2010-11-27 François Dumont <francois.cppdevs@free.fr>
[official-gcc.git] / gcc / ipa-inline.c
blobe8c78f9f5fb7c78afdf6f9b90960bc5ad83b3c6d
1 /* Inlining decision heuristics.
2 Copyright (C) 2003, 2004, 2007, 2008, 2009, 2010
3 Free Software Foundation, Inc.
4 Contributed by Jan Hubicka
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify it under
9 the terms of the GNU General Public License as published by the Free
10 Software Foundation; either version 3, or (at your option) any later
11 version.
13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 for more details.
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
22 /* Inlining decision heuristics
24 We separate inlining decisions from the inliner itself and store it
25 inside callgraph as so called inline plan. Refer to cgraph.c
26 documentation about particular representation of inline plans in the
27 callgraph.
29 There are three major parts of this file:
31 cgraph_mark_inline implementation
33 This function allows to mark given call inline and performs necessary
34 modifications of cgraph (production of the clones and updating overall
35 statistics)
37 inlining heuristics limits
39 These functions allow to check that particular inlining is allowed
40 by the limits specified by user (allowed function growth, overall unit
41 growth and so on).
43 inlining heuristics
45 This is implementation of IPA pass aiming to get as much of benefit
46 from inlining obeying the limits checked above.
48 The implementation of particular heuristics is separated from
49 the rest of code to make it easier to replace it with more complicated
50 implementation in the future. The rest of inlining code acts as a
51 library aimed to modify the callgraph and verify that the parameters
52 on code size growth fits.
54 To mark given call inline, use cgraph_mark_inline function, the
55 verification is performed by cgraph_default_inline_p and
56 cgraph_check_inline_limits.
58 The heuristics implements simple knapsack style algorithm ordering
59 all functions by their "profitability" (estimated by code size growth)
60 and inlining them in priority order.
62 cgraph_decide_inlining implements heuristics taking whole callgraph
63 into account, while cgraph_decide_inlining_incrementally considers
64 only one function at a time and is used by early inliner.
66 The inliner itself is split into several passes:
68 pass_inline_parameters
70 This pass computes local properties of functions that are used by inliner:
71 estimated function body size, whether function is inlinable at all and
72 stack frame consumption.
74 Before executing any of inliner passes, this local pass has to be applied
75 to each function in the callgraph (ie run as subpass of some earlier
76 IPA pass). The results are made out of date by any optimization applied
77 on the function body.
79 pass_early_inlining
81 Simple local inlining pass inlining callees into current function. This
82 pass makes no global whole compilation unit analysis and this when allowed
83 to do inlining expanding code size it might result in unbounded growth of
84 whole unit.
86 The pass is run during conversion into SSA form. Only functions already
87 converted into SSA form are inlined, so the conversion must happen in
88 topological order on the callgraph (that is maintained by pass manager).
89 The functions after inlining are early optimized so the early inliner sees
90 unoptimized function itself, but all considered callees are already
91 optimized allowing it to unfold abstraction penalty on C++ effectively and
92 cheaply.
94 pass_ipa_early_inlining
96 With profiling, the early inlining is also necessary to reduce
97 instrumentation costs on program with high abstraction penalty (doing
98 many redundant calls). This can't happen in parallel with early
99 optimization and profile instrumentation, because we would end up
100 re-instrumenting already instrumented function bodies we brought in via
101 inlining.
103 To avoid this, this pass is executed as IPA pass before profiling. It is
104 simple wrapper to pass_early_inlining and ensures first inlining.
106 pass_ipa_inline
108 This is the main pass implementing simple greedy algorithm to do inlining
109 of small functions that results in overall growth of compilation unit and
110 inlining of functions called once. The pass compute just so called inline
111 plan (representation of inlining to be done in callgraph) and unlike early
112 inlining it is not performing the inlining itself.
114 pass_apply_inline
116 This pass performs actual inlining according to pass_ipa_inline on given
117 function. Possible the function body before inlining is saved when it is
118 needed for further inlining later.
121 #include "config.h"
122 #include "system.h"
123 #include "coretypes.h"
124 #include "tm.h"
125 #include "tree.h"
126 #include "tree-inline.h"
127 #include "langhooks.h"
128 #include "flags.h"
129 #include "cgraph.h"
130 #include "diagnostic.h"
131 #include "gimple-pretty-print.h"
132 #include "timevar.h"
133 #include "params.h"
134 #include "fibheap.h"
135 #include "intl.h"
136 #include "tree-pass.h"
137 #include "hashtab.h"
138 #include "coverage.h"
139 #include "ggc.h"
140 #include "tree-flow.h"
141 #include "rtl.h"
142 #include "ipa-prop.h"
143 #include "except.h"
145 #define MAX_TIME 1000000000
147 /* Mode incremental inliner operate on:
149 In ALWAYS_INLINE only functions marked
150 always_inline are inlined. This mode is used after detecting cycle during
151 flattening.
153 In SIZE mode, only functions that reduce function body size after inlining
154 are inlined, this is used during early inlining.
156 in ALL mode, everything is inlined. This is used during flattening. */
157 enum inlining_mode {
158 INLINE_NONE = 0,
159 INLINE_ALWAYS_INLINE,
160 INLINE_SIZE_NORECURSIVE,
161 INLINE_SIZE,
162 INLINE_ALL
165 static bool
166 cgraph_decide_inlining_incrementally (struct cgraph_node *, enum inlining_mode);
167 static void cgraph_flatten (struct cgraph_node *node);
170 /* Statistics we collect about inlining algorithm. */
171 static int ncalls_inlined;
172 static int nfunctions_inlined;
173 static int overall_size;
174 static gcov_type max_count, max_benefit;
176 /* Holders of ipa cgraph hooks: */
177 static struct cgraph_node_hook_list *function_insertion_hook_holder;
179 static inline struct inline_summary *
180 inline_summary (struct cgraph_node *node)
182 return &node->local.inline_summary;
185 /* Estimate self time of the function after inlining WHAT into TO. */
187 static int
188 cgraph_estimate_time_after_inlining (int frequency, struct cgraph_node *to,
189 struct cgraph_node *what)
191 gcov_type time = (((gcov_type)what->global.time
192 - inline_summary (what)->time_inlining_benefit)
193 * frequency + CGRAPH_FREQ_BASE / 2) / CGRAPH_FREQ_BASE
194 + to->global.time;
195 if (time < 0)
196 time = 0;
197 if (time > MAX_TIME)
198 time = MAX_TIME;
199 return time;
202 /* Estimate self time of the function after inlining WHAT into TO. */
204 static inline int
205 cgraph_estimate_size_after_inlining (int times, struct cgraph_node *to,
206 struct cgraph_node *what)
208 int size = ((what->global.size - inline_summary (what)->size_inlining_benefit)
209 * times + to->global.size);
210 gcc_assert (size >= 0);
211 return size;
214 /* Scale frequency of NODE edges by FREQ_SCALE and increase loop nest
215 by NEST. */
217 static void
218 update_noncloned_frequencies (struct cgraph_node *node,
219 int freq_scale, int nest)
221 struct cgraph_edge *e;
223 /* We do not want to ignore high loop nest after freq drops to 0. */
224 if (!freq_scale)
225 freq_scale = 1;
226 for (e = node->callees; e; e = e->next_callee)
228 e->loop_nest += nest;
229 e->frequency = e->frequency * (gcov_type) freq_scale / CGRAPH_FREQ_BASE;
230 if (e->frequency > CGRAPH_FREQ_MAX)
231 e->frequency = CGRAPH_FREQ_MAX;
232 if (!e->inline_failed)
233 update_noncloned_frequencies (e->callee, freq_scale, nest);
237 /* E is expected to be an edge being inlined. Clone destination node of
238 the edge and redirect it to the new clone.
239 DUPLICATE is used for bookkeeping on whether we are actually creating new
240 clones or re-using node originally representing out-of-line function call.
242 void
243 cgraph_clone_inlined_nodes (struct cgraph_edge *e, bool duplicate,
244 bool update_original)
246 HOST_WIDE_INT peak;
248 if (duplicate)
250 /* We may eliminate the need for out-of-line copy to be output.
251 In that case just go ahead and re-use it. */
252 if (!e->callee->callers->next_caller
253 /* FIXME: When address is taken of DECL_EXTERNAL function we still can remove its
254 offline copy, but we would need to keep unanalyzed node in the callgraph so
255 references can point to it. */
256 && !e->callee->address_taken
257 && cgraph_can_remove_if_no_direct_calls_p (e->callee)
258 /* Inlining might enable more devirtualizing, so we want to remove
259 those only after all devirtualizable virtual calls are processed.
260 Lacking may edges in callgraph we just preserve them post
261 inlining. */
262 && (!DECL_VIRTUAL_P (e->callee->decl)
263 || (!DECL_COMDAT (e->callee->decl) && !DECL_EXTERNAL (e->callee->decl)))
264 /* Don't reuse if more than one function shares a comdat group.
