gcc/tree-vectorizer.c

   1 /* Vectorizer
   2    Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
   3    Free Software 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-pretty-print.h"
  66 #include "tree-flow.h"
  67 #include "tree-dump.h"
  68 #include "cfgloop.h"
  69 #include "cfglayout.h"
  70 #include "tree-vectorizer.h"
  71 #include "tree-pass.h"
  72 #include "timevar.h"
  73
  74 /* vect_dump will be set to stderr or dump_file if exist.  */
  75 FILE *vect_dump;
  76
  77 /* vect_verbosity_level set to an invalid value
  78    to mark that it's uninitialized.  */
  79 static enum verbosity_levels vect_verbosity_level = MAX_VERBOSITY_LEVEL;
  80 static enum verbosity_levels user_vect_verbosity_level = MAX_VERBOSITY_LEVEL;
  81
  82 /* Loop or bb location.  */
  83 LOC vect_location;
  84
  85 /* Vector mapping GIMPLE stmt to stmt_vec_info. */
  86 VEC(vec_void_p,heap) *stmt_vec_info_vec;
  87
  88 \f
  89
  90 /* Function vect_set_verbosity_level.
  91
  92    Called from opts.c upon detection of the
  93    -ftree-vectorizer-verbose=N option.  */
  94
  95 void
  96 vect_set_verbosity_level (const char *val)
  97 {
  98    unsigned int vl;
  99
 100    vl = atoi (val);
 101    if (vl < MAX_VERBOSITY_LEVEL)
 102      user_vect_verbosity_level = (enum verbosity_levels) vl;
 103    else
 104      user_vect_verbosity_level
 105       = (enum verbosity_levels) (MAX_VERBOSITY_LEVEL - 1);
 106 }
 107
 108
 109 /* Function vect_set_dump_settings.
 110
 111    Fix the verbosity level of the vectorizer if the
 112    requested level was not set explicitly using the flag
 113    -ftree-vectorizer-verbose=N.
 114    Decide where to print the debugging information (dump_file/stderr).
 115    If the user defined the verbosity level, but there is no dump file,
 116    print to stderr, otherwise print to the dump file.  */
 117
 118 static void
 119 vect_set_dump_settings (bool slp)
 120 {
 121   vect_dump = dump_file;
 122
 123   /* Check if the verbosity level was defined by the user:  */
 124   if (user_vect_verbosity_level != MAX_VERBOSITY_LEVEL)
 125     {
 126       vect_verbosity_level = user_vect_verbosity_level;
 127       /* Ignore user defined verbosity if dump flags require higher level of
 128          verbosity.  */
 129       if (dump_file)
 130         {
 131           if (((dump_flags & TDF_DETAILS)
 132                 && vect_verbosity_level >= REPORT_DETAILS)
 133                || ((dump_flags & TDF_STATS)
 134                     && vect_verbosity_level >= REPORT_UNVECTORIZED_LOCATIONS))
 135             return;
 136         }
 137       else
 138         {
 139           /* If there is no dump file, print to stderr in case of loop
 140              vectorization.  */
 141           if (!slp)
 142             vect_dump = stderr;
 143
 144           return;
 145         }
 146     }
 147
 148   /* User didn't specify verbosity level:  */
 149   if (dump_file && (dump_flags & TDF_DETAILS))
 150     vect_verbosity_level = REPORT_DETAILS;
 151   else if (dump_file && (dump_flags & TDF_STATS))
 152     vect_verbosity_level = REPORT_UNVECTORIZED_LOCATIONS;
 153   else
 154     vect_verbosity_level = REPORT_NONE;
 155
 156   gcc_assert (dump_file || vect_verbosity_level == REPORT_NONE);
 157 }
 158
 159
 160 /* Function debug_loop_details.
