1 /* Data References Analysis and Manipulation Utilities for Vectorization. 2 Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010, 2011 3 Free Software Foundation, Inc. 4 Contributed by Dorit Naishlos <dorit@il.ibm.com> 5 and 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 "tm_p.h" 30 #include "target.h" 31 #include "basic-block.h" 32 #include "tree-pretty-print.h" 33 #include "gimple-pretty-print.h" 34 #include "tree-flow.h" 35 #include "tree-dump.h" 36 #include "cfgloop.h" 37 #include "tree-chrec.h" 38 #include "tree-scalar-evolution.h" 39 #include "tree-vectorizer.h" 40 #include "diagnostic-core.h" 41 42 /* Need to include rtl.h, expr.h, etc. for optabs. */ 43 #include "expr.h" 44 #include "optabs.h" 45 46 /* Return true if load- or store-lanes optab OPTAB is implemented for 47 COUNT vectors of type VECTYPE. NAME is the name of OPTAB. */ 48 49 static bool 50 vect_lanes_optab_supported_p (const char *name, convert_optab optab, 51 tree vectype, unsigned HOST_WIDE_INT count) 52 { 53 enum machine_mode mode, array_mode; 54 bool limit_p; 55 56 mode = TYPE_MODE (vectype); 57 limit_p = !targetm.array_mode_supported_p (mode, count); 58 array_mode = mode_for_size (count * GET_MODE_BITSIZE (mode), 59 MODE_INT, limit_p); 60 61 if (array_mode == BLKmode) 62 { 63 if (vect_print_dump_info (REPORT_DETAILS)) 64 fprintf (vect_dump, "no array mode for %s[" HOST_WIDE_INT_PRINT_DEC "]", 65 GET_MODE_NAME (mode), count); 66 return false; 67 } 68 69 if (convert_optab_handler (optab, array_mode, mode) == CODE_FOR_nothing) 70 { 71 if (vect_print_dump_info (REPORT_DETAILS)) 72 fprintf (vect_dump, "cannot use %s<%s><%s>", 73 name, GET_MODE_NAME (array_mode), GET_MODE_NAME (mode)); 74 return false; 75 } 76 77 if (vect_print_dump_info (REPORT_DETAILS)) 78 fprintf (vect_dump, "can use %s<%s><%s>", 79 name, GET_MODE_NAME (array_mode), GET_MODE_NAME (mode)); 80 81 return true; 82 } 83 84 85 /* Return the smallest scalar part of STMT. 86 This is used to determine the vectype of the stmt. We generally set the 87 vectype according to the type of the result (lhs). For stmts whose 88 result-type is different than the type of the arguments (e.g., demotion, 89 promotion), vectype will be reset appropriately (later). Note that we have 90 to visit the smallest datatype in this function, because that determines the 91 VF. If the smallest datatype in the loop is present only as the rhs of a 92 promotion operation - we'd miss it. 93 Such a case, where a variable of this datatype does not appear in the lhs 94 anywhere in the loop, can only occur if it's an invariant: e.g.: 95 'int_x = (int) short_inv', which we'd expect to have been optimized away by 96 invariant motion. However, we cannot rely on invariant motion to always 97 take invariants out of the loop, and so in the case of promotion we also 98 have to check the rhs. 99 LHS_SIZE_UNIT and RHS_SIZE_UNIT contain the sizes of the corresponding 100 types. */ 101 102 tree 103 vect_get_smallest_scalar_type (gimple stmt, HOST_WIDE_INT *lhs_size_unit, 104 HOST_WIDE_INT *rhs_size_unit) 105 { 106 tree scalar_type = gimple_expr_type (stmt); 107 HOST_WIDE_INT lhs, rhs; 108 109 lhs = rhs = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (scalar_type)); 110 111 if (is_gimple_assign (stmt) 112 && (gimple_assign_cast_p (stmt) 113 || gimple_assign_rhs_code (stmt) == WIDEN_MULT_EXPR 114 || gimple_assign_rhs_code (stmt) == FLOAT_EXPR)) 115 { 116 tree rhs_type = TREE_TYPE (gimple_assign_rhs1 (stmt)); 117 118 rhs = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (rhs_type)); 119 if (rhs < lhs) 120 scalar_type = rhs_type; 121 } 122 123 *lhs_size_unit = lhs; 124 *rhs_size_unit = rhs; 125 return scalar_type; 126 } 127 128 129 /* Find the place of the data-ref in STMT in the interleaving chain that starts 130 from FIRST_STMT. Return -1 if the data-ref is not a part of the chain. */ 131 132 int 133 vect_get_place_in_interleaving_chain (gimple stmt, gimple first_stmt) 134 { 135 gimple next_stmt = first_stmt; 136 int result = 0; 137 138 if (first_stmt != GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt))) 139 return -1; 140 141 while (next_stmt && next_stmt != stmt) 142 { 143 result++; 144 next_stmt = GROUP_NEXT_ELEMENT (vinfo_for_stmt (next_stmt)); 145 } 146 147 if (next_stmt) 148 return result; 149 else 150 return -1; 151 } 152 153 154 /* Function vect_insert_into_interleaving_chain. 155 156 Insert DRA into the interleaving chain of DRB according to DRA's INIT. */ 157 158 static void 159 vect_insert_into_interleaving_chain (struct data_reference *dra, 160 struct data_reference *drb) 161 { 162 gimple prev, next; 163 tree next_init; 164 stmt_vec_info stmtinfo_a = vinfo_for_stmt (DR_STMT (dra)); 165 stmt_vec_info stmtinfo_b = vinfo_for_stmt (DR_STMT (drb)); 166 167 prev = GROUP_FIRST_ELEMENT (stmtinfo_b); 168 next = GROUP_NEXT_ELEMENT (vinfo_for_stmt (prev)); 169 while (next) 170 { 171 next_init = DR_INIT (STMT_VINFO_DATA_REF (vinfo_for_stmt (next))); 172 if (tree_int_cst_compare (next_init, DR_INIT (dra)) > 0) 173 { 174 /* Insert here. */ 175 GROUP_NEXT_ELEMENT (vinfo_for_stmt (prev)) = DR_STMT (dra); 176 GROUP_NEXT_ELEMENT (stmtinfo_a) = next; 177 return; 178 } 179 prev = next; 180 next = GROUP_NEXT_ELEMENT (vinfo_for_stmt (prev)); 181 } 182 183 /* We got to the end of the list. Insert here. */ 184 GROUP_NEXT_ELEMENT (vinfo_for_stmt (prev)) = DR_STMT (dra); 185 GROUP_NEXT_ELEMENT (stmtinfo_a) = NULL; 186 } 187 188 189 /* Function vect_update_interleaving_chain. 190 191 For two data-refs DRA and DRB that are a part of a chain interleaved data 192 accesses, update the interleaving chain. DRB's INIT is smaller than DRA's. 193 194 There are four possible cases: 195 1. New stmts - both DRA and DRB are not a part of any chain: 196 FIRST_DR = DRB 197 NEXT_DR (DRB) = DRA 198 2. DRB is a part of a chain and DRA is not: 199 no need to update FIRST_DR 200 no need to insert DRB 201 insert DRA according to init 202 3. DRA is a part of a chain and DRB is not: 203 if (init of FIRST_DR > init of DRB) 204 FIRST_DR = DRB 205 NEXT(FIRST_DR) = previous FIRST_DR 206 else 207 insert DRB according to its init 208 4. both DRA and DRB are in some interleaving chains: 209 choose the chain with the smallest init of FIRST_DR 210 insert the nodes of the second chain into the first one. */ 211 212 static void 213 vect_update_interleaving_chain (struct data_reference *drb, 214 struct data_reference *dra) 215 { 216 stmt_vec_info stmtinfo_a = vinfo_for_stmt (DR_STMT (dra)); 217 stmt_vec_info stmtinfo_b = vinfo_for_stmt (DR_STMT (drb)); 218 tree next_init, init_dra_chain, init_drb_chain; 219 gimple first_a, first_b; 220 tree node_init; 221 gimple node, prev, next, first_stmt; 222 223 /* 1. New stmts - both DRA and DRB are not a part of any chain. */ 224 if (!GROUP_FIRST_ELEMENT (stmtinfo_a) && !GROUP_FIRST_ELEMENT (stmtinfo_b)) 225 { 226 GROUP_FIRST_ELEMENT (stmtinfo_a) = DR_STMT (drb); 227 GROUP_FIRST_ELEMENT (stmtinfo_b) = DR_STMT (drb); 228 GROUP_NEXT_ELEMENT (stmtinfo_b) = DR_STMT (dra); 229 return; 230 } 231 232 /* 2. DRB is a part of a chain and DRA is not. */ 233 if (!GROUP_FIRST_ELEMENT (stmtinfo_a) && GROUP_FIRST_ELEMENT (stmtinfo_b)) 234 { 235 GROUP_FIRST_ELEMENT (stmtinfo_a) = GROUP_FIRST_ELEMENT (stmtinfo_b); 236 /* Insert DRA into the chain of DRB. */ 237 vect_insert_into_interleaving_chain (dra, drb); 238 return; 239 } 240 241 /* 3. DRA is a part of a chain and DRB is not. */ 242 if (GROUP_FIRST_ELEMENT (stmtinfo_a) && !GROUP_FIRST_ELEMENT (stmtinfo_b)) 243 { 244 gimple old_first_stmt = GROUP_FIRST_ELEMENT (stmtinfo_a); 245 tree init_old = DR_INIT (STMT_VINFO_DATA_REF (vinfo_for_stmt ( 246 old_first_stmt))); 247 gimple tmp; 248 249 if (tree_int_cst_compare (init_old, DR_INIT (drb)) > 0) 250 { 251 /* DRB's init is smaller than the init of the stmt previously marked 252 as the first stmt of the interleaving chain of DRA. Therefore, we 253 update FIRST_STMT and put DRB in the head of the list. */ 254 GROUP_FIRST_ELEMENT (stmtinfo_b) = DR_STMT (drb); 255 GROUP_NEXT_ELEMENT (stmtinfo_b) = old_first_stmt; 256 257 /* Update all the stmts in the list to point to the new FIRST_STMT. */ 258 tmp = old_first_stmt; 259 while (tmp) 260 { 261 GROUP_FIRST_ELEMENT (vinfo_for_stmt (tmp)) = DR_STMT (drb); 262 tmp = GROUP_NEXT_ELEMENT (vinfo_for_stmt (tmp)); 263 } 264 } 265 else 266 { 267 /* Insert DRB in the list of DRA. */ 268 vect_insert_into_interleaving_chain (drb, dra); 269 GROUP_FIRST_ELEMENT (stmtinfo_b) = GROUP_FIRST_ELEMENT (stmtinfo_a); 270 } 271 return; 272 } 273 274 /* 4. both DRA and DRB are in some interleaving chains. */ 275 first_a = GROUP_FIRST_ELEMENT (stmtinfo_a); 276 first_b = GROUP_FIRST_ELEMENT (stmtinfo_b); 277 if (first_a == first_b) 278 return; 279 init_dra_chain = DR_INIT (STMT_VINFO_DATA_REF (vinfo_for_stmt (first_a))); 280 init_drb_chain = DR_INIT (STMT_VINFO_DATA_REF (vinfo_for_stmt (first_b))); 281 282 if (tree_int_cst_compare (init_dra_chain, init_drb_chain) > 0) 283 { 284 /* Insert the nodes of DRA chain into the DRB chain. 285 After inserting a node, continue from this node of the DRB chain (don't 286 start from the beginning. */ 287 node = GROUP_FIRST_ELEMENT (stmtinfo_a); 288 prev = GROUP_FIRST_ELEMENT (stmtinfo_b); 289 first_stmt = first_b; 290 } 291 else 292 { 293 /* Insert the nodes of DRB chain into the DRA chain. 294 After inserting a node, continue from this node of the DRA chain (don't 295 start from the beginning. */ 296 node = GROUP_FIRST_ELEMENT (stmtinfo_b); 297 prev = GROUP_FIRST_ELEMENT (stmtinfo_a); 298 first_stmt = first_a; 299 } 300 301 while (node) 302 { 303 node_init = DR_INIT (STMT_VINFO_DATA_REF (vinfo_for_stmt (node))); 304 next = GROUP_NEXT_ELEMENT (vinfo_for_stmt (prev)); 305 while (next) 306 { 307 next_init = DR_INIT (STMT_VINFO_DATA_REF (vinfo_for_stmt (next))); 308 if (tree_int_cst_compare (next_init, node_init) > 0) 309 { 310 /* Insert here. */ 311 GROUP_NEXT_ELEMENT (vinfo_for_stmt (prev)) = node; 312 GROUP_NEXT_ELEMENT (vinfo_for_stmt (node)) = next; 313 prev = node; 314 break; 315 } 316 prev = next; 317 next = GROUP_NEXT_ELEMENT (vinfo_for_stmt (prev)); 318 } 319 if (!next) 320 { 321 /* We got to the end of the list. Insert here. */ 322 GROUP_NEXT_ELEMENT (vinfo_for_stmt (prev)) = node; 323 GROUP_NEXT_ELEMENT (vinfo_for_stmt (node)) = NULL; 324 prev = node; 325 } 326 GROUP_FIRST_ELEMENT (vinfo_for_stmt (node)) = first_stmt; 327 node = GROUP_NEXT_ELEMENT (vinfo_for_stmt (node)); 328 } 329 } 330 331 /* Check dependence between DRA and DRB for basic block vectorization. 332 If the accesses share same bases and offsets, we can compare their initial 333 constant offsets to decide whether they differ or not. In case of a read- 334 write dependence we check that the load is before the store to ensure that 335 vectorization will not change the order of the accesses. */ 336 337 static bool 338 vect_drs_dependent_in_basic_block (struct data_reference *dra, 339 struct data_reference *drb) 340 { 341 HOST_WIDE_INT type_size_a, type_size_b, init_a, init_b; 342 gimple earlier_stmt; 343 344 /* We only call this function for pairs of loads and stores, but we verify 345 it here. */ 346 if (DR_IS_READ (dra) == DR_IS_READ (drb)) 347 { 348 if (DR_IS_READ (dra)) 349 return false; 350 else 351 return true; 352 } 353 354 /* Check that the data-refs have same bases and offsets. If not, we can't 355 determine if they are dependent. */ 356 if (!operand_equal_p (DR_BASE_ADDRESS (dra), DR_BASE_ADDRESS (drb), 0) 357 || !dr_equal_offsets_p (dra, drb)) 358 return true; 359 360 /* Check the types. */ 361 type_size_a = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dra)))); 362 type_size_b = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (drb)))); 363 364 if (type_size_a != type_size_b 365 || !types_compatible_p (TREE_TYPE (DR_REF (dra)), 366 TREE_TYPE (DR_REF (drb)))) 367 return true; 368 369 init_a = TREE_INT_CST_LOW (DR_INIT (dra)); 370 init_b = TREE_INT_CST_LOW (DR_INIT (drb)); 371 372 /* Two different locations - no dependence. */ 373 if (init_a != init_b) 374 return false; 375 376 /* We have a read-write dependence. Check that the load is before the store. 377 When we vectorize basic blocks, vector load can be only before 378 corresponding scalar load, and vector store can be only after its 379 corresponding scalar store. So the order of the acceses is preserved in 380 case the load is before the store. */ 381 earlier_stmt = get_earlier_stmt (DR_STMT (dra), DR_STMT (drb)); 382 if (DR_IS_READ (STMT_VINFO_DATA_REF (vinfo_for_stmt (earlier_stmt)))) 383 return false; 384 385 return true; 386 } 387 388 389 /* Function vect_check_interleaving. 390 391 Check if DRA and DRB are a part of interleaving. In case they are, insert 392 DRA and DRB in an interleaving chain. */ 393 394 static bool 395 vect_check_interleaving (struct data_reference *dra, 396 struct data_reference *drb) 397 { 398 HOST_WIDE_INT type_size_a, type_size_b, diff_mod_size, step, init_a, init_b; 399 400 /* Check that the data-refs have same first location (except init) and they 401 are both either store or load (not load and store). */ 402 if (!operand_equal_p (DR_BASE_ADDRESS (dra), DR_BASE_ADDRESS (drb), 0) 403 || !dr_equal_offsets_p (dra, drb) 404 || !tree_int_cst_compare (DR_INIT (dra), DR_INIT (drb)) 405 || DR_IS_READ (dra) != DR_IS_READ (drb)) 406 return false; 407 408 /* Check: 409 1. data-refs are of the same type 410 2. their steps are equal 411 3. the step (if greater than zero) is greater than the difference between 412 data-refs' inits. */ 413 type_size_a = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dra)))); 414 type_size_b = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (drb)))); 415 416 if (type_size_a != type_size_b 417 || tree_int_cst_compare (DR_STEP (dra), DR_STEP (drb)) 418 || !types_compatible_p (TREE_TYPE (DR_REF (dra)), 419 TREE_TYPE (DR_REF (drb)))) 420 return false; 421 422 init_a = TREE_INT_CST_LOW (DR_INIT (dra)); 423 init_b = TREE_INT_CST_LOW (DR_INIT (drb)); 424 step = TREE_INT_CST_LOW (DR_STEP (dra)); 425 426 if (init_a > init_b) 427 { 428 /* If init_a == init_b + the size of the type * k, we have an interleaving, 429 and DRB is accessed before DRA. */ 430 diff_mod_size = (init_a - init_b) % type_size_a; 431 432 if (step && (init_a - init_b) > step) 433 return false; 434 435 if (diff_mod_size == 0) 436 { 437 vect_update_interleaving_chain (drb, dra); 438 if (vect_print_dump_info (REPORT_DR_DETAILS)) 439 { 440 fprintf (vect_dump, "Detected interleaving "); 441 print_generic_expr (vect_dump, DR_REF (dra), TDF_SLIM); 442 fprintf (vect_dump, " and "); 443 print_generic_expr (vect_dump, DR_REF (drb), TDF_SLIM); 444 } 445 return true; 446 } 447 } 448 else 449 { 450 /* If init_b == init_a + the size of the type * k, we have an 451 interleaving, and DRA is accessed before DRB. */ 452 diff_mod_size = (init_b - init_a) % type_size_a; 453 454 if (step && (init_b - init_a) > step) 455 return false; 456 457 if (diff_mod_size == 0) 458 { 459 vect_update_interleaving_chain (dra, drb); 460 if (vect_print_dump_info (REPORT_DR_DETAILS)) 461 { 462 fprintf (vect_dump, "Detected interleaving "); 463 print_generic_expr (vect_dump, DR_REF (dra), TDF_SLIM); 464 fprintf (vect_dump, " and "); 465 print_generic_expr (vect_dump, DR_REF (drb), TDF_SLIM); 466 } 467 return true; 468 } 469 } 470 471 return false; 472 } 473 474 /* Check if data references pointed by DR_I and DR_J are same or 475 belong to same interleaving group. Return FALSE if drs are 476 different, otherwise return TRUE. */ 477 478 static bool 479 vect_same_range_drs (data_reference_p dr_i, data_reference_p dr_j) 480 { 481 gimple stmt_i = DR_STMT (dr_i); 482 gimple stmt_j = DR_STMT (dr_j); 483 484 if (operand_equal_p (DR_REF (dr_i), DR_REF (dr_j), 0) 485 || (GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt_i)) 486 && GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt_j)) 487 && (GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt_i)) 488 == GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt_j))))) 489 return true; 490 else 491 return false; 492 } 493 494 /* If address ranges represented by DDR_I and DDR_J are equal, 495 return TRUE, otherwise return FALSE. */ 496 497 static bool 498 vect_vfa_range_equal (ddr_p ddr_i, ddr_p ddr_j) 499 { 500 if ((vect_same_range_drs (DDR_A (ddr_i), DDR_A (ddr_j)) 501 && vect_same_range_drs (DDR_B (ddr_i), DDR_B (ddr_j))) 502 || (vect_same_range_drs (DDR_A (ddr_i), DDR_B (ddr_j)) 503 && vect_same_range_drs (DDR_B (ddr_i), DDR_A (ddr_j)))) 504 return true; 505 else 506 return false; 507 } 508 509 /* Insert DDR into LOOP_VINFO list of ddrs that may alias and need to be 510 tested at run-time. Return TRUE if DDR was successfully inserted. 511 Return false if versioning is not supported. */ 512 513 static bool 514 vect_mark_for_runtime_alias_test (ddr_p ddr, loop_vec_info loop_vinfo) 515 { 516 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 517 518 if ((unsigned) PARAM_VALUE (PARAM_VECT_MAX_VERSION_FOR_ALIAS_CHECKS) == 0) 519 return false; 520 521 if (vect_print_dump_info (REPORT_DR_DETAILS)) 522 { 523 fprintf (vect_dump, "mark for run-time aliasing test between "); 524 print_generic_expr (vect_dump, DR_REF (DDR_A (ddr)), TDF_SLIM); 525 fprintf (vect_dump, " and "); 526 print_generic_expr (vect_dump, DR_REF (DDR_B (ddr)), TDF_SLIM); 527 } 528 529 if (optimize_loop_nest_for_size_p (loop)) 530 { 531 if (vect_print_dump_info (REPORT_DR_DETAILS)) 532 fprintf (vect_dump, "versioning not supported when optimizing for size."); 533 return false; 534 } 535 536 /* FORNOW: We don't support versioning with outer-loop vectorization. */ 537 if (loop->inner) 538 { 539 if (vect_print_dump_info (REPORT_DR_DETAILS)) 540 fprintf (vect_dump, "versioning not yet supported for outer-loops."); 541 return false; 542 } 543 544 VEC_safe_push (ddr_p, heap, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo), ddr); 545 return true; 546 } 547 548 549 /* Function vect_analyze_data_ref_dependence. 