1 /* Loop Vectorization 2 Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012 3 Free Software Foundation, Inc. 4 Contributed by Dorit Naishlos <dorit@il.ibm.com> and 5 Ira Rosen <irar@il.ibm.com> 6 7 This file is part of GCC. 8 9 GCC is free software; you can redistribute it and/or modify it under 10 the terms of the GNU General Public License as published by the Free 11 Software Foundation; either version 3, or (at your option) any later 12 version. 13 14 GCC is distributed in the hope that it will be useful, but WITHOUT ANY 15 WARRANTY; without even the implied warranty of MERCHANTABILITY or 16 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 17 for more details. 18 19 You should have received a copy of the GNU General Public License 20 along with GCC; see the file COPYING3. If not see 21 <http://www.gnu.org/licenses/>. */ 22 23 #include "config.h" 24 #include "system.h" 25 #include "coretypes.h" 26 #include "tm.h" 27 #include "ggc.h" 28 #include "tree.h" 29 #include "basic-block.h" 30 #include "tree-pretty-print.h" 31 #include "gimple-pretty-print.h" 32 #include "tree-flow.h" 33 #include "tree-dump.h" 34 #include "cfgloop.h" 35 #include "cfglayout.h" 36 #include "expr.h" 37 #include "recog.h" 38 #include "optabs.h" 39 #include "params.h" 40 #include "diagnostic-core.h" 41 #include "tree-chrec.h" 42 #include "tree-scalar-evolution.h" 43 #include "tree-vectorizer.h" 44 #include "target.h" 45 46 /* Loop Vectorization Pass. 47 48 This pass tries to vectorize loops. 49 50 For example, the vectorizer transforms the following simple loop: 51 52 short a[N]; short b[N]; short c[N]; int i; 53 54 for (i=0; i<N; i++){ 55 a[i] = b[i] + c[i]; 56 } 57 58 as if it was manually vectorized by rewriting the source code into: 59 60 typedef int __attribute__((mode(V8HI))) v8hi; 61 short a[N]; short b[N]; short c[N]; int i; 62 v8hi *pa = (v8hi*)a, *pb = (v8hi*)b, *pc = (v8hi*)c; 63 v8hi va, vb, vc; 64 65 for (i=0; i<N/8; i++){ 66 vb = pb[i]; 67 vc = pc[i]; 68 va = vb + vc; 69 pa[i] = va; 70 } 71 72 The main entry to this pass is vectorize_loops(), in which 73 the vectorizer applies a set of analyses on a given set of loops, 74 followed by the actual vectorization transformation for the loops that 75 had successfully passed the analysis phase. 76 Throughout this pass we make a distinction between two types of 77 data: scalars (which are represented by SSA_NAMES), and memory references 78 ("data-refs"). These two types of data require different handling both 79 during analysis and transformation. The types of data-refs that the 80 vectorizer currently supports are ARRAY_REFS which base is an array DECL 81 (not a pointer), and INDIRECT_REFS through pointers; both array and pointer 82 accesses are required to have a simple (consecutive) access pattern. 83 84 Analysis phase: 85 =============== 86 The driver for the analysis phase is vect_analyze_loop(). 87 It applies a set of analyses, some of which rely on the scalar evolution 88 analyzer (scev) developed by Sebastian Pop. 89 90 During the analysis phase the vectorizer records some information 91 per stmt in a "stmt_vec_info" struct which is attached to each stmt in the 92 loop, as well as general information about the loop as a whole, which is 93 recorded in a "loop_vec_info" struct attached to each loop. 94 95 Transformation phase: 96 ===================== 97 The loop transformation phase scans all the stmts in the loop, and 98 creates a vector stmt (or a sequence of stmts) for each scalar stmt S in 99 the loop that needs to be vectorized. It inserts the vector code sequence 100 just before the scalar stmt S, and records a pointer to the vector code 101 in STMT_VINFO_VEC_STMT (stmt_info) (stmt_info is the stmt_vec_info struct 102 attached to S). This pointer will be used for the vectorization of following 103 stmts which use the def of stmt S. Stmt S is removed if it writes to memory; 104 otherwise, we rely on dead code elimination for removing it. 105 106 For example, say stmt S1 was vectorized into stmt VS1: 107 108 VS1: vb = px[i]; 109 S1: b = x[i]; STMT_VINFO_VEC_STMT (stmt_info (S1)) = VS1 110 S2: a = b; 111 112 To vectorize stmt S2, the vectorizer first finds the stmt that defines 113 the operand 'b' (S1), and gets the relevant vector def 'vb' from the 114 vector stmt VS1 pointed to by STMT_VINFO_VEC_STMT (stmt_info (S1)). The 115 resulting sequence would be: 116 117 VS1: vb = px[i]; 118 S1: b = x[i]; STMT_VINFO_VEC_STMT (stmt_info (S1)) = VS1 119 VS2: va = vb; 120 S2: a = b; STMT_VINFO_VEC_STMT (stmt_info (S2)) = VS2 121 122 Operands that are not SSA_NAMEs, are data-refs that appear in 123 load/store operations (like 'x[i]' in S1), and are handled differently. 124 125 Target modeling: 126 ================= 127 Currently the only target specific information that is used is the 128 size of the vector (in bytes) - "TARGET_VECTORIZE_UNITS_PER_SIMD_WORD". 129 Targets that can support different sizes of vectors, for now will need 130 to specify one value for "TARGET_VECTORIZE_UNITS_PER_SIMD_WORD". More 131 flexibility will be added in the future. 132 133 Since we only vectorize operations which vector form can be 134 expressed using existing tree codes, to verify that an operation is 135 supported, the vectorizer checks the relevant optab at the relevant 136 machine_mode (e.g, optab_handler (add_optab, V8HImode)). If 137 the value found is CODE_FOR_nothing, then there's no target support, and 138 we can't vectorize the stmt. 139 140 For additional information on this project see: 141 http://gcc.gnu.org/projects/tree-ssa/vectorization.html 142 */ 143 144 /* Function vect_determine_vectorization_factor 145 146 Determine the vectorization factor (VF). VF is the number of data elements 147 that are operated upon in parallel in a single iteration of the vectorized 148 loop. For example, when vectorizing a loop that operates on 4byte elements, 149 on a target with vector size (VS) 16byte, the VF is set to 4, since 4 150 elements can fit in a single vector register. 151 152 We currently support vectorization of loops in which all types operated upon 153 are of the same size. Therefore this function currently sets VF according to 154 the size of the types operated upon, and fails if there are multiple sizes 155 in the loop. 156 157 VF is also the factor by which the loop iterations are strip-mined, e.g.: 158 original loop: 159 for (i=0; i<N; i++){ 160 a[i] = b[i] + c[i]; 161 } 162 163 vectorized loop: 164 for (i=0; i<N; i+=VF){ 165 a[i:VF] = b[i:VF] + c[i:VF]; 166 } 167 */ 168 169 static bool 170 vect_determine_vectorization_factor (loop_vec_info loop_vinfo) 171 { 172 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 173 basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo); 174 int nbbs = loop->num_nodes; 175 gimple_stmt_iterator si; 176 unsigned int vectorization_factor = 0; 177 tree scalar_type; 178 gimple phi; 179 tree vectype; 180 unsigned int nunits; 181 stmt_vec_info stmt_info; 182 int i; 183 HOST_WIDE_INT dummy; 184 gimple stmt, pattern_stmt = NULL; 185 gimple_seq pattern_def_seq = NULL; 186 gimple_stmt_iterator pattern_def_si = gsi_start (NULL); 187 bool analyze_pattern_stmt = false; 188 189 if (vect_print_dump_info (REPORT_DETAILS)) 190 fprintf (vect_dump, "=== vect_determine_vectorization_factor ==="); 191 192 for (i = 0; i < nbbs; i++) 193 { 194 basic_block bb = bbs[i]; 195 196 for (si = gsi_start_phis (bb); !gsi_end_p (si); gsi_next (&si)) 197 { 198 phi = gsi_stmt (si); 199 stmt_info = vinfo_for_stmt (phi); 200 if (vect_print_dump_info (REPORT_DETAILS)) 201 { 202 fprintf (vect_dump, "==> examining phi: "); 203 print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); 204 } 205 206 gcc_assert (stmt_info); 207 208 if (STMT_VINFO_RELEVANT_P (stmt_info)) 209 { 210 gcc_assert (!STMT_VINFO_VECTYPE (stmt_info)); 211 scalar_type = TREE_TYPE (PHI_RESULT (phi)); 212 213 if (vect_print_dump_info (REPORT_DETAILS)) 214 { 215 fprintf (vect_dump, "get vectype for scalar type: "); 216 print_generic_expr (vect_dump, scalar_type, TDF_SLIM); 217 } 218 219 vectype = get_vectype_for_scalar_type (scalar_type); 220 if (!vectype) 221 { 222 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 223 { 224 fprintf (vect_dump, 225 "not vectorized: unsupported data-type "); 226 print_generic_expr (vect_dump, scalar_type, TDF_SLIM); 227 } 228 return false; 229 } 230 STMT_VINFO_VECTYPE (stmt_info) = vectype; 231 232 if (vect_print_dump_info (REPORT_DETAILS)) 233 { 234 fprintf (vect_dump, "vectype: "); 235 print_generic_expr (vect_dump, vectype, TDF_SLIM); 236 } 237 238 nunits = TYPE_VECTOR_SUBPARTS (vectype); 239 if (vect_print_dump_info (REPORT_DETAILS)) 240 fprintf (vect_dump, "nunits = %d", nunits); 241 242 if (!vectorization_factor 243 || (nunits > vectorization_factor)) 244 vectorization_factor = nunits; 245 } 246 } 247 248 for (si = gsi_start_bb (bb); !gsi_end_p (si) || analyze_pattern_stmt;) 249 { 250 tree vf_vectype; 251 252 if (analyze_pattern_stmt) 253 stmt = pattern_stmt; 254 else 255 stmt = gsi_stmt (si); 256 257 stmt_info = vinfo_for_stmt (stmt); 258 259 if (vect_print_dump_info (REPORT_DETAILS)) 260 { 261 fprintf (vect_dump, "==> examining statement: "); 262 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 263 } 264 265 gcc_assert (stmt_info); 266 267 /* Skip stmts which do not need to be vectorized. */ 268 if (!STMT_VINFO_RELEVANT_P (stmt_info) 269 && !STMT_VINFO_LIVE_P (stmt_info)) 270 { 271 if (STMT_VINFO_IN_PATTERN_P (stmt_info) 272 && (pattern_stmt = STMT_VINFO_RELATED_STMT (stmt_info)) 273 && (STMT_VINFO_RELEVANT_P (vinfo_for_stmt (pattern_stmt)) 274 || STMT_VINFO_LIVE_P (vinfo_for_stmt (pattern_stmt)))) 275 { 276 stmt = pattern_stmt; 277 stmt_info = vinfo_for_stmt (pattern_stmt); 278 if (vect_print_dump_info (REPORT_DETAILS)) 279 { 280 fprintf (vect_dump, "==> examining pattern statement: "); 281 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 282 } 283 } 284 else 285 { 286 if (vect_print_dump_info (REPORT_DETAILS)) 287 fprintf (vect_dump, "skip."); 288 gsi_next (&si); 289 continue; 290 } 291 } 292 else if (STMT_VINFO_IN_PATTERN_P (stmt_info) 293 && (pattern_stmt = STMT_VINFO_RELATED_STMT (stmt_info)) 294 && (STMT_VINFO_RELEVANT_P (vinfo_for_stmt (pattern_stmt)) 295 || STMT_VINFO_LIVE_P (vinfo_for_stmt (pattern_stmt)))) 296 analyze_pattern_stmt = true; 297 298 /* If a pattern statement has def stmts, analyze them too. */ 299 if (is_pattern_stmt_p (stmt_info)) 300 { 301 if (pattern_def_seq == NULL) 302 { 303 pattern_def_seq = STMT_VINFO_PATTERN_DEF_SEQ (stmt_info); 304 pattern_def_si = gsi_start (pattern_def_seq); 305 } 306 else if (!gsi_end_p (pattern_def_si)) 307 gsi_next (&pattern_def_si); 308 if (pattern_def_seq != NULL) 309 { 310 gimple pattern_def_stmt = NULL; 311 stmt_vec_info pattern_def_stmt_info = NULL; 312 313 while (!gsi_end_p (pattern_def_si)) 314 { 315 pattern_def_stmt = gsi_stmt (pattern_def_si); 316 pattern_def_stmt_info 317 = vinfo_for_stmt (pattern_def_stmt); 318 if (STMT_VINFO_RELEVANT_P (pattern_def_stmt_info) 319 || STMT_VINFO_LIVE_P (pattern_def_stmt_info)) 320 break; 321 gsi_next (&pattern_def_si); 322 } 323 324 if (!gsi_end_p (pattern_def_si)) 325 { 326 if (vect_print_dump_info (REPORT_DETAILS)) 327 { 328 fprintf (vect_dump, 329 "==> examining pattern def stmt: "); 330 print_gimple_stmt (vect_dump, pattern_def_stmt, 0, 331 TDF_SLIM); 332 } 333 334 stmt = pattern_def_stmt; 335 stmt_info = pattern_def_stmt_info; 336 } 337 else 338 { 339 pattern_def_si = gsi_start (NULL); 340 analyze_pattern_stmt = false; 341 } 342 } 343 else 344 analyze_pattern_stmt = false; 345 } 346 347 if (gimple_get_lhs (stmt) == NULL_TREE) 348 { 349 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 350 { 351 fprintf (vect_dump, "not vectorized: irregular stmt."); 352 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 353 } 354 return false; 355 } 356 357 if (VECTOR_MODE_P (TYPE_MODE (gimple_expr_type (stmt)))) 358 { 359 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 360 { 361 fprintf (vect_dump, "not vectorized: vector stmt in loop:"); 362 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 363 } 364 return false; 365 } 366 367 if (STMT_VINFO_VECTYPE (stmt_info)) 368 { 369 /* The only case when a vectype had been already set is for stmts 370 that contain a dataref, or for "pattern-stmts" (stmts 371 generated by the vectorizer to represent/replace a certain 372 idiom). */ 373 gcc_assert (STMT_VINFO_DATA_REF (stmt_info) 374 || is_pattern_stmt_p (stmt_info) 375 || !gsi_end_p (pattern_def_si)); 376 vectype = STMT_VINFO_VECTYPE (stmt_info); 377 } 378 else 379 { 380 gcc_assert (!STMT_VINFO_DATA_REF (stmt_info)); 381 scalar_type = TREE_TYPE (gimple_get_lhs (stmt)); 382 if (vect_print_dump_info (REPORT_DETAILS)) 383 { 384 fprintf (vect_dump, "get vectype for scalar type: "); 385 print_generic_expr (vect_dump, scalar_type, TDF_SLIM); 386 } 387 vectype = get_vectype_for_scalar_type (scalar_type); 388 if (!vectype) 389 { 390 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 391 { 392 fprintf (vect_dump, 393 "not vectorized: unsupported data-type "); 394 print_generic_expr (vect_dump, scalar_type, TDF_SLIM); 395 } 396 return false; 397 } 398 399 STMT_VINFO_VECTYPE (stmt_info) = vectype; 400 } 401 402 /* The vectorization factor is according to the smallest 403 scalar type (or the largest vector size, but we only 404 support one vector size per loop). */ 405 scalar_type = vect_get_smallest_scalar_type (stmt, &dummy, 406 &dummy); 407 if (vect_print_dump_info (REPORT_DETAILS)) 408 { 409 fprintf (vect_dump, "get vectype for scalar type: "); 410 print_generic_expr (vect_dump, scalar_type, TDF_SLIM); 411 } 412 vf_vectype = get_vectype_for_scalar_type (scalar_type); 413 if (!vf_vectype) 414 { 415 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 416 { 417 fprintf (vect_dump, 418 "not vectorized: unsupported data-type "); 419 print_generic_expr (vect_dump, scalar_type, TDF_SLIM); 420 } 421 return false; 422 } 423 424 if ((GET_MODE_SIZE (TYPE_MODE (vectype)) 425 != GET_MODE_SIZE (TYPE_MODE (vf_vectype)))) 426 { 427 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 428 { 429 fprintf (vect_dump, 430 "not vectorized: different sized vector " 431 "types in statement, "); 432 print_generic_expr (vect_dump, vectype, TDF_SLIM); 433 fprintf (vect_dump, " and "); 434 print_generic_expr (vect_dump, vf_vectype, TDF_SLIM); 435 } 436 return false; 437 } 438 439 if (vect_print_dump_info (REPORT_DETAILS)) 440 { 441 fprintf (vect_dump, "vectype: "); 442 print_generic_expr (vect_dump, vf_vectype, TDF_SLIM); 443 } 444 445 nunits = TYPE_VECTOR_SUBPARTS (vf_vectype); 446 if (vect_print_dump_info (REPORT_DETAILS)) 447 fprintf (vect_dump, "nunits = %d", nunits); 448 449 if (!vectorization_factor 450 || (nunits > vectorization_factor)) 451 vectorization_factor = nunits; 452 453 if (!analyze_pattern_stmt && gsi_end_p (pattern_def_si)) 454 { 455 pattern_def_seq = NULL; 456 gsi_next (&si); 457 } 458 } 459 } 460 461 /* TODO: Analyze cost. Decide if worth while to vectorize. */ 462 if (vect_print_dump_info (REPORT_DETAILS)) 463 fprintf (vect_dump, "vectorization factor = %d", vectorization_factor); 464 if (vectorization_factor <= 1) 465 { 466 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 467 fprintf (vect_dump, "not vectorized: unsupported data-type"); 468 return false; 469 } 470 LOOP_VINFO_VECT_FACTOR (loop_vinfo) = vectorization_factor; 471 472 return true; 473 } 474 475 476 /* Function vect_is_simple_iv_evolution. 477 478 FORNOW: A simple evolution of an induction variables in the loop is 479 considered a polynomial evolution with constant step. */ 480 481 static bool 482 vect_is_simple_iv_evolution (unsigned loop_nb, tree access_fn, tree * init, 483 tree * step) 484 { 485 tree init_expr; 486 tree step_expr; 487 tree evolution_part = evolution_part_in_loop_num (access_fn, loop_nb); 488 489 /* When there is no evolution in this loop, the evolution function 490 is not "simple". */ 491 if (evolution_part == NULL_TREE) 492 return false; 493 494 /* When the evolution is a polynomial of degree >= 2 495 the evolution function is not "simple". */ 496 if (tree_is_chrec (evolution_part)) 497 return false; 498 499 step_expr = evolution_part; 500 init_expr = unshare_expr (initial_condition_in_loop_num (access_fn, loop_nb)); 501 502 if (vect_print_dump_info (REPORT_DETAILS)) 503 { 504 fprintf (vect_dump, "step: "); 505 print_generic_expr (vect_dump, step_expr, TDF_SLIM); 506 fprintf (vect_dump, ", init: "); 507 print_generic_expr (vect_dump, init_expr, TDF_SLIM); 508 } 509 510 *init = init_expr; 511 *step = step_expr; 512 513 if (TREE_CODE (step_expr) != INTEGER_CST) 514 { 515 if (vect_print_dump_info (REPORT_DETAILS)) 516 fprintf (vect_dump, "step unknown."); 517 return false; 518 } 519 520 return true; 521 } 522 523 /* Function vect_analyze_scalar_cycles_1. 524 525 Examine the cross iteration def-use cycles of scalar variables 526 in LOOP. LOOP_VINFO represents the loop that is now being 527 considered for vectorization (can be LOOP, or an outer-loop 528 enclosing LOOP). */ 529 530 static void 531 vect_analyze_scalar_cycles_1 (loop_vec_info loop_vinfo, struct loop *loop) 532 { 533 basic_block bb = loop->header; 534 tree dumy; 535 VEC(gimple,heap) *worklist = VEC_alloc (gimple, heap, 64); 536 gimple_stmt_iterator gsi; 537 bool double_reduc; 538 539 if (vect_print_dump_info (REPORT_DETAILS)) 540 fprintf (vect_dump, "=== vect_analyze_scalar_cycles ==="); 541 542 /* First - identify all inductions. Reduction detection assumes that all the 543 inductions have been identified, therefore, this order must not be 544 changed. */ 545 for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi)) 546 { 547 gimple phi = gsi_stmt (gsi); 548 tree access_fn = NULL; 549 tree def = PHI_RESULT (phi); 550 stmt_vec_info stmt_vinfo = vinfo_for_stmt (phi); 551 552 if (vect_print_dump_info (REPORT_DETAILS)) 553 { 554 fprintf (vect_dump, "Analyze phi: "); 555 print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); 556 } 557 558 /* Skip virtual phi's. The data dependences that are associated with 559 virtual defs/uses (i.e., memory accesses) are analyzed elsewhere. */ 560 if (!is_gimple_reg (SSA_NAME_VAR (def))) 561 continue; 562 563 STMT_VINFO_DEF_TYPE (stmt_vinfo) = vect_unknown_def_type; 564 565 /* Analyze the evolution function. */ 566 access_fn = analyze_scalar_evolution (loop, def); 567 if (access_fn) 568 { 569 STRIP_NOPS (access_fn); 570 if (vect_print_dump_info (REPORT_DETAILS)) 571 { 572 fprintf (vect_dump, "Access function of PHI: "); 573 print_generic_expr (vect_dump, access_fn, TDF_SLIM); 574 } 575 STMT_VINFO_LOOP_PHI_EVOLUTION_PART (stmt_vinfo) 576 = evolution_part_in_loop_num (access_fn, loop->num); 577 } 578 579 if (!access_fn 580 || !vect_is_simple_iv_evolution (loop->num, access_fn, &dumy, &dumy)) 581 { 582 VEC_safe_push (gimple, heap, worklist, phi); 583 continue; 584 } 585 586 gcc_assert (STMT_VINFO_LOOP_PHI_EVOLUTION_PART (stmt_vinfo) != NULL_TREE); 587 588 if (vect_print_dump_info (REPORT_DETAILS)) 589 fprintf (vect_dump, "Detected induction."); 590 STMT_VINFO_DEF_TYPE (stmt_vinfo) = vect_induction_def; 591 } 592 593 594 /* Second - identify all reductions and nested cycles. */ 595 while (VEC_length (gimple, worklist) > 0) 596 { 597 gimple phi = VEC_pop (gimple, worklist); 598 tree def = PHI_RESULT (phi); 599 stmt_vec_info stmt_vinfo = vinfo_for_stmt (phi); 600 gimple reduc_stmt; 601 bool nested_cycle; 602 603 if (vect_print_dump_info (REPORT_DETAILS)) 604 { 605 fprintf (vect_dump, "Analyze phi: "); 606 print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); 607 } 608 609 gcc_assert (is_gimple_reg (SSA_NAME_VAR (def))); 610 gcc_assert (STMT_VINFO_DEF_TYPE (stmt_vinfo) == vect_unknown_def_type); 611 612 nested_cycle = (loop != LOOP_VINFO_LOOP (loop_vinfo)); 613 reduc_stmt = vect_force_simple_reduction (loop_vinfo, phi, !nested_cycle, 614 &double_reduc); 615 if (reduc_stmt) 616 { 617 if (double_reduc) 618 { 619 if (vect_print_dump_info (REPORT_DETAILS)) 620 fprintf (vect_dump, "Detected double reduction."); 621 622 STMT_VINFO_DEF_TYPE (stmt_vinfo) = vect_double_reduction_def; 623 STMT_VINFO_DEF_TYPE (vinfo_for_stmt (reduc_stmt)) = 624 vect_double_reduction_def; 625 } 626 else 627 { 628 if (nested_cycle) 629 { 630 if (vect_print_dump_info (REPORT_DETAILS)) 631 fprintf (vect_dump, "Detected vectorizable nested cycle."); 632 633 STMT_VINFO_DEF_TYPE (stmt_vinfo) = vect_nested_cycle; 634 STMT_VINFO_DEF_TYPE (vinfo_for_stmt (reduc_stmt)) = 635 vect_nested_cycle; 636 } 637 else 638 { 639 if (vect_print_dump_info (REPORT_DETAILS)) 640 fprintf (vect_dump, "Detected reduction."); 641 642 STMT_VINFO_DEF_TYPE (stmt_vinfo) = vect_reduction_def; 643 STMT_VINFO_DEF_TYPE (vinfo_for_stmt (reduc_stmt)) = 644 vect_reduction_def; 645 /* Store the reduction cycles for possible vectorization in 646 loop-aware SLP. */ 647 VEC_safe_push (gimple, heap, 648 LOOP_VINFO_REDUCTIONS (loop_vinfo), 649 reduc_stmt); 650 } 651 } 652 } 653 else 654 if (vect_print_dump_info (REPORT_DETAILS)) 655 fprintf (vect_dump, "Unknown def-use cycle pattern."); 656 } 657 658 VEC_free (gimple, heap, worklist); 659 } 660 661 662 /* Function vect_analyze_scalar_cycles. 663 664 Examine the cross iteration def-use cycles of scalar variables, by 665 analyzing the loop-header PHIs of scalar variables. Classify each 666 cycle as one of the following: invariant, induction, reduction, unknown. 667 We do that for the loop represented by LOOP_VINFO, and also to its 668 inner-loop, if exists. 669 Examples for scalar cycles: 670 671 Example1: reduction: 672 673 loop1: 674 for (i=0; i<N; i++) 675 sum += a[i]; 676 677 Example2: induction: 678 679 loop2: 680 for (i=0; i<N; i++) 681 a[i] = i; */ 682 683 static void 684 vect_analyze_scalar_cycles (loop_vec_info loop_vinfo) 685 { 686 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 687 688 vect_analyze_scalar_cycles_1 (loop_vinfo, loop); 689 690 /* When vectorizing an outer-loop, the inner-loop is executed sequentially. 691 Reductions in such inner-loop therefore have different properties than 692 the reductions in the nest that gets vectorized: 693 1. When vectorized, they are executed in the same order as in the original 694 scalar loop, so we can't change the order of computation when 695 vectorizing them. 696 2. FIXME: Inner-loop reductions can be used in the inner-loop, so the 697 current checks are too strict. */ 698 699 if (loop->inner) 700 vect_analyze_scalar_cycles_1 (loop_vinfo, loop->inner); 701 } 702 703 /* Function vect_get_loop_niters. 704 705 Determine how many iterations the loop is executed. 706 If an expression that represents the number of iterations 707 can be constructed, place it in NUMBER_OF_ITERATIONS. 