 161
 162    For vectorization debug dumps.  */
 163
 164 bool
 165 vect_print_dump_info (enum verbosity_levels vl)
 166 {
 167   if (vl > vect_verbosity_level)
 168     return false;
 169
 170   if (!current_function_decl || !vect_dump)
 171     return false;
 172
 173   if (vect_location == UNKNOWN_LOC)
 174     fprintf (vect_dump, "\n%s:%d: note: ",
 175              DECL_SOURCE_FILE (current_function_decl),
 176              DECL_SOURCE_LINE (current_function_decl));
 177   else
 178     fprintf (vect_dump, "\n%s:%d: note: ",
 179              LOC_FILE (vect_location), LOC_LINE (vect_location));
 180
 181   return true;
 182 }
 183
 184
 185 /* Function vectorize_loops.
 186
 187    Entry point to loop vectorization phase.  */
 188
 189 unsigned
 190 vectorize_loops (void)
 191 {
 192   unsigned int i;
 193   unsigned int num_vectorized_loops = 0;
 194   unsigned int vect_loops_num;
 195   loop_iterator li;
 196   struct loop *loop;
 197
 198   vect_loops_num = number_of_loops ();
 199
 200   /* Bail out if there are no loops.  */
 201   if (vect_loops_num <= 1)
 202     return 0;
 203
 204   /* Fix the verbosity level if not defined explicitly by the user.  */
 205   vect_set_dump_settings (false);
 206
 207   init_stmt_vec_info_vec ();
 208
 209   /*  ----------- Analyze loops. -----------  */
 210
 211   /* If some loop was duplicated, it gets bigger number
 212      than all previously defined loops. This fact allows us to run
 213      only over initial loops skipping newly generated ones.  */
 214   FOR_EACH_LOOP (li, loop, 0)
 215     if (optimize_loop_nest_for_speed_p (loop))
 216       {
 217         loop_vec_info loop_vinfo;
 218
 219         vect_location = find_loop_location (loop);
 220         loop_vinfo = vect_analyze_loop (loop);
 221         loop->aux = loop_vinfo;
 222
 223         if (!loop_vinfo || !LOOP_VINFO_VECTORIZABLE_P (loop_vinfo))
 224           continue;
 225
 226         vect_transform_loop (loop_vinfo);
 227         num_vectorized_loops++;
 228       }
 229
 230   vect_location = UNKNOWN_LOC;
 231
 232   statistics_counter_event (cfun, "Vectorized loops", num_vectorized_loops);
 233   if (vect_print_dump_info (REPORT_UNVECTORIZED_LOCATIONS)
 234       || (num_vectorized_loops > 0
 235           && vect_print_dump_info (REPORT_VECTORIZED_LOCATIONS)))
 236     fprintf (vect_dump, "vectorized %u loops in function.\n",
 237              num_vectorized_loops);
 238
 239   /*  ----------- Finalize. -----------  */
 240
 241   mark_sym_for_renaming (gimple_vop (cfun));
 242
 243   for (i = 1; i < vect_loops_num; i++)
 244     {
 245       loop_vec_info loop_vinfo;
 246
 247       loop = get_loop (i);
 248       if (!loop)
 249         continue;
 250       loop_vinfo = (loop_vec_info) loop->aux;
 251       destroy_loop_vec_info (loop_vinfo, true);
 252       loop->aux = NULL;
 253     }
 254
 255   free_stmt_vec_info_vec ();
 256
 257   return num_vectorized_loops > 0 ? TODO_cleanup_cfg : 0;
 258 }
 259
 260
 261 /*  Entry point to basic block SLP phase.  */
 262
 263 static unsigned int
 264 execute_vect_slp (void)
 265 {
 266   basic_block bb;
 267
 268   /* Fix the verbosity level if not defined explicitly by the user.  */
 269   vect_set_dump_settings (true);
 270
 271   init_stmt_vec_info_vec ();
 272
 273   FOR_EACH_BB (bb)
 274     {
 275       vect_location = find_bb_location (bb);
 276
 277       if (vect_slp_analyze_bb (bb))
 278         {
 279           vect_slp_transform_bb (bb);
 280
 281           if (vect_print_dump_info (REPORT_VECTORIZED_LOCATIONS))
 282             fprintf (vect_dump, "basic block vectorized using SLP\n");
 283         }
 284     }
 285
 286   free_stmt_vec_info_vec ();
 287   return 0;
 288 }
 289
 290 static bool
 291 gate_vect_slp (void)
 292 {