265 If the other function(s) are needed, we need to emit even
266 this function out of line. */
267 && !e->callee->same_comdat_group
268 && !cgraph_new_nodes)
270 gcc_assert (!e->callee->global.inlined_to);
271 if (e->callee->analyzed && !DECL_EXTERNAL (e->callee->decl))
273 overall_size -= e->callee->global.size;
274 nfunctions_inlined++;
276 duplicate = false;
277 e->callee->local.externally_visible = false;
278 update_noncloned_frequencies (e->callee, e->frequency, e->loop_nest);
280 else
282 struct cgraph_node *n;
283 n = cgraph_clone_node (e->callee, e->callee->decl,
284 e->count, e->frequency, e->loop_nest,
285 update_original, NULL);
286 cgraph_redirect_edge_callee (e, n);
290 if (e->caller->global.inlined_to)
291 e->callee->global.inlined_to = e->caller->global.inlined_to;
292 else
293 e->callee->global.inlined_to = e->caller;
294 e->callee->global.stack_frame_offset
295 = e->caller->global.stack_frame_offset
296 + inline_summary (e->caller)->estimated_self_stack_size;
297 peak = e->callee->global.stack_frame_offset
298 + inline_summary (e->callee)->estimated_self_stack_size;
299 if (e->callee->global.inlined_to->global.estimated_stack_size < peak)
300 e->callee->global.inlined_to->global.estimated_stack_size = peak;
301 cgraph_propagate_frequency (e->callee);
303 /* Recursively clone all bodies. */
304 for (e = e->callee->callees; e; e = e->next_callee)
305 if (!e->inline_failed)
306 cgraph_clone_inlined_nodes (e, duplicate, update_original);
309 /* Mark edge E as inlined and update callgraph accordingly. UPDATE_ORIGINAL
310 specify whether profile of original function should be updated. If any new
311 indirect edges are discovered in the process, add them to NEW_EDGES, unless
312 it is NULL. Return true iff any new callgraph edges were discovered as a
313 result of inlining. */
315 static bool
316 cgraph_mark_inline_edge (struct cgraph_edge *e, bool update_original,
317 VEC (cgraph_edge_p, heap) **new_edges)
319 int old_size = 0, new_size = 0;
320 struct cgraph_node *to = NULL, *what;
321 struct cgraph_edge *curr = e;
322 int freq;
324 gcc_assert (e->inline_failed);
325 e->inline_failed = CIF_OK;
326 DECL_POSSIBLY_INLINED (e->callee->decl) = true;
328 cgraph_clone_inlined_nodes (e, true, update_original);
330 what = e->callee;
332 freq = e->frequency;
333 /* Now update size of caller and all functions caller is inlined into. */
334 for (;e && !e->inline_failed; e = e->caller->callers)
336 to = e->caller;
337 old_size = e->caller->global.size;
338 new_size = cgraph_estimate_size_after_inlining (1, to, what);
339 to->global.size = new_size;
340 to->global.time = cgraph_estimate_time_after_inlining (freq, to, what);
342 gcc_assert (what->global.inlined_to == to);
343 if (new_size > old_size)
344 overall_size += new_size - old_size;
345 ncalls_inlined++;
347 /* FIXME: We should remove the optimize check after we ensure we never run
348 IPA passes when not optimizng. */
349 if (flag_indirect_inlining && optimize)
350 return ipa_propagate_indirect_call_infos (curr, new_edges);
351 else
352 return false;
355 /* Mark all calls of EDGE->CALLEE inlined into EDGE->CALLER. */
357 static void
358 cgraph_mark_inline (struct cgraph_edge *edge)
360 struct cgraph_node *to = edge->caller;
361 struct cgraph_node *what = edge->callee;
362 struct cgraph_edge *e, *next;
364 gcc_assert (!edge->call_stmt_cannot_inline_p);
365 /* Look for all calls, mark them inline and clone recursively
366 all inlined functions. */
367 for (e = what->callers; e; e = next)
369 next = e->next_caller;
370 if (e->caller == to && e->inline_failed)
372 cgraph_mark_inline_edge (e, true, NULL);
373 if (e == edge)
374 edge = next;
379 /* Estimate the growth caused by inlining NODE into all callees. */
381 static int
382 cgraph_estimate_growth (struct cgraph_node *node)
384 int growth = 0;
385 struct cgraph_edge *e;
386 bool self_recursive = false;
388 if (node->global.estimated_growth != INT_MIN)
389 return node->global.estimated_growth;
391 for (e = node->callers; e; e = e->next_caller)
393 if (e->caller == node)
394 self_recursive = true;
395 if (e->inline_failed)
396 growth += (cgraph_estimate_size_after_inlining (1, e->caller, node)
397 - e->caller->global.size);
400 /* ??? Wrong for non-trivially self recursive functions or cases where
401 we decide to not inline for different reasons, but it is not big deal
402 as in that case we will keep the body around, but we will also avoid
403 some inlining. */
404 if (cgraph_will_be_removed_from_program_if_no_direct_calls (node)
405 && !DECL_EXTERNAL (node->decl) && !self_recursive)
406 growth -= node->global.size;
407 /* COMDAT functions are very often not shared across multiple units since they
408 come from various template instantiations. Take this into account. */
409 else if (DECL_COMDAT (node->decl) && !self_recursive
410 && cgraph_can_remove_if_no_direct_calls_p (node))
411 growth -= (node->global.size
412 * (100 - PARAM_VALUE (PARAM_COMDAT_SHARING_PROBABILITY)) + 50) / 100;
414 node->global.estimated_growth = growth;
415 return growth;
418 /* Return false when inlining WHAT into TO is not good idea
419 as it would cause too large growth of function bodies.
420 When ONE_ONLY is true, assume that only one call site is going
421 to be inlined, otherwise figure out how many call sites in
422 TO calls WHAT and verify that all can be inlined.
425 static bool
426 cgraph_check_inline_limits (struct cgraph_node *to, struct cgraph_node *what,
427 cgraph_inline_failed_t *reason, bool one_only)
429 int times = 0;
430 struct cgraph_edge *e;
431 int newsize;
432 int limit;
433 HOST_WIDE_INT stack_size_limit, inlined_stack;
435 if (one_only)
436 times = 1;
437 else
438 for (e = to->callees; e; e = e->next_callee)
439 if (e->callee == what)
440 times++;
442 if (to->global.inlined_to)
443 to = to->global.inlined_to;
445 /* When inlining large function body called once into small function,
446 take the inlined function as base for limiting the growth. */
447 if (inline_summary (to)->self_size > inline_summary(what)->self_size)
448 limit = inline_summary (to)->self_size;
449 else
450 limit = inline_summary (what)->self_size;
452 limit += limit * PARAM_VALUE (PARAM_LARGE_FUNCTION_GROWTH) / 100;
454 /* Check the size after inlining against the function limits. But allow
455 the function to shrink if it went over the limits by forced inlining. */
456 newsize = cgraph_estimate_size_after_inlining (times, to, what);
457 if (newsize >= to->global.size
458 && newsize > PARAM_VALUE (PARAM_LARGE_FUNCTION_INSNS)
459 && newsize > limit)
461 if (reason)
462 *reason = CIF_LARGE_FUNCTION_GROWTH_LIMIT;
463 return false;
466 stack_size_limit = inline_summary (to)->estimated_self_stack_size;
468 stack_size_limit += stack_size_limit * PARAM_VALUE (PARAM_STACK_FRAME_GROWTH) / 100;
470 inlined_stack = (to->global.stack_frame_offset
471 + inline_summary (to)->estimated_self_stack_size
472 + what->global.estimated_stack_size);
473 if (inlined_stack > stack_size_limit
474 && inlined_stack > PARAM_VALUE (PARAM_LARGE_STACK_FRAME))
476 if (reason)
477 *reason = CIF_LARGE_STACK_FRAME_GROWTH_LIMIT;
478 return false;
480 return true;
483 /* Return true when function N is small enough to be inlined. */
485 static bool
486 cgraph_default_inline_p (struct cgraph_node *n, cgraph_inline_failed_t *reason)
488 tree decl = n->decl;
490 if (n->local.disregard_inline_limits)
491 return true;
493 if (!flag_inline_small_functions && !DECL_DECLARED_INLINE_P (decl))
495 if (reason)
496 *reason = CIF_FUNCTION_NOT_INLINE_CANDIDATE;
497 return false;
499 if (!n->analyzed)
501 if (reason)
502 *reason = CIF_BODY_NOT_AVAILABLE;
503 return false;
505 if (cgraph_function_body_availability (n) <= AVAIL_OVERWRITABLE)
507 if (reason)
508 *reason = CIF_OVERWRITABLE;
509 return false;
513 if (DECL_DECLARED_INLINE_P (decl))
515 if (n->global.size >= MAX_INLINE_INSNS_SINGLE)
517 if (reason)
518 *reason = CIF_MAX_INLINE_INSNS_SINGLE_LIMIT;
519 return false;
522 else
524 if (n->global.size >= MAX_INLINE_INSNS_AUTO)
526 if (reason)
527 *reason = CIF_MAX_INLINE_INSNS_AUTO_LIMIT;
528 return false;
532 return true;
535 /* Return true when inlining WHAT would create recursive inlining.
536 We call recursive inlining all cases where same function appears more than
537 once in the single recursion nest path in the inline graph. */
539 static inline bool
540 cgraph_recursive_inlining_p (struct cgraph_node *to,
541 struct cgraph_node *what,
542 cgraph_inline_failed_t *reason)
544 bool recursive;
545 if (to->global.inlined_to)
546 recursive = what->decl == to->global.inlined_to->decl;
547 else
548 recursive = what->decl == to->decl;
549 /* Marking recursive function inline has sane semantic and thus we should
550 not warn on it. */
551 if (recursive && reason)
552 *reason = (what->local.disregard_inline_limits
553 ? CIF_RECURSIVE_INLINING : CIF_UNSPECIFIED);
554 return recursive;
557 /* A cost model driving the inlining heuristics in a way so the edges with
558 smallest badness are inlined first. After each inlining is performed
559 the costs of all caller edges of nodes affected are recomputed so the
560 metrics may accurately depend on values such as number of inlinable callers
561 of the function or function body size. */
563 static int
564 cgraph_edge_badness (struct cgraph_edge *edge, bool dump)
566 gcov_type badness;
567 int growth =
568 (cgraph_estimate_size_after_inlining (1, edge->caller, edge->callee)
569 - edge->caller->global.size);
571 if (edge->callee->local.disregard_inline_limits)
572 return INT_MIN;
574 if (dump)
576 fprintf (dump_file, " Badness calculcation for %s -> %s\n",
577 cgraph_node_name (edge->caller),
578 cgraph_node_name (edge->callee));
579 fprintf (dump_file, " growth %i, time %i-%i, size %i-%i\n",
580 growth,
581 edge->callee->global.time,
582 inline_summary (edge->callee)->time_inlining_benefit,
583 edge->callee->global.size,
584 inline_summary (edge->callee)->size_inlining_benefit);
587 /* Always prefer inlining saving code size. */
588 if (growth <= 0)
590 badness = INT_MIN - growth;
591 if (dump)
592 fprintf (dump_file, " %i: Growth %i < 0\n", (int) badness,
593 growth);
596 /* When profiling is available, base priorities -(#calls / growth).
