summaryrefslogtreecommitdiff
path: root/gcc/tree-loop-distribution.c
blob: 13b78cc37d2f713c4d219d158d103f938e30c5e0 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
/* Loop distribution.
   Copyright (C) 2006-2013 Free Software Foundation, Inc.
   Contributed by Georges-Andre Silber <Georges-Andre.Silber@ensmp.fr>
   and Sebastian Pop <sebastian.pop@amd.com>.

This file is part of GCC.

GCC is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 3, or (at your option) any
later version.

GCC is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
for more details.

You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING3.  If not see
<http://www.gnu.org/licenses/>.  */

/* This pass performs loop distribution: for example, the loop

   |DO I = 2, N
   |    A(I) = B(I) + C
   |    D(I) = A(I-1)*E
   |ENDDO

   is transformed to

   |DOALL I = 2, N
   |   A(I) = B(I) + C
   |ENDDO
   |
   |DOALL I = 2, N
   |   D(I) = A(I-1)*E
   |ENDDO

   This pass uses an RDG, Reduced Dependence Graph built on top of the
   data dependence relations.  The RDG is then topologically sorted to
   obtain a map of information producers/consumers based on which it
   generates the new loops.  */

#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "tree-flow.h"
#include "cfgloop.h"
#include "tree-chrec.h"
#include "tree-data-ref.h"
#include "tree-scalar-evolution.h"
#include "tree-pass.h"

enum partition_kind {
    PKIND_NORMAL, PKIND_REDUCTION, PKIND_MEMSET, PKIND_MEMCPY
};

typedef struct partition_s
{
  bitmap stmts;
  bool has_writes;
  enum partition_kind kind;
  /* data-references a kind != PKIND_NORMAL partition is about.  */
  data_reference_p main_dr;
  data_reference_p secondary_dr;
} *partition_t;


/* Allocate and initialize a partition from BITMAP.  */

static partition_t
partition_alloc (bitmap stmts)
{
  partition_t partition = XCNEW (struct partition_s);
  partition->stmts = stmts ? stmts : BITMAP_ALLOC (NULL);
  partition->has_writes = false;
  partition->kind = PKIND_NORMAL;
  return partition;
}

/* Free PARTITION.  */

static void
partition_free (partition_t partition)
{
  BITMAP_FREE (partition->stmts);
  free (partition);
}

/* Returns true if the partition can be generated as a builtin.  */

static bool
partition_builtin_p (partition_t partition)
{
  return partition->kind > PKIND_REDUCTION;
}

/* Returns true if the partition has an writes.  */

static bool
partition_has_writes (partition_t partition)
{
  return partition->has_writes;
}

/* If bit I is not set, it means that this node represents an
   operation that has already been performed, and that should not be
   performed again.  This is the subgraph of remaining important
   computations that is passed to the DFS algorithm for avoiding to
   include several times the same stores in different loops.  */
static bitmap remaining_stmts;

/* A node of the RDG is marked in this bitmap when it has as a
   predecessor a node that writes to memory.  */
static bitmap upstream_mem_writes;

/* Returns true when DEF is an SSA_NAME defined in LOOP and used after
   the LOOP.  */

static bool
ssa_name_has_uses_outside_loop_p (tree def, loop_p loop)
{
  imm_use_iterator imm_iter;
  use_operand_p use_p;

  FOR_EACH_IMM_USE_FAST (use_p, imm_iter, def)
    {
      gimple use_stmt = USE_STMT (use_p);
      if (!is_gimple_debug (use_stmt)
	  && loop != loop_containing_stmt (use_stmt))
	return true;
    }

  return false;
}

/* Returns true when STMT defines a scalar variable used after the
   loop LOOP.  */

static bool
stmt_has_scalar_dependences_outside_loop (loop_p loop, gimple stmt)
{
  def_operand_p def_p;
  ssa_op_iter op_iter;

  if (gimple_code (stmt) == GIMPLE_PHI)
    return ssa_name_has_uses_outside_loop_p (gimple_phi_result (stmt), loop);

  FOR_EACH_SSA_DEF_OPERAND (def_p, stmt, op_iter, SSA_OP_DEF)
    if (ssa_name_has_uses_outside_loop_p (DEF_FROM_PTR (def_p), loop))
      return true;

  return false;
}

/* Return a copy of LOOP placed before LOOP.  */

static struct loop *
copy_loop_before (struct loop *loop)
{
  struct loop *res;
  edge preheader = loop_preheader_edge (loop);

  initialize_original_copy_tables ();
  res = slpeel_tree_duplicate_loop_to_edge_cfg (loop, preheader);
  gcc_assert (res != NULL);
  free_original_copy_tables ();
  delete_update_ssa ();

  return res;
}

/* Creates an empty basic block after LOOP.  */

static void
create_bb_after_loop (struct loop *loop)
{
  edge exit = single_exit (loop);

  if (!exit)
    return;

  split_edge (exit);
}

/* Generate code for PARTITION from the code in LOOP.  The loop is
   copied when COPY_P is true.  All the statements not flagged in the
   PARTITION bitmap are removed from the loop or from its copy.  The
   statements are indexed in sequence inside a basic block, and the
   basic blocks of a loop are taken in dom order.  */

static void
generate_loops_for_partition (struct loop *loop, partition_t partition,
			      bool copy_p)
{
  unsigned i, x;
  gimple_stmt_iterator bsi;
  basic_block *bbs;

  if (copy_p)
    {
      loop = copy_loop_before (loop);
      gcc_assert (loop != NULL);
      create_preheader (loop, CP_SIMPLE_PREHEADERS);
      create_bb_after_loop (loop);
    }

