gcc/tree-loop-linear.c

   1 /* Linear Loop transforms
   2    Copyright (C) 2003, 2004, 2005 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 2, 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 COPYING.  If not, write to the Free
  19 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
  20 02111-1307, USA.  */
  21
  22
  23 #include "config.h"
  24 #include "system.h"
  25 #include "coretypes.h"
  26 #include "tm.h"
  27 #include "errors.h"
  28 #include "ggc.h"
  29 #include "tree.h"
  30 #include "target.h"
  31
  32 #include "rtl.h"
  33 #include "basic-block.h"
  34 #include "diagnostic.h"
  35 #include "tree-flow.h"
  36 #include "tree-dump.h"
  37 #include "timevar.h"
  38 #include "cfgloop.h"
  39 #include "expr.h"
  40 #include "optabs.h"
  41 #include "tree-chrec.h"
  42 #include "tree-data-ref.h"
  43 #include "tree-scalar-evolution.h"
  44 #include "tree-pass.h"
  45 #include "varray.h"
  46 #include "lambda.h"
  47
  48 /* Linear loop transforms include any composition of interchange,
  49    scaling, skewing, and reversal.  They are used to change the
  50    iteration order of loop nests in order to optimize data locality of
  51    traversals, or remove dependences that prevent
  52    parallelization/vectorization/etc.
  53
  54    TODO: Determine reuse vectors/matrix and use it to determine optimal
  55    transform matrix for locality purposes.
  56    TODO: Completion of partial transforms.  */
  57
  58 /* Gather statistics for loop interchange.  LOOP is the loop being
  59    considered. The first loop in the considered loop nest is
  60    FIRST_LOOP, and consequently, the index of the considered loop is
  61    obtained by LOOP->DEPTH - FIRST_LOOP->DEPTH
  62
  63    Initializes:
  64    - DEPENDENCE_STEPS the sum of all the data dependence distances
  65    carried by loop LOOP,
  66
  67    - NB_DEPS_NOT_CARRIED_BY_LOOP the number of dependence relations
  68    for which the loop LOOP is not carrying any dependence,
  69
  70    - ACCESS_STRIDES the sum of all the strides in LOOP.
  71
  72    Example: for the following loop,
  73
  74    | loop_1 runs 1335 times
  75    |   loop_2 runs 1335 times
  76    |     A[{{0, +, 1}_1, +, 1335}_2]
  77    |     B[{{0, +, 1}_1, +, 1335}_2]
  78    |   endloop_2
  79    |   A[{0, +, 1336}_1]
  80    | endloop_1
  81
  82    gather_interchange_stats (in loop_1) will return
  83    DEPENDENCE_STEPS = 3002
  84    NB_DEPS_NOT_CARRIED_BY_LOOP = 5
  85    ACCESS_STRIDES = 10694
  86
  87    gather_interchange_stats (in loop_2) will return
  88    DEPENDENCE_STEPS = 3000
  89    NB_DEPS_NOT_CARRIED_BY_LOOP = 7
  90    ACCESS_STRIDES = 8010
  91 */
  92
  93 static void
  94 gather_interchange_stats (varray_type dependence_relations,
  95                           varray_type datarefs,
  96                           struct loop *loop,
  97                           struct loop *first_loop,
  98                           unsigned int *dependence_steps,
  99                           unsigned int *nb_deps_not_carried_by_loop,
 100                           unsigned int *access_strides)
 101 {
 102   unsigned int i;
 103
 104   *dependence_steps = 0;
 105   *nb_deps_not_carried_by_loop = 0;
 106   *access_strides = 0;
 107
 108   for (i = 0; i < VARRAY_ACTIVE_SIZE (dependence_relations); i++)
 109     {
 110       int dist;
 111       struct data_dependence_relation *ddr =
 112         (struct data_dependence_relation *)
 113         VARRAY_GENERIC_PTR (dependence_relations, i);
 114
 115       /* If we don't know anything about this dependence, or the distance
 116          vector is NULL, or there is no dependence, then there is no reuse of
 117          data.  */
 118
 119       if (DDR_DIST_VECT (ddr) == NULL
 120           || DDR_ARE_DEPENDENT (ddr) == chrec_dont_know
 121           || DDR_ARE_DEPENDENT (ddr) == chrec_known)
 122         continue;
 123
 124
 125
 126       dist = DDR_DIST_VECT (ddr)[loop->depth - first_loop->depth];
 127       if (dist == 0)
 128         (*nb_deps_not_carried_by_loop) += 1;
 129       else if (dist < 0)
 130         (*dependence_steps) += -dist;
 131       else
 132         (*dependence_steps) += dist;
 133     }
 134
 135   /* Compute the access strides.  */
 136   for (i = 0; i < VARRAY_ACTIVE_SIZE (datarefs); i++)
 137     {
 138       unsigned int it;
 139       struct data_reference *dr = VARRAY_GENERIC_PTR (datarefs, i);
 140       tree stmt = DR_STMT (dr);
 141       struct loop *stmt_loop = loop_containing_stmt (stmt);
 142       struct loop *inner_loop = first_loop->inner;
 143
 144       if (inner_loop != stmt_loop
 145           && !flow_loop_nested_p (inner_loop, stmt_loop))
 146         continue;
 147       for (it = 0; it < DR_NUM_DIMENSIONS (dr); it++)
 148         {
 149           tree chrec = DR_ACCESS_FN (dr, it);
 150           tree tstride = evolution_part_in_loop_num
 151             (chrec, loop->num);
 152
 153           if (tstride == NULL_TREE
 154               || TREE_CODE (tstride) != INTEGER_CST)
 155             continue;
 156
 157           (*access_strides) += int_cst_value (tstride);
 158         }
 159     }
 160 }
 161
 162 /* Attempt to apply interchange transformations to TRANS to maximize the
 163    spatial and temporal locality of the loop.
