gcc/tree-vect-patterns.c

   1 /* Analysis Utilities for Loop Vectorization.
   2    Copyright (C) 2006, 2007 Free Software Foundation, Inc.
   3    Contributed by Dorit Nuzman <dorit@il.ibm.com>
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify it under
   8 the terms of the GNU General Public License as published by the Free
   9 Software Foundation; either version 3, or (at your option) any later
  10 version.
  11
  12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15 for more details.
  16
  17 You should have received a copy of the GNU General Public License
  18 along with GCC; see the file COPYING3.  If not see
  19 <http://www.gnu.org/licenses/>.  */
  20
  21 #include "config.h"
  22 #include "system.h"
  23 #include "coretypes.h"
  24 #include "tm.h"
  25 #include "ggc.h"
  26 #include "tree.h"
  27
  28 #include "target.h"
  29 #include "basic-block.h"
  30 #include "diagnostic.h"
  31 #include "tree-flow.h"
  32 #include "tree-dump.h"
  33 #include "timevar.h"
  34 #include "cfgloop.h"
  35 #include "expr.h"
  36 #include "optabs.h"
  37 #include "params.h"
  38 #include "tree-data-ref.h"
  39 #include "tree-vectorizer.h"
  40 #include "recog.h"
  41 #include "toplev.h"
  42
  43 /* Function prototypes */
  44 static void vect_pattern_recog_1
  45   (tree (* ) (tree, tree *, tree *), block_stmt_iterator);
  46 static bool widened_name_p (tree, tree, tree *, tree *);
  47
  48 /* Pattern recognition functions  */
  49 static tree vect_recog_widen_sum_pattern (tree, tree *, tree *);
  50 static tree vect_recog_widen_mult_pattern (tree, tree *, tree *);
  51 static tree vect_recog_dot_prod_pattern (tree, tree *, tree *);
  52 static tree vect_recog_pow_pattern (tree, tree *, tree *);
  53 static vect_recog_func_ptr vect_vect_recog_func_ptrs[NUM_PATTERNS] = {
  54         vect_recog_widen_mult_pattern,
  55         vect_recog_widen_sum_pattern,
  56         vect_recog_dot_prod_pattern,
  57         vect_recog_pow_pattern};
  58
  59
  60 /* Function widened_name_p
  61
  62    Check whether NAME, an ssa-name used in USE_STMT,
  63    is a result of a type-promotion, such that:
  64      DEF_STMT: NAME = NOP (name0)
  65    where the type of name0 (HALF_TYPE) is smaller than the type of NAME.
  66 */
  67
  68 static bool
  69 widened_name_p (tree name, tree use_stmt, tree *half_type, tree *def_stmt)
  70 {
  71   tree dummy;
  72   loop_vec_info loop_vinfo;
  73   stmt_vec_info stmt_vinfo;
  74   tree expr;
  75   tree type = TREE_TYPE (name);
  76   tree oprnd0;
  77   enum vect_def_type dt;
  78   tree def;
  79
  80   stmt_vinfo = vinfo_for_stmt (use_stmt);
  81   loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
  82
  83   if (!vect_is_simple_use (name, loop_vinfo, def_stmt, &def, &dt))
  84     return false;
  85
  86   if (dt != vect_loop_def
  87       && dt != vect_invariant_def && dt != vect_constant_def)
  88     return false;
  89
  90   if (! *def_stmt)
  91     return false;
  92
  93   if (TREE_CODE (*def_stmt) != GIMPLE_MODIFY_STMT)
  94     return false;
  95
  96   expr = GIMPLE_STMT_OPERAND (*def_stmt, 1);
  97   if (TREE_CODE (expr) != NOP_EXPR)
  98     return false;
  99
 100   oprnd0 = TREE_OPERAND (expr, 0);
 101
 102   *half_type = TREE_TYPE (oprnd0);
 103   if (!INTEGRAL_TYPE_P (type) || !INTEGRAL_TYPE_P (*half_type)
 104       || (TYPE_UNSIGNED (type) != TYPE_UNSIGNED (*half_type))
 105       || (TYPE_PRECISION (type) < (TYPE_PRECISION (*half_type) * 2)))
 106     return false;
 107
 108   if (!vect_is_simple_use (oprnd0, loop_vinfo, &dummy, &dummy, &dt))
 109     return false;
 110
 111   return true;
 112 }
 113
 114
 115 /* Function vect_recog_dot_prod_pattern
 116
 117    Try to find the following pattern:
 118
 119      type x_t, y_t;
 120      TYPE1 prod;
 121      TYPE2 sum = init;
 122    loop:
