gcc/tree-vect-patterns.c

   1 /* Analysis Utilities for Loop Vectorization.
   2    Copyright (C) 2006-2016 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 "backend.h"
  25 #include "rtl.h"
  26 #include "tree.h"
  27 #include "gimple.h"
  28 #include "ssa.h"
  29 #include "expmed.h"
  30 #include "optabs-tree.h"
  31 #include "insn-config.h"
  32 #include "recog.h"              /* FIXME: for insn_data */
  33 #include "fold-const.h"
  34 #include "stor-layout.h"
  35 #include "tree-eh.h"
  36 #include "gimplify.h"
  37 #include "gimple-iterator.h"
  38 #include "cfgloop.h"
  39 #include "tree-vectorizer.h"
  40 #include "dumpfile.h"
  41 #include "builtins.h"
  42 #include "internal-fn.h"
  43 #include "case-cfn-macros.h"
  44
  45 /* Pattern recognition functions  */
  46 static gimple *vect_recog_widen_sum_pattern (vec<gimple *> *, tree *,
  47                                             tree *);
  48 static gimple *vect_recog_widen_mult_pattern (vec<gimple *> *, tree *,
  49                                              tree *);
  50 static gimple *vect_recog_dot_prod_pattern (vec<gimple *> *, tree *,
  51                                            tree *);
  52 static gimple *vect_recog_sad_pattern (vec<gimple *> *, tree *,
  53                                       tree *);
  54 static gimple *vect_recog_pow_pattern (vec<gimple *> *, tree *, tree *);
  55 static gimple *vect_recog_over_widening_pattern (vec<gimple *> *, tree *,
  56                                                  tree *);
  57 static gimple *vect_recog_widen_shift_pattern (vec<gimple *> *,
  58                                         tree *, tree *);
  59 static gimple *vect_recog_rotate_pattern (vec<gimple *> *, tree *, tree *);
  60 static gimple *vect_recog_vector_vector_shift_pattern (vec<gimple *> *,
  61                                                       tree *, tree *);
  62 static gimple *vect_recog_divmod_pattern (vec<gimple *> *,
  63                                          tree *, tree *);
  64
  65 static gimple *vect_recog_mult_pattern (vec<gimple *> *,
  66                                        tree *, tree *);
  67
  68 static gimple *vect_recog_mixed_size_cond_pattern (vec<gimple *> *,
  69                                                   tree *, tree *);
  70 static gimple *vect_recog_bool_pattern (vec<gimple *> *, tree *, tree *);
  71 static gimple *vect_recog_mask_conversion_pattern (vec<gimple *> *, tree *, tree *);
  72
  73 struct vect_recog_func
  74 {
  75   vect_recog_func_ptr fn;
  76   const char *name;
  77 };
  78
  79 /* Note that ordering matters - the first pattern matching on a stmt
  80    is taken which means usually the more complex one needs to preceed
  81    the less comples onex (widen_sum only after dot_prod or sad for example).  */
  82 static vect_recog_func vect_vect_recog_func_ptrs[NUM_PATTERNS] = {
  83       { vect_recog_widen_mult_pattern, "widen_mult" },
  84       { vect_recog_dot_prod_pattern, "dot_prod" },
  85       { vect_recog_sad_pattern, "sad" },
  86       { vect_recog_widen_sum_pattern, "widen_sum" },
  87       { vect_recog_pow_pattern, "pow" },
  88       { vect_recog_widen_shift_pattern, "widen_shift" },
  89       { vect_recog_over_widening_pattern, "over_widening" },
  90       { vect_recog_rotate_pattern, "rotate" },
  91       { vect_recog_vector_vector_shift_pattern, "vector_vector_shift" },
  92       { vect_recog_divmod_pattern, "divmod" },
  93       { vect_recog_mult_pattern, "mult" },
  94       { vect_recog_mixed_size_cond_pattern, "mixed_size_cond" },
  95       { vect_recog_bool_pattern, "bool" },
  96       { vect_recog_mask_conversion_pattern, "mask_conversion" }
  97 };
  98
  99 static inline void
 100 append_pattern_def_seq (stmt_vec_info stmt_info, gimple *stmt)
 101 {
 102   gimple_seq_add_stmt_without_update (&STMT_VINFO_PATTERN_DEF_SEQ (stmt_info),
 103                                       stmt);
 104 }
 105
 106 static inline void
 107 new_pattern_def_seq (stmt_vec_info stmt_info, gimple *stmt)
 108 {
 109   STMT_VINFO_PATTERN_DEF_SEQ (stmt_info) = NULL;
 110   append_pattern_def_seq (stmt_info, stmt);
 111 }
 112
 113 /* Check whether STMT2 is in the same loop or basic block as STMT1.
 114    Which of the two applies depends on whether we're currently doing
 115    loop-based or basic-block-based vectorization, as determined by
 116    the vinfo_for_stmt for STMT1 (which must be defined).
 117
 118    If this returns true, vinfo_for_stmt for STMT2 is guaranteed
 119    to be defined as well.  */
 120
 121 static bool
 122 vect_same_loop_or_bb_p (gimple *stmt1, gimple *stmt2)
 123 {
 124   stmt_vec_info stmt_vinfo = vinfo_for_stmt (stmt1);
 125   return vect_stmt_in_region_p (stmt_vinfo->vinfo, stmt2);
 126 }
 127
 128 /* If the LHS of DEF_STMT has a single use, and that statement is
 129    in the same loop or basic block, return it.  */
 130
 131 static gimple *
 132 vect_single_imm_use (gimple *def_stmt)
 133 {
 134   tree lhs = gimple_assign_lhs (def_stmt);
 135   use_operand_p use_p;
 136   gimple *use_stmt;
 137
 138   if (!single_imm_use (lhs, &use_p, &use_stmt))
 139     return NULL;
 140
 141   if (!vect_same_loop_or_bb_p (def_stmt, use_stmt))
 142     return NULL;
 143
 144   return use_stmt;
 145 }
 146
 147 /* Check whether NAME, an ssa-name used in USE_STMT,
 148    is a result of a type promotion, such that:
 149      DEF_STMT: NAME = NOP (name0)
 150    If CHECK_SIGN is TRUE, check that either both types are signed or both are
 151    unsigned.  */
 152
 153 static bool
 154 type_conversion_p (tree name, gimple *use_stmt, bool check_sign,
 155                    tree *orig_type, gimple **def_stmt, bool *promotion)
 156 {
 157   gimple *dummy_gimple;
 158   stmt_vec_info stmt_vinfo;
 159   tree type = TREE_TYPE (name);
 160   tree oprnd0;
 161   enum vect_def_type dt;
 162
 163   stmt_vinfo = vinfo_for_stmt (use_stmt);
 164   if (!vect_is_simple_use (name, stmt_vinfo->vinfo, def_stmt, &dt))
 165     return false;
 166
 167   if (dt != vect_internal_def
 168       && dt != vect_external_def && dt != vect_constant_def)
 169     return false;
 170
 171   if (!*def_stmt)
 172     return false;
 173
 174   if (dt == vect_internal_def)
 175     {
 176       stmt_vec_info def_vinfo = vinfo_for_stmt (*def_stmt);
 177       if (STMT_VINFO_IN_PATTERN_P (def_vinfo))
 178         return false;
 179     }
 180
 181   if (!is_gimple_assign (*def_stmt))
 182     return false;
 183
 184   if (!CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (*def_stmt)))
 185     return false;
 186
 187   oprnd0 = gimple_assign_rhs1 (*def_stmt);
 188
 189   *orig_type = TREE_TYPE (oprnd0);
 190   if (!INTEGRAL_TYPE_P (type) || !INTEGRAL_TYPE_P (*orig_type)
 191       || ((TYPE_UNSIGNED (type) != TYPE_UNSIGNED (*orig_type)) && check_sign))
 192     return false;
 193
 194   if (TYPE_PRECISION (type) >= (TYPE_PRECISION (*orig_type) * 2))
 195     *promotion = true;
 196   else
 197     *promotion = false;
 198
 199   if (!vect_is_simple_use (oprnd0, stmt_vinfo->vinfo, &dummy_gimple, &dt))
 200     return false;
 201
 202   return true;
 203 }
 204
 205 /* Helper to return a new temporary for pattern of TYPE for STMT.  If STMT
 206    is NULL, the caller must set SSA_NAME_DEF_STMT for the returned SSA var. */
 207
 208 static tree
 209 vect_recog_temp_ssa_var (tree type, gimple *stmt)
 210 {
 211   return make_temp_ssa_name (type, stmt, "patt");
 212 }
 213
 214 /* Function vect_recog_dot_prod_pattern
 215
 216    Try to find the following pattern:
 217
 218      type x_t, y_t;
 219      TYPE1 prod;
 220      TYPE2 sum = init;
 221    loop:
 222      sum_0 = phi <init, sum_1>
 223      S1  x_t = ...
 224      S2  y_t = ...
 225      S3  x_T = (TYPE1) x_t;
 226      S4  y_T = (TYPE1) y_t;
 227      S5  prod = x_T * y_T;
 228      [S6  prod = (TYPE2) prod;  #optional]
 229      S7  sum_1 = prod + sum_0;
 230
 231    where 'TYPE1' is exactly double the size of type 'type', and 'TYPE2' is the
 232    same size of 'TYPE1' or bigger. This is a special case of a reduction
 233    computation.
 234
 235    Input:
 236
 237    * STMTS: Contains a stmt from which the pattern search begins.  In the
 238    example, when this function is called with S7, the pattern {S3,S4,S5,S6,S7}
 239    will be detected.
 240
 241    Output:
 242
 243    * TYPE_IN: The type of the input arguments to the pattern.
 244
 245    * TYPE_OUT: The type of the output  of this pattern.
 246
 247    * Return value: A new stmt that will be used to replace the sequence of
 248    stmts that constitute the pattern. In this case it will be:
 249         WIDEN_DOT_PRODUCT <x_t, y_t, sum_0>
 250
 251    Note: The dot-prod idiom is a widening reduction pattern that is
 252          vectorized without preserving all the intermediate results. It
 253          produces only N/2 (widened) results (by summing up pairs of
 254          intermediate results) rather than all N results.  Therefore, we
 255          cannot allow this pattern when we want to get all the results and in
 256          the correct order (as is the case when this computation is in an
 257          inner-loop nested in an outer-loop that us being vectorized).  */
 258
 259 static gimple *
 260 vect_recog_dot_prod_pattern (vec<gimple *> *stmts, tree *type_in,
 261                              tree *type_out)
 262 {
 263   gimple *stmt, *last_stmt = (*stmts)[0];
 264   tree oprnd0, oprnd1;
 265   tree oprnd00, oprnd01;
 266   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
 267   tree type, half_type;
 268   gimple *pattern_stmt;
 269   tree prod_type;
 270   loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
 271   struct loop *loop;
 272   tree var;
 273   bool promotion;
 274
 275   if (!loop_info)
 276     return NULL;
 277
 278   loop = LOOP_VINFO_LOOP (loop_info);
 279
 280   /* We don't allow changing the order of the computation in the inner-loop
 281      when doing outer-loop vectorization.  */
 282   if (loop && nested_in_vect_loop_p (loop, last_stmt))
 283     return NULL;
 284
 285   if (!is_gimple_assign (last_stmt))
 286     return NULL;
 287
 288   type = gimple_expr_type (last_stmt);
 289
 290   /* Look for the following pattern
 291           DX = (TYPE1) X;
 292           DY = (TYPE1) Y;
 293           DPROD = DX * DY;
 294           DDPROD = (TYPE2) DPROD;
 295           sum_1 = DDPROD + sum_0;
 296      In which
 297      - DX is double the size of X
 298      - DY is double the size of Y
 299      - DX, DY, DPROD all have the same type
 300      - sum is the same size of DPROD or bigger
 301      - sum has been recognized as a reduction variable.
 302
 303      This is equivalent to:
 304        DPROD = X w* Y;          #widen mult
 305        sum_1 = DPROD w+ sum_0;  #widen summation
 306      or
 307        DPROD = X w* Y;          #widen mult
 308        sum_1 = DPROD + sum_0;   #summation
 309    */
 310
 311   /* Starting from LAST_STMT, follow the defs of its uses in search
 312      of the above pattern.  */
 313
 314   if (gimple_assign_rhs_code (last_stmt) != PLUS_EXPR)
 315     return NULL;
 316
 317   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
 318     {
 319       /* Has been detected as widening-summation?  */
 320
 321       stmt = STMT_VINFO_RELATED_STMT (stmt_vinfo);
 322       type = gimple_expr_type (stmt);
 323       if (gimple_assign_rhs_code (stmt) != WIDEN_SUM_EXPR)
 324         return NULL;
 325       oprnd0 = gimple_assign_rhs1 (stmt);
 326       oprnd1 = gimple_assign_rhs2 (stmt);
 327       half_type = TREE_TYPE (oprnd0);
 328     }
 329   else
 330     {
 331       gimple *def_stmt;
 332
 333       if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_reduction_def
 334           && ! STMT_VINFO_GROUP_FIRST_ELEMENT (stmt_vinfo))
 335         return NULL;
 336       oprnd0 = gimple_assign_rhs1 (last_stmt);
 337       oprnd1 = gimple_assign_rhs2 (last_stmt);
 338       if (!types_compatible_p (TREE_TYPE (oprnd0), type)
 339           || !types_compatible_p (TREE_TYPE (oprnd1), type))
 340         return NULL;
 341       stmt = last_stmt;
 342
 343       if (type_conversion_p (oprnd0, stmt, true, &half_type, &def_stmt,
 344                              &promotion)
 345           && promotion)
 346         {
 347           stmt = def_stmt;
 348           oprnd0 = gimple_assign_rhs1 (stmt);
 349         }
 350       else
 351         half_type = type;
 352     }
 353
 354   /* So far so good.  Since last_stmt was detected as a (summation) reduction,
 355      we know that oprnd1 is the reduction variable (defined by a loop-header
 356      phi), and oprnd0 is an ssa-name defined by a stmt in the loop body.
 357      Left to check that oprnd0 is defined by a (widen_)mult_expr  */
 358   if (TREE_CODE (oprnd0) != SSA_NAME)
 359     return NULL;
 360
 361   prod_type = half_type;
 362   stmt = SSA_NAME_DEF_STMT (oprnd0);
 363
 364   /* It could not be the dot_prod pattern if the stmt is outside the loop.  */
 365   if (!gimple_bb (stmt) || !flow_bb_inside_loop_p (loop, gimple_bb (stmt)))
 366     return NULL;
 367
 368   /* FORNOW.  Can continue analyzing the def-use chain when this stmt in a phi
 369      inside the loop (in case we are analyzing an outer-loop).  */
 370   if (!is_gimple_assign (stmt))
 371     return NULL;
 372   stmt_vinfo = vinfo_for_stmt (stmt);
 373   gcc_assert (stmt_vinfo);
 374   if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_internal_def)
 375     return NULL;
 376   if (gimple_assign_rhs_code (stmt) != MULT_EXPR)
 377     return NULL;
 378   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
 379     {
 380       /* Has been detected as a widening multiplication?  */
 381
 382       stmt = STMT_VINFO_RELATED_STMT (stmt_vinfo);
 383       if (gimple_assign_rhs_code (stmt) != WIDEN_MULT_EXPR)
 384         return NULL;
 385       stmt_vinfo = vinfo_for_stmt (stmt);
 386       gcc_assert (stmt_vinfo);
 387       gcc_assert (STMT_VINFO_DEF_TYPE (stmt_vinfo) == vect_internal_def);
 388       oprnd00 = gimple_assign_rhs1 (stmt);
 389       oprnd01 = gimple_assign_rhs2 (stmt);
 390       STMT_VINFO_PATTERN_DEF_SEQ (vinfo_for_stmt (last_stmt))
 391           = STMT_VINFO_PATTERN_DEF_SEQ (stmt_vinfo);
 392     }
 393   else
 394     {
 395       tree half_type0, half_type1;
 396       gimple *def_stmt;
 397       tree oprnd0, oprnd1;
 398
 399       oprnd0 = gimple_assign_rhs1 (stmt);
 400       oprnd1 = gimple_assign_rhs2 (stmt);
 401       if (!types_compatible_p (TREE_TYPE (oprnd0), prod_type)
 402           || !types_compatible_p (TREE_TYPE (oprnd1), prod_type))
 403         return NULL;
 404       if (!type_conversion_p (oprnd0, stmt, true, &half_type0, &def_stmt,
 405                               &promotion)
 406           || !promotion)
 407         return NULL;
 408       oprnd00 = gimple_assign_rhs1 (def_stmt);
 409       if (!type_conversion_p (oprnd1, stmt, true, &half_type1, &def_stmt,
 410                               &promotion)
 411           || !promotion)
 412         return NULL;
 413       oprnd01 = gimple_assign_rhs1 (def_stmt);
 414       if (!types_compatible_p (half_type0, half_type1))
 415         return NULL;
 416       if (TYPE_PRECISION (prod_type) != TYPE_PRECISION (half_type0) * 2)
 417         return NULL;
 418     }
 419
 420   half_type = TREE_TYPE (oprnd00);
 421   *type_in = half_type;
 422   *type_out = type;
 423
 424   /* Pattern detected. 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 (var, DOT_PROD_EXPR,
 427                                       oprnd00, oprnd01, oprnd1);
 428
 429   if (dump_enabled_p ())
 430     {
 431       dump_printf_loc (MSG_NOTE, vect_location,
 432                        "vect_recog_dot_prod_pattern: detected: ");
 433       dump_gimple_stmt (MSG_NOTE, TDF_SLIM, pattern_stmt, 0);
 434     }
 435
 436   return pattern_stmt;
 437 }
 438
 439
 440 /* Function vect_recog_sad_pattern
 441
 442    Try to find the following Sum of Absolute Difference (SAD) pattern:
 443
 444      type x_t, y_t;
 445      signed TYPE1 diff, abs_diff;
 446      TYPE2 sum = init;
 447    loop:
 448      sum_0 = phi <init, sum_1>
 449      S1  x_t = ...
 450      S2  y_t = ...
 451      S3  x_T = (TYPE1) x_t;
 452      S4  y_T = (TYPE1) y_t;
 453      S5  diff = x_T - y_T;
 454      S6  abs_diff = ABS_EXPR <diff>;
 455      [S7  abs_diff = (TYPE2) abs_diff;  #optional]
 456      S8  sum_1 = abs_diff + sum_0;
 457
 458    where 'TYPE1' is at least double the size of type 'type', and 'TYPE2' is the
 459    same size of 'TYPE1' or bigger. This is a special case of a reduction
 460    computation.
 461
 462    Input:
 463
 464    * STMTS: Contains a stmt from which the pattern search begins.  In the
 465    example, when this function is called with S8, the pattern
 466    {S3,S4,S5,S6,S7,S8} will be detected.
 467
 468    Output:
 469
 470    * TYPE_IN: The type of the input arguments to the pattern.
 471
 472    * TYPE_OUT: The type of the output of this pattern.
 473
 474    * Return value: A new stmt that will be used to replace the sequence of
 475    stmts that constitute the pattern. In this case it will be:
 476         SAD_EXPR <x_t, y_t, sum_0>
 477   */
 478
 479 static gimple *
 480 vect_recog_sad_pattern (vec<gimple *> *stmts, tree *type_in,
 481                              tree *type_out)
 482 {
 483   gimple *last_stmt = (*stmts)[0];
 484   tree sad_oprnd0, sad_oprnd1;
 485   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
 486   tree half_type;
 487   loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
 488   struct loop *loop;
 489   bool promotion;
 490
 491   if (!loop_info)
 492     return NULL;
 493
 494   loop = LOOP_VINFO_LOOP (loop_info);
 495
 496   /* We don't allow changing the order of the computation in the inner-loop
 497      when doing outer-loop vectorization.  */
 498   if (loop && nested_in_vect_loop_p (loop, last_stmt))
 499     return NULL;
 500
 501   if (!is_gimple_assign (last_stmt))
 502     return NULL;
 503
 504   tree sum_type = gimple_expr_type (last_stmt);
 505
 506   /* Look for the following pattern
 507           DX = (TYPE1) X;
 508           DY = (TYPE1) Y;
 509           DDIFF = DX - DY;
 510           DAD = ABS_EXPR <DDIFF>;
 511           DDPROD = (TYPE2) DPROD;
 512           sum_1 = DAD + sum_0;
 513      In which
 514      - DX is at least double the size of X
 515      - DY is at least double the size of Y
 516      - DX, DY, DDIFF, DAD all have the same type
 517      - sum is the same size of DAD or bigger
 518      - sum has been recognized as a reduction variable.
