gcc/graphite-blocking.c

   1 /* Heuristics and transform for loop blocking and strip mining on
   2    polyhedral representation.
   3
   4    Copyright (C) 2009-2015 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 /* Workaround for GMP 5.1.3 bug, see PR56019.  */
  28 #include <stddef.h>
  29
  30 #include <isl/set.h>
  31 #include <isl/map.h>
  32 #include <isl/union_map.h>
  33 #include <isl/constraint.h>
  34 #endif
  35
  36 #include "system.h"
  37 #include "coretypes.h"
  38 #include "alias.h"
  39 #include "backend.h"
  40 #include "tree.h"
  41 #include "gimple.h"
  42 #include "hard-reg-set.h"
  43 #include "options.h"
  44 #include "fold-const.h"
  45 #include "dominance.h"
  46 #include "internal-fn.h"
  47 #include "gimple-iterator.h"
  48 #include "tree-ssa-loop.h"
  49 #include "dumpfile.h"
  50 #include "cfgloop.h"
  51 #include "tree-chrec.h"
  52 #include "tree-data-ref.h"
  53 #include "sese.h"
  54
  55 #ifdef HAVE_isl
  56 #include "graphite-poly.h"
  57
  58
  59 /* Strip mines with a factor STRIDE the scattering (time) dimension
  60    around PBB at depth TIME_DEPTH.
  61
  62    The following example comes from the wiki page:
  63    http://gcc.gnu.org/wiki/Graphite/Strip_mine
  64
  65    The strip mine of a loop with a tile of 64 can be obtained with a
  66    scattering function as follows:
  67
  68    $ cat ./albert_strip_mine.cloog
  69    # language: C
  70    c
  71
  72    # parameter {n | n >= 0}
  73    1 3
  74    #  n  1
  75    1  1  0
  76    1
  77    n
  78
  79    1 # Number of statements:
  80
  81    1
  82    # {i | 0 <= i <= n}
  83    2 4
  84    #  i  n   1
  85    1  1  0   0
  86    1 -1  1   0
  87
  88    0  0  0
  89    1
  90    i
  91
  92    1 # Scattering functions
  93
  94    3 6
  95    #  NEW  OLD    i    n    1
  96    1  -64    0    1    0    0
  97    1   64    0   -1    0   63
  98    0    0    1   -1    0    0
  99
 100    1
 101    NEW  OLD
 102
 103    #the output of CLooG is like this:
 104    #$ cloog ./albert_strip_mine.cloog
 105    # for (NEW=0;NEW<=floord(n,64);NEW++) {
 106    #   for (OLD=max(64*NEW,0);OLD<=min(64*NEW+63,n);OLD++) {
 107    #     S1(i = OLD) ;
 108    #   }
 109    # }
 110 */
 111
 112 static void
 113 pbb_strip_mine_time_depth (poly_bb_p pbb, int time_depth, int stride)
 114 {
 115   isl_space *d;
 116   isl_constraint *c;
 117   int iter, strip;
 118   /* STRIP is the dimension that iterates with stride STRIDE.  */
 119   /* ITER is the dimension that enumerates single iterations inside
 120      one strip that has at most STRIDE iterations.  */
 121   strip = time_depth;
 122   iter = strip + 2;
 123
 124   pbb->transformed = isl_map_insert_dims (pbb->transformed, isl_dim_out,
 125                                           strip, 2);
 126
 127   /* Lower bound of the striped loop.  */
 128   d = isl_map_get_space (pbb->transformed);
 129   c = isl_inequality_alloc (isl_local_space_from_space (d));
 130   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, -stride);
 131   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, 1);
 132   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 133
 134   /* Upper bound of the striped loop.  */
 135   d = isl_map_get_space (pbb->transformed);
 136   c = isl_inequality_alloc (isl_local_space_from_space (d));
 137   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, stride);
 138   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, -1);
 139   c = isl_constraint_set_constant_si (c, stride - 1);
 140   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 141
 142   /* Static scheduling for ITER level.
