gcc/tree-vectorizer.c

   1 /* Vectorizer
   2    Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009 Free Software
   3    Foundation, Inc.
   4    Contributed by Dorit Naishlos <dorit@il.ibm.com>
   5
   6 This file is part of GCC.
   7
   8 GCC is free software; you can redistribute it and/or modify it under
   9 the terms of the GNU General Public License as published by the Free
  10 Software Foundation; either version 3, or (at your option) any later
  11 version.
  12
  13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  15 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  16 for more details.
  17
  18 You should have received a copy of the GNU General Public License
  19 along with GCC; see the file COPYING3.  If not see
  20 <http://www.gnu.org/licenses/>.  */
  21
  22 /* Loop and basic block vectorizer.
  23
  24   This file contains drivers for the three vectorizers:
  25   (1) loop vectorizer (inter-iteration parallelism),
  26   (2) loop-aware SLP (intra-iteration parallelism) (invoked by the loop
  27       vectorizer)
  28   (3) BB vectorizer (out-of-loops), aka SLP
  29
  30   The rest of the vectorizer's code is organized as follows:
  31   - tree-vect-loop.c - loop specific parts such as reductions, etc. These are
  32     used by drivers (1) and (2).
  33   - tree-vect-loop-manip.c - vectorizer's loop control-flow utilities, used by
  34     drivers (1) and (2).
  35   - tree-vect-slp.c - BB vectorization specific analysis and transformation,
  36     used by drivers (2) and (3).
  37   - tree-vect-stmts.c - statements analysis and transformation (used by all).
  38   - tree-vect-data-refs.c - vectorizer specific data-refs analysis and
  39     manipulations (used by all).
  40   - tree-vect-patterns.c - vectorizable code patterns detector (used by all)
  41
  42   Here's a poor attempt at illustrating that:
  43
  44      tree-vectorizer.c:
  45      loop_vect()  loop_aware_slp()  slp_vect()
  46           |        /           \          /
  47           |       /             \        /
  48           tree-vect-loop.c  tree-vect-slp.c
  49                 | \      \  /      /   |
  50                 |  \      \/      /    |
  51                 |   \     /\     /     |
  52                 |    \   /  \   /      |
  53          tree-vect-stmts.c  tree-vect-data-refs.c
  54                        \      /
  55                     tree-vect-patterns.c
  56 */
  57
  58 #include "config.h"
  59 #include "system.h"
  60 #include "coretypes.h"
  61 #include "tm.h"
  62 #include "ggc.h"
  63 #include "tree.h"
  64 #include "diagnostic.h"
  65 #include "tree-flow.h"
  66 #include "tree-dump.h"
  67 #include "cfgloop.h"
  68 #include "cfglayout.h"
  69 #include "tree-vectorizer.h"
  70 #include "tree-pass.h"
  71 #include "timevar.h"
  72
  73 /* vect_dump will be set to stderr or dump_file if exist.  */
  74 FILE *vect_dump;
  75
  76 /* vect_verbosity_level set to an invalid value
  77    to mark that it's uninitialized.  */
  78 static enum verbosity_levels vect_verbosity_level = MAX_VERBOSITY_LEVEL;
  79 static enum verbosity_levels user_vect_verbosity_level = MAX_VERBOSITY_LEVEL;
  80
  81 /* Loop or bb location.  */
  82 LOC vect_location;
  83
  84 /* Bitmap of virtual variables to be renamed.  */
  85 bitmap vect_memsyms_to_rename;
  86
  87 /* Vector mapping GIMPLE stmt to stmt_vec_info. */
  88 VEC(vec_void_p,heap) *stmt_vec_info_vec;
  89
  90 \f
  91
  92 /* Function vect_set_verbosity_level.
  93
  94    Called from opts.c upon detection of the
  95    -ftree-vectorizer-verbose=N option.  */
  96
  97 void
  98 vect_set_verbosity_level (const char *val)
  99 {
 100    unsigned int vl;
 101
 102    vl = atoi (val);
 103    if (vl < MAX_VERBOSITY_LEVEL)
 104      user_vect_verbosity_level = (enum verbosity_levels) vl;
 105    else
 106      user_vect_verbosity_level
 107       = (enum verbosity_levels) (MAX_VERBOSITY_LEVEL - 1);
 108 }
 109
 110
 111 /* Function vect_set_dump_settings.
 112
 113    Fix the verbosity level of the vectorizer if the
 114    requested level was not set explicitly using the flag
 115    -ftree-vectorizer-verbose=N.