 519
 520      This is equivalent to:
 521        DDIFF = X w- Y;          #widen sub
 522        DAD = ABS_EXPR <DDIFF>;
 523        sum_1 = DAD w+ sum_0;    #widen summation
 524      or
 525        DDIFF = X w- Y;          #widen sub
 526        DAD = ABS_EXPR <DDIFF>;
 527        sum_1 = DAD + sum_0;     #summation
 528    */
 529
 530   /* Starting from LAST_STMT, follow the defs of its uses in search
 531      of the above pattern.  */
 532
 533   if (gimple_assign_rhs_code (last_stmt) != PLUS_EXPR)
 534     return NULL;
 535
 536   tree plus_oprnd0, plus_oprnd1;
 537
 538   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
 539     {
 540       /* Has been detected as widening-summation?  */
 541
 542       gimple *stmt = STMT_VINFO_RELATED_STMT (stmt_vinfo);
 543       sum_type = gimple_expr_type (stmt);
 544       if (gimple_assign_rhs_code (stmt) != WIDEN_SUM_EXPR)
 545         return NULL;
 546       plus_oprnd0 = gimple_assign_rhs1 (stmt);
 547       plus_oprnd1 = gimple_assign_rhs2 (stmt);
 548       half_type = TREE_TYPE (plus_oprnd0);
 549     }
 550   else
 551     {
 552       gimple *def_stmt;
 553
 554       if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_reduction_def
 555           && ! STMT_VINFO_GROUP_FIRST_ELEMENT (stmt_vinfo))
 556         return NULL;
 557       plus_oprnd0 = gimple_assign_rhs1 (last_stmt);
 558       plus_oprnd1 = gimple_assign_rhs2 (last_stmt);
 559       if (!types_compatible_p (TREE_TYPE (plus_oprnd0), sum_type)
 560           || !types_compatible_p (TREE_TYPE (plus_oprnd1), sum_type))
 561         return NULL;
 562
 563       /* The type conversion could be promotion, demotion,
 564          or just signed -> unsigned.  */
 565       if (type_conversion_p (plus_oprnd0, last_stmt, false,
 566                              &half_type, &def_stmt, &promotion))
 567         plus_oprnd0 = gimple_assign_rhs1 (def_stmt);
 568       else
 569         half_type = sum_type;
 570     }
 571
 572   /* So far so good.  Since last_stmt was detected as a (summation) reduction,
 573      we know that plus_oprnd1 is the reduction variable (defined by a loop-header
 574      phi), and plus_oprnd0 is an ssa-name defined by a stmt in the loop body.
 575      Then check that plus_oprnd0 is defined by an abs_expr.  */
 576
 577   if (TREE_CODE (plus_oprnd0) != SSA_NAME)
 578     return NULL;
 579
 580   tree abs_type = half_type;
 581   gimple *abs_stmt = SSA_NAME_DEF_STMT (plus_oprnd0);
 582
 583   /* It could not be the sad pattern if the abs_stmt is outside the loop.  */
 584   if (!gimple_bb (abs_stmt) || !flow_bb_inside_loop_p (loop, gimple_bb (abs_stmt)))
 585     return NULL;
 586
 587   /* FORNOW.  Can continue analyzing the def-use chain when this stmt in a phi
 588      inside the loop (in case we are analyzing an outer-loop).  */
 589   if (!is_gimple_assign (abs_stmt))
 590     return NULL;
 591
 592   stmt_vec_info abs_stmt_vinfo = vinfo_for_stmt (abs_stmt);
 593   gcc_assert (abs_stmt_vinfo);
 594   if (STMT_VINFO_DEF_TYPE (abs_stmt_vinfo) != vect_internal_def)
 595     return NULL;
 596   if (gimple_assign_rhs_code (abs_stmt) != ABS_EXPR)
 597     return NULL;
 598
 599   tree abs_oprnd = gimple_assign_rhs1 (abs_stmt);
 600   if (!types_compatible_p (TREE_TYPE (abs_oprnd), abs_type))
 601     return NULL;
 602   if (TYPE_UNSIGNED (abs_type))
 603     return NULL;
 604
 605   /* We then detect if the operand of abs_expr is defined by a minus_expr.  */
 606
 607   if (TREE_CODE (abs_oprnd) != SSA_NAME)
 608     return NULL;
 609
 610   gimple *diff_stmt = SSA_NAME_DEF_STMT (abs_oprnd);
 611
 612   /* It could not be the sad pattern if the diff_stmt is outside the loop.  */
 613   if (!gimple_bb (diff_stmt)
 614       || !flow_bb_inside_loop_p (loop, gimple_bb (diff_stmt)))
 615     return NULL;
 616
 617   /* FORNOW.  Can continue analyzing the def-use chain when this stmt in a phi
 618      inside the loop (in case we are analyzing an outer-loop).  */
 619   if (!is_gimple_assign (diff_stmt))
 620     return NULL;
 621
 622   stmt_vec_info diff_stmt_vinfo = vinfo_for_stmt (diff_stmt);
 623   gcc_assert (diff_stmt_vinfo);
 624   if (STMT_VINFO_DEF_TYPE (diff_stmt_vinfo) != vect_internal_def)
 625     return NULL;
 626   if (gimple_assign_rhs_code (diff_stmt) != MINUS_EXPR)
 627     return NULL;
 628
 629   tree half_type0, half_type1;
 630   gimple *def_stmt;
 631
 632   tree minus_oprnd0 = gimple_assign_rhs1 (diff_stmt);
 633   tree minus_oprnd1 = gimple_assign_rhs2 (diff_stmt);
 634
 635   if (!types_compatible_p (TREE_TYPE (minus_oprnd0), abs_type)
 636       || !types_compatible_p (TREE_TYPE (minus_oprnd1), abs_type))
 637     return NULL;
 638   if (!type_conversion_p (minus_oprnd0, diff_stmt, false,
 639                           &half_type0, &def_stmt, &promotion)
 640       || !promotion)
 641     return NULL;
 642   sad_oprnd0 = gimple_assign_rhs1 (def_stmt);
 643
 644   if (!type_conversion_p (minus_oprnd1, diff_stmt, false,
 645                           &half_type1, &def_stmt, &promotion)
 646       || !promotion)
 647     return NULL;
 648   sad_oprnd1 = gimple_assign_rhs1 (def_stmt);
 649
 650   if (!types_compatible_p (half_type0, half_type1))
 651     return NULL;
 652   if (TYPE_PRECISION (abs_type) < TYPE_PRECISION (half_type0) * 2
 653       || TYPE_PRECISION (sum_type) < TYPE_PRECISION (half_type0) * 2)
 654     return NULL;
 655
 656   *type_in = TREE_TYPE (sad_oprnd0);
 657   *type_out = sum_type;
 658
 659   /* Pattern detected. Create a stmt to be used to replace the pattern: */
 660   tree var = vect_recog_temp_ssa_var (sum_type, NULL);
 661   gimple *pattern_stmt = gimple_build_assign (var, SAD_EXPR, sad_oprnd0,
 662                                               sad_oprnd1, plus_oprnd1);
 663
 664   if (dump_enabled_p ())
 665     {
 666       dump_printf_loc (MSG_NOTE, vect_location,
 667                        "vect_recog_sad_pattern: detected: ");
 668       dump_gimple_stmt (MSG_NOTE, TDF_SLIM, pattern_stmt, 0);
 669     }
 670
 671   return pattern_stmt;
 672 }
 673
 674
 675 /* Handle widening operation by a constant.  At the moment we support MULT_EXPR
 676    and LSHIFT_EXPR.
 677
 678    For MULT_EXPR we check that CONST_OPRND fits HALF_TYPE, and for LSHIFT_EXPR
 679    we check that CONST_OPRND is less or equal to the size of HALF_TYPE.
 680
 681    Otherwise, if the type of the result (TYPE) is at least 4 times bigger than
 682    HALF_TYPE, and there is an intermediate type (2 times smaller than TYPE)
 683    that satisfies the above restrictions,  we can perform a widening opeartion
 684    from the intermediate type to TYPE and replace a_T = (TYPE) a_t;
 685    with a_it = (interm_type) a_t;  Store such operation in *WSTMT.  */
 686
 687 static bool
 688 vect_handle_widen_op_by_const (gimple *stmt, enum tree_code code,
 689                                tree const_oprnd, tree *oprnd,
 690                                gimple **wstmt, tree type,
 691                                tree *half_type, gimple *def_stmt)
 692 {
 693   tree new_type, new_oprnd;
 694
 695   if (code != MULT_EXPR && code != LSHIFT_EXPR)
 696     return false;
 697
 698   if (((code == MULT_EXPR && int_fits_type_p (const_oprnd, *half_type))
 699         || (code == LSHIFT_EXPR
 700             && compare_tree_int (const_oprnd, TYPE_PRECISION (*half_type))
 701                 != 1))
 702       && TYPE_PRECISION (type) == (TYPE_PRECISION (*half_type) * 2))
 703     {
 704       /* CONST_OPRND is a constant of HALF_TYPE.  */
 705       *oprnd = gimple_assign_rhs1 (def_stmt);
 706       return true;
 707     }
 708
 709   if (TYPE_PRECISION (type) < (TYPE_PRECISION (*half_type) * 4))
 710     return false;
 711
 712   if (!vect_same_loop_or_bb_p (stmt, def_stmt))
 713     return false;
 714
 715   /* TYPE is 4 times bigger than HALF_TYPE, try widening operation for
 716      a type 2 times bigger than HALF_TYPE.  */
 717   new_type = build_nonstandard_integer_type (TYPE_PRECISION (type) / 2,
 718                                              TYPE_UNSIGNED (type));
 719   if ((code == MULT_EXPR && !int_fits_type_p (const_oprnd, new_type))
 720       || (code == LSHIFT_EXPR
 721           && compare_tree_int (const_oprnd, TYPE_PRECISION (new_type)) == 1))
 722     return false;
 723
 724   /* Use NEW_TYPE for widening operation and create a_T = (NEW_TYPE) a_t;  */
 725   *oprnd = gimple_assign_rhs1 (def_stmt);
 726   new_oprnd = make_ssa_name (new_type);
 727   *wstmt = gimple_build_assign (new_oprnd, NOP_EXPR, *oprnd);
 728   *oprnd = new_oprnd;
 729
 730   *half_type = new_type;
 731   return true;
 732 }
 733
 734
 735 /* Function vect_recog_widen_mult_pattern
 736
 737    Try to find the following pattern:
 738
 739      type1 a_t;
 740      type2 b_t;
 741      TYPE a_T, b_T, prod_T;
 742
 743      S1  a_t = ;
 744      S2  b_t = ;
 745      S3  a_T = (TYPE) a_t;
 746      S4  b_T = (TYPE) b_t;
 747      S5  prod_T = a_T * b_T;
 748
 749    where type 'TYPE' is at least double the size of type 'type1' and 'type2'.
 750
 751    Also detect unsigned cases:
 752
 753      unsigned type1 a_t;
 754      unsigned type2 b_t;
 755      unsigned TYPE u_prod_T;
 756      TYPE a_T, b_T, prod_T;
 757
 758      S1  a_t = ;
 759      S2  b_t = ;
 760      S3  a_T = (TYPE) a_t;
 761      S4  b_T = (TYPE) b_t;
 762      S5  prod_T = a_T * b_T;
 763      S6  u_prod_T = (unsigned TYPE) prod_T;
 764
 765    and multiplication by constants:
 766
 767      type a_t;
 768      TYPE a_T, prod_T;
 769
 770      S1  a_t = ;
 771      S3  a_T = (TYPE) a_t;
 772      S5  prod_T = a_T * CONST;
 773
 774    A special case of multiplication by constants is when 'TYPE' is 4 times
 775    bigger than 'type', but CONST fits an intermediate type 2 times smaller
 776    than 'TYPE'.  In that case we create an additional pattern stmt for S3
 777    to create a variable of the intermediate type, and perform widen-mult
 778    on the intermediate type as well:
 779
 780      type a_t;
 781      interm_type a_it;
 782      TYPE a_T, prod_T,  prod_T';
 783
 784      S1  a_t = ;
 785      S3  a_T = (TYPE) a_t;
 786            '--> a_it = (interm_type) a_t;
 787      S5  prod_T = a_T * CONST;
 788            '--> prod_T' = a_it w* CONST;
 789
 790    Input/Output:
 791
 792    * STMTS: Contains a stmt from which the pattern search begins.  In the
 793    example, when this function is called with S5, the pattern {S3,S4,S5,(S6)}
 794    is detected.  In case of unsigned widen-mult, the original stmt (S5) is
 795    replaced with S6 in STMTS.  In case of multiplication by a constant
 796    of an intermediate type (the last case above), STMTS also contains S3
 797    (inserted before S5).
 798
 799    Output:
 800
 801    * TYPE_IN: The type of the input arguments to the pattern.
 802
 803    * TYPE_OUT: The type of the output of this pattern.
 804
 805    * Return value: A new stmt that will be used to replace the sequence of
 806    stmts that constitute the pattern.  In this case it will be:
 807         WIDEN_MULT <a_t, b_t>
 808    If the result of WIDEN_MULT needs to be converted to a larger type, the
 809    returned stmt will be this type conversion stmt.
 810 */
 811
 812 static gimple *
 813 vect_recog_widen_mult_pattern (vec<gimple *> *stmts,
 814                                tree *type_in, tree *type_out)
 815 {
 816   gimple *last_stmt = stmts->pop ();
 817   gimple *def_stmt0, *def_stmt1;
 818   tree oprnd0, oprnd1;
 819   tree type, half_type0, half_type1;
 820   gimple *new_stmt = NULL, *pattern_stmt = NULL;
 821   tree vectype, vecitype;
 822   tree var;
 823   enum tree_code dummy_code;
 824   int dummy_int;
 825   vec<tree> dummy_vec;
 826   bool op1_ok;
 827   bool promotion;
 828
 829   if (!is_gimple_assign (last_stmt))
 830     return NULL;
 831
 832   type = gimple_expr_type (last_stmt);
 833
 834   /* Starting from LAST_STMT, follow the defs of its uses in search
 835      of the above pattern.  */
 836
 837   if (gimple_assign_rhs_code (last_stmt) != MULT_EXPR)
 838     return NULL;
 839
 840   oprnd0 = gimple_assign_rhs1 (last_stmt);
 841   oprnd1 = gimple_assign_rhs2 (last_stmt);
 842   if (!types_compatible_p (TREE_TYPE (oprnd0), type)
 843       || !types_compatible_p (TREE_TYPE (oprnd1), type))
 844     return NULL;
 845
 846   /* Check argument 0.  */
 847   if (!type_conversion_p (oprnd0, last_stmt, false, &half_type0, &def_stmt0,
 848                          &promotion)
 849       || !promotion)
 850      return NULL;
 851   /* Check argument 1.  */
 852   op1_ok = type_conversion_p (oprnd1, last_stmt, false, &half_type1,
 853                               &def_stmt1, &promotion);
 854
 855   if (op1_ok && promotion)
 856     {
 857       oprnd0 = gimple_assign_rhs1 (def_stmt0);
 858       oprnd1 = gimple_assign_rhs1 (def_stmt1);
 859     }
 860   else
 861     {
 862       if (TREE_CODE (oprnd1) == INTEGER_CST
 863           && TREE_CODE (half_type0) == INTEGER_TYPE
 864           && vect_handle_widen_op_by_const (last_stmt, MULT_EXPR, oprnd1,
 865                                             &oprnd0, &new_stmt, type,
 866                                             &half_type0, def_stmt0))
 867         {
 868           half_type1 = half_type0;
 869           oprnd1 = fold_convert (half_type1, oprnd1);
 870         }
 871       else
 872         return NULL;
 873     }
 874
 875   /* If the two arguments have different sizes, convert the one with
 876      the smaller type into the larger type.  */
 877   if (TYPE_PRECISION (half_type0) != TYPE_PRECISION (half_type1))
 878     {
 879       /* If we already used up the single-stmt slot give up.  */
 880       if (new_stmt)
 881         return NULL;
 882
 883       tree* oprnd = NULL;
 884       gimple *def_stmt = NULL;
 885
 886       if (TYPE_PRECISION (half_type0) < TYPE_PRECISION (half_type1))
 887         {
 888           def_stmt = def_stmt0;
 889           half_type0 = half_type1;
 890           oprnd = &oprnd0;
 891         }
 892       else
 893         {
 894           def_stmt = def_stmt1;
 895           half_type1 = half_type0;
 896           oprnd = &oprnd1;
 897         }
 898
 899       tree old_oprnd = gimple_assign_rhs1 (def_stmt);
 900       tree new_oprnd = make_ssa_name (half_type0);
 901       new_stmt = gimple_build_assign (new_oprnd, NOP_EXPR, old_oprnd);
 902       *oprnd = new_oprnd;
 903     }
 904
 905   /* Handle unsigned case.  Look for
 906      S6  u_prod_T = (unsigned TYPE) prod_T;
 907      Use unsigned TYPE as the type for WIDEN_MULT_EXPR.  */
 908   if (TYPE_UNSIGNED (type) != TYPE_UNSIGNED (half_type0))
 909     {
 910       gimple *use_stmt;
 911       tree use_lhs;
 912       tree use_type;
 913
 914       if (TYPE_UNSIGNED (type) == TYPE_UNSIGNED (half_type1))
 915         return NULL;
 916
 917       use_stmt = vect_single_imm_use (last_stmt);
 918       if (!use_stmt || !is_gimple_assign (use_stmt)
 919           || !CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (use_stmt)))
 920         return NULL;
 921
 922       use_lhs = gimple_assign_lhs (use_stmt);
 923       use_type = TREE_TYPE (use_lhs);
 924       if (!INTEGRAL_TYPE_P (use_type)
 925           || (TYPE_UNSIGNED (type) == TYPE_UNSIGNED (use_type))
 926           || (TYPE_PRECISION (type) != TYPE_PRECISION (use_type)))
 927         return NULL;
 928
 929       type = use_type;
 930       last_stmt = use_stmt;
 931     }
 932
 933   if (!types_compatible_p (half_type0, half_type1))
 934     return NULL;
 935
 936   /* If TYPE is more than twice larger than HALF_TYPE, we use WIDEN_MULT
 937      to get an intermediate result of type ITYPE.  In this case we need
 938      to build a statement to convert this intermediate result to type TYPE.  */
 939   tree itype = type;
 940   if (TYPE_PRECISION (type) > TYPE_PRECISION (half_type0) * 2)
 941     itype = build_nonstandard_integer_type
 942               (GET_MODE_BITSIZE (TYPE_MODE (half_type0)) * 2,
 943                TYPE_UNSIGNED (type));
 944
 945   /* Pattern detected.  */
 946   if (dump_enabled_p ())
 947     dump_printf_loc (MSG_NOTE, vect_location,
 948                      "vect_recog_widen_mult_pattern: detected:\n");
 949
 950   /* Check target support  */
 951   vectype = get_vectype_for_scalar_type (half_type0);
 952   vecitype = get_vectype_for_scalar_type (itype);
 953   if (!vectype
 954       || !vecitype
 955       || !supportable_widening_operation (WIDEN_MULT_EXPR, last_stmt,
 956                                           vecitype, vectype,
 957                                           &dummy_code, &dummy_code,
 958                                           &dummy_int, &dummy_vec))
 959     return NULL;
 960
 961   *type_in = vectype;
 962   *type_out = get_vectype_for_scalar_type (type);
 963
 964   /* Pattern supported. Create a stmt to be used to replace the pattern: */
 965   var = vect_recog_temp_ssa_var (itype, NULL);
 966   pattern_stmt = gimple_build_assign (var, WIDEN_MULT_EXPR, oprnd0, oprnd1);
 967
 968   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
 969   STMT_VINFO_PATTERN_DEF_SEQ (stmt_vinfo) = NULL;
 970
 971   /* If the original two operands have different sizes, we may need to convert
 972      the smaller one into the larget type.  If this is the case, at this point
 973      the new stmt is already built.  */
 974   if (new_stmt)
 975     {
 976       append_pattern_def_seq (stmt_vinfo, new_stmt);
 977       stmt_vec_info new_stmt_info
 978         = new_stmt_vec_info (new_stmt, stmt_vinfo->vinfo);
 979       set_vinfo_for_stmt (new_stmt, new_stmt_info);
 980       STMT_VINFO_VECTYPE (new_stmt_info) = vectype;
 981     }
 982
 983   /* If ITYPE is not TYPE, we need to build a type convertion stmt to convert
 984      the result of the widen-mult operation into type TYPE.  */
 985   if (itype != type)
 986     {
 987       append_pattern_def_seq (stmt_vinfo, pattern_stmt);
 988       stmt_vec_info pattern_stmt_info
 989         = new_stmt_vec_info (pattern_stmt, stmt_vinfo->vinfo);
 990       set_vinfo_for_stmt (pattern_stmt, pattern_stmt_info);
 991       STMT_VINFO_VECTYPE (pattern_stmt_info) = vecitype;
 992       pattern_stmt = gimple_build_assign (vect_recog_temp_ssa_var (type, NULL),
 993                                           NOP_EXPR,
 994                                           gimple_assign_lhs (pattern_stmt));
 995     }
 996
 997   if (dump_enabled_p ())
 998     dump_gimple_stmt_loc (MSG_NOTE, vect_location, TDF_SLIM, pattern_stmt, 0);
 999
1000   stmts->safe_push (last_stmt);
1001   return pattern_stmt;
1002 }
1003
1004
1005 /* Function vect_recog_pow_pattern
1006
1007    Try to find the following pattern:
1008
1009      x = POW (y, N);
1010
1011    with POW being one of pow, powf, powi, powif and N being
1012    either 2 or 0.5.
1013
1014    Input:
1015
1016    * LAST_STMT: A stmt from which the pattern search begins.
1017
1018    Output:
1019
1020    * TYPE_IN: The type of the input arguments to the pattern.
1021
1022    * TYPE_OUT: The type of the output of this pattern.
