gcc/tree-vect-patterns.c

   1 /* Analysis Utilities for Loop Vectorization.
   2    Copyright (C) 2006, 2007, 2008, 2009 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 #include "target.h"
  28 #include "basic-block.h"
  29 #include "diagnostic.h"
  30 #include "tree-flow.h"
  31 #include "tree-dump.h"
  32 #include "cfgloop.h"
  33 #include "expr.h"
  34 #include "optabs.h"
  35 #include "params.h"
  36 #include "tree-data-ref.h"
  37 #include "tree-vectorizer.h"
  38 #include "recog.h"
  39 #include "toplev.h"
  40
  41 /* Function prototypes */
  42 static void vect_pattern_recog_1
  43   (gimple (* ) (gimple, tree *, tree *), gimple_stmt_iterator);
  44 static bool widened_name_p (tree, gimple, tree *, gimple *);
  45
  46 /* Pattern recognition functions  */
  47 static gimple vect_recog_widen_sum_pattern (gimple, tree *, tree *);
  48 static gimple vect_recog_widen_mult_pattern (gimple, tree *, tree *);
  49 static gimple vect_recog_dot_prod_pattern (gimple, tree *, tree *);
  50 static gimple vect_recog_pow_pattern (gimple, tree *, tree *);
  51 static vect_recog_func_ptr vect_vect_recog_func_ptrs[NUM_PATTERNS] = {
  52         vect_recog_widen_mult_pattern,
  53         vect_recog_widen_sum_pattern,
  54         vect_recog_dot_prod_pattern,
  55         vect_recog_pow_pattern};
  56
  57
  58 /* Function widened_name_p
  59
  60    Check whether NAME, an ssa-name used in USE_STMT,
  61    is a result of a type-promotion, such that:
  62      DEF_STMT: NAME = NOP (name0)
  63    where the type of name0 (HALF_TYPE) is smaller than the type of NAME.
  64 */
  65
  66 static bool
  67 widened_name_p (tree name, gimple use_stmt, tree *half_type, gimple *def_stmt)
  68 {
  69   tree dummy;
  70   gimple dummy_gimple;
  71   loop_vec_info loop_vinfo;
  72   stmt_vec_info stmt_vinfo;
  73   tree type = TREE_TYPE (name);
  74   tree oprnd0;
  75   enum vect_def_type dt;
  76   tree def;
  77
  78   stmt_vinfo = vinfo_for_stmt (use_stmt);
  79   loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
  80
  81   if (!vect_is_simple_use (name, loop_vinfo, NULL, def_stmt, &def, &dt))
  82     return false;
  83
  84   if (dt != vect_internal_def
  85       && dt != vect_external_def && dt != vect_constant_def)
  86     return false;
  87
  88   if (! *def_stmt)
  89     return false;
  90
  91   if (!is_gimple_assign (*def_stmt))
  92     return false;
  93
  94   if (gimple_assign_rhs_code (*def_stmt) != NOP_EXPR)
  95     return false;
  96
  97   oprnd0 = gimple_assign_rhs1 (*def_stmt);
  98
  99   *half_type = TREE_TYPE (oprnd0);
 100   if (!INTEGRAL_TYPE_P (type) || !INTEGRAL_TYPE_P (*half_type)
 101       || (TYPE_UNSIGNED (type) != TYPE_UNSIGNED (*half_type))
 102       || (TYPE_PRECISION (type) < (TYPE_PRECISION (*half_type) * 2)))
 103     return false;
 104
 105   if (!vect_is_simple_use (oprnd0, loop_vinfo, NULL, &dummy_gimple, &dummy,
 106                            &dt))
 107     return false;
 108
 109   return true;
 110 }
 111
 112 /* Helper to return a new temporary for pattern of TYPE for STMT.  If STMT
 113    is NULL, the caller must set SSA_NAME_DEF_STMT for the returned SSA var. */
 114
 115 static tree
 116 vect_recog_temp_ssa_var (tree type, gimple stmt)
 117 {
 118   tree var = create_tmp_var (type, "patt");
 119
 120   add_referenced_var (var);
 121   var = make_ssa_name (var, stmt);
 122   return var;
 123 }
 124
 125 /* Function vect_recog_dot_prod_pattern
 126
 127    Try to find the following pattern:
 128
 129      type x_t, y_t;
 130      TYPE1 prod;
 131      TYPE2 sum = init;
 132    loop:
 133      sum_0 = phi <init, sum_1>
 134      S1  x_t = ...
 135      S2  y_t = ...
 136      S3  x_T = (TYPE1) x_t;
 137      S4  y_T = (TYPE1) y_t;
 138      S5  prod = x_T * y_T;
 139      [S6  prod = (TYPE2) prod;  #optional]
 140      S7  sum_1 = prod + sum_0;
 141
 142    where 'TYPE1' is exactly double the size of type 'type', and 'TYPE2' is the
 143    same size of 'TYPE1' or bigger. This is a special case of a reduction
 144    computation.