597 So we optimize for overall number of "executed" inlined calls. */
598 else if (max_count)
600 badness =
601 ((int)
602 ((double) edge->count * INT_MIN / max_count / (max_benefit + 1)) *
603 (inline_summary (edge->callee)->time_inlining_benefit + 1)) / growth;
604 if (dump)
606 fprintf (dump_file,
607 " %i (relative %f): profile info. Relative count %f"
608 " * Relative benefit %f\n",
609 (int) badness, (double) badness / INT_MIN,
610 (double) edge->count / max_count,
611 (double) (inline_summary (edge->callee)->
612 time_inlining_benefit + 1) / (max_benefit + 1));
616 /* When function local profile is available, base priorities on
617 growth / frequency, so we optimize for overall frequency of inlined
618 calls. This is not too accurate since while the call might be frequent
619 within function, the function itself is infrequent.
621 Other objective to optimize for is number of different calls inlined.
622 We add the estimated growth after inlining all functions to bias the
623 priorities slightly in this direction (so fewer times called functions
624 of the same size gets priority). */
625 else if (flag_guess_branch_prob)
627 int div = edge->frequency * 100 / CGRAPH_FREQ_BASE + 1;
628 int benefitperc;
629 int growth_for_all;
630 badness = growth * 10000;
631 benefitperc =
632 MIN (100 * inline_summary (edge->callee)->time_inlining_benefit /
633 (edge->callee->global.time + 1) +1, 100);
634 div *= benefitperc;
637 /* Decrease badness if call is nested. */
638 /* Compress the range so we don't overflow. */
639 if (div > 10000)
640 div = 10000 + ceil_log2 (div) - 8;
641 if (div < 1)
642 div = 1;
643 if (badness > 0)
644 badness /= div;
645 growth_for_all = cgraph_estimate_growth (edge->callee);
646 badness += growth_for_all;
647 if (badness > INT_MAX)
648 badness = INT_MAX;
649 if (dump)
651 fprintf (dump_file,
652 " %i: guessed profile. frequency %i, overall growth %i,"
653 " benefit %i%%, divisor %i\n",
654 (int) badness, edge->frequency, growth_for_all, benefitperc, div);
657 /* When function local profile is not available or it does not give
658 useful information (ie frequency is zero), base the cost on
659 loop nest and overall size growth, so we optimize for overall number
660 of functions fully inlined in program. */
661 else
663 int nest = MIN (edge->loop_nest, 8);
664 badness = cgraph_estimate_growth (edge->callee) * 256;
666 /* Decrease badness if call is nested. */
667 if (badness > 0)
668 badness >>= nest;
669 else
671 badness <<= nest;
673 if (dump)
674 fprintf (dump_file, " %i: no profile. nest %i\n", (int) badness,
675 nest);
678 /* Ensure that we did not overflow in all the fixed point math above. */
679 gcc_assert (badness >= INT_MIN);
680 gcc_assert (badness <= INT_MAX - 1);
681 /* Make recursive inlining happen always after other inlining is done. */
682 if (cgraph_recursive_inlining_p (edge->caller, edge->callee, NULL))
683 return badness + 1;
684 else
685 return badness;
688 /* Recompute badness of EDGE and update its key in HEAP if needed. */
689 static void
690 update_edge_key (fibheap_t heap, struct cgraph_edge *edge)
692 int badness = cgraph_edge_badness (edge, false);
693 if (edge->aux)
695 fibnode_t n = (fibnode_t) edge->aux;
696 gcc_checking_assert (n->data == edge);
698 /* fibheap_replace_key only decrease the keys.
699 When we increase the key we do not update heap
700 and instead re-insert the element once it becomes
701 a minium of heap. */
702 if (badness < n->key)
704 fibheap_replace_key (heap, n, badness);
705 gcc_checking_assert (n->key == badness);
708 else
709 edge->aux = fibheap_insert (heap, badness, edge);
712 /* Recompute heap nodes for each of caller edge. */
714 static void
715 update_caller_keys (fibheap_t heap, struct cgraph_node *node,
716 bitmap updated_nodes)
718 struct cgraph_edge *edge;
719 cgraph_inline_failed_t failed_reason;
721 if (!node->local.inlinable
722 || cgraph_function_body_availability (node) <= AVAIL_OVERWRITABLE
723 || node->global.inlined_to)
724 return;
725 if (!bitmap_set_bit (updated_nodes, node->uid))
726 return;
727 node->global.estimated_growth = INT_MIN;
729 /* See if there is something to do. */
730 for (edge = node->callers; edge; edge = edge->next_caller)
731 if (edge->inline_failed)
732 break;
733 if (!edge)
734 return;
735 /* Prune out edges we won't inline into anymore. */
736 if (!cgraph_default_inline_p (node, &failed_reason))
738 for (; edge; edge = edge->next_caller)
739 if (edge->aux)
741 fibheap_delete_node (heap, (fibnode_t) edge->aux);
742 edge->aux = NULL;
743 if (edge->inline_failed)
744 edge->inline_failed = failed_reason;
746 return;
749 for (; edge; edge = edge->next_caller)
750 if (edge->inline_failed)
751 update_edge_key (heap, edge);
754 /* Recompute heap nodes for each uninlined call.
755 This is used when we know that edge badnesses are going only to increase
756 (we introduced new call site) and thus all we need is to insert newly
757 created edges into heap. */
759 static void
760 update_callee_keys (fibheap_t heap, struct cgraph_node *node,
761 bitmap updated_nodes)
763 struct cgraph_edge *e = node->callees;
764 node->global.estimated_growth = INT_MIN;
766 if (!e)
767 return;
768 while (true)
769 if (!e->inline_failed && e->callee->callees)
770 e = e->callee->callees;
771 else
773 if (e->inline_failed
774 && e->callee->local.inlinable
775 && cgraph_function_body_availability (e->callee) >= AVAIL_AVAILABLE
776 && !bitmap_bit_p (updated_nodes, e->callee->uid))
778 node->global.estimated_growth = INT_MIN;
779 /* If function becomes uninlinable, we need to remove it from the heap. */
780 if (!cgraph_default_inline_p (e->callee, &e->inline_failed))
781 update_caller_keys (heap, e->callee, updated_nodes);
782 else
783 /* Otherwise update just edge E. */
784 update_edge_key (heap, e);
786 if (e->next_callee)
787 e = e->next_callee;
788 else
792 if (e->caller == node)
793 return;
794 e = e->caller->callers;
796 while (!e->next_callee);
797 e = e->next_callee;
802 /* Recompute heap nodes for each of caller edges of each of callees.