 293   /* Apply SLP either if the vectorizer is on and the user didn't specify
 294      whether to run SLP or not, or if the SLP flag was set by the user.  */
 295   return ((flag_tree_vectorize != 0 && flag_tree_slp_vectorize != 0)
 296           || flag_tree_slp_vectorize == 1);
 297 }
 298
 299 struct gimple_opt_pass pass_slp_vectorize =
 300 {
 301  {
 302   GIMPLE_PASS,
 303   "slp",                                /* name */
 304   gate_vect_slp,                        /* gate */
 305   execute_vect_slp,                     /* execute */
 306   NULL,                                 /* sub */
 307   NULL,                                 /* next */
 308   0,                                    /* static_pass_number */
 309   TV_TREE_SLP_VECTORIZATION,            /* tv_id */
 310   PROP_ssa | PROP_cfg,                  /* properties_required */
 311   0,                                    /* properties_provided */
 312   0,                                    /* properties_destroyed */
 313   0,                                    /* todo_flags_start */
 314   TODO_ggc_collect
 315     | TODO_verify_ssa
 316     | TODO_dump_func
 317     | TODO_update_ssa
 318     | TODO_verify_stmts                 /* todo_flags_finish */
 319  }
 320 };
 321
 322
 323 /* Increase alignment of global arrays to improve vectorization potential.
 324    TODO:
 325    - Consider also structs that have an array field.
 326    - Use ipa analysis to prune arrays that can't be vectorized?
 327      This should involve global alignment analysis and in the future also
 328      array padding.  */
 329
 330 static unsigned int
 331 increase_alignment (void)
 332 {
 333   struct varpool_node *vnode;
 334
 335   /* Increase the alignment of all global arrays for vectorization.  */
 336   for (vnode = varpool_nodes_queue;
 337        vnode;
 338        vnode = vnode->next_needed)
 339     {
 340       tree vectype, decl = vnode->decl;
 341       tree t;
 342       unsigned int alignment;
 343
 344       t = TREE_TYPE(decl);
 345       if (TREE_CODE (t) != ARRAY_TYPE)
 346         continue;
 347       vectype = get_vectype_for_scalar_type (strip_array_types (t));
 348       if (!vectype)
 349         continue;
 350       alignment = TYPE_ALIGN (vectype);
 351       if (DECL_ALIGN (decl) >= alignment)
 352         continue;
 353
 354       if (vect_can_force_dr_alignment_p (decl, alignment))
 355         {
 356           DECL_ALIGN (decl) = TYPE_ALIGN (vectype);
 357           DECL_USER_ALIGN (decl) = 1;
 358           if (dump_file)
 359             {
 360               fprintf (dump_file, "Increasing alignment of decl: ");
 361               print_generic_expr (dump_file, decl, TDF_SLIM);
 362               fprintf (dump_file, "\n");
 363             }
 364         }
 365     }
 366   return 0;
 367 }
 368
 369
 370 static bool
 371 gate_increase_alignment (void)
 372 {
 373   return flag_section_anchors && flag_tree_vectorize;
 374 }
 375
 376
 377 struct simple_ipa_opt_pass pass_ipa_increase_alignment =
 378 {
 379  {
 380   SIMPLE_IPA_PASS,
 381   "increase_alignment",                 /* name */
 382   gate_increase_alignment,              /* gate */
 383   increase_alignment,                   /* execute */
 384   NULL,                                 /* sub */
 385   NULL,                                 /* next */
 386   0,                                    /* static_pass_number */
 387   TV_NONE,                              /* tv_id */
 388   0,                                    /* properties_required */
 389   0,                                    /* properties_provided */
 390   0,                                    /* properties_destroyed */
 391   0,                                    /* todo_flags_start */
 392   0                                     /* todo_flags_finish */
 393  }
 394 };