708 Return the loop exit condition. */ 709 710 static gimple 711 vect_get_loop_niters (struct loop *loop, tree *number_of_iterations) 712 { 713 tree niters; 714 715 if (vect_print_dump_info (REPORT_DETAILS)) 716 fprintf (vect_dump, "=== get_loop_niters ==="); 717 718 niters = number_of_exit_cond_executions (loop); 719 720 if (niters != NULL_TREE 721 && niters != chrec_dont_know) 722 { 723 *number_of_iterations = niters; 724 725 if (vect_print_dump_info (REPORT_DETAILS)) 726 { 727 fprintf (vect_dump, "==> get_loop_niters:" ); 728 print_generic_expr (vect_dump, *number_of_iterations, TDF_SLIM); 729 } 730 } 731 732 return get_loop_exit_condition (loop); 733 } 734 735 736 /* Function bb_in_loop_p 737 738 Used as predicate for dfs order traversal of the loop bbs. */ 739 740 static bool 741 bb_in_loop_p (const_basic_block bb, const void *data) 742 { 743 const struct loop *const loop = (const struct loop *)data; 744 if (flow_bb_inside_loop_p (loop, bb)) 745 return true; 746 return false; 747 } 748 749 750 /* Function new_loop_vec_info. 751 752 Create and initialize a new loop_vec_info struct for LOOP, as well as 753 stmt_vec_info structs for all the stmts in LOOP. */ 754 755 static loop_vec_info 756 new_loop_vec_info (struct loop *loop) 757 { 758 loop_vec_info res; 759 basic_block *bbs; 760 gimple_stmt_iterator si; 761 unsigned int i, nbbs; 762 763 res = (loop_vec_info) xcalloc (1, sizeof (struct _loop_vec_info)); 764 LOOP_VINFO_LOOP (res) = loop; 765 766 bbs = get_loop_body (loop); 767 768 /* Create/Update stmt_info for all stmts in the loop. */ 769 for (i = 0; i < loop->num_nodes; i++) 770 { 771 basic_block bb = bbs[i]; 772 773 /* BBs in a nested inner-loop will have been already processed (because 774 we will have called vect_analyze_loop_form for any nested inner-loop). 775 Therefore, for stmts in an inner-loop we just want to update the 776 STMT_VINFO_LOOP_VINFO field of their stmt_info to point to the new 777 loop_info of the outer-loop we are currently considering to vectorize 778 (instead of the loop_info of the inner-loop). 779 For stmts in other BBs we need to create a stmt_info from scratch. */ 780 if (bb->loop_father != loop) 781 { 782 /* Inner-loop bb. */ 783 gcc_assert (loop->inner && bb->loop_father == loop->inner); 784 for (si = gsi_start_phis (bb); !gsi_end_p (si); gsi_next (&si)) 785 { 786 gimple phi = gsi_stmt (si); 787 stmt_vec_info stmt_info = vinfo_for_stmt (phi); 788 loop_vec_info inner_loop_vinfo = 789 STMT_VINFO_LOOP_VINFO (stmt_info); 790 gcc_assert (loop->inner == LOOP_VINFO_LOOP (inner_loop_vinfo)); 791 STMT_VINFO_LOOP_VINFO (stmt_info) = res; 792 } 793 for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si)) 794 { 795 gimple stmt = gsi_stmt (si); 796 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 797 loop_vec_info inner_loop_vinfo = 798 STMT_VINFO_LOOP_VINFO (stmt_info); 799 gcc_assert (loop->inner == LOOP_VINFO_LOOP (inner_loop_vinfo)); 800 STMT_VINFO_LOOP_VINFO (stmt_info) = res; 801 } 802 } 803 else 804 { 805 /* bb in current nest. */ 806 for (si = gsi_start_phis (bb); !gsi_end_p (si); gsi_next (&si)) 807 { 808 gimple phi = gsi_stmt (si); 809 gimple_set_uid (phi, 0); 810 set_vinfo_for_stmt (phi, new_stmt_vec_info (phi, res, NULL)); 811 } 812 813 for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si)) 814 { 815 gimple stmt = gsi_stmt (si); 816 gimple_set_uid (stmt, 0); 817 set_vinfo_for_stmt (stmt, new_stmt_vec_info (stmt, res, NULL)); 818 } 819 } 820 } 821 822 /* CHECKME: We want to visit all BBs before their successors (except for 823 latch blocks, for which this assertion wouldn't hold). In the simple 824 case of the loop forms we allow, a dfs order of the BBs would the same 825 as reversed postorder traversal, so we are safe. */ 826 827 free (bbs); 828 bbs = XCNEWVEC (basic_block, loop->num_nodes); 829 nbbs = dfs_enumerate_from (loop->header, 0, bb_in_loop_p, 830 bbs, loop->num_nodes, loop); 831 gcc_assert (nbbs == loop->num_nodes); 832 833 LOOP_VINFO_BBS (res) = bbs; 834 LOOP_VINFO_NITERS (res) = NULL; 835 LOOP_VINFO_NITERS_UNCHANGED (res) = NULL; 836 LOOP_VINFO_COST_MODEL_MIN_ITERS (res) = 0; 837 LOOP_VINFO_VECTORIZABLE_P (res) = 0; 838 LOOP_PEELING_FOR_ALIGNMENT (res) = 0; 839 LOOP_VINFO_VECT_FACTOR (res) = 0; 840 LOOP_VINFO_LOOP_NEST (res) = VEC_alloc (loop_p, heap, 3); 841 LOOP_VINFO_DATAREFS (res) = VEC_alloc (data_reference_p, heap, 10); 842 LOOP_VINFO_DDRS (res) = VEC_alloc (ddr_p, heap, 10 * 10); 843 LOOP_VINFO_UNALIGNED_DR (res) = NULL; 844 LOOP_VINFO_MAY_MISALIGN_STMTS (res) = 845 VEC_alloc (gimple, heap, 846 PARAM_VALUE (PARAM_VECT_MAX_VERSION_FOR_ALIGNMENT_CHECKS)); 847 LOOP_VINFO_MAY_ALIAS_DDRS (res) = 848 VEC_alloc (ddr_p, heap, 849 PARAM_VALUE (PARAM_VECT_MAX_VERSION_FOR_ALIAS_CHECKS)); 850 LOOP_VINFO_STRIDED_STORES (res) = VEC_alloc (gimple, heap, 10); 851 LOOP_VINFO_REDUCTIONS (res) = VEC_alloc (gimple, heap, 10); 852 LOOP_VINFO_REDUCTION_CHAINS (res) = VEC_alloc (gimple, heap, 10); 853 LOOP_VINFO_SLP_INSTANCES (res) = VEC_alloc (slp_instance, heap, 10); 854 LOOP_VINFO_SLP_UNROLLING_FACTOR (res) = 1; 855 LOOP_VINFO_PEELING_HTAB (res) = NULL; 856 LOOP_VINFO_PEELING_FOR_GAPS (res) = false; 857 858 return res; 859 } 860 861 862 /* Function destroy_loop_vec_info. 863 864 Free LOOP_VINFO struct, as well as all the stmt_vec_info structs of all the 865 stmts in the loop. */ 866 867 void 868 destroy_loop_vec_info (loop_vec_info loop_vinfo, bool clean_stmts) 869 { 870 struct loop *loop; 871 basic_block *bbs; 872 int nbbs; 873 gimple_stmt_iterator si; 874 int j; 875 VEC (slp_instance, heap) *slp_instances; 876 slp_instance instance; 877 878 if (!loop_vinfo) 879 return; 880 881 loop = LOOP_VINFO_LOOP (loop_vinfo); 882 883 bbs = LOOP_VINFO_BBS (loop_vinfo); 884 nbbs = loop->num_nodes; 885 886 if (!clean_stmts) 887 { 888 free (LOOP_VINFO_BBS (loop_vinfo)); 889 free_data_refs (LOOP_VINFO_DATAREFS (loop_vinfo)); 890 free_dependence_relations (LOOP_VINFO_DDRS (loop_vinfo)); 891 VEC_free (loop_p, heap, LOOP_VINFO_LOOP_NEST (loop_vinfo)); 892 VEC_free (gimple, heap, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)); 893 VEC_free (ddr_p, heap, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo)); 894 895 free (loop_vinfo); 896 loop->aux = NULL; 897 return; 898 } 899 900 for (j = 0; j < nbbs; j++) 901 { 902 basic_block bb = bbs[j]; 903 for (si = gsi_start_phis (bb); !gsi_end_p (si); gsi_next (&si)) 904 free_stmt_vec_info (gsi_stmt (si)); 905 906 for (si = gsi_start_bb (bb); !gsi_end_p (si); ) 907 { 908 gimple stmt = gsi_stmt (si); 909 /* Free stmt_vec_info. */ 910 free_stmt_vec_info (stmt); 911 gsi_next (&si); 912 } 913 } 914 915 free (LOOP_VINFO_BBS (loop_vinfo)); 916 free_data_refs (LOOP_VINFO_DATAREFS (loop_vinfo)); 917 free_dependence_relations (LOOP_VINFO_DDRS (loop_vinfo)); 918 VEC_free (loop_p, heap, LOOP_VINFO_LOOP_NEST (loop_vinfo)); 919 VEC_free (gimple, heap, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)); 920 VEC_free (ddr_p, heap, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo)); 921 slp_instances = LOOP_VINFO_SLP_INSTANCES (loop_vinfo); 922 FOR_EACH_VEC_ELT (slp_instance, slp_instances, j, instance) 923 vect_free_slp_instance (instance); 924 925 VEC_free (slp_instance, heap, LOOP_VINFO_SLP_INSTANCES (loop_vinfo)); 926 VEC_free (gimple, heap, LOOP_VINFO_STRIDED_STORES (loop_vinfo)); 927 VEC_free (gimple, heap, LOOP_VINFO_REDUCTIONS (loop_vinfo)); 928 VEC_free (gimple, heap, LOOP_VINFO_REDUCTION_CHAINS (loop_vinfo)); 929 930 if (LOOP_VINFO_PEELING_HTAB (loop_vinfo)) 931 htab_delete (LOOP_VINFO_PEELING_HTAB (loop_vinfo)); 932 933 free (loop_vinfo); 934 loop->aux = NULL; 935 } 936 937 938 /* Function vect_analyze_loop_1. 939 940 Apply a set of analyses on LOOP, and create a loop_vec_info struct 941 for it. The different analyses will record information in the 942 loop_vec_info struct. This is a subset of the analyses applied in 943 vect_analyze_loop, to be applied on an inner-loop nested in the loop 944 that is now considered for (outer-loop) vectorization. */ 945 946 static loop_vec_info 947 vect_analyze_loop_1 (struct loop *loop) 948 { 949 loop_vec_info loop_vinfo; 950 951 if (vect_print_dump_info (REPORT_DETAILS)) 952 fprintf (vect_dump, "===== analyze_loop_nest_1 ====="); 953 954 /* Check the CFG characteristics of the loop (nesting, entry/exit, etc. */ 955 956 loop_vinfo = vect_analyze_loop_form (loop); 957 if (!loop_vinfo) 958 { 959 if (vect_print_dump_info (REPORT_DETAILS)) 960 fprintf (vect_dump, "bad inner-loop form."); 961 return NULL; 962 } 963 964 return loop_vinfo; 965 } 966 967 968 /* Function vect_analyze_loop_form. 969 970 Verify that certain CFG restrictions hold, including: 971 - the loop has a pre-header 972 - the loop has a single entry and exit 973 - the loop exit condition is simple enough, and the number of iterations 974 can be analyzed (a countable loop). */ 975 976 loop_vec_info 977 vect_analyze_loop_form (struct loop *loop) 978 { 979 loop_vec_info loop_vinfo; 980 gimple loop_cond; 981 tree number_of_iterations = NULL; 982 loop_vec_info inner_loop_vinfo = NULL; 983 984 if (vect_print_dump_info (REPORT_DETAILS)) 985 fprintf (vect_dump, "=== vect_analyze_loop_form ==="); 986 987 /* Different restrictions apply when we are considering an inner-most loop, 988 vs. an outer (nested) loop. 989 (FORNOW. May want to relax some of these restrictions in the future). */ 990 991 if (!loop->inner) 992 { 993 /* Inner-most loop. We currently require that the number of BBs is 994 exactly 2 (the header and latch). Vectorizable inner-most loops 995 look like this: 996 997 (pre-header) 998 | 999 header <--------+ 1000 | | | 1001 | +--> latch --+ 1002 | 1003 (exit-bb) */ 1004 1005 if (loop->num_nodes != 2) 1006 { 1007 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1008 fprintf (vect_dump, "not vectorized: control flow in loop."); 1009 return NULL; 1010 } 1011 1012 if (empty_block_p (loop->header)) 1013 { 1014 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1015 fprintf (vect_dump, "not vectorized: empty loop."); 1016 return NULL; 1017 } 1018 } 1019 else 1020 { 1021 struct loop *innerloop = loop->inner; 1022 edge entryedge; 1023 1024 /* Nested loop. We currently require that the loop is doubly-nested, 1025 contains a single inner loop, and the number of BBs is exactly 5. 1026 Vectorizable outer-loops look like this: 1027 1028 (pre-header) 1029 | 1030 header <---+ 1031 | | 1032 inner-loop | 1033 | | 1034 tail ------+ 1035 | 1036 (exit-bb) 1037 1038 The inner-loop has the properties expected of inner-most loops 1039 as described above. */ 1040 1041 if ((loop->inner)->inner || (loop->inner)->next) 1042 { 1043 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1044 fprintf (vect_dump, "not vectorized: multiple nested loops."); 1045 return NULL; 1046 } 1047 1048 /* Analyze the inner-loop. */ 1049 inner_loop_vinfo = vect_analyze_loop_1 (loop->inner); 1050 if (!inner_loop_vinfo) 1051 { 1052 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1053 fprintf (vect_dump, "not vectorized: Bad inner loop."); 1054 return NULL; 1055 } 1056 1057 if (!expr_invariant_in_loop_p (loop, 1058 LOOP_VINFO_NITERS (inner_loop_vinfo))) 1059 { 1060 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1061 fprintf (vect_dump, 1062 "not vectorized: inner-loop count not invariant."); 1063 destroy_loop_vec_info (inner_loop_vinfo, true); 1064 return NULL; 1065 } 1066 1067 if (loop->num_nodes != 5) 1068 { 1069 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1070 fprintf (vect_dump, "not vectorized: control flow in loop."); 1071 destroy_loop_vec_info (inner_loop_vinfo, true); 1072 return NULL; 1073 } 1074 1075 gcc_assert (EDGE_COUNT (innerloop->header->preds) == 2); 1076 entryedge = EDGE_PRED (innerloop->header, 0); 1077 if (EDGE_PRED (innerloop->header, 0)->src == innerloop->latch) 1078 entryedge = EDGE_PRED (innerloop->header, 1); 1079 1080 if (entryedge->src != loop->header 1081 || !single_exit (innerloop) 1082 || single_exit (innerloop)->dest != EDGE_PRED (loop->latch, 0)->src) 1083 { 1084 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1085 fprintf (vect_dump, "not vectorized: unsupported outerloop form."); 1086 destroy_loop_vec_info (inner_loop_vinfo, true); 1087 return NULL; 1088 } 1089 1090 if (vect_print_dump_info (REPORT_DETAILS)) 1091 fprintf (vect_dump, "Considering outer-loop vectorization."); 1092 } 1093 1094 if (!single_exit (loop) 1095 || EDGE_COUNT (loop->header->preds) != 2) 1096 { 1097 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1098 { 1099 if (!single_exit (loop)) 1100 fprintf (vect_dump, "not vectorized: multiple exits."); 1101 else if (EDGE_COUNT (loop->header->preds) != 2) 1102 fprintf (vect_dump, "not vectorized: too many incoming edges."); 1103 } 1104 if (inner_loop_vinfo) 1105 destroy_loop_vec_info (inner_loop_vinfo, true); 1106 return NULL; 1107 } 1108 1109 /* We assume that the loop exit condition is at the end of the loop. i.e, 1110 that the loop is represented as a do-while (with a proper if-guard 1111 before the loop if needed), where the loop header contains all the 1112 executable statements, and the latch is empty. */ 1113 if (!empty_block_p (loop->latch) 1114 || !gimple_seq_empty_p (phi_nodes (loop->latch))) 1115 { 1116 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1117 fprintf (vect_dump, "not vectorized: unexpected loop form."); 1118 if (inner_loop_vinfo) 1119 destroy_loop_vec_info (inner_loop_vinfo, true); 1120 return NULL; 1121 } 1122 1123 /* Make sure there exists a single-predecessor exit bb: */ 1124 if (!single_pred_p (single_exit (loop)->dest)) 1125 { 1126 edge e = single_exit (loop); 1127 if (!(e->flags & EDGE_ABNORMAL)) 1128 { 1129 split_loop_exit_edge (e); 1130 if (vect_print_dump_info (REPORT_DETAILS)) 1131 fprintf (vect_dump, "split exit edge."); 1132 } 1133 else 1134 { 1135 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1136 fprintf (vect_dump, "not vectorized: abnormal loop exit edge."); 1137 if (inner_loop_vinfo) 1138 destroy_loop_vec_info (inner_loop_vinfo, true); 1139 return NULL; 1140 } 1141 } 1142 1143 loop_cond = vect_get_loop_niters (loop, &number_of_iterations); 1144 if (!loop_cond) 1145 { 1146 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1147 fprintf (vect_dump, "not vectorized: complicated exit condition."); 1148 if (inner_loop_vinfo) 1149 destroy_loop_vec_info (inner_loop_vinfo, true); 1150 return NULL; 1151 } 1152 1153 if (!number_of_iterations) 1154 { 1155 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1156 fprintf (vect_dump, 1157 "not vectorized: number of iterations cannot be computed."); 1158 if (inner_loop_vinfo) 1159 destroy_loop_vec_info (inner_loop_vinfo, true); 1160 return NULL; 1161 } 1162 1163 if (chrec_contains_undetermined (number_of_iterations)) 1164 { 1165 if (vect_print_dump_info (REPORT_BAD_FORM_LOOPS)) 1166 fprintf (vect_dump, "Infinite number of iterations."); 1167 if (inner_loop_vinfo) 1168 destroy_loop_vec_info (inner_loop_vinfo, true); 1169 return NULL; 1170 } 1171 1172 if (!NITERS_KNOWN_P (number_of_iterations)) 1173 { 1174 if (vect_print_dump_info (REPORT_DETAILS)) 1175 { 1176 fprintf (vect_dump, "Symbolic number of iterations is "); 1177 print_generic_expr (vect_dump, number_of_iterations, TDF_DETAILS); 1178 } 1179 } 1180 else if (TREE_INT_CST_LOW (number_of_iterations) == 0) 1181 { 1182 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1183 fprintf (vect_dump, "not vectorized: number of iterations = 0."); 1184 if (inner_loop_vinfo) 1185 destroy_loop_vec_info (inner_loop_vinfo, false); 1186 return NULL; 1187 } 1188 1189 loop_vinfo = new_loop_vec_info (loop); 1190 LOOP_VINFO_NITERS (loop_vinfo) = number_of_iterations; 1191 LOOP_VINFO_NITERS_UNCHANGED (loop_vinfo) = number_of_iterations; 1192 1193 STMT_VINFO_TYPE (vinfo_for_stmt (loop_cond)) = loop_exit_ctrl_vec_info_type; 1194 1195 /* CHECKME: May want to keep it around it in the future. */ 1196 if (inner_loop_vinfo) 1197 destroy_loop_vec_info (inner_loop_vinfo, false); 1198 1199 gcc_assert (!loop->aux); 1200 loop->aux = loop_vinfo; 1201 return loop_vinfo; 1202 } 1203 1204 1205 /* Get cost by calling cost target builtin. */ 1206 1207 static inline int 1208 vect_get_cost (enum vect_cost_for_stmt type_of_cost) 1209 { 1210 tree dummy_type = NULL; 1211 int dummy = 0; 1212 1213 return targetm.vectorize.builtin_vectorization_cost (type_of_cost, 1214 dummy_type, dummy); 1215 } 1216 1217 1218 /* Function vect_analyze_loop_operations. 1219 1220 Scan the loop stmts and make sure they are all vectorizable. */ 1221 1222 static bool 1223 vect_analyze_loop_operations (loop_vec_info loop_vinfo, bool slp) 1224 { 1225 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 1226 basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo); 1227 int nbbs = loop->num_nodes; 1228 gimple_stmt_iterator si; 1229 unsigned int vectorization_factor = 0; 1230 int i; 1231 gimple phi; 1232 stmt_vec_info stmt_info; 1233 bool need_to_vectorize = false; 1234 int min_profitable_iters; 1235 int min_scalar_loop_bound; 1236 unsigned int th; 1237 bool only_slp_in_loop = true, ok; 1238 1239 if (vect_print_dump_info (REPORT_DETAILS)) 1240 fprintf (vect_dump, "=== vect_analyze_loop_operations ==="); 1241 1242 gcc_assert (LOOP_VINFO_VECT_FACTOR (loop_vinfo)); 1243 vectorization_factor = LOOP_VINFO_VECT_FACTOR (loop_vinfo); 1244 if (slp) 1245 { 1246 /* If all the stmts in the loop can be SLPed, we perform only SLP, and 1247 vectorization factor of the loop is the unrolling factor required by 1248 the SLP instances. If that unrolling factor is 1, we say, that we 1249 perform pure SLP on loop - cross iteration parallelism is not 1250 exploited. */ 1251 for (i = 0; i < nbbs; i++) 1252 { 1253 basic_block bb = bbs[i]; 1254 for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si)) 1255 { 1256 gimple stmt = gsi_stmt (si); 1257 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 1258 gcc_assert (stmt_info); 1259 if ((STMT_VINFO_RELEVANT_P (stmt_info) 1260 || VECTORIZABLE_CYCLE_DEF (STMT_VINFO_DEF_TYPE (stmt_info))) 1261 && !PURE_SLP_STMT (stmt_info)) 1262 /* STMT needs both SLP and loop-based vectorization. */ 1263 only_slp_in_loop = false; 1264 } 1265 } 1266 1267 if (only_slp_in_loop) 1268 vectorization_factor = LOOP_VINFO_SLP_UNROLLING_FACTOR (loop_vinfo); 1269 else 1270 vectorization_factor = least_common_multiple (vectorization_factor, 1271 LOOP_VINFO_SLP_UNROLLING_FACTOR (loop_vinfo)); 1272 1273 LOOP_VINFO_VECT_FACTOR (loop_vinfo) = vectorization_factor; 1274 if (vect_print_dump_info (REPORT_DETAILS)) 1275 fprintf (vect_dump, "Updating vectorization factor to %d ", 1276 vectorization_factor); 1277 } 1278 1279 for (i = 0; i < nbbs; i++) 1280 { 1281 basic_block bb = bbs[i]; 1282 1283 for (si = gsi_start_phis (bb); !gsi_end_p (si); gsi_next (&si)) 1284 { 1285 phi = gsi_stmt (si); 1286 ok = true; 1287 1288 stmt_info = vinfo_for_stmt (phi); 1289 if (vect_print_dump_info (REPORT_DETAILS)) 1290 { 1291 fprintf (vect_dump, "examining phi: "); 1292 print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); 1293 } 1294 1295 /* Inner-loop loop-closed exit phi in outer-loop vectorization 1296 (i.e., a phi in the tail of the outer-loop). */ 1297 if (! is_loop_header_bb_p (bb)) 1298 { 1299 /* FORNOW: we currently don't support the case that these phis 1300 are not used in the outerloop (unless it is double reduction, 1301 i.e., this phi is vect_reduction_def), cause this case 1302 requires to actually do something here. */ 1303 if ((!STMT_VINFO_RELEVANT_P (stmt_info) 1304 || STMT_VINFO_LIVE_P (stmt_info)) 1305 && STMT_VINFO_DEF_TYPE (stmt_info) 1306 != vect_double_reduction_def) 1307 { 1308 if (vect_print_dump_info (REPORT_DETAILS)) 1309 fprintf (vect_dump, 1310 "Unsupported loop-closed phi in outer-loop."); 1311 return false; 1312 } 1313 1314 /* If PHI is used in the outer loop, we check that its operand 1315 is defined in the inner loop. */ 1316 if (STMT_VINFO_RELEVANT_P (stmt_info)) 1317 { 1318 tree phi_op; 1319 gimple op_def_stmt; 1320 1321 if (gimple_phi_num_args (phi) != 1) 1322 return false; 1323 1324 phi_op = PHI_ARG_DEF (phi, 0); 1325 if (TREE_CODE (phi_op) != SSA_NAME) 1326 return false; 1327 1328 op_def_stmt = SSA_NAME_DEF_STMT (phi_op); 1329 if (!op_def_stmt 1330 || !flow_bb_inside_loop_p (loop, gimple_bb (op_def_stmt)) 1331 || !vinfo_for_stmt (op_def_stmt)) 1332 return false; 1333 1334 if (STMT_VINFO_RELEVANT (vinfo_for_stmt (op_def_stmt)) 1335 != vect_used_in_outer 1336 && STMT_VINFO_RELEVANT (vinfo_for_stmt (op_def_stmt)) 1337 != vect_used_in_outer_by_reduction) 1338 return false; 1339 } 1340 1341 continue; 1342 } 1343 1344 gcc_assert (stmt_info); 1345 1346 if (STMT_VINFO_LIVE_P (stmt_info)) 1347 { 1348 /* FORNOW: not yet supported. */ 1349 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1350 fprintf (vect_dump, "not vectorized: value used after loop."); 1351 return false; 1352 } 1353 1354 if (STMT_VINFO_RELEVANT (stmt_info) == vect_used_in_scope 1355 && STMT_VINFO_DEF_TYPE (stmt_info) != vect_induction_def) 1356 { 1357 /* A scalar-dependence cycle that we don't support. */ 1358 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1359 fprintf (vect_dump, "not vectorized: scalar dependence cycle."); 1360 return false; 1361 } 1362 1363 if (STMT_VINFO_RELEVANT_P (stmt_info)) 1364 { 1365 need_to_vectorize = true; 1366 if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def) 1367 ok = vectorizable_induction (phi, NULL, NULL); 1368 } 1369 1370 if (!ok) 1371 { 1372 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1373 { 1374 fprintf (vect_dump, 1375 "not vectorized: relevant phi not supported: "); 1376 print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); 1377 } 1378 return false; 1379 } 1380 } 1381 1382 for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si)) 1383 { 1384 gimple stmt = gsi_stmt (si); 1385 if (!vect_analyze_stmt (stmt, &need_to_vectorize, NULL)) 1386 return false; 1387 } 1388 } /* bbs */ 1389 1390 /* All operations in the loop are either irrelevant (deal with loop 1391 control, or dead), or only used outside the loop and can be moved 1392 out of the loop (e.g. invariants, inductions). The loop can be 1393 optimized away by scalar optimizations. We're better off not 1394 touching this loop. */ 1395 if (!need_to_vectorize) 1396 { 1397 if (vect_print_dump_info (REPORT_DETAILS)) 1398 fprintf (vect_dump, 1399 "All the computation can be taken out of the loop."); 1400 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1401 fprintf (vect_dump, 1402 "not vectorized: redundant loop. no profit to vectorize."); 1403 return false; 1404 } 1405 1406 if (LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) 1407 && vect_print_dump_info (REPORT_DETAILS)) 1408 fprintf (vect_dump, 1409 "vectorization_factor = %d, niters = " HOST_WIDE_INT_PRINT_DEC, 1410 vectorization_factor, LOOP_VINFO_INT_NITERS (loop_vinfo)); 1411 1412 if (LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) 1413 && (LOOP_VINFO_INT_NITERS (loop_vinfo) < vectorization_factor)) 1414 { 1415 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1416 fprintf (vect_dump, "not vectorized: iteration count too small."); 1417 if (vect_print_dump_info (REPORT_DETAILS)) 1418 fprintf (vect_dump,"not vectorized: iteration count smaller than " 1419 "vectorization factor."); 1420 return false; 1421 } 1422 1423 /* Analyze cost. Decide if worth while to vectorize. */ 1424 1425 /* Once VF is set, SLP costs should be updated since the number of created 1426 vector stmts depends on VF. */ 1427 vect_update_slp_costs_according_to_vf (loop_vinfo); 1428 1429 min_profitable_iters = vect_estimate_min_profitable_iters (loop_vinfo); 1430 LOOP_VINFO_COST_MODEL_MIN_ITERS (loop_vinfo) = min_profitable_iters; 1431 1432 if (min_profitable_iters < 0) 1433 { 1434 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1435 fprintf (vect_dump, "not vectorized: vectorization not profitable."); 1436 if (vect_print_dump_info (REPORT_DETAILS)) 1437 fprintf (vect_dump, "not vectorized: vector version will never be " 1438 "profitable."); 1439 return false; 1440 } 1441 1442 min_scalar_loop_bound = ((PARAM_VALUE (PARAM_MIN_VECT_LOOP_BOUND) 1443 * vectorization_factor) - 1); 1444 1445 /* Use the cost model only if it is more conservative than user specified 1446 threshold. */ 1447 1448 th = (unsigned) min_scalar_loop_bound; 1449 if (min_profitable_iters 1450 && (!min_scalar_loop_bound 1451 || min_profitable_iters > min_scalar_loop_bound)) 1452 th = (unsigned) min_profitable_iters; 1453 1454 if (LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) 1455 && LOOP_VINFO_INT_NITERS (loop_vinfo) <= th) 1456 { 1457 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1458 fprintf (vect_dump, "not vectorized: vectorization not " 1459 "profitable."); 1460 if (vect_print_dump_info (REPORT_DETAILS)) 1461 fprintf (vect_dump, "not vectorized: iteration count smaller than " 1462 "user specified loop bound parameter or minimum " 1463 "profitable iterations (whichever is more conservative)."); 1464 return false; 1465 } 1466 1467 if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) 1468 || LOOP_VINFO_INT_NITERS (loop_vinfo) % vectorization_factor != 0 1469 || LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo)) 1470 { 1471 if (vect_print_dump_info (REPORT_DETAILS)) 1472 fprintf (vect_dump, "epilog loop required."); 1473 if (!vect_can_advance_ivs_p (loop_vinfo)) 1474 { 1475 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1476 fprintf (vect_dump, 1477 "not vectorized: can't create epilog loop 1."); 1478 return false; 1479 } 1480 if (!slpeel_can_duplicate_loop_p (loop, single_exit (loop))) 1481 { 1482 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1483 fprintf (vect_dump, 1484 "not vectorized: can't create epilog loop 2."); 1485 return false; 1486 } 1487 } 1488 1489 return true; 1490 } 1491 1492 1493 /* Function vect_analyze_loop_2. 