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