summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/llvmpipe/lp_state_setup.c
blob: 6a4fbbbf202f5743a5984cdb78617230488a6572 (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
/**************************************************************************
 *
 * Copyright 2010 VMware.
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 **************************************************************************/


#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/simple_list.h"
#include "os/os_time.h"
#include "gallivm/lp_bld_arit.h"
#include "gallivm/lp_bld_bitarit.h"
#include "gallivm/lp_bld_const.h"
#include "gallivm/lp_bld_debug.h"
#include "gallivm/lp_bld_init.h"
#include "gallivm/lp_bld_logic.h"
#include "gallivm/lp_bld_intr.h"
#include "gallivm/lp_bld_flow.h"
#include "gallivm/lp_bld_type.h"

#include "lp_perf.h"
#include "lp_debug.h"
#include "lp_flush.h"
#include "lp_screen.h"
#include "lp_context.h"
#include "lp_state.h"
#include "lp_state_fs.h"
#include "lp_state_setup.h"


/** Setup shader number (for debugging) */
static unsigned setup_no = 0;


/* currently organized to interpolate full float[4] attributes even
 * when some elements are unused.  Later, can pack vertex data more
 * closely.
 */


struct lp_setup_args
{
   /* Function arguments:
    */
   LLVMValueRef v0;
   LLVMValueRef v1;
   LLVMValueRef v2;
   LLVMValueRef facing;		/* boolean */
   LLVMValueRef a0;
   LLVMValueRef dadx;
   LLVMValueRef dady;

   /* Derived:
    */
   LLVMValueRef x0_center;
   LLVMValueRef y0_center;
   LLVMValueRef dy20_ooa;
   LLVMValueRef dy01_ooa;
   LLVMValueRef dx20_ooa;
   LLVMValueRef dx01_ooa;
   struct lp_build_context bld;
};


static void
store_coef(struct gallivm_state *gallivm,
           struct lp_setup_args *args,
           unsigned slot,
           LLVMValueRef a0,
           LLVMValueRef dadx,
           LLVMValueRef dady)
{
   LLVMBuilderRef builder = gallivm->builder;
   LLVMValueRef idx = lp_build_const_int32(gallivm, slot);

   LLVMBuildStore(builder,
                  a0,
                  LLVMBuildGEP(builder, args->a0, &idx, 1, ""));

   LLVMBuildStore(builder,
                  dadx,
                  LLVMBuildGEP(builder, args->dadx, &idx, 1, ""));

   LLVMBuildStore(builder,
                  dady,
                  LLVMBuildGEP(builder, args->dady, &idx, 1, ""));
}



static void 
emit_constant_coef4(struct gallivm_state *gallivm,
                    struct lp_setup_args *args,
                    unsigned slot,
                    LLVMValueRef vert)
{
   store_coef(gallivm, args, slot, vert, args->bld.zero, args->bld.zero);
}



/**
 * Setup the fragment input attribute with the front-facing value.
 * \param frontface  is the triangle front facing?
 */
static void 
emit_facing_coef(struct gallivm_state *gallivm,
                 struct lp_setup_args *args,
                 unsigned slot )
{
   LLVMBuilderRef builder = gallivm->builder;
   LLVMTypeRef float_type = LLVMFloatTypeInContext(gallivm->context);
   LLVMValueRef a0_0 = args->facing;
   LLVMValueRef a0_0f = LLVMBuildSIToFP(builder, a0_0, float_type, "");
   LLVMValueRef a0, face_val;
   const unsigned char swizzles[4] = { PIPE_SWIZZLE_RED, PIPE_SWIZZLE_ZERO,
                                       PIPE_SWIZZLE_ZERO, PIPE_SWIZZLE_ZERO };
   /* Our face val is either 1 or 0 so we do
    * face = (val * 2) - 1
    * to make it 1 or -1
    */
   face_val =
      LLVMBuildFAdd(builder,
                    LLVMBuildFMul(builder, a0_0f,
                                  lp_build_const_float(gallivm, 2.0),
                                  ""),
                    lp_build_const_float(gallivm, -1.0),
                    "facing");
   face_val = lp_build_broadcast_scalar(&args->bld, face_val);
   a0 = lp_build_swizzle_aos(&args->bld, face_val, swizzles);

   store_coef(gallivm, args, slot, a0, args->bld.zero, args->bld.zero);
}


static LLVMValueRef
vert_attrib(struct gallivm_state *gallivm,
            LLVMValueRef vert,
            int attr,
            int elem,
            const char *name)
{
   LLVMBuilderRef b = gallivm->builder;
   LLVMValueRef idx[2];
   idx[0] = lp_build_const_int32(gallivm, attr);
   idx[1] = lp_build_const_int32(gallivm, elem);
   return LLVMBuildLoad(b, LLVMBuildGEP(b, vert, idx, 2, ""), name);
}


static void
lp_twoside(struct gallivm_state *gallivm,
           struct lp_setup_args *args,
           const struct lp_setup_variant_key *key,
           int bcolor_slot,
           LLVMValueRef attribv[3])
{
   LLVMBuilderRef b = gallivm->builder;
   LLVMValueRef a0_back, a1_back, a2_back;
   LLVMValueRef idx2 = lp_build_const_int32(gallivm, bcolor_slot);

