gcc/graphite-blocking.c

   1 /* Heuristics and transform for loop blocking and strip mining on
   2    polyhedral representation.
   3
   4    Copyright (C) 2009-2014 Free Software Foundation, Inc.
   5    Contributed by Sebastian Pop <sebastian.pop@amd.com> and
   6    Pranav Garg  <pranav.garg2107@gmail.com>.
   7
   8 This file is part of GCC.
   9
  10 GCC is free software; you can redistribute it and/or modify
  11 it under the terms of the GNU General Public License as published by
  12 the Free Software Foundation; either version 3, or (at your option)
  13 any later version.
  14
  15 GCC is distributed in the hope that it will be useful,
  16 but WITHOUT ANY WARRANTY; without even the implied warranty of
  17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18 GNU General Public License for more details.
  19
  20 You should have received a copy of the GNU General Public License
  21 along with GCC; see the file COPYING3.  If not see
  22 <http://www.gnu.org/licenses/>.  */
  23
  24 #include "config.h"
  25
  26 #ifdef HAVE_isl
  27 #include <isl/set.h>
  28 #include <isl/map.h>
  29 #include <isl/union_map.h>
  30 #include <isl/constraint.h>
  31 #endif
  32
  33 #include "system.h"
  34 #include "coretypes.h"
  35 #include "tree.h"
  36 #include "predict.h"
  37 #include "vec.h"
  38 #include "hashtab.h"
  39 #include "hash-set.h"
  40 #include "machmode.h"
  41 #include "tm.h"
  42 #include "hard-reg-set.h"
  43 #include "input.h"
  44 #include "function.h"
  45 #include "dominance.h"
  46 #include "basic-block.h"
  47 #include "tree-ssa-alias.h"
  48 #include "internal-fn.h"
  49 #include "gimple-expr.h"
  50 #include "is-a.h"
  51 #include "gimple.h"
  52 #include "gimple-iterator.h"
  53 #include "tree-ssa-loop.h"
  54 #include "dumpfile.h"
  55 #include "cfgloop.h"
  56 #include "tree-chrec.h"
  57 #include "tree-data-ref.h"
  58 #include "sese.h"
  59
  60 #ifdef HAVE_isl
  61 #include "graphite-poly.h"
  62
  63
  64 /* Strip mines with a factor STRIDE the scattering (time) dimension
  65    around PBB at depth TIME_DEPTH.
  66
  67    The following example comes from the wiki page:
  68    http://gcc.gnu.org/wiki/Graphite/Strip_mine
  69
  70    The strip mine of a loop with a tile of 64 can be obtained with a
  71    scattering function as follows:
  72
  73    $ cat ./albert_strip_mine.cloog
  74    # language: C
  75    c
  76
  77    # parameter {n | n >= 0}
  78    1 3
  79    #  n  1
  80    1  1  0
  81    1
  82    n
  83
  84    1 # Number of statements:
  85
  86    1
  87    # {i | 0 <= i <= n}
  88    2 4
  89    #  i  n   1
  90    1  1  0   0
  91    1 -1  1   0
  92
  93    0  0  0
  94    1
  95    i
  96
  97    1 # Scattering functions
  98
  99    3 6
 100    #  NEW  OLD    i    n    1
 101    1  -64    0    1    0    0
 102    1   64    0   -1    0   63
 103    0    0    1   -1    0    0
 104
 105    1
 106    NEW  OLD
 107
 108    #the output of CLooG is like this:
 109    #$ cloog ./albert_strip_mine.cloog
 110    # for (NEW=0;NEW<=floord(n,64);NEW++) {
 111    #   for (OLD=max(64*NEW,0);OLD<=min(64*NEW+63,n);OLD++) {
 112    #     S1(i = OLD) ;
 113    #   }
 114    # }
 115 */
 116
 117 static void
 118 pbb_strip_mine_time_depth (poly_bb_p pbb, int time_depth, int stride)
 119 {
 120   isl_space *d;
 121   isl_constraint *c;
 122   int iter, strip;
 123   /* STRIP is the dimension that iterates with stride STRIDE.  */
 124   /* ITER is the dimension that enumerates single iterations inside
 125      one strip that has at most STRIDE iterations.  */
 126   strip = time_depth;
 127   iter = strip + 2;
 128
 129   pbb->transformed = isl_map_insert_dims (pbb->transformed, isl_dim_out,
 130                                           strip, 2);
 131
 132   /* Lower bound of the striped loop.  */
 133   d = isl_map_get_space (pbb->transformed);
 134   c = isl_inequality_alloc (isl_local_space_from_space (d));
 135   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, -stride);
 136   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, 1);
 137   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 138
 139   /* Upper bound of the striped loop.  */
 140   d = isl_map_get_space (pbb->transformed);
 141   c = isl_inequality_alloc (isl_local_space_from_space (d));
 142   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, stride);
 143   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, -1);
 144   c = isl_constraint_set_constant_si (c, stride - 1);
 145   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 146
 147   /* Static scheduling for ITER level.
