLER_NAME_SET_P (decl) 246 || !TREE_SYMBOL_REFERENCED (DECL_ASSEMBLER_NAME (decl))); 247 248 if (!definition) 249 return false; 250 251 if (DECL_EXTERNAL (decl)) 252 return false; 253 254 /* If the user told us it is used, then it must be so. */ 255 if (force_output) 256 return true; 257 258 /* ABI forced symbols are needed when they are external. */ 259 if (forced_by_abi && TREE_PUBLIC (decl)) 260 return true; 261 262 /* Keep constructors, destructors and virtual functions. */ 263 if (TREE_CODE (decl) == FUNCTION_DECL 264 && (DECL_STATIC_CONSTRUCTOR (decl) || DECL_STATIC_DESTRUCTOR (decl))) 265 return true; 266 267 /* Externally visible variables must be output. The exception is 268 COMDAT variables that must be output only when they are needed. */ 269 if (TREE_PUBLIC (decl) && !DECL_COMDAT (decl)) 270 return true; 271 272 return false; 273 } 274 275 /* Head and terminator of the queue of nodes to be processed while building 276 callgraph. */ 277 278 static symtab_node symtab_terminator (SYMTAB_SYMBOL); 279 static symtab_node *queued_nodes = &symtab_terminator; 280 281 /* Add NODE to queue starting at QUEUED_NODES. 282 The queue is linked via AUX pointers and terminated by pointer to 1. */ 283 284 static void 285 enqueue_node (symtab_node *node) 286 { 287 if (node->aux) 288 return; 289 gcc_checking_assert (queued_nodes); 290 node->aux = queued_nodes; 291 queued_nodes = node; 292 } 293 294 /* Process CGRAPH_NEW_FUNCTIONS and perform actions necessary to add these 295 functions into callgraph in a way so they look like ordinary reachable 296 functions inserted into callgraph already at construction time. */ 297 298 void 299 symbol_table::process_new_functions (void) 300 { 301 tree fndecl; 302 303 if (!cgraph_new_nodes.exists ()) 304 return; 305 306 handle_alias_pairs (); 307 /* Note that this queue may grow as its being processed, as the new 308 functions may generate new ones. */ 309 for (unsigned i = 0; i < cgraph_new_nodes.length (); i++) 310 { 311 cgraph_node *node = cgraph_new_nodes[i]; 312 fndecl = node->decl; 313 switch (state) 314 { 315 case CONSTRUCTION: 316 /* At construction time we just need to finalize function and move 317 it into reachable functions list. */ 318 319 cgraph_node::finalize_function (fndecl, false); 320 call_cgraph_insertion_hooks (node); 321 enqueue_node (node); 322 break; 323 324 case IPA: 325 case IPA_SSA: 326 case IPA_SSA_AFTER_INLINING: 327 /* When IPA optimization already started, do all essential 328 transformations that has been already performed on the whole 329 cgraph but not on this function. */ 330 331 gimple_register_cfg_hooks (); 332 if (!node->analyzed) 333 node->analyze (); 334 push_cfun (DECL_STRUCT_FUNCTION (fndecl)); 335 if ((state == IPA_SSA || state == IPA_SSA_AFTER_INLINING) 336 && !gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl))) 337 { 338 bool summaried_computed = ipa_fn_summaries != NULL; 339 g->get_passes ()->execute_early_local_passes (); 340 /* Early passes compute inline parameters to do inlining 341 and splitting. This is redundant for functions added late. 342 Just throw away whatever it did. */ 343 if (!summaried_computed) 344 { 345 ipa_free_fn_summary (); 346 ipa_free_size_summary (); 347 } 348 } 349 else if (ipa_fn_summaries != NULL) 350 compute_fn_summary (node, true); 351 free_dominance_info (CDI_POST_DOMINATORS); 352 free_dominance_info (CDI_DOMINATORS); 353 pop_cfun (); 354 call_cgraph_insertion_hooks (node); 355 break; 356 357 case EXPANSION: 358 /* Functions created during expansion shall be compiled 359 directly. */ 360 node->process = 0; 361 call_cgraph_insertion_hooks (node); 362 node->expand (); 363 break; 364 365 default: 366 gcc_unreachable (); 367 break; 368 } 369 } 370 371 cgraph_new_nodes.release (); 372 } 373 374 /* As an GCC extension we allow redefinition of the function. The 375 semantics when both copies of bodies differ is not well defined. 376 We replace the old body with new body so in unit at a time mode 377 we always use new body, while in normal mode we may end up with 378 old body inlined into some functions and new body expanded and 379 inlined in others. 380 381 ??? It may make more sense to use one body for inlining and other 382 body for expanding the function but this is difficult to do. */ 383 384 void 385 cgraph_node::reset (void) 386 { 387 /* If process is set, then we have already begun whole-unit analysis. 388 This is *not* testing for whether we've already emitted the function. 389 That case can be sort-of legitimately seen with real function redefinition 390 errors. I would argue that the front end should never present us with 391 such a case, but don't enforce that for now. */ 392 gcc_assert (!process); 393 394 /* Reset our data structures so we can analyze the function again. */ 395 inlined_to = NULL; 396 memset (&rtl, 0, sizeof (rtl)); 397 analyzed = false; 398 definition = false; 399 alias = false; 400 transparent_alias = false; 401 weakref = false; 402 cpp_implicit_alias = false; 403 404 remove_callees (); 405 remove_all_references (); 406 } 407 408 /* Return true when there are references to the node. INCLUDE_SELF is 409 true if a self reference counts as a reference. */ 410 411 bool 412 symtab_node::referred_to_p (bool include_self) 413 { 414 ipa_ref *ref = NULL; 415 416 /* See if there are any references at all. */ 417 if (iterate_referring (0, ref)) 418 return true; 419 /* For functions check also calls. */ 420 cgraph_node *cn = dyn_cast <cgraph_node *> (this); 421 if (cn && cn->callers) 422 { 423 if (include_self) 424 return true; 425 for (cgraph_edge *e = cn->callers; e; e = e->next_caller) 426 if (e->caller != this) 427 return true; 428 } 429 return false; 430 } 431 432 /* DECL has been parsed. Take it, queue it, compile it at the whim of the 433 logic in effect. If NO_COLLECT is true, then our caller cannot stand to have 434 the garbage collector run at the moment. We would need to either create 435 a new GC context, or just not compile right now. */ 436 437 void 438 cgraph_node::finalize_function (tree decl, bool no_collect) 439 { 440 cgraph_node *node = cgraph_node::get_create (decl); 441 442 if (node->definition) 443 { 444 /* Nested functions should only be defined once. */ 445 gcc_assert (!DECL_CONTEXT (decl) 446 || TREE_CODE (DECL_CONTEXT (decl)) != FUNCTION_DECL); 447 node->reset (); 448 node->redefined_extern_inline = true; 449 } 450 451 /* Set definition first before calling notice_global_symbol so that 452 it is available to notice_global_symbol. */ 453 node->definition = true; 454 notice_global_symbol (decl); 455 node->lowered = DECL_STRUCT_FUNCTION (decl)->cfg != NULL; 456 if (!flag_toplevel_reorder) 457 node->no_reorder = true; 458 459 /* With -fkeep-inline-functions we are keeping all inline functions except 460 for extern inline ones. */ 461 if (flag_keep_inline_functions 462 && DECL_DECLARED_INLINE_P (decl) 463 && !DECL_EXTERNAL (decl) 464 && !DECL_DISREGARD_INLINE_LIMITS (decl)) 465 node->force_output = 1; 466 467 /* __RTL functions were already output as soon as they were parsed (due 468 to the large amount of global state in the backend). 469 Mark such functions as "force_output" to reflect the fact that they 470 will be in the asm file when considering the symbols they reference. 471 The attempt to output them later on will bail out immediately. */ 472 if (node->native_rtl_p ()) 473 node->force_output = 1; 474 475 /* When not optimizing, also output the static functions. (see 476 PR24561), but don't do so for always_inline functions, functions 477 declared inline and nested functions. These were optimized out 478 in the original implementation and it is unclear whether we want 479 to change the behavior here. */ 480 if (((!opt_for_fn (decl, optimize) || flag_keep_static_functions 481 || node->no_reorder) 482 && !node->cpp_implicit_alias 483 && !DECL_DISREGARD_INLINE_LIMITS (decl) 484 && !DECL_DECLARED_INLINE_P (decl) 485 && !(DECL_CONTEXT (decl) 486 && TREE_CODE (DECL_CONTEXT (decl)) == FUNCTION_DECL)) 487 && !DECL_COMDAT (decl) && !DECL_EXTERNAL (decl)) 488 node->force_output = 1; 489 490 /* If we've not yet emitted decl, tell the debug info about it. */ 491 if (!TREE_ASM_WRITTEN (decl)) 492 (*debug_hooks->deferred_inline_function) (decl); 493 494 if (!no_collect) 495 ggc_collect (); 496 497 if (symtab->state == CONSTRUCTION 498 && (node->needed_p () || node->referred_to_p ())) 499 enqueue_node (node); 500 } 501 502 /* Add the function FNDECL to the call graph. 503 Unlike finalize_function, this function is intended to be used 504 by middle end and allows insertion of new function at arbitrary point 505 of compilation. The function can be either in high, low or SSA form 506 GIMPLE. 507 508 The function is assumed to be reachable and have address taken (so no 509 API breaking optimizations are performed on it). 510 511 Main work done by this function is to enqueue the function for later 512 processing to avoid need the passes to be re-entrant. */ 513 514 void 515 cgraph_node::add_new_function (tree fndecl, bool lowered) 516 { 517 gcc::pass_manager *passes = g->get_passes (); 518 cgraph_node *node; 519 520 if (dump_file) 521 { 522 struct function *fn = DECL_STRUCT_FUNCTION (fndecl); 523 const char *function_type = ((gimple_has_body_p (fndecl)) 524 ? (lowered 525 ? (gimple_in_ssa_p (fn) 526 ? "ssa gimple" 527 : "low gimple") 528 : "high gimple") 529 : "to-be-gimplified"); 530 fprintf (dump_file, 531 "Added new %s function %s to callgraph\n", 532 function_type, 533 fndecl_name (fndecl)); 534 } 535 536 switch (symtab->state) 537 { 538 case PARSING: 539 cgraph_node::finalize_function (fndecl, false); 540 break; 541 case CONSTRUCTION: 542 /* Just enqueue function to be processed at nearest occurrence. */ 543 node = cgraph_node::get_create (fndecl); 544 if (lowered) 545 node->lowered = true; 546 cgraph_new_nodes.safe_push (node); 547 break; 548 549 case IPA: 550 case IPA_SSA: 551 case IPA_SSA_AFTER_INLINING: 552 case EXPANSION: 553 /* Bring the function into finalized state and enqueue for later 554 analyzing and compilation. */ 555 node = cgraph_node::get_create (fndecl); 556 node->local = false; 557 node->definition = true; 558 node->force_output = true; 559 if (TREE_PUBLIC (fndecl)) 560 node->externally_visible = true; 561 if (!lowered && symtab->state == EXPANSION) 562 { 563 push_cfun (DECL_STRUCT_FUNCTION (fndecl)); 564 gimple_register_cfg_hooks (); 565 bitmap_obstack_initialize (NULL); 566 execute_pass_list (cfun, passes->all_lowering_passes); 567 passes->execute_early_local_passes (); 568 bitmap_obstack_release (NULL); 569 pop_cfun (); 570 571 lowered = true; 572 } 573 if (lowered) 574 node->lowered = true; 575 cgraph_new_nodes.safe_push (node); 576 break; 577 578 case FINISHED: 579 /* At the very end of compilation we have to do all the work up 580 to expansion. */ 581 node = cgraph_node::create (fndecl); 582 if (lowered) 583 node->lowered = true; 584 node->definition = true; 585 node->analyze (); 586 push_cfun (DECL_STRUCT_FUNCTION (fndecl)); 587 gimple_register_cfg_hooks (); 588 bitmap_obstack_initialize (NULL); 589 if (!gimple_in_ssa_p (DECL_STRUCT_FUNCTION (fndecl))) 590 g->get_passes ()->execute_early_local_passes (); 591 bitmap_obstack_release (NULL); 592 