   LLVMValueRef facing = args->facing;
   LLVMValueRef front_facing = LLVMBuildICmp(b, LLVMIntEQ, facing,
                                             lp_build_const_int32(gallivm, 0), ""); /** need i1 for if condition */

   a0_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx2, 1, ""), "v0a_back");
   a1_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx2, 1, ""), "v1a_back");
   a2_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx2, 1, ""), "v2a_back");

   /* Possibly swap the front and back attrib values,
    *
    * Prefer select to if so we don't have to worry about phis or
    * allocas.
    */
   attribv[0] = LLVMBuildSelect(b, front_facing, a0_back, attribv[0], "");
   attribv[1] = LLVMBuildSelect(b, front_facing, a1_back, attribv[1], "");
   attribv[2] = LLVMBuildSelect(b, front_facing, a2_back, attribv[2], "");

}

static void
lp_do_offset_tri(struct gallivm_state *gallivm,
                 struct lp_setup_args *args,
                 const struct lp_setup_variant_key *key,
                 LLVMValueRef inv_det,
                 LLVMValueRef dxyz01,
                 LLVMValueRef dxyz20,
                 LLVMValueRef attribv[3])
{
   LLVMBuilderRef b = gallivm->builder;
   struct lp_build_context flt_scalar_bld;
   struct lp_build_context int_scalar_bld;
   struct lp_build_context *bld = &args->bld;
   LLVMValueRef zoffset, mult;
   LLVMValueRef z0_new, z1_new, z2_new;
   LLVMValueRef dzdxdzdy, dzdx, dzdy, dzxyz20, dyzzx01, dyzzx01_dzxyz20, dzx01_dyz20;
   LLVMValueRef z0z1, z0z1z2;
   LLVMValueRef max, max_value, res12;
   LLVMValueRef shuffles[4];
   LLVMTypeRef shuf_type = LLVMInt32TypeInContext(gallivm->context);
   LLVMValueRef onei = lp_build_const_int32(gallivm, 1);
   LLVMValueRef zeroi = lp_build_const_int32(gallivm, 0);
   LLVMValueRef twoi = lp_build_const_int32(gallivm, 2);
   LLVMValueRef threei  = lp_build_const_int32(gallivm, 3);

   /* (res12) = cross(e,f).xy */
   shuffles[0] = twoi;
   shuffles[1] = zeroi;
   shuffles[2] = onei;
   shuffles[3] = twoi;
   dzxyz20 = LLVMBuildShuffleVector(b, dxyz20, dxyz20, LLVMConstVector(shuffles, 4), "");

   shuffles[0] = onei;
   shuffles[1] = twoi;
   shuffles[2] = twoi;
   shuffles[3] = zeroi;
   dyzzx01 = LLVMBuildShuffleVector(b, dxyz01, dxyz01, LLVMConstVector(shuffles, 4), "");

   dyzzx01_dzxyz20 = LLVMBuildFMul(b, dzxyz20, dyzzx01, "dyzzx01_dzxyz20");

   shuffles[0] = twoi;
   shuffles[1] = threei;
   shuffles[2] = LLVMGetUndef(shuf_type);
   shuffles[3] = LLVMGetUndef(shuf_type);
   dzx01_dyz20 = LLVMBuildShuffleVector(b, dyzzx01_dzxyz20, dyzzx01_dzxyz20,
                                        LLVMConstVector(shuffles, 4), "");

   res12 = LLVMBuildFSub(b, dyzzx01_dzxyz20, dzx01_dyz20, "res12");

   /* dzdx = fabsf(res1 * inv_det), dydx = fabsf(res2 * inv_det)*/
   dzdxdzdy = LLVMBuildFMul(b, res12, inv_det, "dzdxdzdy");
   dzdxdzdy = lp_build_abs(bld, dzdxdzdy);

   dzdx = LLVMBuildExtractElement(b, dzdxdzdy, zeroi, "");
   dzdy = LLVMBuildExtractElement(b, dzdxdzdy, onei, "");