  /* Remove stmts not in the PARTITION bitmap.  The order in which we
     visit the phi nodes and the statements is exactly as in
     stmts_from_loop.  */
  bbs = get_loop_body_in_dom_order (loop);

  if (MAY_HAVE_DEBUG_STMTS)
    for (x = 0, i = 0; i < loop->num_nodes; i++)
      {
	basic_block bb = bbs[i];

	for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi); gsi_next (&bsi))
	  if (!bitmap_bit_p (partition->stmts, x++))
	    reset_debug_uses (gsi_stmt (bsi));

	for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
	  {
	    gimple stmt = gsi_stmt (bsi);
	    if (gimple_code (stmt) != GIMPLE_LABEL
		&& !is_gimple_debug (stmt)
		&& !bitmap_bit_p (partition->stmts, x++))
	      reset_debug_uses (stmt);
	  }
      }

  for (x = 0, i = 0; i < loop->num_nodes; i++)
    {
      basic_block bb = bbs[i];

      for (bsi = gsi_start_phis (bb); !gsi_end_p (bsi);)
	if (!bitmap_bit_p (partition->stmts, x++))
	  {
	    gimple phi = gsi_stmt (bsi);
	    if (virtual_operand_p (gimple_phi_result (phi)))
	      mark_virtual_phi_result_for_renaming (phi);
	    remove_phi_node (&bsi, true);
	  }
	else
	  gsi_next (&bsi);

      for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi);)
	{
	  gimple stmt = gsi_stmt (bsi);
	  if (gimple_code (stmt) != GIMPLE_LABEL
	      && !is_gimple_debug (stmt)
	      && !bitmap_bit_p (partition->stmts, x++))
	    {
	      unlink_stmt_vdef (stmt);
	      gsi_remove (&bsi, true);
	      release_defs (stmt);
	    }
	  else
	    gsi_next (&bsi);
	}
    }

  free (bbs);
}

/* Build the size argument for a memory operation call.  */

static tree
build_size_arg_loc (location_t loc, data_reference_p dr, tree nb_iter)
{
  tree size;
  size = fold_build2_loc (loc, MULT_EXPR, sizetype,
			  fold_convert_loc (loc, sizetype, nb_iter),
			  TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dr))));
  return fold_convert_loc (loc, size_type_node, size);
}

/* Build an address argument for a memory operation call.  */

static tree
build_addr_arg_loc (location_t loc, data_reference_p dr, tree nb_bytes)
{
  tree addr_base;

  addr_base = size_binop_loc (loc, PLUS_EXPR, DR_OFFSET (dr), DR_INIT (dr));
  addr_base = fold_convert_loc (loc, sizetype, addr_base);

  /* Test for a negative stride, iterating over every element.  */
  if (tree_int_cst_sgn (DR_STEP (dr)) == -1)
    {
      addr_base = size_binop_loc (loc, MINUS_EXPR, addr_base,
				  fold_convert_loc (loc, sizetype, nb_bytes));
      addr_base = size_binop_loc (loc, PLUS_EXPR, addr_base,
				  TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dr))));
    }

  return fold_build_pointer_plus_loc (loc, DR_BASE_ADDRESS (dr), addr_base);
}

/* If VAL memory representation contains the same value in all bytes,
   return that value, otherwise return -1.
   E.g. for 0x24242424 return 0x24, for IEEE double
   747708026454360457216.0 return 0x44, etc.  */

static int
const_with_all_bytes_same (tree val)
{
  unsigned char buf[64];
  int i, len;

  if (integer_zerop (val)
      || real_zerop (val)
      || (TREE_CODE (val) == CONSTRUCTOR
          && !TREE_CLOBBER_P (val)
          && CONSTRUCTOR_NELTS (val) == 0))
    return 0;

  if (CHAR_BIT != 8 || BITS_PER_UNIT != 8)
    return -1;

  len = native_encode_expr (val, buf, sizeof (buf));
  if (len == 0)
    return -1;
  for (i = 1; i < len; i++)
    if (buf[i] != buf[0])
      return -1;
  return buf[0];
}

/* Generate a call to memset for PARTITION in LOOP.  */

static void
generate_memset_builtin (struct loop *loop, partition_t partition)
{
  gimple_stmt_iterator gsi;
  gimple stmt, fn_call;
  tree nb_iter, mem, fn, nb_bytes;
  location_t loc;
  tree val;

  stmt = DR_STMT (partition->main_dr);
  loc = gimple_location (stmt);
  if (gimple_bb (stmt) == loop->latch)
    nb_iter = number_of_latch_executions (loop);
  else
    nb_iter = number_of_exit_cond_executions (loop);

  /* The new statements will be placed before LOOP.  */
  gsi = gsi_last_bb (loop_preheader_edge (loop)->src);

  nb_bytes = build_size_arg_loc (loc, partition->main_dr, nb_iter);
  nb_bytes = force_gimple_operand_gsi (&gsi, nb_bytes, true, NULL_TREE,
				       false, GSI_CONTINUE_LINKING);
  mem = build_addr_arg_loc (loc, partition->main_dr, nb_bytes);
  mem = force_gimple_operand_gsi (&gsi, mem, true, NULL_TREE,
				  false, GSI_CONTINUE_LINKING);

  /* This exactly matches the pattern recognition in classify_partition.  */
  val = gimple_assign_rhs1 (stmt);
  /* Handle constants like 0x15151515 and similarly
     floating point constants etc. where all bytes are the same.  */
  int bytev = const_with_all_bytes_same (val);
  if (bytev != -1)
    val = build_int_cst (integer_type_node, bytev);
  else if (TREE_CODE (val) == INTEGER_CST)
    val = fold_convert (integer_type_node, val);
  else if (!useless_type_conversion_p (integer_type_node, TREE_TYPE (val)))
    {
      gimple cstmt;
      tree tem = make_ssa_name (integer_type_node, NULL);
      cstmt = gimple_build_assign_with_ops (NOP_EXPR, tem, val, NULL_TREE);
      gsi_insert_after (&gsi, cstmt, GSI_CONTINUE_LINKING);
      val = tem;
    }

  fn = build_fold_addr_expr (builtin_decl_implicit (BUILT_IN_MEMSET));
  fn_call = gimple_build_call (fn, 3, mem, val, nb_bytes);
  gsi_insert_after (&gsi, fn_call, GSI_CONTINUE_LINKING);

  if (dump_file && (dump_flags & TDF_DETAILS))
    {
      fprintf (dump_file, "generated memset");
      if (bytev == 0)
	fprintf (dump_file, " zero\n");
      else
	fprintf (dump_file, "\n");
    }
}

/* Generate a call to memcpy for PARTITION in LOOP.  */

static void
generate_memcpy_builtin (struct loop *loop, partition_t partition)
{
  gimple_stmt_iterator gsi;
  gimple stmt, fn_call;
  tree nb_iter, dest, src, fn, nb_bytes;
  location_t loc;
  enum built_in_function kind;

  stmt = DR_STMT (partition->main_dr);
  loc = gimple_location (stmt);
  if (gimple_bb (stmt) == loop->latch)
    nb_iter = number_of_latch_executions (loop);
  else
    nb_iter = number_of_exit_cond_executions (loop);

