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 #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 "input.h"
  36 #include "alias.h"
  37 #include "symtab.h"
  38 #include "options.h"
  39 #include "tree.h"
  40 #include "fold-const.h"
  41 #include "predict.h"
  42 #include "tm.h"
  43 #include "hard-reg-set.h"
  44 #include "input.h"
  45 #include "function.h"
  46 #include "dominance.h"
  47 #include "basic-block.h"
  48 #include "tree-ssa-alias.h"
  49 #include "internal-fn.h"
  50 #include "gimple-expr.h"
  51 #include "is-a.h"
  52 #include "gimple.h"
  53 #include "gimple-iterator.h"
  54 #include "tree-ssa-loop.h"
  55 #include "dumpfile.h"
  56 #include "cfgloop.h"
  57 #include "tree-chrec.h"
  58 #include "tree-data-ref.h"
  59 #include "sese.h"
  60
  61 #ifdef HAVE_isl
  62 #include "graphite-poly.h"
  63
  64
  65 /* Strip mines with a factor STRIDE the scattering (time) dimension
  66    around PBB at depth TIME_DEPTH.
  67
  68    The following example comes from the wiki page:
  69    http://gcc.gnu.org/wiki/Graphite/Strip_mine
  70
  71    The strip mine of a loop with a tile of 64 can be obtained with a
  72    scattering function as follows:
  73
  74    $ cat ./albert_strip_mine.cloog
  75    # language: C
  76    c
  77
  78    # parameter {n | n >= 0}
  79    1 3
  80    #  n  1
  81    1  1  0
  82    1
  83    n
  84
  85    1 # Number of statements:
  86
  87    1
  88    # {i | 0 <= i <= n}
  89    2 4
  90    #  i  n   1
  91    1  1  0   0
  92    1 -1  1   0
  93
  94    0  0  0
  95    1
  96    i
  97
  98    1 # Scattering functions
  99
 100    3 6
 101    #  NEW  OLD    i    n    1
 102    1  -64    0    1    0    0
 103    1   64    0   -1    0   63
 104    0    0    1   -1    0    0
 105
 106    1
 107    NEW  OLD
 108
 109    #the output of CLooG is like this:
 110    #$ cloog ./albert_strip_mine.cloog
 111    # for (NEW=0;NEW<=floord(n,64);NEW++) {
 112    #   for (OLD=max(64*NEW,0);OLD<=min(64*NEW+63,n);OLD++) {
 113    #     S1(i = OLD) ;
 114    #   }
 115    # }
 116 */
 117
 118 static void
 119 pbb_strip_mine_time_depth (poly_bb_p pbb, int time_depth, int stride)
 120 {
 121   isl_space *d;
 122   isl_constraint *c;
 123   int iter, strip;
 124   /* STRIP is the dimension that iterates with stride STRIDE.  */
 125   /* ITER is the dimension that enumerates single iterations inside
 126      one strip that has at most STRIDE iterations.  */
 127   strip = time_depth;
 128   iter = strip + 2;
 129
 130   pbb->transformed = isl_map_insert_dims (pbb->transformed, isl_dim_out,
 131                                           strip, 2);
 132
 133   /* Lower bound of the striped loop.  */
 134   d = isl_map_get_space (pbb->transformed);
 135   c = isl_inequality_alloc (isl_local_space_from_space (d));
 136   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, -stride);
 137   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, 1);
 138   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 139
 140   /* Upper bound of the striped loop.  */
 141   d = isl_map_get_space (pbb->transformed);
 142   c = isl_inequality_alloc (isl_local_space_from_space (d));
 143   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, stride);
 144   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, -1);
 145   c = isl_constraint_set_constant_si (c, stride - 1);
 146   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 147
 148   /* Static scheduling for ITER level.
