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 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670
|
/* Detection of Static Control Parts (SCoP) for Graphite.
Copyright (C) 2009-2018 Free Software Foundation, Inc.
Contributed by Sebastian Pop <sebastian.pop@amd.com> and
Tobias Grosser <grosser@fim.uni-passau.de>.
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/>. */
#define USES_ISL
#include "config.h"
#ifdef HAVE_isl
#include "system.h"
#include "coretypes.h"
#include "backend.h"
#include "cfghooks.h"
#include "domwalk.h"
#include "params.h"
#include "tree.h"
#include "gimple.h"
#include "ssa.h"
#include "fold-const.h"
#include "gimple-iterator.h"
#include "tree-cfg.h"
#include "tree-ssa-loop-manip.h"
#include "tree-ssa-loop-niter.h"
#include "tree-ssa-loop.h"
#include "tree-into-ssa.h"
#include "tree-ssa.h"
#include "cfgloop.h"
#include "tree-data-ref.h"
#include "tree-scalar-evolution.h"
#include "tree-pass.h"
#include "tree-ssa-propagate.h"
#include "gimple-pretty-print.h"
#include "cfganal.h"
#include "graphite.h"
class debug_printer
{
private:
FILE *dump_file;
public:
void
set_dump_file (FILE *f)
{
gcc_assert (f);
dump_file = f;
}
friend debug_printer &
operator<< (debug_printer &output, int i)
{
fprintf (output.dump_file, "%d", i);
return output;
}
friend debug_printer &
operator<< (debug_printer &output, const char *s)
{
fprintf (output.dump_file, "%s", s);
return output;
}
} dp;
#define DEBUG_PRINT(args) do \
{ \
if (dump_file && (dump_flags & TDF_DETAILS)) { args; } \
} while (0)
/* Pretty print to FILE all the SCoPs in DOT format and mark them with
different colors. If there are not enough colors, paint the
remaining SCoPs in gray.
Special nodes:
- "*" after the node number denotes the entry of a SCoP,
- "#" after the node number denotes the exit of a SCoP,
- "()" around the node number denotes the entry or the
exit nodes of the SCOP. These are not part of SCoP. */
DEBUG_FUNCTION void
dot_all_sese (FILE *file, vec<sese_l>& scops)
{
/* Disable debugging while printing graph. */
dump_flags_t tmp_dump_flags = dump_flags;
dump_flags = TDF_NONE;
fprintf (file, "digraph all {\n");
basic_block bb;
FOR_ALL_BB_FN (bb, cfun)
{
int part_of_scop = false;
/* Use HTML for every bb label. So we are able to print bbs
which are part of two different SCoPs, with two different
background colors. */
fprintf (file, "%d [label=<\n <TABLE BORDER=\"0\" CELLBORDER=\"1\" ",
bb->index);
fprintf (file, "CELLSPACING=\"0\">\n");
/* Select color for SCoP. */
sese_l *region;
int i;
FOR_EACH_VEC_ELT (scops, i, region)
{
bool sese_in_region = bb_in_sese_p (bb, *region);
if (sese_in_region || (region->exit->dest == bb)
|| (region->entry->dest == bb))
{
const char *color;
switch (i % 17)
{
case 0: /* red */
color = "#e41a1c";
break;
case 1: /* blue */
color = "#377eb8";
break;
case 2: /* green */
color = "#4daf4a";
break;
case 3: /* purple */
color = "#984ea3";
break;
case 4: /* orange */
color = "#ff7f00";
break;
case 5: /* yellow */
color = "#ffff33";
break;
case 6: /* brown */
color = "#a65628";
break;
case 7: /* rose */
color = "#f781bf";
break;
case 8:
color = "#8dd3c7";
break;
case 9:
color = "#ffffb3";
break;
case 10:
color = "#bebada";
break;
case 11:
color = "#fb8072";
break;
case 12:
color = "#80b1d3";
break;
case 13:
color = "#fdb462";
break;
case 14:
color = "#b3de69";
break;
case 15:
color = "#fccde5";
break;
case 16:
color = "#bc80bd";
break;
default: /* gray */
color = "#999999";
}
fprintf (file, " <TR><TD WIDTH=\"50\" BGCOLOR=\"%s\">",
color);
if (!sese_in_region)
fprintf (file, " (");
if (bb == region->entry->dest && bb == region->exit->dest)
fprintf (file, " %d*# ", bb->index);
else if (bb == region->entry->dest)
fprintf (file, " %d* ", bb->index);
else if (bb == region->exit->dest)
fprintf (file, " %d# ", bb->index);
else
fprintf (file, " %d ", bb->index);
fprintf (file, "{lp_%d}", bb->loop_father->num);
if (!sese_in_region)
fprintf (file, ")");
fprintf (file, "</TD></TR>\n");
part_of_scop = true;
}
}
if (!part_of_scop)
{
fprintf (file, " <TR><TD WIDTH=\"50\" BGCOLOR=\"#ffffff\">");
fprintf (file, " %d {lp_%d} </TD></TR>\n", bb->index,
bb->loop_father->num);
}
fprintf (file, " </TABLE>>, shape=box, style=\"setlinewidth(0)\"]\n");
}
FOR_ALL_BB_FN (bb, cfun)
{
edge e;
edge_iterator ei;
FOR_EACH_EDGE (e, ei, bb->succs)
fprintf (file, "%d -> %d;\n", bb->index, e->dest->index);
}
fputs ("}\n\n", file);
/* Enable debugging again. */
dump_flags = tmp_dump_flags;
}
/* Display SCoP on stderr. */
DEBUG_FUNCTION void
dot_sese (sese_l& scop)
{
vec<sese_l> scops;
scops.create (1);
if (scop)
scops.safe_push (scop);
dot_all_sese (stderr, scops);
scops.release ();
}
DEBUG_FUNCTION void
dot_cfg ()
{
vec<sese_l> scops;
scops.create (1);
dot_all_sese (stderr, scops);
scops.release ();
}
/* Returns a COND_EXPR statement when BB has a single predecessor, the
edge between BB and its predecessor is not a loop exit edge, and
the last statement of the single predecessor is a COND_EXPR. */
static gcond *
single_pred_cond_non_loop_exit (basic_block bb)
{
if (single_pred_p (bb))
{
edge e = single_pred_edge (bb);
basic_block pred = e->src;
gimple *stmt;
if (loop_depth (pred->loop_father) > loop_depth (bb->loop_father))
return NULL;
stmt = last_stmt (pred);
if (stmt && gimple_code (stmt) == GIMPLE_COND)
return as_a<gcond *> (stmt);
}
return NULL;
}
namespace
{
/* Build the maximal scop containing LOOPs and add it to SCOPS. */
class scop_detection
{
public:
scop_detection () : scops (vNULL) {}
~scop_detection ()
{
scops.release ();
}
/* A marker for invalid sese_l. */
static sese_l invalid_sese;
/* Return the SCOPS in this SCOP_DETECTION. */
vec<sese_l>
get_scops ()
{
return scops;
}
/* Return an sese_l around the LOOP. */
sese_l get_sese (loop_p loop);
/* Merge scops at same loop depth and returns the new sese.
