Emit .note.GNU-stack for hard-float linux targets.
[official-gcc.git] / gcc / graphite-scop-detection.c
blobed12fea801b5b0089988d55f82b9801e3efef026
1 /* Detection of Static Control Parts (SCoP) for Graphite.
2 Copyright (C) 2009-2020 Free Software Foundation, Inc.
3 Contributed by Sebastian Pop <sebastian.pop@amd.com> and
4 Tobias Grosser <grosser@fim.uni-passau.de>.
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3, or (at your option)
11 any later version.
13 GCC is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License 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 #define USES_ISL
24 #include "config.h"
26 #ifdef HAVE_isl
28 #include "system.h"
29 #include "coretypes.h"
30 #include "backend.h"
31 #include "cfghooks.h"
32 #include "domwalk.h"
33 #include "tree.h"
34 #include "gimple.h"
35 #include "ssa.h"
36 #include "fold-const.h"
37 #include "gimple-iterator.h"
38 #include "tree-cfg.h"
39 #include "tree-ssa-loop-manip.h"
40 #include "tree-ssa-loop-niter.h"
41 #include "tree-ssa-loop.h"
42 #include "tree-into-ssa.h"
43 #include "tree-ssa.h"
44 #include "cfgloop.h"
45 #include "tree-data-ref.h"
46 #include "tree-scalar-evolution.h"
47 #include "tree-pass.h"
48 #include "tree-ssa-propagate.h"
49 #include "gimple-pretty-print.h"
50 #include "cfganal.h"
51 #include "graphite.h"
53 class debug_printer
55 private:
56 FILE *dump_file;
58 public:
59 void
60 set_dump_file (FILE *f)
62 gcc_assert (f);
63 dump_file = f;
66 friend debug_printer &
67 operator<< (debug_printer &output, int i)
69 fprintf (output.dump_file, "%d", i);
70 return output;
72 friend debug_printer &
73 operator<< (debug_printer &output, const char *s)
75 fprintf (output.dump_file, "%s", s);
76 return output;
78 } dp;
80 #define DEBUG_PRINT(args) do \
81 { \
82 if (dump_file && (dump_flags & TDF_DETAILS)) { args; } \
83 } while (0)
85 /* Pretty print to FILE all the SCoPs in DOT format and mark them with
86 different colors. If there are not enough colors, paint the
87 remaining SCoPs in gray.
89 Special nodes:
90 - "*" after the node number denotes the entry of a SCoP,
91 - "#" after the node number denotes the exit of a SCoP,
92 - "()" around the node number denotes the entry or the
93 exit nodes of the SCOP. These are not part of SCoP. */
95 DEBUG_FUNCTION void
96 dot_all_sese (FILE *file, vec<sese_l>& scops)
98 /* Disable debugging while printing graph. */
99 dump_flags_t tmp_dump_flags = dump_flags;
100 dump_flags = TDF_NONE;
102 fprintf (file, "digraph all {\n");
104 basic_block bb;
105 FOR_ALL_BB_FN (bb, cfun)
107 int part_of_scop = false;
109 /* Use HTML for every bb label. So we are able to print bbs
110 which are part of two different SCoPs, with two different
111 background colors. */
112 fprintf (file, "%d [label=<\n <TABLE BORDER=\"0\" CELLBORDER=\"1\" ",
113 bb->index);
114 fprintf (file, "CELLSPACING=\"0\">\n");
116 /* Select color for SCoP. */
117 sese_l *region;
118 int i;
119 FOR_EACH_VEC_ELT (scops, i, region)
121 bool sese_in_region = bb_in_sese_p (bb, *region);
122 if (sese_in_region || (region->exit->dest == bb)
123 || (region->entry->dest == bb))
125 const char *color;
126 switch (i % 17)
128 case 0: /* red */
129 color = "#e41a1c";
130 break;
131 case 1: /* blue */
132 color = "#377eb8";
133 break;
134 case 2: /* green */
135 color = "#4daf4a";
136 break;
137 case 3: /* purple */
138 color = "#984ea3";
139 break;
140 case 4: /* orange */
141 color = "#ff7f00";
142 break;
143 case 5: /* yellow */
144 color = "#ffff33";
145 break;
146 case 6: /* brown */
147 color = "#a65628";
148 break;
149 case 7: /* rose */
150 color = "#f781bf";
151 break;
152 case 8:
153 color = "#8dd3c7";
154 break;
155 case 9:
156 color = "#ffffb3";
157 break;
158 case 10:
159 color = "#bebada";
160 break;
161 case 11:
162 color = "#fb8072";
163 break;
164 case 12:
165 color = "#80b1d3";
166 break;
167 case 13:
168 color = "#fdb462";
169 break;
170 case 14:
171 color = "#b3de69";
172 break;
173 case 15:
174 color = "#fccde5";
175 break;
176 case 16:
177 color = "#bc80bd";
178 break;
179 default: /* gray */
180 color = "#999999";
183 fprintf (file, " <TR><TD WIDTH=\"50\" BGCOLOR=\"%s\">",
184 color);
186 if (!sese_in_region)
187 fprintf (file, " (");
189 if (bb == region->entry->dest && bb == region->exit->dest)
190 fprintf (file, " %d*# ", bb->index);
191 else if (bb == region->entry->dest)
192 fprintf (file, " %d* ", bb->index);
193 else if (bb == region->exit->dest)
194 fprintf (file, " %d# ", bb->index);
195 else
196 fprintf (file, " %d ", bb->index);
198 fprintf (file, "{lp_%d}", bb->loop_father->num);
200 if (!sese_in_region)
201 fprintf (file, ")");
203 fprintf (file, "</TD></TR>\n");
204 part_of_scop = true;
208 if (!part_of_scop)
210 fprintf (file, " <TR><TD WIDTH=\"50\" BGCOLOR=\"#ffffff\">");
211 fprintf (file, " %d {lp_%d} </TD></TR>\n", bb->index,
212 bb->loop_father->num);
214 fprintf (file, " </TABLE>>, shape=box, style=\"setlinewidth(0)\"]\n");
217 FOR_ALL_BB_FN (bb, cfun)
219 edge e;
220 edge_iterator ei;
221 FOR_EACH_EDGE (e, ei, bb->succs)
222 fprintf (file, "%d -> %d;\n", bb->index, e->dest->index);
225 fputs ("}\n\n", file);
227 /* Enable debugging again. */
228 dump_flags = tmp_dump_flags;
231 /* Display SCoP on stderr. */
233 DEBUG_FUNCTION void
234 dot_sese (sese_l& scop)
236 vec<sese_l> scops;
237 scops.create (1);
239 if (scop)
240 scops.safe_push (scop);
242 dot_all_sese (stderr, scops);
244 scops.release ();
247 DEBUG_FUNCTION void
248 dot_cfg ()
250 vec<sese_l> scops;
251 scops.create (1);
252 dot_all_sese (stderr, scops);
253 scops.release ();
256 /* Returns a COND_EXPR statement when BB has a single predecessor, the
257 edge between BB and its predecessor is not a loop exit edge, and
258 the last statement of the single predecessor is a COND_EXPR. */
260 static gcond *
261 single_pred_cond_non_loop_exit (basic_block bb)
263 if (single_pred_p (bb))
265 edge e = single_pred_edge (bb);
266 basic_block pred = e->src;
267 gimple *stmt;
269 if (loop_depth (pred->loop_father) > loop_depth (bb->loop_father))
270 return NULL;
272 stmt = last_stmt (pred);
274 if (stmt && gimple_code (stmt) == GIMPLE_COND)
275 return as_a<gcond *> (stmt);
278 return NULL;
281 namespace
284 /* Build the maximal scop containing LOOPs and add it to SCOPS. */
286 class scop_detection
288 public:
289 scop_detection () : scops (vNULL) {}
291 ~scop_detection ()
293 scops.release ();
296 /* A marker for invalid sese_l. */
297 static sese_l invalid_sese;
299 /* Return the SCOPS in this SCOP_DETECTION. */
301 vec<sese_l>
302 get_scops ()
304 return scops;
307 /* Return an sese_l around the LOOP. */
309 sese_l get_sese (loop_p loop);
311 /* Merge scops at same loop depth and returns the new sese.