 145
 146    Input:
 147
 148    * LAST_STMT: A stmt from which the pattern search begins. In the example,
 149    when this function is called with S7, the pattern {S3,S4,S5,S6,S7} will be
 150    detected.
 151
 152    Output:
 153
 154    * TYPE_IN: The type of the input arguments to the pattern.
 155
 156    * TYPE_OUT: The type of the output  of this pattern.
 157
 158    * Return value: A new stmt that will be used to replace the sequence of
 159    stmts that constitute the pattern. In this case it will be:
 160         WIDEN_DOT_PRODUCT <x_t, y_t, sum_0>
 161
 162    Note: The dot-prod idiom is a widening reduction pattern that is
 163          vectorized without preserving all the intermediate results. It
 164          produces only N/2 (widened) results (by summing up pairs of
 165          intermediate results) rather than all N results.  Therefore, we
 166          cannot allow this pattern when we want to get all the results and in
 167          the correct order (as is the case when this computation is in an
 168          inner-loop nested in an outer-loop that us being vectorized).  */
 169
 170 static gimple
 171 vect_recog_dot_prod_pattern (gimple last_stmt, tree *type_in, tree *type_out)
 172 {
 173   gimple stmt;
 174   tree oprnd0, oprnd1;
 175   tree oprnd00, oprnd01;
 176   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
 177   tree type, half_type;
 178   gimple pattern_stmt;
 179   tree prod_type;
 180   loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
 181   struct loop *loop = LOOP_VINFO_LOOP (loop_info);
 182   tree var, rhs;
 183
 184   if (!is_gimple_assign (last_stmt))
 185     return NULL;
 186
 187   type = gimple_expr_type (last_stmt);
 188
 189   /* Look for the following pattern
 190           DX = (TYPE1) X;
 191           DY = (TYPE1) Y;
 192           DPROD = DX * DY;
 193           DDPROD = (TYPE2) DPROD;
 194           sum_1 = DDPROD + sum_0;
 195      In which
 196      - DX is double the size of X
 197      - DY is double the size of Y
 198      - DX, DY, DPROD all have the same type
 199      - sum is the same size of DPROD or bigger
 200      - sum has been recognized as a reduction variable.
 201
 202      This is equivalent to:
 203        DPROD = X w* Y;          #widen mult
 204        sum_1 = DPROD w+ sum_0;  #widen summation
 205      or
 206        DPROD = X w* Y;          #widen mult
 207        sum_1 = DPROD + sum_0;   #summation
 208    */
 209
 210   /* Starting from LAST_STMT, follow the defs of its uses in search
 211      of the above pattern.  */
 212
 213   if (gimple_assign_rhs_code (last_stmt) != PLUS_EXPR)
 214     return NULL;
 215
 216   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
 217     {
 218       /* Has been detected as widening-summation?  */
 219
 220       stmt = STMT_VINFO_RELATED_STMT (stmt_vinfo);
 221       type = gimple_expr_type (stmt);
 222       if (gimple_assign_rhs_code (stmt) != WIDEN_SUM_EXPR)
 223         return NULL;
 224       oprnd0 = gimple_assign_rhs1 (stmt);
 225       oprnd1 = gimple_assign_rhs2 (stmt);
 226       half_type = TREE_TYPE (oprnd0);
 227     }
 228   else
 229     {
 230       gimple def_stmt;
 231
 232       if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_reduction_def)
 233         return NULL;
 234       oprnd0 = gimple_assign_rhs1 (last_stmt);
 235       oprnd1 = gimple_assign_rhs2 (last_stmt);
 236       if (TYPE_MAIN_VARIANT (TREE_TYPE (oprnd0)) != TYPE_MAIN_VARIANT (type)
 237           || TYPE_MAIN_VARIANT (TREE_TYPE (oprnd1)) != TYPE_MAIN_VARIANT (type))
 238         return NULL;
 239       stmt = last_stmt;
 240
 241       if (widened_name_p (oprnd0, stmt, &half_type, &def_stmt))
 242         {
 243           stmt = def_stmt;
 244           oprnd0 = gimple_assign_rhs1 (stmt);
 245         }
 246       else
 247         half_type = type;
 248     }
 249
 250   /* So far so good. Since last_stmt was detected as a (summation) reduction,
 251      we know that oprnd1 is the reduction variable (defined by a loop-header
 252      phi), and oprnd0 is an ssa-name defined by a stmt in the loop body.