Returns a new SESE when merge was successful, INVALID_SESE otherwise. */
sese_l merge_sese (sese_l first, sese_l second) const;
/* Build scop outer->inner if possible. */
void build_scop_depth (loop_p loop);
/* Return true when BEGIN is the preheader edge of a loop with a single exit
END. */
static bool region_has_one_loop (sese_l s);
/* Add to SCOPS a scop starting at SCOP_BEGIN and ending at SCOP_END. */
void add_scop (sese_l s);
/* Returns true if S1 subsumes/surrounds S2. */
static bool subsumes (sese_l s1, sese_l s2);
/* Remove a SCoP which is subsumed by S1. */
void remove_subscops (sese_l s1);
/* Returns true if S1 intersects with S2. Since we already know that S1 does
not subsume S2 or vice-versa, we only check for entry bbs. */
static bool intersects (sese_l s1, sese_l s2);
/* Remove one of the scops when it intersects with any other. */
void remove_intersecting_scops (sese_l s1);
/* Return true when a statement in SCOP cannot be represented by Graphite. */
bool harmful_loop_in_region (sese_l scop) const;
/* Return true only when STMT is simple enough for being handled by Graphite.
This depends on SCOP, as the parameters are initialized relatively to
this basic block, the linear functions are initialized based on the
outermost loop containing STMT inside the SCOP. BB is the place where we
try to evaluate the STMT. */
bool stmt_simple_for_scop_p (sese_l scop, gimple *stmt,
basic_block bb) const;
/* Something like "n * m" is not allowed. */
static bool graphite_can_represent_init (tree e);
/* Return true when SCEV can be represented in the polyhedral model.
An expression can be represented, if it can be expressed as an
affine expression. For loops (i, j) and parameters (m, n) all
affine expressions are of the form:
x1 * i + x2 * j + x3 * m + x4 * n + x5 * 1 where x1..x5 element of Z
1 i + 20 j + (-2) m + 25
Something like "i * n" or "n * m" is not allowed. */
static bool graphite_can_represent_scev (sese_l scop, tree scev);
/* Return true when EXPR can be represented in the polyhedral model.
This means an expression can be represented, if it is linear with respect
to the loops and the strides are non parametric. LOOP is the place where
the expr will be evaluated. SCOP defines the region we analyse. */
static bool graphite_can_represent_expr (sese_l scop, loop_p loop,
tree expr);
/* Return true if the data references of STMT can be represented by Graphite.
We try to analyze the data references in a loop contained in the SCOP. */
static bool stmt_has_simple_data_refs_p (sese_l scop, gimple *stmt);
/* Remove the close phi node at GSI and replace its rhs with the rhs
of PHI. */
static void remove_duplicate_close_phi (gphi *phi, gphi_iterator *gsi);
/* Returns true when Graphite can represent LOOP in SCOP.
FIXME: For the moment, graphite cannot be used on loops that iterate using
induction variables that wrap. */
static bool can_represent_loop (loop_p loop, sese_l scop);
/* Returns the number of pbbs that are in loops contained in SCOP. */
static int nb_pbbs_in_loops (scop_p scop);
private:
vec<sese_l> scops;
};
sese_l scop_detection::invalid_sese (NULL, NULL);
/* Return an sese_l around the LOOP. */
sese_l
scop_detection::get_sese (loop_p loop)
{
if (!loop)
return invalid_sese;
edge scop_begin = loop_preheader_edge (loop);
edge scop_end = single_exit (loop);
if (!scop_end || (scop_end->flags & (EDGE_COMPLEX|EDGE_FAKE)))
return invalid_sese;
return sese_l (scop_begin, scop_end);
}
/* Merge scops at same loop depth and returns the new sese.
Returns a new SESE when merge was successful, INVALID_SESE otherwise. */
sese_l
scop_detection::merge_sese (sese_l first, sese_l second) const
{
/* In the trivial case first/second may be NULL. */
if (!first)
return second;
if (!second)
return first;
DEBUG_PRINT (dp << "[scop-detection] try merging sese s1: ";
print_sese (dump_file, first);
dp << "[scop-detection] try merging sese s2: ";
print_sese (dump_file, second));
auto_bitmap worklist, in_sese_region;
bitmap_set_bit (worklist, get_entry_bb (first)->index);
bitmap_set_bit (worklist, get_exit_bb (first)->index);
bitmap_set_bit (worklist, get_entry_bb (second)->index);
bitmap_set_bit (worklist, get_exit_bb (second)->index);
edge entry = NULL, exit = NULL;
/* We can optimize the case of adding a loop entry dest or exit
src to the worklist (for single-exit loops) by skipping
directly to the exit dest / entry src. in_sese_region
doesn't have to cover all blocks in the region but merely
its border it acts more like a visited bitmap. */
do
{
int index = bitmap_first_set_bit (worklist);
bitmap_clear_bit (worklist, index);
basic_block bb = BASIC_BLOCK_FOR_FN (cfun, index);
edge_iterator ei;
edge e;
/* With fake exit edges we can end up with no possible exit. */
if (index == EXIT_BLOCK)
{
DEBUG_PRINT (dp << "[scop-detection-fail] cannot merge seses.\n");
return invalid_sese;
}
bitmap_set_bit (in_sese_region, bb->index);
basic_block dom = get_immediate_dominator (CDI_DOMINATORS, bb);
FOR_EACH_EDGE (e, ei, bb->preds)
if (e->src == dom
&& (! entry
|| dominated_by_p (CDI_DOMINATORS, entry->dest, bb)))
{
if (entry
&& ! bitmap_bit_p (in_sese_region, entry->src->index))
bitmap_set_bit (worklist, entry->src->index);
entry = e;
}
else if (! bitmap_bit_p (in_sese_region, e->src->index))
bitmap_set_bit (worklist, e->src->index);
basic_block pdom = get_immediate_dominator (CDI_POST_DOMINATORS, bb);
FOR_EACH_EDGE (e, ei, bb->succs)
if (e->dest == pdom
&& (! exit
|| dominated_by_p (CDI_POST_DOMINATORS, exit->src, bb)))
{
if (exit
&& ! bitmap_bit_p (in_sese_region, exit->dest->index))
bitmap_set_bit (worklist, exit->dest->index);
exit = e;
}
else if (! bitmap_bit_p (in_sese_region, e->dest->index))
bitmap_set_bit (worklist, e->dest->index);
}
while (! bitmap_empty_p (worklist));
sese_l combined (entry, exit);
DEBUG_PRINT (dp << "[merged-sese] s1: "; print_sese (dump_file, combined));
return combined;
}
/* Build scop outer->inner if possible. */
void
scop_detection::build_scop_depth (loop_p loop)
{
sese_l s = invalid_sese;
loop = loop->inner;
while (loop)
{
sese_l next = get_sese (loop);
if (! next
|| harmful_loop_in_region (next))
{
if (s)
add_scop (s);
build_scop_depth (loop);
s = invalid_sese;
}
else if (! s)
s = next;
else
{
sese_l combined = merge_sese (s, next);
if (! combined
|| harmful_loop_in_region (combined))
{
add_scop (s);
s = next;
}
else
s = combined;
}
loop = loop->next;
}
if (s)
add_scop (s);
}
/* Returns true when Graphite can represent LOOP in SCOP.
