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 "hash-set.h"
  36 #include "machmode.h"
  37 #include "vec.h"
  38 #include "double-int.h"
  39 #include "input.h"
  40 #include "alias.h"
  41 #include "symtab.h"
  42 #include "options.h"
  43 #include "wide-int.h"
  44 #include "inchash.h"
  45 #include "tree.h"
  46 #include "fold-const.h"
  47 #include "predict.h"
  48 #include "tm.h"
  49 #include "hard-reg-set.h"
  50 #include "input.h"
  51 #include "function.h"
  52 #include "dominance.h"
  53 #include "basic-block.h"
  54 #include "tree-ssa-alias.h"
  55 #include "internal-fn.h"
  56 #include "gimple-expr.h"
  57 #include "is-a.h"
  58 #include "gimple.h"
  59 #include "gimple-iterator.h"
  60 #include "tree-ssa-loop.h"
  61 #include "dumpfile.h"
  62 #include "cfgloop.h"
  63 #include "tree-chrec.h"
  64 #include "tree-data-ref.h"
  65 #include "sese.h"
  66
  67 #ifdef HAVE_isl
  68 #include "graphite-poly.h"
  69
  70
  71 /* Strip mines with a factor STRIDE the scattering (time) dimension
  72    around PBB at depth TIME_DEPTH.
  73
  74    The following example comes from the wiki page:
  75    http://gcc.gnu.org/wiki/Graphite/Strip_mine
  76
  77    The strip mine of a loop with a tile of 64 can be obtained with a
  78    scattering function as follows:
  79
  80    $ cat ./albert_strip_mine.cloog
  81    # language: C
  82    c
  83
  84    # parameter {n | n >= 0}
  85    1 3
  86    #  n  1
  87    1  1  0
  88    1
  89    n
  90
  91    1 # Number of statements:
  92
  93    1
  94    # {i | 0 <= i <= n}
  95    2 4
  96    #  i  n   1
  97    1  1  0   0
  98    1 -1  1   0
  99
 100    0  0  0
 101    1
 102    i
 103
 104    1 # Scattering functions
 105
 106    3 6
 107    #  NEW  OLD    i    n    1
 108    1  -64    0    1    0    0
 109    1   64    0   -1    0   63
 110    0    0    1   -1    0    0
 111
 112    1
 113    NEW  OLD
 114
 115    #the output of CLooG is like this:
 116    #$ cloog ./albert_strip_mine.cloog
 117    # for (NEW=0;NEW<=floord(n,64);NEW++) {
 118    #   for (OLD=max(64*NEW,0);OLD<=min(64*NEW+63,n);OLD++) {
 119    #     S1(i = OLD) ;
 120    #   }
 121    # }
 122 */
 123
 124 static void
 125 pbb_strip_mine_time_depth (poly_bb_p pbb, int time_depth, int stride)
 126 {
 127   isl_space *d;
 128   isl_constraint *c;
 129   int iter, strip;
 130   /* STRIP is the dimension that iterates with stride STRIDE.  */
 131   /* ITER is the dimension that enumerates single iterations inside
 132      one strip that has at most STRIDE iterations.  */
 133   strip = time_depth;
 134   iter = strip + 2;
 135
 136   pbb->transformed = isl_map_insert_dims (pbb->transformed, isl_dim_out,
 137                                           strip, 2);
 138
 139   /* Lower bound of the striped loop.  */
 140   d = isl_map_get_space (pbb->transformed);
 141   c = isl_inequality_alloc (isl_local_space_from_space (d));
 142   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, -stride);
 143   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, 1);
 144   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 145
 146   /* Upper bound of the striped loop.  */
 147   d = isl_map_get_space (pbb->transformed);
 148   c = isl_inequality_alloc (isl_local_space_from_space (d));
 149   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip, stride);
 150   c = isl_constraint_set_coefficient_si (c, isl_dim_out, iter, -1);
 151   c = isl_constraint_set_constant_si (c, stride - 1);
 152   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 153
 154   /* Static scheduling for ITER level.
