summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/llvmpipe/lp_state_fs.c
blob: cc163ebd4fa987bbf788f578202228c639dc4732 (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
/**************************************************************************
 * 
 * Copyright 2009 VMware, Inc.
 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
 * All Rights Reserved.
 * 
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 * 
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 * 
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 * 
 **************************************************************************/

/**
 * @file
 * Code generate the whole fragment pipeline.
 *
 * The fragment pipeline consists of the following stages:
 * - triangle edge in/out testing
 * - scissor test
 * - stipple (TBI)
 * - early depth test
 * - fragment shader
 * - alpha test
 * - depth/stencil test
 * - blending
 *
 * This file has only the glue to assemble the fragment pipeline.  The actual
 * plumbing of converting Gallium state into LLVM IR is done elsewhere, in the
 * lp_bld_*.[ch] files, and in a complete generic and reusable way. Here we
 * muster the LLVM JIT execution engine to create a function that follows an
 * established binary interface and that can be called from C directly.
 *
 * A big source of complexity here is that we often want to run different
 * stages with different precisions and data types and precisions. For example,
 * the fragment shader needs typically to be done in floats, but the
 * depth/stencil test and blending is better done in the type that most closely
 * matches the depth/stencil and color buffer respectively.
 *
 * Since the width of a SIMD vector register stays the same regardless of the
 * element type, different types imply different number of elements, so we must
 * code generate more instances of the stages with larger types to be able to
 * feed/consume the stages with smaller types.
 *
 * @author Jose Fonseca <jfonseca@vmware.com>
 */

#include <limits.h>
#include "pipe/p_defines.h"
#include "util/u_inlines.h"
#include "util/u_memory.h"
#include "util/u_format.h"
#include "util/u_dump.h"
#include "os/os_time.h"
#include "pipe/p_shader_tokens.h"
#include "draw/draw_context.h"
#include "tgsi/tgsi_dump.h"
#include "tgsi/tgsi_scan.h"
#include "tgsi/tgsi_parse.h"
#include "gallivm/lp_bld_type.h"
#include "gallivm/lp_bld_const.h"
#include "gallivm/lp_bld_conv.h"
#include "gallivm/lp_bld_intr.h"
#include "gallivm/lp_bld_logic.h"
#include "gallivm/lp_bld_tgsi.h"
#include "gallivm/lp_bld_swizzle.h"
#include "gallivm/lp_bld_flow.h"
#include "gallivm/lp_bld_debug.h"

#include "lp_bld_alpha.h"
#include "lp_bld_blend.h"
#include "lp_bld_depth.h"
#include "lp_bld_interp.h"
#include "lp_context.h"
#include "lp_debug.h"
#include "lp_perf.h"
#include "lp_screen.h"
#include "lp_setup.h"
#include "lp_state.h"
#include "lp_tex_sample.h"


#include <llvm-c/Analysis.h>


static const unsigned char quad_offset_x[4] = {0, 1, 0, 1};
static const unsigned char quad_offset_y[4] = {0, 0, 1, 1};


/*
 * Derive from the quad's upper left scalar coordinates the coordinates for
 * all other quad pixels
 */
static void
generate_pos0(LLVMBuilderRef builder,
              LLVMValueRef x,
              LLVMValueRef y,
              LLVMValueRef *x0,
              LLVMValueRef *y0)
{
   LLVMTypeRef int_elem_type = LLVMInt32Type();
   LLVMTypeRef int_vec_type = LLVMVectorType(int_elem_type, QUAD_SIZE);
   LLVMTypeRef elem_type = LLVMFloatType();
   LLVMTypeRef vec_type = LLVMVectorType(elem_type, QUAD_SIZE);
   LLVMValueRef x_offsets[QUAD_SIZE];
   LLVMValueRef y_offsets[QUAD_SIZE];
   unsigned i;

   x = lp_build_broadcast(builder, int_vec_type, x);
   y = lp_build_broadcast(builder, int_vec_type, y);

   for(i = 0; i < QUAD_SIZE; ++i) {
      x_offsets[i] = LLVMConstInt(int_elem_type, quad_offset_x[i], 0);
      y_offsets[i] = LLVMConstInt(int_elem_type, quad_offset_y[i], 0);
   }

   x = LLVMBuildAdd(builder, x, LLVMConstVector(x_offsets, QUAD_SIZE), "");
   y = LLVMBuildAdd(builder, y, LLVMConstVector(y_offsets, QUAD_SIZE), "");

   *x0 = LLVMBuildSIToFP(builder, x, vec_type, "");
   *y0 = LLVMBuildSIToFP(builder, y, vec_type, "");
}


/**
 * Generate the depth /stencil test code.
 */
static void
generate_depth_stencil(LLVMBuilderRef builder,
                       const struct lp_fragment_shader_variant_key *key,
                       struct lp_type src_type,
                       struct lp_build_mask_context *mask,
                       LLVMValueRef stencil_refs[2],
                       LLVMValueRef src,
                       LLVMValueRef dst_ptr,
                       LLVMValueRef facing,
                       LLVMValueRef counter)
{
   const struct util_format_description *format_desc;
   struct lp_type dst_type;

   if (!key->depth.enabled && !key->stencil[0].enabled && !key->stencil[1].enabled)
      return;

   format_desc = util_format_description(key->zsbuf_format);
   assert(format_desc);

   /*
    * Depths are expected to be between 0 and 1, even if they are stored in
    * floats. Setting these bits here will ensure that the lp_build_conv() call
    * below won't try to unnecessarily clamp the incoming values.
    */
   if(src_type.floating) {
      src_type.sign = FALSE;
      src_type.norm = TRUE;
   }
   else {
      assert(!src_type.sign);
      assert(src_type.norm);
   }

   /* Pick the depth type. */
   dst_type = lp_depth_type(format_desc, src_type.width*src_type.length);

