1 /* Top-level LTO routines.
2 Copyright (C) 2009-2024 Free Software Foundation, Inc.
3 Contributed by CodeSourcery, Inc.
5 This file is part of GCC.
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
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
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/>. */
21 #define INCLUDE_STRING
24 #include "coretypes.h"
28 #include "basic-block.h"
32 #include "alloc-pool.h"
33 #include "tree-pass.h"
34 #include "tree-streamer.h"
38 #include "stor-layout.h"
39 #include "symbol-summary.h"
44 #include "lto-section-names.h"
45 #include "splay-tree.h"
46 #include "lto-partition.h"
48 #include "pass_manager.h"
49 #include "ipa-fnsummary.h"
50 #include "ipa-utils.h"
51 #include "gomp-constants.h"
52 #include "lto-symtab.h"
53 #include "stringpool.h"
54 #include "fold-const.h"
57 #include "lto-common.h"
58 #include "opts-jobserver.h"
60 /* Number of parallel tasks to run. */
61 static int lto_parallelism
;
63 /* Number of active WPA streaming processes. */
66 /* GNU make's jobserver info. */
67 static jobserver_info
*jinfo
= NULL
;
69 /* Return true when NODE has a clone that is analyzed (i.e. we need
70 to load its body even if the node itself is not needed). */
73 has_analyzed_clone_p (struct cgraph_node
*node
)
75 struct cgraph_node
*orig
= node
;
84 else if (node
->next_sibling_clone
)
85 node
= node
->next_sibling_clone
;
88 while (node
!= orig
&& !node
->next_sibling_clone
)
89 node
= node
->clone_of
;
91 node
= node
->next_sibling_clone
;
97 /* Read the function body for the function associated with NODE. */
100 lto_materialize_function (struct cgraph_node
*node
)
105 /* Read in functions with body (analyzed nodes)
106 and also functions that are needed to produce virtual clones. */
107 if ((node
->has_gimple_body_p () && node
->analyzed
)
108 || node
->used_as_abstract_origin
109 || has_analyzed_clone_p (node
))
111 /* Clones don't need to be read. */
114 if (DECL_FUNCTION_PERSONALITY (decl
) && !first_personality_decl
)
115 first_personality_decl
= DECL_FUNCTION_PERSONALITY (decl
);
116 /* If the file contains a function with a language specific EH
117 personality set or with EH enabled initialize the backend EH
119 if (DECL_FUNCTION_PERSONALITY (decl
)
120 || opt_for_fn (decl
, flag_exceptions
))
124 /* Let the middle end know about the function. */
125 rest_of_decl_compilation (decl
, 1, 0);
128 /* Materialize all the bodies for all the nodes in the callgraph. */
131 materialize_cgraph (void)
133 struct cgraph_node
*node
;
134 timevar_id_t lto_timer
;
138 flag_wpa
? "Materializing decls:" : "Reading function bodies:");
140 /* Start the appropriate timer depending on the mode that we are
142 lto_timer
= (flag_wpa
) ? TV_WHOPR_WPA
143 : (flag_ltrans
) ? TV_WHOPR_LTRANS
145 timevar_push (lto_timer
);
147 FOR_EACH_FUNCTION (node
)
149 if (node
->lto_file_data
)
151 lto_materialize_function (node
);
152 lto_stats
.num_input_cgraph_nodes
++;
156 current_function_decl
= NULL
;
160 fprintf (stderr
, "\n");
162 timevar_pop (lto_timer
);
165 /* Actually stream out ENCODER into TEMP_FILENAME. */
168 stream_out (char *temp_filename
, lto_symtab_encoder_t encoder
, int part
)
170 lto_file
*file
= lto_obj_file_open (temp_filename
, true);
172 fatal_error (input_location
, "%<lto_obj_file_open()%> failed");
173 lto_set_current_out_file (file
);
175 gcc_assert (!dump_file
);
176 streamer_dump_file
= dump_begin (TDI_lto_stream_out
, NULL
, part
);