 155      This is mandatory to keep the 2d + 1 canonical scheduling format.  */
 156   d = isl_map_get_space (pbb->transformed);
 157   c = isl_equality_alloc (isl_local_space_from_space (d));
 158   c = isl_constraint_set_coefficient_si (c, isl_dim_out, strip + 1, 1);
 159   pbb->transformed = isl_map_add_constraint (pbb->transformed, c);
 160 }
 161
 162 /* Returns true when strip mining with STRIDE of the loop LST is
 163    profitable.  */
 164
 165 static bool
 166 lst_strip_mine_profitable_p (lst_p lst, int stride)
 167 {
 168   mpz_t niter, strip_stride;
 169   bool res;
 170
 171   gcc_assert (LST_LOOP_P (lst));
 172   mpz_init (strip_stride);
 173   mpz_init (niter);
 174
 175   mpz_set_si (strip_stride, stride);
 176   lst_niter_for_loop (lst, niter);
 177   res = (mpz_cmp (niter, strip_stride) > 0);
 178
 179   mpz_clear (strip_stride);
 180   mpz_clear (niter);
 181   return res;
 182 }
 183
 184 /* Strip-mines all the loops of LST with STRIDE.  Return the number of
 185    loops strip-mined.  */
 186
 187 static int
 188 lst_do_strip_mine_loop (lst_p lst, int depth, int stride)
 189 {
 190   int i;
 191   lst_p l;
 192   poly_bb_p pbb;
 193
 194   if (!lst)
 195     return 0;
 196
 197   if (LST_LOOP_P (lst))
 198     {
 199       int res = 0;
 200
 201       FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 202         res += lst_do_strip_mine_loop (l, depth, stride);
 203
 204       return res;
 205     }
 206
 207   pbb = LST_PBB (lst);
 208   pbb_strip_mine_time_depth (pbb, psct_dynamic_dim (pbb, depth), stride);
 209   return 1;
 210 }
 211
 212 /* Strip-mines all the loops of LST with STRIDE.  When STRIDE is zero,
 213    read the stride from the PARAM_LOOP_BLOCK_TILE_SIZE.  Return the
 214    number of strip-mined loops.
 215
 216    Strip mining transforms a loop
 217
 218    | for (i = 0; i < N; i++)
 219    |   S (i);
 220
 221    into the following loop nest:
 222
 223    | for (k = 0; k < N; k += STRIDE)
 224    |   for (j = 0; j < STRIDE; j++)
 225    |     S (i = k + j);
 226 */
 227
 228 static int
 229 lst_do_strip_mine (lst_p lst, int stride)
 230 {
 231   int i;
 232   lst_p l;
 233   int res = 0;
 234   int depth;
 235
 236   if (!stride)
 237     stride = PARAM_VALUE (PARAM_LOOP_BLOCK_TILE_SIZE);
 238
 239   if (!lst
 240       || !LST_LOOP_P (lst))
 241     return false;
 242
 243   FOR_EACH_VEC_ELT (LST_SEQ (lst), i, l)
 244     res += lst_do_strip_mine (l, stride);
 245
 246   depth = lst_depth (lst);
 247   if (depth >= 0
 248       && lst_strip_mine_profitable_p (lst, stride))
 249     {
 250       res += lst_do_strip_mine_loop (lst, lst_depth (lst), stride);
 251       lst_add_loop_under_loop (lst);
 252     }
 253
 254   return res;
 255 }
 256
 257 /* Strip mines all the loops in SCOP.  Returns the number of
 258    strip-mined loops.  */
 259
 260 int
 261 scop_do_strip_mine (scop_p scop, int stride)
 262 {
 263   return lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), stride);
 264 }
 265
 266 /* Loop blocks all the loops in SCOP.  Returns true when we manage to
 267    block some loops.  */
 268
 269 bool
 270 scop_do_block (scop_p scop)
 271 {
 272   store_scattering (scop);
 273
 274   /* If we don't strip mine at least two loops, or not interchange
 275      loops, the strip mine alone will not be profitable, and the
 276      transform is not a loop blocking: so revert the transform.  */
 277   if (lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop), 0) < 2
 278       || scop_do_interchange (scop) == 0)
 279     {
 280       restore_scattering (scop);
 281       return false;
 282     }
 283
 284   if (dump_file && (dump_flags & TDF_DETAILS))
 285     fprintf (dump_file, "SCoP will be loop blocked.\n");
 286
 287   return true;
 288 }
 289
 290 #endif