803 Walk recursively into all inline clones. */
805 static void
806 update_all_callee_keys (fibheap_t heap, struct cgraph_node *node,
807 bitmap updated_nodes)
809 struct cgraph_edge *e = node->callees;
810 node->global.estimated_growth = INT_MIN;
812 if (!e)
813 return;
814 while (true)
815 if (!e->inline_failed && e->callee->callees)
816 e = e->callee->callees;
817 else
819 if (e->inline_failed)
820 update_caller_keys (heap, e->callee, updated_nodes);
821 if (e->next_callee)
822 e = e->next_callee;
823 else
827 if (e->caller == node)
828 return;
829 e = e->caller->callers;
831 while (!e->next_callee);
832 e = e->next_callee;
837 /* Enqueue all recursive calls from NODE into priority queue depending on
838 how likely we want to recursively inline the call. */
840 static void
841 lookup_recursive_calls (struct cgraph_node *node, struct cgraph_node *where,
842 fibheap_t heap)
844 static int priority;
845 struct cgraph_edge *e;
846 for (e = where->callees; e; e = e->next_callee)
847 if (e->callee == node)
849 /* When profile feedback is available, prioritize by expected number
850 of calls. Without profile feedback we maintain simple queue
851 to order candidates via recursive depths. */
852 fibheap_insert (heap,
853 !max_count ? priority++
854 : -(e->count / ((max_count + (1<<24) - 1) / (1<<24))),
857 for (e = where->callees; e; e = e->next_callee)
858 if (!e->inline_failed)
859 lookup_recursive_calls (node, e->callee, heap);
862 /* Decide on recursive inlining: in the case function has recursive calls,
863 inline until body size reaches given argument. If any new indirect edges
864 are discovered in the process, add them to *NEW_EDGES, unless NEW_EDGES
865 is NULL. */
867 static bool
868 cgraph_decide_recursive_inlining (struct cgraph_node *node,
869 VEC (cgraph_edge_p, heap) **new_edges)
871 int limit = PARAM_VALUE (PARAM_MAX_INLINE_INSNS_RECURSIVE_AUTO);
872 int max_depth = PARAM_VALUE (PARAM_MAX_INLINE_RECURSIVE_DEPTH_AUTO);
873 int probability = PARAM_VALUE (PARAM_MIN_INLINE_RECURSIVE_PROBABILITY);
874 fibheap_t heap;
875 struct cgraph_edge *e;
876 struct cgraph_node *master_clone, *next;
877 int depth = 0;
878 int n = 0;
880 /* It does not make sense to recursively inline always-inline functions
881 as we are going to sorry() on the remaining calls anyway. */
882 if (node->local.disregard_inline_limits
883 && lookup_attribute ("always_inline", DECL_ATTRIBUTES (node->decl)))
884 return false;
886 if (optimize_function_for_size_p (DECL_STRUCT_FUNCTION (node->decl))
887 || (!flag_inline_functions && !DECL_DECLARED_INLINE_P (node->decl)))
888 return false;
890 if (DECL_DECLARED_INLINE_P (node->decl))
892 limit = PARAM_VALUE (PARAM_MAX_INLINE_INSNS_RECURSIVE);
893 max_depth = PARAM_VALUE (PARAM_MAX_INLINE_RECURSIVE_DEPTH);
896 /* Make sure that function is small enough to be considered for inlining. */
897 if (!max_depth
898 || cgraph_estimate_size_after_inlining (1, node, node) >= limit)
899 return false;
900 heap = fibheap_new ();
901 lookup_recursive_calls (node, node, heap);
902 if (fibheap_empty (heap))
904 fibheap_delete (heap);
905 return false;
908 if (dump_file)
909 fprintf (dump_file,
910 " Performing recursive inlining on %s\n",
911 cgraph_node_name (node));
913 /* We need original clone to copy around. */
914 master_clone = cgraph_clone_node (node, node->decl,
915 node->count, CGRAPH_FREQ_BASE, 1,
916 false, NULL);
917 master_clone->needed = true;
918 for (e = master_clone->callees; e; e = e->next_callee)
919 if (!e->inline_failed)
920 cgraph_clone_inlined_nodes (e, true, false);
922 /* Do the inlining and update list of recursive call during process. */
923 while (!fibheap_empty (heap)
924 && (cgraph_estimate_size_after_inlining (1, node, master_clone)
925 <= limit))
927 struct cgraph_edge *curr
928 = (struct cgraph_edge *) fibheap_extract_min (heap);
929 struct cgraph_node *cnode;
931 depth = 1;
932 for (cnode = curr->caller;
933 cnode->global.inlined_to; cnode = cnode->callers->caller)
934 if (node->decl == curr->callee->decl)
935 depth++;
936 if (depth > max_depth)
938 if (dump_file)
939 fprintf (dump_file,
940 " maximal depth reached\n");
941 continue;
944 if (max_count)
946 if (!cgraph_maybe_hot_edge_p (curr))
948 if (dump_file)
949 fprintf (dump_file, " Not inlining cold call\n");
950 continue;
952 if (curr->count * 100 / node->count < probability)
954 if (dump_file)
955 fprintf (dump_file,
956 " Probability of edge is too small\n");
957 continue;
961 if (dump_file)
963 fprintf (dump_file,
964 " Inlining call of depth %i", depth);
965 if (node->count)
967 fprintf (dump_file, " called approx. %.2f times per call",
968 (double)curr->count / node->count);
970 fprintf (dump_file, "\n");
972 cgraph_redirect_edge_callee (curr, master_clone);
973 cgraph_mark_inline_edge (curr, false, new_edges);
974 lookup_recursive_calls (node, curr->callee, heap);
975 n++;
977 if (!fibheap_empty (heap) && dump_file)
978 fprintf (dump_file, " Recursive inlining growth limit met.\n");
980 fibheap_delete (heap);
981 if (dump_file)
982 fprintf (dump_file,
983 "\n Inlined %i times, body grown from size %i to %i, time %i to %i\n", n,
984 master_clone->global.size, node->global.size,
985 master_clone->global.time, node->global.time);
987 /* Remove master clone we used for inlining. We rely that clones inlined
988 into master clone gets queued just before master clone so we don't
989 need recursion. */
990 for (node = cgraph_nodes; node != master_clone;
991 node = next)
993 next = node->next;
994 if (node->global.inlined_to == master_clone)
995 cgraph_remove_node (node);
997 cgraph_remove_node (master_clone);
998 /* FIXME: Recursive inlining actually reduces number of calls of the
999 function. At this place we should probably walk the function and
1000 inline clones and compensate the counts accordingly. This probably
1001 doesn't matter much in practice. */
1002 return n > 0;
1005 /* Set inline_failed for all callers of given function to REASON. */
1007 static void
1008 cgraph_set_inline_failed (struct cgraph_node *node,
1009 cgraph_inline_failed_t reason)
1011 struct cgraph_edge *e;
1013 if (dump_file)
1014 fprintf (dump_file, "Inlining failed: %s\n",
1015 cgraph_inline_failed_string (reason));
1016 for (e = node->callers; e; e = e->next_caller)
1017 if (e->inline_failed)
1018 e->inline_failed = reason;
1021 /* Given whole compilation unit estimate of INSNS, compute how large we can
1022 allow the unit to grow. */
1023 static int
1024 compute_max_insns (int insns)
1026 int max_insns = insns;
1027 if (max_insns < PARAM_VALUE (PARAM_LARGE_UNIT_INSNS))
1028 max_insns = PARAM_VALUE (PARAM_LARGE_UNIT_INSNS);
1030 return ((HOST_WIDEST_INT) max_insns
1031 * (100 + PARAM_VALUE (PARAM_INLINE_UNIT_GROWTH)) / 100);
1034 /* Compute badness of all edges in NEW_EDGES and add them to the HEAP. */
1035 static void
1036 add_new_edges_to_heap (fibheap_t heap, VEC (cgraph_edge_p, heap) *new_edges)
1038 while (VEC_length (cgraph_edge_p, new_edges) > 0)
1040 struct cgraph_edge *edge = VEC_pop (cgraph_edge_p, new_edges);
1042 gcc_assert (!edge->aux);
1043 if (edge->callee->local.inlinable
1044 && edge->inline_failed
1045 && cgraph_default_inline_p (edge->callee, &edge->inline_failed))
1046 edge->aux = fibheap_insert (heap, cgraph_edge_badness (edge, false), edge);
1051 /* We use greedy algorithm for inlining of small functions:
1052 All inline candidates are put into prioritized heap based on estimated
1053 growth of the overall number of instructions and then update the estimates.
1055 INLINED and INLINED_CALEES are just pointers to arrays large enough
1056 to be passed to cgraph_inlined_into and cgraph_inlined_callees. */
1058 static void
1059 cgraph_decide_inlining_of_small_functions (void)
1061 struct cgraph_node *node;
1062 struct cgraph_edge *edge;
1063 cgraph_inline_failed_t failed_reason;
1064 fibheap_t heap = fibheap_new ();
1065 bitmap updated_nodes = BITMAP_ALLOC (NULL);
1066 int min_size, max_size;
1067 VEC (cgraph_edge_p, heap) *new_indirect_edges = NULL;
1069 if (flag_indirect_inlining)
1070 new_indirect_edges = VEC_alloc (cgraph_edge_p, heap, 8);
1072 if (dump_file)
1073 fprintf (dump_file, "\nDeciding on smaller functions:\n");
1075 /* Put all inline candidates into the heap. */
1077 for (node = cgraph_nodes; node; node = node->next)
1079 if (!node->local.inlinable || !node->callers)
1080 continue;
1081 if (dump_file)
1082 fprintf (dump_file, "Considering inline candidate %s.\n", cgraph_node_name (node));
1084 node->global.estimated_growth = INT_MIN;
1085 if (!cgraph_default_inline_p (node, &failed_reason))
1087 cgraph_set_inline_failed (node, failed_reason);
1088 continue;
1091 for (edge = node->callers; edge; edge = edge->next_caller)
1092 if (edge->inline_failed)
1094 gcc_assert (!edge->aux);
1095 edge->aux = fibheap_insert (heap, cgraph_edge_badness (edge, false), edge);
1099 max_size = compute_max_insns (overall_size);
1100 min_size = overall_size;
1102 while (overall_size <= max_size
1103 && !fibheap_empty (heap))
1105 int old_size = overall_size;
1106 struct cgraph_node *where, *callee;
1107 int badness = fibheap_min_key (heap);
1108 int current_badness;
1109 int growth;
1110 cgraph_inline_failed_t not_good = CIF_OK;
1112 edge = (struct cgraph_edge *) fibheap_extract_min (heap);
1113 gcc_assert (edge->aux);
1114 edge->aux = NULL;
1115 if (!edge->inline_failed)
1116 continue;
1118 /* When updating the edge costs, we only decrease badness in the keys.
1119 When the badness increase, we keep the heap as it is and re-insert
1120 key now. */
1121 current_badness = cgraph_edge_badness (edge, false);
1122 gcc_assert (current_badness >= badness);
1123 if (current_badness != badness)
1125 edge->aux = fibheap_insert (heap, current_badness, edge);
1126 continue;
1129 callee = edge->callee;
1131 growth = (cgraph_estimate_size_after_inlining (1, edge->caller, edge->callee)
1132 - edge->caller->global.size);
1134 if (dump_file)
1136 fprintf (dump_file,
1137 "\nConsidering %s with %i size\n",
1138 cgraph_node_name (edge->callee),
1139 edge->callee->global.size);
1140 fprintf (dump_file,
1141 " to be inlined into %s in %s:%i\n"
1142 " Estimated growth after inlined into all callees is %+i insns.\n"
1143 " Estimated badness is %i, frequency %.2f.\n",
1144 cgraph_node_name (edge->caller),
1145 flag_wpa ? "unknown"
1146 : gimple_filename ((const_gimple) edge->call_stmt),
1147 flag_wpa ? -1 : gimple_lineno ((const_gimple) edge->call_stmt),
1148 cgraph_estimate_growth (edge->callee),
1149 badness,
1150 edge->frequency / (double)CGRAPH_FREQ_BASE);
1151 if (edge->count)
1152 fprintf (dump_file," Called "HOST_WIDEST_INT_PRINT_DEC"x\n", edge->count);
1153 if (dump_flags & TDF_DETAILS)
1154 cgraph_edge_badness (edge, true);
1157 /* When not having profile info ready we don't weight by any way the
1158 position of call in procedure itself. This means if call of
1159 function A from function B seems profitable to inline, the recursive
1160 call of function A in inline copy of A in B will look profitable too
1161 and we end up inlining until reaching maximal function growth. This
1162 is not good idea so prohibit the recursive inlining.
1164 ??? When the frequencies are taken into account we might not need this
1165 restriction.