312 Returns a new SESE when merge was successful, INVALID_SESE otherwise. */
314 sese_l merge_sese (sese_l first, sese_l second) const;
316 /* Build scop outer->inner if possible. */
318 void build_scop_depth (loop_p loop);
320 /* Return true when BEGIN is the preheader edge of a loop with a single exit
321 END. */
323 static bool region_has_one_loop (sese_l s);
325 /* Add to SCOPS a scop starting at SCOP_BEGIN and ending at SCOP_END. */
327 void add_scop (sese_l s);
329 /* Returns true if S1 subsumes/surrounds S2. */
330 static bool subsumes (sese_l s1, sese_l s2);
332 /* Remove a SCoP which is subsumed by S1. */
333 void remove_subscops (sese_l s1);
335 /* Returns true if S1 intersects with S2. Since we already know that S1 does
336 not subsume S2 or vice-versa, we only check for entry bbs. */
338 static bool intersects (sese_l s1, sese_l s2);
340 /* Remove one of the scops when it intersects with any other. */
342 void remove_intersecting_scops (sese_l s1);
344 /* Return true when a statement in SCOP cannot be represented by Graphite. */
346 bool harmful_loop_in_region (sese_l scop) const;
348 /* Return true only when STMT is simple enough for being handled by Graphite.
349 This depends on SCOP, as the parameters are initialized relatively to
350 this basic block, the linear functions are initialized based on the
351 outermost loop containing STMT inside the SCOP. BB is the place where we
352 try to evaluate the STMT. */
354 bool stmt_simple_for_scop_p (sese_l scop, gimple *stmt,
355 basic_block bb) const;
357 /* Something like "n * m" is not allowed. */
359 static bool graphite_can_represent_init (tree e);
361 /* Return true when SCEV can be represented in the polyhedral model.
363 An expression can be represented, if it can be expressed as an
364 affine expression. For loops (i, j) and parameters (m, n) all
365 affine expressions are of the form:
367 x1 * i + x2 * j + x3 * m + x4 * n + x5 * 1 where x1..x5 element of Z
369 1 i + 20 j + (-2) m + 25
371 Something like "i * n" or "n * m" is not allowed. */
373 static bool graphite_can_represent_scev (sese_l scop, tree scev);
375 /* Return true when EXPR can be represented in the polyhedral model.
377 This means an expression can be represented, if it is linear with respect
378 to the loops and the strides are non parametric. LOOP is the place where
379 the expr will be evaluated. SCOP defines the region we analyse. */
381 static bool graphite_can_represent_expr (sese_l scop, loop_p loop,
382 tree expr);
384 /* Return true if the data references of STMT can be represented by Graphite.
385 We try to analyze the data references in a loop contained in the SCOP. */
387 static bool stmt_has_simple_data_refs_p (sese_l scop, gimple *stmt);
389 /* Remove the close phi node at GSI and replace its rhs with the rhs
390 of PHI. */
392 static void remove_duplicate_close_phi (gphi *phi, gphi_iterator *gsi);
394 /* Returns true when Graphite can represent LOOP in SCOP.
395 FIXME: For the moment, graphite cannot be used on loops that iterate using
396 induction variables that wrap. */
398 static bool can_represent_loop (loop_p loop, sese_l scop);
400 /* Returns the number of pbbs that are in loops contained in SCOP. */
402 static int nb_pbbs_in_loops (scop_p scop);
404 private:
405 vec<sese_l> scops;
408 sese_l scop_detection::invalid_sese (NULL, NULL);
410 /* Return an sese_l around the LOOP. */
412 sese_l
413 scop_detection::get_sese (loop_p loop)
415 if (!loop)
416 return invalid_sese;
418 edge scop_begin = loop_preheader_edge (loop);
419 edge scop_end = single_exit (loop);
420 if (!scop_end || (scop_end->flags & (EDGE_COMPLEX|EDGE_FAKE)))
421 return invalid_sese;
423 return sese_l (scop_begin, scop_end);
426 /* Merge scops at same loop depth and returns the new sese.
427 Returns a new SESE when merge was successful, INVALID_SESE otherwise. */
429 sese_l
430 scop_detection::merge_sese (sese_l first, sese_l second) const
432 /* In the trivial case first/second may be NULL. */
433 if (!first)
434 return second;
435 if (!second)
436 return first;
438 DEBUG_PRINT (dp << "[scop-detection] try merging sese s1: ";
439 print_sese (dump_file, first);
440 dp << "[scop-detection] try merging sese s2: ";
441 print_sese (dump_file, second));
443 auto_bitmap worklist, in_sese_region;
444 bitmap_set_bit (worklist, get_entry_bb (first)->index);
445 bitmap_set_bit (worklist, get_exit_bb (first)->index);
446 bitmap_set_bit (worklist, get_entry_bb (second)->index);
447 bitmap_set_bit (worklist, get_exit_bb (second)->index);
448 edge entry = NULL, exit = NULL;
450 /* We can optimize the case of adding a loop entry dest or exit
451 src to the worklist (for single-exit loops) by skipping
452 directly to the exit dest / entry src. in_sese_region
453 doesn't have to cover all blocks in the region but merely
454 its border it acts more like a visited bitmap. */
457 int index = bitmap_first_set_bit (worklist);
458 bitmap_clear_bit (worklist, index);
459 basic_block bb = BASIC_BLOCK_FOR_FN (cfun, index);
460 edge_iterator ei;
461 edge e;
463 /* With fake exit edges we can end up with no possible exit. */
464 if (index == EXIT_BLOCK)
466 DEBUG_PRINT (dp << "[scop-detection-fail] cannot merge seses.\n");
467 return invalid_sese;
470 bitmap_set_bit (in_sese_region, bb->index);
472 basic_block dom = get_immediate_dominator (CDI_DOMINATORS, bb);
473 FOR_EACH_EDGE (e, ei, bb->preds)
474 if (e->src == dom
475 && (! entry
476 || dominated_by_p (CDI_DOMINATORS, entry->dest, bb)))
478 if (entry
479 && ! bitmap_bit_p (in_sese_region, entry->src->index))
480 bitmap_set_bit (worklist, entry->src->index);
481 entry = e;
483 else if (! bitmap_bit_p (in_sese_region, e->src->index))
484 bitmap_set_bit (worklist, e->src->index);
486 basic_block pdom = get_immediate_dominator (CDI_POST_DOMINATORS, bb);
487 FOR_EACH_EDGE (e, ei, bb->succs)
488 if (e->dest == pdom
489 && (! exit
490 || dominated_by_p (CDI_POST_DOMINATORS, exit->src, bb)))
492 if (exit
493 && ! bitmap_bit_p (in_sese_region, exit->dest->index))
494 bitmap_set_bit (worklist, exit->dest->index);
495 exit = e;
497 else if (! bitmap_bit_p (in_sese_region, e->dest->index))
498 bitmap_set_bit (worklist, e->dest->index);
500 while (! bitmap_empty_p (worklist));
502 sese_l combined (entry, exit);
504 DEBUG_PRINT (dp << "[merged-sese] s1: "; print_sese (dump_file, combined));
506 return combined;
509 /* Build scop outer->inner if possible. */
511 void
512 scop_detection::build_scop_depth (loop_p loop)
514 sese_l s = invalid_sese;
515 loop = loop->inner;
516 while (loop)
518 sese_l next = get_sese (loop);
519 if (! next
520 || harmful_loop_in_region (next))
522 if (s)
523 add_scop (s);
524 build_scop_depth (loop);
525 s = invalid_sese;
527 else if (! s)
528 s = next;
529 else
531 sese_l combined = merge_sese (s, next);
532 if (! combined
533 || harmful_loop_in_region (combined))
535 add_scop (s);
536 s = next;
538 else
539 s = combined;
541 loop = loop->next;
543 if (s)
544 add_scop (s);
547 /* Returns true when Graphite can represent LOOP in SCOP.
