summaryrefslogtreecommitdiff
path: root/src/gallium/auxiliary/draw/draw_llvm.c
blob: 6e1fb407c53b26a3de049e056f18e4484e4a3cff (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
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
/**************************************************************************
 *
 * Copyright 2010 VMware, Inc.
 * All Rights Reserved.
 *
 * 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, sub license, 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 NON-INFRINGEMENT.
 * IN NO EVENT SHALL VMWARE 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.
 *
 **************************************************************************/

#include "draw_llvm.h"

#include "draw_context.h"
#include "draw_vs.h"
#include "draw_gs.h"

#include "gallivm/lp_bld_arit.h"
#include "gallivm/lp_bld_arit_overflow.h"
#include "gallivm/lp_bld_logic.h"
#include "gallivm/lp_bld_const.h"
#include "gallivm/lp_bld_swizzle.h"
#include "gallivm/lp_bld_struct.h"
#include "gallivm/lp_bld_type.h"
#include "gallivm/lp_bld_flow.h"
#include "gallivm/lp_bld_debug.h"
#include "gallivm/lp_bld_tgsi.h"
#include "gallivm/lp_bld_printf.h"
#include "gallivm/lp_bld_intr.h"
#include "gallivm/lp_bld_init.h"
#include "gallivm/lp_bld_type.h"
#include "gallivm/lp_bld_pack.h"
#include "gallivm/lp_bld_format.h"

#include "tgsi/tgsi_exec.h"
#include "tgsi/tgsi_dump.h"

#include "util/u_math.h"
#include "util/u_pointer.h"
#include "util/u_string.h"
#include "util/simple_list.h"


#define DEBUG_STORE 0


static void
draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *var,
                   boolean elts);


struct draw_gs_llvm_iface {
   struct lp_build_tgsi_gs_iface base;

   struct draw_gs_llvm_variant *variant;
   LLVMValueRef input;
};

static INLINE const struct draw_gs_llvm_iface *
draw_gs_llvm_iface(const struct lp_build_tgsi_gs_iface *iface)
{
   return (const struct draw_gs_llvm_iface *)iface;
}

/**
 * Create LLVM type for draw_vertex_buffer.
 */
static LLVMTypeRef
create_jit_dvbuffer_type(struct gallivm_state *gallivm,
                         const char *struct_name)
{
   LLVMTargetDataRef target = gallivm->target;
   LLVMTypeRef dvbuffer_type;
   LLVMTypeRef elem_types[DRAW_JIT_DVBUFFER_NUM_FIELDS];
   LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context);

   elem_types[DRAW_JIT_DVBUFFER_MAP] =
      LLVMPointerType(LLVMIntTypeInContext(gallivm->context, 8), 0);
   elem_types[DRAW_JIT_DVBUFFER_SIZE] = int32_type;

   dvbuffer_type = LLVMStructTypeInContext(gallivm->context, elem_types,
                                           Elements(elem_types), 0);

   LP_CHECK_MEMBER_OFFSET(struct draw_vertex_buffer, map,
                          target, dvbuffer_type,
                          DRAW_JIT_DVBUFFER_MAP);
   LP_CHECK_MEMBER_OFFSET(struct draw_vertex_buffer, size,
                          target, dvbuffer_type,
                          DRAW_JIT_DVBUFFER_SIZE);

   return dvbuffer_type;
}

/**
 * Create LLVM type for struct draw_jit_texture
 */
static LLVMTypeRef
create_jit_texture_type(struct gallivm_state *gallivm, const char *struct_name)
{
   LLVMTargetDataRef target = gallivm->target;
   LLVMTypeRef texture_type;
   LLVMTypeRef elem_types[DRAW_JIT_TEXTURE_NUM_FIELDS];
   LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context);

   elem_types[DRAW_JIT_TEXTURE_WIDTH]  =
   elem_types[DRAW_JIT_TEXTURE_HEIGHT] =
   elem_types[DRAW_JIT_TEXTURE_DEPTH] =
   elem_types[DRAW_JIT_TEXTURE_FIRST_LEVEL] =
   elem_types[DRAW_JIT_TEXTURE_LAST_LEVEL] = int32_type;
   elem_types[DRAW_JIT_TEXTURE_BASE] =
      LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0);
   elem_types[DRAW_JIT_TEXTURE_ROW_STRIDE] =
   elem_types[DRAW_JIT_TEXTURE_IMG_STRIDE] =
   elem_types[DRAW_JIT_TEXTURE_MIP_OFFSETS] =
      LLVMArrayType(int32_type, PIPE_MAX_TEXTURE_LEVELS);

   texture_type = LLVMStructTypeInContext(gallivm->context, elem_types,
                                          Elements(elem_types), 0);

   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, width,
                          target, texture_type,
                          DRAW_JIT_TEXTURE_WIDTH);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, height,
                          target, texture_type,
                          DRAW_JIT_TEXTURE_HEIGHT);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, depth,
                          target, texture_type,
                          DRAW_JIT_TEXTURE_DEPTH);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, first_level,
                          target, texture_type,
                          DRAW_JIT_TEXTURE_FIRST_LEVEL);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, last_level,
                          target, texture_type,
                          DRAW_JIT_TEXTURE_LAST_LEVEL);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, base,
                          target, texture_type,
                          DRAW_JIT_TEXTURE_BASE);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, row_stride,
                          target, texture_type,
                          DRAW_JIT_TEXTURE_ROW_STRIDE);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, img_stride,
                          target, texture_type,
                          DRAW_JIT_TEXTURE_IMG_STRIDE);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, mip_offsets,
                          target, texture_type,
                          DRAW_JIT_TEXTURE_MIP_OFFSETS);

   LP_CHECK_STRUCT_SIZE(struct draw_jit_texture, target, texture_type);

   return texture_type;
}


/**
 * Create LLVM type for struct draw_jit_sampler
 */
static LLVMTypeRef
create_jit_sampler_type(struct gallivm_state *gallivm, const char *struct_name)
{
   LLVMTargetDataRef target = gallivm->target;
   LLVMTypeRef sampler_type;
   LLVMTypeRef elem_types[DRAW_JIT_SAMPLER_NUM_FIELDS];

   elem_types[DRAW_JIT_SAMPLER_MIN_LOD] =
   elem_types[DRAW_JIT_SAMPLER_MAX_LOD] =
   elem_types[DRAW_JIT_SAMPLER_LOD_BIAS] = LLVMFloatTypeInContext(gallivm->context);
   elem_types[DRAW_JIT_SAMPLER_BORDER_COLOR] =
      LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4);

   sampler_type = LLVMStructTypeInContext(gallivm->context, elem_types,
                                          Elements(elem_types), 0);

   LP_CHECK_MEMBER_OFFSET(struct draw_jit_sampler, min_lod,
                          target, sampler_type,
                          DRAW_JIT_SAMPLER_MIN_LOD);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_sampler, max_lod,
                          target, sampler_type,
                          DRAW_JIT_SAMPLER_MAX_LOD);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_sampler, lod_bias,
                          target, sampler_type,
                          DRAW_JIT_SAMPLER_LOD_BIAS);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_sampler, border_color,
                          target, sampler_type,
                          DRAW_JIT_SAMPLER_BORDER_COLOR);

   LP_CHECK_STRUCT_SIZE(struct draw_jit_sampler, target, sampler_type);

   return sampler_type;
}


/**
 * Create LLVM type for struct draw_jit_context
 */
static LLVMTypeRef
create_jit_context_type(struct gallivm_state *gallivm,
                        LLVMTypeRef texture_type, LLVMTypeRef sampler_type,
                        const char *struct_name)
{
   LLVMTargetDataRef target = gallivm->target;
   LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
   LLVMTypeRef int_type = LLVMInt32TypeInContext(gallivm->context);
   LLVMTypeRef elem_types[DRAW_JIT_CTX_NUM_FIELDS];
   LLVMTypeRef context_type;

   elem_types[0] = LLVMArrayType(LLVMPointerType(float_type, 0), /* vs_constants */
                                 LP_MAX_TGSI_CONST_BUFFERS);
   elem_types[1] = LLVMArrayType(int_type, /* num_vs_constants */
                                 LP_MAX_TGSI_CONST_BUFFERS);
   elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4),
                                                 DRAW_TOTAL_CLIP_PLANES), 0);
   elem_types[3] = LLVMPointerType(float_type, 0); /* viewports */
   elem_types[4] = LLVMArrayType(texture_type,
                                 PIPE_MAX_SHADER_SAMPLER_VIEWS); /* textures */
   elem_types[5] = LLVMArrayType(sampler_type,
                                 PIPE_MAX_SAMPLERS); /* samplers */
   context_type = LLVMStructTypeInContext(gallivm->context, elem_types,
                                          Elements(elem_types), 0);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, vs_constants,
                          target, context_type, DRAW_JIT_CTX_CONSTANTS);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, num_vs_constants,
                          target, context_type, DRAW_JIT_CTX_NUM_CONSTANTS);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes,
                          target, context_type, DRAW_JIT_CTX_PLANES);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, viewports,
                          target, context_type, DRAW_JIT_CTX_VIEWPORT);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures,
                          target, context_type,
                          DRAW_JIT_CTX_TEXTURES);
   LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, samplers,
                          target, context_type,
                          DRAW_JIT_CTX_SAMPLERS);
   LP_CHECK_STRUCT_SIZE(struct draw_jit_context,
                        target, context_type);

   return context_type;
}


/**
 * Create LLVM type for struct draw_gs_jit_context
 */
static LLVMTypeRef
create_gs_jit_context_type(struct gallivm_state *gallivm,
                           unsigned vector_length,
                           LLVMTypeRef texture_type, LLVMTypeRef sampler_type,
                           const char *struct_name)
{
   LLVMTargetDataRef target = gallivm->target;
   LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
   LLVMTypeRef int_type = LLVMInt32TypeInContext(gallivm->context);
   LLVMTypeRef elem_types[DRAW_GS_JIT_CTX_NUM_FIELDS];
   LLVMTypeRef context_type;

   elem_types[0] = LLVMArrayType(LLVMPointerType(float_type, 0), /* constants */
                                 LP_MAX_TGSI_CONST_BUFFERS);
   elem_types[1] = LLVMArrayType(int_type, /* num_constants */
                                 LP_MAX_TGSI_CONST_BUFFERS);
   elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(float_type, 4),
                                                 DRAW_TOTAL_CLIP_PLANES), 0);
   elem_types[3] = LLVMPointerType(float_type, 0); /* viewports */

   elem_types[4] = LLVMArrayType(texture_type,
                                 PIPE_MAX_SHADER_SAMPLER_VIEWS); /* textures */
   elem_types[5] = LLVMArrayType(sampler_type,
                                 PIPE_MAX_SAMPLERS); /* samplers */
   
   elem_types[6] = LLVMPointerType(LLVMPointerType(int_type, 0), 0);
   elem_types[7] = LLVMPointerType(LLVMVectorType(int_type,
                                                  vector_length), 0);
   elem_types[8] = LLVMPointerType(LLVMVectorType(int_type,
                                                  vector_length), 0);

   context_type = LLVMStructTypeInContext(gallivm->context, elem_types,
                                          Elements(elem_types), 0);

   LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, constants,
                          target, context_type, DRAW_GS_JIT_CTX_CONSTANTS);
   LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, num_constants,
                          target, context_type, DRAW_GS_JIT_CTX_NUM_CONSTANTS);
   LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, planes,
                          target, context_type, DRAW_GS_JIT_CTX_PLANES);
   LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, viewports,
                          target, context_type, DRAW_GS_JIT_CTX_VIEWPORT);
   LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, textures,
                          target, context_type,
                          DRAW_GS_JIT_CTX_TEXTURES);
   LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, samplers,
                          target, context_type,
                          DRAW_GS_JIT_CTX_SAMPLERS);
   LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, prim_lengths,
                          target, context_type,
                          DRAW_GS_JIT_CTX_PRIM_LENGTHS);
   LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, emitted_vertices,
                          target, context_type,
                          DRAW_GS_JIT_CTX_EMITTED_VERTICES);
   LP_CHECK_MEMBER_OFFSET(struct draw_gs_jit_context, emitted_prims,
                          target, context_type,
                          DRAW_GS_JIT_CTX_EMITTED_PRIMS);
   LP_CHECK_STRUCT_SIZE(struct draw_gs_jit_context,
                        target, context_type);

   return context_type;
}


static LLVMTypeRef
create_gs_jit_input_type(struct gallivm_state *gallivm)
{
   LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
   LLVMTypeRef input_array;

   input_array = LLVMVectorType(float_type, TGSI_NUM_CHANNELS); /* num primitives */
   input_array = LLVMArrayType(input_array, TGSI_NUM_CHANNELS); /* num channels */
   input_array = LLVMArrayType(input_array, PIPE_MAX_SHADER_INPUTS); /* num attrs per vertex */
   input_array = LLVMPointerType(input_array, 0); /* num vertices per prim */

   return input_array;
}

/**
 * Create LLVM type for struct pipe_vertex_buffer
 */
static LLVMTypeRef
create_jit_vertex_buffer_type(struct gallivm_state *gallivm,
                              const char *struct_name)
{
   LLVMTargetDataRef target = gallivm->target;
   LLVMTypeRef elem_types[4];
   LLVMTypeRef vb_type;

   elem_types[0] =
   elem_types[1] = LLVMInt32TypeInContext(gallivm->context);
   elem_types[2] =
   elem_types[3] = LLVMPointerType(LLVMInt8TypeInContext(gallivm->context), 0);

   vb_type = LLVMStructTypeInContext(gallivm->context, elem_types,
                                     Elements(elem_types), 0);

   LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, stride,
                          target, vb_type, 0);
   LP_CHECK_MEMBER_OFFSET(struct pipe_vertex_buffer, buffer_offset,
                          target, vb_type, 1);

   LP_CHECK_STRUCT_SIZE(struct pipe_vertex_buffer, target, vb_type);

   return vb_type;
}


/**
 * Create LLVM type for struct vertex_header;
 */
static LLVMTypeRef
create_jit_vertex_header(struct gallivm_state *gallivm, int data_elems)
{
   LLVMTargetDataRef target = gallivm->target;
   LLVMTypeRef elem_types[4];
   LLVMTypeRef vertex_header;
   char struct_name[24];

   util_snprintf(struct_name, 23, "vertex_header%d", data_elems);

   elem_types[DRAW_JIT_VERTEX_VERTEX_ID]  = LLVMIntTypeInContext(gallivm->context, 32);
   elem_types[DRAW_JIT_VERTEX_CLIP]  = LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4);
   elem_types[DRAW_JIT_VERTEX_PRE_CLIP_POS]  = LLVMArrayType(LLVMFloatTypeInContext(gallivm->context), 4);
   elem_types[DRAW_JIT_VERTEX_DATA]  = LLVMArrayType(elem_types[1], data_elems);

   vertex_header = LLVMStructTypeInContext(gallivm->context, elem_types,
                                           Elements(elem_types), 0);

   /* these are bit-fields and we can't take address of them
      LP_CHECK_MEMBER_OFFSET(struct vertex_header, clipmask,
      target, vertex_header,
      DRAW_JIT_VERTEX_CLIPMASK);
      LP_CHECK_MEMBER_OFFSET(struct vertex_header, edgeflag,
      target, vertex_header,
      DRAW_JIT_VERTEX_EDGEFLAG);
      LP_CHECK_MEMBER_OFFSET(struct vertex_header, pad,
      target, vertex_header,
      DRAW_JIT_VERTEX_PAD);
      LP_CHECK_MEMBER_OFFSET(struct vertex_header, vertex_id,
      target, vertex_header,
      DRAW_JIT_VERTEX_VERTEX_ID);
   */
   LP_CHECK_MEMBER_OFFSET(struct vertex_header, clip,
                          target, vertex_header,
                          DRAW_JIT_VERTEX_CLIP);
   LP_CHECK_MEMBER_OFFSET(struct vertex_header, pre_clip_pos,
                          target, vertex_header,
                          DRAW_JIT_VERTEX_PRE_CLIP_POS);
   LP_CHECK_MEMBER_OFFSET(struct vertex_header, data,
                          target, vertex_header,
                          DRAW_JIT_VERTEX_DATA);

   assert(LLVMABISizeOfType(target, vertex_header) ==
          offsetof(struct vertex_header, data[data_elems]));

   return vertex_header;
}


/**
 * Create LLVM types for various structures.
 */
static void
create_jit_types(struct draw_llvm_variant *variant)
{
   struct gallivm_state *gallivm = variant->gallivm;
   LLVMTypeRef texture_type, sampler_type, context_type, buffer_type,
      vb_type;

   texture_type = create_jit_texture_type(gallivm, "texture");
   sampler_type = create_jit_sampler_type(gallivm, "sampler");

   context_type = create_jit_context_type(gallivm, texture_type, sampler_type,
                                          "draw_jit_context");
   variant->context_ptr_type = LLVMPointerType(context_type, 0);

   buffer_type = create_jit_dvbuffer_type(gallivm, "draw_vertex_buffer");
   variant->buffer_ptr_type = LLVMPointerType(buffer_type, 0);
   
   vb_type = create_jit_vertex_buffer_type(gallivm, "pipe_vertex_buffer");
   variant->vb_ptr_type = LLVMPointerType(vb_type, 0);
}


static LLVMTypeRef
get_context_ptr_type(struct draw_llvm_variant *variant)
{
   if (!variant->context_ptr_type)
      create_jit_types(variant);
   return variant->context_ptr_type;
}


static LLVMTypeRef
get_buffer_ptr_type(struct draw_llvm_variant *variant)
{
   if (!variant->buffer_ptr_type)
      create_jit_types(variant);
   return variant->buffer_ptr_type;
}


static LLVMTypeRef
get_vb_ptr_type(struct draw_llvm_variant *variant)
{
   if (!variant->vb_ptr_type)
      create_jit_types(variant);
   return variant->vb_ptr_type;
}

static LLVMTypeRef
get_vertex_header_ptr_type(struct draw_llvm_variant *variant)
{
   if (!variant->vertex_header_ptr_type)
      create_jit_types(variant);
   return variant->vertex_header_ptr_type;
}


/**
 * Create per-context LLVM info.
 */
struct draw_llvm *
draw_llvm_create(struct draw_context *draw, LLVMContextRef context)
{
   struct draw_llvm *llvm;

   if (!lp_build_init())
      return NULL;

   llvm = CALLOC_STRUCT( draw_llvm );
   if (!llvm)
      return NULL;

   llvm->draw = draw;

   llvm->context = context;
   if (!llvm->context) {
      llvm->context = LLVMContextCreate();
      llvm->context_owned = true;
   }
   if (!llvm->context)
      goto fail;

   llvm->nr_variants = 0;
   make_empty_list(&llvm->vs_variants_list);

   llvm->nr_gs_variants = 0;
   make_empty_list(&llvm->gs_variants_list);

   return llvm;

fail:
   draw_llvm_destroy(llvm);
   return NULL;
}


/**
 * Free per-context LLVM info.
 */
void
draw_llvm_destroy(struct draw_llvm *llvm)
{
   if (llvm->context_owned)
      LLVMContextDispose(llvm->context);
   llvm->context = NULL;

   /* XXX free other draw_llvm data? */
   FREE(llvm);
}


/**
 * Create LLVM-generated code for a vertex shader.
 */
struct draw_llvm_variant *
draw_llvm_create_variant(struct draw_llvm *llvm,
                         unsigned num_inputs,
                         const struct draw_llvm_variant_key *key)
{
   struct draw_llvm_variant *variant;
   struct llvm_vertex_shader *shader =
      llvm_vertex_shader(llvm->draw->vs.vertex_shader);
   LLVMTypeRef vertex_header;
   char module_name[64];

   variant = MALLOC(sizeof *variant +
                    shader->variant_key_size -
                    sizeof variant->key);
   if (variant == NULL)
      return NULL;

   variant->llvm = llvm;
   variant->shader = shader;

   util_snprintf(module_name, sizeof(module_name), "draw_llvm_vs_variant%u",
                 variant->shader->variants_cached);

   variant->gallivm = gallivm_create(module_name, llvm->context);

   create_jit_types(variant);

   memcpy(&variant->key, key, shader->variant_key_size);

   vertex_header = create_jit_vertex_header(variant->gallivm, num_inputs);

   variant->vertex_header_ptr_type = LLVMPointerType(vertex_header, 0);

   draw_llvm_generate(llvm, variant, FALSE);  /* linear */
   draw_llvm_generate(llvm, variant, TRUE);   /* elts */

   gallivm_compile_module(variant->gallivm);

   variant->jit_func = (draw_jit_vert_func)
         gallivm_jit_function(variant->gallivm, variant->function);

   variant->jit_func_elts = (draw_jit_vert_func_elts)
         gallivm_jit_function(variant->gallivm, variant->function_elts);

   gallivm_free_ir(variant->gallivm);

   variant->list_item_global.base = variant;
   variant->list_item_local.base = variant;
   /*variant->no = */shader->variants_created++;
   variant->list_item_global.base = variant;

   return variant;
}


static void
generate_vs(struct draw_llvm_variant *variant,
            LLVMBuilderRef builder,
            struct lp_type vs_type,
            LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
            const LLVMValueRef (*inputs)[TGSI_NUM_CHANNELS],
            const struct lp_bld_tgsi_system_values *system_values,
            LLVMValueRef context_ptr,
            struct lp_build_sampler_soa *draw_sampler,
            boolean clamp_vertex_color)
{
   struct draw_llvm *llvm = variant->llvm;
   const struct tgsi_token *tokens = llvm->draw->vs.vertex_shader->state.tokens;
   LLVMValueRef consts_ptr =
      draw_jit_context_vs_constants(variant->gallivm, context_ptr);
   LLVMValueRef num_consts_ptr =
      draw_jit_context_num_vs_constants(variant->gallivm, context_ptr);

   if (gallivm_debug & (GALLIVM_DEBUG_TGSI | GALLIVM_DEBUG_IR)) {
      tgsi_dump(tokens, 0);
      draw_llvm_dump_variant_key(&variant->key);
   }

   lp_build_tgsi_soa(variant->gallivm,
                     tokens,
                     vs_type,
                     NULL /*struct lp_build_mask_context *mask*/,
                     consts_ptr,
                     num_consts_ptr,
                     system_values,
                     inputs,
                     outputs,
                     draw_sampler,
                     &llvm->draw->vs.vertex_shader->info,
                     NULL);

