aboutsummaryrefslogtreecommitdiffstats
path: root/src/mesa/drivers/dri/i965/brw_vec4.cpp
blob: ee5be56b7132643881f4ab71a40eb104d149fb7a (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
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
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
/*
 * Copyright © 2011 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include "brw_vec4.h"
#include "brw_cfg.h"
#include "brw_vs.h"
#include "brw_dead_control_flow.h"

extern "C" {
#include "main/macros.h"
#include "main/shaderobj.h"
#include "program/prog_print.h"
#include "program/prog_parameter.h"
}

#define MAX_INSTRUCTION (1 << 30)

using namespace brw;

namespace brw {

/**
 * Common helper for constructing swizzles.  When only a subset of
 * channels of a vec4 are used, we don't want to reference the other
 * channels, as that will tell optimization passes that those other
 * channels are used.
 */
unsigned
swizzle_for_size(int size)
{
   static const unsigned size_swizzles[4] = {
      BRW_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
      BRW_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
      BRW_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z),
      BRW_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W),
   };

   assert((size >= 1) && (size <= 4));
   return size_swizzles[size - 1];
}

void
src_reg::init()
{
   memset(this, 0, sizeof(*this));

   this->file = BAD_FILE;
}

src_reg::src_reg(register_file file, int reg, const glsl_type *type)
{
   init();

   this->file = file;
   this->reg = reg;
   if (type && (type->is_scalar() || type->is_vector() || type->is_matrix()))
      this->swizzle = swizzle_for_size(type->vector_elements);
   else
      this->swizzle = BRW_SWIZZLE_XYZW;
}

/** Generic unset register constructor. */
src_reg::src_reg()
{
   init();
}

src_reg::src_reg(float f)
{
   init();

   this->file = IMM;
   this->type = BRW_REGISTER_TYPE_F;
   this->imm.f = f;
}

src_reg::src_reg(uint32_t u)
{
   init();

   this->file = IMM;
   this->type = BRW_REGISTER_TYPE_UD;
   this->imm.u = u;
}

src_reg::src_reg(int32_t i)
{
   init();

   this->file = IMM;
   this->type = BRW_REGISTER_TYPE_D;
   this->imm.i = i;
}

src_reg::src_reg(struct brw_reg reg)
{
   init();

   this->file = HW_REG;
   this->fixed_hw_reg = reg;
   this->type = reg.type;
}

src_reg::src_reg(dst_reg reg)
{
   init();

   this->file = reg.file;
   this->reg = reg.reg;
   this->reg_offset = reg.reg_offset;
   this->type = reg.type;
   this->reladdr = reg.reladdr;
   this->fixed_hw_reg = reg.fixed_hw_reg;

   int swizzles[4];
   int next_chan = 0;
   int last = 0;

   for (int i = 0; i < 4; i++) {
      if (!(reg.writemask & (1 << i)))
         continue;

      swizzles[next_chan++] = last = i;
   }

   for (; next_chan < 4; next_chan++) {
      swizzles[next_chan] = last;
   }

   this->swizzle = BRW_SWIZZLE4(swizzles[0], swizzles[1],
                                swizzles[2], swizzles[3]);
}

bool
src_reg::is_accumulator() const
{
   return file == HW_REG &&
          fixed_hw_reg.file == BRW_ARCHITECTURE_REGISTER_FILE &&
          fixed_hw_reg.nr == BRW_ARF_ACCUMULATOR;
}


void
dst_reg::init()
{
   memset(this, 0, sizeof(*this));
   this->file = BAD_FILE;
   this->writemask = WRITEMASK_XYZW;
}

dst_reg::dst_reg()
{
   init();
}

dst_reg::dst_reg(register_file file, int reg)
{
   init();

   this->file = file;
   this->reg = reg;
}

dst_reg::dst_reg(register_file file, int reg, const glsl_type *type,
                 int writemask)
{
   init();

   this->file = file;
   this->reg = reg;
   this->type = brw_type_for_base_type(type);
   this->writemask = writemask;
}

dst_reg::dst_reg(struct brw_reg reg)
{
   init();

   this->file = HW_REG;
   this->fixed_hw_reg = reg;
   this->type = reg.type;
}

dst_reg::dst_reg(src_reg reg)
{
   init();

   this->file = reg.file;
   this->reg = reg.reg;
   this->reg_offset = reg.reg_offset;
   this->type = reg.type;
   /* How should we do writemasking when converting from a src_reg?  It seems
    * pretty obvious that for src.xxxx the caller wants to write to src.x, but
    * what about for src.wx?  Just special-case src.xxxx for now.
    */
   if (reg.swizzle == BRW_SWIZZLE_XXXX)
      this->writemask = WRITEMASK_X;
   else
      this->writemask = WRITEMASK_XYZW;
   this->reladdr = reg.reladdr;
   this->fixed_hw_reg = reg.fixed_hw_reg;
}

bool
dst_reg::is_null() const
{
   return file == HW_REG &&
          fixed_hw_reg.file == BRW_ARCHITECTURE_REGISTER_FILE &&
          fixed_hw_reg.nr == BRW_ARF_NULL;
}

bool
dst_reg::is_accumulator() const
{
   return file == HW_REG &&
          fixed_hw_reg.file == BRW_ARCHITECTURE_REGISTER_FILE &&
          fixed_hw_reg.nr == BRW_ARF_ACCUMULATOR;
}

bool
vec4_instruction::is_send_from_grf()
{
   switch (opcode) {
   case SHADER_OPCODE_SHADER_TIME_ADD:
   case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
      return true;
   default:
      return false;
   }
}

bool
vec4_visitor::can_do_source_mods(vec4_instruction *inst)
{
   if (brw->gen == 6 && inst->is_math())
      return false;

   if (inst->is_send_from_grf())
      return false;

   if (!inst->can_do_source_mods())
      return false;

   return true;
}

/**
 * Returns how many MRFs an opcode will write over.
 *
 * Note that this is not the 0 or 1 implied writes in an actual gen
 * instruction -- the generate_* functions generate additional MOVs
 * for setup.
 */
int
vec4_visitor::implied_mrf_writes(vec4_instruction *inst)
{
   if (inst->mlen == 0)
      return 0;

   switch (inst->opcode) {
   case SHADER_OPCODE_RCP:
   case SHADER_OPCODE_RSQ:
   case SHADER_OPCODE_SQRT:
   case SHADER_OPCODE_EXP2:
   case SHADER_OPCODE_LOG2:
   case SHADER_OPCODE_SIN:
   case SHADER_OPCODE_COS:
      return 1;
   case SHADER_OPCODE_INT_QUOTIENT:
   case SHADER_OPCODE_INT_REMAINDER:
   case SHADER_OPCODE_POW:
      return 2;
   case VS_OPCODE_URB_WRITE:
      return 1;
   case VS_OPCODE_PULL_CONSTANT_LOAD:
      return 2;
   case SHADER_OPCODE_GEN4_SCRATCH_READ:
      return 2;
   case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
      return 3;
   case GS_OPCODE_URB_WRITE:
   case GS_OPCODE_THREAD_END:
      return 0;
   case SHADER_OPCODE_SHADER_TIME_ADD:
      return 0;
   case SHADER_OPCODE_TEX:
   case SHADER_OPCODE_TXL:
   case SHADER_OPCODE_TXD:
   case SHADER_OPCODE_TXF:
   case SHADER_OPCODE_TXF_CMS:
   case SHADER_OPCODE_TXF_MCS:
   case SHADER_OPCODE_TXS:
   case SHADER_OPCODE_TG4:
   case SHADER_OPCODE_TG4_OFFSET:
      return inst->header_present ? 1 : 0;
   case SHADER_OPCODE_UNTYPED_ATOMIC:
   case SHADER_OPCODE_UNTYPED_SURFACE_READ:
      return 0;
   default:
      assert(!"not reached");
      return inst->mlen;
   }
}