550 551 Return TRUE if there (might) exist a dependence between a memory-reference 552 DRA and a memory-reference DRB. When versioning for alias may check a 553 dependence at run-time, return FALSE. Adjust *MAX_VF according to 554 the data dependence. */ 555 556 static bool 557 vect_analyze_data_ref_dependence (struct data_dependence_relation *ddr, 558 loop_vec_info loop_vinfo, int *max_vf) 559 { 560 unsigned int i; 561 struct loop *loop = NULL; 562 struct data_reference *dra = DDR_A (ddr); 563 struct data_reference *drb = DDR_B (ddr); 564 stmt_vec_info stmtinfo_a = vinfo_for_stmt (DR_STMT (dra)); 565 stmt_vec_info stmtinfo_b = vinfo_for_stmt (DR_STMT (drb)); 566 lambda_vector dist_v; 567 unsigned int loop_depth; 568 569 /* Don't bother to analyze statements marked as unvectorizable. */ 570 if (!STMT_VINFO_VECTORIZABLE (stmtinfo_a) 571 || !STMT_VINFO_VECTORIZABLE (stmtinfo_b)) 572 return false; 573 574 if (DDR_ARE_DEPENDENT (ddr) == chrec_known) 575 { 576 /* Independent data accesses. */ 577 vect_check_interleaving (dra, drb); 578 return false; 579 } 580 581 if (loop_vinfo) 582 loop = LOOP_VINFO_LOOP (loop_vinfo); 583 584 if ((DR_IS_READ (dra) && DR_IS_READ (drb) && loop_vinfo) || dra == drb) 585 return false; 586 587 if (DDR_ARE_DEPENDENT (ddr) == chrec_dont_know) 588 { 589 gimple earlier_stmt; 590 591 if (loop_vinfo) 592 { 593 if (vect_print_dump_info (REPORT_DR_DETAILS)) 594 { 595 fprintf (vect_dump, "versioning for alias required: " 596 "can't determine dependence between "); 597 print_generic_expr (vect_dump, DR_REF (dra), TDF_SLIM); 598 fprintf (vect_dump, " and "); 599 print_generic_expr (vect_dump, DR_REF (drb), TDF_SLIM); 600 } 601 602 /* Add to list of ddrs that need to be tested at run-time. */ 603 return !vect_mark_for_runtime_alias_test (ddr, loop_vinfo); 604 } 605 606 /* When vectorizing a basic block unknown depnedence can still mean 607 strided access. */ 608 if (vect_check_interleaving (dra, drb)) 609 return false; 610 611 /* Read-read is OK (we need this check here, after checking for 612 interleaving). */ 613 if (DR_IS_READ (dra) && DR_IS_READ (drb)) 614 return false; 615 616 if (vect_print_dump_info (REPORT_DR_DETAILS)) 617 { 618 fprintf (vect_dump, "can't determine dependence between "); 619 print_generic_expr (vect_dump, DR_REF (dra), TDF_SLIM); 620 fprintf (vect_dump, " and "); 621 print_generic_expr (vect_dump, DR_REF (drb), TDF_SLIM); 622 } 623 624 /* We do not vectorize basic blocks with write-write dependencies. */ 625 if (DR_IS_WRITE (dra) && DR_IS_WRITE (drb)) 626 return true; 627 628 /* Check that it's not a load-after-store dependence. */ 629 earlier_stmt = get_earlier_stmt (DR_STMT (dra), DR_STMT (drb)); 630 if (DR_IS_WRITE (STMT_VINFO_DATA_REF (vinfo_for_stmt (earlier_stmt)))) 631 return true; 632 633 return false; 634 } 635 636 /* Versioning for alias is not yet supported for basic block SLP, and 637 dependence distance is unapplicable, hence, in case of known data 638 dependence, basic block vectorization is impossible for now. */ 639 if (!loop_vinfo) 640 { 641 if (dra != drb && vect_check_interleaving (dra, drb)) 642 return false; 643 644 if (vect_print_dump_info (REPORT_DR_DETAILS)) 645 { 646 fprintf (vect_dump, "determined dependence between "); 647 print_generic_expr (vect_dump, DR_REF (dra), TDF_SLIM); 648 fprintf (vect_dump, " and "); 649 print_generic_expr (vect_dump, DR_REF (drb), TDF_SLIM); 650 } 651 652 /* Do not vectorize basic blcoks with write-write dependences. */ 653 if (DR_IS_WRITE (dra) && DR_IS_WRITE (drb)) 654 return true; 655 656 /* Check if this dependence is allowed in basic block vectorization. */ 657 return vect_drs_dependent_in_basic_block (dra, drb); 658 } 659 660 /* Loop-based vectorization and known data dependence. */ 661 if (DDR_NUM_DIST_VECTS (ddr) == 0) 662 { 663 if (vect_print_dump_info (REPORT_DR_DETAILS)) 664 { 665 fprintf (vect_dump, "versioning for alias required: bad dist vector for "); 666 print_generic_expr (vect_dump, DR_REF (dra), TDF_SLIM); 667 fprintf (vect_dump, " and "); 668 print_generic_expr (vect_dump, DR_REF (drb), TDF_SLIM); 669 } 670 /* Add to list of ddrs that need to be tested at run-time. */ 671 return !vect_mark_for_runtime_alias_test (ddr, loop_vinfo); 672 } 673 674 loop_depth = index_in_loop_nest (loop->num, DDR_LOOP_NEST (ddr)); 675 FOR_EACH_VEC_ELT (lambda_vector, DDR_DIST_VECTS (ddr), i, dist_v) 676 { 677 int dist = dist_v[loop_depth]; 678 679 if (vect_print_dump_info (REPORT_DR_DETAILS)) 680 fprintf (vect_dump, "dependence distance = %d.", dist); 681 682 if (dist == 0) 683 { 684 if (vect_print_dump_info (REPORT_DR_DETAILS)) 685 { 686 fprintf (vect_dump, "dependence distance == 0 between "); 687 print_generic_expr (vect_dump, DR_REF (dra), TDF_SLIM); 688 fprintf (vect_dump, " and "); 689 print_generic_expr (vect_dump, DR_REF (drb), TDF_SLIM); 690 } 691 692 /* For interleaving, mark that there is a read-write dependency if 693 necessary. We check before that one of the data-refs is store. */ 694 if (DR_IS_READ (dra)) 695 GROUP_READ_WRITE_DEPENDENCE (stmtinfo_a) = true; 696 else 697 { 698 if (DR_IS_READ (drb)) 699 GROUP_READ_WRITE_DEPENDENCE (stmtinfo_b) = true; 700 } 701 702 continue; 703 } 704 705 if (dist > 0 && DDR_REVERSED_P (ddr)) 706 { 707 /* If DDR_REVERSED_P the order of the data-refs in DDR was 708 reversed (to make distance vector positive), and the actual 709 distance is negative. */ 710 if (vect_print_dump_info (REPORT_DR_DETAILS)) 711 fprintf (vect_dump, "dependence distance negative."); 712 continue; 713 } 714 715 if (abs (dist) >= 2 716 && abs (dist) < *max_vf) 717 { 718 /* The dependence distance requires reduction of the maximal 719 vectorization factor. */ 720 *max_vf = abs (dist); 721 if (vect_print_dump_info (REPORT_DR_DETAILS)) 722 fprintf (vect_dump, "adjusting maximal vectorization factor to %i", 723 *max_vf); 724 } 725 726 if (abs (dist) >= *max_vf) 727 { 728 /* Dependence distance does not create dependence, as far as 729 vectorization is concerned, in this case. */ 730 if (vect_print_dump_info (REPORT_DR_DETAILS)) 731 fprintf (vect_dump, "dependence distance >= VF."); 732 continue; 733 } 734 735 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 736 { 737 fprintf (vect_dump, "not vectorized, possible dependence " 738 "between data-refs "); 739 print_generic_expr (vect_dump, DR_REF (dra), TDF_SLIM); 740 fprintf (vect_dump, " and "); 741 print_generic_expr (vect_dump, DR_REF (drb), TDF_SLIM); 742 } 743 744 return true; 745 } 746 747 return false; 748 } 749 750 /* Function vect_analyze_data_ref_dependences. 751 752 Examine all the data references in the loop, and make sure there do not 753 exist any data dependences between them. Set *MAX_VF according to 754 the maximum vectorization factor the data dependences allow. */ 755 756 bool 757 vect_analyze_data_ref_dependences (loop_vec_info loop_vinfo, 758 bb_vec_info bb_vinfo, int *max_vf) 759 { 760 unsigned int i; 761 VEC (ddr_p, heap) *ddrs = NULL; 762 struct data_dependence_relation *ddr; 763 764 if (vect_print_dump_info (REPORT_DETAILS)) 765 fprintf (vect_dump, "=== vect_analyze_dependences ==="); 766 767 if (loop_vinfo) 768 ddrs = LOOP_VINFO_DDRS (loop_vinfo); 769 else 770 ddrs = BB_VINFO_DDRS (bb_vinfo); 771 772 FOR_EACH_VEC_ELT (ddr_p, ddrs, i, ddr) 773 if (vect_analyze_data_ref_dependence (ddr, loop_vinfo, max_vf)) 774 return false; 775 776 return true; 777 } 778 779 780 /* Function vect_compute_data_ref_alignment 781 782 Compute the misalignment of the data reference DR. 783 784 Output: 785 1. If during the misalignment computation it is found that the data reference 786 cannot be vectorized then false is returned. 787 2. DR_MISALIGNMENT (DR) is defined. 788 789 FOR NOW: No analysis is actually performed. Misalignment is calculated 790 only for trivial cases. TODO. */ 791 792 static bool 793 vect_compute_data_ref_alignment (struct data_reference *dr) 794 { 795 gimple stmt = DR_STMT (dr); 796 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 797 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 798 struct loop *loop = NULL; 799 tree ref = DR_REF (dr); 800 tree vectype; 801 tree base, base_addr; 802 bool base_aligned; 803 tree misalign; 804 tree aligned_to, alignment; 805 806 if (vect_print_dump_info (REPORT_DETAILS)) 807 fprintf (vect_dump, "vect_compute_data_ref_alignment:"); 808 809 if (loop_vinfo) 810 loop = LOOP_VINFO_LOOP (loop_vinfo); 811 812 /* Initialize misalignment to unknown. */ 813 SET_DR_MISALIGNMENT (dr, -1); 814 815 misalign = DR_INIT (dr); 816 aligned_to = DR_ALIGNED_TO (dr); 817 base_addr = DR_BASE_ADDRESS (dr); 818 vectype = STMT_VINFO_VECTYPE (stmt_info); 819 820 /* In case the dataref is in an inner-loop of the loop that is being 821 vectorized (LOOP), we use the base and misalignment information 822 relative to the outer-loop (LOOP). This is ok only if the misalignment 823 stays the same throughout the execution of the inner-loop, which is why 824 we have to check that the stride of the dataref in the inner-loop evenly 825 divides by the vector size. */ 826 if (loop && nested_in_vect_loop_p (loop, stmt)) 827 { 828 tree step = DR_STEP (dr); 829 HOST_WIDE_INT dr_step = TREE_INT_CST_LOW (step); 830 831 if (dr_step % GET_MODE_SIZE (TYPE_MODE (vectype)) == 0) 832 { 833 if (vect_print_dump_info (REPORT_ALIGNMENT)) 834 fprintf (vect_dump, "inner step divides the vector-size."); 835 misalign = STMT_VINFO_DR_INIT (stmt_info); 836 aligned_to = STMT_VINFO_DR_ALIGNED_TO (stmt_info); 837 base_addr = STMT_VINFO_DR_BASE_ADDRESS (stmt_info); 838 } 839 else 840 { 841 if (vect_print_dump_info (REPORT_ALIGNMENT)) 842 fprintf (vect_dump, "inner step doesn't divide the vector-size."); 843 misalign = NULL_TREE; 844 } 845 } 846 847 base = build_fold_indirect_ref (base_addr); 848 alignment = ssize_int (TYPE_ALIGN (vectype)/BITS_PER_UNIT); 849 850 if ((aligned_to && tree_int_cst_compare (aligned_to, alignment) < 0) 851 || !misalign) 852 { 853 if (vect_print_dump_info (REPORT_ALIGNMENT)) 854 { 855 fprintf (vect_dump, "Unknown alignment for access: "); 856 print_generic_expr (vect_dump, base, TDF_SLIM); 857 } 858 return true; 859 } 860 861 if ((DECL_P (base) 862 && tree_int_cst_compare (ssize_int (DECL_ALIGN_UNIT (base)), 863 alignment) >= 0) 864 || (TREE_CODE (base_addr) == SSA_NAME 865 && tree_int_cst_compare (ssize_int (TYPE_ALIGN_UNIT (TREE_TYPE ( 866 TREE_TYPE (base_addr)))), 867 alignment) >= 0) 868 || (get_pointer_alignment (base_addr) >= TYPE_ALIGN (vectype))) 869 base_aligned = true; 870 else 871 base_aligned = false; 872 873 if (!base_aligned) 874 { 875 /* Do not change the alignment of global variables if 876 flag_section_anchors is enabled. */ 877 if (!vect_can_force_dr_alignment_p (base, TYPE_ALIGN (vectype)) 878 || (TREE_STATIC (base) && flag_section_anchors)) 879 { 880 if (vect_print_dump_info (REPORT_DETAILS)) 881 { 882 fprintf (vect_dump, "can't force alignment of ref: "); 883 print_generic_expr (vect_dump, ref, TDF_SLIM); 884 } 885 return true; 886 } 887 888 /* Force the alignment of the decl. 889 NOTE: This is the only change to the code we make during 890 the analysis phase, before deciding to vectorize the loop. */ 891 if (vect_print_dump_info (REPORT_DETAILS)) 892 { 893 fprintf (vect_dump, "force alignment of "); 894 print_generic_expr (vect_dump, ref, TDF_SLIM); 895 } 896 897 DECL_ALIGN (base) = TYPE_ALIGN (vectype); 898 DECL_USER_ALIGN (base) = 1; 899 } 900 901 /* At this point we assume that the base is aligned. */ 902 gcc_assert (base_aligned 903 || (TREE_CODE (base) == VAR_DECL 904 && DECL_ALIGN (base) >= TYPE_ALIGN (vectype))); 905 906 /* If this is a backward running DR then first access in the larger 907 vectype actually is N-1 elements before the address in the DR. 908 Adjust misalign accordingly. */ 909 if (tree_int_cst_compare (DR_STEP (dr), size_zero_node) < 0) 910 { 911 tree offset = ssize_int (TYPE_VECTOR_SUBPARTS (vectype) - 1); 912 /* DR_STEP(dr) is the same as -TYPE_SIZE of the scalar type, 913 otherwise we wouldn't be here. */ 914 offset = fold_build2 (MULT_EXPR, ssizetype, offset, DR_STEP (dr)); 915 /* PLUS because DR_STEP was negative. */ 916 misalign = size_binop (PLUS_EXPR, misalign, offset); 917 } 918 919 /* Modulo alignment. */ 920 misalign = size_binop (FLOOR_MOD_EXPR, misalign, alignment); 921 922 if (!host_integerp (misalign, 1)) 923 { 924 /* Negative or overflowed misalignment value. */ 925 if (vect_print_dump_info (REPORT_DETAILS)) 926 fprintf (vect_dump, "unexpected misalign value"); 927 return false; 928 } 929 930 SET_DR_MISALIGNMENT (dr, TREE_INT_CST_LOW (misalign)); 931 932 if (vect_print_dump_info (REPORT_DETAILS)) 933 { 934 fprintf (vect_dump, "misalign = %d bytes of ref ", DR_MISALIGNMENT (dr)); 935 print_generic_expr (vect_dump, ref, TDF_SLIM); 936 } 937 938 return true; 939 } 940 941 942 /* Function vect_compute_data_refs_alignment 943 944 Compute the misalignment of data references in the loop. 945 Return FALSE if a data reference is found that cannot be vectorized. */ 946 947 static bool 948 vect_compute_data_refs_alignment (loop_vec_info loop_vinfo, 949 bb_vec_info bb_vinfo) 950 { 951 VEC (data_reference_p, heap) *datarefs; 952 struct data_reference *dr; 953 unsigned int i; 954 955 if (loop_vinfo) 956 datarefs = LOOP_VINFO_DATAREFS (loop_vinfo); 957 else 958 datarefs = BB_VINFO_DATAREFS (bb_vinfo); 959 960 FOR_EACH_VEC_ELT (data_reference_p, datarefs, i, dr) 961 if (STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (DR_STMT (dr))) 962 && !vect_compute_data_ref_alignment (dr)) 963 { 964 if (bb_vinfo) 965 { 966 /* Mark unsupported statement as unvectorizable. */ 967 STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (DR_STMT (dr))) = false; 968 continue; 969 } 970 else 971 return false; 972 } 973 974 return true; 975 } 976 977 978 /* Function vect_update_misalignment_for_peel 979 980 DR - the data reference whose misalignment is to be adjusted. 981 DR_PEEL - the data reference whose misalignment is being made 982 zero in the vector loop by the peel. 983 NPEEL - the number of iterations in the peel loop if the misalignment 984 of DR_PEEL is known at compile time. */ 985 986 static void 987 vect_update_misalignment_for_peel (struct data_reference *dr, 988 struct data_reference *dr_peel, int npeel) 989 { 990 unsigned int i; 991 VEC(dr_p,heap) *same_align_drs; 992 struct data_reference *current_dr; 993 int dr_size = GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (dr)))); 994 int dr_peel_size = GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (dr_peel)))); 995 stmt_vec_info stmt_info = vinfo_for_stmt (DR_STMT (dr)); 996 stmt_vec_info peel_stmt_info = vinfo_for_stmt (DR_STMT (dr_peel)); 997 998 /* For interleaved data accesses the step in the loop must be multiplied by 999 the size of the interleaving group. */ 1000 if (STMT_VINFO_STRIDED_ACCESS (stmt_info)) 1001 dr_size *= GROUP_SIZE (vinfo_for_stmt (GROUP_FIRST_ELEMENT (stmt_info))); 1002 if (STMT_VINFO_STRIDED_ACCESS (peel_stmt_info)) 1003 dr_peel_size *= GROUP_SIZE (peel_stmt_info); 1004 1005 /* It can be assumed that the data refs with the same alignment as dr_peel 1006 are aligned in the vector loop. */ 1007 same_align_drs 1008 = STMT_VINFO_SAME_ALIGN_REFS (vinfo_for_stmt (DR_STMT (dr_peel))); 1009 FOR_EACH_VEC_ELT (dr_p, same_align_drs, i, current_dr) 1010 { 1011 if (current_dr != dr) 1012 continue; 1013 gcc_assert (DR_MISALIGNMENT (dr) / dr_size == 1014 DR_MISALIGNMENT (dr_peel) / dr_peel_size); 1015 SET_DR_MISALIGNMENT (dr, 0); 1016 return; 1017 } 1018 1019 if (known_alignment_for_access_p (dr) 1020 && known_alignment_for_access_p (dr_peel)) 1021 { 1022 bool negative = tree_int_cst_compare (DR_STEP (dr), size_zero_node) < 0; 1023 int misal = DR_MISALIGNMENT (dr); 1024 tree vectype = STMT_VINFO_VECTYPE (stmt_info); 1025 misal += negative ? -npeel * dr_size : npeel * dr_size; 1026 misal &= (TYPE_ALIGN (vectype) / BITS_PER_UNIT) - 1; 1027 SET_DR_MISALIGNMENT (dr, misal); 1028 return; 1029 } 1030 1031 if (vect_print_dump_info (REPORT_DETAILS)) 1032 fprintf (vect_dump, "Setting misalignment to -1."); 1033 SET_DR_MISALIGNMENT (dr, -1); 1034 } 1035 1036 1037 /* Function vect_verify_datarefs_alignment 1038 1039 Return TRUE if all data references in the loop can be 1040 handled with respect to alignment. */ 1041 1042 bool 1043 vect_verify_datarefs_alignment (loop_vec_info loop_vinfo, bb_vec_info bb_vinfo) 1044 { 1045 VEC (data_reference_p, heap) *datarefs; 1046 struct data_reference *dr; 1047 enum dr_alignment_support supportable_dr_alignment; 1048 unsigned int i; 1049 1050 if (loop_vinfo) 1051 datarefs = LOOP_VINFO_DATAREFS (loop_vinfo); 1052 else 1053 datarefs = BB_VINFO_DATAREFS (bb_vinfo); 1054 1055 FOR_EACH_VEC_ELT (data_reference_p, datarefs, i, dr) 1056 { 1057 gimple stmt = DR_STMT (dr); 1058 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 1059 1060 /* For interleaving, only the alignment of the first access matters. 1061 Skip statements marked as not vectorizable. */ 1062 if ((STMT_VINFO_STRIDED_ACCESS (stmt_info) 1063 && GROUP_FIRST_ELEMENT (stmt_info) != stmt) 1064 || !STMT_VINFO_VECTORIZABLE (stmt_info)) 1065 continue; 1066 1067 supportable_dr_alignment = vect_supportable_dr_alignment (dr, false); 1068 if (!supportable_dr_alignment) 1069 { 1070 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 1071 { 1072 if (DR_IS_READ (dr)) 1073 fprintf (vect_dump, 1074 "not vectorized: unsupported unaligned load."); 1075 else 1076 fprintf (vect_dump, 1077 "not vectorized: unsupported unaligned store."); 1078 1079 print_generic_expr (vect_dump, DR_REF (dr), TDF_SLIM); 1080 } 1081 return false; 1082 } 1083 if (supportable_dr_alignment != dr_aligned 1084 && vect_print_dump_info (REPORT_ALIGNMENT)) 1085 fprintf (vect_dump, "Vectorizing an unaligned access."); 1086 } 1087 return true; 1088 } 1089 1090 1091 /* Function vector_alignment_reachable_p 1092 1093 Return true if vector alignment for DR is reachable by peeling 1094 a few loop iterations. Return false otherwise. */ 1095 1096 static bool 1097 vector_alignment_reachable_p (struct data_reference *dr) 1098 { 1099 gimple stmt = DR_STMT (dr); 1100 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 1101 tree vectype = STMT_VINFO_VECTYPE (stmt_info); 1102 1103 if (STMT_VINFO_STRIDED_ACCESS (stmt_info)) 1104 { 1105 /* For interleaved access we peel only if number of iterations in 1106 the prolog loop ({VF - misalignment}), is a multiple of the 1107 number of the interleaved accesses. */ 1108 int elem_size, mis_in_elements; 1109 int nelements = TYPE_VECTOR_SUBPARTS (vectype); 1110 1111 /* FORNOW: handle only known alignment. */ 1112 if (!known_alignment_for_access_p (dr)) 1113 return false; 1114 1115 elem_size = GET_MODE_SIZE (TYPE_MODE (vectype)) / nelements; 1116 mis_in_elements = DR_MISALIGNMENT (dr) / elem_size; 1117 1118 if ((nelements - mis_in_elements) % GROUP_SIZE (stmt_info)) 1119 return false; 1120 } 1121 1122 /* If misalignment is known at the compile time then allow peeling 1123 only if natural alignment is reachable through peeling. */ 1124 if (known_alignment_for_access_p (dr) && !aligned_access_p (dr)) 1125 { 1126 HOST_WIDE_INT elmsize = 1127 int_cst_value (TYPE_SIZE_UNIT (TREE_TYPE (vectype))); 1128 if (vect_print_dump_info (REPORT_DETAILS)) 1129 { 1130 fprintf (vect_dump, "data size =" HOST_WIDE_INT_PRINT_DEC, elmsize); 1131 fprintf (vect_dump, ". misalignment = %d. ", DR_MISALIGNMENT (dr)); 1132 } 1133 if (DR_MISALIGNMENT (dr) % elmsize) 1134 { 1135 if (vect_print_dump_info (REPORT_DETAILS)) 1136 fprintf (vect_dump, "data size does not divide the misalignment.