 148      This is mandatory to keep the 2d + 1 canonical scheduling format.  */
 149   d = isl_map_get_space (pbb->transformed);
 150   c = isl_equality_alloc (isl_local_space_from_space (d));
 151   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip + 1, 1);
 152   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 153 }
 154
 155 /* Returns true when strip mining with STRIDE of the loop LST is
 156    profitable.  */
 157
 158 static bool
 159 lst_strip_mine_profitable_p (lst_p lst, int stride)
 160 {
 161   mpz_t niter, strip_stride;
 162   bool res;
 163
 164   gcc_assert (LST_LOOP_P (lst));
 165   mpz_init (strip_stride);
 166   mpz_init (niter);
 167
 168   mpz_set_si (strip_stride, stride);
 169   lst_niter_for_loop (lst, niter);
 170   res = (mpz_cmp (niter, strip_stride) > 0);
 171
 172   mpz_clear (strip_stride);
 173   mpz_clear (niter);
 174   return res;
 175 }
 176
 177 /* Strip-mines all the loops of LST with STRIDE.  Return the number of
 178    loops strip-mined.  */
 179
 180 static int
 181 lst_do_strip_mine_loop (lst_p lst, int depth, int stride)
 182 {
 183   int i;
 184   lst_p l;
 185   poly_bb_p pbb;
 186
 187   if (!lst)
 188     return 0;
 189
 190   if (LST_LOOP_P (lst))
 191     {
 192       int res = 0;
 193
 194       FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 195         res += lst_do_strip_mine_loop (l, depth, stride);
 196
 197       return res;
 198     }
 199
 200   pbb = LST_PBB (lst);
 201   pbb_strip_mine_time_depth (pbb, psct_dynamic_dim (pbb, depth), stride);
 202   return 1;
 203 }
 204
 205 /* Strip-mines all the loops of LST with STRIDE.  When STRIDE is zero,
 206    read the stride from the PARAM_LOOP_BLOCK_TILE_SIZE.  Return the
 207    number of strip-mined loops.
 208
 209    Strip mining transforms a loop
 210
 211    | for (i = 0; i < N; i++)
 212    |   S (i);
 213
 214    into the following loop nest:
 215
 216    | for (k = 0; k < N; k += STRIDE)
 217    |   for (j = 0; j < STRIDE; j++)
 218    |     S (i = k + j);
 219 */
 220
 221 static int
 222 lst_do_strip_mine (lst_p lst, int stride)
 223 {
 224   int i;
 225   lst_p l;
 226   int res = 0;
 227   int depth;
 228
 229   if (!stride)
 230     stride = PARAM_VALUE (PARAM_LOOP_BLOCK_TILE_SIZE);
 231
 232   if (!lst
 233       || !LST_LOOP_P (lst))
 234     return false;
 235
 236   FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 237     res += lst_do_strip_mine (l, stride);
 238
 239   depth = lst_depth (lst);
 240   if (depth >= 0
 241       && lst_strip_mine_profitable_p (lst, stride))
 242     {
 243       res += lst_do_strip_mine_loop (lst, lst_depth (lst), stride);
 244       lst_add_loop_under_loop (lst);
 245     }
 246
 247   return res;
 248 }
 249
 250 /* Strip mines all the loops in SCOP.  Returns the number of
 251    strip-mined loops.  */
 252
 253 int
 254 scop_do_strip_mine (scop_p scop, int stride)
 255 {
 256   return lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), stride);
 257 }
 258
 259 /* Loop blocks all the loops in SCOP.  Returns true when we manage to
 260    block some loops.  */
 261
 262 bool
 263 scop_do_block (scop_p scop)
 264 {
 265   store_scattering (scop);
 266
 267   /* If we don't strip mine at least two loops, or not interchange
 268      loops, the strip mine alone will not be profitable, and the
 269      transform is not a loop blocking: so revert the transform.  */
 270   if (lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), 0) < 2
 271       || scop_do_interchange (scop) == 0)
 272     {
 273       restore_scattering (scop);
 274       return false;
 275     }
 276
 277   if (dump_file && (dump_flags & TDF_DETAILS))
 278     fprintf (dump_file, "SCoP will be loop blocked.\n");
 279
 280   return true;
 281 }
 282
 283 #endif