 123      sum_0 = phi <init, sum_1>
 124      S1  x_t = ...
 125      S2  y_t = ...
 126      S3  x_T = (TYPE1) x_t;
 127      S4  y_T = (TYPE1) y_t;
 128      S5  prod = x_T * y_T;
 129      [S6  prod = (TYPE2) prod;  #optional]
 130      S7  sum_1 = prod + sum_0;
 131
 132    where 'TYPE1' is exactly double the size of type 'type', and 'TYPE2' is the
 133    same size of 'TYPE1' or bigger. This is a special case of a reduction
 134    computation.
 135
 136    Input:
 137
 138    * LAST_STMT: A stmt from which the pattern search begins. In the example,
 139    when this function is called with S7, the pattern {S3,S4,S5,S6,S7} will be
 140    detected.
 141
 142    Output:
 143
 144    * TYPE_IN: The type of the input arguments to the pattern.
 145
 146    * TYPE_OUT: The type of the output  of this pattern.
 147
 148    * Return value: A new stmt that will be used to replace the sequence of
 149    stmts that constitute the pattern. In this case it will be:
 150         WIDEN_DOT_PRODUCT <x_t, y_t, sum_0>
 151
 152    Note: The dot-prod idiom is a widening reduction pattern that is
 153          vectorized without preserving all the intermediate results. It
 154          produces only N/2 (widened) results (by summing up pairs of
 155          intermediate results) rather than all N results.  Therefore, we
 156          cannot allow this pattern when we want to get all the results and in
 157          the correct order (as is the case when this computation is in an
 158          inner-loop nested in an outer-loop that us being vectorized).  */
 159
 160 static tree
 161 vect_recog_dot_prod_pattern (tree last_stmt, tree *type_in, tree *type_out)
 162 {
 163   tree stmt, expr;
 164   tree oprnd0, oprnd1;
 165   tree oprnd00, oprnd01;
 166   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
 167   tree type, half_type;
 168   tree pattern_expr;
 169   tree prod_type;
 170   loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
 171   struct loop *loop = LOOP_VINFO_LOOP (loop_info);
 172
 173   if (TREE_CODE (last_stmt) != GIMPLE_MODIFY_STMT)
 174     return NULL;
 175
 176   expr = GIMPLE_STMT_OPERAND (last_stmt, 1);
 177   type = TREE_TYPE (expr);
 178
 179   /* Look for the following pattern
 180           DX = (TYPE1) X;
 181           DY = (TYPE1) Y;
 182           DPROD = DX * DY;
 183           DDPROD = (TYPE2) DPROD;
 184           sum_1 = DDPROD + sum_0;
 185      In which
 186      - DX is double the size of X
 187      - DY is double the size of Y
 188      - DX, DY, DPROD all have the same type
 189      - sum is the same size of DPROD or bigger
 190      - sum has been recognized as a reduction variable.
 191
 192      This is equivalent to:
 193        DPROD = X w* Y;          #widen mult
 194        sum_1 = DPROD w+ sum_0;  #widen summation
 195      or
 196        DPROD = X w* Y;          #widen mult
 197        sum_1 = DPROD + sum_0;   #summation
 198    */
 199
 200   /* Starting from LAST_STMT, follow the defs of its uses in search
 201      of the above pattern.  */
 202
 203   if (TREE_CODE (expr) != PLUS_EXPR)
 204     return NULL;
 205
 206   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
 207     {
 208       /* Has been detected as widening-summation?  */
 209
 210       stmt = STMT_VINFO_RELATED_STMT (stmt_vinfo);
 211       expr = GIMPLE_STMT_OPERAND (stmt, 1);
 212       type = TREE_TYPE (expr);
 213       if (TREE_CODE (expr) != WIDEN_SUM_EXPR)
 214         return NULL;
 215       oprnd0 = TREE_OPERAND (expr, 0);
 216       oprnd1 = TREE_OPERAND (expr, 1);
 217       half_type = TREE_TYPE (oprnd0);
 218     }
 219   else
 220     {
 221       tree def_stmt;
 222
 223       if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_reduction_def)
 224         return NULL;
 225       oprnd0 = TREE_OPERAND (expr, 0);
 226       oprnd1 = TREE_OPERAND (expr, 1);
 227       if (TYPE_MAIN_VARIANT (TREE_TYPE (oprnd0)) != TYPE_MAIN_VARIANT (type)
 228           || TYPE_MAIN_VARIANT (TREE_TYPE (oprnd1)) != TYPE_MAIN_VARIANT (type))
 229         return NULL;
 230       stmt = last_stmt;
 231
 232       if (widened_name_p (oprnd0, stmt, &half_type, &def_stmt))
 233         {
 234           stmt = def_stmt;
 235           expr = GIMPLE_STMT_OPERAND (stmt, 1);
 236           oprnd0 = TREE_OPERAND (expr, 0);
 237         }
 238       else
 239         half_type = type;
 240     }
 241
 242   /* So far so good. Since last_stmt was detected as a (summation) reduction,
 243      we know that oprnd1 is the reduction variable (defined by a loop-header
 244      phi), and oprnd0 is an ssa-name defined by a stmt in the loop body.