\n"); 1137 return false; 1138 } 1139 } 1140 1141 if (!known_alignment_for_access_p (dr)) 1142 { 1143 tree type = (TREE_TYPE (DR_REF (dr))); 1144 bool is_packed = contains_packed_reference (DR_REF (dr)); 1145 1146 if (compare_tree_int (TYPE_SIZE (type), TYPE_ALIGN (type)) > 0) 1147 is_packed = true; 1148 1149 if (vect_print_dump_info (REPORT_DETAILS)) 1150 fprintf (vect_dump, "Unknown misalignment, is_packed = %d",is_packed); 1151 if (targetm.vectorize.vector_alignment_reachable (type, is_packed)) 1152 return true; 1153 else 1154 return false; 1155 } 1156 1157 return true; 1158 } 1159 1160 1161 /* Calculate the cost of the memory access represented by DR. */ 1162 1163 static void 1164 vect_get_data_access_cost (struct data_reference *dr, 1165 unsigned int *inside_cost, 1166 unsigned int *outside_cost) 1167 { 1168 gimple stmt = DR_STMT (dr); 1169 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 1170 int nunits = TYPE_VECTOR_SUBPARTS (STMT_VINFO_VECTYPE (stmt_info)); 1171 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 1172 int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); 1173 int ncopies = vf / nunits; 1174 bool supportable_dr_alignment = vect_supportable_dr_alignment (dr, true); 1175 1176 if (!supportable_dr_alignment) 1177 *inside_cost = VECT_MAX_COST; 1178 else 1179 { 1180 if (DR_IS_READ (dr)) 1181 vect_get_load_cost (dr, ncopies, true, inside_cost, outside_cost); 1182 else 1183 vect_get_store_cost (dr, ncopies, inside_cost); 1184 } 1185 1186 if (vect_print_dump_info (REPORT_COST)) 1187 fprintf (vect_dump, "vect_get_data_access_cost: inside_cost = %d, " 1188 "outside_cost = %d.", *inside_cost, *outside_cost); 1189 } 1190 1191 1192 static hashval_t 1193 vect_peeling_hash (const void *elem) 1194 { 1195 const struct _vect_peel_info *peel_info; 1196 1197 peel_info = (const struct _vect_peel_info *) elem; 1198 return (hashval_t) peel_info->npeel; 1199 } 1200 1201 1202 static int 1203 vect_peeling_hash_eq (const void *elem1, const void *elem2) 1204 { 1205 const struct _vect_peel_info *a, *b; 1206 1207 a = (const struct _vect_peel_info *) elem1; 1208 b = (const struct _vect_peel_info *) elem2; 1209 return (a->npeel == b->npeel); 1210 } 1211 1212 1213 /* Insert DR into peeling hash table with NPEEL as key. */ 1214 1215 static void 1216 vect_peeling_hash_insert (loop_vec_info loop_vinfo, struct data_reference *dr, 1217 int npeel) 1218 { 1219 struct _vect_peel_info elem, *slot; 1220 void **new_slot; 1221 bool supportable_dr_alignment = vect_supportable_dr_alignment (dr, true); 1222 1223 elem.npeel = npeel; 1224 slot = (vect_peel_info) htab_find (LOOP_VINFO_PEELING_HTAB (loop_vinfo), 1225 &elem); 1226 if (slot) 1227 slot->count++; 1228 else 1229 { 1230 slot = XNEW (struct _vect_peel_info); 1231 slot->npeel = npeel; 1232 slot->dr = dr; 1233 slot->count = 1; 1234 new_slot = htab_find_slot (LOOP_VINFO_PEELING_HTAB (loop_vinfo), slot, 1235 INSERT); 1236 *new_slot = slot; 1237 } 1238 1239 if (!supportable_dr_alignment && !flag_vect_cost_model) 1240 slot->count += VECT_MAX_COST; 1241 } 1242 1243 1244 /* Traverse peeling hash table to find peeling option that aligns maximum 1245 number of data accesses. */ 1246 1247 static int 1248 vect_peeling_hash_get_most_frequent (void **slot, void *data) 1249 { 1250 vect_peel_info elem = (vect_peel_info) *slot; 1251 vect_peel_extended_info max = (vect_peel_extended_info) data; 1252 1253 if (elem->count > max->peel_info.count 1254 || (elem->count == max->peel_info.count 1255 && max->peel_info.npeel > elem->npeel)) 1256 { 1257 max->peel_info.npeel = elem->npeel; 1258 max->peel_info.count = elem->count; 1259 max->peel_info.dr = elem->dr; 1260 } 1261 1262 return 1; 1263 } 1264 1265 1266 /* Traverse peeling hash table and calculate cost for each peeling option. 1267 Find the one with the lowest cost. */ 1268 1269 static int 1270 vect_peeling_hash_get_lowest_cost (void **slot, void *data) 1271 { 1272 vect_peel_info elem = (vect_peel_info) *slot; 1273 vect_peel_extended_info min = (vect_peel_extended_info) data; 1274 int save_misalignment, dummy; 1275 unsigned int inside_cost = 0, outside_cost = 0, i; 1276 gimple stmt = DR_STMT (elem->dr); 1277 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 1278 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 1279 VEC (data_reference_p, heap) *datarefs = LOOP_VINFO_DATAREFS (loop_vinfo); 1280 struct data_reference *dr; 1281 1282 FOR_EACH_VEC_ELT (data_reference_p, datarefs, i, dr) 1283 { 1284 stmt = DR_STMT (dr); 1285 stmt_info = vinfo_for_stmt (stmt); 1286 /* For interleaving, only the alignment of the first access 1287 matters. */ 1288 if (STMT_VINFO_STRIDED_ACCESS (stmt_info) 1289 && GROUP_FIRST_ELEMENT (stmt_info) != stmt) 1290 continue; 1291 1292 save_misalignment = DR_MISALIGNMENT (dr); 1293 vect_update_misalignment_for_peel (dr, elem->dr, elem->npeel); 1294 vect_get_data_access_cost (dr, &inside_cost, &outside_cost); 1295 SET_DR_MISALIGNMENT (dr, save_misalignment); 1296 } 1297 1298 outside_cost += vect_get_known_peeling_cost (loop_vinfo, elem->npeel, &dummy, 1299 vect_get_single_scalar_iteration_cost (loop_vinfo)); 1300 1301 if (inside_cost < min->inside_cost 1302 || (inside_cost == min->inside_cost && outside_cost < min->outside_cost)) 1303 { 1304 min->inside_cost = inside_cost; 1305 min->outside_cost = outside_cost; 1306 min->peel_info.dr = elem->dr; 1307 min->peel_info.npeel = elem->npeel; 1308 } 1309 1310 return 1; 1311 } 1312 1313 1314 /* Choose best peeling option by traversing peeling hash table and either 1315 choosing an option with the lowest cost (if cost model is enabled) or the 1316 option that aligns as many accesses as possible. */ 1317 1318 static struct data_reference * 1319 vect_peeling_hash_choose_best_peeling (loop_vec_info loop_vinfo, 1320 unsigned int *npeel) 1321 { 1322 struct _vect_peel_extended_info res; 1323 1324 res.peel_info.dr = NULL; 1325 1326 if (flag_vect_cost_model) 1327 { 1328 res.inside_cost = INT_MAX; 1329 res.outside_cost = INT_MAX; 1330 htab_traverse (LOOP_VINFO_PEELING_HTAB (loop_vinfo), 1331 vect_peeling_hash_get_lowest_cost, &res); 1332 } 1333 else 1334 { 1335 res.peel_info.count = 0; 1336 htab_traverse (LOOP_VINFO_PEELING_HTAB (loop_vinfo), 1337 vect_peeling_hash_get_most_frequent, &res); 1338 } 1339 1340 *npeel = res.peel_info.npeel; 1341 return res.peel_info.dr; 1342 } 1343 1344 1345 /* Function vect_enhance_data_refs_alignment 1346 1347 This pass will use loop versioning and loop peeling in order to enhance 1348 the alignment of data references in the loop. 1349 1350 FOR NOW: we assume that whatever versioning/peeling takes place, only the 1351 original loop is to be vectorized. Any other loops that are created by 1352 the transformations performed in this pass - are not supposed to be 1353 vectorized. This restriction will be relaxed. 1354 1355 This pass will require a cost model to guide it whether to apply peeling 1356 or versioning or a combination of the two. For example, the scheme that 1357 intel uses when given a loop with several memory accesses, is as follows: 1358 choose one memory access ('p') which alignment you want to force by doing 1359 peeling. Then, either (1) generate a loop in which 'p' is aligned and all 1360 other accesses are not necessarily aligned, or (2) use loop versioning to 1361 generate one loop in which all accesses are aligned, and another loop in 1362 which only 'p' is necessarily aligned. 1363 1364 ("Automatic Intra-Register Vectorization for the Intel Architecture", 1365 Aart J.C. Bik, Milind Girkar, Paul M. Grey and Ximmin Tian, International 1366 Journal of Parallel Programming, Vol. 30, No. 2, April 2002.) 1367 1368 Devising a cost model is the most critical aspect of this work. It will 1369 guide us on which access to peel for, whether to use loop versioning, how 1370 many versions to create, etc. The cost model will probably consist of 1371 generic considerations as well as target specific considerations (on 1372 powerpc for example, misaligned stores are more painful than misaligned 1373 loads). 1374 1375 Here are the general steps involved in alignment enhancements: 1376 1377 -- original loop, before alignment analysis: 1378 for (i=0; i<N; i++){ 1379 x = q[i]; # DR_MISALIGNMENT(q) = unknown 1380 p[i] = y; # DR_MISALIGNMENT(p) = unknown 1381 } 1382 1383 -- After vect_compute_data_refs_alignment: 1384 for (i=0; i<N; i++){ 1385 x = q[i]; # DR_MISALIGNMENT(q) = 3 1386 p[i] = y; # DR_MISALIGNMENT(p) = unknown 1387 } 1388 1389 -- Possibility 1: we do loop versioning: 1390 if (p is aligned) { 1391 for (i=0; i<N; i++){ # loop 1A 1392 x = q[i]; # DR_MISALIGNMENT(q) = 3 1393 p[i] = y; # DR_MISALIGNMENT(p) = 0 1394 } 1395 } 1396 else { 1397 for (i=0; i<N; i++){ # loop 1B 1398 x = q[i]; # DR_MISALIGNMENT(q) = 3 1399 p[i] = y; # DR_MISALIGNMENT(p) = unaligned 1400 } 1401 } 1402 1403 -- Possibility 2: we do loop peeling: 1404 for (i = 0; i < 3; i++){ # (scalar loop, not to be vectorized). 1405 x = q[i]; 1406 p[i] = y; 1407 } 1408 for (i = 3; i < N; i++){ # loop 2A 1409 x = q[i]; # DR_MISALIGNMENT(q) = 0 1410 p[i] = y; # DR_MISALIGNMENT(p) = unknown 1411 } 1412 1413 -- Possibility 3: combination of loop peeling and versioning: 1414 for (i = 0; i < 3; i++){ # (scalar loop, not to be vectorized). 1415 x = q[i]; 1416 p[i] = y; 1417 } 1418 if (p is aligned) { 1419 for (i = 3; i<N; i++){ # loop 3A 1420 x = q[i]; # DR_MISALIGNMENT(q) = 0 1421 p[i] = y; # DR_MISALIGNMENT(p) = 0 1422 } 1423 } 1424 else { 1425 for (i = 3; i<N; i++){ # loop 3B 1426 x = q[i]; # DR_MISALIGNMENT(q) = 0 1427 p[i] = y; # DR_MISALIGNMENT(p) = unaligned 1428 } 1429 } 1430 1431 These loops are later passed to loop_transform to be vectorized. The 1432 vectorizer will use the alignment information to guide the transformation 1433 (whether to generate regular loads/stores, or with special handling for 1434 misalignment). */ 1435 1436 bool 1437 vect_enhance_data_refs_alignment (loop_vec_info loop_vinfo) 1438 { 1439 VEC (data_reference_p, heap) *datarefs = LOOP_VINFO_DATAREFS (loop_vinfo); 1440 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 1441 enum dr_alignment_support supportable_dr_alignment; 1442 struct data_reference *dr0 = NULL, *first_store = NULL; 1443 struct data_reference *dr; 1444 unsigned int i, j; 1445 bool do_peeling = false; 1446 bool do_versioning = false; 1447 bool stat; 1448 gimple stmt; 1449 stmt_vec_info stmt_info; 1450 int vect_versioning_for_alias_required; 1451 unsigned int npeel = 0; 1452 bool all_misalignments_unknown = true; 1453 unsigned int vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo); 1454 unsigned possible_npeel_number = 1; 1455 tree vectype; 1456 unsigned int nelements, mis, same_align_drs_max = 0; 1457 1458 if (vect_print_dump_info (REPORT_DETAILS)) 1459 fprintf (vect_dump, "=== vect_enhance_data_refs_alignment ==="); 1460 1461 /* While cost model enhancements are expected in the future, the high level 1462 view of the code at this time is as follows: 1463 1464 A) If there is a misaligned access then see if peeling to align 1465 this access can make all data references satisfy 1466 vect_supportable_dr_alignment. If so, update data structures 1467 as needed and return true. 1468 1469 B) If peeling wasn't possible and there is a data reference with an 1470 unknown misalignment that does not satisfy vect_supportable_dr_alignment 1471 then see if loop versioning checks can be used to make all data 1472 references satisfy vect_supportable_dr_alignment. If so, update 1473 data structures as needed and return true. 1474 1475 C) If neither peeling nor versioning were successful then return false if 1476 any data reference does not satisfy vect_supportable_dr_alignment. 1477 1478 D) Return true (all data references satisfy vect_supportable_dr_alignment). 1479 1480 Note, Possibility 3 above (which is peeling and versioning together) is not 1481 being done at this time. */ 1482 1483 /* (1) Peeling to force alignment. */ 1484 1485 /* (1.1) Decide whether to perform peeling, and how many iterations to peel: 1486 Considerations: 1487 + How many accesses will become aligned due to the peeling 1488 - How many accesses will become unaligned due to the peeling, 1489 and the cost of misaligned accesses. 1490 - The cost of peeling (the extra runtime checks, the increase 1491 in code size). */ 1492 1493 FOR_EACH_VEC_ELT (data_reference_p, datarefs, i, dr) 1494 { 1495 stmt = DR_STMT (dr); 1496 stmt_info = vinfo_for_stmt (stmt); 1497 1498 if (!STMT_VINFO_RELEVANT (stmt_info)) 1499 continue; 1500 1501 /* For interleaving, only the alignment of the first access 1502 matters. */ 1503 if (STMT_VINFO_STRIDED_ACCESS (stmt_info) 1504 && GROUP_FIRST_ELEMENT (stmt_info) != stmt) 1505 continue; 1506 1507 /* For invariant accesses there is nothing to enhance. */ 1508 if (integer_zerop (DR_STEP (dr))) 1509 continue; 1510 1511 supportable_dr_alignment = vect_supportable_dr_alignment (dr, true); 1512 do_peeling = vector_alignment_reachable_p (dr); 1513 if (do_peeling) 1514 { 1515 if (known_alignment_for_access_p (dr)) 1516 { 1517 unsigned int npeel_tmp; 1518 bool negative = tree_int_cst_compare (DR_STEP (dr), 1519 size_zero_node) < 0; 1520 1521 /* Save info about DR in the hash table. */ 1522 if (!LOOP_VINFO_PEELING_HTAB (loop_vinfo)) 1523 LOOP_VINFO_PEELING_HTAB (loop_vinfo) = 1524 htab_create (1, vect_peeling_hash, 1525 vect_peeling_hash_eq, free); 1526 1527 vectype = STMT_VINFO_VECTYPE (stmt_info); 1528 nelements = TYPE_VECTOR_SUBPARTS (vectype); 1529 mis = DR_MISALIGNMENT (dr) / GET_MODE_SIZE (TYPE_MODE ( 1530 TREE_TYPE (DR_REF (dr)))); 1531 npeel_tmp = (negative 1532 ? (mis - nelements) : (nelements - mis)) 1533 & (nelements - 1); 1534 1535 /* For multiple types, it is possible that the bigger type access 1536 will have more than one peeling option. E.g., a loop with two 1537 types: one of size (vector size / 4), and the other one of 1538 size (vector size / 8). Vectorization factor will 8. If both 1539 access are misaligned by 3, the first one needs one scalar 1540 iteration to be aligned, and the second one needs 5. But the 1541 the first one will be aligned also by peeling 5 scalar 1542 iterations, and in that case both accesses will be aligned. 1543 Hence, except for the immediate peeling amount, we also want 1544 to try to add full vector size, while we don't exceed 1545 vectorization factor. 1546 We do this automtically for cost model, since we calculate cost 1547 for every peeling option. */ 1548 if (!flag_vect_cost_model) 1549 possible_npeel_number = vf /nelements; 1550 1551 /* Handle the aligned case. We may decide to align some other 1552 access, making DR unaligned. */ 1553 if (DR_MISALIGNMENT (dr) == 0) 1554 { 1555 npeel_tmp = 0; 1556 if (!flag_vect_cost_model) 1557 possible_npeel_number++; 1558 } 1559 1560 for (j = 0; j < possible_npeel_number; j++) 1561 { 1562 gcc_assert (npeel_tmp <= vf); 1563 vect_peeling_hash_insert (loop_vinfo, dr, npeel_tmp); 1564 npeel_tmp += nelements; 1565 } 1566 1567 all_misalignments_unknown = false; 1568 /* Data-ref that was chosen for the case that all the 1569 misalignments are unknown is not relevant anymore, since we 1570 have a data-ref with known alignment. */ 1571 dr0 = NULL; 1572 } 1573 else 1574 { 1575 /* If we don't know all the misalignment values, we prefer 1576 peeling for data-ref that has maximum number of data-refs 1577 with the same alignment, unless the target prefers to align 1578 stores over load. */ 1579 if (all_misalignments_unknown) 1580 { 1581 if (same_align_drs_max < VEC_length (dr_p, 1582 STMT_VINFO_SAME_ALIGN_REFS (stmt_info)) 1583 || !dr0) 1584 { 1585 same_align_drs_max = VEC_length (dr_p, 1586 STMT_VINFO_SAME_ALIGN_REFS (stmt_info)); 1587 dr0 = dr; 1588 } 1589 1590 if (!first_store && DR_IS_WRITE (dr)) 1591 first_store = dr; 1592 } 1593 1594 /* If there are both known and unknown misaligned accesses in the 1595 loop, we choose peeling amount according to the known 1596 accesses. */ 1597 1598 1599 if (!supportable_dr_alignment) 1600 { 1601 dr0 = dr; 1602 if (!first_store && DR_IS_WRITE (dr)) 1603 first_store = dr; 1604 } 1605 } 1606 } 1607 else 1608 { 1609 if (!aligned_access_p (dr)) 1610 { 1611 if (vect_print_dump_info (REPORT_DETAILS)) 1612 fprintf (vect_dump, "vector alignment may not be reachable"); 1613 1614 break; 1615 } 1616 } 1617 } 1618 1619 vect_versioning_for_alias_required 1620 = LOOP_REQUIRES_VERSIONING_FOR_ALIAS (loop_vinfo); 1621 1622 /* Temporarily, if versioning for alias is required, we disable peeling 1623 until we support peeling and versioning. Often peeling for alignment 1624 will require peeling for loop-bound, which in turn requires that we 1625 know how to adjust the loop ivs after the loop. */ 1626 if (vect_versioning_for_alias_required 1627 || !vect_can_advance_ivs_p (loop_vinfo) 1628 || !slpeel_can_duplicate_loop_p (loop, single_exit (loop))) 1629 do_peeling = false; 1630 1631 if (do_peeling && all_misalignments_unknown 1632 && vect_supportable_dr_alignment (dr0, false)) 1633 { 1634 1635 /* Check if the target requires to prefer stores over loads, i.e., if 1636 misaligned stores are more expensive than misaligned loads (taking 1637 drs with same alignment into account). */ 1638 if (first_store && DR_IS_READ (dr0)) 1639 { 1640 unsigned int load_inside_cost = 0, load_outside_cost = 0; 1641 unsigned int store_inside_cost = 0, store_outside_cost = 0; 1642 unsigned int load_inside_penalty = 0, load_outside_penalty = 0; 1643 unsigned int store_inside_penalty = 0, store_outside_penalty = 0; 1644 1645 vect_get_data_access_cost (dr0, &load_inside_cost, 1646 &load_outside_cost); 1647 vect_get_data_access_cost (first_store, &store_inside_cost, 1648 &store_outside_cost); 1649 1650 /* Calculate the penalty for leaving FIRST_STORE unaligned (by 1651 aligning the load DR0). */ 1652 load_inside_penalty = store_inside_cost; 1653 load_outside_penalty = store_outside_cost; 1654 for (i = 0; VEC_iterate (dr_p, STMT_VINFO_SAME_ALIGN_REFS 1655 (vinfo_for_stmt (DR_STMT (first_store))), 1656 i, dr); 1657 i++) 1658 if (DR_IS_READ (dr)) 1659 { 1660 load_inside_penalty += load_inside_cost; 1661 load_outside_penalty += load_outside_cost; 1662 } 1663 else 1664 { 1665 load_inside_penalty += store_inside_cost; 1666 load_outside_penalty += store_outside_cost; 1667 } 1668 1669 /* Calculate the penalty for leaving DR0 unaligned (by 1670 aligning the FIRST_STORE). */ 1671 store_inside_penalty = load_inside_cost; 1672 store_outside_penalty = load_outside_cost; 1673 for (i = 0; VEC_iterate (dr_p, STMT_VINFO_SAME_ALIGN_REFS 1674 (vinfo_for_stmt (DR_STMT (dr0))), 1675 i, dr); 1676 i++) 1677 if (DR_IS_READ (dr)) 1678 { 1679 store_inside_penalty += load_inside_cost; 1680 store_outside_penalty += load_outside_cost; 1681 } 1682 else 1683 { 1684 store_inside_penalty += store_inside_cost; 1685 store_outside_penalty += store_outside_cost; 1686 } 1687 1688 if (load_inside_penalty > store_inside_penalty 1689 || (load_inside_penalty == store_inside_penalty 1690 && load_outside_penalty > store_outside_penalty)) 1691 dr0 = first_store; 1692 } 1693 1694 /* In case there are only loads with different unknown misalignments, use 1695 peeling only if it may help to align other accesses in the loop. */ 1696 if (!first_store && !VEC_length (dr_p, STMT_VINFO_SAME_ALIGN_REFS 1697 (vinfo_for_stmt (DR_STMT (dr0)))) 1698 && vect_supportable_dr_alignment (dr0, false) 1699 != dr_unaligned_supported) 1700 do_peeling = false; 1701 } 1702 1703 if (do_peeling && !dr0) 1704 { 1705 /* Peeling is possible, but there is no data access that is not supported 1706 unless aligned. So we try to choose the best possible peeling. */ 1707 1708 /* We should get here only if there are drs with known misalignment. */ 1709 gcc_assert (!all_misalignments_unknown); 1710 1711 /* Choose the best peeling from the hash table. */ 1712 dr0 = vect_peeling_hash_choose_best_peeling (loop_vinfo, &npeel); 1713 if (!dr0 || !npeel) 1714 do_peeling = false; 1715 } 1716 1717 if (do_peeling) 1718 { 1719 stmt = DR_STMT (dr0); 1720 stmt_info = vinfo_for_stmt (stmt); 1721 vectype = STMT_VINFO_VECTYPE (stmt_info); 1722 nelements = TYPE_VECTOR_SUBPARTS (vectype); 1723 1724 if (known_alignment_for_access_p (dr0)) 1725 { 1726 bool negative = tree_int_cst_compare (DR_STEP (dr0), 1727 size_zero_node) < 0; 1728 if (!npeel) 1729 { 1730 /* Since it's known at compile time, compute the number of 1731 iterations in the peeled loop (the peeling factor) for use in 1732 updating DR_MISALIGNMENT values. The peeling factor is the 1733 vectorization factor minus the misalignment as an element 1734 count. */ 1735 mis = DR_MISALIGNMENT (dr0); 1736 mis /= GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (dr0)))); 1737 npeel = ((negative ? mis - nelements : nelements - mis) 1738 & (nelements - 1)); 1739 } 1740 1741 /* For interleaved data access every iteration accesses all the 1742 members of the group, therefore we divide the number of iterations 1743 by the group size. */ 1744 stmt_info = vinfo_for_stmt (DR_STMT (dr0)); 1745 if (STMT_VINFO_STRIDED_ACCESS (stmt_info)) 1746 npeel /= GROUP_SIZE (stmt_info); 1747 1748 if (vect_print_dump_info (REPORT_DETAILS)) 1749 fprintf (vect_dump, "Try peeling by %d", npeel); 1750 } 1751 1752 /* Ensure that all data refs can be vectorized after the peel. */ 1753 FOR_EACH_VEC_ELT (data_reference_p, datarefs, i, dr) 1754 { 1755 int save_misalignment; 1756 1757 if (dr == dr0) 1758 continue; 1759 1760 stmt = DR_STMT (dr); 1761 stmt_info = vinfo_for_stmt (stmt); 1762 /* For interleaving, only the alignment of the first access 1763 matters. */ 1764 if (STMT_VINFO_STRIDED_ACCESS (stmt_info) 1765 && GROUP_FIRST_ELEMENT (stmt_info) != stmt) 1766 continue; 1767 1768 save_misalignment = DR_MISALIGNMENT (dr); 1769 vect_update_misalignment_for_peel (dr, dr0, npeel); 1770 supportable_dr_alignment = vect_supportable_dr_alignment (dr, false); 1771 SET_DR_MISALIGNMENT (dr, save_misalignment); 1772 1773 if (!supportable_dr_alignment) 1774 { 1775 do_peeling = false; 1776 break; 1777 } 1778 } 1779 1780 if (do_peeling && known_alignment_for_access_p (dr0) && npeel == 0) 1781 { 1782 stat = vect_verify_datarefs_alignment (loop_vinfo, NULL); 1783 if (!stat) 1784 do_peeling = false; 1785 else 1786 return stat; 1787 } 1788 1789 if (do_peeling) 1790 { 1791 /* (1.2) Update the DR_MISALIGNMENT of each data reference DR_i. 1792 If the misalignment of DR_i is identical to that of dr0 then set 1793 DR_MISALIGNMENT (DR_i) to zero. If the misalignment of DR_i and 1794 dr0 are known at compile time then increment DR_MISALIGNMENT (DR_i) 1795 by the peeling factor times the element size of DR_i (MOD the 1796 vectorization factor times the size). Otherwise, the 1797 misalignment of DR_i must be set to unknown. */ 1798 FOR_EACH_VEC_ELT (data_reference_p, datarefs, i, dr) 1799 if (dr != dr0) 1800 vect_update_misalignment_for_peel (dr, dr0, npeel); 1801 1802 LOOP_VINFO_UNALIGNED_DR (loop_vinfo) = dr0; 1803 if (npeel) 1804 LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo) = npeel; 1805 else 1806 LOOP_PEELING_FOR_ALIGNMENT (loop_vinfo) = DR_MISALIGNMENT (dr0); 1807 SET_DR_MISALIGNMENT (dr0, 0); 1808 if (vect_print_dump_info (REPORT_ALIGNMENT)) 1809 fprintf (vect_dump, "Alignment of access forced using peeling."); 1810 1811 if (vect_print_dump_info (REPORT_DETAILS)) 1812 fprintf (vect_dump, "Peeling for alignment will be applied."); 1813 1814 stat = vect_verify_datarefs_alignment (loop_vinfo, NULL); 1815 gcc_assert (stat); 1816 return stat; 1817 } 1818 } 1819 1820 1821 /* (2) Versioning to force alignment. */ 1822 1823 /* Try versioning if: 1824 1) flag_tree_vect_loop_version is TRUE 1825 2) optimize loop for speed 1826 3) there is at least one unsupported misaligned data ref with an unknown 1827 misalignment, and 1828 4) all misaligned data refs with a known misalignment are supported, and 1829 5) the number of runtime alignment checks is within reason. */ 1830 1831 do_versioning = 1832 flag_tree_vect_loop_version 1833 && optimize_loop_nest_for_speed_p (loop) 1834 && (!loop->inner); /* FORNOW */ 1835 1836 if (do_versioning) 1837 { 1838 FOR_EACH_VEC_ELT (data_reference_p, datarefs, i, dr) 1839 { 1840 stmt = DR_STMT (dr); 1841 stmt_info = vinfo_for_stmt (stmt); 1842 1843 /* For interleaving, only the alignment of the first access 1844 matters. */ 1845 if (aligned_access_p (dr) 1846 || (STMT_VINFO_STRIDED_ACCESS (stmt_info) 1847 && GROUP_FIRST_ELEMENT (stmt_info) != stmt)) 1848 continue; 1849 1850 supportable_dr_alignment = vect_supportable_dr_alignment (dr, false); 1851 1852 if (!supportable_dr_alignment) 1853 { 1854 gimple stmt; 1855 int mask; 1856 tree vectype; 1857 1858 if (known_alignment_for_access_p (dr) 1859 || VEC_length (gimple, 1860 LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo)) 1861 >= (unsigned) PARAM_VALUE (PARAM_VECT_MAX_VERSION_FOR_ALIGNMENT_CHECKS)) 1862 { 1863 do_versioning = false; 1864 break; 1865 } 1866 1867 stmt = DR_STMT (dr); 1868 vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt)); 1869 gcc_assert (vectype); 1870 1871 /* The rightmost bits of an aligned address must be zeros. 1872 Construct the mask needed for this test. For example, 1873 GET_MODE_SIZE for the vector mode V4SI is 16 bytes so the 1874 mask must be 15 = 0xf. */ 1875 mask = GET_MODE_SIZE (TYPE_MODE (vectype)) - 1; 1876 1877 /* FORNOW: use the same mask to test all potentially unaligned 1878 references in the loop. The vectorizer currently supports 1879 a single vector size, see the reference to 1880 GET_MODE_NUNITS (TYPE_MODE (vectype)) where the 1881 vectorization factor is computed. */ 1882 gcc_assert (!LOOP_VINFO_PTR_MASK (loop_vinfo) 1883 || LOOP_VINFO_PTR_MASK (loop_vinfo) == mask); 1884 LOOP_VINFO_PTR_MASK (loop_vinfo) = mask; 1885 VEC_safe_push (gimple, heap, 1886 LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo), 1887 DR_STMT (dr)); 1888 } 1889 } 1890 1891 /* Versioning requires at least one misaligned data reference. */ 1892 if (!LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT (loop_vinfo)) 1893 do_versioning = false; 1894 else if (!do_versioning) 1895 VEC_truncate (gimple, LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo), 0); 1896 } 1897 1898 if (do_versioning) 1899 { 1900 VEC(gimple,heap) *may_misalign_stmts 1901 = LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo); 1902 gimple stmt; 1903 1904 /* It can now be assumed that the data references in the statements 1905 in LOOP_VINFO_MAY_MISALIGN_STMTS will be aligned in the version 1906 of the loop being vectorized. */ 1907 FOR_EACH_VEC_ELT (gimple, may_misalign_stmts, i, stmt) 1908 { 1909 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 1910 dr = STMT_VINFO_DATA_REF (stmt_info); 1911 SET_DR_MISALIGNMENT (dr, 0); 1912 if (vect_print_dump_info (REPORT_ALIGNMENT)) 1913 fprintf (vect_dump, "Alignment of access forced using versioning."); 1914 } 1915 1916 if (vect_print_dump_info (REPORT_DETAILS)) 1917 fprintf (vect_dump, "Versioning for alignment will be applied."); 1918 1919 /* Peeling and versioning can't be done together at this time. */ 1920 gcc_assert (! (do_peeling && do_versioning)); 1921 1922 stat = vect_verify_datarefs_alignment (loop_vinfo, NULL); 1923 gcc_assert (stat); 1924 return stat; 1925 } 1926 1927 /* This point is reached if neither peeling nor versioning is being done. */ 1928 gcc_assert (! (do_peeling || do_versioning)); 1929 1930 stat = vect_verify_datarefs_alignment (loop_vinfo, NULL); 1931 return stat; 1932 } 1933 1934 1935 /* Function vect_find_same_alignment_drs. 1936 1937 Update group and alignment relations according to the chosen 1938 vectorization factor. */ 1939 1940 static void 1941 vect_find_same_alignment_drs (struct data_dependence_relation *ddr, 1942 loop_vec_info loop_vinfo) 1943 { 1944 unsigned int i; 1945 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 1946 int vectorization_factor = LOOP_VINFO_VECT_FACTOR (loop_vinfo); 1947 struct data_reference *dra = DDR_A (ddr); 1948 struct data_reference *drb = DDR_B (ddr); 1949 stmt_vec_info stmtinfo_a = vinfo_for_stmt (DR_STMT (dra)); 1950 stmt_vec_info stmtinfo_b = vinfo_for_stmt (DR_STMT (drb)); 1951 int dra_size = GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (dra)))); 1952 int drb_size = GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (DR_REF (drb)))); 1953 lambda_vector dist_v; 1954 unsigned int loop_depth; 1955 1956 if (DDR_ARE_DEPENDENT (ddr) == chrec_known) 1957 return; 1958 1959 if (dra == drb) 1960 return; 1961 1962 if (DDR_ARE_DEPENDENT (ddr) == chrec_dont_know) 1963 return; 1964 1965 /* Loop-based vectorization and known data dependence. */ 1966 if (DDR_NUM_DIST_VECTS (ddr) == 0) 1967 return; 1968 1969 /* Data-dependence analysis reports a distance vector of zero 1970 for data-references that overlap only in the first iteration 1971 but have different sign step (see PR45764). 1972 So as a sanity check require equal DR_STEP. */ 1973 if (!operand_equal_p (DR_STEP (dra), DR_STEP (drb), 0)) 1974 return; 1975 1976 loop_depth = index_in_loop_nest (loop->num, DDR_LOOP_NEST (ddr)); 1977 FOR_EACH_VEC_ELT (lambda_vector, DDR_DIST_VECTS (ddr), i, dist_v) 1978 { 1979 int dist = dist_v[loop_depth]; 1980 1981 if (vect_print_dump_info (REPORT_DR_DETAILS)) 1982 fprintf (vect_dump, "dependence distance = %d.", dist); 1983 1984 /* Same loop iteration. */ 1985 if (dist == 0 1986 || (dist % vectorization_factor == 0 && dra_size == drb_size)) 1987 { 1988 /* Two references with distance zero have the same alignment. */ 1989 VEC_safe_push (dr_p, heap, STMT_VINFO_SAME_ALIGN_REFS (stmtinfo_a), drb); 1990 VEC_safe_push (dr_p, heap, STMT_VINFO_SAME_ALIGN_REFS (stmtinfo_b), dra); 1991 if (vect_print_dump_info (REPORT_ALIGNMENT)) 1992 fprintf (vect_dump, "accesses have the same alignment."); 1993 if (vect_print_dump_info (REPORT_DR_DETAILS)) 1994 { 1995 fprintf (vect_dump, "dependence distance modulo vf == 0 between "); 1996 print_generic_expr (vect_dump, DR_REF (dra), TDF_SLIM); 1997 fprintf (vect_dump, " and "); 1998 print_generic_expr (vect_dump, DR_REF (drb), TDF_SLIM); 1999 } 2000 } 2001 } 2002 } 2003 2004 2005 /* Function vect_analyze_data_refs_alignment 2006 2007 Analyze the alignment of the data-references in the loop. 2008 Return FALSE if a data reference is found that cannot be vectorized. */ 2009 2010 bool 2011 vect_analyze_data_refs_alignment (loop_vec_info loop_vinfo, 2012 bb_vec_info bb_vinfo) 2013 { 2014 if (vect_print_dump_info (REPORT_DETAILS)) 2015 fprintf (vect_dump, "=== vect_analyze_data_refs_alignment ==="); 2016 2017 /* Mark groups of data references with same alignment using 2018 data dependence information. */ 2019 if (loop_vinfo) 2020 { 2021 VEC (ddr_p, heap) *ddrs = LOOP_VINFO_DDRS (loop_vinfo); 2022 struct data_dependence_relation *ddr; 2023 unsigned int i; 2024 2025 FOR_EACH_VEC_ELT (ddr_p, ddrs, i, ddr) 2026 vect_find_same_alignment_drs (ddr, loop_vinfo); 2027 } 2028 2029 if (!vect_compute_data_refs_alignment (loop_vinfo, bb_vinfo)) 2030 { 2031 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2032 fprintf (vect_dump, 2033 "not vectorized: can't calculate alignment for data ref."); 2034 return false; 2035 } 2036 2037 return true; 2038 } 2039 2040 2041 /* Analyze groups of strided accesses: check that DR belongs to a group of 2042 strided accesses of legal size, step, etc. Detect gaps, single element 2043 interleaving, and other special cases. Set strided access info. 2044 Collect groups of strided stores for further use in SLP analysis. */ 2045 2046 static bool 2047 vect_analyze_group_access (struct data_reference *dr) 2048 { 2049 tree step = DR_STEP (dr); 2050 tree scalar_type = TREE_TYPE (DR_REF (dr)); 2051 HOST_WIDE_INT type_size = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (scalar_type)); 2052 gimple stmt = DR_STMT (dr); 2053 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 2054 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 2055 bb_vec_info bb_vinfo = STMT_VINFO_BB_VINFO (stmt_info); 2056 HOST_WIDE_INT dr_step = TREE_INT_CST_LOW (step); 2057 HOST_WIDE_INT stride, last_accessed_element = 1; 2058 bool slp_impossible = false; 2059 struct loop *loop = NULL; 2060 2061 if (loop_vinfo) 2062 loop = LOOP_VINFO_LOOP (loop_vinfo); 2063 2064 /* For interleaving, STRIDE is STEP counted in elements, i.e., the size of the 2065 interleaving group (including gaps). */ 2066 stride = dr_step / type_size; 2067 2068 /* Not consecutive access is possible only if it is a part of interleaving. */ 2069 if (!GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt))) 2070 { 2071 /* Check if it this DR is a part of interleaving, and is a single 2072 element of the group that is accessed in the loop. */ 2073 2074 /* Gaps are supported only for loads. STEP must be a multiple of the type 2075 size. The size of the group must be a power of 2. */ 2076 if (DR_IS_READ (dr) 2077 && (dr_step % type_size) == 0 2078 && stride > 0 2079 && exact_log2 (stride) != -1) 2080 { 2081 GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) = stmt; 2082 GROUP_SIZE (vinfo_for_stmt (stmt)) = stride; 2083 if (vect_print_dump_info (REPORT_DR_DETAILS)) 2084 { 2085 fprintf (vect_dump, "Detected single element interleaving "); 2086 print_generic_expr (vect_dump, DR_REF (dr), TDF_SLIM); 2087 fprintf (vect_dump, " step "); 2088 print_generic_expr (vect_dump, step, TDF_SLIM); 2089 } 2090 2091 if (loop_vinfo) 2092 { 2093 if (vect_print_dump_info (REPORT_DETAILS)) 2094 fprintf (vect_dump, "Data access with gaps requires scalar " 2095 "epilogue loop"); 2096 if (loop->inner) 2097 { 2098 if (vect_print_dump_info (REPORT_DETAILS)) 2099 fprintf (vect_dump, "Peeling for outer loop is not" 2100 " supported"); 2101 return false; 2102 } 2103 2104 LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo) = true; 2105 } 2106 2107 return true; 2108 } 2109 2110 if (vect_print_dump_info (REPORT_DETAILS)) 2111 { 2112 fprintf (vect_dump, "not consecutive access "); 2113 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 2114 } 2115 2116 if (bb_vinfo) 2117 { 2118 /* Mark the statement as unvectorizable. */ 2119 STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (DR_STMT (dr))) = false; 2120 return true; 2121 } 2122 2123 return false; 2124 } 2125 2126 if (GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) == stmt) 2127 { 2128 /* First stmt in the interleaving chain. Check the chain. */ 2129 gimple next = GROUP_NEXT_ELEMENT (vinfo_for_stmt (stmt)); 2130 struct data_reference *data_ref = dr; 2131 unsigned int count = 1; 2132 tree next_step; 2133 tree prev_init = DR_INIT (data_ref); 2134 gimple prev = stmt; 2135 HOST_WIDE_INT diff, count_in_bytes, gaps = 0; 2136 2137 while (next) 2138 { 2139 /* Skip same data-refs. In case that two or more stmts share 2140 data-ref (supported only for loads), we vectorize only the first 2141 stmt, and the rest get their vectorized loads from the first 2142 one. */ 2143 if (!tree_int_cst_compare (DR_INIT (data_ref), 2144 DR_INIT (STMT_VINFO_DATA_REF ( 2145 vinfo_for_stmt (next))))) 2146 { 2147 if (DR_IS_WRITE (data_ref)) 2148 { 2149 if (vect_print_dump_info (REPORT_DETAILS)) 2150 fprintf (vect_dump, "Two store stmts share the same dr."); 2151 return false; 2152 } 2153 2154 /* Check that there is no load-store dependencies for this loads 2155 to prevent a case of load-store-load to the same location. */ 2156 if (GROUP_READ_WRITE_DEPENDENCE (vinfo_for_stmt (next)) 2157 || GROUP_READ_WRITE_DEPENDENCE (vinfo_for_stmt (prev))) 2158 { 2159 if (vect_print_dump_info (REPORT_DETAILS)) 2160 fprintf (vect_dump, 2161 "READ_WRITE dependence in interleaving."); 2162 return false; 2163 } 2164 2165 /* For load use the same data-ref load. */ 2166 GROUP_SAME_DR_STMT (vinfo_for_stmt (next)) = prev; 2167 2168 prev = next; 2169 next = GROUP_NEXT_ELEMENT (vinfo_for_stmt (next)); 2170 continue; 2171 } 2172 2173 prev = next; 2174 2175 /* Check that all the accesses have the same STEP. */ 2176 next_step = DR_STEP (STMT_VINFO_DATA_REF (vinfo_for_stmt (next))); 2177 if (tree_int_cst_compare (step, next_step)) 2178 { 2179 if (vect_print_dump_info (REPORT_DETAILS)) 2180 fprintf (vect_dump, "not consecutive access in interleaving"); 2181 return false; 2182 } 2183 2184 data_ref = STMT_VINFO_DATA_REF (vinfo_for_stmt (next)); 2185 /* Check that the distance between two accesses is equal to the type 2186 size. Otherwise, we have gaps. */ 2187 diff = (TREE_INT_CST_LOW (DR_INIT (data_ref)) 2188 - TREE_INT_CST_LOW (prev_init)) / type_size; 2189 if (diff != 1) 2190 { 2191 /* FORNOW: SLP of accesses with gaps is not supported. */ 2192 slp_impossible = true; 2193 if (DR_IS_WRITE (data_ref)) 2194 { 2195 if (vect_print_dump_info (REPORT_DETAILS)) 2196 fprintf (vect_dump, "interleaved store with gaps"); 2197 return false; 2198 } 2199 2200 gaps += diff - 1; 2201 } 2202 2203 last_accessed_element += diff; 2204 2205 /* Store the gap from the previous member of the group. If there is no 2206 gap in the access, GROUP_GAP is always 1. */ 2207 GROUP_GAP (vinfo_for_stmt (next)) = diff; 2208 2209 prev_init = DR_INIT (data_ref); 2210 next = GROUP_NEXT_ELEMENT (vinfo_for_stmt (next)); 2211 /* Count the number of data-refs in the chain. */ 2212 count++; 2213 } 2214 2215 /* COUNT is the number of accesses found, we multiply it by the size of 2216 the type to get COUNT_IN_BYTES. */ 2217 count_in_bytes = type_size * count; 2218 2219 /* Check that the size of the interleaving (including gaps) is not 2220 greater than STEP. */ 2221 if (dr_step && dr_step < count_in_bytes + gaps * type_size) 2222 { 2223 if (vect_print_dump_info (REPORT_DETAILS)) 2224 { 2225 fprintf (vect_dump, "interleaving size is greater than step for "); 2226 print_generic_expr (vect_dump, DR_REF (dr), TDF_SLIM); 2227 } 2228 return false; 2229 } 2230 2231 /* Check that the size of the interleaving is equal to STEP for stores, 2232 i.e., that there are no gaps. */ 2233 if (dr_step && dr_step != count_in_bytes) 2234 { 2235 if (DR_IS_READ (dr)) 2236 { 2237 slp_impossible = true; 2238 /* There is a gap after the last load in the group. This gap is a 2239 difference between the stride and the number of elements. When 2240 there is no gap, this difference should be 0. */ 2241 GROUP_GAP (vinfo_for_stmt (stmt)) = stride - count; 2242 } 2243 else 2244 { 2245 if (vect_print_dump_info (REPORT_DETAILS)) 2246 fprintf (vect_dump, "interleaved store with gaps"); 2247 return false; 2248 } 2249 } 2250 2251 /* Check that STEP is a multiple of type size. */ 2252 if (dr_step && (dr_step % type_size) != 0) 2253 { 2254 if (vect_print_dump_info (REPORT_DETAILS)) 2255 { 2256 fprintf (vect_dump, "step is not a multiple of type size: step "); 2257 print_generic_expr (vect_dump, step, TDF_SLIM); 2258 fprintf (vect_dump, " size "); 2259 print_generic_expr (vect_dump, TYPE_SIZE_UNIT (scalar_type), 2260 TDF_SLIM); 2261 } 2262 return false; 2263 } 2264 2265 if (stride == 0) 2266 stride = count; 2267 2268 GROUP_SIZE (vinfo_for_stmt (stmt)) = stride; 2269 if (vect_print_dump_info (REPORT_DETAILS)) 2270 fprintf (vect_dump, "Detected interleaving of size %d", (int)stride); 2271 2272 /* SLP: create an SLP data structure for every interleaving group of 2273 stores for further analysis in vect_analyse_slp. */ 2274 if (DR_IS_WRITE (dr) && !slp_impossible) 2275 { 2276 if (loop_vinfo) 2277 VEC_safe_push (gimple, heap, LOOP_VINFO_STRIDED_STORES (loop_vinfo), 2278 stmt); 2279 if (bb_vinfo) 2280 VEC_safe_push (gimple, heap, BB_VINFO_STRIDED_STORES (bb_vinfo), 2281 stmt); 2282 } 2283 2284 /* There is a gap in the end of the group. */ 2285 if (stride - last_accessed_element > 0 && loop_vinfo) 2286 { 2287 if (vect_print_dump_info (REPORT_DETAILS)) 2288 fprintf (vect_dump, "Data access with gaps requires scalar " 2289 "epilogue loop"); 2290 if (loop->inner) 2291 { 2292 if (vect_print_dump_info (REPORT_DETAILS)) 2293 fprintf (vect_dump, "Peeling for outer loop is not supported"); 2294 return false; 2295 } 2296 2297 LOOP_VINFO_PEELING_FOR_GAPS (loop_vinfo) = true; 2298 } 2299 } 2300 2301 return true; 2302 } 2303 2304 2305 /* Analyze the access pattern of the data-reference DR. 2306 In case of non-consecutive accesses call vect_analyze_group_access() to 2307 analyze groups of strided accesses. */ 2308 2309 static bool 2310 vect_analyze_data_ref_access (struct data_reference *dr) 2311 { 2312 tree step = DR_STEP (dr); 2313 tree scalar_type = TREE_TYPE (DR_REF (dr)); 2314 gimple stmt = DR_STMT (dr); 2315 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 2316 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 2317 struct loop *loop = NULL; 2318 HOST_WIDE_INT dr_step; 2319 2320 if (loop_vinfo) 2321 loop = LOOP_VINFO_LOOP (loop_vinfo); 2322 2323 if (loop_vinfo && !step) 2324 { 2325 if (vect_print_dump_info (REPORT_DETAILS)) 2326 fprintf (vect_dump, "bad data-ref access in loop"); 2327 return false; 2328 } 2329 2330 /* Allow invariant loads in loops. */ 2331 dr_step = TREE_INT_CST_LOW (step); 2332 if (loop_vinfo && dr_step == 0) 2333 { 2334 GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) = NULL; 2335 return DR_IS_READ (dr); 2336 } 2337 2338 if (loop && nested_in_vect_loop_p (loop, stmt)) 2339 { 2340 /* Interleaved accesses are not yet supported within outer-loop 2341 vectorization for references in the inner-loop. */ 2342 GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) = NULL; 2343 2344 /* For the rest of the analysis we use the outer-loop step. */ 2345 step = STMT_VINFO_DR_STEP (stmt_info); 2346 dr_step = TREE_INT_CST_LOW (step); 2347 2348 if (dr_step == 0) 2349 { 2350 if (vect_print_dump_info (REPORT_ALIGNMENT)) 2351 fprintf (vect_dump, "zero step in outer loop."); 2352 if (DR_IS_READ (dr)) 2353 return true; 2354 else 2355 return false; 2356 } 2357 } 2358 2359 /* Consecutive? */ 2360 if (!tree_int_cst_compare (step, TYPE_SIZE_UNIT (scalar_type)) 2361 || (dr_step < 0 2362 && !compare_tree_int (TYPE_SIZE_UNIT (scalar_type), -dr_step))) 2363 { 2364 /* Mark that it is not interleaving. */ 2365 GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) = NULL; 2366 return true; 2367 } 2368 2369 if (loop && nested_in_vect_loop_p (loop, stmt)) 2370 { 2371 if (vect_print_dump_info (REPORT_ALIGNMENT)) 2372 fprintf (vect_dump, "strided access in outer loop."); 2373 return false; 2374 } 2375 2376 /* Not consecutive access - check if it's a part of interleaving group. */ 2377 return vect_analyze_group_access (dr); 2378 } 2379 2380 2381 /* Function vect_analyze_data_ref_accesses. 