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