1494 1495 Apply a set of analyses on LOOP, and create a loop_vec_info struct 1496 for it. The different analyses will record information in the 1497 loop_vec_info struct. */ 1498 static bool 1499 vect_analyze_loop_2 (loop_vec_info loop_vinfo) 1500 { 1501 bool ok, slp = false; 1502 int max_vf = MAX_VECTORIZATION_FACTOR; 1503 int min_vf = 2; 1504 1505 /* Find all data references in the loop (which correspond to vdefs/vuses) 1506 and analyze their evolution in the loop. Also adjust the minimal 1507 vectorization factor according to the loads and stores. 1508 1509 FORNOW: Handle only simple, array references, which 1510 alignment can be forced, and aligned pointer-references. */ 1511 1512 ok = vect_analyze_data_refs (loop_vinfo, NULL, &min_vf); 1513 if (!ok) 1514 { 1515 if (vect_print_dump_info (REPORT_DETAILS)) 1516 fprintf (vect_dump, "bad data references."); 1517 return false; 1518 } 1519 1520 /* Classify all cross-iteration scalar data-flow cycles. 1521 Cross-iteration cycles caused by virtual phis are analyzed separately. */ 1522 1523 vect_analyze_scalar_cycles (loop_vinfo); 1524 1525 vect_pattern_recog (loop_vinfo); 1526 1527 /* Data-flow analysis to detect stmts that do not need to be vectorized. */ 1528 1529 ok = vect_mark_stmts_to_be_vectorized (loop_vinfo); 1530 if (!ok) 1531 { 1532 if (vect_print_dump_info (REPORT_DETAILS)) 1533 fprintf (vect_dump, "unexpected pattern."); 1534 return false; 1535 } 1536 1537 /* Analyze data dependences between the data-refs in the loop 1538 and adjust the maximum vectorization factor according to 1539 the dependences. 1540 FORNOW: fail at the first data dependence that we encounter. */ 1541 1542 ok = vect_analyze_data_ref_dependences (loop_vinfo, NULL, &max_vf); 1543 if (!ok 1544 || max_vf < min_vf) 1545 { 1546 if (vect_print_dump_info (REPORT_DETAILS)) 1547 fprintf (vect_dump, "bad data dependence."); 1548 return false; 1549 } 1550 1551 ok = vect_determine_vectorization_factor (loop_vinfo); 1552 if (!ok) 1553 { 1554 if (vect_print_dump_info (REPORT_DETAILS)) 1555 fprintf (vect_dump, "can't determine vectorization factor."); 1556 return false; 1557 } 1558 if (max_vf < LOOP_VINFO_VECT_FACTOR (loop_vinfo)) 1559 { 1560 if (vect_print_dump_info (REPORT_DETAILS)) 1561 fprintf (vect_dump, "bad data dependence."); 1562 return false; 1563 } 1564 1565 /* Analyze the alignment of the data-refs in the loop. 1566 Fail if a data reference is found that cannot be vectorized. */ 1567 1568 ok = vect_analyze_data_refs_alignment (loop_vinfo, NULL); 1569 if (!ok) 1570 { 1571 if (vect_print_dump_info (REPORT_DETAILS)) 1572 fprintf (vect_dump, "bad data alignment."); 1573 return false; 1574 } 1575 1576 /* Analyze the access patterns of the data-refs in the loop (consecutive, 1577 complex, etc.). FORNOW: Only handle consecutive access pattern. */ 1578 1579 ok = vect_analyze_data_ref_accesses (loop_vinfo, NULL); 1580 if (!ok) 1581 { 1582 if (vect_print_dump_info (REPORT_DETAILS)) 1583 fprintf (vect_dump, "bad data access."); 1584 return false; 1585 } 1586 1587 /* Prune the list of ddrs to be tested at run-time by versioning for alias. 1588 It is important to call pruning after vect_analyze_data_ref_accesses, 1589 since we use grouping information gathered by interleaving analysis. */ 1590 ok = vect_prune_runtime_alias_test_list (loop_vinfo); 1591 if (!ok) 1592 { 1593 if (vect_print_dump_info (REPORT_DETAILS)) 1594 fprintf (vect_dump, "too long list of versioning for alias " 1595 "run-time tests."); 1596 return false; 1597 } 1598 1599 /* This pass will decide on using loop versioning and/or loop peeling in 1600 order to enhance the alignment of data references in the loop. */ 1601 1602 ok = vect_enhance_data_refs_alignment (loop_vinfo); 1603 if (!ok) 1604 { 1605 if (vect_print_dump_info (REPORT_DETAILS)) 1606 fprintf (vect_dump, "bad data alignment."); 1607 return false; 1608 } 1609 1610 /* Check the SLP opportunities in the loop, analyze and build SLP trees. */ 1611 ok = vect_analyze_slp (loop_vinfo, NULL); 1612 if (ok) 1613 { 1614 /* Decide which possible SLP instances to SLP. */ 1615 slp = vect_make_slp_decision (loop_vinfo); 1616 1617 /* Find stmts that need to be both vectorized and SLPed. */ 1618 vect_detect_hybrid_slp (loop_vinfo); 1619 } 1620 else 1621 return false; 1622 1623 /* Scan all the operations in the loop and make sure they are 1624 vectorizable. */ 1625 1626 ok = vect_analyze_loop_operations (loop_vinfo, slp); 1627 if (!ok) 1628 { 1629 if (vect_print_dump_info (REPORT_DETAILS)) 1630 fprintf (vect_dump, "bad operation or unsupported loop bound."); 1631 return false; 1632 } 1633 1634 return true; 1635 } 1636 1637 /* Function vect_analyze_loop. 1638 1639 Apply a set of analyses on LOOP, and create a loop_vec_info struct 1640 for it. The different analyses will record information in the 1641 loop_vec_info struct. */ 1642 loop_vec_info 1643 vect_analyze_loop (struct loop *loop) 1644 { 1645 loop_vec_info loop_vinfo; 1646 unsigned int vector_sizes; 1647 1648 /* Autodetect first vector size we try. */ 1649 current_vector_size = 0; 1650 vector_sizes = targetm.vectorize.autovectorize_vector_sizes (); 1651 1652 if (vect_print_dump_info (REPORT_DETAILS)) 1653 fprintf (vect_dump, "===== analyze_loop_nest ====="); 1654 1655 if (loop_outer (loop) 1656 && loop_vec_info_for_loop (loop_outer (loop)) 1657 && LOOP_VINFO_VECTORIZABLE_P (loop_vec_info_for_loop (loop_outer (loop)))) 1658 { 1659 if (vect_print_dump_info (REPORT_DETAILS)) 1660 fprintf (vect_dump, "outer-loop already vectorized."); 1661 return NULL; 1662 } 1663 1664 while (1) 1665 { 1666 /* Check the CFG characteristics of the loop (nesting, entry/exit). */ 1667 loop_vinfo = vect_analyze_loop_form (loop); 1668 if (!loop_vinfo) 1669 { 1670 if (vect_print_dump_info (REPORT_DETAILS)) 1671 fprintf (vect_dump, "bad loop form."); 1672 return NULL; 1673 } 1674 1675 if (vect_analyze_loop_2 (loop_vinfo)) 1676 { 1677 LOOP_VINFO_VECTORIZABLE_P (loop_vinfo) = 1; 1678 1679 return loop_vinfo; 1680 } 1681 1682 destroy_loop_vec_info (loop_vinfo, true); 1683 1684 vector_sizes &= ~current_vector_size; 1685 if (vector_sizes == 0 1686 || current_vector_size == 0) 1687 return NULL; 1688 1689 /* Try the next biggest vector size. */ 1690 current_vector_size = 1 << floor_log2 (vector_sizes); 1691 if (vect_print_dump_info (REPORT_DETAILS)) 1692 fprintf (vect_dump, "***** Re-trying analysis with " 1693 "vector size %d\n", current_vector_size); 1694 } 1695 } 1696 1697 1698 /* Function reduction_code_for_scalar_code 1699 1700 Input: 1701 CODE - tree_code of a reduction operations. 1702 1703 Output: 1704 REDUC_CODE - the corresponding tree-code to be used to reduce the 1705 vector of partial results into a single scalar result (which 1706 will also reside in a vector) or ERROR_MARK if the operation is 1707 a supported reduction operation, but does not have such tree-code. 1708 1709 Return FALSE if CODE currently cannot be vectorized as reduction. */ 1710 1711 static bool 1712 reduction_code_for_scalar_code (enum tree_code code, 1713 enum tree_code *reduc_code) 1714 { 1715 switch (code) 1716 { 1717 case MAX_EXPR: 1718 *reduc_code = REDUC_MAX_EXPR; 1719 return true; 1720 1721 case MIN_EXPR: 1722 *reduc_code = REDUC_MIN_EXPR; 1723 return true; 1724 1725 case PLUS_EXPR: 1726 *reduc_code = REDUC_PLUS_EXPR; 1727 return true; 1728 1729 case MULT_EXPR: 1730 case MINUS_EXPR: 1731 case BIT_IOR_EXPR: 1732 case BIT_XOR_EXPR: 1733 case BIT_AND_EXPR: 1734 *reduc_code = ERROR_MARK; 1735 return true; 1736 1737 default: 1738 return false; 1739 } 1740 } 1741 1742 1743 /* Error reporting helper for vect_is_simple_reduction below. GIMPLE statement 1744 STMT is printed with a message MSG. */ 1745 1746 static void 1747 report_vect_op (gimple stmt, const char *msg) 1748 { 1749 fprintf (vect_dump, "%s", msg); 1750 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 1751 } 1752 1753 1754 /* Detect SLP reduction of the form: 1755 1756 #a1 = phi <a5, a0> 1757 a2 = operation (a1) 1758 a3 = operation (a2) 1759 a4 = operation (a3) 1760 a5 = operation (a4) 1761 1762 #a = phi <a5> 1763 1764 PHI is the reduction phi node (#a1 = phi <a5, a0> above) 1765 FIRST_STMT is the first reduction stmt in the chain 1766 (a2 = operation (a1)). 1767 1768 Return TRUE if a reduction chain was detected. */ 1769 1770 static bool 1771 vect_is_slp_reduction (loop_vec_info loop_info, gimple phi, gimple first_stmt) 1772 { 1773 struct loop *loop = (gimple_bb (phi))->loop_father; 1774 struct loop *vect_loop = LOOP_VINFO_LOOP (loop_info); 1775 enum tree_code code; 1776 gimple current_stmt = NULL, loop_use_stmt = NULL, first, next_stmt; 1777 stmt_vec_info use_stmt_info, current_stmt_info; 1778 tree lhs; 1779 imm_use_iterator imm_iter; 1780 use_operand_p use_p; 1781 int nloop_uses, size = 0, n_out_of_loop_uses; 1782 bool found = false; 1783 1784 if (loop != vect_loop) 1785 return false; 1786 1787 lhs = PHI_RESULT (phi); 1788 code = gimple_assign_rhs_code (first_stmt); 1789 while (1) 1790 { 1791 nloop_uses = 0; 1792 n_out_of_loop_uses = 0; 1793 FOR_EACH_IMM_USE_FAST (use_p, imm_iter, lhs) 1794 { 1795 gimple use_stmt = USE_STMT (use_p); 1796 if (is_gimple_debug (use_stmt)) 1797 continue; 1798 1799 use_stmt = USE_STMT (use_p); 1800 1801 /* Check if we got back to the reduction phi. */ 1802 if (use_stmt == phi) 1803 { 1804 loop_use_stmt = use_stmt; 1805 found = true; 1806 break; 1807 } 1808 1809 if (flow_bb_inside_loop_p (loop, gimple_bb (use_stmt))) 1810 { 1811 if (vinfo_for_stmt (use_stmt) 1812 && !STMT_VINFO_IN_PATTERN_P (vinfo_for_stmt (use_stmt))) 1813 { 1814 loop_use_stmt = use_stmt; 1815 nloop_uses++; 1816 } 1817 } 1818 else 1819 n_out_of_loop_uses++; 1820 1821 /* There are can be either a single use in the loop or two uses in 1822 phi nodes. */ 1823 if (nloop_uses > 1 || (n_out_of_loop_uses && nloop_uses)) 1824 return false; 1825 } 1826 1827 if (found) 1828 break; 1829 1830 /* We reached a statement with no loop uses. */ 1831 if (nloop_uses == 0) 1832 return false; 1833 1834 /* This is a loop exit phi, and we haven't reached the reduction phi. */ 1835 if (gimple_code (loop_use_stmt) == GIMPLE_PHI) 1836 return false; 1837 1838 if (!is_gimple_assign (loop_use_stmt) 1839 || code != gimple_assign_rhs_code (loop_use_stmt) 1840 || !flow_bb_inside_loop_p (loop, gimple_bb (loop_use_stmt))) 1841 return false; 1842 1843 /* Insert USE_STMT into reduction chain. */ 1844 use_stmt_info = vinfo_for_stmt (loop_use_stmt); 1845 if (current_stmt) 1846 { 1847 current_stmt_info = vinfo_for_stmt (current_stmt); 1848 GROUP_NEXT_ELEMENT (current_stmt_info) = loop_use_stmt; 1849 GROUP_FIRST_ELEMENT (use_stmt_info) 1850 = GROUP_FIRST_ELEMENT (current_stmt_info); 1851 } 1852 else 1853 GROUP_FIRST_ELEMENT (use_stmt_info) = loop_use_stmt; 1854 1855 lhs = gimple_assign_lhs (loop_use_stmt); 1856 current_stmt = loop_use_stmt; 1857 size++; 1858 } 1859 1860 if (!found || loop_use_stmt != phi || size < 2) 1861 return false; 1862 1863 /* Swap the operands, if needed, to make the reduction operand be the second 1864 operand. */ 1865 lhs = PHI_RESULT (phi); 1866 next_stmt = GROUP_FIRST_ELEMENT (vinfo_for_stmt (current_stmt)); 1867 while (next_stmt) 1868 { 1869 if (gimple_assign_rhs2 (next_stmt) == lhs) 1870 { 1871 tree op = gimple_assign_rhs1 (next_stmt); 1872 gimple def_stmt = NULL; 1873 1874 if (TREE_CODE (op) == SSA_NAME) 1875 def_stmt = SSA_NAME_DEF_STMT (op); 1876 1877 /* Check that the other def is either defined in the loop 1878 ("vect_internal_def"), or it's an induction (defined by a 1879 loop-header phi-node). */ 1880 if (def_stmt 1881 && gimple_bb (def_stmt) 1882 && flow_bb_inside_loop_p (loop, gimple_bb (def_stmt)) 1883 && (is_gimple_assign (def_stmt) 1884 || is_gimple_call (def_stmt) 1885 || STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def_stmt)) 1886 == vect_induction_def 1887 || (gimple_code (def_stmt) == GIMPLE_PHI 1888 && STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def_stmt)) 1889 == vect_internal_def 1890 && !is_loop_header_bb_p (gimple_bb (def_stmt))))) 1891 { 1892 lhs = gimple_assign_lhs (next_stmt); 1893 next_stmt = GROUP_NEXT_ELEMENT (vinfo_for_stmt (next_stmt)); 1894 continue; 1895 } 1896 1897 return false; 1898 } 1899 else 1900 { 1901 tree op = gimple_assign_rhs2 (next_stmt); 1902 gimple def_stmt = NULL; 1903 1904 if (TREE_CODE (op) == SSA_NAME) 1905 def_stmt = SSA_NAME_DEF_STMT (op); 1906 1907 /* Check that the other def is either defined in the loop 1908 ("vect_internal_def"), or it's an induction (defined by a 1909 loop-header phi-node). */ 1910 if (def_stmt 1911 && gimple_bb (def_stmt) 1912 && flow_bb_inside_loop_p (loop, gimple_bb (def_stmt)) 1913 && (is_gimple_assign (def_stmt) 1914 || is_gimple_call (def_stmt) 1915 || STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def_stmt)) 1916 == vect_induction_def 1917 || (gimple_code (def_stmt) == GIMPLE_PHI 1918 && STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def_stmt)) 1919 == vect_internal_def 1920 && !is_loop_header_bb_p (gimple_bb (def_stmt))))) 1921 { 1922 if (vect_print_dump_info (REPORT_DETAILS)) 1923 { 1924 fprintf (vect_dump, "swapping oprnds: "); 1925 print_gimple_stmt (vect_dump, next_stmt, 0, TDF_SLIM); 1926 } 1927 1928 swap_tree_operands (next_stmt, 1929 gimple_assign_rhs1_ptr (next_stmt), 1930 gimple_assign_rhs2_ptr (next_stmt)); 1931 mark_symbols_for_renaming (next_stmt); 1932 } 1933 else 1934 return false; 1935 } 1936 1937 lhs = gimple_assign_lhs (next_stmt); 1938 next_stmt = GROUP_NEXT_ELEMENT (vinfo_for_stmt (next_stmt)); 1939 } 1940 1941 /* Save the chain for further analysis in SLP detection. */ 1942 first = GROUP_FIRST_ELEMENT (vinfo_for_stmt (current_stmt)); 1943 VEC_safe_push (gimple, heap, LOOP_VINFO_REDUCTION_CHAINS (loop_info), first); 1944 GROUP_SIZE (vinfo_for_stmt (first)) = size; 1945 1946 return true; 1947 } 1948 1949 1950 /* Function vect_is_simple_reduction_1 1951 1952 (1) Detect a cross-iteration def-use cycle that represents a simple 1953 reduction computation. We look for the following pattern: 1954 1955 loop_header: 1956 a1 = phi < a0, a2 > 1957 a3 = ... 1958 a2 = operation (a3, a1) 1959 1960 such that: 1961 1. operation is commutative and associative and it is safe to 1962 change the order of the computation (if CHECK_REDUCTION is true) 1963 2. no uses for a2 in the loop (a2 is used out of the loop) 1964 3. no uses of a1 in the loop besides the reduction operation 1965 4. no uses of a1 outside the loop. 1966 1967 Conditions 1,4 are tested here. 1968 Conditions 2,3 are tested in vect_mark_stmts_to_be_vectorized. 1969 1970 (2) Detect a cross-iteration def-use cycle in nested loops, i.e., 1971 nested cycles, if CHECK_REDUCTION is false. 1972 1973 (3) Detect cycles of phi nodes in outer-loop vectorization, i.e., double 1974 reductions: 1975 1976 a1 = phi < a0, a2 > 1977 inner loop (def of a3) 1978 a2 = phi < a3 > 1979 1980 If MODIFY is true it tries also to rework the code in-place to enable 1981 detection of more reduction patterns. For the time being we rewrite 1982 "res -= RHS" into "rhs += -RHS" when it seems worthwhile. 1983 */ 1984 1985 static gimple 1986 vect_is_simple_reduction_1 (loop_vec_info loop_info, gimple phi, 1987 bool check_reduction, bool *double_reduc, 1988 bool modify) 1989 { 1990 struct loop *loop = (gimple_bb (phi))->loop_father; 1991 struct loop *vect_loop = LOOP_VINFO_LOOP (loop_info); 1992 edge latch_e = loop_latch_edge (loop); 1993 tree loop_arg = PHI_ARG_DEF_FROM_EDGE (phi, latch_e); 1994 gimple def_stmt, def1 = NULL, def2 = NULL; 1995 enum tree_code orig_code, code; 1996 tree op1, op2, op3 = NULL_TREE, op4 = NULL_TREE; 1997 tree type; 1998 int nloop_uses; 1999 tree name; 2000 imm_use_iterator imm_iter; 2001 use_operand_p use_p; 2002 bool phi_def; 2003 2004 *double_reduc = false; 2005 2006 /* If CHECK_REDUCTION is true, we assume inner-most loop vectorization, 2007 otherwise, we assume outer loop vectorization. */ 2008 gcc_assert ((check_reduction && loop == vect_loop) 2009 || (!check_reduction && flow_loop_nested_p (vect_loop, loop))); 2010 2011 name = PHI_RESULT (phi); 2012 nloop_uses = 0; 2013 FOR_EACH_IMM_USE_FAST (use_p, imm_iter, name) 2014 { 2015 gimple use_stmt = USE_STMT (use_p); 2016 if (is_gimple_debug (use_stmt)) 2017 continue; 2018 2019 if (!flow_bb_inside_loop_p (loop, gimple_bb (use_stmt))) 2020 { 2021 if (vect_print_dump_info (REPORT_DETAILS)) 2022 fprintf (vect_dump, "intermediate value used outside loop."); 2023 2024 return NULL; 2025 } 2026 2027 if (vinfo_for_stmt (use_stmt) 2028 && !is_pattern_stmt_p (vinfo_for_stmt (use_stmt))) 2029 nloop_uses++; 2030 if (nloop_uses > 1) 2031 { 2032 if (vect_print_dump_info (REPORT_DETAILS)) 2033 fprintf (vect_dump, "reduction used in loop."); 2034 return NULL; 2035 } 2036 } 2037 2038 if (TREE_CODE (loop_arg) != SSA_NAME) 2039 { 2040 if (vect_print_dump_info (REPORT_DETAILS)) 2041 { 2042 fprintf (vect_dump, "reduction: not ssa_name: "); 2043 print_generic_expr (vect_dump, loop_arg, TDF_SLIM); 2044 } 2045 return NULL; 2046 } 2047 2048 def_stmt = SSA_NAME_DEF_STMT (loop_arg); 2049 if (!def_stmt) 2050 { 2051 if (vect_print_dump_info (REPORT_DETAILS)) 2052 fprintf (vect_dump, "reduction: no def_stmt."); 2053 return NULL; 2054 } 2055 2056 if (!is_gimple_assign (def_stmt) && gimple_code (def_stmt) != GIMPLE_PHI) 2057 { 2058 if (vect_print_dump_info (REPORT_DETAILS)) 2059 print_gimple_stmt (vect_dump, def_stmt, 0, TDF_SLIM); 2060 return NULL; 2061 } 2062 2063 if (is_gimple_assign (def_stmt)) 2064 { 2065 name = gimple_assign_lhs (def_stmt); 2066 phi_def = false; 2067 } 2068 else 2069 { 2070 name = PHI_RESULT (def_stmt); 2071 phi_def = true; 2072 } 2073 2074 nloop_uses = 0; 2075 FOR_EACH_IMM_USE_FAST (use_p, imm_iter, name) 2076 { 2077 gimple use_stmt = USE_STMT (use_p); 2078 if (is_gimple_debug (use_stmt)) 2079 continue; 2080 if (flow_bb_inside_loop_p (loop, gimple_bb (use_stmt)) 2081 && vinfo_for_stmt (use_stmt) 2082 && !is_pattern_stmt_p (vinfo_for_stmt (use_stmt))) 2083 nloop_uses++; 2084 if (nloop_uses > 1) 2085 { 2086 if (vect_print_dump_info (REPORT_DETAILS)) 2087 fprintf (vect_dump, "reduction used in loop."); 2088 return NULL; 2089 } 2090 } 2091 2092 /* If DEF_STMT is a phi node itself, we expect it to have a single argument 2093 defined in the inner loop. */ 2094 if (phi_def) 2095 { 2096 op1 = PHI_ARG_DEF (def_stmt, 0); 2097 2098 if (gimple_phi_num_args (def_stmt) != 1 2099 || TREE_CODE (op1) != SSA_NAME) 2100 { 2101 if (vect_print_dump_info (REPORT_DETAILS)) 2102 fprintf (vect_dump, "unsupported phi node definition."); 2103 2104 return NULL; 2105 } 2106 2107 def1 = SSA_NAME_DEF_STMT (op1); 2108 if (flow_bb_inside_loop_p (loop, gimple_bb (def_stmt)) 2109 && loop->inner 2110 && flow_bb_inside_loop_p (loop->inner, gimple_bb (def1)) 2111 && is_gimple_assign (def1)) 2112 { 2113 if (vect_print_dump_info (REPORT_DETAILS)) 2114 report_vect_op (def_stmt, "detected double reduction: "); 2115 2116 *double_reduc = true; 2117 return def_stmt; 2118 } 2119 2120 return NULL; 2121 } 2122 2123 code = orig_code = gimple_assign_rhs_code (def_stmt); 2124 2125 /* We can handle "res -= x[i]", which is non-associative by 2126 simply rewriting this into "res += -x[i]". Avoid changing 2127 gimple instruction for the first simple tests and only do this 2128 if we're allowed to change code at all. */ 2129 if (code == MINUS_EXPR 2130 && modify 2131 && (op1 = gimple_assign_rhs1 (def_stmt)) 2132 && TREE_CODE (op1) == SSA_NAME 2133 && SSA_NAME_DEF_STMT (op1) == phi) 2134 code = PLUS_EXPR; 2135 2136 if (check_reduction 2137 && (!commutative_tree_code (code) || !associative_tree_code (code))) 2138 { 2139 if (vect_print_dump_info (REPORT_DETAILS)) 2140 report_vect_op (def_stmt, "reduction: not commutative/associative: "); 2141 return NULL; 2142 } 2143 2144 if (get_gimple_rhs_class (code) != GIMPLE_BINARY_RHS) 2145 { 2146 if (code != COND_EXPR) 2147 { 2148 if (vect_print_dump_info (REPORT_DETAILS)) 2149 report_vect_op (def_stmt, "reduction: not binary operation: "); 2150 2151 return NULL; 2152 } 2153 2154 op3 = gimple_assign_rhs1 (def_stmt); 2155 if (COMPARISON_CLASS_P (op3)) 2156 { 2157 op4 = TREE_OPERAND (op3, 1); 2158 op3 = TREE_OPERAND (op3, 0); 2159 } 2160 2161 op1 = gimple_assign_rhs2 (def_stmt); 2162 op2 = gimple_assign_rhs3 (def_stmt); 2163 2164 if (TREE_CODE (op1) != SSA_NAME && TREE_CODE (op2) != SSA_NAME) 2165 { 2166 if (vect_print_dump_info (REPORT_DETAILS)) 2167 report_vect_op (def_stmt, "reduction: uses not ssa_names: "); 2168 2169 return NULL; 2170 } 2171 } 2172 else 2173 { 2174 op1 = gimple_assign_rhs1 (def_stmt); 2175 op2 = gimple_assign_rhs2 (def_stmt); 2176 2177 if (TREE_CODE (op1) != SSA_NAME && TREE_CODE (op2) != SSA_NAME) 2178 { 2179 if (vect_print_dump_info (REPORT_DETAILS)) 2180 report_vect_op (def_stmt, "reduction: uses not ssa_names: "); 2181 2182 return NULL; 2183 } 2184 } 2185 2186 type = TREE_TYPE (gimple_assign_lhs (def_stmt)); 2187 if ((TREE_CODE (op1) == SSA_NAME 2188 && !types_compatible_p (type,TREE_TYPE (op1))) 2189 || (TREE_CODE (op2) == SSA_NAME 2190 && !types_compatible_p (type, TREE_TYPE (op2))) 2191 || (op3 && TREE_CODE (op3) == SSA_NAME 2192 && !types_compatible_p (type, TREE_TYPE (op3))) 2193 || (op4 && TREE_CODE (op4) == SSA_NAME 2194 && !types_compatible_p (type, TREE_TYPE (op4)))) 2195 { 2196 if (vect_print_dump_info (REPORT_DETAILS)) 2197 { 2198 fprintf (vect_dump, "reduction: multiple types: operation type: "); 2199 print_generic_expr (vect_dump, type, TDF_SLIM); 2200 fprintf (vect_dump, ", operands types: "); 2201 print_generic_expr (vect_dump, TREE_TYPE (op1), TDF_SLIM); 2202 fprintf (vect_dump, ","); 2203 print_generic_expr (vect_dump, TREE_TYPE (op2), TDF_SLIM); 2204 if (op3) 2205 { 2206 fprintf (vect_dump, ","); 2207 print_generic_expr (vect_dump, TREE_TYPE (op3), TDF_SLIM); 2208 } 2209 2210 if (op4) 2211 { 2212 fprintf (vect_dump, ","); 2213 print_generic_expr (vect_dump, TREE_TYPE (op4), TDF_SLIM); 2214 } 2215 } 2216 2217 return NULL; 2218 } 2219 2220 /* Check that it's ok to change the order of the computation. 