2 Copyright (C) 2003-2015 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
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING3. If not see
19 <http://www.gnu.org/licenses/>. */
21 /* 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
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
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:
44 loop_vect() loop_aware_slp() slp_vect()
47 tree-vect-loop.c tree-vect-slp.c
52 tree-vect-stmts.c tree-vect-data-refs.c
59 #include "coretypes.h"
66 #include "fold-const.h"
67 #include "stor-layout.h"
68 #include "tree-pretty-print.h"
70 #include "hard-reg-set.h"
73 #include "dominance.h"
75 #include "basic-block.h"
76 #include "tree-ssa-alias.h"
77 #include "internal-fn.h"
78 #include "gimple-expr.h"
81 #include "gimple-iterator.h"
82 #include "gimple-walk.h"
83 #include "gimple-ssa.h"
84 #include "plugin-api.h"
87 #include "tree-phinodes.h"
88 #include "ssa-iterators.h"
89 #include "tree-ssa-loop-manip.h"
92 #include "tree-vectorizer.h"
93 #include "tree-pass.h"
94 #include "tree-ssa-propagate.h"
96 #include "gimple-fold.h"
97 #include "tree-scalar-evolution.h"
100 /* Loop or bb location. */
101 source_location vect_location
;
103 /* Vector mapping GIMPLE stmt to stmt_vec_info. */
104 vec
<vec_void_p
> stmt_vec_info_vec
;
106 /* For mapping simduid to vectorization factor. */
108 struct simduid_to_vf
: typed_free_remove
<simduid_to_vf
>
110 unsigned int simduid
;
113 /* hash_table support. */
114 typedef simduid_to_vf
*value_type
;
115 typedef simduid_to_vf
*compare_type
;
116 static inline hashval_t
hash (const simduid_to_vf
*);
117 static inline int equal (const simduid_to_vf
*, const simduid_to_vf
*);
121 simduid_to_vf::hash (const simduid_to_vf
*p
)
127 simduid_to_vf::equal (const simduid_to_vf
*p1
, const simduid_to_vf
*p2
)
129 return p1
->simduid
== p2
->simduid
;
132 /* This hash maps the OMP simd array to the corresponding simduid used
133 to index into it. Like thus,
135 _7 = GOMP_SIMD_LANE (simduid.0)
141 This hash maps from the OMP simd array (D.1737[]) to DECL_UID of
144 struct simd_array_to_simduid
: typed_free_remove
<simd_array_to_simduid
>
147 unsigned int simduid
;
149 /* hash_table support. */
150 typedef simd_array_to_simduid
*value_type
;
151 typedef simd_array_to_simduid
*compare_type
;
152 static inline hashval_t
hash (const simd_array_to_simduid
*);
153 static inline int equal (const simd_array_to_simduid
*,
154 const simd_array_to_simduid
*);
158 simd_array_to_simduid::hash (const simd_array_to_simduid
*p
)
160 return DECL_UID (p
->decl
);
164 simd_array_to_simduid::equal (const simd_array_to_simduid
*p1
,
165 const simd_array_to_simduid
*p2
)
167 return p1
->decl
== p2
->decl
;
170 /* Fold IFN_GOMP_SIMD_LANE, IFN_GOMP_SIMD_VF and IFN_GOMP_SIMD_LAST_LANE
171 into their corresponding constants. */
174 adjust_simduid_builtins (hash_table
<simduid_to_vf
> **htab
)
178 FOR_EACH_BB_FN (bb
, cfun
)
180 gimple_stmt_iterator i
;
182 for (i
= gsi_start_bb (bb
); !gsi_end_p (i
); gsi_next (&i
))
185 enum internal_fn ifn
;
186 gimple stmt
= gsi_stmt (i
);
188 if (!is_gimple_call (stmt
)
189 || !gimple_call_internal_p (stmt
))
191 ifn
