summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/etnaviv/etnaviv_compiler.c
blob: 59e1452d80132605b86489feaae6a9a0619c8eec (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
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
/*
 * Copyright (c) 2012-2015 Etnaviv Project
 *
 * 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
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 * DEALINGS IN THE SOFTWARE.
 *
 * Authors:
 *    Wladimir J. van der Laan <laanwj@gmail.com>
 */

/* TGSI->Vivante shader ISA conversion */

/* What does the compiler return (see etna_shader_object)?
 *  1) instruction data
 *  2) input-to-temporary mapping (fixed for ps)
 *      *) in case of ps, semantic -> varying id mapping
 *      *) for each varying: number of components used (r, rg, rgb, rgba)
 *  3) temporary-to-output mapping (in case of vs, fixed for ps)
 *  4) for each input/output: possible semantic (position, color, glpointcoord, ...)
 *  5) immediates base offset, immediates data
 *  6) used texture units (and possibly the TGSI_TEXTURE_* type); not needed to
 *     configure the hw, but useful for error checking
 *  7) enough information to add the z=(z+w)/2.0 necessary for older chips
 *     (output reg id is enough)
 *
 *  Empty shaders are not allowed, should always at least generate a NOP. Also
 *  if there is a label at the end of the shader, an extra NOP should be
 *  generated as jump target.
 *
 * TODO
 * * Use an instruction scheduler
 * * Indirect access to uniforms / temporaries using amode
 */

#include "etnaviv_compiler.h"

#include "etnaviv_asm.h"
#include "etnaviv_context.h"
#include "etnaviv_debug.h"
#include "etnaviv_disasm.h"
#include "etnaviv_uniforms.h"
#include "etnaviv_util.h"

#include "pipe/p_shader_tokens.h"
#include "tgsi/tgsi_info.h"
#include "tgsi/tgsi_iterate.h"
#include "tgsi/tgsi_lowering.h"
#include "tgsi/tgsi_strings.h"
#include "tgsi/tgsi_util.h"
#include "util/u_math.h"
#include "util/u_memory.h"

#include <fcntl.h>
#include <stdio.h>
#include <sys/stat.h>
#include <sys/types.h>

#define ETNA_MAX_INNER_TEMPS 2

static const float sincos_const[2][4] = {
   {
      2., -1., 4., -4.,
   },
   {
      1. / (2. * M_PI), 0.75, 0.5, 0.0,
   },
};

/* Native register description structure */
struct etna_native_reg {
   unsigned valid : 1;
   unsigned is_tex : 1; /* is texture unit, overrides rgroup */
   unsigned rgroup : 3;
   unsigned id : 9;
};

/* Register description */
struct etna_reg_desc {
   enum tgsi_file_type file; /* IN, OUT, TEMP, ... */
   int idx; /* index into file */
   bool active; /* used in program */
   int first_use; /* instruction id of first use (scope begin) */
   int last_use; /* instruction id of last use (scope end, inclusive) */

   struct etna_native_reg native; /* native register to map to */
   unsigned usage_mask : 4; /* usage, per channel */
   bool has_semantic; /* register has associated TGSI semantic */
   struct tgsi_declaration_semantic semantic; /* TGSI semantic */
   struct tgsi_declaration_interp interp; /* Interpolation type */
};

/* Label information structure */
struct etna_compile_label {
   int inst_idx; /* Instruction id that label points to */
};

enum etna_compile_frame_type {
   ETNA_COMPILE_FRAME_IF, /* IF/ELSE/ENDIF */
   ETNA_COMPILE_FRAME_LOOP,
};

/* nesting scope frame (LOOP, IF, ...) during compilation
 */
struct etna_compile_frame {
   enum etna_compile_frame_type type;
   struct etna_compile_label *lbl_else;
   struct etna_compile_label *lbl_endif;
   struct etna_compile_label *lbl_loop_bgn;
   struct etna_compile_label *lbl_loop_end;
};

struct etna_compile_file {
   /* Number of registers in each TGSI file (max register+1) */
   size_t reg_size;
   /* Register descriptions, per register index */
   struct etna_reg_desc *reg;
};

#define array_insert(arr, val)                          \
   do {                                                 \
      if (arr##_count == arr##_sz) {                    \
         arr##_sz = MAX2(2 * arr##_sz, 16);             \
         arr = realloc(arr, arr##_sz * sizeof(arr[0])); \
      }                                                 \
      arr[arr##_count++] = val;                         \
   } while (0)


/* scratch area for compiling shader, freed after compilation finishes */
struct etna_compile {
   const struct tgsi_token *tokens;
   bool free_tokens;

   struct tgsi_shader_info info;

   /* Register descriptions, per TGSI file, per register index */
   struct etna_compile_file file[TGSI_FILE_COUNT];

   /* Keep track of TGSI register declarations */
   struct etna_reg_desc decl[ETNA_MAX_DECL];
   uint total_decls;

   /* Bitmap of dead instructions which are removed in a separate pass */
   bool dead_inst[ETNA_MAX_TOKENS];

   /* Immediate data */
   enum etna_immediate_contents imm_contents[ETNA_MAX_IMM];
   uint32_t imm_data[ETNA_MAX_IMM];
   uint32_t imm_base; /* base of immediates (in 32 bit units) */
   uint32_t imm_size; /* size of immediates (in 32 bit units) */

   /* Next free native register, for register allocation */
   uint32_t next_free_native;

   /* Temporary register for use within translated TGSI instruction,
    * only allocated when needed.
    */
   int inner_temps; /* number of inner temps used; only up to one available at
                       this point */
   struct etna_native_reg inner_temp[ETNA_MAX_INNER_TEMPS];

   /* Fields for handling nested conditionals */
   struct etna_compile_frame frame_stack[ETNA_MAX_DEPTH];
   int frame_sp;
   struct etna_compile_label *lbl_usage[ETNA_MAX_INSTRUCTIONS];

   unsigned labels_count, labels_sz;
   struct etna_compile_label *labels;

   /* Code generation */
   int inst_ptr; /* current instruction pointer */
   uint32_t code[ETNA_MAX_INSTRUCTIONS * ETNA_INST_SIZE];

   /* I/O */

   /* Number of varyings (PS only) */
   int num_varyings;

   /* GPU hardware specs */
   const struct etna_specs *specs;
};

static struct etna_reg_desc *
etna_get_dst_reg(struct etna_compile *c, struct tgsi_dst_register dst)
{
   return &c->file[dst.File].reg[dst.Index];
}

static struct etna_reg_desc *
etna_get_src_reg(struct etna_compile *c, struct tgsi_src_register src)
{
   return &c->file[src.File].reg[src.Index];
}

static struct etna_native_reg
etna_native_temp(unsigned reg)
{
   return (struct etna_native_reg) {
      .valid = 1,
      .rgroup = INST_RGROUP_TEMP,
      .id = reg
   };
}

/** Register allocation **/
enum reg_sort_order {
   FIRST_USE_ASC,
   FIRST_USE_DESC,
   LAST_USE_ASC,
   LAST_USE_DESC
};

/* Augmented register description for sorting */
struct sort_rec {
   struct etna_reg_desc *ptr;
   int key;
};

static int
sort_rec_compar(const struct sort_rec *a, const struct sort_rec *b)
{
   if (a->key < b->key)
      return -1;

   if (a->key > b->key)
      return 1;

   return 0;
}

/* create an index on a register set based on certain criteria. */
static int
sort_registers(struct sort_rec *sorted, struct etna_compile_file *file,
               enum reg_sort_order so)
{
   struct etna_reg_desc *regs = file->reg;
   int ptr = 0;

   /* pre-populate keys from active registers */
   for (int idx = 0; idx < file->reg_size; ++idx) {
      /* only interested in active registers now; will only assign inactive ones
       * if no space in active ones */
      if (regs[idx].active) {
         sorted[ptr].ptr = &regs[idx];

         switch (so) {
         case FIRST_USE_ASC:
            sorted[ptr].key = regs[idx].first_use;
            break;
         case LAST_USE_ASC:
            sorted[ptr].key = regs[idx].last_use;
            break;
         case FIRST_USE_DESC:
            sorted[ptr].key = -regs[idx].first_use;
            break;
         case LAST_USE_DESC:
            sorted[ptr].key = -regs[idx].last_use;
            break;
         }
         ptr++;
      }
   }

   /* sort index by key */
   qsort(sorted, ptr, sizeof(struct sort_rec),
         (int (*)(const void *, const void *))sort_rec_compar);

   return ptr;
}

/* Allocate a new, unused, native temp register */
static struct etna_native_reg
alloc_new_native_reg(struct etna_compile *c)
{
   assert(c->next_free_native < ETNA_MAX_TEMPS);
   return etna_native_temp(c->next_free_native++);
}

/* assign TEMPs to native registers */
static void
assign_temporaries_to_native(struct etna_compile *c,
                             struct etna_compile_file *file)
{
   struct etna_reg_desc *temps = file->reg;

   for (int idx = 0; idx < file->reg_size; ++idx)
      temps[idx].native = alloc_new_native_reg(c);
}

/* assign inputs and outputs to temporaries
 * Gallium assumes that the hardware has separate registers for taking input and
 * output, however Vivante GPUs use temporaries both for passing in inputs and
 * passing back outputs.
 * Try to re-use temporary registers where possible. */
static void
assign_inouts_to_temporaries(struct etna_compile *c, uint file)
{
   bool mode_inputs = (file == TGSI_FILE_INPUT);
   int inout_ptr = 0, num_inouts;
   int temp_ptr = 0, num_temps;
   struct sort_rec inout_order[ETNA_MAX_TEMPS];
   struct sort_rec temps_order[ETNA_MAX_TEMPS];
   num_inouts = sort_registers(inout_order, &c->file[file],
                               mode_inputs ? LAST_USE_ASC : FIRST_USE_ASC);
   num_temps = sort_registers(temps_order, &c->file[TGSI_FILE_TEMPORARY],
                              mode_inputs ? FIRST_USE_ASC : LAST_USE_ASC);

   while (inout_ptr < num_inouts && temp_ptr < num_temps) {
      struct etna_reg_desc *inout = inout_order[inout_ptr].ptr;
      struct etna_reg_desc *temp = temps_order[temp_ptr].ptr;

      if (!inout->active || inout->native.valid) { /* Skip if already a native register assigned */
         inout_ptr++;
         continue;
      }

      /* last usage of this input is before or in same instruction of first use
       * of temporary? */
      if (mode_inputs ? (inout->last_use <= temp->first_use)
                      : (inout->first_use >= temp->last_use)) {
         /* assign it and advance to next input */
         inout->native = temp->native;
         inout_ptr++;
      }

      temp_ptr++;
   }

   /* if we couldn't reuse current ones, allocate new temporaries */
   for (inout_ptr = 0; inout_ptr < num_inouts; ++inout_ptr) {
      struct etna_reg_desc *inout = inout_order[inout_ptr].ptr;

      if (inout->active && !inout->native.valid)
         inout->native = alloc_new_native_reg(c);
   }
}

/* Allocate an immediate with a certain value and return the index. If
 * there is already an immediate with that value, return that.
 */
static struct etna_inst_src
alloc_imm(struct etna_compile *c, enum etna_immediate_contents contents,
          uint32_t value)
{
   int idx;

   /* Could use a hash table to speed this up */
   for (idx = 0; idx < c->imm_size; ++idx) {
      if (c->imm_contents[idx] == contents && c->imm_data[idx] == value)
         break;
   }

   /* look if there is an unused slot */
   if (idx == c->imm_size) {
      for (idx = 0; idx < c->imm_size; ++idx) {
         if (c->imm_contents[idx] == ETNA_IMMEDIATE_UNUSED)
            break;
      }
   }

   /* allocate new immediate */
   if (idx == c->imm_size) {
      assert(c->imm_size < ETNA_MAX_IMM);
      idx = c->imm_size++;
      c->imm_data[idx] = value;
      c->imm_contents[idx] = contents;
   }