 245      Left to check that oprnd0 is defined by a (widen_)mult_expr  */
 246
 247   prod_type = half_type;
 248   stmt = SSA_NAME_DEF_STMT (oprnd0);
 249   gcc_assert (stmt);
 250   stmt_vinfo = vinfo_for_stmt (stmt);
 251   gcc_assert (stmt_vinfo);
 252   if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_loop_def)
 253     return NULL;
 254   /* FORNOW. Can continue analyzing the def-use chain when this stmt in a phi
 255      inside the loop (in case we are analyzing an outer-loop).  */
 256   if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
 257     return NULL;
 258   expr = GIMPLE_STMT_OPERAND (stmt, 1);
 259   if (TREE_CODE (expr) != MULT_EXPR)
 260     return NULL;
 261   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
 262     {
 263       /* Has been detected as a widening multiplication?  */
 264
 265       stmt = STMT_VINFO_RELATED_STMT (stmt_vinfo);
 266       expr = GIMPLE_STMT_OPERAND (stmt, 1);
 267       if (TREE_CODE (expr) != WIDEN_MULT_EXPR)
 268         return NULL;
 269       stmt_vinfo = vinfo_for_stmt (stmt);
 270       gcc_assert (stmt_vinfo);
 271       gcc_assert (STMT_VINFO_DEF_TYPE (stmt_vinfo) == vect_loop_def);
 272       oprnd00 = TREE_OPERAND (expr, 0);
 273       oprnd01 = TREE_OPERAND (expr, 1);
 274     }
 275   else
 276     {
 277       tree half_type0, half_type1;
 278       tree def_stmt;
 279       tree oprnd0, oprnd1;
 280
 281       oprnd0 = TREE_OPERAND (expr, 0);
 282       oprnd1 = TREE_OPERAND (expr, 1);
 283       if (TYPE_MAIN_VARIANT (TREE_TYPE (oprnd0))
 284                                 != TYPE_MAIN_VARIANT (prod_type)
 285           || TYPE_MAIN_VARIANT (TREE_TYPE (oprnd1))
 286                                 != TYPE_MAIN_VARIANT (prod_type))
 287         return NULL;
 288       if (!widened_name_p (oprnd0, stmt, &half_type0, &def_stmt))
 289         return NULL;
 290       oprnd00 = TREE_OPERAND (GIMPLE_STMT_OPERAND (def_stmt, 1), 0);
 291       if (!widened_name_p (oprnd1, stmt, &half_type1, &def_stmt))
 292         return NULL;
 293       oprnd01 = TREE_OPERAND (GIMPLE_STMT_OPERAND (def_stmt, 1), 0);
 294       if (TYPE_MAIN_VARIANT (half_type0) != TYPE_MAIN_VARIANT (half_type1))
 295         return NULL;
 296       if (TYPE_PRECISION (prod_type) != TYPE_PRECISION (half_type0) * 2)
 297         return NULL;
 298     }
 299
 300   half_type = TREE_TYPE (oprnd00);
 301   *type_in = half_type;
 302   *type_out = type;
 303
 304   /* Pattern detected. Create a stmt to be used to replace the pattern: */
 305   pattern_expr = build3 (DOT_PROD_EXPR, type, oprnd00, oprnd01, oprnd1);
 306   if (vect_print_dump_info (REPORT_DETAILS))
 307     {
 308       fprintf (vect_dump, "vect_recog_dot_prod_pattern: detected: ");
 309       print_generic_expr (vect_dump, pattern_expr, TDF_SLIM);
 310     }
 311
 312   /* We don't allow changing the order of the computation in the inner-loop
 313      when doing outer-loop vectorization.  */
 314   if (nested_in_vect_loop_p (loop, last_stmt))
 315     {
 316       if (vect_print_dump_info (REPORT_DETAILS))
 317         fprintf (vect_dump, "vect_recog_dot_prod_pattern: not allowed.");
 318       return NULL;
 319     }
 320
 321   return pattern_expr;
 322 }
 323
 324
 325 /* Function vect_recog_widen_mult_pattern
 326
 327    Try to find the following pattern:
 328
 329      type a_t, b_t;
 330      TYPE a_T, b_T, prod_T;
 331
 332      S1  a_t = ;
 333      S2  b_t = ;
 334      S3  a_T = (TYPE) a_t;
 335      S4  b_T = (TYPE) b_t;
 336      S5  prod_T = a_T * b_T;
 337
 338    where type 'TYPE' is at least double the size of type 'type'.