   /* FIXME: Cope with a depth test type with a different bit width. */
   assert(dst_type.width == src_type.width);
   assert(dst_type.length == src_type.length);

   /* Convert fragment Z from float to integer */
   lp_build_conv(builder, src_type, dst_type, &src, 1, &src, 1);

   dst_ptr = LLVMBuildBitCast(builder,
                              dst_ptr,
                              LLVMPointerType(lp_build_vec_type(dst_type), 0), "");
   lp_build_depth_stencil_test(builder,
                               &key->depth,
                               key->stencil,
                               dst_type,
                               format_desc,
                               mask,
                               stencil_refs,
                               src,
                               dst_ptr,
                               facing,
                               counter);
}


/**
 * Generate the code to do inside/outside triangle testing for the
 * four pixels in a 2x2 quad.  This will set the four elements of the
 * quad mask vector to 0 or ~0.
 * \param i  which quad of the quad group to test, in [0,3]
 */
static void
generate_tri_edge_mask(LLVMBuilderRef builder,
                       unsigned i,
                       LLVMValueRef *mask,      /* ivec4, out */
                       LLVMValueRef c0,         /* int32 */
                       LLVMValueRef c1,         /* int32 */
                       LLVMValueRef c2,         /* int32 */
                       LLVMValueRef step0_ptr,  /* ivec4 */
                       LLVMValueRef step1_ptr,  /* ivec4 */
                       LLVMValueRef step2_ptr)  /* ivec4 */
{
#define OPTIMIZE_IN_OUT_TEST 0
#if OPTIMIZE_IN_OUT_TEST
   struct lp_build_if_state ifctx;
   LLVMValueRef not_draw_all;
#endif
   struct lp_build_flow_context *flow;
   struct lp_type i32_type;
   LLVMTypeRef i32vec4_type;
   LLVMValueRef c0_vec, c1_vec, c2_vec;
   LLVMValueRef in_out_mask;

   assert(i < 4);
   
   /* int32 vector type */
   memset(&i32_type, 0, sizeof i32_type);
   i32_type.floating = FALSE; /* values are integers */
   i32_type.sign = TRUE;      /* values are signed */
   i32_type.norm = FALSE;     /* values are not normalized */
   i32_type.width = 32;       /* 32-bit int values */
   i32_type.length = 4;       /* 4 elements per vector */

   i32vec4_type = lp_build_int32_vec4_type();

   /*
    * Use a conditional here to do detailed pixel in/out testing.
    * We only have to do this if c0 != INT_MIN.
    */
   flow = lp_build_flow_create(builder);
   lp_build_flow_scope_begin(flow);

   {
#if OPTIMIZE_IN_OUT_TEST
      /* not_draw_all = (c0 != INT_MIN) */
      not_draw_all = LLVMBuildICmp(builder,
                                   LLVMIntNE,
                                   c0,
                                   LLVMConstInt(LLVMInt32Type(), INT_MIN, 0),
                                   "");

      in_out_mask = lp_build_const_int_vec(i32_type, ~0);


      lp_build_flow_scope_declare(flow, &in_out_mask);

      /* if (not_draw_all) {... */
      lp_build_if(&ifctx, flow, builder, not_draw_all);
#endif
      {
         LLVMValueRef step0_vec, step1_vec, step2_vec;
         LLVMValueRef m0_vec, m1_vec, m2_vec;
         LLVMValueRef index, m;

         /* c0_vec = {c0, c0, c0, c0}
          * Note that we emit this code four times but LLVM optimizes away
          * three instances of it.
          */
         c0_vec = lp_build_broadcast(builder, i32vec4_type, c0);
         c1_vec = lp_build_broadcast(builder, i32vec4_type, c1);
         c2_vec = lp_build_broadcast(builder, i32vec4_type, c2);
         lp_build_name(c0_vec, "edgeconst0vec");
         lp_build_name(c1_vec, "edgeconst1vec");
         lp_build_name(c2_vec, "edgeconst2vec");

         /* load step0vec, step1, step2 vec from memory */
         index = LLVMConstInt(LLVMInt32Type(), i, 0);
         step0_vec = LLVMBuildLoad(builder, LLVMBuildGEP(builder, step0_ptr, &index, 1, ""), "");
         step1_vec = LLVMBuildLoad(builder, LLVMBuildGEP(builder, step1_ptr, &index, 1, ""), "");
         step2_vec = LLVMBuildLoad(builder, LLVMBuildGEP(builder, step2_ptr, &index, 1, ""), "");
         lp_build_name(step0_vec, "step0vec");
         lp_build_name(step1_vec, "step1vec");
         lp_build_name(step2_vec, "step2vec");

         /* m0_vec = step0_ptr[i] > c0_vec */
         m0_vec = lp_build_compare(builder, i32_type, PIPE_FUNC_GREATER, step0_vec, c0_vec);
         m1_vec = lp_build_compare(builder, i32_type, PIPE_FUNC_GREATER, step1_vec, c1_vec);
         m2_vec = lp_build_compare(builder, i32_type, PIPE_FUNC_GREATER, step2_vec, c2_vec);

         /* in_out_mask = m0_vec & m1_vec & m2_vec */
         m = LLVMBuildAnd(builder, m0_vec, m1_vec, "");
         in_out_mask = LLVMBuildAnd(builder, m, m2_vec, "");
         lp_build_name(in_out_mask, "inoutmaskvec");
      }
#if OPTIMIZE_IN_OUT_TEST
      lp_build_endif(&ifctx);
#endif

   }
   lp_build_flow_scope_end(flow);
   lp_build_flow_destroy(flow);

   /* This is the initial alive/dead pixel mask for a quad of four pixels.
    * It's an int[4] vector with each word set to 0 or ~0.
    * Words will get cleared when pixels faile the Z test, etc.
    */
   *mask = in_out_mask;
}


static LLVMValueRef
generate_scissor_test(LLVMBuilderRef builder,
                      LLVMValueRef context_ptr,
                      const struct lp_build_interp_soa_context *interp,
                      struct lp_type type)
{
   LLVMTypeRef vec_type = lp_build_vec_type(type);
   LLVMValueRef xpos = interp->pos[0], ypos = interp->pos[1];
   LLVMValueRef xmin, ymin, xmax, ymax;
   LLVMValueRef m0, m1, m2, m3, m;