   {
      LLVMValueRef out;
      unsigned chan, attrib;
      struct lp_build_context bld;
      struct tgsi_shader_info* info = &llvm->draw->vs.vertex_shader->info;
      lp_build_context_init(&bld, variant->gallivm, vs_type);

      for (attrib = 0; attrib < info->num_outputs; ++attrib) {
         for (chan = 0; chan < TGSI_NUM_CHANNELS; ++chan) {
            if (outputs[attrib][chan]) {
               switch (info->output_semantic_name[attrib]) {
               case TGSI_SEMANTIC_COLOR:
               case TGSI_SEMANTIC_BCOLOR:
                  if (clamp_vertex_color) {
                     out = LLVMBuildLoad(builder, outputs[attrib][chan], "");
                     out = lp_build_clamp(&bld, out, bld.zero, bld.one);
                     LLVMBuildStore(builder, out, outputs[attrib][chan]);
                  }
                  break;
               }
            }
         }
      }
   }
}

static void
generate_fetch(struct gallivm_state *gallivm,
               struct draw_context *draw,
               LLVMValueRef vbuffers_ptr,
               LLVMValueRef *res,
               struct pipe_vertex_element *velem,
               LLVMValueRef vbuf,
               LLVMValueRef index,
               LLVMValueRef instance_id,
               LLVMValueRef start_instance)
{
   const struct util_format_description *format_desc =
      util_format_description(velem->src_format);
   LLVMValueRef zero = LLVMConstNull(LLVMInt32TypeInContext(gallivm->context));
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef indices =
      LLVMConstInt(LLVMInt64TypeInContext(gallivm->context),
                   velem->vertex_buffer_index, 0);
   LLVMValueRef vbuffer_ptr = LLVMBuildGEP(builder, vbuffers_ptr,
                                           &indices, 1, "");
   LLVMValueRef vb_stride = draw_jit_vbuffer_stride(gallivm, vbuf);
   LLVMValueRef vb_buffer_offset = draw_jit_vbuffer_offset(gallivm, vbuf);
   LLVMValueRef map_ptr = draw_jit_dvbuffer_map(gallivm, vbuffer_ptr);
   LLVMValueRef buffer_size = draw_jit_dvbuffer_size(gallivm, vbuffer_ptr);
   LLVMValueRef stride;
   LLVMValueRef buffer_overflowed;
   LLVMValueRef needed_buffer_size;
   LLVMValueRef temp_ptr =
      lp_build_alloca(gallivm,
                      lp_build_vec_type(gallivm, lp_float32_vec4_type()), "");
   LLVMValueRef ofbit = NULL;
   struct lp_build_if_state if_ctx;

   if (velem->instance_divisor) {
      /* Index is equal to the start instance plus the number of current 
       * instance divided by the divisor. In this case we compute it as:
       * index = start_instance + (instance_id  / divisor)
       */
      LLVMValueRef current_instance;
      current_instance = LLVMBuildUDiv(builder, instance_id,
                                       lp_build_const_int32(gallivm, velem->instance_divisor),
                                       "instance_divisor");
      index = lp_build_uadd_overflow(gallivm, start_instance,
                                     current_instance, &ofbit);
   }

   stride = lp_build_umul_overflow(gallivm, vb_stride, index, &ofbit);
   stride = lp_build_uadd_overflow(gallivm, stride, vb_buffer_offset, &ofbit);
   stride = lp_build_uadd_overflow(
      gallivm, stride,
      lp_build_const_int32(gallivm, velem->src_offset), &ofbit);
   needed_buffer_size = lp_build_uadd_overflow(
      gallivm, stride,
      lp_build_const_int32(gallivm,
                           util_format_get_blocksize(velem->src_format)),
      &ofbit);

   buffer_overflowed = LLVMBuildICmp(builder, LLVMIntUGT,
                                     needed_buffer_size, buffer_size,
                                     "buffer_overflowed");
   buffer_overflowed = LLVMBuildOr(builder, buffer_overflowed, ofbit, "");
#if 0
   lp_build_printf(gallivm, "vbuf index = %u, vb_stride is %u\n",
                   index, vb_stride);
   lp_build_printf(gallivm, "   vb_buffer_offset = %u, src_offset is %u\n",
                   vb_buffer_offset,
                   lp_build_const_int32(gallivm, velem->src_offset));
   lp_build_print_value(gallivm, "   blocksize = ",
                        lp_build_const_int32(
                           gallivm,
                           util_format_get_blocksize(velem->src_format)));
   lp_build_printf(gallivm, "   instance_id = %u\n", instance_id);
   lp_build_printf(gallivm, "   stride = %u\n", stride);
   lp_build_printf(gallivm, "   buffer size = %u\n", buffer_size);
   lp_build_printf(gallivm, "   needed_buffer_size = %u\n", needed_buffer_size);
   lp_build_print_value(gallivm, "   buffer overflowed = ", buffer_overflowed);
#endif

   lp_build_if(&if_ctx, gallivm, buffer_overflowed);
   {
      LLVMValueRef val =
         lp_build_const_vec(gallivm, lp_float32_vec4_type(), 0);
      LLVMBuildStore(builder, val, temp_ptr);
   }
   lp_build_else(&if_ctx);
   {
      LLVMValueRef val;
      map_ptr = LLVMBuildGEP(builder, map_ptr, &stride, 1, "");

      val = lp_build_fetch_rgba_aos(gallivm,
                                    format_desc,
                                    lp_float32_vec4_type(),
                                    FALSE,
                                    map_ptr,
                                    zero, zero, zero);
      LLVMBuildStore(builder, val, temp_ptr);
   }
   lp_build_endif(&if_ctx);

   *res = LLVMBuildLoad(builder, temp_ptr, "aos");
}

static void
convert_to_soa(struct gallivm_state *gallivm,
               LLVMValueRef (*src_aos)[LP_MAX_VECTOR_WIDTH / 32],
               LLVMValueRef (*dst_soa)[TGSI_NUM_CHANNELS],
               unsigned num_attribs, const struct lp_type soa_type)
{
   unsigned i, j, k;
   struct lp_type aos_channel_type = soa_type;

   debug_assert(TGSI_NUM_CHANNELS == 4);
   debug_assert((soa_type.length % TGSI_NUM_CHANNELS) == 0);

   aos_channel_type.length >>= 1;

   for (i = 0; i < num_attribs; ++i) {
      LLVMValueRef aos_channels[TGSI_NUM_CHANNELS];
      unsigned pixels_per_channel = soa_type.length / TGSI_NUM_CHANNELS;

      for (j = 0; j < TGSI_NUM_CHANNELS; ++j) {
         LLVMValueRef channel[LP_MAX_VECTOR_LENGTH] = { 0 };

         assert(pixels_per_channel <= LP_MAX_VECTOR_LENGTH);

         for (k = 0; k < pixels_per_channel; ++k) {
            channel[k] = src_aos[i][j + TGSI_NUM_CHANNELS * k];
         }

         aos_channels[j] = lp_build_concat(gallivm, channel, aos_channel_type, pixels_per_channel);
      }

      lp_build_transpose_aos(gallivm, soa_type, aos_channels, dst_soa[i]);
   }
}


static void
store_aos(struct gallivm_state *gallivm,
          LLVMValueRef io_ptr,
          LLVMValueRef index,
          LLVMValueRef value)
{
   LLVMTypeRef data_ptr_type = LLVMPointerType(lp_build_vec_type(gallivm, lp_float32_vec4_type()), 0);
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef data_ptr = draw_jit_header_data(gallivm, io_ptr);
   LLVMValueRef indices[3];

   indices[0] = lp_build_const_int32(gallivm, 0);
   indices[1] = index;
   indices[2] = lp_build_const_int32(gallivm, 0);

   data_ptr = LLVMBuildGEP(builder, data_ptr, indices, 3, "");
   data_ptr = LLVMBuildPointerCast(builder, data_ptr, data_ptr_type, "");

#if DEBUG_STORE
   lp_build_printf(gallivm, "    ---- %p storing attribute %d (io = %p)\n", data_ptr, index, io_ptr);
#endif

   /* Unaligned store due to the vertex header */
   lp_set_store_alignment(LLVMBuildStore(builder, value, data_ptr), sizeof(float));
}

/**
 * Adjust the mask to architecture endianess. The mask will the store in struct:
 *
 * struct vertex_header {
 *    unsigned clipmask:DRAW_TOTAL_CLIP_PLANES;
 *    unsigned edgeflag:1;
 *    unsigned have_clipdist:1;
 *    unsigned vertex_id:16;
 *    [...]
 * }
 *
 * On little-endian machine nothing needs to done, however on bit-endian machine
 * the mask's fields need to be adjusted with the algorithm:
 *
 * uint32_t reverse (uint32_t x)
 * {
 *   return (x >> 16) |              // vertex_id
 *          ((x & 0x3fff) << 18) |   // clipmask
 *          ((x & 0x4000) << 3) |    // have_clipdist
 *          ((x & 0x8000) << 1);     // edgeflag
 * }
 */
static LLVMValueRef
adjust_mask(struct gallivm_state *gallivm,
            LLVMValueRef mask)
{
#ifdef PIPE_ARCH_BIG_ENDIAN
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef vertex_id;
   LLVMValueRef clipmask;
   LLVMValueRef have_clipdist;
   LLVMValueRef edgeflag;

   vertex_id = LLVMBuildLShr(builder, mask, lp_build_const_int32(gallivm, 16), "");
   clipmask  = LLVMBuildAnd(builder, mask, lp_build_const_int32(gallivm, 0x3fff), "");
   clipmask  = LLVMBuildShl(builder, clipmask, lp_build_const_int32(gallivm, 18), "");
   have_clipdist = LLVMBuildAnd(builder, mask, lp_build_const_int32(gallivm, 0x4000), "");
   have_clipdist = LLVMBuildShl(builder, have_clipdist, lp_build_const_int32(gallivm, 3), "");
   edgeflag = LLVMBuildAnd(builder, mask, lp_build_const_int32(gallivm, 0x8000), "");
   edgeflag = LLVMBuildShl(builder, edgeflag, lp_build_const_int32(gallivm, 1), "");

   mask = LLVMBuildOr(builder, vertex_id, clipmask, "");
   mask = LLVMBuildOr(builder, mask, have_clipdist, "");
   mask = LLVMBuildOr(builder, mask, edgeflag, "");
#endif
   return mask;
}

static void
store_aos_array(struct gallivm_state *gallivm,
                struct lp_type soa_type,
                LLVMValueRef io_ptr,
                LLVMValueRef *indices,
                LLVMValueRef* aos,
                int attrib,
                int num_outputs,
                LLVMValueRef clipmask,
                boolean have_clipdist)
{
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef attr_index = lp_build_const_int32(gallivm, attrib);
   LLVMValueRef inds[LP_MAX_VECTOR_WIDTH / 32];
   LLVMValueRef linear_inds[LP_MAX_VECTOR_WIDTH / 32];
   LLVMValueRef io_ptrs[LP_MAX_VECTOR_WIDTH / 32];
   int vector_length = soa_type.length;
   int i;

   debug_assert(TGSI_NUM_CHANNELS == 4);

   for (i = 0; i < vector_length; i++) {
      linear_inds[i] = lp_build_const_int32(gallivm, i);
      if (indices) {
         inds[i] = indices[i];
      } else {
         inds[i] = linear_inds[i];
      }
      io_ptrs[i] = LLVMBuildGEP(builder, io_ptr, &inds[i], 1, "");
   }

   if (attrib == 0) {
      /* store vertex header for each of the n vertices */
      LLVMValueRef val, cliptmp;
      int vertex_id_pad_edgeflag;

