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 #ifdef HAVE_cloog
  32 #include <cloog/cloog.h>
  33 #include <cloog/isl/domain.h>
  34 #endif
  35 #endif
  36
  37 #include "system.h"
  38 #include "coretypes.h"
  39 #include "tree.h"
  40 #include "predict.h"
  41 #include "vec.h"
  42 #include "hashtab.h"
  43 #include "hash-set.h"
  44 #include "machmode.h"
  45 #include "tm.h"
  46 #include "hard-reg-set.h"
  47 #include "input.h"
  48 #include "function.h"
  49 #include "dominance.h"
  50 #include "basic-block.h"
  51 #include "tree-ssa-alias.h"
  52 #include "internal-fn.h"
  53 #include "gimple-expr.h"
  54 #include "is-a.h"
  55 #include "gimple.h"
  56 #include "gimple-iterator.h"
  57 #include "tree-ssa-loop.h"
  58 #include "dumpfile.h"
  59 #include "cfgloop.h"
  60 #include "tree-chrec.h"
  61 #include "tree-data-ref.h"
  62 #include "sese.h"
  63
  64 #ifdef HAVE_isl
  65 #include "graphite-poly.h"
  66
  67
  68 /* Strip mines with a factor STRIDE the scattering (time) dimension
  69    around PBB at depth TIME_DEPTH.
  70
  71    The following example comes from the wiki page:
  72    http://gcc.gnu.org/wiki/Graphite/Strip_mine
  73
  74    The strip mine of a loop with a tile of 64 can be obtained with a
  75    scattering function as follows:
  76
  77    $ cat ./albert_strip_mine.cloog
  78    # language: C
  79    c
  80
  81    # parameter {n | n >= 0}
  82    1 3
  83    #  n  1
  84    1  1  0
  85    1
  86    n
  87
  88    1 # Number of statements:
  89
  90    1
  91    # {i | 0 <= i <= n}
  92    2 4
  93    #  i  n   1
  94    1  1  0   0
  95    1 -1  1   0
  96
  97    0  0  0
  98    1
  99    i
 100
 101    1 # Scattering functions
 102
 103    3 6
 104    #  NEW  OLD    i    n    1
 105    1  -64    0    1    0    0
 106    1   64    0   -1    0   63
 107    0    0    1   -1    0    0
 108
 109    1
 110    NEW  OLD
 111
 112    #the output of CLooG is like this:
 113    #$ cloog ./albert_strip_mine.cloog
 114    # for (NEW=0;NEW<=floord(n,64);NEW++) {
 115    #   for (OLD=max(64*NEW,0);OLD<=min(64*NEW+63,n);OLD++) {
 116    #     S1(i = OLD) ;
 117    #   }
 118    # }
 119 */
 120
 121 static void
 122 pbb_strip_mine_time_depth (poly_bb_p pbb, int time_depth, int stride)
 123 {
 124   isl_space *d;
 125   isl_constraint *c;
 126   int iter, strip;
 127   /* STRIP is the dimension that iterates with stride STRIDE.  */
 128   /* ITER is the dimension that enumerates single iterations inside
 129      one strip that has at most STRIDE iterations.  */
 130   strip = time_depth;
 131   iter = strip + 2;
 132
 133   pbb->transformed = isl_map_insert_dims (pbb->transformed, isl_dim_out,
 134                                           strip, 2);
 135
 136   /* Lower bound of the striped loop.  */
 137   d = isl_map_get_space (pbb->transformed);
 138   c = isl_inequality_alloc (isl_local_space_from_space (d));
 139   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, -stride);
 140   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, 1);
 141   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 142
 143   /* Upper bound of the striped loop.  */
 144   d = isl_map_get_space (pbb->transformed);
 145   c = isl_inequality_alloc (isl_local_space_from_space (d));
 146   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, stride);
 147   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, -1);
 148   c = isl_constraint_set_constant_si (c, stride - 1);
 149   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 150
 151   /* Static scheduling for ITER level.