2221 Generally, when vectorizing a reduction we change the order of the 2222 computation. This may change the behavior of the program in some 2223 cases, so we need to check that this is ok. One exception is when 2224 vectorizing an outer-loop: the inner-loop is executed sequentially, 2225 and therefore vectorizing reductions in the inner-loop during 2226 outer-loop vectorization is safe. */ 2227 2228 /* CHECKME: check for !flag_finite_math_only too? */ 2229 if (SCALAR_FLOAT_TYPE_P (type) && !flag_associative_math 2230 && check_reduction) 2231 { 2232 /* Changing the order of operations changes the semantics. */ 2233 if (vect_print_dump_info (REPORT_DETAILS)) 2234 report_vect_op (def_stmt, "reduction: unsafe fp math optimization: "); 2235 return NULL; 2236 } 2237 else if (INTEGRAL_TYPE_P (type) && TYPE_OVERFLOW_TRAPS (type) 2238 && check_reduction) 2239 { 2240 /* Changing the order of operations changes the semantics. */ 2241 if (vect_print_dump_info (REPORT_DETAILS)) 2242 report_vect_op (def_stmt, "reduction: unsafe int math optimization: "); 2243 return NULL; 2244 } 2245 else if (SAT_FIXED_POINT_TYPE_P (type) && check_reduction) 2246 { 2247 /* Changing the order of operations changes the semantics. */ 2248 if (vect_print_dump_info (REPORT_DETAILS)) 2249 report_vect_op (def_stmt, 2250 "reduction: unsafe fixed-point math optimization: "); 2251 return NULL; 2252 } 2253 2254 /* If we detected "res -= x[i]" earlier, rewrite it into 2255 "res += -x[i]" now. If this turns out to be useless reassoc 2256 will clean it up again. */ 2257 if (orig_code == MINUS_EXPR) 2258 { 2259 tree rhs = gimple_assign_rhs2 (def_stmt); 2260 tree negrhs = make_ssa_name (SSA_NAME_VAR (rhs), NULL); 2261 gimple negate_stmt = gimple_build_assign_with_ops (NEGATE_EXPR, negrhs, 2262 rhs, NULL); 2263 gimple_stmt_iterator gsi = gsi_for_stmt (def_stmt); 2264 set_vinfo_for_stmt (negate_stmt, new_stmt_vec_info (negate_stmt, 2265 loop_info, NULL)); 2266 gsi_insert_before (&gsi, negate_stmt, GSI_NEW_STMT); 2267 gimple_assign_set_rhs2 (def_stmt, negrhs); 2268 gimple_assign_set_rhs_code (def_stmt, PLUS_EXPR); 2269 update_stmt (def_stmt); 2270 } 2271 2272 /* Reduction is safe. We're dealing with one of the following: 2273 1) integer arithmetic and no trapv 2274 2) floating point arithmetic, and special flags permit this optimization 2275 3) nested cycle (i.e., outer loop vectorization). */ 2276 if (TREE_CODE (op1) == SSA_NAME) 2277 def1 = SSA_NAME_DEF_STMT (op1); 2278 2279 if (TREE_CODE (op2) == SSA_NAME) 2280 def2 = SSA_NAME_DEF_STMT (op2); 2281 2282 if (code != COND_EXPR 2283 && ((!def1 || gimple_nop_p (def1)) && (!def2 || gimple_nop_p (def2)))) 2284 { 2285 if (vect_print_dump_info (REPORT_DETAILS)) 2286 report_vect_op (def_stmt, "reduction: no defs for operands: "); 2287 return NULL; 2288 } 2289 2290 /* Check that one def is the reduction def, defined by PHI, 2291 the other def is either defined in the loop ("vect_internal_def"), 2292 or it's an induction (defined by a loop-header phi-node). */ 2293 2294 if (def2 && def2 == phi 2295 && (code == COND_EXPR 2296 || !def1 || gimple_nop_p (def1) 2297 || (def1 && flow_bb_inside_loop_p (loop, gimple_bb (def1)) 2298 && (is_gimple_assign (def1) 2299 || is_gimple_call (def1) 2300 || STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def1)) 2301 == vect_induction_def 2302 || (gimple_code (def1) == GIMPLE_PHI 2303 && STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def1)) 2304 == vect_internal_def 2305 && !is_loop_header_bb_p (gimple_bb (def1))))))) 2306 { 2307 if (vect_print_dump_info (REPORT_DETAILS)) 2308 report_vect_op (def_stmt, "detected reduction: "); 2309 return def_stmt; 2310 } 2311 2312 if (def1 && def1 == phi 2313 && (code == COND_EXPR 2314 || !def2 || gimple_nop_p (def2) 2315 || (def2 && flow_bb_inside_loop_p (loop, gimple_bb (def2)) 2316 && (is_gimple_assign (def2) 2317 || is_gimple_call (def2) 2318 || STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def2)) 2319 == vect_induction_def 2320 || (gimple_code (def2) == GIMPLE_PHI 2321 && STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def2)) 2322 == vect_internal_def 2323 && !is_loop_header_bb_p (gimple_bb (def2))))))) 2324 { 2325 if (check_reduction) 2326 { 2327 /* Swap operands (just for simplicity - so that the rest of the code 2328 can assume that the reduction variable is always the last (second) 2329 argument). */ 2330 if (vect_print_dump_info (REPORT_DETAILS)) 2331 report_vect_op (def_stmt, 2332 "detected reduction: need to swap operands: "); 2333 2334 swap_tree_operands (def_stmt, gimple_assign_rhs1_ptr (def_stmt), 2335 gimple_assign_rhs2_ptr (def_stmt)); 2336 } 2337 else 2338 { 2339 if (vect_print_dump_info (REPORT_DETAILS)) 2340 report_vect_op (def_stmt, "detected reduction: "); 2341 } 2342 2343 return def_stmt; 2344 } 2345 2346 /* Try to find SLP reduction chain. */ 2347 if (check_reduction && vect_is_slp_reduction (loop_info, phi, def_stmt)) 2348 { 2349 if (vect_print_dump_info (REPORT_DETAILS)) 2350 report_vect_op (def_stmt, "reduction: detected reduction chain: "); 2351 2352 return def_stmt; 2353 } 2354 2355 if (vect_print_dump_info (REPORT_DETAILS)) 2356 report_vect_op (def_stmt, "reduction: unknown pattern: "); 2357 2358 return NULL; 2359 } 2360 2361 /* Wrapper around vect_is_simple_reduction_1, that won't modify code 2362 in-place. Arguments as there. */ 2363 2364 static gimple 2365 vect_is_simple_reduction (loop_vec_info loop_info, gimple phi, 2366 bool check_reduction, bool *double_reduc) 2367 { 2368 return vect_is_simple_reduction_1 (loop_info, phi, check_reduction, 2369 double_reduc, false); 2370 } 2371 2372 /* Wrapper around vect_is_simple_reduction_1, which will modify code 2373 in-place if it enables detection of more reductions. Arguments 2374 as there. */ 2375 2376 gimple 2377 vect_force_simple_reduction (loop_vec_info loop_info, gimple phi, 2378 bool check_reduction, bool *double_reduc) 2379 { 2380 return vect_is_simple_reduction_1 (loop_info, phi, check_reduction, 2381 double_reduc, true); 2382 } 2383 2384 /* Calculate the cost of one scalar iteration of the loop. */ 2385 int 2386 vect_get_single_scalar_iteraion_cost (loop_vec_info loop_vinfo) 2387 { 2388 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 2389 basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo); 2390 int nbbs = loop->num_nodes, factor, scalar_single_iter_cost = 0; 2391 int innerloop_iters, i, stmt_cost; 2392 2393 /* Count statements in scalar loop. Using this as scalar cost for a single 2394 iteration for now. 2395 2396 TODO: Add outer loop support. 2397 2398 TODO: Consider assigning different costs to different scalar 2399 statements. */ 2400 2401 /* FORNOW. */ 2402 innerloop_iters = 1; 2403 if (loop->inner) 2404 innerloop_iters = 50; /* FIXME */ 2405 2406 for (i = 0; i < nbbs; i++) 2407 { 2408 gimple_stmt_iterator si; 2409 basic_block bb = bbs[i]; 2410 2411 if (bb->loop_father == loop->inner) 2412 factor = innerloop_iters; 2413 else 2414 factor = 1; 2415 2416 for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si)) 2417 { 2418 gimple stmt = gsi_stmt (si); 2419 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 2420 2421 if (!is_gimple_assign (stmt) && !is_gimple_call (stmt)) 2422 continue; 2423 2424 /* Skip stmts that are not vectorized inside the loop. */ 2425 if (stmt_info 2426 && !STMT_VINFO_RELEVANT_P (stmt_info) 2427 && (!STMT_VINFO_LIVE_P (stmt_info) 2428 || !VECTORIZABLE_CYCLE_DEF (STMT_VINFO_DEF_TYPE (stmt_info))) 2429 && !STMT_VINFO_IN_PATTERN_P (stmt_info)) 2430 continue; 2431 2432 if (STMT_VINFO_DATA_REF (vinfo_for_stmt (stmt))) 2433 { 2434 if (DR_IS_READ (STMT_VINFO_DATA_REF (vinfo_for_stmt (stmt)))) 2435 stmt_cost = vect_get_cost (scalar_load); 2436 else 2437 stmt_cost = vect_get_cost (scalar_store); 2438 } 2439 else 2440 stmt_cost = vect_get_cost (scalar_stmt); 2441 2442 scalar_single_iter_cost += stmt_cost * factor; 2443 } 2444 } 2445 return scalar_single_iter_cost; 2446 } 2447 2448 /* Calculate cost of peeling the loop PEEL_ITERS_PROLOGUE times. */ 2449 int 2450 vect_get_known_peeling_cost (loop_vec_info loop_vinfo, int peel_iters_prologue, 2451 int *peel_iters_epilogue, 2452 int scalar_single_iter_cost) 2453 { 2454 int peel_guard_costs = 0; 2455 int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); 2456 2457 if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo)) 2458 { 2459 *peel_iters_epilogue = vf/2; 2460 if (vect_print_dump_info (REPORT_COST)) 2461 fprintf (vect_dump, "cost model: " 2462 "epilogue peel iters set to vf/2 because " 2463 "loop iterations are unknown ."); 2464 2465 /* If peeled iterations are known but number of scalar loop 2466 iterations are unknown, count a taken branch per peeled loop. */ 2467 peel_guard_costs = 2 * vect_get_cost (cond_branch_taken); 2468 } 2469 else 2470 { 2471 int niters = LOOP_VINFO_INT_NITERS (loop_vinfo); 2472 peel_iters_prologue = niters < peel_iters_prologue ? 2473 niters : peel_iters_prologue; 2474 *peel_iters_epilogue = (niters - peel_iters_prologue) % vf; 2475 /* If we need to peel for gaps, but no peeling is required, we have to 2476 peel VF iterations. */ 2477 if (LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo) && !*peel_iters_epilogue) 2478 *peel_iters_epilogue = vf; 2479 } 2480 2481 return (peel_iters_prologue * scalar_single_iter_cost) 2482 + (*peel_iters_epilogue * scalar_single_iter_cost) 2483 + peel_guard_costs; 2484 } 2485 2486 /* Function vect_estimate_min_profitable_iters 2487 2488 Return the number of iterations required for the vector version of the 2489 loop to be profitable relative to the cost of the scalar version of the 2490 loop. 2491 2492 TODO: Take profile info into account before making vectorization 2493 decisions, if available. */ 2494 2495 int 2496 vect_estimate_min_profitable_iters (loop_vec_info loop_vinfo) 2497 { 2498 int i; 2499 int min_profitable_iters; 2500 int peel_iters_prologue; 2501 int peel_iters_epilogue; 2502 int vec_inside_cost = 0; 2503 int vec_outside_cost = 0; 2504 int scalar_single_iter_cost = 0; 2505 int scalar_outside_cost = 0; 2506 int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); 2507 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 2508 basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo); 2509 int nbbs = loop->num_nodes; 2510 int npeel = LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo); 2511 int peel_guard_costs = 0; 2512 int innerloop_iters = 0, factor; 2513 VEC (slp_instance, heap) *slp_instances; 2514 slp_instance instance; 2515 2516 /* Cost model disabled. */ 2517 if (!flag_vect_cost_model) 2518 { 2519 if (vect_print_dump_info (REPORT_COST)) 2520 fprintf (vect_dump, "cost model disabled."); 2521 return 0; 2522 } 2523 2524 /* Requires loop versioning tests to handle misalignment. */ 2525 if (LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT (loop_vinfo)) 2526 { 2527 /* FIXME: Make cost depend on complexity of individual check. */ 2528 vec_outside_cost += 2529 VEC_length (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)); 2530 if (vect_print_dump_info (REPORT_COST)) 2531 fprintf (vect_dump, "cost model: Adding cost of checks for loop " 2532 "versioning to treat misalignment.\n"); 2533 } 2534 2535 /* Requires loop versioning with alias checks. */ 2536 if (LOOP_REQUIRES_VERSIONING_FOR_ALIAS (loop_vinfo)) 2537 { 2538 /* FIXME: Make cost depend on complexity of individual check. */ 2539 vec_outside_cost += 2540 VEC_length (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo)); 2541 if (vect_print_dump_info (REPORT_COST)) 2542 fprintf (vect_dump, "cost model: Adding cost of checks for loop " 2543 "versioning aliasing.\n"); 2544 } 2545 2546 if (LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT (loop_vinfo) 2547 || LOOP_REQUIRES_VERSIONING_FOR_ALIAS (loop_vinfo)) 2548 vec_outside_cost += vect_get_cost (cond_branch_taken); 2549 2550 /* Count statements in scalar loop. Using this as scalar cost for a single 2551 iteration for now. 2552 2553 TODO: Add outer loop support. 2554 2555 TODO: Consider assigning different costs to different scalar 2556 statements. */ 2557 2558 /* FORNOW. */ 2559 if (loop->inner) 2560 innerloop_iters = 50; /* FIXME */ 2561 2562 for (i = 0; i < nbbs; i++) 2563 { 2564 gimple_stmt_iterator si; 2565 basic_block bb = bbs[i]; 2566 2567 if (bb->loop_father == loop->inner) 2568 factor = innerloop_iters; 2569 else 2570 factor = 1; 2571 2572 for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si)) 2573 { 2574 gimple stmt = gsi_stmt (si); 2575 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 2576 2577 if (STMT_VINFO_IN_PATTERN_P (stmt_info)) 2578 { 2579 stmt = STMT_VINFO_RELATED_STMT (stmt_info); 2580 stmt_info = vinfo_for_stmt (stmt); 2581 } 2582 2583 /* Skip stmts that are not vectorized inside the loop. */ 2584 if (!STMT_VINFO_RELEVANT_P (stmt_info) 2585 && (!STMT_VINFO_LIVE_P (stmt_info) 2586 || !VECTORIZABLE_CYCLE_DEF (STMT_VINFO_DEF_TYPE (stmt_info)))) 2587 continue; 2588 2589 vec_inside_cost += STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) * factor; 2590 /* FIXME: for stmts in the inner-loop in outer-loop vectorization, 2591 some of the "outside" costs are generated inside the outer-loop. */ 2592 vec_outside_cost += STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info); 2593 if (is_pattern_stmt_p (stmt_info) 2594 && STMT_VINFO_PATTERN_DEF_SEQ (stmt_info)) 2595 { 2596 gimple_stmt_iterator gsi; 2597 2598 for (gsi = gsi_start (STMT_VINFO_PATTERN_DEF_SEQ (stmt_info)); 2599 !gsi_end_p (gsi); gsi_next (&gsi)) 2600 { 2601 gimple pattern_def_stmt = gsi_stmt (gsi); 2602 stmt_vec_info pattern_def_stmt_info 2603 = vinfo_for_stmt (pattern_def_stmt); 2604 if (STMT_VINFO_RELEVANT_P (pattern_def_stmt_info) 2605 || STMT_VINFO_LIVE_P (pattern_def_stmt_info)) 2606 { 2607 vec_inside_cost 2608 += STMT_VINFO_INSIDE_OF_LOOP_COST 2609 (pattern_def_stmt_info) * factor; 2610 vec_outside_cost 2611 += STMT_VINFO_OUTSIDE_OF_LOOP_COST 2612 (pattern_def_stmt_info); 2613 } 2614 } 2615 } 2616 } 2617 } 2618 2619 scalar_single_iter_cost = vect_get_single_scalar_iteraion_cost (loop_vinfo); 2620 2621 /* Add additional cost for the peeled instructions in prologue and epilogue 2622 loop. 2623 2624 FORNOW: If we don't know the value of peel_iters for prologue or epilogue 2625 at compile-time - we assume it's vf/2 (the worst would be vf-1). 2626 2627 TODO: Build an expression that represents peel_iters for prologue and 2628 epilogue to be used in a run-time test. */ 2629 2630 if (npeel < 0) 2631 { 2632 peel_iters_prologue = vf/2; 2633 if (vect_print_dump_info (REPORT_COST)) 2634 fprintf (vect_dump, "cost model: " 2635 "prologue peel iters set to vf/2."); 2636 2637 /* If peeling for alignment is unknown, loop bound of main loop becomes 2638 unknown. */ 2639 peel_iters_epilogue = vf/2; 2640 if (vect_print_dump_info (REPORT_COST)) 2641 fprintf (vect_dump, "cost model: " 2642 "epilogue peel iters set to vf/2 because " 2643 "peeling for alignment is unknown ."); 2644 2645 /* If peeled iterations are unknown, count a taken branch and a not taken 2646 branch per peeled loop. Even if scalar loop iterations are known, 2647 vector iterations are not known since peeled prologue iterations are 2648 not known. Hence guards remain the same. */ 2649 peel_guard_costs += 2 * (vect_get_cost (cond_branch_taken) 2650 + vect_get_cost (cond_branch_not_taken)); 2651 vec_outside_cost += (peel_iters_prologue * scalar_single_iter_cost) 2652 + (peel_iters_epilogue * scalar_single_iter_cost) 2653 + peel_guard_costs; 2654 } 2655 else 2656 { 2657 peel_iters_prologue = npeel; 2658 vec_outside_cost += vect_get_known_peeling_cost (loop_vinfo, 2659 peel_iters_prologue, &peel_iters_epilogue, 2660 scalar_single_iter_cost); 2661 } 2662 2663 /* FORNOW: The scalar outside cost is incremented in one of the 2664 following ways: 2665 2666 1. The vectorizer checks for alignment and aliasing and generates 2667 a condition that allows dynamic vectorization. A cost model 2668 check is ANDED with the versioning condition. Hence scalar code 2669 path now has the added cost of the versioning check. 2670 2671 if (cost > th & versioning_check) 2672 jmp to vector code 2673 2674 Hence run-time scalar is incremented by not-taken branch cost. 2675 2676 2. The vectorizer then checks if a prologue is required. If the 2677 cost model check was not done before during versioning, it has to 2678 be done before the prologue check. 2679 2680 if (cost <= th) 2681 prologue = scalar_iters 2682 if (prologue == 0) 2683 jmp to vector code 2684 else 2685 execute prologue 2686 if (prologue == num_iters) 2687 go to exit 2688 2689 Hence the run-time scalar cost is incremented by a taken branch, 2690 plus a not-taken branch, plus a taken branch cost. 2691 2692 3. The vectorizer then checks if an epilogue is required. If the 2693 cost model check was not done before during prologue check, it 2694 has to be done with the epilogue check. 2695 2696 if (prologue == 0) 2697 jmp to vector code 2698 else 2699 execute prologue 2700 if (prologue == num_iters) 2701 go to exit 2702 vector code: 2703 if ((cost <= th) | (scalar_iters-prologue-epilogue == 0)) 2704 jmp to epilogue 2705 2706 Hence the run-time scalar cost should be incremented by 2 taken 2707 branches. 2708 2709 TODO: The back end may reorder the BBS's differently and reverse 2710 conditions/branch directions. Change the estimates below to 2711 something more reasonable. */ 2712 2713 /* If the number of iterations is known and we do not do versioning, we can 2714 decide whether to vectorize at compile time. Hence the scalar version 2715 do not carry cost model guard costs. */ 2716 if (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) 2717 || LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT (loop_vinfo) 2718 || LOOP_REQUIRES_VERSIONING_FOR_ALIAS (loop_vinfo)) 2719 { 2720 /* Cost model check occurs at versioning. */ 2721 if (LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT (loop_vinfo) 2722 || LOOP_REQUIRES_VERSIONING_FOR_ALIAS (loop_vinfo)) 2723 scalar_outside_cost += vect_get_cost (cond_branch_not_taken); 2724 else 2725 { 2726 /* Cost model check occurs at prologue generation. */ 2727 if (LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo) < 0) 2728 scalar_outside_cost += 2 * vect_get_cost (cond_branch_taken) 2729 + vect_get_cost (cond_branch_not_taken); 2730 /* Cost model check occurs at epilogue generation. */ 2731 else 2732 scalar_outside_cost += 2 * vect_get_cost (cond_branch_taken); 2733 } 2734 } 2735 2736 /* Add SLP costs. */ 2737 slp_instances = LOOP_VINFO_SLP_INSTANCES (loop_vinfo); 2738 FOR_EACH_VEC_ELT (slp_instance, slp_instances, i, instance) 2739 { 2740 vec_outside_cost += SLP_INSTANCE_OUTSIDE_OF_LOOP_COST (instance); 2741 vec_inside_cost += SLP_INSTANCE_INSIDE_OF_LOOP_COST (instance); 2742 } 2743 2744 /* Calculate number of iterations required to make the vector version 2745 profitable, relative to the loop bodies only. The following condition 2746 must hold true: 2747 SIC * niters + SOC > VIC * ((niters-PL_ITERS-EP_ITERS)/VF) + VOC 2748 where 2749 SIC = scalar iteration cost, VIC = vector iteration cost, 2750 VOC = vector outside cost, VF = vectorization factor, 2751 PL_ITERS = prologue iterations, EP_ITERS= epilogue iterations 2752 SOC = scalar outside cost for run time cost model check. */ 2753 2754 if ((scalar_single_iter_cost * vf) > vec_inside_cost) 2755 { 2756 if (vec_outside_cost <= 0) 2757 min_profitable_iters = 1; 2758 else 2759 { 2760 min_profitable_iters = ((vec_outside_cost - scalar_outside_cost) * vf 2761 - vec_inside_cost * peel_iters_prologue 2762 - vec_inside_cost * peel_iters_epilogue) 2763 / ((scalar_single_iter_cost * vf) 2764 - vec_inside_cost); 2765 2766 if ((scalar_single_iter_cost * vf * min_profitable_iters) 2767 <= ((vec_inside_cost * min_profitable_iters) 2768 + ((vec_outside_cost - scalar_outside_cost) * vf))) 2769 min_profitable_iters++; 2770 } 2771 } 2772 /* vector version will never be profitable. */ 2773 else 2774 { 2775 if (vect_print_dump_info (REPORT_COST)) 2776 fprintf (vect_dump, "cost model: the vector iteration cost = %d " 2777 "divided by the scalar iteration cost = %d " 2778 "is greater or equal to the vectorization factor = %d.", 2779 vec_inside_cost, scalar_single_iter_cost, vf); 2780 return -1; 2781 } 2782 2783 if (vect_print_dump_info (REPORT_COST)) 2784 { 2785 fprintf (vect_dump, "Cost model analysis: \n"); 2786 fprintf (vect_dump, " Vector inside of loop cost: %d\n", 2787 vec_inside_cost); 2788 fprintf (vect_dump, " Vector outside of loop cost: %d\n", 2789 vec_outside_cost); 2790 fprintf (vect_dump, " Scalar iteration cost: %d\n", 2791 scalar_single_iter_cost); 2792 fprintf (vect_dump, " Scalar outside cost: %d\n", scalar_outside_cost); 2793 fprintf (vect_dump, " prologue iterations: %d\n", 2794 peel_iters_prologue); 2795 fprintf (vect_dump, " epilogue iterations: %d\n", 2796 peel_iters_epilogue); 2797 fprintf (vect_dump, " Calculated minimum iters for profitability: %d\n", 2798 min_profitable_iters); 2799 } 2800 2801 min_profitable_iters = 2802 min_profitable_iters < vf ? vf : min_profitable_iters; 2803 2804 /* Because the condition we create is: 2805 if (niters <= min_profitable_iters) 2806 then skip the vectorized loop. */ 2807 min_profitable_iters--; 2808 2809 if (vect_print_dump_info (REPORT_COST)) 2810 fprintf (vect_dump, " Profitability threshold = %d\n", 2811 min_profitable_iters); 2812 2813 return min_profitable_iters; 2814 } 2815 2816 2817 /* TODO: Close dependency between vect_model_*_cost and vectorizable_* 2818 functions. Design better to avoid maintenance issues. */ 2819 2820 /* Function vect_model_reduction_cost. 2821 2822 Models cost for a reduction operation, including the vector ops 2823 generated within the strip-mine loop, the initial definition before 2824 the loop, and the epilogue code that must be generated. */ 2825 2826 static bool 2827 vect_model_reduction_cost (stmt_vec_info stmt_info, enum tree_code reduc_code, 2828 int ncopies) 2829 { 2830 int outer_cost = 0; 2831 enum tree_code code; 2832 optab optab; 2833 tree vectype; 2834 gimple stmt, orig_stmt; 2835 tree reduction_op; 2836 enum machine_mode mode; 2837 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 2838 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 2839 2840 2841 /* Cost of reduction op inside loop. */ 2842 STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) 2843 += ncopies * vect_get_cost (vector_stmt); 2844 2845 stmt = STMT_VINFO_STMT (stmt_info); 2846 2847 switch (get_gimple_rhs_class (gimple_assign_rhs_code (stmt))) 2848 { 2849 case GIMPLE_SINGLE_RHS: 2850 gcc_assert (TREE_OPERAND_LENGTH (gimple_assign_rhs1 (stmt)) == ternary_op); 2851 reduction_op = TREE_OPERAND (gimple_assign_rhs1 (stmt), 2); 2852 break; 2853 case GIMPLE_UNARY_RHS: 2854 reduction_op = gimple_assign_rhs1 (stmt); 2855 break; 2856 case GIMPLE_BINARY_RHS: 2857 reduction_op = gimple_assign_rhs2 (stmt); 2858 break; 2859 case GIMPLE_TERNARY_RHS: 2860 reduction_op = gimple_assign_rhs3 (stmt); 2861 break; 2862 default: 2863 gcc_unreachable (); 2864 } 2865 2866 vectype = get_vectype_for_scalar_type (TREE_TYPE (reduction_op)); 2867 if (!vectype) 2868 { 2869 if (vect_print_dump_info (REPORT_COST)) 2870 { 2871 fprintf (vect_dump, "unsupported data-type "); 2872 print_generic_expr (vect_dump, TREE_TYPE (reduction_op), TDF_SLIM); 2873 } 2874 return false; 2875 } 2876 2877 mode = TYPE_MODE (vectype); 2878 orig_stmt = STMT_VINFO_RELATED_STMT (stmt_info); 2879 2880 if (!orig_stmt) 2881 orig_stmt = STMT_VINFO_STMT (stmt_info); 2882 2883 code = gimple_assign_rhs_code (orig_stmt); 2884 2885 /* Add in cost for initial definition. */ 2886 outer_cost += vect_get_cost (scalar_to_vec); 2887 2888 /* Determine cost of epilogue code. 2889 2890 We have a reduction operator that will reduce the vector in one statement. 