pop_cfun (); 593 node->expand (); 594 break; 595 596 default: 597 gcc_unreachable (); 598 } 599 600 /* Set a personality if required and we already passed EH lowering. */ 601 if (lowered 602 && (function_needs_eh_personality (DECL_STRUCT_FUNCTION (fndecl)) 603 == eh_personality_lang)) 604 DECL_FUNCTION_PERSONALITY (fndecl) = lang_hooks.eh_personality (); 605 } 606 607 /* Analyze the function scheduled to be output. */ 608 void 609 cgraph_node::analyze (void) 610 { 611 if (native_rtl_p ()) 612 { 613 analyzed = true; 614 return; 615 } 616 617 tree decl = this->decl; 618 location_t saved_loc = input_location; 619 input_location = DECL_SOURCE_LOCATION (decl); 620 621 if (thunk.thunk_p) 622 { 623 cgraph_node *t = cgraph_node::get (thunk.alias); 624 625 create_edge (t, NULL, t->count); 626 callees->can_throw_external = !TREE_NOTHROW (t->decl); 627 /* Target code in expand_thunk may need the thunk's target 628 to be analyzed, so recurse here. */ 629 if (!t->analyzed && t->definition) 630 t->analyze (); 631 if (t->alias) 632 { 633 t = t->get_alias_target (); 634 if (!t->analyzed && t->definition) 635 t->analyze (); 636 } 637 bool ret = expand_thunk (false, false); 638 thunk.alias = NULL; 639 if (!ret) 640 return; 641 } 642 if (alias) 643 resolve_alias (cgraph_node::get (alias_target), transparent_alias); 644 else if (dispatcher_function) 645 { 646 /* Generate the dispatcher body of multi-versioned functions. */ 647 cgraph_function_version_info *dispatcher_version_info 648 = function_version (); 649 if (dispatcher_version_info != NULL 650 && (dispatcher_version_info->dispatcher_resolver 651 == NULL_TREE)) 652 { 653 tree resolver = NULL_TREE; 654 gcc_assert (targetm.generate_version_dispatcher_body); 655 resolver = targetm.generate_version_dispatcher_body (this); 656 gcc_assert (resolver != NULL_TREE); 657 } 658 } 659 else 660 { 661 push_cfun (DECL_STRUCT_FUNCTION (decl)); 662 663 assign_assembler_name_if_needed (decl); 664 665 /* Make sure to gimplify bodies only once. During analyzing a 666 function we lower it, which will require gimplified nested 667 functions, so we can end up here with an already gimplified 668 body. */ 669 if (!gimple_has_body_p (decl)) 670 gimplify_function_tree (decl); 671 672 /* Lower the function. */ 673 if (!lowered) 674 { 675 if (nested) 676 lower_nested_functions (decl); 677 gcc_assert (!nested); 678 679 gimple_register_cfg_hooks (); 680 bitmap_obstack_initialize (NULL); 681 execute_pass_list (cfun, g->get_passes ()->all_lowering_passes); 682 free_dominance_info (CDI_POST_DOMINATORS); 683 free_dominance_info (CDI_DOMINATORS); 684 compact_blocks (); 685 bitmap_obstack_release (NULL); 686 lowered = true; 687 } 688 689 pop_cfun (); 690 } 691 analyzed = true; 692 693 input_location = saved_loc; 694 } 695 696 /* C++ frontend produce same body aliases all over the place, even before PCH 697 gets streamed out. It relies on us linking the aliases with their function 698 in order to do the fixups, but ipa-ref is not PCH safe. Consequently we 699 first produce aliases without links, but once C++ FE is sure he won't stream 700 PCH we build the links via this function. */ 701 702 void 703 symbol_table::process_same_body_aliases (void) 704 { 705 symtab_node *node; 706 FOR_EACH_SYMBOL (node) 707 if (node->cpp_implicit_alias && !node->analyzed) 708 node->resolve_alias 709 (VAR_P (node->alias_target) 710 ? (symtab_node *)varpool_node::get_create (node->alias_target) 711 : (symtab_node *)cgraph_node::get_create (node->alias_target)); 712 cpp_implicit_aliases_done = true; 713 } 714 715 /* Process a symver attribute. */ 716 717 static void 718 process_symver_attribute (symtab_node *n) 719 { 720 tree value = lookup_attribute ("symver", DECL_ATTRIBUTES (n->decl)); 721 722 if (!value) 723 return; 724 if (lookup_attribute ("symver", TREE_CHAIN (value))) 725 { 726 error_at (DECL_SOURCE_LOCATION (n->decl), 727 "multiple versions for one symbol"); 728 return; 729 } 730 tree symver = get_identifier_with_length 731 (TREE_STRING_POINTER (TREE_VALUE (TREE_VALUE (value))), 732 TREE_STRING_LENGTH (TREE_VALUE (TREE_VALUE (value)))); 733 symtab_node *def = symtab_node::get_for_asmname (symver); 734 735 if (def) 736 { 737 error_at (DECL_SOURCE_LOCATION (n->decl), 738 "duplicate definition of a symbol version"); 739 inform (DECL_SOURCE_LOCATION (def->decl), 740 "same version was previously defined here"); 741 return; 742 } 743 if (!n->definition) 744 { 745 error_at (DECL_SOURCE_LOCATION (n->decl), 746 "symbol needs to be defined to have a version"); 747 return; 748 } 749 if (DECL_COMMON (n->decl)) 750 { 751 error_at (DECL_SOURCE_LOCATION (n->decl), 752 "common symbol cannot be versioned"); 753 return; 754 } 755 if (DECL_COMDAT (n->decl)) 756 { 757 error_at (DECL_SOURCE_LOCATION (n->decl), 758 "comdat symbol cannot be versioned"); 759 return; 760 } 761 if (n->weakref) 762 { 763 error_at (DECL_SOURCE_LOCATION (n->decl), 764 "weakref cannot be versioned"); 765 return; 766 } 767 if (!TREE_PUBLIC (n->decl)) 768 { 769 error_at (DECL_SOURCE_LOCATION (n->decl), 770 "versioned symbol must be public"); 771 return; 772 } 773 if (DECL_VISIBILITY (n->decl) != VISIBILITY_DEFAULT) 774 { 775 error_at (DECL_SOURCE_LOCATION (n->decl), 776 "versioned symbol must have default visibility"); 777 return; 778 } 779 780 /* Create new symbol table entry representing the version. */ 781 tree new_decl = copy_node (n->decl); 782 783 DECL_INITIAL (new_decl) = NULL_TREE; 784 if (TREE_CODE (new_decl) == FUNCTION_DECL) 785 DECL_STRUCT_FUNCTION (new_decl) = NULL; 786 SET_DECL_ASSEMBLER_NAME (new_decl, symver); 787 TREE_PUBLIC (new_decl) = 1; 788 DECL_ATTRIBUTES (new_decl) = NULL; 789 790 symtab_node *symver_node = symtab_node::get_create (new_decl); 791 symver_node->alias = true; 792 symver_node->definition = true; 793 symver_node->symver = true; 794 symver_node->create_reference (n, IPA_REF_ALIAS, NULL); 795 symver_node->analyzed = true; 796 } 797 798 /* Process attributes common for vars and functions. */ 799 800 static void 801 process_common_attributes (symtab_node *node, tree decl) 802 { 803 tree weakref = lookup_attribute ("weakref", DECL_ATTRIBUTES (decl)); 804 805 if (weakref && !lookup_attribute ("alias", DECL_ATTRIBUTES (decl))) 806 { 807 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes, 808 "%<weakref%> attribute should be accompanied with" 809 " an %<alias%> attribute"); 810 DECL_WEAK (decl) = 0; 811 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref", 812 DECL_ATTRIBUTES (decl)); 813 } 814 815 if (lookup_attribute ("no_reorder", DECL_ATTRIBUTES (decl))) 816 node->no_reorder = 1; 817 process_symver_attribute (node); 818 } 819 820 /* Look for externally_visible and used attributes and mark cgraph nodes 821 accordingly. 822 823 We cannot mark the nodes at the point the attributes are processed (in 824 handle_*_attribute) because the copy of the declarations available at that 825 point may not be canonical. For example, in: 826 827 void f(); 828 void f() __attribute__((used)); 829 830 the declaration we see in handle_used_attribute will be the second 831 declaration -- but the front end will subsequently merge that declaration 832 with the original declaration and discard the second declaration. 833 834 Furthermore, we can't mark these nodes in finalize_function because: 835 836 void f() {} 837 void f() __attribute__((externally_visible)); 838 839 is valid. 840 841 So, we walk the nodes at the end of the translation unit, applying the 842 attributes at that point. */ 843 844 static void 845 process_function_and_variable_attributes (cgraph_node *first, 846 varpool_node *first_var) 847 { 848 cgraph_node *node; 849 varpool_node *vnode; 850 851 for (node = symtab->first_function (); node != first; 852 node = symtab->next_function (node)) 853 { 854 tree decl = node->decl; 855 856 if (node->alias 857 && lookup_attribute ("flatten", DECL_ATTRIBUTES (decl))) 858 { 859 tree tdecl = node->get_alias_target_tree (); 860 if (!tdecl || !DECL_P (tdecl) 861 || !lookup_attribute ("flatten", DECL_ATTRIBUTES (tdecl))) 862 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes, 863 "%<flatten%> attribute is ignored on aliases"); 864 } 865 if (DECL_PRESERVE_P (decl)) 866 node->mark_force_output (); 867 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl))) 868 { 869 if (! TREE_PUBLIC (node->decl)) 870 warning_at (DECL_SOURCE_LOCATION (node->decl), OPT_Wattributes, 871 "%<externally_visible%>" 872 " attribute have effect only on public objects"); 873 } 874 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl)) 875 && node->definition 876 && (!node->alias || DECL_INITIAL (decl) != error_mark_node)) 877 { 878 /* NODE->DEFINITION && NODE->ALIAS is nonzero for valid weakref 879 function declarations; DECL_INITIAL is non-null for invalid 880 weakref functions that are also defined. */ 881 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes, 882 "%<weakref%> attribute ignored" 883 " because function is defined"); 884 DECL_WEAK (decl) = 0; 885 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref", 886 DECL_ATTRIBUTES (decl)); 887 DECL_ATTRIBUTES (decl) = remove_attribute ("alias", 888 DECL_ATTRIBUTES (decl)); 889 node->alias = false; 890 node->weakref = false; 891 node->transparent_alias = false; 892 } 893 else if (lookup_attribute ("alias", DECL_ATTRIBUTES (decl)) 894 && node->definition 895 && !node->alias) 896 warning_at (DECL_SOURCE_LOCATION (node->decl), OPT_Wattributes, 897 "%<alias%> attribute ignored" 898 " because function is defined"); 899 900 if (lookup_attribute ("always_inline", DECL_ATTRIBUTES (decl)) 901 && !DECL_DECLARED_INLINE_P (decl) 902 /* redefining extern inline function makes it DECL_UNINLINABLE. */ 903 && !DECL_UNINLINABLE (decl)) 904 warning_at (DECL_SOURCE_LOCATION (decl), OPT_Wattributes, 905 "%<always_inline%> function might not be inlinable"); 906 907 process_common_attributes (node, decl); 908 } 909 for (vnode = symtab->first_variable (); vnode != first_var; 910 vnode = symtab->next_variable (vnode)) 911 { 912 tree decl = vnode->decl; 913 if (DECL_EXTERNAL (decl) 914 && DECL_INITIAL (decl)) 915 varpool_node::finalize_decl (decl); 916 if (DECL_PRESERVE_P (decl)) 917 vnode->force_output = true; 918 else if (lookup_attribute ("externally_visible", DECL_ATTRIBUTES (decl))) 919 { 920 if (! TREE_PUBLIC (vnode->decl)) 921 warning_at (DECL_SOURCE_LOCATION (vnode->decl), OPT_Wattributes, 922 "%<externally_visible%>" 923 " attribute have effect only on public objects"); 924 } 925 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl)) 926 && vnode->definition 927 && DECL_INITIAL (decl)) 928 { 929 warning_at (DECL_SOURCE_LOCATION (vnode->decl), OPT_Wattributes, 930 "%<weakref%> attribute ignored" 931 " because variable is initialized"); 932 DECL_WEAK (decl) = 0; 933 DECL_ATTRIBUTES (decl) = remove_attribute ("weakref", 934 DECL_ATTRIBUTES (decl)); 935 } 936 process_common_attributes (vnode, decl); 937 } 938 } 939 940 /* Mark DECL as finalized. By finalizing the declaration, frontend instruct the 941 middle end to output the variable to asm file, if needed or externally 942 visible. */ 943 944 void 945 varpool_node::finalize_decl (tree decl) 946 { 947 varpool_node *node = varpool_node::get_create (decl); 948 949 gcc_assert (TREE_STATIC (decl) || DECL_EXTERNAL (decl)); 950 951 if (node->definition) 952 return; 953 /* Set definition first before calling notice_global_symbol so that 954 it is available to notice_global_symbol. */ 955 node->definition = true; 956 notice_global_symbol (decl); 957 if (!flag_toplevel_reorder) 958 node->no_reorder = true; 959 if (TREE_THIS_VOLATILE (decl) || DECL_PRESERVE_P (decl) 960 /* Traditionally we do not eliminate static variables when not 961 optimizing and when not doing toplevel reorder. */ 962 || (node->no_reorder && !DECL_COMDAT (node->decl) 963 && !DECL_ARTIFICIAL (node->decl))) 964 node->force_output = true; 965 966 if (symtab->state == CONSTRUCTION 967 && (node->needed_p () || node->referred_to_p ())) 968 enqueue_node (node); 969 if (symtab->state >= IPA_SSA) 970 node->analyze (); 971 /* Some frontends produce various interface variables after compilation 972 finished. */ 973 if (symtab->state == FINISHED 974 || (node->no_reorder 975 && symtab->state == EXPANSION)) 976 node->assemble_decl (); 977 } 978 979 /* EDGE is an polymorphic call. Mark all possible targets as reachable 980 and if there is only one target, perform trivial devirtualization. 