gcc/tree-loop-linear.c

   1 /* Linear Loop transforms
   2    Copyright (C) 2003, 2004, 2005, 2007 Free Software Foundation, Inc.
   3    Contributed by Daniel Berlin <dberlin@dberlin.org>.
   4
   5 This file is part of GCC.
   6
   7 GCC is free software; you can redistribute it and/or modify it under
   8 the terms of the GNU General Public License as published by the Free
   9 Software Foundation; either version 3, or (at your option) any later
  10 version.
  11
  12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14 FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15 for more details.
  16
  17 You should have received a copy of the GNU General Public License
  18 along with GCC; see the file COPYING3.  If not see
  19 <http://www.gnu.org/licenses/>.  */
  20
  21
  22 #include "config.h"
  23 #include "system.h"
  24 #include "coretypes.h"
  25 #include "tm.h"
  26 #include "ggc.h"
  27 #include "tree.h"
  28 #include "target.h"
  29
  30 #include "rtl.h"
  31 #include "basic-block.h"
  32 #include "diagnostic.h"
  33 #include "obstack.h"
  34 #include "tree-flow.h"
  35 #include "tree-dump.h"
  36 #include "timevar.h"
  37 #include "cfgloop.h"
  38 #include "expr.h"
  39 #include "optabs.h"
  40 #include "tree-chrec.h"
  41 #include "tree-data-ref.h"
  42 #include "tree-scalar-evolution.h"
  43 #include "tree-pass.h"
  44 #include "lambda.h"
  45
  46 /* Linear loop transforms include any composition of interchange,
  47    scaling, skewing, and reversal.  They are used to change the
  48    iteration order of loop nests in order to optimize data locality of
  49    traversals, or remove dependences that prevent
  50    parallelization/vectorization/etc.
  51
  52    TODO: Determine reuse vectors/matrix and use it to determine optimal
  53    transform matrix for locality purposes.
  54    TODO: Completion of partial transforms.  */
  55
  56 /* Gather statistics for loop interchange.  LOOP is the loop being
  57    considered. The first loop in the considered loop nest is
  58    FIRST_LOOP, and consequently, the index of the considered loop is
  59    obtained by LOOP->DEPTH - FIRST_LOOP->DEPTH
  60
  61    Initializes:
  62    - DEPENDENCE_STEPS the sum of all the data dependence distances
  63    carried by loop LOOP,
  64
  65    - NB_DEPS_NOT_CARRIED_BY_LOOP the number of dependence relations
  66    for which the loop LOOP is not carrying any dependence,
  67
  68    - ACCESS_STRIDES the sum of all the strides in LOOP.
  69
  70    Example: for the following loop,
  71
  72    | loop_1 runs 1335 times
  73    |   loop_2 runs 1335 times
  74    |     A[{{0, +, 1}_1, +, 1335}_2]
  75    |     B[{{0, +, 1}_1, +, 1335}_2]
  76    |   endloop_2
  77    |   A[{0, +, 1336}_1]
  78    | endloop_1
  79
  80    gather_interchange_stats (in loop_1) will return
  81    DEPENDENCE_STEPS = 3002
  82    NB_DEPS_NOT_CARRIED_BY_LOOP = 5
  83    ACCESS_STRIDES = 10694
  84
  85    gather_interchange_stats (in loop_2) will return
  86    DEPENDENCE_STEPS = 3000
  87    NB_DEPS_NOT_CARRIED_BY_LOOP = 7
  88    ACCESS_STRIDES = 8010
  89 */
  90
  91 static void