   /* mult = MAX2(dzdx, dzdy) * pgon_offset_scale */
   max = LLVMBuildFCmp(b, LLVMRealUGT, dzdx, dzdy, "");
   max_value = LLVMBuildSelect(b, max, dzdx, dzdy, "max"); 

   mult = LLVMBuildFMul(b, max_value,
                        lp_build_const_float(gallivm, key->pgon_offset_scale), "");

   lp_build_context_init(&flt_scalar_bld, gallivm, lp_type_float_vec(32, 32));

   if (key->floating_point_depth) {
      /*
       * bias = pgon_offset_units * 2^(exponent(max(z0, z1, z2)) - mantissa_bits) +
       *           MAX2(dzdx, dzdy) * pgon_offset_scale
       *
       * NOTE: Assumes IEEE float32.
       */
      LLVMValueRef c23_shifted, exp_mask, bias, exp;
      LLVMValueRef maxz_value, maxz0z1_value;

      lp_build_context_init(&int_scalar_bld, gallivm, lp_type_int_vec(32, 32));

      c23_shifted = lp_build_const_int32(gallivm, 23 << 23);
      exp_mask = lp_build_const_int32(gallivm, 0xff << 23);

      maxz0z1_value = lp_build_max(&flt_scalar_bld,
                         LLVMBuildExtractElement(b, attribv[0], twoi, ""),
                         LLVMBuildExtractElement(b, attribv[1], twoi, ""));

      maxz_value = lp_build_max(&flt_scalar_bld,
                      LLVMBuildExtractElement(b, attribv[2], twoi, ""),
                      maxz0z1_value);

      exp = LLVMBuildBitCast(b, maxz_value, int_scalar_bld.vec_type, "");
      exp = lp_build_and(&int_scalar_bld, exp, exp_mask);
      exp = lp_build_sub(&int_scalar_bld, exp, c23_shifted);
      /* Clamping to zero means mrd will be zero for very small numbers,
       * but specs do not indicate this should be prevented by clamping
       * mrd to smallest normal number instead. */
      exp = lp_build_max(&int_scalar_bld, exp, int_scalar_bld.zero);
      exp = LLVMBuildBitCast(b, exp, flt_scalar_bld.vec_type, "");

      bias = LLVMBuildFMul(b, exp,
                           lp_build_const_float(gallivm, key->pgon_offset_units),
                           "bias");

      zoffset = LLVMBuildFAdd(b, bias, mult, "zoffset");
   } else {
      /*
       * bias = pgon_offset_units + MAX2(dzdx, dzdy) * pgon_offset_scale
       */
      zoffset = LLVMBuildFAdd(b,
                              lp_build_const_float(gallivm, key->pgon_offset_units),
                              mult, "zoffset");
   }

   if (key->pgon_offset_clamp > 0) {
      zoffset = lp_build_min(&flt_scalar_bld,
                             lp_build_const_float(gallivm, key->pgon_offset_clamp),
                             zoffset);
   }
   else if (key->pgon_offset_clamp < 0) {
      zoffset = lp_build_max(&flt_scalar_bld,
                             lp_build_const_float(gallivm, key->pgon_offset_clamp),
                             zoffset);
   }

   /* yuck */
   shuffles[0] = twoi;
   shuffles[1] = lp_build_const_int32(gallivm, 6);
   shuffles[2] = LLVMGetUndef(shuf_type);
   shuffles[3] = LLVMGetUndef(shuf_type);
   z0z1 = LLVMBuildShuffleVector(b, attribv[0], attribv[1], LLVMConstVector(shuffles, 4), "");
   shuffles[0] = zeroi;
   shuffles[1] = onei;
   shuffles[2] = lp_build_const_int32(gallivm, 6);
   shuffles[3] = LLVMGetUndef(shuf_type);
   z0z1z2 = LLVMBuildShuffleVector(b, z0z1, attribv[2], LLVMConstVector(shuffles, 4), "");
   zoffset = lp_build_broadcast_scalar(bld, zoffset);

   /* clamp and do offset */
   /*
    * FIXME I suspect the clamp (is that even right to always clamp to fixed
    * 0.0/1.0?) should really be per fragment?
    */
   z0z1z2 = lp_build_clamp(bld, LLVMBuildFAdd(b, z0z1z2, zoffset, ""), bld->zero, bld->one);

   /* insert into args->a0.z, a1.z, a2.z:
    */
   z0_new = LLVMBuildExtractElement(b, z0z1z2, zeroi, "");
   z1_new = LLVMBuildExtractElement(b, z0z1z2, onei, "");
   z2_new = LLVMBuildExtractElement(b, z0z1z2, twoi, "");
   attribv[0] = LLVMBuildInsertElement(b, attribv[0], z0_new, twoi, "");
   attribv[1] = LLVMBuildInsertElement(b, attribv[1], z1_new, twoi, "");
   attribv[2] = LLVMBuildInsertElement(b, attribv[2], z2_new, twoi, "");
}

static void
load_attribute(struct gallivm_state *gallivm,
               struct lp_setup_args *args,
               const struct lp_setup_variant_key *key,
               unsigned vert_attr,
               LLVMValueRef attribv[3])
{
   LLVMBuilderRef b = gallivm->builder;
   LLVMValueRef idx = lp_build_const_int32(gallivm, vert_attr);