   /* swizzle so that component with value is returned in all components */
   idx += c->imm_base;
   struct etna_inst_src imm_src = {
      .use = 1,
      .rgroup = INST_RGROUP_UNIFORM_0,
      .reg = idx / 4,
      .swiz = INST_SWIZ_BROADCAST(idx & 3)
   };

   return imm_src;
}

static struct etna_inst_src
alloc_imm_u32(struct etna_compile *c, uint32_t value)
{
   return alloc_imm(c, ETNA_IMMEDIATE_CONSTANT, value);
}

static struct etna_inst_src
alloc_imm_vec4u(struct etna_compile *c, enum etna_immediate_contents contents,
                const uint32_t *values)
{
   struct etna_inst_src imm_src = { };
   int idx, i;

   for (idx = 0; idx + 3 < c->imm_size; idx += 4) {
      /* What if we can use a uniform with a different swizzle? */
      for (i = 0; i < 4; i++)
         if (c->imm_contents[idx + i] != contents || c->imm_data[idx + i] != values[i])
            break;
      if (i == 4)
         break;
   }

   if (idx + 3 >= c->imm_size) {
      idx = align(c->imm_size, 4);
      assert(idx + 4 <= ETNA_MAX_IMM);

      for (i = 0; i < 4; i++) {
         c->imm_data[idx + i] = values[i];
         c->imm_contents[idx + i] = contents;
      }

      c->imm_size = idx + 4;
   }

   assert((c->imm_base & 3) == 0);
   idx += c->imm_base;
   imm_src.use = 1;
   imm_src.rgroup = INST_RGROUP_UNIFORM_0;
   imm_src.reg = idx / 4;
   imm_src.swiz = INST_SWIZ_IDENTITY;

   return imm_src;
}

static uint32_t
get_imm_u32(struct etna_compile *c, const struct etna_inst_src *imm,
            unsigned swiz_idx)
{
   assert(imm->use == 1 && imm->rgroup == INST_RGROUP_UNIFORM_0);
   unsigned int idx = imm->reg * 4 + ((imm->swiz >> (swiz_idx * 2)) & 3);

   return c->imm_data[idx];
}

/* Allocate immediate with a certain float value. If there is already an
 * immediate with that value, return that.
 */
static struct etna_inst_src
alloc_imm_f32(struct etna_compile *c, float value)
{
   return alloc_imm_u32(c, fui(value));
}

static struct etna_inst_src
etna_imm_vec4f(struct etna_compile *c, const float *vec4)
{
   uint32_t val[4];

   for (int i = 0; i < 4; i++)
      val[i] = fui(vec4[i]);

   return alloc_imm_vec4u(c, ETNA_IMMEDIATE_CONSTANT, val);
}

/* Pass -- check register file declarations and immediates */
static void
etna_compile_parse_declarations(struct etna_compile *c)
{
   struct tgsi_parse_context ctx = { };
   unsigned status = TGSI_PARSE_OK;
   status = tgsi_parse_init(&ctx, c->tokens);
   assert(status == TGSI_PARSE_OK);

   while (!tgsi_parse_end_of_tokens(&ctx)) {
      tgsi_parse_token(&ctx);

      switch (ctx.FullToken.Token.Type) {
      case TGSI_TOKEN_TYPE_IMMEDIATE: {
         /* immediates are handled differently from other files; they are
          * not declared explicitly, and always add four components */
         const struct tgsi_full_immediate *imm = &ctx.FullToken.FullImmediate;
         assert(c->imm_size <= (ETNA_MAX_IMM - 4));

         for (int i = 0; i < 4; ++i) {
            unsigned idx = c->imm_size++;

            c->imm_data[idx] = imm->u[i].Uint;
            c->imm_contents[idx] = ETNA_IMMEDIATE_CONSTANT;
         }
      }
      break;
      }
   }

   tgsi_parse_free(&ctx);
}

/* Allocate register declarations for the registers in all register files */
static void
etna_allocate_decls(struct etna_compile *c)
{
   uint idx = 0;

   for (int x = 0; x < TGSI_FILE_COUNT; ++x) {
      c->file[x].reg = &c->decl[idx];
      c->file[x].reg_size = c->info.file_max[x] + 1;

      for (int sub = 0; sub < c->file[x].reg_size; ++sub) {
         c->decl[idx].file = x;
         c->decl[idx].idx = sub;
         idx++;
      }
   }

   c->total_decls = idx;
}

/* Pass -- check and record usage of temporaries, inputs, outputs */
static void
etna_compile_pass_check_usage(struct etna_compile *c)
{
   struct tgsi_parse_context ctx = { };
   unsigned status = TGSI_PARSE_OK;
   status = tgsi_parse_init(&ctx, c->tokens);
   assert(status == TGSI_PARSE_OK);

   for (int idx = 0; idx < c->total_decls; ++idx) {
      c->decl[idx].active = false;
      c->decl[idx].first_use = c->decl[idx].last_use = -1;
   }

   int inst_idx = 0;
   while (!tgsi_parse_end_of_tokens(&ctx)) {
      tgsi_parse_token(&ctx);
      /* find out max register #s used
       * For every register mark first and last instruction index where it's
       * used this allows finding ranges where the temporary can be borrowed
       * as input and/or output register
       *
       * XXX in the case of loops this needs special care, or even be completely
       * disabled, as
       * the last usage of a register inside a loop means it can still be used
       * on next loop
       * iteration (execution is no longer * chronological). The register can
       * only be
       * declared "free" after the loop finishes.
       *
       * Same for inputs: the first usage of a register inside a loop doesn't
       * mean that the register
       * won't have been overwritten in previous iteration. The register can
       * only be declared free before the loop
       * starts.
       * The proper way would be to do full dominator / post-dominator analysis
       * (especially with more complicated
       * control flow such as direct branch instructions) but not for now...
       */
      switch (ctx.FullToken.Token.Type) {
      case TGSI_TOKEN_TYPE_DECLARATION: {
         /* Declaration: fill in file details */
         const struct tgsi_full_declaration *decl = &ctx.FullToken.FullDeclaration;
         struct etna_compile_file *file = &c->file[decl->Declaration.File];

         for (int idx = decl->Range.First; idx <= decl->Range.Last; ++idx) {
            file->reg[idx].usage_mask = 0; // we'll compute this ourselves
            file->reg[idx].has_semantic = decl->Declaration.Semantic;
            file->reg[idx].semantic = decl->Semantic;
            file->reg[idx].interp = decl->Interp;
         }
      } break;
      case TGSI_TOKEN_TYPE_INSTRUCTION: {
         /* Instruction: iterate over operands of instruction */
         const struct tgsi_full_instruction *inst = &ctx.FullToken.FullInstruction;

         /* iterate over destination registers */
         for (int idx = 0; idx < inst->Instruction.NumDstRegs; ++idx) {
            struct etna_reg_desc *reg_desc = &c->file[inst->Dst[idx].Register.File].reg[inst->Dst[idx].Register.Index];

            if (reg_desc->first_use == -1)
               reg_desc->first_use = inst_idx;

            reg_desc->last_use = inst_idx;
            reg_desc->active = true;
         }

         /* iterate over source registers */
         for (int idx = 0; idx < inst->Instruction.NumSrcRegs; ++idx) {
            struct etna_reg_desc *reg_desc = &c->file[inst->Src[idx].Register.File].reg[inst->Src[idx].Register.Index];

            if (reg_desc->first_use == -1)
               reg_desc->first_use = inst_idx;

            reg_desc->last_use = inst_idx;
            reg_desc->active = true;
            /* accumulate usage mask for register, this is used to determine how
             * many slots for varyings
             * should be allocated */
            reg_desc->usage_mask |= tgsi_util_get_inst_usage_mask(inst, idx);
         }
         inst_idx += 1;
      } break;
      default:
         break;
      }
   }

   tgsi_parse_free(&ctx);
}

/* assign inputs that need to be assigned to specific registers */
static void
assign_special_inputs(struct etna_compile *c)
{
   if (c->info.processor == PIPE_SHADER_FRAGMENT) {
      /* never assign t0 as it is the position output, start assigning at t1 */
      c->next_free_native = 1;

      /* hardwire TGSI_SEMANTIC_POSITION (input and output) to t0 */
      for (int idx = 0; idx < c->total_decls; ++idx) {
         struct etna_reg_desc *reg = &c->decl[idx];

         if (reg->active && reg->semantic.Name == TGSI_SEMANTIC_POSITION)
            reg->native = etna_native_temp(0);
      }
   }
}

/* Check that a move instruction does not swizzle any of the components
 * that it writes.
 */
static bool
etna_mov_check_no_swizzle(const struct tgsi_dst_register dst,
                          const struct tgsi_src_register src)
{
   return (!(dst.WriteMask & TGSI_WRITEMASK_X) || src.SwizzleX == TGSI_SWIZZLE_X) &&
          (!(dst.WriteMask & TGSI_WRITEMASK_Y) || src.SwizzleY == TGSI_SWIZZLE_Y) &&
          (!(dst.WriteMask & TGSI_WRITEMASK_Z) || src.SwizzleZ == TGSI_SWIZZLE_Z) &&
          (!(dst.WriteMask & TGSI_WRITEMASK_W) || src.SwizzleW == TGSI_SWIZZLE_W);
}

/* Pass -- optimize outputs
 * Mesa tends to generate code like this at the end if their shaders
 *   MOV OUT[1], TEMP[2]
 *   MOV OUT[0], TEMP[0]
 *   MOV OUT[2], TEMP[1]
 * Recognize if
 * a) there is only a single assignment to an output register and
 * b) the temporary is not used after that
 * Also recognize direct assignment of IN to OUT (passthrough)
 **/
static void
etna_compile_pass_optimize_outputs(struct etna_compile *c)
{
   struct tgsi_parse_context ctx = { };
   int inst_idx = 0;
   unsigned status = TGSI_PARSE_OK;
   status = tgsi_parse_init(&ctx, c->tokens);
   assert(status == TGSI_PARSE_OK);

   while (!tgsi_parse_end_of_tokens(&ctx)) {
      tgsi_parse_token(&ctx);

      switch (ctx.FullToken.Token.Type) {
      case TGSI_TOKEN_TYPE_INSTRUCTION: {
         const struct tgsi_full_instruction *inst = &ctx.FullToken.FullInstruction;

         /* iterate over operands */
         switch (inst->Instruction.Opcode) {
         case TGSI_OPCODE_MOV: {
            /* We are only interested in eliminating MOVs which write to
             * the shader outputs. Test for this early. */
            if (inst->Dst[0].Register.File != TGSI_FILE_OUTPUT)
               break;
            /* Elimination of a MOV must have no visible effect on the
             * resulting shader: this means the MOV must not swizzle or
             * saturate, and its source must not have the negate or
             * absolute modifiers. */
            if (!etna_mov_check_no_swizzle(inst->Dst[0].Register, inst->Src[0].Register) ||
                inst->Instruction.Saturate || inst->Src[0].Register.Negate ||
                inst->Src[0].Register.Absolute)
               break;

            uint out_idx = inst->Dst[0].Register.Index;
            uint in_idx = inst->Src[0].Register.Index;
            /* assignment of temporary to output --
             * and the output doesn't yet have a native register assigned
             * and the last use of the temporary is this instruction
             * and the MOV does not do a swizzle
             */
            if (inst->Src[0].Register.File == TGSI_FILE_TEMPORARY &&
                !c->file[TGSI_FILE_OUTPUT].reg[out_idx].native.valid &&
                c->file[TGSI_FILE_TEMPORARY].reg[in_idx].last_use == inst_idx) {
               c->file[TGSI_FILE_OUTPUT].reg[out_idx].native =
                  c->file[TGSI_FILE_TEMPORARY].reg[in_idx].native;
               /* prevent temp from being re-used for the rest of the shader */
               c->file[TGSI_FILE_TEMPORARY].reg[in_idx].last_use = ETNA_MAX_TOKENS;
               /* mark this MOV instruction as a no-op */
               c->dead_inst[inst_idx] = true;
            }
            /* direct assignment of input to output --
             * and the input or output doesn't yet have a native register
             * assigned
             * and the output is only used in this instruction,
             * allocate a new register, and associate both input and output to
             * it
             * and the MOV does not do a swizzle
             */
            if (inst->Src[0].Register.File == TGSI_FILE_INPUT &&
                !c->file[TGSI_FILE_INPUT].reg[in_idx].native.valid &&
                !c->file[TGSI_FILE_OUTPUT].reg[out_idx].native.valid &&
                c->file[TGSI_FILE_OUTPUT].reg[out_idx].last_use == inst_idx &&
                c->file[TGSI_FILE_OUTPUT].reg[out_idx].first_use == inst_idx) {
               c->file[TGSI_FILE_OUTPUT].reg[out_idx].native =
                  c->file[TGSI_FILE_INPUT].reg[in_idx].native =
                     alloc_new_native_reg(c);
               /* mark this MOV instruction as a no-op */
               c->dead_inst[inst_idx] = true;
            }
         } break;
         default:;
         }
         inst_idx += 1;
      } break;
      }
   }

   tgsi_parse_free(&ctx);
}

/* Get a temporary to be used within one TGSI instruction.
 * The first time that this function is called the temporary will be allocated.
 * Each call to this function will return the same temporary.
 */
static struct etna_native_reg
etna_compile_get_inner_temp(struct etna_compile *c)
{
   int inner_temp = c->inner_temps;

   if (inner_temp < ETNA_MAX_INNER_TEMPS) {
      if (!c->inner_temp[inner_temp].valid)
         c->inner_temp[inner_temp] = alloc_new_native_reg(c);