  92 gather_interchange_stats (VEC (ddr_p, heap) *dependence_relations,
  93                           VEC (data_reference_p, heap) *datarefs,
  94                           struct loop *loop,
  95                           struct loop *first_loop,
  96                           unsigned int *dependence_steps,
  97                           unsigned int *nb_deps_not_carried_by_loop,
  98                           double_int *access_strides)
  99 {
 100   unsigned int i, j;
 101   struct data_dependence_relation *ddr;
 102   struct data_reference *dr;
 103
 104   *dependence_steps = 0;
 105   *nb_deps_not_carried_by_loop = 0;
 106   *access_strides = double_int_zero;
 107
 108   for (i = 0; VEC_iterate (ddr_p, dependence_relations, i, ddr); i++)
 109     {
 110       /* If we don't know anything about this dependence, or the distance
 111          vector is NULL, or there is no dependence, then there is no reuse of
 112          data.  */
 113       if (DDR_ARE_DEPENDENT (ddr) == chrec_dont_know
 114           || DDR_ARE_DEPENDENT (ddr) == chrec_known
 115           || DDR_NUM_DIST_VECTS (ddr) == 0)
 116         continue;
 117
 118       for (j = 0; j < DDR_NUM_DIST_VECTS (ddr); j++)
 119         {
 120           int dist = DDR_DIST_VECT (ddr, j)[loop_depth (loop) - loop_depth (first_loop)];
 121
 122           if (dist == 0)
 123             (*nb_deps_not_carried_by_loop) += 1;
 124
 125           else if (dist < 0)
 126             (*dependence_steps) += -dist;
 127
 128           else
 129             (*dependence_steps) += dist;
 130         }
 131     }
 132
 133   /* Compute the access strides.  */
 134   for (i = 0; VEC_iterate (data_reference_p, datarefs, i, dr); i++)
 135     {
 136       unsigned int it;
 137       tree ref = DR_REF (dr);
 138       tree stmt = DR_STMT (dr);
 139       struct loop *stmt_loop = loop_containing_stmt (stmt);
 140       struct loop *inner_loop = first_loop->inner;
 141
 142       if (inner_loop != stmt_loop
 143           && !flow_loop_nested_p (inner_loop, stmt_loop))
 144         continue;
 145
 146       for (it = 0; it < DR_NUM_DIMENSIONS (dr);
 147            it++, ref = TREE_OPERAND (ref, 0))
 148         {
 149           tree chrec = DR_ACCESS_FN (dr, it);
 150           tree tstride = evolution_part_in_loop_num (chrec, loop->num);
 151           tree array_size = TYPE_SIZE (TREE_TYPE (ref));
 152           double_int dstride;
 153
 154           if (tstride == NULL_TREE
 155               || array_size == NULL_TREE
 156               || TREE_CODE (tstride) != INTEGER_CST
 157               || TREE_CODE (array_size) != INTEGER_CST)
 158             continue;
 159
 160           dstride = double_int_mul (tree_to_double_int (array_size),
 161                                     tree_to_double_int (tstride));
 162           (*access_strides) = double_int_add (*access_strides, dstride);
 163         }
 164     }
 165 }
 166
 167 /* Attempt to apply interchange transformations to TRANS to maximize the
 168    spatial and temporal locality of the loop.
 169    Returns the new transform matrix.  The smaller the reuse vector
 170    distances in the inner loops, the fewer the cache misses.