 339
 340    Input:
 341
 342    * LAST_STMT: A stmt from which the pattern search begins. In the example,
 343    when this function is called with S5, the pattern {S3,S4,S5} is be detected.
 344
 345    Output:
 346
 347    * TYPE_IN: The type of the input arguments to the pattern.
 348
 349    * TYPE_OUT: The type of the output  of this pattern.
 350
 351    * Return value: A new stmt that will be used to replace the sequence of
 352    stmts that constitute the pattern. In this case it will be:
 353         WIDEN_MULT <a_t, b_t>
 354 */
 355
 356 static tree
 357 vect_recog_widen_mult_pattern (tree last_stmt,
 358                                tree *type_in,
 359                                tree *type_out)
 360 {
 361   tree expr;
 362   tree def_stmt0, def_stmt1;
 363   tree oprnd0, oprnd1;
 364   tree type, half_type0, half_type1;
 365   tree pattern_expr;
 366   tree vectype;
 367   tree dummy;
 368   enum tree_code dummy_code;
 369
 370   if (TREE_CODE (last_stmt) != GIMPLE_MODIFY_STMT)
 371     return NULL;
 372
 373   expr = GIMPLE_STMT_OPERAND (last_stmt, 1);
 374   type = TREE_TYPE (expr);
 375
 376   /* Starting from LAST_STMT, follow the defs of its uses in search
 377      of the above pattern.  */
 378
 379   if (TREE_CODE (expr) != MULT_EXPR)
 380     return NULL;
 381
 382   oprnd0 = TREE_OPERAND (expr, 0);
 383   oprnd1 = TREE_OPERAND (expr, 1);
 384   if (TYPE_MAIN_VARIANT (TREE_TYPE (oprnd0)) != TYPE_MAIN_VARIANT (type)
 385       || TYPE_MAIN_VARIANT (TREE_TYPE (oprnd1)) != TYPE_MAIN_VARIANT (type))
 386     return NULL;
 387
 388   /* Check argument 0 */
 389   if (!widened_name_p (oprnd0, last_stmt, &half_type0, &def_stmt0))
 390     return NULL;
 391   oprnd0 = TREE_OPERAND (GIMPLE_STMT_OPERAND (def_stmt0, 1), 0);
 392
 393   /* Check argument 1 */
 394   if (!widened_name_p (oprnd1, last_stmt, &half_type1, &def_stmt1))
 395     return NULL;
 396   oprnd1 = TREE_OPERAND (GIMPLE_STMT_OPERAND (def_stmt1, 1), 0);
 397
 398   if (TYPE_MAIN_VARIANT (half_type0) != TYPE_MAIN_VARIANT (half_type1))
 399     return NULL;
 400
 401   /* Pattern detected.  */
 402   if (vect_print_dump_info (REPORT_DETAILS))
 403     fprintf (vect_dump, "vect_recog_widen_mult_pattern: detected: ");
 404
 405   /* Check target support  */
 406   vectype = get_vectype_for_scalar_type (half_type0);
 407   if (!vectype
 408       || !supportable_widening_operation (WIDEN_MULT_EXPR, last_stmt, vectype,
 409                                        &dummy, &dummy, &dummy_code,
 410                                        &dummy_code))
 411     return NULL;
 412
 413   *type_in = vectype;
 414   *type_out = NULL_TREE;
 415
 416   /* Pattern supported. Create a stmt to be used to replace the pattern: */
 417   pattern_expr = build2 (WIDEN_MULT_EXPR, type, oprnd0, oprnd1);
 418   if (vect_print_dump_info (REPORT_DETAILS))
 419     print_generic_expr (vect_dump, pattern_expr, TDF_SLIM);
 420   return pattern_expr;
 421 }
 422
 423
 424 /* Function vect_recog_pow_pattern
 425
 426    Try to find the following pattern:
 427
 428      x = POW (y, N);
 429
 430    with POW being one of pow, powf, powi, powif and N being
 431    either 2 or 0.5.