1023
1024    * Return value: A new stmt that will be used to replace the sequence of
1025    stmts that constitute the pattern. In this case it will be:
1026         x = x * x
1027    or
1028         x = sqrt (x)
1029 */
1030
1031 static gimple *
1032 vect_recog_pow_pattern (vec<gimple *> *stmts, tree *type_in,
1033                         tree *type_out)
1034 {
1035   gimple *last_stmt = (*stmts)[0];
1036   tree base, exp = NULL;
1037   gimple *stmt;
1038   tree var;
1039
1040   if (!is_gimple_call (last_stmt) || gimple_call_lhs (last_stmt) == NULL)
1041     return NULL;
1042
1043   switch (gimple_call_combined_fn (last_stmt))
1044     {
1045     CASE_CFN_POW:
1046     CASE_CFN_POWI:
1047       base = gimple_call_arg (last_stmt, 0);
1048       exp = gimple_call_arg (last_stmt, 1);
1049       if (TREE_CODE (exp) != REAL_CST
1050           && TREE_CODE (exp) != INTEGER_CST)
1051         return NULL;
1052       break;
1053
1054     default:
1055       return NULL;
1056     }
1057
1058   /* We now have a pow or powi builtin function call with a constant
1059      exponent.  */
1060
1061   *type_out = NULL_TREE;
1062
1063   /* Catch squaring.  */
1064   if ((tree_fits_shwi_p (exp)
1065        && tree_to_shwi (exp) == 2)
1066       || (TREE_CODE (exp) == REAL_CST
1067           && real_equal (&TREE_REAL_CST (exp), &dconst2)))
1068     {
1069       *type_in = TREE_TYPE (base);
1070
1071       var = vect_recog_temp_ssa_var (TREE_TYPE (base), NULL);
1072       stmt = gimple_build_assign (var, MULT_EXPR, base, base);
1073       return stmt;
1074     }
1075
1076   /* Catch square root.  */
1077   if (TREE_CODE (exp) == REAL_CST
1078       && real_equal (&TREE_REAL_CST (exp), &dconsthalf))
1079     {
1080       *type_in = get_vectype_for_scalar_type (TREE_TYPE (base));
1081       if (*type_in
1082           && direct_internal_fn_supported_p (IFN_SQRT, *type_in,
1083                                              OPTIMIZE_FOR_SPEED))
1084         {
1085           gcall *stmt = gimple_build_call_internal (IFN_SQRT, 1, base);
1086           var = vect_recog_temp_ssa_var (TREE_TYPE (base), stmt);
1087           gimple_call_set_lhs (stmt, var);
1088           return stmt;
1089         }
1090     }
1091
1092   return NULL;
1093 }
1094
1095
1096 /* Function vect_recog_widen_sum_pattern
1097
1098    Try to find the following pattern:
1099
1100      type x_t;
1101      TYPE x_T, sum = init;
1102    loop:
1103      sum_0 = phi <init, sum_1>
1104      S1  x_t = *p;
1105      S2  x_T = (TYPE) x_t;
1106      S3  sum_1 = x_T + sum_0;
1107
1108    where type 'TYPE' is at least double the size of type 'type', i.e - we're
1109    summing elements of type 'type' into an accumulator of type 'TYPE'. This is
1110    a special case of a reduction computation.
1111
1112    Input:
1113
1114    * LAST_STMT: A stmt from which the pattern search begins. In the example,
1115    when this function is called with S3, the pattern {S2,S3} will be detected.
1116
1117    Output:
1118
1119    * TYPE_IN: The type of the input arguments to the pattern.
1120
1121    * TYPE_OUT: The type of the output of this pattern.
1122
1123    * Return value: A new stmt that will be used to replace the sequence of
1124    stmts that constitute the pattern. In this case it will be:
1125         WIDEN_SUM <x_t, sum_0>
1126
1127    Note: The widening-sum idiom is a widening reduction pattern that is
1128          vectorized without preserving all the intermediate results. It
1129          produces only N/2 (widened) results (by summing up pairs of
1130          intermediate results) rather than all N results.  Therefore, we
1131          cannot allow this pattern when we want to get all the results and in
1132          the correct order (as is the case when this computation is in an
1133          inner-loop nested in an outer-loop that us being vectorized).  */
1134
1135 static gimple *
1136 vect_recog_widen_sum_pattern (vec<gimple *> *stmts, tree *type_in,
1137                               tree *type_out)
1138 {
1139   gimple *stmt, *last_stmt = (*stmts)[0];
1140   tree oprnd0, oprnd1;
1141   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
1142   tree type, half_type;
1143   gimple *pattern_stmt;
1144   loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_vinfo);
1145   struct loop *loop;
1146   tree var;
1147   bool promotion;
1148
1149   if (!loop_info)
1150     return NULL;
1151
1152   loop = LOOP_VINFO_LOOP (loop_info);
1153
1154   /* We don't allow changing the order of the computation in the inner-loop
1155      when doing outer-loop vectorization.  */
1156   if (loop && nested_in_vect_loop_p (loop, last_stmt))
1157     return NULL;
1158
1159   if (!is_gimple_assign (last_stmt))
1160     return NULL;
1161
1162   type = gimple_expr_type (last_stmt);
1163
1164   /* Look for the following pattern
1165           DX = (TYPE) X;
1166           sum_1 = DX + sum_0;
1167      In which DX is at least double the size of X, and sum_1 has been
1168      recognized as a reduction variable.
1169    */
1170
1171   /* Starting from LAST_STMT, follow the defs of its uses in search
1172      of the above pattern.  */
1173
1174   if (gimple_assign_rhs_code (last_stmt) != PLUS_EXPR)
1175     return NULL;
1176
1177   if (STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_reduction_def
1178       && ! STMT_VINFO_GROUP_FIRST_ELEMENT (stmt_vinfo))
1179     return NULL;
1180
1181   oprnd0 = gimple_assign_rhs1 (last_stmt);
1182   oprnd1 = gimple_assign_rhs2 (last_stmt);
1183   if (!types_compatible_p (TREE_TYPE (oprnd0), type)
1184       || !types_compatible_p (TREE_TYPE (oprnd1), type))
1185     return NULL;
1186
1187   /* So far so good.  Since last_stmt was detected as a (summation) reduction,
1188      we know that oprnd1 is the reduction variable (defined by a loop-header
1189      phi), and oprnd0 is an ssa-name defined by a stmt in the loop body.
1190      Left to check that oprnd0 is defined by a cast from type 'type' to type
1191      'TYPE'.  */
1192
1193   if (!type_conversion_p (oprnd0, last_stmt, true, &half_type, &stmt,
1194                           &promotion)
1195       || !promotion)
1196      return NULL;
1197
1198   oprnd0 = gimple_assign_rhs1 (stmt);
1199   *type_in = half_type;
1200   *type_out = type;
1201
1202   /* Pattern detected. Create a stmt to be used to replace the pattern: */
1203   var = vect_recog_temp_ssa_var (type, NULL);
1204   pattern_stmt = gimple_build_assign (var, WIDEN_SUM_EXPR, oprnd0, oprnd1);
1205
1206   if (dump_enabled_p ())
1207     {
1208       dump_printf_loc (MSG_NOTE, vect_location,
1209                        "vect_recog_widen_sum_pattern: detected: ");
1210       dump_gimple_stmt (MSG_NOTE, TDF_SLIM, pattern_stmt, 0);
1211     }
1212
1213   return pattern_stmt;
1214 }
1215
1216
1217 /* Return TRUE if the operation in STMT can be performed on a smaller type.
1218
1219    Input:
1220    STMT - a statement to check.
1221    DEF - we support operations with two operands, one of which is constant.
1222          The other operand can be defined by a demotion operation, or by a
1223          previous statement in a sequence of over-promoted operations.  In the
1224          later case DEF is used to replace that operand.  (It is defined by a
1225          pattern statement we created for the previous statement in the
1226          sequence).
1227
1228    Input/output:
1229    NEW_TYPE - Output: a smaller type that we are trying to use.  Input: if not
1230          NULL, it's the type of DEF.
1231    STMTS - additional pattern statements.  If a pattern statement (type
1232          conversion) is created in this function, its original statement is
1233          added to STMTS.
1234
1235    Output:
1236    OP0, OP1 - if the operation fits a smaller type, OP0 and OP1 are the new
1237          operands to use in the new pattern statement for STMT (will be created
1238          in vect_recog_over_widening_pattern ()).
1239    NEW_DEF_STMT - in case DEF has to be promoted, we create two pattern
1240          statements for STMT: the first one is a type promotion and the second
1241          one is the operation itself.  We return the type promotion statement
1242          in NEW_DEF_STMT and further store it in STMT_VINFO_PATTERN_DEF_SEQ of
1243          the second pattern statement.  */
1244
1245 static bool
1246 vect_operation_fits_smaller_type (gimple *stmt, tree def, tree *new_type,
1247                                   tree *op0, tree *op1, gimple **new_def_stmt,
1248                                   vec<gimple *> *stmts)
1249 {
1250   enum tree_code code;
1251   tree const_oprnd, oprnd;
1252   tree interm_type = NULL_TREE, half_type, new_oprnd, type;
1253   gimple *def_stmt, *new_stmt;
1254   bool first = false;
1255   bool promotion;
1256
1257   *op0 = NULL_TREE;
1258   *op1 = NULL_TREE;
1259   *new_def_stmt = NULL;
1260
1261   if (!is_gimple_assign (stmt))
1262     return false;
1263
1264   code = gimple_assign_rhs_code (stmt);
1265   if (code != LSHIFT_EXPR && code != RSHIFT_EXPR
1266       && code != BIT_IOR_EXPR && code != BIT_XOR_EXPR && code != BIT_AND_EXPR)
1267     return false;
1268
1269   oprnd = gimple_assign_rhs1 (stmt);
1270   const_oprnd = gimple_assign_rhs2 (stmt);
1271   type = gimple_expr_type (stmt);
1272
1273   if (TREE_CODE (oprnd) != SSA_NAME
1274       || TREE_CODE (const_oprnd) != INTEGER_CST)
1275     return false;
1276
1277   /* If oprnd has other uses besides that in stmt we cannot mark it
1278      as being part of a pattern only.  */
1279   if (!has_single_use (oprnd))
1280     return false;
1281
1282   /* If we are in the middle of a sequence, we use DEF from a previous
1283      statement.  Otherwise, OPRND has to be a result of type promotion.  */
1284   if (*new_type)
1285     {
1286       half_type = *new_type;
1287       oprnd = def;
1288     }
1289   else
1290     {
1291       first = true;
1292       if (!type_conversion_p (oprnd, stmt, false, &half_type, &def_stmt,
1293                               &promotion)
1294           || !promotion
1295           || !vect_same_loop_or_bb_p (stmt, def_stmt))
1296         return false;
1297     }
1298
1299   /* Can we perform the operation on a smaller type?  */
1300   switch (code)
1301     {
1302       case BIT_IOR_EXPR:
1303       case BIT_XOR_EXPR:
1304       case BIT_AND_EXPR:
1305         if (!int_fits_type_p (const_oprnd, half_type))
1306           {
1307             /* HALF_TYPE is not enough.  Try a bigger type if possible.  */
1308             if (TYPE_PRECISION (type) < (TYPE_PRECISION (half_type) * 4))
1309               return false;
1310
1311             interm_type = build_nonstandard_integer_type (
1312                         TYPE_PRECISION (half_type) * 2, TYPE_UNSIGNED (type));
1313             if (!int_fits_type_p (const_oprnd, interm_type))
1314               return false;
1315           }
1316
1317         break;
1318
1319       case LSHIFT_EXPR:
1320         /* Try intermediate type - HALF_TYPE is not enough for sure.  */
1321         if (TYPE_PRECISION (type) < (TYPE_PRECISION (half_type) * 4))
1322           return false;
1323
1324         /* Check that HALF_TYPE size + shift amount <= INTERM_TYPE size.
1325           (e.g., if the original value was char, the shift amount is at most 8
1326            if we want to use short).  */
1327         if (compare_tree_int (const_oprnd, TYPE_PRECISION (half_type)) == 1)
1328           return false;
1329
1330         interm_type = build_nonstandard_integer_type (
1331                         TYPE_PRECISION (half_type) * 2, TYPE_UNSIGNED (type));
1332
1333         if (!vect_supportable_shift (code, interm_type))
1334           return false;
1335
1336         break;
1337
1338       case RSHIFT_EXPR:
1339         if (vect_supportable_shift (code, half_type))
1340           break;
1341
1342         /* Try intermediate type - HALF_TYPE is not supported.  */
1343         if (TYPE_PRECISION (type) < (TYPE_PRECISION (half_type) * 4))
1344           return false;
1345
1346         interm_type = build_nonstandard_integer_type (
1347                         TYPE_PRECISION (half_type) * 2, TYPE_UNSIGNED (type));
1348
1349         if (!vect_supportable_shift (code, interm_type))
1350           return false;
1351
1352         break;
1353
1354       default:
1355         gcc_unreachable ();
1356     }
1357
1358   /* There are four possible cases:
1359      1. OPRND is defined by a type promotion (in that case FIRST is TRUE, it's
1360         the first statement in the sequence)
1361         a. The original, HALF_TYPE, is not enough - we replace the promotion
1362            from HALF_TYPE to TYPE with a promotion to INTERM_TYPE.
1363         b. HALF_TYPE is sufficient, OPRND is set as the RHS of the original
1364            promotion.
1365      2. OPRND is defined by a pattern statement we created.
1366         a. Its type is not sufficient for the operation, we create a new stmt:
1367            a type conversion for OPRND from HALF_TYPE to INTERM_TYPE.  We store
1368            this statement in NEW_DEF_STMT, and it is later put in
1369            STMT_VINFO_PATTERN_DEF_SEQ of the pattern statement for STMT.
1370         b. OPRND is good to use in the new statement.  */
1371   if (first)
1372     {
1373       if (interm_type)
1374         {
1375           /* Replace the original type conversion HALF_TYPE->TYPE with
1376              HALF_TYPE->INTERM_TYPE.  */
1377           if (STMT_VINFO_RELATED_STMT (vinfo_for_stmt (def_stmt)))
1378             {
1379               new_stmt = STMT_VINFO_RELATED_STMT (vinfo_for_stmt (def_stmt));
1380               /* Check if the already created pattern stmt is what we need.  */
1381               if (!is_gimple_assign (new_stmt)
1382                   || !CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (new_stmt))
1383                   || TREE_TYPE (gimple_assign_lhs (new_stmt)) != interm_type)
1384                 return false;
1385
1386               stmts->safe_push (def_stmt);
1387               oprnd = gimple_assign_lhs (new_stmt);
1388             }
1389           else
1390             {
1391               /* Create NEW_OPRND = (INTERM_TYPE) OPRND.  */
1392               oprnd = gimple_assign_rhs1 (def_stmt);
1393               new_oprnd = make_ssa_name (interm_type);
1394               new_stmt = gimple_build_assign (new_oprnd, NOP_EXPR, oprnd);
1395               STMT_VINFO_RELATED_STMT (vinfo_for_stmt (def_stmt)) = new_stmt;
1396               stmts->safe_push (def_stmt);
1397               oprnd = new_oprnd;
1398             }
1399         }
1400       else
1401         {
1402           /* Retrieve the operand before the type promotion.  */
1403           oprnd = gimple_assign_rhs1 (def_stmt);
1404         }
1405     }
1406   else
1407     {
1408       if (interm_type)
1409         {
1410           /* Create a type conversion HALF_TYPE->INTERM_TYPE.  */
1411           new_oprnd = make_ssa_name (interm_type);
1412           new_stmt = gimple_build_assign (new_oprnd, NOP_EXPR, oprnd);
1413           oprnd = new_oprnd;
1414           *new_def_stmt = new_stmt;
1415         }
1416
1417       /* Otherwise, OPRND is already set.  */
1418     }
1419
1420   if (interm_type)
1421     *new_type = interm_type;
1422   else
1423     *new_type = half_type;
1424
1425   *op0 = oprnd;
1426   *op1 = fold_convert (*new_type, const_oprnd);
1427
1428   return true;
1429 }
1430
1431
1432 /* Try to find a statement or a sequence of statements that can be performed
1433    on a smaller type:
1434
1435      type x_t;
1436      TYPE x_T, res0_T, res1_T;
1437    loop:
1438      S1  x_t = *p;
1439      S2  x_T = (TYPE) x_t;
1440      S3  res0_T = op (x_T, C0);
1441      S4  res1_T = op (res0_T, C1);
1442      S5  ... = () res1_T;  - type demotion
1443
1444    where type 'TYPE' is at least double the size of type 'type', C0 and C1 are
1445    constants.
1446    Check if S3 and S4 can be done on a smaller type than 'TYPE', it can either
1447    be 'type' or some intermediate type.  For now, we expect S5 to be a type
1448    demotion operation.  We also check that S3 and S4 have only one use.  */
1449
1450 static gimple *
1451 vect_recog_over_widening_pattern (vec<gimple *> *stmts,
1452                                   tree *type_in, tree *type_out)
1453 {
1454   gimple *stmt = stmts->pop ();
1455   gimple *pattern_stmt = NULL, *new_def_stmt, *prev_stmt = NULL,
1456          *use_stmt = NULL;
1457   tree op0, op1, vectype = NULL_TREE, use_lhs, use_type;
1458   tree var = NULL_TREE, new_type = NULL_TREE, new_oprnd;
1459   bool first;
1460   tree type = NULL;
1461
1462   first = true;
1463   while (1)
1464     {
1465       if (!vinfo_for_stmt (stmt)
1466           || STMT_VINFO_IN_PATTERN_P (vinfo_for_stmt (stmt)))
1467         return NULL;
1468
1469       new_def_stmt = NULL;
1470       if (!vect_operation_fits_smaller_type (stmt, var, &new_type,
1471                                              &op0, &op1, &new_def_stmt,
1472                                              stmts))
1473         {
1474           if (first)
1475             return NULL;
1476           else
1477             break;
1478         }
1479
1480       /* STMT can be performed on a smaller type.  Check its uses.  */
1481       use_stmt = vect_single_imm_use (stmt);
1482       if (!use_stmt || !is_gimple_assign (use_stmt))
1483         return NULL;
1484
1485       /* Create pattern statement for STMT.  */
1486       vectype = get_vectype_for_scalar_type (new_type);
1487       if (!vectype)
1488         return NULL;
1489
1490       /* We want to collect all the statements for which we create pattern
1491          statetments, except for the case when the last statement in the
1492          sequence doesn't have a corresponding pattern statement.  In such
1493          case we associate the last pattern statement with the last statement
1494          in the sequence.  Therefore, we only add the original statement to
1495          the list if we know that it is not the last.  */
1496       if (prev_stmt)
1497         stmts->safe_push (prev_stmt);
1498
1499       var = vect_recog_temp_ssa_var (new_type, NULL);
1500       pattern_stmt
1501         = gimple_build_assign (var, gimple_assign_rhs_code (stmt), op0, op1);
1502       STMT_VINFO_RELATED_STMT (vinfo_for_stmt (stmt)) = pattern_stmt;
1503       new_pattern_def_seq (vinfo_for_stmt (stmt), new_def_stmt);
1504
1505       if (dump_enabled_p ())
1506         {
1507           dump_printf_loc (MSG_NOTE, vect_location,
1508                            "created pattern stmt: ");
1509           dump_gimple_stmt (MSG_NOTE, TDF_SLIM, pattern_stmt, 0);
1510         }
1511
1512       type = gimple_expr_type (stmt);
1513       prev_stmt = stmt;
1514       stmt = use_stmt;
1515
1516       first = false;
1517     }
1518
1519   /* We got a sequence.  We expect it to end with a type demotion operation.
1520      Otherwise, we quit (for now).  There are three possible cases: the
1521      conversion is to NEW_TYPE (we don't do anything), the conversion is to
1522      a type bigger than NEW_TYPE and/or the signedness of USE_TYPE and
1523      NEW_TYPE differs (we create a new conversion statement).  */
1524   if (CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (use_stmt)))
1525     {
1526       use_lhs = gimple_assign_lhs (use_stmt);
1527       use_type = TREE_TYPE (use_lhs);
1528       /* Support only type demotion or signedess change.  */
1529       if (!INTEGRAL_TYPE_P (use_type)
1530           || TYPE_PRECISION (type) <= TYPE_PRECISION (use_type))
1531         return NULL;
1532
1533       /* Check that NEW_TYPE is not bigger than the conversion result.  */
1534       if (TYPE_PRECISION (new_type) > TYPE_PRECISION (use_type))
1535         return NULL;
1536
1537       if (TYPE_UNSIGNED (new_type) != TYPE_UNSIGNED (use_type)
1538           || TYPE_PRECISION (new_type) != TYPE_PRECISION (use_type))
1539         {
1540           /* Create NEW_TYPE->USE_TYPE conversion.  */
1541           new_oprnd = make_ssa_name (use_type);
1542           pattern_stmt = gimple_build_assign (new_oprnd, NOP_EXPR, var);
1543           STMT_VINFO_RELATED_STMT (vinfo_for_stmt (use_stmt)) = pattern_stmt;
1544
1545           *type_in = get_vectype_for_scalar_type (new_type);
1546           *type_out = get_vectype_for_scalar_type (use_type);
1547
1548           /* We created a pattern statement for the last statement in the
1549              sequence, so we don't need to associate it with the pattern
1550              statement created for PREV_STMT.  Therefore, we add PREV_STMT
1551              to the list in order to mark it later in vect_pattern_recog_1.  */
1552           if (prev_stmt)
1553             stmts->safe_push (prev_stmt);
1554         }
1555       else
1556         {
1557           if (prev_stmt)
1558             STMT_VINFO_PATTERN_DEF_SEQ (vinfo_for_stmt (use_stmt))
1559                = STMT_VINFO_PATTERN_DEF_SEQ (vinfo_for_stmt (prev_stmt));
1560
1561           *type_in = vectype;
1562           *type_out = NULL_TREE;
1563         }
1564
1565       stmts->safe_push (use_stmt);
1566     }
1567   else
1568     /* TODO: support general case, create a conversion to the correct type.  */
1569     return NULL;
1570
1571   /* Pattern detected.  */
1572   if (dump_enabled_p ())
1573     {
1574       dump_printf_loc (MSG_NOTE, vect_location,
1575                        "vect_recog_over_widening_pattern: detected: ");
1576       dump_gimple_stmt (MSG_NOTE, TDF_SLIM, pattern_stmt, 0);
1577     }
1578
1579   return pattern_stmt;
1580 }
1581
1582 /* Detect widening shift pattern:
1583
1584    type a_t;
1585    TYPE a_T, res_T;
1586
1587    S1 a_t = ;
1588    S2 a_T = (TYPE) a_t;
1589    S3 res_T = a_T << CONST;
1590
1591   where type 'TYPE' is at least double the size of type 'type'.
1592
1593   Also detect cases where the shift result is immediately converted
1594   to another type 'result_type' that is no larger in size than 'TYPE'.