= gimple_call_internal_fn (stmt
);
194 case IFN_GOMP_SIMD_LANE
:
195 case IFN_GOMP_SIMD_VF
:
196 case IFN_GOMP_SIMD_LAST_LANE
:
201 tree arg
= gimple_call_arg (stmt
, 0);
202 gcc_assert (arg
!= NULL_TREE
);
203 gcc_assert (TREE_CODE (arg
) == SSA_NAME
);
204 simduid_to_vf
*p
= NULL
, data
;
205 data
.simduid
= DECL_UID (SSA_NAME_VAR (arg
));
207 p
= (*htab
)->find (&data
);
212 case IFN_GOMP_SIMD_VF
:
213 t
= build_int_cst (unsigned_type_node
, vf
);
215 case IFN_GOMP_SIMD_LANE
:
216 t
= build_int_cst (unsigned_type_node
, 0);
218 case IFN_GOMP_SIMD_LAST_LANE
:
219 t
= gimple_call_arg (stmt
, 1);
224 update_call_from_tree (&i
, t
);
229 /* Helper structure for note_simd_array_uses. */
231 struct note_simd_array_uses_struct
233 hash_table
<simd_array_to_simduid
> **htab
;
234 unsigned int simduid
;
237 /* Callback for note_simd_array_uses, called through walk_gimple_op. */
240 note_simd_array_uses_cb (tree
*tp
, int *walk_subtrees
, void *data
)
242 struct walk_stmt_info
*wi
= (struct walk_stmt_info
*) data
;
243 struct note_simd_array_uses_struct
*ns
244 = (struct note_simd_array_uses_struct
*) wi
->info
;
249 && lookup_attribute ("omp simd array", DECL_ATTRIBUTES (*tp
))
250 && DECL_CONTEXT (*tp
) == current_function_decl
)
252 simd_array_to_simduid data
;
254 *ns
->htab
= new hash_table
<simd_array_to_simduid
> (15);
256 data
.simduid
= ns
->simduid
;
257 simd_array_to_simduid
**slot
= (*ns
->htab
)->find_slot (&data
, INSERT
);
260 simd_array_to_simduid
*p
= XNEW (simd_array_to_simduid
);
264 else if ((*slot
)->simduid
!= ns
->simduid
)
265 (*slot
)->simduid
= -1U;
271 /* Find "omp simd array" temporaries and map them to corresponding
275 note_simd_array_uses (hash_table
<simd_array_to_simduid
> **htab
)
278 gimple_stmt_iterator gsi
;
279 struct walk_stmt_info wi
;
280 struct note_simd_array_uses_struct ns
;
282 memset (&wi
, 0, sizeof (wi
));
286 FOR_EACH_BB_FN (bb
, cfun
)
287 for (gsi
= gsi_start_bb (bb
); !gsi_end_p (gsi
); gsi_next (&gsi
))
289 gimple stmt
= gsi_stmt (gsi
);
290 if (!is_gimple_call (stmt
) || !gimple_call_internal_p (stmt
))
292 switch (gimple_call_internal_fn (stmt
))
294 case IFN_GOMP_SIMD_LANE
:
295 case IFN_GOMP_SIMD_VF
:
296 case IFN_GOMP_SIMD_LAST_LANE
:
301 tree lhs
= gimple_call_lhs (stmt
);
302 if (lhs
== NULL_TREE
)
304 imm_use_iterator use_iter
;
306 ns
.simduid
= DECL_UID (SSA_NAME_VAR (gimple_call_arg (stmt
, 0)));
307 FOR_EACH_IMM_USE_STMT (use_stmt
, use_iter
, lhs
)
308 if (!is_gimple_debug (use_stmt
))
309 walk_gimple_op (use_stmt
, note_simd_array_uses_cb
, &wi
);
313 /* A helper function to free data refs. */
316 vect_destroy_datarefs (loop_vec_info loop_vinfo
, bb_vec_info bb_vinfo
)
318 vec
<data_reference_p
> datarefs
;
319 struct data_reference
*dr
;
323 datarefs
= LOOP_VINFO_DATAREFS (loop_vinfo
);
325 datarefs
= BB_VINFO_DATAREFS (bb_vinfo
);
327 FOR_EACH_VEC_ELT (datarefs
, i
, dr
)
334 free_data_refs (datarefs
);
338 /* If LOOP has been versioned during ifcvt, return the internal call
342 vect_loop_vectorized_call (struct loop
*loop
)
344 basic_block bb
= loop_preheader_edge (loop
)->src
;
351 if (!single_pred_p (bb
))
353 bb
= single_pred (bb
);
356 if (g
