summaryrefslogtreecommitdiff
path: root/src/mesa/drivers/dri/i965/brw_vs_emit.c
blob: b6c9e5a1ceb610496d803f4e6184951d646ad3f5 (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
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
/*
 Copyright (C) Intel Corp.  2006.  All Rights Reserved.
 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
 develop this 3D driver.
 
 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 COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS 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.
 
 **********************************************************************/
 /*
  * Authors:
  *   Keith Whitwell <keith@tungstengraphics.com>
  */
            

#include "main/macros.h"
#include "program/program.h"
#include "program/prog_parameter.h"
#include "program/prog_print.h"
#include "brw_context.h"
#include "brw_vs.h"

/* Return the SrcReg index of the channels that can be immediate float operands
 * instead of usage of PROGRAM_CONSTANT values through push/pull.
 */
static GLboolean
brw_vs_arg_can_be_immediate(enum prog_opcode opcode, int arg)
{
   int opcode_array[] = {
      [OPCODE_MOV] = 1,
      [OPCODE_ADD] = 2,
      [OPCODE_CMP] = 3,
      [OPCODE_DP2] = 2,
      [OPCODE_DP3] = 2,
      [OPCODE_DP4] = 2,
      [OPCODE_DPH] = 2,
      [OPCODE_MAX] = 2,
      [OPCODE_MIN] = 2,
      [OPCODE_MUL] = 2,
      [OPCODE_SEQ] = 2,
      [OPCODE_SGE] = 2,
      [OPCODE_SGT] = 2,
      [OPCODE_SLE] = 2,
      [OPCODE_SLT] = 2,
      [OPCODE_SNE] = 2,
      [OPCODE_XPD] = 2,
   };

   /* These opcodes get broken down in a way that allow two
    * args to be immediates.
    */
   if (opcode == OPCODE_MAD || opcode == OPCODE_LRP) {
      if (arg == 1 || arg == 2)
	 return GL_TRUE;
   }

   if (opcode > ARRAY_SIZE(opcode_array))
      return GL_FALSE;

   return arg == opcode_array[opcode] - 1;
}

static struct brw_reg get_tmp( struct brw_vs_compile *c )
{
   struct brw_reg tmp = brw_vec8_grf(c->last_tmp, 0);

   if (++c->last_tmp > c->prog_data.total_grf)
      c->prog_data.total_grf = c->last_tmp;

   return tmp;
}

static void release_tmp( struct brw_vs_compile *c, struct brw_reg tmp )
{
   if (tmp.nr == c->last_tmp-1)
      c->last_tmp--;
}
			       
static void release_tmps( struct brw_vs_compile *c )
{
   c->last_tmp = c->first_tmp;
}

static int
get_first_reladdr_output(struct gl_vertex_program *vp)
{
   int i;
   int first_reladdr_output = VERT_RESULT_MAX;

   for (i = 0; i < vp->Base.NumInstructions; i++) {
      struct prog_instruction *inst = vp->Base.Instructions + i;

      if (inst->DstReg.File == PROGRAM_OUTPUT &&
	  inst->DstReg.RelAddr &&
	  inst->DstReg.Index < first_reladdr_output)
	 first_reladdr_output = inst->DstReg.Index;
   }

   return first_reladdr_output;
}

/* Clears the record of which vp_const_buffer elements have been
 * loaded into our constant buffer registers, for the starts of new
 * blocks after control flow.
 */
static void
clear_current_const(struct brw_vs_compile *c)
{
   unsigned int i;

   if (c->vp->use_const_buffer) {
      for (i = 0; i < 3; i++) {
         c->current_const[i].index = -1;
      }
   }
}

/**
 * Preallocate GRF register before code emit.
 * Do things as simply as possible.  Allocate and populate all regs
 * ahead of time.
 */
static void brw_vs_alloc_regs( struct brw_vs_compile *c )
{
   struct intel_context *intel = &c->func.brw->intel;
   GLuint i, reg = 0, mrf, j;
   int attributes_in_vue;
   int first_reladdr_output;
   int max_constant;
   int constant = 0;
   int vert_result_reoder[VERT_RESULT_MAX];
   int bfc = 0;

   /* Determine whether to use a real constant buffer or use a block
    * of GRF registers for constants.  The later is faster but only
    * works if everything fits in the GRF.
    * XXX this heuristic/check may need some fine tuning...
    */
   if (c->vp->program.Base.Parameters->NumParameters +
       c->vp->program.Base.NumTemporaries + 20 > BRW_MAX_GRF)
      c->vp->use_const_buffer = GL_TRUE;
   else
      c->vp->use_const_buffer = GL_FALSE;

   /*printf("use_const_buffer = %d\n", c->vp->use_const_buffer);*/

   /* r0 -- reserved as usual
    */
   c->r0 = brw_vec8_grf(reg, 0);
   reg++;

   /* User clip planes from curbe: 
    */
   if (c->key.nr_userclip) {
      if (intel->gen >= 6) {
	 for (i = 0; i < c->key.nr_userclip; i++) {
	    c->userplane[i] = stride(brw_vec4_grf(reg + i / 2,
						  (i % 2) * 4), 0, 4, 1);
	 }
	 reg += ALIGN(c->key.nr_userclip, 2) / 2;
      } else {
	 for (i = 0; i < c->key.nr_userclip; i++) {
	    c->userplane[i] = stride(brw_vec4_grf(reg + (6 + i) / 2,
						  (i % 2) * 4), 0, 4, 1);
	 }
	 reg += (ALIGN(6 + c->key.nr_userclip, 4) / 4) * 2;
      }

   }

   /* Assign some (probably all) of the vertex program constants to
    * the push constant buffer/CURBE.
    *
    * There's an obvious limit to the numer of push constants equal to
    * the number of register available, and that number is smaller
    * than the minimum maximum number of vertex program parameters, so
    * support for pull constants is required if we overflow.
    * Additionally, on gen6 the number of push constants is even
    * lower.
    *
    * When there's relative addressing, we don't know what range of
    * Mesa IR registers can be accessed.  And generally, when relative
    * addressing is used we also have too many constants to load them
    * all as push constants.  So, we'll just support relative
    * addressing out of the pull constant buffers, and try to load as
    * many statically-accessed constants into the push constant buffer
    * as we can.
    */
   if (intel->gen >= 6) {
      /* We can only load 32 regs of push constants. */
      max_constant = 32 * 2 - c->key.nr_userclip;
   } else {
      max_constant = BRW_MAX_GRF - 20 - c->vp->program.Base.NumTemporaries;
   }

   /* constant_map maps from ParameterValues[] index to index in the
    * push constant buffer, or -1 if it's only in the pull constant
    * buffer.
    */
   memset(c->constant_map, -1, c->vp->program.Base.Parameters->NumParameters);
   for (i = 0;
	i < c->vp->program.Base.NumInstructions && constant < max_constant;
	i++) {
      struct prog_instruction *inst = &c->vp->program.Base.Instructions[i];
      int arg;

      for (arg = 0; arg < 3 && constant < max_constant; arg++) {
	 if (inst->SrcReg[arg].File != PROGRAM_STATE_VAR &&
	     inst->SrcReg[arg].File != PROGRAM_CONSTANT &&
	     inst->SrcReg[arg].File != PROGRAM_UNIFORM &&
	     inst->SrcReg[arg].File != PROGRAM_ENV_PARAM &&
	     inst->SrcReg[arg].File != PROGRAM_LOCAL_PARAM) {
	    continue;
	 }

	 if (inst->SrcReg[arg].RelAddr) {
	    c->vp->use_const_buffer = GL_TRUE;
	    continue;
	 }

	 if (c->constant_map[inst->SrcReg[arg].Index] == -1) {
	    c->constant_map[inst->SrcReg[arg].Index] = constant++;
	 }
      }
   }

   /* If we ran out of push constant space, then we'll also upload all
    * constants through the pull constant buffer so that they can be
    * accessed no matter what.  For relative addressing (the common
    * case) we need them all in place anyway.
    */
   if (constant == max_constant)
      c->vp->use_const_buffer = GL_TRUE;

   for (i = 0; i < constant; i++) {
      c->regs[PROGRAM_STATE_VAR][i] = stride(brw_vec4_grf(reg + i / 2,
							  (i % 2) * 4),
					     0, 4, 1);
   }
   reg += (constant + 1) / 2;
   c->prog_data.curb_read_length = reg - 1;
   c->prog_data.nr_params = constant * 4;
   /* XXX 0 causes a bug elsewhere... */
   if (intel->gen < 6 && c->prog_data.nr_params == 0)
      c->prog_data.nr_params = 4;

   /* Allocate input regs:  
    */
   c->nr_inputs = 0;
   for (i = 0; i < VERT_ATTRIB_MAX; i++) {
      if (c->prog_data.inputs_read & (1 << i)) {
	 c->nr_inputs++;
	 c->regs[PROGRAM_INPUT][i] = brw_vec8_grf(reg, 0);
	 reg++;
      }
   }
   /* If there are no inputs, we'll still be reading one attribute's worth
    * because it's required -- see urb_read_length setting.
    */
   if (c->nr_inputs == 0)
      reg++;

   /* Allocate outputs.  The non-position outputs go straight into message regs.
    */
   c->nr_outputs = 0;
   c->first_output = reg;
   c->first_overflow_output = 0;

   if (intel->gen >= 6) {
      mrf = 3;
      if (c->key.nr_userclip)
	 mrf += 2;
   } else if (intel->gen == 5)
      mrf = 8;
   else
      mrf = 4;

   first_reladdr_output = get_first_reladdr_output(&c->vp->program);

   for (i = 0; i < VERT_RESULT_MAX; i++)
       vert_result_reoder[i] = i;

