2014-04-14 Martin Jambor <mjambor@suse.cz>
[official-gcc.git] / gcc / tree-vectorizer.c
blob765e38f689aba652fb329b7074c9fbe313c2b8d3
1 /* Vectorizer
2 Copyright (C) 2003-2014 Free Software Foundation, Inc.
3 Contributed by Dorit Naishlos <dorit@il.ibm.com>
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 3, or (at your option) any later
10 version.
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
15 for more details.
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING3. If not see
19 <http://www.gnu.org/licenses/>. */
21 /* Loop and basic block vectorizer.
23 This file contains drivers for the three vectorizers:
24 (1) loop vectorizer (inter-iteration parallelism),
25 (2) loop-aware SLP (intra-iteration parallelism) (invoked by the loop
26 vectorizer)
27 (3) BB vectorizer (out-of-loops), aka SLP
29 The rest of the vectorizer's code is organized as follows:
30 - tree-vect-loop.c - loop specific parts such as reductions, etc. These are
31 used by drivers (1) and (2).
32 - tree-vect-loop-manip.c - vectorizer's loop control-flow utilities, used by
33 drivers (1) and (2).
34 - tree-vect-slp.c - BB vectorization specific analysis and transformation,
35 used by drivers (2) and (3).
36 - tree-vect-stmts.c - statements analysis and transformation (used by all).
37 - tree-vect-data-refs.c - vectorizer specific data-refs analysis and
38 manipulations (used by all).
39 - tree-vect-patterns.c - vectorizable code patterns detector (used by all)
41 Here's a poor attempt at illustrating that:
43 tree-vectorizer.c:
44 loop_vect() loop_aware_slp() slp_vect()
45 | / \ /
46 | / \ /
47 tree-vect-loop.c tree-vect-slp.c
48 | \ \ / / |
49 | \ \/ / |
50 | \ /\ / |
51 | \ / \ / |
52 tree-vect-stmts.c tree-vect-data-refs.c
53 \ /
54 tree-vect-patterns.c
57 #include "config.h"
58 #include "system.h"
59 #include "coretypes.h"
60 #include "dumpfile.h"
61 #include "tm.h"
62 #include "tree.h"
63 #include "stor-layout.h"
64 #include "tree-pretty-print.h"
65 #include "basic-block.h"
66 #include "tree-ssa-alias.h"
67 #include "internal-fn.h"
68 #include "gimple-expr.h"
69 #include "is-a.h"
70 #include "gimple.h"
71 #include "gimple-iterator.h"
72 #include "gimple-walk.h"
73 #include "gimple-ssa.h"
74 #include "cgraph.h"
75 #include "tree-phinodes.h"
76 #include "ssa-iterators.h"
77 #include "tree-ssa-loop-manip.h"
78 #include "tree-cfg.h"
79 #include "cfgloop.h"
80 #include "tree-vectorizer.h"
81 #include "tree-pass.h"
82 #include "tree-ssa-propagate.h"
83 #include "dbgcnt.h"
84 #include "gimple-fold.h"
86 /* Loop or bb location. */
87 source_location vect_location;
89 /* Vector mapping GIMPLE stmt to stmt_vec_info. */
90 vec<vec_void_p> stmt_vec_info_vec;
92 /* For mapping simduid to vectorization factor. */
94 struct simduid_to_vf : typed_free_remove<simduid_to_vf>
96 unsigned int simduid;
97 int vf;
99 /* hash_table support. */
100 typedef simduid_to_vf value_type;
101 typedef simduid_to_vf compare_type;
102 static inline hashval_t hash (const value_type *);
103 static inline int equal (const value_type *, const compare_type *);
106 inline hashval_t
107 simduid_to_vf::hash (const value_type *p)
109 return p->simduid;
112 inline int