      /* If this assertion fails, it means we need to update the bit twidding
       * code here.  See struct vertex_header in draw_private.h.
       */
      assert(DRAW_TOTAL_CLIP_PLANES==14);
      /* initialize vertex id:16 = 0xffff, have_clipdist:1 = 0, edgeflag:1 = 1 */
      vertex_id_pad_edgeflag = (0xffff << 16) | (1 << DRAW_TOTAL_CLIP_PLANES);
      if (have_clipdist)
         vertex_id_pad_edgeflag |= 1 << (DRAW_TOTAL_CLIP_PLANES+1);
      val = lp_build_const_int_vec(gallivm, lp_int_type(soa_type), vertex_id_pad_edgeflag);
      /* OR with the clipmask */
      cliptmp = LLVMBuildOr(builder, val, clipmask, "");
      for (i = 0; i < vector_length; i++) {
         LLVMValueRef id_ptr = draw_jit_header_id(gallivm, io_ptrs[i]);
         val = LLVMBuildExtractElement(builder, cliptmp, linear_inds[i], "");
         val = adjust_mask(gallivm, val);
#if DEBUG_STORE
         lp_build_printf(gallivm, "io = %p, index %d, clipmask = %x\n",
                         io_ptrs[i], inds[i], val);
#endif
         LLVMBuildStore(builder, val, id_ptr);
      }
   }

   /* store for each of the n vertices */
   for (i = 0; i < vector_length; i++) {
      store_aos(gallivm, io_ptrs[i], attr_index, aos[i]);
   }
}


static void
convert_to_aos(struct gallivm_state *gallivm,
               LLVMValueRef io,
               LLVMValueRef *indices,
               LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
               LLVMValueRef clipmask,
               int num_outputs,
               struct lp_type soa_type,
               boolean have_clipdist)
{
   LLVMBuilderRef builder = gallivm->builder;
   unsigned chan, attrib, i;

#if DEBUG_STORE
   lp_build_printf(gallivm, "   # storing begin\n");
#endif
   for (attrib = 0; attrib < num_outputs; ++attrib) {
      LLVMValueRef soa[TGSI_NUM_CHANNELS];
      LLVMValueRef aos[LP_MAX_VECTOR_WIDTH / 32];
      for (chan = 0; chan < TGSI_NUM_CHANNELS; ++chan) {
         if (outputs[attrib][chan]) {
            LLVMValueRef out = LLVMBuildLoad(builder, outputs[attrib][chan], "");
            lp_build_name(out, "output%u.%c", attrib, "xyzw"[chan]);
#if DEBUG_STORE
            lp_build_printf(gallivm, "output %d : %d ",
                            LLVMConstInt(LLVMInt32TypeInContext(gallivm->context),
                                         attrib, 0),
                            LLVMConstInt(LLVMInt32TypeInContext(gallivm->context),
                                         chan, 0));
            lp_build_print_value(gallivm, "val = ", out);
            {
               LLVMValueRef iv =
                  LLVMBuildBitCast(builder, out, lp_build_int_vec_type(gallivm, soa_type), "");
               
               lp_build_print_value(gallivm, "  ival = ", iv);
            }
#endif
            soa[chan] = out;
         }
         else {
            soa[chan] = 0;
         }
      }


      if (soa_type.length == TGSI_NUM_CHANNELS) {
         lp_build_transpose_aos(gallivm, soa_type, soa, aos);
      } else {
         lp_build_transpose_aos(gallivm, soa_type, soa, soa);

         for (i = 0; i < soa_type.length; ++i) {
            aos[i] = lp_build_extract_range(gallivm,
                                            soa[i % TGSI_NUM_CHANNELS],
                                            (i / TGSI_NUM_CHANNELS) * TGSI_NUM_CHANNELS,
                                            TGSI_NUM_CHANNELS);
         }
      }

      store_aos_array(gallivm,
                      soa_type,
                      io, indices,
                      aos,
                      attrib,
                      num_outputs,
                      clipmask, have_clipdist);
   }
#if DEBUG_STORE
   lp_build_printf(gallivm, "   # storing end\n");
#endif
}


/**
 * Stores original vertex positions in clip coordinates
 */
static void
store_clip(struct gallivm_state *gallivm,
           const struct lp_type vs_type,
           LLVMValueRef io_ptr,
           LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
           boolean pre_clip_pos, int idx)
{
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef soa[4];
   LLVMValueRef aos[LP_MAX_VECTOR_LENGTH];
   LLVMValueRef indices[2];
   LLVMValueRef io_ptrs[LP_MAX_VECTOR_WIDTH / 32];
   LLVMValueRef inds[LP_MAX_VECTOR_WIDTH / 32];
   LLVMValueRef clip_ptrs[LP_MAX_VECTOR_WIDTH / 32];
   LLVMTypeRef clip_ptr_type =
      LLVMPointerType(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context),
                                     4), 0);
   int i, j;

   indices[0] =
   indices[1] = lp_build_const_int32(gallivm, 0);

   for (i = 0; i < vs_type.length; i++) {
      inds[i] = lp_build_const_int32(gallivm, i);
      io_ptrs[i] = LLVMBuildGEP(builder, io_ptr, &inds[i], 1, "");
   }

   soa[0] = LLVMBuildLoad(builder, outputs[idx][0], ""); /*x0 x1 .. xn*/
   soa[1] = LLVMBuildLoad(builder, outputs[idx][1], ""); /*y0 y1 .. yn*/
   soa[2] = LLVMBuildLoad(builder, outputs[idx][2], ""); /*z0 z1 .. zn*/
   soa[3] = LLVMBuildLoad(builder, outputs[idx][3], ""); /*w0 w1 .. wn*/

   if (!pre_clip_pos) {
      for (i = 0; i < vs_type.length; i++) {
         clip_ptrs[i] = draw_jit_header_clip(gallivm, io_ptrs[i]);
      }
   } else {
      for (i = 0; i < vs_type.length; i++) {
         clip_ptrs[i] = draw_jit_header_pre_clip_pos(gallivm, io_ptrs[i]);
      }
   }

   lp_build_transpose_aos(gallivm, vs_type, soa, soa);
   for (i = 0; i < vs_type.length; ++i) {
      aos[i] = lp_build_extract_range(gallivm,
                                      soa[i % TGSI_NUM_CHANNELS],
                                      (i / TGSI_NUM_CHANNELS) * TGSI_NUM_CHANNELS,
                                      TGSI_NUM_CHANNELS);
   }

   for (j = 0; j < vs_type.length; j++) {
      LLVMValueRef clip_ptr;

      clip_ptr = LLVMBuildGEP(builder, clip_ptrs[j], indices, 2, "clipo");
      clip_ptr = LLVMBuildPointerCast(builder, clip_ptr, clip_ptr_type, "");

      /* Unaligned store */
      lp_set_store_alignment(LLVMBuildStore(builder, aos[j], clip_ptr), sizeof(float));
   }
}


/**
 * Transforms the outputs for viewport mapping
 */
static void
generate_viewport(struct draw_llvm_variant *variant,
                  LLVMBuilderRef builder,
                  struct lp_type vs_type,
                  LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
                  LLVMValueRef context_ptr)
{
   int i;
   struct gallivm_state *gallivm = variant->gallivm;
   struct lp_type f32_type = vs_type;
   const unsigned pos = variant->llvm->draw->vs.position_output;
   LLVMTypeRef vs_type_llvm = lp_build_vec_type(gallivm, vs_type);
   LLVMValueRef out3 = LLVMBuildLoad(builder, outputs[pos][3], ""); /*w0 w1 .. wn*/
   LLVMValueRef const1 = lp_build_const_vec(gallivm, f32_type, 1.0);       /*1.0 1.0 1.0 1.0*/
   LLVMValueRef vp_ptr = draw_jit_context_viewports(gallivm, context_ptr);

   /* We treat pipe_viewport_state as a float array */
   const int scale_index_offset = offsetof(struct pipe_viewport_state, scale) / sizeof(float);
   const int trans_index_offset = offsetof(struct pipe_viewport_state, translate) / sizeof(float);

   /* for 1/w convention*/
   out3 = LLVMBuildFDiv(builder, const1, out3, "");
   LLVMBuildStore(builder, out3, outputs[pos][3]);

   /* Viewport Mapping */
   for (i=0; i<3; i++) {
      LLVMValueRef out = LLVMBuildLoad(builder, outputs[pos][i], ""); /*x0 x1 .. xn*/
      LLVMValueRef scale;
      LLVMValueRef trans;
      LLVMValueRef scale_i;
      LLVMValueRef trans_i;
      LLVMValueRef index;

      index = lp_build_const_int32(gallivm, i + scale_index_offset);
      scale_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");

      index = lp_build_const_int32(gallivm, i + trans_index_offset);
      trans_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");

      scale = lp_build_broadcast(gallivm, vs_type_llvm,
                                 LLVMBuildLoad(builder, scale_i, "scale"));
      trans = lp_build_broadcast(gallivm, vs_type_llvm,
                                 LLVMBuildLoad(builder, trans_i, "trans"));

      /* divide by w */
      out = LLVMBuildFMul(builder, out, out3, "");
      /* mult by scale */
      out = LLVMBuildFMul(builder, out, scale, "");
      /* add translation */
      out = LLVMBuildFAdd(builder, out, trans, "");

      /* store transformed outputs */
      LLVMBuildStore(builder, out, outputs[pos][i]);
   }

}


/**
 * Returns clipmask as nxi32 bitmask for the n vertices
 */
static LLVMValueRef
generate_clipmask(struct draw_llvm *llvm,
                  struct gallivm_state *gallivm,
                  struct lp_type vs_type,
                  LLVMValueRef (*outputs)[TGSI_NUM_CHANNELS],
                  boolean clip_xy,
                  boolean clip_z,
                  boolean clip_user,
                  boolean clip_halfz,
                  unsigned ucp_enable,
                  LLVMValueRef context_ptr,
                  boolean *have_clipdist)
{
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef mask; /* stores the <nxi32> clipmasks */
   LLVMValueRef test, temp;
   LLVMValueRef zero, shift;
   LLVMValueRef pos_x, pos_y, pos_z, pos_w;
   LLVMValueRef cv_x, cv_y, cv_z, cv_w;
   LLVMValueRef plane1, planes, plane_ptr, sum;
   struct lp_type f32_type = vs_type;
   struct lp_type i32_type = lp_int_type(vs_type);
   const unsigned pos = llvm->draw->vs.position_output;
   const unsigned cv = llvm->draw->vs.clipvertex_output;
   int num_written_clipdistance = llvm->draw->vs.vertex_shader->info.num_written_clipdistance;
   bool have_cd = false;
   unsigned cd[2];

   cd[0] = llvm->draw->vs.clipdistance_output[0];
   cd[1] = llvm->draw->vs.clipdistance_output[1];

   if (cd[0] != pos || cd[1] != pos)
      have_cd = true;

   if (num_written_clipdistance && !clip_user) {
      clip_user = true;
      ucp_enable = (1 << num_written_clipdistance) - 1;
   }

   mask = lp_build_const_int_vec(gallivm, i32_type, 0);
   temp = lp_build_const_int_vec(gallivm, i32_type, 0);
   zero = lp_build_const_vec(gallivm, f32_type, 0);         /* 0.0f 0.0f 0.0f 0.0f */
   shift = lp_build_const_int_vec(gallivm, i32_type, 1);    /* 1 1 1 1 */

   /*
    * load clipvertex and position from correct locations.
    * if they are the same just load them once.
    */
   pos_x = LLVMBuildLoad(builder, outputs[pos][0], ""); /*x0 x1 .. xn */
   pos_y = LLVMBuildLoad(builder, outputs[pos][1], ""); /*y0 y1 .. yn */
   pos_z = LLVMBuildLoad(builder, outputs[pos][2], ""); /*z0 z1 .. zn */
   pos_w = LLVMBuildLoad(builder, outputs[pos][3], ""); /*w0 w1 .. wn */

   if (clip_user && cv != pos) {
      cv_x = LLVMBuildLoad(builder, outputs[cv][0], ""); /*x0 x1 .. xn */
      cv_y = LLVMBuildLoad(builder, outputs[cv][1], ""); /*y0 y1 .. yn */
      cv_z = LLVMBuildLoad(builder, outputs[cv][2], ""); /*z0 z1 .. zn */
      cv_w = LLVMBuildLoad(builder, outputs[cv][3], ""); /*w0 w1 .. wn */
   } else {
      cv_x = pos_x;
      cv_y = pos_y;
      cv_z = pos_z;
      cv_w = pos_w;
   }