 432
 433    Input:
 434
 435    * LAST_STMT: A stmt from which the pattern search begins.
 436
 437    Output:
 438
 439    * TYPE_IN: The type of the input arguments to the pattern.
 440
 441    * TYPE_OUT: The type of the output of this pattern.
 442
 443    * Return value: A new stmt that will be used to replace the sequence of
 444    stmts that constitute the pattern. In this case it will be:
 445         x * x
 446    or
 447         sqrt (x)
 448 */
 449
 450 static tree
 451 vect_recog_pow_pattern (tree last_stmt, tree *type_in, tree *type_out)
 452 {
 453   tree expr;
 454   tree type;
 455   tree fn, base, exp;
 456
 457   if (TREE_CODE (last_stmt) != GIMPLE_MODIFY_STMT)
 458     return NULL;
 459
 460   expr = GIMPLE_STMT_OPERAND (last_stmt, 1);
 461   type = TREE_TYPE (expr);
 462
 463   if (TREE_CODE (expr) != CALL_EXPR)
 464     return NULL_TREE;
 465
 466   fn = get_callee_fndecl (expr);
 467   switch (DECL_FUNCTION_CODE (fn))
 468     {
 469     case BUILT_IN_POWIF:
 470     case BUILT_IN_POWI:
 471     case BUILT_IN_POWF:
 472     case BUILT_IN_POW:
 473       base = CALL_EXPR_ARG (expr, 0);
 474       exp = CALL_EXPR_ARG (expr, 1);
 475       if (TREE_CODE (exp) != REAL_CST
 476           && TREE_CODE (exp) != INTEGER_CST)
 477         return NULL_TREE;
 478       break;
 479
 480     default:;
 481       return NULL_TREE;
 482     }
 483
 484   /* We now have a pow or powi builtin function call with a constant
 485      exponent.  */
 486
 487   *type_out = NULL_TREE;
 488
 489   /* Catch squaring.  */
 490   if ((host_integerp (exp, 0)
 491        && tree_low_cst (exp, 0) == 2)
 492       || (TREE_CODE (exp) == REAL_CST
 493           && REAL_VALUES_EQUAL (TREE_REAL_CST (exp), dconst2)))
 494     {
 495       *type_in = TREE_TYPE (base);
 496       return build2 (MULT_EXPR, TREE_TYPE (base), base, base);
 497     }
 498
 499   /* Catch square root.  */
 500   if (TREE_CODE (exp) == REAL_CST
 501       && REAL_VALUES_EQUAL (TREE_REAL_CST (exp), dconsthalf))
 502     {
 503       tree newfn = mathfn_built_in (TREE_TYPE (base), BUILT_IN_SQRT);
 504       *type_in = get_vectype_for_scalar_type (TREE_TYPE (base));
 505       if (*type_in)
 506         {
 507           newfn = build_call_expr (newfn, 1, base);
 508           if (vectorizable_function (newfn, *type_in, *type_in) != NULL_TREE)
 509             return newfn;
 510         }
 511     }
 512
 513   return NULL_TREE;
 514 }
 515
 516
 517 /* Function vect_recog_widen_sum_pattern
 518
 519    Try to find the following pattern:
 520
 521      type x_t;
 522      TYPE x_T, sum = init;
 523    loop:
 524      sum_0 = phi <init, sum_1>
 525      S1  x_t = *p;
 526      S2  x_T = (TYPE) x_t;
 527      S3  sum_1 = x_T + sum_0;
 528
 529    where type 'TYPE' is at least double the size of type 'type', i.e - we're
 530    summing elements of type 'type' into an accumulator of type 'TYPE'. This is
 531    a special case of a reduction computation.
 532
 533    Input:
 534
 535    * LAST_STMT: A stmt from which the pattern search begins. In the example,
 536    when this function is called with S3, the pattern {S2,S3} will be detected.
 537
 538    Output:
 539
 540    * TYPE_IN: The type of the input arguments to the pattern.
