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