 253      Left to check that oprnd0 is defined by a (widen_)mult_expr  */
 254
 255   prod_type = half_type;
 256   stmt = SSA_NAME_DEF_STMT (oprnd0);
 257   /* FORNOW.  Can continue analyzing the def-use chain when this stmt in a phi
 258      inside the loop (in case we are analyzing an outer-loop).  */
 259   if (!is_gimple_assign (stmt))
 260     return NULL;
 261   stmt_vinfo = vinfo_for_stmt (stmt);
 262   gcc_assert (stmt_vinfo);
 263   if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_internal_def)
 264     return NULL;
 265   if (gimple_assign_rhs_code (stmt) != MULT_EXPR)
 266     return NULL;
 267   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
 268     {
 269       /* Has been detected as a widening multiplication?  */
 270
 271       stmt = STMT_VINFO_RELATED_STMT (stmt_vinfo);
 272       if (gimple_assign_rhs_code (stmt) != WIDEN_MULT_EXPR)
 273         return NULL;
 274       stmt_vinfo = vinfo_for_stmt (stmt);
 275       gcc_assert (stmt_vinfo);
 276       gcc_assert (STMT_VINFO_DEF_TYPE (stmt_vinfo) == vect_internal_def);
 277       oprnd00 = gimple_assign_rhs1 (stmt);
 278       oprnd01 = gimple_assign_rhs2 (stmt);
 279     }
 280   else
 281     {
 282       tree half_type0, half_type1;
 283       gimple def_stmt;
 284       tree oprnd0, oprnd1;
 285
 286       oprnd0 = gimple_assign_rhs1 (stmt);
 287       oprnd1 = gimple_assign_rhs2 (stmt);
 288       if (TYPE_MAIN_VARIANT (TREE_TYPE (oprnd0))
 289                                 != TYPE_MAIN_VARIANT (prod_type)
 290           || TYPE_MAIN_VARIANT (TREE_TYPE (oprnd1))
 291                                 != TYPE_MAIN_VARIANT (prod_type))
 292         return NULL;
 293       if (!widened_name_p (oprnd0, stmt, &half_type0, &def_stmt))
 294         return NULL;
 295       oprnd00 = gimple_assign_rhs1 (def_stmt);
 296       if (!widened_name_p (oprnd1, stmt, &half_type1, &def_stmt))
 297         return NULL;
 298       oprnd01 = gimple_assign_rhs1 (def_stmt);
 299       if (TYPE_MAIN_VARIANT (half_type0) != TYPE_MAIN_VARIANT (half_type1))
 300         return NULL;
 301       if (TYPE_PRECISION (prod_type) != TYPE_PRECISION (half_type0) * 2)
 302         return NULL;
 303     }
 304
 305   half_type = TREE_TYPE (oprnd00);
 306   *type_in = half_type;
 307   *type_out = type;
 308
 309   /* Pattern detected. Create a stmt to be used to replace the pattern: */
 310   var = vect_recog_temp_ssa_var (type, NULL);
 311   rhs = build3 (DOT_PROD_EXPR, type, oprnd00, oprnd01, oprnd1),
 312   pattern_stmt = gimple_build_assign (var, rhs);
 313
 314   if (vect_print_dump_info (REPORT_DETAILS))
 315     {
 316       fprintf (vect_dump, "vect_recog_dot_prod_pattern: detected: ");
 317       print_gimple_stmt (vect_dump, pattern_stmt, 0, TDF_SLIM);
 318     }
 319
 320   /* We don't allow changing the order of the computation in the inner-loop
 321      when doing outer-loop vectorization.  */
 322   if (nested_in_vect_loop_p (loop, last_stmt))
 323     {
 324       if (vect_print_dump_info (REPORT_DETAILS))
 325         fprintf (vect_dump, "vect_recog_dot_prod_pattern: not allowed.");
 326       return NULL;
 327     }
 328
 329   return pattern_stmt;
 330 }
 331
 332 /* Function vect_recog_widen_mult_pattern
 333
 334    Try to find the following pattern:
 335
 336      type a_t, b_t;
 337      TYPE a_T, b_T, prod_T;
 338
 339      S1  a_t = ;
 340      S2  b_t = ;
 341      S3  a_T = (TYPE) a_t;
 342      S4  b_T = (TYPE) b_t;
 343      S5  prod_T = a_T * b_T;
 344
 345    where type 'TYPE' is at least double the size of type 'type'.