113 simduid_to_vf::equal (const value_type *p1, const value_type *p2)
115 return p1->simduid == p2->simduid;
118 /* This hash maps the OMP simd array to the corresponding simduid used
119 to index into it. Like thus,
121 _7 = GOMP_SIMD_LANE (simduid.0)
124 D.1737[_7] = stuff;
127 This hash maps from the OMP simd array (D.1737[]) to DECL_UID of
128 simduid.0. */
130 struct simd_array_to_simduid : typed_free_remove<simd_array_to_simduid>
132 tree decl;
133 unsigned int simduid;
135 /* hash_table support. */
136 typedef simd_array_to_simduid value_type;
137 typedef simd_array_to_simduid compare_type;
138 static inline hashval_t hash (const value_type *);
139 static inline int equal (const value_type *, const compare_type *);
142 inline hashval_t
143 simd_array_to_simduid::hash (const value_type *p)
145 return DECL_UID (p->decl);
148 inline int
149 simd_array_to_simduid::equal (const value_type *p1, const value_type *p2)
151 return p1->decl == p2->decl;
154 /* Fold IFN_GOMP_SIMD_LANE, IFN_GOMP_SIMD_VF and IFN_GOMP_SIMD_LAST_LANE
155 into their corresponding constants. */
157 static void
158 adjust_simduid_builtins (hash_table <simduid_to_vf> &htab)
160 basic_block bb;
162 FOR_EACH_BB_FN (bb, cfun)
164 gimple_stmt_iterator i;
166 for (i = gsi_start_bb (bb); !gsi_end_p (i); gsi_next (&i))
168 unsigned int vf = 1;
169 enum internal_fn ifn;
170 gimple stmt = gsi_stmt (i);
171 tree t;
172 if (!is_gimple_call (stmt)
173 || !gimple_call_internal_p (stmt))
174 continue;
175 ifn = gimple_call_internal_fn (stmt);
176 switch (ifn)
178 case IFN_GOMP_SIMD_LANE:
179 case IFN_GOMP_SIMD_VF:
180 case IFN_GOMP_SIMD_LAST_LANE:
181 break;
182 default:
183 continue;
185 tree arg = gimple_call_arg (stmt, 0);
186 gcc_assert (arg != NULL_TREE);
187 gcc_assert (TREE_CODE (arg) == SSA_NAME);
188 simduid_to_vf *p = NULL, data;
189 data.simduid = DECL_UID (SSA_NAME_VAR (arg));
190 if (htab.is_created ())
191 p = htab.find (&data);
192 if (p)
193 vf = p->vf;
194 switch (ifn)
196 case IFN_GOMP_SIMD_VF:
197 t = build_int_cst (unsigned_type_node, vf);
198 break;
199 case IFN_GOMP_SIMD_LANE:
200 t = build_int_cst (unsigned_type_node, 0);
201 break;
202 case IFN_GOMP_SIMD_LAST_LANE:
203 t = gimple_call_arg (stmt, 1);
204 break;
205 default:
206 gcc_unreachable ();
208 update_call_from_tree (&i, t);
213 /* Helper structure for note_simd_array_uses. */
215 struct note_simd_array_uses_struct
217 hash_table <simd_array_to_simduid> *htab;
218 unsigned int simduid;
221 /* Callback for note_simd_array_uses, called through walk_gimple_op. */
223 static tree
224 note_simd_array_uses_cb (tree *tp, int *walk_subtrees, void *data)
226 struct walk_stmt_info *wi = (struct walk_stmt_info *) data;
227 struct note_simd_array_uses_struct *ns
228 = (struct note_simd_array_uses_struct *) wi->info;
230 if (TYPE_P (*tp))
231 *walk_subtrees = 0;
232 else if (VAR_P (*tp)
233 && lookup_attribute ("omp simd array", DECL_ATTRIBUTES (*tp))
234 && DECL_CONTEXT (*tp) == current_function_decl)
236 simd_array_to_simduid data;
237 if (!ns->htab->is_created ())
238 ns->htab->create (15);
239 data.decl = *tp;
240 data.simduid = ns->simduid;
241 simd_array_to_simduid **slot = ns->htab->find_slot (&data, INSERT);