548 FIXME: For the moment, graphite cannot be used on loops that iterate using
549 induction variables that wrap. */
551 bool
552 scop_detection::can_represent_loop (loop_p loop, sese_l scop)
554 tree niter;
555 struct tree_niter_desc niter_desc;
557 /* We can only handle do {} while () style loops correctly. */
558 edge exit = single_exit (loop);
559 if (!exit
560 || !single_pred_p (loop->latch)
561 || exit->src != single_pred (loop->latch)
562 || !empty_block_p (loop->latch))
563 return false;
565 return !(loop_preheader_edge (loop)->flags & EDGE_IRREDUCIBLE_LOOP)
566 && number_of_iterations_exit (loop, single_exit (loop), &niter_desc, false)
567 && niter_desc.control.no_overflow
568 && (niter = number_of_latch_executions (loop))
569 && !chrec_contains_undetermined (niter)
570 && graphite_can_represent_expr (scop, loop, niter);
573 /* Return true when BEGIN is the preheader edge of a loop with a single exit
574 END. */
576 bool
577 scop_detection::region_has_one_loop (sese_l s)
579 edge begin = s.entry;
580 edge end = s.exit;
581 /* Check for a single perfectly nested loop. */
582 if (begin->dest->loop_father->inner)
583 return false;
585 /* Otherwise, check whether we have adjacent loops. */
586 return (single_pred_p (end->src)
587 && begin->dest->loop_father == single_pred (end->src)->loop_father);
590 /* Add to SCOPS a scop starting at SCOP_BEGIN and ending at SCOP_END. */
592 void
593 scop_detection::add_scop (sese_l s)
595 gcc_assert (s);
597 /* If the exit edge is fake discard the SCoP for now as we're removing the
598 fake edges again after analysis. */
599 if (s.exit->flags & EDGE_FAKE)
601 DEBUG_PRINT (dp << "[scop-detection-fail] Discarding infinite loop SCoP: ";
602 print_sese (dump_file, s));
603 return;
606 /* Include the BB with the loop-closed SSA PHI nodes, we need this
607 block in the region for code-generating out-of-SSA copies.
608 canonicalize_loop_closed_ssa makes sure that is in proper shape. */
609 if (s.exit->dest != EXIT_BLOCK_PTR_FOR_FN (cfun)
610 && loop_exit_edge_p (s.exit->src->loop_father, s.exit))
612 gcc_assert (single_pred_p (s.exit->dest)
613 && single_succ_p (s.exit->dest)
614 && sese_trivially_empty_bb_p (s.exit->dest));
615 s.exit = single_succ_edge (s.exit->dest);
618 /* Do not add scops with only one loop. */
619 if (region_has_one_loop (s))
621 DEBUG_PRINT (dp << "[scop-detection-fail] Discarding one loop SCoP: ";
622 print_sese (dump_file, s));
623 return;
626 if (get_exit_bb (s) == EXIT_BLOCK_PTR_FOR_FN (cfun))
628 DEBUG_PRINT (dp << "[scop-detection-fail] "
629 << "Discarding SCoP exiting to return: ";
630 print_sese (dump_file, s));
631 return;
634 /* Remove all the scops which are subsumed by s. */
635 remove_subscops (s);
637 /* Remove intersecting scops. FIXME: It will be a good idea to keep
638 the non-intersecting part of the scop already in the list. */
639 remove_intersecting_scops (s);
641 scops.safe_push (s);
642 DEBUG_PRINT (dp << "[scop-detection] Adding SCoP: "; print_sese (dump_file, s));
645 /* Return true when a statement in SCOP cannot be represented by Graphite. */
647 bool
648 scop_detection::harmful_loop_in_region (sese_l scop) const
650 basic_block exit_bb = get_exit_bb (scop);
651 basic_block entry_bb = get_entry_bb (scop);
653 DEBUG_PRINT (dp << "[checking-harmful-bbs] ";
654 print_sese (dump_file, scop));
655 gcc_assert (dominated_by_p (CDI_DOMINATORS, exit_bb, entry_bb));
657 auto_vec<basic_block> worklist;
658 auto_bitmap loops;
660 worklist.safe_push (entry_bb);
661 while (! worklist.is_empty ())
663 basic_block bb = worklist.pop ();
664 DEBUG_PRINT (dp << "Visiting bb_" << bb->index << "\n");
666 /* The basic block should not be part of an irreducible loop. */
667 if (bb->flags & BB_IRREDUCIBLE_LOOP)
668 return true;
670 /* Check for unstructured control flow: CFG not generated by structured
671 if-then-else. */
672 if (bb->succs->length () > 1)
674 edge e;
675 edge_iterator ei;
676 FOR_EACH_EDGE (e, ei, bb->succs)
677 if (!dominated_by_p (CDI_POST_DOMINATORS, bb, e->dest)
678 && !dominated_by_p (CDI_DOMINATORS, e->dest, bb))
679 return true;
682 /* Collect all loops in the current region. */
683 loop_p loop = bb->loop_father;
684 if (loop_in_sese_p (loop, scop))
685 bitmap_set_bit (loops, loop->num);
687 /* Check for harmful statements in basic blocks part of the region. */
688 for (gimple_stmt_iterator gsi = gsi_start_bb (bb);
689 !gsi_end_p (gsi); gsi_next (&gsi))
690 if (!stmt_simple_for_scop_p (scop, gsi_stmt (gsi), bb))
691 return true;
693 for (basic_block dom = first_dom_son (CDI_DOMINATORS, bb);
694 dom;
695 dom = next_dom_son (CDI_DOMINATORS, dom))
696 if (dom != scop.exit->dest)
697 worklist.safe_push (dom);
700 /* Go through all loops and check that they are still valid in the combined
701 scop. */
702 unsigned j;
703 bitmap_iterator bi;
704 EXECUTE_IF_SET_IN_BITMAP (loops, 0, j, bi)
706 loop_p loop = (*current_loops->larray)[j];
707 gcc_assert (loop->num == (int) j);
709 /* Check if the loop nests are to be optimized for speed. */
710 if (! loop->inner
711 && ! optimize_loop_for_speed_p (loop))
713 DEBUG_PRINT (dp << "[scop-detection-fail] loop_"
714 << loop->num << " is not on a hot path.\n");
715 return true;
718 if (! can_represent_loop (loop, scop))
720 DEBUG_PRINT (dp << "[scop-detection-fail] cannot represent loop_"
721 << loop->num << "\n");
722 return true;
725 /* Check if all loop nests have at least one data reference.