FIXME: For the moment, graphite cannot be used on loops that iterate using
induction variables that wrap. */
bool
scop_detection::can_represent_loop (loop_p loop, sese_l scop)
{
tree niter;
struct tree_niter_desc niter_desc;
/* We can only handle do {} while () style loops correctly. */
edge exit = single_exit (loop);
if (!exit
|| !single_pred_p (loop->latch)
|| exit->src != single_pred (loop->latch)
|| !empty_block_p (loop->latch))
return false;
return !(loop_preheader_edge (loop)->flags & EDGE_IRREDUCIBLE_LOOP)
&& number_of_iterations_exit (loop, single_exit (loop), &niter_desc, false)
&& niter_desc.control.no_overflow
&& (niter = number_of_latch_executions (loop))
&& !chrec_contains_undetermined (niter)
&& !chrec_contains_undetermined (scalar_evolution_in_region (scop,
loop, niter))
&& graphite_can_represent_expr (scop, loop, niter);
}
/* Return true when BEGIN is the preheader edge of a loop with a single exit
END. */
bool
scop_detection::region_has_one_loop (sese_l s)
{
edge begin = s.entry;
edge end = s.exit;
/* Check for a single perfectly nested loop. */
if (begin->dest->loop_father->inner)
return false;
/* Otherwise, check whether we have adjacent loops. */
return (single_pred_p (end->src)
&& begin->dest->loop_father == single_pred (end->src)->loop_father);
}
/* Add to SCOPS a scop starting at SCOP_BEGIN and ending at SCOP_END. */
void
scop_detection::add_scop (sese_l s)
{
gcc_assert (s);
/* If the exit edge is fake discard the SCoP for now as we're removing the
fake edges again after analysis. */
if (s.exit->flags & EDGE_FAKE)
{
DEBUG_PRINT (dp << "[scop-detection-fail] Discarding infinite loop SCoP: ";
print_sese (dump_file, s));
return;
}
/* Include the BB with the loop-closed SSA PHI nodes, we need this
block in the region for code-generating out-of-SSA copies.
canonicalize_loop_closed_ssa makes sure that is in proper shape. */
if (s.exit->dest != EXIT_BLOCK_PTR_FOR_FN (cfun)
&& loop_exit_edge_p (s.exit->src->loop_father, s.exit))
{
gcc_assert (single_pred_p (s.exit->dest)
&& single_succ_p (s.exit->dest)
&& sese_trivially_empty_bb_p (s.exit->dest));
s.exit = single_succ_edge (s.exit->dest);
}
/* Do not add scops with only one loop. */
if (region_has_one_loop (s))
{
DEBUG_PRINT (dp << "[scop-detection-fail] Discarding one loop SCoP: ";
print_sese (dump_file, s));
return;
}
if (get_exit_bb (s) == EXIT_BLOCK_PTR_FOR_FN (cfun))
{
DEBUG_PRINT (dp << "[scop-detection-fail] "
<< "Discarding SCoP exiting to return: ";
print_sese (dump_file, s));
return;
}
/* Remove all the scops which are subsumed by s. */
remove_subscops (s);
/* Remove intersecting scops. FIXME: It will be a good idea to keep
the non-intersecting part of the scop already in the list. */
remove_intersecting_scops (s);
scops.safe_push (s);
DEBUG_PRINT (dp << "[scop-detection] Adding SCoP: "; print_sese (dump_file, s));
}
/* Return true when a statement in SCOP cannot be represented by Graphite. */
bool
scop_detection::harmful_loop_in_region (sese_l scop) const
{
basic_block exit_bb = get_exit_bb (scop);
basic_block entry_bb = get_entry_bb (scop);
DEBUG_PRINT (dp << "[checking-harmful-bbs] ";
print_sese (dump_file, scop));
gcc_assert (dominated_by_p (CDI_DOMINATORS, exit_bb, entry_bb));
auto_vec<basic_block> worklist;
auto_bitmap loops;
worklist.safe_push (entry_bb);
while (! worklist.is_empty ())
{
basic_block bb = worklist.pop ();
DEBUG_PRINT (dp << "Visiting bb_" << bb->index << "\n");
/* The basic block should not be part of an irreducible loop. */
if (bb->flags & BB_IRREDUCIBLE_LOOP)
return true;
/* Check for unstructured control flow: CFG not generated by structured
if-then-else. */
if (bb->succs->length () > 1)
{
edge e;
edge_iterator ei;
FOR_EACH_EDGE (e, ei, bb->succs)
if (!dominated_by_p (CDI_POST_DOMINATORS, bb, e->dest)
&& !dominated_by_p (CDI_DOMINATORS, e->dest, bb))
return true;
}
/* Collect all loops in the current region. */
loop_p loop = bb->loop_father;
if (loop_in_sese_p (loop, scop))
bitmap_set_bit (loops, loop->num);
/* Check for harmful statements in basic blocks part of the region. */
for (gimple_stmt_iterator gsi = gsi_start_bb (bb);
!gsi_end_p (gsi); gsi_next (&gsi))
if (!stmt_simple_for_scop_p (scop, gsi_stmt (gsi), bb))
return true;
for (basic_block dom = first_dom_son (CDI_DOMINATORS, bb);
dom;
dom = next_dom_son (CDI_DOMINATORS, dom))
if (dom != scop.exit->dest)
worklist.safe_push (dom);
}
/* Go through all loops and check that they are still valid in the combined
scop. */
unsigned j;
bitmap_iterator bi;
EXECUTE_IF_SET_IN_BITMAP (loops, 0, j, bi)
{
loop_p loop = (*current_loops->larray)[j];
gcc_assert (loop->num == (int) j);
/* Check if the loop nests are to be optimized for speed. */
if (! loop->inner
&& ! optimize_loop_for_speed_p (loop))
{
DEBUG_PRINT (dp << "[scop-detection-fail] loop_"
<< loop->num << " is not on a hot path.\n");
return true;
}
if (! can_represent_loop (loop, scop))
{
DEBUG_PRINT (dp << "[scop-detection-fail] cannot represent loop_"
<< loop->num << "\n");
return true;
}
/* Check if all loop nests have at least one data reference.