   /* xpos, ypos contain the window coords for the four pixels in the quad */
   assert(xpos);
   assert(ypos);

   /* get the current scissor bounds, convert to vectors */
   xmin = lp_jit_context_scissor_xmin_value(builder, context_ptr);
   xmin = lp_build_broadcast(builder, vec_type, xmin);

   ymin = lp_jit_context_scissor_ymin_value(builder, context_ptr);
   ymin = lp_build_broadcast(builder, vec_type, ymin);

   xmax = lp_jit_context_scissor_xmax_value(builder, context_ptr);
   xmax = lp_build_broadcast(builder, vec_type, xmax);

   ymax = lp_jit_context_scissor_ymax_value(builder, context_ptr);
   ymax = lp_build_broadcast(builder, vec_type, ymax);

   /* compare the fragment's position coordinates against the scissor bounds */
   m0 = lp_build_compare(builder, type, PIPE_FUNC_GEQUAL, xpos, xmin);
   m1 = lp_build_compare(builder, type, PIPE_FUNC_GEQUAL, ypos, ymin);
   m2 = lp_build_compare(builder, type, PIPE_FUNC_LESS, xpos, xmax);
   m3 = lp_build_compare(builder, type, PIPE_FUNC_LESS, ypos, ymax);

   /* AND all the masks together */
   m = LLVMBuildAnd(builder, m0, m1, "");
   m = LLVMBuildAnd(builder, m, m2, "");
   m = LLVMBuildAnd(builder, m, m3, "");

   lp_build_name(m, "scissormask");

   return m;
}


static LLVMValueRef
build_int32_vec_const(int value)
{
   struct lp_type i32_type;

   memset(&i32_type, 0, sizeof i32_type);
   i32_type.floating = FALSE; /* values are integers */
   i32_type.sign = TRUE;      /* values are signed */
   i32_type.norm = FALSE;     /* values are not normalized */
   i32_type.width = 32;       /* 32-bit int values */
   i32_type.length = 4;       /* 4 elements per vector */
   return lp_build_const_int_vec(i32_type, value);
}



/**
 * Generate the fragment shader, depth/stencil test, and alpha tests.
 * \param i  which quad in the tile, in range [0,3]
 * \param do_tri_test  if 1, do triangle edge in/out testing
 */
static void
generate_fs(struct llvmpipe_context *lp,
            struct lp_fragment_shader *shader,
            const struct lp_fragment_shader_variant_key *key,
            LLVMBuilderRef builder,
            struct lp_type type,
            LLVMValueRef context_ptr,
            unsigned i,
            const struct lp_build_interp_soa_context *interp,
            struct lp_build_sampler_soa *sampler,
            LLVMValueRef *pmask,
            LLVMValueRef (*color)[4],
            LLVMValueRef depth_ptr,
            LLVMValueRef facing,
            unsigned do_tri_test,
            LLVMValueRef c0,
            LLVMValueRef c1,
            LLVMValueRef c2,
            LLVMValueRef step0_ptr,
            LLVMValueRef step1_ptr,
            LLVMValueRef step2_ptr,
            LLVMValueRef counter)
{
   const struct tgsi_token *tokens = shader->base.tokens;
   LLVMTypeRef vec_type;
   LLVMValueRef consts_ptr;
   LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS];
   LLVMValueRef z = interp->pos[2];
   LLVMValueRef stencil_refs[2];
   struct lp_build_flow_context *flow;
   struct lp_build_mask_context mask;
   boolean early_depth_stencil_test;
   unsigned attrib;
   unsigned chan;
   unsigned cbuf;

   assert(i < 4);

   stencil_refs[0] = lp_jit_context_stencil_ref_front_value(builder, context_ptr);
   stencil_refs[1] = lp_jit_context_stencil_ref_back_value(builder, context_ptr);

   vec_type = lp_build_vec_type(type);

   consts_ptr = lp_jit_context_constants(builder, context_ptr);

   flow = lp_build_flow_create(builder);

   memset(outputs, 0, sizeof outputs);

   lp_build_flow_scope_begin(flow);

   /* Declare the color and z variables */
   for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
      for(chan = 0; chan < NUM_CHANNELS; ++chan) {
	 color[cbuf][chan] = LLVMGetUndef(vec_type);
	 lp_build_flow_scope_declare(flow, &color[cbuf][chan]);
      }
   }
   lp_build_flow_scope_declare(flow, &z);

   /* do triangle edge testing */
   if (do_tri_test) {
      generate_tri_edge_mask(builder, i, pmask,
                             c0, c1, c2, step0_ptr, step1_ptr, step2_ptr);
   }
   else {
      *pmask = build_int32_vec_const(~0);
   }

   /* 'mask' will control execution based on quad's pixel alive/killed state */
   lp_build_mask_begin(&mask, flow, type, *pmask);

   if (key->scissor) {
      LLVMValueRef smask =
         generate_scissor_test(builder, context_ptr, interp, type);
      lp_build_mask_update(&mask, smask);
   }

   early_depth_stencil_test =
      (key->depth.enabled || key->stencil[0].enabled) &&
      !key->alpha.enabled &&
      !shader->info.uses_kill &&
      !shader->info.writes_z;

   if (early_depth_stencil_test)
      generate_depth_stencil(builder, key,
                             type, &mask,
                             stencil_refs, z, depth_ptr, facing, counter);

   lp_build_tgsi_soa(builder, tokens, type, &mask,
                     consts_ptr, interp->pos, interp->inputs,
                     outputs, sampler, &shader->info);