      /* alloc_new_native_reg() handles lack of registers */
      c->inner_temps += 1;
   } else {
      BUG("Too many inner temporaries (%i) requested in one instruction",
          inner_temp + 1);
   }

   return c->inner_temp[inner_temp];
}

static struct etna_inst_dst
etna_native_to_dst(struct etna_native_reg native, unsigned comps)
{
   /* Can only assign to temporaries */
   assert(native.valid && !native.is_tex && native.rgroup == INST_RGROUP_TEMP);

   struct etna_inst_dst rv = {
      .comps = comps,
      .use = 1,
      .reg = native.id,
   };

   return rv;
}

static struct etna_inst_src
etna_native_to_src(struct etna_native_reg native, uint32_t swizzle)
{
   assert(native.valid && !native.is_tex);

   struct etna_inst_src rv = {
      .use = 1,
      .swiz = swizzle,
      .rgroup = native.rgroup,
      .reg = native.id,
      .amode = INST_AMODE_DIRECT,
   };

   return rv;
}

static inline struct etna_inst_src
negate(struct etna_inst_src src)
{
   src.neg = !src.neg;

   return src;
}

static inline struct etna_inst_src
absolute(struct etna_inst_src src)
{
   src.abs = 1;

   return src;
}

static inline struct etna_inst_src
swizzle(struct etna_inst_src src, unsigned swizzle)
{
   src.swiz = inst_swiz_compose(src.swiz, swizzle);

   return src;
}

/* Emit instruction and append it to program */
static void
emit_inst(struct etna_compile *c, struct etna_inst *inst)
{
   assert(c->inst_ptr <= ETNA_MAX_INSTRUCTIONS);

   /* Check for uniform conflicts (each instruction can only access one
    * uniform),
    * if detected, use an intermediate temporary */
   unsigned uni_rgroup = -1;
   unsigned uni_reg = -1;

   for (int src = 0; src < ETNA_NUM_SRC; ++src) {
      if (etna_rgroup_is_uniform(inst->src[src].rgroup)) {
         if (uni_reg == -1) { /* first unique uniform used */
            uni_rgroup = inst->src[src].rgroup;
            uni_reg = inst->src[src].reg;
         } else { /* second or later; check that it is a re-use */
            if (uni_rgroup != inst->src[src].rgroup ||
                uni_reg != inst->src[src].reg) {
               DBG_F(ETNA_DBG_COMPILER_MSGS, "perf warning: instruction that "
                                             "accesses different uniforms, "
                                             "need to generate extra MOV");
               struct etna_native_reg inner_temp = etna_compile_get_inner_temp(c);

               /* Generate move instruction to temporary */
               etna_assemble(&c->code[c->inst_ptr * 4], &(struct etna_inst) {
                  .opcode = INST_OPCODE_MOV,
                  .dst = etna_native_to_dst(inner_temp, INST_COMPS_X | INST_COMPS_Y |
                                                        INST_COMPS_Z | INST_COMPS_W),
                  .src[2] = inst->src[src]
               });

               c->inst_ptr++;

               /* Modify instruction to use temp register instead of uniform */
               inst->src[src].use = 1;
               inst->src[src].rgroup = INST_RGROUP_TEMP;
               inst->src[src].reg = inner_temp.id;
               inst->src[src].swiz = INST_SWIZ_IDENTITY; /* swizzling happens on MOV */
               inst->src[src].neg = 0; /* negation happens on MOV */
               inst->src[src].abs = 0; /* abs happens on MOV */
               inst->src[src].amode = 0; /* amode effects happen on MOV */
            }
         }
      }
   }

   /* Finally assemble the actual instruction */
   etna_assemble(&c->code[c->inst_ptr * 4], inst);
   c->inst_ptr++;
}

static unsigned int
etna_amode(struct tgsi_ind_register indirect)
{
   assert(indirect.File == TGSI_FILE_ADDRESS);
   assert(indirect.Index == 0);

   switch (indirect.Swizzle) {
   case TGSI_SWIZZLE_X:
      return INST_AMODE_ADD_A_X;
   case TGSI_SWIZZLE_Y:
      return INST_AMODE_ADD_A_Y;
   case TGSI_SWIZZLE_Z:
      return INST_AMODE_ADD_A_Z;
   case TGSI_SWIZZLE_W:
      return INST_AMODE_ADD_A_W;
   default:
      assert(!"Invalid swizzle");
   }
}

/* convert destination operand */
static struct etna_inst_dst
convert_dst(struct etna_compile *c, const struct tgsi_full_dst_register *in)
{
   struct etna_inst_dst rv = {
      /// XXX .amode
      .comps = in->Register.WriteMask,
   };

   if (in->Register.File == TGSI_FILE_ADDRESS) {
      assert(in->Register.Index == 0);
      rv.reg = in->Register.Index;
      rv.use = 0;
   } else {
      rv = etna_native_to_dst(etna_get_dst_reg(c, in->Register)->native,
                              in->Register.WriteMask);
   }

   if (in->Register.Indirect)
      rv.amode = etna_amode(in->Indirect);

   return rv;
}

/* convert texture operand */
static struct etna_inst_tex
convert_tex(struct etna_compile *c, const struct tgsi_full_src_register *in,
            const struct tgsi_instruction_texture *tex)
{
   struct etna_native_reg native_reg = etna_get_src_reg(c, in->Register)->native;
   struct etna_inst_tex rv = {
      // XXX .amode (to allow for an array of samplers?)
      .swiz = INST_SWIZ_IDENTITY
   };

   assert(native_reg.is_tex && native_reg.valid);
   rv.id = native_reg.id;

   return rv;
}

/* convert source operand */
static struct etna_inst_src
etna_create_src(const struct tgsi_full_src_register *tgsi,
                const struct etna_native_reg *native)
{
   const struct tgsi_src_register *reg = &tgsi->Register;
   struct etna_inst_src rv = {
      .use = 1,
      .swiz = INST_SWIZ(reg->SwizzleX, reg->SwizzleY, reg->SwizzleZ, reg->SwizzleW),
      .neg = reg->Negate,
      .abs = reg->Absolute,
      .rgroup = native->rgroup,
      .reg = native->id,
      .amode = INST_AMODE_DIRECT,
   };

   assert(native->valid && !native->is_tex);

   if (reg->Indirect)
      rv.amode = etna_amode(tgsi->Indirect);

   return rv;
}

static struct etna_inst_src
etna_mov_src_to_temp(struct etna_compile *c, struct etna_inst_src src,
                     struct etna_native_reg temp)
{
   struct etna_inst mov = { };

   mov.opcode = INST_OPCODE_MOV;
   mov.sat = 0;
   mov.dst = etna_native_to_dst(temp, INST_COMPS_X | INST_COMPS_Y |
                                      INST_COMPS_Z | INST_COMPS_W);
   mov.src[2] = src;
   emit_inst(c, &mov);

   src.swiz = INST_SWIZ_IDENTITY;
   src.neg = src.abs = 0;
   src.rgroup = temp.rgroup;
   src.reg = temp.id;

   return src;
}

static struct etna_inst_src
etna_mov_src(struct etna_compile *c, struct etna_inst_src src)
{
   struct etna_native_reg temp = etna_compile_get_inner_temp(c);

   return etna_mov_src_to_temp(c, src, temp);
}

static bool
etna_src_uniforms_conflict(struct etna_inst_src a, struct etna_inst_src b)
{
   return etna_rgroup_is_uniform(a.rgroup) &&
          etna_rgroup_is_uniform(b.rgroup) &&
          (a.rgroup != b.rgroup || a.reg != b.reg);
}

/* create a new label */
static struct etna_compile_label *
alloc_new_label(struct etna_compile *c)
{
   struct etna_compile_label label = {
      .inst_idx = -1, /* start by point to no specific instruction */
   };

   array_insert(c->labels, label);

   return &c->labels[c->labels_count - 1];
}

/* place label at current instruction pointer */
static void
label_place(struct etna_compile *c, struct etna_compile_label *label)
{
   label->inst_idx = c->inst_ptr;
}

/* mark label use at current instruction.
 * target of the label will be filled in in the marked instruction's src2.imm
 * slot as soon
 * as the value becomes known.
 */
static void
label_mark_use(struct etna_compile *c, struct etna_compile_label *label)
{
   assert(c->inst_ptr < ETNA_MAX_INSTRUCTIONS);
   c->lbl_usage[c->inst_ptr] = label;
}

/* walk the frame stack and return first frame with matching type */
static struct etna_compile_frame *
find_frame(struct etna_compile *c, enum etna_compile_frame_type type)
{
   for (unsigned sp = c->frame_sp; sp >= 0; sp--)
      if (c->frame_stack[sp].type == type)
         return &c->frame_stack[sp];

   assert(0);
   return NULL;
}

struct instr_translater {
   void (*fxn)(const struct instr_translater *t, struct etna_compile *c,
               const struct tgsi_full_instruction *inst,
               struct etna_inst_src *src);
   unsigned tgsi_opc;
   uint8_t opc;

   /* tgsi src -> etna src swizzle */
   int src[3];

   unsigned cond;
};

static void
trans_instr(const struct instr_translater *t, struct etna_compile *c,
            const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   const struct tgsi_opcode_info *info = tgsi_get_opcode_info(inst->Instruction.Opcode);
   struct etna_inst instr = { };

   instr.opcode = t->opc;
   instr.cond = t->cond;
   instr.sat = inst->Instruction.Saturate;

   assert(info->num_dst <= 1);
   if (info->num_dst)
      instr.dst = convert_dst(c, &inst->Dst[0]);

   assert(info->num_src <= ETNA_NUM_SRC);

   for (unsigned i = 0; i < info->num_src; i++) {
      int swizzle = t->src[i];

      assert(swizzle != -1);
      instr.src[swizzle] = src[i];
   }

   emit_inst(c, &instr);
}

static void
trans_min_max(const struct instr_translater *t, struct etna_compile *c,
              const struct tgsi_full_instruction *inst,
              struct etna_inst_src *src)
{
   emit_inst(c, &(struct etna_inst) {
      .opcode = INST_OPCODE_SELECT,
       .cond = t->cond,
       .sat = inst->Instruction.Saturate,
       .dst = convert_dst(c, &inst->Dst[0]),
       .src[0] = src[0],
       .src[1] = src[1],
       .src[2] = src[0],
    });
}

static void
trans_if(const struct instr_translater *t, struct etna_compile *c,
         const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   struct etna_compile_frame *f = &c->frame_stack[c->frame_sp++];
   struct etna_inst_src imm_0 = alloc_imm_f32(c, 0.0f);