   /* Load the vertex data
    */
   attribv[0] = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
   attribv[1] = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
   attribv[2] = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");


   /* Potentially modify it according to twoside, etc:
    */
   if (key->twoside) {
      if (vert_attr == key->color_slot && key->bcolor_slot >= 0)
         lp_twoside(gallivm, args, key, key->bcolor_slot, attribv);
      else if (vert_attr == key->spec_slot && key->bspec_slot >= 0)
         lp_twoside(gallivm, args, key, key->bspec_slot, attribv);
   }
}

/*
 * FIXME: interpolation is always done wrt fb origin (0/0).
 * However, if some (small) tri is far away from the origin and gradients
 * are large, this can lead to HUGE errors, since the a0 value calculated
 * here can get very large (with the actual values inside the triangle way
 * smaller), leading to complete loss of accuracy. This could be prevented
 * by using some point inside (or at corner) of the tri as interpolation
 * origin, or just use barycentric interpolation (which GL suggests and is
 * what real hw does - you can get the barycentric coordinates from the
 * edge functions in rasterization in principle (though we skip these
 * sometimes completely in case of tris covering a block fully,
 * which obviously wouldn't work)).
 */
static void 
emit_coef4( struct gallivm_state *gallivm,
            struct lp_setup_args *args,
            unsigned slot,
            LLVMValueRef a0,
            LLVMValueRef a1,
            LLVMValueRef a2)
{
   LLVMBuilderRef b = gallivm->builder;
   LLVMValueRef attr_0;
   LLVMValueRef dy20_ooa = args->dy20_ooa;
   LLVMValueRef dy01_ooa = args->dy01_ooa;
   LLVMValueRef dx20_ooa = args->dx20_ooa;
   LLVMValueRef dx01_ooa = args->dx01_ooa;
   LLVMValueRef x0_center = args->x0_center;
   LLVMValueRef y0_center = args->y0_center;
   LLVMValueRef da01 = LLVMBuildFSub(b, a0, a1, "da01");
   LLVMValueRef da20 = LLVMBuildFSub(b, a2, a0, "da20");

   /* Calculate dadx (vec4f)
    */
   LLVMValueRef da01_dy20_ooa = LLVMBuildFMul(b, da01, dy20_ooa, "da01_dy20_ooa");
   LLVMValueRef da20_dy01_ooa = LLVMBuildFMul(b, da20, dy01_ooa, "da20_dy01_ooa");
   LLVMValueRef dadx          = LLVMBuildFSub(b, da01_dy20_ooa, da20_dy01_ooa, "dadx");

   /* Calculate dady (vec4f)
    */
   LLVMValueRef da01_dx20_ooa = LLVMBuildFMul(b, da01, dx20_ooa, "da01_dx20_ooa");
   LLVMValueRef da20_dx01_ooa = LLVMBuildFMul(b, da20, dx01_ooa, "da20_dx01_ooa");
   LLVMValueRef dady          = LLVMBuildFSub(b, da20_dx01_ooa, da01_dx20_ooa, "dady");

   /* Calculate a0 - the attribute value at the origin
    */
   LLVMValueRef dadx_x0    = LLVMBuildFMul(b, dadx, x0_center, "dadx_x0");
   LLVMValueRef dady_y0    = LLVMBuildFMul(b, dady, y0_center, "dady_y0");
   LLVMValueRef attr_v0    = LLVMBuildFAdd(b, dadx_x0, dady_y0, "attr_v0");
   attr_0                  = LLVMBuildFSub(b, a0, attr_v0, "attr_0");

   store_coef(gallivm, args, slot, attr_0, dadx, dady);
}


static void 
emit_linear_coef( struct gallivm_state *gallivm,
                  struct lp_setup_args *args,
                  unsigned slot,
                  LLVMValueRef attribv[3])
{
   /* nothing to do anymore */
   emit_coef4(gallivm,
              args, slot, 
              attribv[0],
              attribv[1],
              attribv[2]);
}


/**
 * Compute a0, dadx and dady for a perspective-corrected interpolant,
 * for a triangle.
 * We basically multiply the vertex value by 1/w before computing
 * the plane coefficients (a0, dadx, dady).
 * Later, when we compute the value at a particular fragment position we'll
 * divide the interpolated value by the interpolated W at that fragment.
 */
static void 
apply_perspective_corr( struct gallivm_state *gallivm,
                        struct lp_setup_args *args,
                        unsigned slot,
                        LLVMValueRef attribv[3])
{
   LLVMBuilderRef b = gallivm->builder;