   /* loop over fragment shader outputs/results */
   for (attrib = 0; attrib < shader->info.num_outputs; ++attrib) {
      for(chan = 0; chan < NUM_CHANNELS; ++chan) {
         if(outputs[attrib][chan]) {
            LLVMValueRef out = LLVMBuildLoad(builder, outputs[attrib][chan], "");
            lp_build_name(out, "output%u.%u.%c", i, attrib, "xyzw"[chan]);

            switch (shader->info.output_semantic_name[attrib]) {
            case TGSI_SEMANTIC_COLOR:
               {
                  unsigned cbuf = shader->info.output_semantic_index[attrib];

                  lp_build_name(out, "color%u.%u.%c", i, attrib, "rgba"[chan]);

                  /* Alpha test */
                  /* XXX: should the alpha reference value be passed separately? */
		  /* XXX: should only test the final assignment to alpha */
                  if(cbuf == 0 && chan == 3) {
                     LLVMValueRef alpha = out;
                     LLVMValueRef alpha_ref_value;
                     alpha_ref_value = lp_jit_context_alpha_ref_value(builder, context_ptr);
                     alpha_ref_value = lp_build_broadcast(builder, vec_type, alpha_ref_value);
                     lp_build_alpha_test(builder, &key->alpha, type,
                                         &mask, alpha, alpha_ref_value);
                  }

		  color[cbuf][chan] = out;
                  break;
               }

            case TGSI_SEMANTIC_POSITION:
               if(chan == 2)
                  z = out;
               break;
            }
         }
      }
   }

   if (!early_depth_stencil_test)
      generate_depth_stencil(builder, key,
                             type, &mask,
                             stencil_refs, z, depth_ptr, facing, counter);

   lp_build_mask_end(&mask);

   lp_build_flow_scope_end(flow);

   lp_build_flow_destroy(flow);

   *pmask = mask.value;

}


/**
 * Generate color blending and color output.
 * \param rt  the render target index (to index blend, colormask state)
 * \param type  the pixel color type
 * \param context_ptr  pointer to the runtime JIT context
 * \param mask  execution mask (active fragment/pixel mask)
 * \param src  colors from the fragment shader
 * \param dst_ptr  the destination color buffer pointer
 */
static void
generate_blend(const struct pipe_blend_state *blend,
               unsigned rt,
               LLVMBuilderRef builder,
               struct lp_type type,
               LLVMValueRef context_ptr,
               LLVMValueRef mask,
               LLVMValueRef *src,
               LLVMValueRef dst_ptr)
{
   struct lp_build_context bld;
   struct lp_build_flow_context *flow;
   struct lp_build_mask_context mask_ctx;
   LLVMTypeRef vec_type;
   LLVMValueRef const_ptr;
   LLVMValueRef con[4];
   LLVMValueRef dst[4];
   LLVMValueRef res[4];
   unsigned chan;

   lp_build_context_init(&bld, builder, type);

   flow = lp_build_flow_create(builder);

   /* we'll use this mask context to skip blending if all pixels are dead */
   lp_build_mask_begin(&mask_ctx, flow, type, mask);

   vec_type = lp_build_vec_type(type);

   const_ptr = lp_jit_context_blend_color(builder, context_ptr);
   const_ptr = LLVMBuildBitCast(builder, const_ptr,
                                LLVMPointerType(vec_type, 0), "");

   /* load constant blend color and colors from the dest color buffer */
   for(chan = 0; chan < 4; ++chan) {
      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
      con[chan] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, const_ptr, &index, 1, ""), "");

      dst[chan] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dst_ptr, &index, 1, ""), "");

      lp_build_name(con[chan], "con.%c", "rgba"[chan]);
      lp_build_name(dst[chan], "dst.%c", "rgba"[chan]);
   }

   /* do blend */
   lp_build_blend_soa(builder, blend, type, rt, src, dst, con, res);

   /* store results to color buffer */
   for(chan = 0; chan < 4; ++chan) {
      if(blend->rt[rt].colormask & (1 << chan)) {
         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
         lp_build_name(res[chan], "res.%c", "rgba"[chan]);
         res[chan] = lp_build_select(&bld, mask, res[chan], dst[chan]);
         LLVMBuildStore(builder, res[chan], LLVMBuildGEP(builder, dst_ptr, &index, 1, ""));
      }
   }

   lp_build_mask_end(&mask_ctx);
   lp_build_flow_destroy(flow);
}


/**
 * Generate the runtime callable function for the whole fragment pipeline.
 * Note that the function which we generate operates on a block of 16
 * pixels at at time.  The block contains 2x2 quads.  Each quad contains
 * 2x2 pixels.
 */
static void
generate_fragment(struct llvmpipe_context *lp,
                  struct lp_fragment_shader *shader,
                  struct lp_fragment_shader_variant *variant,
                  unsigned do_tri_test)
{
   struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen);
   const struct lp_fragment_shader_variant_key *key = &variant->key;
   struct lp_type fs_type;
   struct lp_type blend_type;
   LLVMTypeRef fs_elem_type;
   LLVMTypeRef fs_int_vec_type;
   LLVMTypeRef blend_vec_type;
   LLVMTypeRef arg_types[16];
   LLVMTypeRef func_type;
   LLVMTypeRef int32_vec4_type = lp_build_int32_vec4_type();
   LLVMValueRef context_ptr;
   LLVMValueRef x;
   LLVMValueRef y;
   LLVMValueRef a0_ptr;
   LLVMValueRef dadx_ptr;
   LLVMValueRef dady_ptr;
   LLVMValueRef color_ptr_ptr;
   LLVMValueRef depth_ptr;
   LLVMValueRef c0, c1, c2, step0_ptr, step1_ptr, step2_ptr, counter = NULL;
   LLVMBasicBlockRef block;
   LLVMBuilderRef builder;
   LLVMValueRef x0;
   LLVMValueRef y0;
   struct lp_build_sampler_soa *sampler;
   struct lp_build_interp_soa_context interp;
   LLVMValueRef fs_mask[LP_MAX_VECTOR_LENGTH];
   LLVMValueRef fs_out_color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][LP_MAX_VECTOR_LENGTH];
   LLVMValueRef blend_mask;
   LLVMValueRef function;
   LLVMValueRef facing;
   unsigned num_fs;
   unsigned i;
   unsigned chan;
   unsigned cbuf;