1595   In those cases we perform a widen-shift that directly results in
1596   'result_type', to avoid a possible over-widening situation:
1597
1598   type a_t;
1599   TYPE a_T, res_T;
1600   result_type res_result;
1601
1602   S1 a_t = ;
1603   S2 a_T = (TYPE) a_t;
1604   S3 res_T = a_T << CONST;
1605   S4 res_result = (result_type) res_T;
1606       '--> res_result' = a_t w<< CONST;
1607
1608   And a case when 'TYPE' is 4 times bigger than 'type'.  In that case we
1609   create an additional pattern stmt for S2 to create a variable of an
1610   intermediate type, and perform widen-shift on the intermediate type:
1611
1612   type a_t;
1613   interm_type a_it;
1614   TYPE a_T, res_T, res_T';
1615
1616   S1 a_t = ;
1617   S2 a_T = (TYPE) a_t;
1618       '--> a_it = (interm_type) a_t;
1619   S3 res_T = a_T << CONST;
1620       '--> res_T' = a_it <<* CONST;
1621
1622   Input/Output:
1623
1624   * STMTS: Contains a stmt from which the pattern search begins.
1625     In case of unsigned widen-shift, the original stmt (S3) is replaced with S4
1626     in STMTS.  When an intermediate type is used and a pattern statement is
1627     created for S2, we also put S2 here (before S3).
1628
1629   Output:
1630
1631   * TYPE_IN: The type of the input arguments to the pattern.
1632
1633   * TYPE_OUT: The type of the output of this pattern.
1634
1635   * Return value: A new stmt that will be used to replace the sequence of
1636     stmts that constitute the pattern.  In this case it will be:
1637     WIDEN_LSHIFT_EXPR <a_t, CONST>.  */
1638
1639 static gimple *
1640 vect_recog_widen_shift_pattern (vec<gimple *> *stmts,
1641                                 tree *type_in, tree *type_out)
1642 {
1643   gimple *last_stmt = stmts->pop ();
1644   gimple *def_stmt0;
1645   tree oprnd0, oprnd1;
1646   tree type, half_type0;
1647   gimple *pattern_stmt;
1648   tree vectype, vectype_out = NULL_TREE;
1649   tree var;
1650   enum tree_code dummy_code;
1651   int dummy_int;
1652   vec<tree>  dummy_vec;
1653   gimple *use_stmt;
1654   bool promotion;
1655
1656   if (!is_gimple_assign (last_stmt) || !vinfo_for_stmt (last_stmt))
1657     return NULL;
1658
1659   if (STMT_VINFO_IN_PATTERN_P (vinfo_for_stmt (last_stmt)))
1660     return NULL;
1661
1662   if (gimple_assign_rhs_code (last_stmt) != LSHIFT_EXPR)
1663     return NULL;
1664
1665   oprnd0 = gimple_assign_rhs1 (last_stmt);
1666   oprnd1 = gimple_assign_rhs2 (last_stmt);
1667   if (TREE_CODE (oprnd0) != SSA_NAME || TREE_CODE (oprnd1) != INTEGER_CST)
1668     return NULL;
1669
1670   /* Check operand 0: it has to be defined by a type promotion.  */
1671   if (!type_conversion_p (oprnd0, last_stmt, false, &half_type0, &def_stmt0,
1672                           &promotion)
1673       || !promotion)
1674      return NULL;
1675
1676   /* Check operand 1: has to be positive.  We check that it fits the type
1677      in vect_handle_widen_op_by_const ().  */
1678   if (tree_int_cst_compare (oprnd1, size_zero_node) <= 0)
1679     return NULL;
1680
1681   oprnd0 = gimple_assign_rhs1 (def_stmt0);
1682   type = gimple_expr_type (last_stmt);
1683
1684   /* Check for subsequent conversion to another type.  */
1685   use_stmt = vect_single_imm_use (last_stmt);
1686   if (use_stmt && is_gimple_assign (use_stmt)
1687       && CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (use_stmt))
1688       && !STMT_VINFO_IN_PATTERN_P (vinfo_for_stmt (use_stmt)))
1689     {
1690       tree use_lhs = gimple_assign_lhs (use_stmt);
1691       tree use_type = TREE_TYPE (use_lhs);
1692
1693       if (INTEGRAL_TYPE_P (use_type)
1694           && TYPE_PRECISION (use_type) <= TYPE_PRECISION (type))
1695         {
1696           last_stmt = use_stmt;
1697           type = use_type;
1698         }
1699     }
1700
1701   /* Check if this a widening operation.  */
1702   gimple *wstmt = NULL;
1703   if (!vect_handle_widen_op_by_const (last_stmt, LSHIFT_EXPR, oprnd1,
1704                                       &oprnd0, &wstmt,
1705                                       type, &half_type0, def_stmt0))
1706     return NULL;
1707
1708   /* Pattern detected.  */
1709   if (dump_enabled_p ())
1710     dump_printf_loc (MSG_NOTE, vect_location,
1711                      "vect_recog_widen_shift_pattern: detected:\n");
1712
1713   /* Check target support.  */
1714   vectype = get_vectype_for_scalar_type (half_type0);
1715   vectype_out = get_vectype_for_scalar_type (type);
1716
1717   if (!vectype
1718       || !vectype_out
1719       || !supportable_widening_operation (WIDEN_LSHIFT_EXPR, last_stmt,
1720                                           vectype_out, vectype,
1721                                           &dummy_code, &dummy_code,
1722                                           &dummy_int, &dummy_vec))
1723     return NULL;
1724
1725   *type_in = vectype;
1726   *type_out = vectype_out;
1727
1728   /* Pattern supported.  Create a stmt to be used to replace the pattern.  */
1729   var = vect_recog_temp_ssa_var (type, NULL);
1730   pattern_stmt =
1731     gimple_build_assign (var, WIDEN_LSHIFT_EXPR, oprnd0, oprnd1);
1732   if (wstmt)
1733     {
1734       stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
1735       new_pattern_def_seq (stmt_vinfo, wstmt);
1736       stmt_vec_info new_stmt_info
1737         = new_stmt_vec_info (wstmt, stmt_vinfo->vinfo);
1738       set_vinfo_for_stmt (wstmt, new_stmt_info);
1739       STMT_VINFO_VECTYPE (new_stmt_info) = vectype;
1740     }
1741
1742   if (dump_enabled_p ())
1743     dump_gimple_stmt_loc (MSG_NOTE, vect_location, TDF_SLIM, pattern_stmt, 0);
1744
1745   stmts->safe_push (last_stmt);
1746   return pattern_stmt;
1747 }
1748
1749 /* Detect a rotate pattern wouldn't be otherwise vectorized:
1750
1751    type a_t, b_t, c_t;
1752
1753    S0 a_t = b_t r<< c_t;
1754
1755   Input/Output:
1756
1757   * STMTS: Contains a stmt from which the pattern search begins,
1758     i.e. the shift/rotate stmt.  The original stmt (S0) is replaced
1759     with a sequence:
1760
1761    S1 d_t = -c_t;
1762    S2 e_t = d_t & (B - 1);
1763    S3 f_t = b_t << c_t;
1764    S4 g_t = b_t >> e_t;
1765    S0 a_t = f_t | g_t;
1766
1767     where B is element bitsize of type.
1768
1769   Output:
1770
1771   * TYPE_IN: The type of the input arguments to the pattern.
1772
1773   * TYPE_OUT: The type of the output of this pattern.
1774
1775   * Return value: A new stmt that will be used to replace the rotate
1776     S0 stmt.  */
1777
1778 static gimple *
1779 vect_recog_rotate_pattern (vec<gimple *> *stmts, tree *type_in, tree *type_out)
1780 {
1781   gimple *last_stmt = stmts->pop ();
1782   tree oprnd0, oprnd1, lhs, var, var1, var2, vectype, type, stype, def, def2;
1783   gimple *pattern_stmt, *def_stmt;
1784   enum tree_code rhs_code;
1785   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
1786   vec_info *vinfo = stmt_vinfo->vinfo;
1787   enum vect_def_type dt;
1788   optab optab1, optab2;
1789   edge ext_def = NULL;
1790
1791   if (!is_gimple_assign (last_stmt))
1792     return NULL;
1793
1794   rhs_code = gimple_assign_rhs_code (last_stmt);
1795   switch (rhs_code)
1796     {
1797     case LROTATE_EXPR:
1798     case RROTATE_EXPR:
1799       break;
1800     default:
1801       return NULL;
1802     }
1803
1804   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
1805     return NULL;
1806
1807   lhs = gimple_assign_lhs (last_stmt);
1808   oprnd0 = gimple_assign_rhs1 (last_stmt);
1809   type = TREE_TYPE (oprnd0);
1810   oprnd1 = gimple_assign_rhs2 (last_stmt);
1811   if (TREE_CODE (oprnd0) != SSA_NAME
1812       || TYPE_PRECISION (TREE_TYPE (lhs)) != TYPE_PRECISION (type)
1813       || !INTEGRAL_TYPE_P (type)
1814       || !TYPE_UNSIGNED (type))
1815     return NULL;
1816
1817   if (!vect_is_simple_use (oprnd1, vinfo, &def_stmt, &dt))
1818     return NULL;
1819
1820   if (dt != vect_internal_def
1821       && dt != vect_constant_def
1822       && dt != vect_external_def)
1823     return NULL;
1824
1825   vectype = get_vectype_for_scalar_type (type);
1826   if (vectype == NULL_TREE)
1827     return NULL;
1828
1829   /* If vector/vector or vector/scalar rotate is supported by the target,
1830      don't do anything here.  */
1831   optab1 = optab_for_tree_code (rhs_code, vectype, optab_vector);
1832   if (optab1
1833       && optab_handler (optab1, TYPE_MODE (vectype)) != CODE_FOR_nothing)
1834     return NULL;
1835
1836   if (is_a <bb_vec_info> (vinfo) || dt != vect_internal_def)
1837     {
1838       optab2 = optab_for_tree_code (rhs_code, vectype, optab_scalar);
1839       if (optab2
1840           && optab_handler (optab2, TYPE_MODE (vectype)) != CODE_FOR_nothing)
1841         return NULL;
1842     }
1843
1844   /* If vector/vector or vector/scalar shifts aren't supported by the target,
1845      don't do anything here either.  */
1846   optab1 = optab_for_tree_code (LSHIFT_EXPR, vectype, optab_vector);
1847   optab2 = optab_for_tree_code (RSHIFT_EXPR, vectype, optab_vector);
1848   if (!optab1
1849       || optab_handler (optab1, TYPE_MODE (vectype)) == CODE_FOR_nothing
1850       || !optab2
1851       || optab_handler (optab2, TYPE_MODE (vectype)) == CODE_FOR_nothing)
1852     {
1853       if (! is_a <bb_vec_info> (vinfo) && dt == vect_internal_def)
1854         return NULL;
1855       optab1 = optab_for_tree_code (LSHIFT_EXPR, vectype, optab_scalar);
1856       optab2 = optab_for_tree_code (RSHIFT_EXPR, vectype, optab_scalar);
1857       if (!optab1
1858           || optab_handler (optab1, TYPE_MODE (vectype)) == CODE_FOR_nothing
1859           || !optab2
1860           || optab_handler (optab2, TYPE_MODE (vectype)) == CODE_FOR_nothing)
1861         return NULL;
1862     }
1863
1864   *type_in = vectype;
1865   *type_out = vectype;
1866   if (*type_in == NULL_TREE)
1867     return NULL;
1868
1869   if (dt == vect_external_def
1870       && TREE_CODE (oprnd1) == SSA_NAME
1871       && is_a <loop_vec_info> (vinfo))
1872     {
1873       struct loop *loop = as_a <loop_vec_info> (vinfo)->loop;
1874       ext_def = loop_preheader_edge (loop);
1875       if (!SSA_NAME_IS_DEFAULT_DEF (oprnd1))
1876         {
1877           basic_block bb = gimple_bb (SSA_NAME_DEF_STMT (oprnd1));
1878           if (bb == NULL
1879               || !dominated_by_p (CDI_DOMINATORS, ext_def->dest, bb))
1880             ext_def = NULL;
1881         }
1882     }
1883
1884   def = NULL_TREE;
1885   if (TREE_CODE (oprnd1) == INTEGER_CST
1886       || TYPE_MODE (TREE_TYPE (oprnd1)) == TYPE_MODE (type))
1887     def = oprnd1;
1888   else if (def_stmt && gimple_assign_cast_p (def_stmt))
1889     {
1890       tree rhs1 = gimple_assign_rhs1 (def_stmt);
1891       if (TYPE_MODE (TREE_TYPE (rhs1)) == TYPE_MODE (type)
1892           && TYPE_PRECISION (TREE_TYPE (rhs1))
1893              == TYPE_PRECISION (type))
1894         def = rhs1;
1895     }
1896
1897   STMT_VINFO_PATTERN_DEF_SEQ (stmt_vinfo) = NULL;
1898   if (def == NULL_TREE)
1899     {
1900       def = vect_recog_temp_ssa_var (type, NULL);
1901       def_stmt = gimple_build_assign (def, NOP_EXPR, oprnd1);
1902       if (ext_def)
1903         {
1904           basic_block new_bb
1905             = gsi_insert_on_edge_immediate (ext_def, def_stmt);
1906           gcc_assert (!new_bb);
1907         }
1908       else
1909         append_pattern_def_seq (stmt_vinfo, def_stmt);
1910     }
1911   stype = TREE_TYPE (def);
1912
1913   if (TREE_CODE (def) == INTEGER_CST)
1914     {
1915       if (!tree_fits_uhwi_p (def)
1916           || tree_to_uhwi (def) >= GET_MODE_PRECISION (TYPE_MODE (type))
1917           || integer_zerop (def))
1918         return NULL;
1919       def2 = build_int_cst (stype,
1920                             GET_MODE_PRECISION (TYPE_MODE (type))
1921                             - tree_to_uhwi (def));
1922     }
1923   else
1924     {
1925       tree vecstype = get_vectype_for_scalar_type (stype);
1926       stmt_vec_info def_stmt_vinfo;
1927
1928       if (vecstype == NULL_TREE)
1929         return NULL;
1930       def2 = vect_recog_temp_ssa_var (stype, NULL);
1931       def_stmt = gimple_build_assign (def2, NEGATE_EXPR, def);
1932       if (ext_def)
1933         {
1934           basic_block new_bb
1935             = gsi_insert_on_edge_immediate (ext_def, def_stmt);
1936           gcc_assert (!new_bb);
1937         }
1938       else
1939         {
1940           def_stmt_vinfo = new_stmt_vec_info (def_stmt, vinfo);
1941           set_vinfo_for_stmt (def_stmt, def_stmt_vinfo);
1942           STMT_VINFO_VECTYPE (def_stmt_vinfo) = vecstype;
1943           append_pattern_def_seq (stmt_vinfo, def_stmt);
1944         }
1945
1946       def2 = vect_recog_temp_ssa_var (stype, NULL);
1947       tree mask
1948         = build_int_cst (stype, GET_MODE_PRECISION (TYPE_MODE (stype)) - 1);
1949       def_stmt = gimple_build_assign (def2, BIT_AND_EXPR,
1950                                       gimple_assign_lhs (def_stmt), mask);
1951       if (ext_def)
1952         {
1953           basic_block new_bb
1954             = gsi_insert_on_edge_immediate (ext_def, def_stmt);
1955           gcc_assert (!new_bb);
1956         }
1957       else
1958         {
1959           def_stmt_vinfo = new_stmt_vec_info (def_stmt, vinfo);
1960           set_vinfo_for_stmt (def_stmt, def_stmt_vinfo);
1961           STMT_VINFO_VECTYPE (def_stmt_vinfo) = vecstype;
1962           append_pattern_def_seq (stmt_vinfo, def_stmt);
1963         }
1964     }
1965
1966   var1 = vect_recog_temp_ssa_var (type, NULL);
1967   def_stmt = gimple_build_assign (var1, rhs_code == LROTATE_EXPR
1968                                         ? LSHIFT_EXPR : RSHIFT_EXPR,
1969                                   oprnd0, def);
1970   append_pattern_def_seq (stmt_vinfo, def_stmt);
1971
1972   var2 = vect_recog_temp_ssa_var (type, NULL);
1973   def_stmt = gimple_build_assign (var2, rhs_code == LROTATE_EXPR
1974                                         ? RSHIFT_EXPR : LSHIFT_EXPR,
1975                                   oprnd0, def2);
1976   append_pattern_def_seq (stmt_vinfo, def_stmt);
1977
1978   /* Pattern detected.  */
1979   if (dump_enabled_p ())
1980     dump_printf_loc (MSG_NOTE, vect_location,
1981                      "vect_recog_rotate_pattern: detected:\n");
1982
1983   /* Pattern supported.  Create a stmt to be used to replace the pattern.  */
1984   var = vect_recog_temp_ssa_var (type, NULL);
1985   pattern_stmt = gimple_build_assign (var, BIT_IOR_EXPR, var1, var2);
1986
1987   if (dump_enabled_p ())
1988     dump_gimple_stmt_loc (MSG_NOTE, vect_location, TDF_SLIM, pattern_stmt, 0);
1989
1990   stmts->safe_push (last_stmt);
1991   return pattern_stmt;
1992 }
1993
1994 /* Detect a vector by vector shift pattern that wouldn't be otherwise
1995    vectorized:
1996
1997    type a_t;
1998    TYPE b_T, res_T;
1999
2000    S1 a_t = ;
2001    S2 b_T = ;
2002    S3 res_T = b_T op a_t;
2003
2004   where type 'TYPE' is a type with different size than 'type',
2005   and op is <<, >> or rotate.
2006
2007   Also detect cases:
2008
2009    type a_t;
2010    TYPE b_T, c_T, res_T;
2011
2012    S0 c_T = ;
2013    S1 a_t = (type) c_T;
2014    S2 b_T = ;
2015    S3 res_T = b_T op a_t;
2016
2017   Input/Output:
2018
2019   * STMTS: Contains a stmt from which the pattern search begins,
2020     i.e. the shift/rotate stmt.  The original stmt (S3) is replaced
2021     with a shift/rotate which has same type on both operands, in the
2022     second case just b_T op c_T, in the first case with added cast
2023     from a_t to c_T in STMT_VINFO_PATTERN_DEF_SEQ.
2024
2025   Output:
2026
2027   * TYPE_IN: The type of the input arguments to the pattern.
2028
2029   * TYPE_OUT: The type of the output of this pattern.
2030
2031   * Return value: A new stmt that will be used to replace the shift/rotate
2032     S3 stmt.  */
2033
2034 static gimple *
2035 vect_recog_vector_vector_shift_pattern (vec<gimple *> *stmts,
2036                                         tree *type_in, tree *type_out)
2037 {
2038   gimple *last_stmt = stmts->pop ();
2039   tree oprnd0, oprnd1, lhs, var;
2040   gimple *pattern_stmt, *def_stmt;
2041   enum tree_code rhs_code;
2042   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
2043   vec_info *vinfo = stmt_vinfo->vinfo;
2044   enum vect_def_type dt;
2045
2046   if (!is_gimple_assign (last_stmt))
2047     return NULL;
2048
2049   rhs_code = gimple_assign_rhs_code (last_stmt);
2050   switch (rhs_code)
2051     {
2052     case LSHIFT_EXPR:
2053     case RSHIFT_EXPR:
2054     case LROTATE_EXPR:
2055     case RROTATE_EXPR:
2056       break;
2057     default:
2058       return NULL;
2059     }
2060
2061   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
2062     return NULL;
2063
2064   lhs = gimple_assign_lhs (last_stmt);
2065   oprnd0 = gimple_assign_rhs1 (last_stmt);
2066   oprnd1 = gimple_assign_rhs2 (last_stmt);
2067   if (TREE_CODE (oprnd0) != SSA_NAME
2068       || TREE_CODE (oprnd1) != SSA_NAME
2069       || TYPE_MODE (TREE_TYPE (oprnd0)) == TYPE_MODE (TREE_TYPE (oprnd1))
2070       || TYPE_PRECISION (TREE_TYPE (oprnd1))
2071          != GET_MODE_PRECISION (TYPE_MODE (TREE_TYPE (oprnd1)))
2072       || TYPE_PRECISION (TREE_TYPE (lhs))
2073          != TYPE_PRECISION (TREE_TYPE (oprnd0)))
2074     return NULL;
2075
2076   if (!vect_is_simple_use (oprnd1, vinfo, &def_stmt, &dt))
2077     return NULL;
2078
2079   if (dt != vect_internal_def)
2080     return NULL;
2081
2082   *type_in = get_vectype_for_scalar_type (TREE_TYPE (oprnd0));
2083   *type_out = *type_in;
2084   if (*type_in == NULL_TREE)
2085     return NULL;
2086
2087   tree def = NULL_TREE;
2088   stmt_vec_info def_vinfo = vinfo_for_stmt (def_stmt);
2089   if (!STMT_VINFO_IN_PATTERN_P (def_vinfo) && gimple_assign_cast_p (def_stmt))
2090     {
2091       tree rhs1 = gimple_assign_rhs1 (def_stmt);
2092       if (TYPE_MODE (TREE_TYPE (rhs1)) == TYPE_MODE (TREE_TYPE (oprnd0))
2093           && TYPE_PRECISION (TREE_TYPE (rhs1))
2094              == TYPE_PRECISION (TREE_TYPE (oprnd0)))
2095         {
2096           if (TYPE_PRECISION (TREE_TYPE (oprnd1))
2097               >= TYPE_PRECISION (TREE_TYPE (rhs1)))
2098             def = rhs1;
2099           else
2100             {
2101               tree mask
2102                 = build_low_bits_mask (TREE_TYPE (rhs1),
2103                                        TYPE_PRECISION (TREE_TYPE (oprnd1)));
2104               def = vect_recog_temp_ssa_var (TREE_TYPE (rhs1), NULL);
2105               def_stmt = gimple_build_assign (def, BIT_AND_EXPR, rhs1, mask);
2106               new_pattern_def_seq (stmt_vinfo, def_stmt);
2107             }
2108         }
2109     }
2110
2111   if (def == NULL_TREE)
2112     {
2113       def = vect_recog_temp_ssa_var (TREE_TYPE (oprnd0), NULL);
2114       def_stmt = gimple_build_assign (def, NOP_EXPR, oprnd1);
2115       new_pattern_def_seq (stmt_vinfo, def_stmt);
2116     }
2117
2118   /* Pattern detected.  */
2119   if (dump_enabled_p ())
2120     dump_printf_loc (MSG_NOTE, vect_location,
2121                      "vect_recog_vector_vector_shift_pattern: detected:\n");
2122
2123   /* Pattern supported.  Create a stmt to be used to replace the pattern.  */
2124   var = vect_recog_temp_ssa_var (TREE_TYPE (oprnd0), NULL);
2125   pattern_stmt = gimple_build_assign (var, rhs_code, oprnd0, def);
2126
2127   if (dump_enabled_p ())
2128     dump_gimple_stmt_loc (MSG_NOTE, vect_location, TDF_SLIM, pattern_stmt, 0);
2129
2130   stmts->safe_push (last_stmt);
2131   return pattern_stmt;
2132 }
2133
2134 /* Detect multiplication by constant which are postive or negatives of power 2,
2135    and convert them to shift patterns.