  /* The new statements will be placed before LOOP.  */
  gsi = gsi_last_bb (loop_preheader_edge (loop)->src);

  nb_bytes = build_size_arg_loc (loc, partition->main_dr, nb_iter);
  nb_bytes = force_gimple_operand_gsi (&gsi, nb_bytes, true, NULL_TREE,
				       false, GSI_CONTINUE_LINKING);
  dest = build_addr_arg_loc (loc, partition->main_dr, nb_bytes);
  src = build_addr_arg_loc (loc, partition->secondary_dr, nb_bytes);
  if (ptr_derefs_may_alias_p (dest, src))
    kind = BUILT_IN_MEMMOVE;
  else
    kind = BUILT_IN_MEMCPY;

  dest = force_gimple_operand_gsi (&gsi, dest, true, NULL_TREE,
				   false, GSI_CONTINUE_LINKING);
  src = force_gimple_operand_gsi (&gsi, src, true, NULL_TREE,
				  false, GSI_CONTINUE_LINKING);
  fn = build_fold_addr_expr (builtin_decl_implicit (kind));
  fn_call = gimple_build_call (fn, 3, dest, src, nb_bytes);
  gsi_insert_after (&gsi, fn_call, GSI_CONTINUE_LINKING);

  if (dump_file && (dump_flags & TDF_DETAILS))
    {
      if (kind == BUILT_IN_MEMCPY)
	fprintf (dump_file, "generated memcpy\n");
      else
	fprintf (dump_file, "generated memmove\n");
    }
}

/* Remove and destroy the loop LOOP.  */

static void
destroy_loop (struct loop *loop)
{
  unsigned nbbs = loop->num_nodes;
  edge exit = single_exit (loop);
  basic_block src = loop_preheader_edge (loop)->src, dest = exit->dest;
  basic_block *bbs;
  unsigned i;

  bbs = get_loop_body_in_dom_order (loop);

  redirect_edge_pred (exit, src);
  exit->flags &= ~(EDGE_TRUE_VALUE|EDGE_FALSE_VALUE);
  exit->flags |= EDGE_FALLTHRU;
  cancel_loop_tree (loop);
  rescan_loop_exit (exit, false, true);

  for (i = 0; i < nbbs; i++)
    {
      /* We have made sure to not leave any dangling uses of SSA
         names defined in the loop.  With the exception of virtuals.
	 Make sure we replace all uses of virtual defs that will remain
	 outside of the loop with the bare symbol as delete_basic_block
	 will release them.  */
      gimple_stmt_iterator gsi;
      for (gsi = gsi_start_phis (bbs[i]); !gsi_end_p (gsi); gsi_next (&gsi))
	{
	  gimple phi = gsi_stmt (gsi);
	  if (virtual_operand_p (gimple_phi_result (phi)))
	    mark_virtual_phi_result_for_renaming (phi);
	}
      for (gsi = gsi_start_bb (bbs[i]); !gsi_end_p (gsi); gsi_next (&gsi))
	{
	  gimple stmt = gsi_stmt (gsi);
	  tree vdef = gimple_vdef (stmt);
	  if (vdef && TREE_CODE (vdef) == SSA_NAME)
	    mark_virtual_operand_for_renaming (vdef);
	}
      delete_basic_block (bbs[i]);
    }
  free (bbs);

  set_immediate_dominator (CDI_DOMINATORS, dest,
			   recompute_dominator (CDI_DOMINATORS, dest));
}

/* Generates code for PARTITION.  */

static void
generate_code_for_partition (struct loop *loop,
			     partition_t partition, bool copy_p)
{
  switch (partition->kind)
    {
    case PKIND_MEMSET:
      generate_memset_builtin (loop, partition);
      /* If this is the last partition for which we generate code, we have
	 to destroy the loop.  */
      if (!copy_p)
	destroy_loop (loop);
      break;

    case PKIND_MEMCPY:
      generate_memcpy_builtin (loop, partition);
      /* If this is the last partition for which we generate code, we have
	 to destroy the loop.  */
      if (!copy_p)
	destroy_loop (loop);
      break;

    case PKIND_REDUCTION:
      /* Reductions all have to be in the last partition.  */
      gcc_assert (!copy_p);
    case PKIND_NORMAL:
      generate_loops_for_partition (loop, partition, copy_p);
      break;

    default:
      gcc_unreachable ();
    }
}


/* Returns true if the node V of RDG cannot be recomputed.  */

static bool
rdg_cannot_recompute_vertex_p (struct graph *rdg, int v)
{
  if (RDG_MEM_WRITE_STMT (rdg, v))
    return true;

  return false;
}

/* Returns true when the vertex V has already been generated in the
   current partition (V is in PROCESSED), or when V belongs to another
   partition and cannot be recomputed (V is not in REMAINING_STMTS).  */

static inline bool
already_processed_vertex_p (bitmap processed, int v)
{
  return (bitmap_bit_p (processed, v)
	  || !bitmap_bit_p (remaining_stmts, v));
}

/* Returns NULL when there is no anti-dependence among the successors
   of vertex V, otherwise returns the edge with the anti-dep.  */

static struct graph_edge *
has_anti_dependence (struct vertex *v)
{
  struct graph_edge *e;

  if (v->succ)
    for (e = v->succ; e; e = e->succ_next)
      if (RDGE_TYPE (e) == anti_dd)
	return e;

  return NULL;
}

/* Returns true when V has an anti-dependence edge among its successors.  */

static bool
predecessor_has_mem_write (struct graph *rdg, struct vertex *v)
{
  struct graph_edge *e;

  if (v->pred)
    for (e = v->pred; e; e = e->pred_next)
      if (bitmap_bit_p (upstream_mem_writes, e->src)
	  /* Don't consider flow channels: a write to memory followed
	     by a read from memory.  These channels allow the split of
	     the RDG in different partitions.  */
	  && !RDG_MEM_WRITE_STMT (rdg, e->src))
	return true;

  return false;
}

/* Initializes the upstream_mem_writes bitmap following the
   information from RDG.  */

static void
mark_nodes_having_upstream_mem_writes (struct graph *rdg)
{
  int v, x;
  bitmap seen = BITMAP_ALLOC (NULL);

  for (v = rdg->n_vertices - 1; v >= 0; v--)
    if (!bitmap_bit_p (seen, v))
      {
	unsigned i;
	vec<int> nodes;
	nodes.create (3);

	graphds_dfs (rdg, &v, 1, &nodes, false, NULL);