   /* TODO: actually pick these based on the fs and color buffer
    * characteristics. */

   memset(&fs_type, 0, sizeof fs_type);
   fs_type.floating = TRUE; /* floating point values */
   fs_type.sign = TRUE;     /* values are signed */
   fs_type.norm = FALSE;    /* values are not limited to [0,1] or [-1,1] */
   fs_type.width = 32;      /* 32-bit float */
   fs_type.length = 4;      /* 4 elements per vector */
   num_fs = 4;              /* number of quads per block */

   memset(&blend_type, 0, sizeof blend_type);
   blend_type.floating = FALSE; /* values are integers */
   blend_type.sign = FALSE;     /* values are unsigned */
   blend_type.norm = TRUE;      /* values are in [0,1] or [-1,1] */
   blend_type.width = 8;        /* 8-bit ubyte values */
   blend_type.length = 16;      /* 16 elements per vector */

   /* 
    * Generate the function prototype. Any change here must be reflected in
    * lp_jit.h's lp_jit_frag_func function pointer type, and vice-versa.
    */

   fs_elem_type = lp_build_elem_type(fs_type);
   fs_int_vec_type = lp_build_int_vec_type(fs_type);

   blend_vec_type = lp_build_vec_type(blend_type);

   arg_types[0] = screen->context_ptr_type;            /* context */
   arg_types[1] = LLVMInt32Type();                     /* x */
   arg_types[2] = LLVMInt32Type();                     /* y */
   arg_types[3] = LLVMFloatType();                     /* facing */
   arg_types[4] = LLVMPointerType(fs_elem_type, 0);    /* a0 */
   arg_types[5] = LLVMPointerType(fs_elem_type, 0);    /* dadx */
   arg_types[6] = LLVMPointerType(fs_elem_type, 0);    /* dady */
   arg_types[7] = LLVMPointerType(LLVMPointerType(blend_vec_type, 0), 0);  /* color */
   arg_types[8] = LLVMPointerType(fs_int_vec_type, 0); /* depth */
   arg_types[9] = LLVMInt32Type();                     /* c0 */
   arg_types[10] = LLVMInt32Type();                    /* c1 */
   arg_types[11] = LLVMInt32Type();                    /* c2 */
   /* Note: the step arrays are built as int32[16] but we interpret
    * them here as int32_vec4[4].
    */
   arg_types[12] = LLVMPointerType(int32_vec4_type, 0);/* step0 */
   arg_types[13] = LLVMPointerType(int32_vec4_type, 0);/* step1 */
   arg_types[14] = LLVMPointerType(int32_vec4_type, 0);/* step2 */
   arg_types[15] = LLVMPointerType(LLVMInt32Type(), 0);/* counter */

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

   function = LLVMAddFunction(screen->module, "shader", func_type);
   LLVMSetFunctionCallConv(function, LLVMCCallConv);

   variant->function[do_tri_test] = function;


   /* XXX: need to propagate noalias down into color param now we are
    * passing a pointer-to-pointer?
    */
   for(i = 0; i < Elements(arg_types); ++i)
      if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
         LLVMAddAttribute(LLVMGetParam(function, i), LLVMNoAliasAttribute);

   context_ptr  = LLVMGetParam(function, 0);
   x            = LLVMGetParam(function, 1);
   y            = LLVMGetParam(function, 2);
   facing       = LLVMGetParam(function, 3);
   a0_ptr       = LLVMGetParam(function, 4);
   dadx_ptr     = LLVMGetParam(function, 5);
   dady_ptr     = LLVMGetParam(function, 6);
   color_ptr_ptr = LLVMGetParam(function, 7);
   depth_ptr    = LLVMGetParam(function, 8);
   c0           = LLVMGetParam(function, 9);
   c1           = LLVMGetParam(function, 10);
   c2           = LLVMGetParam(function, 11);
   step0_ptr    = LLVMGetParam(function, 12);
   step1_ptr    = LLVMGetParam(function, 13);
   step2_ptr    = LLVMGetParam(function, 14);

   lp_build_name(context_ptr, "context");
   lp_build_name(x, "x");
   lp_build_name(y, "y");
   lp_build_name(a0_ptr, "a0");
   lp_build_name(dadx_ptr, "dadx");
   lp_build_name(dady_ptr, "dady");
   lp_build_name(color_ptr_ptr, "color_ptr_ptr");
   lp_build_name(depth_ptr, "depth");
   lp_build_name(c0, "c0");
   lp_build_name(c1, "c1");
   lp_build_name(c2, "c2");
   lp_build_name(step0_ptr, "step0");
   lp_build_name(step1_ptr, "step1");
   lp_build_name(step2_ptr, "step2");

   if (key->occlusion_count) {
      counter = LLVMGetParam(function, 15);
      lp_build_name(counter, "counter");
   }

   /*
    * Function body
    */

   block = LLVMAppendBasicBlock(function, "entry");
   builder = LLVMCreateBuilder();
   LLVMPositionBuilderAtEnd(builder, block);

   generate_pos0(builder, x, y, &x0, &y0);

   lp_build_interp_soa_init(&interp, 
                            shader->base.tokens,
                            key->flatshade,
                            builder, fs_type,
                            a0_ptr, dadx_ptr, dady_ptr,
                            x0, y0);

   /* code generated texture sampling */
   sampler = lp_llvm_sampler_soa_create(key->sampler, context_ptr);