2891 Also requires scalar extract. */ 2892 2893 if (!nested_in_vect_loop_p (loop, orig_stmt)) 2894 { 2895 if (reduc_code != ERROR_MARK) 2896 outer_cost += vect_get_cost (vector_stmt) 2897 + vect_get_cost (vec_to_scalar); 2898 else 2899 { 2900 int vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1); 2901 tree bitsize = 2902 TYPE_SIZE (TREE_TYPE (gimple_assign_lhs (orig_stmt))); 2903 int element_bitsize = tree_low_cst (bitsize, 1); 2904 int nelements = vec_size_in_bits / element_bitsize; 2905 2906 optab = optab_for_tree_code (code, vectype, optab_default); 2907 2908 /* We have a whole vector shift available. */ 2909 if (VECTOR_MODE_P (mode) 2910 && optab_handler (optab, mode) != CODE_FOR_nothing 2911 && optab_handler (vec_shr_optab, mode) != CODE_FOR_nothing) 2912 /* Final reduction via vector shifts and the reduction operator. Also 2913 requires scalar extract. */ 2914 outer_cost += ((exact_log2(nelements) * 2) 2915 * vect_get_cost (vector_stmt) 2916 + vect_get_cost (vec_to_scalar)); 2917 else 2918 /* Use extracts and reduction op for final reduction. For N elements, 2919 we have N extracts and N-1 reduction ops. */ 2920 outer_cost += ((nelements + nelements - 1) 2921 * vect_get_cost (vector_stmt)); 2922 } 2923 } 2924 2925 STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info) = outer_cost; 2926 2927 if (vect_print_dump_info (REPORT_COST)) 2928 fprintf (vect_dump, "vect_model_reduction_cost: inside_cost = %d, " 2929 "outside_cost = %d .", STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info), 2930 STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info)); 2931 2932 return true; 2933 } 2934 2935 2936 /* Function vect_model_induction_cost. 2937 2938 Models cost for induction operations. */ 2939 2940 static void 2941 vect_model_induction_cost (stmt_vec_info stmt_info, int ncopies) 2942 { 2943 /* loop cost for vec_loop. */ 2944 STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info) 2945 = ncopies * vect_get_cost (vector_stmt); 2946 /* prologue cost for vec_init and vec_step. */ 2947 STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info) 2948 = 2 * vect_get_cost (scalar_to_vec); 2949 2950 if (vect_print_dump_info (REPORT_COST)) 2951 fprintf (vect_dump, "vect_model_induction_cost: inside_cost = %d, " 2952 "outside_cost = %d .", STMT_VINFO_INSIDE_OF_LOOP_COST (stmt_info), 2953 STMT_VINFO_OUTSIDE_OF_LOOP_COST (stmt_info)); 2954 } 2955 2956 2957 /* Function get_initial_def_for_induction 2958 2959 Input: 2960 STMT - a stmt that performs an induction operation in the loop. 2961 IV_PHI - the initial value of the induction variable 2962 2963 Output: 2964 Return a vector variable, initialized with the first VF values of 2965 the induction variable. E.g., for an iv with IV_PHI='X' and 2966 evolution S, for a vector of 4 units, we want to return: 2967 [X, X + S, X + 2*S, X + 3*S]. */ 2968 2969 static tree 2970 get_initial_def_for_induction (gimple iv_phi) 2971 { 2972 stmt_vec_info stmt_vinfo = vinfo_for_stmt (iv_phi); 2973 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); 2974 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 2975 tree scalar_type; 2976 tree vectype; 2977 int nunits; 2978 edge pe = loop_preheader_edge (loop); 2979 struct loop *iv_loop; 2980 basic_block new_bb; 2981 tree vec, vec_init, vec_step, t; 2982 tree access_fn; 2983 tree new_var; 2984 tree new_name; 2985 gimple init_stmt, induction_phi, new_stmt; 2986 tree induc_def, vec_def, vec_dest; 2987 tree init_expr, step_expr; 2988 int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); 2989 int i; 2990 bool ok; 2991 int ncopies; 2992 tree expr; 2993 stmt_vec_info phi_info = vinfo_for_stmt (iv_phi); 2994 bool nested_in_vect_loop = false; 2995 gimple_seq stmts = NULL; 2996 imm_use_iterator imm_iter; 2997 use_operand_p use_p; 2998 gimple exit_phi; 2999 edge latch_e; 3000 tree loop_arg; 3001 gimple_stmt_iterator si; 3002 basic_block bb = gimple_bb (iv_phi); 3003 tree stepvectype; 3004 tree resvectype; 3005 3006 /* Is phi in an inner-loop, while vectorizing an enclosing outer-loop? */ 3007 if (nested_in_vect_loop_p (loop, iv_phi)) 3008 { 3009 nested_in_vect_loop = true; 3010 iv_loop = loop->inner; 3011 } 3012 else 3013 iv_loop = loop; 3014 gcc_assert (iv_loop == (gimple_bb (iv_phi))->loop_father); 3015 3016 latch_e = loop_latch_edge (iv_loop); 3017 loop_arg = PHI_ARG_DEF_FROM_EDGE (iv_phi, latch_e); 3018 3019 access_fn = analyze_scalar_evolution (iv_loop, PHI_RESULT (iv_phi)); 3020 gcc_assert (access_fn); 3021 STRIP_NOPS (access_fn); 3022 ok = vect_is_simple_iv_evolution (iv_loop->num, access_fn, 3023 &init_expr, &step_expr); 3024 gcc_assert (ok); 3025 pe = loop_preheader_edge (iv_loop); 3026 3027 scalar_type = TREE_TYPE (init_expr); 3028 vectype = get_vectype_for_scalar_type (scalar_type); 3029 resvectype = get_vectype_for_scalar_type (TREE_TYPE (PHI_RESULT (iv_phi))); 3030 gcc_assert (vectype); 3031 nunits = TYPE_VECTOR_SUBPARTS (vectype); 3032 ncopies = vf / nunits; 3033 3034 gcc_assert (phi_info); 3035 gcc_assert (ncopies >= 1); 3036 3037 /* Find the first insertion point in the BB. */ 3038 si = gsi_after_labels (bb); 3039 3040 /* Create the vector that holds the initial_value of the induction. */ 3041 if (nested_in_vect_loop) 3042 { 3043 /* iv_loop is nested in the loop to be vectorized. init_expr had already 3044 been created during vectorization of previous stmts. We obtain it 3045 from the STMT_VINFO_VEC_STMT of the defining stmt. */ 3046 tree iv_def = PHI_ARG_DEF_FROM_EDGE (iv_phi, 3047 loop_preheader_edge (iv_loop)); 3048 vec_init = vect_get_vec_def_for_operand (iv_def, iv_phi, NULL); 3049 } 3050 else 3051 { 3052 /* iv_loop is the loop to be vectorized. Create: 3053 vec_init = [X, X+S, X+2*S, X+3*S] (S = step_expr, X = init_expr) */ 3054 new_var = vect_get_new_vect_var (scalar_type, vect_scalar_var, "var_"); 3055 add_referenced_var (new_var); 3056 3057 new_name = force_gimple_operand (init_expr, &stmts, false, new_var); 3058 if (stmts) 3059 { 3060 new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); 3061 gcc_assert (!new_bb); 3062 } 3063 3064 t = NULL_TREE; 3065 t = tree_cons (NULL_TREE, new_name, t); 3066 for (i = 1; i < nunits; i++) 3067 { 3068 /* Create: new_name_i = new_name + step_expr */ 3069 enum tree_code code = POINTER_TYPE_P (scalar_type) 3070 ? POINTER_PLUS_EXPR : PLUS_EXPR; 3071 init_stmt = gimple_build_assign_with_ops (code, new_var, 3072 new_name, step_expr); 3073 new_name = make_ssa_name (new_var, init_stmt); 3074 gimple_assign_set_lhs (init_stmt, new_name); 3075 3076 new_bb = gsi_insert_on_edge_immediate (pe, init_stmt); 3077 gcc_assert (!new_bb); 3078 3079 if (vect_print_dump_info (REPORT_DETAILS)) 3080 { 3081 fprintf (vect_dump, "created new init_stmt: "); 3082 print_gimple_stmt (vect_dump, init_stmt, 0, TDF_SLIM); 3083 } 3084 t = tree_cons (NULL_TREE, new_name, t); 3085 } 3086 /* Create a vector from [new_name_0, new_name_1, ..., new_name_nunits-1] */ 3087 vec = build_constructor_from_list (vectype, nreverse (t)); 3088 vec_init = vect_init_vector (iv_phi, vec, vectype, NULL); 3089 } 3090 3091 3092 /* Create the vector that holds the step of the induction. */ 3093 if (nested_in_vect_loop) 3094 /* iv_loop is nested in the loop to be vectorized. Generate: 3095 vec_step = [S, S, S, S] */ 3096 new_name = step_expr; 3097 else 3098 { 3099 /* iv_loop is the loop to be vectorized. Generate: 3100 vec_step = [VF*S, VF*S, VF*S, VF*S] */ 3101 expr = build_int_cst (TREE_TYPE (step_expr), vf); 3102 new_name = fold_build2 (MULT_EXPR, TREE_TYPE (step_expr), 3103 expr, step_expr); 3104 } 3105 3106 t = unshare_expr (new_name); 3107 gcc_assert (CONSTANT_CLASS_P (new_name)); 3108 stepvectype = get_vectype_for_scalar_type (TREE_TYPE (new_name)); 3109 gcc_assert (stepvectype); 3110 vec = build_vector_from_val (stepvectype, t); 3111 vec_step = vect_init_vector (iv_phi, vec, stepvectype, NULL); 3112 3113 3114 /* Create the following def-use cycle: 3115 loop prolog: 3116 vec_init = ... 3117 vec_step = ... 3118 loop: 3119 vec_iv = PHI <vec_init, vec_loop> 3120 ... 3121 STMT 3122 ... 3123 vec_loop = vec_iv + vec_step; */ 3124 3125 /* Create the induction-phi that defines the induction-operand. */ 3126 vec_dest = vect_get_new_vect_var (vectype, vect_simple_var, "vec_iv_"); 3127 add_referenced_var (vec_dest); 3128 induction_phi = create_phi_node (vec_dest, iv_loop->header); 3129 set_vinfo_for_stmt (induction_phi, 3130 new_stmt_vec_info (induction_phi, loop_vinfo, NULL)); 3131 induc_def = PHI_RESULT (induction_phi); 3132 3133 /* Create the iv update inside the loop */ 3134 new_stmt = gimple_build_assign_with_ops (PLUS_EXPR, vec_dest, 3135 induc_def, vec_step); 3136 vec_def = make_ssa_name (vec_dest, new_stmt); 3137 gimple_assign_set_lhs (new_stmt, vec_def); 3138 gsi_insert_before (&si, new_stmt, GSI_SAME_STMT); 3139 set_vinfo_for_stmt (new_stmt, new_stmt_vec_info (new_stmt, loop_vinfo, 3140 NULL)); 3141 3142 /* Set the arguments of the phi node: */ 3143 add_phi_arg (induction_phi, vec_init, pe, UNKNOWN_LOCATION); 3144 add_phi_arg (induction_phi, vec_def, loop_latch_edge (iv_loop), 3145 UNKNOWN_LOCATION); 3146 3147 3148 /* In case that vectorization factor (VF) is bigger than the number 3149 of elements that we can fit in a vectype (nunits), we have to generate 3150 more than one vector stmt - i.e - we need to "unroll" the 3151 vector stmt by a factor VF/nunits. For more details see documentation 3152 in vectorizable_operation. */ 3153 3154 if (ncopies > 1) 3155 { 3156 stmt_vec_info prev_stmt_vinfo; 3157 /* FORNOW. This restriction should be relaxed. */ 3158 gcc_assert (!nested_in_vect_loop); 3159 3160 /* Create the vector that holds the step of the induction. */ 3161 expr = build_int_cst (TREE_TYPE (step_expr), nunits); 3162 new_name = fold_build2 (MULT_EXPR, TREE_TYPE (step_expr), 3163 expr, step_expr); 3164 t = unshare_expr (new_name); 3165 gcc_assert (CONSTANT_CLASS_P (new_name)); 3166 vec = build_vector_from_val (stepvectype, t); 3167 vec_step = vect_init_vector (iv_phi, vec, stepvectype, NULL); 3168 3169 vec_def = induc_def; 3170 prev_stmt_vinfo = vinfo_for_stmt (induction_phi); 3171 for (i = 1; i < ncopies; i++) 3172 { 3173 /* vec_i = vec_prev + vec_step */ 3174 new_stmt = gimple_build_assign_with_ops (PLUS_EXPR, vec_dest, 3175 vec_def, vec_step); 3176 vec_def = make_ssa_name (vec_dest, new_stmt); 3177 gimple_assign_set_lhs (new_stmt, vec_def); 3178 3179 gsi_insert_before (&si, new_stmt, GSI_SAME_STMT); 3180 if (!useless_type_conversion_p (resvectype, vectype)) 3181 { 3182 new_stmt = gimple_build_assign_with_ops 3183 (VIEW_CONVERT_EXPR, 3184 vect_get_new_vect_var (resvectype, vect_simple_var, 3185 "vec_iv_"), 3186 build1 (VIEW_CONVERT_EXPR, resvectype, 3187 gimple_assign_lhs (new_stmt)), NULL_TREE); 3188 gimple_assign_set_lhs (new_stmt, 3189 make_ssa_name 3190 (gimple_assign_lhs (new_stmt), new_stmt)); 3191 gsi_insert_before (&si, new_stmt, GSI_SAME_STMT); 3192 } 3193 set_vinfo_for_stmt (new_stmt, 3194 new_stmt_vec_info (new_stmt, loop_vinfo, NULL)); 3195 STMT_VINFO_RELATED_STMT (prev_stmt_vinfo) = new_stmt; 3196 prev_stmt_vinfo = vinfo_for_stmt (new_stmt); 3197 } 3198 } 3199 3200 if (nested_in_vect_loop) 3201 { 3202 /* Find the loop-closed exit-phi of the induction, and record 3203 the final vector of induction results: */ 3204 exit_phi = NULL; 3205 FOR_EACH_IMM_USE_FAST (use_p, imm_iter, loop_arg) 3206 { 3207 if (!flow_bb_inside_loop_p (iv_loop, gimple_bb (USE_STMT (use_p)))) 3208 { 3209 exit_phi = USE_STMT (use_p); 3210 break; 3211 } 3212 } 3213 if (exit_phi) 3214 { 3215 stmt_vec_info stmt_vinfo = vinfo_for_stmt (exit_phi); 3216 /* FORNOW. Currently not supporting the case that an inner-loop induction 3217 is not used in the outer-loop (i.e. only outside the outer-loop). */ 3218 gcc_assert (STMT_VINFO_RELEVANT_P (stmt_vinfo) 3219 && !STMT_VINFO_LIVE_P (stmt_vinfo)); 3220 3221 STMT_VINFO_VEC_STMT (stmt_vinfo) = new_stmt; 3222 if (vect_print_dump_info (REPORT_DETAILS)) 3223 { 3224 fprintf (vect_dump, "vector of inductions after inner-loop:"); 3225 print_gimple_stmt (vect_dump, new_stmt, 0, TDF_SLIM); 3226 } 3227 } 3228 } 3229 3230 3231 if (vect_print_dump_info (REPORT_DETAILS)) 3232 { 3233 fprintf (vect_dump, "transform induction: created def-use cycle: "); 3234 print_gimple_stmt (vect_dump, induction_phi, 0, TDF_SLIM); 3235 fprintf (vect_dump, "\n"); 3236 print_gimple_stmt (vect_dump, SSA_NAME_DEF_STMT (vec_def), 0, TDF_SLIM); 3237 } 3238 3239 STMT_VINFO_VEC_STMT (phi_info) = induction_phi; 3240 if (!useless_type_conversion_p (resvectype, vectype)) 3241 { 3242 new_stmt = gimple_build_assign_with_ops 3243 (VIEW_CONVERT_EXPR, 3244 vect_get_new_vect_var (resvectype, vect_simple_var, "vec_iv_"), 3245 build1 (VIEW_CONVERT_EXPR, resvectype, induc_def), NULL_TREE); 3246 induc_def = make_ssa_name (gimple_assign_lhs (new_stmt), new_stmt); 3247 gimple_assign_set_lhs (new_stmt, induc_def); 3248 si = gsi_start_bb (bb); 3249 gsi_insert_before (&si, new_stmt, GSI_SAME_STMT); 3250 set_vinfo_for_stmt (new_stmt, 3251 new_stmt_vec_info (new_stmt, loop_vinfo, NULL)); 3252 STMT_VINFO_RELATED_STMT (vinfo_for_stmt (new_stmt)) 3253 = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (induction_phi)); 3254 } 3255 3256 return induc_def; 3257 } 3258 3259 3260 /* Function get_initial_def_for_reduction 3261 3262 Input: 3263 STMT - a stmt that performs a reduction operation in the loop. 3264 INIT_VAL - the initial value of the reduction variable 3265 3266 Output: 3267 ADJUSTMENT_DEF - a tree that holds a value to be added to the final result 3268 of the reduction (used for adjusting the epilog - see below). 3269 Return a vector variable, initialized according to the operation that STMT 3270 performs. This vector will be used as the initial value of the 3271 vector of partial results. 3272 3273 Option1 (adjust in epilog): Initialize the vector as follows: 3274 add/bit or/xor: [0,0,...,0,0] 3275 mult/bit and: [1,1,...,1,1] 3276 min/max/cond_expr: [init_val,init_val,..,init_val,init_val] 3277 and when necessary (e.g. add/mult case) let the caller know 3278 that it needs to adjust the result by init_val. 3279 3280 Option2: Initialize the vector as follows: 3281 add/bit or/xor: [init_val,0,0,...,0] 3282 mult/bit and: [init_val,1,1,...,1] 3283 min/max/cond_expr: [init_val,init_val,...,init_val] 3284 and no adjustments are needed. 3285 3286 For example, for the following code: 3287 3288 s = init_val; 3289 for (i=0;i<n;i++) 3290 s = s + a[i]; 3291 3292 STMT is 's = s + a[i]', and the reduction variable is 's'. 3293 For a vector of 4 units, we want to return either [0,0,0,init_val], 3294 or [0,0,0,0] and let the caller know that it needs to adjust 3295 the result at the end by 'init_val'. 3296 3297 FORNOW, we are using the 'adjust in epilog' scheme, because this way the 3298 initialization vector is simpler (same element in all entries), if 3299 ADJUSTMENT_DEF is not NULL, and Option2 otherwise. 3300 3301 A cost model should help decide between these two schemes. */ 3302 3303 tree 3304 get_initial_def_for_reduction (gimple stmt, tree init_val, 3305 tree *adjustment_def) 3306 { 3307 stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt); 3308 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo); 3309 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 3310 tree scalar_type = TREE_TYPE (init_val); 3311 tree vectype = get_vectype_for_scalar_type (scalar_type); 3312 int nunits; 3313 enum tree_code code = gimple_assign_rhs_code (stmt); 3314 tree def_for_init; 3315 tree init_def; 3316 tree t = NULL_TREE; 3317 int i; 3318 bool nested_in_vect_loop = false; 3319 tree init_value; 3320 REAL_VALUE_TYPE real_init_val = dconst0; 3321 int int_init_val = 0; 3322 gimple def_stmt = NULL; 3323 3324 gcc_assert (vectype); 3325 nunits = TYPE_VECTOR_SUBPARTS (vectype); 3326 3327 gcc_assert (POINTER_TYPE_P (scalar_type) || INTEGRAL_TYPE_P (scalar_type) 3328 || SCALAR_FLOAT_TYPE_P (scalar_type)); 3329 3330 if (nested_in_vect_loop_p (loop, stmt)) 3331 nested_in_vect_loop = true; 3332 else 3333 gcc_assert (loop == (gimple_bb (stmt))->loop_father); 3334 3335 /* In case of double reduction we only create a vector variable to be put 3336 in the reduction phi node. The actual statement creation is done in 3337 vect_create_epilog_for_reduction. */ 3338 if (adjustment_def && nested_in_vect_loop 3339 && TREE_CODE (init_val) == SSA_NAME 3340 && (def_stmt = SSA_NAME_DEF_STMT (init_val)) 3341 && gimple_code (def_stmt) == GIMPLE_PHI 3342 && flow_bb_inside_loop_p (loop, gimple_bb (def_stmt)) 3343 && vinfo_for_stmt (def_stmt) 3344 && STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def_stmt)) 3345 == vect_double_reduction_def) 3346 { 3347 *adjustment_def = NULL; 3348 return vect_create_destination_var (init_val, vectype); 3349 } 3350 3351 if (TREE_CONSTANT (init_val)) 3352 { 3353 if (SCALAR_FLOAT_TYPE_P (scalar_type)) 3354 init_value = build_real (scalar_type, TREE_REAL_CST (init_val)); 3355 else 3356 init_value = build_int_cst (scalar_type, TREE_INT_CST_LOW (init_val)); 3357 } 3358 else 3359 init_value = init_val; 3360 3361 switch (code) 3362 { 3363 case WIDEN_SUM_EXPR: 3364 case DOT_PROD_EXPR: 3365 case PLUS_EXPR: 3366 case MINUS_EXPR: 3367 case BIT_IOR_EXPR: 3368 case BIT_XOR_EXPR: 3369 case MULT_EXPR: 3370 case BIT_AND_EXPR: 3371 /* ADJUSMENT_DEF is NULL when called from 3372 vect_create_epilog_for_reduction to vectorize double reduction. */ 3373 if (adjustment_def) 3374 { 3375 if (nested_in_vect_loop) 3376 *adjustment_def = vect_get_vec_def_for_operand (init_val, stmt, 3377 NULL); 3378 else 3379 *adjustment_def = init_val; 3380 } 3381 3382 if (code == MULT_EXPR) 3383 { 3384 real_init_val = dconst1; 3385 int_init_val = 1; 3386 } 3387 3388 if (code == BIT_AND_EXPR) 3389 int_init_val = -1; 3390 3391 if (SCALAR_FLOAT_TYPE_P (scalar_type)) 3392 def_for_init = build_real (scalar_type, real_init_val); 3393 else 3394 def_for_init = build_int_cst (scalar_type, int_init_val); 3395 3396 /* Create a vector of '0' or '1' except the first element. */ 3397 for (i = nunits - 2; i >= 0; --i) 3398 t = tree_cons (NULL_TREE, def_for_init, t); 3399 3400 /* Option1: the first element is '0' or '1' as well. */ 3401 if (adjustment_def) 3402 { 3403 t = tree_cons (NULL_TREE, def_for_init, t); 3404 init_def = build_vector (vectype, t); 3405 break; 3406 } 3407 3408 /* Option2: the first element is INIT_VAL. */ 3409 t = tree_cons (NULL_TREE, init_value, t); 3410 if (TREE_CONSTANT (init_val)) 3411 init_def = build_vector (vectype, t); 3412 else 3413 init_def = build_constructor_from_list (vectype, t); 3414 3415 break; 3416 3417 case MIN_EXPR: 3418 case MAX_EXPR: 3419 case COND_EXPR: 3420 if (adjustment_def) 3421 { 3422 *adjustment_def = NULL_TREE; 3423 init_def = vect_get_vec_def_for_operand (init_val, stmt, NULL); 3424 break; 3425 } 3426 3427 init_def = build_vector_from_val (vectype, init_value); 3428 break; 3429 3430 default: 3431 gcc_unreachable (); 3432 } 3433 3434 return init_def; 3435 } 3436 3437 3438 /* Function vect_create_epilog_for_reduction 3439 3440 Create code at the loop-epilog to finalize the result of a reduction 3441 computation. 3442 3443 VECT_DEFS is list of vector of partial results, i.e., the lhs's of vector 3444 reduction statements. 3445 STMT is the scalar reduction stmt that is being vectorized. 3446 NCOPIES is > 1 in case the vectorization factor (VF) is bigger than the 3447 number of elements that we can fit in a vectype (nunits). In this case 3448 we have to generate more than one vector stmt - i.e - we need to "unroll" 3449 the vector stmt by a factor VF/nunits. For more details see documentation 3450 in vectorizable_operation. 3451 REDUC_CODE is the tree-code for the epilog reduction. 3452 REDUCTION_PHIS is a list of the phi-nodes that carry the reduction 3453 computation. 3454 REDUC_INDEX is the index of the operand in the right hand side of the 3455 statement that is defined by REDUCTION_PHI. 3456 DOUBLE_REDUC is TRUE if double reduction phi nodes should be handled. 3457 SLP_NODE is an SLP node containing a group of reduction statements. The 3458 first one in this group is STMT. 3459 3460 This function: 3461 1. Creates the reduction def-use cycles: sets the arguments for 3462 REDUCTION_PHIS: 3463 The loop-entry argument is the vectorized initial-value of the reduction. 3464 The loop-latch argument is taken from VECT_DEFS - the vector of partial 3465 sums. 3466 2. "Reduces" each vector of partial results VECT_DEFS into a single result, 3467 by applying the operation specified by REDUC_CODE if available, or by 3468 other means (whole-vector shifts or a scalar loop). 3469 The function also creates a new phi node at the loop exit to preserve 3470 loop-closed form, as illustrated below. 