242 if (*slot == NULL)
244 simd_array_to_simduid *p = XNEW (simd_array_to_simduid);
245 *p = data;
246 *slot = p;
248 else if ((*slot)->simduid != ns->simduid)
249 (*slot)->simduid = -1U;
250 *walk_subtrees = 0;
252 return NULL_TREE;
255 /* Find "omp simd array" temporaries and map them to corresponding
256 simduid. */
258 static void
259 note_simd_array_uses (hash_table <simd_array_to_simduid> *htab)
261 basic_block bb;
262 gimple_stmt_iterator gsi;
263 struct walk_stmt_info wi;
264 struct note_simd_array_uses_struct ns;
266 memset (&wi, 0, sizeof (wi));
267 wi.info = &ns;
268 ns.htab = htab;
270 FOR_EACH_BB_FN (bb, cfun)
271 for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
273 gimple stmt = gsi_stmt (gsi);
274 if (!is_gimple_call (stmt) || !gimple_call_internal_p (stmt))
275 continue;
276 switch (gimple_call_internal_fn (stmt))
278 case IFN_GOMP_SIMD_LANE:
279 case IFN_GOMP_SIMD_VF:
280 case IFN_GOMP_SIMD_LAST_LANE:
281 break;
282 default:
283 continue;
285 tree lhs = gimple_call_lhs (stmt);
286 if (lhs == NULL_TREE)
287 continue;
288 imm_use_iterator use_iter;
289 gimple use_stmt;
290 ns.simduid = DECL_UID (SSA_NAME_VAR (gimple_call_arg (stmt, 0)));
291 FOR_EACH_IMM_USE_STMT (use_stmt, use_iter, lhs)
292 if (!is_gimple_debug (use_stmt))
293 walk_gimple_op (use_stmt, note_simd_array_uses_cb, &wi);
297 /* A helper function to free data refs. */
299 void
300 vect_destroy_datarefs (loop_vec_info loop_vinfo, bb_vec_info bb_vinfo)
302 vec<data_reference_p> datarefs;
303 struct data_reference *dr;
304 unsigned int i;
306 if (loop_vinfo)
307 datarefs = LOOP_VINFO_DATAREFS (loop_vinfo);
308 else
309 datarefs = BB_VINFO_DATAREFS (bb_vinfo);
311 FOR_EACH_VEC_ELT (datarefs, i, dr)
312 if (dr->aux)
314 free (dr->aux);
315 dr->aux = NULL;
318 free_data_refs (datarefs);
322 /* If LOOP has been versioned during ifcvt, return the internal call
323 guarding it. */
325 static gimple
326 vect_loop_vectorized_call (struct loop *loop)
328 basic_block bb = loop_preheader_edge (loop)->src;
329 gimple g;
332 g = last_stmt (bb);
333 if (g)
334 break;
335 if (!single_pred_p (bb))
336 break;
337 bb = single_pred (bb);
339 while (1);
340 if (g && gimple_code (g) == GIMPLE_COND)
342 gimple_stmt_iterator gsi = gsi_for_stmt (g);
343 gsi_prev (&gsi);
344 if (!gsi_end_p (gsi))
346 g = gsi_stmt (gsi);
347 if (is_gimple_call (g)
348 && gimple_call_internal_p (g)
349 && gimple_call_internal_fn (g) == IFN_LOOP_VECTORIZED
350 && (tree_to_shwi (gimple_call_arg (g, 0)) == loop->num
351 || tree_to_shwi (gimple_call_arg (g, 1)) == loop->num))
352 return g;
355 return NULL;
358 /* Fold LOOP_VECTORIZED internal call G to VALUE and
359 update any immediate uses of it's LHS. */
361 static void
362 fold_loop_vectorized_call (gimple g, tree value)
364 tree lhs = gimple_call_lhs (g);
365 use_operand_p use_p;
366 imm_use_iterator iter;
367 gimple use_stmt;
368 gimple_stmt_iterator gsi = gsi_for_stmt (g);
370 update_call_from_tree (&gsi, value);
371 FOR_EACH_IMM_USE_STMT (use_stmt, iter, lhs)
373 FOR_EACH_IMM_USE_ON_STMT (use_p, iter)
374 SET_USE (use_p, value);
375 update_stmt (use_stmt);
379 /* Function vectorize_loops.
