gcc/tree-vectorizer.c

   1 /* Vectorizer
   2    Copyright (C) 2003-2013 Free Software Foundation, Inc.
   3    Contributed by Dorit Naishlos <dorit@il.ibm.com>
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify it under
   8 the terms of the GNU General Public License as published by the Free
   9 Software Foundation; either version 3, or (at your option) any later
  10 version.
  11
  12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15 for more details.
  16
  17 You should have received a copy of the GNU General Public License
  18 along with GCC; see the file COPYING3.  If not see
  19 <http://www.gnu.org/licenses/>.  */
  20
  21 /* Loop and basic block vectorizer.
  22
  23   This file contains drivers for the three vectorizers:
  24   (1) loop vectorizer (inter-iteration parallelism),
  25   (2) loop-aware SLP (intra-iteration parallelism) (invoked by the loop
  26       vectorizer)
  27   (3) BB vectorizer (out-of-loops), aka SLP
  28
  29   The rest of the vectorizer's code is organized as follows:
  30   - tree-vect-loop.c - loop specific parts such as reductions, etc. These are
  31     used by drivers (1) and (2).
  32   - tree-vect-loop-manip.c - vectorizer's loop control-flow utilities, used by
  33     drivers (1) and (2).
  34   - tree-vect-slp.c - BB vectorization specific analysis and transformation,
  35     used by drivers (2) and (3).
  36   - tree-vect-stmts.c - statements analysis and transformation (used by all).
  37   - tree-vect-data-refs.c - vectorizer specific data-refs analysis and
  38     manipulations (used by all).
  39   - tree-vect-patterns.c - vectorizable code patterns detector (used by all)
  40
  41   Here's a poor attempt at illustrating that:
  42
  43      tree-vectorizer.c:
  44      loop_vect()  loop_aware_slp()  slp_vect()
  45           |        /           \          /
  46           |       /             \        /
  47           tree-vect-loop.c  tree-vect-slp.c
  48                 | \      \  /      /   |
  49                 |  \      \/      /    |
  50                 |   \     /\     /     |
  51                 |    \   /  \   /      |
  52          tree-vect-stmts.c  tree-vect-data-refs.c
  53                        \      /
  54                     tree-vect-patterns.c
  55 */
  56
  57 #include "config.h"
  58 #include "system.h"
  59 #include "coretypes.h"
  60 #include "dumpfile.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 /* Loop or bb location.  */
  71 LOC vect_location;
  72
  73 /* Vector mapping GIMPLE stmt to stmt_vec_info. */
  74 vec<vec_void_p> stmt_vec_info_vec;
  75
  76 \f
  77 /* Function vectorize_loops.
  78
  79    Entry point to loop vectorization phase.  */
  80
  81 unsigned
  82 vectorize_loops (void)
  83 {
  84   unsigned int i;
  85   unsigned int num_vectorized_loops = 0;
  86   unsigned int vect_loops_num;
  87   loop_iterator li;
  88   struct loop *loop;
  89
  90   vect_loops_num = number_of_loops (cfun);
  91
  92   /* Bail out if there are no loops.  */
  93   if (vect_loops_num <= 1)
  94     return 0;
  95
  96   init_stmt_vec_info_vec ();
  97
  98   /*  ----------- Analyze loops. -----------  */
  99
 100   /* If some loop was duplicated, it gets bigger number
 101      than all previously defined loops.  This fact allows us to run
 102      only over initial loops skipping newly generated ones.  */
 103   FOR_EACH_LOOP (li, loop, 0)
 104     if (optimize_loop_nest_for_speed_p (loop))
 105       {
 106         loop_vec_info loop_vinfo;
 107         vect_location = find_loop_location (loop);
 108         if (LOCATION_LOCUS (vect_location) != UNKNOWN_LOC
 109             && dump_enabled_p ())
 110           dump_printf (MSG_NOTE, "\nAnalyzing loop at %s:%d\n",
 111                        LOC_FILE (vect_location), LOC_LINE (vect_location));
 112
 113         loop_vinfo = vect_analyze_loop (loop);
 114         loop->aux = loop_vinfo;
 115
 116         if (!loop_vinfo || !LOOP_VINFO_VECTORIZABLE_P (loop_vinfo))
 117           continue;
 118
 119         if (LOCATION_LOCUS (vect_location) != UNKNOWN_LOC
 120             && dump_enabled_p ())
 121           dump_printf (MSG_NOTE, "\n\nVectorizing loop at %s:%d\n",
 122                        LOC_FILE (vect_location), LOC_LINE (vect_location));
 123         vect_transform_loop (loop_vinfo);
 124         num_vectorized_loops++;
 125       }
 126
 127   vect_location = UNKNOWN_LOC;
 128
 129   statistics_counter_event (cfun, "Vectorized loops", num_vectorized_loops);
 130   if (dump_enabled_p ()
 131       || (num_vectorized_loops > 0 && dump_enabled_p ()))
 132     dump_printf_loc (MSG_NOTE, vect_location,
 133                      "vectorized %u loops in function.\n",
 134                      num_vectorized_loops);
 135
 136   /*  ----------- Finalize. -----------  */
 137
 138   for (i = 1; i < vect_loops_num; i++)
 139     {
 140       loop_vec_info loop_vinfo;
 141
 142       loop = get_loop (cfun, i);
 143       if (!loop)
 144         continue;
 145       loop_vinfo = (loop_vec_info) loop->aux;
 146       destroy_loop_vec_info (loop_vinfo, true);
 147       loop->aux = NULL;
 148     }
 149
 150   free_stmt_vec_info_vec ();
 151
 152   if (num_vectorized_loops > 0)
 153     {
 154       /* If we vectorized any loop only virtual SSA form needs to be updated.