 541
 542    * TYPE_OUT: The type of the output of this pattern.
 543
 544    * Return value: A new stmt that will be used to replace the sequence of
 545    stmts that constitute the pattern. In this case it will be:
 546         WIDEN_SUM <x_t, sum_0>
 547
 548    Note: The widening-sum idiom is a widening reduction pattern that is
 549          vectorized without preserving all the intermediate results. It
 550          produces only N/2 (widened) results (by summing up pairs of
 551          intermediate results) rather than all N results.  Therefore, we
 552          cannot allow this pattern when we want to get all the results and in
 553          the correct order (as is the case when this computation is in an
 554          inner-loop nested in an outer-loop that us being vectorized).  */
 555
 556 static tree
 557 vect_recog_widen_sum_pattern (tree last_stmt, tree *type_in, tree *type_out)
 558 {
 559   tree stmt, expr;
 560   tree oprnd0, oprnd1;
 561   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
 562   tree type, half_type;
 563   tree pattern_expr;
 564   loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
 565   struct loop *loop = LOOP_VINFO_LOOP (loop_info);
 566
 567   if (TREE_CODE (last_stmt) != GIMPLE_MODIFY_STMT)
 568     return NULL;
 569
 570   expr = GIMPLE_STMT_OPERAND (last_stmt, 1);
 571   type = TREE_TYPE (expr);
 572
 573   /* Look for the following pattern
 574           DX = (TYPE) X;
 575           sum_1 = DX + sum_0;
 576      In which DX is at least double the size of X, and sum_1 has been
 577      recognized as a reduction variable.
 578    */
 579
 580   /* Starting from LAST_STMT, follow the defs of its uses in search
 581      of the above pattern.  */
 582
 583   if (TREE_CODE (expr) != PLUS_EXPR)
 584     return NULL;
 585
 586   if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_reduction_def)
 587     return NULL;
 588
 589   oprnd0 = TREE_OPERAND (expr, 0);
 590   oprnd1 = TREE_OPERAND (expr, 1);
 591   if (TYPE_MAIN_VARIANT (TREE_TYPE (oprnd0)) != TYPE_MAIN_VARIANT (type)
 592       || TYPE_MAIN_VARIANT (TREE_TYPE (oprnd1)) != TYPE_MAIN_VARIANT (type))
 593     return NULL;
 594
 595   /* So far so good. Since last_stmt was detected as a (summation) reduction,
 596      we know that oprnd1 is the reduction variable (defined by a loop-header
 597      phi), and oprnd0 is an ssa-name defined by a stmt in the loop body.
 598      Left to check that oprnd0 is defined by a cast from type 'type' to type
 599      'TYPE'.  */
 600
 601   if (!widened_name_p (oprnd0, last_stmt, &half_type, &stmt))
 602     return NULL;
 603
 604   oprnd0 = TREE_OPERAND (GIMPLE_STMT_OPERAND (stmt, 1), 0);
 605   *type_in = half_type;
 606   *type_out = type;
 607
 608   /* Pattern detected. Create a stmt to be used to replace the pattern: */
 609   pattern_expr = build2 (WIDEN_SUM_EXPR, type, oprnd0, oprnd1);
 610   if (vect_print_dump_info (REPORT_DETAILS))
 611     {
 612       fprintf (vect_dump, "vect_recog_widen_sum_pattern: detected: ");
 613       print_generic_expr (vect_dump, pattern_expr, TDF_SLIM);
 614     }
 615
 616   /* We don't allow changing the order of the computation in the inner-loop
 617      when doing outer-loop vectorization.  */
 618   if (nested_in_vect_loop_p (loop, last_stmt))
 619     {
 620       if (vect_print_dump_info (REPORT_DETAILS))
 621         fprintf (vect_dump, "vect_recog_widen_sum_pattern: not allowed.");
 622       return NULL;
 623     }
 624
 625   return pattern_expr;
 626 }
 627
 628
 629 /* Function vect_pattern_recog_1
 630
 631    Input:
 632    PATTERN_RECOG_FUNC: A pointer to a function that detects a certain
 633         computation pattern.
 634    STMT: A stmt from which the pattern search should start.
 635
 636    If PATTERN_RECOG_FUNC successfully detected the pattern, it creates an
 637    expression that computes the same functionality and can be used to
 638    replace the sequence of stmts that are involved in the pattern.