1167 We need to be cureful here, in some testcases, e.g. directivec.c in
1168 libcpp, we can estimate self recursive function to have negative growth
1169 for inlining completely.
1171 if (!edge->count)
1173 where = edge->caller;
1174 while (where->global.inlined_to)
1176 if (where->decl == edge->callee->decl)
1177 break;
1178 where = where->callers->caller;
1180 if (where->global.inlined_to)
1182 edge->inline_failed
1183 = (edge->callee->local.disregard_inline_limits
1184 ? CIF_RECURSIVE_INLINING : CIF_UNSPECIFIED);
1185 if (dump_file)
1186 fprintf (dump_file, " inline_failed:Recursive inlining performed only for function itself.\n");
1187 continue;
1191 if (edge->callee->local.disregard_inline_limits)
1193 else if (!cgraph_maybe_hot_edge_p (edge))
1194 not_good = CIF_UNLIKELY_CALL;
1195 else if (!flag_inline_functions
1196 && !DECL_DECLARED_INLINE_P (edge->callee->decl))
1197 not_good = CIF_NOT_DECLARED_INLINED;
1198 else if (optimize_function_for_size_p (DECL_STRUCT_FUNCTION(edge->caller->decl)))
1199 not_good = CIF_OPTIMIZING_FOR_SIZE;
1200 if (not_good && growth > 0 && cgraph_estimate_growth (edge->callee) > 0)
1202 if (!cgraph_recursive_inlining_p (edge->caller, edge->callee,
1203 &edge->inline_failed))
1205 edge->inline_failed = not_good;
1206 if (dump_file)
1207 fprintf (dump_file, " inline_failed:%s.\n",
1208 cgraph_inline_failed_string (edge->inline_failed));
1210 continue;
1212 if (!cgraph_default_inline_p (edge->callee, &edge->inline_failed))
1214 if (!cgraph_recursive_inlining_p (edge->caller, edge->callee,
1215 &edge->inline_failed))
1217 if (dump_file)
1218 fprintf (dump_file, " inline_failed:%s.\n",
1219 cgraph_inline_failed_string (edge->inline_failed));
1221 continue;
1223 if (!tree_can_inline_p (edge))
1225 if (dump_file)
1226 fprintf (dump_file, " inline_failed:%s.\n",
1227 cgraph_inline_failed_string (edge->inline_failed));
1228 continue;
1230 if (cgraph_recursive_inlining_p (edge->caller, edge->callee,
1231 &edge->inline_failed))
1233 where = edge->caller;
1234 if (where->global.inlined_to)
1235 where = where->global.inlined_to;
1236 if (!cgraph_decide_recursive_inlining (where,
1237 flag_indirect_inlining
1238 ? &new_indirect_edges : NULL))
1239 continue;
1240 if (flag_indirect_inlining)
1241 add_new_edges_to_heap (heap, new_indirect_edges);
1242 update_all_callee_keys (heap, where, updated_nodes);
1244 else
1246 struct cgraph_node *callee;
1247 if (edge->call_stmt_cannot_inline_p
1248 || !cgraph_check_inline_limits (edge->caller, edge->callee,
1249 &edge->inline_failed, true))
1251 if (dump_file)
1252 fprintf (dump_file, " Not inlining into %s:%s.\n",
1253 cgraph_node_name (edge->caller),
1254 cgraph_inline_failed_string (edge->inline_failed));
1255 continue;
1257 callee = edge->callee;
1258 gcc_checking_assert (!callee->global.inlined_to);
1259 cgraph_mark_inline_edge (edge, true, &new_indirect_edges);
1260 if (flag_indirect_inlining)
1261 add_new_edges_to_heap (heap, new_indirect_edges);
1263 /* We inlined last offline copy to the body. This might lead
1264 to callees of function having fewer call sites and thus they
1265 may need updating. */
1266 if (callee->global.inlined_to)
1267 update_all_callee_keys (heap, callee, updated_nodes);
1268 else
1269 update_callee_keys (heap, edge->callee, updated_nodes);
1271 where = edge->caller;
1272 if (where->global.inlined_to)
1273 where = where->global.inlined_to;
1275 /* Our profitability metric can depend on local properties
1276 such as number of inlinable calls and size of the function body.
1277 After inlining these properties might change for the function we
1278 inlined into (since it's body size changed) and for the functions
1279 called by function we inlined (since number of it inlinable callers
1280 might change). */
1281 update_caller_keys (heap, where, updated_nodes);
1283 /* We removed one call of the function we just inlined. If offline
1284 copy is still needed, be sure to update the keys. */
1285 if (callee != where && !callee->global.inlined_to)
1286 update_caller_keys (heap, callee, updated_nodes);
1287 bitmap_clear (updated_nodes);
1289 if (dump_file)
1291 fprintf (dump_file,
1292 " Inlined into %s which now has size %i and self time %i,"
1293 "net change of %+i.\n",
1294 cgraph_node_name (edge->caller),
1295 edge->caller->global.time,
1296 edge->caller->global.size,
1297 overall_size - old_size);
1299 if (min_size > overall_size)
1301 min_size = overall_size;
1302 max_size = compute_max_insns (min_size);
1304 if (dump_file)
1305 fprintf (dump_file, "New minimal size reached: %i\n", min_size);
1308 while (!fibheap_empty (heap))
1310 int badness = fibheap_min_key (heap);
1312 edge = (struct cgraph_edge *) fibheap_extract_min (heap);
1313 gcc_assert (edge->aux);
1314 edge->aux = NULL;
1315 if (!edge->inline_failed)
1316 continue;
1317 #ifdef ENABLE_CHECKING
1318 gcc_assert (cgraph_edge_badness (edge, false) >= badness);
1319 #endif
1320 if (dump_file)
1322 fprintf (dump_file,
1323 "\nSkipping %s with %i size\n",
1324 cgraph_node_name (edge->callee),
1325 edge->callee->global.size);
1326 fprintf (dump_file,
1327 " called by %s in %s:%i\n"
1328 " Estimated growth after inlined into all callees is %+i insns.\n"
1329 " Estimated badness is %i, frequency %.2f.\n",
1330 cgraph_node_name (edge->caller),
1331 flag_wpa ? "unknown"
1332 : gimple_filename ((const_gimple) edge->call_stmt),
1333 flag_wpa ? -1 : gimple_lineno ((const_gimple) edge->call_stmt),
1334 cgraph_estimate_growth (edge->callee),
1335 badness,
1336 edge->frequency / (double)CGRAPH_FREQ_BASE);
1337 if (edge->count)
1338 fprintf (dump_file," Called "HOST_WIDEST_INT_PRINT_DEC"x\n", edge->count);
1339 if (dump_flags & TDF_DETAILS)
1340 cgraph_edge_badness (edge, true);
1342 if (!edge->callee->local.disregard_inline_limits && edge->inline_failed
1343 && !cgraph_recursive_inlining_p (edge->caller, edge->callee,
1344 &edge->inline_failed))
1345 edge->inline_failed = CIF_INLINE_UNIT_GROWTH_LIMIT;
1348 if (new_indirect_edges)
1349 VEC_free (cgraph_edge_p, heap, new_indirect_edges);
1350 fibheap_delete (heap);
1351 BITMAP_FREE (updated_nodes);
1354 /* Flatten NODE from the IPA inliner. */
1356 static void
1357 cgraph_flatten (struct cgraph_node *node)
1359 struct cgraph_edge *e;
1361 /* We shouldn't be called recursively when we are being processed. */
1362 gcc_assert (node->aux == NULL);
1364 node->aux = (void *)(size_t) INLINE_ALL;
1366 for (e = node->callees; e; e = e->next_callee)
1368 struct cgraph_node *orig_callee;
1370 if (e->call_stmt_cannot_inline_p)
1371 continue;
1373 if (!e->callee->analyzed)
1375 if (dump_file)
1376 fprintf (dump_file,
1377 "Not inlining: Function body not available.\n");
1378 continue;
1381 /* We've hit cycle? It is time to give up. */
1382 if (e->callee->aux)
1384 if (dump_file)
1385 fprintf (dump_file,
1386 "Not inlining %s into %s to avoid cycle.\n",
1387 cgraph_node_name (e->callee),
1388 cgraph_node_name (e->caller));
1389 e->inline_failed = CIF_RECURSIVE_INLINING;
1390 continue;
1393 /* When the edge is already inlined, we just need to recurse into
1394 it in order to fully flatten the leaves. */
1395 if (!e->inline_failed)
1397 cgraph_flatten (e->callee);
1398 continue;
1401 if (cgraph_recursive_inlining_p (node, e->callee, &e->inline_failed))
1403 if (dump_file)
1404 fprintf (dump_file, "Not inlining: recursive call.\n");
1405 continue;
1408 if (!tree_can_inline_p (e))
1410 if (dump_file)
1411 fprintf (dump_file, "Not inlining: %s",
1412 cgraph_inline_failed_string (e->inline_failed));
1413 continue;
1416 if (gimple_in_ssa_p (DECL_STRUCT_FUNCTION (node->decl))
1417 != gimple_in_ssa_p (DECL_STRUCT_FUNCTION (e->callee->decl)))
1419 if (dump_file)
1420 fprintf (dump_file, "Not inlining: SSA form does not match.\n");
1421 continue;
1424 /* Inline the edge and flatten the inline clone. Avoid
1425 recursing through the original node if the node was cloned. */
1426 if (dump_file)
1427 fprintf (dump_file, " Inlining %s into %s.\n",
1428 cgraph_node_name (e->callee),
1429 cgraph_node_name (e->caller));
1430 orig_callee = e->callee;
1431 cgraph_mark_inline_edge (e, true, NULL);
1432 if (e->callee != orig_callee)
1433 orig_callee->aux = (void *)(size_t) INLINE_ALL;
1434 cgraph_flatten (e->callee);
1435 if (e->callee != orig_callee)
1436 orig_callee->aux = NULL;
1439 node->aux = NULL;
1442 /* Decide on the inlining. We do so in the topological order to avoid
1443 expenses on updating data structures. */
1445 static unsigned int
1446 cgraph_decide_inlining (void)
1448 struct cgraph_node *node;
1449 int nnodes;
1450 struct cgraph_node **order =
1451 XCNEWVEC (struct cgraph_node *, cgraph_n_nodes);
1452 int old_size = 0;
1453 int i;
1454 int initial_size = 0;
1456 cgraph_remove_function_insertion_hook (function_insertion_hook_holder);
1457 if (in_lto_p && flag_indirect_inlining)
1458 ipa_update_after_lto_read ();
1459 if (flag_indirect_inlining)
1460 ipa_create_all_structures_for_iinln ();
1462 max_count = 0;
1463 max_benefit = 0;
1464 for (node = cgraph_nodes; node; node = node->next)
1465 if (node->analyzed)
1467 struct cgraph_edge *e;
1469 gcc_assert (inline_summary (node)->self_size == node->global.size);
1470 if (!DECL_EXTERNAL (node->decl))
1471 initial_size += node->global.size;
1472 for (e = node->callees; e; e = e->next_callee)
1473 if (max_count < e->count)
1474 max_count = e->count;
1475 if (max_benefit < inline_summary (node)->time_inlining_benefit)
1476 max_benefit = inline_summary (node)->time_inlining_benefit;
1478 gcc_assert (in_lto_p
1479 || !max_count
1480 || (profile_info && flag_branch_probabilities));
1481 overall_size = initial_size;
1483 nnodes = cgraph_postorder (order);
1485 if (dump_file)
1486 fprintf (dump_file,
1487 "\nDeciding on inlining. Starting with size %i.\n",
1488 initial_size);
1490 for (node = cgraph_nodes; node; node = node->next)
1491 node->aux = 0;
1493 if (dump_file)
1494 fprintf (dump_file, "\nFlattening functions:\n");
1496 /* In the first pass handle functions to be flattened. Do this with
1497 a priority so none of our later choices will make this impossible. */
1498 for (i = nnodes - 1; i >= 0; i--)
1500 node = order[i];
1502 /* Handle nodes to be flattened, but don't update overall unit
1503 size. Calling the incremental inliner here is lame,
1504 a simple worklist should be enough. What should be left
1505 here from the early inliner (if it runs) is cyclic cases.