 155          ???  Also while we try hard to update loop-closed SSA form we fail
 156          to properly do this in some corner-cases (see PR56286).  */
 157       rewrite_into_loop_closed_ssa (NULL, TODO_update_ssa_only_virtuals);
 158       return TODO_cleanup_cfg;
 159     }
 160
 161   return 0;
 162 }
 163
 164
 165 /*  Entry point to basic block SLP phase.  */
 166
 167 static unsigned int
 168 execute_vect_slp (void)
 169 {
 170   basic_block bb;
 171
 172   init_stmt_vec_info_vec ();
 173
 174   FOR_EACH_BB (bb)
 175     {
 176       vect_location = find_bb_location (bb);
 177
 178       if (vect_slp_analyze_bb (bb))
 179         {
 180           vect_slp_transform_bb (bb);
 181           if (dump_enabled_p ())
 182             dump_printf_loc (MSG_NOTE, vect_location,
 183                              "basic block vectorized using SLP\n");
 184         }
 185     }
 186
 187   free_stmt_vec_info_vec ();
 188   return 0;
 189 }
 190
 191 static bool
 192 gate_vect_slp (void)
 193 {
 194   /* Apply SLP either if the vectorizer is on and the user didn't specify
 195      whether to run SLP or not, or if the SLP flag was set by the user.  */
 196   return ((flag_tree_vectorize != 0 && flag_tree_slp_vectorize != 0)
 197           || flag_tree_slp_vectorize == 1);
 198 }
 199
 200 struct gimple_opt_pass pass_slp_vectorize =
 201 {
 202  {
 203   GIMPLE_PASS,
 204   "slp",                                /* name */
 205   OPTGROUP_LOOP
 206   | OPTGROUP_VEC,                       /* optinfo_flags */
 207   gate_vect_slp,                        /* gate */
 208   execute_vect_slp,                     /* execute */
 209   NULL,                                 /* sub */
 210   NULL,                                 /* next */
 211   0,                                    /* static_pass_number */
 212   TV_TREE_SLP_VECTORIZATION,            /* tv_id */
 213   PROP_ssa | PROP_cfg,                  /* properties_required */
 214   0,                                    /* properties_provided */
 215   0,                                    /* properties_destroyed */
 216   0,                                    /* todo_flags_start */
 217   TODO_verify_ssa
 218     | TODO_update_ssa
 219     | TODO_verify_stmts                 /* todo_flags_finish */
 220  }
 221 };
 222
 223
 224 /* Increase alignment of global arrays to improve vectorization potential.
 225    TODO:
 226    - Consider also structs that have an array field.
 227    - Use ipa analysis to prune arrays that can't be vectorized?
 228      This should involve global alignment analysis and in the future also
 229      array padding.  */
 230
 231 static unsigned int
 232 increase_alignment (void)
 233 {
 234   struct varpool_node *vnode;
 235
 236   vect_location = UNKNOWN_LOC;
 237
 238   /* Increase the alignment of all global arrays for vectorization.  */
 239   FOR_EACH_DEFINED_VARIABLE (vnode)
 240     {
 241       tree vectype, decl = vnode->symbol.decl;
 242       tree t;
 243       unsigned int alignment;
 244
 245       t = TREE_TYPE(decl);
 246       if (TREE_CODE (t) != ARRAY_TYPE)
 247         continue;
 248       vectype = get_vectype_for_scalar_type (strip_array_types (t));
 249       if (!vectype)
 250         continue;
 251       alignment = TYPE_ALIGN (vectype);
 252       if (DECL_ALIGN (decl) >= alignment)
 253         continue;
 254
 255       if (vect_can_force_dr_alignment_p (decl, alignment))
 256         {
 257           DECL_ALIGN (decl) = TYPE_ALIGN (vectype);
 258           DECL_USER_ALIGN (decl) = 1;
 259           dump_printf (MSG_NOTE, "Increasing alignment of decl: ");
 260           dump_generic_expr (MSG_NOTE, TDF_SLIM, decl);
 261           dump_printf (MSG_NOTE, "\n");
 262         }
 263     }
 264   return 0;
 265 }
 266
 267
 268 static bool
 269 gate_increase_alignment (void)
 270 {
 271   return flag_section_anchors && flag_tree_vectorize;
 272 }
 273
 274
 275 struct simple_ipa_opt_pass pass_ipa_increase_alignment =
 276 {
 277  {
 278   SIMPLE_IPA_PASS,
 279   "increase_alignment",                 /* name */
 280   OPTGROUP_LOOP
 281   | OPTGROUP_VEC,                       /* optinfo_flags */
 282   gate_increase_alignment,              /* gate */
 283   increase_alignment,                   /* execute */
 284   NULL,                                 /* sub */
 285   NULL,                                 /* next */
 286   0,                                    /* static_pass_number */
 287   TV_IPA_OPT,                           /* tv_id */
 288   0,                                    /* properties_required */
 289   0,                                    /* properties_provided */
 290   0,                                    /* properties_destroyed */
 291   0,                                    /* todo_flags_start */
 292   0                                     /* todo_flags_finish */
 293  }
 294 };