3471 3472 The flow at the entry to this function: 3473 3474 loop: 3475 vec_def = phi <null, null> # REDUCTION_PHI 3476 VECT_DEF = vector_stmt # vectorized form of STMT 3477 s_loop = scalar_stmt # (scalar) STMT 3478 loop_exit: 3479 s_out0 = phi <s_loop> # (scalar) EXIT_PHI 3480 use <s_out0> 3481 use <s_out0> 3482 3483 The above is transformed by this function into: 3484 3485 loop: 3486 vec_def = phi <vec_init, VECT_DEF> # REDUCTION_PHI 3487 VECT_DEF = vector_stmt # vectorized form of STMT 3488 s_loop = scalar_stmt # (scalar) STMT 3489 loop_exit: 3490 s_out0 = phi <s_loop> # (scalar) EXIT_PHI 3491 v_out1 = phi <VECT_DEF> # NEW_EXIT_PHI 3492 v_out2 = reduce <v_out1> 3493 s_out3 = extract_field <v_out2, 0> 3494 s_out4 = adjust_result <s_out3> 3495 use <s_out4> 3496 use <s_out4> 3497 */ 3498 3499 static void 3500 vect_create_epilog_for_reduction (VEC (tree, heap) *vect_defs, gimple stmt, 3501 int ncopies, enum tree_code reduc_code, 3502 VEC (gimple, heap) *reduction_phis, 3503 int reduc_index, bool double_reduc, 3504 slp_tree slp_node) 3505 { 3506 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 3507 stmt_vec_info prev_phi_info; 3508 tree vectype; 3509 enum machine_mode mode; 3510 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 3511 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo), *outer_loop = NULL; 3512 basic_block exit_bb; 3513 tree scalar_dest; 3514 tree scalar_type; 3515 gimple new_phi = NULL, phi; 3516 gimple_stmt_iterator exit_gsi; 3517 tree vec_dest; 3518 tree new_temp = NULL_TREE, new_dest, new_name, new_scalar_dest; 3519 gimple epilog_stmt = NULL; 3520 enum tree_code code = gimple_assign_rhs_code (stmt); 3521 gimple exit_phi; 3522 tree bitsize, bitpos; 3523 tree adjustment_def = NULL; 3524 tree vec_initial_def = NULL; 3525 tree reduction_op, expr, def; 3526 tree orig_name, scalar_result; 3527 imm_use_iterator imm_iter, phi_imm_iter; 3528 use_operand_p use_p, phi_use_p; 3529 bool extract_scalar_result = false; 3530 gimple use_stmt, orig_stmt, reduction_phi = NULL; 3531 bool nested_in_vect_loop = false; 3532 VEC (gimple, heap) *new_phis = NULL; 3533 VEC (gimple, heap) *inner_phis = NULL; 3534 enum vect_def_type dt = vect_unknown_def_type; 3535 int j, i; 3536 VEC (tree, heap) *scalar_results = NULL; 3537 unsigned int group_size = 1, k, ratio; 3538 VEC (tree, heap) *vec_initial_defs = NULL; 3539 VEC (gimple, heap) *phis; 3540 bool slp_reduc = false; 3541 tree new_phi_result; 3542 gimple inner_phi = NULL; 3543 3544 if (slp_node) 3545 group_size = VEC_length (gimple, SLP_TREE_SCALAR_STMTS (slp_node)); 3546 3547 if (nested_in_vect_loop_p (loop, stmt)) 3548 { 3549 outer_loop = loop; 3550 loop = loop->inner; 3551 nested_in_vect_loop = true; 3552 gcc_assert (!slp_node); 3553 } 3554 3555 switch (get_gimple_rhs_class (gimple_assign_rhs_code (stmt))) 3556 { 3557 case GIMPLE_SINGLE_RHS: 3558 gcc_assert (TREE_OPERAND_LENGTH (gimple_assign_rhs1 (stmt)) 3559 == ternary_op); 3560 reduction_op = TREE_OPERAND (gimple_assign_rhs1 (stmt), reduc_index); 3561 break; 3562 case GIMPLE_UNARY_RHS: 3563 reduction_op = gimple_assign_rhs1 (stmt); 3564 break; 3565 case GIMPLE_BINARY_RHS: 3566 reduction_op = reduc_index ? 3567 gimple_assign_rhs2 (stmt) : gimple_assign_rhs1 (stmt); 3568 break; 3569 case GIMPLE_TERNARY_RHS: 3570 reduction_op = gimple_op (stmt, reduc_index + 1); 3571 break; 3572 default: 3573 gcc_unreachable (); 3574 } 3575 3576 vectype = get_vectype_for_scalar_type (TREE_TYPE (reduction_op)); 3577 gcc_assert (vectype); 3578 mode = TYPE_MODE (vectype); 3579 3580 /* 1. Create the reduction def-use cycle: 3581 Set the arguments of REDUCTION_PHIS, i.e., transform 3582 3583 loop: 3584 vec_def = phi <null, null> # REDUCTION_PHI 3585 VECT_DEF = vector_stmt # vectorized form of STMT 3586 ... 3587 3588 into: 3589 3590 loop: 3591 vec_def = phi <vec_init, VECT_DEF> # REDUCTION_PHI 3592 VECT_DEF = vector_stmt # vectorized form of STMT 3593 ... 3594 3595 (in case of SLP, do it for all the phis). */ 3596 3597 /* Get the loop-entry arguments. */ 3598 if (slp_node) 3599 vect_get_vec_defs (reduction_op, NULL_TREE, stmt, &vec_initial_defs, 3600 NULL, slp_node, reduc_index); 3601 else 3602 { 3603 vec_initial_defs = VEC_alloc (tree, heap, 1); 3604 /* For the case of reduction, vect_get_vec_def_for_operand returns 3605 the scalar def before the loop, that defines the initial value 3606 of the reduction variable. */ 3607 vec_initial_def = vect_get_vec_def_for_operand (reduction_op, stmt, 3608 &adjustment_def); 3609 VEC_quick_push (tree, vec_initial_defs, vec_initial_def); 3610 } 3611 3612 /* Set phi nodes arguments. */ 3613 FOR_EACH_VEC_ELT (gimple, reduction_phis, i, phi) 3614 { 3615 tree vec_init_def = VEC_index (tree, vec_initial_defs, i); 3616 tree def = VEC_index (tree, vect_defs, i); 3617 for (j = 0; j < ncopies; j++) 3618 { 3619 /* Set the loop-entry arg of the reduction-phi. */ 3620 add_phi_arg (phi, vec_init_def, loop_preheader_edge (loop), 3621 UNKNOWN_LOCATION); 3622 3623 /* Set the loop-latch arg for the reduction-phi. */ 3624 if (j > 0) 3625 def = vect_get_vec_def_for_stmt_copy (vect_unknown_def_type, def); 3626 3627 add_phi_arg (phi, def, loop_latch_edge (loop), UNKNOWN_LOCATION); 3628 3629 if (vect_print_dump_info (REPORT_DETAILS)) 3630 { 3631 fprintf (vect_dump, "transform reduction: created def-use" 3632 " cycle: "); 3633 print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); 3634 fprintf (vect_dump, "\n"); 3635 print_gimple_stmt (vect_dump, SSA_NAME_DEF_STMT (def), 0, 3636 TDF_SLIM); 3637 } 3638 3639 phi = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (phi)); 3640 } 3641 } 3642 3643 VEC_free (tree, heap, vec_initial_defs); 3644 3645 /* 2. Create epilog code. 3646 The reduction epilog code operates across the elements of the vector 3647 of partial results computed by the vectorized loop. 3648 The reduction epilog code consists of: 3649 3650 step 1: compute the scalar result in a vector (v_out2) 3651 step 2: extract the scalar result (s_out3) from the vector (v_out2) 3652 step 3: adjust the scalar result (s_out3) if needed. 3653 3654 Step 1 can be accomplished using one the following three schemes: 3655 (scheme 1) using reduc_code, if available. 3656 (scheme 2) using whole-vector shifts, if available. 3657 (scheme 3) using a scalar loop. In this case steps 1+2 above are 3658 combined. 3659 3660 The overall epilog code looks like this: 3661 3662 s_out0 = phi <s_loop> # original EXIT_PHI 3663 v_out1 = phi <VECT_DEF> # NEW_EXIT_PHI 3664 v_out2 = reduce <v_out1> # step 1 3665 s_out3 = extract_field <v_out2, 0> # step 2 3666 s_out4 = adjust_result <s_out3> # step 3 3667 3668 (step 3 is optional, and steps 1 and 2 may be combined). 3669 Lastly, the uses of s_out0 are replaced by s_out4. */ 3670 3671 3672 /* 2.1 Create new loop-exit-phis to preserve loop-closed form: 3673 v_out1 = phi <VECT_DEF> 3674 Store them in NEW_PHIS. */ 3675 3676 exit_bb = single_exit (loop)->dest; 3677 prev_phi_info = NULL; 3678 new_phis = VEC_alloc (gimple, heap, VEC_length (tree, vect_defs)); 3679 FOR_EACH_VEC_ELT (tree, vect_defs, i, def) 3680 { 3681 for (j = 0; j < ncopies; j++) 3682 { 3683 phi = create_phi_node (SSA_NAME_VAR (def), exit_bb); 3684 set_vinfo_for_stmt (phi, new_stmt_vec_info (phi, loop_vinfo, NULL)); 3685 if (j == 0) 3686 VEC_quick_push (gimple, new_phis, phi); 3687 else 3688 { 3689 def = vect_get_vec_def_for_stmt_copy (dt, def); 3690 STMT_VINFO_RELATED_STMT (prev_phi_info) = phi; 3691 } 3692 3693 SET_PHI_ARG_DEF (phi, single_exit (loop)->dest_idx, def); 3694 prev_phi_info = vinfo_for_stmt (phi); 3695 } 3696 } 3697 3698 /* The epilogue is created for the outer-loop, i.e., for the loop being 3699 vectorized. Create exit phis for the outer loop. */ 3700 if (double_reduc) 3701 { 3702 loop = outer_loop; 3703 exit_bb = single_exit (loop)->dest; 3704 inner_phis = VEC_alloc (gimple, heap, VEC_length (tree, vect_defs)); 3705 FOR_EACH_VEC_ELT (gimple, new_phis, i, phi) 3706 { 3707 gimple outer_phi = create_phi_node (SSA_NAME_VAR (PHI_RESULT (phi)), 3708 exit_bb); 3709 SET_PHI_ARG_DEF (outer_phi, single_exit (loop)->dest_idx, 3710 PHI_RESULT (phi)); 3711 set_vinfo_for_stmt (outer_phi, new_stmt_vec_info (outer_phi, 3712 loop_vinfo, NULL)); 3713 VEC_quick_push (gimple, inner_phis, phi); 3714 VEC_replace (gimple, new_phis, i, outer_phi); 3715 prev_phi_info = vinfo_for_stmt (outer_phi); 3716 while (STMT_VINFO_RELATED_STMT (vinfo_for_stmt (phi))) 3717 { 3718 phi = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (phi)); 3719 outer_phi = create_phi_node (SSA_NAME_VAR (PHI_RESULT (phi)), 3720 exit_bb); 3721 SET_PHI_ARG_DEF (outer_phi, single_exit (loop)->dest_idx, 3722 PHI_RESULT (phi)); 3723 set_vinfo_for_stmt (outer_phi, new_stmt_vec_info (outer_phi, 3724 loop_vinfo, NULL)); 3725 STMT_VINFO_RELATED_STMT (prev_phi_info) = outer_phi; 3726 prev_phi_info = vinfo_for_stmt (outer_phi); 3727 } 3728 } 3729 } 3730 3731 exit_gsi = gsi_after_labels (exit_bb); 3732 3733 /* 2.2 Get the relevant tree-code to use in the epilog for schemes 2,3 3734 (i.e. when reduc_code is not available) and in the final adjustment 3735 code (if needed). Also get the original scalar reduction variable as 3736 defined in the loop. In case STMT is a "pattern-stmt" (i.e. - it 3737 represents a reduction pattern), the tree-code and scalar-def are 3738 taken from the original stmt that the pattern-stmt (STMT) replaces. 3739 Otherwise (it is a regular reduction) - the tree-code and scalar-def 3740 are taken from STMT. */ 3741 3742 orig_stmt = STMT_VINFO_RELATED_STMT (stmt_info); 3743 if (!orig_stmt) 3744 { 3745 /* Regular reduction */ 3746 orig_stmt = stmt; 3747 } 3748 else 3749 { 3750 /* Reduction pattern */ 3751 stmt_vec_info stmt_vinfo = vinfo_for_stmt (orig_stmt); 3752 gcc_assert (STMT_VINFO_IN_PATTERN_P (stmt_vinfo)); 3753 gcc_assert (STMT_VINFO_RELATED_STMT (stmt_vinfo) == stmt); 3754 } 3755 3756 code = gimple_assign_rhs_code (orig_stmt); 3757 /* For MINUS_EXPR the initial vector is [init_val,0,...,0], therefore, 3758 partial results are added and not subtracted. */ 3759 if (code == MINUS_EXPR) 3760 code = PLUS_EXPR; 3761 3762 scalar_dest = gimple_assign_lhs (orig_stmt); 3763 scalar_type = TREE_TYPE (scalar_dest); 3764 scalar_results = VEC_alloc (tree, heap, group_size); 3765 new_scalar_dest = vect_create_destination_var (scalar_dest, NULL); 3766 bitsize = TYPE_SIZE (scalar_type); 3767 3768 /* In case this is a reduction in an inner-loop while vectorizing an outer 3769 loop - we don't need to extract a single scalar result at the end of the 3770 inner-loop (unless it is double reduction, i.e., the use of reduction is 3771 outside the outer-loop). The final vector of partial results will be used 3772 in the vectorized outer-loop, or reduced to a scalar result at the end of 3773 the outer-loop. */ 3774 if (nested_in_vect_loop && !double_reduc) 3775 goto vect_finalize_reduction; 3776 3777 /* SLP reduction without reduction chain, e.g., 3778 # a1 = phi <a2, a0> 3779 # b1 = phi <b2, b0> 3780 a2 = operation (a1) 3781 b2 = operation (b1) */ 3782 slp_reduc = (slp_node && !GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt))); 3783 3784 /* In case of reduction chain, e.g., 3785 # a1 = phi <a3, a0> 3786 a2 = operation (a1) 3787 a3 = operation (a2), 3788 3789 we may end up with more than one vector result. Here we reduce them to 3790 one vector. */ 3791 if (GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt))) 3792 { 3793 tree first_vect = PHI_RESULT (VEC_index (gimple, new_phis, 0)); 3794 tree tmp; 3795 gimple new_vec_stmt = NULL; 3796 3797 vec_dest = vect_create_destination_var (scalar_dest, vectype); 3798 for (k = 1; k < VEC_length (gimple, new_phis); k++) 3799 { 3800 gimple next_phi = VEC_index (gimple, new_phis, k); 3801 tree second_vect = PHI_RESULT (next_phi); 3802 3803 tmp = build2 (code, vectype, first_vect, second_vect); 3804 new_vec_stmt = gimple_build_assign (vec_dest, tmp); 3805 first_vect = make_ssa_name (vec_dest, new_vec_stmt); 3806 gimple_assign_set_lhs (new_vec_stmt, first_vect); 3807 gsi_insert_before (&exit_gsi, new_vec_stmt, GSI_SAME_STMT); 3808 } 3809 3810 new_phi_result = first_vect; 3811 if (new_vec_stmt) 3812 { 3813 VEC_truncate (gimple, new_phis, 0); 3814 VEC_safe_push (gimple, heap, new_phis, new_vec_stmt); 3815 } 3816 } 3817 else 3818 new_phi_result = PHI_RESULT (VEC_index (gimple, new_phis, 0)); 3819 3820 /* 2.3 Create the reduction code, using one of the three schemes described 3821 above. In SLP we simply need to extract all the elements from the 3822 vector (without reducing them), so we use scalar shifts. */ 3823 if (reduc_code != ERROR_MARK && !slp_reduc) 3824 { 3825 tree tmp; 3826 3827 /*** Case 1: Create: 3828 v_out2 = reduc_expr <v_out1> */ 3829 3830 if (vect_print_dump_info (REPORT_DETAILS)) 3831 fprintf (vect_dump, "Reduce using direct vector reduction."); 3832 3833 vec_dest = vect_create_destination_var (scalar_dest, vectype); 3834 tmp = build1 (reduc_code, vectype, new_phi_result); 3835 epilog_stmt = gimple_build_assign (vec_dest, tmp); 3836 new_temp = make_ssa_name (vec_dest, epilog_stmt); 3837 gimple_assign_set_lhs (epilog_stmt, new_temp); 3838 gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); 3839 3840 extract_scalar_result = true; 3841 } 3842 else 3843 { 3844 enum tree_code shift_code = ERROR_MARK; 3845 bool have_whole_vector_shift = true; 3846 int bit_offset; 3847 int element_bitsize = tree_low_cst (bitsize, 1); 3848 int vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1); 3849 tree vec_temp; 3850 3851 if (optab_handler (vec_shr_optab, mode) != CODE_FOR_nothing) 3852 shift_code = VEC_RSHIFT_EXPR; 3853 else 3854 have_whole_vector_shift = false; 3855 3856 /* Regardless of whether we have a whole vector shift, if we're 3857 emulating the operation via tree-vect-generic, we don't want 3858 to use it. Only the first round of the reduction is likely 3859 to still be profitable via emulation. */ 3860 /* ??? It might be better to emit a reduction tree code here, so that 3861 tree-vect-generic can expand the first round via bit tricks. */ 3862 if (!VECTOR_MODE_P (mode)) 3863 have_whole_vector_shift = false; 3864 else 3865 { 3866 optab optab = optab_for_tree_code (code, vectype, optab_default); 3867 if (optab_handler (optab, mode) == CODE_FOR_nothing) 3868 have_whole_vector_shift = false; 3869 } 3870 3871 if (have_whole_vector_shift && !slp_reduc) 3872 { 3873 /*** Case 2: Create: 3874 for (offset = VS/2; offset >= element_size; offset/=2) 3875 { 3876 Create: va' = vec_shift <va, offset> 3877 Create: va = vop <va, va'> 3878 } */ 3879 3880 if (vect_print_dump_info (REPORT_DETAILS)) 3881 fprintf (vect_dump, "Reduce using vector shifts"); 3882 3883 vec_dest = vect_create_destination_var (scalar_dest, vectype); 3884 new_temp = new_phi_result; 3885 for (bit_offset = vec_size_in_bits/2; 3886 bit_offset >= element_bitsize; 3887 bit_offset /= 2) 3888 { 3889 tree bitpos = size_int (bit_offset); 3890 3891 epilog_stmt = gimple_build_assign_with_ops (shift_code, 3892 vec_dest, new_temp, bitpos); 3893 new_name = make_ssa_name (vec_dest, epilog_stmt); 3894 gimple_assign_set_lhs (epilog_stmt, new_name); 3895 gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); 3896 3897 epilog_stmt = gimple_build_assign_with_ops (code, vec_dest, 3898 new_name, new_temp); 3899 new_temp = make_ssa_name (vec_dest, epilog_stmt); 3900 gimple_assign_set_lhs (epilog_stmt, new_temp); 3901 gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); 3902 } 3903 3904 extract_scalar_result = true; 3905 } 3906 else 3907 { 3908 tree rhs; 3909 3910 /*** Case 3: Create: 3911 s = extract_field <v_out2, 0> 3912 for (offset = element_size; 3913 offset < vector_size; 3914 offset += element_size;) 3915 { 3916 Create: s' = extract_field <v_out2, offset> 3917 Create: s = op <s, s'> // For non SLP cases 3918 } */ 3919 3920 if (vect_print_dump_info (REPORT_DETAILS)) 3921 fprintf (vect_dump, "Reduce using scalar code. "); 3922 3923 vec_size_in_bits = tree_low_cst (TYPE_SIZE (vectype), 1); 3924 FOR_EACH_VEC_ELT (gimple, new_phis, i, new_phi) 3925 { 3926 if (gimple_code (new_phi) == GIMPLE_PHI) 3927 vec_temp = PHI_RESULT (new_phi); 3928 else 3929 vec_temp = gimple_assign_lhs (new_phi); 3930 rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, bitsize, 3931 bitsize_zero_node); 3932 epilog_stmt = gimple_build_assign (new_scalar_dest, rhs); 3933 new_temp = make_ssa_name (new_scalar_dest, epilog_stmt); 3934 gimple_assign_set_lhs (epilog_stmt, new_temp); 3935 gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); 3936 3937 /* In SLP we don't need to apply reduction operation, so we just 3938 collect s' values in SCALAR_RESULTS. */ 3939 if (slp_reduc) 3940 VEC_safe_push (tree, heap, scalar_results, new_temp); 3941 3942 for (bit_offset = element_bitsize; 3943 bit_offset < vec_size_in_bits; 3944 bit_offset += element_bitsize) 3945 { 3946 tree bitpos = bitsize_int (bit_offset); 3947 tree rhs = build3 (BIT_FIELD_REF, scalar_type, vec_temp, 3948 bitsize, bitpos); 3949 3950 epilog_stmt = gimple_build_assign (new_scalar_dest, rhs); 3951 new_name = make_ssa_name (new_scalar_dest, epilog_stmt); 3952 gimple_assign_set_lhs (epilog_stmt, new_name); 3953 gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); 3954 3955 if (slp_reduc) 3956 { 3957 /* In SLP we don't need to apply reduction operation, so 3958 we just collect s' values in SCALAR_RESULTS. */ 3959 new_temp = new_name; 3960 VEC_safe_push (tree, heap, scalar_results, new_name); 3961 } 3962 else 3963 { 3964 epilog_stmt = gimple_build_assign_with_ops (code, 3965 new_scalar_dest, new_name, new_temp); 3966 new_temp = make_ssa_name (new_scalar_dest, epilog_stmt); 3967 gimple_assign_set_lhs (epilog_stmt, new_temp); 3968 gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); 3969 } 3970 } 3971 } 3972 3973 /* The only case where we need to reduce scalar results in SLP, is 3974 unrolling. If the size of SCALAR_RESULTS is greater than 3975 GROUP_SIZE, we reduce them combining elements modulo 3976 GROUP_SIZE. */ 3977 if (slp_reduc) 3978 { 3979 tree res, first_res, new_res; 3980 gimple new_stmt; 3981 3982 /* Reduce multiple scalar results in case of SLP unrolling. */ 3983 for (j = group_size; VEC_iterate (tree, scalar_results, j, res); 3984 j++) 3985 { 3986 first_res = VEC_index (tree, scalar_results, j % group_size); 3987 new_stmt = gimple_build_assign_with_ops (code, 3988 new_scalar_dest, first_res, res); 3989 new_res = make_ssa_name (new_scalar_dest, new_stmt); 3990 gimple_assign_set_lhs (new_stmt, new_res); 3991 gsi_insert_before (&exit_gsi, new_stmt, GSI_SAME_STMT); 3992 VEC_replace (tree, scalar_results, j % group_size, new_res); 3993 } 3994 } 3995 else 3996 /* Not SLP - we have one scalar to keep in SCALAR_RESULTS. */ 3997 VEC_safe_push (tree, heap, scalar_results, new_temp); 3998 3999 extract_scalar_result = false; 4000 } 4001 } 4002 4003 /* 2.4 Extract the final scalar result. Create: 4004 s_out3 = extract_field <v_out2, bitpos> */ 4005 4006 if (extract_scalar_result) 4007 { 4008 tree rhs; 4009 4010 if (vect_print_dump_info (REPORT_DETAILS)) 4011 fprintf (vect_dump, "extract scalar result"); 4012 4013 if (BYTES_BIG_ENDIAN) 4014 bitpos = size_binop (MULT_EXPR, 4015 bitsize_int (TYPE_VECTOR_SUBPARTS (vectype) - 1), 4016 TYPE_SIZE (scalar_type)); 4017 else 4018 bitpos = bitsize_zero_node; 4019 4020 rhs = build3 (BIT_FIELD_REF, scalar_type, new_temp, bitsize, bitpos); 4021 epilog_stmt = gimple_build_assign (new_scalar_dest, rhs); 4022 new_temp = make_ssa_name (new_scalar_dest, epilog_stmt); 4023 gimple_assign_set_lhs (epilog_stmt, new_temp); 4024 gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); 4025 VEC_safe_push (tree, heap, scalar_results, new_temp); 4026 } 4027 4028 vect_finalize_reduction: 4029 4030 if (double_reduc) 4031 loop = loop->inner; 4032 4033 /* 2.5 Adjust the final result by the initial value of the reduction 4034 variable. (When such adjustment is not needed, then 4035 'adjustment_def' is zero). For example, if code is PLUS we create: 4036 new_temp = loop_exit_def + adjustment_def */ 4037 4038 if (adjustment_def) 4039 { 4040 gcc_assert (!slp_reduc); 4041 if (nested_in_vect_loop) 4042 { 4043 new_phi = VEC_index (gimple, new_phis, 0); 4044 gcc_assert (TREE_CODE (TREE_TYPE (adjustment_def)) == VECTOR_TYPE); 4045 expr = build2 (code, vectype, PHI_RESULT (new_phi), adjustment_def); 4046 new_dest = vect_create_destination_var (scalar_dest, vectype); 4047 } 4048 else 4049 { 4050 new_temp = VEC_index (tree, scalar_results, 0); 4051 gcc_assert (TREE_CODE (TREE_TYPE (adjustment_def)) != VECTOR_TYPE); 4052 expr = build2 (code, scalar_type, new_temp, adjustment_def); 4053 new_dest = vect_create_destination_var (scalar_dest, scalar_type); 4054 } 4055 4056 epilog_stmt = gimple_build_assign (new_dest, expr); 4057 new_temp = make_ssa_name (new_dest, epilog_stmt); 4058 gimple_assign_set_lhs (epilog_stmt, new_temp); 4059 SSA_NAME_DEF_STMT (new_temp) = epilog_stmt; 4060 gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT); 4061 if (nested_in_vect_loop) 4062 { 4063 set_vinfo_for_stmt (epilog_stmt, 4064 new_stmt_vec_info (epilog_stmt, loop_vinfo, 4065 NULL)); 4066 STMT_VINFO_RELATED_STMT (vinfo_for_stmt (epilog_stmt)) = 4067 STMT_VINFO_RELATED_STMT (vinfo_for_stmt (new_phi)); 4068 4069 if (!double_reduc) 4070 VEC_quick_push (tree, scalar_results, new_temp); 4071 else 4072 VEC_replace (tree, scalar_results, 0, new_temp); 4073 } 4074 else 4075 VEC_replace (tree, scalar_results, 0, new_temp); 4076 4077 VEC_replace (gimple, new_phis, 0, epilog_stmt); 4078 } 4079 4080 /* 2.6 Handle the loop-exit phis. Replace the uses of scalar loop-exit 4081 phis with new adjusted scalar results, i.e., replace use <s_out0> 4082 with use <s_out4>. 4083 4084 Transform: 4085 loop_exit: 4086 s_out0 = phi <s_loop> # (scalar) EXIT_PHI 4087 v_out1 = phi <VECT_DEF> # NEW_EXIT_PHI 4088 v_out2 = reduce <v_out1> 4089 s_out3 = extract_field <v_out2, 0> 4090 s_out4 = adjust_result <s_out3> 4091 use <s_out0> 4092 use <s_out0> 4093 4094 into: 4095 4096 loop_exit: 4097 s_out0 = phi <s_loop> # (scalar) EXIT_PHI 4098 v_out1 = phi <VECT_DEF> # NEW_EXIT_PHI 4099 v_out2 = reduce <v_out1> 4100 s_out3 = extract_field <v_out2, 0> 4101 s_out4 = adjust_result <s_out3> 4102 use <s_out4> 4103 use <s_out4> */ 4104 4105 4106 /* In SLP reduction chain we reduce vector results into one vector if 4107 necessary, hence we set here GROUP_SIZE to 1. SCALAR_DEST is the LHS of 4108 the last stmt in the reduction chain, since we are looking for the loop 4109 exit phi node. */ 4110 if (GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt))) 4111 { 4112 scalar_dest = gimple_assign_lhs (VEC_index (gimple, 4113 SLP_TREE_SCALAR_STMTS (slp_node), 4114 group_size - 1)); 4115 group_size = 1; 4116 } 4117 4118 /* In SLP we may have several statements in NEW_PHIS and REDUCTION_PHIS (in 4119 case that GROUP_SIZE is greater than vectorization factor). Therefore, we 4120 need to match SCALAR_RESULTS with corresponding statements. The first 4121 (GROUP_SIZE / number of new vector stmts) scalar results correspond to 4122 the first vector stmt, etc. 4123 (RATIO is equal to (GROUP_SIZE / number of new vector stmts)). */ 4124 if (group_size > VEC_length (gimple, new_phis)) 4125 { 4126 ratio = group_size / VEC_length (gimple, new_phis); 4127 gcc_assert (!