??? This check is expensive and loops premature at this point.
If important to retain we can pre-compute this for all innermost
loops and reject those when we build a SESE region for a loop
during SESE discovery. */
if (! loop->inner
&& ! loop_nest_has_data_refs (loop))
{
DEBUG_PRINT (dp << "[scop-detection-fail] loop_" << loop->num
<< "does not have any data reference.\n");
return true;
}
}
return false;
}
/* Returns true if S1 subsumes/surrounds S2. */
bool
scop_detection::subsumes (sese_l s1, sese_l s2)
{
if (dominated_by_p (CDI_DOMINATORS, get_entry_bb (s2),
get_entry_bb (s1))
&& dominated_by_p (CDI_POST_DOMINATORS, s2.exit->dest,
s1.exit->dest))
return true;
return false;
}
/* Remove a SCoP which is subsumed by S1. */
void
scop_detection::remove_subscops (sese_l s1)
{
int j;
sese_l *s2;
FOR_EACH_VEC_ELT_REVERSE (scops, j, s2)
{
if (subsumes (s1, *s2))
{
DEBUG_PRINT (dp << "Removing sub-SCoP";
print_sese (dump_file, *s2));
scops.unordered_remove (j);
}
}
}
/* Returns true if S1 intersects with S2. Since we already know that S1 does
not subsume S2 or vice-versa, we only check for entry bbs. */
bool
scop_detection::intersects (sese_l s1, sese_l s2)
{
if (dominated_by_p (CDI_DOMINATORS, get_entry_bb (s2),
get_entry_bb (s1))
&& !dominated_by_p (CDI_DOMINATORS, get_entry_bb (s2),
get_exit_bb (s1)))
return true;
if ((s1.exit == s2.entry) || (s2.exit == s1.entry))
return true;
return false;
}
/* Remove one of the scops when it intersects with any other. */
void
scop_detection::remove_intersecting_scops (sese_l s1)
{
int j;
sese_l *s2;
FOR_EACH_VEC_ELT_REVERSE (scops, j, s2)
{
if (intersects (s1, *s2))
{
DEBUG_PRINT (dp << "Removing intersecting SCoP";
print_sese (dump_file, *s2);
dp << "Intersects with:";
print_sese (dump_file, s1));
scops.unordered_remove (j);
}
}
}
/* Something like "n * m" is not allowed. */
bool
scop_detection::graphite_can_represent_init (tree e)
{
switch (TREE_CODE (e))
{
case POLYNOMIAL_CHREC:
return graphite_can_represent_init (CHREC_LEFT (e))
&& graphite_can_represent_init (CHREC_RIGHT (e));
case MULT_EXPR:
if (chrec_contains_symbols (TREE_OPERAND (e, 0)))
return graphite_can_represent_init (TREE_OPERAND (e, 0))
&& tree_fits_shwi_p (TREE_OPERAND (e, 1));
else
return graphite_can_represent_init (TREE_OPERAND (e, 1))
&& tree_fits_shwi_p (TREE_OPERAND (e, 0));
case PLUS_EXPR:
case POINTER_PLUS_EXPR:
case MINUS_EXPR:
return graphite_can_represent_init (TREE_OPERAND (e, 0))
&& graphite_can_represent_init (TREE_OPERAND (e, 1));
case NEGATE_EXPR:
case BIT_NOT_EXPR:
CASE_CONVERT:
case NON_LVALUE_EXPR:
return graphite_can_represent_init (TREE_OPERAND (e, 0));
default:
break;
}
return true;
}
/* Return true when SCEV can be represented in the polyhedral model.
An expression can be represented, if it can be expressed as an
affine expression. For loops (i, j) and parameters (m, n) all
affine expressions are of the form:
x1 * i + x2 * j + x3 * m + x4 * n + x5 * 1 where x1..x5 element of Z
1 i + 20 j + (-2) m + 25
Something like "i * n" or "n * m" is not allowed. */
bool
scop_detection::graphite_can_represent_scev (sese_l scop, tree scev)
{
if (chrec_contains_undetermined (scev))
return false;
switch (TREE_CODE (scev))
{
case NEGATE_EXPR:
case BIT_NOT_EXPR:
CASE_CONVERT:
case NON_LVALUE_EXPR:
return graphite_can_represent_scev (scop, TREE_OPERAND (scev, 0));
case PLUS_EXPR:
case POINTER_PLUS_EXPR:
case MINUS_EXPR:
return graphite_can_represent_scev (scop, TREE_OPERAND (scev, 0))
&& graphite_can_represent_scev (scop, TREE_OPERAND (scev, 1));
case MULT_EXPR:
return !CONVERT_EXPR_CODE_P (TREE_CODE (TREE_OPERAND (scev, 0)))
&& !CONVERT_EXPR_CODE_P (TREE_CODE (TREE_OPERAND (scev, 1)))
&& !(chrec_contains_symbols (TREE_OPERAND (scev, 0))
&& chrec_contains_symbols (TREE_OPERAND (scev, 1)))
&& graphite_can_represent_init (scev)
&& graphite_can_represent_scev (scop, TREE_OPERAND (scev, 0))
&& graphite_can_represent_scev (scop, TREE_OPERAND (scev, 1));
case POLYNOMIAL_CHREC:
/* Check for constant strides. With a non constant stride of
'n' we would have a value of 'iv * n'. Also check that the
initial value can represented: for example 'n * m' cannot be
represented. */
gcc_assert (loop_in_sese_p (get_loop (cfun,
CHREC_VARIABLE (scev)), scop));
if (!evolution_function_right_is_integer_cst (scev)
|| !graphite_can_represent_init (scev))
return false;
return graphite_can_represent_scev (scop, CHREC_LEFT (scev));
default:
break;
}
/* Only affine functions can be represented. */
if (tree_contains_chrecs (scev, NULL) || !scev_is_linear_expression (scev))
return false;
return true;
}
/* Return true when EXPR can be represented in the polyhedral model.