   /* premultiply by 1/w  (v[0][3] is always 1/w):
    */
   LLVMValueRef v0_oow = lp_build_broadcast_scalar(&args->bld,
                            vert_attrib(gallivm, args->v0, 0, 3, "v0_oow"));
   LLVMValueRef v1_oow = lp_build_broadcast_scalar(&args->bld,
                            vert_attrib(gallivm, args->v1, 0, 3, "v1_oow"));
   LLVMValueRef v2_oow = lp_build_broadcast_scalar(&args->bld,
                            vert_attrib(gallivm, args->v2, 0, 3, "v2_oow"));

   attribv[0] = LLVMBuildFMul(b, attribv[0], v0_oow, "v0_oow_v0a");
   attribv[1] = LLVMBuildFMul(b, attribv[1], v1_oow, "v1_oow_v1a");
   attribv[2] = LLVMBuildFMul(b, attribv[2], v2_oow, "v2_oow_v2a");
}


/**
 * Applys cylindrical wrapping to vertex attributes if enabled.
 * Input coordinates must be in [0, 1] range, otherwise results are undefined.
 *
 * @param cyl_wrap  TGSI_CYLINDRICAL_WRAP_x flags
 */
static void
emit_apply_cyl_wrap(struct gallivm_state *gallivm,
                    struct lp_setup_args *args,
                    uint cyl_wrap,
                    LLVMValueRef attribv[3])

{
   LLVMBuilderRef builder = gallivm->builder;
   struct lp_type type = args->bld.type;
   LLVMTypeRef float_vec_type = args->bld.vec_type;
   LLVMValueRef pos_half;
   LLVMValueRef neg_half;
   LLVMValueRef cyl_mask;
   LLVMValueRef offset;
   LLVMValueRef delta;
   LLVMValueRef one;

   if (!cyl_wrap)
      return;

   /* Constants */
   pos_half = lp_build_const_vec(gallivm, type, +0.5f);
   neg_half = lp_build_const_vec(gallivm, type, -0.5f);
   cyl_mask = lp_build_const_mask_aos(gallivm, type, cyl_wrap, 4);

   one = lp_build_const_vec(gallivm, type, 1.0f);
   one = LLVMBuildBitCast(builder, one, lp_build_int_vec_type(gallivm, type), "");
   one = LLVMBuildAnd(builder, one, cyl_mask, "");

   /* Edge v0 -> v1 */
   delta = LLVMBuildFSub(builder, attribv[1], attribv[0], "");

   offset     = lp_build_compare(gallivm, type, PIPE_FUNC_GREATER, delta, pos_half);
   offset     = LLVMBuildAnd(builder, offset, one, "");
   offset     = LLVMBuildBitCast(builder, offset, float_vec_type, "");
   attribv[0] = LLVMBuildFAdd(builder, attribv[0], offset, "");

   offset     = lp_build_compare(gallivm, type, PIPE_FUNC_LESS, delta, neg_half);
   offset     = LLVMBuildAnd(builder, offset, one, "");
   offset     = LLVMBuildBitCast(builder, offset, float_vec_type, "");
   attribv[1] = LLVMBuildFAdd(builder, attribv[1], offset, "");

   /* Edge v1 -> v2 */
   delta = LLVMBuildFSub(builder, attribv[2], attribv[1], "");

   offset     = lp_build_compare(gallivm, type, PIPE_FUNC_GREATER, delta, pos_half);
   offset     = LLVMBuildAnd(builder, offset, one, "");
   offset     = LLVMBuildBitCast(builder, offset, float_vec_type, "");
   attribv[1] = LLVMBuildFAdd(builder, attribv[1], offset, "");

   offset     = lp_build_compare(gallivm, type, PIPE_FUNC_LESS, delta, neg_half);
   offset     = LLVMBuildAnd(builder, offset, one, "");
   offset     = LLVMBuildBitCast(builder, offset, float_vec_type, "");
   attribv[2] = LLVMBuildFAdd(builder, attribv[2], offset, "");

   /* Edge v2 -> v0 */
   delta = LLVMBuildFSub(builder, attribv[0], attribv[2], "");

   offset     = lp_build_compare(gallivm, type, PIPE_FUNC_GREATER, delta, pos_half);
   offset     = LLVMBuildAnd(builder, offset, one, "");
   offset     = LLVMBuildBitCast(builder, offset, float_vec_type, "");
   attribv[2] = LLVMBuildFAdd(builder, attribv[2], offset, "");

   offset     = lp_build_compare(gallivm, type, PIPE_FUNC_LESS, delta, neg_half);
   offset     = LLVMBuildAnd(builder, offset, one, "");
   offset     = LLVMBuildBitCast(builder, offset, float_vec_type, "");
   attribv[0] = LLVMBuildFAdd(builder, attribv[0], offset, "");
}