2382 2383 Analyze the access pattern of all the data references in the loop. 2384 2385 FORNOW: the only access pattern that is considered vectorizable is a 2386 simple step 1 (consecutive) access. 2387 2388 FORNOW: handle only arrays and pointer accesses. */ 2389 2390 bool 2391 vect_analyze_data_ref_accesses (loop_vec_info loop_vinfo, bb_vec_info bb_vinfo) 2392 { 2393 unsigned int i; 2394 VEC (data_reference_p, heap) *datarefs; 2395 struct data_reference *dr; 2396 2397 if (vect_print_dump_info (REPORT_DETAILS)) 2398 fprintf (vect_dump, "=== vect_analyze_data_ref_accesses ==="); 2399 2400 if (loop_vinfo) 2401 datarefs = LOOP_VINFO_DATAREFS (loop_vinfo); 2402 else 2403 datarefs = BB_VINFO_DATAREFS (bb_vinfo); 2404 2405 FOR_EACH_VEC_ELT (data_reference_p, datarefs, i, dr) 2406 if (STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (DR_STMT (dr))) 2407 && !vect_analyze_data_ref_access (dr)) 2408 { 2409 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2410 fprintf (vect_dump, "not vectorized: complicated access pattern."); 2411 2412 if (bb_vinfo) 2413 { 2414 /* Mark the statement as not vectorizable. */ 2415 STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (DR_STMT (dr))) = false; 2416 continue; 2417 } 2418 else 2419 return false; 2420 } 2421 2422 return true; 2423 } 2424 2425 /* Function vect_prune_runtime_alias_test_list. 2426 2427 Prune a list of ddrs to be tested at run-time by versioning for alias. 2428 Return FALSE if resulting list of ddrs is longer then allowed by 2429 PARAM_VECT_MAX_VERSION_FOR_ALIAS_CHECKS, otherwise return TRUE. */ 2430 2431 bool 2432 vect_prune_runtime_alias_test_list (loop_vec_info loop_vinfo) 2433 { 2434 VEC (ddr_p, heap) * ddrs = 2435 LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo); 2436 unsigned i, j; 2437 2438 if (vect_print_dump_info (REPORT_DETAILS)) 2439 fprintf (vect_dump, "=== vect_prune_runtime_alias_test_list ==="); 2440 2441 for (i = 0; i < VEC_length (ddr_p, ddrs); ) 2442 { 2443 bool found; 2444 ddr_p ddr_i; 2445 2446 ddr_i = VEC_index (ddr_p, ddrs, i); 2447 found = false; 2448 2449 for (j = 0; j < i; j++) 2450 { 2451 ddr_p ddr_j = VEC_index (ddr_p, ddrs, j); 2452 2453 if (vect_vfa_range_equal (ddr_i, ddr_j)) 2454 { 2455 if (vect_print_dump_info (REPORT_DR_DETAILS)) 2456 { 2457 fprintf (vect_dump, "found equal ranges "); 2458 print_generic_expr (vect_dump, DR_REF (DDR_A (ddr_i)), TDF_SLIM); 2459 fprintf (vect_dump, ", "); 2460 print_generic_expr (vect_dump, DR_REF (DDR_B (ddr_i)), TDF_SLIM); 2461 fprintf (vect_dump, " and "); 2462 print_generic_expr (vect_dump, DR_REF (DDR_A (ddr_j)), TDF_SLIM); 2463 fprintf (vect_dump, ", "); 2464 print_generic_expr (vect_dump, DR_REF (DDR_B (ddr_j)), TDF_SLIM); 2465 } 2466 found = true; 2467 break; 2468 } 2469 } 2470 2471 if (found) 2472 { 2473 VEC_ordered_remove (ddr_p, ddrs, i); 2474 continue; 2475 } 2476 i++; 2477 } 2478 2479 if (VEC_length (ddr_p, ddrs) > 2480 (unsigned) PARAM_VALUE (PARAM_VECT_MAX_VERSION_FOR_ALIAS_CHECKS)) 2481 { 2482 if (vect_print_dump_info (REPORT_DR_DETAILS)) 2483 { 2484 fprintf (vect_dump, 2485 "disable versioning for alias - max number of generated " 2486 "checks exceeded."); 2487 } 2488 2489 VEC_truncate (ddr_p, LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo), 0); 2490 2491 return false; 2492 } 2493 2494 return true; 2495 } 2496 2497 /* Check whether a non-affine read in stmt is suitable for gather load 2498 and if so, return a builtin decl for that operation. */ 2499 2500 tree 2501 vect_check_gather (gimple stmt, loop_vec_info loop_vinfo, tree *basep, 2502 tree *offp, int *scalep) 2503 { 2504 HOST_WIDE_INT scale = 1, pbitpos, pbitsize; 2505 struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo); 2506 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 2507 struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info); 2508 tree offtype = NULL_TREE; 2509 tree decl, base, off; 2510 enum machine_mode pmode; 2511 int punsignedp, pvolatilep; 2512 2513 /* The gather builtins need address of the form 2514 loop_invariant + vector * {1, 2, 4, 8} 2515 or 2516 loop_invariant + sign_extend (vector) * { 1, 2, 4, 8 }. 2517 Unfortunately DR_BASE_ADDRESS/DR_OFFSET can be a mixture 2518 of loop invariants/SSA_NAMEs defined in the loop, with casts, 2519 multiplications and additions in it. To get a vector, we need 2520 a single SSA_NAME that will be defined in the loop and will 2521 contain everything that is not loop invariant and that can be 2522 vectorized. The following code attempts to find such a preexistng 2523 SSA_NAME OFF and put the loop invariants into a tree BASE 2524 that can be gimplified before the loop. */ 2525 base = get_inner_reference (DR_REF (dr), &pbitsize, &pbitpos, &off, 2526 &pmode, &punsignedp, &pvolatilep, false); 2527 gcc_assert (base != NULL_TREE && (pbitpos % BITS_PER_UNIT) == 0); 2528 2529 if (TREE_CODE (base) == MEM_REF) 2530 { 2531 if (!integer_zerop (TREE_OPERAND (base, 1))) 2532 { 2533 if (off == NULL_TREE) 2534 { 2535 double_int moff = mem_ref_offset (base); 2536 off = double_int_to_tree (sizetype, moff); 2537 } 2538 else 2539 off = size_binop (PLUS_EXPR, off, 2540 fold_convert (sizetype, TREE_OPERAND (base, 1))); 2541 } 2542 base = TREE_OPERAND (base, 0); 2543 } 2544 else 2545 base = build_fold_addr_expr (base); 2546 2547 if (off == NULL_TREE) 2548 off = size_zero_node; 2549 2550 /* If base is not loop invariant, either off is 0, then we start with just 2551 the constant offset in the loop invariant BASE and continue with base 2552 as OFF, otherwise give up. 2553 We could handle that case by gimplifying the addition of base + off 2554 into some SSA_NAME and use that as off, but for now punt. */ 2555 if (!expr_invariant_in_loop_p (loop, base)) 2556 { 2557 if (!integer_zerop (off)) 2558 return NULL_TREE; 2559 off = base; 2560 base = size_int (pbitpos / BITS_PER_UNIT); 2561 } 2562 /* Otherwise put base + constant offset into the loop invariant BASE 2563 and continue with OFF. */ 2564 else 2565 { 2566 base = fold_convert (sizetype, base); 2567 base = size_binop (PLUS_EXPR, base, size_int (pbitpos / BITS_PER_UNIT)); 2568 } 2569 2570 /* OFF at this point may be either a SSA_NAME or some tree expression 2571 from get_inner_reference. Try to peel off loop invariants from it 2572 into BASE as long as possible. */ 2573 STRIP_NOPS (off); 2574 while (offtype == NULL_TREE) 2575 { 2576 enum tree_code code; 2577 tree op0, op1, add = NULL_TREE; 2578 2579 if (TREE_CODE (off) == SSA_NAME) 2580 { 2581 gimple def_stmt = SSA_NAME_DEF_STMT (off); 2582 2583 if (expr_invariant_in_loop_p (loop, off)) 2584 return NULL_TREE; 2585 2586 if (gimple_code (def_stmt) != GIMPLE_ASSIGN) 2587 break; 2588 2589 op0 = gimple_assign_rhs1 (def_stmt); 2590 code = gimple_assign_rhs_code (def_stmt); 2591 op1 = gimple_assign_rhs2 (def_stmt); 2592 } 2593 else 2594 { 2595 if (get_gimple_rhs_class (TREE_CODE (off)) == GIMPLE_TERNARY_RHS) 2596 return NULL_TREE; 2597 code = TREE_CODE (off); 2598 extract_ops_from_tree (off, &code, &op0, &op1); 2599 } 2600 switch (code) 2601 { 2602 case POINTER_PLUS_EXPR: 2603 case PLUS_EXPR: 2604 if (expr_invariant_in_loop_p (loop, op0)) 2605 { 2606 add = op0; 2607 off = op1; 2608 do_add: 2609 add = fold_convert (sizetype, add); 2610 if (scale != 1) 2611 add = size_binop (MULT_EXPR, add, size_int (scale)); 2612 base = size_binop (PLUS_EXPR, base, add); 2613 continue; 2614 } 2615 if (expr_invariant_in_loop_p (loop, op1)) 2616 { 2617 add = op1; 2618 off = op0; 2619 goto do_add; 2620 } 2621 break; 2622 case MINUS_EXPR: 2623 if (expr_invariant_in_loop_p (loop, op1)) 2624 { 2625 add = fold_convert (sizetype, op1); 2626 add = size_binop (MINUS_EXPR, size_zero_node, add); 2627 off = op0; 2628 goto do_add; 2629 } 2630 break; 2631 case MULT_EXPR: 2632 if (scale == 1 && host_integerp (op1, 0)) 2633 { 2634 scale = tree_low_cst (op1, 0); 2635 off = op0; 2636 continue; 2637 } 2638 break; 2639 case SSA_NAME: 2640 off = op0; 2641 continue; 2642 CASE_CONVERT: 2643 if (!POINTER_TYPE_P (TREE_TYPE (op0)) 2644 && !INTEGRAL_TYPE_P (TREE_TYPE (op0))) 2645 break; 2646 if (TYPE_PRECISION (TREE_TYPE (op0)) 2647 == TYPE_PRECISION (TREE_TYPE (off))) 2648 { 2649 off = op0; 2650 continue; 2651 } 2652 if (TYPE_PRECISION (TREE_TYPE (op0)) 2653 < TYPE_PRECISION (TREE_TYPE (off))) 2654 { 2655 off = op0; 2656 offtype = TREE_TYPE (off); 2657 STRIP_NOPS (off); 2658 continue; 2659 } 2660 break; 2661 default: 2662 break; 2663 } 2664 break; 2665 } 2666 2667 /* If at the end OFF still isn't a SSA_NAME or isn't 2668 defined in the loop, punt. */ 2669 if (TREE_CODE (off) != SSA_NAME 2670 || expr_invariant_in_loop_p (loop, off)) 2671 return NULL_TREE; 2672 2673 if (offtype == NULL_TREE) 2674 offtype = TREE_TYPE (off); 2675 2676 decl = targetm.vectorize.builtin_gather (STMT_VINFO_VECTYPE (stmt_info), 2677 offtype, scale); 2678 if (decl == NULL_TREE) 2679 return NULL_TREE; 2680 2681 if (basep) 2682 *basep = base; 2683 if (offp) 2684 *offp = off; 2685 if (scalep) 2686 *scalep = scale; 2687 return decl; 2688 } 2689 2690 2691 /* Function vect_analyze_data_refs. 2692 2693 Find all the data references in the loop or basic block. 2694 2695 The general structure of the analysis of data refs in the vectorizer is as 2696 follows: 2697 1- vect_analyze_data_refs(loop/bb): call 2698 compute_data_dependences_for_loop/bb to find and analyze all data-refs 2699 in the loop/bb and their dependences. 2700 2- vect_analyze_dependences(): apply dependence testing using ddrs. 2701 3- vect_analyze_drs_alignment(): check that ref_stmt.alignment is ok. 2702 4- vect_analyze_drs_access(): check that ref_stmt.step is ok. 2703 2704 */ 2705 2706 bool 2707 vect_analyze_data_refs (loop_vec_info loop_vinfo, 2708 bb_vec_info bb_vinfo, 2709 int *min_vf) 2710 { 2711 struct loop *loop = NULL; 2712 basic_block bb = NULL; 2713 unsigned int i; 2714 VEC (data_reference_p, heap) *datarefs; 2715 struct data_reference *dr; 2716 tree scalar_type; 2717 bool res, stop_bb_analysis = false; 2718 2719 if (vect_print_dump_info (REPORT_DETAILS)) 2720 fprintf (vect_dump, "=== vect_analyze_data_refs ===\n"); 2721 2722 if (loop_vinfo) 2723 { 2724 loop = LOOP_VINFO_LOOP (loop_vinfo); 2725 res = compute_data_dependences_for_loop 2726 (loop, true, 2727 &LOOP_VINFO_LOOP_NEST (loop_vinfo), 2728 &LOOP_VINFO_DATAREFS (loop_vinfo), 2729 &LOOP_VINFO_DDRS (loop_vinfo)); 2730 2731 if (!res) 2732 { 2733 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2734 fprintf (vect_dump, "not vectorized: loop contains function calls" 2735 " or data references that cannot be analyzed"); 2736 return false; 2737 } 2738 2739 datarefs = LOOP_VINFO_DATAREFS (loop_vinfo); 2740 } 2741 else 2742 { 2743 bb = BB_VINFO_BB (bb_vinfo); 2744 res = compute_data_dependences_for_bb (bb, true, 2745 &BB_VINFO_DATAREFS (bb_vinfo), 2746 &BB_VINFO_DDRS (bb_vinfo)); 2747 if (!res) 2748 { 2749 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2750 fprintf (vect_dump, "not vectorized: basic block contains function" 2751 " calls or data references that cannot be analyzed"); 2752 return false; 2753 } 2754 2755 datarefs = BB_VINFO_DATAREFS (bb_vinfo); 2756 } 2757 2758 /* Go through the data-refs, check that the analysis succeeded. Update 2759 pointer from stmt_vec_info struct to DR and vectype. */ 2760 2761 FOR_EACH_VEC_ELT (data_reference_p, datarefs, i, dr) 2762 { 2763 gimple stmt; 2764 stmt_vec_info stmt_info; 2765 tree base, offset, init; 2766 bool gather = false; 2767 int vf; 2768 2769 if (!dr || !DR_REF (dr)) 2770 { 2771 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2772 fprintf (vect_dump, "not vectorized: unhandled data-ref "); 2773 2774 return false; 2775 } 2776 2777 stmt = DR_STMT (dr); 2778 stmt_info = vinfo_for_stmt (stmt); 2779 2780 if (stop_bb_analysis) 2781 { 2782 STMT_VINFO_VECTORIZABLE (stmt_info) = false; 2783 continue; 2784 } 2785 2786 /* Check that analysis of the data-ref succeeded. */ 2787 if (!DR_BASE_ADDRESS (dr) || !DR_OFFSET (dr) || !DR_INIT (dr) 2788 || !DR_STEP (dr)) 2789 { 2790 /* If target supports vector gather loads, see if they can't 2791 be used. */ 2792 if (loop_vinfo 2793 && DR_IS_READ (dr) 2794 && !TREE_THIS_VOLATILE (DR_REF (dr)) 2795 && targetm.vectorize.builtin_gather != NULL 2796 && !nested_in_vect_loop_p (loop, stmt)) 2797 { 2798 struct data_reference *newdr 2799 = create_data_ref (NULL, loop_containing_stmt (stmt), 2800 DR_REF (dr), stmt, true); 2801 gcc_assert (newdr != NULL && DR_REF (newdr)); 2802 if (DR_BASE_ADDRESS (newdr) 2803 && DR_OFFSET (newdr) 2804 && DR_INIT (newdr) 2805 && DR_STEP (newdr) 2806 && integer_zerop (DR_STEP (newdr))) 2807 { 2808 dr = newdr; 2809 gather = true; 2810 } 2811 else 2812 free_data_ref (newdr); 2813 } 2814 2815 if (!gather) 2816 { 2817 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2818 { 2819 fprintf (vect_dump, "not vectorized: data ref analysis " 2820 "failed "); 2821 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 2822 } 2823 2824 if (bb_vinfo) 2825 { 2826 STMT_VINFO_VECTORIZABLE (stmt_info) = false; 2827 stop_bb_analysis = true; 2828 continue; 2829 } 2830 2831 return false; 2832 } 2833 } 2834 2835 if (TREE_CODE (DR_BASE_ADDRESS (dr)) == INTEGER_CST) 2836 { 2837 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2838 fprintf (vect_dump, "not vectorized: base addr of dr is a " 2839 "constant"); 2840 2841 if (bb_vinfo) 2842 { 2843 STMT_VINFO_VECTORIZABLE (stmt_info) = false; 2844 stop_bb_analysis = true; 2845 continue; 2846 } 2847 2848 if (gather) 2849 free_data_ref (dr); 2850 return false; 2851 } 2852 2853 if (TREE_THIS_VOLATILE (DR_REF (dr))) 2854 { 2855 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2856 { 2857 fprintf (vect_dump, "not vectorized: volatile type "); 2858 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 2859 } 2860 2861 if (bb_vinfo) 2862 { 2863 STMT_VINFO_VECTORIZABLE (stmt_info) = false; 2864 stop_bb_analysis = true; 2865 continue; 2866 } 2867 2868 return false; 2869 } 2870 2871 if (stmt_can_throw_internal (stmt)) 2872 { 2873 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2874 { 2875 fprintf (vect_dump, "not vectorized: statement can throw an " 2876 "exception "); 2877 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 2878 } 2879 2880 if (bb_vinfo) 2881 { 2882 STMT_VINFO_VECTORIZABLE (stmt_info) = false; 2883 stop_bb_analysis = true; 2884 continue; 2885 } 2886 2887 if (gather) 2888 free_data_ref (dr); 2889 return false; 2890 } 2891 2892 if (TREE_CODE (DR_REF (dr)) == COMPONENT_REF 2893 && DECL_BIT_FIELD (TREE_OPERAND (DR_REF (dr), 1))) 2894 { 2895 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2896 { 2897 fprintf (vect_dump, "not vectorized: statement is bitfield " 2898 "access "); 2899 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 2900 } 2901 2902 if (bb_vinfo) 2903 { 2904 STMT_VINFO_VECTORIZABLE (stmt_info) = false; 2905 stop_bb_analysis = true; 2906 continue; 2907 } 2908 2909 if (gather) 2910 free_data_ref (dr); 2911 return false; 2912 } 2913 2914 base = unshare_expr (DR_BASE_ADDRESS (dr)); 2915 offset = unshare_expr (DR_OFFSET (dr)); 2916 init = unshare_expr (DR_INIT (dr)); 2917 2918 if (is_gimple_call (stmt)) 2919 { 2920 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 2921 { 2922 fprintf (vect_dump, "not vectorized: dr in a call "); 2923 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 2924 } 2925 2926 if (bb_vinfo) 2927 { 2928 STMT_VINFO_VECTORIZABLE (stmt_info) = false; 2929 stop_bb_analysis = true; 2930 continue; 2931 } 2932 2933 if (gather) 2934 free_data_ref (dr); 2935 return false; 2936 } 2937 2938 /* Update DR field in stmt_vec_info struct. */ 2939 2940 /* If the dataref is in an inner-loop of the loop that is considered for 2941 for vectorization, we also want to analyze the access relative to 2942 the outer-loop (DR contains information only relative to the 2943 inner-most enclosing loop). We do that by building a reference to the 2944 first location accessed by the inner-loop, and analyze it relative to 2945 the outer-loop. */ 2946 if (loop && nested_in_vect_loop_p (loop, stmt)) 2947 { 2948 tree outer_step, outer_base, outer_init; 2949 HOST_WIDE_INT pbitsize, pbitpos; 2950 tree poffset; 2951 enum machine_mode pmode; 2952 int punsignedp, pvolatilep; 2953 affine_iv base_iv, offset_iv; 2954 tree dinit; 2955 2956 /* Build a reference to the first location accessed by the 2957 inner-loop: *(BASE+INIT). (The first location is actually 2958 BASE+INIT+OFFSET, but we add OFFSET separately later). */ 2959 tree inner_base = build_fold_indirect_ref 2960 (fold_build_pointer_plus (base, init)); 2961 2962 if (vect_print_dump_info (REPORT_DETAILS)) 2963 { 2964 fprintf (vect_dump, "analyze in outer-loop: "); 2965 print_generic_expr (vect_dump, inner_base, TDF_SLIM); 2966 } 2967 2968 outer_base = get_inner_reference (inner_base, &pbitsize, &pbitpos, 2969 &poffset, &pmode, &punsignedp, &pvolatilep, false); 2970 gcc_assert (outer_base != NULL_TREE); 2971 2972 if (pbitpos % BITS_PER_UNIT != 0) 2973 { 2974 if (vect_print_dump_info (REPORT_DETAILS)) 2975 fprintf (vect_dump, "failed: bit offset alignment.\n"); 2976 return false; 2977 } 2978 2979 outer_base = build_fold_addr_expr (outer_base); 2980 if (!simple_iv (loop, loop_containing_stmt (stmt), outer_base, 2981 &base_iv, false)) 2982 { 2983 if (vect_print_dump_info (REPORT_DETAILS)) 2984 fprintf (vect_dump, "failed: evolution of base is not affine.\n"); 2985 return false; 2986 } 2987 2988 if (offset) 2989 { 2990 if (poffset) 2991 poffset = fold_build2 (PLUS_EXPR, TREE_TYPE (offset), offset, 2992 poffset); 2993 else 2994 poffset = offset; 2995 } 2996 2997 if (!poffset) 2998 { 2999 offset_iv.base = ssize_int (0); 3000 offset_iv.step = ssize_int (0); 3001 } 3002 else if (!simple_iv (loop, loop_containing_stmt (stmt), poffset, 3003 &offset_iv, false)) 3004 { 3005 if (vect_print_dump_info (REPORT_DETAILS)) 3006 fprintf (vect_dump, "evolution of offset is not affine.