2136
2137    Mult with constants that are postive power of two.
2138    type a_t;
2139    type b_t
2140    S1: b_t = a_t * n
2141
2142    or
2143
2144    Mult with constants that are negative power of two.
2145    S2: b_t = a_t * -n
2146
2147    Input/Output:
2148
2149    STMTS: Contains a stmt from which the pattern search begins,
2150    i.e. the mult stmt.  Convert the mult operation to LSHIFT if
2151    constant operand is a power of 2.
2152    type a_t, b_t
2153    S1': b_t = a_t << log2 (n)
2154
2155    Convert the mult operation to LSHIFT and followed by a NEGATE
2156    if constant operand is a negative power of 2.
2157    type a_t, b_t, res_T;
2158    S2': b_t = a_t << log2 (n)
2159    S3': res_T  = - (b_t)
2160
2161  Output:
2162
2163   * TYPE_IN: The type of the input arguments to the pattern.
2164
2165   * TYPE_OUT: The type of the output of this pattern.
2166
2167   * Return value: A new stmt that will be used to replace the multiplication
2168     S1 or S2 stmt.  */
2169
2170 static gimple *
2171 vect_recog_mult_pattern (vec<gimple *> *stmts,
2172                          tree *type_in, tree *type_out)
2173 {
2174   gimple *last_stmt = stmts->pop ();
2175   tree oprnd0, oprnd1, vectype, itype;
2176   gimple *pattern_stmt, *def_stmt;
2177   optab optab;
2178   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
2179   int power2_val, power2_neg_val;
2180   tree shift;
2181
2182   if (!is_gimple_assign (last_stmt))
2183     return NULL;
2184
2185   if (gimple_assign_rhs_code (last_stmt) != MULT_EXPR)
2186     return NULL;
2187
2188   oprnd0 = gimple_assign_rhs1 (last_stmt);
2189   oprnd1 = gimple_assign_rhs2 (last_stmt);
2190   itype = TREE_TYPE (oprnd0);
2191
2192   if (TREE_CODE (oprnd0) != SSA_NAME
2193       || TREE_CODE (oprnd1) != INTEGER_CST
2194       || !INTEGRAL_TYPE_P (itype)
2195       || TYPE_PRECISION (itype) != GET_MODE_PRECISION (TYPE_MODE (itype)))
2196     return NULL;
2197
2198   vectype = get_vectype_for_scalar_type (itype);
2199   if (vectype == NULL_TREE)
2200     return NULL;
2201
2202   /* If the target can handle vectorized multiplication natively,
2203      don't attempt to optimize this.  */
2204   optab = optab_for_tree_code (MULT_EXPR, vectype, optab_default);
2205   if (optab != unknown_optab)
2206     {
2207       machine_mode vec_mode = TYPE_MODE (vectype);
2208       int icode = (int) optab_handler (optab, vec_mode);
2209       if (icode != CODE_FOR_nothing)
2210         return NULL;
2211     }
2212
2213   /* If target cannot handle vector left shift then we cannot
2214      optimize and bail out.  */
2215   optab = optab_for_tree_code (LSHIFT_EXPR, vectype, optab_vector);
2216   if (!optab
2217       || optab_handler (optab, TYPE_MODE (vectype)) == CODE_FOR_nothing)
2218     return NULL;
2219
2220   power2_val = wi::exact_log2 (oprnd1);
2221   power2_neg_val = wi::exact_log2 (wi::neg (oprnd1));
2222
2223   /* Handle constant operands that are postive or negative powers of 2.  */
2224   if (power2_val != -1)
2225     {
2226       shift = build_int_cst (itype, power2_val);
2227       pattern_stmt
2228         = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
2229                                LSHIFT_EXPR, oprnd0, shift);
2230     }
2231   else if (power2_neg_val != -1)
2232     {
2233       /* If the target cannot handle vector NEGATE then we cannot
2234          do the optimization.  */
2235       optab = optab_for_tree_code (NEGATE_EXPR, vectype, optab_vector);
2236       if (!optab
2237           || optab_handler (optab, TYPE_MODE (vectype)) == CODE_FOR_nothing)
2238         return NULL;
2239
2240       shift = build_int_cst (itype, power2_neg_val);
2241       def_stmt
2242         = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
2243                                LSHIFT_EXPR, oprnd0, shift);
2244       new_pattern_def_seq (stmt_vinfo, def_stmt);
2245       pattern_stmt
2246          = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
2247                                 NEGATE_EXPR, gimple_assign_lhs (def_stmt));
2248     }
2249   else
2250     return NULL;
2251
2252   /* Pattern detected.  */
2253   if (dump_enabled_p ())
2254     dump_printf_loc (MSG_NOTE, vect_location,
2255                      "vect_recog_mult_pattern: detected:\n");
2256
2257   if (dump_enabled_p ())
2258     dump_gimple_stmt_loc (MSG_NOTE, vect_location, TDF_SLIM,
2259                           pattern_stmt,0);
2260
2261   stmts->safe_push (last_stmt);
2262   *type_in = vectype;
2263   *type_out = vectype;
2264
2265   return pattern_stmt;
2266 }
2267
2268 /* Detect a signed division by a constant that wouldn't be
2269    otherwise vectorized:
2270
2271    type a_t, b_t;
2272
2273    S1 a_t = b_t / N;
2274
2275   where type 'type' is an integral type and N is a constant.
2276
2277   Similarly handle modulo by a constant:
2278
2279    S4 a_t = b_t % N;
2280
2281   Input/Output:
2282
2283   * STMTS: Contains a stmt from which the pattern search begins,
2284     i.e. the division stmt.  S1 is replaced by if N is a power
2285     of two constant and type is signed:
2286   S3  y_t = b_t < 0 ? N - 1 : 0;
2287   S2  x_t = b_t + y_t;
2288   S1' a_t = x_t >> log2 (N);
2289
2290     S4 is replaced if N is a power of two constant and
2291     type is signed by (where *_T temporaries have unsigned type):
2292   S9  y_T = b_t < 0 ? -1U : 0U;
2293   S8  z_T = y_T >> (sizeof (type_t) * CHAR_BIT - log2 (N));
2294   S7  z_t = (type) z_T;
2295   S6  w_t = b_t + z_t;
2296   S5  x_t = w_t & (N - 1);
2297   S4' a_t = x_t - z_t;
2298
2299   Output:
2300
2301   * TYPE_IN: The type of the input arguments to the pattern.
2302
2303   * TYPE_OUT: The type of the output of this pattern.
2304
2305   * Return value: A new stmt that will be used to replace the division
2306     S1 or modulo S4 stmt.  */
2307
2308 static gimple *
2309 vect_recog_divmod_pattern (vec<gimple *> *stmts,
2310                            tree *type_in, tree *type_out)
2311 {
2312   gimple *last_stmt = stmts->pop ();
2313   tree oprnd0, oprnd1, vectype, itype, cond;
2314   gimple *pattern_stmt, *def_stmt;
2315   enum tree_code rhs_code;
2316   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
2317   vec_info *vinfo = stmt_vinfo->vinfo;
2318   optab optab;
2319   tree q;
2320   int dummy_int, prec;
2321   stmt_vec_info def_stmt_vinfo;
2322
2323   if (!is_gimple_assign (last_stmt))
2324     return NULL;
2325
2326   rhs_code = gimple_assign_rhs_code (last_stmt);
2327   switch (rhs_code)
2328     {
2329     case TRUNC_DIV_EXPR:
2330     case TRUNC_MOD_EXPR:
2331       break;
2332     default:
2333       return NULL;
2334     }
2335
2336   if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
2337     return NULL;
2338
2339   oprnd0 = gimple_assign_rhs1 (last_stmt);
2340   oprnd1 = gimple_assign_rhs2 (last_stmt);
2341   itype = TREE_TYPE (oprnd0);
2342   if (TREE_CODE (oprnd0) != SSA_NAME
2343       || TREE_CODE (oprnd1) != INTEGER_CST
2344       || TREE_CODE (itype) != INTEGER_TYPE
2345       || TYPE_PRECISION (itype) != GET_MODE_PRECISION (TYPE_MODE (itype)))
2346     return NULL;
2347
2348   vectype = get_vectype_for_scalar_type (itype);
2349   if (vectype == NULL_TREE)
2350     return NULL;
2351
2352   /* If the target can handle vectorized division or modulo natively,
2353      don't attempt to optimize this.  */
2354   optab = optab_for_tree_code (rhs_code, vectype, optab_default);
2355   if (optab != unknown_optab)
2356     {
2357       machine_mode vec_mode = TYPE_MODE (vectype);
2358       int icode = (int) optab_handler (optab, vec_mode);
2359       if (icode != CODE_FOR_nothing)
2360         return NULL;
2361     }
2362
2363   prec = TYPE_PRECISION (itype);
2364   if (integer_pow2p (oprnd1))
2365     {
2366       if (TYPE_UNSIGNED (itype) || tree_int_cst_sgn (oprnd1) != 1)
2367         return NULL;
2368
2369       /* Pattern detected.  */
2370       if (dump_enabled_p ())
2371         dump_printf_loc (MSG_NOTE, vect_location,
2372                          "vect_recog_divmod_pattern: detected:\n");
2373
2374       cond = build2 (LT_EXPR, boolean_type_node, oprnd0,
2375                      build_int_cst (itype, 0));
2376       if (rhs_code == TRUNC_DIV_EXPR)
2377         {
2378           tree var = vect_recog_temp_ssa_var (itype, NULL);
2379           tree shift;
2380           def_stmt
2381             = gimple_build_assign (var, COND_EXPR, cond,
2382                                    fold_build2 (MINUS_EXPR, itype, oprnd1,
2383                                                 build_int_cst (itype, 1)),
2384                                    build_int_cst (itype, 0));
2385           new_pattern_def_seq (stmt_vinfo, def_stmt);
2386           var = vect_recog_temp_ssa_var (itype, NULL);
2387           def_stmt
2388             = gimple_build_assign (var, PLUS_EXPR, oprnd0,
2389                                    gimple_assign_lhs (def_stmt));
2390           append_pattern_def_seq (stmt_vinfo, def_stmt);
2391
2392           shift = build_int_cst (itype, tree_log2 (oprnd1));
2393           pattern_stmt
2394             = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
2395                                    RSHIFT_EXPR, var, shift);
2396         }
2397       else
2398         {
2399           tree signmask;
2400           STMT_VINFO_PATTERN_DEF_SEQ (stmt_vinfo) = NULL;
2401           if (compare_tree_int (oprnd1, 2) == 0)
2402             {
2403               signmask = vect_recog_temp_ssa_var (itype, NULL);
2404               def_stmt = gimple_build_assign (signmask, COND_EXPR, cond,
2405                                               build_int_cst (itype, 1),
2406                                               build_int_cst (itype, 0));
2407               append_pattern_def_seq (stmt_vinfo, def_stmt);
2408             }
2409           else
2410             {
2411               tree utype
2412                 = build_nonstandard_integer_type (prec, 1);
2413               tree vecutype = get_vectype_for_scalar_type (utype);
2414               tree shift
2415                 = build_int_cst (utype, GET_MODE_BITSIZE (TYPE_MODE (itype))
2416                                         - tree_log2 (oprnd1));
2417               tree var = vect_recog_temp_ssa_var (utype, NULL);
2418
2419               def_stmt = gimple_build_assign (var, COND_EXPR, cond,
2420                                               build_int_cst (utype, -1),
2421                                               build_int_cst (utype, 0));
2422               def_stmt_vinfo = new_stmt_vec_info (def_stmt, vinfo);
2423               set_vinfo_for_stmt (def_stmt, def_stmt_vinfo);
2424               STMT_VINFO_VECTYPE (def_stmt_vinfo) = vecutype;
2425               append_pattern_def_seq (stmt_vinfo, def_stmt);
2426               var = vect_recog_temp_ssa_var (utype, NULL);
2427               def_stmt = gimple_build_assign (var, RSHIFT_EXPR,
2428                                               gimple_assign_lhs (def_stmt),
2429                                               shift);
2430               def_stmt_vinfo = new_stmt_vec_info (def_stmt, vinfo);
2431               set_vinfo_for_stmt (def_stmt, def_stmt_vinfo);
2432               STMT_VINFO_VECTYPE (def_stmt_vinfo) = vecutype;
2433               append_pattern_def_seq (stmt_vinfo, def_stmt);
2434               signmask = vect_recog_temp_ssa_var (itype, NULL);
2435               def_stmt
2436                 = gimple_build_assign (signmask, NOP_EXPR, var);
2437               append_pattern_def_seq (stmt_vinfo, def_stmt);
2438             }
2439           def_stmt
2440             = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
2441                                    PLUS_EXPR, oprnd0, signmask);
2442           append_pattern_def_seq (stmt_vinfo, def_stmt);
2443           def_stmt
2444             = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
2445                                    BIT_AND_EXPR, gimple_assign_lhs (def_stmt),
2446                                    fold_build2 (MINUS_EXPR, itype, oprnd1,
2447                                                 build_int_cst (itype, 1)));
2448           append_pattern_def_seq (stmt_vinfo, def_stmt);
2449
2450           pattern_stmt
2451             = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
2452                                    MINUS_EXPR, gimple_assign_lhs (def_stmt),
2453                                    signmask);
2454         }
2455
2456       if (dump_enabled_p ())
2457         dump_gimple_stmt_loc (MSG_NOTE, vect_location, TDF_SLIM, pattern_stmt,
2458                               0);
2459
2460       stmts->safe_push (last_stmt);
2461
2462       *type_in = vectype;
2463       *type_out = vectype;
2464       return pattern_stmt;
2465     }
2466
2467   if (prec > HOST_BITS_PER_WIDE_INT
2468       || integer_zerop (oprnd1))
2469     return NULL;
2470
2471   if (!can_mult_highpart_p (TYPE_MODE (vectype), TYPE_UNSIGNED (itype)))
2472     return NULL;
2473
2474   STMT_VINFO_PATTERN_DEF_SEQ (stmt_vinfo) = NULL;
2475
2476   if (TYPE_UNSIGNED (itype))
2477     {
2478       unsigned HOST_WIDE_INT mh, ml;
2479       int pre_shift, post_shift;
2480       unsigned HOST_WIDE_INT d = (TREE_INT_CST_LOW (oprnd1)
2481                                   & GET_MODE_MASK (TYPE_MODE (itype)));
2482       tree t1, t2, t3, t4;
2483
2484       if (d >= ((unsigned HOST_WIDE_INT) 1 << (prec - 1)))
2485         /* FIXME: Can transform this into oprnd0 >= oprnd1 ? 1 : 0.  */
2486         return NULL;
2487
2488       /* Find a suitable multiplier and right shift count
2489          instead of multiplying with D.  */
2490       mh = choose_multiplier (d, prec, prec, &ml, &post_shift, &dummy_int);
2491
2492       /* If the suggested multiplier is more than SIZE bits, we can do better
2493          for even divisors, using an initial right shift.  */
2494       if (mh != 0 && (d & 1) == 0)
2495         {
2496           pre_shift = floor_log2 (d & -d);
2497           mh = choose_multiplier (d >> pre_shift, prec, prec - pre_shift,
2498                                   &ml, &post_shift, &dummy_int);
2499           gcc_assert (!mh);
2500         }
2501       else
2502         pre_shift = 0;
2503
2504       if (mh != 0)
2505         {
2506           if (post_shift - 1 >= prec)
2507             return NULL;
2508
2509           /* t1 = oprnd0 h* ml;
2510              t2 = oprnd0 - t1;
2511              t3 = t2 >> 1;
2512              t4 = t1 + t3;
2513              q = t4 >> (post_shift - 1);  */
2514           t1 = vect_recog_temp_ssa_var (itype, NULL);
2515           def_stmt = gimple_build_assign (t1, MULT_HIGHPART_EXPR, oprnd0,
2516                                           build_int_cst (itype, ml));
2517           append_pattern_def_seq (stmt_vinfo, def_stmt);
2518
2519           t2 = vect_recog_temp_ssa_var (itype, NULL);
2520           def_stmt
2521             = gimple_build_assign (t2, MINUS_EXPR, oprnd0, t1);
2522           append_pattern_def_seq (stmt_vinfo, def_stmt);
2523
2524           t3 = vect_recog_temp_ssa_var (itype, NULL);
2525           def_stmt
2526             = gimple_build_assign (t3, RSHIFT_EXPR, t2, integer_one_node);
2527           append_pattern_def_seq (stmt_vinfo, def_stmt);
2528
2529           t4 = vect_recog_temp_ssa_var (itype, NULL);
2530           def_stmt
2531             = gimple_build_assign (t4, PLUS_EXPR, t1, t3);
2532
2533           if (post_shift != 1)
2534             {
2535               append_pattern_def_seq (stmt_vinfo, def_stmt);
2536
2537               q = vect_recog_temp_ssa_var (itype, NULL);
2538               pattern_stmt
2539                 = gimple_build_assign (q, RSHIFT_EXPR, t4,
2540                                        build_int_cst (itype, post_shift - 1));
2541             }
2542           else
2543             {
2544               q = t4;
2545               pattern_stmt = def_stmt;
2546             }
2547         }
2548       else
2549         {
2550           if (pre_shift >= prec || post_shift >= prec)
2551             return NULL;
2552
2553           /* t1 = oprnd0 >> pre_shift;
2554              t2 = t1 h* ml;
2555              q = t2 >> post_shift;  */
2556           if (pre_shift)
2557             {
2558               t1 = vect_recog_temp_ssa_var (itype, NULL);
2559               def_stmt
2560                 = gimple_build_assign (t1, RSHIFT_EXPR, oprnd0,
2561                                        build_int_cst (NULL, pre_shift));
2562               append_pattern_def_seq (stmt_vinfo, def_stmt);
2563             }
2564           else
2565             t1 = oprnd0;
2566
2567           t2 = vect_recog_temp_ssa_var (itype, NULL);
2568           def_stmt = gimple_build_assign (t2, MULT_HIGHPART_EXPR, t1,
2569                                           build_int_cst (itype, ml));
2570
2571           if (post_shift)
2572             {
2573               append_pattern_def_seq (stmt_vinfo, def_stmt);
2574
2575               q = vect_recog_temp_ssa_var (itype, NULL);
2576               def_stmt
2577                 = gimple_build_assign (q, RSHIFT_EXPR, t2,
2578                                        build_int_cst (itype, post_shift));
2579             }
2580           else
2581             q = t2;
2582
2583           pattern_stmt = def_stmt;
2584         }
2585     }
2586   else
2587     {
2588       unsigned HOST_WIDE_INT ml;
2589       int post_shift;
2590       HOST_WIDE_INT d = TREE_INT_CST_LOW (oprnd1);
2591       unsigned HOST_WIDE_INT abs_d;
2592       bool add = false;
2593       tree t1, t2, t3, t4;
2594
2595       /* Give up for -1.  */
2596       if (d == -1)
2597         return NULL;
2598
2599       /* Since d might be INT_MIN, we have to cast to
2600          unsigned HOST_WIDE_INT before negating to avoid
2601          undefined signed overflow.  */
2602       abs_d = (d >= 0
2603                ? (unsigned HOST_WIDE_INT) d
2604                : - (unsigned HOST_WIDE_INT) d);
2605
2606       /* n rem d = n rem -d */
2607       if (rhs_code == TRUNC_MOD_EXPR && d < 0)
2608         {
2609           d = abs_d;
2610           oprnd1 = build_int_cst (itype, abs_d);
2611         }
2612       else if (HOST_BITS_PER_WIDE_INT >= prec
2613                && abs_d == (unsigned HOST_WIDE_INT) 1 << (prec - 1))
2614         /* This case is not handled correctly below.  */
2615         return NULL;
2616
2617       choose_multiplier (abs_d, prec, prec - 1, &ml, &post_shift, &dummy_int);
2618       if (ml >= (unsigned HOST_WIDE_INT) 1 << (prec - 1))
2619         {
2620           add = true;
2621           ml |= (~(unsigned HOST_WIDE_INT) 0) << (prec - 1);
2622         }
2623       if (post_shift >= prec)
2624         return NULL;
2625
2626       /* t1 = oprnd0 h* ml;  */
2627       t1 = vect_recog_temp_ssa_var (itype, NULL);
2628       def_stmt = gimple_build_assign (t1, MULT_HIGHPART_EXPR, oprnd0,
2629                                       build_int_cst (itype, ml));
2630
2631       if (add)
2632         {
2633           /* t2 = t1 + oprnd0;  */
2634           append_pattern_def_seq (stmt_vinfo, def_stmt);
2635           t2 = vect_recog_temp_ssa_var (itype, NULL);
2636           def_stmt = gimple_build_assign (t2, PLUS_EXPR, t1, oprnd0);
2637         }
2638       else
2639         t2 = t1;
2640
2641       if (post_shift)
2642         {
2643           /* t3 = t2 >> post_shift;  */
2644           append_pattern_def_seq (stmt_vinfo, def_stmt);
2645           t3 = vect_recog_temp_ssa_var (itype, NULL);
2646           def_stmt = gimple_build_assign (t3, RSHIFT_EXPR, t2,
2647                                           build_int_cst (itype, post_shift));
2648         }
2649       else
2650         t3 = t2;
2651
2652       wide_int oprnd0_min, oprnd0_max;
2653       int msb = 1;
2654       if (get_range_info (oprnd0, &oprnd0_min, &oprnd0_max) == VR_RANGE)
2655         {
2656           if (!wi::neg_p (oprnd0_min, TYPE_SIGN (itype)))
2657             msb = 0;
2658           else if (wi::neg_p (oprnd0_max, TYPE_SIGN (itype)))
2659             msb = -1;
2660         }
2661
2662       if (msb == 0 && d >= 0)
2663         {
2664           /* q = t3;  */
2665           q = t3;
2666           pattern_stmt = def_stmt;
2667         }
2668       else
2669         {
2670           /* t4 = oprnd0 >> (prec - 1);
2671              or if we know from VRP that oprnd0 >= 0
2672              t4 = 0;
2673              or if we know from VRP that oprnd0 < 0
2674              t4 = -1;  */
2675           append_pattern_def_seq (stmt_vinfo, def_stmt);
2676           t4 = vect_recog_temp_ssa_var (itype, NULL);
2677           if (msb != 1)
2678             def_stmt = gimple_build_assign (t4, INTEGER_CST,
2679                                             build_int_cst (itype, msb));
2680           else
2681             def_stmt = gimple_build_assign (t4, RSHIFT_EXPR, oprnd0,
2682                                             build_int_cst (itype, prec - 1));
2683           append_pattern_def_seq (stmt_vinfo, def_stmt);
2684
2685           /* q = t3 - t4;  or q = t4 - t3;  */
2686           q = vect_recog_temp_ssa_var (itype, NULL);
2687           pattern_stmt = gimple_build_assign (q, MINUS_EXPR, d < 0 ? t4 : t3,
2688                                               d < 0 ? t3 : t4);
2689         }
2690     }
2691
2692   if (rhs_code == TRUNC_MOD_EXPR)
2693     {
2694       tree r, t1;
2695
2696       /* We divided.  Now finish by:
2697          t1 = q * oprnd1;
2698          r = oprnd0 - t1;  */
2699       append_pattern_def_seq (stmt_vinfo, pattern_stmt);
2700
2701       t1 = vect_recog_temp_ssa_var (itype, NULL);
2702       def_stmt = gimple_build_assign (t1, MULT_EXPR, q, oprnd1);
2703       append_pattern_def_seq (stmt_vinfo, def_stmt);
2704
2705       r = vect_recog_temp_ssa_var (itype, NULL);
2706       pattern_stmt = gimple_build_assign (r, MINUS_EXPR, oprnd0, t1);
2707     }
2708
2709   /* Pattern detected.  */
2710   if (dump_enabled_p ())
2711     {
2712       dump_printf_loc (MSG_NOTE, vect_location,
2713                        "vect_recog_divmod_pattern: detected: ");
2714       dump_gimple_stmt (MSG_NOTE, TDF_SLIM, pattern_stmt, 0);
2715     }
2716
2717   stmts->safe_push (last_stmt);
2718
2719   *type_in = vectype;
2720   *type_out = vectype;
2721   return pattern_stmt;
2722 }
2723
2724 /* Function vect_recog_mixed_size_cond_pattern
2725
2726    Try to find the following pattern:
2727
2728      type x_t, y_t;
2729      TYPE a_T, b_T, c_T;
2730    loop:
2731      S1  a_T = x_t CMP y_t ? b_T : c_T;
2732
2733    where type 'TYPE' is an integral type which has different size
2734    from 'type'.  b_T and c_T are either constants (and if 'TYPE' is wider
2735    than 'type', the constants need to fit into an integer type
2736    with the same width as 'type') or results of conversion from 'type'.