bool
src_reg::equals(const src_reg &r) const
{
   return (file == r.file &&
	   reg == r.reg &&
	   reg_offset == r.reg_offset &&
	   type == r.type &&
	   negate == r.negate &&
	   abs == r.abs &&
	   swizzle == r.swizzle &&
	   !reladdr && !r.reladdr &&
	   memcmp(&fixed_hw_reg, &r.fixed_hw_reg,
		  sizeof(fixed_hw_reg)) == 0 &&
	   imm.u == r.imm.u);
}

static bool
try_eliminate_instruction(vec4_instruction *inst, int new_writemask,
                          const struct brw_context *brw)
{
   if (inst->has_side_effects())
      return false;

   if (new_writemask == 0) {
      /* Don't dead code eliminate instructions that write to the
       * accumulator as a side-effect. Instead just set the destination
       * to the null register to free it.
       */
      if (inst->writes_accumulator || inst->writes_flag()) {
         inst->dst = dst_reg(retype(brw_null_reg(), inst->dst.type));
      } else {
         inst->remove();
      }

      return true;
   } else if (inst->dst.writemask != new_writemask) {
      switch (inst->opcode) {
      case SHADER_OPCODE_TXF_CMS:
      case SHADER_OPCODE_GEN4_SCRATCH_READ:
      case VS_OPCODE_PULL_CONSTANT_LOAD:
      case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
         break;
      default:
         /* Do not set a writemask on Gen6 for math instructions, those are
          * executed using align1 mode that does not support a destination mask.
          */
         if (!(brw->gen == 6 && inst->is_math()) && !inst->is_tex()) {
            inst->dst.writemask = new_writemask;
            return true;
         }
      }
   }

   return false;
}

/**
 * Must be called after calculate_live_intervals() to remove unused
 * writes to registers -- register allocation will fail otherwise
 * because something deffed but not used won't be considered to
 * interfere with other regs.
 */
bool
vec4_visitor::dead_code_eliminate()
{
   bool progress = false;
   int pc = -1;

   calculate_live_intervals();

   foreach_list_safe(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      pc++;

      bool inst_writes_flag = false;
      if (inst->dst.file != GRF) {
         if (inst->dst.is_null() && inst->writes_flag()) {
            inst_writes_flag = true;
         } else {
            continue;
         }
      }

      if (inst->dst.file == GRF) {
         int write_mask = inst->dst.writemask;

         for (int c = 0; c < 4; c++) {
            if (write_mask & (1 << c)) {
               assert(this->virtual_grf_end[inst->dst.reg * 4 + c] >= pc);
               if (this->virtual_grf_end[inst->dst.reg * 4 + c] == pc) {
                  write_mask &= ~(1 << c);
               }
            }
         }

         progress = try_eliminate_instruction(inst, write_mask, brw) ||
                    progress;
      }

      if (inst->predicate || inst->prev == NULL)
         continue;

      int dead_channels;
      if (inst_writes_flag) {
/* Arbitrarily chosen, other than not being an xyzw writemask. */
#define FLAG_WRITEMASK (1 << 5)
         dead_channels = inst->reads_flag() ? 0 : FLAG_WRITEMASK;
      } else {
         dead_channels = inst->dst.writemask;

         for (int i = 0; i < 3; i++) {
            if (inst->src[i].file != GRF ||
                inst->src[i].reg != inst->dst.reg)
                  continue;

            for (int j = 0; j < 4; j++) {
               int swiz = BRW_GET_SWZ(inst->src[i].swizzle, j);
               dead_channels &= ~(1 << swiz);
            }
         }
      }

      for (exec_node *node = inst->prev, *prev = node->prev;
           prev != NULL && dead_channels != 0;
           node = prev, prev = prev->prev) {
         vec4_instruction *scan_inst = (vec4_instruction  *)node;

         if (scan_inst->is_control_flow())
            break;

         if (inst_writes_flag) {
            if (scan_inst->dst.is_null() && scan_inst->writes_flag()) {
               scan_inst->remove();
               progress = true;
               continue;
            } else if (scan_inst->reads_flag()) {
               break;
            }
         }

         if (inst->dst.file == scan_inst->dst.file &&
             inst->dst.reg == scan_inst->dst.reg &&
             inst->dst.reg_offset == scan_inst->dst.reg_offset) {
            int new_writemask = scan_inst->dst.writemask & ~dead_channels;

            progress = try_eliminate_instruction(scan_inst, new_writemask, brw) ||
                       progress;
         }

         for (int i = 0; i < 3; i++) {
            if (scan_inst->src[i].file != inst->dst.file ||
                scan_inst->src[i].reg != inst->dst.reg)
               continue;

            for (int j = 0; j < 4; j++) {
               int swiz = BRW_GET_SWZ(scan_inst->src[i].swizzle, j);
               dead_channels &= ~(1 << swiz);
            }
         }
      }
   }

   if (progress)
      invalidate_live_intervals();

   return progress;
}

void
vec4_visitor::split_uniform_registers()
{
   /* Prior to this, uniforms have been in an array sized according to
    * the number of vector uniforms present, sparsely filled (so an
    * aggregate results in reg indices being skipped over).  Now we're
    * going to cut those aggregates up so each .reg index is one
    * vector.  The goal is to make elimination of unused uniform
    * components easier later.
    */
   foreach_list(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      for (int i = 0 ; i < 3; i++) {
	 if (inst->src[i].file != UNIFORM)
	    continue;

	 assert(!inst->src[i].reladdr);

	 inst->src[i].reg += inst->src[i].reg_offset;
	 inst->src[i].reg_offset = 0;
      }
   }

   /* Update that everything is now vector-sized. */
   for (int i = 0; i < this->uniforms; i++) {
      this->uniform_size[i] = 1;
   }
}

void
vec4_visitor::pack_uniform_registers()
{
   bool uniform_used[this->uniforms];
   int new_loc[this->uniforms];
   int new_chan[this->uniforms];

   memset(uniform_used, 0, sizeof(uniform_used));
   memset(new_loc, 0, sizeof(new_loc));
   memset(new_chan, 0, sizeof(new_chan));

   /* Find which uniform vectors are actually used by the program.  We
    * expect unused vector elements when we've moved array access out
    * to pull constants, and from some GLSL code generators like wine.
    */
   foreach_list(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      for (int i = 0 ; i < 3; i++) {
	 if (inst->src[i].file != UNIFORM)
	    continue;

	 uniform_used[inst->src[i].reg] = true;
      }
   }

   int new_uniform_count = 0;