This means an expression can be represented, if it is linear with respect to
the loops and the strides are non parametric. LOOP is the place where the
expr will be evaluated. SCOP defines the region we analyse. */
bool
scop_detection::graphite_can_represent_expr (sese_l scop, loop_p loop,
tree expr)
{
tree scev = scalar_evolution_in_region (scop, loop, expr);
return graphite_can_represent_scev (scop, scev);
}
/* Return true if the data references of STMT can be represented by Graphite.
We try to analyze the data references in a loop contained in the SCOP. */
bool
scop_detection::stmt_has_simple_data_refs_p (sese_l scop, gimple *stmt)
{
edge nest = scop.entry;
loop_p loop = loop_containing_stmt (stmt);
if (!loop_in_sese_p (loop, scop))
loop = NULL;
auto_vec<data_reference_p> drs;
if (! graphite_find_data_references_in_stmt (nest, loop, stmt, &drs))
return false;
int j;
data_reference_p dr;
FOR_EACH_VEC_ELT (drs, j, dr)
{
for (unsigned i = 0; i < DR_NUM_DIMENSIONS (dr); ++i)
if (! graphite_can_represent_scev (scop, DR_ACCESS_FN (dr, i)))
return false;
}
return true;
}
/* GIMPLE_ASM and GIMPLE_CALL may embed arbitrary side effects.
Calls have side-effects, except those to const or pure
functions. */
static bool
stmt_has_side_effects (gimple *stmt)
{
if (gimple_has_volatile_ops (stmt)
|| (gimple_code (stmt) == GIMPLE_CALL
&& !(gimple_call_flags (stmt) & (ECF_CONST | ECF_PURE)))
|| (gimple_code (stmt) == GIMPLE_ASM))
{
DEBUG_PRINT (dp << "[scop-detection-fail] "
<< "Statement has side-effects:\n";
print_gimple_stmt (dump_file, stmt, 0, TDF_VOPS | TDF_MEMSYMS));
return true;
}
return false;
}
/* Return true only when STMT is simple enough for being handled by Graphite.
This depends on SCOP, as the parameters are initialized relatively to
this basic block, the linear functions are initialized based on the outermost
loop containing STMT inside the SCOP. BB is the place where we try to
evaluate the STMT. */
bool
scop_detection::stmt_simple_for_scop_p (sese_l scop, gimple *stmt,
basic_block bb) const
{
gcc_assert (scop);
if (is_gimple_debug (stmt))
return true;
if (stmt_has_side_effects (stmt))
return false;
if (!stmt_has_simple_data_refs_p (scop, stmt))
{
DEBUG_PRINT (dp << "[scop-detection-fail] "
<< "Graphite cannot handle data-refs in stmt:\n";
print_gimple_stmt (dump_file, stmt, 0, TDF_VOPS|TDF_MEMSYMS););
return false;
}
switch (gimple_code (stmt))
{
case GIMPLE_LABEL:
return true;
case GIMPLE_COND:
{
/* We can handle all binary comparisons. Inequalities are
also supported as they can be represented with union of
polyhedra. */
enum tree_code code = gimple_cond_code (stmt);
if (!(code == LT_EXPR
|| code == GT_EXPR
|| code == LE_EXPR
|| code == GE_EXPR
|| code == EQ_EXPR
|| code == NE_EXPR))
{
DEBUG_PRINT (dp << "[scop-detection-fail] "
<< "Graphite cannot handle cond stmt:\n";
print_gimple_stmt (dump_file, stmt, 0,
TDF_VOPS | TDF_MEMSYMS));
return false;
}
loop_p loop = bb->loop_father;
for (unsigned i = 0; i < 2; ++i)
{
tree op = gimple_op (stmt, i);
if (!graphite_can_represent_expr (scop, loop, op)
/* We can only constrain on integer type. */
|| ! INTEGRAL_TYPE_P (TREE_TYPE (op)))
{
DEBUG_PRINT (dp << "[scop-detection-fail] "
<< "Graphite cannot represent stmt:\n";
print_gimple_stmt (dump_file, stmt, 0,
TDF_VOPS | TDF_MEMSYMS));
return false;
}
}
return true;
}
case GIMPLE_ASSIGN:
case GIMPLE_CALL:
{
tree op, lhs = gimple_get_lhs (stmt);
ssa_op_iter i;
/* If we are not going to instantiate the stmt do not require
its operands to be instantiatable at this point. */
if (lhs
&& TREE_CODE (lhs) == SSA_NAME
&& scev_analyzable_p (lhs, scop))
return true;
/* Verify that if we can analyze operands at their def site we
also can represent them when analyzed at their uses. */
FOR_EACH_SSA_TREE_OPERAND (op, stmt, i, SSA_OP_USE)
if (scev_analyzable_p (op, scop)
&& chrec_contains_undetermined
(scalar_evolution_in_region (scop, bb->loop_father, op)))
{
DEBUG_PRINT (dp << "[scop-detection-fail] "
<< "Graphite cannot code-gen stmt:\n";
print_gimple_stmt (dump_file, stmt, 0,
TDF_VOPS | TDF_MEMSYMS));
return false;
}
return true;
}
default:
/* These nodes cut a new scope. */
DEBUG_PRINT (
dp << "[scop-detection-fail] "
<< "Gimple stmt not handled in Graphite:\n";
print_gimple_stmt (dump_file, stmt, 0, TDF_VOPS | TDF_MEMSYMS));
return false;
}
}
/* Returns the number of pbbs that are in loops contained in SCOP. */
int
scop_detection::nb_pbbs_in_loops (scop_p scop)
{
int i;
poly_bb_p pbb;
int res = 0;
FOR_EACH_VEC_ELT (scop->pbbs, i, pbb)
if (loop_in_sese_p (gbb_loop (PBB_BLACK_BOX (pbb)), scop->scop_info->region))
res++;
return res;
}
/* Assigns the parameter NAME an index in REGION. */
static void
assign_parameter_index_in_region (tree name, sese_info_p region)
{
gcc_assert (TREE_CODE (name) == SSA_NAME
&& INTEGRAL_TYPE_P (TREE_TYPE (name))