/**
 * Compute the inputs-> dadx, dady, a0 values.
 */
static void 
emit_tri_coef( struct gallivm_state *gallivm,
               const struct lp_setup_variant_key *key,
               struct lp_setup_args *args)
{
   unsigned slot;

   LLVMValueRef attribs[3];

  /* setup interpolation for all the remaining attributes:
    */
   for (slot = 0; slot < key->num_inputs; slot++) {
      switch (key->inputs[slot].interp) {
      case LP_INTERP_CONSTANT:
         load_attribute(gallivm, args, key, key->inputs[slot].src_index, attribs);
         if (key->flatshade_first) {
            emit_constant_coef4(gallivm, args, slot+1, attribs[0]);
         }
         else {
            emit_constant_coef4(gallivm, args, slot+1, attribs[2]);
         }
         break;

      case LP_INTERP_LINEAR:
         load_attribute(gallivm, args, key, key->inputs[slot].src_index, attribs);
         emit_apply_cyl_wrap(gallivm, args, key->inputs[slot].cyl_wrap, attribs);
         emit_linear_coef(gallivm, args, slot+1, attribs);
         break;

      case LP_INTERP_PERSPECTIVE:
         load_attribute(gallivm, args, key, key->inputs[slot].src_index, attribs);
         emit_apply_cyl_wrap(gallivm, args, key->inputs[slot].cyl_wrap, attribs);
         apply_perspective_corr(gallivm, args, slot+1, attribs);
         emit_linear_coef(gallivm, args, slot+1, attribs);
         break;

      case LP_INTERP_POSITION:
         /*
          * The generated pixel interpolators will pick up the coeffs from
          * slot 0.
          */
         break;

      case LP_INTERP_FACING:
         emit_facing_coef(gallivm, args, slot+1);
         break;

      default:
         assert(0);
      }
   }
}


/* XXX: generic code:
 */
static void
set_noalias(LLVMBuilderRef builder,
            LLVMValueRef function,
            const LLVMTypeRef *arg_types,
            int nr_args)
{
   int i;
   for(i = 0; i < nr_args; ++i)
      if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
         LLVMAddAttribute(LLVMGetParam(function, i),
            LLVMNoAliasAttribute);
}

static void
init_args(struct gallivm_state *gallivm,
          const struct lp_setup_variant_key *key,
          struct lp_setup_args *args)
{
   LLVMBuilderRef b = gallivm->builder;
   LLVMTypeRef shuf_type = LLVMInt32TypeInContext(gallivm->context);
   LLVMValueRef onef = lp_build_const_float(gallivm, 1.0);
   LLVMValueRef onei = lp_build_const_int32(gallivm, 1);
   LLVMValueRef zeroi = lp_build_const_int32(gallivm, 0);
   LLVMValueRef pixel_center, xy0_center, dxy01, dxy20, dyx20;
   LLVMValueRef e, f, ef, ooa;
   LLVMValueRef shuffles[4], shuf10;
   LLVMValueRef attr_pos[3];
   struct lp_type typef4 = lp_type_float_vec(32, 128);
   struct lp_build_context bld;

   lp_build_context_init(&bld, gallivm, typef4);
   args->bld = bld;

   /* The internal position input is in slot zero:
    */
   load_attribute(gallivm, args, key, 0, attr_pos);

   pixel_center = lp_build_const_vec(gallivm, typef4,
                                     key->pixel_center_half ? 0.5 : 0.0);

   /*
    * xy are first two elems in v0a/v1a/v2a but just use vec4 arit
    * also offset_tri uses actually xyz in them
    */
   xy0_center = LLVMBuildFSub(b, attr_pos[0], pixel_center, "xy0_center" );

   dxy01 = LLVMBuildFSub(b, attr_pos[0], attr_pos[1], "dxy01");
   dxy20 = LLVMBuildFSub(b, attr_pos[2], attr_pos[0], "dxy20");

   shuffles[0] = onei;
   shuffles[1] = zeroi;
   shuffles[2] = LLVMGetUndef(shuf_type);
   shuffles[3] = LLVMGetUndef(shuf_type);
   shuf10 = LLVMConstVector(shuffles, 4);

   dyx20 = LLVMBuildShuffleVector(b, dxy20, dxy20, shuf10, "");

   ef = LLVMBuildFMul(b, dxy01, dyx20, "ef");
   e = LLVMBuildExtractElement(b, ef, zeroi, "");
   f = LLVMBuildExtractElement(b, ef, onei, "");

   ooa  = LLVMBuildFDiv(b, onef, LLVMBuildFSub(b, e, f, ""), "ooa");

   ooa = lp_build_broadcast_scalar(&bld, ooa);