   /* push IF to stack */
   f->type = ETNA_COMPILE_FRAME_IF;
   /* create "else" label */
   f->lbl_else = alloc_new_label(c);
   f->lbl_endif = NULL;

   /* We need to avoid the emit_inst() below becoming two instructions */
   if (etna_src_uniforms_conflict(src[0], imm_0))
      src[0] = etna_mov_src(c, src[0]);

   /* mark position in instruction stream of label reference so that it can be
    * filled in in next pass */
   label_mark_use(c, f->lbl_else);

   /* create conditional branch to label if src0 EQ 0 */
   emit_inst(c, &(struct etna_inst){
      .opcode = INST_OPCODE_BRANCH,
      .cond = INST_CONDITION_EQ,
      .src[0] = src[0],
      .src[1] = imm_0,
    /* imm is filled in later */
   });
}

static void
trans_else(const struct instr_translater *t, struct etna_compile *c,
           const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   assert(c->frame_sp > 0);
   struct etna_compile_frame *f = &c->frame_stack[c->frame_sp - 1];
   assert(f->type == ETNA_COMPILE_FRAME_IF);

   /* create "endif" label, and branch to endif label */
   f->lbl_endif = alloc_new_label(c);
   label_mark_use(c, f->lbl_endif);
   emit_inst(c, &(struct etna_inst) {
      .opcode = INST_OPCODE_BRANCH,
      .cond = INST_CONDITION_TRUE,
      /* imm is filled in later */
   });

   /* mark "else" label at this position in instruction stream */
   label_place(c, f->lbl_else);
}

static void
trans_endif(const struct instr_translater *t, struct etna_compile *c,
            const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   assert(c->frame_sp > 0);
   struct etna_compile_frame *f = &c->frame_stack[--c->frame_sp];
   assert(f->type == ETNA_COMPILE_FRAME_IF);

   /* assign "endif" or "else" (if no ELSE) label to current position in
    * instruction stream, pop IF */
   if (f->lbl_endif != NULL)
      label_place(c, f->lbl_endif);
   else
      label_place(c, f->lbl_else);
}

static void
trans_loop_bgn(const struct instr_translater *t, struct etna_compile *c,
               const struct tgsi_full_instruction *inst,
               struct etna_inst_src *src)
{
   struct etna_compile_frame *f = &c->frame_stack[c->frame_sp++];

   /* push LOOP to stack */
   f->type = ETNA_COMPILE_FRAME_LOOP;
   f->lbl_loop_bgn = alloc_new_label(c);
   f->lbl_loop_end = alloc_new_label(c);

   label_place(c, f->lbl_loop_bgn);
}

static void
trans_loop_end(const struct instr_translater *t, struct etna_compile *c,
               const struct tgsi_full_instruction *inst,
               struct etna_inst_src *src)
{
   assert(c->frame_sp > 0);
   struct etna_compile_frame *f = &c->frame_stack[--c->frame_sp];
   assert(f->type == ETNA_COMPILE_FRAME_LOOP);

   /* mark position in instruction stream of label reference so that it can be
    * filled in in next pass */
   label_mark_use(c, f->lbl_loop_bgn);

   /* create branch to loop_bgn label */
   emit_inst(c, &(struct etna_inst) {
      .opcode = INST_OPCODE_BRANCH,
      .cond = INST_CONDITION_TRUE,
      .src[0] = src[0],
      /* imm is filled in later */
   });

   label_place(c, f->lbl_loop_end);
}

static void
trans_brk(const struct instr_translater *t, struct etna_compile *c,
          const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   assert(c->frame_sp > 0);
   struct etna_compile_frame *f = find_frame(c, ETNA_COMPILE_FRAME_LOOP);

   /* mark position in instruction stream of label reference so that it can be
    * filled in in next pass */
   label_mark_use(c, f->lbl_loop_end);

   /* create branch to loop_end label */
   emit_inst(c, &(struct etna_inst) {
      .opcode = INST_OPCODE_BRANCH,
      .cond = INST_CONDITION_TRUE,
      .src[0] = src[0],
      /* imm is filled in later */
   });
}

static void
trans_cont(const struct instr_translater *t, struct etna_compile *c,
           const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   assert(c->frame_sp > 0);
   struct etna_compile_frame *f = find_frame(c, ETNA_COMPILE_FRAME_LOOP);

   /* mark position in instruction stream of label reference so that it can be
    * filled in in next pass */
   label_mark_use(c, f->lbl_loop_bgn);

   /* create branch to loop_end label */
   emit_inst(c, &(struct etna_inst) {
      .opcode = INST_OPCODE_BRANCH,
      .cond = INST_CONDITION_TRUE,
      .src[0] = src[0],
      /* imm is filled in later */
   });
}

static void
trans_deriv(const struct instr_translater *t, struct etna_compile *c,
            const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   emit_inst(c, &(struct etna_inst) {
      .opcode = t->opc,
      .sat = inst->Instruction.Saturate,
      .dst = convert_dst(c, &inst->Dst[0]),
      .src[0] = src[0],
      .src[2] = src[0],
   });
}

static void
trans_arl(const struct instr_translater *t, struct etna_compile *c,
          const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   struct etna_native_reg temp = etna_compile_get_inner_temp(c);
   struct etna_inst arl = { };
   struct etna_inst_dst dst;

   dst = etna_native_to_dst(temp, INST_COMPS_X | INST_COMPS_Y | INST_COMPS_Z |
                                  INST_COMPS_W);

   if (c->specs->has_sign_floor_ceil) {
      struct etna_inst floor = { };

      floor.opcode = INST_OPCODE_FLOOR;
      floor.src[2] = src[0];
      floor.dst = dst;

      emit_inst(c, &floor);
   } else {
      struct etna_inst floor[2] = { };

      floor[0].opcode = INST_OPCODE_FRC;
      floor[0].sat = inst->Instruction.Saturate;
      floor[0].dst = dst;
      floor[0].src[2] = src[0];

      floor[1].opcode = INST_OPCODE_ADD;
      floor[1].sat = inst->Instruction.Saturate;
      floor[1].dst = dst;
      floor[1].src[0] = src[0];
      floor[1].src[2].use = 1;
      floor[1].src[2].swiz = INST_SWIZ_IDENTITY;
      floor[1].src[2].neg = 1;
      floor[1].src[2].rgroup = temp.rgroup;
      floor[1].src[2].reg = temp.id;

      emit_inst(c, &floor[0]);
      emit_inst(c, &floor[1]);
   }

   arl.opcode = INST_OPCODE_MOVAR;
   arl.sat = inst->Instruction.Saturate;
   arl.dst = convert_dst(c, &inst->Dst[0]);
   arl.src[2] = etna_native_to_src(temp, INST_SWIZ_IDENTITY);

   emit_inst(c, &arl);
}

static void
trans_lrp(const struct instr_translater *t, struct etna_compile *c,
          const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   /* dst = src0 * src1 + (1 - src0) * src2
    *     => src0 * src1 - (src0 - 1) * src2
    *     => src0 * src1 - (src0 * src2 - src2)
    * MAD tTEMP.xyzw, tSRC0.xyzw, tSRC2.xyzw, -tSRC2.xyzw
    * MAD tDST.xyzw, tSRC0.xyzw, tSRC1.xyzw, -tTEMP.xyzw
    */
   struct etna_native_reg temp = etna_compile_get_inner_temp(c);
   if (etna_src_uniforms_conflict(src[0], src[1]) ||
       etna_src_uniforms_conflict(src[0], src[2])) {
      src[0] = etna_mov_src(c, src[0]);
   }

   struct etna_inst mad[2] = { };
   mad[0].opcode = INST_OPCODE_MAD;
   mad[0].sat = 0;
   mad[0].dst = etna_native_to_dst(temp, INST_COMPS_X | INST_COMPS_Y |
                                         INST_COMPS_Z | INST_COMPS_W);
   mad[0].src[0] = src[0];
   mad[0].src[1] = src[2];
   mad[0].src[2] = negate(src[2]);
   mad[1].opcode = INST_OPCODE_MAD;
   mad[1].sat = inst->Instruction.Saturate;
   mad[1].dst = convert_dst(c, &inst->Dst[0]), mad[1].src[0] = src[0];
   mad[1].src[1] = src[1];
   mad[1].src[2] = negate(etna_native_to_src(temp, INST_SWIZ_IDENTITY));

   emit_inst(c, &mad[0]);
   emit_inst(c, &mad[1]);
}

static void
trans_lit(const struct instr_translater *t, struct etna_compile *c,
          const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   /* SELECT.LT tmp._y__, 0, src.yyyy, 0
    *  - can be eliminated if src.y is a uniform and >= 0
    * SELECT.GT tmp.___w, 128, src.wwww, 128
    * SELECT.LT tmp.___w, -128, tmp.wwww, -128
    *  - can be eliminated if src.w is a uniform and fits clamp
    * LOG tmp.x, void, void, tmp.yyyy
    * MUL tmp.x, tmp.xxxx, tmp.wwww, void
    * LITP dst, undef, src.xxxx, tmp.xxxx
    */
   struct etna_native_reg inner_temp = etna_compile_get_inner_temp(c);
   struct etna_inst_src src_y = { };

   if (!etna_rgroup_is_uniform(src[0].rgroup)) {
      src_y = etna_native_to_src(inner_temp, SWIZZLE(Y, Y, Y, Y));

      struct etna_inst ins = { };
      ins.opcode = INST_OPCODE_SELECT;
      ins.cond = INST_CONDITION_LT;
      ins.dst = etna_native_to_dst(inner_temp, INST_COMPS_Y);
      ins.src[0] = ins.src[2] = alloc_imm_f32(c, 0.0);
      ins.src[1] = swizzle(src[0], SWIZZLE(Y, Y, Y, Y));
      emit_inst(c, &ins);
   } else if (uif(get_imm_u32(c, &src[0], 1)) < 0)
      src_y = alloc_imm_f32(c, 0.0);
   else
      src_y = swizzle(src[0], SWIZZLE(Y, Y, Y, Y));

   struct etna_inst_src src_w = { };

   if (!etna_rgroup_is_uniform(src[0].rgroup)) {
      src_w = etna_native_to_src(inner_temp, SWIZZLE(W, W, W, W));

      struct etna_inst ins = { };
      ins.opcode = INST_OPCODE_SELECT;
      ins.cond = INST_CONDITION_GT;
      ins.dst = etna_native_to_dst(inner_temp, INST_COMPS_W);
      ins.src[0] = ins.src[2] = alloc_imm_f32(c, 128.);
      ins.src[1] = swizzle(src[0], SWIZZLE(W, W, W, W));
      emit_inst(c, &ins);
      ins.cond = INST_CONDITION_LT;
      ins.src[0].neg = !ins.src[0].neg;
      ins.src[2].neg = !ins.src[2].neg;
      ins.src[1] = src_w;
      emit_inst(c, &ins);
   } else if (uif(get_imm_u32(c, &src[0], 3)) < -128.)
      src_w = alloc_imm_f32(c, -128.);
   else if (uif(get_imm_u32(c, &src[0], 3)) > 128.)
      src_w = alloc_imm_f32(c, 128.);
   else
      src_w = swizzle(src[0], SWIZZLE(W, W, W, W));

   struct etna_inst ins[3] = { };
   ins[0].opcode = INST_OPCODE_LOG;
   ins[0].dst = etna_native_to_dst(inner_temp, INST_COMPS_X);
   ins[0].src[2] = src_y;

   emit_inst(c, &ins[0]);
   emit_inst(c, &(struct etna_inst) {
      .opcode = INST_OPCODE_MUL,
      .sat = 0,
      .dst = etna_native_to_dst(inner_temp, INST_COMPS_X),
      .src[0] = etna_native_to_src(inner_temp, SWIZZLE(X, X, X, X)),
      .src[1] = src_w,
   });
   emit_inst(c, &(struct etna_inst) {
      .opcode = INST_OPCODE_LITP,
      .sat = 0,
      .dst = convert_dst(c, &inst->Dst[0]),
      .src[0] = swizzle(src[0], SWIZZLE(X, X, X, X)),
      .src[1] = swizzle(src[0], SWIZZLE(X, X, X, X)),
      .src[2] = etna_native_to_src(inner_temp, SWIZZLE(X, X, X, X)),
   });
}