   /* adjust attribute order in VUE for BFC0/BFC1 on Gen6+ */
   if (intel->gen >= 6 && c->key.two_side_color) {
       if ((c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_COL1)) &&
           (c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_BFC1))) {
           assert(c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_COL0));
           assert(c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_BFC0));
           bfc = 2;
       } else if ((c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_COL0)) &&
           (c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_BFC0)))
           bfc = 1;

       if (bfc) {
           for (i = 0; i < bfc; i++) {
               vert_result_reoder[VERT_RESULT_COL0 + i * 2 + 0] = VERT_RESULT_COL0 + i;
               vert_result_reoder[VERT_RESULT_COL0 + i * 2 + 1] = VERT_RESULT_BFC0 + i;
           }

           for (i = VERT_RESULT_COL0 + bfc * 2; i < VERT_RESULT_BFC0 + bfc; i++) {
               vert_result_reoder[i] = i - bfc;
           }
       }
   }

   for (j = 0; j < VERT_RESULT_MAX; j++) {
      i = vert_result_reoder[j];

      if (c->prog_data.outputs_written & BITFIELD64_BIT(i)) {
	 c->nr_outputs++;
         assert(i < Elements(c->regs[PROGRAM_OUTPUT]));
	 if (i == VERT_RESULT_HPOS) {
	    c->regs[PROGRAM_OUTPUT][i] = brw_vec8_grf(reg, 0);
	    reg++;
	 }
	 else if (i == VERT_RESULT_PSIZ) {
	    c->regs[PROGRAM_OUTPUT][i] = brw_vec8_grf(reg, 0);
	    reg++;
	 }
	 else {
	    /* Two restrictions on our compute-to-MRF here.  The
	     * message length for all SEND messages is restricted to
	     * [1,15], so we can't use mrf 15, as that means a length
	     * of 16.
	     *
	     * Additionally, URB writes are aligned to URB rows, so we
	     * need to put an even number of registers of URB data in
	     * each URB write so that the later write is aligned.  A
	     * message length of 15 means 1 message header reg plus 14
	     * regs of URB data.
	     *
	     * For attributes beyond the compute-to-MRF, we compute to
	     * GRFs and they will be written in the second URB_WRITE.
	     */
            if (first_reladdr_output > i && mrf < 15) {
               c->regs[PROGRAM_OUTPUT][i] = brw_message_reg(mrf);
               mrf++;
            }
            else {
               if (mrf >= 15 && !c->first_overflow_output)
                  c->first_overflow_output = i;
               c->regs[PROGRAM_OUTPUT][i] = brw_vec8_grf(reg, 0);
               reg++;
	       mrf++;
            }
	 }
      }
   }     

   /* Allocate program temporaries:
    */
   for (i = 0; i < c->vp->program.Base.NumTemporaries; i++) {
      c->regs[PROGRAM_TEMPORARY][i] = brw_vec8_grf(reg, 0);
      reg++;
   }

   /* Address reg(s).  Don't try to use the internal address reg until
    * deref time.
    */
   for (i = 0; i < c->vp->program.Base.NumAddressRegs; i++) {
      c->regs[PROGRAM_ADDRESS][i] =  brw_reg(BRW_GENERAL_REGISTER_FILE,
					     reg,
					     0,
					     BRW_REGISTER_TYPE_D,
					     BRW_VERTICAL_STRIDE_8,
					     BRW_WIDTH_8,
					     BRW_HORIZONTAL_STRIDE_1,
					     BRW_SWIZZLE_XXXX,
					     WRITEMASK_X);
      reg++;
   }

   if (c->vp->use_const_buffer) {
      for (i = 0; i < 3; i++) {
         c->current_const[i].reg = brw_vec8_grf(reg, 0);
         reg++;
      }
      clear_current_const(c);
   }

   for (i = 0; i < 128; i++) {
      if (c->output_regs[i].used_in_src) {
         c->output_regs[i].reg = brw_vec8_grf(reg, 0);
         reg++;
      }
   }

   if (c->needs_stack) {
      c->stack =  brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, reg, 0);
      reg += 2;
   }

   /* Some opcodes need an internal temporary:
    */
   c->first_tmp = reg;
   c->last_tmp = reg;		/* for allocation purposes */

   /* Each input reg holds data from two vertices.  The
    * urb_read_length is the number of registers read from *each*
    * vertex urb, so is half the amount:
    */
   c->prog_data.urb_read_length = (c->nr_inputs + 1) / 2;
   /* Setting this field to 0 leads to undefined behavior according to the
    * the VS_STATE docs.  Our VUEs will always have at least one attribute
    * sitting in them, even if it's padding.
    */
   if (c->prog_data.urb_read_length == 0)
      c->prog_data.urb_read_length = 1;

   /* The VS VUEs are shared by VF (outputting our inputs) and VS, so size
    * them to fit the biggest thing they need to.
    */
   attributes_in_vue = MAX2(c->nr_outputs, c->nr_inputs);

   /* See emit_vertex_write() for where the VUE's overhead on top of the
    * attributes comes from.
    */
   if (intel->gen >= 7) {
      int header_regs = 2;
      if (c->key.nr_userclip)
	 header_regs += 2;

      /* Each attribute is 16 bytes (1 vec4), so dividing by 4 gives us the
       * number of 64-byte (512-bit) units.
       */
      c->prog_data.urb_entry_size = (attributes_in_vue + header_regs + 3) / 4;
   } else if (intel->gen == 6) {
      int header_regs = 2;
      if (c->key.nr_userclip)
	 header_regs += 2;

      /* Each attribute is 16 bytes (1 vec4), so dividing by 8 gives us the
       * number of 128-byte (1024-bit) units.
       */
      c->prog_data.urb_entry_size = (attributes_in_vue + header_regs + 7) / 8;
   } else if (intel->gen == 5)
      /* Each attribute is 16 bytes (1 vec4), so dividing by 4 gives us the
       * number of 64-byte (512-bit) units.
       */
      c->prog_data.urb_entry_size = (attributes_in_vue + 6 + 3) / 4;
   else
      c->prog_data.urb_entry_size = (attributes_in_vue + 2 + 3) / 4;

   c->prog_data.total_grf = reg;

   if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
      printf("%s NumAddrRegs %d\n", __FUNCTION__, c->vp->program.Base.NumAddressRegs);
      printf("%s NumTemps %d\n", __FUNCTION__, c->vp->program.Base.NumTemporaries);
      printf("%s reg = %d\n", __FUNCTION__, reg);
   }
}


/**
 * If an instruction uses a temp reg both as a src and the dest, we
 * sometimes need to allocate an intermediate temporary.
 */
static void unalias1( struct brw_vs_compile *c,
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      void (*func)( struct brw_vs_compile *,
				    struct brw_reg,
				    struct brw_reg ))
{
   if (dst.file == arg0.file && dst.nr == arg0.nr) {
      struct brw_compile *p = &c->func;
      struct brw_reg tmp = brw_writemask(get_tmp(c), dst.dw1.bits.writemask);
      func(c, tmp, arg0);
      brw_MOV(p, dst, tmp);
      release_tmp(c, tmp);
   }
   else {
      func(c, dst, arg0);
   }
}

/**
 * \sa unalias2
 * Checkes if 2-operand instruction needs an intermediate temporary.
 */
static void unalias2( struct brw_vs_compile *c,
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      struct brw_reg arg1,
		      void (*func)( struct brw_vs_compile *,
				    struct brw_reg,
				    struct brw_reg,
				    struct brw_reg ))
{
   if ((dst.file == arg0.file && dst.nr == arg0.nr) ||
       (dst.file == arg1.file && dst.nr == arg1.nr)) {
      struct brw_compile *p = &c->func;
      struct brw_reg tmp = brw_writemask(get_tmp(c), dst.dw1.bits.writemask);
      func(c, tmp, arg0, arg1);
      brw_MOV(p, dst, tmp);
      release_tmp(c, tmp);
   }
   else {
      func(c, dst, arg0, arg1);
   }
}

/**
 * \sa unalias2
 * Checkes if 3-operand instruction needs an intermediate temporary.
 */
static void unalias3( struct brw_vs_compile *c,
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      struct brw_reg arg1,
		      struct brw_reg arg2,
		      void (*func)( struct brw_vs_compile *,
				    struct brw_reg,
				    struct brw_reg,
				    struct brw_reg,
				    struct brw_reg ))
{
   if ((dst.file == arg0.file && dst.nr == arg0.nr) ||
       (dst.file == arg1.file && dst.nr == arg1.nr) ||
       (dst.file == arg2.file && dst.nr == arg2.nr)) {
      struct brw_compile *p = &c->func;
      struct brw_reg tmp = brw_writemask(get_tmp(c), dst.dw1.bits.writemask);
      func(c, tmp, arg0, arg1, arg2);
      brw_MOV(p, dst, tmp);
      release_tmp(c, tmp);
   }
   else {
      func(c, dst, arg0, arg1, arg2);
   }
}

static void emit_sop( struct brw_vs_compile *c,
                      struct brw_reg dst,
                      struct brw_reg arg0,
                      struct brw_reg arg1, 
		      GLuint cond)
{
   struct brw_compile *p = &c->func;

   brw_MOV(p, dst, brw_imm_f(0.0f));
   brw_CMP(p, brw_null_reg(), cond, arg0, arg1);
   brw_MOV(p, dst, brw_imm_f(1.0f));
   brw_set_predicate_control_flag_value(p, 0xff);
}

static void emit_seq( struct brw_vs_compile *c,
                      struct brw_reg dst,
                      struct brw_reg arg0,
                      struct brw_reg arg1 )
{
   emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_EQ);
}

static void emit_sne( struct brw_vs_compile *c,
                      struct brw_reg dst,
                      struct brw_reg arg0,
                      struct brw_reg arg1 )
{
   emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_NEQ);
}
static void emit_slt( struct brw_vs_compile *c,
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      struct brw_reg arg1 )
{
   emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_L);
}

static void emit_sle( struct brw_vs_compile *c,
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      struct brw_reg arg1 )
{
   emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_LE);
}

static void emit_sgt( struct brw_vs_compile *c,
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      struct brw_reg arg1 )
{
   emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_G);
}

static void emit_sge( struct brw_vs_compile *c,
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      struct brw_reg arg1 )
{
  emit_sop(c, dst, arg0, arg1, BRW_CONDITIONAL_GE);
}

static void emit_cmp( struct brw_compile *p,
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      struct brw_reg arg1,
		      struct brw_reg arg2 )
{
   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0, brw_imm_f(0));
   brw_SEL(p, dst, arg1, arg2);
   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}

static void emit_sign(struct brw_vs_compile *c,
		      struct brw_reg dst,
		      struct brw_reg arg0)
{
   struct brw_compile *p = &c->func;

   brw_MOV(p, dst, brw_imm_f(0));