381 Entry point to loop vectorization phase. */
383 unsigned
384 vectorize_loops (void)
386 unsigned int i;
387 unsigned int num_vectorized_loops = 0;
388 unsigned int vect_loops_num;
389 struct loop *loop;
390 hash_table <simduid_to_vf> simduid_to_vf_htab;
391 hash_table <simd_array_to_simduid> simd_array_to_simduid_htab;
392 bool any_ifcvt_loops = false;
393 unsigned ret = 0;
395 vect_loops_num = number_of_loops (cfun);
397 /* Bail out if there are no loops. */
398 if (vect_loops_num <= 1)
400 if (cfun->has_simduid_loops)
401 adjust_simduid_builtins (simduid_to_vf_htab);
402 return 0;
405 if (cfun->has_simduid_loops)
406 note_simd_array_uses (&simd_array_to_simduid_htab);
408 init_stmt_vec_info_vec ();
410 /* ----------- Analyze loops. ----------- */
412 /* If some loop was duplicated, it gets bigger number
413 than all previously defined loops. This fact allows us to run
414 only over initial loops skipping newly generated ones. */
415 FOR_EACH_LOOP (loop, 0)
416 if (loop->dont_vectorize)
417 any_ifcvt_loops = true;
418 else if ((flag_tree_loop_vectorize
419 && optimize_loop_nest_for_speed_p (loop))
420 || loop->force_vectorize)
422 loop_vec_info loop_vinfo;
423 vect_location = find_loop_location (loop);
424 if (LOCATION_LOCUS (vect_location) != UNKNOWN_LOCATION
425 && dump_enabled_p ())
426 dump_printf (MSG_NOTE, "\nAnalyzing loop at %s:%d\n",
427 LOCATION_FILE (vect_location),
428 LOCATION_LINE (vect_location));
430 loop_vinfo = vect_analyze_loop (loop);
431 loop->aux = loop_vinfo;
433 if (!loop_vinfo || !LOOP_VINFO_VECTORIZABLE_P (loop_vinfo))
434 continue;
436 if (!dbg_cnt (vect_loop))
437 break;
439 gimple loop_vectorized_call = vect_loop_vectorized_call (loop);
440 if (loop_vectorized_call)
442 tree arg = gimple_call_arg (loop_vectorized_call, 1);
443 basic_block *bbs;
444 unsigned int i;
445 struct loop *scalar_loop = get_loop (cfun, tree_to_shwi (arg));
447 LOOP_VINFO_SCALAR_LOOP (loop_vinfo) = scalar_loop;
448 gcc_checking_assert (vect_loop_vectorized_call
449 (LOOP_VINFO_SCALAR_LOOP (loop_vinfo))
450 == loop_vectorized_call);
451 bbs = get_loop_body (scalar_loop);
452 for (i = 0; i < scalar_loop->num_nodes; i++)
454 basic_block bb = bbs[i];
455 gimple_stmt_iterator gsi;
456 for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi);
457 gsi_next (&gsi))
459 gimple phi = gsi_stmt (gsi);
460 gimple_set_uid (phi, 0);
462 for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi);
463 gsi_next (&gsi))
465 gimple stmt = gsi_stmt (gsi);
466 gimple_set_uid (stmt, 0);
469 free (bbs);
472 if (LOCATION_LOCUS (vect_location) != UNKNOWN_LOCATION
473 && dump_enabled_p ())
474 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, vect_location,