2737
2738    Input:
2739
2740    * LAST_STMT: A stmt from which the pattern search begins.
2741
2742    Output:
2743
2744    * TYPE_IN: The type of the input arguments to the pattern.
2745
2746    * TYPE_OUT: The type of the output of this pattern.
2747
2748    * Return value: A new stmt that will be used to replace the pattern.
2749         Additionally a def_stmt is added.
2750
2751         a_it = x_t CMP y_t ? b_it : c_it;
2752         a_T = (TYPE) a_it;  */
2753
2754 static gimple *
2755 vect_recog_mixed_size_cond_pattern (vec<gimple *> *stmts, tree *type_in,
2756                                     tree *type_out)
2757 {
2758   gimple *last_stmt = (*stmts)[0];
2759   tree cond_expr, then_clause, else_clause;
2760   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt), def_stmt_info;
2761   tree type, vectype, comp_vectype, itype = NULL_TREE, vecitype;
2762   gimple *pattern_stmt, *def_stmt;
2763   vec_info *vinfo = stmt_vinfo->vinfo;
2764   tree orig_type0 = NULL_TREE, orig_type1 = NULL_TREE;
2765   gimple *def_stmt0 = NULL, *def_stmt1 = NULL;
2766   bool promotion;
2767   tree comp_scalar_type;
2768
2769   if (!is_gimple_assign (last_stmt)
2770       || gimple_assign_rhs_code (last_stmt) != COND_EXPR
2771       || STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_internal_def)
2772     return NULL;
2773
2774   cond_expr = gimple_assign_rhs1 (last_stmt);
2775   then_clause = gimple_assign_rhs2 (last_stmt);
2776   else_clause = gimple_assign_rhs3 (last_stmt);
2777
2778   if (!COMPARISON_CLASS_P (cond_expr))
2779     return NULL;
2780
2781   comp_scalar_type = TREE_TYPE (TREE_OPERAND (cond_expr, 0));
2782   comp_vectype = get_vectype_for_scalar_type (comp_scalar_type);
2783   if (comp_vectype == NULL_TREE)
2784     return NULL;
2785
2786   type = gimple_expr_type (last_stmt);
2787   if (types_compatible_p (type, comp_scalar_type)
2788       || ((TREE_CODE (then_clause) != INTEGER_CST
2789            || TREE_CODE (else_clause) != INTEGER_CST)
2790           && !INTEGRAL_TYPE_P (comp_scalar_type))
2791       || !INTEGRAL_TYPE_P (type))
2792     return NULL;
2793
2794   if ((TREE_CODE (then_clause) != INTEGER_CST
2795        && !type_conversion_p (then_clause, last_stmt, false, &orig_type0,
2796                               &def_stmt0, &promotion))
2797       || (TREE_CODE (else_clause) != INTEGER_CST
2798           && !type_conversion_p (else_clause, last_stmt, false, &orig_type1,
2799                                  &def_stmt1, &promotion)))
2800     return NULL;
2801
2802   if (orig_type0 && orig_type1
2803       && !types_compatible_p (orig_type0, orig_type1))
2804     return NULL;
2805
2806   if (orig_type0)
2807     {
2808       if (!types_compatible_p (orig_type0, comp_scalar_type))
2809         return NULL;
2810       then_clause = gimple_assign_rhs1 (def_stmt0);
2811       itype = orig_type0;
2812     }
2813
2814   if (orig_type1)
2815     {
2816       if (!types_compatible_p (orig_type1, comp_scalar_type))
2817         return NULL;
2818       else_clause = gimple_assign_rhs1 (def_stmt1);
2819       itype = orig_type1;
2820     }
2821
2822
2823   HOST_WIDE_INT cmp_mode_size
2824     = GET_MODE_UNIT_BITSIZE (TYPE_MODE (comp_vectype));
2825
2826   if (GET_MODE_BITSIZE (TYPE_MODE (type)) == cmp_mode_size)
2827     return NULL;
2828
2829   vectype = get_vectype_for_scalar_type (type);
2830   if (vectype == NULL_TREE)
2831     return NULL;
2832
2833   if (expand_vec_cond_expr_p (vectype, comp_vectype))
2834     return NULL;
2835
2836   if (itype == NULL_TREE)
2837     itype = build_nonstandard_integer_type (cmp_mode_size,
2838                                             TYPE_UNSIGNED (type));
2839
2840   if (itype == NULL_TREE
2841       || GET_MODE_BITSIZE (TYPE_MODE (itype)) != cmp_mode_size)
2842     return NULL;
2843
2844   vecitype = get_vectype_for_scalar_type (itype);
2845   if (vecitype == NULL_TREE)
2846     return NULL;
2847
2848   if (!expand_vec_cond_expr_p (vecitype, comp_vectype))
2849     return NULL;
2850
2851   if (GET_MODE_BITSIZE (TYPE_MODE (type)) > cmp_mode_size)
2852     {
2853       if ((TREE_CODE (then_clause) == INTEGER_CST
2854            && !int_fits_type_p (then_clause, itype))
2855           || (TREE_CODE (else_clause) == INTEGER_CST
2856               && !int_fits_type_p (else_clause, itype)))
2857         return NULL;
2858     }
2859
2860   def_stmt = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
2861                                   COND_EXPR, unshare_expr (cond_expr),
2862                                   fold_convert (itype, then_clause),
2863                                   fold_convert (itype, else_clause));
2864   pattern_stmt = gimple_build_assign (vect_recog_temp_ssa_var (type, NULL),
2865                                       NOP_EXPR, gimple_assign_lhs (def_stmt));
2866
2867   new_pattern_def_seq (stmt_vinfo, def_stmt);
2868   def_stmt_info = new_stmt_vec_info (def_stmt, vinfo);
2869   set_vinfo_for_stmt (def_stmt, def_stmt_info);
2870   STMT_VINFO_VECTYPE (def_stmt_info) = vecitype;
2871   *type_in = vecitype;
2872   *type_out = vectype;
2873
2874   if (dump_enabled_p ())
2875     dump_printf_loc (MSG_NOTE, vect_location,
2876                      "vect_recog_mixed_size_cond_pattern: detected:\n");
2877
2878   return pattern_stmt;
2879 }
2880
2881
2882 /* Helper function of vect_recog_bool_pattern.  Called recursively, return
2883    true if bool VAR can and should be optimized that way.  Assume it shouldn't
2884    in case it's a result of a comparison which can be directly vectorized into
2885    a vector comparison.  Fills in STMTS with all stmts visited during the
2886    walk.  */
2887
2888 static bool
2889 check_bool_pattern (tree var, vec_info *vinfo, hash_set<gimple *> &stmts)
2890 {
2891   gimple *def_stmt;
2892   enum vect_def_type dt;
2893   tree rhs1;
2894   enum tree_code rhs_code;
2895
2896   if (!vect_is_simple_use (var, vinfo, &def_stmt, &dt))
2897     return false;
2898
2899   if (dt != vect_internal_def)
2900     return false;
2901
2902   if (!is_gimple_assign (def_stmt))
2903     return false;
2904
2905   if (stmts.contains (def_stmt))
2906     return true;
2907
2908   rhs1 = gimple_assign_rhs1 (def_stmt);
2909   rhs_code = gimple_assign_rhs_code (def_stmt);
2910   switch (rhs_code)
2911     {
2912     case SSA_NAME:
2913       if (! check_bool_pattern (rhs1, vinfo, stmts))
2914         return false;
2915       break;
2916
2917     CASE_CONVERT:
2918       if ((TYPE_PRECISION (TREE_TYPE (rhs1)) != 1
2919            || !TYPE_UNSIGNED (TREE_TYPE (rhs1)))
2920           && TREE_CODE (TREE_TYPE (rhs1)) != BOOLEAN_TYPE)
2921         return false;
2922       if (! check_bool_pattern (rhs1, vinfo, stmts))
2923         return false;
2924       break;
2925
2926     case BIT_NOT_EXPR:
2927       if (! check_bool_pattern (rhs1, vinfo, stmts))
2928         return false;
2929       break;
2930
2931     case BIT_AND_EXPR:
2932     case BIT_IOR_EXPR:
2933     case BIT_XOR_EXPR:
2934       if (! check_bool_pattern (rhs1, vinfo, stmts)
2935           || ! check_bool_pattern (gimple_assign_rhs2 (def_stmt), vinfo, stmts))
2936         return false;
2937       break;
2938
2939     default:
2940       if (TREE_CODE_CLASS (rhs_code) == tcc_comparison)
2941         {
2942           tree vecitype, comp_vectype;
2943
2944           /* If the comparison can throw, then is_gimple_condexpr will be
2945              false and we can't make a COND_EXPR/VEC_COND_EXPR out of it.  */
2946           if (stmt_could_throw_p (def_stmt))
2947             return false;
2948
2949           comp_vectype = get_vectype_for_scalar_type (TREE_TYPE (rhs1));
2950           if (comp_vectype == NULL_TREE)
2951             return false;
2952
2953           tree mask_type = get_mask_type_for_scalar_type (TREE_TYPE (rhs1));
2954           if (mask_type
2955               && expand_vec_cmp_expr_p (comp_vectype, mask_type))
2956             return false;
2957
2958           if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE)
2959             {
2960               machine_mode mode = TYPE_MODE (TREE_TYPE (rhs1));
2961               tree itype
2962                 = build_nonstandard_integer_type (GET_MODE_BITSIZE (mode), 1);
2963               vecitype = get_vectype_for_scalar_type (itype);
2964               if (vecitype == NULL_TREE)
2965                 return false;
2966             }
2967           else
2968             vecitype = comp_vectype;
2969           if (! expand_vec_cond_expr_p (vecitype, comp_vectype))
2970             return false;
2971         }
2972       else
2973         return false;
2974       break;
2975     }
2976
2977   bool res = stmts.add (def_stmt);
2978   /* We can't end up recursing when just visiting SSA defs but not PHIs.  */
2979   gcc_assert (!res);
2980
2981   return true;
2982 }
2983
2984
2985 /* Helper function of adjust_bool_pattern.  Add a cast to TYPE to a previous
2986    stmt (SSA_NAME_DEF_STMT of VAR) adding a cast to STMT_INFOs
2987    pattern sequence.  */
2988
2989 static tree
2990 adjust_bool_pattern_cast (tree type, tree var, stmt_vec_info stmt_info)
2991 {
2992   gimple *cast_stmt = gimple_build_assign (vect_recog_temp_ssa_var (type, NULL),
2993                                            NOP_EXPR, var);
2994   stmt_vec_info patt_vinfo = new_stmt_vec_info (cast_stmt, stmt_info->vinfo);
2995   set_vinfo_for_stmt (cast_stmt, patt_vinfo);
2996   STMT_VINFO_VECTYPE (patt_vinfo) = get_vectype_for_scalar_type (type);
2997   append_pattern_def_seq (stmt_info, cast_stmt);
2998   return gimple_assign_lhs (cast_stmt);
2999 }
3000
3001 /* Helper function of vect_recog_bool_pattern.  Do the actual transformations.
3002    VAR is an SSA_NAME that should be transformed from bool to a wider integer
3003    type, OUT_TYPE is the desired final integer type of the whole pattern.
3004    STMT_INFO is the info of the pattern root and is where pattern stmts should
3005    be associated with.  DEFS is a map of pattern defs.  */
3006
3007 static void
3008 adjust_bool_pattern (tree var, tree out_type,
3009                      stmt_vec_info stmt_info, hash_map <tree, tree> &defs)
3010 {
3011   gimple *stmt = SSA_NAME_DEF_STMT (var);
3012   enum tree_code rhs_code, def_rhs_code;
3013   tree itype, cond_expr, rhs1, rhs2, irhs1, irhs2;
3014   location_t loc;
3015   gimple *pattern_stmt, *def_stmt;
3016   tree trueval = NULL_TREE;
3017
3018   rhs1 = gimple_assign_rhs1 (stmt);
3019   rhs2 = gimple_assign_rhs2 (stmt);
3020   rhs_code = gimple_assign_rhs_code (stmt);
3021   loc = gimple_location (stmt);
3022   switch (rhs_code)
3023     {
3024     case SSA_NAME:
3025     CASE_CONVERT:
3026       irhs1 = *defs.get (rhs1);
3027       itype = TREE_TYPE (irhs1);
3028       pattern_stmt
3029         = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
3030                                SSA_NAME, irhs1);
3031       break;
3032
3033     case BIT_NOT_EXPR:
3034       irhs1 = *defs.get (rhs1);
3035       itype = TREE_TYPE (irhs1);
3036       pattern_stmt
3037         = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
3038                                BIT_XOR_EXPR, irhs1, build_int_cst (itype, 1));
3039       break;
3040
3041     case BIT_AND_EXPR:
3042       /* Try to optimize x = y & (a < b ? 1 : 0); into
3043          x = (a < b ? y : 0);
3044
3045          E.g. for:
3046            bool a_b, b_b, c_b;
3047            TYPE d_T;
3048
3049            S1  a_b = x1 CMP1 y1;
3050            S2  b_b = x2 CMP2 y2;
3051            S3  c_b = a_b & b_b;
3052            S4  d_T = (TYPE) c_b;
3053
3054          we would normally emit:
3055
3056            S1'  a_T = x1 CMP1 y1 ? 1 : 0;
3057            S2'  b_T = x2 CMP2 y2 ? 1 : 0;
3058            S3'  c_T = a_T & b_T;
3059            S4'  d_T = c_T;
3060
3061          but we can save one stmt by using the
3062          result of one of the COND_EXPRs in the other COND_EXPR and leave
3063          BIT_AND_EXPR stmt out:
3064
3065            S1'  a_T = x1 CMP1 y1 ? 1 : 0;
3066            S3'  c_T = x2 CMP2 y2 ? a_T : 0;
3067            S4'  f_T = c_T;
3068
3069          At least when VEC_COND_EXPR is implemented using masks
3070          cond ? 1 : 0 is as expensive as cond ? var : 0, in both cases it
3071          computes the comparison masks and ands it, in one case with
3072          all ones vector, in the other case with a vector register.