 639
 640    Output:
 641    This function checks if the expression returned by PATTERN_RECOG_FUNC is
 642    supported in vector form by the target.  We use 'TYPE_IN' to obtain the
 643    relevant vector type. If 'TYPE_IN' is already a vector type, then this
 644    indicates that target support had already been checked by PATTERN_RECOG_FUNC.
 645    If 'TYPE_OUT' is also returned by PATTERN_RECOG_FUNC, we check that it fits
 646    to the available target pattern.
 647
 648    This function also does some bookkeeping, as explained in the documentation
 649    for vect_recog_pattern.  */
 650
 651 static void
 652 vect_pattern_recog_1 (
 653         tree (* vect_recog_func) (tree, tree *, tree *),
 654         block_stmt_iterator si)
 655 {
 656   tree stmt = bsi_stmt (si);
 657   stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
 658   stmt_vec_info pattern_stmt_info;
 659   loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
 660   tree pattern_expr;
 661   tree pattern_vectype;
 662   tree type_in, type_out;
 663   tree pattern_type;
 664   enum tree_code code;
 665   tree var, var_name;
 666   stmt_ann_t ann;
 667
 668   pattern_expr = (* vect_recog_func) (stmt, &type_in, &type_out);
 669   if (!pattern_expr)
 670     return;
 671
 672   if (VECTOR_MODE_P (TYPE_MODE (type_in)))
 673     {
 674       /* No need to check target support (already checked by the pattern
 675          recognition function).  */
 676       pattern_vectype = type_in;
 677     }
 678   else
 679     {
 680       enum tree_code vec_mode;
 681       enum insn_code icode;
 682       optab optab;
 683
 684       /* Check target support  */
 685       pattern_vectype = get_vectype_for_scalar_type (type_in);
 686       if (!pattern_vectype)
 687         return;
 688
 689       optab = optab_for_tree_code (TREE_CODE (pattern_expr), pattern_vectype);
 690       vec_mode = TYPE_MODE (pattern_vectype);
 691       if (!optab
 692           || (icode = optab_handler (optab, vec_mode)->insn_code) ==
 693               CODE_FOR_nothing
 694           || (type_out
 695               && (!get_vectype_for_scalar_type (type_out)
 696                   || (insn_data[icode].operand[0].mode !=
 697                       TYPE_MODE (get_vectype_for_scalar_type (type_out))))))
 698         return;
 699     }
 700
 701   /* Found a vectorizable pattern.  */
 702   if (vect_print_dump_info (REPORT_DETAILS))
 703     {
 704       fprintf (vect_dump, "pattern recognized: ");
 705       print_generic_expr (vect_dump, pattern_expr, TDF_SLIM);
 706     }
 707
 708   /* Mark the stmts that are involved in the pattern,
 709      create a new stmt to express the pattern and insert it.  */
 710   code = TREE_CODE (pattern_expr);
 711   pattern_type = TREE_TYPE (pattern_expr);
 712   var = create_tmp_var (pattern_type, "patt");
 713   add_referenced_var (var);
 714   var_name = make_ssa_name (var, NULL_TREE);
 715   pattern_expr = build_gimple_modify_stmt (var_name, pattern_expr);
 716   SSA_NAME_DEF_STMT (var_name) = pattern_expr;
 717   bsi_insert_before (&si, pattern_expr, BSI_SAME_STMT);
 718   ann = stmt_ann (pattern_expr);
 719   set_stmt_info (ann, new_stmt_vec_info (pattern_expr, loop_vinfo));
 720   pattern_stmt_info = vinfo_for_stmt (pattern_expr);
 721
 722   STMT_VINFO_RELATED_STMT (pattern_stmt_info) = stmt;
 723   STMT_VINFO_DEF_TYPE (pattern_stmt_info) = STMT_VINFO_DEF_TYPE (stmt_info);
 724   STMT_VINFO_VECTYPE (pattern_stmt_info) = pattern_vectype;
 725   STMT_VINFO_IN_PATTERN_P (stmt_info) = true;
 726   STMT_VINFO_RELATED_STMT (stmt_info) = pattern_expr;
 727
 728   return;
 729 }
 730
 731
 732 /* Function vect_pattern_recog
 733
 734    Input:
 735    LOOP_VINFO - a struct_loop_info of a loop in which we want to look for
 736         computation idioms.