static void
trans_ssg(const struct instr_translater *t, struct etna_compile *c,
          const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   if (c->specs->has_sign_floor_ceil) {
      emit_inst(c, &(struct etna_inst){
         .opcode = INST_OPCODE_SIGN,
         .sat = inst->Instruction.Saturate,
         .dst = convert_dst(c, &inst->Dst[0]),
         .src[2] = src[0],
      });
   } else {
      struct etna_native_reg temp = etna_compile_get_inner_temp(c);
      struct etna_inst ins[2] = { };

      ins[0].opcode = INST_OPCODE_SET;
      ins[0].cond = INST_CONDITION_NZ;
      ins[0].dst = etna_native_to_dst(temp, INST_COMPS_X | INST_COMPS_Y |
                                            INST_COMPS_Z | INST_COMPS_W);
      ins[0].src[0] = src[0];

      ins[1].opcode = INST_OPCODE_SELECT;
      ins[1].cond = INST_CONDITION_LZ;
      ins[1].sat = inst->Instruction.Saturate;
      ins[1].dst = convert_dst(c, &inst->Dst[0]);
      ins[1].src[0] = src[0];
      ins[1].src[2] = etna_native_to_src(temp, INST_SWIZ_IDENTITY);
      ins[1].src[1] = negate(ins[1].src[2]);

      emit_inst(c, &ins[0]);
      emit_inst(c, &ins[1]);
   }
}

static void
trans_trig(const struct instr_translater *t, struct etna_compile *c,
           const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   if (c->specs->has_sin_cos_sqrt) {
      /* TGSI lowering should deal with SCS */
      assert(inst->Instruction.Opcode != TGSI_OPCODE_SCS);

      struct etna_native_reg temp = etna_compile_get_inner_temp(c);
      /* add divide by PI/2, using a temp register. GC2000
       * fails with src==dst for the trig instruction. */
      emit_inst(c, &(struct etna_inst) {
         .opcode = INST_OPCODE_MUL,
         .sat = 0,
         .dst = etna_native_to_dst(temp, INST_COMPS_X | INST_COMPS_Y |
                                         INST_COMPS_Z | INST_COMPS_W),
         .src[0] = src[0], /* any swizzling happens here */
         .src[1] = alloc_imm_f32(c, 2.0f / M_PI),
      });
      emit_inst(c, &(struct etna_inst) {
         .opcode = inst->Instruction.Opcode == TGSI_OPCODE_COS
                    ? INST_OPCODE_COS
                    : INST_OPCODE_SIN,
         .sat = inst->Instruction.Saturate,
         .dst = convert_dst(c, &inst->Dst[0]),
         .src[2] = etna_native_to_src(temp, INST_SWIZ_IDENTITY),
      });
   } else {
      /* Implement Nick's fast sine/cosine. Taken from:
       * http://forum.devmaster.net/t/fast-and-accurate-sine-cosine/9648
       * A=(1/2*PI 0 1/2*PI 0) B=(0.75 0 0.5 0) C=(-4 4 X X)
       *  MAD t.x_zw, src.xxxx, A, B
       *  FRC t.x_z_, void, void, t.xwzw
       *  MAD t.x_z_, t.xwzw, 2, -1
       *  MUL t._y__, t.wzww, |t.wzww|, void  (for sin/scs)
       *  DP3 t.x_z_, t.zyww, C, void         (for sin)
       *  DP3 t.__z_, t.zyww, C, void         (for scs)
       *  MUL t._y__, t.wxww, |t.wxww|, void  (for cos/scs)
       *  DP3 t.x_z_, t.xyww, C, void         (for cos)
       *  DP3 t.x___, t.xyww, C, void         (for scs)
       *  MAD t._y_w, t,xxzz, |t.xxzz|, -t.xxzz
       *  MAD dst, t.ywyw, .2225, t.xzxz
       *
       * TODO: we don't set dst.zw correctly for SCS.
       */
      struct etna_inst *p, ins[9] = { };
      struct etna_native_reg t0 = etna_compile_get_inner_temp(c);
      struct etna_inst_src t0s = etna_native_to_src(t0, INST_SWIZ_IDENTITY);
      struct etna_inst_src sincos[3], in = src[0];
      sincos[0] = etna_imm_vec4f(c, sincos_const[0]);
      sincos[1] = etna_imm_vec4f(c, sincos_const[1]);

      /* A uniform source will cause the inner temp limit to
       * be exceeded.  Explicitly deal with that scenario.
       */
      if (etna_rgroup_is_uniform(src[0].rgroup)) {
         struct etna_inst ins = { };
         ins.opcode = INST_OPCODE_MOV;
         ins.dst = etna_native_to_dst(t0, INST_COMPS_X);
         ins.src[2] = in;
         emit_inst(c, &ins);
         in = t0s;
      }

      ins[0].opcode = INST_OPCODE_MAD;
      ins[0].dst = etna_native_to_dst(t0, INST_COMPS_X | INST_COMPS_Z | INST_COMPS_W);
      ins[0].src[0] = swizzle(in, SWIZZLE(X, X, X, X));
      ins[0].src[1] = swizzle(sincos[1], SWIZZLE(X, W, X, W)); /* 1/2*PI */
      ins[0].src[2] = swizzle(sincos[1], SWIZZLE(Y, W, Z, W)); /* 0.75, 0, 0.5, 0 */

      ins[1].opcode = INST_OPCODE_FRC;
      ins[1].dst = etna_native_to_dst(t0, INST_COMPS_X | INST_COMPS_Z);
      ins[1].src[2] = swizzle(t0s, SWIZZLE(X, W, Z, W));

      ins[2].opcode = INST_OPCODE_MAD;
      ins[2].dst = etna_native_to_dst(t0, INST_COMPS_X | INST_COMPS_Z);
      ins[2].src[0] = swizzle(t0s, SWIZZLE(X, W, Z, W));
      ins[2].src[1] = swizzle(sincos[0], SWIZZLE(X, X, X, X)); /* 2 */
      ins[2].src[2] = swizzle(sincos[0], SWIZZLE(Y, Y, Y, Y)); /* -1 */

      unsigned mul_swiz, dp3_swiz;
      if (inst->Instruction.Opcode == TGSI_OPCODE_SIN) {
         mul_swiz = SWIZZLE(W, Z, W, W);
         dp3_swiz = SWIZZLE(Z, Y, W, W);
      } else {
         mul_swiz = SWIZZLE(W, X, W, W);
         dp3_swiz = SWIZZLE(X, Y, W, W);
      }

      ins[3].opcode = INST_OPCODE_MUL;
      ins[3].dst = etna_native_to_dst(t0, INST_COMPS_Y);
      ins[3].src[0] = swizzle(t0s, mul_swiz);
      ins[3].src[1] = absolute(ins[3].src[0]);

      ins[4].opcode = INST_OPCODE_DP3;
      ins[4].dst = etna_native_to_dst(t0, INST_COMPS_X | INST_COMPS_Z);
      ins[4].src[0] = swizzle(t0s, dp3_swiz);
      ins[4].src[1] = swizzle(sincos[0], SWIZZLE(Z, W, W, W));

      if (inst->Instruction.Opcode == TGSI_OPCODE_SCS) {
         ins[5] = ins[3];
         ins[6] = ins[4];
         ins[4].dst.comps = INST_COMPS_X;
         ins[6].dst.comps = INST_COMPS_Z;
         ins[5].src[0] = swizzle(t0s, SWIZZLE(W, Z, W, W));
         ins[6].src[0] = swizzle(t0s, SWIZZLE(Z, Y, W, W));
         ins[5].src[1] = absolute(ins[5].src[0]);
         p = &ins[7];
      } else {
         p = &ins[5];
      }

      p->opcode = INST_OPCODE_MAD;
      p->dst = etna_native_to_dst(t0, INST_COMPS_Y | INST_COMPS_W);
      p->src[0] = swizzle(t0s, SWIZZLE(X, X, Z, Z));
      p->src[1] = absolute(p->src[0]);
      p->src[2] = negate(p->src[0]);

      p++;
      p->opcode = INST_OPCODE_MAD;
      p->sat = inst->Instruction.Saturate;
      p->dst = convert_dst(c, &inst->Dst[0]),
      p->src[0] = swizzle(t0s, SWIZZLE(Y, W, Y, W));
      p->src[1] = alloc_imm_f32(c, 0.2225);
      p->src[2] = swizzle(t0s, SWIZZLE(X, Z, X, Z));

      for (int i = 0; &ins[i] <= p; i++)
         emit_inst(c, &ins[i]);
   }
}

static void
trans_dph(const struct instr_translater *t, struct etna_compile *c,
          const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   /*
   DP3 tmp.xyzw, src0.xyzw, src1,xyzw, void
   ADD dst.xyzw, tmp.xyzw, void, src1.wwww
   */
   struct etna_native_reg temp = etna_compile_get_inner_temp(c);
   struct etna_inst ins[2] = { };

   ins[0].opcode = INST_OPCODE_DP3;
   ins[0].dst = etna_native_to_dst(temp, INST_COMPS_X | INST_COMPS_Y |
                                         INST_COMPS_Z | INST_COMPS_W);
   ins[0].src[0] = src[0];
   ins[0].src[1] = src[1];

   ins[1].opcode = INST_OPCODE_ADD;
   ins[1].sat = inst->Instruction.Saturate;
   ins[1].dst = convert_dst(c, &inst->Dst[0]);
   ins[1].src[0] = etna_native_to_src(temp, INST_SWIZ_IDENTITY);
   ins[1].src[2] = swizzle(src[1], SWIZZLE(W, W, W, W));

   emit_inst(c, &ins[0]);
   emit_inst(c, &ins[1]);
}

static void
trans_sampler(const struct instr_translater *t, struct etna_compile *c,
              const struct tgsi_full_instruction *inst,
              struct etna_inst_src *src)
{
   /* There is no native support for GL texture rectangle coordinates, so
    * we have to rescale from ([0, width], [0, height]) to ([0, 1], [0, 1]). */
   if (inst->Texture.Texture == TGSI_TEXTURE_RECT) {
      uint32_t unit = inst->Src[1].Register.Index;
      struct etna_inst ins[2] = { };
      struct etna_native_reg temp = etna_compile_get_inner_temp(c);

      ins[0].opcode = INST_OPCODE_MUL;
      ins[0].dst = etna_native_to_dst(temp, INST_COMPS_X);
      ins[0].src[0] = src[0];
      ins[0].src[1] = alloc_imm(c, ETNA_IMMEDIATE_TEXRECT_SCALE_X, unit);

      ins[1].opcode = INST_OPCODE_MUL;
      ins[1].dst = etna_native_to_dst(temp, INST_COMPS_Y);
      ins[1].src[0] = src[0];
      ins[1].src[1] = alloc_imm(c, ETNA_IMMEDIATE_TEXRECT_SCALE_Y, unit);

      emit_inst(c, &ins[0]);
      emit_inst(c, &ins[1]);

      src[0] = etna_native_to_src(temp, INST_SWIZ_IDENTITY); /* temp.xyzw */
   }

   switch (inst->Instruction.Opcode) {
   case TGSI_OPCODE_TEX:
      emit_inst(c, &(struct etna_inst) {
         .opcode = INST_OPCODE_TEXLD,
         .sat = 0,
         .dst = convert_dst(c, &inst->Dst[0]),
         .tex = convert_tex(c, &inst->Src[1], &inst->Texture),
         .src[0] = src[0],
      });
      break;

   case TGSI_OPCODE_TXB:
      emit_inst(c, &(struct etna_inst) {
         .opcode = INST_OPCODE_TEXLDB,
         .sat = 0,
         .dst = convert_dst(c, &inst->Dst[0]),
         .tex = convert_tex(c, &inst->Src[1], &inst->Texture),
         .src[0] = src[0],
      });
      break;