981 REACHABLE_CALL_TARGETS collects target lists we already walked to 982 avoid duplicate work. */ 983 984 static void 985 walk_polymorphic_call_targets (hash_set<void *> *reachable_call_targets, 986 cgraph_edge *edge) 987 { 988 unsigned int i; 989 void *cache_token; 990 bool final; 991 vec <cgraph_node *>targets 992 = possible_polymorphic_call_targets 993 (edge, &final, &cache_token); 994 995 if (!reachable_call_targets->add (cache_token)) 996 { 997 if (symtab->dump_file) 998 dump_possible_polymorphic_call_targets 999 (symtab->dump_file, edge); 1000 1001 for (i = 0; i < targets.length (); i++) 1002 { 1003 /* Do not bother to mark virtual methods in anonymous namespace; 1004 either we will find use of virtual table defining it, or it is 1005 unused. */ 1006 if (targets[i]->definition 1007 && TREE_CODE 1008 (TREE_TYPE (targets[i]->decl)) 1009 == METHOD_TYPE 1010 && !type_in_anonymous_namespace_p 1011 (TYPE_METHOD_BASETYPE (TREE_TYPE (targets[i]->decl)))) 1012 enqueue_node (targets[i]); 1013 } 1014 } 1015 1016 /* Very trivial devirtualization; when the type is 1017 final or anonymous (so we know all its derivation) 1018 and there is only one possible virtual call target, 1019 make the edge direct. */ 1020 if (final) 1021 { 1022 if (targets.length () <= 1 && dbg_cnt (devirt)) 1023 { 1024 cgraph_node *target; 1025 if (targets.length () == 1) 1026 target = targets[0]; 1027 else 1028 target = cgraph_node::create 1029 (builtin_decl_implicit (BUILT_IN_UNREACHABLE)); 1030 1031 if (symtab->dump_file) 1032 { 1033 fprintf (symtab->dump_file, 1034 "Devirtualizing call: "); 1035 print_gimple_stmt (symtab->dump_file, 1036 edge->call_stmt, 0, 1037 TDF_SLIM); 1038 } 1039 if (dump_enabled_p ()) 1040 { 1041 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, edge->call_stmt, 1042 "devirtualizing call in %s to %s\n", 1043 edge->caller->dump_name (), 1044 target->dump_name ()); 1045 } 1046 1047 edge = cgraph_edge::make_direct (edge, target); 1048 gimple *new_call = cgraph_edge::redirect_call_stmt_to_callee (edge); 1049 1050 if (symtab->dump_file) 1051 { 1052 fprintf (symtab->dump_file, "Devirtualized as: "); 1053 print_gimple_stmt (symtab->dump_file, new_call, 0, TDF_SLIM); 1054 } 1055 } 1056 } 1057 } 1058 1059 /* Issue appropriate warnings for the global declaration DECL. */ 1060 1061 static void 1062 check_global_declaration (symtab_node *snode) 1063 { 1064 const char *decl_file; 1065 tree decl = snode->decl; 1066 1067 /* Warn about any function declared static but not defined. We don't 1068 warn about variables, because many programs have static variables 1069 that exist only to get some text into the object file. */ 1070 if (TREE_CODE (decl) == FUNCTION_DECL 1071 && DECL_INITIAL (decl) == 0 1072 && DECL_EXTERNAL (decl) 1073 && ! DECL_ARTIFICIAL (decl) 1074 && ! TREE_PUBLIC (decl)) 1075 { 1076 if (TREE_NO_WARNING (decl)) 1077 ; 1078 else if (snode->referred_to_p (/*include_self=*/false)) 1079 pedwarn (input_location, 0, "%q+F used but never defined", decl); 1080 else 1081 warning (OPT_Wunused_function, "%q+F declared %<static%> but never " 1082 "defined", decl); 1083 /* This symbol is effectively an "extern" declaration now. */ 1084 TREE_PUBLIC (decl) = 1; 1085 } 1086 1087 /* Warn about static fns or vars defined but not used. */ 1088 if (((warn_unused_function && TREE_CODE (decl) == FUNCTION_DECL) 1089 || (((warn_unused_variable && ! TREE_READONLY (decl)) 1090 || (warn_unused_const_variable > 0 && TREE_READONLY (decl) 1091 && (warn_unused_const_variable == 2 1092 || (main_input_filename != NULL 1093 && (decl_file = DECL_SOURCE_FILE (decl)) != NULL 1094 && filename_cmp (main_input_filename, 1095 decl_file) == 0)))) 1096 && VAR_P (decl))) 1097 && ! DECL_IN_SYSTEM_HEADER (decl) 1098 && ! snode->referred_to_p (/*include_self=*/false) 1099 /* This TREE_USED check is needed in addition to referred_to_p 1100 above, because the `__unused__' attribute is not being 1101 considered for referred_to_p. */ 1102 && ! TREE_USED (decl) 1103 /* The TREE_USED bit for file-scope decls is kept in the identifier, 1104 to handle multiple external decls in different scopes. */ 1105 && ! (DECL_NAME (decl) && TREE_USED (DECL_NAME (decl))) 1106 && ! DECL_EXTERNAL (decl) 1107 && ! DECL_ARTIFICIAL (decl) 1108 && ! DECL_ABSTRACT_ORIGIN (decl) 1109 && ! TREE_PUBLIC (decl) 1110 /* A volatile variable might be used in some non-obvious way. */ 1111 && (! VAR_P (decl) || ! TREE_THIS_VOLATILE (decl)) 1112 /* Global register variables must be declared to reserve them. */ 1113 && ! (VAR_P (decl) && DECL_REGISTER (decl)) 1114 /* Global ctors and dtors are called by the runtime. */ 1115 && (TREE_CODE (decl) != FUNCTION_DECL 1116 || (!DECL_STATIC_CONSTRUCTOR (decl) 1117 && !DECL_STATIC_DESTRUCTOR (decl))) 1118 /* Otherwise, ask the language. */ 1119 && lang_hooks.decls.warn_unused_global (decl)) 1120 warning_at (DECL_SOURCE_LOCATION (decl), 1121 (TREE_CODE (decl) == FUNCTION_DECL) 1122 ? OPT_Wunused_function 1123 : (TREE_READONLY (decl) 1124 ? OPT_Wunused_const_variable_ 1125 : OPT_Wunused_variable), 1126 "%qD defined but not used", decl); 1127 } 1128 1129 /* Discover all functions and variables that are trivially needed, analyze 1130 them as well as all functions and variables referred by them */ 1131 static cgraph_node *first_analyzed; 1132 static varpool_node *first_analyzed_var; 1133 1134 /* FIRST_TIME is set to TRUE for the first time we are called for a 1135 translation unit from finalize_compilation_unit() or false 1136 otherwise. */ 1137 1138 static void 1139 analyze_functions (bool first_time) 1140 { 1141 /* Keep track of already processed nodes when called multiple times for 1142 intermodule optimization. */ 1143 cgraph_node *first_handled = first_analyzed; 1144 varpool_node *first_handled_var = first_analyzed_var; 1145 hash_set<void *> reachable_call_targets; 1146 1147 symtab_node *node; 1148 symtab_node *next; 1149 int i; 1150 ipa_ref *ref; 1151 bool changed = true; 1152 location_t saved_loc = input_location; 1153 1154 bitmap_obstack_initialize (NULL); 1155 symtab->state = CONSTRUCTION; 1156 input_location = UNKNOWN_LOCATION; 1157 1158 /* Ugly, but the fixup cannot happen at a time same body alias is created; 1159 C++ FE is confused about the COMDAT groups being right. */ 1160 if (symtab->cpp_implicit_aliases_done) 1161 FOR_EACH_SYMBOL (node) 1162 if (node->cpp_implicit_alias) 1163 node->fixup_same_cpp_alias_visibility (node->get_alias_target ()); 1164 build_type_inheritance_graph (); 1165 1166 /* Analysis adds static variables that in turn adds references to new functions. 1167 So we need to iterate the process until it stabilize. */ 1168 while (changed) 1169 { 1170 changed = false; 1171 process_function_and_variable_attributes (first_analyzed, 1172 first_analyzed_var); 1173 1174 /* First identify the trivially needed symbols. */ 1175 for (node = symtab->first_symbol (); 1176 node != first_analyzed 1177 && node != first_analyzed_var; node = node->next) 1178 { 1179 /* Convert COMDAT group designators to IDENTIFIER_NODEs. */ 1180 node->get_comdat_group_id (); 1181 if (node->needed_p ()) 1182 { 1183 enqueue_node (node); 1184 if (!changed && symtab->dump_file) 1185 fprintf (symtab->dump_file, "Trivially needed symbols:"); 1186 changed = true; 1187 if (symtab->dump_file) 1188 fprintf (symtab->dump_file, " %s", node->dump_asm_name ()); 1189 if (!changed && symtab->dump_file) 1190 fprintf (symtab->dump_file, "\n"); 1191 } 1192 if (node == first_analyzed 1193 || node == first_analyzed_var) 1194 break; 1195 } 1196 symtab->process_new_functions (); 1197 first_analyzed_var = symtab->first_variable (); 1198 first_analyzed = symtab->first_function (); 1199 1200 if (changed && symtab->dump_file) 1201 fprintf (symtab->dump_file, "\n"); 1202 1203 /* Lower representation, build callgraph edges and references for all trivially 1204 needed symbols and all symbols referred by them. */ 1205 while (queued_nodes != &symtab_terminator) 1206 { 1207 changed = true; 1208 node = queued_nodes; 1209 queued_nodes = (symtab_node *)queued_nodes->aux; 1210 cgraph_node *cnode = dyn_cast <cgraph_node *> (node); 1211 if (cnode && cnode->definition) 1212 { 1213 cgraph_edge *edge; 1214 tree decl = cnode->decl; 1215 1216 /* ??? It is possible to create extern inline function 1217 and later using weak alias attribute to kill its body. 1218 See gcc.c-torture/compile/20011119-1.c */ 1219 if (!DECL_STRUCT_FUNCTION (decl) 1220 && !cnode->alias 1221 && !cnode->thunk.thunk_p 1222 && !cnode->dispatcher_function) 1223 { 1224 cnode->reset (); 1225 cnode->redefined_extern_inline = true; 1226 continue; 1227 } 1228 1229 if (!cnode->analyzed) 1230 cnode->analyze (); 1231 1232 for (edge = cnode->callees; edge; edge = edge->next_callee) 1233 if (edge->callee->definition 1234 && (!DECL_EXTERNAL (edge->callee->decl) 1235 /* When not optimizing, do not try to analyze extern 1236 inline functions. Doing so is pointless. */ 1237 || opt_for_fn (edge->callee->decl, optimize) 1238 /* Weakrefs needs to be preserved. */ 1239 || edge->callee->alias 1240 /* always_inline functions are inlined even at -O0. */ 1241 || lookup_attribute 1242 ("always_inline", 1243 DECL_ATTRIBUTES (edge->callee->decl)) 1244 /* Multiversioned functions needs the dispatcher to 1245 be produced locally even for extern functions. */ 1246 || edge->callee->function_version ())) 1247 enqueue_node (edge->callee); 1248 if (opt_for_fn (cnode->decl, optimize) 1249 && opt_for_fn (cnode->decl, flag_devirtualize)) 1250 { 1251 cgraph_edge *next; 1252 1253 for (edge = cnode->indirect_calls; edge; edge = next) 1254 { 1255 next = edge->next_callee; 1256 if (edge->indirect_info->polymorphic) 1257 walk_polymorphic_call_targets (&reachable_call_targets, 1258 edge); 1259 } 1260 } 1261 1262 /* If decl is a clone of an abstract function, 1263 mark that abstract function so that we don't release its body. 1264 The DECL_INITIAL() of that abstract function declaration 1265 will be later needed to output debug info. */ 1266 if (DECL_ABSTRACT_ORIGIN (decl)) 1267 { 1268 cgraph_node *origin_node 1269 = cgraph_node::get_create (DECL_ABSTRACT_ORIGIN (decl)); 1270 origin_node->used_as_abstract_origin = true; 1271 } 1272 /* Preserve a functions function context node. It will 1273 later be needed to output debug info. */ 1274 if (tree fn = decl_function_context (decl)) 1275 { 1276 cgraph_node *origin_node = cgraph_node::get_create (fn); 1277 enqueue_node (origin_node); 1278 } 1279 } 1280 else 1281 { 1282 varpool_node *vnode = dyn_cast <varpool_node *> (node); 1283 if (vnode && vnode->definition && !vnode->analyzed) 1284 vnode->analyze (); 1285 } 1286 1287 if (node->same_comdat_group) 1288 { 1289 symtab_node *next; 1290 for (next = node->same_comdat_group; 1291 next != node; 1292 next = next->same_comdat_group) 1293 if (!next->comdat_local_p ()) 1294 enqueue_node (next); 1295 } 1296 for (i = 0; node->iterate_reference (i, ref); i++) 1297 if (ref->referred->definition 1298 && (!DECL_EXTERNAL (ref->referred->decl) 1299 || ((TREE_CODE (ref->referred->decl) != FUNCTION_DECL 1300 && optimize) 1301 || (TREE_CODE (ref->referred->decl) == FUNCTION_DECL 1302 && opt_for_fn (ref->referred->decl, optimize)) 1303 || node->alias 1304 || ref->referred->alias))) 1305 enqueue_node (ref->referred); 1306 symtab->process_new_functions (); 1307 } 1308 } 1309 update_type_inheritance_graph (); 1310 1311 /* Collect entry points to the unit. */ 1312 if (symtab->dump_file) 1313 { 1314 fprintf (symtab->dump_file, "\n\nInitial "); 1315 symtab->dump (symtab->dump_file); 1316 } 1317 1318 if (first_time) 1319 { 1320 symtab_node *snode; 1321 FOR_EACH_SYMBOL (snode) 1322 check_global_declaration (snode); 1323 } 1324 1325 if (symtab->dump_file) 1326 fprintf (symtab->dump_file, "\nRemoving unused symbols:"); 1327 1328 for (node = symtab->first_symbol (); 1329 node != first_handled 1330 && node != first_handled_var; node = next) 1331 { 1332 next = node->next; 1333 /* For symbols declared locally we clear TREE_READONLY when emitting 1334 the constructor (if one is needed). For external declarations we can 1335 not safely assume that the type is readonly because we may be called 1336 during its construction. */ 1337 if (TREE_CODE (node->decl) == VAR_DECL 1338 && TYPE_P (TREE_TYPE (node->decl)) 1339 && TYPE_NEEDS_CONSTRUCTING (TREE_TYPE (node->decl)) 1340 && DECL_EXTERNAL (node->decl)) 1341 TREE_READONLY (node->decl) = 0; 1342 if (!node->aux && !node->referred_to_p ()) 1343 { 1344 if (symtab->dump_file) 1345 fprintf (symtab->dump_file, " %s", node->dump_name ()); 1346 1347 /* See if the debugger can use anything before the DECL 1348 passes away. Perhaps it can notice a DECL that is now a 1349 constant and can tag the early DIE with an appropriate 1350 attribute. 1351 1352 Otherwise, this is the last chance the debug_hooks have 1353 at looking at optimized away DECLs, since 1354 late_global_decl will subsequently be called from the 1355 contents of the now pruned symbol table. */ 1356 if (VAR_P (node->decl) 1357 && !decl_function_context (node->decl)) 1358 { 1359 /* We are reclaiming totally unreachable code and variables 1360 so they effectively appear as readonly. Show that to 1361 the debug machinery. */ 1362 TREE_READONLY (node->decl) = 1; 1363 node->definition = false; 1364 (*debug_hooks->late_global_decl) (node->decl); 1365 } 1366 1367 node->remove (); 1368 continue; 1369 } 1370 if (cgraph_node *cnode = dyn_cast <cgraph_node *> (node)) 1371 { 1372 tree decl = node->decl; 1373 1374 if (cnode->definition && !gimple_has_body_p (decl) 1375 && !cnode->alias 1376 && !cnode->thunk.thunk_p) 1377 cnode->reset (); 1378 1379 gcc_assert (!cnode->definition || cnode->thunk.thunk_p 1380 || cnode->alias 1381 || gimple_has_body_p (decl) 1382 || cnode->native_rtl_p ()); 1383 gcc_assert (cnode->analyzed == cnode->definition); 1384 } 1385 node->aux = NULL; 1386 } 1387 for (;node; node = node->next) 1388 node->aux = NULL; 1389 first_analyzed = symtab->first_function (); 1390 first_analyzed_var = symtab->first_variable (); 1391 if (symtab->dump_file) 1392 { 1393 fprintf (symtab->dump_file, "\n\nReclaimed "); 1394 symtab->dump (symtab->dump_file); 1395 } 1396 bitmap_obstack_release (NULL); 1397 ggc_collect (); 1398 /* Initialize assembler name hash, in particular we want to trigger C++ 1399 mangling and same body alias creation before we free DECL_ARGUMENTS 1400 used by it. */ 1401 if (!seen_error ()) 1402 symtab->symtab_initialize_asm_name_hash (); 1403 1404 input_location = saved_loc; 1405 } 1406 1407 /* Check declaration of the type of ALIAS for compatibility with its TARGET 1408 (which may be an ifunc resolver) and issue a diagnostic when they are 1409 not compatible according to language rules (plus a C++ extension for 1410 non-static member functions). */ 1411 1412 static void 1413 maybe_diag_incompatible_alias (tree alias, tree target) 1414 { 1415 tree altype = TREE_TYPE (alias); 1416 tree targtype = TREE_TYPE (target); 1417 1418 bool ifunc = cgraph_node::get (alias)->ifunc_resolver; 1419 tree funcptr = altype; 1420 1421 if (ifunc) 1422 { 1423 /* Handle attribute ifunc first. */ 1424 if (TREE_CODE (altype) == METHOD_TYPE) 1425 { 1426 /* Set FUNCPTR to the type of the alias target. If the type 1427 is a non-static member function of class C, construct a type 1428 of an ordinary function taking C* as the first argument, 1429 followed by the member function argument list, and use it 1430 instead to check for incompatibility. This conversion is 1431 not defined by the language but an extension provided by 1432 G++. */ 1433 1434 tree rettype = TREE_TYPE (altype); 1435 tree args = TYPE_ARG_TYPES (altype); 1436 altype = build_function_type (rettype, args); 1437 funcptr = altype; 1438 } 1439 1440 targtype = TREE_TYPE (targtype); 1441 1442 if (POINTER_TYPE_P (targtype)) 1443 { 1444 targtype = TREE_TYPE (targtype); 1445 1446 /* Only issue Wattribute-alias for conversions to void* with 1447 -Wextra. */ 1448 if (VOID_TYPE_P (targtype) && !extra_warnings) 1449 return; 1450 1451 /* Proceed to handle incompatible ifunc resolvers below. */ 1452 } 1453 else 1454 { 1455 funcptr = build_pointer_type (funcptr); 1456 1457 error_at (DECL_SOURCE_LOCATION (target), 1458 "%<ifunc%> resolver for %qD must return %qT", 1459 alias, funcptr); 1460 inform (DECL_SOURCE_LOCATION (alias), 1461 "resolver indirect function declared here"); 1462 return; 1463 } 1464 } 1465 1466 if ((!FUNC_OR_METHOD_TYPE_P (targtype) 1467 || (prototype_p (altype) 1468 && prototype_p (targtype) 1469 && !types_compatible_p (altype, targtype)))) 1470 { 1471 /* Warn for incompatibilities. Avoid warning for functions 1472 without a prototype to make it possible to declare aliases 1473 without knowing the exact type, as libstdc++ does. */ 1474 if (ifunc) 1475 { 1476 funcptr = build_pointer_type (funcptr); 1477 1478 auto_diagnostic_group d; 1479 if (warning_at (DECL_SOURCE_LOCATION (target), 1480 OPT_Wattribute_alias_, 1481 "%<ifunc%> resolver for %qD should return %qT", 1482 alias, funcptr)) 1483 inform (DECL_SOURCE_LOCATION (alias), 1484 "resolver indirect function declared here"); 1485 } 1486 else 1487 { 1488 auto_diagnostic_group d; 1489 if (warning_at (DECL_SOURCE_LOCATION (alias), 1490 OPT_Wattribute_alias_, 1491 "%qD alias between functions of incompatible " 1492 "types %qT and %qT", alias, altype, targtype)) 1493 inform (DECL_SOURCE_LOCATION (target), 1494 "aliased declaration here"); 1495 } 1496 } 1497 } 1498 1499 /* Translate the ugly representation of aliases as alias pairs into nice 1500 representation in callgraph. We don't handle all cases yet, 1501 unfortunately. */ 1502 1503 static void 1504 handle_alias_pairs (void) 1505 { 1506 alias_pair *p; 1507 unsigned i; 1508 1509 for (i = 0; alias_pairs && alias_pairs->iterate (i, &p);) 1510 { 1511 symtab_node *target_node = symtab_node::get_for_asmname (p->target); 1512 1513 /* Weakrefs with target not defined in current unit are easy to handle: 1514 they behave just as external variables except we need to note the 1515 alias flag to later output the weakref pseudo op into asm file. */ 1516 if (!target_node 1517 && lookup_attribute ("weakref", DECL_ATTRIBUTES (p->decl)) != NULL) 1518 { 1519 symtab_node *node = symtab_node::get (p->decl); 1520 if (node) 1521 { 1522 node->alias_target = p->target; 1523 node->weakref = true; 1524 node->alias = true; 1525 node->transparent_alias = true; 1526 } 1527 alias_pairs->unordered_remove (i); 1528 continue; 1529 } 1530 else if (!target_node) 1531 { 1532 error ("%q+D aliased to undefined symbol %qE", p->decl, p->target); 1533 symtab_node *node = symtab_node::get (p->decl); 1534 if (node) 1535 node->alias = false; 1536 alias_pairs->unordered_remove (i); 1537 continue; 1538 } 1539 1540 if (DECL_EXTERNAL (target_node->decl) 1541 /* We use local aliases for C++ thunks to force the tailcall 1542 to bind locally. This is a hack - to keep it working do 1543 the following (which is not strictly correct). */ 1544 && (TREE_CODE (target_node->decl) != FUNCTION_DECL 1545 || ! DECL_VIRTUAL_P (target_node->decl)) 1546 && ! lookup_attribute ("weakref", DECL_ATTRIBUTES (p->decl))) 1547 { 1548 error ("%q+D aliased to external symbol %qE", 1549 p->decl, p->target); 1550 } 1551 1552 if (TREE_CODE (p->decl) == FUNCTION_DECL 1553 && target_node && is_a <cgraph_node *> (target_node)) 1554 { 1555 maybe_diag_incompatible_alias (p->decl, target_node->decl); 1556 1557 maybe_diag_alias_attributes (p->decl, target_node->decl); 1558 1559 cgraph_node *src_node = cgraph_node::get (p->decl); 1560 if (src_node && src_node->definition) 1561 src_node->reset (); 1562 cgraph_node::create_alias (p->decl, target_node->decl); 1563 alias_pairs->unordered_remove (i); 1564 } 1565 else if (VAR_P (p->decl) 1566 && target_node && is_a <varpool_node *> (target_node)) 1567 { 1568 varpool_node::create_alias (p->decl, target_node->decl); 1569 alias_pairs->unordered_remove (i); 1570 } 1571 else 1572 { 1573 error ("%q+D alias between function and variable is not supported", 1574 p->decl); 1575 inform (DECL_SOURCE_LOCATION (target_node->decl), 1576 "aliased declaration here"); 1577 1578 alias_pairs->unordered_remove (i); 1579 } 1580 } 1581 vec_free (alias_pairs); 1582 } 1583 1584 1585 /* Figure out what functions we want to assemble. */ 1586 1587 static void 1588 mark_functions_to_output (void) 1589 { 1590 bool check_same_comdat_groups = false; 1591 cgraph_node *node; 1592 1593 if (flag_checking) 1594 FOR_EACH_FUNCTION (node) 1595 gcc_assert (!node->process); 1596 1597 FOR_EACH_FUNCTION (node) 1598 { 1599 tree decl = node->decl; 1600 1601 gcc_assert (!node->process || node->same_comdat_group); 1602 if (node->process) 1603 continue; 1604 1605 /* We need to output all local functions that are used and not 1606 always inlined, as well as those that are reachable from 1607 outside the current compilation unit. */ 1608 if (node->analyzed 1609 && !node->thunk.thunk_p 1610 && !node->alias 1611 && !node->inlined_to 1612 && !TREE_ASM_WRITTEN (decl) 1613 && !DECL_EXTERNAL (decl)) 1614 { 1615 node->process = 1; 1616 if (node->same_comdat_group) 1617 { 1618 cgraph_node *next; 1619 for (next = dyn_cast<cgraph_node *> (node->same_comdat_group); 1620 next != node; 1621 next = dyn_cast<cgraph_node *> (next->same_comdat_group)) 1622 if (!next->thunk.thunk_p && !next->alias 1623 && !next->comdat_local_p ()) 1624 next->process = 1; 1625 } 1626 } 1627 else if (node->same_comdat_group) 1628 { 1629 if (flag_checking) 1630 check_same_comdat_groups = true; 1631 } 1632 else 1633 { 1634 /* We should've reclaimed all functions that are not needed. */ 1635 if (flag_checking 1636 && !node->inlined_to 1637 && gimple_has_body_p (decl) 1638 /* FIXME: in ltrans unit when offline copy is outside partition but inline copies 1639 are inside partition, we can end up not removing the body since we no longer 1640 have analyzed node pointing to it. */ 1641 && !node->in_other_partition 1642 && !node->alias 1643 && !node->clones 1644 && !DECL_EXTERNAL (decl)) 1645 { 1646 node->debug (); 1647 internal_error ("failed to reclaim unneeded function"); 1648 } 1649 gcc_assert (node->inlined_to 1650 || !gimple_has_body_p (decl) 1651 || node->in_other_partition 1652 || node->clones 1653 || DECL_ARTIFICIAL (decl) 1654 || DECL_EXTERNAL (decl)); 1655 1656 } 1657 1658 } 1659 if (flag_checking && check_same_comdat_groups) 1660 FOR_EACH_FUNCTION (node) 1661 if (node->same_comdat_group && !node->process) 1662 { 1663 tree decl = node->decl; 1664 if (!node->inlined_to 1665 && gimple_has_body_p (decl) 1666 /* FIXME: in an ltrans unit when the offline copy is outside a 1667 partition but inline copies are inside a partition, we can 1668 end up not removing the body since we no longer have an 1669 analyzed node pointing to it. */ 1670 && !node->in_other_partition 1671 && !node->clones 1672 && !DECL_EXTERNAL (decl)) 1673 { 1674 node->debug (); 1675 internal_error ("failed to reclaim unneeded function in same " 1676 "comdat group"); 1677 } 1678 } 1679 } 1680 1681 /* DECL is FUNCTION_DECL. Initialize datastructures so DECL is a function 1682 in lowered gimple form. IN_SSA is true if the gimple is in SSA. 1683 1684 Set current_function_decl and cfun to newly constructed empty function body. 1685 return basic block in the function body. */ 1686 1687 basic_block 1688 init_lowered_empty_function (tree decl, bool in_ssa, profile_count count) 1689 { 1690 basic_block bb; 1691 edge e; 1692 1693 current_function_decl = decl; 1694 allocate_struct_function (decl, false); 1695 gimple_register_cfg_hooks (); 1696 init_empty_tree_cfg (); 1697 init_tree_ssa (cfun); 1698 1699 if (in_ssa) 1700 { 1701 init_ssa_operands (cfun); 1702 cfun->gimple_df->in_ssa_p = true; 1703 cfun->curr_properties |= PROP_ssa; 1704 } 1705 1706 DECL_INITIAL (decl) = make_node (BLOCK); 1707 BLOCK_SUPERCONTEXT (DECL_INITIAL (decl)) = decl; 1708 1709 DECL_SAVED_TREE (decl) = error_mark_node; 1710 cfun->curr_properties |= (PROP_gimple_lcf | PROP_gimple_leh | PROP_gimple_any 1711 | PROP_cfg | PROP_loops); 1712 1713 set_loops_for_fn (cfun, ggc_cleared_alloc<loops> ()); 1714 init_loops_structure (cfun, loops_for_fn (cfun), 1); 1715 loops_for_fn (cfun)->state |= LOOPS_MAY_HAVE_MULTIPLE_LATCHES; 1716 1717 /* Create BB for body of the function and connect it properly. */ 1718 ENTRY_BLOCK_PTR_FOR_FN (cfun)->count = count; 1719 EXIT_BLOCK_PTR_FOR_FN (cfun)->count = count; 1720 bb = create_basic_block (NULL, ENTRY_BLOCK_PTR_FOR_FN (cfun)); 1721 bb->count = count; 1722 e = make_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun), bb, EDGE_FALLTHRU); 1723 e->probability = profile_probability::always (); 1724 e = make_edge (bb, EXIT_BLOCK_PTR_FOR_FN (cfun), 0); 1725 e->probability = profile_probability::always (); 1726 add_bb_to_loop (bb, ENTRY_BLOCK_PTR_FOR_FN (cfun)->loop_father); 1727 1728 return bb; 1729 } 1730 1731 /* Adjust PTR by the constant FIXED_OFFSET, by the vtable offset indicated by 1732 VIRTUAL_OFFSET, and by the indirect offset indicated by INDIRECT_OFFSET, if 1733 it is non-null. THIS_ADJUSTING is nonzero for a this adjusting thunk and zero 1734 for a result adjusting thunk. */ 1735 1736 tree 1737 thunk_adjust (gimple_stmt_iterator * bsi, 1738 tree ptr, bool this_adjusting, 1739 HOST_WIDE_INT fixed_offset, tree virtual_offset, 1740 HOST_WIDE_INT indirect_offset) 1741 { 1742 gassign *stmt; 1743 tree ret; 1744 1745 if (this_adjusting 1746 && fixed_offset != 0) 1747 { 1748 stmt = gimple_build_assign 1749 (ptr, fold_build_pointer_plus_hwi_loc (input_location, 1750 ptr, 1751 fixed_offset)); 1752 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1753 } 1754 1755 if (!vtable_entry_type && (virtual_offset || indirect_offset != 0)) 1756 { 1757 tree vfunc_type = make_node (FUNCTION_TYPE); 1758 TREE_TYPE (vfunc_type) = integer_type_node; 1759 TYPE_ARG_TYPES (vfunc_type) = NULL_TREE; 1760 layout_type (vfunc_type); 1761 1762 vtable_entry_type = build_pointer_type (vfunc_type); 1763 } 1764 1765 /* If there's a virtual offset, look up that value in the vtable and 1766 adjust the pointer again. */ 1767 if (virtual_offset) 1768 { 1769 tree vtabletmp; 1770 tree vtabletmp2; 1771 tree vtabletmp3; 1772 1773 vtabletmp = 1774 create_tmp_reg (build_pointer_type 1775 (build_pointer_type (vtable_entry_type)), "vptr"); 1776 1777 /* The vptr is always at offset zero in the object. */ 1778 stmt = gimple_build_assign (vtabletmp, 1779 build1 (NOP_EXPR, TREE_TYPE (vtabletmp), 1780 ptr)); 1781 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1782 1783 /* Form the vtable address. */ 1784 vtabletmp2 = create_tmp_reg (TREE_TYPE (TREE_TYPE (vtabletmp)), 1785 "vtableaddr"); 1786 stmt = gimple_build_assign (vtabletmp2, 1787 build_simple_mem_ref (vtabletmp)); 1788 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1789 1790 /* Find the entry with the vcall offset. */ 1791 stmt = gimple_build_assign (vtabletmp2, 1792 fold_build_pointer_plus_loc (input_location, 1793 vtabletmp2, 1794 virtual_offset)); 1795 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1796 1797 /* Get the offset itself. */ 1798 vtabletmp3 = create_tmp_reg (TREE_TYPE (TREE_TYPE (vtabletmp2)), 1799 "vcalloffset"); 1800 stmt = gimple_build_assign (vtabletmp3, 1801 build_simple_mem_ref (vtabletmp2)); 1802 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1803 1804 /* Adjust the `this' pointer. */ 1805 ptr = fold_build_pointer_plus_loc (input_location, ptr, vtabletmp3); 1806 ptr = force_gimple_operand_gsi (bsi, ptr, true, NULL_TREE, false, 1807 GSI_CONTINUE_LINKING); 1808 } 1809 1810 /* Likewise for an offset that is stored in the object that contains the 1811 vtable. */ 1812 if (indirect_offset != 0) 1813 { 1814 tree offset_ptr, offset_tree; 1815 1816 /* Get the address of the offset. */ 1817 offset_ptr 1818 = create_tmp_reg (build_pointer_type 1819 (build_pointer_type (vtable_entry_type)), 1820 "offset_ptr"); 1821 stmt = gimple_build_assign (offset_ptr, 1822 build1 (NOP_EXPR, TREE_TYPE (offset_ptr), 1823 ptr)); 1824 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1825 1826 stmt = gimple_build_assign 1827 (offset_ptr, 1828 fold_build_pointer_plus_hwi_loc (input_location, offset_ptr, 1829 indirect_offset)); 1830 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1831 1832 /* Get the offset itself. */ 1833 offset_tree = create_tmp_reg (TREE_TYPE (TREE_TYPE (offset_ptr)), 1834 "offset"); 1835 stmt = gimple_build_assign (offset_tree, 1836 build_simple_mem_ref (offset_ptr)); 1837 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1838 1839 /* Adjust the `this' pointer. */ 1840 ptr = fold_build_pointer_plus_loc (input_location, ptr, offset_tree); 1841 ptr = force_gimple_operand_gsi (bsi, ptr, true, NULL_TREE, false, 1842 GSI_CONTINUE_LINKING); 1843 } 1844 1845 if (!this_adjusting 1846 && fixed_offset != 0) 1847 /* Adjust the pointer by the constant. */ 1848 { 1849 tree ptrtmp; 1850 1851 if (VAR_P (ptr)) 1852 ptrtmp = ptr; 1853 else 1854 { 1855 ptrtmp = create_tmp_reg (TREE_TYPE (ptr), "ptr"); 1856 stmt = gimple_build_assign (ptrtmp, ptr); 1857 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1858 } 1859 ptr = fold_build_pointer_plus_hwi_loc (input_location, 1860 ptrtmp, fixed_offset); 1861 } 1862 1863 /* Emit the statement and gimplify the adjustment expression. */ 1864 ret = create_tmp_reg (TREE_TYPE (ptr), "adjusted_this"); 1865 stmt = gimple_build_assign (ret, ptr); 1866 gsi_insert_after (bsi, stmt, GSI_NEW_STMT); 1867 1868 return ret; 1869 } 1870 1871 /* Expand thunk NODE to gimple if possible. 1872 When FORCE_GIMPLE_THUNK is true, gimple thunk is created and 1873 no assembler is produced. 1874 When OUTPUT_ASM_THUNK is true, also produce assembler for 1875 thunks that are not lowered. */ 1876 1877 bool 1878 cgraph_node::expand_thunk (bool output_asm_thunks, bool force_gimple_thunk) 1879 { 1880 bool this_adjusting = thunk.this_adjusting; 1881 HOST_WIDE_INT fixed_offset = thunk.fixed_offset; 1882 HOST_WIDE_INT virtual_value = thunk.virtual_value; 1883 HOST_WIDE_INT indirect_offset = thunk.indirect_offset; 1884 tree virtual_offset = NULL; 1885 tree alias = callees->callee->decl; 1886 tree thunk_fndecl = decl; 1887 tree a; 1888 1889 if (!force_gimple_thunk 1890 && this_adjusting 1891 && indirect_offset == 0 1892 && !DECL_EXTERNAL (alias) 1893 && !DECL_STATIC_CHAIN (alias) 1894 && targetm.asm_out.can_output_mi_thunk (thunk_fndecl, fixed_offset, 1895 virtual_value, alias)) 1896 { 1897 tree fn_block; 1898 tree restype = TREE_TYPE (TREE_TYPE (thunk_fndecl)); 1899 1900 if (!output_asm_thunks) 1901 { 1902 analyzed = true; 1903 return false; 1904 } 1905 1906 if (in_lto_p) 1907 get_untransformed_body (); 1908 a = DECL_ARGUMENTS (thunk_fndecl); 1909 1910 current_function_decl = thunk_fndecl; 1911 1912 /* Ensure thunks are emitted in their correct sections. */ 1913 resolve_unique_section (thunk_fndecl, 0, 1914 flag_function_sections); 1915 1916 DECL_RESULT (thunk_fndecl) 1917 = build_decl (DECL_SOURCE_LOCATION (thunk_fndecl), 1918 RESULT_DECL, 0, restype); 1919 DECL_CONTEXT (DECL_RESULT (thunk_fndecl)) = thunk_fndecl; 1920 1921 /* The back end expects DECL_INITIAL to contain a BLOCK, so we 1922 create one. */ 1923 fn_block = make_node (BLOCK); 1924 BLOCK_VARS (fn_block) = a; 1925 DECL_INITIAL (thunk_fndecl) = fn_block; 1926 BLOCK_SUPERCONTEXT (fn_block) = thunk_fndecl; 1927 allocate_struct_function (thunk_fndecl, false); 1928 init_function_start (thunk_fndecl); 1929 cfun->is_thunk = 1; 1930 insn_locations_init (); 1931 set_curr_insn_location (DECL_SOURCE_LOCATION (thunk_fndecl)); 1932 prologue_location = curr_insn_location (); 1933 1934 targetm.asm_out.output_mi_thunk (asm_out_file, thunk_fndecl, 1935 fixed_offset, virtual_value, alias); 1936 1937 insn_locations_finalize (); 1938 init_insn_lengths (); 1939 free_after_compilation (cfun); 1940 TREE_ASM_WRITTEN (thunk_fndecl) = 1; 1941 thunk.thunk_p = false; 1942 analyzed = false; 1943 } 1944 else if (stdarg_p (TREE_TYPE (thunk_fndecl))) 1945 { 1946 error ("generic thunk code fails for method %qD which uses %<...