&& ! defined_in_sese_p (name, region->region));
int i;
tree p;
FOR_EACH_VEC_ELT (region->params, i, p)
if (p == name)
return;
i = region->params.length ();
region->params.safe_push (name);
}
/* In the context of sese S, scan the expression E and translate it to
a linear expression C. When parsing a symbolic multiplication, K
represents the constant multiplier of an expression containing
parameters. */
static void
scan_tree_for_params (sese_info_p s, tree e)
{
if (e == chrec_dont_know)
return;
switch (TREE_CODE (e))
{
case POLYNOMIAL_CHREC:
scan_tree_for_params (s, CHREC_LEFT (e));
break;
case MULT_EXPR:
if (chrec_contains_symbols (TREE_OPERAND (e, 0)))
scan_tree_for_params (s, TREE_OPERAND (e, 0));
else
scan_tree_for_params (s, TREE_OPERAND (e, 1));
break;
case PLUS_EXPR:
case POINTER_PLUS_EXPR:
case MINUS_EXPR:
scan_tree_for_params (s, TREE_OPERAND (e, 0));
scan_tree_for_params (s, TREE_OPERAND (e, 1));
break;
case NEGATE_EXPR:
case BIT_NOT_EXPR:
CASE_CONVERT:
case NON_LVALUE_EXPR:
scan_tree_for_params (s, TREE_OPERAND (e, 0));
break;
case SSA_NAME:
assign_parameter_index_in_region (e, s);
break;
case INTEGER_CST:
case ADDR_EXPR:
case REAL_CST:
case COMPLEX_CST:
case VECTOR_CST:
break;
default:
gcc_unreachable ();
break;
}
}
/* Find parameters with respect to REGION in BB. We are looking in memory
access functions, conditions and loop bounds. */
static void
find_params_in_bb (sese_info_p region, gimple_poly_bb_p gbb)
{
/* Find parameters in the access functions of data references. */
int i;
data_reference_p dr;
FOR_EACH_VEC_ELT (GBB_DATA_REFS (gbb), i, dr)
for (unsigned j = 0; j < DR_NUM_DIMENSIONS (dr); j++)
scan_tree_for_params (region, DR_ACCESS_FN (dr, j));
/* Find parameters in conditional statements. */
gimple *stmt;
FOR_EACH_VEC_ELT (GBB_CONDITIONS (gbb), i, stmt)
{
loop_p loop = gimple_bb (stmt)->loop_father;
tree lhs = scalar_evolution_in_region (region->region, loop,
gimple_cond_lhs (stmt));
tree rhs = scalar_evolution_in_region (region->region, loop,
gimple_cond_rhs (stmt));
gcc_assert (!chrec_contains_undetermined (lhs)
&& !chrec_contains_undetermined (rhs));
scan_tree_for_params (region, lhs);
scan_tree_for_params (region, rhs);
}
}
/* Record the parameters used in the SCOP BBs. A variable is a parameter
in a scop if it does not vary during the execution of that scop. */
static void
find_scop_parameters (scop_p scop)
{
unsigned i;
sese_info_p region = scop->scop_info;
/* Parameters used in loop bounds are processed during gather_bbs. */
/* Find the parameters used in data accesses. */
poly_bb_p pbb;
FOR_EACH_VEC_ELT (scop->pbbs, i, pbb)
find_params_in_bb (region, PBB_BLACK_BOX (pbb));
int nbp = sese_nb_params (region);
scop_set_nb_params (scop, nbp);
}
static void
add_write (vec<tree> *writes, tree def)
{
writes->safe_push (def);
DEBUG_PRINT (dp << "Adding scalar write: ";
print_generic_expr (dump_file, def);
dp << "\nFrom stmt: ";
print_gimple_stmt (dump_file,
SSA_NAME_DEF_STMT (def), 0));
}
static void
add_read (vec<scalar_use> *reads, tree use, gimple *use_stmt)
{
DEBUG_PRINT (dp << "Adding scalar read: ";
print_generic_expr (dump_file, use);
dp << "\nFrom stmt: ";
print_gimple_stmt (dump_file, use_stmt, 0));
reads->safe_push (std::make_pair (use_stmt, use));
}
/* Record DEF if it is used in other bbs different than DEF_BB in the SCOP. */
static void
build_cross_bb_scalars_def (scop_p scop, tree def, basic_block def_bb,
vec<tree> *writes)
{
if (!is_gimple_reg (def))
return;
bool scev_analyzable = scev_analyzable_p (def, scop->scop_info->region);
gimple *use_stmt;
imm_use_iterator imm_iter;
FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, def)
/* Do not gather scalar variables that can be analyzed by SCEV as they can
be generated out of the induction variables. */
if ((! scev_analyzable
/* But gather SESE liveouts as we otherwise fail to rewrite their
exit PHIs. */
|| ! bb_in_sese_p (gimple_bb (use_stmt), scop->scop_info->region))
&& (def_bb != gimple_bb (use_stmt) && !is_gimple_debug (use_stmt)))
{
add_write (writes, def);
/* This is required by the FOR_EACH_IMM_USE_STMT when we want to break
before all the uses have been visited. */
BREAK_FROM_IMM_USE_STMT (imm_iter);
}
}
/* Record USE if it is defined in other bbs different than USE_STMT
in the SCOP. */
static void
build_cross_bb_scalars_use (scop_p scop, tree use, gimple *use_stmt,
vec<scalar_use> *reads)
{
if (!is_gimple_reg (use))
return;
/* Do not gather scalar variables that can be analyzed by SCEV as they can be
generated out of the induction variables. */
if (scev_analyzable_p (use, scop->scop_info->region))
return;
gimple *def_stmt = SSA_NAME_DEF_STMT (use);
if (gimple_bb (def_stmt) != gimple_bb (use_stmt))
add_read (reads, use, use_stmt);
}
/* Generates a polyhedral black box only if the bb contains interesting
information. */