   case TGSI_OPCODE_TXL:
      emit_inst(c, &(struct etna_inst) {
         .opcode = INST_OPCODE_TEXLDL,
         .sat = 0,
         .dst = convert_dst(c, &inst->Dst[0]),
         .tex = convert_tex(c, &inst->Src[1], &inst->Texture),
         .src[0] = src[0],
      });
      break;

   case TGSI_OPCODE_TXP: { /* divide src.xyz by src.w */
      struct etna_native_reg temp = etna_compile_get_inner_temp(c);

      emit_inst(c, &(struct etna_inst) {
         .opcode = INST_OPCODE_RCP,
         .sat = 0,
         .dst = etna_native_to_dst(temp, INST_COMPS_W), /* tmp.w */
         .src[2] = swizzle(src[0], SWIZZLE(W, W, W, W)),
      });
      emit_inst(c, &(struct etna_inst) {
         .opcode = INST_OPCODE_MUL,
         .sat = 0,
         .dst = etna_native_to_dst(temp, INST_COMPS_X | INST_COMPS_Y |
                                         INST_COMPS_Z), /* tmp.xyz */
         .src[0] = etna_native_to_src(temp, SWIZZLE(W, W, W, W)),
         .src[1] = src[0], /* src.xyzw */
      });
      emit_inst(c, &(struct etna_inst) {
         .opcode = INST_OPCODE_TEXLD,
         .sat = 0,
         .dst = convert_dst(c, &inst->Dst[0]),
         .tex = convert_tex(c, &inst->Src[1], &inst->Texture),
         .src[0] = etna_native_to_src(temp, INST_SWIZ_IDENTITY), /* tmp.xyzw */
      });
   } break;

   default:
      BUG("Unhandled instruction %s",
          tgsi_get_opcode_name(inst->Instruction.Opcode));
      assert(0);
      break;
   }
}

static void
trans_dummy(const struct instr_translater *t, struct etna_compile *c,
            const struct tgsi_full_instruction *inst, struct etna_inst_src *src)
{
   /* nothing to do */
}

static const struct instr_translater translaters[TGSI_OPCODE_LAST] = {
#define INSTR(n, f, ...) \
   [TGSI_OPCODE_##n] = {.fxn = (f), .tgsi_opc = TGSI_OPCODE_##n, ##__VA_ARGS__}

   INSTR(MOV, trans_instr, .opc = INST_OPCODE_MOV, .src = {2, -1, -1}),
   INSTR(RCP, trans_instr, .opc = INST_OPCODE_RCP, .src = {2, -1, -1}),
   INSTR(RSQ, trans_instr, .opc = INST_OPCODE_RSQ, .src = {2, -1, -1}),
   INSTR(MUL, trans_instr, .opc = INST_OPCODE_MUL, .src = {0, 1, -1}),
   INSTR(ADD, trans_instr, .opc = INST_OPCODE_ADD, .src = {0, 2, -1}),
   INSTR(DP3, trans_instr, .opc = INST_OPCODE_DP3, .src = {0, 1, -1}),
   INSTR(DP4, trans_instr, .opc = INST_OPCODE_DP4, .src = {0, 1, -1}),
   INSTR(DST, trans_instr, .opc = INST_OPCODE_DST, .src = {0, 1, -1}),
   INSTR(MAD, trans_instr, .opc = INST_OPCODE_MAD, .src = {0, 1, 2}),
   INSTR(EX2, trans_instr, .opc = INST_OPCODE_EXP, .src = {2, -1, -1}),
   INSTR(LG2, trans_instr, .opc = INST_OPCODE_LOG, .src = {2, -1, -1}),
   INSTR(SQRT, trans_instr, .opc = INST_OPCODE_SQRT, .src = {2, -1, -1}),
   INSTR(FRC, trans_instr, .opc = INST_OPCODE_FRC, .src = {2, -1, -1}),
   INSTR(CEIL, trans_instr, .opc = INST_OPCODE_CEIL, .src = {2, -1, -1}),
   INSTR(FLR, trans_instr, .opc = INST_OPCODE_FLOOR, .src = {2, -1, -1}),
   INSTR(CMP, trans_instr, .opc = INST_OPCODE_SELECT, .src = {0, 1, 2}, .cond = INST_CONDITION_LZ),

   INSTR(KILL, trans_instr, .opc = INST_OPCODE_TEXKILL),
   INSTR(KILL_IF, trans_instr, .opc = INST_OPCODE_TEXKILL, .src = {0, -1, -1}, .cond = INST_CONDITION_LZ),

   INSTR(DDX, trans_deriv, .opc = INST_OPCODE_DSX),
   INSTR(DDY, trans_deriv, .opc = INST_OPCODE_DSY),

   INSTR(IF, trans_if),
   INSTR(ELSE, trans_else),
   INSTR(ENDIF, trans_endif),

   INSTR(BGNLOOP, trans_loop_bgn),
   INSTR(ENDLOOP, trans_loop_end),
   INSTR(BRK, trans_brk),
   INSTR(CONT, trans_cont),

   INSTR(MIN, trans_min_max, .opc = INST_OPCODE_SELECT, .cond = INST_CONDITION_GT),
   INSTR(MAX, trans_min_max, .opc = INST_OPCODE_SELECT, .cond = INST_CONDITION_LT),

   INSTR(ARL, trans_arl),
   INSTR(LRP, trans_lrp),
   INSTR(LIT, trans_lit),
   INSTR(SSG, trans_ssg),
   INSTR(DPH, trans_dph),

   INSTR(SIN, trans_trig),
   INSTR(COS, trans_trig),
   INSTR(SCS, trans_trig),

   INSTR(SLT, trans_instr, .opc = INST_OPCODE_SET, .src = {0, 1, -1}, .cond = INST_CONDITION_LT),
   INSTR(SGE, trans_instr, .opc = INST_OPCODE_SET, .src = {0, 1, -1}, .cond = INST_CONDITION_GE),
   INSTR(SEQ, trans_instr, .opc = INST_OPCODE_SET, .src = {0, 1, -1}, .cond = INST_CONDITION_EQ),
   INSTR(SGT, trans_instr, .opc = INST_OPCODE_SET, .src = {0, 1, -1}, .cond = INST_CONDITION_GT),
   INSTR(SLE, trans_instr, .opc = INST_OPCODE_SET, .src = {0, 1, -1}, .cond = INST_CONDITION_LE),
   INSTR(SNE, trans_instr, .opc = INST_OPCODE_SET, .src = {0, 1, -1}, .cond = INST_CONDITION_NE),

   INSTR(TEX, trans_sampler),
   INSTR(TXB, trans_sampler),
   INSTR(TXL, trans_sampler),
   INSTR(TXP, trans_sampler),

   INSTR(NOP, trans_dummy),
   INSTR(END, trans_dummy),
};

/* Pass -- compile instructions */
static void
etna_compile_pass_generate_code(struct etna_compile *c)
{
   struct tgsi_parse_context ctx = { };
   unsigned status = tgsi_parse_init(&ctx, c->tokens);
   assert(status == TGSI_PARSE_OK);

   int inst_idx = 0;
   while (!tgsi_parse_end_of_tokens(&ctx)) {
      const struct tgsi_full_instruction *inst = 0;

      /* No inner temps used yet for this instruction, clear counter */
      c->inner_temps = 0;

      tgsi_parse_token(&ctx);

      switch (ctx.FullToken.Token.Type) {
      case TGSI_TOKEN_TYPE_INSTRUCTION:
         /* iterate over operands */
         inst = &ctx.FullToken.FullInstruction;
         if (c->dead_inst[inst_idx]) { /* skip dead instructions */
            inst_idx++;
            continue;
         }

         /* Lookup the TGSI information and generate the source arguments */
         struct etna_inst_src src[ETNA_NUM_SRC];
         memset(src, 0, sizeof(src));

         const struct tgsi_opcode_info *tgsi = tgsi_get_opcode_info(inst->Instruction.Opcode);

         for (int i = 0; i < tgsi->num_src && i < ETNA_NUM_SRC; i++) {
            const struct tgsi_full_src_register *reg = &inst->Src[i];
            const struct etna_native_reg *n = &etna_get_src_reg(c, reg->Register)->native;

            if (!n->valid || n->is_tex)
               continue;

            src[i] = etna_create_src(reg, n);
         }

         const unsigned opc = inst->Instruction.Opcode;
         const struct instr_translater *t = &translaters[opc];

         if (t->fxn) {
            t->fxn(t, c, inst, src);

            inst_idx += 1;
         } else {
            BUG("Unhandled instruction %s", tgsi_get_opcode_name(opc));
            assert(0);
         }
         break;
      }
   }
   tgsi_parse_free(&ctx);
}

/* Look up register by semantic */
static struct etna_reg_desc *
find_decl_by_semantic(struct etna_compile *c, uint file, uint name, uint index)
{
   for (int idx = 0; idx < c->file[file].reg_size; ++idx) {
      struct etna_reg_desc *reg = &c->file[file].reg[idx];

      if (reg->semantic.Name == name && reg->semantic.Index == index)
         return reg;
   }

   return NULL; /* not found */
}

/** Add ADD and MUL instruction to bring Z/W to 0..1 if -1..1 if needed:
 * - this is a vertex shader
 * - and this is an older GPU
 */
static void
etna_compile_add_z_div_if_needed(struct etna_compile *c)
{
   if (c->info.processor == PIPE_SHADER_VERTEX && c->specs->vs_need_z_div) {
      /* find position out */
      struct etna_reg_desc *pos_reg =
         find_decl_by_semantic(c, TGSI_FILE_OUTPUT, TGSI_SEMANTIC_POSITION, 0);

      if (pos_reg != NULL) {
         /*
          * ADD tX.__z_, tX.zzzz, void, tX.wwww
          * MUL tX.__z_, tX.zzzz, 0.5, void
         */
         emit_inst(c, &(struct etna_inst) {
            .opcode = INST_OPCODE_ADD,
            .dst = etna_native_to_dst(pos_reg->native, INST_COMPS_Z),
            .src[0] = etna_native_to_src(pos_reg->native, SWIZZLE(Z, Z, Z, Z)),
            .src[2] = etna_native_to_src(pos_reg->native, SWIZZLE(W, W, W, W)),
         });
         emit_inst(c, &(struct etna_inst) {
            .opcode = INST_OPCODE_MUL,
            .dst = etna_native_to_dst(pos_reg->native, INST_COMPS_Z),
            .src[0] = etna_native_to_src(pos_reg->native, SWIZZLE(Z, Z, Z, Z)),
            .src[1] = alloc_imm_f32(c, 0.5f),
         });
      }
   }
}

/** add a NOP to the shader if
 * a) the shader is empty
 * or
 * b) there is a label at the end of the shader
 */
static void
etna_compile_add_nop_if_needed(struct etna_compile *c)
{
   bool label_at_last_inst = false;

   for (int idx = 0; idx < c->labels_count; ++idx) {
      if (c->labels[idx].inst_idx == c->inst_ptr)
         label_at_last_inst = true;

   }

   if (c->inst_ptr == 0 || label_at_last_inst)
      emit_inst(c, &(struct etna_inst){.opcode = INST_OPCODE_NOP});
}

static void
assign_uniforms(struct etna_compile_file *file, unsigned base)
{
   for (int idx = 0; idx < file->reg_size; ++idx) {
      file->reg[idx].native.valid = 1;
      file->reg[idx].native.rgroup = INST_RGROUP_UNIFORM_0;
      file->reg[idx].native.id = base + idx;
   }
}

/* Allocate CONST and IMM to native ETNA_RGROUP_UNIFORM(x).
 * CONST must be consecutive as const buffers are supposed to be consecutive,
 * and before IMM, as this is
 * more convenient because is possible for the compilation process itself to
 * generate extra
 * immediates for constants such as pi, one, zero.
 */
static void
assign_constants_and_immediates(struct etna_compile *c)
{
   assign_uniforms(&c->file[TGSI_FILE_CONSTANT], 0);
   /* immediates start after the constants */
   c->imm_base = c->file[TGSI_FILE_CONSTANT].reg_size * 4;
   assign_uniforms(&c->file[TGSI_FILE_IMMEDIATE], c->imm_base / 4);
   DBG_F(ETNA_DBG_COMPILER_MSGS, "imm base: %i size: %i", c->imm_base,
         c->imm_size);
}