 116    Decide where to print the debugging information (dump_file/stderr).
 117    If the user defined the verbosity level, but there is no dump file,
 118    print to stderr, otherwise print to the dump file.  */
 119
 120 static void
 121 vect_set_dump_settings (bool slp)
 122 {
 123   vect_dump = dump_file;
 124
 125   /* Check if the verbosity level was defined by the user:  */
 126   if (user_vect_verbosity_level != MAX_VERBOSITY_LEVEL)
 127     {
 128       vect_verbosity_level = user_vect_verbosity_level;
 129       /* Ignore user defined verbosity if dump flags require higher level of
 130          verbosity.  */
 131       if (dump_file)
 132         {
 133           if (((dump_flags & TDF_DETAILS)
 134                 && vect_verbosity_level >= REPORT_DETAILS)
 135                || ((dump_flags & TDF_STATS)
 136                     && vect_verbosity_level >= REPORT_UNVECTORIZED_LOCATIONS))
 137             return;
 138         }
 139       else
 140         {
 141           /* If there is no dump file, print to stderr in case of loop
 142              vectorization.  */
 143           if (!slp)
 144             vect_dump = stderr;
 145
 146           return;
 147         }
 148     }
 149
 150   /* User didn't specify verbosity level:  */
 151   if (dump_file && (dump_flags & TDF_DETAILS))
 152     vect_verbosity_level = REPORT_DETAILS;
 153   else if (dump_file && (dump_flags & TDF_STATS))
 154     vect_verbosity_level = REPORT_UNVECTORIZED_LOCATIONS;
 155   else
 156     vect_verbosity_level = REPORT_NONE;
 157
 158   gcc_assert (dump_file || vect_verbosity_level == REPORT_NONE);
 159 }
 160
 161
 162 /* Function debug_loop_details.
 163
 164    For vectorization debug dumps.  */
 165
 166 bool
 167 vect_print_dump_info (enum verbosity_levels vl)
 168 {
 169   if (vl > vect_verbosity_level)
 170     return false;
 171
 172   if (!current_function_decl || !vect_dump)
 173     return false;
 174
 175   if (vect_location == UNKNOWN_LOC)
 176     fprintf (vect_dump, "\n%s:%d: note: ",
 177              DECL_SOURCE_FILE (current_function_decl),
 178              DECL_SOURCE_LINE (current_function_decl));
 179   else
 180     fprintf (vect_dump, "\n%s:%d: note: ",
 181              LOC_FILE (vect_location), LOC_LINE (vect_location));
 182
 183   return true;
 184 }
 185
 186
 187 /* Function vectorize_loops.
 188
 189    Entry point to loop vectorization phase.  */
 190
 191 unsigned
 192 vectorize_loops (void)
 193 {
 194   unsigned int i;
 195   unsigned int num_vectorized_loops = 0;
 196   unsigned int vect_loops_num;
 197   loop_iterator li;
 198   struct loop *loop;
 199
 200   vect_loops_num = number_of_loops ();
 201
 202   /* Bail out if there are no loops.  */
 203   if (vect_loops_num <= 1)
 204     return 0;
 205
 206   /* Fix the verbosity level if not defined explicitly by the user.  */
 207   vect_set_dump_settings (false);
 208
 209   /* Allocate the bitmap that records which virtual variables
 210      need to be renamed.  */
 211   vect_memsyms_to_rename = BITMAP_ALLOC (NULL);
 212
 213   init_stmt_vec_info_vec ();
 214
 215   /*  ----------- Analyze loops. -----------  */
 216
 217   /* If some loop was duplicated, it gets bigger number
 218      than all previously defined loops. This fact allows us to run
 219      only over initial loops skipping newly generated ones.  */
 220   FOR_EACH_LOOP (li, loop, 0)
 221     if (optimize_loop_nest_for_speed_p (loop))
 222       {
 223         loop_vec_info loop_vinfo;
 224
 225         vect_location = find_loop_location (loop);
 226         loop_vinfo = vect_analyze_loop (loop);
 227         loop->aux = loop_vinfo;
 228
 229         if (!loop_vinfo || !LOOP_VINFO_VECTORIZABLE_P (loop_vinfo))
 230           continue;
 231
 232         vect_transform_loop (loop_vinfo);
 233         num_vectorized_loops++;
 234       }
 235
 236   vect_location = UNKNOWN_LOC;
 237
 238   statistics_counter_event (cfun, "Vectorized loops", num_vectorized_loops);
 239   if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)
 240       || (vect_print_dump_info (REPORT_VECTORIZED_LOCATIONS)
 241           && num_vectorized_loops > 0))
 242     fprintf (vect_dump, "vectorized %u loops in function.\n",
 243              num_vectorized_loops);
 244
 245   /*  ----------- Finalize. -----------  */
 246
 247   BITMAP_FREE (vect_memsyms_to_rename);
 248
 249   for (i = 1; i < vect_loops_num; i++)
 250     {
 251       loop_vec_info loop_vinfo;
 252
 253       loop = get_loop (i);
 254       if (!loop)
 255         continue;
 256       loop_vinfo = (loop_vec_info) loop->aux;
 257       destroy_loop_vec_info (loop_vinfo, true);
 258       loop->aux = NULL;