177 ipa_write_optimization_summaries (encoder
);
179 free (CONST_CAST (char *, file
->filename
));
181 lto_set_current_out_file (NULL
);
182 lto_obj_file_close (file
);
184 if (streamer_dump_file
)
186 dump_end (TDI_lto_stream_out
, streamer_dump_file
);
187 streamer_dump_file
= NULL
;
191 /* Wait for forked process and signal errors. */
192 #ifdef HAVE_WORKING_FORK
202 int w
= waitpid (0, &status
, WUNTRACED
| WCONTINUED
);
204 fatal_error (input_location
, "waitpid failed");
206 if (WIFEXITED (status
) && WEXITSTATUS (status
))
207 fatal_error (input_location
, "streaming subprocess failed");
208 else if (WIFSIGNALED (status
))
209 fatal_error (input_location
,
210 "streaming subprocess was killed by signal");
212 while (!WIFEXITED (status
) && !WIFSIGNALED (status
));
216 /* Return token to the jobserver if active. */
217 if (jinfo
!= NULL
&& jinfo
->is_connected
)
218 jinfo
->return_token ();
223 stream_out_partitions_1 (char *temp_filename
, int blen
, int min
, int max
)
225 /* Write all the nodes in SET. */
226 for (int p
= min
; p
< max
; p
++)
228 sprintf (temp_filename
+ blen
, "%u.o", p
);
229 stream_out (temp_filename
, ltrans_partitions
[p
]->encoder
, p
);
230 ltrans_partitions
[p
]->encoder
= NULL
;
234 /* Stream out ENCODER into TEMP_FILENAME
235 Fork if that seems to help. */
238 stream_out_partitions (char *temp_filename
, int blen
, int min
, int max
,
239 bool ARG_UNUSED (last
))
241 #ifdef HAVE_WORKING_FORK
242 if (lto_parallelism
<= 1)
244 stream_out_partitions_1 (temp_filename
, blen
, min
, max
);
248 if (lto_parallelism
> 0 && nruns
>= lto_parallelism
)
251 /* If this is not the last parallel partition, execute new
252 streaming process. */
255 if (jinfo
!= NULL
&& jinfo
->is_connected
)
258 if (jinfo
->get_token ())
264 /* There are no free tokens, lets do the job outselves. */
265 stream_out_partitions_1 (temp_filename
, blen
, min
, max
);
266 asm_nodes_output
= true;
271 pid_t cpid
= fork ();
275 setproctitle ("lto1-wpa-streaming");
276 stream_out_partitions_1 (temp_filename
, blen
, min
, max
);
279 /* Fork failed; lets do the job ourseleves. */
281 stream_out_partitions_1 (temp_filename
, blen
, min
, max
);
285 /* Last partition; stream it and wait for all children to die. */
288 stream_out_partitions_1 (temp_filename
, blen
, min
, max
);
292 if (jinfo
!= NULL
&& jinfo
->is_connected
)
293 jinfo
->disconnect ();
295 asm_nodes_output
= true;
297 stream_out_partitions_1 (temp_filename
, blen
, min
, max
);
301 /* Write all output files in WPA mode and the file with the list of
305 lto_wpa_write_files (void)
308 ltrans_partition part
;
309 FILE *ltrans_output_list_stream
;
311 auto_vec
<char *>temp_filenames
;
312 auto_vec
<int>temp_priority
;
315 /* Open the LTRANS output list. */
316 if (!ltrans_output_list
)
317 fatal_error (input_location
, "no LTRANS output list filename provided");
319 timevar_push (TV_WHOPR_WPA
);
321 FOR_EACH_VEC_ELT (ltrans_partitions
, i
, part
)
322 lto_stats
.num_output_symtab_nodes
323 += lto_symtab_encoder_size (part
->encoder
);
325 timevar_pop (TV_WHOPR_WPA
);
327 timevar_push (TV_WHOPR_WPA_IO
);
330 /* Do body modifications needed for streaming before we fork out
332 FOR_EACH_FUNCTION (node
)
333 if (!node
->clone_of
&& gimple_has_body_p (node
->decl
))
334 lto_prepare_function_for_streaming (node
);
337 report_heap_memory_use ();
339 /* Generate a prefix for the LTRANS unit files. */
340 blen
= strlen (ltrans_output_list
);