	FOR_EACH_VEC_ELT (nodes, i, x)
	  {
	    if (!bitmap_set_bit (seen, x))
	      continue;

	    if (RDG_MEM_WRITE_STMT (rdg, x)
		|| predecessor_has_mem_write (rdg, &(rdg->vertices[x]))
		/* In anti dependences the read should occur before
		   the write, this is why both the read and the write
		   should be placed in the same partition.  */
		|| has_anti_dependence (&(rdg->vertices[x])))
	      {
		bitmap_set_bit (upstream_mem_writes, x);
	      }
	  }

	nodes.release ();
      }
}

/* Returns true when vertex u has a memory write node as a predecessor
   in RDG.  */

static bool
has_upstream_mem_writes (int u)
{
  return bitmap_bit_p (upstream_mem_writes, u);
}

static void rdg_flag_vertex_and_dependent (struct graph *, int, partition_t,
					   bitmap, bitmap);

/* Flag the uses of U stopping following the information from
   upstream_mem_writes.  */

static void
rdg_flag_uses (struct graph *rdg, int u, partition_t partition, bitmap loops,
	       bitmap processed)
{
  use_operand_p use_p;
  struct vertex *x = &(rdg->vertices[u]);
  gimple stmt = RDGV_STMT (x);
  struct graph_edge *anti_dep = has_anti_dependence (x);

  /* Keep in the same partition the destination of an antidependence,
     because this is a store to the exact same location.  Putting this
     in another partition is bad for cache locality.  */
  if (anti_dep)
    {
      int v = anti_dep->dest;

      if (!already_processed_vertex_p (processed, v))
	rdg_flag_vertex_and_dependent (rdg, v, partition, loops,
				       processed);
    }

  if (gimple_code (stmt) != GIMPLE_PHI)
    {
      if ((use_p = gimple_vuse_op (stmt)) != NULL_USE_OPERAND_P)
	{
	  tree use = USE_FROM_PTR (use_p);

	  if (TREE_CODE (use) == SSA_NAME
	      && !SSA_NAME_IS_DEFAULT_DEF (use))
	    {
	      gimple def_stmt = SSA_NAME_DEF_STMT (use);
	      int v = rdg_vertex_for_stmt (rdg, def_stmt);

	      if (v >= 0
		  && !already_processed_vertex_p (processed, v))
		rdg_flag_vertex_and_dependent (rdg, v, partition, loops,
					       processed);
	    }
	}
    }

  if (is_gimple_assign (stmt) && has_upstream_mem_writes (u))
    {
      tree op0 = gimple_assign_lhs (stmt);

      /* Scalar channels don't have enough space for transmitting data
	 between tasks, unless we add more storage by privatizing.  */
      if (is_gimple_reg (op0))
	{
	  use_operand_p use_p;
	  imm_use_iterator iter;

	  FOR_EACH_IMM_USE_FAST (use_p, iter, op0)
	    {
	      int v = rdg_vertex_for_stmt (rdg, USE_STMT (use_p));

	      if (!already_processed_vertex_p (processed, v))
		rdg_flag_vertex_and_dependent (rdg, v, partition, loops,
					       processed);
	    }
	}
    }
}

/* Flag V from RDG as part of PARTITION, and also flag its loop number
   in LOOPS.  */

static void
rdg_flag_vertex (struct graph *rdg, int v, partition_t partition, bitmap loops)
{
  struct loop *loop;

  if (!bitmap_set_bit (partition->stmts, v))
    return;

  loop = loop_containing_stmt (RDG_STMT (rdg, v));
  bitmap_set_bit (loops, loop->num);

  if (rdg_cannot_recompute_vertex_p (rdg, v))
    {
      partition->has_writes = true;
      bitmap_clear_bit (remaining_stmts, v);
    }
}

/* Flag in the bitmap PARTITION the vertex V and all its predecessors.
   Also flag their loop number in LOOPS.  */

static void
rdg_flag_vertex_and_dependent (struct graph *rdg, int v, partition_t partition,
			       bitmap loops, bitmap processed)
{
  unsigned i;
  vec<int> nodes;
  nodes.create (3);
  int x;

  bitmap_set_bit (processed, v);
  rdg_flag_uses (rdg, v, partition, loops, processed);
  graphds_dfs (rdg, &v, 1, &nodes, false, remaining_stmts);
  rdg_flag_vertex (rdg, v, partition, loops);

  FOR_EACH_VEC_ELT (nodes, i, x)
    if (!already_processed_vertex_p (processed, x))
      rdg_flag_vertex_and_dependent (rdg, x, partition, loops, processed);

  nodes.release ();
}

/* Initialize CONDS with all the condition statements from the basic
   blocks of LOOP.  */

static void
collect_condition_stmts (struct loop *loop, vec<gimple> *conds)
{
  unsigned i;
  edge e;
  vec<edge> exits = get_loop_exit_edges (loop);

  FOR_EACH_VEC_ELT (exits, i, e)
    {
      gimple cond = last_stmt (e->src);

      if (cond)
	conds->safe_push (cond);
    }

  exits.release ();
}

/* Add to PARTITION all the exit condition statements for LOOPS
   together with all their dependent statements determined from
   RDG.  */

static void
rdg_flag_loop_exits (struct graph *rdg, bitmap loops, partition_t partition,
		     bitmap processed)
{
  unsigned i;
  bitmap_iterator bi;
  vec<gimple> conds;
  conds.create (3);

  EXECUTE_IF_SET_IN_BITMAP (loops, 0, i, bi)
    collect_condition_stmts (get_loop (cfun, i), &conds);

  while (!conds.is_empty ())
    {
      gimple cond = conds.pop ();
      int v = rdg_vertex_for_stmt (rdg, cond);
      bitmap new_loops = BITMAP_ALLOC (NULL);

      if (!already_processed_vertex_p (processed, v))
	rdg_flag_vertex_and_dependent (rdg, v, partition, new_loops, processed);

