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