%>", 1947 thunk_fndecl); 1948 TREE_ASM_WRITTEN (thunk_fndecl) = 1; 1949 analyzed = true; 1950 return false; 1951 } 1952 else 1953 { 1954 tree restype; 1955 basic_block bb, then_bb, else_bb, return_bb; 1956 gimple_stmt_iterator bsi; 1957 int nargs = 0; 1958 tree arg; 1959 int i; 1960 tree resdecl; 1961 tree restmp = NULL; 1962 1963 gcall *call; 1964 greturn *ret; 1965 bool alias_is_noreturn = TREE_THIS_VOLATILE (alias); 1966 1967 /* We may be called from expand_thunk that releases body except for 1968 DECL_ARGUMENTS. In this case force_gimple_thunk is true. */ 1969 if (in_lto_p && !force_gimple_thunk) 1970 get_untransformed_body (); 1971 1972 /* We need to force DECL_IGNORED_P when the thunk is created 1973 after early debug was run. */ 1974 if (force_gimple_thunk) 1975 DECL_IGNORED_P (thunk_fndecl) = 1; 1976 1977 a = DECL_ARGUMENTS (thunk_fndecl); 1978 1979 current_function_decl = thunk_fndecl; 1980 1981 /* Ensure thunks are emitted in their correct sections. */ 1982 resolve_unique_section (thunk_fndecl, 0, 1983 flag_function_sections); 1984 1985 bitmap_obstack_initialize (NULL); 1986 1987 if (thunk.virtual_offset_p) 1988 virtual_offset = size_int (virtual_value); 1989 1990 /* Build the return declaration for the function. */ 1991 restype = TREE_TYPE (TREE_TYPE (thunk_fndecl)); 1992 if (DECL_RESULT (thunk_fndecl) == NULL_TREE) 1993 { 1994 resdecl = build_decl (input_location, RESULT_DECL, 0, restype); 1995 DECL_ARTIFICIAL (resdecl) = 1; 1996 DECL_IGNORED_P (resdecl) = 1; 1997 DECL_CONTEXT (resdecl) = thunk_fndecl; 1998 DECL_RESULT (thunk_fndecl) = resdecl; 1999 } 2000 else 2001 resdecl = DECL_RESULT (thunk_fndecl); 2002 2003 profile_count cfg_count = count; 2004 if (!cfg_count.initialized_p ()) 2005 cfg_count = profile_count::from_gcov_type (BB_FREQ_MAX).guessed_local (); 2006 2007 bb = then_bb = else_bb = return_bb 2008 = init_lowered_empty_function (thunk_fndecl, true, cfg_count); 2009 2010 bsi = gsi_start_bb (bb); 2011 2012 /* Build call to the function being thunked. */ 2013 if (!VOID_TYPE_P (restype) 2014 && (!alias_is_noreturn 2015 || TREE_ADDRESSABLE (restype) 2016 || TREE_CODE (TYPE_SIZE_UNIT (restype)) != INTEGER_CST)) 2017 { 2018 if (DECL_BY_REFERENCE (resdecl)) 2019 { 2020 restmp = gimple_fold_indirect_ref (resdecl); 2021 if (!restmp) 2022 restmp = build2 (MEM_REF, 2023 TREE_TYPE (TREE_TYPE (resdecl)), 2024 resdecl, 2025 build_int_cst (TREE_TYPE (resdecl), 0)); 2026 } 2027 else if (!is_gimple_reg_type (restype)) 2028 { 2029 if (aggregate_value_p (resdecl, TREE_TYPE (thunk_fndecl))) 2030 { 2031 restmp = resdecl; 2032 2033 if (VAR_P (restmp)) 2034 { 2035 add_local_decl (cfun, restmp); 2036 BLOCK_VARS (DECL_INITIAL (current_function_decl)) 2037 = restmp; 2038 } 2039 } 2040 else 2041 restmp = create_tmp_var (restype, "retval"); 2042 } 2043 else 2044 restmp = create_tmp_reg (restype, "retval"); 2045 } 2046 2047 for (arg = a; arg; arg = DECL_CHAIN (arg)) 2048 nargs++; 2049 auto_vec<tree> vargs (nargs); 2050 i = 0; 2051 arg = a; 2052 if (this_adjusting) 2053 { 2054 vargs.quick_push (thunk_adjust (&bsi, a, 1, fixed_offset, 2055 virtual_offset, indirect_offset)); 2056 arg = DECL_CHAIN (a); 2057 i = 1; 2058 } 2059 2060 if (nargs) 2061 for (; i < nargs; i++, arg = DECL_CHAIN (arg)) 2062 { 2063 tree tmp = arg; 2064 if (VECTOR_TYPE_P (TREE_TYPE (arg)) 2065 || TREE_CODE (TREE_TYPE (arg)) == COMPLEX_TYPE) 2066 DECL_GIMPLE_REG_P (arg) = 1; 2067 2068 if (!is_gimple_val (arg)) 2069 { 2070 tmp = create_tmp_reg (TYPE_MAIN_VARIANT 2071 (TREE_TYPE (arg)), "arg"); 2072 gimple *stmt = gimple_build_assign (tmp, arg); 2073 gsi_insert_after (&bsi, stmt, GSI_NEW_STMT); 2074 } 2075 vargs.quick_push (tmp); 2076 } 2077 call = gimple_build_call_vec (build_fold_addr_expr_loc (0, alias), vargs); 2078 callees->call_stmt = call; 2079 gimple_call_set_from_thunk (call, true); 2080 if (DECL_STATIC_CHAIN (alias)) 2081 { 2082 tree p = DECL_STRUCT_FUNCTION (alias)->static_chain_decl; 2083 tree type = TREE_TYPE (p); 2084 tree decl = build_decl (DECL_SOURCE_LOCATION (thunk_fndecl), 2085 PARM_DECL, create_tmp_var_name ("CHAIN"), 2086 type); 2087 DECL_ARTIFICIAL (decl) = 1; 2088 DECL_IGNORED_P (decl) = 1; 2089 TREE_USED (decl) = 1; 2090 DECL_CONTEXT (decl) = thunk_fndecl; 2091 DECL_ARG_TYPE (decl) = type; 2092 TREE_READONLY (decl) = 1; 2093 2094 struct function *sf = DECL_STRUCT_FUNCTION (thunk_fndecl); 2095 sf->static_chain_decl = decl; 2096 2097 gimple_call_set_chain (call, decl); 2098 } 2099 2100 /* Return slot optimization is always possible and in fact required to 2101 return values with DECL_BY_REFERENCE. */ 2102 if (aggregate_value_p (resdecl, TREE_TYPE (thunk_fndecl)) 2103 && (!is_gimple_reg_type (TREE_TYPE (resdecl)) 2104 || DECL_BY_REFERENCE (resdecl))) 2105 gimple_call_set_return_slot_opt (call, true); 2106 2107 if (restmp) 2108 { 2109 gimple_call_set_lhs (call, restmp); 2110 gcc_assert (useless_type_conversion_p (TREE_TYPE (restmp), 2111 TREE_TYPE (TREE_TYPE (alias)))); 2112 } 2113 gsi_insert_after (&bsi, call, GSI_NEW_STMT); 2114 if (!alias_is_noreturn) 2115 { 2116 if (restmp && !this_adjusting 2117 && (fixed_offset || virtual_offset)) 2118 { 2119 tree true_label = NULL_TREE; 2120 2121 if (TREE_CODE (TREE_TYPE (restmp)) == POINTER_TYPE) 2122 { 2123 gimple *stmt; 2124 edge e; 2125 /* If the return type is a pointer, we need to 2126 protect against NULL. We know there will be an 2127 adjustment, because that's why we're emitting a 2128 thunk. */ 2129 then_bb = create_basic_block (NULL, bb); 2130 then_bb->count = cfg_count - cfg_count.apply_scale (1, 16); 2131 return_bb = create_basic_block (NULL, then_bb); 2132 return_bb->count = cfg_count; 2133 else_bb = create_basic_block (NULL, else_bb); 2134 else_bb->count = cfg_count.apply_scale (1, 16); 2135 add_bb_to_loop (then_bb, bb->loop_father); 2136 add_bb_to_loop (return_bb, bb->loop_father); 2137 add_bb_to_loop (else_bb, bb->loop_father); 2138 remove_edge (single_succ_edge (bb)); 2139 true_label = gimple_block_label (then_bb); 2140 stmt = gimple_build_cond (NE_EXPR, restmp, 2141 build_zero_cst (TREE_TYPE (restmp)), 2142 NULL_TREE, NULL_TREE); 2143 gsi_insert_after (&bsi, stmt, GSI_NEW_STMT); 2144 e = make_edge (bb, then_bb, EDGE_TRUE_VALUE); 2145 e->probability = profile_probability::guessed_always () 2146 .apply_scale (1, 16); 2147 e = make_edge (bb, else_bb, EDGE_FALSE_VALUE); 2148 e->probability = profile_probability::guessed_always () 2149 .apply_scale (1, 16); 2150 make_single_succ_edge (return_bb, 2151 EXIT_BLOCK_PTR_FOR_FN (cfun), 0); 2152 make_single_succ_edge (then_bb, return_bb, EDGE_FALLTHRU); 2153 e = make_edge (else_bb, return_bb, EDGE_FALLTHRU); 2154 e->probability = profile_probability::always (); 2155 bsi = gsi_last_bb (then_bb); 2156 } 2157 2158 restmp = thunk_adjust (&bsi, restmp, /*this_adjusting=*/0, 2159 fixed_offset, virtual_offset, 2160 indirect_offset); 2161 if (true_label) 2162 { 2163 gimple *stmt; 2164 bsi = gsi_last_bb (else_bb); 2165 stmt = gimple_build_assign (restmp, 2166 build_zero_cst (TREE_TYPE (restmp))); 2167 gsi_insert_after (&bsi, stmt, GSI_NEW_STMT); 2168 bsi = gsi_last_bb (return_bb); 2169 } 2170 } 2171 else 2172 gimple_call_set_tail (call, true); 2173 2174 /* Build return value. */ 2175 if (!DECL_BY_REFERENCE (resdecl)) 2176 ret = gimple_build_return (restmp); 2177 else 2178 ret = gimple_build_return (resdecl); 2179 2180 gsi_insert_after (&bsi, ret, GSI_NEW_STMT); 2181 } 2182 else 2183 { 2184 gimple_call_set_tail (call, true); 2185 remove_edge (single_succ_edge (bb)); 2186 } 2187 2188 cfun->gimple_df->in_ssa_p = true; 2189 update_max_bb_count (); 2190 profile_status_for_fn (cfun) 2191 = cfg_count.initialized_p () && cfg_count.ipa_p () 2192 ? PROFILE_READ : PROFILE_GUESSED; 2193 /* FIXME: C++ FE should stop setting TREE_ASM_WRITTEN on thunks. */ 2194 TREE_ASM_WRITTEN (thunk_fndecl) = false; 2195 delete_unreachable_blocks (); 2196 update_ssa (TODO_update_ssa); 2197 checking_verify_flow_info (); 2198 free_dominance_info (CDI_DOMINATORS); 2199 2200 /* Since we want to emit the thunk, we explicitly mark its name as 2201 referenced. */ 2202 thunk.thunk_p = false; 2203 lowered = true; 2204 bitmap_obstack_release (NULL); 2205 } 2206 current_function_decl = NULL; 2207 set_cfun (NULL); 2208 return true; 2209 } 2210 2211 /* Assemble thunks and aliases associated to node. */ 2212 2213 void 2214 cgraph_node::assemble_thunks_and_aliases (void) 2215 { 2216 cgraph_edge *e; 2217 ipa_ref *ref; 2218 2219 for (e = callers; e;) 2220 if (e->caller->thunk.thunk_p 2221 && !e->caller->inlined_to) 2222 { 2223 cgraph_node *thunk = e->caller; 2224 2225 e = e->next_caller; 2226 thunk->expand_thunk (!rtl_dump_and_exit, false); 2227 thunk->assemble_thunks_and_aliases (); 2228 } 2229 else 2230 e = e->next_caller; 2231 2232 FOR_EACH_ALIAS (this, ref) 2233 { 2234 cgraph_node *alias = dyn_cast <cgraph_node *> (ref->referring); 2235 if (!alias->transparent_alias) 2236 { 2237 bool saved_written = TREE_ASM_WRITTEN (decl); 2238 2239 /* Force assemble_alias to really output the alias this time instead 2240 of buffering it in same alias pairs. */ 2241 TREE_ASM_WRITTEN (decl) = 1; 2242 if (alias->symver) 2243 do_assemble_symver (alias->decl, 2244 DECL_ASSEMBLER_NAME (decl)); 2245 else 2246 do_assemble_alias (alias->decl, 2247 DECL_ASSEMBLER_NAME (decl)); 2248 alias->assemble_thunks_and_aliases (); 2249 TREE_ASM_WRITTEN (decl) = saved_written; 2250 } 2251 } 2252 } 2253 2254 /* Expand function specified by node. */ 2255 2256 void 2257 cgraph_node::expand (void) 2258 { 2259 location_t saved_loc; 2260 2261 /* We ought to not compile any inline clones. */ 2262 gcc_assert (!inlined_to); 2263 2264 /* __RTL functions are compiled as soon as they are parsed, so don't 2265 do it again. */ 2266 if (native_rtl_p ()) 2267 return; 2268 2269 announce_function (decl); 2270 process = 0; 2271 gcc_assert (lowered); 2272 get_untransformed_body (); 2273 2274 /* Generate RTL for the body of DECL. */ 2275 2276 timevar_push (TV_REST_OF_COMPILATION); 2277 2278 gcc_assert (symtab->global_info_ready); 2279 2280 /* Initialize the default bitmap obstack. */ 2281 bitmap_obstack_initialize (NULL); 2282 2283 /* Initialize the RTL code for the function. */ 2284 saved_loc = input_location; 2285 input_location = DECL_SOURCE_LOCATION (decl); 2286 2287 gcc_assert (DECL_STRUCT_FUNCTION (decl)); 2288 push_cfun (DECL_STRUCT_FUNCTION (decl)); 2289 init_function_start (decl); 2290 2291 gimple_register_cfg_hooks (); 2292 2293 bitmap_obstack_initialize (®_obstack); /* FIXME, only at RTL generation*/ 2294 2295 update_ssa (TODO_update_ssa_only_virtuals); 2296 execute_all_ipa_transforms (false); 2297 2298 /* Perform all tree transforms and optimizations. */ 2299 2300 /* Signal the start of passes. */ 2301 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_START, NULL); 2302 2303 execute_pass_list (cfun, g->get_passes ()->all_passes); 2304 2305 /* Signal the end of passes. */ 2306 invoke_plugin_callbacks (PLUGIN_ALL_PASSES_END, NULL); 2307 2308 bitmap_obstack_release (®_obstack); 2309 2310 /* Release the default bitmap obstack. */ 2311 bitmap_obstack_release (NULL); 2312 2313 /* If requested, warn about function definitions where the function will 2314 return a value (usually of some struct or union type) which itself will 2315 take up a lot of stack space. */ 2316 if (!DECL_EXTERNAL (decl) && TREE_TYPE (decl)) 2317 { 2318 tree ret_type = TREE_TYPE (TREE_TYPE (decl)); 2319 2320 if (ret_type && TYPE_SIZE_UNIT (ret_type) 2321 && TREE_CODE (TYPE_SIZE_UNIT (ret_type)) == INTEGER_CST 2322 && compare_tree_int (TYPE_SIZE_UNIT (ret_type), 2323 warn_larger_than_size) > 0) 2324 { 2325 unsigned int size_as_int 2326 = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (ret_type)); 2327 2328 if (compare_tree_int (TYPE_SIZE_UNIT (ret_type), size_as_int) == 0) 2329 warning (OPT_Wlarger_than_, 2330 "size of return value of %q+D is %u bytes", 2331 decl, size_as_int); 2332 else 2333 warning (OPT_Wlarger_than_, 2334 "size of return value of %q+D is larger than %wu bytes", 2335 decl, warn_larger_than_size); 2336 } 2337 } 2338 2339 gimple_set_body (decl, NULL); 2340 if (DECL_STRUCT_FUNCTION (decl) == 0 2341 && !cgraph_node::get (decl)->origin) 2342 { 2343 /* Stop pointing to the local nodes about to be freed. 