341 temp_filename
= (char *) xmalloc (blen
+ sizeof ("2147483648.o"));
342 strcpy (temp_filename
, ltrans_output_list
);
343 if (blen
> sizeof (".out")
344 && strcmp (temp_filename
+ blen
- sizeof (".out") + 1,
346 temp_filename
[blen
- sizeof (".out") + 1] = '\0';
347 blen
= strlen (temp_filename
);
349 n_sets
= ltrans_partitions
.length ();
350 unsigned sets_per_worker
= n_sets
;
351 if (lto_parallelism
> 1)
353 if (lto_parallelism
> (int)n_sets
)
354 lto_parallelism
= n_sets
;
355 sets_per_worker
= (n_sets
+ lto_parallelism
- 1) / lto_parallelism
;
358 for (i
= 0; i
< n_sets
; i
++)
360 ltrans_partition part
= ltrans_partitions
[i
];
362 /* Write all the nodes in SET. */
363 sprintf (temp_filename
+ blen
, "%u.o", i
);
366 fprintf (stderr
, " %s (%s %i insns)", temp_filename
, part
->name
,
368 if (symtab
->dump_file
)
370 lto_symtab_encoder_iterator lsei
;
372 fprintf (symtab
->dump_file
,
373 "Writing partition %s to file %s, %i insns\n",
374 part
->name
, temp_filename
, part
->insns
);
375 fprintf (symtab
->dump_file
, " Symbols in partition: ");
376 for (lsei
= lsei_start_in_partition (part
->encoder
);
378 lsei_next_in_partition (&lsei
))
380 symtab_node
*node
= lsei_node (lsei
);
381 fprintf (symtab
->dump_file
, "%s ", node
->dump_asm_name ());
383 fprintf (symtab
->dump_file
, "\n Symbols in boundary: ");
384 for (lsei
= lsei_start (part
->encoder
); !lsei_end_p (lsei
);
387 symtab_node
*node
= lsei_node (lsei
);
388 if (!lto_symtab_encoder_in_partition_p (part
->encoder
, node
))
390 fprintf (symtab
->dump_file
, "%s ", node
->dump_asm_name ());
391 cgraph_node
*cnode
= dyn_cast
<cgraph_node
*> (node
);
393 && lto_symtab_encoder_encode_body_p (part
->encoder
,
395 fprintf (symtab
->dump_file
, "(body included)");
398 varpool_node
*vnode
= dyn_cast
<varpool_node
*> (node
);
400 && lto_symtab_encoder_encode_initializer_p (part
->encoder
,
402 fprintf (symtab
->dump_file
, "(initializer included)");
406 fprintf (symtab
->dump_file
, "\n");
408 gcc_checking_assert (lto_symtab_encoder_size (part
->encoder
) || !i
);
410 temp_priority
.safe_push (part
->insns
);
411 temp_filenames
.safe_push (xstrdup (temp_filename
));
413 memory_block_pool::trim (0);
415 for (int set
= 0; set
< MAX (lto_parallelism
, 1); set
++)
417 stream_out_partitions (temp_filename
, blen
, set
* sets_per_worker
,
418 MIN ((set
+ 1) * sets_per_worker
, n_sets
),
419 set
== MAX (lto_parallelism
, 1) - 1);
422 ltrans_output_list_stream
= fopen (ltrans_output_list
, "w");
423 if (ltrans_output_list_stream
== NULL
)
424 fatal_error (input_location
,
425 "opening LTRANS output list %s: %m", ltrans_output_list
);
426 for (i
= 0; i
< n_sets
; i
++)
428 unsigned int len
= strlen (temp_filenames
[i
]);
429 if (fprintf (ltrans_output_list_stream
, "%i\n", temp_priority
[i
]) < 0
430 || fwrite (temp_filenames
[i
], 1, len
, ltrans_output_list_stream
) < len
431 || fwrite ("\n", 1, 1, ltrans_output_list_stream
) < 1)
432 fatal_error (input_location
, "writing to LTRANS output list %s: %m",
434 free (temp_filenames
[i
]);
437 lto_stats
.num_output_files
+= n_sets
;
439 /* Close the LTRANS output list. */
440 if (fclose (ltrans_output_list_stream
))
441 fatal_error (input_location
,
442 "closing LTRANS output list %s: %m", ltrans_output_list
);
444 free_ltrans_partitions ();
445 free (temp_filename
);
447 timevar_pop (TV_WHOPR_WPA_IO
);
450 /* Create artificial pointers for "omp declare target link" vars. */
453 offload_handle_link_vars (void)