\n"); 3007 return false; 3008 } 3009 3010 outer_init = ssize_int (pbitpos / BITS_PER_UNIT); 3011 split_constant_offset (base_iv.base, &base_iv.base, &dinit); 3012 outer_init = size_binop (PLUS_EXPR, outer_init, dinit); 3013 split_constant_offset (offset_iv.base, &offset_iv.base, &dinit); 3014 outer_init = size_binop (PLUS_EXPR, outer_init, dinit); 3015 3016 outer_step = size_binop (PLUS_EXPR, 3017 fold_convert (ssizetype, base_iv.step), 3018 fold_convert (ssizetype, offset_iv.step)); 3019 3020 STMT_VINFO_DR_STEP (stmt_info) = outer_step; 3021 /* FIXME: Use canonicalize_base_object_address (base_iv.base); */ 3022 STMT_VINFO_DR_BASE_ADDRESS (stmt_info) = base_iv.base; 3023 STMT_VINFO_DR_INIT (stmt_info) = outer_init; 3024 STMT_VINFO_DR_OFFSET (stmt_info) = 3025 fold_convert (ssizetype, offset_iv.base); 3026 STMT_VINFO_DR_ALIGNED_TO (stmt_info) = 3027 size_int (highest_pow2_factor (offset_iv.base)); 3028 3029 if (vect_print_dump_info (REPORT_DETAILS)) 3030 { 3031 fprintf (vect_dump, "\touter base_address: "); 3032 print_generic_expr (vect_dump, STMT_VINFO_DR_BASE_ADDRESS (stmt_info), TDF_SLIM); 3033 fprintf (vect_dump, "\n\touter offset from base address: "); 3034 print_generic_expr (vect_dump, STMT_VINFO_DR_OFFSET (stmt_info), TDF_SLIM); 3035 fprintf (vect_dump, "\n\touter constant offset from base address: "); 3036 print_generic_expr (vect_dump, STMT_VINFO_DR_INIT (stmt_info), TDF_SLIM); 3037 fprintf (vect_dump, "\n\touter step: "); 3038 print_generic_expr (vect_dump, STMT_VINFO_DR_STEP (stmt_info), TDF_SLIM); 3039 fprintf (vect_dump, "\n\touter aligned to: "); 3040 print_generic_expr (vect_dump, STMT_VINFO_DR_ALIGNED_TO (stmt_info), TDF_SLIM); 3041 } 3042 } 3043 3044 if (STMT_VINFO_DATA_REF (stmt_info)) 3045 { 3046 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 3047 { 3048 fprintf (vect_dump, 3049 "not vectorized: more than one data ref in stmt: "); 3050 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 3051 } 3052 3053 if (bb_vinfo) 3054 { 3055 STMT_VINFO_VECTORIZABLE (stmt_info) = false; 3056 stop_bb_analysis = true; 3057 continue; 3058 } 3059 3060 if (gather) 3061 free_data_ref (dr); 3062 return false; 3063 } 3064 3065 STMT_VINFO_DATA_REF (stmt_info) = dr; 3066 3067 /* Set vectype for STMT. */ 3068 scalar_type = TREE_TYPE (DR_REF (dr)); 3069 STMT_VINFO_VECTYPE (stmt_info) = 3070 get_vectype_for_scalar_type (scalar_type); 3071 if (!STMT_VINFO_VECTYPE (stmt_info)) 3072 { 3073 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 3074 { 3075 fprintf (vect_dump, 3076 "not vectorized: no vectype for stmt: "); 3077 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 3078 fprintf (vect_dump, " scalar_type: "); 3079 print_generic_expr (vect_dump, scalar_type, TDF_DETAILS); 3080 } 3081 3082 if (bb_vinfo) 3083 { 3084 /* Mark the statement as not vectorizable. */ 3085 STMT_VINFO_VECTORIZABLE (stmt_info) = false; 3086 stop_bb_analysis = true; 3087 continue; 3088 } 3089 3090 if (gather) 3091 { 3092 STMT_VINFO_DATA_REF (stmt_info) = NULL; 3093 free_data_ref (dr); 3094 } 3095 return false; 3096 } 3097 3098 /* Adjust the minimal vectorization factor according to the 3099 vector type. */ 3100 vf = TYPE_VECTOR_SUBPARTS (STMT_VINFO_VECTYPE (stmt_info)); 3101 if (vf > *min_vf) 3102 *min_vf = vf; 3103 3104 if (gather) 3105 { 3106 unsigned int j, k, n; 3107 struct data_reference *olddr 3108 = VEC_index (data_reference_p, datarefs, i); 3109 VEC (ddr_p, heap) *ddrs = LOOP_VINFO_DDRS (loop_vinfo); 3110 struct data_dependence_relation *ddr, *newddr; 3111 bool bad = false; 3112 tree off; 3113 VEC (loop_p, heap) *nest = LOOP_VINFO_LOOP_NEST (loop_vinfo); 3114 3115 if (!vect_check_gather (stmt, loop_vinfo, NULL, &off, NULL) 3116 || get_vectype_for_scalar_type (TREE_TYPE (off)) == NULL_TREE) 3117 { 3118 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 3119 { 3120 fprintf (vect_dump, 3121 "not vectorized: not suitable for gather "); 3122 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 3123 } 3124 return false; 3125 } 3126 3127 n = VEC_length (data_reference_p, datarefs) - 1; 3128 for (j = 0, k = i - 1; j < i; j++) 3129 { 3130 ddr = VEC_index (ddr_p, ddrs, k); 3131 gcc_assert (DDR_B (ddr) == olddr); 3132 newddr = initialize_data_dependence_relation (DDR_A (ddr), dr, 3133 nest); 3134 VEC_replace (ddr_p, ddrs, k, newddr); 3135 free_dependence_relation (ddr); 3136 if (!bad 3137 && DR_IS_WRITE (DDR_A (newddr)) 3138 && DDR_ARE_DEPENDENT (newddr) != chrec_known) 3139 bad = true; 3140 k += --n; 3141 } 3142 3143 k++; 3144 n = k + VEC_length (data_reference_p, datarefs) - i - 1; 3145 for (; k < n; k++) 3146 { 3147 ddr = VEC_index (ddr_p, ddrs, k); 3148 gcc_assert (DDR_A (ddr) == olddr); 3149 newddr = initialize_data_dependence_relation (dr, DDR_B (ddr), 3150 nest); 3151 VEC_replace (ddr_p, ddrs, k, newddr); 3152 free_dependence_relation (ddr); 3153 if (!bad 3154 && DR_IS_WRITE (DDR_B (newddr)) 3155 && DDR_ARE_DEPENDENT (newddr) != chrec_known) 3156 bad = true; 3157 } 3158 3159 k = VEC_length (ddr_p, ddrs) 3160 - VEC_length (data_reference_p, datarefs) + i; 3161 ddr = VEC_index (ddr_p, ddrs, k); 3162 gcc_assert (DDR_A (ddr) == olddr && DDR_B (ddr) == olddr); 3163 newddr = initialize_data_dependence_relation (dr, dr, nest); 3164 VEC_replace (ddr_p, ddrs, k, newddr); 3165 free_dependence_relation (ddr); 3166 VEC_replace (data_reference_p, datarefs, i, dr); 3167 3168 if (bad) 3169 { 3170 if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)) 3171 { 3172 fprintf (vect_dump, 3173 "not vectorized: data dependence conflict" 3174 " prevents gather"); 3175 print_gimple_stmt (vect_dump, stmt, 0, TDF_SLIM); 3176 } 3177 return false; 3178 } 3179 3180 STMT_VINFO_GATHER_P (stmt_info) = true; 3181 } 3182 } 3183 3184 return true; 3185 } 3186 3187 3188 /* Function vect_get_new_vect_var. 3189 3190 Returns a name for a new variable. The current naming scheme appends the 3191 prefix "vect_" or "vect_p" (depending on the value of VAR_KIND) to 3192 the name of vectorizer generated variables, and appends that to NAME if 3193 provided. */ 3194 3195 tree 3196 vect_get_new_vect_var (tree type, enum vect_var_kind var_kind, const char *name) 3197 { 3198 const char *prefix; 3199 tree new_vect_var; 3200 3201 switch (var_kind) 3202 { 3203 case vect_simple_var: 3204 prefix = "vect_"; 3205 break; 3206 case vect_scalar_var: 3207 prefix = "stmp_"; 3208 break; 3209 case vect_pointer_var: 3210 prefix = "vect_p"; 3211 break; 3212 default: 3213 gcc_unreachable (); 3214 } 3215 3216 if (name) 3217 { 3218 char* tmp = concat (prefix, name, NULL); 3219 new_vect_var = create_tmp_var (type, tmp); 3220 free (tmp); 3221 } 3222 else 3223 new_vect_var = create_tmp_var (type, prefix); 3224 3225 /* Mark vector typed variable as a gimple register variable. */ 3226 if (TREE_CODE (type) == VECTOR_TYPE) 3227 DECL_GIMPLE_REG_P (new_vect_var) = true; 3228 3229 return new_vect_var; 3230 } 3231 3232 3233 /* Function vect_create_addr_base_for_vector_ref. 3234 3235 Create an expression that computes the address of the first memory location 3236 that will be accessed for a data reference. 3237 3238 Input: 3239 STMT: The statement containing the data reference. 3240 NEW_STMT_LIST: Must be initialized to NULL_TREE or a statement list. 3241 OFFSET: Optional. If supplied, it is be added to the initial address. 3242 LOOP: Specify relative to which loop-nest should the address be computed. 3243 For example, when the dataref is in an inner-loop nested in an 3244 outer-loop that is now being vectorized, LOOP can be either the 3245 outer-loop, or the inner-loop. The first memory location accessed 3246 by the following dataref ('in' points to short): 3247 3248 for (i=0; i<N; i++) 3249 for (j=0; j<M; j++) 3250 s += in[i+j] 3251 3252 is as follows: 3253 if LOOP=i_loop: &in (relative to i_loop) 3254 if LOOP=j_loop: &in+i*2B (relative to j_loop) 3255 3256 Output: 3257 1. Return an SSA_NAME whose value is the address of the memory location of 3258 the first vector of the data reference. 3259 2. If new_stmt_list is not NULL_TREE after return then the caller must insert 3260 these statement(s) which define the returned SSA_NAME. 3261 3262 FORNOW: We are only handling array accesses with step 1. */ 3263 3264 tree 3265 vect_create_addr_base_for_vector_ref (gimple stmt, 3266 gimple_seq *new_stmt_list, 3267 tree offset, 3268 struct loop *loop) 3269 { 3270 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 3271 struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info); 3272 tree data_ref_base = unshare_expr (DR_BASE_ADDRESS (dr)); 3273 tree base_name; 3274 tree data_ref_base_var; 3275 tree vec_stmt; 3276 tree addr_base, addr_expr; 3277 tree dest; 3278 gimple_seq seq = NULL; 3279 tree base_offset = unshare_expr (DR_OFFSET (dr)); 3280 tree init = unshare_expr (DR_INIT (dr)); 3281 tree vect_ptr_type; 3282 tree step = TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dr))); 3283 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 3284 tree base; 3285 3286 if (loop_vinfo && loop && loop != (gimple_bb (stmt))->loop_father) 3287 { 3288 struct loop *outer_loop = LOOP_VINFO_LOOP (loop_vinfo); 3289 3290 gcc_assert (nested_in_vect_loop_p (outer_loop, stmt)); 3291 3292 data_ref_base = unshare_expr (STMT_VINFO_DR_BASE_ADDRESS (stmt_info)); 3293 base_offset = unshare_expr (STMT_VINFO_DR_OFFSET (stmt_info)); 3294 init = unshare_expr (STMT_VINFO_DR_INIT (stmt_info)); 3295 } 3296 3297 if (loop_vinfo) 3298 base_name = build_fold_indirect_ref (data_ref_base); 3299 else 3300 { 3301 base_offset = ssize_int (0); 3302 init = ssize_int (0); 3303 base_name = build_fold_indirect_ref (unshare_expr (DR_REF (dr))); 3304 } 3305 3306 data_ref_base_var = create_tmp_var (TREE_TYPE (data_ref_base), "batmp"); 3307 add_referenced_var (data_ref_base_var); 3308 data_ref_base = force_gimple_operand (data_ref_base, &seq, true, 3309 data_ref_base_var); 3310 gimple_seq_add_seq (new_stmt_list, seq); 3311 3312 /* Create base_offset */ 3313 base_offset = size_binop (PLUS_EXPR, 3314 fold_convert (sizetype, base_offset), 3315 fold_convert (sizetype, init)); 3316 dest = create_tmp_var (sizetype, "base_off"); 3317 add_referenced_var (dest); 3318 base_offset = force_gimple_operand (base_offset, &seq, true, dest); 3319 gimple_seq_add_seq (new_stmt_list, seq); 3320 3321 if (offset) 3322 { 3323 tree tmp = create_tmp_var (sizetype, "offset"); 3324 3325 add_referenced_var (tmp); 3326 offset = fold_build2 (MULT_EXPR, sizetype, 3327 fold_convert (sizetype, offset), step); 3328 base_offset = fold_build2 (PLUS_EXPR, sizetype, 3329 base_offset, offset); 3330 base_offset = force_gimple_operand (base_offset, &seq, false, tmp); 3331 gimple_seq_add_seq (new_stmt_list, seq); 3332 } 3333 3334 /* base + base_offset */ 3335 if (loop_vinfo) 3336 addr_base = fold_build_pointer_plus (data_ref_base, base_offset); 3337 else 3338 { 3339 addr_base = build1 (ADDR_EXPR, 3340 build_pointer_type (TREE_TYPE (DR_REF (dr))), 3341 unshare_expr (DR_REF (dr))); 3342 } 3343 3344 vect_ptr_type = build_pointer_type (STMT_VINFO_VECTYPE (stmt_info)); 3345 base = get_base_address (DR_REF (dr)); 3346 if (base 3347 && TREE_CODE (base) == MEM_REF) 3348 vect_ptr_type 3349 = build_qualified_type (vect_ptr_type, 3350 TYPE_QUALS (TREE_TYPE (TREE_OPERAND (base, 0)))); 3351 3352 vec_stmt = fold_convert (vect_ptr_type, addr_base); 3353 addr_expr = vect_get_new_vect_var (vect_ptr_type, vect_pointer_var, 3354 get_name (base_name)); 3355 add_referenced_var (addr_expr); 3356 vec_stmt = force_gimple_operand (vec_stmt, &seq, false, addr_expr); 3357 gimple_seq_add_seq (new_stmt_list, seq); 3358 3359 if (DR_PTR_INFO (dr) 3360 && TREE_CODE (vec_stmt) == SSA_NAME) 3361 { 3362 duplicate_ssa_name_ptr_info (vec_stmt, DR_PTR_INFO (dr)); 3363 if (offset) 3364 { 3365 SSA_NAME_PTR_INFO (vec_stmt)->align = 1; 3366 SSA_NAME_PTR_INFO (vec_stmt)->misalign = 0; 3367 } 3368 } 3369 3370 if (vect_print_dump_info (REPORT_DETAILS)) 3371 { 3372 fprintf (vect_dump, "created "); 3373 print_generic_expr (vect_dump, vec_stmt, TDF_SLIM); 3374 } 3375 3376 return vec_stmt; 3377 } 3378 3379 3380 /* Function vect_create_data_ref_ptr. 3381 3382 Create a new pointer-to-AGGR_TYPE variable (ap), that points to the first 3383 location accessed in the loop by STMT, along with the def-use update 3384 chain to appropriately advance the pointer through the loop iterations. 3385 Also set aliasing information for the pointer. This pointer is used by 3386 the callers to this function to create a memory reference expression for 3387 vector load/store access. 3388 3389 Input: 3390 1. STMT: a stmt that references memory. Expected to be of the form 3391 GIMPLE_ASSIGN <name, data-ref> or 3392 GIMPLE_ASSIGN <data-ref, name>. 3393 2. AGGR_TYPE: the type of the reference, which should be either a vector 3394 or an array. 3395 3. AT_LOOP: the loop where the vector memref is to be created. 3396 4. OFFSET (optional): an offset to be added to the initial address accessed 3397 by the data-ref in STMT. 3398 5. BSI: location where the new stmts are to be placed if there is no loop 3399 6. ONLY_INIT: indicate if ap is to be updated in the loop, or remain 3400 pointing to the initial address. 3401 3402 Output: 3403 1. Declare a new ptr to vector_type, and have it point to the base of the 3404 data reference (initial addressed accessed by the data reference). 3405 For example, for vector of type V8HI, the following code is generated: 3406 3407 v8hi *ap; 3408 ap = (v8hi *)initial_address; 3409 3410 if OFFSET is not supplied: 3411 initial_address = &a[init]; 3412 if OFFSET is supplied: 3413 initial_address = &a[init + OFFSET]; 3414 3415 Return the initial_address in INITIAL_ADDRESS. 3416 3417 2. If ONLY_INIT is true, just return the initial pointer. Otherwise, also 3418 update the pointer in each iteration of the loop. 3419 3420 Return the increment stmt that updates the pointer in PTR_INCR. 3421 3422 3. Set INV_P to true if the access pattern of the data reference in the 3423 vectorized loop is invariant. Set it to false otherwise. 3424 3425 4. Return the pointer. */ 3426 3427 tree 3428 vect_create_data_ref_ptr (gimple stmt, tree aggr_type, struct loop *at_loop, 3429 tree offset, tree *initial_address, 3430 gimple_stmt_iterator *gsi, gimple *ptr_incr, 3431 bool only_init, bool *inv_p) 3432 { 3433 tree base_name; 3434 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 3435 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 3436 struct loop *loop = NULL; 3437 bool nested_in_vect_loop = false; 3438 struct loop *containing_loop = NULL; 3439 tree aggr_ptr_type; 3440 tree aggr_ptr; 3441 tree new_temp; 3442 gimple vec_stmt; 3443 gimple_seq new_stmt_list = NULL; 3444 edge pe = NULL; 3445 basic_block new_bb; 3446 tree aggr_ptr_init; 3447 struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info); 3448 tree aptr; 3449 gimple_stmt_iterator incr_gsi; 3450 bool insert_after; 3451 bool negative; 3452 tree indx_before_incr, indx_after_incr; 3453 gimple incr; 3454 tree step; 3455 bb_vec_info bb_vinfo = STMT_VINFO_BB_VINFO (stmt_info); 3456 tree base; 3457 3458 gcc_assert (TREE_CODE (aggr_type) == ARRAY_TYPE 3459 || TREE_CODE (aggr_type) == VECTOR_TYPE); 3460 3461 if (loop_vinfo) 3462 { 3463 loop = LOOP_VINFO_LOOP (loop_vinfo); 3464 nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt); 3465 containing_loop = (gimple_bb (stmt))->loop_father; 3466 pe = loop_preheader_edge (loop); 3467 } 3468 else 3469 { 3470 gcc_assert (bb_vinfo); 3471 only_init = true; 3472 *ptr_incr = NULL; 3473 } 3474 3475 /* Check the step (evolution) of the load in LOOP, and record 3476 whether it's invariant. */ 3477 if (nested_in_vect_loop) 3478 step = STMT_VINFO_DR_STEP (stmt_info); 3479 else 3480 step = DR_STEP (STMT_VINFO_DATA_REF (stmt_info)); 3481 3482 if (tree_int_cst_compare (step, size_zero_node) == 0) 3483 *inv_p = true; 3484 else 3485 *inv_p = false; 3486 negative = tree_int_cst_compare (step, size_zero_node) < 0; 3487 3488 /* Create an expression for the first address accessed by this load 3489 in LOOP. */ 3490 base_name = build_fold_indirect_ref (unshare_expr (DR_BASE_ADDRESS (dr))); 3491 3492 if (vect_print_dump_info (REPORT_DETAILS)) 3493 { 3494 tree data_ref_base = base_name; 3495 fprintf (vect_dump, "create %s-pointer variable to type: ", 3496 tree_code_name[(int) TREE_CODE (aggr_type)]); 3497 print_generic_expr (vect_dump, aggr_type, TDF_SLIM); 3498 if (TREE_CODE (data_ref_base) == VAR_DECL 3499 || TREE_CODE (data_ref_base) == ARRAY_REF) 3500 fprintf (vect_dump, " vectorizing an array ref: "); 3501 else if (TREE_CODE (data_ref_base) == COMPONENT_REF) 3502 fprintf (vect_dump, " vectorizing a record based array ref: "); 3503 else if (TREE_CODE (data_ref_base) == SSA_NAME) 3504 fprintf (vect_dump, " vectorizing a pointer ref: "); 3505 print_generic_expr (vect_dump, base_name, TDF_SLIM); 3506 } 3507 3508 /* (1) Create the new aggregate-pointer variable. */ 3509 aggr_ptr_type = build_pointer_type (aggr_type); 3510 base = get_base_address (DR_REF (dr)); 3511 if (base 3512 && TREE_CODE (base) == MEM_REF) 3513 aggr_ptr_type 3514 = build_qualified_type (aggr_ptr_type, 3515 TYPE_QUALS (TREE_TYPE (TREE_OPERAND (base, 0)))); 3516 aggr_ptr = vect_get_new_vect_var (aggr_ptr_type, vect_pointer_var, 3517 get_name (base_name)); 3518 3519 /* Vector and array types inherit the alias set of their component 3520 type by default so we need to use a ref-all pointer if the data 3521 reference does not conflict with the created aggregated data 3522 reference because it is not addressable. */ 3523 if (!alias_sets_conflict_p (get_deref_alias_set (aggr_ptr), 3524 get_alias_set (DR_REF (dr)))) 3525 { 3526 aggr_ptr_type 3527 = build_pointer_type_for_mode (aggr_type, 3528 TYPE_MODE (aggr_ptr_type), true); 3529 aggr_ptr = vect_get_new_vect_var (aggr_ptr_type, vect_pointer_var, 3530 get_name (base_name)); 3531 } 3532 3533 /* Likewise for any of the data references in the stmt group. */ 3534 else if (STMT_VINFO_GROUP_SIZE (stmt_info) > 1) 3535 { 3536 gimple orig_stmt = STMT_VINFO_GROUP_FIRST_ELEMENT (stmt_info); 3537 do 3538 { 3539 tree lhs = gimple_assign_lhs (orig_stmt); 3540 if (!alias_sets_conflict_p (get_deref_alias_set (aggr_ptr), 3541 get_alias_set (lhs))) 3542 { 3543 aggr_ptr_type 3544 = build_pointer_type_for_mode (aggr_type, 3545 TYPE_MODE (aggr_ptr_type), true); 3546 aggr_ptr 3547 = vect_get_new_vect_var (aggr_ptr_type, vect_pointer_var, 3548 get_name (base_name)); 3549 break; 3550 } 3551 3552 orig_stmt = STMT_VINFO_GROUP_NEXT_ELEMENT (vinfo_for_stmt (orig_stmt)); 3553 } 3554 while (orig_stmt); 3555 } 3556 3557 add_referenced_var (aggr_ptr); 3558 3559 /* Note: If the dataref is in an inner-loop nested in LOOP, and we are 3560 vectorizing LOOP (i.e., outer-loop vectorization), we need to create two 3561 def-use update cycles for the pointer: one relative to the outer-loop 3562 (LOOP), which is what steps (3) and (4) below do. The other is relative 3563 to the inner-loop (which is the inner-most loop containing the dataref), 3564 and this is done be step (5) below. 3565 3566 When vectorizing inner-most loops, the vectorized loop (LOOP) is also the 3567 inner-most loop, and so steps (3),(4) work the same, and step (5) is 3568 redundant. Steps (3),(4) create the following: 3569 3570 vp0 = &base_addr; 3571 LOOP: vp1 = phi(vp0,vp2) 3572 ... 3573 ... 3574 vp2 = vp1 + step 3575 goto LOOP 3576 3577 If there is an inner-loop nested in loop, then step (5) will also be 3578 applied, and an additional update in the inner-loop will be created: 3579 3580 vp0 = &base_addr; 3581 LOOP: vp1 = phi(vp0,vp2) 3582 ... 3583 inner: vp3 = phi(vp1,vp4) 3584 vp4 = vp3 + inner_step 3585 if () goto inner 3586 ... 