   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0, brw_imm_f(0));
   brw_MOV(p, dst, brw_imm_f(-1.0));
   brw_set_predicate_control(p, BRW_PREDICATE_NONE);

   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, arg0, brw_imm_f(0));
   brw_MOV(p, dst, brw_imm_f(1.0));
   brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}

static void emit_max( struct brw_compile *p, 
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      struct brw_reg arg1 )
{
   struct intel_context *intel = &p->brw->intel;

   if (intel->gen >= 6) {
      brw_set_conditionalmod(p, BRW_CONDITIONAL_GE);
      brw_SEL(p, dst, arg0, arg1);
      brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE);
      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
   } else {
      brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_GE, arg0, arg1);
      brw_SEL(p, dst, arg0, arg1);
      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
   }
}

static void emit_min( struct brw_compile *p, 
		      struct brw_reg dst,
		      struct brw_reg arg0,
		      struct brw_reg arg1 )
{
   struct intel_context *intel = &p->brw->intel;

   if (intel->gen >= 6) {
      brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
      brw_SEL(p, dst, arg0, arg1);
      brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE);
      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
   } else {
      brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, arg0, arg1);
      brw_SEL(p, dst, arg0, arg1);
      brw_set_predicate_control(p, BRW_PREDICATE_NONE);
   }
}

static void emit_arl(struct brw_compile *p,
		     struct brw_reg dst,
		     struct brw_reg src)
{
   struct intel_context *intel = &p->brw->intel;

   if (intel->gen >= 6) {
      struct brw_reg dst_f = retype(dst, BRW_REGISTER_TYPE_F);

      brw_RNDD(p, dst_f, src);
      brw_MOV(p, dst, dst_f);
   } else {
      brw_RNDD(p, dst, src);
   }
}

static void emit_math1_gen4(struct brw_vs_compile *c,
			    GLuint function,
			    struct brw_reg dst,
			    struct brw_reg arg0,
			    GLuint precision)
{
   /* There are various odd behaviours with SEND on the simulator.  In
    * addition there are documented issues with the fact that the GEN4
    * processor doesn't do dependency control properly on SEND
    * results.  So, on balance, this kludge to get around failures
    * with writemasked math results looks like it might be necessary
    * whether that turns out to be a simulator bug or not:
    */
   struct brw_compile *p = &c->func;
   struct brw_reg tmp = dst;
   GLboolean need_tmp = GL_FALSE;

   if (dst.file != BRW_GENERAL_REGISTER_FILE ||
       dst.dw1.bits.writemask != 0xf)
      need_tmp = GL_TRUE;

   if (need_tmp)
      tmp = get_tmp(c);

   brw_math(p, 
	    tmp,
	    function,
	    BRW_MATH_SATURATE_NONE,
	    2,
	    arg0,
	    BRW_MATH_DATA_SCALAR,
	    precision);

   if (need_tmp) {
      brw_MOV(p, dst, tmp);
      release_tmp(c, tmp);
   }
}

static void
emit_math1_gen6(struct brw_vs_compile *c,
		GLuint function,
		struct brw_reg dst,
		struct brw_reg arg0,
		GLuint precision)
{
   struct brw_compile *p = &c->func;
   struct brw_reg tmp_src, tmp_dst;

   /* Something is strange on gen6 math in 16-wide mode, though the
    * docs say it's supposed to work.  Punt to using align1 mode,
    * which doesn't do writemasking and swizzles.
    */
   tmp_src = get_tmp(c);
   tmp_dst = get_tmp(c);

   brw_MOV(p, tmp_src, arg0);

   brw_set_access_mode(p, BRW_ALIGN_1);
   brw_math(p,
	    tmp_dst,
	    function,
	    BRW_MATH_SATURATE_NONE,
	    2,
	    tmp_src,
	    BRW_MATH_DATA_SCALAR,
	    precision);
   brw_set_access_mode(p, BRW_ALIGN_16);

   brw_MOV(p, dst, tmp_dst);

   release_tmp(c, tmp_src);
   release_tmp(c, tmp_dst);
}

static void
emit_math1(struct brw_vs_compile *c,
	   GLuint function,
	   struct brw_reg dst,
	   struct brw_reg arg0,
	   GLuint precision)
{
   struct brw_compile *p = &c->func;
   struct intel_context *intel = &p->brw->intel;

   if (intel->gen >= 6)
      emit_math1_gen6(c, function, dst, arg0, precision);
   else
      emit_math1_gen4(c, function, dst, arg0, precision);
}

static void emit_math2_gen4( struct brw_vs_compile *c, 
			GLuint function,
			struct brw_reg dst,
			struct brw_reg arg0,
			struct brw_reg arg1,
			GLuint precision)
{
   struct brw_compile *p = &c->func;
   struct brw_reg tmp = dst;
   GLboolean need_tmp = GL_FALSE;

   if (dst.file != BRW_GENERAL_REGISTER_FILE ||
       dst.dw1.bits.writemask != 0xf)
      need_tmp = GL_TRUE;

   if (need_tmp) 
      tmp = get_tmp(c);

   brw_MOV(p, brw_message_reg(3), arg1);
   
   brw_math(p, 
	    tmp,
	    function,
	    BRW_MATH_SATURATE_NONE,
	    2,
 	    arg0,
	    BRW_MATH_DATA_SCALAR,
	    precision);

   if (need_tmp) {
      brw_MOV(p, dst, tmp);
      release_tmp(c, tmp);
   }
}

static void emit_math2_gen6( struct brw_vs_compile *c, 
			GLuint function,
			struct brw_reg dst,
			struct brw_reg arg0,
			struct brw_reg arg1,
			GLuint precision)
{
   struct brw_compile *p = &c->func;
   struct brw_reg tmp_src0, tmp_src1, tmp_dst;

   tmp_src0 = get_tmp(c);
   tmp_src1 = get_tmp(c);
   tmp_dst = get_tmp(c);

   brw_MOV(p, tmp_src0, arg0);
   brw_MOV(p, tmp_src1, arg1);
   
   brw_set_access_mode(p, BRW_ALIGN_1);
   brw_math2(p,
	    tmp_dst,
	    function,
	    tmp_src0,
	    tmp_src1);
   brw_set_access_mode(p, BRW_ALIGN_16);

   brw_MOV(p, dst, tmp_dst);

   release_tmp(c, tmp_src0);
   release_tmp(c, tmp_src1);
   release_tmp(c, tmp_dst);
}

static void emit_math2( struct brw_vs_compile *c, 
			GLuint function,
			struct brw_reg dst,
			struct brw_reg arg0,
			struct brw_reg arg1,
			GLuint precision)
{
   struct brw_compile *p = &c->func;
   struct intel_context *intel = &p->brw->intel;

   if (intel->gen >= 6)
      emit_math2_gen6(c, function, dst, arg0, arg1, precision);
   else
      emit_math2_gen4(c, function, dst, arg0, arg1, precision);
}

static void emit_exp_noalias( struct brw_vs_compile *c,
			      struct brw_reg dst,
			      struct brw_reg arg0 )
{
   struct brw_compile *p = &c->func;
   

   if (dst.dw1.bits.writemask & WRITEMASK_X) {
      struct brw_reg tmp = get_tmp(c);
      struct brw_reg tmp_d = retype(tmp, BRW_REGISTER_TYPE_D);

      /* tmp_d = floor(arg0.x) */
      brw_RNDD(p, tmp_d, brw_swizzle1(arg0, 0));

      /* result[0] = 2.0 ^ tmp */

      /* Adjust exponent for floating point: 
       * exp += 127 
       */
      brw_ADD(p, brw_writemask(tmp_d, WRITEMASK_X), tmp_d, brw_imm_d(127));

      /* Install exponent and sign.  
       * Excess drops off the edge: 
       */
      brw_SHL(p, brw_writemask(retype(dst, BRW_REGISTER_TYPE_D), WRITEMASK_X), 
	      tmp_d, brw_imm_d(23));

      release_tmp(c, tmp);
   }

   if (dst.dw1.bits.writemask & WRITEMASK_Y) {
      /* result[1] = arg0.x - floor(arg0.x) */
      brw_FRC(p, brw_writemask(dst, WRITEMASK_Y), brw_swizzle1(arg0, 0));
   }
   
   if (dst.dw1.bits.writemask & WRITEMASK_Z) {
      /* As with the LOG instruction, we might be better off just
       * doing a taylor expansion here, seeing as we have to do all
       * the prep work.
       *
       * If mathbox partial precision is too low, consider also:
       * result[3] = result[0] * EXP(result[1])
       */
      emit_math1(c, 
		 BRW_MATH_FUNCTION_EXP, 
		 brw_writemask(dst, WRITEMASK_Z),
		 brw_swizzle1(arg0, 0), 
		 BRW_MATH_PRECISION_FULL);
   }  

   if (dst.dw1.bits.writemask & WRITEMASK_W) {
      /* result[3] = 1.0; */
      brw_MOV(p, brw_writemask(dst, WRITEMASK_W), brw_imm_f(1));
   }
}


static void emit_log_noalias( struct brw_vs_compile *c,
			      struct brw_reg dst,
			      struct brw_reg arg0 )
{
   struct brw_compile *p = &c->func;
   struct brw_reg tmp = dst;
   struct brw_reg tmp_ud = retype(tmp, BRW_REGISTER_TYPE_UD);
   struct brw_reg arg0_ud = retype(arg0, BRW_REGISTER_TYPE_UD);
   GLboolean need_tmp = (dst.dw1.bits.writemask != 0xf ||
			 dst.file != BRW_GENERAL_REGISTER_FILE);

   if (need_tmp) {
      tmp = get_tmp(c);
      tmp_ud = retype(tmp, BRW_REGISTER_TYPE_UD);
   }
   