455 #ifdef ACCEL_COMPILER
457 FOR_EACH_VARIABLE (var
)
458 if (lookup_attribute ("omp declare target link",
459 DECL_ATTRIBUTES (var
->decl
)))
461 tree type
= build_pointer_type (TREE_TYPE (var
->decl
));
462 tree link_ptr_var
= build_decl (UNKNOWN_LOCATION
, VAR_DECL
,
463 clone_function_name (var
->decl
,
465 TREE_USED (link_ptr_var
) = 1;
466 TREE_STATIC (link_ptr_var
) = 1;
467 TREE_PUBLIC (link_ptr_var
) = TREE_PUBLIC (var
->decl
);
468 DECL_ARTIFICIAL (link_ptr_var
) = 1;
469 SET_DECL_ASSEMBLER_NAME (link_ptr_var
, DECL_NAME (link_ptr_var
));
470 SET_DECL_VALUE_EXPR (var
->decl
, build_simple_mem_ref (link_ptr_var
));
471 DECL_HAS_VALUE_EXPR_P (var
->decl
) = 1;
476 /* Perform whole program analysis (WPA) on the callgraph and write out the
477 optimization plan. */
480 do_whole_program_analysis (void)
486 if (!strcmp (flag_wpa
, "jobserver"))
488 jinfo
= new jobserver_info ();
489 if (jinfo
->is_active
)
492 lto_parallelism
= param_max_lto_streaming_parallelism
;
496 lto_parallelism
= atoi (flag_wpa
);
497 if (lto_parallelism
<= 0)
499 if (lto_parallelism
>= param_max_lto_streaming_parallelism
)
500 lto_parallelism
= param_max_lto_streaming_parallelism
;
503 timevar_start (TV_PHASE_OPT_GEN
);
505 /* Note that since we are in WPA mode, materialize_cgraph will not
506 actually read in all the function bodies. It only materializes
507 the decls and cgraph nodes so that analysis can be performed. */
508 materialize_cgraph ();
510 /* Reading in the cgraph uses different timers, start timing WPA now. */
511 timevar_push (TV_WHOPR_WPA
);
513 if (pre_ipa_mem_report
)
514 dump_memory_report ("Memory consumption before IPA");
516 symtab
->function_flags_ready
= true;
518 if (symtab
->dump_file
)
519 symtab
->dump (symtab
->dump_file
);
520 bitmap_obstack_initialize (NULL
);
521 symtab
->state
= IPA_SSA
;
523 execute_ipa_pass_list (g
->get_passes ()->all_regular_ipa_passes
);
525 /* When WPA analysis raises errors, do not bother to output anything. */
529 /* We are about to launch the final LTRANS phase, stop the WPA timer. */
530 timevar_pop (TV_WHOPR_WPA
);
532 /* We are no longer going to stream in anything. Free some memory. */
533 lto_free_file_name_hash ();
536 timevar_push (TV_WHOPR_PARTITIONING
);
538 gcc_assert (!dump_file
);
539 dump_file
= dump_begin (partition_dump_id
, NULL
);
542 symtab
->dump (dump_file
);
544 symtab_node::checking_verify_symtab_nodes ();
545 bitmap_obstack_release (NULL
);
546 if (flag_lto_partition
== LTO_PARTITION_1TO1
)
548 else if (flag_lto_partition
== LTO_PARTITION_MAX
)
550 else if (flag_lto_partition
== LTO_PARTITION_ONE
)
551 lto_balanced_map (1, INT_MAX
);
552 else if (flag_lto_partition
== LTO_PARTITION_BALANCED
)
553 lto_balanced_map (param_lto_partitions
,
554 param_max_partition_size
);
558 /* Size summaries are needed for balanced partitioning. Free them now so
559 the memory can be used for streamer caches. */
560 ipa_free_size_summary ();
562 /* AUX pointers are used by partitioning code to bookkeep number of
563 partitions symbol is in. This is no longer needed. */
564 FOR_EACH_SYMBOL (node
)
567 lto_stats
.num_cgraph_partitions
+= ltrans_partitions
.length ();
569 /* Find out statics that need to be promoted
570 to globals with hidden visibility because they are accessed from multiple
572 lto_promote_cross_file_statics ();
573 offload_handle_link_vars ();
575 dump_end (partition_dump_id
, dump_file
);
577 timevar_pop (TV_WHOPR_PARTITIONING