   /* loop over quads in the block */
   for(i = 0; i < num_fs; ++i) {
      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
      LLVMValueRef out_color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS];
      LLVMValueRef depth_ptr_i;

      if(i != 0)
         lp_build_interp_soa_update(&interp, i);

      depth_ptr_i = LLVMBuildGEP(builder, depth_ptr, &index, 1, "");

      generate_fs(lp, shader, key,
                  builder,
                  fs_type,
                  context_ptr,
                  i,
                  &interp,
                  sampler,
                  &fs_mask[i], /* output */
                  out_color,
                  depth_ptr_i,
                  facing,
                  do_tri_test,
                  c0, c1, c2,
                  step0_ptr, step1_ptr, step2_ptr, counter);

      for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++)
	 for(chan = 0; chan < NUM_CHANNELS; ++chan)
	    fs_out_color[cbuf][chan][i] = out_color[cbuf][chan];
   }

   sampler->destroy(sampler);

   /* Loop over color outputs / color buffers to do blending.
    */
   for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
      LLVMValueRef color_ptr;
      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), cbuf, 0);
      LLVMValueRef blend_in_color[NUM_CHANNELS];
      unsigned rt;

      /* 
       * Convert the fs's output color and mask to fit to the blending type. 
       */
      for(chan = 0; chan < NUM_CHANNELS; ++chan) {
	 lp_build_conv(builder, fs_type, blend_type,
		       fs_out_color[cbuf][chan], num_fs,
		       &blend_in_color[chan], 1);
	 lp_build_name(blend_in_color[chan], "color%d.%c", cbuf, "rgba"[chan]);
      }

      lp_build_conv_mask(builder, fs_type, blend_type,
			 fs_mask, num_fs,
			 &blend_mask, 1);

      color_ptr = LLVMBuildLoad(builder, 
				LLVMBuildGEP(builder, color_ptr_ptr, &index, 1, ""),
				"");
      lp_build_name(color_ptr, "color_ptr%d", cbuf);

      /* which blend/colormask state to use */
      rt = key->blend.independent_blend_enable ? cbuf : 0;

      /*
       * Blending.
       */
      generate_blend(&key->blend,
                     rt,
		     builder,
		     blend_type,
		     context_ptr,
		     blend_mask,
		     blend_in_color,
		     color_ptr);
   }

   LLVMBuildRetVoid(builder);

   LLVMDisposeBuilder(builder);


   /* Verify the LLVM IR.  If invalid, dump and abort */
#ifdef DEBUG
   if(LLVMVerifyFunction(function, LLVMPrintMessageAction)) {
      if (1)
         lp_debug_dump_value(function);
      abort();
   }
#endif

   /* Apply optimizations to LLVM IR */
   if (1)
      LLVMRunFunctionPassManager(screen->pass, function);

   if (LP_DEBUG & DEBUG_JIT) {
      /* Print the LLVM IR to stderr */
      lp_debug_dump_value(function);
      debug_printf("\n");
   }

   /*
    * Translate the LLVM IR into machine code.
    */
   {
      void *f = LLVMGetPointerToGlobal(screen->engine, function);

      variant->jit_function[do_tri_test] = cast_voidptr_to_lp_jit_frag_func(f);

      if (LP_DEBUG & DEBUG_ASM)
         lp_disassemble(f);
   }
}


static struct lp_fragment_shader_variant *
generate_variant(struct llvmpipe_context *lp,
                 struct lp_fragment_shader *shader,
                 const struct lp_fragment_shader_variant_key *key)
{
   struct lp_fragment_shader_variant *variant;

   if (LP_DEBUG & DEBUG_JIT) {
      unsigned i;

      tgsi_dump(shader->base.tokens, 0);
      if(key->depth.enabled) {
         debug_printf("depth.format = %s\n", util_format_name(key->zsbuf_format));
         debug_printf("depth.func = %s\n", util_dump_func(key->depth.func, TRUE));
         debug_printf("depth.writemask = %u\n", key->depth.writemask);
      }
      for (i = 0; i < 2; ++i) {
         if(key->stencil[i].enabled) {
            debug_printf("stencil[%u].func = %s\n", i, util_dump_func(key->stencil[i].func, TRUE));
            debug_printf("stencil[%u].fail_op = %s\n", i, util_dump_stencil_op(key->stencil[i].fail_op, TRUE));
            debug_printf("stencil[%u].zpass_op = %s\n", i, util_dump_stencil_op(key->stencil[i].zpass_op, TRUE));
            debug_printf("stencil[%u].zfail_op = %s\n", i, util_dump_stencil_op(key->stencil[i].zfail_op, TRUE));
            debug_printf("stencil[%u].valuemask = 0x%x\n", i, key->stencil[i].valuemask);
            debug_printf("stencil[%u].writemask = 0x%x\n", i, key->stencil[i].writemask);
         }
      }
      if(key->alpha.enabled) {
         debug_printf("alpha.func = %s\n", util_dump_func(key->alpha.func, TRUE));
         debug_printf("alpha.ref_value = %f\n", key->alpha.ref_value);
      }
      if(key->blend.logicop_enable) {
         debug_printf("blend.logicop_func = %u\n", key->blend.logicop_func);
      }
      else if(key->blend.rt[0].blend_enable) {
         debug_printf("blend.rgb_func = %s\n",   util_dump_blend_func  (key->blend.rt[0].rgb_func, TRUE));
         debug_printf("rgb_src_factor = %s\n",   util_dump_blend_factor(key->blend.rt[0].rgb_src_factor, TRUE));
         debug_printf("rgb_dst_factor = %s\n",   util_dump_blend_factor(key->blend.rt[0].rgb_dst_factor, TRUE));
         debug_printf("alpha_func = %s\n",       util_dump_blend_func  (key->blend.rt[0].alpha_func, TRUE));
         debug_printf("alpha_src_factor = %s\n", util_dump_blend_factor(key->blend.rt[0].alpha_src_factor, TRUE));
         debug_printf("alpha_dst_factor = %s\n", util_dump_blend_factor(key->blend.rt[0].alpha_dst_factor, TRUE));
      }
      debug_printf("blend.colormask = 0x%x\n", key->blend.rt[0].colormask);
      for(i = 0; i < PIPE_MAX_SAMPLERS; ++i) {
         if(key->sampler[i].format) {
            debug_printf("sampler[%u] = \n", i);
            debug_printf("  .format = %s\n",
                         util_format_name(key->sampler[i].format));
            debug_printf("  .target = %s\n",
                         util_dump_tex_target(key->sampler[i].target, TRUE));
            debug_printf("  .pot = %u %u %u\n",
                         key->sampler[i].pot_width,
                         key->sampler[i].pot_height,
                         key->sampler[i].pot_depth);
            debug_printf("  .wrap = %s %s %s\n",
                         util_dump_tex_wrap(key->sampler[i].wrap_s, TRUE),
                         util_dump_tex_wrap(key->sampler[i].wrap_t, TRUE),
                         util_dump_tex_wrap(key->sampler[i].wrap_r, TRUE));
            debug_printf("  .min_img_filter = %s\n",
                         util_dump_tex_filter(key->sampler[i].min_img_filter, TRUE));
            debug_printf("  .min_mip_filter = %s\n",
                         util_dump_tex_mipfilter(key->sampler[i].min_mip_filter, TRUE));
            debug_printf("  .mag_img_filter = %s\n",
                         util_dump_tex_filter(key->sampler[i].mag_img_filter, TRUE));
            if(key->sampler[i].compare_mode != PIPE_TEX_COMPARE_NONE)
               debug_printf("  .compare_func = %s\n", util_dump_func(key->sampler[i].compare_func, TRUE));
            debug_printf("  .normalized_coords = %u\n", key->sampler[i].normalized_coords);
         }
      }
   }