      EXECUTE_IF_SET_IN_BITMAP (new_loops, 0, i, bi)
	if (bitmap_set_bit (loops, i))
	  collect_condition_stmts (get_loop (cfun, i), &conds);

      BITMAP_FREE (new_loops);
    }

  conds.release ();
}

/* Returns a bitmap in which all the statements needed for computing
   the strongly connected component C of the RDG are flagged, also
   including the loop exit conditions.  */

static partition_t
build_rdg_partition_for_component (struct graph *rdg, rdgc c)
{
  int i, v;
  partition_t partition = partition_alloc (NULL);
  bitmap loops = BITMAP_ALLOC (NULL);
  bitmap processed = BITMAP_ALLOC (NULL);

  FOR_EACH_VEC_ELT (c->vertices, i, v)
    if (!already_processed_vertex_p (processed, v))
      rdg_flag_vertex_and_dependent (rdg, v, partition, loops, processed);

  rdg_flag_loop_exits (rdg, loops, partition, processed);

  BITMAP_FREE (processed);
  BITMAP_FREE (loops);
  return partition;
}

/* Free memory for COMPONENTS.  */

static void
free_rdg_components (vec<rdgc> components)
{
  int i;
  rdgc x;

  FOR_EACH_VEC_ELT (components, i, x)
    {
      x->vertices.release ();
      free (x);
    }

  components.release ();
}

/* Build the COMPONENTS vector with the strongly connected components
   of RDG in which the STARTING_VERTICES occur.  */

static void
rdg_build_components (struct graph *rdg, vec<int> starting_vertices,
		      vec<rdgc> *components)
{
  int i, v;
  bitmap saved_components = BITMAP_ALLOC (NULL);
  int n_components = graphds_scc (rdg, NULL);
  /* ??? Macros cannot process template types with more than one
     argument, so we need this typedef.  */
  typedef vec<int> vec_int_heap;
  vec<int> *all_components = XNEWVEC (vec_int_heap, n_components);

  for (i = 0; i < n_components; i++)
    all_components[i].create (3);

  for (i = 0; i < rdg->n_vertices; i++)
    all_components[rdg->vertices[i].component].safe_push (i);

  FOR_EACH_VEC_ELT (starting_vertices, i, v)
    {
      int c = rdg->vertices[v].component;

      if (bitmap_set_bit (saved_components, c))
	{
	  rdgc x = XCNEW (struct rdg_component);
	  x->num = c;
	  x->vertices = all_components[c];

	  components->safe_push (x);
	}
    }

  for (i = 0; i < n_components; i++)
    if (!bitmap_bit_p (saved_components, i))
      all_components[i].release ();

  free (all_components);
  BITMAP_FREE (saved_components);
}

/* Classifies the builtin kind we can generate for PARTITION of RDG and LOOP.
   For the moment we detect only the memset zero pattern.  */

static void
classify_partition (loop_p loop, struct graph *rdg, partition_t partition)
{
  bitmap_iterator bi;
  unsigned i;
  tree nb_iter;
  data_reference_p single_load, single_store;
  bool volatiles_p = false;

  partition->kind = PKIND_NORMAL;
  partition->main_dr = NULL;
  partition->secondary_dr = NULL;

  EXECUTE_IF_SET_IN_BITMAP (partition->stmts, 0, i, bi)
    {
      gimple stmt = RDG_STMT (rdg, i);

      if (gimple_has_volatile_ops (stmt))
	volatiles_p = true;

      /* If the stmt has uses outside of the loop fail.
	 ???  If the stmt is generated in another partition that
	 is not created as builtin we can ignore this.  */
      if (stmt_has_scalar_dependences_outside_loop (loop, stmt))
	{
	  if (dump_file && (dump_flags & TDF_DETAILS))
	    fprintf (dump_file, "not generating builtin, partition has "
		     "scalar uses outside of the loop\n");
	  partition->kind = PKIND_REDUCTION;
	  return;
	}
    }

  /* Perform general partition disqualification for builtins.  */
  if (volatiles_p
      || !flag_tree_loop_distribute_patterns)
    return;

  nb_iter = number_of_exit_cond_executions (loop);
  if (!nb_iter || nb_iter == chrec_dont_know)
    return;

  /* Detect memset and memcpy.  */
  single_load = NULL;
  single_store = NULL;
  EXECUTE_IF_SET_IN_BITMAP (partition->stmts, 0, i, bi)
    {
      gimple stmt = RDG_STMT (rdg, i);
      data_reference_p dr;
      unsigned j;

      if (gimple_code (stmt) == GIMPLE_PHI)
	continue;

      /* Any scalar stmts are ok.  */
      if (!gimple_vuse (stmt))
	continue;

      /* Otherwise just regular loads/stores.  */
      if (!gimple_assign_single_p (stmt))
	return;

      /* But exactly one store and/or load.  */
      for (j = 0; RDG_DATAREFS (rdg, i).iterate (j, &dr); ++j)
	{
	  if (DR_IS_READ (dr))
	    {
	      if (single_load != NULL)
		return;
	      single_load = dr;
	    }
	  else
	    {
	      if (single_store != NULL)
		return;
	      single_store = dr;
	    }
	}
    }