   /* Now, figure out a packing of the live uniform vectors into our
    * push constants.
    */
   for (int src = 0; src < uniforms; src++) {
      assert(src < uniform_array_size);
      int size = this->uniform_vector_size[src];

      if (!uniform_used[src]) {
	 this->uniform_vector_size[src] = 0;
	 continue;
      }

      int dst;
      /* Find the lowest place we can slot this uniform in. */
      for (dst = 0; dst < src; dst++) {
	 if (this->uniform_vector_size[dst] + size <= 4)
	    break;
      }

      if (src == dst) {
	 new_loc[src] = dst;
	 new_chan[src] = 0;
      } else {
	 new_loc[src] = dst;
	 new_chan[src] = this->uniform_vector_size[dst];

	 /* Move the references to the data */
	 for (int j = 0; j < size; j++) {
	    stage_prog_data->param[dst * 4 + new_chan[src] + j] =
	       stage_prog_data->param[src * 4 + j];
	 }

	 this->uniform_vector_size[dst] += size;
	 this->uniform_vector_size[src] = 0;
      }

      new_uniform_count = MAX2(new_uniform_count, dst + 1);
   }

   this->uniforms = new_uniform_count;

   /* Now, update the instructions for our repacked uniforms. */
   foreach_list(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      for (int i = 0 ; i < 3; i++) {
	 int src = inst->src[i].reg;

	 if (inst->src[i].file != UNIFORM)
	    continue;

	 inst->src[i].reg = new_loc[src];

	 int sx = BRW_GET_SWZ(inst->src[i].swizzle, 0) + new_chan[src];
	 int sy = BRW_GET_SWZ(inst->src[i].swizzle, 1) + new_chan[src];
	 int sz = BRW_GET_SWZ(inst->src[i].swizzle, 2) + new_chan[src];
	 int sw = BRW_GET_SWZ(inst->src[i].swizzle, 3) + new_chan[src];
	 inst->src[i].swizzle = BRW_SWIZZLE4(sx, sy, sz, sw);
      }
   }
}

bool
src_reg::is_zero() const
{
   if (file != IMM)
      return false;

   if (type == BRW_REGISTER_TYPE_F) {
      return imm.f == 0.0;
   } else {
      return imm.i == 0;
   }
}

bool
src_reg::is_one() const
{
   if (file != IMM)
      return false;

   if (type == BRW_REGISTER_TYPE_F) {
      return imm.f == 1.0;
   } else {
      return imm.i == 1;
   }
}

/**
 * Does algebraic optimizations (0 * a = 0, 1 * a = a, a + 0 = a).
 *
 * While GLSL IR also performs this optimization, we end up with it in
 * our instruction stream for a couple of reasons.  One is that we
 * sometimes generate silly instructions, for example in array access
 * where we'll generate "ADD offset, index, base" even if base is 0.
 * The other is that GLSL IR's constant propagation doesn't track the
 * components of aggregates, so some VS patterns (initialize matrix to
 * 0, accumulate in vertex blending factors) end up breaking down to
 * instructions involving 0.
 */
bool
vec4_visitor::opt_algebraic()
{
   bool progress = false;

   foreach_list(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      switch (inst->opcode) {
      case BRW_OPCODE_ADD:
	 if (inst->src[1].is_zero()) {
	    inst->opcode = BRW_OPCODE_MOV;
	    inst->src[1] = src_reg();
	    progress = true;
	 }
	 break;

      case BRW_OPCODE_MUL:
	 if (inst->src[1].is_zero()) {
	    inst->opcode = BRW_OPCODE_MOV;
	    switch (inst->src[0].type) {
	    case BRW_REGISTER_TYPE_F:
	       inst->src[0] = src_reg(0.0f);
	       break;
	    case BRW_REGISTER_TYPE_D:
	       inst->src[0] = src_reg(0);
	       break;
	    case BRW_REGISTER_TYPE_UD:
	       inst->src[0] = src_reg(0u);
	       break;
	    default:
	       assert(!"not reached");
	       inst->src[0] = src_reg(0.0f);
	       break;
	    }
	    inst->src[1] = src_reg();
	    progress = true;
	 } else if (inst->src[1].is_one()) {
	    inst->opcode = BRW_OPCODE_MOV;
	    inst->src[1] = src_reg();
	    progress = true;
	 }
	 break;
      default:
	 break;
      }
   }

   if (progress)
      invalidate_live_intervals();

   return progress;
}

/**
 * Only a limited number of hardware registers may be used for push
 * constants, so this turns access to the overflowed constants into
 * pull constants.
 */
void
vec4_visitor::move_push_constants_to_pull_constants()
{
   int pull_constant_loc[this->uniforms];

   /* Only allow 32 registers (256 uniform components) as push constants,
    * which is the limit on gen6.
    */
   int max_uniform_components = 32 * 8;
   if (this->uniforms * 4 <= max_uniform_components)
      return;

   /* Make some sort of choice as to which uniforms get sent to pull
    * constants.  We could potentially do something clever here like
    * look for the most infrequently used uniform vec4s, but leave
    * that for later.
    */
   for (int i = 0; i < this->uniforms * 4; i += 4) {
      pull_constant_loc[i / 4] = -1;

      if (i >= max_uniform_components) {
	 const float **values = &stage_prog_data->param[i];

	 /* Try to find an existing copy of this uniform in the pull
	  * constants if it was part of an array access already.
	  */
	 for (unsigned int j = 0; j < stage_prog_data->nr_pull_params; j += 4) {
	    int matches;

	    for (matches = 0; matches < 4; matches++) {
	       if (stage_prog_data->pull_param[j + matches] != values[matches])
		  break;
	    }

	    if (matches == 4) {
	       pull_constant_loc[i / 4] = j / 4;
	       break;
	    }
	 }

	 if (pull_constant_loc[i / 4] == -1) {
	    assert(stage_prog_data->nr_pull_params % 4 == 0);
	    pull_constant_loc[i / 4] = stage_prog_data->nr_pull_params / 4;

	    for (int j = 0; j < 4; j++) {
	       stage_prog_data->pull_param[stage_prog_data->nr_pull_params++] =
                  values[j];
	    }
	 }
      }
   }

   /* Now actually rewrite usage of the things we've moved to pull
    * constants.
    */
   foreach_list_safe(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      for (int i = 0 ; i < 3; i++) {
	 if (inst->src[i].file != UNIFORM ||
	     pull_constant_loc[inst->src[i].reg] == -1)
	    continue;

	 int uniform = inst->src[i].reg;

	 dst_reg temp = dst_reg(this, glsl_type::vec4_type);

	 emit_pull_constant_load(inst, temp, inst->src[i],
				 pull_constant_loc[uniform]);

	 inst->src[i].file = temp.file;
	 inst->src[i].reg = temp.reg;
	 inst->src[i].reg_offset = temp.reg_offset;
	 inst->src[i].reladdr = NULL;
      }
   }

   /* Repack push constants to remove the now-unused ones. */
   pack_uniform_registers();
}

