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