 143      This is mandatory to keep the 2d + 1 canonical scheduling format.  */
 144   d = isl_map_get_space (pbb->transformed);
 145   c = isl_equality_alloc (isl_local_space_from_space (d));
 146   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip + 1, 1);
 147   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 148 }
 149
 150 /* Returns true when strip mining with STRIDE of the loop LST is
 151    profitable.  */
 152
 153 static bool
 154 lst_strip_mine_profitable_p (lst_p lst, int stride)
 155 {
 156   mpz_t niter, strip_stride;
 157   bool res;
 158
 159   gcc_assert (LST_LOOP_P (lst));
 160   mpz_init (strip_stride);
 161   mpz_init (niter);
 162
 163   mpz_set_si (strip_stride, stride);
 164   lst_niter_for_loop (lst, niter);
 165   res = (mpz_cmp (niter, strip_stride) > 0);
 166
 167   mpz_clear (strip_stride);
 168   mpz_clear (niter);
 169   return res;
 170 }
 171
 172 /* Strip-mines all the loops of LST with STRIDE.  Return the number of
 173    loops strip-mined.  */
 174
 175 static int
 176 lst_do_strip_mine_loop (lst_p lst, int depth, int stride)
 177 {
 178   int i;
 179   lst_p l;
 180   poly_bb_p pbb;
 181
 182   if (!lst)
 183     return 0;
 184
 185   if (LST_LOOP_P (lst))
 186     {
 187       int res = 0;
 188
 189       FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 190         res += lst_do_strip_mine_loop (l, depth, stride);
 191
 192       return res;
 193     }
 194
 195   pbb = LST_PBB (lst);
 196   pbb_strip_mine_time_depth (pbb, psct_dynamic_dim (pbb, depth), stride);
 197   return 1;
 198 }
 199
 200 /* Strip-mines all the loops of LST with STRIDE.  When STRIDE is zero,
 201    read the stride from the PARAM_LOOP_BLOCK_TILE_SIZE.  Return the
 202    number of strip-mined loops.
 203
 204    Strip mining transforms a loop
 205
 206    | for (i = 0; i < N; i++)
 207    |   S (i);
 208
 209    into the following loop nest:
 210
 211    | for (k = 0; k < N; k += STRIDE)
 212    |   for (j = 0; j < STRIDE; j++)
 213    |     S (i = k + j);
 214 */
 215
 216 static int
 217 lst_do_strip_mine (lst_p lst, int stride)
 218 {
 219   int i;
 220   lst_p l;
 221   int res = 0;
 222   int depth;
 223
 224   if (!stride)
 225     stride = PARAM_VALUE (PARAM_LOOP_BLOCK_TILE_SIZE);
 226
 227   if (!lst
 228       || !LST_LOOP_P (lst))
 229     return false;
 230
 231   FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 232     res += lst_do_strip_mine (l, stride);
 233
 234   depth = lst_depth (lst);
 235   if (depth >= 0
 236       && lst_strip_mine_profitable_p (lst, stride))
 237     {
 238       res += lst_do_strip_mine_loop (lst, lst_depth (lst), stride);
 239       lst_add_loop_under_loop (lst);
 240     }
 241
 242   return res;
 243 }
 244
 245 /* Strip mines all the loops in SCOP.  Returns the number of
 246    strip-mined loops.  */
 247
 248 int
 249 scop_do_strip_mine (scop_p scop, int stride)
 250 {
 251   return lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), stride);
 252 }
 253
 254 /* Loop blocks all the loops in SCOP.  Returns true when we manage to
 255    block some loops.  */
 256
 257 bool
 258 scop_do_block (scop_p scop)
 259 {
 260   store_scattering (scop);
 261
 262   /* If we don't strip mine at least two loops, or not interchange
 263      loops, the strip mine alone will not be profitable, and the
 264      transform is not a loop blocking: so revert the transform.  */
 265   if (lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), 0) < 2
 266       || scop_do_interchange (scop) == 0)
 267     {
 268       restore_scattering (scop);
 269       return false;
 270     }
 271
 272   if (dump_file && (dump_flags & TDF_DETAILS))
 273     fprintf (dump_file, "SCoP will be loop blocked.\n");
 274
 275   return true;
 276 }
 277
 278 #endif