   /* Cliptest, for hardwired planes */
   if (clip_xy) {
      /* plane 1 */
      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w);
      temp = shift;
      test = LLVMBuildAnd(builder, test, temp, "");
      mask = test;

      /* plane 2 */
      test = LLVMBuildFAdd(builder, pos_x, pos_w, "");
      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test);
      temp = LLVMBuildShl(builder, temp, shift, "");
      test = LLVMBuildAnd(builder, test, temp, "");
      mask = LLVMBuildOr(builder, mask, test, "");

      /* plane 3 */
      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w);
      temp = LLVMBuildShl(builder, temp, shift, "");
      test = LLVMBuildAnd(builder, test, temp, "");
      mask = LLVMBuildOr(builder, mask, test, "");

      /* plane 4 */
      test = LLVMBuildFAdd(builder, pos_y, pos_w, "");
      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test);
      temp = LLVMBuildShl(builder, temp, shift, "");
      test = LLVMBuildAnd(builder, test, temp, "");
      mask = LLVMBuildOr(builder, mask, test, "");
   }

   if (clip_z) {
      temp = lp_build_const_int_vec(gallivm, i32_type, 16);
      if (clip_halfz) {
         /* plane 5 */
         test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, pos_z);
         test = LLVMBuildAnd(builder, test, temp, "");
         mask = LLVMBuildOr(builder, mask, test, "");
      }
      else {
         /* plane 5 */
         test = LLVMBuildFAdd(builder, pos_z, pos_w, "");
         test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, test);
         test = LLVMBuildAnd(builder, test, temp, "");
         mask = LLVMBuildOr(builder, mask, test, "");
      }
      /* plane 6 */
      test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, pos_z, pos_w);
      temp = LLVMBuildShl(builder, temp, shift, "");
      test = LLVMBuildAnd(builder, test, temp, "");
      mask = LLVMBuildOr(builder, mask, test, "");
   }

   if (clip_user) {
      LLVMValueRef planes_ptr = draw_jit_context_planes(gallivm, context_ptr);
      LLVMValueRef indices[3];
      LLVMValueRef is_nan_or_inf;

      /* userclip planes */
      while (ucp_enable) {
         unsigned plane_idx = ffs(ucp_enable)-1;
         ucp_enable &= ~(1 << plane_idx);
         plane_idx += 6;

         if (have_cd && num_written_clipdistance) {
            LLVMValueRef clipdist;
            int i;
            i = plane_idx - 6;

            *have_clipdist = TRUE;
            if (i < 4) {
               clipdist = LLVMBuildLoad(builder, outputs[cd[0]][i], "");
            } else {
               clipdist = LLVMBuildLoad(builder, outputs[cd[1]][i-4], "");
            }
            test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, clipdist);
            is_nan_or_inf = lp_build_is_inf_or_nan(gallivm, vs_type, clipdist);
            test = LLVMBuildOr(builder, test, is_nan_or_inf, "");
            temp = lp_build_const_int_vec(gallivm, i32_type, 1LL << plane_idx);
            test = LLVMBuildAnd(builder, test, temp, "");
            mask = LLVMBuildOr(builder, mask, test, "");
         } else {
            LLVMTypeRef vs_type_llvm = lp_build_vec_type(gallivm, vs_type);
            indices[0] = lp_build_const_int32(gallivm, 0);
            indices[1] = lp_build_const_int32(gallivm, plane_idx);

            indices[2] = lp_build_const_int32(gallivm, 0);
            plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
            plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_x");
            planes = lp_build_broadcast(gallivm, vs_type_llvm, plane1);
            sum = LLVMBuildFMul(builder, planes, cv_x, "");

            indices[2] = lp_build_const_int32(gallivm, 1);
            plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
            plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_y");
            planes = lp_build_broadcast(gallivm, vs_type_llvm, plane1);
            test = LLVMBuildFMul(builder, planes, cv_y, "");
            sum = LLVMBuildFAdd(builder, sum, test, "");

            indices[2] = lp_build_const_int32(gallivm, 2);
            plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
            plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_z");
            planes = lp_build_broadcast(gallivm, vs_type_llvm, plane1);
            test = LLVMBuildFMul(builder, planes, cv_z, "");
            sum = LLVMBuildFAdd(builder, sum, test, "");

            indices[2] = lp_build_const_int32(gallivm, 3);
            plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
            plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_w");
            planes = lp_build_broadcast(gallivm, vs_type_llvm, plane1);
            test = LLVMBuildFMul(builder, planes, cv_w, "");
            sum = LLVMBuildFAdd(builder, sum, test, "");

            test = lp_build_compare(gallivm, f32_type, PIPE_FUNC_GREATER, zero, sum);
            temp = lp_build_const_int_vec(gallivm, i32_type, 1LL << plane_idx);
            test = LLVMBuildAnd(builder, test, temp, "");
            mask = LLVMBuildOr(builder, mask, test, "");
         }
      }
   }
   return mask;
}


/**
 * Returns boolean if any clipping has occurred
 * Used zero/non-zero i32 value to represent boolean
 */
static LLVMValueRef
clipmask_booli32(struct gallivm_state *gallivm,
                 const struct lp_type vs_type,
                 LLVMValueRef clipmask_bool_ptr)
{
   LLVMBuilderRef builder = gallivm->builder;
   LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context);
   LLVMValueRef clipmask_bool = LLVMBuildLoad(builder, clipmask_bool_ptr, "");
   LLVMValueRef ret = LLVMConstNull(int32_type);
   LLVMValueRef temp;
   int i;

   /*
    * Can do this with log2(vector length) pack instructions and one extract
    * (as we don't actually need a or) with sse2 which would be way better.
    */
   for (i=0; i < vs_type.length; i++) {
      temp = LLVMBuildExtractElement(builder, clipmask_bool,
                                     lp_build_const_int32(gallivm, i) , "");
      ret = LLVMBuildOr(builder, ret, temp, "");
   }
   return ret;
}

static LLVMValueRef
draw_gs_llvm_fetch_input(const struct lp_build_tgsi_gs_iface *gs_iface,
                         struct lp_build_tgsi_context * bld_base,
                         boolean is_vindex_indirect,
                         LLVMValueRef vertex_index,
                         boolean is_aindex_indirect,
                         LLVMValueRef attrib_index,
                         LLVMValueRef swizzle_index)
{
   const struct draw_gs_llvm_iface *gs = draw_gs_llvm_iface(gs_iface);
   struct gallivm_state *gallivm = bld_base->base.gallivm;
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef indices[3];
   LLVMValueRef res;
   struct lp_type type = bld_base->base.type;

   if (is_vindex_indirect || is_aindex_indirect) {
      int i;
      res = bld_base->base.zero;
      for (i = 0; i < type.length; ++i) {
         LLVMValueRef idx = lp_build_const_int32(gallivm, i);
         LLVMValueRef vert_chan_index = vertex_index;
         LLVMValueRef attr_chan_index = attrib_index;
         LLVMValueRef channel_vec, value;

         if (is_vindex_indirect) {
            vert_chan_index = LLVMBuildExtractElement(builder,
                                                      vertex_index, idx, "");
         }
         if (is_aindex_indirect) {
            attr_chan_index = LLVMBuildExtractElement(builder,
                                                      attrib_index, idx, "");
         }

         indices[0] = vert_chan_index;
         indices[1] = attr_chan_index;
         indices[2] = swizzle_index;

         channel_vec = LLVMBuildGEP(builder, gs->input, indices, 3, "");
         channel_vec = LLVMBuildLoad(builder, channel_vec, "");
         value = LLVMBuildExtractElement(builder, channel_vec, idx, "");

         res = LLVMBuildInsertElement(builder, res, value, idx, "");
      }
   } else {
      indices[0] = vertex_index;
      indices[1] = attrib_index;
      indices[2] = swizzle_index;

      res = LLVMBuildGEP(builder, gs->input, indices, 3, "");
      res = LLVMBuildLoad(builder, res, "");
   }

   return res;
}

static void
draw_gs_llvm_emit_vertex(const struct lp_build_tgsi_gs_iface *gs_base,
                         struct lp_build_tgsi_context * bld_base,
                         LLVMValueRef (*outputs)[4],
                         LLVMValueRef emitted_vertices_vec)
{
   const struct draw_gs_llvm_iface *gs_iface = draw_gs_llvm_iface(gs_base);
   struct draw_gs_llvm_variant *variant = gs_iface->variant;
   struct gallivm_state *gallivm = variant->gallivm;
   LLVMBuilderRef builder = gallivm->builder;
   struct lp_type gs_type = bld_base->base.type;
   LLVMValueRef clipmask = lp_build_const_int_vec(gallivm,
                                                  lp_int_type(gs_type), 0);
   LLVMValueRef indices[LP_MAX_VECTOR_LENGTH];
   LLVMValueRef next_prim_offset =
      lp_build_const_int32(gallivm, variant->shader->base.primitive_boundary);
   LLVMValueRef io = variant->io_ptr;
   unsigned i;
   const struct tgsi_shader_info *gs_info = &variant->shader->base.info;

   for (i = 0; i < gs_type.length; ++i) {
      LLVMValueRef ind = lp_build_const_int32(gallivm, i);
      LLVMValueRef currently_emitted =
         LLVMBuildExtractElement(builder, emitted_vertices_vec, ind, "");
      indices[i] = LLVMBuildMul(builder, ind, next_prim_offset, "");
      indices[i] = LLVMBuildAdd(builder, indices[i], currently_emitted, "");
   }

   convert_to_aos(gallivm, io, indices,
                  outputs, clipmask,
                  gs_info->num_outputs, gs_type,
                  FALSE);
}

static void
draw_gs_llvm_end_primitive(const struct lp_build_tgsi_gs_iface *gs_base,
                           struct lp_build_tgsi_context * bld_base,
                           LLVMValueRef verts_per_prim_vec,
                           LLVMValueRef emitted_prims_vec)
{
   const struct draw_gs_llvm_iface *gs_iface = draw_gs_llvm_iface(gs_base);
   struct draw_gs_llvm_variant *variant = gs_iface->variant;
   struct gallivm_state *gallivm = variant->gallivm;
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef prim_lengts_ptr =
      draw_gs_jit_prim_lengths(variant->gallivm, variant->context_ptr);
   unsigned i;

   for (i = 0; i < bld_base->base.type.length; ++i) {
      LLVMValueRef ind = lp_build_const_int32(gallivm, i);
      LLVMValueRef prims_emitted =
         LLVMBuildExtractElement(builder, emitted_prims_vec, ind, "");
      LLVMValueRef store_ptr;
      LLVMValueRef num_vertices =
         LLVMBuildExtractElement(builder, verts_per_prim_vec, ind, "");

      store_ptr = LLVMBuildGEP(builder, prim_lengts_ptr, &prims_emitted, 1, "");
      store_ptr = LLVMBuildLoad(builder, store_ptr, "");
      store_ptr = LLVMBuildGEP(builder, store_ptr, &ind, 1, "");
      LLVMBuildStore(builder, num_vertices, store_ptr);
   }
}

static void
draw_gs_llvm_epilogue(const struct lp_build_tgsi_gs_iface *gs_base,
                      struct lp_build_tgsi_context * bld_base,
                      LLVMValueRef total_emitted_vertices_vec,
                      LLVMValueRef emitted_prims_vec)
{
   const struct draw_gs_llvm_iface *gs_iface = draw_gs_llvm_iface(gs_base);
   struct draw_gs_llvm_variant *variant = gs_iface->variant;
   struct gallivm_state *gallivm = variant->gallivm;
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef emitted_verts_ptr =
      draw_gs_jit_emitted_vertices(gallivm, variant->context_ptr);
   LLVMValueRef emitted_prims_ptr =
      draw_gs_jit_emitted_prims(gallivm, variant->context_ptr);
   LLVMValueRef zero = lp_build_const_int32(gallivm, 0);
   
   emitted_verts_ptr = LLVMBuildGEP(builder, emitted_verts_ptr, &zero, 0, "");
   emitted_prims_ptr = LLVMBuildGEP(builder, emitted_prims_ptr, &zero, 0, "");