  if (single_store && !single_load)
    {
      gimple stmt = DR_STMT (single_store);
      tree rhs = gimple_assign_rhs1 (stmt);
      if (const_with_all_bytes_same (rhs) == -1
	  && (!INTEGRAL_TYPE_P (TREE_TYPE (rhs))
	      || (TYPE_MODE (TREE_TYPE (rhs))
		  != TYPE_MODE (unsigned_char_type_node))))
	return;
      if (TREE_CODE (rhs) == SSA_NAME
	  && !SSA_NAME_IS_DEFAULT_DEF (rhs)
	  && flow_bb_inside_loop_p (loop, gimple_bb (SSA_NAME_DEF_STMT (rhs))))
	return;
      if (!adjacent_dr_p (single_store))
	return;
      partition->kind = PKIND_MEMSET;
      partition->main_dr = single_store;
    }
  else if (single_store && single_load)
    {
      gimple store = DR_STMT (single_store);
      gimple load = DR_STMT (single_load);
      /* Direct aggregate copy or via an SSA name temporary.  */
      if (load != store
	  && gimple_assign_lhs (load) != gimple_assign_rhs1 (store))
	return;
      if (!adjacent_dr_p (single_store)
	  || !adjacent_dr_p (single_load)
	  || !operand_equal_p (DR_STEP (single_store),
			       DR_STEP (single_load), 0))
	return;
      /* Now check that if there is a dependence this dependence is
         of a suitable form for memmove.  */
      vec<loop_p> loops = vNULL;
      ddr_p ddr;
      loops.safe_push (loop);
      ddr = initialize_data_dependence_relation (single_load, single_store,
						 loops);
      compute_affine_dependence (ddr, loop);
      if (DDR_ARE_DEPENDENT (ddr) == chrec_dont_know)
	{
	  free_dependence_relation (ddr);
	  loops.release ();
	  return;
	}
      if (DDR_ARE_DEPENDENT (ddr) != chrec_known)
	{
	  if (DDR_NUM_DIST_VECTS (ddr) == 0)
	    {
	      free_dependence_relation (ddr);
	      loops.release ();
	      return;
	    }
	  lambda_vector dist_v;
	  FOR_EACH_VEC_ELT (DDR_DIST_VECTS (ddr), i, dist_v)
	    {
	      int dist = dist_v[index_in_loop_nest (loop->num,
						    DDR_LOOP_NEST (ddr))];
	      if (dist > 0 && !DDR_REVERSED_P (ddr))
		{
		  free_dependence_relation (ddr);
		  loops.release ();
		  return;
		}
	    }
	}
      free_dependence_relation (ddr);
      loops.release ();
      partition->kind = PKIND_MEMCPY;
      partition->main_dr = single_store;
      partition->secondary_dr = single_load;
    }
}

/* For a data reference REF, return the declaration of its base
   address or NULL_TREE if the base is not determined.  */

static tree
ref_base_address (data_reference_p dr)
{
  tree base_address = DR_BASE_ADDRESS (dr);
  if (base_address
      && TREE_CODE (base_address) == ADDR_EXPR)
    return TREE_OPERAND (base_address, 0);

  return base_address;
}

/* Returns true when PARTITION1 and PARTITION2 have similar memory
   accesses in RDG.  */

static bool
similar_memory_accesses (struct graph *rdg, partition_t partition1,
			 partition_t partition2)
{
  unsigned i, j, k, l;
  bitmap_iterator bi, bj;
  data_reference_p ref1, ref2;

  /* First check whether in the intersection of the two partitions are
     any loads or stores.  Common loads are the situation that happens
     most often.  */
  EXECUTE_IF_AND_IN_BITMAP (partition1->stmts, partition2->stmts, 0, i, bi)
    if (RDG_MEM_WRITE_STMT (rdg, i)
	|| RDG_MEM_READS_STMT (rdg, i))
      return true;

  /* Then check all data-references against each other.  */
  EXECUTE_IF_SET_IN_BITMAP (partition1->stmts, 0, i, bi)
    if (RDG_MEM_WRITE_STMT (rdg, i)
	|| RDG_MEM_READS_STMT (rdg, i))
      EXECUTE_IF_SET_IN_BITMAP (partition2->stmts, 0, j, bj)
	if (RDG_MEM_WRITE_STMT (rdg, j)
	    || RDG_MEM_READS_STMT (rdg, j))
	  {
	    FOR_EACH_VEC_ELT (RDG_DATAREFS (rdg, i), k, ref1)
	      {
		tree base1 = ref_base_address (ref1);
		if (base1)
		  FOR_EACH_VEC_ELT (RDG_DATAREFS (rdg, j), l, ref2)
		    if (base1 == ref_base_address (ref2))
		      return true;
	      }
	  }

  return false;
}

/* Aggregate several components into a useful partition that is
   registered in the PARTITIONS vector.  Partitions will be
   distributed in different loops.  */

static void
rdg_build_partitions (struct graph *rdg, vec<rdgc> components,
		      vec<int> *other_stores,
		      vec<partition_t> *partitions, bitmap processed)
{
  int i;
  rdgc x;
  partition_t partition = partition_alloc (NULL);

  FOR_EACH_VEC_ELT (components, i, x)
    {
      partition_t np;
      int v = x->vertices[0];

      if (bitmap_bit_p (processed, v))
	continue;

      np = build_rdg_partition_for_component (rdg, x);
      bitmap_ior_into (partition->stmts, np->stmts);
      partition->has_writes = partition_has_writes (np);
      bitmap_ior_into (processed, np->stmts);
      partition_free (np);

      if (partition_has_writes (partition))
	{
	  if (dump_file && (dump_flags & TDF_DETAILS))
	    {
	      fprintf (dump_file, "ldist useful partition:\n");
	      dump_bitmap (dump_file, partition->stmts);
	    }

	  partitions->safe_push (partition);
	  partition = partition_alloc (NULL);
	}
    }

  /* Add the nodes from the RDG that were not marked as processed, and
     that are used outside the current loop.  These are scalar
     computations that are not yet part of previous partitions.  */
  for (i = 0; i < rdg->n_vertices; i++)
    if (!bitmap_bit_p (processed, i)
	&& rdg_defs_used_in_other_loops_p (rdg, i))
      other_stores->safe_push (i);

  /* If there are still statements left in the OTHER_STORES array,
     create other components and partitions with these stores and
     their dependences.  */
  if (other_stores->length () > 0)
    {
      vec<rdgc> comps;
      comps.create (3);
      vec<int> foo;
      foo.create (3);

      rdg_build_components (rdg, *other_stores, &comps);
      rdg_build_partitions (rdg, comps, &foo, partitions, processed);

      foo.release ();
      free_rdg_components (comps);
    }

  /* If there is something left in the last partition, save it.  */
  if (bitmap_count_bits (partition->stmts) > 0)
    partitions->safe_push (partition);
  else
    partition_free (partition);
}

/* Dump to FILE the PARTITIONS.  */

static void
dump_rdg_partitions (FILE *file, vec<partition_t> partitions)
{
  int i;
  partition_t partition;

  FOR_EACH_VEC_ELT (partitions, i, partition)
    debug_bitmap_file (file, partition->stmts);
}

/* Debug PARTITIONS.  */
extern void debug_rdg_partitions (vec<partition_t> );