);
579 timevar_stop (TV_PHASE_OPT_GEN
);
581 /* Collect a last time - in lto_wpa_write_files we may end up forking
582 with the idea that this doesn't increase memory usage. So we
583 absoultely do not want to collect after that. */
586 timevar_start (TV_PHASE_STREAM_OUT
);
589 fprintf (stderr
, "\nStreaming out");
592 lto_wpa_write_files ();
594 fprintf (stderr
, "\n");
595 timevar_stop (TV_PHASE_STREAM_OUT
);
597 if (post_ipa_mem_report
)
598 dump_memory_report ("Memory consumption after IPA");
600 /* Show the LTO report before launching LTRANS. */
601 if (flag_lto_report
|| (flag_wpa
&& flag_lto_report_wpa
))
602 print_lto_report_1 ();
604 dump_memory_report ("Final");
608 lto_option_lang_mask (void)
613 /* Main entry point for the GIMPLE front end. This front end has
614 three main personalities:
616 - LTO (-flto). All the object files on the command line are
617 loaded in memory and processed as a single translation unit.
618 This is the traditional link-time optimization behavior.
620 - WPA (-fwpa). Only the callgraph and summary information for
621 files in the command file are loaded. A single callgraph
622 (without function bodies) is instantiated for the whole set of
623 files. IPA passes are only allowed to analyze the call graph
624 and make transformation decisions. The callgraph is
625 partitioned, each partition is written to a new object file
626 together with the transformation decisions.
628 - LTRANS (-fltrans). Similar to -flto but it prevents the IPA
629 summary files from running again. Since WPA computed summary
630 information and decided what transformations to apply, LTRANS
631 simply applies them. */
636 /* LTO is called as a front end, even though it is not a front end.
637 Because it is called as a front end, TV_PHASE_PARSING and
638 TV_PARSE_GLOBAL are active, and we need to turn them off while
639 doing LTO. Later we turn them back on so they are active up in
641 timevar_pop (TV_PARSE_GLOBAL
);
642 timevar_stop (TV_PHASE_PARSING
);
644 timevar_start (TV_PHASE_SETUP
);
646 /* Initialize the LTO front end. */
649 timevar_stop (TV_PHASE_SETUP
);
650 timevar_start (TV_PHASE_STREAM_IN
);
652 /* Read all the symbols and call graph from all the files in the
654 read_cgraph_and_symbols (num_in_fnames
, in_fnames
);
656 timevar_stop (TV_PHASE_STREAM_IN
);
660 offload_handle_link_vars ();
662 /* If WPA is enabled analyze the whole call graph and create an
663 optimization plan. Otherwise, read in all the function
664 bodies and continue with optimization. */
666 do_whole_program_analysis ();
669 timevar_start (TV_PHASE_OPT_GEN
);
671 materialize_cgraph ();
674 lto_promote_statics_nonwpa ();
675 offload_handle_link_vars ();
678 /* Annotate the CU DIE and mark the early debug phase as finished. */
679 debuginfo_early_start ();
680 debug_hooks
->early_finish ("<artificial>");
681 debuginfo_early_stop ();
683 /* Let the middle end know that we have read and merged all of
687 timevar_stop (TV_PHASE_OPT_GEN
);
689 /* FIXME lto, if the processes spawned by WPA fail, we miss
690 the chance to print WPA's report, so WPA will call
691 print_lto_report before launching LTRANS. If LTRANS was
692 launched directly by the driver we would not need to do
694 if (flag_lto_report
|| (flag_wpa
&& flag_lto_report_wpa
))
695 print_lto_report_1 ();
699 /* Here we make LTO pretend to be a parser. */
700 timevar_start (TV_PHASE_PARSING
);
701 timevar_push (TV_PARSE_GLOBAL
);