 164    Returns the new transform matrix.  The smaller the reuse vector
 165    distances in the inner loops, the fewer the cache misses.
 166    FIRST_LOOP is the loop->num of the first loop in the analyzed loop
 167    nest.  */
 168
 169
 170 static lambda_trans_matrix
 171 try_interchange_loops (lambda_trans_matrix trans,
 172                        unsigned int depth,
 173                        varray_type dependence_relations,
 174                        varray_type datarefs,
 175                        struct loop *first_loop)
 176 {
 177   struct loop *loop_i;
 178   struct loop *loop_j;
 179   unsigned int dependence_steps_i, dependence_steps_j;
 180   unsigned int access_strides_i, access_strides_j;
 181   unsigned int nb_deps_not_carried_by_i, nb_deps_not_carried_by_j;
 182   struct data_dependence_relation *ddr;
 183
 184   /* When there is an unknown relation in the dependence_relations, we
 185      know that it is no worth looking at this loop nest: give up.  */
 186   ddr = (struct data_dependence_relation *)
 187     VARRAY_GENERIC_PTR (dependence_relations, 0);
 188   if (ddr == NULL || DDR_ARE_DEPENDENT (ddr) == chrec_dont_know)
 189     return trans;
 190
 191   /* LOOP_I is always the outer loop.  */
 192   for (loop_j = first_loop->inner;
 193        loop_j;
 194        loop_j = loop_j->inner)
 195     for (loop_i = first_loop;
 196          loop_i->depth < loop_j->depth;
 197          loop_i = loop_i->inner)
 198       {
 199         gather_interchange_stats (dependence_relations, datarefs,
 200                                   loop_i, first_loop,
 201                                   &dependence_steps_i,
 202                                   &nb_deps_not_carried_by_i,
 203                                   &access_strides_i);
 204         gather_interchange_stats (dependence_relations, datarefs,
 205                                   loop_j, first_loop,
 206                                   &dependence_steps_j,
 207                                   &nb_deps_not_carried_by_j,
 208                                   &access_strides_j);
 209
 210         /* Heuristics for loop interchange profitability:
 211
 212            1. (spatial locality) Inner loops should have smallest
 213               dependence steps.
 214
 215            2. (spatial locality) Inner loops should contain more
 216            dependence relations not carried by the loop.
 217
 218            3. (temporal locality) Inner loops should have smallest
 219               array access strides.
 220         */
 221         if (dependence_steps_i < dependence_steps_j
 222             || nb_deps_not_carried_by_i > nb_deps_not_carried_by_j
 223             || access_strides_i < access_strides_j)
 224           {
 225             lambda_matrix_row_exchange (LTM_MATRIX (trans),
 226                                         loop_i->depth - first_loop->depth,
 227                                         loop_j->depth - first_loop->depth);
 228             /* Validate the resulting matrix.  When the transformation
 229                is not valid, reverse to the previous transformation.  */
 230             if (!lambda_transform_legal_p (trans, depth, dependence_relations))
 231               lambda_matrix_row_exchange (LTM_MATRIX (trans),
 232                                           loop_i->depth - first_loop->depth,
 233                                           loop_j->depth - first_loop->depth);
 234           }
 235       }
 236
 237   return trans;
 238 }
 239
 240 /* Perform a set of linear transforms on LOOPS.  */
 241
 242 void
 243 linear_transform_loops (struct loops *loops)
 244 {
 245   unsigned int i;
 246   VEC(tree,gc) *oldivs = NULL;  /* FIXME:These should really be on the
 247                                    heap.  (nathan 2005/04/15)*/
 248   VEC(tree,gc) *invariants = NULL;  /* FIXME:Likewise. */
 249
 250   for (i = 1; i < loops->num; i++)
 251     {
 252       unsigned int depth = 0;
 253       varray_type datarefs;
 254       varray_type dependence_relations;
 255       struct loop *loop_nest = loops->parray[i];
 256       struct loop *temp;
 257       lambda_loopnest before, after;
 258       lambda_trans_matrix trans;
 259       bool problem = false;
 260       bool need_perfect_nest = false;
 261       /* If it's not a loop nest, we don't want it.