 346
 347    Input:
 348
 349    * LAST_STMT: A stmt from which the pattern search begins. In the example,
 350    when this function is called with S5, the pattern {S3,S4,S5} is be detected.
 351
 352    Output:
 353
 354    * TYPE_IN: The type of the input arguments to the pattern.
 355
 356    * TYPE_OUT: The type of the output  of this pattern.
 357
 358    * Return value: A new stmt that will be used to replace the sequence of
 359    stmts that constitute the pattern. In this case it will be:
 360         WIDEN_MULT <a_t, b_t>
 361 */
 362
 363 static gimple
 364 vect_recog_widen_mult_pattern (gimple last_stmt,
 365                                tree *type_in,
 366                                tree *type_out)
 367 {
 368   gimple def_stmt0, def_stmt1;
 369   tree oprnd0, oprnd1;
 370   tree type, half_type0, half_type1;
 371   gimple pattern_stmt;
 372   tree vectype;
 373   tree dummy;
 374   tree var;
 375   enum tree_code dummy_code;
 376   int dummy_int;
 377   VEC (tree, heap) *dummy_vec;
 378
 379   if (!is_gimple_assign (last_stmt))
 380     return NULL;
 381
 382   type = gimple_expr_type (last_stmt);
 383
 384   /* Starting from LAST_STMT, follow the defs of its uses in search
 385      of the above pattern.  */
 386
 387   if (gimple_assign_rhs_code (last_stmt) != MULT_EXPR)
 388     return NULL;
 389
 390   oprnd0 = gimple_assign_rhs1 (last_stmt);
 391   oprnd1 = gimple_assign_rhs2 (last_stmt);
 392   if (TYPE_MAIN_VARIANT (TREE_TYPE (oprnd0)) != TYPE_MAIN_VARIANT (type)
 393       || TYPE_MAIN_VARIANT (TREE_TYPE (oprnd1)) != TYPE_MAIN_VARIANT (type))
 394     return NULL;
 395
 396   /* Check argument 0 */
 397   if (!widened_name_p (oprnd0, last_stmt, &half_type0, &def_stmt0))
 398     return NULL;
 399   oprnd0 = gimple_assign_rhs1 (def_stmt0);
 400
 401   /* Check argument 1 */
 402   if (!widened_name_p (oprnd1, last_stmt, &half_type1, &def_stmt1))
 403     return NULL;
 404   oprnd1 = gimple_assign_rhs1 (def_stmt1);
 405
 406   if (TYPE_MAIN_VARIANT (half_type0) != TYPE_MAIN_VARIANT (half_type1))
 407     return NULL;
 408
 409   /* Pattern detected.  */
 410   if (vect_print_dump_info (REPORT_DETAILS))
 411     fprintf (vect_dump, "vect_recog_widen_mult_pattern: detected: ");
 412
 413   /* Check target support  */
 414   vectype = get_vectype_for_scalar_type (half_type0);
 415   if (!vectype
 416       || !supportable_widening_operation (WIDEN_MULT_EXPR, last_stmt, vectype,
 417                                           &dummy, &dummy, &dummy_code,
 418                                           &dummy_code, &dummy_int, &dummy_vec))
 419     return NULL;
 420
 421   *type_in = vectype;
 422   *type_out = NULL_TREE;
 423
 424   /* Pattern supported. Create a stmt to be used to replace the pattern: */
 425   var = vect_recog_temp_ssa_var (type, NULL);
 426   pattern_stmt = gimple_build_assign_with_ops (WIDEN_MULT_EXPR, var, oprnd0,
 427                                                oprnd1);
 428   SSA_NAME_DEF_STMT (var) = pattern_stmt;
 429
 430   if (vect_print_dump_info (REPORT_DETAILS))
 431     print_gimple_stmt (vect_dump, pattern_stmt, 0, TDF_SLIM);
 432
 433   return pattern_stmt;
 434 }
 435
 436
 437 /* Function vect_recog_pow_pattern
 438
 439    Try to find the following pattern:
 440
 441      x = POW (y, N);
 442
 443    with POW being one of pow, powf, powi, powif and N being
 444    either 2 or 0.5.
