gcc/graphite-blocking.c

   1 /* Heuristics and transform for loop blocking and strip mining on
   2    polyhedral representation.
   3
   4    Copyright (C) 2009-2013 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_cloog
  27 #include <isl/set.h>
  28 #include <isl/map.h>
  29 #include <isl/union_map.h>
  30 #include <isl/constraint.h>
  31 #include <cloog/cloog.h>
  32 #include <cloog/isl/domain.h>
  33 #endif
  34
  35 #include "system.h"
  36 #include "coretypes.h"
  37 #include "tree.h"
  38 #include "tree-ssa.h"
  39 #include "dumpfile.h"
  40 #include "cfgloop.h"
  41 #include "tree-chrec.h"
  42 #include "tree-data-ref.h"
  43 #include "sese.h"
  44
  45 #ifdef HAVE_cloog
  46 #include "graphite-poly.h"
  47
  48
  49 /* Strip mines with a factor STRIDE the scattering (time) dimension
  50    around PBB at depth TIME_DEPTH.
  51
  52    The following example comes from the wiki page:
  53    http://gcc.gnu.org/wiki/Graphite/Strip_mine
  54
  55    The strip mine of a loop with a tile of 64 can be obtained with a
  56    scattering function as follows:
  57
  58    $ cat ./albert_strip_mine.cloog
  59    # language: C
  60    c
  61
  62    # parameter {n | n >= 0}
  63    1 3
  64    #  n  1
  65    1  1  0
  66    1
  67    n
  68
  69    1 # Number of statements:
  70
  71    1
  72    # {i | 0 <= i <= n}
  73    2 4
  74    #  i  n   1
  75    1  1  0   0
  76    1 -1  1   0
  77
  78    0  0  0
  79    1
  80    i
  81
  82    1 # Scattering functions
  83
  84    3 6
  85    #  NEW  OLD    i    n    1
  86    1  -64    0    1    0    0
  87    1   64    0   -1    0   63
  88    0    0    1   -1    0    0
  89
  90    1
  91    NEW  OLD
  92
  93    #the output of CLooG is like this:
  94    #$ cloog ./albert_strip_mine.cloog
  95    # for (NEW=0;NEW<=floord(n,64);NEW++) {
  96    #   for (OLD=max(64*NEW,0);OLD<=min(64*NEW+63,n);OLD++) {
  97    #     S1(i = OLD) ;
  98    #   }
  99    # }
 100 */
 101
 102 static void
 103 pbb_strip_mine_time_depth (poly_bb_p pbb, int time_depth, int stride)
 104 {
 105   isl_space *d;
 106   isl_constraint *c;
 107   int iter, strip;
 108   /* STRIP is the dimension that iterates with stride STRIDE.  */
 109   /* ITER is the dimension that enumerates single iterations inside
 110      one strip that has at most STRIDE iterations.  */
 111   strip = time_depth;
 112   iter = strip + 2;
 113
 114   pbb->transformed = isl_map_insert_dims (pbb->transformed, isl_dim_out,
 115                                           strip, 2);
 116
 117   /* Lower bound of the striped loop.  */
 118   d = isl_map_get_space (pbb->transformed);
 119   c = isl_inequality_alloc (isl_local_space_from_space (d));
 120   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, -stride);
 121   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, 1);
 122   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 123
 124   /* Upper bound of the striped loop.  */
 125   d = isl_map_get_space (pbb->transformed);
 126   c = isl_inequality_alloc (isl_local_space_from_space (d));
 127   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, stride);
 128   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, -1);
 129   c = isl_constraint_set_constant_si (c, stride - 1);
 130   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 131
 132   /* Static scheduling for ITER level.
