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
  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 enum verbosity_levels vect_verbosity_level = MAX_VERBOSITY_LEVEL;
  78
  79 /* Loop location.  */
  80 LOC vect_loop_location;
  81
  82 /* Bitmap of virtual variables to be renamed.  */
  83 bitmap vect_memsyms_to_rename;
  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 toplev.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      vect_verbosity_level = vl;
 103    else
 104      vect_verbosity_level = MAX_VERBOSITY_LEVEL - 1;
 105 }
 106
 107
 108 /* Function vect_set_dump_settings.
 109
 110    Fix the verbosity level of the vectorizer if the
 111    requested level was not set explicitly using the flag
 112    -ftree-vectorizer-verbose=N.
 113    Decide where to print the debugging information (dump_file/stderr).
 114    If the user defined the verbosity level, but there is no dump file,
 115    print to stderr, otherwise print to the dump file.  */
 116
 117 static void
 118 vect_set_dump_settings (void)
 119 {
 120   vect_dump = dump_file;
 121
 122   /* Check if the verbosity level was defined by the user:  */
 123   if (vect_verbosity_level != MAX_VERBOSITY_LEVEL)
 124     {
 125       /* If there is no dump file, print to stderr.  */
 126       if (!dump_file)
 127         vect_dump = stderr;
 128       return;
 129     }
 130
 131   /* User didn't specify verbosity level:  */
 132   if (dump_file && (dump_flags & TDF_DETAILS))
 133     vect_verbosity_level = REPORT_DETAILS;
 134   else if (dump_file && (dump_flags & TDF_STATS))
 135     vect_verbosity_level = REPORT_UNVECTORIZED_LOOPS;
 136   else
 137     vect_verbosity_level = REPORT_NONE;
 138
 139   gcc_assert (dump_file || vect_verbosity_level == REPORT_NONE);
 140 }
 141
 142
 143 /* Function debug_loop_details.
 144
 145    For vectorization debug dumps.  */
 146
 147 bool
 148 vect_print_dump_info (enum verbosity_levels vl)
 149 {
 150   if (vl > vect_verbosity_level)
 151     return false;
 152
 153   if (!current_function_decl || !vect_dump)
 154     return false;
 155
 156   if (vect_loop_location == UNKNOWN_LOC)
 157     fprintf (vect_dump, "\n%s:%d: note: ",
 158              DECL_SOURCE_FILE (current_function_decl),
 159              DECL_SOURCE_LINE (current_function_decl));
 160   else
 161     fprintf (vect_dump, "\n%s:%d: note: ",
 162              LOC_FILE (vect_loop_location), LOC_LINE (vect_loop_location));
 163
 164   return true;
 165 }
 166
 167
 168 /* Function vectorize_loops.
 169
 170    Entry Point to loop vectorization phase.  */
 171
 172 unsigned
 173 vectorize_loops (void)
 174 {
 175   unsigned int i;
 176   unsigned int num_vectorized_loops = 0;
 177   unsigned int vect_loops_num;
 178   loop_iterator li;
 179   struct loop *loop;
 180
 181   vect_loops_num = number_of_loops ();
 182
 183   /* Bail out if there are no loops.  */
 184   if (vect_loops_num <= 1)
 185     return 0;
 186
 187   /* Fix the verbosity level if not defined explicitly by the user.  */
 188   vect_set_dump_settings ();
 189
 190   /* Allocate the bitmap that records which virtual variables that
 191      need to be renamed.  */
 192   vect_memsyms_to_rename = BITMAP_ALLOC (NULL);
 193
 194   init_stmt_vec_info_vec ();
 195
 196   /*  ----------- Analyze loops. -----------  */
 197
 198   /* If some loop was duplicated, it gets bigger number
 199      than all previously defined loops. This fact allows us to run
 200      only over initial loops skipping newly generated ones.  */
 201   FOR_EACH_LOOP (li, loop, 0)
 202     if (optimize_loop_nest_for_speed_p (loop))
 203       {
 204         loop_vec_info loop_vinfo;
 205
 206         vect_loop_location = find_loop_location (loop);