 445
 446    Input:
 447
 448    * LAST_STMT: A stmt from which the pattern search begins.
 449
 450    Output:
 451
 452    * TYPE_IN: The type of the input arguments to the pattern.
 453
 454    * TYPE_OUT: The type of the output of this pattern.
 455
 456    * Return value: A new stmt that will be used to replace the sequence of
 457    stmts that constitute the pattern. In this case it will be:
 458         x = x * x
 459    or
 460         x = sqrt (x)
 461 */
 462
 463 static gimple
 464 vect_recog_pow_pattern (gimple last_stmt, tree *type_in, tree *type_out)
 465 {
 466   tree type;
 467   tree fn, base, exp = NULL;
 468   gimple stmt;
 469   tree var;
 470
 471   if (!is_gimple_call (last_stmt) || gimple_call_lhs (last_stmt) == NULL)
 472     return NULL;
 473
 474   type = gimple_expr_type (last_stmt);
 475
 476   fn = gimple_call_fndecl (last_stmt);
 477   switch (DECL_FUNCTION_CODE (fn))
 478     {
 479     case BUILT_IN_POWIF:
 480     case BUILT_IN_POWI:
 481     case BUILT_IN_POWF:
 482     case BUILT_IN_POW:
 483       base = gimple_call_arg (last_stmt, 0);
 484       exp = gimple_call_arg (last_stmt, 1);
 485       if (TREE_CODE (exp) != REAL_CST
 486           && TREE_CODE (exp) != INTEGER_CST)
 487         return NULL;
 488       break;
 489
 490     default:
 491       return NULL;
 492     }
 493
 494   /* We now have a pow or powi builtin function call with a constant
 495      exponent.  */
 496
 497   *type_out = NULL_TREE;
 498
 499   /* Catch squaring.  */
 500   if ((host_integerp (exp, 0)
 501        && tree_low_cst (exp, 0) == 2)
 502       || (TREE_CODE (exp) == REAL_CST
 503           && REAL_VALUES_EQUAL (TREE_REAL_CST (exp), dconst2)))
 504     {
 505       *type_in = TREE_TYPE (base);
 506
 507       var = vect_recog_temp_ssa_var (TREE_TYPE (base), NULL);
 508       stmt = gimple_build_assign_with_ops (MULT_EXPR, var, base, base);
 509       SSA_NAME_DEF_STMT (var) = stmt;
 510       return stmt;
 511     }
 512
 513   /* Catch square root.  */
 514   if (TREE_CODE (exp) == REAL_CST
 515       && REAL_VALUES_EQUAL (TREE_REAL_CST (exp), dconsthalf))
 516     {
 517       tree newfn = mathfn_built_in (TREE_TYPE (base), BUILT_IN_SQRT);
 518       *type_in = get_vectype_for_scalar_type (TREE_TYPE (base));
 519       if (*type_in)
 520         {
 521           gimple stmt = gimple_build_call (newfn, 1, base);
 522           if (vectorizable_function (stmt, *type_in, *type_in)
 523               != NULL_TREE)
 524             {
 525               var = vect_recog_temp_ssa_var (TREE_TYPE (base), stmt);
 526               gimple_call_set_lhs (stmt, var);
 527               return stmt;
 528             }
 529         }
 530     }
 531
 532   return NULL;
 533 }
 534
 535
 536 /* Function vect_recog_widen_sum_pattern
 537
 538    Try to find the following pattern:
 539
 540      type x_t;
 541      TYPE x_T, sum = init;
 542    loop:
 543      sum_0 = phi <init, sum_1>
 544      S1  x_t = *p;
 545      S2  x_T = (TYPE) x_t;
 546      S3  sum_1 = x_T + sum_0;
 547
 548    where type 'TYPE' is at least double the size of type 'type', i.e - we're
 549    summing elements of type 'type' into an accumulator of type 'TYPE'. This is
 550    a special case of a reduction computation.
 551
 552    Input:
 553
 554    * LAST_STMT: A stmt from which the pattern search begins. In the example,
 555    when this function is called with S3, the pattern {S2,S3} will be detected.