&& gimple_code (g
) == GIMPLE_COND
)
358 gimple_stmt_iterator gsi
= gsi_for_stmt (g
);
360 if (!gsi_end_p (gsi
))
363 if (is_gimple_call (g
)
364 && gimple_call_internal_p (g
)
365 && gimple_call_internal_fn (g
) == IFN_LOOP_VECTORIZED
366 && (tree_to_shwi (gimple_call_arg (g
, 0)) == loop
->num
367 || tree_to_shwi (gimple_call_arg (g
, 1)) == loop
->num
))
374 /* Fold LOOP_VECTORIZED internal call G to VALUE and
375 update any immediate uses of it's LHS. */
378 fold_loop_vectorized_call (gimple g
, tree value
)
380 tree lhs
= gimple_call_lhs (g
);
382 imm_use_iterator iter
;
384 gimple_stmt_iterator gsi
= gsi_for_stmt (g
);
386 update_call_from_tree (&gsi
, value
);
387 FOR_EACH_IMM_USE_STMT (use_stmt
, iter
, lhs
)
389 FOR_EACH_IMM_USE_ON_STMT (use_p
, iter
)
390 SET_USE (use_p
, value
);
391 update_stmt (use_stmt
);
394 /* Set the uids of all the statements in basic blocks inside loop
395 represented by LOOP_VINFO. LOOP_VECTORIZED_CALL is the internal
396 call guarding the loop which has been if converted. */
398 set_uid_loop_bbs (loop_vec_info loop_vinfo
, gimple loop_vectorized_call
)
400 tree arg
= gimple_call_arg (loop_vectorized_call
, 1);
403 struct loop
*scalar_loop
= get_loop (cfun
, tree_to_shwi (arg
));
405 LOOP_VINFO_SCALAR_LOOP (loop_vinfo
) = scalar_loop
;
406 gcc_checking_assert (vect_loop_vectorized_call
407 (LOOP_VINFO_SCALAR_LOOP (loop_vinfo
))
408 == loop_vectorized_call
);
409 bbs
= get_loop_body (scalar_loop
);
410 for (i
= 0; i
< scalar_loop
->num_nodes
; i
++)
412 basic_block bb
= bbs
[i
];
413 gimple_stmt_iterator gsi
;
414 for (gsi
= gsi_start_phis (bb
); !gsi_end_p (gsi
); gsi_next (&gsi
))
416 gimple phi
= gsi_stmt (gsi
);
417 gimple_set_uid (phi
, 0);
419 for (gsi
= gsi_start_bb (bb
); !gsi_end_p (gsi
); gsi_next (&gsi
))
421 gimple stmt
= gsi_stmt (gsi
);
422 gimple_set_uid (stmt
, 0);
428 /* Function vectorize_loops.
430 Entry point to loop vectorization phase. */
433 vectorize_loops (void)
436 unsigned int num_vectorized_loops
= 0;
437 unsigned int vect_loops_num
;
439 hash_table
<simduid_to_vf
> *simduid_to_vf_htab
= NULL
;
440 hash_table
<simd_array_to_simduid
> *simd_array_to_simduid_htab
= NULL
;
441 bool any_ifcvt_loops
= false;
444 vect_loops_num
= number_of_loops (cfun
);
446 /* Bail out if there are no loops. */
447 if (vect_loops_num
<= 1)
449 if (cfun
->has_simduid_loops
)
450 adjust_simduid_builtins (&simduid_to_vf_htab
);
454 if (cfun
->has_simduid_loops
)
455 note_simd_array_uses (&simd_array_to_simduid_htab
);
457 init_stmt_vec_info_vec ();
459 /* ----------- Analyze loops. ----------- */
461 /* If some loop was duplicated, it gets bigger number
462 than all previously defined loops. This fact allows us to run
463 only over initial loops skipping newly generated ones. */
464 FOR_EACH_LOOP (loop
, 0)
465 if (loop
->dont_vectorize
)
466 any_ifcvt_loops
= true;
467 else if ((flag_tree_loop_vectorize
468 && optimize_loop_nest_for_speed_p (loop
))
469 || loop
->force_vectorize
)
471 loop_vec_info loop_vinfo
;
472 vect_location
= find_loop_location (loop
);
473 if (LOCATION_LOCUS (vect_location
) != UNKNOWN_LOCATION
474 && dump_enabled_p ())
475 dump_printf (MSG_NOTE