 737
 738    Output - for each computation idiom that is detected we insert a new stmt
 739         that provides the same functionality and that can be vectorized. We
 740         also record some information in the struct_stmt_info of the relevant
 741         stmts, as explained below:
 742
 743    At the entry to this function we have the following stmts, with the
 744    following initial value in the STMT_VINFO fields:
 745
 746          stmt                     in_pattern_p  related_stmt    vec_stmt
 747          S1: a_i = ....                 -       -               -
 748          S2: a_2 = ..use(a_i)..         -       -               -
 749          S3: a_1 = ..use(a_2)..         -       -               -
 750          S4: a_0 = ..use(a_1)..         -       -               -
 751          S5: ... = ..use(a_0)..         -       -               -
 752
 753    Say the sequence {S1,S2,S3,S4} was detected as a pattern that can be
 754    represented by a single stmt. We then:
 755    - create a new stmt S6 that will replace the pattern.
 756    - insert the new stmt S6 before the last stmt in the pattern
 757    - fill in the STMT_VINFO fields as follows:
 758
 759                                   in_pattern_p  related_stmt    vec_stmt
 760          S1: a_i = ....                 -       -               -
 761          S2: a_2 = ..use(a_i)..         -       -               -
 762          S3: a_1 = ..use(a_2)..         -       -               -
 763        > S6: a_new = ....               -       S4              -
 764          S4: a_0 = ..use(a_1)..         true    S6              -
 765          S5: ... = ..use(a_0)..         -       -               -
 766
 767    (the last stmt in the pattern (S4) and the new pattern stmt (S6) point
 768     to each other through the RELATED_STMT field).
 769
 770    S6 will be marked as relevant in vect_mark_stmts_to_be_vectorized instead
 771    of S4 because it will replace all its uses.  Stmts {S1,S2,S3} will
 772    remain irrelevant unless used by stmts other than S4.
 773
 774    If vectorization succeeds, vect_transform_stmt will skip over {S1,S2,S3}
 775    (because they are marked as irrelevant). It will vectorize S6, and record
 776    a pointer to the new vector stmt VS6 both from S6 (as usual), and also
 777    from S4. We do that so that when we get to vectorizing stmts that use the
 778    def of S4 (like S5 that uses a_0), we'll know where to take the relevant
 779    vector-def from. S4 will be skipped, and S5 will be vectorized as usual:
 780
 781                                   in_pattern_p  related_stmt    vec_stmt
 782          S1: a_i = ....                 -       -               -
 783          S2: a_2 = ..use(a_i)..         -       -               -
 784          S3: a_1 = ..use(a_2)..         -       -               -
 785        > VS6: va_new = ....             -       -               -
 786          S6: a_new = ....               -       S4              VS6
 787          S4: a_0 = ..use(a_1)..         true    S6              VS6
 788        > VS5: ... = ..vuse(va_new)..    -       -               -
 789          S5: ... = ..use(a_0)..         -       -               -
 790
 791    DCE could then get rid of {S1,S2,S3,S4,S5,S6} (if their defs are not used
 792    elsewhere), and we'll end up with:
 793
 794         VS6: va_new = ....
 795         VS5: ... = ..vuse(va_new)..
 796
 797    If vectorization does not succeed, DCE will clean S6 away (its def is
 798    not used), and we'll end up with the original sequence.
 799 */
 800
 801 void
 802 vect_pattern_recog (loop_vec_info loop_vinfo)
 803 {
 804   struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
 805   basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo);
 806   unsigned int nbbs = loop->num_nodes;
 807   block_stmt_iterator si;
 808   tree stmt;
 809   unsigned int i, j;
 810   tree (* vect_recog_func_ptr) (tree, tree *, tree *);
 811
 812   if (vect_print_dump_info (REPORT_DETAILS))
 813     fprintf (vect_dump, "=== vect_pattern_recog ===");
 814
 815   /* Scan through the loop stmts, applying the pattern recognition
 816      functions starting at each stmt visited:  */
 817   for (i = 0; i < nbbs; i++)
 818     {
 819       basic_block bb = bbs[i];
 820       for (si = bsi_start (bb); !bsi_end_p (si); bsi_next (&si))
 821         {
 822           stmt = bsi_stmt (si);
 823
 824           /* Scan over all generic vect_recog_xxx_pattern functions.  */
 825           for (j = 0; j < NUM_PATTERNS; j++)
 826             {
 827               vect_recog_func_ptr = vect_vect_recog_func_ptrs[j];
 828               vect_pattern_recog_1 (vect_recog_func_ptr, si);
 829             }
 830         }
 831     }
 832 }