3073          Don't do this for BIT_IOR_EXPR, because cond ? 1 : var; is
3074          often more expensive.  */
3075       def_stmt = SSA_NAME_DEF_STMT (rhs2);
3076       def_rhs_code = gimple_assign_rhs_code (def_stmt);
3077       if (TREE_CODE_CLASS (def_rhs_code) == tcc_comparison)
3078         {
3079           irhs1 = *defs.get (rhs1);
3080           tree def_rhs1 = gimple_assign_rhs1 (def_stmt);
3081           if (TYPE_PRECISION (TREE_TYPE (irhs1))
3082               == GET_MODE_BITSIZE (TYPE_MODE (TREE_TYPE (def_rhs1))))
3083             {
3084               rhs_code = def_rhs_code;
3085               rhs1 = def_rhs1;
3086               rhs2 = gimple_assign_rhs2 (def_stmt);
3087               trueval = irhs1;
3088               goto do_compare;
3089             }
3090           else
3091             irhs2 = *defs.get (rhs2);
3092           goto and_ior_xor;
3093         }
3094       def_stmt = SSA_NAME_DEF_STMT (rhs1);
3095       def_rhs_code = gimple_assign_rhs_code (def_stmt);
3096       if (TREE_CODE_CLASS (def_rhs_code) == tcc_comparison)
3097         {
3098           irhs2 = *defs.get (rhs2);
3099           tree def_rhs1 = gimple_assign_rhs1 (def_stmt);
3100           if (TYPE_PRECISION (TREE_TYPE (irhs2))
3101               == GET_MODE_BITSIZE (TYPE_MODE (TREE_TYPE (def_rhs1))))
3102             {
3103               rhs_code = def_rhs_code;
3104               rhs1 = def_rhs1;
3105               rhs2 = gimple_assign_rhs2 (def_stmt);
3106               trueval = irhs2;
3107               goto do_compare;
3108             }
3109           else
3110             irhs1 = *defs.get (rhs1);
3111           goto and_ior_xor;
3112         }
3113       /* FALLTHRU */
3114     case BIT_IOR_EXPR:
3115     case BIT_XOR_EXPR:
3116       irhs1 = *defs.get (rhs1);
3117       irhs2 = *defs.get (rhs2);
3118     and_ior_xor:
3119       if (TYPE_PRECISION (TREE_TYPE (irhs1))
3120           != TYPE_PRECISION (TREE_TYPE (irhs2)))
3121         {
3122           int prec1 = TYPE_PRECISION (TREE_TYPE (irhs1));
3123           int prec2 = TYPE_PRECISION (TREE_TYPE (irhs2));
3124           int out_prec = TYPE_PRECISION (out_type);
3125           if (absu_hwi (out_prec - prec1) < absu_hwi (out_prec - prec2))
3126             irhs2 = adjust_bool_pattern_cast (TREE_TYPE (irhs1), irhs2,
3127                                               stmt_info);
3128           else if (absu_hwi (out_prec - prec1) > absu_hwi (out_prec - prec2))
3129             irhs1 = adjust_bool_pattern_cast (TREE_TYPE (irhs2), irhs1,
3130                                               stmt_info);
3131           else
3132             {
3133               irhs1 = adjust_bool_pattern_cast (out_type, irhs1, stmt_info);
3134               irhs2 = adjust_bool_pattern_cast (out_type, irhs2, stmt_info);
3135             }
3136         }
3137       itype = TREE_TYPE (irhs1);
3138       pattern_stmt
3139         = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
3140                                rhs_code, irhs1, irhs2);
3141       break;
3142
3143     default:
3144     do_compare:
3145       gcc_assert (TREE_CODE_CLASS (rhs_code) == tcc_comparison);
3146       if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE
3147           || !TYPE_UNSIGNED (TREE_TYPE (rhs1))
3148           || (TYPE_PRECISION (TREE_TYPE (rhs1))
3149               != GET_MODE_BITSIZE (TYPE_MODE (TREE_TYPE (rhs1)))))
3150         {
3151           machine_mode mode = TYPE_MODE (TREE_TYPE (rhs1));
3152           itype
3153             = build_nonstandard_integer_type (GET_MODE_BITSIZE (mode), 1);
3154         }
3155       else
3156         itype = TREE_TYPE (rhs1);
3157       cond_expr = build2_loc (loc, rhs_code, itype, rhs1, rhs2);
3158       if (trueval == NULL_TREE)
3159         trueval = build_int_cst (itype, 1);
3160       else
3161         gcc_checking_assert (useless_type_conversion_p (itype,
3162                                                         TREE_TYPE (trueval)));
3163       pattern_stmt
3164         = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
3165                                COND_EXPR, cond_expr, trueval,
3166                                build_int_cst (itype, 0));
3167       break;
3168     }
3169
3170   gimple_set_location (pattern_stmt, loc);
3171   /* ???  Why does vect_mark_pattern_stmts set the vector type on all
3172      pattern def seq stmts instead of just letting auto-detection do
3173      its work?  */
3174   stmt_vec_info patt_vinfo = new_stmt_vec_info (pattern_stmt, stmt_info->vinfo);
3175   set_vinfo_for_stmt (pattern_stmt, patt_vinfo);
3176   STMT_VINFO_VECTYPE (patt_vinfo) = get_vectype_for_scalar_type (itype);
3177   append_pattern_def_seq (stmt_info, pattern_stmt);
3178   defs.put (var, gimple_assign_lhs (pattern_stmt));
3179 }
3180
3181 /* Comparison function to qsort a vector of gimple stmts after UID.  */
3182
3183 static int
3184 sort_after_uid (const void *p1, const void *p2)
3185 {
3186   const gimple *stmt1 = *(const gimple * const *)p1;
3187   const gimple *stmt2 = *(const gimple * const *)p2;
3188   return gimple_uid (stmt1) - gimple_uid (stmt2);
3189 }
3190
3191 /* Create pattern stmts for all stmts participating in the bool pattern
3192    specified by BOOL_STMT_SET and its root STMT with the desired type
3193    OUT_TYPE.  Return the def of the pattern root.  */
3194
3195 static tree
3196 adjust_bool_stmts (hash_set <gimple *> &bool_stmt_set,
3197                    tree out_type, gimple *stmt)
3198 {
3199   /* Gather original stmts in the bool pattern in their order of appearance
3200      in the IL.  */
3201   auto_vec<gimple *> bool_stmts (bool_stmt_set.elements ());
3202   for (hash_set <gimple *>::iterator i = bool_stmt_set.begin ();
3203        i != bool_stmt_set.end (); ++i)
3204     bool_stmts.quick_push (*i);
3205   bool_stmts.qsort (sort_after_uid);
3206
3207   /* Now process them in that order, producing pattern stmts.  */
3208   hash_map <tree, tree> defs;
3209   for (unsigned i = 0; i < bool_stmts.length (); ++i)
3210     adjust_bool_pattern (gimple_assign_lhs (bool_stmts[i]),
3211                          out_type, vinfo_for_stmt (stmt), defs);
3212
3213   /* Pop the last pattern seq stmt and install it as pattern root for STMT.  */
3214   gimple *pattern_stmt
3215     = gimple_seq_last_stmt (STMT_VINFO_PATTERN_DEF_SEQ (vinfo_for_stmt (stmt)));
3216   return gimple_assign_lhs (pattern_stmt);
3217 }
3218
3219 /* Return the proper type for converting bool VAR into
3220    an integer value or NULL_TREE if no such type exists.
3221    The type is chosen so that converted value has the
3222    same number of elements as VAR's vector type.  */
3223
3224 static tree
3225 search_type_for_mask (tree var, vec_info *vinfo)
3226 {
3227   gimple *def_stmt;
3228   enum vect_def_type dt;
3229   tree rhs1;
3230   enum tree_code rhs_code;
3231   tree res = NULL_TREE, res2;
3232
3233   if (TREE_CODE (var) != SSA_NAME)
3234     return NULL_TREE;
3235
3236   if ((TYPE_PRECISION (TREE_TYPE (var)) != 1
3237        || !TYPE_UNSIGNED (TREE_TYPE (var)))
3238       && TREE_CODE (TREE_TYPE (var)) != BOOLEAN_TYPE)
3239     return NULL_TREE;
3240
3241   if (!vect_is_simple_use (var, vinfo, &def_stmt, &dt))
3242     return NULL_TREE;
3243
3244   if (dt != vect_internal_def)
3245     return NULL_TREE;
3246
3247   if (!is_gimple_assign (def_stmt))
3248     return NULL_TREE;
3249
3250   rhs_code = gimple_assign_rhs_code (def_stmt);
3251   rhs1 = gimple_assign_rhs1 (def_stmt);
3252
3253   switch (rhs_code)
3254     {
3255     case SSA_NAME:
3256     case BIT_NOT_EXPR:
3257     CASE_CONVERT:
3258       res = search_type_for_mask (rhs1, vinfo);
3259       break;
3260
3261     case BIT_AND_EXPR:
3262     case BIT_IOR_EXPR:
3263     case BIT_XOR_EXPR:
3264       res = search_type_for_mask (rhs1, vinfo);
3265       res2 = search_type_for_mask (gimple_assign_rhs2 (def_stmt), vinfo);
3266       if (!res || (res2 && TYPE_PRECISION (res) > TYPE_PRECISION (res2)))
3267         res = res2;
3268       break;
3269
3270     default:
3271       if (TREE_CODE_CLASS (rhs_code) == tcc_comparison)
3272         {
3273           tree comp_vectype, mask_type;
3274
3275           if (TREE_CODE (TREE_TYPE (rhs1)) == BOOLEAN_TYPE)
3276             {
3277               res = search_type_for_mask (rhs1, vinfo);
3278               res2 = search_type_for_mask (gimple_assign_rhs2 (def_stmt), vinfo);
3279               if (!res || (res2 && TYPE_PRECISION (res) > TYPE_PRECISION (res2)))
3280                 res = res2;
3281               break;
3282             }
3283
3284           comp_vectype = get_vectype_for_scalar_type (TREE_TYPE (rhs1));
3285           if (comp_vectype == NULL_TREE)
3286             return NULL_TREE;
3287
3288           mask_type = get_mask_type_for_scalar_type (TREE_TYPE (rhs1));
3289           if (!mask_type
3290               || !expand_vec_cmp_expr_p (comp_vectype, mask_type))
3291             return NULL_TREE;
3292
3293           if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE
3294               || !TYPE_UNSIGNED (TREE_TYPE (rhs1)))
3295             {
3296               machine_mode mode = TYPE_MODE (TREE_TYPE (rhs1));
3297               res = build_nonstandard_integer_type (GET_MODE_BITSIZE (mode), 1);
3298             }
3299           else
3300             res = TREE_TYPE (rhs1);
3301         }
3302     }
3303
3304   return res;
3305 }
3306
3307
3308 /* Function vect_recog_bool_pattern
3309
3310    Try to find pattern like following:
3311
3312      bool a_b, b_b, c_b, d_b, e_b;
3313      TYPE f_T;
3314    loop:
3315      S1  a_b = x1 CMP1 y1;
3316      S2  b_b = x2 CMP2 y2;
3317      S3  c_b = a_b & b_b;
3318      S4  d_b = x3 CMP3 y3;
3319      S5  e_b = c_b | d_b;
3320      S6  f_T = (TYPE) e_b;
3321
3322    where type 'TYPE' is an integral type.  Or a similar pattern
3323    ending in
3324
3325      S6  f_Y = e_b ? r_Y : s_Y;
3326
3327    as results from if-conversion of a complex condition.
3328
3329    Input:
3330
3331    * LAST_STMT: A stmt at the end from which the pattern
3332                 search begins, i.e. cast of a bool to
3333                 an integer type.
3334
3335    Output:
3336
3337    * TYPE_IN: The type of the input arguments to the pattern.
3338
3339    * TYPE_OUT: The type of the output of this pattern.
3340
3341    * Return value: A new stmt that will be used to replace the pattern.
3342
3343         Assuming size of TYPE is the same as size of all comparisons
3344         (otherwise some casts would be added where needed), the above
3345         sequence we create related pattern stmts:
3346         S1'  a_T = x1 CMP1 y1 ? 1 : 0;
3347         S3'  c_T = x2 CMP2 y2 ? a_T : 0;
3348         S4'  d_T = x3 CMP3 y3 ? 1 : 0;
3349         S5'  e_T = c_T | d_T;
3350         S6'  f_T = e_T;
3351
3352         Instead of the above S3' we could emit:
3353         S2'  b_T = x2 CMP2 y2 ? 1 : 0;
3354         S3'  c_T = a_T | b_T;
3355         but the above is more efficient.  */
3356
3357 static gimple *
3358 vect_recog_bool_pattern (vec<gimple *> *stmts, tree *type_in,
3359                          tree *type_out)
3360 {
3361   gimple *last_stmt = stmts->pop ();
3362   enum tree_code rhs_code;
3363   tree var, lhs, rhs, vectype;
3364   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
3365   stmt_vec_info new_stmt_info;
3366   vec_info *vinfo = stmt_vinfo->vinfo;
3367   gimple *pattern_stmt;
3368
3369   if (!is_gimple_assign (last_stmt))
3370     return NULL;
3371
3372   var = gimple_assign_rhs1 (last_stmt);
3373   lhs = gimple_assign_lhs (last_stmt);
3374
3375   if ((TYPE_PRECISION (TREE_TYPE (var)) != 1
3376        || !TYPE_UNSIGNED (TREE_TYPE (var)))
3377       && TREE_CODE (TREE_TYPE (var)) != BOOLEAN_TYPE)
3378     return NULL;
3379
3380   hash_set<gimple *> bool_stmts;
3381
3382   rhs_code = gimple_assign_rhs_code (last_stmt);
3383   if (CONVERT_EXPR_CODE_P (rhs_code))
3384     {
3385       if (TREE_CODE (TREE_TYPE (lhs)) != INTEGER_TYPE
3386           || TYPE_PRECISION (TREE_TYPE (lhs)) == 1)
3387         return NULL;
3388       vectype = get_vectype_for_scalar_type (TREE_TYPE (lhs));
3389       if (vectype == NULL_TREE)
3390         return NULL;
3391
3392       if (check_bool_pattern (var, vinfo, bool_stmts))
3393         {
3394           rhs = adjust_bool_stmts (bool_stmts, TREE_TYPE (lhs), last_stmt);
3395           lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
3396           if (useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
3397             pattern_stmt = gimple_build_assign (lhs, SSA_NAME, rhs);
3398           else
3399             pattern_stmt
3400               = gimple_build_assign (lhs, NOP_EXPR, rhs);
3401         }
3402       else
3403         {
3404           tree type = search_type_for_mask (var, vinfo);
3405           tree cst0, cst1, tmp;
3406
3407           if (!type)
3408             return NULL;
3409
3410           /* We may directly use cond with narrowed type to avoid
3411              multiple cond exprs with following result packing and
3412              perform single cond with packed mask instead.  In case
3413              of widening we better make cond first and then extract
3414              results.  */
3415           if (TYPE_MODE (type) == TYPE_MODE (TREE_TYPE (lhs)))
3416             type = TREE_TYPE (lhs);
3417
3418           cst0 = build_int_cst (type, 0);
3419           cst1 = build_int_cst (type, 1);
3420           tmp = vect_recog_temp_ssa_var (type, NULL);
3421           pattern_stmt = gimple_build_assign (tmp, COND_EXPR, var, cst1, cst0);
3422
3423           if (!useless_type_conversion_p (type, TREE_TYPE (lhs)))
3424             {
3425               tree new_vectype = get_vectype_for_scalar_type (type);
3426               new_stmt_info = new_stmt_vec_info (pattern_stmt, vinfo);
3427               set_vinfo_for_stmt (pattern_stmt, new_stmt_info);
3428               STMT_VINFO_VECTYPE (new_stmt_info) = new_vectype;
3429               new_pattern_def_seq (stmt_vinfo, pattern_stmt);
3430
3431               lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
3432               pattern_stmt = gimple_build_assign (lhs, CONVERT_EXPR, tmp);
3433             }
3434         }
3435
3436       *type_out = vectype;
3437       *type_in = vectype;
3438       stmts->safe_push (last_stmt);
3439       if (dump_enabled_p ())
3440         dump_printf_loc (MSG_NOTE, vect_location,
3441                          "vect_recog_bool_pattern: detected:\n");
3442
3443       return pattern_stmt;
3444     }
3445   else if (rhs_code == COND_EXPR
3446            && TREE_CODE (var) == SSA_NAME)
3447     {
3448       vectype = get_vectype_for_scalar_type (TREE_TYPE (lhs));
3449       if (vectype == NULL_TREE)
3450         return NULL;
3451
3452       /* Build a scalar type for the boolean result that when
3453          vectorized matches the vector type of the result in
3454          size and number of elements.  */
3455       unsigned prec
3456         = wi::udiv_trunc (TYPE_SIZE (vectype),
3457                           TYPE_VECTOR_SUBPARTS (vectype)).to_uhwi ();
3458       tree type
3459         = build_nonstandard_integer_type (prec,
3460                                           TYPE_UNSIGNED (TREE_TYPE (var)));
3461       if (get_vectype_for_scalar_type (type) == NULL_TREE)
3462         return NULL;
3463
3464       if (!check_bool_pattern (var, vinfo, bool_stmts))
3465         return NULL;
3466
3467       rhs = adjust_bool_stmts (bool_stmts, type, last_stmt);
3468
3469       lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
3470       pattern_stmt
3471           = gimple_build_assign (lhs, COND_EXPR,
3472                                  build2 (NE_EXPR, boolean_type_node,
3473                                          rhs, build_int_cst (type, 0)),
3474                                  gimple_assign_rhs2 (last_stmt),
3475                                  gimple_assign_rhs3 (last_stmt));
3476       *type_out = vectype;
3477       *type_in = vectype;
3478       stmts->safe_push (last_stmt);
3479       if (dump_enabled_p ())
3480         dump_printf_loc (MSG_NOTE, vect_location,
3481                          "vect_recog_bool_pattern: detected:\n");
3482
3483       return pattern_stmt;
3484     }
3485   else if (rhs_code == SSA_NAME
3486            && STMT_VINFO_DATA_REF (stmt_vinfo))
3487     {
3488       stmt_vec_info pattern_stmt_info;
3489       vectype = STMT_VINFO_VECTYPE (stmt_vinfo);
3490       gcc_assert (vectype != NULL_TREE);
3491       if (!VECTOR_MODE_P (TYPE_MODE (vectype)))
3492         return NULL;
3493
3494       if (check_bool_pattern (var, vinfo, bool_stmts))
3495         rhs = adjust_bool_stmts (bool_stmts, TREE_TYPE (vectype), last_stmt);
3496       else
3497         {
3498           tree type = search_type_for_mask (var, vinfo);
3499           tree cst0, cst1, new_vectype;
3500
3501           if (!type)
3502             return NULL;
3503
3504           if (TYPE_MODE (type) == TYPE_MODE (TREE_TYPE (vectype)))
3505             type = TREE_TYPE (vectype);
3506
3507           cst0 = build_int_cst (type, 0);
3508           cst1 = build_int_cst (type, 1);
3509           new_vectype = get_vectype_for_scalar_type (type);
3510
3511           rhs = vect_recog_temp_ssa_var (type, NULL);
3512           pattern_stmt = gimple_build_assign (rhs, COND_EXPR, var, cst1, cst0);
3513
3514           pattern_stmt_info = new_stmt_vec_info (pattern_stmt, vinfo);
3515           set_vinfo_for_stmt (pattern_stmt, pattern_stmt_info);
3516           STMT_VINFO_VECTYPE (pattern_stmt_info) = new_vectype;
3517           append_pattern_def_seq (stmt_vinfo, pattern_stmt);
3518         }
3519
3520       lhs = build1 (VIEW_CONVERT_EXPR, TREE_TYPE (vectype), lhs);
3521       if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
3522         {
3523           tree rhs2 = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
3524           gimple *cast_stmt = gimple_build_assign (rhs2, NOP_EXPR, rhs);
3525           append_pattern_def_seq (stmt_vinfo, cast_stmt);
3526           rhs = rhs2;
3527         }
3528       pattern_stmt = gimple_build_assign (lhs, SSA_NAME, rhs);
3529       pattern_stmt_info = new_stmt_vec_info (pattern_stmt, vinfo);
3530       set_vinfo_for_stmt (pattern_stmt, pattern_stmt_info);
3531       STMT_VINFO_DATA_REF (pattern_stmt_info)
3532         = STMT_VINFO_DATA_REF (stmt_vinfo);
3533       STMT_VINFO_DR_BASE_ADDRESS (pattern_stmt_info)
3534         = STMT_VINFO_DR_BASE_ADDRESS (stmt_vinfo);
3535       STMT_VINFO_DR_INIT (pattern_stmt_info) = STMT_VINFO_DR_INIT (stmt_vinfo);
3536       STMT_VINFO_DR_OFFSET (pattern_stmt_info)
3537         = STMT_VINFO_DR_OFFSET (stmt_vinfo);
3538       STMT_VINFO_DR_STEP (pattern_stmt_info) = STMT_VINFO_DR_STEP (stmt_vinfo);
3539       STMT_VINFO_DR_ALIGNED_TO (pattern_stmt_info)
3540         = STMT_VINFO_DR_ALIGNED_TO (stmt_vinfo);
3541       DR_STMT (STMT_VINFO_DATA_REF (stmt_vinfo)) = pattern_stmt;
3542       *type_out = vectype;
3543       *type_in = vectype;
3544       stmts->safe_push (last_stmt);
3545       if (dump_enabled_p ())
3546         dump_printf_loc (MSG_NOTE, vect_location,
3547                          "vect_recog_bool_pattern: detected:\n");
3548       return pattern_stmt;
3549     }
3550   else
3551     return NULL;
3552 }
3553
3554
3555 /* A helper for vect_recog_mask_conversion_pattern.  Build
3556    conversion of MASK to a type suitable for masking VECTYPE.
3557    Built statement gets required vectype and is appended to
3558    a pattern sequence of STMT_VINFO.