, "\nAnalyzing loop at %s:%d\n",
476 LOCATION_FILE (vect_location
),
477 LOCATION_LINE (vect_location
));
479 loop_vinfo
= vect_analyze_loop (loop
);
480 loop
->aux
= loop_vinfo
;
482 if (!loop_vinfo
|| !LOOP_VINFO_VECTORIZABLE_P (loop_vinfo
))
485 if (!dbg_cnt (vect_loop
))
488 gimple loop_vectorized_call
= vect_loop_vectorized_call (loop
);
489 if (loop_vectorized_call
)
490 set_uid_loop_bbs (loop_vinfo
, loop_vectorized_call
);
491 if (LOCATION_LOCUS (vect_location
) != UNKNOWN_LOCATION
492 && dump_enabled_p ())
493 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS
, vect_location
,
494 "loop vectorized\n");
495 vect_transform_loop (loop_vinfo
);
496 num_vectorized_loops
++;
497 /* Now that the loop has been vectorized, allow it to be unrolled
499 loop
->force_vectorize
= false;
503 simduid_to_vf
*simduid_to_vf_data
= XNEW (simduid_to_vf
);
504 if (!simduid_to_vf_htab
)
505 simduid_to_vf_htab
= new hash_table
<simduid_to_vf
> (15);
506 simduid_to_vf_data
->simduid
= DECL_UID (loop
->simduid
);
507 simduid_to_vf_data
->vf
= loop_vinfo
->vectorization_factor
;
508 *simduid_to_vf_htab
->find_slot (simduid_to_vf_data
, INSERT
)
509 = simduid_to_vf_data
;
512 if (loop_vectorized_call
)
514 fold_loop_vectorized_call (loop_vectorized_call
, boolean_true_node
);
515 ret
|= TODO_cleanup_cfg
;
519 vect_location
= UNKNOWN_LOCATION
;
521 statistics_counter_event (cfun
, "Vectorized loops", num_vectorized_loops
);
522 if (dump_enabled_p ()
523 || (num_vectorized_loops
> 0 && dump_enabled_p ()))
524 dump_printf_loc (MSG_NOTE
, vect_location
,
525 "vectorized %u loops in function.\n",
526 num_vectorized_loops
);
528 /* ----------- Finalize. ----------- */
531 for (i
= 1; i
< vect_loops_num
; i
++)
533 loop
= get_loop (cfun
, i
);
534 if (loop
&& loop
->dont_vectorize
)
536 gimple g
= vect_loop_vectorized_call (loop
);
539 fold_loop_vectorized_call (g
, boolean_false_node
);
540 ret
|= TODO_cleanup_cfg
;
545 for (i
= 1; i
< vect_loops_num
; i
++)
547 loop_vec_info loop_vinfo
;
549 loop
= get_loop (cfun
, i
);
552 loop_vinfo
= (loop_vec_info
) loop
->aux
;
553 destroy_loop_vec_info (loop_vinfo
, true);
557 free_stmt_vec_info_vec ();
559 /* Fold IFN_GOMP_SIMD_{VF,LANE,LAST_LANE} builtins. */
560 if (cfun
->has_simduid_loops
)
561 adjust_simduid_builtins (&simduid_to_vf_htab
);
563 /* Shrink any "omp array simd" temporary arrays to the
564 actual vectorization factors. */
565 if (simd_array_to_simduid_htab
)
567 for (hash_table
<simd_array_to_simduid
>::iterator iter
568 = simd_array_to_simduid_htab
->begin ();
569 iter
!= simd_array_to_simduid_htab
->end (); ++iter
)
570 if ((*iter
)->simduid
!= -1U)
572 tree decl
= (*iter
)->decl
;
574 if (simduid_to_vf_htab
)
576 simduid_to_vf
*p
= NULL
, data
;
577 data
.simduid
= (*iter
)->simduid
;
578 p
= simduid_to_vf_htab
->find (&data
);
583 = build_array_type_nelts (TREE_TYPE (TREE_TYPE (decl
)), vf
);
584 TREE_TYPE (decl
) = atype
;
585 relayout_decl (decl
);
588 delete simd_array_to_simduid_htab
;
590 delete simduid_to_vf_htab
;
591 simduid_to_vf_htab
= NULL
;
593 if (num_vectorized_loops
> 0)
595 /* If we vectorized any loop only virtual SSA form needs to be updated.