 259     }
 260
 261   free_stmt_vec_info_vec ();
 262
 263   return num_vectorized_loops > 0 ? TODO_cleanup_cfg : 0;
 264 }
 265
 266
 267 /*  Entry point to basic block SLP phase.  */
 268
 269 static unsigned int
 270 execute_vect_slp (void)
 271 {
 272   basic_block bb;
 273
 274   /* Fix the verbosity level if not defined explicitly by the user.  */
 275   vect_set_dump_settings (true);
 276
 277   init_stmt_vec_info_vec ();
 278
 279   FOR_EACH_BB (bb)
 280     {
 281       vect_location = find_bb_location (bb);
 282
 283       if (vect_slp_analyze_bb (bb))
 284         {
 285           vect_slp_transform_bb (bb);
 286
 287           if (vect_print_dump_info (REPORT_VECTORIZED_LOCATIONS))
 288             fprintf (vect_dump, "basic block vectorized using SLP\n");
 289         }
 290     }
 291
 292   free_stmt_vec_info_vec ();
 293   return 0;
 294 }
 295
 296 static bool
 297 gate_vect_slp (void)
 298 {
 299   /* Apply SLP either if the vectorizer is on and the user didn't specify
 300      whether to run SLP or not, or if the SLP flag was set by the user.  */
 301   return ((flag_tree_vectorize != 0 && flag_tree_slp_vectorize != 0)
 302           || flag_tree_slp_vectorize == 1);
 303 }
 304
 305 struct gimple_opt_pass pass_slp_vectorize =
 306 {
 307  {
 308   GIMPLE_PASS,
 309   "slp",                                /* name */
 310   gate_vect_slp,                        /* gate */
 311   execute_vect_slp,                     /* execute */
 312   NULL,                                 /* sub */
 313   NULL,                                 /* next */
 314   0,                                    /* static_pass_number */
 315   TV_TREE_SLP_VECTORIZATION,            /* tv_id */
 316   PROP_ssa | PROP_cfg,                  /* properties_required */
 317   0,                                    /* properties_provided */
 318   0,                                    /* properties_destroyed */
 319   0,                                    /* todo_flags_start */
 320   TODO_ggc_collect
 321     | TODO_verify_ssa
 322     | TODO_dump_func
 323     | TODO_update_ssa
 324     | TODO_verify_stmts                 /* todo_flags_finish */
 325  }
 326 };
 327
 328
 329 /* Increase alignment of global arrays to improve vectorization potential.
 330    TODO:
 331    - Consider also structs that have an array field.
 332    - Use ipa analysis to prune arrays that can't be vectorized?
 333      This should involve global alignment analysis and in the future also
 334      array padding.  */
 335
 336 static unsigned int
 337 increase_alignment (void)
 338 {
 339   struct varpool_node *vnode;
 340
 341   /* Increase the alignment of all global arrays for vectorization.  */
 342   for (vnode = varpool_nodes_queue;
 343        vnode;
 344        vnode = vnode->next_needed)
 345     {
 346       tree vectype, decl = vnode->decl;
 347       unsigned int alignment;
 348
 349       if (TREE_CODE (TREE_TYPE (decl)) != ARRAY_TYPE)
 350         continue;
 351       vectype = get_vectype_for_scalar_type (TREE_TYPE (TREE_TYPE (decl)));
 352       if (!vectype)
 353         continue;
 354       alignment = TYPE_ALIGN (vectype);
 355       if (DECL_ALIGN (decl) >= alignment)
 356         continue;
 357
 358       if (vect_can_force_dr_alignment_p (decl, alignment))
 359         {
 360           DECL_ALIGN (decl) = TYPE_ALIGN (vectype);
 361           DECL_USER_ALIGN (decl) = 1;
 362           if (dump_file)
 363             {
 364               fprintf (dump_file, "Increasing alignment of decl: ");
 365               print_generic_expr (dump_file, decl, TDF_SLIM);
 366             }
 367         }
 368     }
 369   return 0;
 370 }
 371
 372
 373 static bool
 374 gate_increase_alignment (void)
 375 {
 376   return flag_section_anchors && flag_tree_vectorize;
 377 }
 378
 379
 380 struct simple_ipa_opt_pass pass_ipa_increase_alignment =
 381 {
 382  {
 383   SIMPLE_IPA_PASS,
 384   "increase_alignment",                 /* name */
 385   gate_increase_alignment,              /* gate */
 386   increase_alignment,                   /* execute */
 387   NULL,                                 /* sub */
 388   NULL,                                 /* next */
 389   0,                                    /* static_pass_number */
 390   TV_NONE,                              /* tv_id */
 391   0,                                    /* properties_required */
 392   0,                                    /* properties_provided */
 393   0,                                    /* properties_destroyed */
 394   0,                                    /* todo_flags_start */
 395   0                                     /* todo_flags_finish */
 396  }
 397 };