3559
3560    Return converted mask.  */
3561
3562 static tree
3563 build_mask_conversion (tree mask, tree vectype, stmt_vec_info stmt_vinfo,
3564                        vec_info *vinfo)
3565 {
3566   gimple *stmt;
3567   tree masktype, tmp;
3568   stmt_vec_info new_stmt_info;
3569
3570   masktype = build_same_sized_truth_vector_type (vectype);
3571   tmp = vect_recog_temp_ssa_var (TREE_TYPE (masktype), NULL);
3572   stmt = gimple_build_assign (tmp, CONVERT_EXPR, mask);
3573   new_stmt_info = new_stmt_vec_info (stmt, vinfo);
3574   set_vinfo_for_stmt (stmt, new_stmt_info);
3575   STMT_VINFO_VECTYPE (new_stmt_info) = masktype;
3576   append_pattern_def_seq (stmt_vinfo, stmt);
3577
3578   return tmp;
3579 }
3580
3581
3582 /* Function vect_recog_mask_conversion_pattern
3583
3584    Try to find statements which require boolean type
3585    converison.  Additional conversion statements are
3586    added to handle such cases.  For example:
3587
3588    bool m_1, m_2, m_3;
3589    int i_4, i_5;
3590    double d_6, d_7;
3591    char c_1, c_2, c_3;
3592
3593    S1   m_1 = i_4 > i_5;
3594    S2   m_2 = d_6 < d_7;
3595    S3   m_3 = m_1 & m_2;
3596    S4   c_1 = m_3 ? c_2 : c_3;
3597
3598    Will be transformed into:
3599
3600    S1   m_1 = i_4 > i_5;
3601    S2   m_2 = d_6 < d_7;
3602    S3'' m_2' = (_Bool[bitsize=32])m_2
3603    S3'  m_3' = m_1 & m_2';
3604    S4'' m_3'' = (_Bool[bitsize=8])m_3'
3605    S4'  c_1' = m_3'' ? c_2 : c_3;  */
3606
3607 static gimple *
3608 vect_recog_mask_conversion_pattern (vec<gimple *> *stmts, tree *type_in,
3609                                     tree *type_out)
3610 {
3611   gimple *last_stmt = stmts->pop ();
3612   enum tree_code rhs_code;
3613   tree lhs = NULL_TREE, rhs1, rhs2, tmp, rhs1_type, rhs2_type;
3614   tree vectype1, vectype2;
3615   stmt_vec_info stmt_vinfo = vinfo_for_stmt (last_stmt);
3616   stmt_vec_info pattern_stmt_info;
3617   vec_info *vinfo = stmt_vinfo->vinfo;
3618   gimple *pattern_stmt;
3619
3620   /* Check for MASK_LOAD ans MASK_STORE calls requiring mask conversion.  */
3621   if (is_gimple_call (last_stmt)
3622       && gimple_call_internal_p (last_stmt)
3623       && (gimple_call_internal_fn (last_stmt) == IFN_MASK_STORE
3624           || gimple_call_internal_fn (last_stmt) == IFN_MASK_LOAD))
3625     {
3626       bool load = (gimple_call_internal_fn (last_stmt) == IFN_MASK_LOAD);
3627
3628       if (load)
3629         {
3630           lhs = gimple_call_lhs (last_stmt);
3631           vectype1 = get_vectype_for_scalar_type (TREE_TYPE (lhs));
3632         }
3633       else
3634         {
3635           rhs2 = gimple_call_arg (last_stmt, 3);
3636           vectype1 = get_vectype_for_scalar_type (TREE_TYPE (rhs2));
3637         }
3638
3639       rhs1 = gimple_call_arg (last_stmt, 2);
3640       rhs1_type = search_type_for_mask (rhs1, vinfo);
3641       if (!rhs1_type)
3642         return NULL;
3643       vectype2 = get_mask_type_for_scalar_type (rhs1_type);
3644
3645       if (!vectype1 || !vectype2
3646           || TYPE_VECTOR_SUBPARTS (vectype1) == TYPE_VECTOR_SUBPARTS (vectype2))
3647         return NULL;
3648
3649       tmp = build_mask_conversion (rhs1, vectype1, stmt_vinfo, vinfo);
3650
3651       if (load)
3652         {
3653           lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
3654           pattern_stmt
3655             = gimple_build_call_internal (IFN_MASK_LOAD, 3,
3656                                           gimple_call_arg (last_stmt, 0),
3657                                           gimple_call_arg (last_stmt, 1),
3658                                           tmp);
3659           gimple_call_set_lhs (pattern_stmt, lhs);
3660         }
3661       else
3662           pattern_stmt
3663             = gimple_build_call_internal (IFN_MASK_STORE, 4,
3664                                           gimple_call_arg (last_stmt, 0),
3665                                           gimple_call_arg (last_stmt, 1),
3666                                           tmp,
3667                                           gimple_call_arg (last_stmt, 3));
3668
3669
3670       pattern_stmt_info = new_stmt_vec_info (pattern_stmt, vinfo);
3671       set_vinfo_for_stmt (pattern_stmt, pattern_stmt_info);
3672       STMT_VINFO_DATA_REF (pattern_stmt_info)
3673         = STMT_VINFO_DATA_REF (stmt_vinfo);
3674       STMT_VINFO_DR_BASE_ADDRESS (pattern_stmt_info)
3675         = STMT_VINFO_DR_BASE_ADDRESS (stmt_vinfo);
3676       STMT_VINFO_DR_INIT (pattern_stmt_info) = STMT_VINFO_DR_INIT (stmt_vinfo);
3677       STMT_VINFO_DR_OFFSET (pattern_stmt_info)
3678         = STMT_VINFO_DR_OFFSET (stmt_vinfo);
3679       STMT_VINFO_DR_STEP (pattern_stmt_info) = STMT_VINFO_DR_STEP (stmt_vinfo);
3680       STMT_VINFO_DR_ALIGNED_TO (pattern_stmt_info)
3681         = STMT_VINFO_DR_ALIGNED_TO (stmt_vinfo);
3682       DR_STMT (STMT_VINFO_DATA_REF (stmt_vinfo)) = pattern_stmt;
3683
3684       *type_out = vectype1;
3685       *type_in = vectype1;
3686       stmts->safe_push (last_stmt);
3687       if (dump_enabled_p ())
3688         dump_printf_loc (MSG_NOTE, vect_location,
3689                          "vect_recog_mask_conversion_pattern: detected:\n");
3690
3691       return pattern_stmt;
3692     }
3693
3694   if (!is_gimple_assign (last_stmt))
3695     return NULL;
3696
3697   lhs = gimple_assign_lhs (last_stmt);
3698   rhs1 = gimple_assign_rhs1 (last_stmt);
3699   rhs_code = gimple_assign_rhs_code (last_stmt);
3700
3701   /* Check for cond expression requiring mask conversion.  */
3702   if (rhs_code == COND_EXPR)
3703     {
3704       /* vect_recog_mixed_size_cond_pattern could apply.
3705          Do nothing then.  */
3706       if (STMT_VINFO_IN_PATTERN_P (stmt_vinfo))
3707         return NULL;
3708
3709       vectype1 = get_vectype_for_scalar_type (TREE_TYPE (lhs));
3710
3711       if (TREE_CODE (rhs1) == SSA_NAME)
3712         {
3713           rhs1_type = search_type_for_mask (rhs1, vinfo);
3714           if (!rhs1_type)
3715             return NULL;
3716         }
3717       else if (COMPARISON_CLASS_P (rhs1))
3718         rhs1_type = TREE_TYPE (TREE_OPERAND (rhs1, 0));
3719       else
3720         return NULL;
3721
3722       vectype2 = get_mask_type_for_scalar_type (rhs1_type);
3723
3724       if (!vectype1 || !vectype2
3725           || TYPE_VECTOR_SUBPARTS (vectype1) == TYPE_VECTOR_SUBPARTS (vectype2))
3726         return NULL;
3727
3728       /* If rhs1 is a comparison we need to move it into a
3729          separate statement.  */
3730       if (TREE_CODE (rhs1) != SSA_NAME)
3731         {
3732           tmp = vect_recog_temp_ssa_var (TREE_TYPE (rhs1), NULL);
3733           pattern_stmt = gimple_build_assign (tmp, rhs1);
3734           rhs1 = tmp;
3735
3736           pattern_stmt_info = new_stmt_vec_info (pattern_stmt, vinfo);
3737           set_vinfo_for_stmt (pattern_stmt, pattern_stmt_info);
3738           STMT_VINFO_VECTYPE (pattern_stmt_info) = vectype2;
3739           append_pattern_def_seq (stmt_vinfo, pattern_stmt);
3740         }
3741
3742       tmp = build_mask_conversion (rhs1, vectype1, stmt_vinfo, vinfo);
3743
3744       lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
3745       pattern_stmt = gimple_build_assign (lhs, COND_EXPR, tmp,
3746                                           gimple_assign_rhs2 (last_stmt),
3747                                           gimple_assign_rhs3 (last_stmt));
3748
3749       *type_out = vectype1;
3750       *type_in = vectype1;
3751       stmts->safe_push (last_stmt);
3752       if (dump_enabled_p ())
3753         dump_printf_loc (MSG_NOTE, vect_location,
3754                          "vect_recog_mask_conversion_pattern: detected:\n");
3755
3756       return pattern_stmt;
3757     }
3758
3759   /* Now check for binary boolean operations requiring conversion for
3760      one of operands.  */
3761   if (TREE_CODE (TREE_TYPE (lhs)) != BOOLEAN_TYPE)
3762     return NULL;
3763
3764   if (rhs_code != BIT_IOR_EXPR
3765       && rhs_code != BIT_XOR_EXPR
3766       && rhs_code != BIT_AND_EXPR)
3767     return NULL;
3768
3769   rhs2 = gimple_assign_rhs2 (last_stmt);
3770
3771   rhs1_type = search_type_for_mask (rhs1, vinfo);
3772   rhs2_type = search_type_for_mask (rhs2, vinfo);
3773
3774   if (!rhs1_type || !rhs2_type
3775       || TYPE_PRECISION (rhs1_type) == TYPE_PRECISION (rhs2_type))
3776     return NULL;
3777
3778   if (TYPE_PRECISION (rhs1_type) < TYPE_PRECISION (rhs2_type))
3779     {
3780       vectype1 = get_mask_type_for_scalar_type (rhs1_type);
3781       if (!vectype1)
3782         return NULL;
3783       rhs2 = build_mask_conversion (rhs2, vectype1, stmt_vinfo, vinfo);
3784     }
3785   else
3786     {
3787       vectype1 = get_mask_type_for_scalar_type (rhs2_type);
3788       if (!vectype1)
3789         return NULL;
3790       rhs1 = build_mask_conversion (rhs1, vectype1, stmt_vinfo, vinfo);
3791     }
3792
3793   lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
3794   pattern_stmt = gimple_build_assign (lhs, rhs_code, rhs1, rhs2);
3795
3796   *type_out = vectype1;
3797   *type_in = vectype1;
3798   stmts->safe_push (last_stmt);
3799   if (dump_enabled_p ())
3800     dump_printf_loc (MSG_NOTE, vect_location,
3801                      "vect_recog_mask_conversion_pattern: detected:\n");
3802
3803   return pattern_stmt;
3804 }
3805
3806
3807 /* Mark statements that are involved in a pattern.  */
3808
3809 static inline void
3810 vect_mark_pattern_stmts (gimple *orig_stmt, gimple *pattern_stmt,
3811                          tree pattern_vectype)
3812 {
3813   stmt_vec_info pattern_stmt_info, def_stmt_info;
3814   stmt_vec_info orig_stmt_info = vinfo_for_stmt (orig_stmt);
3815   vec_info *vinfo = orig_stmt_info->vinfo;
3816   gimple *def_stmt;
3817
3818   pattern_stmt_info = vinfo_for_stmt (pattern_stmt);
3819   if (pattern_stmt_info == NULL)
3820     {
3821       pattern_stmt_info = new_stmt_vec_info (pattern_stmt, vinfo);
3822       set_vinfo_for_stmt (pattern_stmt, pattern_stmt_info);
3823     }
3824   gimple_set_bb (pattern_stmt, gimple_bb (orig_stmt));
3825
3826   STMT_VINFO_RELATED_STMT (pattern_stmt_info) = orig_stmt;
3827   STMT_VINFO_DEF_TYPE (pattern_stmt_info)
3828     = STMT_VINFO_DEF_TYPE (orig_stmt_info);
3829   STMT_VINFO_VECTYPE (pattern_stmt_info) = pattern_vectype;
3830   STMT_VINFO_IN_PATTERN_P (orig_stmt_info) = true;
3831   STMT_VINFO_RELATED_STMT (orig_stmt_info) = pattern_stmt;
3832   STMT_VINFO_PATTERN_DEF_SEQ (pattern_stmt_info)
3833     = STMT_VINFO_PATTERN_DEF_SEQ (orig_stmt_info);
3834   if (STMT_VINFO_PATTERN_DEF_SEQ (pattern_stmt_info))
3835     {
3836       gimple_stmt_iterator si;
3837       for (si = gsi_start (STMT_VINFO_PATTERN_DEF_SEQ (pattern_stmt_info));
3838            !gsi_end_p (si); gsi_next (&si))
3839         {
3840           def_stmt = gsi_stmt (si);
3841           def_stmt_info = vinfo_for_stmt (def_stmt);
3842           if (def_stmt_info == NULL)
3843             {
3844               def_stmt_info = new_stmt_vec_info (def_stmt, vinfo);
3845               set_vinfo_for_stmt (def_stmt, def_stmt_info);
3846             }
3847           gimple_set_bb (def_stmt, gimple_bb (orig_stmt));
3848           STMT_VINFO_RELATED_STMT (def_stmt_info) = orig_stmt;
3849           STMT_VINFO_DEF_TYPE (def_stmt_info) = vect_internal_def;
3850           if (STMT_VINFO_VECTYPE (def_stmt_info) == NULL_TREE)
3851             STMT_VINFO_VECTYPE (def_stmt_info) = pattern_vectype;
3852         }
3853     }
3854 }
3855
3856 /* Function vect_pattern_recog_1
3857
3858    Input:
3859    PATTERN_RECOG_FUNC: A pointer to a function that detects a certain
3860         computation pattern.
3861    STMT: A stmt from which the pattern search should start.
3862
3863    If PATTERN_RECOG_FUNC successfully detected the pattern, it creates an
3864    expression that computes the same functionality and can be used to
3865    replace the sequence of stmts that are involved in the pattern.
3866
3867    Output:
3868    This function checks if the expression returned by PATTERN_RECOG_FUNC is
3869    supported in vector form by the target.  We use 'TYPE_IN' to obtain the
3870    relevant vector type. If 'TYPE_IN' is already a vector type, then this
3871    indicates that target support had already been checked by PATTERN_RECOG_FUNC.
3872    If 'TYPE_OUT' is also returned by PATTERN_RECOG_FUNC, we check that it fits
3873    to the available target pattern.
3874
3875    This function also does some bookkeeping, as explained in the documentation
3876    for vect_recog_pattern.  */
3877
3878 static bool
3879 vect_pattern_recog_1 (vect_recog_func *recog_func,
3880                       gimple_stmt_iterator si,
3881                       vec<gimple *> *stmts_to_replace)
3882 {
3883   gimple *stmt = gsi_stmt (si), *pattern_stmt;
3884   stmt_vec_info stmt_info;
3885   loop_vec_info loop_vinfo;
3886   tree pattern_vectype;
3887   tree type_in, type_out;
3888   enum tree_code code;
3889   int i;
3890   gimple *next;
3891
3892   stmts_to_replace->truncate (0);
3893   stmts_to_replace->quick_push (stmt);
3894   pattern_stmt = recog_func->fn (stmts_to_replace, &type_in, &type_out);
3895   if (!pattern_stmt)
3896     return false;
3897
3898   stmt = stmts_to_replace->last ();
3899   stmt_info = vinfo_for_stmt (stmt);
3900   loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
3901
3902   if (VECTOR_BOOLEAN_TYPE_P (type_in)
3903       || VECTOR_MODE_P (TYPE_MODE (type_in)))
3904     {
3905       /* No need to check target support (already checked by the pattern
3906          recognition function).  */
3907       pattern_vectype = type_out ? type_out : type_in;
3908     }
3909   else
3910     {
3911       machine_mode vec_mode;
3912       enum insn_code icode;
3913       optab optab;
3914
3915       /* Check target support  */
3916       type_in = get_vectype_for_scalar_type (type_in);
3917       if (!type_in)
3918         return false;
3919       if (type_out)
3920         type_out = get_vectype_for_scalar_type (type_out);
3921       else
3922         type_out = type_in;
3923       if (!type_out)
3924         return false;
3925       pattern_vectype = type_out;
3926
3927       if (is_gimple_assign (pattern_stmt))
3928         code = gimple_assign_rhs_code (pattern_stmt);
3929       else
3930         {
3931           gcc_assert (is_gimple_call (pattern_stmt));
3932           code = CALL_EXPR;
3933         }
3934
3935       optab = optab_for_tree_code (code, type_in, optab_default);
3936       vec_mode = TYPE_MODE (type_in);
3937       if (!optab
3938           || (icode = optab_handler (optab, vec_mode)) == CODE_FOR_nothing
3939           || (insn_data[icode].operand[0].mode != TYPE_MODE (type_out)))
3940         return false;
3941     }
3942
3943   /* Found a vectorizable pattern.  */
3944   if (dump_enabled_p ())
3945     {
3946       dump_printf_loc (MSG_NOTE, vect_location,
3947                        "%s pattern recognized: ", recog_func->name);
3948       dump_gimple_stmt (MSG_NOTE, TDF_SLIM, pattern_stmt, 0);
3949     }
3950
3951   /* Mark the stmts that are involved in the pattern. */
3952   vect_mark_pattern_stmts (stmt, pattern_stmt, pattern_vectype);
3953
3954   /* Patterns cannot be vectorized using SLP, because they change the order of
3955      computation.  */
3956   if (loop_vinfo)
3957     FOR_EACH_VEC_ELT (LOOP_VINFO_REDUCTIONS (loop_vinfo), i, next)
3958       if (next == stmt)
3959         LOOP_VINFO_REDUCTIONS (loop_vinfo).ordered_remove (i);
3960
3961   /* It is possible that additional pattern stmts are created and inserted in
3962      STMTS_TO_REPLACE.  We create a stmt_info for each of them, and mark the
3963      relevant statements.  */
3964   for (i = 0; stmts_to_replace->iterate (i, &stmt)
3965               && (unsigned) i < (stmts_to_replace->length () - 1);
3966        i++)
3967     {
3968       stmt_info = vinfo_for_stmt (stmt);
3969       pattern_stmt = STMT_VINFO_RELATED_STMT (stmt_info);
3970       if (dump_enabled_p ())
3971         {
3972           dump_printf_loc (MSG_NOTE, vect_location,
3973                            "additional pattern stmt: ");
3974           dump_gimple_stmt (MSG_NOTE, TDF_SLIM, pattern_stmt, 0);
3975         }
3976
3977       vect_mark_pattern_stmts (stmt, pattern_stmt, NULL_TREE);
3978     }
3979
3980   return true;
3981 }
3982
3983
3984 /* Function vect_pattern_recog
3985
3986    Input:
3987    LOOP_VINFO - a struct_loop_info of a loop in which we want to look for
3988         computation idioms.
3989
3990    Output - for each computation idiom that is detected we create a new stmt
3991         that provides the same functionality and that can be vectorized.  We
3992         also record some information in the struct_stmt_info of the relevant
3993         stmts, as explained below:
3994
3995    At the entry to this function we have the following stmts, with the
3996    following initial value in the STMT_VINFO fields:
3997
3998          stmt                     in_pattern_p  related_stmt    vec_stmt
3999          S1: a_i = ....                 -       -               -
4000          S2: a_2 = ..use(a_i)..         -       -               -
4001          S3: a_1 = ..use(a_2)..         -       -               -
4002          S4: a_0 = ..use(a_1)..         -       -               -
4003          S5: ... = ..use(a_0)..         -       -               -
4004
4005    Say the sequence {S1,S2,S3,S4} was detected as a pattern that can be
4006    represented by a single stmt.  We then:
4007    - create a new stmt S6 equivalent to the pattern (the stmt is not
4008      inserted into the code)
4009    - fill in the STMT_VINFO fields as follows:
4010
4011                                   in_pattern_p  related_stmt    vec_stmt
4012          S1: a_i = ....                 -       -               -
4013          S2: a_2 = ..use(a_i)..         -       -               -
4014          S3: a_1 = ..use(a_2)..         -       -               -
4015          S4: a_0 = ..use(a_1)..         true    S6              -
4016           '---> S6: a_new = ....        -       S4              -
4017          S5: ... = ..use(a_0)..         -       -               -
4018
4019    (the last stmt in the pattern (S4) and the new pattern stmt (S6) point
4020    to each other through the RELATED_STMT field).
4021
4022    S6 will be marked as relevant in vect_mark_stmts_to_be_vectorized instead
4023    of S4 because it will replace all its uses.  Stmts {S1,S2,S3} will
4024    remain irrelevant unless used by stmts other than S4.
4025
4026    If vectorization succeeds, vect_transform_stmt will skip over {S1,S2,S3}
4027    (because they are marked as irrelevant).  It will vectorize S6, and record
4028    a pointer to the new vector stmt VS6 from S6 (as usual).
4029    S4 will be skipped, and S5 will be vectorized as usual:
4030
4031                                   in_pattern_p  related_stmt    vec_stmt
4032          S1: a_i = ....                 -       -               -
4033          S2: a_2 = ..use(a_i)..         -       -               -
4034          S3: a_1 = ..use(a_2)..         -       -               -
4035        > VS6: va_new = ....             -       -               -
4036          S4: a_0 = ..use(a_1)..         true    S6              VS6
4037           '---> S6: a_new = ....        -       S4              VS6
4038        > VS5: ... = ..vuse(va_new)..    -       -               -
4039          S5: ... = ..use(a_0)..         -       -               -
4040
4041    DCE could then get rid of {S1,S2,S3,S4,S5} (if their defs are not used
4042    elsewhere), and we'll end up with:
4043
4044         VS6: va_new = ....
4045         VS5: ... = ..vuse(va_new)..
4046
4047    In case of more than one pattern statements, e.g., widen-mult with
4048    intermediate type:
4049
4050      S1  a_t = ;
4051      S2  a_T = (TYPE) a_t;
4052            '--> S3: a_it = (interm_type) a_t;
4053      S4  prod_T = a_T * CONST;
4054            '--> S5: prod_T' = a_it w* CONST;
4055
4056    there may be other users of a_T outside the pattern.  In that case S2 will
4057    be marked as relevant (as well as S3), and both S2 and S3 will be analyzed
4058    and vectorized.  The vector stmt VS2 will be recorded in S2, and VS3 will
4059    be recorded in S3.  */
4060
4061 void
4062 vect_pattern_recog (vec_info *vinfo)
4063 {
4064   struct loop *loop;
4065   basic_block *bbs;
4066   unsigned int nbbs;
4067   gimple_stmt_iterator si;
4068   unsigned int i, j;
4069   auto_vec<gimple *, 1> stmts_to_replace;
4070   gimple *stmt;
4071
4072   if (dump_enabled_p ())
4073     dump_printf_loc (MSG_NOTE, vect_location,
4074                      "=== vect_pattern_recog ===\n");
4075
4076   if (loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo))
4077     {
4078       loop = LOOP_VINFO_LOOP (loop_vinfo);
4079       bbs = LOOP_VINFO_BBS (loop_vinfo);
4080       nbbs = loop->num_nodes;
4081
4082       /* Scan through the loop stmts, applying the pattern recognition
4083          functions starting at each stmt visited:  */
4084       for (i = 0; i < nbbs; i++)
4085         {
4086           basic_block bb = bbs[i];
4087           for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si))
4088             {
4089               /* Scan over all generic vect_recog_xxx_pattern functions.  */
4090               for (j = 0; j < NUM_PATTERNS; j++)
4091                 if (vect_pattern_recog_1 (&vect_vect_recog_func_ptrs[j], si,
4092                                           &stmts_to_replace))
4093                   break;
4094             }
4095         }
4096     }
4097   else
4098     {
4099       bb_vec_info bb_vinfo = as_a <bb_vec_info> (vinfo);
4100       for (si = bb_vinfo->region_begin;
4101            gsi_stmt (si) != gsi_stmt (bb_vinfo->region_end); gsi_next (&si))
4102         {
4103           if ((stmt = gsi_stmt (si))
4104               && vinfo_for_stmt (stmt)
4105               && !STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (stmt)))
4106             continue;
4107
4108           /* Scan over all generic vect_recog_xxx_pattern functions.  */
4109           for (j = 0; j < NUM_PATTERNS; j++)
4110             if (vect_pattern_recog_1 (&vect_vect_recog_func_ptrs[j], si,
4111                                       &stmts_to_replace))
4112               break;
4113         }
4114     }
4115 }