   variant = CALLOC_STRUCT(lp_fragment_shader_variant);
   if(!variant)
      return NULL;

   memcpy(&variant->key, key, sizeof *key);

   generate_fragment(lp, shader, variant, RAST_WHOLE);
   generate_fragment(lp, shader, variant, RAST_EDGE_TEST);

   /* TODO: most of these can be relaxed, in particular the colormask */
   variant->opaque =
         !key->blend.logicop_enable &&
         !key->blend.rt[0].blend_enable &&
         key->blend.rt[0].colormask == 0xf &&
         !key->stencil[0].enabled &&
         !key->alpha.enabled &&
         !key->depth.enabled &&
         !key->scissor &&
         !shader->info.uses_kill
         ? TRUE : FALSE;

   /* insert new variant into linked list */
   variant->next = shader->variants;
   shader->variants = variant;

   return variant;
}


static void *
llvmpipe_create_fs_state(struct pipe_context *pipe,
                         const struct pipe_shader_state *templ)
{
   struct lp_fragment_shader *shader;

   shader = CALLOC_STRUCT(lp_fragment_shader);
   if (!shader)
      return NULL;

   /* get/save the summary info for this shader */
   tgsi_scan_shader(templ->tokens, &shader->info);

   /* we need to keep a local copy of the tokens */
   shader->base.tokens = tgsi_dup_tokens(templ->tokens);

   if (LP_DEBUG & DEBUG_TGSI) {
      debug_printf("llvmpipe: Create fragment shader %p:\n", (void *) shader);
      tgsi_dump(templ->tokens, 0);
   }

   return shader;
}


static void
llvmpipe_bind_fs_state(struct pipe_context *pipe, void *fs)
{
   struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);

   if (llvmpipe->fs == fs)
      return;

   draw_flush(llvmpipe->draw);

   llvmpipe->fs = fs;

   llvmpipe->dirty |= LP_NEW_FS;
}


static void
llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
{
   struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
   struct llvmpipe_screen *screen = llvmpipe_screen(pipe->screen);
   struct lp_fragment_shader *shader = fs;
   struct lp_fragment_shader_variant *variant;

   assert(fs != llvmpipe->fs);
   (void) llvmpipe;

   /*
    * XXX: we need to flush the context until we have some sort of reference
    * counting in fragment shaders as they may still be binned
    */
   draw_flush(llvmpipe->draw);
   lp_setup_flush(llvmpipe->setup, 0);

   variant = shader->variants;
   while(variant) {
      struct lp_fragment_shader_variant *next = variant->next;
      unsigned i;

      for (i = 0; i < Elements(variant->function); i++) {
         if (variant->function[i]) {
            if (variant->jit_function[i])
               LLVMFreeMachineCodeForFunction(screen->engine,
                                              variant->function[i]);
            LLVMDeleteFunction(variant->function[i]);
         }
      }

      FREE(variant);

      variant = next;
   }

   FREE((void *) shader->base.tokens);
   FREE(shader);
}



static void
llvmpipe_set_constant_buffer(struct pipe_context *pipe,
                             uint shader, uint index,
                             struct pipe_resource *constants)
{
   struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
   unsigned size = constants ? constants->width0 : 0;
   const void *data = constants ? llvmpipe_resource_data(constants) : NULL;

   assert(shader < PIPE_SHADER_TYPES);
   assert(index == 0);

   if(llvmpipe->constants[shader] == constants)
      return;

   draw_flush(llvmpipe->draw);