/**
 * Sets the dependency control fields on instructions after register
 * allocation and before the generator is run.
 *
 * When you have a sequence of instructions like:
 *
 * DP4 temp.x vertex uniform[0]
 * DP4 temp.y vertex uniform[0]
 * DP4 temp.z vertex uniform[0]
 * DP4 temp.w vertex uniform[0]
 *
 * The hardware doesn't know that it can actually run the later instructions
 * while the previous ones are in flight, producing stalls.  However, we have
 * manual fields we can set in the instructions that let it do so.
 */
void
vec4_visitor::opt_set_dependency_control()
{
   vec4_instruction *last_grf_write[BRW_MAX_GRF];
   uint8_t grf_channels_written[BRW_MAX_GRF];
   vec4_instruction *last_mrf_write[BRW_MAX_GRF];
   uint8_t mrf_channels_written[BRW_MAX_GRF];

   cfg_t cfg(&instructions);

   assert(prog_data->total_grf ||
          !"Must be called after register allocation");

   for (int i = 0; i < cfg.num_blocks; i++) {
      bblock_t *bblock = cfg.blocks[i];
      vec4_instruction *inst;

      memset(last_grf_write, 0, sizeof(last_grf_write));
      memset(last_mrf_write, 0, sizeof(last_mrf_write));

      for (inst = (vec4_instruction *)bblock->start;
           inst != (vec4_instruction *)bblock->end->next;
           inst = (vec4_instruction *)inst->next) {
         /* If we read from a register that we were doing dependency control
          * on, don't do dependency control across the read.
          */
         for (int i = 0; i < 3; i++) {
            int reg = inst->src[i].reg + inst->src[i].reg_offset;
            if (inst->src[i].file == GRF) {
               last_grf_write[reg] = NULL;
            } else if (inst->src[i].file == HW_REG) {
               memset(last_grf_write, 0, sizeof(last_grf_write));
               break;
            }
            assert(inst->src[i].file != MRF);
         }

         /* In the presence of send messages, totally interrupt dependency
          * control.  They're long enough that the chance of dependency
          * control around them just doesn't matter.
          */
         if (inst->mlen) {
            memset(last_grf_write, 0, sizeof(last_grf_write));
            memset(last_mrf_write, 0, sizeof(last_mrf_write));
            continue;
         }

         /* It looks like setting dependency control on a predicated
          * instruction hangs the GPU.
          */
         if (inst->predicate) {
            memset(last_grf_write, 0, sizeof(last_grf_write));
            memset(last_mrf_write, 0, sizeof(last_mrf_write));
            continue;
         }

         /* Dependency control does not work well over math instructions.
          */
         if (inst->is_math()) {
            memset(last_grf_write, 0, sizeof(last_grf_write));
            memset(last_mrf_write, 0, sizeof(last_mrf_write));
            continue;
         }

         /* Now, see if we can do dependency control for this instruction
          * against a previous one writing to its destination.
          */
         int reg = inst->dst.reg + inst->dst.reg_offset;
         if (inst->dst.file == GRF) {
            if (last_grf_write[reg] &&
                !(inst->dst.writemask & grf_channels_written[reg])) {
               last_grf_write[reg]->no_dd_clear = true;
               inst->no_dd_check = true;
            } else {
               grf_channels_written[reg] = 0;
            }

            last_grf_write[reg] = inst;
            grf_channels_written[reg] |= inst->dst.writemask;
         } else if (inst->dst.file == MRF) {
            if (last_mrf_write[reg] &&
                !(inst->dst.writemask & mrf_channels_written[reg])) {
               last_mrf_write[reg]->no_dd_clear = true;
               inst->no_dd_check = true;
            } else {
               mrf_channels_written[reg] = 0;
            }

            last_mrf_write[reg] = inst;
            mrf_channels_written[reg] |= inst->dst.writemask;
         } else if (inst->dst.reg == HW_REG) {
            if (inst->dst.fixed_hw_reg.file == BRW_GENERAL_REGISTER_FILE)
               memset(last_grf_write, 0, sizeof(last_grf_write));
            if (inst->dst.fixed_hw_reg.file == BRW_MESSAGE_REGISTER_FILE)
               memset(last_mrf_write, 0, sizeof(last_mrf_write));
         }
      }
   }
}

bool
vec4_instruction::can_reswizzle_dst(int dst_writemask,
                                    int swizzle,
                                    int swizzle_mask)
{
   /* If this instruction sets anything not referenced by swizzle, then we'd
    * totally break it when we reswizzle.
    */
   if (dst.writemask & ~swizzle_mask)
      return false;

   switch (opcode) {
   case BRW_OPCODE_DP4:
   case BRW_OPCODE_DP3:
   case BRW_OPCODE_DP2:
      return true;
   default:
      /* Check if there happens to be no reswizzling required. */
      for (int c = 0; c < 4; c++) {
         int bit = 1 << BRW_GET_SWZ(swizzle, c);
         /* Skip components of the swizzle not used by the dst. */
         if (!(dst_writemask & (1 << c)))
            continue;

         /* We don't do the reswizzling yet, so just sanity check that we
          * don't have to.
          */
         if (bit != (1 << c))
            return false;
      }
      return true;
   }
}

/**
 * For any channels in the swizzle's source that were populated by this
 * instruction, rewrite the instruction to put the appropriate result directly
 * in those channels.
 *
 * e.g. for swizzle=yywx, MUL a.xy b c -> MUL a.yy_x b.yy z.yy_x
 */
void
vec4_instruction::reswizzle_dst(int dst_writemask, int swizzle)
{
   int new_writemask = 0;

   switch (opcode) {
   case BRW_OPCODE_DP4:
   case BRW_OPCODE_DP3:
   case BRW_OPCODE_DP2:
      for (int c = 0; c < 4; c++) {
         int bit = 1 << BRW_GET_SWZ(swizzle, c);
         /* Skip components of the swizzle not used by the dst. */
         if (!(dst_writemask & (1 << c)))
            continue;
         /* If we were populating this component, then populate the
          * corresponding channel of the new dst.
          */
         if (dst.writemask & bit)
            new_writemask |= (1 << c);
      }
      dst.writemask = new_writemask;
      break;
   default:
      for (int c = 0; c < 4; c++) {
         /* Skip components of the swizzle not used by the dst. */
         if (!(dst_writemask & (1 << c)))
            continue;

         /* We don't do the reswizzling yet, so just sanity check that we
          * don't have to.
          */
         assert((1 << BRW_GET_SWZ(swizzle, c)) == (1 << c));
      }
      break;
   }
}

/*
 * Tries to reduce extra MOV instructions by taking temporary GRFs that get
 * just written and then MOVed into another reg and making the original write
 * of the GRF write directly to the final destination instead.
 */
bool
vec4_visitor::opt_register_coalesce()
{
   bool progress = false;
   int next_ip = 0;

   calculate_live_intervals();

   foreach_list_safe(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      int ip = next_ip;
      next_ip++;

      if (inst->opcode != BRW_OPCODE_MOV ||
          (inst->dst.file != GRF && inst->dst.file != MRF) ||
	  inst->predicate ||
	  inst->src[0].file != GRF ||
	  inst->dst.type != inst->src[0].type ||
	  inst->src[0].abs || inst->src[0].negate || inst->src[0].reladdr)
	 continue;

      bool to_mrf = (inst->dst.file == MRF);