 171    FIRST_LOOP is the loop->num of the first loop in the analyzed loop
 172    nest.  */
 173
 174
 175 static lambda_trans_matrix
 176 try_interchange_loops (lambda_trans_matrix trans,
 177                        unsigned int depth,
 178                        VEC (ddr_p, heap) *dependence_relations,
 179                        VEC (data_reference_p, heap) *datarefs,
 180                        struct loop *first_loop)
 181 {
 182   struct loop *loop_i;
 183   struct loop *loop_j;
 184   unsigned int dependence_steps_i, dependence_steps_j;
 185   double_int access_strides_i, access_strides_j;
 186   unsigned int nb_deps_not_carried_by_i, nb_deps_not_carried_by_j;
 187   struct data_dependence_relation *ddr;
 188
 189   if (VEC_length (ddr_p, dependence_relations) == 0)
 190     return trans;
 191
 192   /* When there is an unknown relation in the dependence_relations, we
 193      know that it is no worth looking at this loop nest: give up.  */
 194   ddr = VEC_index (ddr_p, dependence_relations, 0);
 195   if (ddr == NULL || DDR_ARE_DEPENDENT (ddr) == chrec_dont_know)
 196     return trans;
 197
 198   /* LOOP_I is always the outer loop.  */
 199   for (loop_j = first_loop->inner;
 200        loop_j;
 201        loop_j = loop_j->inner)
 202     for (loop_i = first_loop;
 203          loop_depth (loop_i) < loop_depth (loop_j);
 204          loop_i = loop_i->inner)
 205       {
 206         gather_interchange_stats (dependence_relations, datarefs,
 207                                   loop_i, first_loop,
 208                                   &dependence_steps_i,
 209                                   &nb_deps_not_carried_by_i,
 210                                   &access_strides_i);
 211         gather_interchange_stats (dependence_relations, datarefs,
 212                                   loop_j, first_loop,
 213                                   &dependence_steps_j,
 214                                   &nb_deps_not_carried_by_j,
 215                                   &access_strides_j);
 216
 217         /* Heuristics for loop interchange profitability:
 218
 219            1. (spatial locality) Inner loops should have smallest
 220               dependence steps.
 221
 222            2. (spatial locality) Inner loops should contain more
 223            dependence relations not carried by the loop.
 224
 225            3. (temporal locality) Inner loops should have smallest
 226               array access strides.
 227         */
 228         if (dependence_steps_i < dependence_steps_j
 229             || nb_deps_not_carried_by_i > nb_deps_not_carried_by_j
 230             || double_int_ucmp (access_strides_i, access_strides_j) < 0)
 231           {
 232             lambda_matrix_row_exchange (LTM_MATRIX (trans),
 233                                         loop_depth (loop_i) - loop_depth (first_loop),
 234                                         loop_depth (loop_j) - loop_depth (first_loop));
 235             /* Validate the resulting matrix.  When the transformation
 236                is not valid, reverse to the previous transformation.  */
 237             if (!lambda_transform_legal_p (trans, depth, dependence_relations))
 238               lambda_matrix_row_exchange (LTM_MATRIX (trans),
 239                                           loop_depth (loop_i) - loop_depth (first_loop),
 240                                           loop_depth (loop_j) - loop_depth (first_loop));
 241           }
 242       }
 243
 244   return trans;
 245 }
 246
 247 /* Return the number of nested loops in LOOP_NEST, or 0 if the loops
 248    are not perfectly nested.  */
 249
 250 static unsigned int
 251 perfect_loop_nest_depth (struct loop *loop_nest)
 252 {
 253   struct loop *temp;
 254   unsigned int depth = 1;
 255
 256   /* If it's not a loop nest, we don't want it.  We also don't handle
 257      sibling loops properly, which are loops of the following form:
 258
 259      | for (i = 0; i < 50; i++)
 260      |   {
 261      |     for (j = 0; j < 50; j++)
 262      |       {
 263      |        ...
 264      |       }
 265      |     for (j = 0; j < 50; j++)
 266      |       {
 267      |        ...