 556
 557    Output:
 558
 559    * TYPE_IN: The type of the input arguments to the pattern.
 560
 561    * TYPE_OUT: The type of the output of this pattern.
 562
 563    * Return value: A new stmt that will be used to replace the sequence of
 564    stmts that constitute the pattern. In this case it will be:
 565         WIDEN_SUM <x_t, sum_0>
 566
 567    Note: The widening-sum idiom is a widening reduction pattern that is
 568          vectorized without preserving all the intermediate results. It
 569          produces only N/2 (widened) results (by summing up pairs of
 570          intermediate results) rather than all N results.  Therefore, we
 571          cannot allow this pattern when we want to get all the results and in
 572          the correct order (as is the case when this computation is in an
 573          inner-loop nested in an outer-loop that us being vectorized).  */
 574
 575 static gimple
 576 vect_recog_widen_sum_pattern (gimple last_stmt, tree *type_in, tree *type_out)
 577 {
 578   gimple stmt;
 579   tree oprnd0, oprnd1;
 580   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
 581   tree type, half_type;
 582   gimple pattern_stmt;
 583   loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
 584   struct loop *loop = LOOP_VINFO_LOOP (loop_info);
 585   tree var;
 586
 587   if (!is_gimple_assign (last_stmt))
 588     return NULL;
 589
 590   type = gimple_expr_type (last_stmt);
 591
 592   /* Look for the following pattern
 593           DX = (TYPE) X;
 594           sum_1 = DX + sum_0;
 595      In which DX is at least double the size of X, and sum_1 has been
 596      recognized as a reduction variable.
 597    */
 598
 599   /* Starting from LAST_STMT, follow the defs of its uses in search
 600      of the above pattern.  */
 601
 602   if (gimple_assign_rhs_code (last_stmt) != PLUS_EXPR)
 603     return NULL;
 604
 605   if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_reduction_def)
 606     return NULL;
 607
 608   oprnd0 = gimple_assign_rhs1 (last_stmt);
 609   oprnd1 = gimple_assign_rhs2 (last_stmt);
 610   if (TYPE_MAIN_VARIANT (TREE_TYPE (oprnd0)) != TYPE_MAIN_VARIANT (type)
 611       || TYPE_MAIN_VARIANT (TREE_TYPE (oprnd1)) != TYPE_MAIN_VARIANT (type))
 612     return NULL;
 613
 614   /* So far so good. Since last_stmt was detected as a (summation) reduction,
 615      we know that oprnd1 is the reduction variable (defined by a loop-header
 616      phi), and oprnd0 is an ssa-name defined by a stmt in the loop body.
 617      Left to check that oprnd0 is defined by a cast from type 'type' to type
 618      'TYPE'.  */
 619
 620   if (!widened_name_p (oprnd0, last_stmt, &half_type, &stmt))
 621     return NULL;
 622
 623   oprnd0 = gimple_assign_rhs1 (stmt);
 624   *type_in = half_type;
 625   *type_out = type;
 626
 627   /* Pattern detected. Create a stmt to be used to replace the pattern: */
 628   var = vect_recog_temp_ssa_var (type, NULL);
 629   pattern_stmt = gimple_build_assign_with_ops (WIDEN_SUM_EXPR, var,
 630                                                oprnd0, oprnd1);
 631   SSA_NAME_DEF_STMT (var) = pattern_stmt;
 632
 633   if (vect_print_dump_info (REPORT_DETAILS))
 634     {
 635       fprintf (vect_dump, "vect_recog_widen_sum_pattern: detected: ");
 636       print_gimple_stmt (vect_dump, pattern_stmt, 0, TDF_SLIM);
 637     }
 638
 639   /* We don't allow changing the order of the computation in the inner-loop
 640      when doing outer-loop vectorization.  */
 641   if (nested_in_vect_loop_p (loop, last_stmt))
 642     {
 643       if (vect_print_dump_info (REPORT_DETAILS))
 644         fprintf (vect_dump, "vect_recog_widen_sum_pattern: not allowed.");
 645       return NULL;
 646     }
 647
 648   return pattern_stmt;
 649 }
 650
 651
 652 /* Function vect_pattern_recog_1
 653
 654    Input:
 655    PATTERN_RECOG_FUNC: A pointer to a function that detects a certain
 656         computation pattern.