726 ??? This check is expensive and loops premature at this point.
727 If important to retain we can pre-compute this for all innermost
728 loops and reject those when we build a SESE region for a loop
729 during SESE discovery. */
730 if (! loop->inner
731 && ! loop_nest_has_data_refs (loop))
733 DEBUG_PRINT (dp << "[scop-detection-fail] loop_" << loop->num
734 << "does not have any data reference.\n");
735 return true;
739 return false;
742 /* Returns true if S1 subsumes/surrounds S2. */
743 bool
744 scop_detection::subsumes (sese_l s1, sese_l s2)
746 if (dominated_by_p (CDI_DOMINATORS, get_entry_bb (s2),
747 get_entry_bb (s1))
748 && dominated_by_p (CDI_POST_DOMINATORS, s2.exit->dest,
749 s1.exit->dest))
750 return true;
751 return false;
754 /* Remove a SCoP which is subsumed by S1. */
755 void
756 scop_detection::remove_subscops (sese_l s1)
758 int j;
759 sese_l *s2;
760 FOR_EACH_VEC_ELT_REVERSE (scops, j, s2)
762 if (subsumes (s1, *s2))
764 DEBUG_PRINT (dp << "Removing sub-SCoP";
765 print_sese (dump_file, *s2));
766 scops.unordered_remove (j);
771 /* Returns true if S1 intersects with S2. Since we already know that S1 does
772 not subsume S2 or vice-versa, we only check for entry bbs. */
774 bool
775 scop_detection::intersects (sese_l s1, sese_l s2)
777 if (dominated_by_p (CDI_DOMINATORS, get_entry_bb (s2),
778 get_entry_bb (s1))
779 && !dominated_by_p (CDI_DOMINATORS, get_entry_bb (s2),
780 get_exit_bb (s1)))
781 return true;
782 if ((s1.exit == s2.entry) || (s2.exit == s1.entry))
783 return true;
785 return false;
788 /* Remove one of the scops when it intersects with any other. */
790 void
791 scop_detection::remove_intersecting_scops (sese_l s1)
793 int j;
794 sese_l *s2;
795 FOR_EACH_VEC_ELT_REVERSE (scops, j, s2)
797 if (intersects (s1, *s2))
799 DEBUG_PRINT (dp << "Removing intersecting SCoP";
800 print_sese (dump_file, *s2);
801 dp << "Intersects with:";
802 print_sese (dump_file, s1));
803 scops.unordered_remove (j);
808 /* Something like "n * m" is not allowed. */
810 bool
811 scop_detection::graphite_can_represent_init (tree e)
813 switch (TREE_CODE (e))
815 case POLYNOMIAL_CHREC:
816 return graphite_can_represent_init (CHREC_LEFT (e))
817 && graphite_can_represent_init (CHREC_RIGHT (e));
819 case MULT_EXPR:
820 if (chrec_contains_symbols (TREE_OPERAND (e, 0)))
821 return graphite_can_represent_init (TREE_OPERAND (e, 0))
822 && tree_fits_shwi_p (TREE_OPERAND (e, 1));
823 else
824 return graphite_can_represent_init (TREE_OPERAND (e, 1))
825 && tree_fits_shwi_p (TREE_OPERAND (e, 0));
827 case PLUS_EXPR:
828 case POINTER_PLUS_EXPR:
829 case MINUS_EXPR:
830 return graphite_can_represent_init (TREE_OPERAND (e, 0))
831 && graphite_can_represent_init (TREE_OPERAND (e, 1));
833 case NEGATE_EXPR:
834 case BIT_NOT_EXPR:
835 CASE_CONVERT:
836 case NON_LVALUE_EXPR:
837 return graphite_can_represent_init (TREE_OPERAND (e, 0));
839 default:
840 break;
843 return true;
846 /* Return true when SCEV can be represented in the polyhedral model.
848 An expression can be represented, if it can be expressed as an
849 affine expression. For loops (i, j) and parameters (m, n) all
850 affine expressions are of the form:
852 x1 * i + x2 * j + x3 * m + x4 * n + x5 * 1 where x1..x5 element of Z
854 1 i + 20 j + (-2) m + 25
856 Something like "i * n" or "n * m" is not allowed. */
858 bool
859 scop_detection::graphite_can_represent_scev (sese_l scop, tree scev)
861 if (chrec_contains_undetermined (scev))
862 return false;
864 switch (TREE_CODE (scev))
866 case NEGATE_EXPR:
867 case BIT_NOT_EXPR:
868 CASE_CONVERT:
869 case NON_LVALUE_EXPR:
870 return graphite_can_represent_scev (scop, TREE_OPERAND (scev, 0));
872 case PLUS_EXPR:
873 case POINTER_PLUS_EXPR:
874 case MINUS_EXPR:
875 return graphite_can_represent_scev (scop, TREE_OPERAND (scev, 0))
876 && graphite_can_represent_scev (scop, TREE_OPERAND (scev, 1));
878 case MULT_EXPR:
879 return !CONVERT_EXPR_CODE_P (TREE_CODE (TREE_OPERAND (scev, 0)))
880 && !CONVERT_EXPR_CODE_P (TREE_CODE (TREE_OPERAND (scev, 1)))
881 && !(chrec_contains_symbols (TREE_OPERAND (scev, 0))
882 && chrec_contains_symbols (TREE_OPERAND (scev, 1)))
883 && graphite_can_represent_init (scev)
884 && graphite_can_represent_scev (scop, TREE_OPERAND (scev, 0))
885 && graphite_can_represent_scev (scop, TREE_OPERAND (scev, 1));
887 case POLYNOMIAL_CHREC:
888 /* Check for constant strides. With a non constant stride of
889 'n' we would have a value of 'iv * n'. Also check that the
890 initial value can represented: for example 'n * m' cannot be
891 represented. */
892 gcc_assert (loop_in_sese_p (get_loop (cfun,
893 CHREC_VARIABLE (scev)), scop));
894 if (!evolution_function_right_is_integer_cst (scev)
895 || !graphite_can_represent_init (scev))
896 return false;
897 return graphite_can_represent_scev (scop, CHREC_LEFT (scev));
899 case ADDR_EXPR:
900 /* We cannot encode addresses for ISL. */
901 return false;
903 default:
904 break;
907 /* Only affine functions can be represented. */
908 if (tree_contains_chrecs (scev, NULL) || !scev_is_linear_expression (scev))
909 return false;
911 return true;
914 /* Return true when EXPR can be represented in the polyhedral model.