   LLVMBuildStore(builder, total_emitted_vertices_vec, emitted_verts_ptr);
   LLVMBuildStore(builder, emitted_prims_vec, emitted_prims_ptr);
}

static void
draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
                   boolean elts)
{
   struct gallivm_state *gallivm = variant->gallivm;
   LLVMContextRef context = gallivm->context;
   LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
   LLVMTypeRef arg_types[11];
   unsigned num_arg_types =
      elts ? Elements(arg_types) : Elements(arg_types) - 1;
   LLVMTypeRef func_type;
   LLVMValueRef context_ptr;
   LLVMBasicBlockRef block;
   LLVMBuilderRef builder;
   char func_name[64];
   struct lp_type vs_type;
   LLVMValueRef end, start;
   LLVMValueRef count, fetch_elts, fetch_elt_max, fetch_count;
   LLVMValueRef vertex_id_offset, start_instance;
   LLVMValueRef stride, step, io_itr;
   LLVMValueRef io_ptr, vbuffers_ptr, vb_ptr;
   LLVMValueRef zero = lp_build_const_int32(gallivm, 0);
   LLVMValueRef one = lp_build_const_int32(gallivm, 1);
   struct draw_context *draw = llvm->draw;
   const struct tgsi_shader_info *vs_info = &draw->vs.vertex_shader->info;
   unsigned i, j;
   struct lp_build_context bld;
   struct lp_build_loop_state lp_loop;
   const int vector_length = lp_native_vector_width / 32;
   LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][TGSI_NUM_CHANNELS];
   LLVMValueRef fetch_max;
   struct lp_build_sampler_soa *sampler = 0;
   LLVMValueRef ret, clipmask_bool_ptr;
   struct draw_llvm_variant_key *key = &variant->key;
   /* If geometry shader is present we need to skip both the viewport
    * transformation and clipping otherwise the inputs to the geometry
    * shader will be incorrect.
    * The code can't handle vp transform when vs writes vp index neither
    * (though this would be fixable here, but couldn't just broadcast
    * the values).
    */
   const boolean bypass_viewport = key->has_gs || key->bypass_viewport ||
                                   llvm->draw->vs.vertex_shader->info.writes_viewport_index;
   const boolean enable_cliptest = !key->has_gs && (key->clip_xy ||
                                                    key->clip_z  ||
                                                    key->clip_user);
   LLVMValueRef variant_func;
   const unsigned pos = llvm->draw->vs.position_output;
   const unsigned cv = llvm->draw->vs.clipvertex_output;
   boolean have_clipdist = FALSE;
   struct lp_bld_tgsi_system_values system_values;

   memset(&system_values, 0, sizeof(system_values));

   util_snprintf(func_name, sizeof(func_name), "draw_llvm_vs_variant%u_%s",
                 variant->shader->variants_cached, elts ? "elts" : "linear");

   i = 0;
   arg_types[i++] = get_context_ptr_type(variant);       /* context */
   arg_types[i++] = get_vertex_header_ptr_type(variant); /* vertex_header */
   arg_types[i++] = get_buffer_ptr_type(variant);        /* vbuffers */
   if (elts) {
      arg_types[i++] = LLVMPointerType(int32_type, 0);/* fetch_elts  */
      arg_types[i++] = int32_type;                  /* fetch_elt_max */
   } else
      arg_types[i++] = int32_type;                  /* start */
   arg_types[i++] = int32_type;                     /* fetch_count / count */
   arg_types[i++] = int32_type;                     /* stride */
   arg_types[i++] = get_vb_ptr_type(variant);       /* pipe_vertex_buffer's */
   arg_types[i++] = int32_type;                     /* instance_id */
   arg_types[i++] = int32_type;                     /* vertex_id_offset */
   arg_types[i++] = int32_type;                     /* start_instance */

   func_type = LLVMFunctionType(int32_type, arg_types, num_arg_types, 0);

   variant_func = LLVMAddFunction(gallivm->module, func_name, func_type);

   if (elts)
      variant->function_elts = variant_func;
   else
      variant->function = variant_func;

   LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);
   for (i = 0; i < num_arg_types; ++i)
      if (LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
         LLVMAddAttribute(LLVMGetParam(variant_func, i),
                          LLVMNoAliasAttribute);

   context_ptr               = LLVMGetParam(variant_func, 0);
   io_ptr                    = LLVMGetParam(variant_func, 1);
   vbuffers_ptr              = LLVMGetParam(variant_func, 2);
   stride                    = LLVMGetParam(variant_func, 5 + (elts ? 1 : 0));
   vb_ptr                    = LLVMGetParam(variant_func, 6 + (elts ? 1 : 0));
   system_values.instance_id = LLVMGetParam(variant_func, 7 + (elts ? 1 : 0));
   vertex_id_offset          = LLVMGetParam(variant_func, 8 + (elts ? 1 : 0));
   start_instance            = LLVMGetParam(variant_func, 9 + (elts ? 1 : 0));

   lp_build_name(context_ptr, "context");
   lp_build_name(io_ptr, "io");
   lp_build_name(vbuffers_ptr, "vbuffers");
   lp_build_name(stride, "stride");
   lp_build_name(vb_ptr, "vb");
   lp_build_name(system_values.instance_id, "instance_id");
   lp_build_name(vertex_id_offset, "vertex_id_offset");
   lp_build_name(start_instance, "start_instance");

   if (elts) {
      fetch_elts    = LLVMGetParam(variant_func, 3);
      fetch_elt_max = LLVMGetParam(variant_func, 4);
      fetch_count   = LLVMGetParam(variant_func, 5);
      lp_build_name(fetch_elts, "fetch_elts");
      lp_build_name(fetch_elt_max, "fetch_elt_max");
      lp_build_name(fetch_count, "fetch_count");
      start = count = NULL;
   }
   else {
      start        = LLVMGetParam(variant_func, 3);
      count        = LLVMGetParam(variant_func, 4);
      lp_build_name(start, "start");
      lp_build_name(count, "count");
      fetch_elts = fetch_count = NULL;
   }

   /*
    * Function body
    */

   block = LLVMAppendBasicBlockInContext(gallivm->context, variant_func, "entry");
   builder = gallivm->builder;
   LLVMPositionBuilderAtEnd(builder, block);

   lp_build_context_init(&bld, gallivm, lp_type_int(32));

   memset(&vs_type, 0, sizeof vs_type);
   vs_type.floating = TRUE; /* floating point values */
   vs_type.sign = TRUE;     /* values are signed */
   vs_type.norm = FALSE;    /* values are not limited to [0,1] or [-1,1] */
   vs_type.width = 32;      /* 32-bit float */
   vs_type.length = vector_length;

   /* hold temporary "bool" clipmask */
   clipmask_bool_ptr = lp_build_alloca(gallivm, lp_build_int_vec_type(gallivm, vs_type), "");
   LLVMBuildStore(builder, lp_build_zero(gallivm, lp_int_type(vs_type)), clipmask_bool_ptr);

   /* code generated texture sampling */
   sampler = draw_llvm_sampler_soa_create(
      draw_llvm_variant_key_samplers(key),
      context_ptr);

   if (elts) {
      start = zero;
      end = fetch_count;
      count = fetch_count;
   }
   else {
      end = lp_build_add(&bld, start, count);
   }

   step = lp_build_const_int32(gallivm, vector_length);

   fetch_max = LLVMBuildSub(builder, end, one, "fetch_max");

   lp_build_loop_begin(&lp_loop, gallivm, zero);
   {
      LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][TGSI_NUM_CHANNELS];
      LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][LP_MAX_VECTOR_WIDTH / 32] = { { 0 } };
      LLVMValueRef io;
      LLVMValueRef clipmask;   /* holds the clipmask value */
      LLVMValueRef true_index_array = lp_build_zero(gallivm,
                                                    lp_type_uint_vec(32, 32*vector_length));
      const LLVMValueRef (*ptr_aos)[TGSI_NUM_CHANNELS];

      io_itr = lp_loop.counter;

      io = LLVMBuildGEP(builder, io_ptr, &io_itr, 1, "");
#if DEBUG_STORE
      lp_build_printf(gallivm, " --- io %d = %p, loop counter %d\n",
                      io_itr, io, lp_loop.counter);
#endif
      for (i = 0; i < vector_length; ++i) {
         LLVMValueRef vert_index =
            LLVMBuildAdd(builder,
                         lp_loop.counter,
                         lp_build_const_int32(gallivm, i), "");
         LLVMValueRef true_index =
            LLVMBuildAdd(builder, start, vert_index, "");

         /* make sure we're not out of bounds which can happen
          * if fetch_count % 4 != 0, because on the last iteration
          * a few of the 4 vertex fetches will be out of bounds */
         true_index = lp_build_min(&bld, true_index, fetch_max);

         if (elts) {
            LLVMValueRef fetch_ptr;
            LLVMValueRef index_overflowed;
            LLVMValueRef index_ptr =
               lp_build_alloca(
                  gallivm,
                  lp_build_vec_type(gallivm, lp_type_int(32)), "");
            struct lp_build_if_state if_ctx;
            index_overflowed = LLVMBuildICmp(builder, LLVMIntUGT,
                                             true_index, fetch_elt_max,
                                             "index_overflowed");

            lp_build_if(&if_ctx, gallivm, index_overflowed);
            {
               /* Generate maximum possible index so that
                * generate_fetch can treat it just like
                * any other overflow and return zeros.
                * We don't have to worry about the restart
                * primitive index because it has already been 
                * handled
                */
               LLVMValueRef val =
                  lp_build_const_int32(gallivm, 0xffffffff);
               LLVMBuildStore(builder, val, index_ptr);
            }
            lp_build_else(&if_ctx);
            {
               LLVMValueRef val;
               fetch_ptr = LLVMBuildGEP(builder, fetch_elts,
                                        &true_index, 1, "");
               val = LLVMBuildLoad(builder, fetch_ptr, "");
               LLVMBuildStore(builder, val, index_ptr);
            }
            lp_build_endif(&if_ctx);
            true_index = LLVMBuildLoad(builder, index_ptr, "true_index");
         }
         true_index_array = LLVMBuildInsertElement(
            gallivm->builder, true_index_array, true_index,
            lp_build_const_int32(gallivm, i), "");

         for (j = 0; j < draw->pt.nr_vertex_elements; ++j) {
            struct pipe_vertex_element *velem = &draw->pt.vertex_element[j];
            LLVMValueRef vb_index =
               lp_build_const_int32(gallivm, velem->vertex_buffer_index);
            LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr, &vb_index, 1, "");
            generate_fetch(gallivm, draw, vbuffers_ptr,
                           &aos_attribs[j][i], velem, vb, true_index,
                           system_values.instance_id, start_instance);
         }
      }
      convert_to_soa(gallivm, aos_attribs, inputs,
                     draw->pt.nr_vertex_elements, vs_type);