(group_size % VEC_length (gimple, new_phis))); 4128 } 4129 else 4130 ratio = 1; 4131 4132 for (k = 0; k < group_size; k++) 4133 { 4134 if (k % ratio == 0) 4135 { 4136 epilog_stmt = VEC_index (gimple, new_phis, k / ratio); 4137 reduction_phi = VEC_index (gimple, reduction_phis, k / ratio); 4138 if (double_reduc) 4139 inner_phi = VEC_index (gimple, inner_phis, k / ratio); 4140 } 4141 4142 if (slp_reduc) 4143 { 4144 gimple current_stmt = VEC_index (gimple, 4145 SLP_TREE_SCALAR_STMTS (slp_node), k); 4146 4147 orig_stmt = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (current_stmt)); 4148 /* SLP statements can't participate in patterns. */ 4149 gcc_assert (!orig_stmt); 4150 scalar_dest = gimple_assign_lhs (current_stmt); 4151 } 4152 4153 phis = VEC_alloc (gimple, heap, 3); 4154 /* Find the loop-closed-use at the loop exit of the original scalar 4155 result. (The reduction result is expected to have two immediate uses - 4156 one at the latch block, and one at the loop exit). */ 4157 FOR_EACH_IMM_USE_FAST (use_p, imm_iter, scalar_dest) 4158 if (!flow_bb_inside_loop_p (loop, gimple_bb (USE_STMT (use_p)))) 4159 VEC_safe_push (gimple, heap, phis, USE_STMT (use_p)); 4160 4161 /* We expect to have found an exit_phi because of loop-closed-ssa 4162 form. */ 4163 gcc_assert (!VEC_empty (gimple, phis)); 4164 4165 FOR_EACH_VEC_ELT (gimple, phis, i, exit_phi) 4166 { 4167 if (outer_loop) 4168 { 4169 stmt_vec_info exit_phi_vinfo = vinfo_for_stmt (exit_phi); 4170 gimple vect_phi; 4171 4172 /* FORNOW. Currently not supporting the case that an inner-loop 4173 reduction is not used in the outer-loop (but only outside the 4174 outer-loop), unless it is double reduction. */ 4175 gcc_assert ((STMT_VINFO_RELEVANT_P (exit_phi_vinfo) 4176 && !STMT_VINFO_LIVE_P (exit_phi_vinfo)) 4177 || double_reduc); 4178 4179 STMT_VINFO_VEC_STMT (exit_phi_vinfo) = epilog_stmt; 4180 if (!double_reduc 4181 || STMT_VINFO_DEF_TYPE (exit_phi_vinfo) 4182 != vect_double_reduction_def) 4183 continue; 4184 4185 /* Handle double reduction: 4186 4187 stmt1: s1 = phi <s0, s2> - double reduction phi (outer loop) 4188 stmt2: s3 = phi <s1, s4> - (regular) reduc phi (inner loop) 4189 stmt3: s4 = use (s3) - (regular) reduc stmt (inner loop) 4190 stmt4: s2 = phi <s4> - double reduction stmt (outer loop) 4191 4192 At that point the regular reduction (stmt2 and stmt3) is 4193 already vectorized, as well as the exit phi node, stmt4. 4194 Here we vectorize the phi node of double reduction, stmt1, and 4195 update all relevant statements. */ 4196 4197 /* Go through all the uses of s2 to find double reduction phi 4198 node, i.e., stmt1 above. */ 4199 orig_name = PHI_RESULT (exit_phi); 4200 FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, orig_name) 4201 { 4202 stmt_vec_info use_stmt_vinfo = vinfo_for_stmt (use_stmt); 4203 stmt_vec_info new_phi_vinfo; 4204 tree vect_phi_init, preheader_arg, vect_phi_res, init_def; 4205 basic_block bb = gimple_bb (use_stmt); 4206 gimple use; 4207 4208 /* Check that USE_STMT is really double reduction phi 4209 node. */ 4210 if (gimple_code (use_stmt) != GIMPLE_PHI 4211 || gimple_phi_num_args (use_stmt) != 2 4212 || !use_stmt_vinfo 4213 || STMT_VINFO_DEF_TYPE (use_stmt_vinfo) 4214 != vect_double_reduction_def 4215 || bb->loop_father != outer_loop) 4216 continue; 4217 4218 /* Create vector phi node for double reduction: 4219 vs1 = phi <vs0, vs2> 4220 vs1 was created previously in this function by a call to 4221 vect_get_vec_def_for_operand and is stored in 4222 vec_initial_def; 4223 vs2 is defined by INNER_PHI, the vectorized EXIT_PHI; 4224 vs0 is created here. */ 4225 4226 /* Create vector phi node. */ 4227 vect_phi = create_phi_node (vec_initial_def, bb); 4228 new_phi_vinfo = new_stmt_vec_info (vect_phi, 4229 loop_vec_info_for_loop (outer_loop), NULL); 4230 set_vinfo_for_stmt (vect_phi, new_phi_vinfo); 4231 4232 /* Create vs0 - initial def of the double reduction phi. */ 4233 preheader_arg = PHI_ARG_DEF_FROM_EDGE (use_stmt, 4234 loop_preheader_edge (outer_loop)); 4235 init_def = get_initial_def_for_reduction (stmt, 4236 preheader_arg, NULL); 4237 vect_phi_init = vect_init_vector (use_stmt, init_def, 4238 vectype, NULL); 4239 4240 /* Update phi node arguments with vs0 and vs2. */ 4241 add_phi_arg (vect_phi, vect_phi_init, 4242 loop_preheader_edge (outer_loop), 4243 UNKNOWN_LOCATION); 4244 add_phi_arg (vect_phi, PHI_RESULT (inner_phi), 4245 loop_latch_edge (outer_loop), UNKNOWN_LOCATION); 4246 if (vect_print_dump_info (REPORT_DETAILS)) 4247 { 4248 fprintf (vect_dump, "created double reduction phi " 4249 "node: "); 4250 print_gimple_stmt (vect_dump, vect_phi, 0, TDF_SLIM); 4251 } 4252 4253 vect_phi_res = PHI_RESULT (vect_phi); 4254 4255 /* Replace the use, i.e., set the correct vs1 in the regular 4256 reduction phi node. FORNOW, NCOPIES is always 1, so the 4257 loop is redundant. */ 4258 use = reduction_phi; 4259 for (j = 0; j < ncopies; j++) 4260 { 4261 edge pr_edge = loop_preheader_edge (loop); 4262 SET_PHI_ARG_DEF (use, pr_edge->dest_idx, vect_phi_res); 4263 use = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (use)); 4264 } 4265 } 4266 } 4267 } 4268 4269 VEC_free (gimple, heap, phis); 4270 if (nested_in_vect_loop) 4271 { 4272 if (double_reduc) 4273 loop = outer_loop; 4274 else 4275 continue; 4276 } 4277 4278 phis = VEC_alloc (gimple, heap, 3); 4279 /* Find the loop-closed-use at the loop exit of the original scalar 4280 result. (The reduction result is expected to have two immediate uses, 4281 one at the latch block, and one at the loop exit). For double 4282 reductions we are looking for exit phis of the outer loop. */ 4283 FOR_EACH_IMM_USE_FAST (use_p, imm_iter, scalar_dest) 4284 { 4285 if (!flow_bb_inside_loop_p (loop, gimple_bb (USE_STMT (use_p)))) 4286 VEC_safe_push (gimple, heap, phis, USE_STMT (use_p)); 4287 else 4288 { 4289 if (double_reduc && gimple_code (USE_STMT (use_p)) == GIMPLE_PHI) 4290 { 4291 tree phi_res = PHI_RESULT (USE_STMT (use_p)); 4292 4293 FOR_EACH_IMM_USE_FAST (phi_use_p, phi_imm_iter, phi_res) 4294 { 4295 if (!flow_bb_inside_loop_p (loop, 4296 gimple_bb (USE_STMT (phi_use_p)))) 4297 VEC_safe_push (gimple, heap, phis, 4298 USE_STMT (phi_use_p)); 4299 } 4300 } 4301 } 4302 } 4303 4304 FOR_EACH_VEC_ELT (gimple, phis, i, exit_phi) 4305 { 4306 /* Replace the uses: */ 4307 orig_name = PHI_RESULT (exit_phi); 4308 scalar_result = VEC_index (tree, scalar_results, k); 4309 FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, orig_name) 4310 FOR_EACH_IMM_USE_ON_STMT (use_p, imm_iter) 4311 SET_USE (use_p, scalar_result); 4312 } 4313 4314 VEC_free (gimple, heap, phis); 4315 } 4316 4317 VEC_free (tree, heap, scalar_results); 4318 VEC_free (gimple, heap, new_phis); 4319 } 4320 4321 4322 /* Function vectorizable_reduction. 4323 4324 Check if STMT performs a reduction operation that can be vectorized. 4325 If VEC_STMT is also passed, vectorize the STMT: create a vectorized 4326 stmt to replace it, put it in VEC_STMT, and insert it at GSI. 4327 Return FALSE if not a vectorizable STMT, TRUE otherwise. 4328 4329 This function also handles reduction idioms (patterns) that have been 4330 recognized in advance during vect_pattern_recog. In this case, STMT may be 4331 of this form: 4332 X = pattern_expr (arg0, arg1, ..., X) 4333 and it's STMT_VINFO_RELATED_STMT points to the last stmt in the original 4334 sequence that had been detected and replaced by the pattern-stmt (STMT). 4335 4336 In some cases of reduction patterns, the type of the reduction variable X is 4337 different than the type of the other arguments of STMT. 4338 In such cases, the vectype that is used when transforming STMT into a vector 4339 stmt is different than the vectype that is used to determine the 4340 vectorization factor, because it consists of a different number of elements 4341 than the actual number of elements that are being operated upon in parallel. 4342 4343 For example, consider an accumulation of shorts into an int accumulator. 4344 On some targets it's possible to vectorize this pattern operating on 8 4345 shorts at a time (hence, the vectype for purposes of determining the 4346 vectorization factor should be V8HI); on the other hand, the vectype that 4347 is used to create the vector form is actually V4SI (the type of the result). 4348 4349 Upon entry to this function, STMT_VINFO_VECTYPE records the vectype that 4350 indicates what is the actual level of parallelism (V8HI in the example), so 4351 that the right vectorization factor would be derived. This vectype 4352 corresponds to the type of arguments to the reduction stmt, and should *NOT* 4353 be used to create the vectorized stmt. The right vectype for the vectorized 4354 stmt is obtained from the type of the result X: 4355 get_vectype_for_scalar_type (TREE_TYPE (X)) 4356 4357 This means that, contrary to "regular" reductions (or "regular" stmts in 4358 general), the following equation: 4359 STMT_VINFO_VECTYPE == get_vectype_for_scalar_type (TREE_TYPE (X)) 4360 does *NOT* necessarily hold for reduction patterns. */ 4361 4362 bool 4363 vectorizable_reduction (gimple stmt, gimple_stmt_iterator *gsi, 4364 gimple *vec_stmt, slp_tree slp_node) 4365 { 4366 tree vec_dest; 4367 tree scalar_dest; 4368 tree loop_vec_def0 = NULL_TREE, loop_vec_def1 = NULL_TREE; 4369 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 4370 tree vectype_out = STMT_VINFO_VECTYPE (stmt_info); 4371 tree vectype_in = NULL_TREE; 4372 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 4373 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 4374 enum tree_code code, orig_code, epilog_reduc_code; 4375 enum machine_mode vec_mode; 4376 int op_type; 4377 optab optab, reduc_optab; 4378 tree new_temp = NULL_TREE; 4379 tree def; 4380 gimple def_stmt; 4381 enum vect_def_type dt; 4382 gimple new_phi = NULL; 4383 tree scalar_type; 4384 bool is_simple_use; 4385 gimple orig_stmt; 4386 stmt_vec_info orig_stmt_info; 4387 tree expr = NULL_TREE; 4388 int i; 4389 int ncopies; 4390 int epilog_copies; 4391 stmt_vec_info prev_stmt_info, prev_phi_info; 4392 bool single_defuse_cycle = false; 4393 tree reduc_def = NULL_TREE; 4394 gimple new_stmt = NULL; 4395 int j; 4396 tree ops[3]; 4397 bool nested_cycle = false, found_nested_cycle_def = false; 4398 gimple reduc_def_stmt = NULL; 4399 /* The default is that the reduction variable is the last in statement. */ 4400 int reduc_index = 2; 4401 bool double_reduc = false, dummy; 4402 basic_block def_bb; 4403 struct loop * def_stmt_loop, *outer_loop = NULL; 4404 tree def_arg; 4405 gimple def_arg_stmt; 4406 VEC (tree, heap) *vec_oprnds0 = NULL, *vec_oprnds1 = NULL, *vect_defs = NULL; 4407 VEC (gimple, heap) *phis = NULL; 4408 int vec_num; 4409 tree def0, def1, tem, op0, op1 = NULL_TREE; 4410 4411 /* In case of reduction chain we switch to the first stmt in the chain, but 4412 we don't update STMT_INFO, since only the last stmt is marked as reduction 4413 and has reduction properties. */ 4414 if (GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt))) 4415 stmt = GROUP_FIRST_ELEMENT (stmt_info); 4416 4417 if (nested_in_vect_loop_p (loop, stmt)) 4418 { 4419 outer_loop = loop; 4420 loop = loop->inner; 4421 nested_cycle = true; 4422 } 4423 4424 /* 1. Is vectorizable reduction? */ 4425 /* Not supportable if the reduction variable is used in the loop, unless 4426 it's a reduction chain. */ 4427 if (STMT_VINFO_RELEVANT (stmt_info) > vect_used_in_outer 4428 && !GROUP_FIRST_ELEMENT (stmt_info)) 4429 return false; 4430 4431 /* Reductions that are not used even in an enclosing outer-loop, 4432 are expected to be "live" (used out of the loop). */ 4433 if (STMT_VINFO_RELEVANT (stmt_info) == vect_unused_in_scope 4434 && !STMT_VINFO_LIVE_P (stmt_info)) 4435 return false; 4436 4437 /* Make sure it was already recognized as a reduction computation. */ 4438 if (STMT_VINFO_DEF_TYPE (stmt_info) != vect_reduction_def 4439 && STMT_VINFO_DEF_TYPE (stmt_info) != vect_nested_cycle) 4440 return false; 4441 4442 /* 2. Has this been recognized as a reduction pattern? 4443 4444 Check if STMT represents a pattern that has been recognized 4445 in earlier analysis stages. For stmts that represent a pattern, 4446 the STMT_VINFO_RELATED_STMT field records the last stmt in 4447 the original sequence that constitutes the pattern. */ 4448 4449 orig_stmt = STMT_VINFO_RELATED_STMT (stmt_info); 4450 if (orig_stmt) 4451 { 4452 orig_stmt_info = vinfo_for_stmt (orig_stmt); 4453 gcc_assert (STMT_VINFO_RELATED_STMT (orig_stmt_info) == stmt); 4454 gcc_assert (STMT_VINFO_IN_PATTERN_P (orig_stmt_info)); 4455 gcc_assert (!STMT_VINFO_IN_PATTERN_P (stmt_info)); 4456 } 4457 4458 /* 3. Check the operands of the operation. The first operands are defined 4459 inside the loop body. The last operand is the reduction variable, 4460 which is defined by the loop-header-phi. */ 4461 4462 gcc_assert (is_gimple_assign (stmt)); 4463 4464 /* Flatten RHS. */ 4465 switch (get_gimple_rhs_class (gimple_assign_rhs_code (stmt))) 4466 { 4467 case GIMPLE_SINGLE_RHS: 4468 op_type = TREE_OPERAND_LENGTH (gimple_assign_rhs1 (stmt)); 4469 if (op_type == ternary_op) 4470 { 4471 tree rhs = gimple_assign_rhs1 (stmt); 4472 ops[0] = TREE_OPERAND (rhs, 0); 4473 ops[1] = TREE_OPERAND (rhs, 1); 4474 ops[2] = TREE_OPERAND (rhs, 2); 4475 code = TREE_CODE (rhs); 4476 } 4477 else 4478 return false; 4479 break; 4480 4481 case GIMPLE_BINARY_RHS: 4482 code = gimple_assign_rhs_code (stmt); 4483 op_type = TREE_CODE_LENGTH (code); 4484 gcc_assert (op_type == binary_op); 4485 ops[0] = gimple_assign_rhs1 (stmt); 4486 ops[1] = gimple_assign_rhs2 (stmt); 4487 break; 4488 4489 case GIMPLE_TERNARY_RHS: 4490 code = gimple_assign_rhs_code (stmt); 4491 op_type = TREE_CODE_LENGTH (code); 4492 gcc_assert (op_type == ternary_op); 4493 ops[0] = gimple_assign_rhs1 (stmt); 4494 ops[1] = gimple_assign_rhs2 (stmt); 4495 ops[2] = gimple_assign_rhs3 (stmt); 4496 break; 4497 4498 case GIMPLE_UNARY_RHS: 4499 return false; 4500 4501 default: 4502 gcc_unreachable (); 4503 } 4504 4505 if (code == COND_EXPR && slp_node) 4506 return false; 4507 4508 scalar_dest = gimple_assign_lhs (stmt); 4509 scalar_type = TREE_TYPE (scalar_dest); 4510 if (!POINTER_TYPE_P (scalar_type) && !INTEGRAL_TYPE_P (scalar_type) 4511 && !SCALAR_FLOAT_TYPE_P (scalar_type)) 4512 return false; 4513 4514 /* Do not try to vectorize bit-precision reductions. */ 4515 if ((TYPE_PRECISION (scalar_type) 4516 != GET_MODE_PRECISION (TYPE_MODE (scalar_type)))) 4517 return false; 4518 4519 /* All uses but the last are expected to be defined in the loop. 4520 The last use is the reduction variable. In case of nested cycle this 4521 assumption is not true: we use reduc_index to record the index of the 4522 reduction variable. */ 4523 for (i = 0; i < op_type-1; i++) 4524 { 4525 /* The condition of COND_EXPR is checked in vectorizable_condition(). */ 4526 if (i == 0 && code == COND_EXPR) 4527 continue; 4528 4529 is_simple_use = vect_is_simple_use_1 (ops[i], stmt, loop_vinfo, NULL, 4530 &def_stmt, &def, &dt, &tem); 4531 if (!vectype_in) 4532 vectype_in = tem; 4533 gcc_assert (is_simple_use); 4534 4535 if (dt != vect_internal_def 4536 && dt != vect_external_def 4537 && dt != vect_constant_def 4538 && dt != vect_induction_def 4539 && !(dt == vect_nested_cycle && nested_cycle)) 4540 return false; 4541 4542 if (dt == vect_nested_cycle) 4543 { 4544 found_nested_cycle_def = true; 4545 reduc_def_stmt = def_stmt; 4546 reduc_index = i; 4547 } 4548 } 4549 4550 is_simple_use = vect_is_simple_use_1 (ops[i], stmt, loop_vinfo, NULL, 4551 &def_stmt, &def, &dt, &tem); 4552 if (!vectype_in) 4553 vectype_in = tem; 4554 gcc_assert (is_simple_use); 4555 gcc_assert (dt == vect_reduction_def 4556 || dt == vect_nested_cycle 4557 || ((dt == vect_internal_def || dt == vect_external_def 4558 || dt == vect_constant_def || dt == vect_induction_def) 4559 && nested_cycle && found_nested_cycle_def)); 4560 if (!found_nested_cycle_def) 4561 reduc_def_stmt = def_stmt; 4562 4563 gcc_assert (gimple_code (reduc_def_stmt) == GIMPLE_PHI); 4564 if (orig_stmt) 4565 gcc_assert (orig_stmt == vect_is_simple_reduction (loop_vinfo, 4566 reduc_def_stmt, 4567 !nested_cycle, 4568 &dummy)); 4569 else 4570 { 4571 gimple tmp = vect_is_simple_reduction (loop_vinfo, reduc_def_stmt, 4572 !nested_cycle, &dummy); 4573 /* We changed STMT to be the first stmt in reduction chain, hence we 4574 check that in this case the first element in the chain is STMT. */ 4575 gcc_assert (stmt == tmp 4576 || GROUP_FIRST_ELEMENT (vinfo_for_stmt (tmp)) == stmt); 4577 } 4578 4579 if (STMT_VINFO_LIVE_P (vinfo_for_stmt (reduc_def_stmt))) 4580 return false; 4581 4582 if (slp_node || PURE_SLP_STMT (stmt_info)) 4583 ncopies = 1; 4584 else 4585 ncopies = (LOOP_VINFO_VECT_FACTOR (loop_vinfo) 4586 / TYPE_VECTOR_SUBPARTS (vectype_in)); 4587 4588 gcc_assert (ncopies >= 1); 4589 4590 vec_mode = TYPE_MODE (vectype_in); 4591 4592 if (code == COND_EXPR) 4593 { 4594 if (!vectorizable_condition (stmt, gsi, NULL, ops[reduc_index], 0, NULL)) 4595 { 4596 if (vect_print_dump_info (REPORT_DETAILS)) 4597 fprintf (vect_dump, "unsupported condition in reduction"); 4598 4599 return false; 4600 } 4601 } 4602 else 4603 { 4604 /* 4. Supportable by target? */ 4605 4606 /* 4.1. check support for the operation in the loop */ 4607 optab = optab_for_tree_code (code, vectype_in, optab_default); 4608 if (!optab) 4609 { 4610 if (vect_print_dump_info (REPORT_DETAILS)) 4611 fprintf (vect_dump, "no optab."); 4612 4613 return false; 4614 } 4615 4616 if (optab_handler (optab, vec_mode) == CODE_FOR_nothing) 4617 { 4618 if (vect_print_dump_info (REPORT_DETAILS)) 4619 fprintf (vect_dump, "op not supported by target."); 4620 4621 if (GET_MODE_SIZE (vec_mode) != UNITS_PER_WORD 4622 || LOOP_VINFO_VECT_FACTOR (loop_vinfo) 4623 < vect_min_worthwhile_factor (code)) 4624 return false; 4625 4626 if (vect_print_dump_info (REPORT_DETAILS)) 4627 fprintf (vect_dump, "proceeding using word mode."); 4628 } 4629 4630 /* Worthwhile without SIMD support? */ 4631 if (!VECTOR_MODE_P (TYPE_MODE (vectype_in)) 4632 && LOOP_VINFO_VECT_FACTOR (loop_vinfo) 4633 < vect_min_worthwhile_factor (code)) 4634 { 4635 if (vect_print_dump_info (REPORT_DETAILS)) 4636 fprintf (vect_dump, "not worthwhile without SIMD support."); 4637 4638 return false; 4639 } 4640 } 4641 4642 /* 4.2. Check support for the epilog operation. 4643 4644 If STMT represents a reduction pattern, then the type of the 4645 reduction variable may be different than the type of the rest 4646 of the arguments. For example, consider the case of accumulation 4647 of shorts into an int accumulator; The original code: 4648 S1: int_a = (int) short_a; 4649 orig_stmt-> S2: int_acc = plus <int_a ,int_acc>; 4650 4651 was replaced with: 4652 STMT: int_acc = widen_sum <short_a, int_acc> 4653 4654 This means that: 4655 1. The tree-code that is used to create the vector operation in the 4656 epilog code (that reduces the partial results) is not the 4657 tree-code of STMT, but is rather the tree-code of the original 4658 stmt from the pattern that STMT is replacing. I.e, in the example 4659 above we want to use 'widen_sum' in the loop, but 'plus' in the 4660 epilog. 4661 2. The type (mode) we use to check available target support 4662 for the vector operation to be created in the *epilog*, is 4663 determined by the type of the reduction variable (in the example 4664 above we'd check this: optab_handler (plus_optab, vect_int_mode])). 4665 However the type (mode) we use to check available target support 4666 for the vector operation to be created *inside the loop*, is 4667 determined by the type of the other arguments to STMT (in the 4668 example we'd check this: optab_handler (widen_sum_optab, 4669 vect_short_mode)). 4670 4671 This is contrary to "regular" reductions, in which the types of all 4672 the arguments are the same as the type of the reduction variable. 4673 For "regular" reductions we can therefore use the same vector type 4674 (and also the same tree-code) when generating the epilog code and 4675 when generating the code inside the loop. */ 4676 4677 if (orig_stmt) 4678 { 4679 /* This is a reduction pattern: get the vectype from the type of the 4680 reduction variable, and get the tree-code from orig_stmt. */ 4681 orig_code = gimple_assign_rhs_code (orig_stmt); 4682 gcc_assert (vectype_out); 4683 vec_mode = TYPE_MODE (vectype_out); 4684 } 4685 else 4686 { 4687 /* Regular reduction: use the same vectype and tree-code as used for 4688 the vector code inside the loop can be used for the epilog code. */ 4689 orig_code = code; 4690 } 4691 4692 if (nested_cycle) 4693 { 4694 def_bb = gimple_bb (reduc_def_stmt); 4695 def_stmt_loop = def_bb->loop_father; 4696 def_arg = PHI_ARG_DEF_FROM_EDGE (reduc_def_stmt, 4697 loop_preheader_edge (def_stmt_loop)); 4698 if (TREE_CODE (def_arg) == SSA_NAME 4699 && (def_arg_stmt = SSA_NAME_DEF_STMT (def_arg)) 4700 && gimple_code (def_arg_stmt) == GIMPLE_PHI 4701 && flow_bb_inside_loop_p (outer_loop, gimple_bb (def_arg_stmt)) 4702 && vinfo_for_stmt (def_arg_stmt) 4703 && STMT_VINFO_DEF_TYPE (vinfo_for_stmt (def_arg_stmt)) 4704 == vect_double_reduction_def) 4705 double_reduc = true; 4706 } 4707 4708 epilog_reduc_code = ERROR_MARK; 4709 if (reduction_code_for_scalar_code (orig_code, &epilog_reduc_code)) 4710 { 4711 reduc_optab = optab_for_tree_code (epilog_reduc_code, vectype_out, 4712 optab_default); 4713 if (!reduc_optab) 4714 { 4715 if (vect_print_dump_info (REPORT_DETAILS)) 4716 fprintf (vect_dump, "no optab for reduction."); 4717 4718 epilog_reduc_code = ERROR_MARK; 4719 } 4720 4721 if (reduc_optab 4722 && optab_handler (reduc_optab, vec_mode) == CODE_FOR_nothing) 4723 { 4724 if (vect_print_dump_info (REPORT_DETAILS)) 4725 fprintf (vect_dump, "reduc op not supported by target."); 4726 4727 epilog_reduc_code = ERROR_MARK; 4728 } 4729 } 4730 else 4731 { 4732 if (!nested_cycle || double_reduc) 4733 { 4734 if (vect_print_dump_info (REPORT_DETAILS)) 4735 fprintf (vect_dump, "no reduc code for scalar code."); 4736 4737 return false; 4738 } 4739 } 4740 4741 if (double_reduc && ncopies > 1) 4742 { 4743 if (vect_print_dump_info (REPORT_DETAILS)) 4744 fprintf (vect_dump, "multiple types in double reduction"); 4745 4746 return false; 4747 } 4748 4749 /* In case of widenning multiplication by a constant, we update the type 4750 of the constant to be the type of the other operand. We check that the 4751 constant fits the type in the pattern recognition pass. */ 4752 if (code == DOT_PROD_EXPR 4753 && !types_compatible_p (TREE_TYPE (ops[0]), TREE_TYPE (ops[1]))) 4754 { 4755 if (TREE_CODE (ops[0]) == INTEGER_CST) 4756 ops[0] = fold_convert (TREE_TYPE (ops[1]), ops[0]); 4757 else if (TREE_CODE (ops[1]) == INTEGER_CST) 4758 ops[1] = fold_convert (TREE_TYPE (ops[0]), ops[1]); 4759 else 4760 { 4761 if (vect_print_dump_info (REPORT_DETAILS)) 4762 fprintf (vect_dump, "invalid types in dot-prod"); 4763 4764 return false; 4765 } 4766 } 4767 4768 if (!vec_stmt) /* transformation not required. */ 4769 { 4770 if (!vect_model_reduction_cost (stmt_info, epilog_reduc_code, ncopies)) 4771 return false; 4772 STMT_VINFO_TYPE (stmt_info) = reduc_vec_info_type; 4773 return true; 4774 } 4775 4776 /** Transform. **/ 4777 4778 if (vect_print_dump_info (REPORT_DETAILS)) 4779 fprintf (vect_dump, "transform reduction."); 4780 4781 /* FORNOW: Multiple types are not supported for condition. */ 4782 if (code == COND_EXPR) 4783 gcc_assert (ncopies == 1); 4784 4785 /* Create the destination vector */ 4786 vec_dest = vect_create_destination_var (scalar_dest, vectype_out); 4787 4788 /* In case the vectorization factor (VF) is bigger than the number 4789 of elements that we can fit in a vectype (nunits), we have to generate 4790 more than one vector stmt - i.e - we need to "unroll" the 4791 vector stmt by a factor VF/nunits. For more details see documentation 4792 in vectorizable_operation. */ 4793 4794 /* If the reduction is used in an outer loop we need to generate 4795 VF intermediate results, like so (e.g. for ncopies=2): 4796 r0 = phi (init, r0) 4797 r1 = phi (init, r1) 4798 r0 = x0 + r0; 4799 r1 = x1 + r1; 4800 (i.e. we generate VF results in 2 registers). 