2344 But DECL_INITIAL must remain nonzero so we know this 2345 was an actual function definition. 2346 For a nested function, this is done in c_pop_function_context. 2347 If rest_of_compilation set this to 0, leave it 0. */ 2348 if (DECL_INITIAL (decl) != 0) 2349 DECL_INITIAL (decl) = error_mark_node; 2350 } 2351 2352 input_location = saved_loc; 2353 2354 ggc_collect (); 2355 timevar_pop (TV_REST_OF_COMPILATION); 2356 2357 /* Make sure that BE didn't give up on compiling. */ 2358 gcc_assert (TREE_ASM_WRITTEN (decl)); 2359 if (cfun) 2360 pop_cfun (); 2361 2362 /* It would make a lot more sense to output thunks before function body to 2363 get more forward and fewer backward jumps. This however would need 2364 solving problem with comdats. See PR48668. Also aliases must come after 2365 function itself to make one pass assemblers, like one on AIX, happy. 2366 See PR 50689. 2367 FIXME: Perhaps thunks should be move before function IFF they are not in 2368 comdat groups. */ 2369 assemble_thunks_and_aliases (); 2370 release_body (); 2371 /* Eliminate all call edges. This is important so the GIMPLE_CALL no longer 2372 points to the dead function body. */ 2373 remove_callees (); 2374 remove_all_references (); 2375 } 2376 2377 /* Node comparator that is responsible for the order that corresponds 2378 to time when a function was launched for the first time. */ 2379 2380 int 2381 tp_first_run_node_cmp (const void *pa, const void *pb) 2382 { 2383 const cgraph_node *a = *(const cgraph_node * const *) pa; 2384 const cgraph_node *b = *(const cgraph_node * const *) pb; 2385 unsigned int tp_first_run_a = a->tp_first_run; 2386 unsigned int tp_first_run_b = b->tp_first_run; 2387 2388 if (!opt_for_fn (a->decl, flag_profile_reorder_functions) 2389 || a->no_reorder) 2390 tp_first_run_a = 0; 2391 if (!opt_for_fn (b->decl, flag_profile_reorder_functions) 2392 || b->no_reorder) 2393 tp_first_run_b = 0; 2394 2395 if (tp_first_run_a == tp_first_run_b) 2396 return a->order - b->order; 2397 2398 /* Functions with time profile must be before these without profile. */ 2399 tp_first_run_a = (tp_first_run_a - 1) & INT_MAX; 2400 tp_first_run_b = (tp_first_run_b - 1) & INT_MAX; 2401 2402 return tp_first_run_a - tp_first_run_b; 2403 } 2404 2405 /* Expand all functions that must be output. 2406 2407 Attempt to topologically sort the nodes so function is output when 2408 all called functions are already assembled to allow data to be 2409 propagated across the callgraph. Use a stack to get smaller distance 2410 between a function and its callees (later we may choose to use a more 2411 sophisticated algorithm for function reordering; we will likely want 2412 to use subsections to make the output functions appear in top-down 2413 order). */ 2414 2415 static void 2416 expand_all_functions (void) 2417 { 2418 cgraph_node *node; 2419 cgraph_node **order = XCNEWVEC (cgraph_node *, 2420 symtab->cgraph_count); 2421 cgraph_node **tp_first_run_order = XCNEWVEC (cgraph_node *, 2422 symtab->cgraph_count); 2423 unsigned int expanded_func_count = 0, profiled_func_count = 0; 2424 int order_pos, tp_first_run_order_pos = 0, new_order_pos = 0; 2425 int i; 2426 2427 order_pos = ipa_reverse_postorder (order); 2428 gcc_assert (order_pos == symtab->cgraph_count); 2429 2430 /* Garbage collector may remove inline clones we eliminate during 2431 optimization. So we must be sure to not reference them. */ 2432 for (i = 0; i < order_pos; i++) 2433 if (order[i]->process) 2434 { 2435 if (order[i]->tp_first_run 2436 && opt_for_fn (order[i]->decl, flag_profile_reorder_functions)) 2437 tp_first_run_order[tp_first_run_order_pos++] = order[i]; 2438 else 2439 order[new_order_pos++] = order[i]; 2440 } 2441 2442 /* First output functions with time profile in specified order. */ 2443 qsort (tp_first_run_order, tp_first_run_order_pos, 2444 sizeof (cgraph_node *), tp_first_run_node_cmp); 2445 for (i = 0; i < tp_first_run_order_pos; i++) 2446 { 2447 node = tp_first_run_order[i]; 2448 2449 if (node->process) 2450 { 2451 expanded_func_count++; 2452 profiled_func_count++; 2453 2454 if (symtab->dump_file) 2455 fprintf (symtab->dump_file, 2456 "Time profile order in expand_all_functions:%s:%d\n", 2457 node->dump_asm_name (), node->tp_first_run); 2458 node->process = 0; 2459 node->expand (); 2460 } 2461 } 2462 2463 /* Output functions in RPO so callees get optimized before callers. This 2464 makes ipa-ra and other propagators to work. 2465 FIXME: This is far from optimal code layout. */ 2466 for (i = new_order_pos - 1; i >= 0; i--) 2467 { 2468 node = order[i]; 2469 2470 if (node->process) 2471 { 2472 expanded_func_count++; 2473 node->process = 0; 2474 node->expand (); 2475 } 2476 } 2477 2478 if (dump_file) 2479 fprintf (dump_file, "Expanded functions with time profile (%s):%u/%u\n", 2480 main_input_filename, profiled_func_count, expanded_func_count); 2481 2482 if (symtab->dump_file && tp_first_run_order_pos) 2483 fprintf (symtab->dump_file, "Expanded functions with time profile:%u/%u\n", 2484 profiled_func_count, expanded_func_count); 2485 2486 symtab->process_new_functions (); 2487 free_gimplify_stack (); 2488 delete ipa_saved_clone_sources; 2489 ipa_saved_clone_sources = NULL; 2490 free (order); 2491 } 2492 2493 /* This is used to sort the node types by the cgraph order number. */ 2494 2495 enum cgraph_order_sort_kind 2496 { 2497 ORDER_FUNCTION, 2498 ORDER_VAR, 2499 ORDER_VAR_UNDEF, 2500 ORDER_ASM 2501 }; 2502 2503 struct cgraph_order_sort 2504 { 2505 /* Construct from a cgraph_node. */ 2506 cgraph_order_sort (cgraph_node *node) 2507 : kind (ORDER_FUNCTION), order (node->order) 2508 { 2509 u.f = node; 2510 } 2511 2512 /* Construct from a varpool_node. */ 2513 cgraph_order_sort (varpool_node *node) 2514 : kind (node->definition ? ORDER_VAR : ORDER_VAR_UNDEF), order (node->order) 2515 { 2516 u.v = node; 2517 } 2518 2519 /* Construct from a asm_node. */ 2520 cgraph_order_sort (asm_node *node) 2521 : kind (ORDER_ASM), order (node->order) 2522 { 2523 u.a = node; 2524 } 2525 2526 /* Assembly cgraph_order_sort based on its type. */ 2527 void process (); 2528 2529 enum cgraph_order_sort_kind kind; 2530 union 2531 { 2532 cgraph_node *f; 2533 varpool_node *v; 2534 asm_node *a; 2535 } u; 2536 int order; 2537 }; 2538 2539 /* Assembly cgraph_order_sort based on its type. */ 2540 2541 void 2542 cgraph_order_sort::process () 2543 { 2544 switch (kind) 2545 { 2546 case ORDER_FUNCTION: 2547 u.f->process = 0; 2548 u.f->expand (); 2549 break; 2550 case ORDER_VAR: 2551 u.v->assemble_decl (); 2552 break; 2553 case ORDER_VAR_UNDEF: 2554 assemble_undefined_decl (u.v->decl); 2555 break; 2556 case ORDER_ASM: 2557 assemble_asm (u.a->asm_str); 2558 break; 2559 default: 2560 gcc_unreachable (); 2561 } 2562 } 2563 2564 /* Compare cgraph_order_sort by order. */ 2565 2566 static int 2567 cgraph_order_cmp (const void *a_p, const void *b_p) 2568 { 2569 const cgraph_order_sort *nodea = (const cgraph_order_sort *)a_p; 2570 const cgraph_order_sort *nodeb = (const cgraph_order_sort *)b_p; 2571 2572 return nodea->order - nodeb->order; 2573 } 2574 2575 /* Output all functions, variables, and asm statements in the order 2576 according to their order fields, which is the order in which they 2577 appeared in the file. This implements -fno-toplevel-reorder. In 2578 this mode we may output functions and variables which don't really 2579 need to be output. */ 2580 2581 static void 2582 output_in_order (void) 2583 { 2584 int i; 2585 cgraph_node *cnode; 2586 varpool_node *vnode; 2587 asm_node *anode; 2588 auto_vec<cgraph_order_sort> nodes; 2589 cgraph_order_sort *node; 2590 2591 FOR_EACH_DEFINED_FUNCTION (cnode) 2592 if (cnode->process && !cnode->thunk.thunk_p 2593 && !cnode->alias && cnode->no_reorder) 2594 nodes.safe_push (cgraph_order_sort (cnode)); 2595 2596 /* There is a similar loop in symbol_table::output_variables. 2597 Please keep them in sync. */ 2598 FOR_EACH_VARIABLE (vnode) 2599 if (vnode->no_reorder 2600 && !DECL_HARD_REGISTER (vnode->decl) 2601 && !DECL_HAS_VALUE_EXPR_P (vnode->decl)) 2602 nodes.safe_push (cgraph_order_sort (vnode)); 2603 2604 for (anode = symtab->first_asm_symbol (); anode; anode = anode->next) 2605 nodes.safe_push (cgraph_order_sort (anode)); 2606 2607 /* Sort nodes by order. */ 2608 nodes.qsort (cgraph_order_cmp); 2609 2610 /* In toplevel reorder mode we output all statics; mark them as needed. */ 2611 FOR_EACH_VEC_ELT (nodes, i, node) 2612 if (node->kind == ORDER_VAR) 2613 node->u.v->finalize_named_section_flags (); 2614 2615 FOR_EACH_VEC_ELT (nodes, i, node) 2616 node->process (); 2617 2618 symtab->clear_asm_symbols (); 2619 } 2620 2621 static void 2622 ipa_passes (void) 2623 { 2624 gcc::pass_manager *passes = g->get_passes (); 2625 2626 set_cfun (NULL); 2627 current_function_decl = NULL; 2628 gimple_register_cfg_hooks (); 2629 bitmap_obstack_initialize (NULL); 2630 2631 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_START, NULL); 2632 2633 if (!in_lto_p) 2634 { 2635 execute_ipa_pass_list (passes->all_small_ipa_passes); 2636 if (seen_error ()) 2637 return; 2638 } 2639 2640 /* This extra symtab_remove_unreachable_nodes pass tends to catch some 2641 devirtualization and other changes where removal iterate. */ 2642 symtab->remove_unreachable_nodes (symtab->dump_file); 2643 2644 /* If pass_all_early_optimizations was not scheduled, the state of 2645 the cgraph will not be properly updated. Update it now. */ 2646 if (symtab->state < IPA_SSA) 2647 symtab->state = IPA_SSA; 2648 2649 if (!in_lto_p) 2650 { 2651 /* Generate coverage variables and constructors. */ 2652 coverage_finish (); 2653 2654 /* Process new functions added. */ 2655 set_cfun (NULL); 2656 current_function_decl = NULL; 2657 symtab->process_new_functions (); 2658 2659 execute_ipa_summary_passes 2660 ((ipa_opt_pass_d *) passes->all_regular_ipa_passes); 2661 } 2662 2663 /* Some targets need to handle LTO assembler output specially. */ 2664 if (flag_generate_lto || flag_generate_offload) 2665 targetm.asm_out.lto_start (); 2666 2667 if (!in_lto_p 2668 || flag_incremental_link == INCREMENTAL_LINK_LTO) 2669 { 2670 if (!quiet_flag) 2671 fprintf (stderr, "Streaming LTO\n"); 2672 if (g->have_offload) 2673 { 2674 section_name_prefix = OFFLOAD_SECTION_NAME_PREFIX; 2675 lto_stream_offload_p = true; 2676 ipa_write_summaries (); 2677 lto_stream_offload_p = false; 2678 } 2679 if (flag_lto) 2680 { 2681 section_name_prefix = LTO_SECTION_NAME_PREFIX; 2682 lto_stream_offload_p = false; 2683 ipa_write_summaries (); 2684 } 2685 } 2686 2687 if (flag_generate_lto || flag_generate_offload) 2688 targetm.asm_out.lto_end (); 2689 2690 if (!flag_ltrans 2691 && ((in_lto_p && flag_incremental_link != INCREMENTAL_LINK_LTO) 2692 || !flag_lto || flag_fat_lto_objects)) 2693 execute_ipa_pass_list (passes->all_regular_ipa_passes); 2694 invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_END, NULL); 2695 2696 bitmap_obstack_release (NULL); 2697 } 2698 2699 2700 /* Return string alias is alias of. */ 2701 2702 static tree 2703 get_alias_symbol (tree decl) 2704 { 2705 tree alias = lookup_attribute ("alias", DECL_ATTRIBUTES (decl)); 2706 return get_identifier (TREE_STRING_POINTER 2707 (TREE_VALUE (TREE_VALUE (alias)))); 2708 } 2709 2710 2711 /* Weakrefs may be associated to external decls and thus not output 2712 at expansion time. Emit all necessary aliases. */ 2713 2714 void 2715 symbol_table::output_weakrefs (void) 2716 { 2717 symtab_node *node; 2718 FOR_EACH_SYMBOL (node) 2719 if (node->alias 2720 && !TREE_ASM_WRITTEN (node->decl) 2721 && node->weakref) 2722 { 2723 tree target; 2724 2725 /* Weakrefs are special by not requiring target definition in current 2726 compilation unit. It is thus bit hard to work out what we want to 2727 alias. 