      /* In the paths with elts vertex id has to be unaffected by the
       * index bias and because indices inside our elements array have
       * already had index bias applied we need to subtract it here to
       * get back to the original index.
       * in the linear paths vertex id has to be unaffected by the
       * original start index and because we abuse the 'start' variable
       * to either represent the actual start index or the index at which
       * the primitive was split (we split rendering into chunks of at
       * most 4095-vertices) we need to back out the original start
       * index out of our vertex id here.
       */
      system_values.basevertex = lp_build_broadcast(gallivm, lp_build_vec_type(gallivm,
                                                       lp_type_uint_vec(32, 32*vector_length)),
                                                    vertex_id_offset);
      system_values.vertex_id = true_index_array;
      system_values.vertex_id_nobase = LLVMBuildSub(builder, true_index_array,
                                                      system_values.basevertex, "");

      ptr_aos = (const LLVMValueRef (*)[TGSI_NUM_CHANNELS]) inputs;
      generate_vs(variant,
                  builder,
                  vs_type,
                  outputs,
                  ptr_aos,
                  &system_values,
                  context_ptr,
                  sampler,
                  key->clamp_vertex_color);

      if (pos != -1 && cv != -1) {
         /* store original positions in clip before further manipulation */
         store_clip(gallivm, vs_type, io, outputs, FALSE, key->clip_user ? cv : pos);
         store_clip(gallivm, vs_type, io, outputs, TRUE, pos);

         /* do cliptest */
         if (enable_cliptest) {
            LLVMValueRef temp = LLVMBuildLoad(builder, clipmask_bool_ptr, "");
            /* allocate clipmask, assign it integer type */
            clipmask = generate_clipmask(llvm,
                                         gallivm,
                                         vs_type,
                                         outputs,
                                         key->clip_xy,
                                         key->clip_z,
                                         key->clip_user,
                                         key->clip_halfz,
                                         key->ucp_enable,
                                         context_ptr, &have_clipdist);
            temp = LLVMBuildOr(builder, clipmask, temp, "");
            /* store temporary clipping boolean value */
            LLVMBuildStore(builder, temp, clipmask_bool_ptr);
         }
         else {
            clipmask = lp_build_const_int_vec(gallivm, lp_int_type(vs_type), 0);
         }

         /* do viewport mapping */
         if (!bypass_viewport) {
            generate_viewport(variant, builder, vs_type, outputs, context_ptr);
         }
      }
      else {
         clipmask = lp_build_const_int_vec(gallivm, lp_int_type(vs_type), 0);
      }

      /* store clipmask in vertex header,
       * original positions in clip
       * and transformed positions in data
       */
      convert_to_aos(gallivm, io, NULL, outputs, clipmask,
                     vs_info->num_outputs, vs_type,
                     have_clipdist);
   }
   lp_build_loop_end_cond(&lp_loop, count, step, LLVMIntUGE);

   sampler->destroy(sampler);

   /* return clipping boolean value for function */
   ret = clipmask_booli32(gallivm, vs_type, clipmask_bool_ptr);

   LLVMBuildRet(builder, ret);

   gallivm_verify_function(gallivm, variant_func);
}


struct draw_llvm_variant_key *
draw_llvm_make_variant_key(struct draw_llvm *llvm, char *store)
{
   unsigned i;
   struct draw_llvm_variant_key *key;
   struct draw_sampler_static_state *draw_sampler;

   key = (struct draw_llvm_variant_key *)store;

   key->clamp_vertex_color = llvm->draw->rasterizer->clamp_vertex_color; /**/

   /* Presumably all variants of the shader should have the same
    * number of vertex elements - ie the number of shader inputs.
    * NOTE: we NEED to store the needed number of needed inputs
    * here, not the number of provided elements to match keysize
    * (and the offset of sampler state in the key).
    */
   key->nr_vertex_elements = llvm->draw->vs.vertex_shader->info.file_max[TGSI_FILE_INPUT] + 1;
   assert(key->nr_vertex_elements <= llvm->draw->pt.nr_vertex_elements);

   /* will have to rig this up properly later */
   key->clip_xy = llvm->draw->clip_xy;
   key->clip_z = llvm->draw->clip_z;
   key->clip_user = llvm->draw->clip_user;
   key->bypass_viewport = llvm->draw->bypass_viewport;
   key->clip_halfz = llvm->draw->rasterizer->clip_halfz;
   key->need_edgeflags = (llvm->draw->vs.edgeflag_output ? TRUE : FALSE);
   key->ucp_enable = llvm->draw->rasterizer->clip_plane_enable;
   key->has_gs = llvm->draw->gs.geometry_shader != NULL;
   key->num_outputs = draw_total_vs_outputs(llvm->draw);
   key->pad1 = 0;

   /* All variants of this shader will have the same value for
    * nr_samplers.  Not yet trying to compact away holes in the
    * sampler array.
    */
   key->nr_samplers = llvm->draw->vs.vertex_shader->info.file_max[TGSI_FILE_SAMPLER] + 1;
   if (llvm->draw->vs.vertex_shader->info.file_max[TGSI_FILE_SAMPLER_VIEW] != -1) {
      key->nr_sampler_views =
         llvm->draw->vs.vertex_shader->info.file_max[TGSI_FILE_SAMPLER_VIEW] + 1;
   }
   else {
      key->nr_sampler_views = key->nr_samplers;
   }

   draw_sampler = draw_llvm_variant_key_samplers(key);

   memcpy(key->vertex_element,
          llvm->draw->pt.vertex_element,
          sizeof(struct pipe_vertex_element) * key->nr_vertex_elements);

   memset(draw_sampler, 0, MAX2(key->nr_samplers, key->nr_sampler_views) * sizeof *draw_sampler);

   for (i = 0 ; i < key->nr_samplers; i++) {
      lp_sampler_static_sampler_state(&draw_sampler[i].sampler_state,
                                      llvm->draw->samplers[PIPE_SHADER_VERTEX][i]);
   }
   for (i = 0 ; i < key->nr_sampler_views; i++) {
      lp_sampler_static_texture_state(&draw_sampler[i].texture_state,
                                      llvm->draw->sampler_views[PIPE_SHADER_VERTEX][i]);
   }

   return key;
}


void
draw_llvm_dump_variant_key(struct draw_llvm_variant_key *key)
{
   unsigned i;
   struct draw_sampler_static_state *sampler = draw_llvm_variant_key_samplers(key);

   debug_printf("clamp_vertex_color = %u\n", key->clamp_vertex_color);
   debug_printf("clip_xy = %u\n", key->clip_xy);
   debug_printf("clip_z = %u\n", key->clip_z);
   debug_printf("clip_user = %u\n", key->clip_user);
   debug_printf("bypass_viewport = %u\n", key->bypass_viewport);
   debug_printf("clip_halfz = %u\n", key->clip_halfz);
   debug_printf("need_edgeflags = %u\n", key->need_edgeflags);
   debug_printf("has_gs = %u\n", key->has_gs);
   debug_printf("ucp_enable = %u\n", key->ucp_enable);

   for (i = 0 ; i < key->nr_vertex_elements; i++) {
      debug_printf("vertex_element[%i].src_offset = %u\n", i, key->vertex_element[i].src_offset);
      debug_printf("vertex_element[%i].instance_divisor = %u\n", i, key->vertex_element[i].instance_divisor);
      debug_printf("vertex_element[%i].vertex_buffer_index = %u\n", i, key->vertex_element[i].vertex_buffer_index);
      debug_printf("vertex_element[%i].src_format = %s\n", i, util_format_name(key->vertex_element[i].src_format));
   }

   for (i = 0 ; i < key->nr_sampler_views; i++) {
      debug_printf("sampler[%i].src_format = %s\n", i, util_format_name(sampler[i].texture_state.format));
   }
}


void
draw_llvm_set_mapped_texture(struct draw_context *draw,
                             unsigned shader_stage,
                             unsigned sview_idx,
                             uint32_t width, uint32_t height, uint32_t depth,
                             uint32_t first_level, uint32_t last_level,
                             const void *base_ptr,
                             uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
                             uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
                             uint32_t mip_offsets[PIPE_MAX_TEXTURE_LEVELS])
{
   unsigned j;
   struct draw_jit_texture *jit_tex;

   assert(shader_stage == PIPE_SHADER_VERTEX ||
          shader_stage == PIPE_SHADER_GEOMETRY);

   if (shader_stage == PIPE_SHADER_VERTEX) {
      assert(sview_idx < Elements(draw->llvm->jit_context.textures));

      jit_tex = &draw->llvm->jit_context.textures[sview_idx];
   } else if (shader_stage == PIPE_SHADER_GEOMETRY) {
      assert(sview_idx < Elements(draw->llvm->gs_jit_context.textures));

      jit_tex = &draw->llvm->gs_jit_context.textures[sview_idx];
   } else {
      assert(0);
      return;
   }

   jit_tex->width = width;
   jit_tex->height = height;
   jit_tex->depth = depth;
   jit_tex->first_level = first_level;
   jit_tex->last_level = last_level;
   jit_tex->base = base_ptr;

   for (j = first_level; j <= last_level; j++) {
      jit_tex->mip_offsets[j] = mip_offsets[j];
      jit_tex->row_stride[j] = row_stride[j];
      jit_tex->img_stride[j] = img_stride[j];
   }
}


void
draw_llvm_set_sampler_state(struct draw_context *draw, 
                            unsigned shader_type)
{
   unsigned i;

   if (shader_type == PIPE_SHADER_VERTEX) {
      for (i = 0; i < draw->num_samplers[PIPE_SHADER_VERTEX]; i++) {
         struct draw_jit_sampler *jit_sam = &draw->llvm->jit_context.samplers[i];

         if (draw->samplers[i]) {
            const struct pipe_sampler_state *s
               = draw->samplers[PIPE_SHADER_VERTEX][i];
            jit_sam->min_lod = s->min_lod;
            jit_sam->max_lod = s->max_lod;
            jit_sam->lod_bias = s->lod_bias;
            COPY_4V(jit_sam->border_color, s->border_color.f);
         }
      }
   } else if (shader_type == PIPE_SHADER_GEOMETRY) {
      for (i = 0; i < draw->num_samplers[PIPE_SHADER_GEOMETRY]; i++) {
         struct draw_jit_sampler *jit_sam = &draw->llvm->gs_jit_context.samplers[i];

         if (draw->samplers[i]) {
            const struct pipe_sampler_state *s
               = draw->samplers[PIPE_SHADER_GEOMETRY][i];
            jit_sam->min_lod = s->min_lod;
            jit_sam->max_lod = s->max_lod;
            jit_sam->lod_bias = s->lod_bias;
            COPY_4V(jit_sam->border_color, s->border_color.f);
         }
      }
   }
}


void
draw_llvm_destroy_variant(struct draw_llvm_variant *variant)
{
   struct draw_llvm *llvm = variant->llvm;

   gallivm_destroy(variant->gallivm);

   remove_from_list(&variant->list_item_local);
   variant->shader->variants_cached--;
   remove_from_list(&variant->list_item_global);
   llvm->nr_variants--;
   FREE(variant);
}