916 This means an expression can be represented, if it is linear with respect to
917 the loops and the strides are non parametric. LOOP is the place where the
918 expr will be evaluated. SCOP defines the region we analyse. */
920 bool
921 scop_detection::graphite_can_represent_expr (sese_l scop, loop_p loop,
922 tree expr)
924 tree scev = cached_scalar_evolution_in_region (scop, loop, expr);
925 return graphite_can_represent_scev (scop, scev);
928 /* Return true if the data references of STMT can be represented by Graphite.
929 We try to analyze the data references in a loop contained in the SCOP. */
931 bool
932 scop_detection::stmt_has_simple_data_refs_p (sese_l scop, gimple *stmt)
934 edge nest = scop.entry;
935 loop_p loop = loop_containing_stmt (stmt);
936 if (!loop_in_sese_p (loop, scop))
937 loop = NULL;
939 auto_vec<data_reference_p> drs;
940 if (! graphite_find_data_references_in_stmt (nest, loop, stmt, &drs))
941 return false;
943 int j;
944 data_reference_p dr;
945 FOR_EACH_VEC_ELT (drs, j, dr)
947 for (unsigned i = 0; i < DR_NUM_DIMENSIONS (dr); ++i)
948 if (! graphite_can_represent_scev (scop, DR_ACCESS_FN (dr, i)))
949 return false;
952 return true;
955 /* GIMPLE_ASM and GIMPLE_CALL may embed arbitrary side effects.
956 Calls have side-effects, except those to const or pure
957 functions. */
959 static bool
960 stmt_has_side_effects (gimple *stmt)
962 if (gimple_has_volatile_ops (stmt)
963 || (gimple_code (stmt) == GIMPLE_CALL
964 && !(gimple_call_flags (stmt) & (ECF_CONST | ECF_PURE)))
965 || (gimple_code (stmt) == GIMPLE_ASM))
967 DEBUG_PRINT (dp << "[scop-detection-fail] "
968 << "Statement has side-effects:\n";
969 print_gimple_stmt (dump_file, stmt, 0, TDF_VOPS | TDF_MEMSYMS));
970 return true;
972 return false;
975 /* Return true only when STMT is simple enough for being handled by Graphite.
976 This depends on SCOP, as the parameters are initialized relatively to
977 this basic block, the linear functions are initialized based on the outermost
978 loop containing STMT inside the SCOP. BB is the place where we try to
979 evaluate the STMT. */
981 bool
982 scop_detection::stmt_simple_for_scop_p (sese_l scop, gimple *stmt,
983 basic_block bb) const
985 gcc_assert (scop);
987 if (is_gimple_debug (stmt))
988 return true;
990 if (stmt_has_side_effects (stmt))
991 return false;
993 if (!stmt_has_simple_data_refs_p (scop, stmt))
995 DEBUG_PRINT (dp << "[scop-detection-fail] "
996 << "Graphite cannot handle data-refs in stmt:\n";
997 print_gimple_stmt (dump_file, stmt, 0, TDF_VOPS|TDF_MEMSYMS););
998 return false;
1001 switch (gimple_code (stmt))
1003 case GIMPLE_LABEL:
1004 return true;
1006 case GIMPLE_COND:
1008 /* We can handle all binary comparisons. Inequalities are
1009 also supported as they can be represented with union of
1010 polyhedra. */
1011 enum tree_code code = gimple_cond_code (stmt);
1012 if (!(code == LT_EXPR
1013 || code == GT_EXPR
1014 || code == LE_EXPR
1015 || code == GE_EXPR
1016 || code == EQ_EXPR
1017 || code == NE_EXPR))
1019 DEBUG_PRINT (dp << "[scop-detection-fail] "
1020 << "Graphite cannot handle cond stmt:\n";
1021 print_gimple_stmt (dump_file, stmt, 0,
1022 TDF_VOPS | TDF_MEMSYMS));
1023 return false;
1026 loop_p loop = bb->loop_father;
1027 for (unsigned i = 0; i < 2; ++i)
1029 tree op = gimple_op (stmt, i);
1030 if (!graphite_can_represent_expr (scop, loop, op)
1031 /* We can only constrain on integer type. */
1032 || ! INTEGRAL_TYPE_P (TREE_TYPE (op)))
1034 DEBUG_PRINT (dp << "[scop-detection-fail] "
1035 << "Graphite cannot represent stmt:\n";
1036 print_gimple_stmt (dump_file, stmt, 0,
1037 TDF_VOPS | TDF_MEMSYMS));
1038 return false;
1042 return true;
1045 case GIMPLE_ASSIGN:
1046 case GIMPLE_CALL:
1048 tree op, lhs = gimple_get_lhs (stmt);
1049 ssa_op_iter i;
1050 /* If we are not going to instantiate the stmt do not require
1051 its operands to be instantiatable at this point. */
1052 if (lhs
1053 && TREE_CODE (lhs) == SSA_NAME
1054 && scev_analyzable_p (lhs, scop))
1055 return true;
1056 /* Verify that if we can analyze operands at their def site we
1057 also can represent them when analyzed at their uses. */
1058 FOR_EACH_SSA_TREE_OPERAND (op, stmt, i, SSA_OP_USE)
1059 if (scev_analyzable_p (op, scop)
1060 && chrec_contains_undetermined
1061 (cached_scalar_evolution_in_region (scop,
1062 bb->loop_father, op)))
1064 DEBUG_PRINT (dp << "[scop-detection-fail] "
1065 << "Graphite cannot code-gen stmt:\n";
1066 print_gimple_stmt (dump_file, stmt, 0,
1067 TDF_VOPS | TDF_MEMSYMS));
1068 return false;
1070 return true;
1073 default:
1074 /* These nodes cut a new scope. */
1075 DEBUG_PRINT (
1076 dp << "[scop-detection-fail] "
1077 << "Gimple stmt not handled in Graphite:\n";
1078 print_gimple_stmt (dump_file, stmt, 0, TDF_VOPS | TDF_MEMSYMS));
1079 return false;
1083 /* Returns the number of pbbs that are in loops contained in SCOP. */
1086 scop_detection::nb_pbbs_in_loops (scop_p scop)
1088 int i;
1089 poly_bb_p pbb;
1090 int res = 0;
1092 FOR_EACH_VEC_ELT (scop->pbbs, i, pbb)
1093 if (loop_in_sese_p (gbb_loop (PBB_BLACK_BOX (pbb)), scop->scop_info->region))
1094 res++;
1096 return res;
1099 /* Assigns the parameter NAME an index in REGION. */
1101 static void
1102 assign_parameter_index_in_region (tree name, sese_info_p region)
1104 gcc_assert (TREE_CODE (name) == SSA_NAME
1105 && INTEGRAL_TYPE_P (TREE_TYPE (name))
1106 && ! defined_in_sese_p (name, region->region));
1107 int i;