1506 Ideally when processing callees we stop inlining at the
1507 entry of cycles, possibly cloning that entry point and
1508 try to flatten itself turning it into a self-recursive
1509 function. */
1510 if (lookup_attribute ("flatten",
1511 DECL_ATTRIBUTES (node->decl)) != NULL)
1513 if (dump_file)
1514 fprintf (dump_file,
1515 "Flattening %s\n", cgraph_node_name (node));
1516 cgraph_flatten (node);
1520 cgraph_decide_inlining_of_small_functions ();
1522 if (flag_inline_functions_called_once)
1524 if (dump_file)
1525 fprintf (dump_file, "\nDeciding on functions called once:\n");
1527 /* And finally decide what functions are called once. */
1528 for (i = nnodes - 1; i >= 0; i--)
1530 node = order[i];
1532 if (node->callers
1533 && !node->callers->next_caller
1534 && !node->global.inlined_to
1535 && cgraph_will_be_removed_from_program_if_no_direct_calls (node)
1536 && node->local.inlinable
1537 && cgraph_function_body_availability (node) >= AVAIL_AVAILABLE
1538 && node->callers->inline_failed
1539 && node->callers->caller != node
1540 && node->callers->caller->global.inlined_to != node
1541 && !node->callers->call_stmt_cannot_inline_p
1542 && !DECL_EXTERNAL (node->decl))
1544 cgraph_inline_failed_t reason;
1545 old_size = overall_size;
1546 if (dump_file)
1548 fprintf (dump_file,
1549 "\nConsidering %s size %i.\n",
1550 cgraph_node_name (node), node->global.size);
1551 fprintf (dump_file,
1552 " Called once from %s %i insns.\n",
1553 cgraph_node_name (node->callers->caller),
1554 node->callers->caller->global.size);
1557 if (cgraph_check_inline_limits (node->callers->caller, node,
1558 &reason, false))
1560 struct cgraph_node *caller = node->callers->caller;
1561 cgraph_mark_inline (node->callers);
1562 if (dump_file)
1563 fprintf (dump_file,
1564 " Inlined into %s which now has %i size"
1565 " for a net change of %+i size.\n",
1566 cgraph_node_name (caller),
1567 caller->global.size,
1568 overall_size - old_size);
1570 else
1572 if (dump_file)
1573 fprintf (dump_file,
1574 " Not inlining: %s.\n",
1575 cgraph_inline_failed_string (reason));
1581 /* Free ipa-prop structures if they are no longer needed. */
1582 if (flag_indirect_inlining)
1583 ipa_free_all_structures_after_iinln ();
1585 if (dump_file)
1586 fprintf (dump_file,
1587 "\nInlined %i calls, eliminated %i functions, "
1588 "size %i turned to %i size.\n\n",
1589 ncalls_inlined, nfunctions_inlined, initial_size,
1590 overall_size);
1591 free (order);
1592 return 0;
1595 /* Return true when N is leaf function. Accept cheap builtins
1596 in leaf functions. */
1598 static bool
1599 leaf_node_p (struct cgraph_node *n)
1601 struct cgraph_edge *e;
1602 for (e = n->callees; e; e = e->next_callee)
1603 if (!is_inexpensive_builtin (e->callee->decl))
1604 return false;
1605 return true;
1608 /* Decide on the inlining. We do so in the topological order to avoid
1609 expenses on updating data structures. */
1611 static bool
1612 cgraph_decide_inlining_incrementally (struct cgraph_node *node,
1613 enum inlining_mode mode)
1615 struct cgraph_edge *e;
1616 bool inlined = false;
1617 cgraph_inline_failed_t failed_reason;
1619 #ifdef ENABLE_CHECKING
1620 verify_cgraph_node (node);
1621 #endif
1623 if (mode != INLINE_ALWAYS_INLINE && mode != INLINE_SIZE_NORECURSIVE
1624 && lookup_attribute ("flatten", DECL_ATTRIBUTES (node->decl)) != NULL)
1626 if (dump_file)
1627 fprintf (dump_file, "Incrementally flattening %s\n",
1628 cgraph_node_name (node));
1629 mode = INLINE_ALL;
1632 /* First of all look for always inline functions. */
1633 if (mode != INLINE_SIZE_NORECURSIVE)
1634 for (e = node->callees; e; e = e->next_callee)
1636 if (!e->callee->local.disregard_inline_limits
1637 && (mode != INLINE_ALL || !e->callee->local.inlinable))
1638 continue;
1639 if (e->call_stmt_cannot_inline_p)
1640 continue;
1641 if (dump_file)
1642 fprintf (dump_file,
1643 "Considering to always inline inline candidate %s.\n",
1644 cgraph_node_name (e->callee));
1645 if (cgraph_recursive_inlining_p (node, e->callee, &e->inline_failed))
1647 if (dump_file)
1648 fprintf (dump_file, "Not inlining: recursive call.\n");
1649 continue;
1651 if (!tree_can_inline_p (e))
1653 if (dump_file)
1654 fprintf (dump_file,
1655 "Not inlining: %s",
1656 cgraph_inline_failed_string (e->inline_failed));
1657 continue;
1659 if (gimple_in_ssa_p (DECL_STRUCT_FUNCTION (node->decl))
1660 != gimple_in_ssa_p (DECL_STRUCT_FUNCTION (e->callee->decl)))
1662 if (dump_file)
1663 fprintf (dump_file, "Not inlining: SSA form does not match.\n");
1664 continue;
1666 if (!e->callee->analyzed)
1668 if (dump_file)
1669 fprintf (dump_file,
1670 "Not inlining: Function body no longer available.\n");
1671 continue;
1674 if (dump_file)
1675 fprintf (dump_file, " Inlining %s into %s.\n",
1676 cgraph_node_name (e->callee),
1677 cgraph_node_name (e->caller));
1678 cgraph_mark_inline (e);
1679 inlined = true;
1682 /* Now do the automatic inlining. */
1683 if (mode != INLINE_ALL && mode != INLINE_ALWAYS_INLINE
1684 /* Never inline regular functions into always-inline functions
1685 during incremental inlining. */
1686 && !node->local.disregard_inline_limits)
1688 bitmap visited = BITMAP_ALLOC (NULL);
1689 for (e = node->callees; e; e = e->next_callee)
1691 int allowed_growth = 0;
1692 if (!e->callee->local.inlinable
1693 || !e->inline_failed
1694 || e->callee->local.disregard_inline_limits)
1695 continue;
1696 /* We are inlining a function to all call-sites in node
1697 or to none. So visit each candidate only once. */
1698 if (!bitmap_set_bit (visited, e->callee->uid))
1699 continue;
1700 if (dump_file)
1701 fprintf (dump_file, "Considering inline candidate %s.\n",
1702 cgraph_node_name (e->callee));
1703 if (cgraph_recursive_inlining_p (node, e->callee, &e->inline_failed))
1705 if (dump_file)
1706 fprintf (dump_file, "Not inlining: recursive call.\n");
1707 continue;
1709 if (gimple_in_ssa_p (DECL_STRUCT_FUNCTION (node->decl))
1710 != gimple_in_ssa_p (DECL_STRUCT_FUNCTION (e->callee->decl)))
1712 if (dump_file)
1713 fprintf (dump_file,
1714 "Not inlining: SSA form does not match.\n");
1715 continue;
1718 if (cgraph_maybe_hot_edge_p (e) && leaf_node_p (e->callee)
1719 && optimize_function_for_speed_p (cfun))
1720 allowed_growth = PARAM_VALUE (PARAM_EARLY_INLINING_INSNS);
1722 /* When the function body would grow and inlining the function
1723 won't eliminate the need for offline copy of the function,
1724 don't inline. */
1725 if (((mode == INLINE_SIZE || mode == INLINE_SIZE_NORECURSIVE)
1726 || (!flag_inline_functions
1727 && !DECL_DECLARED_INLINE_P (e->callee->decl)))
1728 && (cgraph_estimate_size_after_inlining (1, e->caller, e->callee)
1729 > e->caller->global.size + allowed_growth)
1730 && cgraph_estimate_growth (e->callee) > allowed_growth)
1732 if (dump_file)
1733 fprintf (dump_file,
1734 "Not inlining: code size would grow by %i.\n",
1735 cgraph_estimate_size_after_inlining (1, e->caller,
1736 e->callee)
1737 - e->caller->global.size);
1738 continue;
1740 if (!cgraph_check_inline_limits (node, e->callee, &e->inline_failed,
1741 false)
1742 || e->call_stmt_cannot_inline_p)
1744 if (dump_file)
1745 fprintf (dump_file, "Not inlining: %s.\n",
1746 cgraph_inline_failed_string (e->inline_failed));
1747 continue;
1749 if (!e->callee->analyzed)
1751 if (dump_file)
1752 fprintf (dump_file,
1753 "Not inlining: Function body no longer available.\n");
1754 continue;
1756 if (!tree_can_inline_p (e))
1758 if (dump_file)
1759 fprintf (dump_file,
1760 "Not inlining: %s.",
1761 cgraph_inline_failed_string (e->inline_failed));
1762 continue;