/**
 * Create LLVM types for various structures.
 */
static void
create_gs_jit_types(struct draw_gs_llvm_variant *var)
{
   struct gallivm_state *gallivm = var->gallivm;
   LLVMTypeRef texture_type, sampler_type, context_type;

   texture_type = create_jit_texture_type(gallivm, "texture");
   sampler_type = create_jit_sampler_type(gallivm, "sampler");

   context_type = create_gs_jit_context_type(gallivm,
                                             var->shader->base.vector_length,
                                             texture_type, sampler_type,
                                             "draw_gs_jit_context");
   var->context_ptr_type = LLVMPointerType(context_type, 0);

   var->input_array_type = create_gs_jit_input_type(gallivm);
}

static LLVMTypeRef
get_gs_context_ptr_type(struct draw_gs_llvm_variant *variant)
{
   if (!variant->context_ptr_type)
      create_gs_jit_types(variant);
   return variant->context_ptr_type;
}

static LLVMValueRef
generate_mask_value(struct draw_gs_llvm_variant *variant,
                    struct lp_type gs_type)
{
   struct gallivm_state *gallivm = variant->gallivm;
   LLVMBuilderRef builder = gallivm->builder;
   struct lp_type mask_type = lp_int_type(gs_type);
   LLVMValueRef num_prims;
   LLVMValueRef mask_val = lp_build_const_vec(gallivm, mask_type, 0);
   unsigned i;

   num_prims = lp_build_broadcast(gallivm, lp_build_vec_type(gallivm, mask_type),
                                  variant->num_prims);
   for (i = 0; i <= gs_type.length; i++) {
      LLVMValueRef idx = lp_build_const_int32(gallivm, i);
      mask_val = LLVMBuildInsertElement(builder, mask_val, idx, idx, "");
   }
   mask_val = lp_build_compare(gallivm, mask_type,
                               PIPE_FUNC_GREATER, num_prims, mask_val);

   return mask_val;
}

static void
draw_gs_llvm_generate(struct draw_llvm *llvm,
                      struct draw_gs_llvm_variant *variant)
{
   struct gallivm_state *gallivm = variant->gallivm;
   LLVMContextRef context = gallivm->context;
   LLVMTypeRef int32_type = LLVMInt32TypeInContext(context);
   LLVMTypeRef arg_types[6];
   LLVMTypeRef func_type;
   LLVMValueRef variant_func;
   LLVMValueRef context_ptr;
   LLVMValueRef prim_id_ptr;
   LLVMBasicBlockRef block;
   LLVMBuilderRef builder;
   LLVMValueRef io_ptr, input_array, num_prims, mask_val;
   struct lp_build_sampler_soa *sampler = 0;
   struct lp_build_context bld;
   struct lp_bld_tgsi_system_values system_values;
   char func_name[64];
   struct lp_type gs_type;
   unsigned i;
   struct draw_gs_llvm_iface gs_iface;
   const struct tgsi_token *tokens = variant->shader->base.state.tokens;
   LLVMValueRef consts_ptr, num_consts_ptr;
   LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][TGSI_NUM_CHANNELS];
   struct lp_build_mask_context mask;
   const struct tgsi_shader_info *gs_info = &variant->shader->base.info;
   unsigned vector_length = variant->shader->base.vector_length;

   memset(&system_values, 0, sizeof(system_values));

   util_snprintf(func_name, sizeof(func_name), "draw_llvm_gs_variant%u",
                 variant->shader->variants_cached);

   assert(variant->vertex_header_ptr_type);

   arg_types[0] = get_gs_context_ptr_type(variant);    /* context */
   arg_types[1] = variant->input_array_type;           /* input */
   arg_types[2] = variant->vertex_header_ptr_type;     /* vertex_header */
   arg_types[3] = int32_type;                          /* num_prims */
   arg_types[4] = int32_type;                          /* instance_id */
   arg_types[5] = LLVMPointerType(
      LLVMVectorType(int32_type, vector_length), 0);   /* prim_id_ptr */

   func_type = LLVMFunctionType(int32_type, arg_types, Elements(arg_types), 0);

   variant_func = LLVMAddFunction(gallivm->module, func_name, func_type);

   variant->function = variant_func;

   LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);

   for (i = 0; i < Elements(arg_types); ++i)
      if (LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
         LLVMAddAttribute(LLVMGetParam(variant_func, i),
                          LLVMNoAliasAttribute);

   context_ptr               = LLVMGetParam(variant_func, 0);
   input_array               = LLVMGetParam(variant_func, 1);
   io_ptr                    = LLVMGetParam(variant_func, 2);
   num_prims                 = LLVMGetParam(variant_func, 3);
   system_values.instance_id = LLVMGetParam(variant_func, 4);
   prim_id_ptr               = LLVMGetParam(variant_func, 5);

   lp_build_name(context_ptr, "context");
   lp_build_name(input_array, "input");
   lp_build_name(io_ptr, "io");
   lp_build_name(num_prims, "num_prims");
   lp_build_name(system_values.instance_id, "instance_id");
   lp_build_name(prim_id_ptr, "prim_id_ptr");

   variant->context_ptr = context_ptr;
   variant->io_ptr = io_ptr;
   variant->num_prims = num_prims;

   gs_iface.base.fetch_input = draw_gs_llvm_fetch_input;
   gs_iface.base.emit_vertex = draw_gs_llvm_emit_vertex;
   gs_iface.base.end_primitive = draw_gs_llvm_end_primitive;
   gs_iface.base.gs_epilogue = draw_gs_llvm_epilogue;
   gs_iface.input = input_array;
   gs_iface.variant = variant;

   /*
    * Function body
    */

   block = LLVMAppendBasicBlockInContext(gallivm->context, variant_func, "entry");
   builder = gallivm->builder;
   LLVMPositionBuilderAtEnd(builder, block);

   lp_build_context_init(&bld, gallivm, lp_type_int(32));

   memset(&gs_type, 0, sizeof gs_type);
   gs_type.floating = TRUE; /* floating point values */
   gs_type.sign = TRUE;     /* values are signed */
   gs_type.norm = FALSE;    /* values are not limited to [0,1] or [-1,1] */
   gs_type.width = 32;      /* 32-bit float */
   gs_type.length = vector_length;

   consts_ptr = draw_gs_jit_context_constants(variant->gallivm, context_ptr);
   num_consts_ptr =
      draw_gs_jit_context_num_constants(variant->gallivm, context_ptr);

   /* code generated texture sampling */
   sampler = draw_llvm_sampler_soa_create(variant->key.samplers,
                                          context_ptr);

   mask_val = generate_mask_value(variant, gs_type);
   lp_build_mask_begin(&mask, gallivm, gs_type, mask_val);

   if (gs_info->uses_primid) {
      system_values.prim_id = LLVMBuildLoad(builder, prim_id_ptr, "prim_id");
   }

   if (gallivm_debug & (GALLIVM_DEBUG_TGSI | GALLIVM_DEBUG_IR)) {
      tgsi_dump(tokens, 0);
      draw_gs_llvm_dump_variant_key(&variant->key);
   }

   lp_build_tgsi_soa(variant->gallivm,
                     tokens,
                     gs_type,
                     &mask,
                     consts_ptr,
                     num_consts_ptr,
                     &system_values,
                     NULL,
                     outputs,
                     sampler,
                     &llvm->draw->gs.geometry_shader->info,
                     (const struct lp_build_tgsi_gs_iface *)&gs_iface);

   sampler->destroy(sampler);

   lp_build_mask_end(&mask);

   LLVMBuildRet(builder, lp_build_zero(gallivm, lp_type_uint(32)));

   gallivm_verify_function(gallivm, variant_func);
}


struct draw_gs_llvm_variant *
draw_gs_llvm_create_variant(struct draw_llvm *llvm,
                            unsigned num_outputs,
                            const struct draw_gs_llvm_variant_key *key)
{
   struct draw_gs_llvm_variant *variant;
   struct llvm_geometry_shader *shader =
      llvm_geometry_shader(llvm->draw->gs.geometry_shader);
   LLVMTypeRef vertex_header;
   char module_name[64];

   variant = MALLOC(sizeof *variant +
                    shader->variant_key_size -
                    sizeof variant->key);
   if (variant == NULL)
      return NULL;

   variant->llvm = llvm;
   variant->shader = shader;

   util_snprintf(module_name, sizeof(module_name), "draw_llvm_gs_variant%u",
                 variant->shader->variants_cached);

   variant->gallivm = gallivm_create(module_name, llvm->context);

   create_gs_jit_types(variant);

   memcpy(&variant->key, key, shader->variant_key_size);

   vertex_header = create_jit_vertex_header(variant->gallivm, num_outputs);

   variant->vertex_header_ptr_type = LLVMPointerType(vertex_header, 0);

   draw_gs_llvm_generate(llvm, variant);

   gallivm_compile_module(variant->gallivm);

   variant->jit_func = (draw_gs_jit_func)
         gallivm_jit_function(variant->gallivm, variant->function);

   gallivm_free_ir(variant->gallivm);

   variant->list_item_global.base = variant;
   variant->list_item_local.base = variant;
   /*variant->no = */shader->variants_created++;
   variant->list_item_global.base = variant;

   return variant;
}

void
draw_gs_llvm_destroy_variant(struct draw_gs_llvm_variant *variant)
{
   struct draw_llvm *llvm = variant->llvm;

   gallivm_destroy(variant->gallivm);

   remove_from_list(&variant->list_item_local);
   variant->shader->variants_cached--;
   remove_from_list(&variant->list_item_global);
   llvm->nr_gs_variants--;
   FREE(variant);
}

struct draw_gs_llvm_variant_key *
draw_gs_llvm_make_variant_key(struct draw_llvm *llvm, char *store)
{
   unsigned i;
   struct draw_gs_llvm_variant_key *key;
   struct draw_sampler_static_state *draw_sampler;

   key = (struct draw_gs_llvm_variant_key *)store;

   key->num_outputs = draw_total_gs_outputs(llvm->draw);

   /* All variants of this shader will have the same value for
    * nr_samplers.  Not yet trying to compact away holes in the
    * sampler array.
    */
   key->nr_samplers = llvm->draw->gs.geometry_shader->info.file_max[TGSI_FILE_SAMPLER] + 1;
   if (llvm->draw->gs.geometry_shader->info.file_max[TGSI_FILE_SAMPLER_VIEW] != -1) {
      key->nr_sampler_views =
         llvm->draw->gs.geometry_shader->info.file_max[TGSI_FILE_SAMPLER_VIEW] + 1;
   }
   else {
      key->nr_sampler_views = key->nr_samplers;
   }

   draw_sampler = key->samplers;

   memset(draw_sampler, 0, MAX2(key->nr_samplers, key->nr_sampler_views) * sizeof *draw_sampler);

   for (i = 0 ; i < key->nr_samplers; i++) {
      lp_sampler_static_sampler_state(&draw_sampler[i].sampler_state,
                                      llvm->draw->samplers[PIPE_SHADER_GEOMETRY][i]);
   }
   for (i = 0 ; i < key->nr_sampler_views; i++) {
      lp_sampler_static_texture_state(&draw_sampler[i].texture_state,
                                      llvm->draw->sampler_views[PIPE_SHADER_GEOMETRY][i]);
   }

   return key;
}

void
draw_gs_llvm_dump_variant_key(struct draw_gs_llvm_variant_key *key)
{
   unsigned i;
   struct draw_sampler_static_state *sampler = key->samplers;

   for (i = 0 ; i < key->nr_sampler_views; i++) {
      debug_printf("sampler[%i].src_format = %s\n", i,
                   util_format_name(sampler[i].texture_state.format));
   }
}