1108 tree p;
1109 FOR_EACH_VEC_ELT (region->params, i, p)
1110 if (p == name)
1111 return;
1113 region->params.safe_push (name);
1116 /* In the context of sese S, scan the expression E and translate it to
1117 a linear expression C. When parsing a symbolic multiplication, K
1118 represents the constant multiplier of an expression containing
1119 parameters. */
1121 static void
1122 scan_tree_for_params (sese_info_p s, tree e)
1124 if (e == chrec_dont_know)
1125 return;
1127 switch (TREE_CODE (e))
1129 case POLYNOMIAL_CHREC:
1130 scan_tree_for_params (s, CHREC_LEFT (e));
1131 break;
1133 case MULT_EXPR:
1134 if (chrec_contains_symbols (TREE_OPERAND (e, 0)))
1135 scan_tree_for_params (s, TREE_OPERAND (e, 0));
1136 else
1137 scan_tree_for_params (s, TREE_OPERAND (e, 1));
1138 break;
1140 case PLUS_EXPR:
1141 case POINTER_PLUS_EXPR:
1142 case MINUS_EXPR:
1143 scan_tree_for_params (s, TREE_OPERAND (e, 0));
1144 scan_tree_for_params (s, TREE_OPERAND (e, 1));
1145 break;
1147 case NEGATE_EXPR:
1148 case BIT_NOT_EXPR:
1149 CASE_CONVERT:
1150 case NON_LVALUE_EXPR:
1151 scan_tree_for_params (s, TREE_OPERAND (e, 0));
1152 break;
1154 case SSA_NAME:
1155 assign_parameter_index_in_region (e, s);
1156 break;
1158 case INTEGER_CST:
1159 case ADDR_EXPR:
1160 case REAL_CST:
1161 case COMPLEX_CST:
1162 case VECTOR_CST:
1163 break;
1165 default:
1166 gcc_unreachable ();
1167 break;
1171 /* Find parameters with respect to REGION in BB. We are looking in memory
1172 access functions, conditions and loop bounds. */
1174 static void
1175 find_params_in_bb (sese_info_p region, gimple_poly_bb_p gbb)
1177 /* Find parameters in the access functions of data references. */
1178 int i;
1179 data_reference_p dr;
1180 FOR_EACH_VEC_ELT (GBB_DATA_REFS (gbb), i, dr)
1181 for (unsigned j = 0; j < DR_NUM_DIMENSIONS (dr); j++)
1182 scan_tree_for_params (region, DR_ACCESS_FN (dr, j));
1184 /* Find parameters in conditional statements. */
1185 gimple *stmt;
1186 FOR_EACH_VEC_ELT (GBB_CONDITIONS (gbb), i, stmt)
1188 loop_p loop = gimple_bb (stmt)->loop_father;
1189 tree lhs = cached_scalar_evolution_in_region (region->region, loop,
1190 gimple_cond_lhs (stmt));
1191 tree rhs = cached_scalar_evolution_in_region (region->region, loop,
1192 gimple_cond_rhs (stmt));
1193 gcc_assert (!chrec_contains_undetermined (lhs)
1194 && !chrec_contains_undetermined (rhs));
1196 scan_tree_for_params (region, lhs);
1197 scan_tree_for_params (region, rhs);
1201 /* Record the parameters used in the SCOP BBs. A variable is a parameter
1202 in a scop if it does not vary during the execution of that scop. */
1204 static void
1205 find_scop_parameters (scop_p scop)
1207 unsigned i;
1208 sese_info_p region = scop->scop_info;
1210 /* Parameters used in loop bounds are processed during gather_bbs. */
1212 /* Find the parameters used in data accesses. */
1213 poly_bb_p pbb;
1214 FOR_EACH_VEC_ELT (scop->pbbs, i, pbb)
1215 find_params_in_bb (region, PBB_BLACK_BOX (pbb));
1217 int nbp = sese_nb_params (region);
1218 scop_set_nb_params (scop, nbp);
1221 static void
1222 add_write (vec<tree> *writes, tree def)
1224 writes->safe_push (def);
1225 DEBUG_PRINT (dp << "Adding scalar write: ";
1226 print_generic_expr (dump_file, def);
1227 dp << "\nFrom stmt: ";
1228 print_gimple_stmt (dump_file,
1229 SSA_NAME_DEF_STMT (def), 0));
1232 static void
1233 add_read (vec<scalar_use> *reads, tree use, gimple *use_stmt)
1235 DEBUG_PRINT (dp << "Adding scalar read: ";
1236 print_generic_expr (dump_file, use);
1237 dp << "\nFrom stmt: ";
1238 print_gimple_stmt (dump_file, use_stmt, 0));
1239 reads->safe_push (std::make_pair (use_stmt, use));
1243 /* Record DEF if it is used in other bbs different than DEF_BB in the SCOP. */
1245 static void
1246 build_cross_bb_scalars_def (scop_p scop, tree def, basic_block def_bb,
1247 vec<tree> *writes)
1249 if (!is_gimple_reg (def))
1250 return;
1252 bool scev_analyzable = scev_analyzable_p (def, scop->scop_info->region);
1254 gimple *use_stmt;
1255 imm_use_iterator imm_iter;
1256 FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, def)
1257 /* Do not gather scalar variables that can be analyzed by SCEV as they can
1258 be generated out of the induction variables. */
1259 if ((! scev_analyzable
1260 /* But gather SESE liveouts as we otherwise fail to rewrite their
1261 exit PHIs. */
1262 || ! bb_in_sese_p (gimple_bb (use_stmt), scop->scop_info->region))
1263 && (def_bb != gimple_bb (use_stmt) && !is_gimple_debug (use_stmt)))
1265 add_write (writes, def);
1266 /* This is required by the FOR_EACH_IMM_USE_STMT when we want to break
1267 before all the uses have been visited. */
1268 BREAK_FROM_IMM_USE_STMT (imm_iter);
1272 /* Record USE if it is defined in other bbs different than USE_STMT
1273 in the SCOP. */
1275 static void
1276 build_cross_bb_scalars_use (scop_p scop, tree use, gimple *use_stmt,
1277 vec<scalar_use> *reads)
1279 if (!is_gimple_reg (use))
1280 return;
1282 /* Do not gather scalar variables that can be analyzed by SCEV as they can be
1283 generated out of the induction variables. */
1284 if (scev_analyzable_p (use, scop->scop_info->region))
1285 return;
1287 gimple *def_stmt = SSA_NAME_DEF_STMT (use);
1288 if (gimple_bb (def_stmt) != gimple_bb (use_stmt))
1289 add_read (reads, use, use_stmt);
1292 /* Generates a polyhedral black box only if the bb contains interesting
1293 information. */
1295 static gimple_poly_bb_p
1296 try_generate_gimple_bb (scop_p scop, basic_block bb)