/* Assign declared samplers to native texture units */
static void
assign_texture_units(struct etna_compile *c)
{
   uint tex_base = 0;

   if (c->info.processor == PIPE_SHADER_VERTEX)
      tex_base = c->specs->vertex_sampler_offset;

   for (int idx = 0; idx < c->file[TGSI_FILE_SAMPLER].reg_size; ++idx) {
      c->file[TGSI_FILE_SAMPLER].reg[idx].native.valid = 1;
      c->file[TGSI_FILE_SAMPLER].reg[idx].native.is_tex = 1; // overrides rgroup
      c->file[TGSI_FILE_SAMPLER].reg[idx].native.id = tex_base + idx;
   }
}

/* Additional pass to fill in branch targets. This pass should be last
 * as no instruction reordering or removing/addition can be done anymore
 * once the branch targets are computed.
 */
static void
etna_compile_fill_in_labels(struct etna_compile *c)
{
   for (int idx = 0; idx < c->inst_ptr; ++idx) {
      if (c->lbl_usage[idx])
         etna_assemble_set_imm(&c->code[idx * 4], c->lbl_usage[idx]->inst_idx);
   }
}

/* compare two etna_native_reg structures, return true if equal */
static bool
cmp_etna_native_reg(const struct etna_native_reg to,
                    const struct etna_native_reg from)
{
   return to.valid == from.valid && to.is_tex == from.is_tex &&
          to.rgroup == from.rgroup && to.id == from.id;
}

/* go through all declarations and swap native registers *to* and *from* */
static void
swap_native_registers(struct etna_compile *c, const struct etna_native_reg to,
                      const struct etna_native_reg from)
{
   if (cmp_etna_native_reg(from, to))
      return; /* Nothing to do */

   for (int idx = 0; idx < c->total_decls; ++idx) {
      if (cmp_etna_native_reg(c->decl[idx].native, from)) {
         c->decl[idx].native = to;
      } else if (cmp_etna_native_reg(c->decl[idx].native, to)) {
         c->decl[idx].native = from;
      }
   }
}

/* For PS we need to permute so that inputs are always in temporary 0..N-1.
 * Semantic POS is always t0. If that semantic is not used, avoid t0.
 */
static void
permute_ps_inputs(struct etna_compile *c)
{
   /* Special inputs:
    * gl_FragCoord  VARYING_SLOT_POS   TGSI_SEMANTIC_POSITION
    * gl_PointCoord VARYING_SLOT_PNTC  TGSI_SEMANTIC_PCOORD
    */
   uint native_idx = 1;

   for (int idx = 0; idx < c->file[TGSI_FILE_INPUT].reg_size; ++idx) {
      struct etna_reg_desc *reg = &c->file[TGSI_FILE_INPUT].reg[idx];
      uint input_id;
      assert(reg->has_semantic);

      if (!reg->active || reg->semantic.Name == TGSI_SEMANTIC_POSITION)
         continue;

      input_id = native_idx++;
      swap_native_registers(c, etna_native_temp(input_id),
                            c->file[TGSI_FILE_INPUT].reg[idx].native);
   }

   c->num_varyings = native_idx - 1;

   if (native_idx > c->next_free_native)
      c->next_free_native = native_idx;
}

/* fill in ps inputs into shader object */
static void
fill_in_ps_inputs(struct etna_shader *sobj, struct etna_compile *c)
{
   struct etna_shader_io_file *sf = &sobj->infile;

   sf->num_reg = 0;

   for (int idx = 0; idx < c->file[TGSI_FILE_INPUT].reg_size; ++idx) {
      struct etna_reg_desc *reg = &c->file[TGSI_FILE_INPUT].reg[idx];

      if (reg->native.id > 0) {
         assert(sf->num_reg < ETNA_NUM_INPUTS);
         sf->reg[sf->num_reg].reg = reg->native.id;
         sf->reg[sf->num_reg].semantic = reg->semantic;
         /* convert usage mask to number of components (*=wildcard)
          *   .r    (0..1)  -> 1 component
          *   .*g   (2..3)  -> 2 component
          *   .**b  (4..7)  -> 3 components
          *   .***a (8..15) -> 4 components
          */
         sf->reg[sf->num_reg].num_components = util_last_bit(reg->usage_mask);
         sf->num_reg++;
      }
   }

   assert(sf->num_reg == c->num_varyings);
   sobj->input_count_unk8 = 31; /* XXX what is this */
}

/* fill in output mapping for ps into shader object */
static void
fill_in_ps_outputs(struct etna_shader *sobj, struct etna_compile *c)
{
   sobj->outfile.num_reg = 0;

   for (int idx = 0; idx < c->file[TGSI_FILE_OUTPUT].reg_size; ++idx) {
      struct etna_reg_desc *reg = &c->file[TGSI_FILE_OUTPUT].reg[idx];

      switch (reg->semantic.Name) {
      case TGSI_SEMANTIC_COLOR: /* FRAG_RESULT_COLOR */
         sobj->ps_color_out_reg = reg->native.id;
         break;
      case TGSI_SEMANTIC_POSITION: /* FRAG_RESULT_DEPTH */
         sobj->ps_depth_out_reg = reg->native.id; /* =always native reg 0, only z component should be assigned */
         break;
      default:
         assert(0); /* only outputs supported are COLOR and POSITION at the moment */
      }
   }
}

/* fill in inputs for vs into shader object */
static void
fill_in_vs_inputs(struct etna_shader *sobj, struct etna_compile *c)
{
   struct etna_shader_io_file *sf = &sobj->infile;

   sf->num_reg = 0;
   for (int idx = 0; idx < c->file[TGSI_FILE_INPUT].reg_size; ++idx) {
      struct etna_reg_desc *reg = &c->file[TGSI_FILE_INPUT].reg[idx];
      assert(sf->num_reg < ETNA_NUM_INPUTS);
      /* XXX exclude inputs with special semantics such as gl_frontFacing */
      sf->reg[sf->num_reg].reg = reg->native.id;
      sf->reg[sf->num_reg].semantic = reg->semantic;
      sf->reg[sf->num_reg].num_components = util_last_bit(reg->usage_mask);
      sf->num_reg++;
   }

   sobj->input_count_unk8 = (sf->num_reg + 19) / 16; /* XXX what is this */
}

/* build two-level output index [Semantic][Index] for fast linking */
static void
build_output_index(struct etna_shader *sobj)
{
   int total = 0;
   int offset = 0;

   for (int name = 0; name < TGSI_SEMANTIC_COUNT; ++name)
      total += sobj->output_count_per_semantic[name];

   sobj->output_per_semantic_list = CALLOC(total, sizeof(struct etna_shader_inout *));

   for (int name = 0; name < TGSI_SEMANTIC_COUNT; ++name) {
      sobj->output_per_semantic[name] = &sobj->output_per_semantic_list[offset];
      offset += sobj->output_count_per_semantic[name];
   }

   for (int idx = 0; idx < sobj->outfile.num_reg; ++idx) {
      sobj->output_per_semantic[sobj->outfile.reg[idx].semantic.Name]
                               [sobj->outfile.reg[idx].semantic.Index] =
         &sobj->outfile.reg[idx];
   }
}

/* fill in outputs for vs into shader object */
static void
fill_in_vs_outputs(struct etna_shader *sobj, struct etna_compile *c)
{
   struct etna_shader_io_file *sf = &sobj->outfile;

   sf->num_reg = 0;
   for (int idx = 0; idx < c->file[TGSI_FILE_OUTPUT].reg_size; ++idx) {
      struct etna_reg_desc *reg = &c->file[TGSI_FILE_OUTPUT].reg[idx];
      assert(sf->num_reg < ETNA_NUM_INPUTS);

      switch (reg->semantic.Name) {
      case TGSI_SEMANTIC_POSITION:
         sobj->vs_pos_out_reg = reg->native.id;
         break;
      case TGSI_SEMANTIC_PSIZE:
         sobj->vs_pointsize_out_reg = reg->native.id;
         break;
      default:
         sf->reg[sf->num_reg].reg = reg->native.id;
         sf->reg[sf->num_reg].semantic = reg->semantic;
         sf->reg[sf->num_reg].num_components = 4; // XXX reg->num_components;
         sf->num_reg++;
         sobj->output_count_per_semantic[reg->semantic.Name] =
            MAX2(reg->semantic.Index + 1,
                 sobj->output_count_per_semantic[reg->semantic.Name]);
      }
   }

   /* build two-level index for linking */
   build_output_index(sobj);

   /* fill in "mystery meat" load balancing value. This value determines how
    * work is scheduled between VS and PS
    * in the unified shader architecture. More precisely, it is determined from
    * the number of VS outputs, as well as chip-specific
    * vertex output buffer size, vertex cache size, and the number of shader
    * cores.
    *
    * XXX this is a conservative estimate, the "optimal" value is only known for
    * sure at link time because some
    * outputs may be unused and thus unmapped. Then again, in the general use
    * case with GLSL the vertex and fragment
    * shaders are linked already before submitting to Gallium, thus all outputs
    * are used.
    */
   int half_out = (c->file[TGSI_FILE_OUTPUT].reg_size + 1) / 2;
   assert(half_out);

   uint32_t b = ((20480 / (c->specs->vertex_output_buffer_size -
                           2 * half_out * c->specs->vertex_cache_size)) +
                 9) /
                10;
   uint32_t a = (b + 256 / (c->specs->shader_core_count * half_out)) / 2;
   sobj->vs_load_balancing = VIVS_VS_LOAD_BALANCING_A(MIN2(a, 255)) |
                             VIVS_VS_LOAD_BALANCING_B(MIN2(b, 255)) |
                             VIVS_VS_LOAD_BALANCING_C(0x3f) |
                             VIVS_VS_LOAD_BALANCING_D(0x0f);
}

static bool
etna_compile_check_limits(struct etna_compile *c)
{
   int max_uniforms = (c->info.processor == PIPE_SHADER_VERTEX)
                         ? c->specs->max_vs_uniforms
                         : c->specs->max_ps_uniforms;
   /* round up number of uniforms, including immediates, in units of four */
   int num_uniforms = c->imm_base / 4 + (c->imm_size + 3) / 4;

   if (c->inst_ptr > c->specs->max_instructions) {
      DBG("Number of instructions (%d) exceeds maximum %d", c->inst_ptr,
          c->specs->max_instructions);
      return false;
   }

   if (c->next_free_native > c->specs->max_registers) {
      DBG("Number of registers (%d) exceeds maximum %d", c->next_free_native,
          c->specs->max_registers);
      return false;
   }

   if (num_uniforms > max_uniforms) {
      DBG("Number of uniforms (%d) exceeds maximum %d", num_uniforms,
          max_uniforms);
      return false;
   }

   if (c->num_varyings > c->specs->max_varyings) {
      DBG("Number of varyings (%d) exceeds maximum %d", c->num_varyings,
          c->specs->max_varyings);
      return false;
   }

   if (c->imm_base > c->specs->num_constants) {
      DBG("Number of constants (%d) exceeds maximum %d", c->imm_base,
          c->specs->num_constants);
   }

   return true;
}

static void
copy_uniform_state_to_shader(struct etna_compile *c, struct etna_shader *sobj)
{
   uint32_t count = c->imm_size;
   struct etna_shader_uniform_info *uinfo = &sobj->uniforms;

   uinfo->const_count = c->imm_base;
   uinfo->imm_count = count;
   uinfo->imm_data = mem_dup(c->imm_data, count * sizeof(*c->imm_data));
   uinfo->imm_contents = mem_dup(c->imm_contents, count * sizeof(*c->imm_contents));

   etna_set_shader_uniforms_dirty_flags(sobj);
}

struct etna_shader *
etna_compile_shader(const struct etna_specs *specs,
                    const struct tgsi_token *tokens)
{
   /* Create scratch space that may be too large to fit on stack
    */
   bool ret;
   struct etna_compile *c;
   struct etna_shader *shader;

   struct tgsi_lowering_config lconfig = {
      .lower_SCS = specs->has_sin_cos_sqrt,
      .lower_FLR = !specs->has_sign_floor_ceil,
      .lower_CEIL = !specs->has_sign_floor_ceil,
      .lower_POW = true,
      .lower_EXP = true,
      .lower_LOG = true,
      .lower_DP2 = true,
      .lower_DP2A = true,
      .lower_TRUNC = true,
      .lower_XPD = true
   };

   c = CALLOC_STRUCT(etna_compile);
   if (!c)
      return NULL;

   shader = CALLOC_STRUCT(etna_shader);
   if (!shader)
      goto out;

   c->specs = specs;
   c->tokens = tgsi_transform_lowering(&lconfig, tokens, &c->info);
   c->free_tokens = !!c->tokens;
   if (!c->tokens) {
      /* no lowering */
      c->tokens = tokens;
   }

