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