 657    STMT: A stmt from which the pattern search should start.
 658
 659    If PATTERN_RECOG_FUNC successfully detected the pattern, it creates an
 660    expression that computes the same functionality and can be used to
 661    replace the sequence of stmts that are involved in the pattern.
 662
 663    Output:
 664    This function checks if the expression returned by PATTERN_RECOG_FUNC is
 665    supported in vector form by the target.  We use 'TYPE_IN' to obtain the
 666    relevant vector type. If 'TYPE_IN' is already a vector type, then this
 667    indicates that target support had already been checked by PATTERN_RECOG_FUNC.
 668    If 'TYPE_OUT' is also returned by PATTERN_RECOG_FUNC, we check that it fits
 669    to the available target pattern.
 670
 671    This function also does some bookkeeping, as explained in the documentation
 672    for vect_recog_pattern.  */
 673
 674 static void
 675 vect_pattern_recog_1 (
 676         gimple (* vect_recog_func) (gimple, tree *, tree *),
 677         gimple_stmt_iterator si)
 678 {
 679   gimple stmt = gsi_stmt (si), pattern_stmt;
 680   stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
 681   stmt_vec_info pattern_stmt_info;
 682   loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
 683   tree pattern_vectype;
 684   tree type_in, type_out;
 685   enum tree_code code;
 686
 687   pattern_stmt = (* vect_recog_func) (stmt, &type_in, &type_out);
 688   if (!pattern_stmt)
 689     return;
 690
 691   if (VECTOR_MODE_P (TYPE_MODE (type_in)))
 692     {
 693       /* No need to check target support (already checked by the pattern
 694          recognition function).  */
 695       pattern_vectype = type_in;
 696     }
 697   else
 698     {
 699       enum machine_mode vec_mode;
 700       enum insn_code icode;
 701       optab optab;
 702
 703       /* Check target support  */
 704       pattern_vectype = get_vectype_for_scalar_type (type_in);
 705       if (!pattern_vectype)
 706         return;
 707
 708       if (is_gimple_assign (pattern_stmt))
 709         code = gimple_assign_rhs_code (pattern_stmt);
 710       else
 711         {
 712           gcc_assert (is_gimple_call (pattern_stmt));
 713           code = CALL_EXPR;
 714         }
 715
 716       optab = optab_for_tree_code (code, pattern_vectype, optab_default);
 717       vec_mode = TYPE_MODE (pattern_vectype);
 718       if (!optab
 719           || (icode = optab_handler (optab, vec_mode)->insn_code) ==
 720               CODE_FOR_nothing
 721           || (type_out
 722               && (!get_vectype_for_scalar_type (type_out)
 723                   || (insn_data[icode].operand[0].mode !=
 724                       TYPE_MODE (get_vectype_for_scalar_type (type_out))))))
 725         return;
 726     }
 727
 728   /* Found a vectorizable pattern.  */
 729   if (vect_print_dump_info (REPORT_DETAILS))
 730     {
 731       fprintf (vect_dump, "pattern recognized: ");
 732       print_gimple_stmt (vect_dump, pattern_stmt, 0, TDF_SLIM);
 733     }
 734
 735   /* Mark the stmts that are involved in the pattern. */
 736   gsi_insert_before (&si, pattern_stmt, GSI_SAME_STMT);
 737   set_vinfo_for_stmt (pattern_stmt,
 738                       new_stmt_vec_info (pattern_stmt, loop_vinfo, NULL));
 739   pattern_stmt_info = vinfo_for_stmt (pattern_stmt);
 740
 741   STMT_VINFO_RELATED_STMT (pattern_stmt_info) = stmt;
 742   STMT_VINFO_DEF_TYPE (pattern_stmt_info) = STMT_VINFO_DEF_TYPE (stmt_info);
 743   STMT_VINFO_VECTYPE (pattern_stmt_info) = pattern_vectype;
 744   STMT_VINFO_IN_PATTERN_P (stmt_info) = true;
 745   STMT_VINFO_RELATED_STMT (stmt_info) = pattern_stmt;
 746
 747   return;
 748 }
 749
 750
 751 /* Function vect_pattern_recog
 752
 753    Input:
 754    LOOP_VINFO - a struct_loop_info of a loop in which we want to look for
 755         computation idioms.