   /* tri offset calc shares a lot of arithmetic, do it here */
   if (key->pgon_offset_scale != 0.0f || key->pgon_offset_units != 0.0f) {
      lp_do_offset_tri(gallivm, args, key, ooa, dxy01, dxy20, attr_pos);
   }

   dxy20 = LLVMBuildFMul(b, dxy20, ooa, "");
   dxy01 = LLVMBuildFMul(b, dxy01, ooa, "");

   args->dy20_ooa  = lp_build_extract_broadcast(gallivm, typef4, typef4, dxy20, onei);
   args->dy01_ooa  = lp_build_extract_broadcast(gallivm, typef4, typef4, dxy01, onei);

   args->dx20_ooa  = lp_build_extract_broadcast(gallivm, typef4, typef4, dxy20, zeroi);
   args->dx01_ooa  = lp_build_extract_broadcast(gallivm, typef4, typef4, dxy01, zeroi);

   args->x0_center = lp_build_extract_broadcast(gallivm, typef4, typef4, xy0_center, zeroi);
   args->y0_center = lp_build_extract_broadcast(gallivm, typef4, typef4, xy0_center, onei);

   emit_linear_coef(gallivm, args, 0, attr_pos);
}

/**
 * Generate the runtime callable function for the coefficient calculation.
 *
 */
static struct lp_setup_variant *
generate_setup_variant(struct lp_setup_variant_key *key,
                       struct llvmpipe_context *lp)
{
   struct lp_setup_variant *variant = NULL;
   struct gallivm_state *gallivm;
   struct lp_setup_args args;
   char func_name[64];
   LLVMTypeRef vec4f_type;
   LLVMTypeRef func_type;
   LLVMTypeRef arg_types[7];
   LLVMBasicBlockRef block;
   LLVMBuilderRef builder;
   int64_t t0 = 0, t1;

   if (0)
      goto fail;

   variant = CALLOC_STRUCT(lp_setup_variant);
   if (!variant)
      goto fail;

   variant->no = setup_no++;

   util_snprintf(func_name, sizeof(func_name), "setup_variant_%u",
                 variant->no);

   variant->gallivm = gallivm = gallivm_create(func_name, lp->context);
   if (!variant->gallivm) {
      goto fail;
   }

   builder = gallivm->builder;

   if (LP_DEBUG & DEBUG_COUNTERS) {
      t0 = os_time_get();
   }

   memcpy(&variant->key, key, key->size);
   variant->list_item_global.base = variant;

   /* Currently always deal with full 4-wide vertex attributes from
    * the vertices.
    */

   vec4f_type = LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), 4);

   arg_types[0] = LLVMPointerType(vec4f_type, 0);        /* v0 */
   arg_types[1] = LLVMPointerType(vec4f_type, 0);        /* v1 */
   arg_types[2] = LLVMPointerType(vec4f_type, 0);        /* v2 */
   arg_types[3] = LLVMInt32TypeInContext(gallivm->context); /* facing */
   arg_types[4] = LLVMPointerType(vec4f_type, 0);	/* a0, aligned */
   arg_types[5] = LLVMPointerType(vec4f_type, 0);	/* dadx, aligned */
   arg_types[6] = LLVMPointerType(vec4f_type, 0);	/* dady, aligned */

   func_type = LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context),
                                arg_types, Elements(arg_types), 0);

   variant->function = LLVMAddFunction(gallivm->module, func_name, func_type);
   if (!variant->function)
      goto fail;

   LLVMSetFunctionCallConv(variant->function, LLVMCCallConv);

   args.v0       = LLVMGetParam(variant->function, 0);
   args.v1       = LLVMGetParam(variant->function, 1);
   args.v2       = LLVMGetParam(variant->function, 2);
   args.facing   = LLVMGetParam(variant->function, 3);
   args.a0       = LLVMGetParam(variant->function, 4);
   args.dadx     = LLVMGetParam(variant->function, 5);
   args.dady     = LLVMGetParam(variant->function, 6);

   lp_build_name(args.v0, "in_v0");
   lp_build_name(args.v1, "in_v1");
   lp_build_name(args.v2, "in_v2");
   lp_build_name(args.facing, "in_facing");
   lp_build_name(args.a0, "out_a0");
   lp_build_name(args.dadx, "out_dadx");
   lp_build_name(args.dady, "out_dady");

   /*
    * Function body
    */
   block = LLVMAppendBasicBlockInContext(gallivm->context,
                                         variant->function, "entry");
   LLVMPositionBuilderAtEnd(builder, block);

   set_noalias(builder, variant->function, arg_types, Elements(arg_types));
   init_args(gallivm, &variant->key, &args);
   emit_tri_coef(gallivm, &variant->key, &args);

   LLVMBuildRetVoid(builder);

   gallivm_verify_function(gallivm, variant->function);

   gallivm_compile_module(gallivm);

   variant->jit_function = (lp_jit_setup_triangle)
      gallivm_jit_function(gallivm, variant->function);
   if (!variant->jit_function)
      goto fail;

   gallivm_free_ir(variant->gallivm);