1298 vec<data_reference_p> drs = vNULL;
1299 vec<tree> writes = vNULL;
1300 vec<scalar_use> reads = vNULL;
1302 sese_l region = scop->scop_info->region;
1303 edge nest = region.entry;
1304 loop_p loop = bb->loop_father;
1305 if (!loop_in_sese_p (loop, region))
1306 loop = NULL;
1308 for (gimple_stmt_iterator gsi = gsi_start_bb (bb); !gsi_end_p (gsi);
1309 gsi_next (&gsi))
1311 gimple *stmt = gsi_stmt (gsi);
1312 if (is_gimple_debug (stmt))
1313 continue;
1315 graphite_find_data_references_in_stmt (nest, loop, stmt, &drs);
1317 tree def = gimple_get_lhs (stmt);
1318 if (def)
1319 build_cross_bb_scalars_def (scop, def, gimple_bb (stmt), &writes);
1321 ssa_op_iter iter;
1322 tree use;
1323 FOR_EACH_SSA_TREE_OPERAND (use, stmt, iter, SSA_OP_USE)
1324 build_cross_bb_scalars_use (scop, use, stmt, &reads);
1327 /* Handle defs and uses in PHIs. Those need special treatment given
1328 that we have to present ISL with sth that looks like we've rewritten
1329 the IL out-of-SSA. */
1330 for (gphi_iterator psi = gsi_start_phis (bb); !gsi_end_p (psi);
1331 gsi_next (&psi))
1333 gphi *phi = psi.phi ();
1334 tree res = gimple_phi_result (phi);
1335 if (virtual_operand_p (res)
1336 || scev_analyzable_p (res, scop->scop_info->region))
1337 continue;
1338 /* To simulate out-of-SSA the block containing the PHI node has
1339 reads of the PHI destination. And to preserve SSA dependences
1340 we also write to it (the out-of-SSA decl and the SSA result
1341 are coalesced for dependence purposes which is good enough). */
1342 add_read (&reads, res, phi);
1343 add_write (&writes, res);
1345 basic_block bb_for_succs = bb;
1346 if (bb_for_succs == bb_for_succs->loop_father->latch
1347 && bb_in_sese_p (bb_for_succs, scop->scop_info->region)
1348 && sese_trivially_empty_bb_p (bb_for_succs))
1349 bb_for_succs = NULL;
1350 while (bb_for_succs)
1352 basic_block latch = NULL;
1353 edge_iterator ei;
1354 edge e;
1355 FOR_EACH_EDGE (e, ei, bb_for_succs->succs)
1357 for (gphi_iterator psi = gsi_start_phis (e->dest); !gsi_end_p (psi);
1358 gsi_next (&psi))
1360 gphi *phi = psi.phi ();
1361 tree res = gimple_phi_result (phi);
1362 if (virtual_operand_p (res))
1363 continue;
1364 /* To simulate out-of-SSA the predecessor of edges into PHI nodes
1365 has a copy from the PHI argument to the PHI destination. */
1366 if (! scev_analyzable_p (res, scop->scop_info->region))
1367 add_write (&writes, res);
1368 tree use = PHI_ARG_DEF_FROM_EDGE (phi, e);
1369 if (TREE_CODE (use) == SSA_NAME
1370 && ! SSA_NAME_IS_DEFAULT_DEF (use)
1371 && gimple_bb (SSA_NAME_DEF_STMT (use)) != bb_for_succs
1372 && ! scev_analyzable_p (use, scop->scop_info->region))
1373 add_read (&reads, use, phi);
1375 if (e->dest == bb_for_succs->loop_father->latch
1376 && bb_in_sese_p (e->dest, scop->scop_info->region)
1377 && sese_trivially_empty_bb_p (e->dest))
1378 latch = e->dest;
1380 /* Handle empty latch block PHIs here, otherwise we confuse ISL
1381 with extra conditional code where it then peels off the last
1382 iteration just because of that. It would be simplest if we
1383 just didn't force simple latches (thus remove the forwarder). */
1384 bb_for_succs = latch;
1387 /* For the region exit block add reads for all live-out vars. */
1388 if (bb == scop->scop_info->region.exit->src)
1390 sese_build_liveouts (scop->scop_info);
1391 unsigned i;
1392 bitmap_iterator bi;
1393 EXECUTE_IF_SET_IN_BITMAP (scop->scop_info->liveout, 0, i, bi)
1395 tree use = ssa_name (i);
1396 add_read (&reads, use, NULL);
1400 if (drs.is_empty () && writes.is_empty () && reads.is_empty ())
1401 return NULL;
1403 return new_gimple_poly_bb (bb, drs, reads, writes);
1406 /* Compute alias-sets for all data references in DRS. */
1408 static bool
1409 build_alias_set (scop_p scop)
1411 int num_vertices = scop->drs.length ();
1412 struct graph *g = new_graph (num_vertices);
1413 dr_info *dr1, *dr2;
1414 int i, j;
1415 int *all_vertices;
1417 struct loop *nest
1418 = find_common_loop (scop->scop_info->region.entry->dest->loop_father,
1419 scop->scop_info->region.exit->src->loop_father);
1421 FOR_EACH_VEC_ELT (scop->drs, i, dr1)
1422 for (j = i+1; scop->drs.iterate (j, &dr2); j++)
1423 if (dr_may_alias_p (dr1->dr, dr2->dr, nest))
1425 /* Dependences in the same alias set need to be handled
1426 by just looking at DR_ACCESS_FNs. */
1427 if (DR_NUM_DIMENSIONS (dr1->dr) == 0
1428 || DR_NUM_DIMENSIONS (dr1->dr) != DR_NUM_DIMENSIONS (dr2->dr)
1429 || ! operand_equal_p (DR_BASE_OBJECT (dr1->dr),
1430 DR_BASE_OBJECT (dr2->dr),
1431 OEP_ADDRESS_OF)
1432 || ! types_compatible_p (TREE_TYPE (DR_BASE_OBJECT (dr1->dr)),
1433 TREE_TYPE (DR_BASE_OBJECT (dr2->dr))))
1435 free_graph (g);
1436 return false;
1438 add_edge (g, i, j);
1439 add_edge (g, j, i);
1442 all_vertices = XNEWVEC (int, num_vertices);
1443 for (i = 0; i < num_vertices; i++)
1444 all_vertices[i] = i;
1446 scop->max_alias_set
1447 = graphds_dfs (g, all_vertices, num_vertices, NULL, true, NULL) + 1;
1448 free (all_vertices);
1450 for (i = 0; i < g->n_vertices; i++)
1451 scop->drs[i].alias_set = g->vertices[i].component + 1;
1453 free_graph (g);
1454 return true;
1457 /* Gather BBs and conditions for a SCOP. */
1458 class gather_bbs : public dom_walker
1460 public:
1461 gather_bbs (cdi_direction, scop_p, int *);
1463 virtual edge before_dom_children (basic_block);
1464 virtual void after_dom_children (basic_block);
1466 private:
1467 auto_vec<gimple *, 3> conditions, cases;