 152      This is mandatory to keep the 2d + 1 canonical scheduling format.  */
 153   d = isl_map_get_space (pbb->transformed);
 154   c = isl_equality_alloc (isl_local_space_from_space (d));
 155   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip + 1, 1);
 156   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 157 }
 158
 159 /* Returns true when strip mining with STRIDE of the loop LST is
 160    profitable.  */
 161
 162 static bool
 163 lst_strip_mine_profitable_p (lst_p lst, int stride)
 164 {
 165   mpz_t niter, strip_stride;
 166   bool res;
 167
 168   gcc_assert (LST_LOOP_P (lst));
 169   mpz_init (strip_stride);
 170   mpz_init (niter);
 171
 172   mpz_set_si (strip_stride, stride);
 173   lst_niter_for_loop (lst, niter);
 174   res = (mpz_cmp (niter, strip_stride) > 0);
 175
 176   mpz_clear (strip_stride);
 177   mpz_clear (niter);
 178   return res;
 179 }
 180
 181 /* Strip-mines all the loops of LST with STRIDE.  Return the number of
 182    loops strip-mined.  */
 183
 184 static int
 185 lst_do_strip_mine_loop (lst_p lst, int depth, int stride)
 186 {
 187   int i;
 188   lst_p l;
 189   poly_bb_p pbb;
 190
 191   if (!lst)
 192     return 0;
 193
 194   if (LST_LOOP_P (lst))
 195     {
 196       int res = 0;
 197
 198       FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 199         res += lst_do_strip_mine_loop (l, depth, stride);
 200
 201       return res;
 202     }
 203
 204   pbb = LST_PBB (lst);
 205   pbb_strip_mine_time_depth (pbb, psct_dynamic_dim (pbb, depth), stride);
 206   return 1;
 207 }
 208
 209 /* Strip-mines all the loops of LST with STRIDE.  When STRIDE is zero,
 210    read the stride from the PARAM_LOOP_BLOCK_TILE_SIZE.  Return the
 211    number of strip-mined loops.
 212
 213    Strip mining transforms a loop
 214
 215    | for (i = 0; i < N; i++)
 216    |   S (i);
 217
 218    into the following loop nest:
 219
 220    | for (k = 0; k < N; k += STRIDE)
 221    |   for (j = 0; j < STRIDE; j++)
 222    |     S (i = k + j);
 223 */
 224
 225 static int
 226 lst_do_strip_mine (lst_p lst, int stride)
 227 {
 228   int i;
 229   lst_p l;
 230   int res = 0;
 231   int depth;
 232
 233   if (!stride)
 234     stride = PARAM_VALUE (PARAM_LOOP_BLOCK_TILE_SIZE);
 235
 236   if (!lst
 237       || !LST_LOOP_P (lst))
 238     return false;
 239
 240   FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 241     res += lst_do_strip_mine (l, stride);
 242
 243   depth = lst_depth (lst);
 244   if (depth >= 0
 245       && lst_strip_mine_profitable_p (lst, stride))
 246     {
 247       res += lst_do_strip_mine_loop (lst, lst_depth (lst), stride);
 248       lst_add_loop_under_loop (lst);
 249     }
 250
 251   return res;
 252 }
 253
 254 /* Strip mines all the loops in SCOP.  Returns the number of
 255    strip-mined loops.  */
 256
 257 int
 258 scop_do_strip_mine (scop_p scop, int stride)
 259 {
 260   return lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), stride);
 261 }
 262
 263 /* Loop blocks all the loops in SCOP.  Returns true when we manage to
 264    block some loops.  */
 265
 266 bool
 267 scop_do_block (scop_p scop)
 268 {
 269   store_scattering (scop);
 270
 271   /* If we don't strip mine at least two loops, or not interchange
 272      loops, the strip mine alone will not be profitable, and the
 273      transform is not a loop blocking: so revert the transform.  */
 274   if (lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), 0) < 2
 275       || scop_do_interchange (scop) == 0)
 276     {
 277       restore_scattering (scop);
 278       return false;
 279     }
 280
 281   if (dump_file && (dump_flags & TDF_DETAILS))
 282     fprintf (dump_file, "SCoP will be loop blocked.\n");
 283
 284   return true;
 285 }
 286
 287 #endif