 262          We also don't handle sibling loops properly,
 263          which are loops of the following form:
 264          for (i = 0; i < 50; i++)
 265            {
 266              for (j = 0; j < 50; j++)
 267                {
 268                 ...
 269                }
 270            for (j = 0; j < 50; j++)
 271                {
 272                 ...
 273                }
 274            } */
 275       if (!loop_nest || !loop_nest->inner)
 276         continue;
 277       VEC_truncate (tree, oldivs, 0);
 278       VEC_truncate (tree, invariants, 0);
 279       depth = 1;
 280       for (temp = loop_nest->inner; temp; temp = temp->inner)
 281         {
 282           /* If we have a sibling loop or multiple exit edges, jump ship.  */
 283           if (temp->next || !temp->single_exit)
 284             {
 285               problem = true;
 286               break;
 287             }
 288           depth ++;
 289         }
 290       if (problem)
 291         continue;
 292
 293       /* Analyze data references and dependence relations using scev.  */
 294
 295       VARRAY_GENERIC_PTR_INIT (datarefs, 10, "datarefs");
 296       VARRAY_GENERIC_PTR_INIT (dependence_relations, 10,
 297                                "dependence_relations");
 298
 299
 300       compute_data_dependences_for_loop (depth, loop_nest,
 301                                          &datarefs, &dependence_relations);
 302       if (dump_file && (dump_flags & TDF_DETAILS))
 303         {
 304           unsigned int j;
 305           for (j = 0; j < VARRAY_ACTIVE_SIZE (dependence_relations); j++)
 306             {
 307               struct data_dependence_relation *ddr =
 308                 (struct data_dependence_relation *)
 309                 VARRAY_GENERIC_PTR (dependence_relations, j);
 310
 311               if (DDR_ARE_DEPENDENT (ddr) == NULL_TREE)
 312                 {
 313                   fprintf (dump_file, "DISTANCE_V (");
 314                   print_lambda_vector (dump_file, DDR_DIST_VECT (ddr),
 315                                        DDR_SIZE_VECT (ddr));
 316                   fprintf (dump_file, ")\n");
 317                   fprintf (dump_file, "DIRECTION_V (");
 318                   print_lambda_vector (dump_file, DDR_DIR_VECT (ddr),
 319                                        DDR_SIZE_VECT (ddr));
 320                   fprintf (dump_file, ")\n");
 321                 }
 322             }
 323           fprintf (dump_file, "\n\n");
 324         }
 325       /* Build the transformation matrix.  */
 326       trans = lambda_trans_matrix_new (depth, depth);
 327       lambda_matrix_id (LTM_MATRIX (trans), depth);
 328
 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           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           continue;
 345         }
 346       if (!perfect_nest_p (loop_nest))
 347         need_perfect_nest = true;
 348       before = gcc_loopnest_to_lambda_loopnest (loops,
 349                                                 loop_nest, &oldivs,
 350                                                 &invariants,
 351                                                 need_perfect_nest);
 352       if (!before)
 353         continue;
 354
 355       if (dump_file)
 356         {
 357           fprintf (dump_file, "Before:\n");
 358           print_lambda_loopnest (dump_file, before, 'i');
 359         }
 360
 361       after = lambda_loopnest_transform (before, trans);
 362       if (dump_file)
 363         {
 364           fprintf (dump_file, "After:\n");
 365           print_lambda_loopnest (dump_file, after, 'u');
 366         }
 367       lambda_loopnest_to_gcc_loopnest (loop_nest, oldivs, invariants,
 368                                        after, trans);
 369       if (dump_file)
 370         fprintf (dump_file, "Successfully transformed loop.\n");
 371       free_dependence_relations (dependence_relations);
 372       free_data_refs (datarefs);
 373     }
 374   VEC_free (tree, gc, oldivs);
 375   VEC_free (tree, gc, invariants);
 376   scev_reset ();
 377   update_ssa (TODO_update_ssa);
 378   rewrite_into_loop_closed_ssa (NULL);
 379 }