   /* Perform mant = frexpf(fabsf(x), &exp), adjust exp and mnt
    * according to spec:
    *
    * These almost look likey they could be joined up, but not really
    * practical:
    *
    * result[0].f = (x.i & ((1<<31)-1) >> 23) - 127
    * result[1].i = (x.i & ((1<<23)-1)        + (127<<23)
    */
   if (dst.dw1.bits.writemask & WRITEMASK_XZ) {
      brw_AND(p, 
	      brw_writemask(tmp_ud, WRITEMASK_X),
	      brw_swizzle1(arg0_ud, 0),
	      brw_imm_ud((1U<<31)-1));

      brw_SHR(p, 
	      brw_writemask(tmp_ud, WRITEMASK_X), 
	      tmp_ud,
	      brw_imm_ud(23));

      brw_ADD(p, 
	      brw_writemask(tmp, WRITEMASK_X), 
	      retype(tmp_ud, BRW_REGISTER_TYPE_D),	/* does it matter? */
	      brw_imm_d(-127));
   }

   if (dst.dw1.bits.writemask & WRITEMASK_YZ) {
      brw_AND(p, 
	      brw_writemask(tmp_ud, WRITEMASK_Y),
	      brw_swizzle1(arg0_ud, 0),
	      brw_imm_ud((1<<23)-1));

      brw_OR(p, 
	     brw_writemask(tmp_ud, WRITEMASK_Y), 
	     tmp_ud,
	     brw_imm_ud(127<<23));
   }
   
   if (dst.dw1.bits.writemask & WRITEMASK_Z) {
      /* result[2] = result[0] + LOG2(result[1]); */

      /* Why bother?  The above is just a hint how to do this with a
       * taylor series.  Maybe we *should* use a taylor series as by
       * the time all the above has been done it's almost certainly
       * quicker than calling the mathbox, even with low precision.
       * 
       * Options are:
       *    - result[0] + mathbox.LOG2(result[1])
       *    - mathbox.LOG2(arg0.x)
       *    - result[0] + inline_taylor_approx(result[1])
       */
      emit_math1(c, 
		 BRW_MATH_FUNCTION_LOG, 
		 brw_writemask(tmp, WRITEMASK_Z), 
		 brw_swizzle1(tmp, 1), 
		 BRW_MATH_PRECISION_FULL);
      
      brw_ADD(p, 
	      brw_writemask(tmp, WRITEMASK_Z), 
	      brw_swizzle1(tmp, 2), 
	      brw_swizzle1(tmp, 0));
   }  

   if (dst.dw1.bits.writemask & WRITEMASK_W) {
      /* result[3] = 1.0; */
      brw_MOV(p, brw_writemask(tmp, WRITEMASK_W), brw_imm_f(1));
   }

   if (need_tmp) {
      brw_MOV(p, dst, tmp);
      release_tmp(c, tmp);
   }
}


/* Need to unalias - consider swizzles:   r0 = DST r0.xxxx r1
 */
static void emit_dst_noalias( struct brw_vs_compile *c, 
			      struct brw_reg dst,
			      struct brw_reg arg0,
			      struct brw_reg arg1)
{
   struct brw_compile *p = &c->func;

   /* There must be a better way to do this: 
    */
   if (dst.dw1.bits.writemask & WRITEMASK_X)
      brw_MOV(p, brw_writemask(dst, WRITEMASK_X), brw_imm_f(1.0));
   if (dst.dw1.bits.writemask & WRITEMASK_Y)
      brw_MUL(p, brw_writemask(dst, WRITEMASK_Y), arg0, arg1);
   if (dst.dw1.bits.writemask & WRITEMASK_Z)
      brw_MOV(p, brw_writemask(dst, WRITEMASK_Z), arg0);
   if (dst.dw1.bits.writemask & WRITEMASK_W)
      brw_MOV(p, brw_writemask(dst, WRITEMASK_W), arg1);
}


static void emit_xpd( struct brw_compile *p,
		      struct brw_reg dst,
		      struct brw_reg t,
		      struct brw_reg u)
{
   brw_MUL(p, brw_null_reg(), brw_swizzle(t, 1,2,0,3),  brw_swizzle(u,2,0,1,3));
   brw_MAC(p, dst,     negate(brw_swizzle(t, 2,0,1,3)), brw_swizzle(u,1,2,0,3));
}


static void emit_lit_noalias( struct brw_vs_compile *c, 
			      struct brw_reg dst,
			      struct brw_reg arg0 )
{
   struct brw_compile *p = &c->func;
   struct brw_reg tmp = dst;
   GLboolean need_tmp = (dst.file != BRW_GENERAL_REGISTER_FILE);

   if (need_tmp) 
      tmp = get_tmp(c);
   
   brw_MOV(p, brw_writemask(dst, WRITEMASK_YZ), brw_imm_f(0)); 
   brw_MOV(p, brw_writemask(dst, WRITEMASK_XW), brw_imm_f(1)); 

   /* Need to use BRW_EXECUTE_8 and also do an 8-wide compare in order
    * to get all channels active inside the IF.  In the clipping code
    * we run with NoMask, so it's not an option and we can use
    * BRW_EXECUTE_1 for all comparisions.
    */
   brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, brw_swizzle1(arg0,0), brw_imm_f(0));
   brw_IF(p, BRW_EXECUTE_8);
   {
      brw_MOV(p, brw_writemask(dst, WRITEMASK_Y), brw_swizzle1(arg0,0));

      brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_G, brw_swizzle1(arg0,1), brw_imm_f(0));
      brw_MOV(p, brw_writemask(tmp, WRITEMASK_Z),  brw_swizzle1(arg0,1));
      brw_set_predicate_control(p, BRW_PREDICATE_NONE);

      emit_math2(c, 
		 BRW_MATH_FUNCTION_POW, 
		 brw_writemask(dst, WRITEMASK_Z),
		 brw_swizzle1(tmp, 2),
		 brw_swizzle1(arg0, 3),
		 BRW_MATH_PRECISION_PARTIAL);      
   }
   brw_ENDIF(p);

   release_tmp(c, tmp);
}

static void emit_lrp_noalias(struct brw_vs_compile *c,
			     struct brw_reg dst,
			     struct brw_reg arg0,
			     struct brw_reg arg1,
			     struct brw_reg arg2)
{
   struct brw_compile *p = &c->func;

   brw_ADD(p, dst, negate(arg0), brw_imm_f(1.0));
   brw_MUL(p, brw_null_reg(), dst, arg2);
   brw_MAC(p, dst, arg0, arg1);
}

/** 3 or 4-component vector normalization */
static void emit_nrm( struct brw_vs_compile *c, 
                      struct brw_reg dst,
                      struct brw_reg arg0,
                      int num_comps)
{
   struct brw_compile *p = &c->func;
   struct brw_reg tmp = get_tmp(c);

   /* tmp = dot(arg0, arg0) */
   if (num_comps == 3)
      brw_DP3(p, tmp, arg0, arg0);
   else
      brw_DP4(p, tmp, arg0, arg0);

   /* tmp = 1 / sqrt(tmp) */
   emit_math1(c, BRW_MATH_FUNCTION_RSQ, tmp, tmp, BRW_MATH_PRECISION_FULL);

   /* dst = arg0 * tmp */
   brw_MUL(p, dst, arg0, tmp);

   release_tmp(c, tmp);
}


static struct brw_reg
get_constant(struct brw_vs_compile *c,
             const struct prog_instruction *inst,
             GLuint argIndex)
{
   const struct prog_src_register *src = &inst->SrcReg[argIndex];
   struct brw_compile *p = &c->func;
   struct brw_reg const_reg = c->current_const[argIndex].reg;

   assert(argIndex < 3);

   if (c->current_const[argIndex].index != src->Index) {
      /* Keep track of the last constant loaded in this slot, for reuse. */
      c->current_const[argIndex].index = src->Index;

#if 0
      printf("  fetch const[%d] for arg %d into reg %d\n",
             src->Index, argIndex, c->current_const[argIndex].reg.nr);
#endif
      /* need to fetch the constant now */
      brw_dp_READ_4_vs(p,
                       const_reg,                     /* writeback dest */
                       16 * src->Index,               /* byte offset */
                       SURF_INDEX_VERT_CONST_BUFFER   /* binding table index */
                       );
   }

   /* replicate lower four floats into upper half (to get XYZWXYZW) */
   const_reg = stride(const_reg, 0, 4, 1);
   const_reg.subnr = 0;

   return const_reg;
}

static struct brw_reg
get_reladdr_constant(struct brw_vs_compile *c,
		     const struct prog_instruction *inst,
		     GLuint argIndex)
{
   const struct prog_src_register *src = &inst->SrcReg[argIndex];
   struct brw_compile *p = &c->func;
   struct brw_context *brw = p->brw;
   struct intel_context *intel = &brw->intel;
   struct brw_reg const_reg = c->current_const[argIndex].reg;
   struct brw_reg addr_reg = c->regs[PROGRAM_ADDRESS][0];
   uint32_t offset;

   assert(argIndex < 3);

   /* Can't reuse a reladdr constant load. */
   c->current_const[argIndex].index = -1;

 #if 0
   printf("  fetch const[a0.x+%d] for arg %d into reg %d\n",
	  src->Index, argIndex, c->current_const[argIndex].reg.nr);
#endif

   if (intel->gen >= 6) {
      offset = src->Index;
   } else {
      struct brw_reg byte_addr_reg = retype(get_tmp(c), BRW_REGISTER_TYPE_D);
      brw_MUL(p, byte_addr_reg, addr_reg, brw_imm_d(16));
      addr_reg = byte_addr_reg;
      offset = 16 * src->Index;
   }

   /* fetch the first vec4 */
   brw_dp_READ_4_vs_relative(p,
			     const_reg,
			     addr_reg,
			     offset,
			     SURF_INDEX_VERT_CONST_BUFFER);

   return const_reg;
}



/* TODO: relative addressing!
 */
static struct brw_reg get_reg( struct brw_vs_compile *c,
			       gl_register_file file,
			       GLuint index )
{
   switch (file) {
   case PROGRAM_TEMPORARY:
   case PROGRAM_INPUT:
   case PROGRAM_OUTPUT:
      assert(c->regs[file][index].nr != 0);
      return c->regs[file][index];
   case PROGRAM_STATE_VAR:
   case PROGRAM_CONSTANT:
   case PROGRAM_UNIFORM:
      assert(c->regs[PROGRAM_STATE_VAR][index].nr != 0);
      return c->regs[PROGRAM_STATE_VAR][index];
   case PROGRAM_ADDRESS:
      assert(index == 0);
      return c->regs[file][index];

   case PROGRAM_UNDEFINED:			/* undef values */
      return brw_null_reg();

   case PROGRAM_LOCAL_PARAM: 
   case PROGRAM_ENV_PARAM: 
   case PROGRAM_WRITE_ONLY:
   default:
      assert(0);
      return brw_null_reg();
   }
}