      /* Can't coalesce this GRF if someone else was going to
       * read it later.
       */
      if (this->virtual_grf_end[inst->src[0].reg * 4 + 0] > ip ||
          this->virtual_grf_end[inst->src[0].reg * 4 + 1] > ip ||
          this->virtual_grf_end[inst->src[0].reg * 4 + 2] > ip ||
          this->virtual_grf_end[inst->src[0].reg * 4 + 3] > ip)
	 continue;

      /* We need to check interference with the final destination between this
       * instruction and the earliest instruction involved in writing the GRF
       * we're eliminating.  To do that, keep track of which of our source
       * channels we've seen initialized.
       */
      bool chans_needed[4] = {false, false, false, false};
      int chans_remaining = 0;
      int swizzle_mask = 0;
      for (int i = 0; i < 4; i++) {
	 int chan = BRW_GET_SWZ(inst->src[0].swizzle, i);

	 if (!(inst->dst.writemask & (1 << i)))
	    continue;

         swizzle_mask |= (1 << chan);

	 if (!chans_needed[chan]) {
	    chans_needed[chan] = true;
	    chans_remaining++;
	 }
      }

      /* Now walk up the instruction stream trying to see if we can rewrite
       * everything writing to the temporary to write into the destination
       * instead.
       */
      vec4_instruction *scan_inst;
      for (scan_inst = (vec4_instruction *)inst->prev;
	   scan_inst->prev != NULL;
	   scan_inst = (vec4_instruction *)scan_inst->prev) {
	 if (scan_inst->dst.file == GRF &&
	     scan_inst->dst.reg == inst->src[0].reg &&
	     scan_inst->dst.reg_offset == inst->src[0].reg_offset) {
            /* Found something writing to the reg we want to coalesce away. */
            if (to_mrf) {
               /* SEND instructions can't have MRF as a destination. */
               if (scan_inst->mlen)
                  break;

               if (brw->gen == 6) {
                  /* gen6 math instructions must have the destination be
                   * GRF, so no compute-to-MRF for them.
                   */
                  if (scan_inst->is_math()) {
                     break;
                  }
               }
            }

            /* If we can't handle the swizzle, bail. */
            if (!scan_inst->can_reswizzle_dst(inst->dst.writemask,
                                              inst->src[0].swizzle,
                                              swizzle_mask)) {
               break;
            }

	    /* Mark which channels we found unconditional writes for. */
	    if (!scan_inst->predicate) {
	       for (int i = 0; i < 4; i++) {
		  if (scan_inst->dst.writemask & (1 << i) &&
		      chans_needed[i]) {
		     chans_needed[i] = false;
		     chans_remaining--;
		  }
	       }
	    }

	    if (chans_remaining == 0)
	       break;
	 }

	 /* We don't handle flow control here.  Most computation of values
	  * that could be coalesced happens just before their use.
	  */
	 if (scan_inst->opcode == BRW_OPCODE_DO ||
	     scan_inst->opcode == BRW_OPCODE_WHILE ||
	     scan_inst->opcode == BRW_OPCODE_ELSE ||
	     scan_inst->opcode == BRW_OPCODE_ENDIF) {
	    break;
	 }

         /* You can't read from an MRF, so if someone else reads our MRF's
          * source GRF that we wanted to rewrite, that stops us.  If it's a
          * GRF we're trying to coalesce to, we don't actually handle
          * rewriting sources so bail in that case as well.
          */
	 bool interfered = false;
	 for (int i = 0; i < 3; i++) {
	    if (scan_inst->src[i].file == GRF &&
		scan_inst->src[i].reg == inst->src[0].reg &&
		scan_inst->src[i].reg_offset == inst->src[0].reg_offset) {
	       interfered = true;
	    }
	 }
	 if (interfered)
	    break;

         /* If somebody else writes our destination here, we can't coalesce
          * before that.
          */
         if (scan_inst->dst.file == inst->dst.file &&
             scan_inst->dst.reg == inst->dst.reg) {
	    break;
         }

         /* Check for reads of the register we're trying to coalesce into.  We
          * can't go rewriting instructions above that to put some other value
          * in the register instead.
          */
         if (to_mrf && scan_inst->mlen > 0) {
            if (inst->dst.reg >= scan_inst->base_mrf &&
                inst->dst.reg < scan_inst->base_mrf + scan_inst->mlen) {
               break;
            }
         } else {
            for (int i = 0; i < 3; i++) {
               if (scan_inst->src[i].file == inst->dst.file &&
                   scan_inst->src[i].reg == inst->dst.reg &&
                   scan_inst->src[i].reg_offset == inst->src[0].reg_offset) {
                  interfered = true;
               }
            }
            if (interfered)
               break;
         }
      }

      if (chans_remaining == 0) {
	 /* If we've made it here, we have an MOV we want to coalesce out, and
	  * a scan_inst pointing to the earliest instruction involved in
	  * computing the value.  Now go rewrite the instruction stream
	  * between the two.
	  */

	 while (scan_inst != inst) {
	    if (scan_inst->dst.file == GRF &&
		scan_inst->dst.reg == inst->src[0].reg &&
		scan_inst->dst.reg_offset == inst->src[0].reg_offset) {
               scan_inst->reswizzle_dst(inst->dst.writemask,
                                        inst->src[0].swizzle);
	       scan_inst->dst.file = inst->dst.file;
	       scan_inst->dst.reg = inst->dst.reg;
	       scan_inst->dst.reg_offset = inst->dst.reg_offset;
	       scan_inst->saturate |= inst->saturate;
	    }
	    scan_inst = (vec4_instruction *)scan_inst->next;
	 }
	 inst->remove();
	 progress = true;
      }
   }

   if (progress)
      invalidate_live_intervals();

   return progress;
}

/**
 * Splits virtual GRFs requesting more than one contiguous physical register.
 *
 * We initially create large virtual GRFs for temporary structures, arrays,
 * and matrices, so that the dereference visitor functions can add reg_offsets
 * to work their way down to the actual member being accessed.  But when it
 * comes to optimization, we'd like to treat each register as individual
 * storage if possible.
 *
 * So far, the only thing that might prevent splitting is a send message from
 * a GRF on IVB.
 */
void
vec4_visitor::split_virtual_grfs()
{
   int num_vars = this->virtual_grf_count;
   int new_virtual_grf[num_vars];
   bool split_grf[num_vars];

   memset(new_virtual_grf, 0, sizeof(new_virtual_grf));

   /* Try to split anything > 0 sized. */
   for (int i = 0; i < num_vars; i++) {
      split_grf[i] = this->virtual_grf_sizes[i] != 1;
   }

   /* Check that the instructions are compatible with the registers we're trying
    * to split.
    */
   foreach_list(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      /* If there's a SEND message loading from a GRF on gen7+, it needs to be
       * contiguous.
       */
      if (inst->is_send_from_grf()) {
         for (int i = 0; i < 3; i++) {
            if (inst->src[i].file == GRF) {
               split_grf[inst->src[i].reg] = false;
            }
         }
      }
   }