1764 if (cgraph_default_inline_p (e->callee, &failed_reason))
1766 if (dump_file)
1767 fprintf (dump_file, " Inlining %s into %s.\n",
1768 cgraph_node_name (e->callee),
1769 cgraph_node_name (e->caller));
1770 cgraph_mark_inline (e);
1771 inlined = true;
1774 BITMAP_FREE (visited);
1776 return inlined;
1779 /* Because inlining might remove no-longer reachable nodes, we need to
1780 keep the array visible to garbage collector to avoid reading collected
1781 out nodes. */
1782 static int nnodes;
1783 static GTY ((length ("nnodes"))) struct cgraph_node **order;
1785 /* Do inlining of small functions. Doing so early helps profiling and other
1786 passes to be somewhat more effective and avoids some code duplication in
1787 later real inlining pass for testcases with very many function calls. */
1788 static unsigned int
1789 cgraph_early_inlining (void)
1791 struct cgraph_node *node = cgraph_node (current_function_decl);
1792 unsigned int todo = 0;
1793 int iterations = 0;
1795 if (seen_error ())
1796 return 0;
1798 if (!optimize
1799 || flag_no_inline
1800 || !flag_early_inlining)
1802 /* When not optimizing or not inlining inline only always-inline
1803 functions. */
1804 cgraph_decide_inlining_incrementally (node, INLINE_ALWAYS_INLINE);
1805 timevar_push (TV_INTEGRATION);
1806 todo |= optimize_inline_calls (current_function_decl);
1807 timevar_pop (TV_INTEGRATION);
1809 else
1811 if (lookup_attribute ("flatten",
1812 DECL_ATTRIBUTES (node->decl)) != NULL)
1814 if (dump_file)
1815 fprintf (dump_file,
1816 "Flattening %s\n", cgraph_node_name (node));
1817 cgraph_flatten (node);
1818 timevar_push (TV_INTEGRATION);
1819 todo |= optimize_inline_calls (current_function_decl);
1820 timevar_pop (TV_INTEGRATION);
1822 /* We iterate incremental inlining to get trivial cases of indirect
1823 inlining. */
1824 while (iterations < PARAM_VALUE (PARAM_EARLY_INLINER_MAX_ITERATIONS)
1825 && cgraph_decide_inlining_incrementally (node,
1826 iterations
1827 ? INLINE_SIZE_NORECURSIVE
1828 : INLINE_SIZE))
1830 timevar_push (TV_INTEGRATION);
1831 todo |= optimize_inline_calls (current_function_decl);
1832 iterations++;
1833 timevar_pop (TV_INTEGRATION);
1835 if (dump_file)
1836 fprintf (dump_file, "Iterations: %i\n", iterations);
1839 cfun->always_inline_functions_inlined = true;
1841 return todo;
1844 struct gimple_opt_pass pass_early_inline =
1847 GIMPLE_PASS,
1848 "einline", /* name */
1849 NULL, /* gate */
1850 cgraph_early_inlining, /* execute */
1851 NULL, /* sub */
1852 NULL, /* next */
1853 0, /* static_pass_number */
1854 TV_INLINE_HEURISTICS, /* tv_id */
1855 0, /* properties_required */
1856 0, /* properties_provided */
1857 0, /* properties_destroyed */
1858 0, /* todo_flags_start */
1859 TODO_dump_func /* todo_flags_finish */
1864 /* See if statement might disappear after inlining.
1865 0 - means not eliminated
1866 1 - half of statements goes away
1867 2 - for sure it is eliminated.
1868 We are not terribly sophisficated, basically looking for simple abstraction
1869 penalty wrappers. */
1871 static int
1872 eliminated_by_inlining_prob (gimple stmt)
1874 enum gimple_code code = gimple_code (stmt);
1875 switch (code)
1877 case GIMPLE_RETURN:
1878 return 2;
1879 case GIMPLE_ASSIGN:
1880 if (gimple_num_ops (stmt) != 2)
1881 return 0;
1883 /* Casts of parameters, loads from parameters passed by reference
1884 and stores to return value or parameters are often free after
1885 inlining dua to SRA and further combining.
1886 Assume that half of statements goes away. */
1887 if (gimple_assign_rhs_code (stmt) == CONVERT_EXPR
1888 || gimple_assign_rhs_code (stmt) == NOP_EXPR
1889 || gimple_assign_rhs_code (stmt) == VIEW_CONVERT_EXPR
1890 || gimple_assign_rhs_class (stmt) == GIMPLE_SINGLE_RHS)
1892 tree rhs = gimple_assign_rhs1 (stmt);
1893 tree lhs = gimple_assign_lhs (stmt);
1894 tree inner_rhs = rhs;
1895 tree inner_lhs = lhs;
1896 bool rhs_free = false;
1897 bool lhs_free = false;
1899 while (handled_component_p (inner_lhs)
1900 || TREE_CODE (inner_lhs) == MEM_REF)
1901 inner_lhs = TREE_OPERAND (inner_lhs, 0);
1902 while (handled_component_p (inner_rhs)
1903 || TREE_CODE (inner_rhs) == ADDR_EXPR
1904 || TREE_CODE (inner_rhs) == MEM_REF)
1905 inner_rhs = TREE_OPERAND (inner_rhs, 0);
1908 if (TREE_CODE (inner_rhs) == PARM_DECL
1909 || (TREE_CODE (inner_rhs) == SSA_NAME
1910 && SSA_NAME_IS_DEFAULT_DEF (inner_rhs)
1911 && TREE_CODE (SSA_NAME_VAR (inner_rhs)) == PARM_DECL))
1912 rhs_free = true;
1913 if (rhs_free && is_gimple_reg (lhs))
1914 lhs_free = true;
1915 if (((TREE_CODE (inner_lhs) == PARM_DECL
1916 || (TREE_CODE (inner_lhs) == SSA_NAME
1917 && SSA_NAME_IS_DEFAULT_DEF (inner_lhs)
1918 && TREE_CODE (SSA_NAME_VAR (inner_lhs)) == PARM_DECL))
1919 && inner_lhs != lhs)
1920 || TREE_CODE (inner_lhs) == RESULT_DECL
1921 || (TREE_CODE (inner_lhs) == SSA_NAME
1922 && TREE_CODE (SSA_NAME_VAR (inner_lhs)) == RESULT_DECL))
1923 lhs_free = true;
1924 if (lhs_free
1925 && (is_gimple_reg (rhs) || is_gimple_min_invariant (rhs)))
1926 rhs_free = true;
1927 if (lhs_free && rhs_free)
1928 return 1;
1930 return 0;
1931 default:
1932 return 0;
1936 /* Compute function body size parameters for NODE. */
1938 static void
1939 estimate_function_body_sizes (struct cgraph_node *node)
1941 gcov_type time = 0;
1942 gcov_type time_inlining_benefit = 0;
1943 /* Estimate static overhead for function prologue/epilogue and alignment. */
1944 int size = 2;
1945 /* Benefits are scaled by probability of elimination that is in range
1946 <0,2>. */
1947 int size_inlining_benefit = 2 * 2;
1948 basic_block bb;
1949 gimple_stmt_iterator bsi;
1950 struct function *my_function = DECL_STRUCT_FUNCTION (node->decl);
1951 tree arg;
1952 int freq;
1953 tree funtype = TREE_TYPE (node->decl);
1955 if (dump_file)
1956 fprintf (dump_file, "Analyzing function body size: %s\n",
1957 cgraph_node_name (node));
1959 gcc_assert (my_function && my_function->cfg);
1960 FOR_EACH_BB_FN (bb, my_function)
1962 freq = compute_call_stmt_bb_frequency (node->decl, bb);
1963 for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
1965 gimple stmt = gsi_stmt (bsi);
1966 int this_size = estimate_num_insns (stmt, &eni_size_weights);
1967 int this_time = estimate_num_insns (stmt, &eni_time_weights);
1968 int prob;
1970 if (dump_file && (dump_flags & TDF_DETAILS))
1972 fprintf (dump_file, " freq:%6i size:%3i time:%3i ",
1973 freq, this_size, this_time);
1974 print_gimple_stmt (dump_file, stmt, 0, 0);
1976 this_time *= freq;
1977 time += this_time;
1978 size += this_size;
1979 prob = eliminated_by_inlining_prob (stmt);
1980 if (prob == 1 && dump_file && (dump_flags & TDF_DETAILS))
1981 fprintf (dump_file, " 50%% will be eliminated by inlining\n");
1982 if (prob == 2 && dump_file && (dump_flags & TDF_DETAILS))
1983 fprintf (dump_file, " will eliminated by inlining\n");
1984 size_inlining_benefit += this_size * prob;
1985 time_inlining_benefit += this_time * prob;
1986 gcc_assert (time >= 0);
1987 gcc_assert (size >= 0);
1990 time = (time + CGRAPH_FREQ_BASE / 2) / CGRAPH_FREQ_BASE;
1991 time_inlining_benefit = ((time_inlining_benefit + CGRAPH_FREQ_BASE)
1992 / (CGRAPH_FREQ_BASE * 2));
1993 size_inlining_benefit = (size_inlining_benefit + 1) / 2;
1994 if (dump_file)
1995 fprintf (dump_file, "Overall function body time: %i-%i size: %i-%i\n",
1996 (int)time, (int)time_inlining_benefit,
1997 size, size_inlining_benefit);
1998 time_inlining_benefit += eni_time_weights.call_cost;
1999 size_inlining_benefit += eni_size_weights.call_cost;
2000 if (!VOID_TYPE_P (TREE_TYPE (funtype)))
2002 int cost = estimate_move_cost (TREE_TYPE (funtype));