596 ??? Also while we try hard to update loop-closed SSA form we fail
597 to properly do this in some corner-cases (see PR56286). */
598 rewrite_into_loop_closed_ssa (NULL
, TODO_update_ssa_only_virtuals
);
599 return TODO_cleanup_cfg
;
606 /* Entry point to basic block SLP phase. */
610 const pass_data pass_data_slp_vectorize
=
612 GIMPLE_PASS
, /* type */
614 OPTGROUP_LOOP
| OPTGROUP_VEC
, /* optinfo_flags */
615 TV_TREE_SLP_VECTORIZATION
, /* tv_id */
616 ( PROP_ssa
| PROP_cfg
), /* properties_required */
617 0, /* properties_provided */
618 0, /* properties_destroyed */
619 0, /* todo_flags_start */
620 TODO_update_ssa
, /* todo_flags_finish */
623 class pass_slp_vectorize
: public gimple_opt_pass
626 pass_slp_vectorize (gcc::context
*ctxt
)
627 : gimple_opt_pass (pass_data_slp_vectorize
, ctxt
)
630 /* opt_pass methods: */
631 opt_pass
* clone () { return new pass_slp_vectorize (m_ctxt
); }
632 virtual bool gate (function
*) { return flag_tree_slp_vectorize
!= 0; }
633 virtual unsigned int execute (function
*);
635 }; // class pass_slp_vectorize
638 pass_slp_vectorize::execute (function
*fun
)
642 bool in_loop_pipeline
= scev_initialized_p ();
643 if (!in_loop_pipeline
)
645 loop_optimizer_init (LOOPS_NORMAL
);
649 init_stmt_vec_info_vec ();
651 FOR_EACH_BB_FN (bb
, fun
)
653 vect_location
= find_bb_location (bb
);
655 if (vect_slp_analyze_bb (bb
))
657 if (!dbg_cnt (vect_slp
))
660 vect_slp_transform_bb (bb
);
661 if (dump_enabled_p ())
662 dump_printf_loc (MSG_OPTIMIZED_LOCATIONS
, vect_location
,
663 "basic block vectorized\n");
667 free_stmt_vec_info_vec ();
669 if (!in_loop_pipeline
)
672 loop_optimizer_finalize ();
681 make_pass_slp_vectorize (gcc::context
*ctxt
)
683 return new pass_slp_vectorize (ctxt
);
687 /* Increase alignment of global arrays to improve vectorization potential.
689 - Consider also structs that have an array field.
690 - Use ipa analysis to prune arrays that can't be vectorized?
691 This should involve global alignment analysis and in the future also
695 increase_alignment (void)
699 vect_location
= UNKNOWN_LOCATION
;
701 /* Increase the alignment of all global arrays for vectorization. */
702 FOR_EACH_DEFINED_VARIABLE (vnode
)
704 tree vectype
, decl
= vnode
->decl
;
706 unsigned int alignment
;
708 t
= TREE_TYPE (decl
);
709 if (TREE_CODE (t
) != ARRAY_TYPE
)
711 vectype
= get_vectype_for_scalar_type (strip_array_types (t
));
714 alignment
= TYPE_ALIGN (vectype
);
715 if (DECL_ALIGN (decl
) >= alignment
)
718 if (vect_can_force_dr_alignment_p (decl
, alignment
))
720 vnode
->increase_alignment (TYPE_ALIGN (vectype
));
721 dump_printf (MSG_NOTE
, "Increasing alignment of decl: ");
722 dump_generic_expr (MSG_NOTE
, TDF_SLIM
, decl
);
723 dump_printf (MSG_NOTE
, "\n");
732 const pass_data pass_data_ipa_increase_alignment
=
734 SIMPLE_IPA_PASS
, /* type */
735 "increase_alignment", /* name */
736 OPTGROUP_LOOP
| OPTGROUP_VEC
, /* optinfo_flags */
737 TV_IPA_OPT
, /* tv_id */
738 0, /* properties_required */
739 0, /* properties_provided */
740 0, /* properties_destroyed */
741 0, /* todo_flags_start */
742 0, /* todo_flags_finish */
745 class pass_ipa_increase_alignment
: public simple_ipa_opt_pass
748 pass_ipa_increase_alignment (gcc::context
*ctxt
)
749 : simple_ipa_opt_pass (pass_data_ipa_increase_alignment
, ctxt
)
752 /* opt_pass methods: */
753 virtual bool gate (function
*)
755 return flag_section_anchors
&& flag_tree_loop_vectorize
;
758 virtual unsigned int execute (function
*) { return increase_alignment (); }
760 }; // class pass_ipa_increase_alignment
764 simple_ipa_opt_pass
*
765 make_pass_ipa_increase_alignment (gcc::context
*ctxt
)
767 return new pass_ipa_increase_alignment (ctxt
);