3587 vp2 = vp1 + step 3588 if () goto LOOP */ 3589 3590 /* (2) Calculate the initial address of the aggregate-pointer, and set 3591 the aggregate-pointer to point to it before the loop. */ 3592 3593 /* Create: (&(base[init_val+offset]) in the loop preheader. */ 3594 3595 new_temp = vect_create_addr_base_for_vector_ref (stmt, &new_stmt_list, 3596 offset, loop); 3597 if (new_stmt_list) 3598 { 3599 if (pe) 3600 { 3601 new_bb = gsi_insert_seq_on_edge_immediate (pe, new_stmt_list); 3602 gcc_assert (!new_bb); 3603 } 3604 else 3605 gsi_insert_seq_before (gsi, new_stmt_list, GSI_SAME_STMT); 3606 } 3607 3608 *initial_address = new_temp; 3609 3610 /* Create: p = (aggr_type *) initial_base */ 3611 if (TREE_CODE (new_temp) != SSA_NAME 3612 || !useless_type_conversion_p (aggr_ptr_type, TREE_TYPE (new_temp))) 3613 { 3614 vec_stmt = gimple_build_assign (aggr_ptr, 3615 fold_convert (aggr_ptr_type, new_temp)); 3616 aggr_ptr_init = make_ssa_name (aggr_ptr, vec_stmt); 3617 /* Copy the points-to information if it exists. */ 3618 if (DR_PTR_INFO (dr)) 3619 duplicate_ssa_name_ptr_info (aggr_ptr_init, DR_PTR_INFO (dr)); 3620 gimple_assign_set_lhs (vec_stmt, aggr_ptr_init); 3621 if (pe) 3622 { 3623 new_bb = gsi_insert_on_edge_immediate (pe, vec_stmt); 3624 gcc_assert (!new_bb); 3625 } 3626 else 3627 gsi_insert_before (gsi, vec_stmt, GSI_SAME_STMT); 3628 } 3629 else 3630 aggr_ptr_init = new_temp; 3631 3632 /* (3) Handle the updating of the aggregate-pointer inside the loop. 3633 This is needed when ONLY_INIT is false, and also when AT_LOOP is the 3634 inner-loop nested in LOOP (during outer-loop vectorization). */ 3635 3636 /* No update in loop is required. */ 3637 if (only_init && (!loop_vinfo || at_loop == loop)) 3638 aptr = aggr_ptr_init; 3639 else 3640 { 3641 /* The step of the aggregate pointer is the type size. */ 3642 tree step = TYPE_SIZE_UNIT (aggr_type); 3643 /* One exception to the above is when the scalar step of the load in 3644 LOOP is zero. In this case the step here is also zero. */ 3645 if (*inv_p) 3646 step = size_zero_node; 3647 else if (negative) 3648 step = fold_build1 (NEGATE_EXPR, TREE_TYPE (step), step); 3649 3650 standard_iv_increment_position (loop, &incr_gsi, &insert_after); 3651 3652 create_iv (aggr_ptr_init, 3653 fold_convert (aggr_ptr_type, step), 3654 aggr_ptr, loop, &incr_gsi, insert_after, 3655 &indx_before_incr, &indx_after_incr); 3656 incr = gsi_stmt (incr_gsi); 3657 set_vinfo_for_stmt (incr, new_stmt_vec_info (incr, loop_vinfo, NULL)); 3658 3659 /* Copy the points-to information if it exists. */ 3660 if (DR_PTR_INFO (dr)) 3661 { 3662 duplicate_ssa_name_ptr_info (indx_before_incr, DR_PTR_INFO (dr)); 3663 duplicate_ssa_name_ptr_info (indx_after_incr, DR_PTR_INFO (dr)); 3664 } 3665 if (ptr_incr) 3666 *ptr_incr = incr; 3667 3668 aptr = indx_before_incr; 3669 } 3670 3671 if (!nested_in_vect_loop || only_init) 3672 return aptr; 3673 3674 3675 /* (4) Handle the updating of the aggregate-pointer inside the inner-loop 3676 nested in LOOP, if exists. */ 3677 3678 gcc_assert (nested_in_vect_loop); 3679 if (!only_init) 3680 { 3681 standard_iv_increment_position (containing_loop, &incr_gsi, 3682 &insert_after); 3683 create_iv (aptr, fold_convert (aggr_ptr_type, DR_STEP (dr)), aggr_ptr, 3684 containing_loop, &incr_gsi, insert_after, &indx_before_incr, 3685 &indx_after_incr); 3686 incr = gsi_stmt (incr_gsi); 3687 set_vinfo_for_stmt (incr, new_stmt_vec_info (incr, loop_vinfo, NULL)); 3688 3689 /* Copy the points-to information if it exists. */ 3690 if (DR_PTR_INFO (dr)) 3691 { 3692 duplicate_ssa_name_ptr_info (indx_before_incr, DR_PTR_INFO (dr)); 3693 duplicate_ssa_name_ptr_info (indx_after_incr, DR_PTR_INFO (dr)); 3694 } 3695 if (ptr_incr) 3696 *ptr_incr = incr; 3697 3698 return indx_before_incr; 3699 } 3700 else 3701 gcc_unreachable (); 3702 } 3703 3704 3705 /* Function bump_vector_ptr 3706 3707 Increment a pointer (to a vector type) by vector-size. If requested, 3708 i.e. if PTR-INCR is given, then also connect the new increment stmt 3709 to the existing def-use update-chain of the pointer, by modifying 3710 the PTR_INCR as illustrated below: 3711 3712 The pointer def-use update-chain before this function: 3713 DATAREF_PTR = phi (p_0, p_2) 3714 .... 3715 PTR_INCR: p_2 = DATAREF_PTR + step 3716 3717 The pointer def-use update-chain after this function: 3718 DATAREF_PTR = phi (p_0, p_2) 3719 .... 3720 NEW_DATAREF_PTR = DATAREF_PTR + BUMP 3721 .... 3722 PTR_INCR: p_2 = NEW_DATAREF_PTR + step 3723 3724 Input: 3725 DATAREF_PTR - ssa_name of a pointer (to vector type) that is being updated 3726 in the loop. 3727 PTR_INCR - optional. The stmt that updates the pointer in each iteration of 3728 the loop. The increment amount across iterations is expected 3729 to be vector_size. 3730 BSI - location where the new update stmt is to be placed. 3731 STMT - the original scalar memory-access stmt that is being vectorized. 3732 BUMP - optional. The offset by which to bump the pointer. If not given, 3733 the offset is assumed to be vector_size. 3734 3735 Output: Return NEW_DATAREF_PTR as illustrated above. 3736 3737 */ 3738 3739 tree 3740 bump_vector_ptr (tree dataref_ptr, gimple ptr_incr, gimple_stmt_iterator *gsi, 3741 gimple stmt, tree bump) 3742 { 3743 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 3744 struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info); 3745 tree vectype = STMT_VINFO_VECTYPE (stmt_info); 3746 tree ptr_var = SSA_NAME_VAR (dataref_ptr); 3747 tree update = TYPE_SIZE_UNIT (vectype); 3748 gimple incr_stmt; 3749 ssa_op_iter iter; 3750 use_operand_p use_p; 3751 tree new_dataref_ptr; 3752 3753 if (bump) 3754 update = bump; 3755 3756 incr_stmt = gimple_build_assign_with_ops (POINTER_PLUS_EXPR, ptr_var, 3757 dataref_ptr, update); 3758 new_dataref_ptr = make_ssa_name (ptr_var, incr_stmt); 3759 gimple_assign_set_lhs (incr_stmt, new_dataref_ptr); 3760 vect_finish_stmt_generation (stmt, incr_stmt, gsi); 3761 3762 /* Copy the points-to information if it exists. */ 3763 if (DR_PTR_INFO (dr)) 3764 { 3765 duplicate_ssa_name_ptr_info (new_dataref_ptr, DR_PTR_INFO (dr)); 3766 SSA_NAME_PTR_INFO (new_dataref_ptr)->align = 1; 3767 SSA_NAME_PTR_INFO (new_dataref_ptr)->misalign = 0; 3768 } 3769 3770 if (!ptr_incr) 3771 return new_dataref_ptr; 3772 3773 /* Update the vector-pointer's cross-iteration increment. */ 3774 FOR_EACH_SSA_USE_OPERAND (use_p, ptr_incr, iter, SSA_OP_USE) 3775 { 3776 tree use = USE_FROM_PTR (use_p); 3777 3778 if (use == dataref_ptr) 3779 SET_USE (use_p, new_dataref_ptr); 3780 else 3781 gcc_assert (tree_int_cst_compare (use, update) == 0); 3782 } 3783 3784 return new_dataref_ptr; 3785 } 3786 3787 3788 /* Function vect_create_destination_var. 3789 3790 Create a new temporary of type VECTYPE. */ 3791 3792 tree 3793 vect_create_destination_var (tree scalar_dest, tree vectype) 3794 { 3795 tree vec_dest; 3796 const char *new_name; 3797 tree type; 3798 enum vect_var_kind kind; 3799 3800 kind = vectype ? vect_simple_var : vect_scalar_var; 3801 type = vectype ? vectype : TREE_TYPE (scalar_dest); 3802 3803 gcc_assert (TREE_CODE (scalar_dest) == SSA_NAME); 3804 3805 new_name = get_name (scalar_dest); 3806 if (!new_name) 3807 new_name = "var_"; 3808 vec_dest = vect_get_new_vect_var (type, kind, new_name); 3809 add_referenced_var (vec_dest); 3810 3811 return vec_dest; 3812 } 3813 3814 /* Function vect_strided_store_supported. 3815 3816 Returns TRUE if interleave high and interleave low permutations 3817 are supported, and FALSE otherwise. */ 3818 3819 bool 3820 vect_strided_store_supported (tree vectype, unsigned HOST_WIDE_INT count) 3821 { 3822 enum machine_mode mode = TYPE_MODE (vectype); 3823 3824 /* vect_permute_store_chain requires the group size to be a power of two. */ 3825 if (exact_log2 (count) == -1) 3826 { 3827 if (vect_print_dump_info (REPORT_DETAILS)) 3828 fprintf (vect_dump, "the size of the group of strided accesses" 3829 " is not a power of 2"); 3830 return false; 3831 } 3832 3833 /* Check that the permutation is supported. */ 3834 if (VECTOR_MODE_P (mode)) 3835 { 3836 unsigned int i, nelt = GET_MODE_NUNITS (mode); 3837 unsigned char *sel = XALLOCAVEC (unsigned char, nelt); 3838 for (i = 0; i < nelt / 2; i++) 3839 { 3840 sel[i * 2] = i; 3841 sel[i * 2 + 1] = i + nelt; 3842 } 3843 if (can_vec_perm_p (mode, false, sel)) 3844 { 3845 for (i = 0; i < nelt; i++) 3846 sel[i] += nelt / 2; 3847 if (can_vec_perm_p (mode, false, sel)) 3848 return true; 3849 } 3850 } 3851 3852 if (vect_print_dump_info (REPORT_DETAILS)) 3853 fprintf (vect_dump, "interleave op not supported by target."); 3854 return false; 3855 } 3856 3857 3858 /* Return TRUE if vec_store_lanes is available for COUNT vectors of 3859 type VECTYPE. */ 3860 3861 bool 3862 vect_store_lanes_supported (tree vectype, unsigned HOST_WIDE_INT count) 3863 { 3864 return vect_lanes_optab_supported_p ("vec_store_lanes", 3865 vec_store_lanes_optab, 3866 vectype, count); 3867 } 3868 3869 3870 /* Function vect_permute_store_chain. 3871 3872 Given a chain of interleaved stores in DR_CHAIN of LENGTH that must be 3873 a power of 2, generate interleave_high/low stmts to reorder the data 3874 correctly for the stores. Return the final references for stores in 3875 RESULT_CHAIN. 3876 3877 E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8. 3878 The input is 4 vectors each containing 8 elements. We assign a number to 3879 each element, the input sequence is: 3880 3881 1st vec: 0 1 2 3 4 5 6 7 3882 2nd vec: 8 9 10 11 12 13 14 15 3883 3rd vec: 16 17 18 19 20 21 22 23 3884 4th vec: 24 25 26 27 28 29 30 31 3885 3886 The output sequence should be: 3887 3888 1st vec: 0 8 16 24 1 9 17 25 3889 2nd vec: 2 10 18 26 3 11 19 27 3890 3rd vec: 4 12 20 28 5 13 21 30 3891 4th vec: 6 14 22 30 7 15 23 31 3892 3893 i.e., we interleave the contents of the four vectors in their order. 3894 3895 We use interleave_high/low instructions to create such output. The input of 3896 each interleave_high/low operation is two vectors: 3897 1st vec 2nd vec 3898 0 1 2 3 4 5 6 7 3899 the even elements of the result vector are obtained left-to-right from the 3900 high/low elements of the first vector. The odd elements of the result are 3901 obtained left-to-right from the high/low elements of the second vector. 3902 The output of interleave_high will be: 0 4 1 5 3903 and of interleave_low: 2 6 3 7 3904 3905 3906 The permutation is done in log LENGTH stages. In each stage interleave_high 3907 and interleave_low stmts are created for each pair of vectors in DR_CHAIN, 3908 where the first argument is taken from the first half of DR_CHAIN and the 3909 second argument from it's second half. 3910 In our example, 3911 3912 I1: interleave_high (1st vec, 3rd vec) 3913 I2: interleave_low (1st vec, 3rd vec) 3914 I3: interleave_high (2nd vec, 4th vec) 3915 I4: interleave_low (2nd vec, 4th vec) 3916 3917 The output for the first stage is: 3918 3919 I1: 0 16 1 17 2 18 3 19 3920 I2: 4 20 5 21 6 22 7 23 3921 I3: 8 24 9 25 10 26 11 27 3922 I4: 12 28 13 29 14 30 15 31 3923 3924 The output of the second stage, i.e. the final result is: 3925 3926 I1: 0 8 16 24 1 9 17 25 3927 I2: 2 10 18 26 3 11 19 27 3928 I3: 4 12 20 28 5 13 21 30 3929 I4: 6 14 22 30 7 15 23 31. */ 3930 3931 void 3932 vect_permute_store_chain (VEC(tree,heap) *dr_chain, 3933 unsigned int length, 3934 gimple stmt, 3935 gimple_stmt_iterator *gsi, 3936 VEC(tree,heap) **result_chain) 3937 { 3938 tree perm_dest, vect1, vect2, high, low; 3939 gimple perm_stmt; 3940 tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt)); 3941 tree perm_mask_low, perm_mask_high; 3942 unsigned int i, n; 3943 unsigned int j, nelt = TYPE_VECTOR_SUBPARTS (vectype); 3944 unsigned char *sel = XALLOCAVEC (unsigned char, nelt); 3945 3946 *result_chain = VEC_copy (tree, heap, dr_chain); 3947 3948 for (i = 0, n = nelt / 2; i < n; i++) 3949 { 3950 sel[i * 2] = i; 3951 sel[i * 2 + 1] = i + nelt; 3952 } 3953 perm_mask_high = vect_gen_perm_mask (vectype, sel); 3954 gcc_assert (perm_mask_high != NULL); 3955 3956 for (i = 0; i < nelt; i++) 3957 sel[i] += nelt / 2; 3958 perm_mask_low = vect_gen_perm_mask (vectype, sel); 3959 gcc_assert (perm_mask_low != NULL); 3960 3961 for (i = 0, n = exact_log2 (length); i < n; i++) 3962 { 3963 for (j = 0; j < length/2; j++) 3964 { 3965 vect1 = VEC_index (tree, dr_chain, j); 3966 vect2 = VEC_index (tree, dr_chain, j+length/2); 3967 3968 /* Create interleaving stmt: 3969 high = VEC_PERM_EXPR <vect1, vect2, {0, nelt, 1, nelt+1, ...}> */ 3970 perm_dest = create_tmp_var (vectype, "vect_inter_high"); 3971 DECL_GIMPLE_REG_P (perm_dest) = 1; 3972 add_referenced_var (perm_dest); 3973 high = make_ssa_name (perm_dest, NULL); 3974 perm_stmt 3975 = gimple_build_assign_with_ops3 (VEC_PERM_EXPR, high, 3976 vect1, vect2, perm_mask_high); 3977 vect_finish_stmt_generation (stmt, perm_stmt, gsi); 3978 VEC_replace (tree, *result_chain, 2*j, high); 3979 3980 /* Create interleaving stmt: 3981 low = VEC_PERM_EXPR <vect1, vect2, {nelt/2, nelt*3/2, nelt/2+1, 3982 nelt*3/2+1, ...}> */ 3983 perm_dest = create_tmp_var (vectype, "vect_inter_low"); 3984 DECL_GIMPLE_REG_P (perm_dest) = 1; 3985 add_referenced_var (perm_dest); 3986 low = make_ssa_name (perm_dest, NULL); 3987 perm_stmt 3988 = gimple_build_assign_with_ops3 (VEC_PERM_EXPR, low, 3989 vect1, vect2, perm_mask_low); 3990 vect_finish_stmt_generation (stmt, perm_stmt, gsi); 3991 VEC_replace (tree, *result_chain, 2*j+1, low); 3992 } 3993 dr_chain = VEC_copy (tree, heap, *result_chain); 3994 } 3995 } 3996 3997 /* Function vect_setup_realignment 3998 3999 This function is called when vectorizing an unaligned load using 4000 the dr_explicit_realign[_optimized] scheme. 4001 This function generates the following code at the loop prolog: 4002 4003 p = initial_addr; 4004 x msq_init = *(floor(p)); # prolog load 4005 realignment_token = call target_builtin; 4006 loop: 4007 x msq = phi (msq_init, ---) 4008 4009 The stmts marked with x are generated only for the case of 4010 dr_explicit_realign_optimized. 4011 4012 The code above sets up a new (vector) pointer, pointing to the first 4013 location accessed by STMT, and a "floor-aligned" load using that pointer. 4014 It also generates code to compute the "realignment-token" (if the relevant 4015 target hook was defined), and creates a phi-node at the loop-header bb 4016 whose arguments are the result of the prolog-load (created by this 4017 function) and the result of a load that takes place in the loop (to be 4018 created by the caller to this function). 4019 4020 For the case of dr_explicit_realign_optimized: 4021 The caller to this function uses the phi-result (msq) to create the 4022 realignment code inside the loop, and sets up the missing phi argument, 4023 as follows: 4024 loop: 4025 msq = phi (msq_init, lsq) 4026 lsq = *(floor(p')); # load in loop 4027 result = realign_load (msq, lsq, realignment_token); 4028 4029 For the case of dr_explicit_realign: 4030 loop: 4031 msq = *(floor(p)); # load in loop 4032 p' = p + (VS-1); 4033 lsq = *(floor(p')); # load in loop 4034 result = realign_load (msq, lsq, realignment_token); 4035 4036 Input: 4037 STMT - (scalar) load stmt to be vectorized. This load accesses 4038 a memory location that may be unaligned. 4039 BSI - place where new code is to be inserted. 4040 ALIGNMENT_SUPPORT_SCHEME - which of the two misalignment handling schemes 4041 is used. 4042 4043 Output: 4044 REALIGNMENT_TOKEN - the result of a call to the builtin_mask_for_load 4045 target hook, if defined. 4046 Return value - the result of the loop-header phi node. */ 4047 4048 tree 4049 vect_setup_realignment (gimple stmt, gimple_stmt_iterator *gsi, 4050 tree *realignment_token, 4051 enum dr_alignment_support alignment_support_scheme, 4052 tree init_addr, 4053 struct loop **at_loop) 4054 { 4055 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 4056 tree vectype = STMT_VINFO_VECTYPE (stmt_info); 4057 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 4058 struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info); 4059 struct loop *loop = NULL; 4060 edge pe = NULL; 4061 tree scalar_dest = gimple_assign_lhs (stmt); 4062 tree vec_dest; 4063 gimple inc; 4064 tree ptr; 4065 tree data_ref; 4066 gimple new_stmt; 4067 basic_block new_bb; 4068 tree msq_init = NULL_TREE; 4069 tree new_temp; 4070 gimple phi_stmt; 4071 tree msq = NULL_TREE; 4072 gimple_seq stmts = NULL; 4073 bool inv_p; 4074 bool compute_in_loop = false; 4075 bool nested_in_vect_loop = false; 4076 struct loop *containing_loop = (gimple_bb (stmt))->loop_father; 4077 struct loop *loop_for_initial_load = NULL; 4078 4079 if (loop_vinfo) 4080 { 4081 loop = LOOP_VINFO_LOOP (loop_vinfo); 4082 nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt); 4083 } 4084 4085 gcc_assert (alignment_support_scheme == dr_explicit_realign 4086 || alignment_support_scheme == dr_explicit_realign_optimized); 4087 4088 /* We need to generate three things: 4089 1. the misalignment computation 4090 2. the extra vector load (for the optimized realignment scheme). 4091 3. the phi node for the two vectors from which the realignment is 4092 done (for the optimized realignment scheme). */ 4093 4094 /* 1. Determine where to generate the misalignment computation. 4095 4096 If INIT_ADDR is NULL_TREE, this indicates that the misalignment 4097 calculation will be generated by this function, outside the loop (in the 4098 preheader). Otherwise, INIT_ADDR had already been computed for us by the 4099 caller, inside the loop. 4100 4101 Background: If the misalignment remains fixed throughout the iterations of 4102 the loop, then both realignment schemes are applicable, and also the 4103 misalignment computation can be done outside LOOP. This is because we are 4104 vectorizing LOOP, and so the memory accesses in LOOP advance in steps that 4105 are a multiple of VS (the Vector Size), and therefore the misalignment in 4106 different vectorized LOOP iterations is always the same. 4107 The problem arises only if the memory access is in an inner-loop nested 4108 inside LOOP, which is now being vectorized using outer-loop vectorization. 4109 This is the only case when the misalignment of the memory access may not 4110 remain fixed throughout the iterations of the inner-loop (as explained in 4111 detail in vect_supportable_dr_alignment). In this case, not only is the 4112 optimized realignment scheme not applicable, but also the misalignment 4113 computation (and generation of the realignment token that is passed to 4114 REALIGN_LOAD) have to be done inside the loop. 4115 4116 In short, INIT_ADDR indicates whether we are in a COMPUTE_IN_LOOP mode 4117 or not, which in turn determines if the misalignment is computed inside 4118 the inner-loop, or outside LOOP. */ 4119 4120 if (init_addr != NULL_TREE || !loop_vinfo) 4121 { 4122 compute_in_loop = true; 4123 gcc_assert (alignment_support_scheme == dr_explicit_realign); 4124 } 4125 4126 4127 /* 2. Determine where to generate the extra vector load. 4128 4129 For the optimized realignment scheme, instead of generating two vector 4130 loads in each iteration, we generate a single extra vector load in the 4131 preheader of the loop, and in each iteration reuse the result of the 4132 vector load from the previous iteration. In case the memory access is in 4133 an inner-loop nested inside LOOP, which is now being vectorized using 4134 outer-loop vectorization, we need to determine whether this initial vector 4135 load should be generated at the preheader of the inner-loop, or can be 4136 generated at the preheader of LOOP. If the memory access has no evolution 4137 in LOOP, it can be generated in the preheader of LOOP. Otherwise, it has 4138 to be generated inside LOOP (in the preheader of the inner-loop). */ 4139 4140 if (nested_in_vect_loop) 4141 { 4142 tree outerloop_step = STMT_VINFO_DR_STEP (stmt_info); 4143 bool invariant_in_outerloop = 4144 (tree_int_cst_compare (outerloop_step, size_zero_node) == 0); 4145 loop_for_initial_load = (invariant_in_outerloop ? loop : loop->inner); 4146 } 4147 else 4148 loop_for_initial_load = loop; 4149 if (at_loop) 4150 *at_loop = loop_for_initial_load; 4151 4152 if (loop_for_initial_load) 4153 pe = loop_preheader_edge (loop_for_initial_load); 4154 4155 /* 3. For the case of the optimized realignment, create the first vector 4156 load at the loop preheader. */ 4157 4158 if (alignment_support_scheme == dr_explicit_realign_optimized) 4159 { 4160 /* Create msq_init = *(floor(p1)) in the loop preheader */ 4161 4162 gcc_assert (!compute_in_loop); 4163 vec_dest = vect_create_destination_var (scalar_dest, vectype); 4164 ptr = vect_create_data_ref_ptr (stmt, vectype, loop_for_initial_load, 4165 NULL_TREE, &init_addr, NULL, &inc, 4166 true, &inv_p); 4167 new_stmt = gimple_build_assign_with_ops 4168 (BIT_AND_EXPR, NULL_TREE, ptr, 4169 build_int_cst (TREE_TYPE (ptr), 4170 -(HOST_WIDE_INT)TYPE_ALIGN_UNIT (vectype))); 4171 new_temp = make_ssa_name (SSA_NAME_VAR (ptr), new_stmt); 4172 gimple_assign_set_lhs (new_stmt, new_temp); 4173 new_bb = gsi_insert_on_edge_immediate (pe, new_stmt); 4174 gcc_assert (!new_bb); 4175 data_ref 4176 = build2 (MEM_REF, TREE_TYPE (vec_dest), new_temp, 4177 build_int_cst (reference_alias_ptr_type (DR_REF (dr)), 0)); 4178 new_stmt = gimple_build_assign (vec_dest, data_ref); 4179 new_temp = make_ssa_name (vec_dest, new_stmt); 4180 gimple_assign_set_lhs (new_stmt, new_temp); 4181 mark_symbols_for_renaming (new_stmt); 4182 if (pe) 4183 { 4184 new_bb = gsi_insert_on_edge_immediate (pe, new_stmt); 4185 gcc_assert (!new_bb); 4186 } 4187 else 4188 gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT); 4189 4190 msq_init = gimple_assign_lhs (new_stmt); 4191 } 4192 4193 /* 4. Create realignment token using a target builtin, if available. 