/**
 * Indirect addressing:  get reg[[arg] + offset].
 */
static struct brw_reg deref( struct brw_vs_compile *c,
			     struct brw_reg arg,
			     GLint offset,
			     GLuint reg_size )
{
   struct brw_compile *p = &c->func;
   struct brw_reg tmp = get_tmp(c);
   struct brw_reg addr_reg = c->regs[PROGRAM_ADDRESS][0];
   struct brw_reg vp_address = retype(vec1(addr_reg), BRW_REGISTER_TYPE_D);
   GLuint byte_offset = arg.nr * 32 + arg.subnr + offset * reg_size;
   struct brw_reg indirect = brw_vec4_indirect(0,0);
   struct brw_reg acc = retype(vec1(get_tmp(c)), BRW_REGISTER_TYPE_UW);

   /* Set the vertical stride on the register access so that the first
    * 4 components come from a0.0 and the second 4 from a0.1.
    */
   indirect.vstride = BRW_VERTICAL_STRIDE_ONE_DIMENSIONAL;

   {
      brw_push_insn_state(p);
      brw_set_access_mode(p, BRW_ALIGN_1);

      brw_MUL(p, acc, vp_address, brw_imm_uw(reg_size));
      brw_ADD(p, brw_address_reg(0), acc, brw_imm_uw(byte_offset));

      brw_MUL(p, acc, suboffset(vp_address, 4), brw_imm_uw(reg_size));
      brw_ADD(p, brw_address_reg(1), acc, brw_imm_uw(byte_offset));

      brw_MOV(p, tmp, indirect);

      brw_pop_insn_state(p);
   }

   /* NOTE: tmp not released */
   return tmp;
}

static void
move_to_reladdr_dst(struct brw_vs_compile *c,
		    const struct prog_instruction *inst,
		    struct brw_reg val)
{
   struct brw_compile *p = &c->func;
   int reg_size = 32;
   struct brw_reg addr_reg = c->regs[PROGRAM_ADDRESS][0];
   struct brw_reg vp_address = retype(vec1(addr_reg), BRW_REGISTER_TYPE_D);
   struct brw_reg base = c->regs[inst->DstReg.File][inst->DstReg.Index];
   GLuint byte_offset = base.nr * 32 + base.subnr;
   struct brw_reg indirect = brw_vec4_indirect(0,0);
   struct brw_reg acc = retype(vec1(get_tmp(c)), BRW_REGISTER_TYPE_UW);

   /* Because destination register indirect addressing can only use
    * one index, we'll write each vertex's vec4 value separately.
    */
   val.width = BRW_WIDTH_4;
   val.vstride = BRW_VERTICAL_STRIDE_4;

   brw_push_insn_state(p);
   brw_set_access_mode(p, BRW_ALIGN_1);

   brw_MUL(p, acc, vp_address, brw_imm_uw(reg_size));
   brw_ADD(p, brw_address_reg(0), acc, brw_imm_uw(byte_offset));
   brw_MOV(p, indirect, val);

   brw_MUL(p, acc, suboffset(vp_address, 4), brw_imm_uw(reg_size));
   brw_ADD(p, brw_address_reg(0), acc,
	   brw_imm_uw(byte_offset + reg_size / 2));
   brw_MOV(p, indirect, suboffset(val, 4));

   brw_pop_insn_state(p);
}

/**
 * Get brw reg corresponding to the instruction's [argIndex] src reg.
 * TODO: relative addressing!
 */
static struct brw_reg
get_src_reg( struct brw_vs_compile *c,
             const struct prog_instruction *inst,
             GLuint argIndex )
{
   const GLuint file = inst->SrcReg[argIndex].File;
   const GLint index = inst->SrcReg[argIndex].Index;
   const GLboolean relAddr = inst->SrcReg[argIndex].RelAddr;

   if (brw_vs_arg_can_be_immediate(inst->Opcode, argIndex)) {
      const struct prog_src_register *src = &inst->SrcReg[argIndex];

      if (src->Swizzle == MAKE_SWIZZLE4(SWIZZLE_ZERO,
					SWIZZLE_ZERO,
					SWIZZLE_ZERO,
					SWIZZLE_ZERO)) {
	  return brw_imm_f(0.0f);
      } else if (src->Swizzle == MAKE_SWIZZLE4(SWIZZLE_ONE,
					       SWIZZLE_ONE,
					       SWIZZLE_ONE,
					       SWIZZLE_ONE)) {
	 if (src->Negate)
	    return brw_imm_f(-1.0F);
	 else
	    return brw_imm_f(1.0F);
      } else if (src->File == PROGRAM_CONSTANT) {
	 const struct gl_program_parameter_list *params;
	 float f;
	 int component = -1;

	 switch (src->Swizzle) {
	 case SWIZZLE_XXXX:
	    component = 0;
	    break;
	 case SWIZZLE_YYYY:
	    component = 1;
	    break;
	 case SWIZZLE_ZZZZ:
	    component = 2;
	    break;
	 case SWIZZLE_WWWW:
	    component = 3;
	    break;
	 }

	 if (component >= 0) {
	    params = c->vp->program.Base.Parameters;
	    f = params->ParameterValues[src->Index][component];

	    if (src->Abs)
	       f = fabs(f);
	    if (src->Negate)
	       f = -f;
	    return brw_imm_f(f);
	 }
      }
   }

   switch (file) {
   case PROGRAM_TEMPORARY:
   case PROGRAM_INPUT:
   case PROGRAM_OUTPUT:
      if (relAddr) {
         return deref(c, c->regs[file][0], index, 32);
      }
      else {
         assert(c->regs[file][index].nr != 0);
         return c->regs[file][index];
      }

   case PROGRAM_STATE_VAR:
   case PROGRAM_CONSTANT:
   case PROGRAM_UNIFORM:
   case PROGRAM_ENV_PARAM:
   case PROGRAM_LOCAL_PARAM:
      if (!relAddr && c->constant_map[index] != -1) {
	 /* Take from the push constant buffer if possible. */
	 assert(c->regs[PROGRAM_STATE_VAR][c->constant_map[index]].nr != 0);
	 return c->regs[PROGRAM_STATE_VAR][c->constant_map[index]];
      } else {
	 /* Must be in the pull constant buffer then .*/
	 assert(c->vp->use_const_buffer);
	 if (relAddr)
	    return get_reladdr_constant(c, inst, argIndex);
	 else
	    return get_constant(c, inst, argIndex);
      }
   case PROGRAM_ADDRESS:
      assert(index == 0);
      return c->regs[file][index];

   case PROGRAM_UNDEFINED:
      /* this is a normal case since we loop over all three src args */
      return brw_null_reg();

   case PROGRAM_WRITE_ONLY:
   default:
      assert(0);
      return brw_null_reg();
   }
}

/**
 * Return the brw reg for the given instruction's src argument.
 * Will return mangled results for SWZ op.  The emit_swz() function
 * ignores this result and recalculates taking extended swizzles into
 * account.
 */
static struct brw_reg get_arg( struct brw_vs_compile *c,
                               const struct prog_instruction *inst,
                               GLuint argIndex )
{
   const struct prog_src_register *src = &inst->SrcReg[argIndex];
   struct brw_reg reg;

   if (src->File == PROGRAM_UNDEFINED)
      return brw_null_reg();

   reg = get_src_reg(c, inst, argIndex);

   /* Convert 3-bit swizzle to 2-bit.  
    */
   if (reg.file != BRW_IMMEDIATE_VALUE) {
      reg.dw1.bits.swizzle = BRW_SWIZZLE4(GET_SWZ(src->Swizzle, 0),
					  GET_SWZ(src->Swizzle, 1),
					  GET_SWZ(src->Swizzle, 2),
					  GET_SWZ(src->Swizzle, 3));

      /* Note this is ok for non-swizzle ARB_vp instructions */
      reg.negate = src->Negate ? 1 : 0;
   }

   return reg;
}


/**
 * Get brw register for the given program dest register.
 */
static struct brw_reg get_dst( struct brw_vs_compile *c,
			       struct prog_dst_register dst )
{
   struct brw_reg reg;

   switch (dst.File) {
   case PROGRAM_TEMPORARY:
   case PROGRAM_OUTPUT:
      /* register-indirect addressing is only 1x1, not VxH, for
       * destination regs.  So, for RelAddr we'll return a temporary
       * for the dest and do a move of the result to the RelAddr
       * register after the instruction emit.
       */
      if (dst.RelAddr) {
	 reg = get_tmp(c);
      } else {
	 assert(c->regs[dst.File][dst.Index].nr != 0);
	 reg = c->regs[dst.File][dst.Index];
      }
      break;
   case PROGRAM_ADDRESS:
      assert(dst.Index == 0);
      reg = c->regs[dst.File][dst.Index];
      break;
   case PROGRAM_UNDEFINED:
      /* we may hit this for OPCODE_END, OPCODE_KIL, etc */
      reg = brw_null_reg();
      break;
   default:
      assert(0);
      reg = brw_null_reg();
   }

   assert(reg.type != BRW_IMMEDIATE_VALUE);
   reg.dw1.bits.writemask = dst.WriteMask;

   return reg;
}


static void emit_swz( struct brw_vs_compile *c, 
		      struct brw_reg dst,
                      const struct prog_instruction *inst)
{
   const GLuint argIndex = 0;
   const struct prog_src_register src = inst->SrcReg[argIndex];
   struct brw_compile *p = &c->func;
   GLuint zeros_mask = 0;
   GLuint ones_mask = 0;
   GLuint src_mask = 0;
   GLubyte src_swz[4];
   GLboolean need_tmp = (src.Negate &&
			 dst.file != BRW_GENERAL_REGISTER_FILE);
   struct brw_reg tmp = dst;
   GLuint i;

   if (need_tmp)
      tmp = get_tmp(c);

   for (i = 0; i < 4; i++) {
      if (dst.dw1.bits.writemask & (1<<i)) {
	 GLubyte s = GET_SWZ(src.Swizzle, i);
	 switch (s) {
	 case SWIZZLE_X:
	 case SWIZZLE_Y:
	 case SWIZZLE_Z:
	 case SWIZZLE_W:
	    src_mask |= 1<<i;
	    src_swz[i] = s;
	    break;
	 case SWIZZLE_ZERO:
	    zeros_mask |= 1<<i;
	    break;
	 case SWIZZLE_ONE:
	    ones_mask |= 1<<i;
	    break;
	 }
      }
   }
   