 756
 757    Output - for each computation idiom that is detected we insert a new stmt
 758         that provides the same functionality and that can be vectorized. We
 759         also record some information in the struct_stmt_info of the relevant
 760         stmts, as explained below:
 761
 762    At the entry to this function we have the following stmts, with the
 763    following initial value in the STMT_VINFO fields:
 764
 765          stmt                     in_pattern_p  related_stmt    vec_stmt
 766          S1: a_i = ....                 -       -               -
 767          S2: a_2 = ..use(a_i)..         -       -               -
 768          S3: a_1 = ..use(a_2)..         -       -               -
 769          S4: a_0 = ..use(a_1)..         -       -               -
 770          S5: ... = ..use(a_0)..         -       -               -
 771
 772    Say the sequence {S1,S2,S3,S4} was detected as a pattern that can be
 773    represented by a single stmt. We then:
 774    - create a new stmt S6 that will replace the pattern.
 775    - insert the new stmt S6 before the last stmt in the pattern
 776    - fill in the STMT_VINFO fields as follows:
 777
 778                                   in_pattern_p  related_stmt    vec_stmt
 779          S1: a_i = ....                 -       -               -
 780          S2: a_2 = ..use(a_i)..         -       -               -
 781          S3: a_1 = ..use(a_2)..         -       -               -
 782        > S6: a_new = ....               -       S4              -
 783          S4: a_0 = ..use(a_1)..         true    S6              -
 784          S5: ... = ..use(a_0)..         -       -               -
 785
 786    (the last stmt in the pattern (S4) and the new pattern stmt (S6) point
 787     to each other through the RELATED_STMT field).
 788
 789    S6 will be marked as relevant in vect_mark_stmts_to_be_vectorized instead
 790    of S4 because it will replace all its uses.  Stmts {S1,S2,S3} will
 791    remain irrelevant unless used by stmts other than S4.
 792
 793    If vectorization succeeds, vect_transform_stmt will skip over {S1,S2,S3}
 794    (because they are marked as irrelevant). It will vectorize S6, and record
 795    a pointer to the new vector stmt VS6 both from S6 (as usual), and also
 796    from S4. We do that so that when we get to vectorizing stmts that use the
 797    def of S4 (like S5 that uses a_0), we'll know where to take the relevant
 798    vector-def from. S4 will be skipped, and S5 will be vectorized as usual:
 799
 800                                   in_pattern_p  related_stmt    vec_stmt
 801          S1: a_i = ....                 -       -               -
 802          S2: a_2 = ..use(a_i)..         -       -               -
 803          S3: a_1 = ..use(a_2)..         -       -               -
 804        > VS6: va_new = ....             -       -               -
 805          S6: a_new = ....               -       S4              VS6
 806          S4: a_0 = ..use(a_1)..         true    S6              VS6
 807        > VS5: ... = ..vuse(va_new)..    -       -               -
 808          S5: ... = ..use(a_0)..         -       -               -
 809
 810    DCE could then get rid of {S1,S2,S3,S4,S5,S6} (if their defs are not used
 811    elsewhere), and we'll end up with:
 812
 813         VS6: va_new = ....
 814         VS5: ... = ..vuse(va_new)..
 815
 816    If vectorization does not succeed, DCE will clean S6 away (its def is
 817    not used), and we'll end up with the original sequence.
 818 */
 819
 820 void
 821 vect_pattern_recog (loop_vec_info loop_vinfo)
 822 {
 823   struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
 824   basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo);
 825   unsigned int nbbs = loop->num_nodes;
 826   gimple_stmt_iterator si;
 827   gimple stmt;
 828   unsigned int i, j;
 829   gimple (* vect_recog_func_ptr) (gimple, tree *, tree *);
 830
 831   if (vect_print_dump_info (REPORT_DETAILS))
 832     fprintf (vect_dump, "=== vect_pattern_recog ===");
 833
 834   /* Scan through the loop stmts, applying the pattern recognition
 835      functions starting at each stmt visited:  */
 836   for (i = 0; i < nbbs; i++)
 837     {
 838       basic_block bb = bbs[i];
 839       for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si))
 840         {
 841           stmt = gsi_stmt (si);
 842
 843           /* Scan over all generic vect_recog_xxx_pattern functions.  */
 844           for (j = 0; j < NUM_PATTERNS; j++)
 845             {
 846               vect_recog_func_ptr = vect_vect_recog_func_ptrs[j];
 847               vect_pattern_recog_1 (vect_recog_func_ptr, si);
 848             }
 849         }
 850     }
 851 }