DEBUG_FUNCTION void
debug_rdg_partitions (vec<partition_t> partitions)
{
  dump_rdg_partitions (stderr, partitions);
}

/* Returns the number of read and write operations in the RDG.  */

static int
number_of_rw_in_rdg (struct graph *rdg)
{
  int i, res = 0;

  for (i = 0; i < rdg->n_vertices; i++)
    {
      if (RDG_MEM_WRITE_STMT (rdg, i))
	++res;

      if (RDG_MEM_READS_STMT (rdg, i))
	++res;
    }

  return res;
}

/* Returns the number of read and write operations in a PARTITION of
   the RDG.  */

static int
number_of_rw_in_partition (struct graph *rdg, partition_t partition)
{
  int res = 0;
  unsigned i;
  bitmap_iterator ii;

  EXECUTE_IF_SET_IN_BITMAP (partition->stmts, 0, i, ii)
    {
      if (RDG_MEM_WRITE_STMT (rdg, i))
	++res;

      if (RDG_MEM_READS_STMT (rdg, i))
	++res;
    }

  return res;
}

/* Returns true when one of the PARTITIONS contains all the read or
   write operations of RDG.  */

static bool
partition_contains_all_rw (struct graph *rdg,
			   vec<partition_t> partitions)
{
  int i;
  partition_t partition;
  int nrw = number_of_rw_in_rdg (rdg);

  FOR_EACH_VEC_ELT (partitions, i, partition)
    if (nrw == number_of_rw_in_partition (rdg, partition))
      return true;

  return false;
}

/* Generate code from STARTING_VERTICES in RDG.  Returns the number of
   distributed loops.  */

static int
ldist_gen (struct loop *loop, struct graph *rdg,
	   vec<int> starting_vertices)
{
  int i, nbp;
  vec<rdgc> components;
  components.create (3);
  vec<partition_t> partitions;
  partitions.create (3);
  vec<int> other_stores;
  other_stores.create (3);
  partition_t partition;
  bitmap processed = BITMAP_ALLOC (NULL);
  bool any_builtin;

  remaining_stmts = BITMAP_ALLOC (NULL);
  upstream_mem_writes = BITMAP_ALLOC (NULL);

  for (i = 0; i < rdg->n_vertices; i++)
    {
      bitmap_set_bit (remaining_stmts, i);

      /* Save in OTHER_STORES all the memory writes that are not in
	 STARTING_VERTICES.  */
      if (RDG_MEM_WRITE_STMT (rdg, i))
	{
	  int v;
	  unsigned j;
	  bool found = false;

	  FOR_EACH_VEC_ELT (starting_vertices, j, v)
	    if (i == v)
	      {
		found = true;
		break;
	      }

	  if (!found)
	    other_stores.safe_push (i);
	}
    }

  mark_nodes_having_upstream_mem_writes (rdg);
  rdg_build_components (rdg, starting_vertices, &components);
  rdg_build_partitions (rdg, components, &other_stores, &partitions,
			processed);
  BITMAP_FREE (processed);

  any_builtin = false;
  FOR_EACH_VEC_ELT (partitions, i, partition)
    {
      classify_partition (loop, rdg, partition);
      any_builtin |= partition_builtin_p (partition);
    }

  /* If we are only distributing patterns fuse all partitions that
     were not properly classified as builtins.  Else fuse partitions
     with similar memory accesses.  */
  if (!flag_tree_loop_distribution)
    {
      partition_t into;
      /* If we did not detect any builtin simply bail out.  */
      if (!any_builtin)
	{
	  nbp = 0;
	  goto ldist_done;
	}
      /* Only fuse adjacent non-builtin partitions, see PR53616.
         ???  Use dependence information to improve partition ordering.  */
      i = 0;
      do
	{
	  for (; partitions.iterate (i, &into); ++i)
	    if (!partition_builtin_p (into))
	      break;
	  for (++i; partitions.iterate (i, &partition); ++i)
	    if (!partition_builtin_p (partition))
	      {
		bitmap_ior_into (into->stmts, partition->stmts);
		if (partition->kind == PKIND_REDUCTION)
		  into->kind = PKIND_REDUCTION;
		partitions.ordered_remove (i);
		partition_free (partition);
		i--;
	      }
	    else
	      break;
	}
      while ((unsigned) i < partitions.length ());
    }
  else
    {
      partition_t into;
      int j;
      for (i = 0; partitions.iterate (i, &into); ++i)
	{
	  if (partition_builtin_p (into))
	    continue;
	  for (j = i + 1;
	       partitions.iterate (j, &partition); ++j)
	    {
	      if (!partition_builtin_p (partition)
		  /* ???  The following is horribly inefficient,
		     we are re-computing and analyzing data-references
		     of the stmts in the partitions all the time.  */
		  && similar_memory_accesses (rdg, into, partition))
		{
		  if (dump_file && (dump_flags & TDF_DETAILS))
		    {
		      fprintf (dump_file, "fusing partitions\n");
		      dump_bitmap (dump_file, into->stmts);
		      dump_bitmap (dump_file, partition->stmts);
		      fprintf (dump_file, "because they have similar "
			       "memory accesses\n");
		    }
		  bitmap_ior_into (into->stmts, partition->stmts);
		  if (partition->kind == PKIND_REDUCTION)
		    into->kind = PKIND_REDUCTION;
		  partitions.ordered_remove (j);
		  partition_free (partition);
		  j--;
		}
	    }
	}
    }

  /* Fuse all reduction partitions into the last.  */
  if (partitions.length () > 1)
    {
      partition_t into = partitions.last ();
      for (i = partitions.length () - 2; i >= 0; --i)
	{
	  partition_t what = partitions[i];
	  if (what->kind == PKIND_REDUCTION)
	    {
	      if (dump_file && (dump_flags & TDF_DETAILS))
		{
		  fprintf (dump_file, "fusing partitions\n");
		  dump_bitmap (dump_file, into->stmts);
		  dump_bitmap (dump_file, what->stmts);
		  fprintf (dump_file, "because the latter has reductions\n");
		}
	      bitmap_ior_into (into->stmts, what->stmts);
	      into->kind = PKIND_REDUCTION;
	      partitions.ordered_remove (i);
	      partition_free (what);
	    }
	}
    }

  nbp = partitions.length ();
  if (nbp == 0
      || (nbp == 1 && !partition_builtin_p (partitions[0]))
      || (nbp > 1 && partition_contains_all_rw (rdg, partitions)))
    {
      nbp = 0;
      goto ldist_done;
    }

  if (dump_file && (dump_flags & TDF_DETAILS))
    dump_rdg_partitions (dump_file, partitions);