 149      This is mandatory to keep the 2d + 1 canonical scheduling format.  */
 150   d = isl_map_get_space (pbb->transformed);
 151   c = isl_equality_alloc (isl_local_space_from_space (d));
 152   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip + 1, 1);
 153   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 154 }
 155
 156 /* Returns true when strip mining with STRIDE of the loop LST is
 157    profitable.  */
 158
 159 static bool
 160 lst_strip_mine_profitable_p (lst_p lst, int stride)
 161 {
 162   mpz_t niter, strip_stride;
 163   bool res;
 164
 165   gcc_assert (LST_LOOP_P (lst));
 166   mpz_init (strip_stride);
 167   mpz_init (niter);
 168
 169   mpz_set_si (strip_stride, stride);
 170   lst_niter_for_loop (lst, niter);
 171   res = (mpz_cmp (niter, strip_stride) > 0);
 172
 173   mpz_clear (strip_stride);
 174   mpz_clear (niter);
 175   return res;
 176 }
 177
 178 /* Strip-mines all the loops of LST with STRIDE.  Return the number of
 179    loops strip-mined.  */
 180
 181 static int
 182 lst_do_strip_mine_loop (lst_p lst, int depth, int stride)
 183 {
 184   int i;
 185   lst_p l;
 186   poly_bb_p pbb;
 187
 188   if (!lst)
 189     return 0;
 190
 191   if (LST_LOOP_P (lst))
 192     {
 193       int res = 0;
 194
 195       FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 196         res += lst_do_strip_mine_loop (l, depth, stride);
 197
 198       return res;
 199     }
 200
 201   pbb = LST_PBB (lst);
 202   pbb_strip_mine_time_depth (pbb, psct_dynamic_dim (pbb, depth), stride);
 203   return 1;
 204 }
 205
 206 /* Strip-mines all the loops of LST with STRIDE.  When STRIDE is zero,
 207    read the stride from the PARAM_LOOP_BLOCK_TILE_SIZE.  Return the
 208    number of strip-mined loops.
 209
 210    Strip mining transforms a loop
 211
 212    | for (i = 0; i < N; i++)
 213    |   S (i);
 214
 215    into the following loop nest:
 216
 217    | for (k = 0; k < N; k += STRIDE)
 218    |   for (j = 0; j < STRIDE; j++)
 219    |     S (i = k + j);
 220 */
 221
 222 static int
 223 lst_do_strip_mine (lst_p lst, int stride)
 224 {
 225   int i;
 226   lst_p l;
 227   int res = 0;
 228   int depth;
 229
 230   if (!stride)
 231     stride = PARAM_VALUE (PARAM_LOOP_BLOCK_TILE_SIZE);
 232
 233   if (!lst
 234       || !LST_LOOP_P (lst))
 235     return false;
 236
 237   FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 238     res += lst_do_strip_mine (l, stride);
 239
 240   depth = lst_depth (lst);
 241   if (depth >= 0
 242       && lst_strip_mine_profitable_p (lst, stride))
 243     {
 244       res += lst_do_strip_mine_loop (lst, lst_depth (lst), stride);
 245       lst_add_loop_under_loop (lst);
 246     }
 247
 248   return res;
 249 }
 250
 251 /* Strip mines all the loops in SCOP.  Returns the number of
 252    strip-mined loops.  */
 253
 254 int
 255 scop_do_strip_mine (scop_p scop, int stride)
 256 {
 257   return lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), stride);
 258 }
 259
 260 /* Loop blocks all the loops in SCOP.  Returns true when we manage to
 261    block some loops.  */
 262
 263 bool
 264 scop_do_block (scop_p scop)
 265 {
 266   store_scattering (scop);
 267
 268   /* If we don't strip mine at least two loops, or not interchange
 269      loops, the strip mine alone will not be profitable, and the
 270      transform is not a loop blocking: so revert the transform.  */
 271   if (lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), 0) < 2
 272       || scop_do_interchange (scop) == 0)
 273     {
 274       restore_scattering (scop);
 275       return false;
 276     }
 277
 278   if (dump_file && (dump_flags & TDF_DETAILS))
 279     fprintf (dump_file, "SCoP will be loop blocked.\n");
 280
 281   return true;
 282 }
 283
 284 #endif