static gimple_poly_bb_p
try_generate_gimple_bb (scop_p scop, basic_block bb)
{
vec<data_reference_p> drs = vNULL;
vec<tree> writes = vNULL;
vec<scalar_use> reads = vNULL;
sese_l region = scop->scop_info->region;
edge nest = region.entry;
loop_p loop = bb->loop_father;
if (!loop_in_sese_p (loop, region))
loop = NULL;
for (gimple_stmt_iterator gsi = gsi_start_bb (bb); !gsi_end_p (gsi);
gsi_next (&gsi))
{
gimple *stmt = gsi_stmt (gsi);
if (is_gimple_debug (stmt))
continue;
graphite_find_data_references_in_stmt (nest, loop, stmt, &drs);
tree def = gimple_get_lhs (stmt);
if (def)
build_cross_bb_scalars_def (scop, def, gimple_bb (stmt), &writes);
ssa_op_iter iter;
tree use;
FOR_EACH_SSA_TREE_OPERAND (use, stmt, iter, SSA_OP_USE)
build_cross_bb_scalars_use (scop, use, stmt, &reads);
}
/* Handle defs and uses in PHIs. Those need special treatment given
that we have to present ISL with sth that looks like we've rewritten
the IL out-of-SSA. */
for (gphi_iterator psi = gsi_start_phis (bb); !gsi_end_p (psi);
gsi_next (&psi))
{
gphi *phi = psi.phi ();
tree res = gimple_phi_result (phi);
if (virtual_operand_p (res)
|| scev_analyzable_p (res, scop->scop_info->region))
continue;
/* To simulate out-of-SSA the block containing the PHI node has
reads of the PHI destination. And to preserve SSA dependences
we also write to it (the out-of-SSA decl and the SSA result
are coalesced for dependence purposes which is good enough). */
add_read (&reads, res, phi);
add_write (&writes, res);
}
basic_block bb_for_succs = bb;
if (bb_for_succs == bb_for_succs->loop_father->latch
&& bb_in_sese_p (bb_for_succs, scop->scop_info->region)
&& sese_trivially_empty_bb_p (bb_for_succs))
bb_for_succs = NULL;
while (bb_for_succs)
{
basic_block latch = NULL;
edge_iterator ei;
edge e;
FOR_EACH_EDGE (e, ei, bb_for_succs->succs)
{
for (gphi_iterator psi = gsi_start_phis (e->dest); !gsi_end_p (psi);
gsi_next (&psi))
{
gphi *phi = psi.phi ();
tree res = gimple_phi_result (phi);
if (virtual_operand_p (res))
continue;
/* To simulate out-of-SSA the predecessor of edges into PHI nodes
has a copy from the PHI argument to the PHI destination. */
if (! scev_analyzable_p (res, scop->scop_info->region))
add_write (&writes, res);
tree use = PHI_ARG_DEF_FROM_EDGE (phi, e);
if (TREE_CODE (use) == SSA_NAME
&& ! SSA_NAME_IS_DEFAULT_DEF (use)
&& gimple_bb (SSA_NAME_DEF_STMT (use)) != bb_for_succs
&& ! scev_analyzable_p (use, scop->scop_info->region))
add_read (&reads, use, phi);
}
if (e->dest == bb_for_succs->loop_father->latch
&& bb_in_sese_p (e->dest, scop->scop_info->region)
&& sese_trivially_empty_bb_p (e->dest))
latch = e->dest;
}
/* Handle empty latch block PHIs here, otherwise we confuse ISL
with extra conditional code where it then peels off the last
iteration just because of that. It would be simplest if we
just didn't force simple latches (thus remove the forwarder). */
bb_for_succs = latch;
}
/* For the region exit block add reads for all live-out vars. */
if (bb == scop->scop_info->region.exit->src)
{
sese_build_liveouts (scop->scop_info);
unsigned i;
bitmap_iterator bi;
EXECUTE_IF_SET_IN_BITMAP (scop->scop_info->liveout, 0, i, bi)
{
tree use = ssa_name (i);
add_read (&reads, use, NULL);
}
}
if (drs.is_empty () && writes.is_empty () && reads.is_empty ())
return NULL;
return new_gimple_poly_bb (bb, drs, reads, writes);
}
/* Compute alias-sets for all data references in DRS. */
static bool
build_alias_set (scop_p scop)
{
int num_vertices = scop->drs.length ();
struct graph *g = new_graph (num_vertices);
dr_info *dr1, *dr2;
int i, j;
int *all_vertices;
FOR_EACH_VEC_ELT (scop->drs, i, dr1)
for (j = i+1; scop->drs.iterate (j, &dr2); j++)
if (dr_may_alias_p (dr1->dr, dr2->dr, true))
{
/* Dependences in the same alias set need to be handled
by just looking at DR_ACCESS_FNs. */
if (DR_NUM_DIMENSIONS (dr1->dr) == 0
|| DR_NUM_DIMENSIONS (dr1->dr) != DR_NUM_DIMENSIONS (dr2->dr)
|| ! operand_equal_p (DR_BASE_OBJECT (dr1->dr),
DR_BASE_OBJECT (dr2->dr),
OEP_ADDRESS_OF)
|| ! types_compatible_p (TREE_TYPE (DR_BASE_OBJECT (dr1->dr)),
TREE_TYPE (DR_BASE_OBJECT (dr2->dr))))
{
free_graph (g);
return false;
}
add_edge (g, i, j);
add_edge (g, j, i);
}
all_vertices = XNEWVEC (int, num_vertices);
for (i = 0; i < num_vertices; i++)
all_vertices[i] = i;
scop->max_alias_set
= graphds_dfs (g, all_vertices, num_vertices, NULL, true, NULL) + 1;
free (all_vertices);
for (i = 0; i < g->n_vertices; i++)
scop->drs[i].alias_set = g->vertices[i].component + 1;
free_graph (g);
return true;
}
/* Gather BBs and conditions for a SCOP. */
class gather_bbs : public dom_walker
{
public:
gather_bbs (cdi_direction, scop_p, int *);
virtual edge before_dom_children (basic_block);
virtual void after_dom_children (basic_block);
private:
auto_vec<gimple *, 3> conditions, cases;
scop_p scop;
};
}