2003 time_inlining_benefit += cost;
2004 size_inlining_benefit += cost;
2006 for (arg = DECL_ARGUMENTS (node->decl); arg; arg = DECL_CHAIN (arg))
2007 if (!VOID_TYPE_P (TREE_TYPE (arg)))
2009 int cost = estimate_move_cost (TREE_TYPE (arg));
2010 time_inlining_benefit += cost;
2011 size_inlining_benefit += cost;
2013 if (time_inlining_benefit > MAX_TIME)
2014 time_inlining_benefit = MAX_TIME;
2015 if (time > MAX_TIME)
2016 time = MAX_TIME;
2017 inline_summary (node)->self_time = time;
2018 inline_summary (node)->self_size = size;
2019 if (dump_file)
2020 fprintf (dump_file, "With function call overhead time: %i-%i size: %i-%i\n",
2021 (int)time, (int)time_inlining_benefit,
2022 size, size_inlining_benefit);
2023 inline_summary (node)->time_inlining_benefit = time_inlining_benefit;
2024 inline_summary (node)->size_inlining_benefit = size_inlining_benefit;
2027 /* Compute parameters of functions used by inliner. */
2028 unsigned int
2029 compute_inline_parameters (struct cgraph_node *node)
2031 HOST_WIDE_INT self_stack_size;
2033 gcc_assert (!node->global.inlined_to);
2035 /* Estimate the stack size for the function. But not at -O0
2036 because estimated_stack_frame_size is a quadratic problem. */
2037 self_stack_size = optimize ? estimated_stack_frame_size (node->decl) : 0;
2038 inline_summary (node)->estimated_self_stack_size = self_stack_size;
2039 node->global.estimated_stack_size = self_stack_size;
2040 node->global.stack_frame_offset = 0;
2042 /* Can this function be inlined at all? */
2043 node->local.inlinable = tree_inlinable_function_p (node->decl);
2044 if (node->local.inlinable && !node->local.disregard_inline_limits)
2045 node->local.disregard_inline_limits
2046 = DECL_DISREGARD_INLINE_LIMITS (node->decl);
2047 estimate_function_body_sizes (node);
2048 /* Inlining characteristics are maintained by the cgraph_mark_inline. */
2049 node->global.time = inline_summary (node)->self_time;
2050 node->global.size = inline_summary (node)->self_size;
2051 return 0;
2055 /* Compute parameters of functions used by inliner using
2056 current_function_decl. */
2057 static unsigned int
2058 compute_inline_parameters_for_current (void)
2060 compute_inline_parameters (cgraph_node (current_function_decl));
2061 return 0;
2064 struct gimple_opt_pass pass_inline_parameters =
2067 GIMPLE_PASS,
2068 "inline_param", /* name */
2069 NULL, /* gate */
2070 compute_inline_parameters_for_current,/* execute */
2071 NULL, /* sub */
2072 NULL, /* next */
2073 0, /* static_pass_number */
2074 TV_INLINE_HEURISTICS, /* tv_id */
2075 0, /* properties_required */
2076 0, /* properties_provided */
2077 0, /* properties_destroyed */
2078 0, /* todo_flags_start */
2079 0 /* todo_flags_finish */
2083 /* This function performs intraprocedural analyzis in NODE that is required to
2084 inline indirect calls. */
2085 static void
2086 inline_indirect_intraprocedural_analysis (struct cgraph_node *node)
2088 ipa_analyze_node (node);
2089 if (dump_file && (dump_flags & TDF_DETAILS))
2091 ipa_print_node_params (dump_file, node);
2092 ipa_print_node_jump_functions (dump_file, node);
2096 /* Note function body size. */
2097 static void
2098 analyze_function (struct cgraph_node *node)
2100 push_cfun (DECL_STRUCT_FUNCTION (node->decl));
2101 current_function_decl = node->decl;
2103 compute_inline_parameters (node);
2104 /* FIXME: We should remove the optimize check after we ensure we never run
2105 IPA passes when not optimizng. */
2106 if (flag_indirect_inlining && optimize)
2107 inline_indirect_intraprocedural_analysis (node);
2109 current_function_decl = NULL;
2110 pop_cfun ();
2113 /* Called when new function is inserted to callgraph late. */
2114 static void
2115 add_new_function (struct cgraph_node *node, void *data ATTRIBUTE_UNUSED)
2117 analyze_function (node);
2120 /* Note function body size. */
2121 static void
2122 inline_generate_summary (void)
2124 struct cgraph_node *node;
2126 function_insertion_hook_holder =
2127 cgraph_add_function_insertion_hook (&add_new_function, NULL);
2129 if (flag_indirect_inlining)
2131 ipa_register_cgraph_hooks ();
2132 ipa_check_create_node_params ();
2133 ipa_check_create_edge_args ();
2136 for (node = cgraph_nodes; node; node = node->next)
2137 if (node->analyzed)
2138 analyze_function (node);
2140 return;
2143 /* Apply inline plan to function. */
2144 static unsigned int
2145 inline_transform (struct cgraph_node *node)
2147 unsigned int todo = 0;
2148 struct cgraph_edge *e;
2149 bool inline_p = false;
2151 /* FIXME: Currently the passmanager is adding inline transform more than once to some
2152 clones. This needs revisiting after WPA cleanups. */
2153 if (cfun->after_inlining)
2154 return 0;
2156 /* We might need the body of this function so that we can expand
2157 it inline somewhere else. */
2158 if (cgraph_preserve_function_body_p (node->decl))
2159 save_inline_function_body (node);
2161 for (e = node->callees; e; e = e->next_callee)
2163 cgraph_redirect_edge_call_stmt_to_callee (e);
2164 if (!e->inline_failed || warn_inline)
2165 inline_p = true;
2168 if (inline_p)
2170 timevar_push (TV_INTEGRATION);
2171 todo = optimize_inline_calls (current_function_decl);
2172 timevar_pop (TV_INTEGRATION);
2174 cfun->always_inline_functions_inlined = true;
2175 cfun->after_inlining = true;
2176 return todo | execute_fixup_cfg ();
2179 /* Read inline summary. Jump functions are shared among ipa-cp
2180 and inliner, so when ipa-cp is active, we don't need to write them
2181 twice. */
2183 static void
2184 inline_read_summary (void)
2186 if (flag_indirect_inlining)
2188 ipa_register_cgraph_hooks ();
2189 if (!flag_ipa_cp)
2190 ipa_prop_read_jump_functions ();
2192 function_insertion_hook_holder =
2193 cgraph_add_function_insertion_hook (&add_new_function, NULL);
2196 /* Write inline summary for node in SET.
2197 Jump functions are shared among ipa-cp and inliner, so when ipa-cp is
2198 active, we don't need to write them twice. */
2200 static void
2201 inline_write_summary (cgraph_node_set set,
2202 varpool_node_set vset ATTRIBUTE_UNUSED)
2204 if (flag_indirect_inlining && !flag_ipa_cp)
2205 ipa_prop_write_jump_functions (set);
2208 /* When to run IPA inlining. Inlining of always-inline functions
2209 happens during early inlining. */
2211 static bool
2212 gate_cgraph_decide_inlining (void)
2214 /* ??? We'd like to skip this if not optimizing or not inlining as
2215 all always-inline functions have been processed by early
2216 inlining already. But this at least breaks EH with C++ as
2217 we need to unconditionally run fixup_cfg even at -O0.
2218 So leave it on unconditionally for now. */
2219 return 1;
2222 struct ipa_opt_pass_d pass_ipa_inline =
2225 IPA_PASS,
2226 "inline", /* name */
2227 gate_cgraph_decide_inlining, /* gate */
2228 cgraph_decide_inlining, /* execute */
2229 NULL, /* sub */
2230 NULL, /* next */
2231 0, /* static_pass_number */
2232 TV_INLINE_HEURISTICS, /* tv_id */
2233 0, /* properties_required */
2234 0, /* properties_provided */
2235 0, /* properties_destroyed */
2236 TODO_remove_functions, /* todo_flags_finish */
2237 TODO_dump_cgraph | TODO_dump_func
2238 | TODO_remove_functions | TODO_ggc_collect /* todo_flags_finish */
2240 inline_generate_summary, /* generate_summary */
2241 inline_write_summary, /* write_summary */
2242 inline_read_summary, /* read_summary */
2243 NULL, /* write_optimization_summary */
2244 NULL, /* read_optimization_summary */
2245 NULL, /* stmt_fixup */
2246 0, /* TODOs */
2247 inline_transform, /* function_transform */
2248 NULL, /* variable_transform */
2252 #include "gt-ipa-inline.h"