1468 scop_p scop;
1471 gather_bbs::gather_bbs (cdi_direction direction, scop_p scop, int *bb_to_rpo)
1472 : dom_walker (direction, ALL_BLOCKS, bb_to_rpo), scop (scop)
1476 /* Call-back for dom_walk executed before visiting the dominated
1477 blocks. */
1479 edge
1480 gather_bbs::before_dom_children (basic_block bb)
1482 sese_info_p region = scop->scop_info;
1483 if (!bb_in_sese_p (bb, region->region))
1484 return dom_walker::STOP;
1486 /* For loops fully contained in the region record parameters in the
1487 loop bounds. */
1488 loop_p loop = bb->loop_father;
1489 if (loop->header == bb
1490 && loop_in_sese_p (loop, region->region))
1492 tree nb_iters = number_of_latch_executions (loop);
1493 if (chrec_contains_symbols (nb_iters))
1495 nb_iters = cached_scalar_evolution_in_region (region->region,
1496 loop, nb_iters);
1497 scan_tree_for_params (region, nb_iters);
1501 if (gcond *stmt = single_pred_cond_non_loop_exit (bb))
1503 edge e = single_pred_edge (bb);
1504 /* Make sure the condition is in the region and thus was verified
1505 to be handled. */
1506 if (e != region->region.entry)
1508 conditions.safe_push (stmt);
1509 if (e->flags & EDGE_TRUE_VALUE)
1510 cases.safe_push (stmt);
1511 else
1512 cases.safe_push (NULL);
1516 scop->scop_info->bbs.safe_push (bb);
1518 gimple_poly_bb_p gbb = try_generate_gimple_bb (scop, bb);
1519 if (!gbb)
1520 return NULL;
1522 GBB_CONDITIONS (gbb) = conditions.copy ();
1523 GBB_CONDITION_CASES (gbb) = cases.copy ();
1525 poly_bb_p pbb = new_poly_bb (scop, gbb);
1526 scop->pbbs.safe_push (pbb);
1528 int i;
1529 data_reference_p dr;
1530 FOR_EACH_VEC_ELT (gbb->data_refs, i, dr)
1532 DEBUG_PRINT (dp << "Adding memory ";
1533 if (dr->is_read)
1534 dp << "read: ";
1535 else
1536 dp << "write: ";
1537 print_generic_expr (dump_file, dr->ref);
1538 dp << "\nFrom stmt: ";
1539 print_gimple_stmt (dump_file, dr->stmt, 0));
1541 scop->drs.safe_push (dr_info (dr, pbb));
1544 return NULL;
1547 /* Call-back for dom_walk executed after visiting the dominated
1548 blocks. */
1550 void
1551 gather_bbs::after_dom_children (basic_block bb)
1553 if (!bb_in_sese_p (bb, scop->scop_info->region))
1554 return;
1556 if (single_pred_cond_non_loop_exit (bb))
1558 edge e = single_pred_edge (bb);
1559 if (e != scop->scop_info->region.entry)
1561 conditions.pop ();
1562 cases.pop ();
1568 /* Compute sth like an execution order, dominator order with first executing
1569 edges that stay inside the current loop, delaying processing exit edges. */
1571 static int *bb_to_rpo;
1573 /* Helper for qsort, sorting after order above. */
1575 static int
1576 cmp_pbbs (const void *pa, const void *pb)
1578 poly_bb_p bb1 = *((const poly_bb_p *)pa);
1579 poly_bb_p bb2 = *((const poly_bb_p *)pb);
1580 if (bb_to_rpo[bb1->black_box->bb->index]
1581 < bb_to_rpo[bb2->black_box->bb->index])
1582 return -1;
1583 else if (bb_to_rpo[bb1->black_box->bb->index]
1584 > bb_to_rpo[bb2->black_box->bb->index])
1585 return 1;
1586 else
1587 return 0;
1590 /* Find Static Control Parts (SCoP) in the current function and pushes
1591 them to SCOPS. */
1593 void
1594 build_scops (vec<scop_p> *scops)
1596 if (dump_file)
1597 dp.set_dump_file (dump_file);
1599 scop_detection sb;
1600 sb.build_scop_depth (current_loops->tree_root);
1602 /* Now create scops from the lightweight SESEs. */
1603 vec<sese_l> scops_l = sb.get_scops ();
1605 /* Domwalk needs a bb to RPO mapping. Compute it once here. */
1606 int *postorder = XNEWVEC (int, n_basic_blocks_for_fn (cfun));
1607 int postorder_num = pre_and_rev_post_order_compute (NULL, postorder, true);
1608 bb_to_rpo = XNEWVEC (int, last_basic_block_for_fn (cfun));
1609 for (int i = 0; i < postorder_num; ++i)
1610 bb_to_rpo[postorder[i]] = i;
1611 free (postorder);
1613 int i;
1614 sese_l *s;
1615 FOR_EACH_VEC_ELT (scops_l, i, s)
1617 scop_p scop = new_scop (s->entry, s->exit);
1619 /* Record all basic blocks and their conditions in REGION. */
1620 gather_bbs (CDI_DOMINATORS, scop, bb_to_rpo).walk (s->entry->dest);
1622 /* Sort pbbs after execution order for initial schedule generation. */
1623 scop->pbbs.qsort (cmp_pbbs);
1625 if (! build_alias_set (scop))
1627 DEBUG_PRINT (dp << "[scop-detection-fail] cannot handle dependences\n");
1628 free_scop (scop);
1629 continue;
1632 /* Do not optimize a scop containing only PBBs that do not belong
1633 to any loops. */
1634 if (sb.nb_pbbs_in_loops (scop) == 0)
1636 DEBUG_PRINT (dp << "[scop-detection-fail] no data references.\n");
1637 free_scop (scop);
1638 continue;
1641 unsigned max_arrays = param_graphite_max_arrays_per_scop;
1642 if (max_arrays > 0
1643 && scop->drs.length () >= max_arrays)
1645 DEBUG_PRINT (dp << "[scop-detection-fail] too many data references: "
1646 << scop->drs.length ()
1647 << " is larger than --param graphite-max-arrays-per-scop="
1648 << max_arrays << ".\n");
1649 free_scop (scop);
1650 continue;
1653 find_scop_parameters (scop);
1654 graphite_dim_t max_dim = param_graphite_max_nb_scop_params;
1655 if (max_dim > 0
1656 && scop_nb_params (scop) > max_dim)
1658 DEBUG_PRINT (dp << "[scop-detection-fail] too many parameters: "
1659 << scop_nb_params (scop)
1660 << " larger than --param graphite-max-nb-scop-params="
1661 << max_dim << ".\n");
1662 free_scop (scop);
1663 continue;
1666 scops->safe_push (scop);
1669 free (bb_to_rpo);
1670 bb_to_rpo = NULL;
1671 DEBUG_PRINT (dp << "number of SCoPs: " << (scops ? scops->length () : 0););
1674 #endif /* HAVE_isl */