   /* Do src first, in case dst aliases src:
    */
   if (src_mask) {
      struct brw_reg arg0;

      arg0 = get_src_reg(c, inst, argIndex);

      arg0 = brw_swizzle(arg0, 
			 src_swz[0], src_swz[1], 
			 src_swz[2], src_swz[3]);

      brw_MOV(p, brw_writemask(tmp, src_mask), arg0);
   } 
   
   if (zeros_mask) 
      brw_MOV(p, brw_writemask(tmp, zeros_mask), brw_imm_f(0));

   if (ones_mask) 
      brw_MOV(p, brw_writemask(tmp, ones_mask), brw_imm_f(1));

   if (src.Negate)
      brw_MOV(p, brw_writemask(tmp, src.Negate), negate(tmp));
   
   if (need_tmp) {
      brw_MOV(p, dst, tmp);
      release_tmp(c, tmp);
   }
}

static int
align_interleaved_urb_mlen(struct brw_context *brw, int mlen)
{
   struct intel_context *intel = &brw->intel;

   if (intel->gen >= 6) {
      /* URB data written (does not include the message header reg) must
       * be a multiple of 256 bits, or 2 VS registers.  See vol5c.5,
       * section 5.4.3.2.2: URB_INTERLEAVED.
       *
       * URB entries are allocated on a multiple of 1024 bits, so an
       * extra 128 bits written here to make the end align to 256 is
       * no problem.
       */
      if ((mlen % 2) != 1)
	 mlen++;
   }

   return mlen;
}

/**
 * Post-vertex-program processing.  Send the results to the URB.
 */
static void emit_vertex_write( struct brw_vs_compile *c)
{
   struct brw_compile *p = &c->func;
   struct brw_context *brw = p->brw;
   struct intel_context *intel = &brw->intel;
   struct brw_reg pos = c->regs[PROGRAM_OUTPUT][VERT_RESULT_HPOS];
   struct brw_reg ndc;
   int eot;
   GLuint len_vertex_header = 2;
   int next_mrf, i;
   int msg_len;

   if (c->key.copy_edgeflag) {
      brw_MOV(p, 
	      get_reg(c, PROGRAM_OUTPUT, VERT_RESULT_EDGE),
	      get_reg(c, PROGRAM_INPUT, VERT_ATTRIB_EDGEFLAG));
   }

   if (intel->gen < 6) {
      /* Build ndc coords */
      ndc = get_tmp(c);
      /* ndc = 1.0 / pos.w */
      emit_math1(c, BRW_MATH_FUNCTION_INV, ndc, brw_swizzle1(pos, 3), BRW_MATH_PRECISION_FULL);
      /* ndc.xyz = pos * ndc */
      brw_MUL(p, brw_writemask(ndc, WRITEMASK_XYZ), pos, ndc);
   }

   /* Update the header for point size, user clipping flags, and -ve rhw
    * workaround.
    */
   if (intel->gen >= 6) {
      struct brw_reg m1 = brw_message_reg(1);

      /* On gen6, m1 has each value in a separate dword, so we never
       * need to mess with a temporary for computing the m1 value.
       */
      brw_MOV(p, retype(m1, BRW_REGISTER_TYPE_UD), brw_imm_ud(0));
      if (c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_PSIZ)) {
	 brw_MOV(p, brw_writemask(m1, WRITEMASK_W),
		 brw_swizzle1(c->regs[PROGRAM_OUTPUT][VERT_RESULT_PSIZ], 0));
      }

      /* Set the user clip distances in dword 8-15. (m3-4)*/
      if (c->key.nr_userclip) {
	 for (i = 0; i < c->key.nr_userclip; i++) {
	    struct brw_reg m;
	    if (i < 4)
	       m = brw_message_reg(3);
	    else
	       m = brw_message_reg(4);

	    brw_DP4(p, brw_writemask(m, (1 << (i & 7))),pos, c->userplane[i]);
	 }
      }
   } else if ((c->prog_data.outputs_written &
	       BITFIELD64_BIT(VERT_RESULT_PSIZ)) ||
	      c->key.nr_userclip || brw->has_negative_rhw_bug) {
      struct brw_reg header1 = retype(get_tmp(c), BRW_REGISTER_TYPE_UD);
      GLuint i;

      brw_MOV(p, header1, brw_imm_ud(0));

      brw_set_access_mode(p, BRW_ALIGN_16);	

      if (c->prog_data.outputs_written & BITFIELD64_BIT(VERT_RESULT_PSIZ)) {
	 struct brw_reg psiz = c->regs[PROGRAM_OUTPUT][VERT_RESULT_PSIZ];
	 brw_MUL(p, brw_writemask(header1, WRITEMASK_W),
		 brw_swizzle1(psiz, 0), brw_imm_f(1<<11));
	 brw_AND(p, brw_writemask(header1, WRITEMASK_W),
		 header1, brw_imm_ud(0x7ff<<8));
      }

      for (i = 0; i < c->key.nr_userclip; i++) {
	 brw_set_conditionalmod(p, BRW_CONDITIONAL_L);
	 brw_DP4(p, brw_null_reg(), pos, c->userplane[i]);
	 brw_OR(p, brw_writemask(header1, WRITEMASK_W), header1, brw_imm_ud(1<<i));
	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
      }

      /* i965 clipping workaround: 
       * 1) Test for -ve rhw
       * 2) If set, 
       *      set ndc = (0,0,0,0)
       *      set ucp[6] = 1
       *
       * Later, clipping will detect ucp[6] and ensure the primitive is
       * clipped against all fixed planes.
       */
      if (brw->has_negative_rhw_bug) {
	 brw_CMP(p,
		 vec8(brw_null_reg()),
		 BRW_CONDITIONAL_L,
		 brw_swizzle1(ndc, 3),
		 brw_imm_f(0));
   
	 brw_OR(p, brw_writemask(header1, WRITEMASK_W), header1, brw_imm_ud(1<<6));
	 brw_MOV(p, ndc, brw_imm_f(0));
	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
      }

      brw_set_access_mode(p, BRW_ALIGN_1);	/* why? */
      brw_MOV(p, retype(brw_message_reg(1), BRW_REGISTER_TYPE_UD), header1);
      brw_set_access_mode(p, BRW_ALIGN_16);

      release_tmp(c, header1);
   }
   else {
      brw_MOV(p, retype(brw_message_reg(1), BRW_REGISTER_TYPE_UD), brw_imm_ud(0));
   }

   /* Emit the (interleaved) headers for the two vertices - an 8-reg
    * of zeros followed by two sets of NDC coordinates:
    */
   brw_set_access_mode(p, BRW_ALIGN_1);
   brw_set_acc_write_control(p, 0);

   /* The VUE layout is documented in Volume 2a. */
   if (intel->gen >= 6) {
      /* There are 8 or 16 DWs (D0-D15) in VUE header on Sandybridge:
       * dword 0-3 (m1) of the header is indices, point width, clip flags.
       * dword 4-7 (m2) is the 4D space position
       * dword 8-15 (m3,m4) of the vertex header is the user clip distance if
       * enabled.
       * m3 or 5 is the first vertex element data we fill, which is
       * the vertex position.
       */
      brw_MOV(p, brw_message_reg(2), pos);
      len_vertex_header = 1;
      if (c->key.nr_userclip > 0)
	 len_vertex_header += 2;
   } else if (intel->gen == 5) {
      /* There are 20 DWs (D0-D19) in VUE header on Ironlake:
       * dword 0-3 (m1) of the header is indices, point width, clip flags.
       * dword 4-7 (m2) is the ndc position (set above)
       * dword 8-11 (m3) of the vertex header is the 4D space position
       * dword 12-19 (m4,m5) of the vertex header is the user clip distance.
       * m6 is a pad so that the vertex element data is aligned
       * m7 is the first vertex data we fill, which is the vertex position.
       */
      brw_MOV(p, brw_message_reg(2), ndc);
      brw_MOV(p, brw_message_reg(3), pos);
      brw_MOV(p, brw_message_reg(7), pos);
      len_vertex_header = 6;
   } else {
      /* There are 8 dwords in VUE header pre-Ironlake:
       * dword 0-3 (m1) is indices, point width, clip flags.
       * dword 4-7 (m2) is ndc position (set above)
       *
       * dword 8-11 (m3) is the first vertex data, which we always have be the
       * vertex position.
       */
      brw_MOV(p, brw_message_reg(2), ndc);
      brw_MOV(p, brw_message_reg(3), pos);
      len_vertex_header = 2;
   }

   /* Move variable-addressed, non-overflow outputs to their MRFs. */
   next_mrf = 2 + len_vertex_header;
   for (i = 0; i < VERT_RESULT_MAX; i++) {
      if (c->first_overflow_output > 0 && i >= c->first_overflow_output)
	 break;
      if (!(c->prog_data.outputs_written & BITFIELD64_BIT(i)))
	 continue;
      if (i == VERT_RESULT_PSIZ)
	 continue;

      if (i >= VERT_RESULT_TEX0 &&
	  c->regs[PROGRAM_OUTPUT][i].file == BRW_GENERAL_REGISTER_FILE) {
	 brw_MOV(p, brw_message_reg(next_mrf), c->regs[PROGRAM_OUTPUT][i]);
	 next_mrf++;
      } else if (c->regs[PROGRAM_OUTPUT][i].file == BRW_MESSAGE_REGISTER_FILE) {
	 next_mrf = c->regs[PROGRAM_OUTPUT][i].nr + 1;
      }
   }

   eot = (c->first_overflow_output == 0);

