gcc/graphite-blocking.c

   1 /* Heuristics and transform for loop blocking and strip mining on
   2    polyhedral representation.
   3
   4    Copyright (C) 2009 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 #include "config.h"
  24 #include "system.h"
  25 #include "coretypes.h"
  26 #include "tm.h"
  27 #include "ggc.h"
  28 #include "tree.h"
  29 #include "rtl.h"
  30 #include "output.h"
  31 #include "basic-block.h"
  32 #include "diagnostic.h"
  33 #include "tree-flow.h"
  34 #include "toplev.h"
  35 #include "tree-dump.h"
  36 #include "timevar.h"
  37 #include "cfgloop.h"
  38 #include "tree-chrec.h"
  39 #include "tree-data-ref.h"
  40 #include "tree-scalar-evolution.h"
  41 #include "tree-pass.h"
  42 #include "domwalk.h"
  43 #include "value-prof.h"
  44 #include "pointer-set.h"
  45 #include "gimple.h"
  46 #include "params.h"
  47
  48 #ifdef HAVE_cloog
  49 #include "cloog/cloog.h"
  50 #include "ppl_c.h"
  51 #include "sese.h"
  52 #include "graphite-ppl.h"
  53 #include "graphite.h"
  54 #include "graphite-poly.h"
  55
  56
  57 /* Strip mines with a factor STRIDE the scattering (time) dimension
  58    around PBB at depth TIME_DEPTH.
  59
  60    The following example comes from the wiki page:
  61    http://gcc.gnu.org/wiki/Graphite/Strip_mine
  62
  63    The strip mine of a loop with a tile of 64 can be obtained with a
  64    scattering function as follows:
  65
  66    $ cat ./albert_strip_mine.cloog
  67    # language: C
  68    c
  69
  70    # parameter {n | n >= 0}
  71    1 3
  72    #  n  1
  73    1  1  0
  74    1
  75    n
  76
  77    1 # Number of statements:
  78
  79    1
  80    # {i | 0 <= i <= n}
  81    2 4
  82    #  i  n   1
  83    1  1  0   0
  84    1 -1  1   0
  85
  86    0  0  0
  87    1
  88    i
  89
  90    1 # Scattering functions
  91
  92    3 6
  93    #  NEW  OLD    i    n    1
  94    1  -64    0    1    0    0
  95    1   64    0   -1    0   63
  96    0    0    1   -1    0    0
  97
  98    1
  99    NEW  OLD
 100
 101    #the output of CLooG is like this:
 102    #$ cloog ./albert_strip_mine.cloog
 103    # for (NEW=0;NEW<=floord(n,64);NEW++) {
 104    #   for (OLD=max(64*NEW,0);OLD<=min(64*NEW+63,n);OLD++) {
 105    #     S1(i = OLD) ;
 106    #   }
 107    # }
 108 */
 109
 110 static bool
 111 pbb_strip_mine_time_depth (poly_bb_p pbb, int time_depth, int stride)
 112 {
 113   ppl_dimension_type iter, dim, strip;
 114   ppl_Polyhedron_t res = PBB_TRANSFORMED_SCATTERING (pbb);
 115   /* STRIP is the dimension that iterates with stride STRIDE.  */
 116   /* ITER is the dimension that enumerates single iterations inside
 117      one strip that has at most STRIDE iterations.  */
 118   strip = time_depth;
 119   iter = strip + 2;
 120
 121   psct_add_scattering_dimension (pbb, strip);
 122   psct_add_scattering_dimension (pbb, strip + 1);
 123
 124   ppl_Polyhedron_space_dimension (res, &dim);
 125
 126   /* Lower bound of the striped loop.  */
 127   {
 128     ppl_Constraint_t new_cstr;
 129     ppl_Linear_Expression_t expr;
 130
 131     ppl_new_Linear_Expression_with_dimension (&expr, dim);
 132     ppl_set_coef (expr, strip, -1 * stride);
 133     ppl_set_coef (expr, iter, 1);
 134
 135     ppl_new_Constraint (&new_cstr, expr, PPL_CONSTRAINT_TYPE_GREATER_OR_EQUAL);
 136     ppl_delete_Linear_Expression (expr);
 137     ppl_Polyhedron_add_constraint (res, new_cstr);
 138     ppl_delete_Constraint (new_cstr);
 139   }
 140
 141   /* Upper bound of the striped loop.  */
 142   {
 143     ppl_Constraint_t new_cstr;
 144     ppl_Linear_Expression_t expr;
 145
 146     ppl_new_Linear_Expression_with_dimension (&expr, dim);
 147     ppl_set_coef (expr, strip, stride);
 148     ppl_set_coef (expr, iter, -1);
 149     ppl_set_inhomogeneous (expr, stride - 1);
 150
 151     ppl_new_Constraint (&new_cstr, expr, PPL_CONSTRAINT_TYPE_GREATER_OR_EQUAL);
 152     ppl_delete_Linear_Expression (expr);
 153     ppl_Polyhedron_add_constraint (res, new_cstr);
 154     ppl_delete_Constraint (new_cstr);
 155   }
 156
 157   /* Static scheduling for ITER level.