475 "loop vectorized\n");
476 vect_transform_loop (loop_vinfo);
477 num_vectorized_loops++;
478 /* Now that the loop has been vectorized, allow it to be unrolled
479 etc. */
480 loop->force_vectorize = false;
482 if (loop->simduid)
484 simduid_to_vf *simduid_to_vf_data = XNEW (simduid_to_vf);
485 if (!simduid_to_vf_htab.is_created ())
486 simduid_to_vf_htab.create (15);
487 simduid_to_vf_data->simduid = DECL_UID (loop->simduid);
488 simduid_to_vf_data->vf = loop_vinfo->vectorization_factor;
489 *simduid_to_vf_htab.find_slot (simduid_to_vf_data, INSERT)
490 = simduid_to_vf_data;
493 if (loop_vectorized_call)
495 fold_loop_vectorized_call (loop_vectorized_call, boolean_true_node);
496 ret |= TODO_cleanup_cfg;
500 vect_location = UNKNOWN_LOCATION;
502 statistics_counter_event (cfun, "Vectorized loops", num_vectorized_loops);
503 if (dump_enabled_p ()
504 || (num_vectorized_loops > 0 && dump_enabled_p ()))
505 dump_printf_loc (MSG_NOTE, vect_location,
506 "vectorized %u loops in function.\n",
507 num_vectorized_loops);
509 /* ----------- Finalize. ----------- */
511 if (any_ifcvt_loops)
512 for (i = 1; i < vect_loops_num; i++)
514 loop = get_loop (cfun, i);
515 if (loop && loop->dont_vectorize)
517 gimple g = vect_loop_vectorized_call (loop);
518 if (g)
520 fold_loop_vectorized_call (g, boolean_false_node);
521 ret |= TODO_cleanup_cfg;
526 for (i = 1; i < vect_loops_num; i++)
528 loop_vec_info loop_vinfo;
530 loop = get_loop (cfun, i);
531 if (!loop)
532 continue;
533 loop_vinfo = (loop_vec_info) loop->aux;
534 destroy_loop_vec_info (loop_vinfo, true);
535 loop->aux = NULL;
538 free_stmt_vec_info_vec ();
540 /* Fold IFN_GOMP_SIMD_{VF,LANE,LAST_LANE} builtins. */
541 if (cfun->has_simduid_loops)
542 adjust_simduid_builtins (simduid_to_vf_htab);
544 /* Shrink any "omp array simd" temporary arrays to the
545 actual vectorization factors. */
546 if (simd_array_to_simduid_htab.is_created ())
548 for (hash_table <simd_array_to_simduid>::iterator iter
549 = simd_array_to_simduid_htab.begin ();
550 iter != simd_array_to_simduid_htab.end (); ++iter)
551 if ((*iter).simduid != -1U)
553 tree decl = (*iter).decl;
554 int vf = 1;
555 if (simduid_to_vf_htab.is_created ())
557 simduid_to_vf *p = NULL, data;
558 data.simduid = (*iter).simduid;
559 p = simduid_to_vf_htab.find (&data);
560 if (p)
561 vf = p->vf;
563 tree atype
564 = build_array_type_nelts (TREE_TYPE (TREE_TYPE (decl)), vf);
565 TREE_TYPE (decl) = atype;
566 relayout_decl (decl);
569 simd_array_to_simduid_htab.dispose ();
571 if (simduid_to_vf_htab.is_created ())
572 simduid_to_vf_htab.dispose ();
574 if (num_vectorized_loops > 0)
576 /* If we vectorized any loop only virtual SSA form needs to be updated.