   /* Message header, plus VUE header, plus the (first set of) outputs. */
   msg_len = 1 + len_vertex_header + c->nr_outputs;
   msg_len = align_interleaved_urb_mlen(brw, msg_len);
   /* Any outputs beyond BRW_MAX_MRF should be past first_overflow_output */
   msg_len = MIN2(msg_len, (BRW_MAX_MRF - 1)),

   brw_urb_WRITE(p, 
		 brw_null_reg(), /* dest */
		 0,		/* starting mrf reg nr */
		 c->r0,		/* src */
		 0,		/* allocate */
		 1,		/* used */
		 msg_len,
		 0,		/* response len */
		 eot, 		/* eot */
		 eot, 		/* writes complete */
		 0, 		/* urb destination offset */
		 BRW_URB_SWIZZLE_INTERLEAVE);

   if (c->first_overflow_output > 0) {
      /* Not all of the vertex outputs/results fit into the MRF.
       * Move the overflowed attributes from the GRF to the MRF and
       * issue another brw_urb_WRITE().
       */
      GLuint i, mrf = 1;
      for (i = c->first_overflow_output; i < VERT_RESULT_MAX; i++) {
         if (c->prog_data.outputs_written & BITFIELD64_BIT(i)) {
            /* move from GRF to MRF */
            brw_MOV(p, brw_message_reg(mrf), c->regs[PROGRAM_OUTPUT][i]);
            mrf++;
         }
      }

      brw_urb_WRITE(p,
                    brw_null_reg(), /* dest */
                    0,              /* starting mrf reg nr */
                    c->r0,          /* src */
                    0,              /* allocate */
                    1,              /* used */
                    align_interleaved_urb_mlen(brw, mrf),
                    0,              /* response len */
                    1,              /* eot */
                    1,              /* writes complete */
                    14 / 2,  /* urb destination offset */
                    BRW_URB_SWIZZLE_INTERLEAVE);
   }
}

static GLboolean
accumulator_contains(struct brw_vs_compile *c, struct brw_reg val)
{
   struct brw_compile *p = &c->func;
   struct brw_instruction *prev_insn = &p->store[p->nr_insn - 1];

   if (p->nr_insn == 0)
      return GL_FALSE;

   if (val.address_mode != BRW_ADDRESS_DIRECT)
      return GL_FALSE;

   switch (prev_insn->header.opcode) {
   case BRW_OPCODE_MOV:
   case BRW_OPCODE_MAC:
   case BRW_OPCODE_MUL:
      if (prev_insn->header.access_mode == BRW_ALIGN_16 &&
	  prev_insn->header.execution_size == val.width &&
	  prev_insn->bits1.da1.dest_reg_file == val.file &&
	  prev_insn->bits1.da1.dest_reg_type == val.type &&
	  prev_insn->bits1.da1.dest_address_mode == val.address_mode &&
	  prev_insn->bits1.da1.dest_reg_nr == val.nr &&
	  prev_insn->bits1.da16.dest_subreg_nr == val.subnr / 16 &&
	  prev_insn->bits1.da16.dest_writemask == 0xf)
	 return GL_TRUE;
      else
	 return GL_FALSE;
   default:
      return GL_FALSE;
   }
}

static uint32_t
get_predicate(const struct prog_instruction *inst)
{
   if (inst->DstReg.CondMask == COND_TR)
      return BRW_PREDICATE_NONE;

   /* All of GLSL only produces predicates for COND_NE and one channel per
    * vector.  Fail badly if someone starts doing something else, as it might
    * mean infinite looping or something.
    *
    * We'd like to support all the condition codes, but our hardware doesn't
    * quite match the Mesa IR, which is modeled after the NV extensions.  For
    * those, the instruction may update the condition codes or not, then any
    * later instruction may use one of those condition codes.  For gen4, the
    * instruction may update the flags register based on one of the condition
    * codes output by the instruction, and then further instructions may
    * predicate on that.  We can probably support this, but it won't
    * necessarily be easy.
    */
   assert(inst->DstReg.CondMask == COND_NE);

   switch (inst->DstReg.CondSwizzle) {
   case SWIZZLE_XXXX:
      return BRW_PREDICATE_ALIGN16_REPLICATE_X;
   case SWIZZLE_YYYY:
      return BRW_PREDICATE_ALIGN16_REPLICATE_Y;
   case SWIZZLE_ZZZZ:
      return BRW_PREDICATE_ALIGN16_REPLICATE_Z;
   case SWIZZLE_WWWW:
      return BRW_PREDICATE_ALIGN16_REPLICATE_W;
   default:
      _mesa_problem(NULL, "Unexpected predicate: 0x%08x\n",
		    inst->DstReg.CondMask);
      return BRW_PREDICATE_NORMAL;
   }
}

static void
brw_vs_rescale_gl_fixed(struct brw_vs_compile *c)
{
   struct brw_compile *p = &c->func;
   int i;

   for (i = 0; i < VERT_ATTRIB_MAX; i++) {
      if (!(c->prog_data.inputs_read & (1 << i)))
	 continue;

      if (c->key.gl_fixed_input_size[i] != 0) {
	 struct brw_reg reg = c->regs[PROGRAM_INPUT][i];

	 brw_MUL(p,
		 brw_writemask(reg, (1 << c->key.gl_fixed_input_size[i]) - 1),
		 reg, brw_imm_f(1.0 / 65536.0));
      }
   }
}

/* Emit the vertex program instructions here.
 */
void brw_vs_emit(struct brw_vs_compile *c )
{
#define MAX_IF_DEPTH 32
#define MAX_LOOP_DEPTH 32
   struct brw_compile *p = &c->func;
   struct brw_context *brw = p->brw;
   struct intel_context *intel = &brw->intel;
   const GLuint nr_insns = c->vp->program.Base.NumInstructions;
   GLuint insn, loop_depth = 0;
   struct brw_instruction *loop_inst[MAX_LOOP_DEPTH] = { 0 };
   int if_depth_in_loop[MAX_LOOP_DEPTH];
   const struct brw_indirect stack_index = brw_indirect(0, 0);   
   GLuint index;
   GLuint file;

   if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
      printf("vs-mesa:\n");
      _mesa_fprint_program_opt(stdout, &c->vp->program.Base, PROG_PRINT_DEBUG,
			       GL_TRUE);
      printf("\n");
   }

   brw_set_compression_control(p, BRW_COMPRESSION_NONE);
   brw_set_access_mode(p, BRW_ALIGN_16);
   if_depth_in_loop[loop_depth] = 0;

   brw_set_acc_write_control(p, 1);

   for (insn = 0; insn < nr_insns; insn++) {
       GLuint i;
       struct prog_instruction *inst = &c->vp->program.Base.Instructions[insn];

       /* Message registers can't be read, so copy the output into GRF
	* register if they are used in source registers
	*/
       for (i = 0; i < 3; i++) {
	   struct prog_src_register *src = &inst->SrcReg[i];
	   GLuint index = src->Index;
	   GLuint file = src->File;	
	   if (file == PROGRAM_OUTPUT && index != VERT_RESULT_HPOS)
	       c->output_regs[index].used_in_src = GL_TRUE;
       }

       switch (inst->Opcode) {
       case OPCODE_CAL:
       case OPCODE_RET:
	  c->needs_stack = GL_TRUE;
	  break;
       default:
	  break;
       }
   }

   /* Static register allocation
    */
   brw_vs_alloc_regs(c);

   brw_vs_rescale_gl_fixed(c);

   if (c->needs_stack)
      brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack));

   for (insn = 0; insn < nr_insns; insn++) {

      const struct prog_instruction *inst = &c->vp->program.Base.Instructions[insn];
      struct brw_reg args[3], dst;
      GLuint i;

#if 0
      printf("%d: ", insn);
      _mesa_print_instruction(inst);
#endif

      /* Get argument regs.  SWZ is special and does this itself.
       */
      if (inst->Opcode != OPCODE_SWZ)
	  for (i = 0; i < 3; i++) {
	      const struct prog_src_register *src = &inst->SrcReg[i];
	      index = src->Index;
	      file = src->File;	
	      if (file == PROGRAM_OUTPUT && c->output_regs[index].used_in_src)
		  args[i] = c->output_regs[index].reg;
	      else
                  args[i] = get_arg(c, inst, i);
	  }

      /* Get dest regs.  Note that it is possible for a reg to be both
       * dst and arg, given the static allocation of registers.  So
       * care needs to be taken emitting multi-operation instructions.
       */ 
      index = inst->DstReg.Index;
      file = inst->DstReg.File;
      if (file == PROGRAM_OUTPUT && c->output_regs[index].used_in_src)
	  dst = c->output_regs[index].reg;
      else
	  dst = get_dst(c, inst->DstReg);

      if (inst->SaturateMode != SATURATE_OFF) {
	 _mesa_problem(NULL, "Unsupported saturate %d in vertex shader",
                       inst->SaturateMode);
      }