 158      This is mandatory to keep the 2d + 1 canonical scheduling format.  */
 159   {
 160     ppl_Constraint_t new_cstr;
 161     ppl_Linear_Expression_t expr;
 162
 163     ppl_new_Linear_Expression_with_dimension (&expr, dim);
 164     ppl_set_coef (expr, strip + 1, 1);
 165     ppl_set_inhomogeneous (expr, 0);
 166
 167     ppl_new_Constraint (&new_cstr, expr, PPL_CONSTRAINT_TYPE_EQUAL);
 168     ppl_delete_Linear_Expression (expr);
 169     ppl_Polyhedron_add_constraint (res, new_cstr);
 170     ppl_delete_Constraint (new_cstr);
 171   }
 172
 173   return true;
 174 }
 175
 176 /* Returns true when strip mining with STRIDE of the loop around PBB
 177    at DEPTH is profitable.  */
 178
 179 static bool
 180 pbb_strip_mine_profitable_p (poly_bb_p pbb,
 181                              graphite_dim_t depth,
 182                              int stride)
 183 {
 184   mpz_t niter, strip_stride;
 185   bool res;
 186
 187   mpz_init (strip_stride);
 188   mpz_init (niter);
 189   mpz_set_si (strip_stride, stride);
 190   pbb_number_of_iterations_at_time (pbb, psct_dynamic_dim (pbb, depth), niter);
 191   res = (mpz_cmp (niter, strip_stride) > 0);
 192   mpz_clear (strip_stride);
 193   mpz_clear (niter);
 194
 195   return res;
 196 }
 197
 198 /* Strip-mines all the loops of LST that are considered profitable to
 199    strip-mine.  Return true if it did strip-mined some loops.  */
 200
 201 static bool
 202 lst_do_strip_mine_loop (lst_p lst, int depth)
 203 {
 204   int i;
 205   lst_p l;
 206   int stride = PARAM_VALUE (PARAM_LOOP_BLOCK_TILE_SIZE);
 207   poly_bb_p pbb;
 208
 209   if (!lst)
 210     return false;
 211
 212   if (LST_LOOP_P (lst))
 213     {
 214       bool res = false;
 215
 216       for (i = 0; VEC_iterate (lst_p, LST_SEQ (lst), i, l); i++)
 217         res |= lst_do_strip_mine_loop (l, depth);
 218
 219       return res;
 220     }
 221
 222   pbb = LST_PBB (lst);
 223   return pbb_strip_mine_time_depth (pbb, psct_dynamic_dim (pbb, depth),
 224                                     stride);
 225 }
 226
 227 /* Strip-mines all the loops of LST that are considered profitable to
 228    strip-mine.  Return true if it did strip-mined some loops.  */
 229
 230 static bool
 231 lst_do_strip_mine (lst_p lst)
 232 {
 233   int i;
 234   lst_p l;
 235   bool res = false;
 236   int stride = PARAM_VALUE (PARAM_LOOP_BLOCK_TILE_SIZE);
 237   int depth;
 238
 239   if (!lst
 240       || !LST_LOOP_P (lst))
 241     return false;
 242
 243   for (i = 0; VEC_iterate (lst_p, LST_SEQ (lst), i, l); i++)
 244     res |= lst_do_strip_mine (l);
 245
 246   depth = lst_depth (lst);
 247   if (depth >= 0
 248       && pbb_strip_mine_profitable_p (LST_PBB (lst_find_first_pbb (lst)),
 249                                       depth, stride))
 250     {
 251       res |= lst_do_strip_mine_loop (lst, lst_depth (lst));
 252       lst_add_loop_under_loop (lst);
 253     }
 254
 255   return res;
 256 }
 257
 258 /* Strip mines all the loops in SCOP.  Nothing profitable in all this:
 259    this is just a driver function.  */
 260
 261 bool
 262 scop_do_strip_mine (scop_p scop)
 263 {
 264   bool transform_done = false;
 265
 266   store_scattering (scop);
 267
 268   transform_done = lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop));
 269
 270   if (!transform_done)
 271     return false;
 272
 273   if (!graphite_legal_transform (scop))
 274     {
 275       restore_scattering (scop);
 276       return false;
 277     }
 278
 279   return transform_done;
 280 }
 281
 282 /* Loop blocks all the loops in SCOP.  Returns true when we manage to
 283    block some loops.  */
 284
 285 bool
 286 scop_do_block (scop_p scop)
 287 {
 288   bool strip_mined = false;
 289   bool interchanged = false;
 290
 291   store_scattering (scop);
 292
 293   strip_mined = lst_do_strip_mine (SCOP_TRANSFORMED_SCHEDULE (scop));
 294   interchanged = scop_do_interchange (scop);
 295
 296   /* If we don't interchange loops, then the strip mine is not
 297      profitable, and the transform is not a loop blocking.  */
 298   if (!interchanged
 299       || !graphite_legal_transform (scop))
 300     {
 301       restore_scattering (scop);
 302       return false;
 303     }
 304   else if (strip_mined && interchanged
 305            && dump_file && (dump_flags & TDF_DETAILS))
 306     fprintf (dump_file, "SCoP will be loop blocked.\n");
 307
 308   return strip_mined || interchanged;
 309 }
 310
 311 #endif