   /* Allocate new space for split regs.  Note that the virtual
    * numbers will be contiguous.
    */
   for (int i = 0; i < num_vars; i++) {
      if (!split_grf[i])
         continue;

      new_virtual_grf[i] = virtual_grf_alloc(1);
      for (int j = 2; j < this->virtual_grf_sizes[i]; j++) {
         int reg = virtual_grf_alloc(1);
         assert(reg == new_virtual_grf[i] + j - 1);
         (void) reg;
      }
      this->virtual_grf_sizes[i] = 1;
   }

   foreach_list(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      if (inst->dst.file == GRF && split_grf[inst->dst.reg] &&
          inst->dst.reg_offset != 0) {
         inst->dst.reg = (new_virtual_grf[inst->dst.reg] +
                          inst->dst.reg_offset - 1);
         inst->dst.reg_offset = 0;
      }
      for (int i = 0; i < 3; i++) {
         if (inst->src[i].file == GRF && split_grf[inst->src[i].reg] &&
             inst->src[i].reg_offset != 0) {
            inst->src[i].reg = (new_virtual_grf[inst->src[i].reg] +
                                inst->src[i].reg_offset - 1);
            inst->src[i].reg_offset = 0;
         }
      }
   }
   invalidate_live_intervals();
}

void
vec4_visitor::dump_instruction(backend_instruction *be_inst)
{
   dump_instruction(be_inst, stderr);
}

void
vec4_visitor::dump_instruction(backend_instruction *be_inst, FILE *file)
{
   vec4_instruction *inst = (vec4_instruction *)be_inst;

   if (inst->predicate) {
      fprintf(file, "(%cf0) ",
             inst->predicate_inverse ? '-' : '+');
   }

   fprintf(file, "%s", brw_instruction_name(inst->opcode));
   if (inst->conditional_mod) {
      fprintf(file, "%s", conditional_modifier[inst->conditional_mod]);
   }
   fprintf(file, " ");

   switch (inst->dst.file) {
   case GRF:
      fprintf(file, "vgrf%d.%d", inst->dst.reg, inst->dst.reg_offset);
      break;
   case MRF:
      fprintf(file, "m%d", inst->dst.reg);
      break;
   case HW_REG:
      if (inst->dst.fixed_hw_reg.file == BRW_ARCHITECTURE_REGISTER_FILE) {
         switch (inst->dst.fixed_hw_reg.nr) {
         case BRW_ARF_NULL:
            fprintf(file, "null");
            break;
         case BRW_ARF_ADDRESS:
            fprintf(file, "a0.%d", inst->dst.fixed_hw_reg.subnr);
            break;
         case BRW_ARF_ACCUMULATOR:
            fprintf(file, "acc%d", inst->dst.fixed_hw_reg.subnr);
            break;
         case BRW_ARF_FLAG:
            fprintf(file, "f%d.%d", inst->dst.fixed_hw_reg.nr & 0xf,
                             inst->dst.fixed_hw_reg.subnr);
            break;
         default:
            fprintf(file, "arf%d.%d", inst->dst.fixed_hw_reg.nr & 0xf,
                               inst->dst.fixed_hw_reg.subnr);
            break;
         }
      } else {
         fprintf(file, "hw_reg%d", inst->dst.fixed_hw_reg.nr);
      }
      if (inst->dst.fixed_hw_reg.subnr)
         fprintf(file, "+%d", inst->dst.fixed_hw_reg.subnr);
      break;
   case BAD_FILE:
      fprintf(file, "(null)");
      break;
   default:
      fprintf(file, "???");
      break;
   }
   if (inst->dst.writemask != WRITEMASK_XYZW) {
      fprintf(file, ".");
      if (inst->dst.writemask & 1)
         fprintf(file, "x");
      if (inst->dst.writemask & 2)
         fprintf(file, "y");
      if (inst->dst.writemask & 4)
         fprintf(file, "z");
      if (inst->dst.writemask & 8)
         fprintf(file, "w");
   }
   fprintf(file, ":%s, ", brw_reg_type_letters(inst->dst.type));

   for (int i = 0; i < 3 && inst->src[i].file != BAD_FILE; i++) {
      if (inst->src[i].negate)
         fprintf(file, "-");
      if (inst->src[i].abs)
         fprintf(file, "|");
      switch (inst->src[i].file) {
      case GRF:
         fprintf(file, "vgrf%d", inst->src[i].reg);
         break;
      case ATTR:
         fprintf(file, "attr%d", inst->src[i].reg);
         break;
      case UNIFORM:
         fprintf(file, "u%d", inst->src[i].reg);
         break;
      case IMM:
         switch (inst->src[i].type) {
         case BRW_REGISTER_TYPE_F:
            fprintf(file, "%fF", inst->src[i].imm.f);
            break;
         case BRW_REGISTER_TYPE_D:
            fprintf(file, "%dD", inst->src[i].imm.i);
            break;
         case BRW_REGISTER_TYPE_UD:
            fprintf(file, "%uU", inst->src[i].imm.u);
            break;
         default:
            fprintf(file, "???");
            break;
         }
         break;
      case HW_REG:
         if (inst->src[i].fixed_hw_reg.negate)
            fprintf(file, "-");
         if (inst->src[i].fixed_hw_reg.abs)
            fprintf(file, "|");
         if (inst->src[i].fixed_hw_reg.file == BRW_ARCHITECTURE_REGISTER_FILE) {
            switch (inst->src[i].fixed_hw_reg.nr) {
            case BRW_ARF_NULL:
               fprintf(file, "null");
               break;
            case BRW_ARF_ADDRESS:
               fprintf(file, "a0.%d", inst->src[i].fixed_hw_reg.subnr);
               break;
            case BRW_ARF_ACCUMULATOR:
               fprintf(file, "acc%d", inst->src[i].fixed_hw_reg.subnr);
               break;
            case BRW_ARF_FLAG:
               fprintf(file, "f%d.%d", inst->src[i].fixed_hw_reg.nr & 0xf,
                                inst->src[i].fixed_hw_reg.subnr);
               break;
            default:
               fprintf(file, "arf%d.%d", inst->src[i].fixed_hw_reg.nr & 0xf,
                                  inst->src[i].fixed_hw_reg.subnr);
               break;
            }
         } else {
            fprintf(file, "hw_reg%d", inst->src[i].fixed_hw_reg.nr);
         }
         if (inst->src[i].fixed_hw_reg.subnr)
            fprintf(file, "+%d", inst->src[i].fixed_hw_reg.subnr);
         if (inst->src[i].fixed_hw_reg.abs)
            fprintf(file, "|");
         break;
      case BAD_FILE:
         fprintf(file, "(null)");
         break;
      default:
         fprintf(file, "???");
         break;
      }

      if (virtual_grf_sizes[inst->src[i].reg] != 1)
         fprintf(file, ".%d", inst->src[i].reg_offset);

      if (inst->src[i].file != IMM) {
         static const char *chans[4] = {"x", "y", "z", "w"};
         fprintf(file, ".");
         for (int c = 0; c < 4; c++) {
            fprintf(file, "%s", chans[BRW_GET_SWZ(inst->src[i].swizzle, c)]);
         }
      }

      if (inst->src[i].abs)
         fprintf(file, "|");

      if (inst->src[i].file != IMM) {
         fprintf(file, ":%s", brw_reg_type_letters(inst->src[i].type));
      }

      if (i < 2 && inst->src[i + 1].file != BAD_FILE)
         fprintf(file, ", ");
   }

   fprintf(file, "\n");
}


static inline struct brw_reg
attribute_to_hw_reg(int attr, bool interleaved)
{
   if (interleaved)
      return stride(brw_vec4_grf(attr / 2, (attr % 2) * 4), 0, 4, 1);
   else
      return brw_vec8_grf(attr, 0);
}