4194 It is done either inside the containing loop, or before LOOP (as 4195 determined above). */ 4196 4197 if (targetm.vectorize.builtin_mask_for_load) 4198 { 4199 tree builtin_decl; 4200 4201 /* Compute INIT_ADDR - the initial addressed accessed by this memref. */ 4202 if (!init_addr) 4203 { 4204 /* Generate the INIT_ADDR computation outside LOOP. */ 4205 init_addr = vect_create_addr_base_for_vector_ref (stmt, &stmts, 4206 NULL_TREE, loop); 4207 if (loop) 4208 { 4209 pe = loop_preheader_edge (loop); 4210 new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts); 4211 gcc_assert (!new_bb); 4212 } 4213 else 4214 gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT); 4215 } 4216 4217 builtin_decl = targetm.vectorize.builtin_mask_for_load (); 4218 new_stmt = gimple_build_call (builtin_decl, 1, init_addr); 4219 vec_dest = 4220 vect_create_destination_var (scalar_dest, 4221 gimple_call_return_type (new_stmt)); 4222 new_temp = make_ssa_name (vec_dest, new_stmt); 4223 gimple_call_set_lhs (new_stmt, new_temp); 4224 4225 if (compute_in_loop) 4226 gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT); 4227 else 4228 { 4229 /* Generate the misalignment computation outside LOOP. */ 4230 pe = loop_preheader_edge (loop); 4231 new_bb = gsi_insert_on_edge_immediate (pe, new_stmt); 4232 gcc_assert (!new_bb); 4233 } 4234 4235 *realignment_token = gimple_call_lhs (new_stmt); 4236 4237 /* The result of the CALL_EXPR to this builtin is determined from 4238 the value of the parameter and no global variables are touched 4239 which makes the builtin a "const" function. Requiring the 4240 builtin to have the "const" attribute makes it unnecessary 4241 to call mark_call_clobbered. */ 4242 gcc_assert (TREE_READONLY (builtin_decl)); 4243 } 4244 4245 if (alignment_support_scheme == dr_explicit_realign) 4246 return msq; 4247 4248 gcc_assert (!compute_in_loop); 4249 gcc_assert (alignment_support_scheme == dr_explicit_realign_optimized); 4250 4251 4252 /* 5. Create msq = phi <msq_init, lsq> in loop */ 4253 4254 pe = loop_preheader_edge (containing_loop); 4255 vec_dest = vect_create_destination_var (scalar_dest, vectype); 4256 msq = make_ssa_name (vec_dest, NULL); 4257 phi_stmt = create_phi_node (msq, containing_loop->header); 4258 SSA_NAME_DEF_STMT (msq) = phi_stmt; 4259 add_phi_arg (phi_stmt, msq_init, pe, UNKNOWN_LOCATION); 4260 4261 return msq; 4262 } 4263 4264 4265 /* Function vect_strided_load_supported. 4266 4267 Returns TRUE if even and odd permutations are supported, 4268 and FALSE otherwise. */ 4269 4270 bool 4271 vect_strided_load_supported (tree vectype, unsigned HOST_WIDE_INT count) 4272 { 4273 enum machine_mode mode = TYPE_MODE (vectype); 4274 4275 /* vect_permute_load_chain requires the group size to be a power of two. */ 4276 if (exact_log2 (count) == -1) 4277 { 4278 if (vect_print_dump_info (REPORT_DETAILS)) 4279 fprintf (vect_dump, "the size of the group of strided accesses" 4280 " is not a power of 2"); 4281 return false; 4282 } 4283 4284 /* Check that the permutation is supported. */ 4285 if (VECTOR_MODE_P (mode)) 4286 { 4287 unsigned int i, nelt = GET_MODE_NUNITS (mode); 4288 unsigned char *sel = XALLOCAVEC (unsigned char, nelt); 4289 4290 for (i = 0; i < nelt; i++) 4291 sel[i] = i * 2; 4292 if (can_vec_perm_p (mode, false, sel)) 4293 { 4294 for (i = 0; i < nelt; i++) 4295 sel[i] = i * 2 + 1; 4296 if (can_vec_perm_p (mode, false, sel)) 4297 return true; 4298 } 4299 } 4300 4301 if (vect_print_dump_info (REPORT_DETAILS)) 4302 fprintf (vect_dump, "extract even/odd not supported by target"); 4303 return false; 4304 } 4305 4306 /* Return TRUE if vec_load_lanes is available for COUNT vectors of 4307 type VECTYPE. */ 4308 4309 bool 4310 vect_load_lanes_supported (tree vectype, unsigned HOST_WIDE_INT count) 4311 { 4312 return vect_lanes_optab_supported_p ("vec_load_lanes", 4313 vec_load_lanes_optab, 4314 vectype, count); 4315 } 4316 4317 /* Function vect_permute_load_chain. 4318 4319 Given a chain of interleaved loads in DR_CHAIN of LENGTH that must be 4320 a power of 2, generate extract_even/odd stmts to reorder the input data 4321 correctly. Return the final references for loads in RESULT_CHAIN. 4322 4323 E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8. 4324 The input is 4 vectors each containing 8 elements. We assign a number to each 4325 element, the input sequence is: 4326 4327 1st vec: 0 1 2 3 4 5 6 7 4328 2nd vec: 8 9 10 11 12 13 14 15 4329 3rd vec: 16 17 18 19 20 21 22 23 4330 4th vec: 24 25 26 27 28 29 30 31 4331 4332 The output sequence should be: 4333 4334 1st vec: 0 4 8 12 16 20 24 28 4335 2nd vec: 1 5 9 13 17 21 25 29 4336 3rd vec: 2 6 10 14 18 22 26 30 4337 4th vec: 3 7 11 15 19 23 27 31 4338 4339 i.e., the first output vector should contain the first elements of each 4340 interleaving group, etc. 4341 4342 We use extract_even/odd instructions to create such output. The input of 4343 each extract_even/odd operation is two vectors 4344 1st vec 2nd vec 4345 0 1 2 3 4 5 6 7 4346 4347 and the output is the vector of extracted even/odd elements. The output of 4348 extract_even will be: 0 2 4 6 4349 and of extract_odd: 1 3 5 7 4350 4351 4352 The permutation is done in log LENGTH stages. In each stage extract_even 4353 and extract_odd stmts are created for each pair of vectors in DR_CHAIN in 4354 their order. In our example, 4355 4356 E1: extract_even (1st vec, 2nd vec) 4357 E2: extract_odd (1st vec, 2nd vec) 4358 E3: extract_even (3rd vec, 4th vec) 4359 E4: extract_odd (3rd vec, 4th vec) 4360 4361 The output for the first stage will be: 4362 4363 E1: 0 2 4 6 8 10 12 14 4364 E2: 1 3 5 7 9 11 13 15 4365 E3: 16 18 20 22 24 26 28 30 4366 E4: 17 19 21 23 25 27 29 31 4367 4368 In order to proceed and create the correct sequence for the next stage (or 4369 for the correct output, if the second stage is the last one, as in our 4370 example), we first put the output of extract_even operation and then the 4371 output of extract_odd in RESULT_CHAIN (which is then copied to DR_CHAIN). 4372 The input for the second stage is: 4373 4374 1st vec (E1): 0 2 4 6 8 10 12 14 4375 2nd vec (E3): 16 18 20 22 24 26 28 30 4376 3rd vec (E2): 1 3 5 7 9 11 13 15 4377 4th vec (E4): 17 19 21 23 25 27 29 31 4378 4379 The output of the second stage: 4380 4381 E1: 0 4 8 12 16 20 24 28 4382 E2: 2 6 10 14 18 22 26 30 4383 E3: 1 5 9 13 17 21 25 29 4384 E4: 3 7 11 15 19 23 27 31 4385 4386 And RESULT_CHAIN after reordering: 4387 4388 1st vec (E1): 0 4 8 12 16 20 24 28 4389 2nd vec (E3): 1 5 9 13 17 21 25 29 4390 3rd vec (E2): 2 6 10 14 18 22 26 30 4391 4th vec (E4): 3 7 11 15 19 23 27 31. */ 4392 4393 static void 4394 vect_permute_load_chain (VEC(tree,heap) *dr_chain, 4395 unsigned int length, 4396 gimple stmt, 4397 gimple_stmt_iterator *gsi, 4398 VEC(tree,heap) **result_chain) 4399 { 4400 tree perm_dest, data_ref, first_vect, second_vect; 4401 tree perm_mask_even, perm_mask_odd; 4402 gimple perm_stmt; 4403 tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt)); 4404 unsigned int i, j, log_length = exact_log2 (length); 4405 unsigned nelt = TYPE_VECTOR_SUBPARTS (vectype); 4406 unsigned char *sel = XALLOCAVEC (unsigned char, nelt); 4407 4408 *result_chain = VEC_copy (tree, heap, dr_chain); 4409 4410 for (i = 0; i < nelt; ++i) 4411 sel[i] = i * 2; 4412 perm_mask_even = vect_gen_perm_mask (vectype, sel); 4413 gcc_assert (perm_mask_even != NULL); 4414 4415 for (i = 0; i < nelt; ++i) 4416 sel[i] = i * 2 + 1; 4417 perm_mask_odd = vect_gen_perm_mask (vectype, sel); 4418 gcc_assert (perm_mask_odd != NULL); 4419 4420 for (i = 0; i < log_length; i++) 4421 { 4422 for (j = 0; j < length; j += 2) 4423 { 4424 first_vect = VEC_index (tree, dr_chain, j); 4425 second_vect = VEC_index (tree, dr_chain, j+1); 4426 4427 /* data_ref = permute_even (first_data_ref, second_data_ref); */ 4428 perm_dest = create_tmp_var (vectype, "vect_perm_even"); 4429 DECL_GIMPLE_REG_P (perm_dest) = 1; 4430 add_referenced_var (perm_dest); 4431 4432 perm_stmt = gimple_build_assign_with_ops3 (VEC_PERM_EXPR, perm_dest, 4433 first_vect, second_vect, 4434 perm_mask_even); 4435 4436 data_ref = make_ssa_name (perm_dest, perm_stmt); 4437 gimple_assign_set_lhs (perm_stmt, data_ref); 4438 vect_finish_stmt_generation (stmt, perm_stmt, gsi); 4439 mark_symbols_for_renaming (perm_stmt); 4440 4441 VEC_replace (tree, *result_chain, j/2, data_ref); 4442 4443 /* data_ref = permute_odd (first_data_ref, second_data_ref); */ 4444 perm_dest = create_tmp_var (vectype, "vect_perm_odd"); 4445 DECL_GIMPLE_REG_P (perm_dest) = 1; 4446 add_referenced_var (perm_dest); 4447 4448 perm_stmt = gimple_build_assign_with_ops3 (VEC_PERM_EXPR, perm_dest, 4449 first_vect, second_vect, 4450 perm_mask_odd); 4451 4452 data_ref = make_ssa_name (perm_dest, perm_stmt); 4453 gimple_assign_set_lhs (perm_stmt, data_ref); 4454 vect_finish_stmt_generation (stmt, perm_stmt, gsi); 4455 mark_symbols_for_renaming (perm_stmt); 4456 4457 VEC_replace (tree, *result_chain, j/2+length/2, data_ref); 4458 } 4459 dr_chain = VEC_copy (tree, heap, *result_chain); 4460 } 4461 } 4462 4463 4464 /* Function vect_transform_strided_load. 4465 4466 Given a chain of input interleaved data-refs (in DR_CHAIN), build statements 4467 to perform their permutation and ascribe the result vectorized statements to 4468 the scalar statements. 4469 */ 4470 4471 void 4472 vect_transform_strided_load (gimple stmt, VEC(tree,heap) *dr_chain, int size, 4473 gimple_stmt_iterator *gsi) 4474 { 4475 VEC(tree,heap) *result_chain = NULL; 4476 4477 /* DR_CHAIN contains input data-refs that are a part of the interleaving. 4478 RESULT_CHAIN is the output of vect_permute_load_chain, it contains permuted 4479 vectors, that are ready for vector computation. */ 4480 result_chain = VEC_alloc (tree, heap, size); 4481 vect_permute_load_chain (dr_chain, size, stmt, gsi, &result_chain); 4482 vect_record_strided_load_vectors (stmt, result_chain); 4483 VEC_free (tree, heap, result_chain); 4484 } 4485 4486 /* RESULT_CHAIN contains the output of a group of strided loads that were 4487 generated as part of the vectorization of STMT. Assign the statement 4488 for each vector to the associated scalar statement. */ 4489 4490 void 4491 vect_record_strided_load_vectors (gimple stmt, VEC(tree,heap) *result_chain) 4492 { 4493 gimple first_stmt = GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)); 4494 gimple next_stmt, new_stmt; 4495 unsigned int i, gap_count; 4496 tree tmp_data_ref; 4497 4498 /* Put a permuted data-ref in the VECTORIZED_STMT field. 4499 Since we scan the chain starting from it's first node, their order 4500 corresponds the order of data-refs in RESULT_CHAIN. */ 4501 next_stmt = first_stmt; 4502 gap_count = 1; 4503 FOR_EACH_VEC_ELT (tree, result_chain, i, tmp_data_ref) 4504 { 4505 if (!next_stmt) 4506 break; 4507 4508 /* Skip the gaps. Loads created for the gaps will be removed by dead 4509 code elimination pass later. No need to check for the first stmt in 4510 the group, since it always exists. 4511 GROUP_GAP is the number of steps in elements from the previous 4512 access (if there is no gap GROUP_GAP is 1). We skip loads that 4513 correspond to the gaps. */ 4514 if (next_stmt != first_stmt 4515 && gap_count < GROUP_GAP (vinfo_for_stmt (next_stmt))) 4516 { 4517 gap_count++; 4518 continue; 4519 } 4520 4521 while (next_stmt) 4522 { 4523 new_stmt = SSA_NAME_DEF_STMT (tmp_data_ref); 4524 /* We assume that if VEC_STMT is not NULL, this is a case of multiple 4525 copies, and we put the new vector statement in the first available 4526 RELATED_STMT. */ 4527 if (!STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt))) 4528 STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)) = new_stmt; 4529 else 4530 { 4531 if (!GROUP_SAME_DR_STMT (vinfo_for_stmt (next_stmt))) 4532 { 4533 gimple prev_stmt = 4534 STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)); 4535 gimple rel_stmt = 4536 STMT_VINFO_RELATED_STMT (vinfo_for_stmt (prev_stmt)); 4537 while (rel_stmt) 4538 { 4539 prev_stmt = rel_stmt; 4540 rel_stmt = 4541 STMT_VINFO_RELATED_STMT (vinfo_for_stmt (rel_stmt)); 4542 } 4543 4544 STMT_VINFO_RELATED_STMT (vinfo_for_stmt (prev_stmt)) = 4545 new_stmt; 4546 } 4547 } 4548 4549 next_stmt = GROUP_NEXT_ELEMENT (vinfo_for_stmt (next_stmt)); 4550 gap_count = 1; 4551 /* If NEXT_STMT accesses the same DR as the previous statement, 4552 put the same TMP_DATA_REF as its vectorized statement; otherwise 4553 get the next data-ref from RESULT_CHAIN. */ 4554 if (!next_stmt || !GROUP_SAME_DR_STMT (vinfo_for_stmt (next_stmt))) 4555 break; 4556 } 4557 } 4558 } 4559 4560 /* Function vect_force_dr_alignment_p. 4561 4562 Returns whether the alignment of a DECL can be forced to be aligned 4563 on ALIGNMENT bit boundary. */ 4564 4565 bool 4566 vect_can_force_dr_alignment_p (const_tree decl, unsigned int alignment) 4567 { 4568 if (TREE_CODE (decl) != VAR_DECL) 4569 return false; 4570 4571 if (DECL_EXTERNAL (decl)) 4572 return false; 4573 4574 if (TREE_ASM_WRITTEN (decl)) 4575 return false; 4576 4577 /* Do not override explicit alignment set by the user when an explicit 4578 section name is also used. This is a common idiom used by many 4579 software projects. */ 4580 if (DECL_SECTION_NAME (decl) != NULL_TREE 4581 && !DECL_HAS_IMPLICIT_SECTION_NAME_P (decl)) 4582 return false; 4583 4584 if (TREE_STATIC (decl)) 4585 return (alignment <= MAX_OFILE_ALIGNMENT); 4586 else 4587 return (alignment <= MAX_STACK_ALIGNMENT); 4588 } 4589 4590 4591 /* Return whether the data reference DR is supported with respect to its 4592 alignment. 4593 If CHECK_ALIGNED_ACCESSES is TRUE, check if the access is supported even 4594 it is aligned, i.e., check if it is possible to vectorize it with different 4595 alignment. */ 4596 4597 enum dr_alignment_support 4598 vect_supportable_dr_alignment (struct data_reference *dr, 4599 bool check_aligned_accesses) 4600 { 4601 gimple stmt = DR_STMT (dr); 4602 stmt_vec_info stmt_info = vinfo_for_stmt (stmt); 4603 tree vectype = STMT_VINFO_VECTYPE (stmt_info); 4604 enum machine_mode mode = TYPE_MODE (vectype); 4605 loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info); 4606 struct loop *vect_loop = NULL; 4607 bool nested_in_vect_loop = false; 4608 4609 if (aligned_access_p (dr) && !check_aligned_accesses) 4610 return dr_aligned; 4611 4612 if (loop_vinfo) 4613 { 4614 vect_loop = LOOP_VINFO_LOOP (loop_vinfo); 4615 nested_in_vect_loop = nested_in_vect_loop_p (vect_loop, stmt); 4616 } 4617 4618 /* Possibly unaligned access. */ 4619 4620 /* We can choose between using the implicit realignment scheme (generating 4621 a misaligned_move stmt) and the explicit realignment scheme (generating 4622 aligned loads with a REALIGN_LOAD). There are two variants to the 4623 explicit realignment scheme: optimized, and unoptimized. 4624 We can optimize the realignment only if the step between consecutive 4625 vector loads is equal to the vector size. Since the vector memory 4626 accesses advance in steps of VS (Vector Size) in the vectorized loop, it 4627 is guaranteed that the misalignment amount remains the same throughout the 4628 execution of the vectorized loop. Therefore, we can create the 4629 "realignment token" (the permutation mask that is passed to REALIGN_LOAD) 4630 at the loop preheader. 4631 4632 However, in the case of outer-loop vectorization, when vectorizing a 4633 memory access in the inner-loop nested within the LOOP that is now being 4634 vectorized, while it is guaranteed that the misalignment of the 4635 vectorized memory access will remain the same in different outer-loop 4636 iterations, it is *not* guaranteed that is will remain the same throughout 4637 the execution of the inner-loop. This is because the inner-loop advances 4638 with the original scalar step (and not in steps of VS). If the inner-loop 4639 step happens to be a multiple of VS, then the misalignment remains fixed 4640 and we can use the optimized realignment scheme. For example: 4641 4642 for (i=0; i<N; i++) 4643 for (j=0; j<M; j++) 4644 s += a[i+j]; 4645 4646 When vectorizing the i-loop in the above example, the step between 4647 consecutive vector loads is 1, and so the misalignment does not remain 4648 fixed across the execution of the inner-loop, and the realignment cannot 4649 be optimized (as illustrated in the following pseudo vectorized loop): 4650 4651 for (i=0; i<N; i+=4) 4652 for (j=0; j<M; j++){ 4653 vs += vp[i+j]; // misalignment of &vp[i+j] is {0,1,2,3,0,1,2,3,...} 4654 // when j is {0,1,2,3,4,5,6,7,...} respectively. 4655 // (assuming that we start from an aligned address). 4656 } 4657 4658 We therefore have to use the unoptimized realignment scheme: 4659 4660 for (i=0; i<N; i+=4) 4661 for (j=k; j<M; j+=4) 4662 vs += vp[i+j]; // misalignment of &vp[i+j] is always k (assuming 4663 // that the misalignment of the initial address is 4664 // 0). 4665 4666 The loop can then be vectorized as follows: 4667 4668 for (k=0; k<4; k++){ 4669 rt = get_realignment_token (&vp[k]); 4670 for (i=0; i<N; i+=4){ 4671 v1 = vp[i+k]; 4672 for (j=k; j<M; j+=4){ 4673 v2 = vp[i+j+VS-1]; 4674 va = REALIGN_LOAD <v1,v2,rt>; 4675 vs += va; 4676 v1 = v2; 4677 } 4678 } 4679 } */ 4680 4681 if (DR_IS_READ (dr)) 4682 { 4683 bool is_packed = false; 4684 tree type = (TREE_TYPE (DR_REF (dr))); 4685 4686 if (optab_handler (vec_realign_load_optab, mode) != CODE_FOR_nothing 4687 && (!targetm.vectorize.builtin_mask_for_load 4688 || targetm.vectorize.builtin_mask_for_load ())) 4689 { 4690 tree vectype = STMT_VINFO_VECTYPE (stmt_info); 4691 if ((nested_in_vect_loop 4692 && (TREE_INT_CST_LOW (DR_STEP (dr)) 4693 != GET_MODE_SIZE (TYPE_MODE (vectype)))) 4694 || !loop_vinfo) 4695 return dr_explicit_realign; 4696 else 4697 return dr_explicit_realign_optimized; 4698 } 4699 if (!known_alignment_for_access_p (dr)) 4700 is_packed = contains_packed_reference (DR_REF (dr)); 4701 4702 if (targetm.vectorize. 4703 support_vector_misalignment (mode, type, 4704 DR_MISALIGNMENT (dr), is_packed)) 4705 /* Can't software pipeline the loads, but can at least do them. */ 4706 return dr_unaligned_supported; 4707 } 4708 else 4709 { 4710 bool is_packed = false; 4711 tree type = (TREE_TYPE (DR_REF (dr))); 4712 4713 if (!known_alignment_for_access_p (dr)) 4714 is_packed = contains_packed_reference (DR_REF (dr)); 4715 4716 if (targetm.vectorize. 4717 support_vector_misalignment (mode, type, 4718 DR_MISALIGNMENT (dr), is_packed)) 4719 return dr_unaligned_supported; 4720 } 4721 4722 /* Unsupported. */ 4723 return dr_unaligned_unsupported; 4724 } 4725