 268      |       }
 269      |   }
 270   */
 271
 272   if (!loop_nest->inner || !single_exit (loop_nest))
 273     return 0;
 274
 275   for (temp = loop_nest->inner; temp; temp = temp->inner)
 276     {
 277       /* If we have a sibling loop or multiple exit edges, jump ship.  */
 278       if (temp->next || !single_exit (temp))
 279         return 0;
 280
 281       depth++;
 282     }
 283
 284   return depth;
 285 }
 286
 287 /* Perform a set of linear transforms on loops.  */
 288
 289 void
 290 linear_transform_loops (void)
 291 {
 292   bool modified = false;
 293   loop_iterator li;
 294   VEC(tree,heap) *oldivs = NULL;
 295   VEC(tree,heap) *invariants = NULL;
 296   VEC(tree,heap) *remove_ivs = VEC_alloc (tree, heap, 3);
 297   struct loop *loop_nest;
 298   tree oldiv_stmt;
 299   unsigned i;
 300
 301   FOR_EACH_LOOP (li, loop_nest, 0)
 302     {
 303       unsigned int depth = 0;
 304       VEC (ddr_p, heap) *dependence_relations;
 305       VEC (data_reference_p, heap) *datarefs;
 306       lambda_loopnest before, after;
 307       lambda_trans_matrix trans;
 308       struct obstack lambda_obstack;
 309       gcc_obstack_init (&lambda_obstack);
 310
 311       depth = perfect_loop_nest_depth (loop_nest);
 312       if (depth == 0)
 313         continue;
 314
 315       VEC_truncate (tree, oldivs, 0);
 316       VEC_truncate (tree, invariants, 0);
 317
 318       datarefs = VEC_alloc (data_reference_p, heap, 10);
 319       dependence_relations = VEC_alloc (ddr_p, heap, 10 * 10);
 320       compute_data_dependences_for_loop (loop_nest, true, &datarefs,
 321                                          &dependence_relations);
 322
 323       if (dump_file && (dump_flags & TDF_DETAILS))
 324         dump_ddrs (dump_file, dependence_relations);
 325
 326       /* Build the transformation matrix.  */
 327       trans = lambda_trans_matrix_new (depth, depth);
 328       lambda_matrix_id (LTM_MATRIX (trans), depth);
 329       trans = try_interchange_loops (trans, depth, dependence_relations,
 330                                      datarefs, loop_nest);
 331
 332       if (lambda_trans_matrix_id_p (trans))
 333         {
 334           if (dump_file)
 335            fprintf (dump_file, "Won't transform loop. Optimal transform is the identity transform\n");
 336           goto free_and_continue;
 337         }
 338
 339       /* Check whether the transformation is legal.  */
 340       if (!lambda_transform_legal_p (trans, depth, dependence_relations))
 341         {
 342           if (dump_file)
 343             fprintf (dump_file, "Can't transform loop, transform is illegal:\n");
 344           goto free_and_continue;
 345         }
 346
 347       before = gcc_loopnest_to_lambda_loopnest (loop_nest, &oldivs,
 348                                                 &invariants, &lambda_obstack);
 349
 350       if (!before)
 351         goto free_and_continue;
 352
 353       if (dump_file)
 354         {
 355           fprintf (dump_file, "Before:\n");
 356           print_lambda_loopnest (dump_file, before, 'i');
 357         }
 358
 359       after = lambda_loopnest_transform (before, trans, &lambda_obstack);
 360
 361       if (dump_file)
 362         {
 363           fprintf (dump_file, "After:\n");
 364           print_lambda_loopnest (dump_file, after, 'u');
 365         }
 366
 367       lambda_loopnest_to_gcc_loopnest (loop_nest, oldivs, invariants,
 368                                        &remove_ivs,
 369                                        after, trans, &lambda_obstack);
 370       modified = true;
 371
 372       if (dump_file)
 373         fprintf (dump_file, "Successfully transformed loop.\n");
 374
 375     free_and_continue:
 376       obstack_free (&lambda_obstack, NULL);
 377       free_dependence_relations (dependence_relations);
 378       free_data_refs (datarefs);
 379     }
 380
 381   for (i = 0; VEC_iterate (tree, remove_ivs, i, oldiv_stmt); i++)
 382     remove_iv (oldiv_stmt);
 383
 384   VEC_free (tree, heap, oldivs);
 385   VEC_free (tree, heap, invariants);
 386   VEC_free (tree, heap, remove_ivs);
 387   scev_reset ();
 388
 389   if (modified)
 390     rewrite_into_loop_closed_ssa (NULL, TODO_update_ssa_full_phi);
 391 }