   /* Build a map from gallium register to native registers for files
    * CONST, SAMP, IMM, OUT, IN, TEMP.
    * SAMP will map as-is for fragment shaders, there will be a +8 offset for
    * vertex shaders.
    */
   /* Pass one -- check register file declarations and immediates */
   etna_compile_parse_declarations(c);

   etna_allocate_decls(c);

   /* Pass two -- check usage of temporaries, inputs, outputs */
   etna_compile_pass_check_usage(c);

   assign_special_inputs(c);

   /* Assign native temp register to TEMPs */
   assign_temporaries_to_native(c, &c->file[TGSI_FILE_TEMPORARY]);

   /* optimize outputs */
   etna_compile_pass_optimize_outputs(c);

   /* XXX assign special inputs: gl_FrontFacing (VARYING_SLOT_FACE)
    *     this is part of RGROUP_INTERNAL
    */

   /* assign inputs: last usage of input should be <= first usage of temp */
   /*   potential optimization case:
    *     if single MOV TEMP[y], IN[x] before which temp y is not used, and
    * after which IN[x]
    *     is not read, temp[y] can be used as input register as-is
    */
   /*   sort temporaries by first use
    *   sort inputs by last usage
    *   iterate over inputs, temporaries
    *     if last usage of input <= first usage of temp:
    *       assign input to temp
    *       advance input, temporary pointer
    *     else
    *       advance temporary pointer
    *
    *   potential problem: instruction with multiple inputs of which one is the
    * temp and the other is the input;
    *      however, as the temp is not used before this, how would this make
    * sense? uninitialized temporaries have an undefined
    *      value, so this would be ok
    */
   assign_inouts_to_temporaries(c, TGSI_FILE_INPUT);

   /* assign outputs: first usage of output should be >= last usage of temp */
   /*   potential optimization case:
    *      if single MOV OUT[x], TEMP[y] (with full write mask, or at least
    * writing all components that are used in
    *        the shader) after which temp y is no longer used temp[y] can be
    * used as output register as-is
    *
    *   potential problem: instruction with multiple outputs of which one is the
    * temp and the other is the output;
    *      however, as the temp is not used after this, how would this make
    * sense? could just discard the output value
    */
   /*   sort temporaries by last use
    *   sort outputs by first usage
    *   iterate over outputs, temporaries
    *     if first usage of output >= last usage of temp:
    *       assign output to temp
    *       advance output, temporary pointer
    *     else
    *       advance temporary pointer
    */
   assign_inouts_to_temporaries(c, TGSI_FILE_OUTPUT);

   assign_constants_and_immediates(c);
   assign_texture_units(c);

   /* list declarations */
   for (int x = 0; x < c->total_decls; ++x) {
      DBG_F(ETNA_DBG_COMPILER_MSGS, "%i: %s,%d active=%i first_use=%i "
                                    "last_use=%i native=%i usage_mask=%x "
                                    "has_semantic=%i",
            x, tgsi_file_name(c->decl[x].file), c->decl[x].idx,
            c->decl[x].active, c->decl[x].first_use, c->decl[x].last_use,
            c->decl[x].native.valid ? c->decl[x].native.id : -1,
            c->decl[x].usage_mask, c->decl[x].has_semantic);
      if (c->decl[x].has_semantic)
         DBG_F(ETNA_DBG_COMPILER_MSGS, " semantic_name=%s semantic_idx=%i",
               tgsi_semantic_names[c->decl[x].semantic.Name],
               c->decl[x].semantic.Index);
   }
   /* XXX for PS we need to permute so that inputs are always in temporary
    * 0..N-1.
    * There is no "switchboard" for varyings (AFAIK!). The output color,
    * however, can be routed
    * from an arbitrary temporary.
    */
   if (c->info.processor == PIPE_SHADER_FRAGMENT)
      permute_ps_inputs(c);


   /* list declarations */
   for (int x = 0; x < c->total_decls; ++x) {
      DBG_F(ETNA_DBG_COMPILER_MSGS, "%i: %s,%d active=%i first_use=%i "
                                    "last_use=%i native=%i usage_mask=%x "
                                    "has_semantic=%i",
            x, tgsi_file_name(c->decl[x].file), c->decl[x].idx,
            c->decl[x].active, c->decl[x].first_use, c->decl[x].last_use,
            c->decl[x].native.valid ? c->decl[x].native.id : -1,
            c->decl[x].usage_mask, c->decl[x].has_semantic);
      if (c->decl[x].has_semantic)
         DBG_F(ETNA_DBG_COMPILER_MSGS, " semantic_name=%s semantic_idx=%i",
               tgsi_semantic_names[c->decl[x].semantic.Name],
               c->decl[x].semantic.Index);
   }

   /* pass 3: generate instructions */
   etna_compile_pass_generate_code(c);
   etna_compile_add_z_div_if_needed(c);
   etna_compile_add_nop_if_needed(c);
   etna_compile_fill_in_labels(c);

   ret = etna_compile_check_limits(c);
   if (!ret) {
      FREE(shader);
      shader = NULL;
      goto out;
   }

   /* fill in output structure */
   shader->processor = c->info.processor;
   shader->code_size = c->inst_ptr * 4;
   shader->code = mem_dup(c->code, c->inst_ptr * 16);
   shader->num_temps = c->next_free_native;
   shader->vs_pos_out_reg = -1;
   shader->vs_pointsize_out_reg = -1;
   shader->ps_color_out_reg = -1;
   shader->ps_depth_out_reg = -1;
   copy_uniform_state_to_shader(c, shader);

   if (c->info.processor == PIPE_SHADER_VERTEX) {
      fill_in_vs_inputs(shader, c);
      fill_in_vs_outputs(shader, c);
   } else if (c->info.processor == PIPE_SHADER_FRAGMENT) {
      fill_in_ps_inputs(shader, c);
      fill_in_ps_outputs(shader, c);
   }

out:
   if (c->free_tokens)
      FREE((void *)c->tokens);

   FREE(c->labels);
   FREE(c);

   return shader;
}

extern const char *tgsi_swizzle_names[];
void
etna_dump_shader(const struct etna_shader *shader)
{
   if (shader->processor == PIPE_SHADER_VERTEX)
      printf("VERT\n");
   else
      printf("FRAG\n");


   etna_disasm(shader->code, shader->code_size, PRINT_RAW);

   printf("num temps: %i\n", shader->num_temps);
   printf("num const: %i\n", shader->uniforms.const_count);
   printf("immediates:\n");
   for (int idx = 0; idx < shader->uniforms.imm_count; ++idx) {
      printf(" [%i].%s = %f (0x%08x)\n",
             (idx + shader->uniforms.const_count) / 4,
             tgsi_swizzle_names[idx % 4],
             *((float *)&shader->uniforms.imm_data[idx]),
             shader->uniforms.imm_data[idx]);
   }
   printf("inputs:\n");
   for (int idx = 0; idx < shader->infile.num_reg; ++idx) {
      printf(" [%i] name=%s index=%i comps=%i\n", shader->infile.reg[idx].reg,
             tgsi_semantic_names[shader->infile.reg[idx].semantic.Name],
             shader->infile.reg[idx].semantic.Index,
             shader->infile.reg[idx].num_components);
   }
   printf("outputs:\n");
   for (int idx = 0; idx < shader->outfile.num_reg; ++idx) {
      printf(" [%i] name=%s index=%i comps=%i\n", shader->outfile.reg[idx].reg,
             tgsi_semantic_names[shader->outfile.reg[idx].semantic.Name],
             shader->outfile.reg[idx].semantic.Index,
             shader->outfile.reg[idx].num_components);
   }
   printf("special:\n");
   if (shader->processor == PIPE_SHADER_VERTEX) {
      printf("  vs_pos_out_reg=%i\n", shader->vs_pos_out_reg);
      printf("  vs_pointsize_out_reg=%i\n", shader->vs_pointsize_out_reg);
      printf("  vs_load_balancing=0x%08x\n", shader->vs_load_balancing);
   } else {
      printf("  ps_color_out_reg=%i\n", shader->ps_color_out_reg);
      printf("  ps_depth_out_reg=%i\n", shader->ps_depth_out_reg);
   }
   printf("  input_count_unk8=0x%08x\n", shader->input_count_unk8);
}

void
etna_destroy_shader(struct etna_shader *shader)
{
   assert(shader);

   FREE(shader->code);
   FREE(shader->uniforms.imm_data);
   FREE(shader->uniforms.imm_contents);
   FREE(shader->output_per_semantic_list);
   FREE(shader);
}

static const struct etna_shader_inout *
etna_shader_vs_lookup(const struct etna_shader *sobj,
                      const struct etna_shader_inout *in)
{
   if (in->semantic.Index < sobj->output_count_per_semantic[in->semantic.Name])
      return sobj->output_per_semantic[in->semantic.Name][in->semantic.Index];

   return NULL;
}

bool
etna_link_shader(struct etna_shader_link_info *info,
                 const struct etna_shader *vs, const struct etna_shader *fs)
{
   /* For each fragment input we need to find the associated vertex shader
    * output, which can be found by matching on semantic name and index. A
    * binary search could be used because the vs outputs are sorted by their
    * semantic index and grouped by semantic type by fill_in_vs_outputs.
    */
   assert(fs->infile.num_reg < ETNA_NUM_INPUTS);

   for (int idx = 0; idx < fs->infile.num_reg; ++idx) {
      const struct etna_shader_inout *fsio = &fs->infile.reg[idx];
      const struct etna_shader_inout *vsio = etna_shader_vs_lookup(vs, fsio);
      struct etna_varying *varying;

      assert(fsio->reg > 0 && fsio->reg <= ARRAY_SIZE(info->varyings));

      if (fsio->reg > info->num_varyings)
         info->num_varyings = fsio->reg;

      varying = &info->varyings[fsio->reg - 1];
      varying->num_components = fsio->num_components;

      if (fsio->semantic.Name == TGSI_SEMANTIC_COLOR) /* colors affected by flat shading */
         varying->pa_attributes = 0x200;
      else /* texture coord or other bypasses flat shading */
         varying->pa_attributes = 0x2f1;

      if (fsio->semantic.Name == TGSI_SEMANTIC_PCOORD) {
         varying->use[0] = VARYING_COMPONENT_USE_POINTCOORD_X;
         varying->use[1] = VARYING_COMPONENT_USE_POINTCOORD_Y;
         varying->use[2] = VARYING_COMPONENT_USE_USED;
         varying->use[3] = VARYING_COMPONENT_USE_USED;
         varying->reg = 0; /* replaced by point coord -- doesn't matter */
         continue;
      }

      if (vsio == NULL)
         return true; /* not found -- link error */

      varying->use[0] = VARYING_COMPONENT_USE_USED;
      varying->use[1] = VARYING_COMPONENT_USE_USED;
      varying->use[2] = VARYING_COMPONENT_USE_USED;
      varying->use[3] = VARYING_COMPONENT_USE_USED;
      varying->reg = vsio->reg;
   }

   assert(info->num_varyings == fs->infile.num_reg);

   return false;
}