      switch (inst->Opcode) {
      case OPCODE_ABS:
	 args[0].negate = false;
	 brw_MOV(p, dst, brw_abs(args[0]));
	 break;
      case OPCODE_ADD:
	 brw_ADD(p, dst, args[0], args[1]);
	 break;
      case OPCODE_COS:
	 emit_math1(c, BRW_MATH_FUNCTION_COS, dst, args[0], BRW_MATH_PRECISION_FULL);
	 break;
      case OPCODE_DP2:
	 brw_DP2(p, dst, args[0], args[1]);
	 break;
      case OPCODE_DP3:
	 brw_DP3(p, dst, args[0], args[1]);
	 break;
      case OPCODE_DP4:
	 brw_DP4(p, dst, args[0], args[1]);
	 break;
      case OPCODE_DPH:
	 brw_DPH(p, dst, args[0], args[1]);
	 break;
      case OPCODE_NRM3:
	 emit_nrm(c, dst, args[0], 3);
	 break;
      case OPCODE_NRM4:
	 emit_nrm(c, dst, args[0], 4);
	 break;
      case OPCODE_DST:
	 unalias2(c, dst, args[0], args[1], emit_dst_noalias); 
	 break;
      case OPCODE_EXP:
	 unalias1(c, dst, args[0], emit_exp_noalias);
	 break;
      case OPCODE_EX2:
	 emit_math1(c, BRW_MATH_FUNCTION_EXP, dst, args[0], BRW_MATH_PRECISION_FULL);
	 break;
      case OPCODE_ARL:
	 emit_arl(p, dst, args[0]);
	 break;
      case OPCODE_FLR:
	 brw_RNDD(p, dst, args[0]);
	 break;
      case OPCODE_FRC:
	 brw_FRC(p, dst, args[0]);
	 break;
      case OPCODE_LOG:
	 unalias1(c, dst, args[0], emit_log_noalias);
	 break;
      case OPCODE_LG2:
	 emit_math1(c, BRW_MATH_FUNCTION_LOG, dst, args[0], BRW_MATH_PRECISION_FULL);
	 break;
      case OPCODE_LIT:
	 unalias1(c, dst, args[0], emit_lit_noalias);
	 break;
      case OPCODE_LRP:
	 unalias3(c, dst, args[0], args[1], args[2], emit_lrp_noalias);
	 break;
      case OPCODE_MAD:
	 if (!accumulator_contains(c, args[2]))
	    brw_MOV(p, brw_acc_reg(), args[2]);
	 brw_MAC(p, dst, args[0], args[1]);
	 break;
      case OPCODE_CMP:
	 emit_cmp(p, dst, args[0], args[1], args[2]);
	 break;
      case OPCODE_MAX:
	 emit_max(p, dst, args[0], args[1]);
	 break;
      case OPCODE_MIN:
	 emit_min(p, dst, args[0], args[1]);
	 break;
      case OPCODE_MOV:
	 brw_MOV(p, dst, args[0]);
	 break;
      case OPCODE_MUL:
	 brw_MUL(p, dst, args[0], args[1]);
	 break;
      case OPCODE_POW:
	 emit_math2(c, BRW_MATH_FUNCTION_POW, dst, args[0], args[1], BRW_MATH_PRECISION_FULL); 
	 break;
      case OPCODE_RCP:
	 emit_math1(c, BRW_MATH_FUNCTION_INV, dst, args[0], BRW_MATH_PRECISION_FULL);
	 break;
      case OPCODE_RSQ:
	 emit_math1(c, BRW_MATH_FUNCTION_RSQ, dst, brw_abs(args[0]), BRW_MATH_PRECISION_FULL);
	 break;

      case OPCODE_SEQ:
         unalias2(c, dst, args[0], args[1], emit_seq);
         break;
      case OPCODE_SIN:
	 emit_math1(c, BRW_MATH_FUNCTION_SIN, dst, args[0], BRW_MATH_PRECISION_FULL);
	 break;
      case OPCODE_SNE:
         unalias2(c, dst, args[0], args[1], emit_sne);
         break;
      case OPCODE_SGE:
         unalias2(c, dst, args[0], args[1], emit_sge);
	 break;
      case OPCODE_SGT:
         unalias2(c, dst, args[0], args[1], emit_sgt);
         break;
      case OPCODE_SLT:
         unalias2(c, dst, args[0], args[1], emit_slt);
	 break;
      case OPCODE_SLE:
         unalias2(c, dst, args[0], args[1], emit_sle);
         break;
      case OPCODE_SSG:
         unalias1(c, dst, args[0], emit_sign);
         break;
      case OPCODE_SUB:
	 brw_ADD(p, dst, args[0], negate(args[1]));
	 break;
      case OPCODE_SWZ:
	 /* The args[0] value can't be used here as it won't have
	  * correctly encoded the full swizzle:
	  */
	 emit_swz(c, dst, inst);
	 break;
      case OPCODE_TRUNC:
         /* round toward zero */
	 brw_RNDZ(p, dst, args[0]);
	 break;
      case OPCODE_XPD:
	 emit_xpd(p, dst, args[0], args[1]);
	 break;
      case OPCODE_IF: {
	 struct brw_instruction *if_inst = brw_IF(p, BRW_EXECUTE_8);
	 /* Note that brw_IF smashes the predicate_control field. */
	 if_inst->header.predicate_control = get_predicate(inst);
	 if_depth_in_loop[loop_depth]++;
	 break;
      }
      case OPCODE_ELSE:
	 clear_current_const(c);
	 brw_ELSE(p);
	 break;
      case OPCODE_ENDIF:
	 clear_current_const(c);
	 brw_ENDIF(p);
	 if_depth_in_loop[loop_depth]--;
	 break;			
      case OPCODE_BGNLOOP:
	 clear_current_const(c);
         loop_inst[loop_depth++] = brw_DO(p, BRW_EXECUTE_8);
	 if_depth_in_loop[loop_depth] = 0;
         break;
      case OPCODE_BRK:
	 brw_set_predicate_control(p, get_predicate(inst));
	 brw_BREAK(p, if_depth_in_loop[loop_depth]);
	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
         break;
      case OPCODE_CONT:
	 brw_set_predicate_control(p, get_predicate(inst));
	 if (intel->gen >= 6) {
	    gen6_CONT(p, loop_inst[loop_depth - 1]);
	 } else {
	    brw_CONT(p, if_depth_in_loop[loop_depth]);
	 }
         brw_set_predicate_control(p, BRW_PREDICATE_NONE);
         break;

      case OPCODE_ENDLOOP: {
	 clear_current_const(c);
	 struct brw_instruction *inst0, *inst1;
	 GLuint br = 1;

	 loop_depth--;

	 if (intel->gen == 5)
	    br = 2;

	 inst0 = inst1 = brw_WHILE(p, loop_inst[loop_depth]);

	 if (intel->gen < 6) {
	    /* patch all the BREAK/CONT instructions from last BEGINLOOP */
	    while (inst0 > loop_inst[loop_depth]) {
	       inst0--;
	       if (inst0->header.opcode == BRW_OPCODE_BREAK &&
		   inst0->bits3.if_else.jump_count == 0) {
		  inst0->bits3.if_else.jump_count = br * (inst1 - inst0 + 1);
	       } else if (inst0->header.opcode == BRW_OPCODE_CONTINUE &&
			  inst0->bits3.if_else.jump_count == 0) {
		  inst0->bits3.if_else.jump_count = br * (inst1 - inst0);
	       }
	    }
	 }
      }
         break;

      case OPCODE_BRA:
	 brw_set_predicate_control(p, get_predicate(inst));
         brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
	 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
         break;
      case OPCODE_CAL:
	 brw_set_access_mode(p, BRW_ALIGN_1);
	 brw_ADD(p, deref_1d(stack_index, 0), brw_ip_reg(), brw_imm_d(3*16));
	 brw_set_access_mode(p, BRW_ALIGN_16);
	 brw_ADD(p, get_addr_reg(stack_index),
			 get_addr_reg(stack_index), brw_imm_d(4));
         brw_save_call(p, inst->Comment, p->nr_insn);
	 brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
         break;
      case OPCODE_RET:
	 brw_ADD(p, get_addr_reg(stack_index),
			 get_addr_reg(stack_index), brw_imm_d(-4));
	 brw_set_access_mode(p, BRW_ALIGN_1);
         brw_MOV(p, brw_ip_reg(), deref_1d(stack_index, 0));
	 brw_set_access_mode(p, BRW_ALIGN_16);
	 break;
      case OPCODE_END:
	 emit_vertex_write(c);
         break;
      case OPCODE_PRINT:
         /* no-op */
         break;
      case OPCODE_BGNSUB:
         brw_save_label(p, inst->Comment, p->nr_insn);
         break;
      case OPCODE_ENDSUB:
         /* no-op */
         break;
      default:
	 _mesa_problem(NULL, "Unsupported opcode %i (%s) in vertex shader",
                       inst->Opcode, inst->Opcode < MAX_OPCODE ?
				    _mesa_opcode_string(inst->Opcode) :
				    "unknown");
      }

      /* Set the predication update on the last instruction of the native
       * instruction sequence.
       *
       * This would be problematic if it was set on a math instruction,
       * but that shouldn't be the case with the current GLSL compiler.
       */
      if (inst->CondUpdate) {
	 struct brw_instruction *hw_insn = &p->store[p->nr_insn - 1];

	 assert(hw_insn->header.destreg__conditionalmod == 0);
	 hw_insn->header.destreg__conditionalmod = BRW_CONDITIONAL_NZ;
      }

      if ((inst->DstReg.File == PROGRAM_OUTPUT)
          && (inst->DstReg.Index != VERT_RESULT_HPOS)
          && c->output_regs[inst->DstReg.Index].used_in_src) {
         brw_MOV(p, get_dst(c, inst->DstReg), dst);
      }

      /* Result color clamping.
       *
       * When destination register is an output register and
       * it's primary/secondary front/back color, we have to clamp
       * the result to [0,1]. This is done by enabling the
       * saturation bit for the last instruction.
       *
       * We don't use brw_set_saturate() as it modifies
       * p->current->header.saturate, which affects all the subsequent
       * instructions. Instead, we directly modify the header
       * of the last (already stored) instruction.
       */
      if (inst->DstReg.File == PROGRAM_OUTPUT &&
	  c->key.clamp_vertex_color) {
         if ((inst->DstReg.Index == VERT_RESULT_COL0)
             || (inst->DstReg.Index == VERT_RESULT_COL1)
             || (inst->DstReg.Index == VERT_RESULT_BFC0)
             || (inst->DstReg.Index == VERT_RESULT_BFC1)) {
            p->store[p->nr_insn-1].header.saturate = 1;
         }
      }

      if (inst->DstReg.RelAddr) {
	 assert(inst->DstReg.File == PROGRAM_TEMPORARY||
		inst->DstReg.File == PROGRAM_OUTPUT);
	 move_to_reladdr_dst(c, inst, dst);
      }

      release_tmps(c);
   }

   brw_resolve_cals(p);
   brw_set_uip_jip(p);

   brw_optimize(p);

   if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
      int i;

      printf("vs-native:\n");
      for (i = 0; i < p->nr_insn; i++)
	 brw_disasm(stdout, &p->store[i], intel->gen);
      printf("\n");
   }
}