2728 When alias target is defined, we need to fetch it from symtab reference, 2729 otherwise it is pointed to by alias_target. */ 2730 if (node->alias_target) 2731 target = (DECL_P (node->alias_target) 2732 ? DECL_ASSEMBLER_NAME (node->alias_target) 2733 : node->alias_target); 2734 else if (node->analyzed) 2735 target = DECL_ASSEMBLER_NAME (node->get_alias_target ()->decl); 2736 else 2737 { 2738 gcc_unreachable (); 2739 target = get_alias_symbol (node->decl); 2740 } 2741 do_assemble_alias (node->decl, target); 2742 } 2743 } 2744 2745 /* Perform simple optimizations based on callgraph. */ 2746 2747 void 2748 symbol_table::compile (void) 2749 { 2750 if (seen_error ()) 2751 return; 2752 2753 symtab_node::checking_verify_symtab_nodes (); 2754 2755 timevar_push (TV_CGRAPHOPT); 2756 if (pre_ipa_mem_report) 2757 dump_memory_report ("Memory consumption before IPA"); 2758 if (!quiet_flag) 2759 fprintf (stderr, "Performing interprocedural optimizations\n"); 2760 state = IPA; 2761 2762 /* If LTO is enabled, initialize the streamer hooks needed by GIMPLE. */ 2763 if (flag_generate_lto || flag_generate_offload) 2764 lto_streamer_hooks_init (); 2765 2766 /* Don't run the IPA passes if there was any error or sorry messages. */ 2767 if (!seen_error ()) 2768 { 2769 timevar_start (TV_CGRAPH_IPA_PASSES); 2770 ipa_passes (); 2771 timevar_stop (TV_CGRAPH_IPA_PASSES); 2772 } 2773 /* Do nothing else if any IPA pass found errors or if we are just streaming LTO. */ 2774 if (seen_error () 2775 || ((!in_lto_p || flag_incremental_link == INCREMENTAL_LINK_LTO) 2776 && flag_lto && !flag_fat_lto_objects)) 2777 { 2778 timevar_pop (TV_CGRAPHOPT); 2779 return; 2780 } 2781 2782 global_info_ready = true; 2783 if (dump_file) 2784 { 2785 fprintf (dump_file, "Optimized "); 2786 symtab->dump (dump_file); 2787 } 2788 if (post_ipa_mem_report) 2789 dump_memory_report ("Memory consumption after IPA"); 2790 timevar_pop (TV_CGRAPHOPT); 2791 2792 /* Output everything. */ 2793 switch_to_section (text_section); 2794 (*debug_hooks->assembly_start) (); 2795 if (!quiet_flag) 2796 fprintf (stderr, "Assembling functions:\n"); 2797 symtab_node::checking_verify_symtab_nodes (); 2798 2799 bitmap_obstack_initialize (NULL); 2800 execute_ipa_pass_list (g->get_passes ()->all_late_ipa_passes); 2801 bitmap_obstack_release (NULL); 2802 mark_functions_to_output (); 2803 2804 /* When weakref support is missing, we automatically translate all 2805 references to NODE to references to its ultimate alias target. 2806 The renaming mechanism uses flag IDENTIFIER_TRANSPARENT_ALIAS and 2807 TREE_CHAIN. 2808 2809 Set up this mapping before we output any assembler but once we are sure 2810 that all symbol renaming is done. 2811 2812 FIXME: All this ugliness can go away if we just do renaming at gimple 2813 level by physically rewriting the IL. At the moment we can only redirect 2814 calls, so we need infrastructure for renaming references as well. */ 2815 #ifndef ASM_OUTPUT_WEAKREF 2816 symtab_node *node; 2817 2818 FOR_EACH_SYMBOL (node) 2819 if (node->alias 2820 && lookup_attribute ("weakref", DECL_ATTRIBUTES (node->decl))) 2821 { 2822 IDENTIFIER_TRANSPARENT_ALIAS 2823 (DECL_ASSEMBLER_NAME (node->decl)) = 1; 2824 TREE_CHAIN (DECL_ASSEMBLER_NAME (node->decl)) 2825 = (node->alias_target ? node->alias_target 2826 : DECL_ASSEMBLER_NAME (node->get_alias_target ()->decl)); 2827 } 2828 #endif 2829 2830 state = EXPANSION; 2831 2832 /* Output first asm statements and anything ordered. The process 2833 flag is cleared for these nodes, so we skip them later. */ 2834 output_in_order (); 2835 2836 timevar_start (TV_CGRAPH_FUNC_EXPANSION); 2837 expand_all_functions (); 2838 timevar_stop (TV_CGRAPH_FUNC_EXPANSION); 2839 2840 output_variables (); 2841 2842 process_new_functions (); 2843 state = FINISHED; 2844 output_weakrefs (); 2845 2846 if (dump_file) 2847 { 2848 fprintf (dump_file, "\nFinal "); 2849 symtab->dump (dump_file); 2850 } 2851 if (!flag_checking) 2852 return; 2853 symtab_node::verify_symtab_nodes (); 2854 /* Double check that all inline clones are gone and that all 2855 function bodies have been released from memory. */ 2856 if (!seen_error ()) 2857 { 2858 cgraph_node *node; 2859 bool error_found = false; 2860 2861 FOR_EACH_DEFINED_FUNCTION (node) 2862 if (node->inlined_to 2863 || gimple_has_body_p (node->decl)) 2864 { 2865 error_found = true; 2866 node->debug (); 2867 } 2868 if (error_found) 2869 internal_error ("nodes with unreleased memory found"); 2870 } 2871 } 2872 2873 /* Earlydebug dump file, flags, and number. */ 2874 2875 static int debuginfo_early_dump_nr; 2876 static FILE *debuginfo_early_dump_file; 2877 static dump_flags_t debuginfo_early_dump_flags; 2878 2879 /* Debug dump file, flags, and number. */ 2880 2881 static int debuginfo_dump_nr; 2882 static FILE *debuginfo_dump_file; 2883 static dump_flags_t debuginfo_dump_flags; 2884 2885 /* Register the debug and earlydebug dump files. */ 2886 2887 void 2888 debuginfo_early_init (void) 2889 { 2890 gcc::dump_manager *dumps = g->get_dumps (); 2891 debuginfo_early_dump_nr = dumps->dump_register (".earlydebug", "earlydebug", 2892 "earlydebug", DK_tree, 2893 OPTGROUP_NONE, 2894 false); 2895 debuginfo_dump_nr = dumps->dump_register (".debug", "debug", 2896 "debug", DK_tree, 2897 OPTGROUP_NONE, 2898 false); 2899 } 2900 2901 /* Initialize the debug and earlydebug dump files. */ 2902 2903 void 2904 debuginfo_init (void) 2905 { 2906 gcc::dump_manager *dumps = g->get_dumps (); 2907 debuginfo_dump_file = dump_begin (debuginfo_dump_nr, NULL); 2908 debuginfo_dump_flags = dumps->get_dump_file_info (debuginfo_dump_nr)->pflags; 2909 debuginfo_early_dump_file = dump_begin (debuginfo_early_dump_nr, NULL); 2910 debuginfo_early_dump_flags 2911 = dumps->get_dump_file_info (debuginfo_early_dump_nr)->pflags; 2912 } 2913 2914 /* Finalize the debug and earlydebug dump files. */ 2915 2916 void 2917 debuginfo_fini (void) 2918 { 2919 if (debuginfo_dump_file) 2920 dump_end (debuginfo_dump_nr, debuginfo_dump_file); 2921 if (debuginfo_early_dump_file) 2922 dump_end (debuginfo_early_dump_nr, debuginfo_early_dump_file); 2923 } 2924 2925 /* Set dump_file to the debug dump file. */ 2926 2927 void 2928 debuginfo_start (void) 2929 { 2930 set_dump_file (debuginfo_dump_file); 2931 } 2932 2933 /* Undo setting dump_file to the debug dump file. */ 2934 2935 void 2936 debuginfo_stop (void) 2937 { 2938 set_dump_file (NULL); 2939 } 2940 2941 /* Set dump_file to the earlydebug dump file. */ 2942 2943 void 2944 debuginfo_early_start (void) 2945 { 2946 set_dump_file (debuginfo_early_dump_file); 2947 } 2948 2949 /* Undo setting dump_file to the earlydebug dump file. */ 2950 2951 void 2952 debuginfo_early_stop (void) 2953 { 2954 set_dump_file (NULL); 2955 } 2956 2957 /* Analyze the whole compilation unit once it is parsed completely. */ 2958 2959 void 2960 symbol_table::finalize_compilation_unit (void) 2961 { 2962 timevar_push (TV_CGRAPH); 2963 2964 /* If we're here there's no current function anymore. Some frontends 2965 are lazy in clearing these. */ 2966 current_function_decl = NULL; 2967 set_cfun (NULL); 2968 2969 /* Do not skip analyzing the functions if there were errors, we 2970 miss diagnostics for following functions otherwise. */ 2971 2972 /* Emit size functions we didn't inline. */ 2973 finalize_size_functions (); 2974 2975 /* Mark alias targets necessary and emit diagnostics. */ 2976 handle_alias_pairs (); 2977 2978 if (!quiet_flag) 2979 { 2980 fprintf (stderr, "\nAnalyzing compilation unit\n"); 2981 fflush (stderr); 2982 } 2983 2984 if (flag_dump_passes) 2985 dump_passes (); 2986 2987 /* Gimplify and lower all functions, compute reachability and 2988 remove unreachable nodes. */ 2989 analyze_functions (/*first_time=*/true); 2990 2991 /* Mark alias targets necessary and emit diagnostics. */ 2992 handle_alias_pairs (); 2993 2994 /* Gimplify and lower thunks. */ 2995 analyze_functions (/*first_time=*/false); 2996 2997 /* Offloading requires LTO infrastructure. */ 2998 if (!in_lto_p && g->have_offload) 2999 flag_generate_offload = 1; 3000 3001 if (!seen_error ()) 3002 { 3003 /* Emit early debug for reachable functions, and by consequence, 3004 locally scoped symbols. */ 3005 struct cgraph_node *cnode; 3006 FOR_EACH_FUNCTION_WITH_GIMPLE_BODY (cnode) 3007 (*debug_hooks->early_global_decl) (cnode->decl); 3008 3009 /* Clean up anything that needs cleaning up after initial debug 3010 generation. */ 3011 debuginfo_early_start (); 3012 (*debug_hooks->early_finish) (main_input_filename); 3013 debuginfo_early_stop (); 3014 } 3015 3016 /* Finally drive the pass manager. */ 3017 compile (); 3018 3019 timevar_pop (TV_CGRAPH); 3020 } 3021 3022 /* Reset all state within cgraphunit.c so that we can rerun the compiler 3023 within the same process. For use by toplev::finalize. */ 3024 3025 void 3026 cgraphunit_c_finalize (void) 3027 { 3028 gcc_assert (cgraph_new_nodes.length () == 0); 3029 cgraph_new_nodes.truncate (0); 3030 3031 vtable_entry_type = NULL; 3032 queued_nodes = &symtab_terminator; 3033 3034 first_analyzed = NULL; 3035 first_analyzed_var = NULL; 3036 } 3037 3038 /* Creates a wrapper from cgraph_node to TARGET node. Thunk is used for this 3039 kind of wrapper method. */ 3040 3041 void 3042 cgraph_node::create_wrapper (cgraph_node *target) 3043 { 3044 /* Preserve DECL_RESULT so we get right by reference flag. */ 3045 tree decl_result = DECL_RESULT (decl); 3046 3047 /* Remove the function's body but keep arguments to be reused 3048 for thunk. */ 3049 release_body (true); 3050 reset (); 3051 3052 DECL_UNINLINABLE (decl) = false; 3053 DECL_RESULT (decl) = decl_result; 3054 DECL_INITIAL (decl) = NULL; 3055 allocate_struct_function (decl, false); 3056 set_cfun (NULL); 3057 3058 /* Turn alias into thunk and expand it into GIMPLE representation. */ 3059 definition = true; 3060 3061 memset (&thunk, 0, sizeof (cgraph_thunk_info)); 3062 thunk.thunk_p = true; 3063 create_edge (target, NULL, count); 3064 callees->can_throw_external = !TREE_NOTHROW (target->decl); 3065 3066 tree arguments = DECL_ARGUMENTS (decl); 3067 3068 while (arguments) 3069 { 3070 TREE_ADDRESSABLE (arguments) = false; 3071 arguments = TREE_CHAIN (arguments); 3072 } 3073 3074 expand_thunk (false, true); 3075 3076 /* Inline summary set-up. */ 3077 analyze (); 3078 inline_analyze_function (this); 3079 } 3080 3081 #include "gt-cgraphunit.h" 3082