/**
 * Replace each register of type ATTR in this->instructions with a reference
 * to a fixed HW register.
 *
 * If interleaved is true, then each attribute takes up half a register, with
 * register N containing attribute 2*N in its first half and attribute 2*N+1
 * in its second half (this corresponds to the payload setup used by geometry
 * shaders in "single" or "dual instanced" dispatch mode).  If interleaved is
 * false, then each attribute takes up a whole register, with register N
 * containing attribute N (this corresponds to the payload setup used by
 * vertex shaders, and by geometry shaders in "dual object" dispatch mode).
 */
void
vec4_visitor::lower_attributes_to_hw_regs(const int *attribute_map,
                                          bool interleaved)
{
   foreach_list(node, &this->instructions) {
      vec4_instruction *inst = (vec4_instruction *)node;

      /* We have to support ATTR as a destination for GL_FIXED fixup. */
      if (inst->dst.file == ATTR) {
	 int grf = attribute_map[inst->dst.reg + inst->dst.reg_offset];

         /* All attributes used in the shader need to have been assigned a
          * hardware register by the caller
          */
         assert(grf != 0);

	 struct brw_reg reg = attribute_to_hw_reg(grf, interleaved);
	 reg.type = inst->dst.type;
	 reg.dw1.bits.writemask = inst->dst.writemask;

	 inst->dst.file = HW_REG;
	 inst->dst.fixed_hw_reg = reg;
      }

      for (int i = 0; i < 3; i++) {
	 if (inst->src[i].file != ATTR)
	    continue;

	 int grf = attribute_map[inst->src[i].reg + inst->src[i].reg_offset];

         /* All attributes used in the shader need to have been assigned a
          * hardware register by the caller
          */
         assert(grf != 0);

	 struct brw_reg reg = attribute_to_hw_reg(grf, interleaved);
	 reg.dw1.bits.swizzle = inst->src[i].swizzle;
         reg.type = inst->src[i].type;
	 if (inst->src[i].abs)
	    reg = brw_abs(reg);
	 if (inst->src[i].negate)
	    reg = negate(reg);

	 inst->src[i].file = HW_REG;
	 inst->src[i].fixed_hw_reg = reg;
      }
   }
}

int
vec4_vs_visitor::setup_attributes(int payload_reg)
{
   int nr_attributes;
   int attribute_map[VERT_ATTRIB_MAX + 1];
   memset(attribute_map, 0, sizeof(attribute_map));

   nr_attributes = 0;
   for (int i = 0; i < VERT_ATTRIB_MAX; i++) {
      if (vs_prog_data->inputs_read & BITFIELD64_BIT(i)) {
	 attribute_map[i] = payload_reg + nr_attributes;
	 nr_attributes++;
      }
   }

   /* VertexID is stored by the VF as the last vertex element, but we
    * don't represent it with a flag in inputs_read, so we call it
    * VERT_ATTRIB_MAX.
    */
   if (vs_prog_data->uses_vertexid) {
      attribute_map[VERT_ATTRIB_MAX] = payload_reg + nr_attributes;
      nr_attributes++;
   }

   lower_attributes_to_hw_regs(attribute_map, false /* interleaved */);

   /* The BSpec says we always have to read at least one thing from
    * the VF, and it appears that the hardware wedges otherwise.
    */
   if (nr_attributes == 0)
      nr_attributes = 1;

   prog_data->urb_read_length = (nr_attributes + 1) / 2;

   unsigned vue_entries =
      MAX2(nr_attributes, prog_data->vue_map.num_slots);

   if (brw->gen == 6)
      prog_data->urb_entry_size = ALIGN(vue_entries, 8) / 8;
   else
      prog_data->urb_entry_size = ALIGN(vue_entries, 4) / 4;

   return payload_reg + nr_attributes;
}

int
vec4_visitor::setup_uniforms(int reg)
{
   prog_data->dispatch_grf_start_reg = reg;

   /* The pre-gen6 VS requires that some push constants get loaded no
    * matter what, or the GPU would hang.
    */
   if (brw->gen < 6 && this->uniforms == 0) {
      assert(this->uniforms < this->uniform_array_size);
      this->uniform_vector_size[this->uniforms] = 1;

      stage_prog_data->param =
         reralloc(NULL, stage_prog_data->param, const float *, 4);
      for (unsigned int i = 0; i < 4; i++) {
	 unsigned int slot = this->uniforms * 4 + i;
	 static float zero = 0.0;
	 stage_prog_data->param[slot] = &zero;
      }

      this->uniforms++;
      reg++;
   } else {
      reg += ALIGN(uniforms, 2) / 2;
   }

   stage_prog_data->nr_params = this->uniforms * 4;

   prog_data->curb_read_length = reg - prog_data->dispatch_grf_start_reg;

   return reg;
}

void
vec4_vs_visitor::setup_payload(void)
{
   int reg = 0;

   /* The payload always contains important data in g0, which contains
    * the URB handles that are passed on to the URB write at the end
    * of the thread.  So, we always start push constants at g1.
    */
   reg++;

   reg = setup_uniforms(reg);

   reg = setup_attributes(reg);

   this->first_non_payload_grf = reg;
}

src_reg
vec4_visitor::get_timestamp()
{
   assert(brw->gen >= 7);

   src_reg ts = src_reg(brw_reg(BRW_ARCHITECTURE_REGISTER_FILE,
                                BRW_ARF_TIMESTAMP,
                                0,
                                BRW_REGISTER_TYPE_UD,
                                BRW_VERTICAL_STRIDE_0,
                                BRW_WIDTH_4,
                                BRW_HORIZONTAL_STRIDE_4,
                                BRW_SWIZZLE_XYZW,
                                WRITEMASK_XYZW));

   dst_reg dst = dst_reg(this, glsl_type::uvec4_type);

   vec4_instruction *mov = emit(MOV(dst, ts));
   /* We want to read the 3 fields we care about (mostly field 0, but also 2)
    * even if it's not enabled in the dispatch.
    */
   mov->force_writemask_all = true;

   return src_reg(dst);
}

void
vec4_visitor::emit_shader_time_begin()
{
   current_annotation = "shader time start";
   shader_start_time = get_timestamp();
}

void
vec4_visitor::emit_shader_time_end()
{
   current_annotation = "shader time end";
   src_reg shader_end_time = get_timestamp();


   /* Check that there weren't any timestamp reset events (assuming these
    * were the only two timestamp reads that happened).
    */
   src_reg reset_end = shader_end_time;
   reset_end.swizzle = BRW_SWIZZLE_ZZZZ;
   vec4_instruction *test = emit(AND(dst_null_d(), reset_end, src_reg(1u)));
   test->conditional_mod = BRW_CONDITIONAL_Z;

   emit(IF(BRW_PREDICATE_NORMAL));