  FOR_EACH_VEC_ELT (partitions, i, partition)
    generate_code_for_partition (loop, partition, i < nbp - 1);

 ldist_done:

  BITMAP_FREE (remaining_stmts);
  BITMAP_FREE (upstream_mem_writes);

  FOR_EACH_VEC_ELT (partitions, i, partition)
    partition_free (partition);

  other_stores.release ();
  partitions.release ();
  free_rdg_components (components);
  return nbp;
}

/* Distributes the code from LOOP in such a way that producer
   statements are placed before consumer statements.  When STMTS is
   NULL, performs the maximal distribution, if STMTS is not NULL,
   tries to separate only these statements from the LOOP's body.
   Returns the number of distributed loops.  */

static int
distribute_loop (struct loop *loop, vec<gimple> stmts)
{
  int res = 0;
  struct graph *rdg;
  gimple s;
  unsigned i;
  vec<int> vertices;
  vec<ddr_p> dependence_relations;
  vec<data_reference_p> datarefs;
  vec<loop_p> loop_nest;

  datarefs.create (10);
  dependence_relations.create (100);
  loop_nest.create (3);
  rdg = build_rdg (loop, &loop_nest, &dependence_relations, &datarefs);

  if (!rdg)
    {
      if (dump_file && (dump_flags & TDF_DETAILS))
	fprintf (dump_file,
		 "FIXME: Loop %d not distributed: failed to build the RDG.\n",
		 loop->num);

      free_dependence_relations (dependence_relations);
      free_data_refs (datarefs);
      loop_nest.release ();
      return res;
    }

  vertices.create (3);

  if (dump_file && (dump_flags & TDF_DETAILS))
    dump_rdg (dump_file, rdg);

  FOR_EACH_VEC_ELT (stmts, i, s)
    {
      int v = rdg_vertex_for_stmt (rdg, s);

      if (v >= 0)
	{
	  vertices.safe_push (v);

	  if (dump_file && (dump_flags & TDF_DETAILS))
	    fprintf (dump_file,
		     "ldist asked to generate code for vertex %d\n", v);
	}
    }

  res = ldist_gen (loop, rdg, vertices);
  vertices.release ();
  free_rdg (rdg);
  free_dependence_relations (dependence_relations);
  free_data_refs (datarefs);
  loop_nest.release ();
  return res;
}

/* Distribute all loops in the current function.  */

static unsigned int
tree_loop_distribution (void)
{
  struct loop *loop;
  loop_iterator li;
  bool changed = false;
  basic_block bb;

  FOR_ALL_BB (bb)
    {
      gimple_stmt_iterator gsi;
      for (gsi = gsi_start_phis (bb); !gsi_end_p (gsi); gsi_next (&gsi))
	gimple_set_uid (gsi_stmt (gsi), -1);
      for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
	gimple_set_uid (gsi_stmt (gsi), -1);
    }

  /* We can at the moment only distribute non-nested loops, thus restrict
     walking to innermost loops.  */
  FOR_EACH_LOOP (li, loop, LI_ONLY_INNERMOST)
    {
      vec<gimple> work_list = vNULL;
      basic_block *bbs;
      int num = loop->num;
      int nb_generated_loops = 0;
      unsigned int i;

      /* If the loop doesn't have a single exit we will fail anyway,
	 so do that early.  */
      if (!single_exit (loop))
	continue;

      /* Only optimize hot loops.  */
      if (!optimize_loop_for_speed_p (loop))
	continue;

      /* Only distribute loops with a header and latch for now.  */
      if (loop->num_nodes > 2)
	continue;

      /* Initialize the worklist with stmts we seed the partitions with.  */
      bbs = get_loop_body_in_dom_order (loop);
      for (i = 0; i < loop->num_nodes; ++i)
	{
	  gimple_stmt_iterator gsi;
	  for (gsi = gsi_start_bb (bbs[i]); !gsi_end_p (gsi); gsi_next (&gsi))
	    {
	      gimple stmt = gsi_stmt (gsi);
	      /* Distribute stmts which have defs that are used outside of
	         the loop.  */
	      if (stmt_has_scalar_dependences_outside_loop (loop, stmt))
		;
	      /* Otherwise only distribute stores for now.  */
	      else if (!gimple_assign_single_p (stmt)
		       || is_gimple_reg (gimple_assign_lhs (stmt)))
		continue;

	      work_list.safe_push (stmt);
	    }
	}
      free (bbs);

      if (work_list.length () > 0)
	nb_generated_loops = distribute_loop (loop, work_list);

      if (nb_generated_loops > 0)
	changed = true;

      if (dump_file && (dump_flags & TDF_DETAILS))
	{
	  if (nb_generated_loops > 1)
	    fprintf (dump_file, "Loop %d distributed: split to %d loops.\n",
		     num, nb_generated_loops);
	  else
	    fprintf (dump_file, "Loop %d is the same.\n", num);
	}

      work_list.release ();
    }

  if (changed)
    {
      mark_virtual_operands_for_renaming (cfun);
      rewrite_into_loop_closed_ssa (NULL, TODO_update_ssa);
    }

#ifdef ENABLE_CHECKING
  verify_loop_structure ();
#endif

  return 0;
}

static bool
gate_tree_loop_distribution (void)
{
  return flag_tree_loop_distribution
    || flag_tree_loop_distribute_patterns;
}

struct gimple_opt_pass pass_loop_distribution =
{
 {
  GIMPLE_PASS,
  "ldist",			/* name */
  OPTGROUP_LOOP,                /* optinfo_flags */
  gate_tree_loop_distribution,  /* gate */
  tree_loop_distribution,       /* execute */
  NULL,				/* sub */
  NULL,				/* next */
  0,				/* static_pass_number */
  TV_TREE_LOOP_DISTRIBUTION,    /* tv_id */
  PROP_cfg | PROP_ssa,		/* properties_required */
  0,				/* properties_provided */
  0,				/* properties_destroyed */
  0,				/* todo_flags_start */
  TODO_verify_ssa               /* todo_flags_finish */
 }
};