gather_bbs::gather_bbs (cdi_direction direction, scop_p scop, int *bb_to_rpo)
: dom_walker (direction, ALL_BLOCKS, bb_to_rpo), scop (scop)
{
}
/* Call-back for dom_walk executed before visiting the dominated
blocks. */
edge
gather_bbs::before_dom_children (basic_block bb)
{
sese_info_p region = scop->scop_info;
if (!bb_in_sese_p (bb, region->region))
return dom_walker::STOP;
/* For loops fully contained in the region record parameters in the
loop bounds. */
loop_p loop = bb->loop_father;
if (loop->header == bb
&& loop_in_sese_p (loop, region->region))
{
tree nb_iters = number_of_latch_executions (loop);
if (chrec_contains_symbols (nb_iters))
{
nb_iters = scalar_evolution_in_region (region->region,
loop, nb_iters);
scan_tree_for_params (region, nb_iters);
}
}
if (gcond *stmt = single_pred_cond_non_loop_exit (bb))
{
edge e = single_pred_edge (bb);
/* Make sure the condition is in the region and thus was verified
to be handled. */
if (e != region->region.entry)
{
conditions.safe_push (stmt);
if (e->flags & EDGE_TRUE_VALUE)
cases.safe_push (stmt);
else
cases.safe_push (NULL);
}
}
scop->scop_info->bbs.safe_push (bb);
gimple_poly_bb_p gbb = try_generate_gimple_bb (scop, bb);
if (!gbb)
return NULL;
GBB_CONDITIONS (gbb) = conditions.copy ();
GBB_CONDITION_CASES (gbb) = cases.copy ();
poly_bb_p pbb = new_poly_bb (scop, gbb);
scop->pbbs.safe_push (pbb);
int i;
data_reference_p dr;
FOR_EACH_VEC_ELT (gbb->data_refs, i, dr)
{
DEBUG_PRINT (dp << "Adding memory ";
if (dr->is_read)
dp << "read: ";
else
dp << "write: ";
print_generic_expr (dump_file, dr->ref);
dp << "\nFrom stmt: ";
print_gimple_stmt (dump_file, dr->stmt, 0));
scop->drs.safe_push (dr_info (dr, pbb));
}
return NULL;
}
/* Call-back for dom_walk executed after visiting the dominated
blocks. */
void
gather_bbs::after_dom_children (basic_block bb)
{
if (!bb_in_sese_p (bb, scop->scop_info->region))
return;
if (single_pred_cond_non_loop_exit (bb))
{
edge e = single_pred_edge (bb);
if (e != scop->scop_info->region.entry)
{
conditions.pop ();
cases.pop ();
}
}
}
/* Compute sth like an execution order, dominator order with first executing
edges that stay inside the current loop, delaying processing exit edges. */
static int *bb_to_rpo;
/* Helper for qsort, sorting after order above. */
static int
cmp_pbbs (const void *pa, const void *pb)
{
poly_bb_p bb1 = *((const poly_bb_p *)pa);
poly_bb_p bb2 = *((const poly_bb_p *)pb);
if (bb_to_rpo[bb1->black_box->bb->index]
< bb_to_rpo[bb2->black_box->bb->index])
return -1;
else if (bb_to_rpo[bb1->black_box->bb->index]
> bb_to_rpo[bb2->black_box->bb->index])
return 1;
else
return 0;
}
/* Find Static Control Parts (SCoP) in the current function and pushes
them to SCOPS. */
void
build_scops (vec<scop_p> *scops)
{
if (dump_file)
dp.set_dump_file (dump_file);
scop_detection sb;
sb.build_scop_depth (current_loops->tree_root);
/* Now create scops from the lightweight SESEs. */
vec<sese_l> scops_l = sb.get_scops ();
/* Domwalk needs a bb to RPO mapping. Compute it once here. */
int *postorder = XNEWVEC (int, n_basic_blocks_for_fn (cfun));
int postorder_num = pre_and_rev_post_order_compute (NULL, postorder, true);
bb_to_rpo = XNEWVEC (int, last_basic_block_for_fn (cfun));
for (int i = 0; i < postorder_num; ++i)
bb_to_rpo[postorder[i]] = i;
free (postorder);
int i;
sese_l *s;
FOR_EACH_VEC_ELT (scops_l, i, s)
{
scop_p scop = new_scop (s->entry, s->exit);
/* Record all basic blocks and their conditions in REGION. */
gather_bbs (CDI_DOMINATORS, scop, bb_to_rpo).walk (s->entry->dest);
/* Sort pbbs after execution order for initial schedule generation. */
scop->pbbs.qsort (cmp_pbbs);
if (! build_alias_set (scop))
{
DEBUG_PRINT (dp << "[scop-detection-fail] cannot handle dependences\n");
free_scop (scop);
continue;
}
/* Do not optimize a scop containing only PBBs that do not belong
to any loops. */
if (sb.nb_pbbs_in_loops (scop) == 0)
{
DEBUG_PRINT (dp << "[scop-detection-fail] no data references.\n");
free_scop (scop);
continue;
}
unsigned max_arrays = PARAM_VALUE (PARAM_GRAPHITE_MAX_ARRAYS_PER_SCOP);
if (max_arrays > 0
&& scop->drs.length () >= max_arrays)
{
DEBUG_PRINT (dp << "[scop-detection-fail] too many data references: "
<< scop->drs.length ()
<< " is larger than --param graphite-max-arrays-per-scop="
<< max_arrays << ".\n");
free_scop (scop);
continue;
}
find_scop_parameters (scop);
graphite_dim_t max_dim = PARAM_VALUE (PARAM_GRAPHITE_MAX_NB_SCOP_PARAMS);
if (max_dim > 0
&& scop_nb_params (scop) > max_dim)
{
DEBUG_PRINT (dp << "[scop-detection-fail] too many parameters: "
<< scop_nb_params (scop)
<< " larger than --param graphite-max-nb-scop-params="
<< max_dim << ".\n");
free_scop (scop);
continue;
}
scops->safe_push (scop);
}
free (bb_to_rpo);
bb_to_rpo = NULL;
DEBUG_PRINT (dp << "number of SCoPs: " << (scops ? scops->length () : 0););
}
#endif /* HAVE_isl */
|