   /* Take the current timestamp and get the delta. */
   shader_start_time.negate = true;
   dst_reg diff = dst_reg(this, glsl_type::uint_type);
   emit(ADD(diff, shader_start_time, shader_end_time));

   /* If there were no instructions between the two timestamp gets, the diff
    * is 2 cycles.  Remove that overhead, so I can forget about that when
    * trying to determine the time taken for single instructions.
    */
   emit(ADD(diff, src_reg(diff), src_reg(-2u)));

   emit_shader_time_write(st_base, src_reg(diff));
   emit_shader_time_write(st_written, src_reg(1u));
   emit(BRW_OPCODE_ELSE);
   emit_shader_time_write(st_reset, src_reg(1u));
   emit(BRW_OPCODE_ENDIF);
}

void
vec4_visitor::emit_shader_time_write(enum shader_time_shader_type type,
                                     src_reg value)
{
   int shader_time_index =
      brw_get_shader_time_index(brw, shader_prog, prog, type);

   dst_reg dst =
      dst_reg(this, glsl_type::get_array_instance(glsl_type::vec4_type, 2));

   dst_reg offset = dst;
   dst_reg time = dst;
   time.reg_offset++;

   offset.type = BRW_REGISTER_TYPE_UD;
   emit(MOV(offset, src_reg(shader_time_index * SHADER_TIME_STRIDE)));

   time.type = BRW_REGISTER_TYPE_UD;
   emit(MOV(time, src_reg(value)));

   emit(SHADER_OPCODE_SHADER_TIME_ADD, dst_reg(), src_reg(dst));
}

bool
vec4_visitor::run()
{
   sanity_param_count = prog->Parameters->NumParameters;

   if (INTEL_DEBUG & DEBUG_SHADER_TIME)
      emit_shader_time_begin();

   assign_common_binding_table_offsets(0);

   emit_prolog();

   /* Generate VS IR for main().  (the visitor only descends into
    * functions called "main").
    */
   if (shader) {
      visit_instructions(shader->base.ir);
   } else {
      emit_program_code();
   }
   base_ir = NULL;

   if (key->userclip_active && !prog->UsesClipDistanceOut)
      setup_uniform_clipplane_values();

   emit_thread_end();

   /* Before any optimization, push array accesses out to scratch
    * space where we need them to be.  This pass may allocate new
    * virtual GRFs, so we want to do it early.  It also makes sure
    * that we have reladdr computations available for CSE, since we'll
    * often do repeated subexpressions for those.
    */
   if (shader) {
      move_grf_array_access_to_scratch();
      move_uniform_array_access_to_pull_constants();
   } else {
      /* The ARB_vertex_program frontend emits pull constant loads directly
       * rather than using reladdr, so we don't need to walk through all the
       * instructions looking for things to move.  There isn't anything.
       *
       * We do still need to split things to vec4 size.
       */
      split_uniform_registers();
   }
   pack_uniform_registers();
   move_push_constants_to_pull_constants();
   split_virtual_grfs();

   bool progress;
   do {
      progress = false;
      progress = dead_code_eliminate() || progress;
      progress = dead_control_flow_eliminate(this) || progress;
      progress = opt_copy_propagation() || progress;
      progress = opt_algebraic() || progress;
      progress = opt_register_coalesce() || progress;
   } while (progress);


   if (failed)
      return false;

   setup_payload();

   if (false) {
      /* Debug of register spilling: Go spill everything. */
      const int grf_count = virtual_grf_count;
      float spill_costs[virtual_grf_count];
      bool no_spill[virtual_grf_count];
      evaluate_spill_costs(spill_costs, no_spill);
      for (int i = 0; i < grf_count; i++) {
         if (no_spill[i])
            continue;
         spill_reg(i);
      }
   }

   while (!reg_allocate()) {
      if (failed)
         return false;
   }

   opt_schedule_instructions();

   opt_set_dependency_control();

   /* If any state parameters were appended, then ParameterValues could have
    * been realloced, in which case the driver uniform storage set up by
    * _mesa_associate_uniform_storage() would point to freed memory.  Make
    * sure that didn't happen.
    */
   assert(sanity_param_count == prog->Parameters->NumParameters);

   return !failed;
}

} /* namespace brw */

extern "C" {

/**
 * Compile a vertex shader.
 *
 * Returns the final assembly and the program's size.
 */
const unsigned *
brw_vs_emit(struct brw_context *brw,
            struct gl_shader_program *prog,
            struct brw_vs_compile *c,
            struct brw_vs_prog_data *prog_data,
            void *mem_ctx,
            unsigned *final_assembly_size)
{
   bool start_busy = false;
   double start_time = 0;

   if (unlikely(brw->perf_debug)) {
      start_busy = (brw->batch.last_bo &&
                    drm_intel_bo_busy(brw->batch.last_bo));
      start_time = get_time();
   }

   struct brw_shader *shader = NULL;
   if (prog)
      shader = (brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];

   if (unlikely(INTEL_DEBUG & DEBUG_VS))
      brw_dump_ir(brw, "vertex", prog, &shader->base, &c->vp->program.Base);

   vec4_vs_visitor v(brw, c, prog_data, prog, mem_ctx);
   if (!v.run()) {
      if (prog) {
         prog->LinkStatus = false;
         ralloc_strcat(&prog->InfoLog, v.fail_msg);
      }

      _mesa_problem(NULL, "Failed to compile vertex shader: %s\n",
                    v.fail_msg);

      return NULL;
   }

   const unsigned *assembly = NULL;
   if (brw->gen >= 8) {
      gen8_vec4_generator g(brw, prog, &c->vp->program.Base, &prog_data->base,
                            mem_ctx, INTEL_DEBUG & DEBUG_VS);
      assembly = g.generate_assembly(&v.instructions, final_assembly_size);
   } else {
      vec4_generator g(brw, prog, &c->vp->program.Base, &prog_data->base,
                       mem_ctx, INTEL_DEBUG & DEBUG_VS);
      assembly = g.generate_assembly(&v.instructions, final_assembly_size);
   }

   if (unlikely(brw->perf_debug) && shader) {
      if (shader->compiled_once) {
         brw_vs_debug_recompile(brw, prog, &c->key);
      }
      if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
         perf_debug("VS compile took %.03f ms and stalled the GPU\n",
                    (get_time() - start_time) * 1000);
      }
      shader->compiled_once = true;
   }

   return assembly;
}


void
brw_vec4_setup_prog_key_for_precompile(struct gl_context *ctx,
                                       struct brw_vec4_prog_key *key,
                                       GLuint id, struct gl_program *prog)
{
   key->program_string_id = id;
   key->clamp_vertex_color = ctx->API == API_OPENGL_COMPAT;

   unsigned sampler_count = _mesa_fls(prog->SamplersUsed);
   for (unsigned i = 0; i < sampler_count; i++) {
      if (prog->ShadowSamplers & (1 << i)) {
         /* Assume DEPTH_TEXTURE_MODE is the default: X, X, X, 1 */
         key->tex.swizzles[i] =
            MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_ONE);
      } else {
         /* Color sampler: assume no swizzling. */
         key->tex.swizzles[i] = SWIZZLE_XYZW;
      }
   }
}

} /* extern "C" */