   /* note: reference counting */
   pipe_resource_reference(&llvmpipe->constants[shader], constants);

   if(shader == PIPE_SHADER_VERTEX) {
      draw_set_mapped_constant_buffer(llvmpipe->draw, PIPE_SHADER_VERTEX, 0,
                                      data, size);
   }

   llvmpipe->dirty |= LP_NEW_CONSTANTS;
}


/**
 * Return the blend factor equivalent to a destination alpha of one.
 */
static INLINE unsigned
force_dst_alpha_one(unsigned factor, boolean alpha)
{
   switch(factor) {
   case PIPE_BLENDFACTOR_DST_ALPHA:
      return PIPE_BLENDFACTOR_ONE;
   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
      return PIPE_BLENDFACTOR_ZERO;
   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
      return PIPE_BLENDFACTOR_ZERO;
   }

   if (alpha) {
      switch(factor) {
      case PIPE_BLENDFACTOR_DST_COLOR:
         return PIPE_BLENDFACTOR_ONE;
      case PIPE_BLENDFACTOR_INV_DST_COLOR:
         return PIPE_BLENDFACTOR_ZERO;
      }
   }

   return factor;
}


/**
 * We need to generate several variants of the fragment pipeline to match
 * all the combinations of the contributing state atoms.
 *
 * TODO: there is actually no reason to tie this to context state -- the
 * generated code could be cached globally in the screen.
 */
static void
make_variant_key(struct llvmpipe_context *lp,
                 struct lp_fragment_shader *shader,
                 struct lp_fragment_shader_variant_key *key)
{
   unsigned i;

   memset(key, 0, sizeof *key);

   if (lp->framebuffer.zsbuf) {
      if (lp->depth_stencil->depth.enabled) {
         key->zsbuf_format = lp->framebuffer.zsbuf->format;
         memcpy(&key->depth, &lp->depth_stencil->depth, sizeof key->depth);
      }
      if (lp->depth_stencil->stencil[0].enabled) {
         key->zsbuf_format = lp->framebuffer.zsbuf->format;
         memcpy(&key->stencil, &lp->depth_stencil->stencil, sizeof key->stencil);
      }
   }

   key->alpha.enabled = lp->depth_stencil->alpha.enabled;
   if(key->alpha.enabled)
      key->alpha.func = lp->depth_stencil->alpha.func;
   /* alpha.ref_value is passed in jit_context */

   key->flatshade = lp->rasterizer->flatshade;
   key->scissor = lp->rasterizer->scissor;
   if (lp->active_query_count) {
      key->occlusion_count = TRUE;
   }

   if (lp->framebuffer.nr_cbufs) {
      memcpy(&key->blend, lp->blend, sizeof key->blend);
   }

   key->nr_cbufs = lp->framebuffer.nr_cbufs;
   for (i = 0; i < lp->framebuffer.nr_cbufs; i++) {
      struct pipe_rt_blend_state *blend_rt = &key->blend.rt[i];
      const struct util_format_description *format_desc;
      unsigned chan;

      format_desc = util_format_description(lp->framebuffer.cbufs[i]->format);
      assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_RGB ||
             format_desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB);

      blend_rt->colormask = lp->blend->rt[i].colormask;

      /* mask out color channels not present in the color buffer.
       * Should be simple to incorporate per-cbuf writemasks:
       */
      for(chan = 0; chan < 4; ++chan) {
         enum util_format_swizzle swizzle = format_desc->swizzle[chan];

         if(swizzle > UTIL_FORMAT_SWIZZLE_W)
            blend_rt->colormask &= ~(1 << chan);
      }

      /*
       * Our swizzled render tiles always have an alpha channel, but the linear
       * render target format often does not, so force here the dst alpha to be
       * one.
       *
       * This is not a mere optimization. Wrong results will be produced if the
       * dst alpha is used, the dst format does not have alpha, and the previous
       * rendering was not flushed from the swizzled to linear buffer. For
       * example, NonPowTwo DCT.
       *
       * TODO: This should be generalized to all channels for better
       * performance, but only alpha causes correctness issues.
       */
      if (format_desc->swizzle[3] > UTIL_FORMAT_SWIZZLE_W) {
         blend_rt->rgb_src_factor = force_dst_alpha_one(blend_rt->rgb_src_factor, FALSE);
         blend_rt->rgb_dst_factor = force_dst_alpha_one(blend_rt->rgb_dst_factor, FALSE);
         blend_rt->alpha_src_factor = force_dst_alpha_one(blend_rt->alpha_src_factor, TRUE);
         blend_rt->alpha_dst_factor = force_dst_alpha_one(blend_rt->alpha_dst_factor, TRUE);
      }
   }

   for(i = 0; i < PIPE_MAX_SAMPLERS; ++i)
      if(shader->info.file_mask[TGSI_FILE_SAMPLER] & (1 << i))
         lp_sampler_static_state(&key->sampler[i], lp->fragment_sampler_views[i], lp->sampler[i]);
}


/**
 * Update fragment state.  This is called just prior to drawing
 * something when some fragment-related state has changed.
 */
void 
llvmpipe_update_fs(struct llvmpipe_context *lp)
{
   struct lp_fragment_shader *shader = lp->fs;
   struct lp_fragment_shader_variant_key key;
   struct lp_fragment_shader_variant *variant;

   make_variant_key(lp, shader, &key);

   variant = shader->variants;
   while(variant) {
      if(memcmp(&variant->key, &key, sizeof key) == 0)
         break;

      variant = variant->next;
   }

   if (!variant) {
      int64_t t0, t1;
      int64_t dt;
      t0 = os_time_get();

      variant = generate_variant(lp, shader, &key);

      t1 = os_time_get();
      dt = t1 - t0;
      LP_COUNT_ADD(llvm_compile_time, dt);
      LP_COUNT_ADD(nr_llvm_compiles, 2);  /* emit vs. omit in/out test */
   }

   lp_setup_set_fs_functions(lp->setup, 
                             variant->jit_function[RAST_WHOLE],
                             variant->jit_function[RAST_EDGE_TEST],
                             variant->opaque);
}



void
llvmpipe_init_fs_funcs(struct llvmpipe_context *llvmpipe)
{
   llvmpipe->pipe.create_fs_state = llvmpipe_create_fs_state;
   llvmpipe->pipe.bind_fs_state   = llvmpipe_bind_fs_state;
   llvmpipe->pipe.delete_fs_state = llvmpipe_delete_fs_state;

   llvmpipe->pipe.set_constant_buffer = llvmpipe_set_constant_buffer;
}