 207         loop_vinfo = vect_analyze_loop (loop);
 208         loop->aux = loop_vinfo;
 209
 210         if (!loop_vinfo || !LOOP_VINFO_VECTORIZABLE_P (loop_vinfo))
 211           continue;
 212
 213         vect_transform_loop (loop_vinfo);
 214         num_vectorized_loops++;
 215       }
 216   vect_loop_location = UNKNOWN_LOC;
 217
 218   statistics_counter_event (cfun, "Vectorized loops", num_vectorized_loops);
 219   if (vect_print_dump_info (REPORT_UNVECTORIZED_LOOPS)
 220       || (vect_print_dump_info (REPORT_VECTORIZED_LOOPS)
 221           && num_vectorized_loops > 0))
 222     fprintf (vect_dump, "vectorized %u loops in function.\n",
 223              num_vectorized_loops);
 224
 225   /*  ----------- Finalize. -----------  */
 226
 227   BITMAP_FREE (vect_memsyms_to_rename);
 228
 229   for (i = 1; i < vect_loops_num; i++)
 230     {
 231       loop_vec_info loop_vinfo;
 232
 233       loop = get_loop (i);
 234       if (!loop)
 235         continue;
 236       loop_vinfo = (loop_vec_info) loop->aux;
 237       destroy_loop_vec_info (loop_vinfo, true);
 238       loop->aux = NULL;
 239     }
 240
 241   free_stmt_vec_info_vec ();
 242
 243   return num_vectorized_loops > 0 ? TODO_cleanup_cfg : 0;
 244 }
 245
 246
 247 /* Increase alignment of global arrays to improve vectorization potential.
 248    TODO:
 249    - Consider also structs that have an array field.
 250    - Use ipa analysis to prune arrays that can't be vectorized?
 251      This should involve global alignment analysis and in the future also
 252      array padding.  */
 253
 254 static unsigned int
 255 increase_alignment (void)
 256 {
 257   struct varpool_node *vnode;
 258
 259   /* Increase the alignment of all global arrays for vectorization.  */
 260   for (vnode = varpool_nodes_queue;
 261        vnode;
 262        vnode = vnode->next_needed)
 263     {
 264       tree vectype, decl = vnode->decl;
 265       unsigned int alignment;
 266
 267       if (TREE_CODE (TREE_TYPE (decl)) != ARRAY_TYPE)
 268         continue;
 269       vectype = get_vectype_for_scalar_type (TREE_TYPE (TREE_TYPE (decl)));
 270       if (!vectype)
 271         continue;
 272       alignment = TYPE_ALIGN (vectype);
 273       if (DECL_ALIGN (decl) >= alignment)
 274         continue;
 275
 276       if (vect_can_force_dr_alignment_p (decl, alignment))
 277         {
 278           DECL_ALIGN (decl) = TYPE_ALIGN (vectype);
 279           DECL_USER_ALIGN (decl) = 1;
 280           if (dump_file)
 281             {
 282               fprintf (dump_file, "Increasing alignment of decl: ");
 283               print_generic_expr (dump_file, decl, TDF_SLIM);
 284             }
 285         }
 286     }
 287   return 0;
 288 }
 289
 290
 291 static bool
 292 gate_increase_alignment (void)
 293 {
 294   return flag_section_anchors && flag_tree_vectorize;
 295 }
 296
 297
 298 struct simple_ipa_opt_pass pass_ipa_increase_alignment =
 299 {
 300  {
 301   SIMPLE_IPA_PASS,
 302   "increase_alignment",                 /* name */
 303   gate_increase_alignment,              /* gate */
 304   increase_alignment,                   /* execute */
 305   NULL,                                 /* sub */
 306   NULL,                                 /* next */
 307   0,                                    /* static_pass_number */
 308   0,                                    /* tv_id */
 309   0,                                    /* properties_required */
 310   0,                                    /* properties_provided */
 311   0,                                    /* properties_destroyed */
 312   0,                                    /* todo_flags_start */
 313   0                                     /* todo_flags_finish */
 314  }
 315 };
 316
 317