 133      This is mandatory to keep the 2d + 1 canonical scheduling format.  */
 134   d = isl_map_get_space (pbb->transformed);
 135   c = isl_equality_alloc (isl_local_space_from_space (d));
 136   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip + 1, 1);
 137   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 138 }
 139
 140 /* Returns true when strip mining with STRIDE of the loop LST is
 141    profitable.  */
 142
 143 static bool
 144 lst_strip_mine_profitable_p (lst_p lst, int stride)
 145 {
 146   mpz_t niter, strip_stride;
 147   bool res;
 148
 149   gcc_assert (LST_LOOP_P (lst));
 150   mpz_init (strip_stride);
 151   mpz_init (niter);
 152
 153   mpz_set_si (strip_stride, stride);
 154   lst_niter_for_loop (lst, niter);
 155   res = (mpz_cmp (niter, strip_stride) > 0);
 156
 157   mpz_clear (strip_stride);
 158   mpz_clear (niter);
 159   return res;
 160 }
 161
 162 /* Strip-mines all the loops of LST with STRIDE.  Return the number of
 163    loops strip-mined.  */
 164
 165 static int
 166 lst_do_strip_mine_loop (lst_p lst, int depth, int stride)
 167 {
 168   int i;
 169   lst_p l;
 170   poly_bb_p pbb;
 171
 172   if (!lst)
 173     return 0;
 174
 175   if (LST_LOOP_P (lst))
 176     {
 177       int res = 0;
 178
 179       FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 180         res += lst_do_strip_mine_loop (l, depth, stride);
 181
 182       return res;
 183     }
 184
 185   pbb = LST_PBB (lst);
 186   pbb_strip_mine_time_depth (pbb, psct_dynamic_dim (pbb, depth), stride);
 187   return 1;
 188 }
 189
 190 /* Strip-mines all the loops of LST with STRIDE.  When STRIDE is zero,
 191    read the stride from the PARAM_LOOP_BLOCK_TILE_SIZE.  Return the
 192    number of strip-mined loops.
 193
 194    Strip mining transforms a loop
 195
 196    | for (i = 0; i < N; i++)
 197    |   S (i);
 198
 199    into the following loop nest:
 200
 201    | for (k = 0; k < N; k += STRIDE)
 202    |   for (j = 0; j < STRIDE; j++)
 203    |     S (i = k + j);
 204 */
 205
 206 static int
 207 lst_do_strip_mine (lst_p lst, int stride)
 208 {
 209   int i;
 210   lst_p l;
 211   int res = 0;
 212   int depth;
 213
 214   if (!stride)
 215     stride = PARAM_VALUE (PARAM_LOOP_BLOCK_TILE_SIZE);
 216
 217   if (!lst
 218       || !LST_LOOP_P (lst))
 219     return false;
 220
 221   FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 222     res += lst_do_strip_mine (l, stride);
 223
 224   depth = lst_depth (lst);
 225   if (depth >= 0
 226       && lst_strip_mine_profitable_p (lst, stride))
 227     {
 228       res += lst_do_strip_mine_loop (lst, lst_depth (lst), stride);
 229       lst_add_loop_under_loop (lst);
 230     }
 231
 232   return res;
 233 }
 234
 235 /* Strip mines all the loops in SCOP.  Returns the number of
 236    strip-mined loops.  */
 237
 238 int
 239 scop_do_strip_mine (scop_p scop, int stride)
 240 {
 241   return lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), stride);
 242 }
 243
 244 /* Loop blocks all the loops in SCOP.  Returns true when we manage to
 245    block some loops.  */
 246
 247 bool
 248 scop_do_block (scop_p scop)
 249 {
 250   store_scattering (scop);
 251
 252   /* If we don't strip mine at least two loops, or not interchange
 253      loops, the strip mine alone will not be profitable, and the
 254      transform is not a loop blocking: so revert the transform.  */
 255   if (lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), 0) < 2
 256       || scop_do_interchange (scop) == 0)
 257     {
 258       restore_scattering (scop);
 259       return false;
 260     }
 261
 262   if (dump_file && (dump_flags & TDF_DETAILS))
 263     fprintf (dump_file, "SCoP will be loop blocked.\n");
 264
 265   return true;
 266 }
 267
 268 #endif