4801 In this case we have a separate def-use cycle for each copy, and therefore 4802 for each copy we get the vector def for the reduction variable from the 4803 respective phi node created for this copy. 4804 4805 Otherwise (the reduction is unused in the loop nest), we can combine 4806 together intermediate results, like so (e.g. for ncopies=2): 4807 r = phi (init, r) 4808 r = x0 + r; 4809 r = x1 + r; 4810 (i.e. we generate VF/2 results in a single register). 4811 In this case for each copy we get the vector def for the reduction variable 4812 from the vectorized reduction operation generated in the previous iteration. 4813 */ 4814 4815 if (STMT_VINFO_RELEVANT (stmt_info) == vect_unused_in_scope) 4816 { 4817 single_defuse_cycle = true; 4818 epilog_copies = 1; 4819 } 4820 else 4821 epilog_copies = ncopies; 4822 4823 prev_stmt_info = NULL; 4824 prev_phi_info = NULL; 4825 if (slp_node) 4826 { 4827 vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node); 4828 gcc_assert (TYPE_VECTOR_SUBPARTS (vectype_out) 4829 == TYPE_VECTOR_SUBPARTS (vectype_in)); 4830 } 4831 else 4832 { 4833 vec_num = 1; 4834 vec_oprnds0 = VEC_alloc (tree, heap, 1); 4835 if (op_type == ternary_op) 4836 vec_oprnds1 = VEC_alloc (tree, heap, 1); 4837 } 4838 4839 phis = VEC_alloc (gimple, heap, vec_num); 4840 vect_defs = VEC_alloc (tree, heap, vec_num); 4841 if (!slp_node) 4842 VEC_quick_push (tree, vect_defs, NULL_TREE); 4843 4844 for (j = 0; j < ncopies; j++) 4845 { 4846 if (j == 0 || !single_defuse_cycle) 4847 { 4848 for (i = 0; i < vec_num; i++) 4849 { 4850 /* Create the reduction-phi that defines the reduction 4851 operand. */ 4852 new_phi = create_phi_node (vec_dest, loop->header); 4853 set_vinfo_for_stmt (new_phi, 4854 new_stmt_vec_info (new_phi, loop_vinfo, 4855 NULL)); 4856 if (j == 0 || slp_node) 4857 VEC_quick_push (gimple, phis, new_phi); 4858 } 4859 } 4860 4861 if (code == COND_EXPR) 4862 { 4863 gcc_assert (!slp_node); 4864 vectorizable_condition (stmt, gsi, vec_stmt, 4865 PHI_RESULT (VEC_index (gimple, phis, 0)), 4866 reduc_index, NULL); 4867 /* Multiple types are not supported for condition. */ 4868 break; 4869 } 4870 4871 /* Handle uses. */ 4872 if (j == 0) 4873 { 4874 op0 = ops[!reduc_index]; 4875 if (op_type == ternary_op) 4876 { 4877 if (reduc_index == 0) 4878 op1 = ops[2]; 4879 else 4880 op1 = ops[1]; 4881 } 4882 4883 if (slp_node) 4884 vect_get_vec_defs (op0, op1, stmt, &vec_oprnds0, &vec_oprnds1, 4885 slp_node, -1); 4886 else 4887 { 4888 loop_vec_def0 = vect_get_vec_def_for_operand (ops[!reduc_index], 4889 stmt, NULL); 4890 VEC_quick_push (tree, vec_oprnds0, loop_vec_def0); 4891 if (op_type == ternary_op) 4892 { 4893 loop_vec_def1 = vect_get_vec_def_for_operand (op1, stmt, 4894 NULL); 4895 VEC_quick_push (tree, vec_oprnds1, loop_vec_def1); 4896 } 4897 } 4898 } 4899 else 4900 { 4901 if (!slp_node) 4902 { 4903 enum vect_def_type dt; 4904 gimple dummy_stmt; 4905 tree dummy; 4906 4907 vect_is_simple_use (ops[!reduc_index], stmt, loop_vinfo, NULL, 4908 &dummy_stmt, &dummy, &dt); 4909 loop_vec_def0 = vect_get_vec_def_for_stmt_copy (dt, 4910 loop_vec_def0); 4911 VEC_replace (tree, vec_oprnds0, 0, loop_vec_def0); 4912 if (op_type == ternary_op) 4913 { 4914 vect_is_simple_use (op1, stmt, loop_vinfo, NULL, &dummy_stmt, 4915 &dummy, &dt); 4916 loop_vec_def1 = vect_get_vec_def_for_stmt_copy (dt, 4917 loop_vec_def1); 4918 VEC_replace (tree, vec_oprnds1, 0, loop_vec_def1); 4919 } 4920 } 4921 4922 if (single_defuse_cycle) 4923 reduc_def = gimple_assign_lhs (new_stmt); 4924 4925 STMT_VINFO_RELATED_STMT (prev_phi_info) = new_phi; 4926 } 4927 4928 FOR_EACH_VEC_ELT (tree, vec_oprnds0, i, def0) 4929 { 4930 if (slp_node) 4931 reduc_def = PHI_RESULT (VEC_index (gimple, phis, i)); 4932 else 4933 { 4934 if (!single_defuse_cycle || j == 0) 4935 reduc_def = PHI_RESULT (new_phi); 4936 } 4937 4938 def1 = ((op_type == ternary_op) 4939 ? VEC_index (tree, vec_oprnds1, i) : NULL); 4940 if (op_type == binary_op) 4941 { 4942 if (reduc_index == 0) 4943 expr = build2 (code, vectype_out, reduc_def, def0); 4944 else 4945 expr = build2 (code, vectype_out, def0, reduc_def); 4946 } 4947 else 4948 { 4949 if (reduc_index == 0) 4950 expr = build3 (code, vectype_out, reduc_def, def0, def1); 4951 else 4952 { 4953 if (reduc_index == 1) 4954 expr = build3 (code, vectype_out, def0, reduc_def, def1); 4955 else 4956 expr = build3 (code, vectype_out, def0, def1, reduc_def); 4957 } 4958 } 4959 4960 new_stmt = gimple_build_assign (vec_dest, expr); 4961 new_temp = make_ssa_name (vec_dest, new_stmt); 4962 gimple_assign_set_lhs (new_stmt, new_temp); 4963 vect_finish_stmt_generation (stmt, new_stmt, gsi); 4964 4965 if (slp_node) 4966 { 4967 VEC_quick_push (gimple, SLP_TREE_VEC_STMTS (slp_node), new_stmt); 4968 VEC_quick_push (tree, vect_defs, new_temp); 4969 } 4970 else 4971 VEC_replace (tree, vect_defs, 0, new_temp); 4972 } 4973 4974 if (slp_node) 4975 continue; 4976 4977 if (j == 0) 4978 STMT_VINFO_VEC_STMT (stmt_info) = *vec_stmt = new_stmt; 4979 else 4980 STMT_VINFO_RELATED_STMT (prev_stmt_info) = new_stmt; 4981 4982 prev_stmt_info = vinfo_for_stmt (new_stmt); 4983 prev_phi_info = vinfo_for_stmt (new_phi); 4984 } 4985 4986 /* Finalize the reduction-phi (set its arguments) and create the 4987 epilog reduction code. */ 4988 if ((!single_defuse_cycle || code == COND_EXPR) && !slp_node) 4989 { 4990 new_temp = gimple_assign_lhs (*vec_stmt); 4991 VEC_replace (tree, vect_defs, 0, new_temp); 4992 } 4993 4994 vect_create_epilog_for_reduction (vect_defs, stmt, epilog_copies, 4995 epilog_reduc_code, phis, reduc_index, 4996 double_reduc, slp_node); 4997 4998 VEC_free (gimple, heap, phis); 4999 VEC_free (tree, heap, vec_oprnds0); 5000 if (vec_oprnds1) 5001 VEC_free (tree, heap, vec_oprnds1); 5002 5003 return true; 5004 } 5005 5006 /* Function vect_min_worthwhile_factor. 5007 5008 For a loop where we could vectorize the operation indicated by CODE, 5009 return the minimum vectorization factor that makes it worthwhile 5010 to use generic vectors. */ 5011 int 5012 vect_min_worthwhile_factor (enum tree_code code) 5013 { 5014 switch (code) 5015 { 5016 case PLUS_EXPR: 5017 case MINUS_EXPR: 5018 case NEGATE_EXPR: 5019 return 4; 5020 5021 case BIT_AND_EXPR: 5022 case BIT_IOR_EXPR: 5023 case BIT_XOR_EXPR: 5024 case BIT_NOT_EXPR: 5025 return 2; 5026 5027 default: 5028 return INT_MAX; 5029 } 5030 } 5031 5032 5033 /* Function vectorizable_induction 5034 5035 Check if PHI performs an induction computation that can be vectorized. 5036 If VEC_STMT is also passed, vectorize the induction PHI: create a vectorized 5037 phi to replace it, put it in VEC_STMT, and add it to the same basic block. 5038 Return FALSE if not a vectorizable STMT, TRUE otherwise. */ 5039 5040 bool 5041 vectorizable_induction (gimple phi, gimple_stmt_iterator *gsi ATTRIBUTE_UNUSED, 5042 gimple *vec_stmt) 5043 { 5044 stmt_vec_info stmt_info = vinfo_for_stmt (phi); 5045 tree vectype = STMT_VINFO_VECTYPE (stmt_info); 5046 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 5047 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 5048 int nunits = TYPE_VECTOR_SUBPARTS (vectype); 5049 int ncopies = LOOP_VINFO_VECT_FACTOR (loop_vinfo) / nunits; 5050 tree vec_def; 5051 5052 gcc_assert (ncopies >= 1); 5053 /* FORNOW. These restrictions should be relaxed. */ 5054 if (nested_in_vect_loop_p (loop, phi)) 5055 { 5056 imm_use_iterator imm_iter; 5057 use_operand_p use_p; 5058 gimple exit_phi; 5059 edge latch_e; 5060 tree loop_arg; 5061 5062 if (ncopies > 1) 5063 { 5064 if (vect_print_dump_info (REPORT_DETAILS)) 5065 fprintf (vect_dump, "multiple types in nested loop."); 5066 return false; 5067 } 5068 5069 exit_phi = NULL; 5070 latch_e = loop_latch_edge (loop->inner); 5071 loop_arg = PHI_ARG_DEF_FROM_EDGE (phi, latch_e); 5072 FOR_EACH_IMM_USE_FAST (use_p, imm_iter, loop_arg) 5073 { 5074 if (!flow_bb_inside_loop_p (loop->inner, 5075 gimple_bb (USE_STMT (use_p)))) 5076 { 5077 exit_phi = USE_STMT (use_p); 5078 break; 5079 } 5080 } 5081 if (exit_phi) 5082 { 5083 stmt_vec_info exit_phi_vinfo = vinfo_for_stmt (exit_phi); 5084 if (!(STMT_VINFO_RELEVANT_P (exit_phi_vinfo) 5085 && !STMT_VINFO_LIVE_P (exit_phi_vinfo))) 5086 { 5087 if (vect_print_dump_info (REPORT_DETAILS)) 5088 fprintf (vect_dump, "inner-loop induction only used outside " 5089 "of the outer vectorized loop."); 5090 return false; 5091 } 5092 } 5093 } 5094 5095 if (!STMT_VINFO_RELEVANT_P (stmt_info)) 5096 return false; 5097 5098 /* FORNOW: SLP not supported. */ 5099 if (STMT_SLP_TYPE (stmt_info)) 5100 return false; 5101 5102 gcc_assert (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def); 5103 5104 if (gimple_code (phi) != GIMPLE_PHI) 5105 return false; 5106 5107 if (!vec_stmt) /* transformation not required. */ 5108 { 5109 STMT_VINFO_TYPE (stmt_info) = induc_vec_info_type; 5110 if (vect_print_dump_info (REPORT_DETAILS)) 5111 fprintf (vect_dump, "=== vectorizable_induction ==="); 5112 vect_model_induction_cost (stmt_info, ncopies); 5113 return true; 5114 } 5115 5116 /** Transform. **/ 5117 5118 if (vect_print_dump_info (REPORT_DETAILS)) 5119 fprintf (vect_dump, "transform induction phi."); 5120 5121 vec_def = get_initial_def_for_induction (phi); 5122 *vec_stmt = SSA_NAME_DEF_STMT (vec_def); 5123 return true; 5124 } 5125 5126 /* Function vectorizable_live_operation. 5127 5128 STMT computes a value that is used outside the loop. Check if 5129 it can be supported. */ 5130 5131 bool 5132 vectorizable_live_operation (gimple stmt, 5133 gimple_stmt_iterator *gsi ATTRIBUTE_UNUSED, 5134 gimple *vec_stmt ATTRIBUTE_UNUSED) 5135 { 5136 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 5137 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 5138 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 5139 int i; 5140 int op_type; 5141 tree op; 5142 tree def; 5143 gimple def_stmt; 5144 enum vect_def_type dt; 5145 enum tree_code code; 5146 enum gimple_rhs_class rhs_class; 5147 5148 gcc_assert (STMT_VINFO_LIVE_P (stmt_info)); 5149 5150 if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_reduction_def) 5151 return false; 5152 5153 if (!is_gimple_assign (stmt)) 5154 return false; 5155 5156 if (TREE_CODE (gimple_assign_lhs (stmt)) != SSA_NAME) 5157 return false; 5158 5159 /* FORNOW. CHECKME. */ 5160 if (nested_in_vect_loop_p (loop, stmt)) 5161 return false; 5162 5163 code = gimple_assign_rhs_code (stmt); 5164 op_type = TREE_CODE_LENGTH (code); 5165 rhs_class = get_gimple_rhs_class (code); 5166 gcc_assert (rhs_class != GIMPLE_UNARY_RHS || op_type == unary_op); 5167 gcc_assert (rhs_class != GIMPLE_BINARY_RHS || op_type == binary_op); 5168 5169 /* FORNOW: support only if all uses are invariant. This means 5170 that the scalar operations can remain in place, unvectorized. 5171 The original last scalar value that they compute will be used. */ 5172 5173 for (i = 0; i < op_type; i++) 5174 { 5175 if (rhs_class == GIMPLE_SINGLE_RHS) 5176 op = TREE_OPERAND (gimple_op (stmt, 1), i); 5177 else 5178 op = gimple_op (stmt, i + 1); 5179 if (op 5180 && !vect_is_simple_use (op, stmt, loop_vinfo, NULL, &def_stmt, &def, 5181 &dt)) 5182 { 5183 if (vect_print_dump_info (REPORT_DETAILS)) 5184 fprintf (vect_dump, "use not simple."); 5185 return false; 5186 } 5187 5188 if (dt != vect_external_def && dt != vect_constant_def) 5189 return false; 5190 } 5191 5192 /* No transformation is required for the cases we currently support. */ 5193 return true; 5194 } 5195 5196 /* Kill any debug uses outside LOOP of SSA names defined in STMT. */ 5197 5198 static void 5199 vect_loop_kill_debug_uses (struct loop *loop, gimple stmt) 5200 { 5201 ssa_op_iter op_iter; 5202 imm_use_iterator imm_iter; 5203 def_operand_p def_p; 5204 gimple ustmt; 5205 5206 FOR_EACH_PHI_OR_STMT_DEF (def_p, stmt, op_iter, SSA_OP_DEF) 5207 { 5208 FOR_EACH_IMM_USE_STMT (ustmt, imm_iter, DEF_FROM_PTR (def_p)) 5209 { 5210 basic_block bb; 5211 5212 if (!is_gimple_debug (ustmt)) 5213 continue; 5214 5215 bb = gimple_bb (ustmt); 5216 5217 if (!flow_bb_inside_loop_p (loop, bb)) 5218 { 5219 if (gimple_debug_bind_p (ustmt)) 5220 { 5221 if (vect_print_dump_info (REPORT_DETAILS)) 5222 fprintf (vect_dump, "killing debug use"); 5223 5224 gimple_debug_bind_reset_value (ustmt); 5225 update_stmt (ustmt); 5226 } 5227 else 5228 gcc_unreachable (); 5229 } 5230 } 5231 } 5232 } 5233 5234 /* Function vect_transform_loop. 5235 5236 The analysis phase has determined that the loop is vectorizable. 5237 Vectorize the loop - created vectorized stmts to replace the scalar 5238 stmts in the loop, and update the loop exit condition. */ 5239 5240 void 5241 vect_transform_loop (loop_vec_info loop_vinfo) 5242 { 5243 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 5244 basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo); 5245 int nbbs = loop->num_nodes; 5246 gimple_stmt_iterator si; 5247 int i; 5248 tree ratio = NULL; 5249 int vectorization_factor = LOOP_VINFO_VECT_FACTOR (loop_vinfo); 5250 bool strided_store; 5251 bool slp_scheduled = false; 5252 unsigned int nunits; 5253 tree cond_expr = NULL_TREE; 5254 gimple_seq cond_expr_stmt_list = NULL; 5255 bool do_peeling_for_loop_bound; 5256 gimple stmt, pattern_stmt; 5257 gimple_seq pattern_def_seq = NULL; 5258 gimple_stmt_iterator pattern_def_si = gsi_start (NULL); 5259 bool transform_pattern_stmt = false; 5260 5261 if (vect_print_dump_info (REPORT_DETAILS)) 5262 fprintf (vect_dump, "=== vec_transform_loop ==="); 5263 5264 /* Peel the loop if there are data refs with unknown alignment. 5265 Only one data ref with unknown store is allowed. */ 5266 5267 if (LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo)) 5268 vect_do_peeling_for_alignment (loop_vinfo); 5269 5270 do_peeling_for_loop_bound 5271 = (!LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) 5272 || (LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo) 5273 && LOOP_VINFO_INT_NITERS (loop_vinfo) % vectorization_factor != 0) 5274 || LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo)); 5275 5276 if (LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT (loop_vinfo) 5277 || LOOP_REQUIRES_VERSIONING_FOR_ALIAS (loop_vinfo)) 5278 vect_loop_versioning (loop_vinfo, 5279 !do_peeling_for_loop_bound, 5280 &cond_expr, &cond_expr_stmt_list); 5281 5282 /* If the loop has a symbolic number of iterations 'n' (i.e. it's not a 5283 compile time constant), or it is a constant that doesn't divide by the 5284 vectorization factor, then an epilog loop needs to be created. 5285 We therefore duplicate the loop: the original loop will be vectorized, 5286 and will compute the first (n/VF) iterations. The second copy of the loop 5287 will remain scalar and will compute the remaining (n%VF) iterations. 5288 (VF is the vectorization factor). */ 5289 5290 if (do_peeling_for_loop_bound) 5291 vect_do_peeling_for_loop_bound (loop_vinfo, &ratio, 5292 cond_expr, cond_expr_stmt_list); 5293 else 5294 ratio = build_int_cst (TREE_TYPE (LOOP_VINFO_NITERS (loop_vinfo)), 5295 LOOP_VINFO_INT_NITERS (loop_vinfo) / vectorization_factor); 5296 5297 /* 1) Make sure the loop header has exactly two entries 5298 2) Make sure we have a preheader basic block. */ 5299 5300 gcc_assert (EDGE_COUNT (loop->header->preds) == 2); 5301 5302 split_edge (loop_preheader_edge (loop)); 5303 5304 /* FORNOW: the vectorizer supports only loops which body consist 5305 of one basic block (header + empty latch). When the vectorizer will 5306 support more involved loop forms, the order by which the BBs are 5307 traversed need to be reconsidered. */ 5308 5309 for (i = 0; i < nbbs; i++) 5310 { 5311 basic_block bb = bbs[i]; 5312 stmt_vec_info stmt_info; 5313 gimple phi; 5314 5315 for (si = gsi_start_phis (bb); !gsi_end_p (si); gsi_next (&si)) 5316 { 5317 phi = gsi_stmt (si); 5318 if (vect_print_dump_info (REPORT_DETAILS)) 5319 { 5320 fprintf (vect_dump, "------>vectorizing phi: "); 5321 print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM); 5322 } 5323 stmt_info = vinfo_for_stmt (phi); 5324 if (!stmt_info) 5325 continue; 5326 5327 if (MAY_HAVE_DEBUG_STMTS && !STMT_VINFO_LIVE_P (stmt_info)) 5328 vect_loop_kill_debug_uses (loop, phi); 5329 5330 if (!STMT_VINFO_RELEVANT_P (stmt_info) 5331 && !STMT_VINFO_LIVE_P (stmt_info)) 5332 continue; 5333 5334 if ((TYPE_VECTOR_SUBPARTS (STMT_VINFO_VECTYPE (stmt_info)) 5335 != (unsigned HOST_WIDE_INT) vectorization_factor) 5336 && vect_print_dump_info (REPORT_DETAILS)) 5337 fprintf (vect_dump, "multiple-types."); 5338 5339 if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_induction_def) 5340 { 5341 if (vect_print_dump_info (REPORT_DETAILS)) 5342 fprintf (vect_dump, "transform phi."); 5343 vect_transform_stmt (phi, NULL, NULL, NULL, NULL); 5344 } 5345 } 5346 5347 pattern_stmt = NULL; 5348 for (si = gsi_start_bb (bb); !gsi_end_p (si) || transform_pattern_stmt;) 5349 { 5350 bool is_store; 5351 5352 if (transform_pattern_stmt) 5353 stmt = pattern_stmt; 5354 else 5355 stmt = gsi_stmt (si); 5356 5357 if (vect_print_dump_info (REPORT_DETAILS)) 5358 { 5359 fprintf (vect_dump, "------>vectorizing statement: "); 5360 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 5361 } 5362 5363 stmt_info = vinfo_for_stmt (stmt); 5364 5365 /* vector stmts created in the outer-loop during vectorization of 5366 stmts in an inner-loop may not have a stmt_info, and do not 5367 need to be vectorized. */ 5368 if (!stmt_info) 5369 { 5370 gsi_next (&si); 5371 continue; 5372 } 5373 5374 if (MAY_HAVE_DEBUG_STMTS && !STMT_VINFO_LIVE_P (stmt_info)) 5375 vect_loop_kill_debug_uses (loop, stmt); 5376 5377 if (!STMT_VINFO_RELEVANT_P (stmt_info) 5378 && !STMT_VINFO_LIVE_P (stmt_info)) 5379 { 5380 if (STMT_VINFO_IN_PATTERN_P (stmt_info) 5381 && (pattern_stmt = STMT_VINFO_RELATED_STMT (stmt_info)) 5382 && (STMT_VINFO_RELEVANT_P (vinfo_for_stmt (pattern_stmt)) 5383 || STMT_VINFO_LIVE_P (vinfo_for_stmt (pattern_stmt)))) 5384 { 5385 stmt = pattern_stmt; 5386 stmt_info = vinfo_for_stmt (stmt); 5387 } 5388 else 5389 { 5390 gsi_next (&si); 5391 continue; 5392 } 5393 } 5394 else if (STMT_VINFO_IN_PATTERN_P (stmt_info) 5395 && (pattern_stmt = STMT_VINFO_RELATED_STMT (stmt_info)) 5396 && (STMT_VINFO_RELEVANT_P (vinfo_for_stmt (pattern_stmt)) 5397 || STMT_VINFO_LIVE_P (vinfo_for_stmt (pattern_stmt)))) 5398 transform_pattern_stmt = true; 5399 5400 /* If pattern statement has def stmts, vectorize them too. */ 5401 if (is_pattern_stmt_p (stmt_info)) 5402 { 5403 if (pattern_def_seq == NULL) 5404 { 5405 pattern_def_seq = STMT_VINFO_PATTERN_DEF_SEQ (stmt_info); 5406 pattern_def_si = gsi_start (pattern_def_seq); 5407 } 5408 else if (!gsi_end_p (pattern_def_si)) 5409 gsi_next (&pattern_def_si); 5410 if (pattern_def_seq != NULL) 5411 { 5412 gimple pattern_def_stmt = NULL; 5413 stmt_vec_info pattern_def_stmt_info = NULL; 5414 5415 while (!gsi_end_p (pattern_def_si)) 5416 { 5417 pattern_def_stmt = gsi_stmt (pattern_def_si); 5418 pattern_def_stmt_info 5419 = vinfo_for_stmt (pattern_def_stmt); 5420 if (STMT_VINFO_RELEVANT_P (pattern_def_stmt_info) 5421 || STMT_VINFO_LIVE_P (pattern_def_stmt_info)) 5422 break; 5423 gsi_next (&pattern_def_si); 5424 } 5425 5426 if (!gsi_end_p (pattern_def_si)) 5427 { 5428 if (vect_print_dump_info (REPORT_DETAILS)) 5429 { 5430 fprintf (vect_dump, "==> vectorizing pattern def" 5431 " stmt: "); 5432 print_gimple_stmt (vect_dump, pattern_def_stmt, 0, 5433 TDF_SLIM); 5434 } 5435 5436 stmt = pattern_def_stmt; 5437 stmt_info = pattern_def_stmt_info; 5438 } 5439 else 5440 { 5441 pattern_def_si = gsi_start (NULL); 5442 transform_pattern_stmt = false; 5443 } 5444 } 5445 else 5446 transform_pattern_stmt = false; 5447 } 5448 5449 gcc_assert (STMT_VINFO_VECTYPE (stmt_info)); 5450 nunits = (unsigned int) TYPE_VECTOR_SUBPARTS ( 5451 STMT_VINFO_VECTYPE (stmt_info)); 5452 if (!STMT_SLP_TYPE (stmt_info) 5453 && nunits != (unsigned int) vectorization_factor 5454 && vect_print_dump_info (REPORT_DETAILS)) 5455 /* For SLP VF is set according to unrolling factor, and not to 5456 vector size, hence for SLP this print is not valid. */ 5457 fprintf (vect_dump, "multiple-types."); 5458 5459 /* SLP. Schedule all the SLP instances when the first SLP stmt is 5460 reached. */ 5461 if (STMT_SLP_TYPE (stmt_info)) 5462 { 5463 if (!slp_scheduled) 5464 { 5465 slp_scheduled = true; 5466 5467 if (vect_print_dump_info (REPORT_DETAILS)) 5468 fprintf (vect_dump, "=== scheduling SLP instances ==="); 5469 5470 vect_schedule_slp (loop_vinfo, NULL); 5471 } 5472 5473 /* Hybrid SLP stmts must be vectorized in addition to SLP. */ 5474 if (!vinfo_for_stmt (stmt) || PURE_SLP_STMT (stmt_info)) 5475 { 5476 if (!transform_pattern_stmt && gsi_end_p (pattern_def_si)) 5477 { 5478 pattern_def_seq = NULL; 5479 gsi_next (&si); 5480 } 5481 continue; 5482 } 5483 } 5484 5485 /* -------- vectorize statement ------------ */ 5486 if (vect_print_dump_info (REPORT_DETAILS)) 5487 fprintf (vect_dump, "transform statement."); 5488 5489 strided_store = false; 5490 is_store = vect_transform_stmt (stmt, &si, &strided_store, NULL, NULL); 5491 if (is_store) 5492 { 5493 if (STMT_VINFO_STRIDED_ACCESS (stmt_info)) 5494 { 5495 /* Interleaving. If IS_STORE is TRUE, the vectorization of the 5496 interleaving chain was completed - free all the stores in 5497 the chain. */ 5498 gsi_next (&si); 5499 vect_remove_stores (GROUP_FIRST_ELEMENT (stmt_info)); 5500 continue; 5501 } 5502 else 5503 { 5504 /* Free the attached stmt_vec_info and remove the stmt. */ 5505 free_stmt_vec_info (gsi_stmt (si)); 5506 gsi_remove (&si, true); 5507 continue; 5508 } 5509 } 5510 5511 if (!transform_pattern_stmt && gsi_end_p (pattern_def_si)) 5512 { 5513 pattern_def_seq = NULL; 5514 gsi_next (&si); 5515 } 5516 } /* stmts in BB */ 5517 } /* BBs in loop */ 5518 5519 slpeel_make_loop_iterate_ntimes (loop, ratio); 5520 5521 /* The memory tags and pointers in vectorized statements need to 5522 have their SSA forms updated. FIXME, why can't this be delayed 5523 until all the loops have been transformed? */ 5524 update_ssa (TODO_update_ssa); 5525 5526 if (vect_print_dump_info (REPORT_VECTORIZED_LOCATIONS)) 5527 fprintf (vect_dump, "LOOP VECTORIZED."); 5528 if (loop->inner && vect_print_dump_info (REPORT_VECTORIZED_LOCATIONS)) 5529 fprintf (vect_dump, "OUTER LOOP VECTORIZED."); 5530 } 5531