577 ??? Also while we try hard to update loop-closed SSA form we fail
578 to properly do this in some corner-cases (see PR56286). */
579 rewrite_into_loop_closed_ssa (NULL, TODO_update_ssa_only_virtuals);
580 return TODO_cleanup_cfg;
583 return ret;
587 /* Entry point to basic block SLP phase. */
589 static unsigned int
590 execute_vect_slp (void)
592 basic_block bb;
594 init_stmt_vec_info_vec ();
596 FOR_EACH_BB_FN (bb, cfun)
598 vect_location = find_bb_location (bb);
600 if (vect_slp_analyze_bb (bb))
602 if (!dbg_cnt (vect_slp))
603 break;
605 vect_slp_transform_bb (bb);
606 if (dump_enabled_p ())
607 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, vect_location,
608 "basic block vectorized\n");
612 free_stmt_vec_info_vec ();
613 return 0;
616 static bool
617 gate_vect_slp (void)
619 return flag_tree_slp_vectorize != 0;
622 namespace {
624 const pass_data pass_data_slp_vectorize =
626 GIMPLE_PASS, /* type */
627 "slp", /* name */
628 OPTGROUP_LOOP | OPTGROUP_VEC, /* optinfo_flags */
629 true, /* has_gate */
630 true, /* has_execute */
631 TV_TREE_SLP_VECTORIZATION, /* tv_id */
632 ( PROP_ssa | PROP_cfg ), /* properties_required */
633 0, /* properties_provided */
634 0, /* properties_destroyed */
635 0, /* todo_flags_start */
636 ( TODO_verify_ssa | TODO_update_ssa
637 | TODO_verify_stmts ), /* todo_flags_finish */
640 class pass_slp_vectorize : public gimple_opt_pass
642 public:
643 pass_slp_vectorize (gcc::context *ctxt)
644 : gimple_opt_pass (pass_data_slp_vectorize, ctxt)
647 /* opt_pass methods: */
648 bool gate () { return gate_vect_slp (); }
649 unsigned int execute () { return execute_vect_slp (); }
651 }; // class pass_slp_vectorize
653 } // anon namespace
655 gimple_opt_pass *
656 make_pass_slp_vectorize (gcc::context *ctxt)
658 return new pass_slp_vectorize (ctxt);
662 /* Increase alignment of global arrays to improve vectorization potential.
663 TODO:
664 - Consider also structs that have an array field.
665 - Use ipa analysis to prune arrays that can't be vectorized?
666 This should involve global alignment analysis and in the future also
667 array padding. */
669 static unsigned int
670 increase_alignment (void)
672 varpool_node *vnode;
674 vect_location = UNKNOWN_LOCATION;
676 /* Increase the alignment of all global arrays for vectorization. */
677 FOR_EACH_DEFINED_VARIABLE (vnode)
679 tree vectype, decl = vnode->decl;
680 tree t;
681 unsigned int alignment;
683 t = TREE_TYPE (decl);
684 if (TREE_CODE (t) != ARRAY_TYPE)
685 continue;
686 vectype = get_vectype_for_scalar_type (strip_array_types (t));
687 if (!vectype)
688 continue;
689 alignment = TYPE_ALIGN (vectype);
690 if (DECL_ALIGN (decl) >= alignment)
691 continue;
693 if (vect_can_force_dr_alignment_p (decl, alignment))
695 DECL_ALIGN (decl) = TYPE_ALIGN (vectype);
696 DECL_USER_ALIGN (decl) = 1;
697 dump_printf (MSG_NOTE, "Increasing alignment of decl: ");
698 dump_generic_expr (MSG_NOTE, TDF_SLIM, decl);
699 dump_printf (MSG_NOTE, "\n");
702 return 0;
706 static bool
707 gate_increase_alignment (void)
709 return flag_section_anchors && flag_tree_loop_vectorize;
713 namespace {
715 const pass_data pass_data_ipa_increase_alignment =
717 SIMPLE_IPA_PASS, /* type */
718 "increase_alignment", /* name */
719 OPTGROUP_LOOP | OPTGROUP_VEC, /* optinfo_flags */
720 true, /* has_gate */
721 true, /* has_execute */
722 TV_IPA_OPT, /* tv_id */
723 0, /* properties_required */
724 0, /* properties_provided */
725 0, /* properties_destroyed */
726 0, /* todo_flags_start */
727 0, /* todo_flags_finish */
730 class pass_ipa_increase_alignment : public simple_ipa_opt_pass
732 public:
733 pass_ipa_increase_alignment (gcc::context *ctxt)
734 : simple_ipa_opt_pass (pass_data_ipa_increase_alignment, ctxt)
737 /* opt_pass methods: */
738 bool gate () { return gate_increase_alignment (); }
739 unsigned int execute () { return increase_alignment (); }
741 }; // class pass_ipa_increase_alignment
743 } // anon namespace
745 simple_ipa_opt_pass *
746 make_pass_ipa_increase_alignment (gcc::context *ctxt)
748 return new pass_ipa_increase_alignment (ctxt);