   /*
    * Update timing information:
    */
   if (LP_DEBUG & DEBUG_COUNTERS) {
      t1 = os_time_get();
      LP_COUNT_ADD(llvm_compile_time, t1 - t0);
      LP_COUNT_ADD(nr_llvm_compiles, 1);
   }

   return variant;

fail:
   if (variant) {
      if (variant->gallivm) {
         gallivm_destroy(variant->gallivm);
      }
      FREE(variant);
   }

   return NULL;
}



static void
lp_make_setup_variant_key(struct llvmpipe_context *lp,
                          struct lp_setup_variant_key *key)
{
   struct lp_fragment_shader *fs = lp->fs;
   unsigned i;

   assert(sizeof key->inputs[0] == sizeof(uint));

   key->num_inputs = fs->info.base.num_inputs;
   key->flatshade_first = lp->rasterizer->flatshade_first;
   key->pixel_center_half = lp->rasterizer->half_pixel_center;
   key->twoside = lp->rasterizer->light_twoside;
   key->size = Offset(struct lp_setup_variant_key,
                      inputs[key->num_inputs]);

   key->color_slot = lp->color_slot[0];
   key->bcolor_slot = lp->bcolor_slot[0];
   key->spec_slot = lp->color_slot[1];
   key->bspec_slot = lp->bcolor_slot[1];

   /*
    * If depth is floating point, depth bias is calculated with respect
    * to the primitive's maximum Z value. Retain the original depth bias
    * value until that stage.
    */
   key->floating_point_depth = lp->floating_point_depth;

   if (key->floating_point_depth) {
      key->pgon_offset_units = (float) lp->rasterizer->offset_units;
   } else {
      key->pgon_offset_units =
         (float) (lp->rasterizer->offset_units * lp->mrd);
   }

   key->pgon_offset_scale = lp->rasterizer->offset_scale;
   key->pgon_offset_clamp = lp->rasterizer->offset_clamp;
   key->pad = 0;
   memcpy(key->inputs, fs->inputs, key->num_inputs * sizeof key->inputs[0]);
   for (i = 0; i < key->num_inputs; i++) {
      if (key->inputs[i].interp == LP_INTERP_COLOR) {
         if (lp->rasterizer->flatshade)
            key->inputs[i].interp = LP_INTERP_CONSTANT;
         else
            key->inputs[i].interp = LP_INTERP_PERSPECTIVE;
      }
   }

}


static void
remove_setup_variant(struct llvmpipe_context *lp,
                     struct lp_setup_variant *variant)
{
   if (gallivm_debug & GALLIVM_DEBUG_IR) {
      debug_printf("llvmpipe: del setup_variant #%u total %u\n",
                   variant->no, lp->nr_setup_variants);
   }

   if (variant->gallivm) {
      gallivm_destroy(variant->gallivm);
   }

   remove_from_list(&variant->list_item_global);
   lp->nr_setup_variants--;
   FREE(variant);
}



/* When the number of setup variants exceeds a threshold, cull a
 * fraction (currently a quarter) of them.
 */
static void
cull_setup_variants(struct llvmpipe_context *lp)
{
   struct pipe_context *pipe = &lp->pipe;
   int i;

   /*
    * XXX: we need to flush the context until we have some sort of reference
    * counting in fragment shaders as they may still be binned
    * Flushing alone might not be sufficient we need to wait on it too.
    */
   llvmpipe_finish(pipe, __FUNCTION__);

   for (i = 0; i < LP_MAX_SETUP_VARIANTS / 4; i++) {
      struct lp_setup_variant_list_item *item;
      if (is_empty_list(&lp->setup_variants_list)) {
         break;
      }
      item = last_elem(&lp->setup_variants_list);
      assert(item);
      assert(item->base);
      remove_setup_variant(lp, item->base);
   }
}


/**
 * Update fragment/vertex shader linkage state.  This is called just
 * prior to drawing something when some fragment-related state has
 * changed.
 */
void 
llvmpipe_update_setup(struct llvmpipe_context *lp)
{
   struct lp_setup_variant_key *key = &lp->setup_variant.key;
   struct lp_setup_variant *variant = NULL;
   struct lp_setup_variant_list_item *li;

   lp_make_setup_variant_key(lp, key);

   foreach(li, &lp->setup_variants_list) {
      if(li->base->key.size == key->size &&
         memcmp(&li->base->key, key, key->size) == 0) {
         variant = li->base;
         break;
      }
   }

   if (variant) {
      move_to_head(&lp->setup_variants_list, &variant->list_item_global);
   }
   else {
      if (lp->nr_setup_variants >= LP_MAX_SETUP_VARIANTS) {
         cull_setup_variants(lp);
      }

      variant = generate_setup_variant(key, lp);
      if (variant) {
         insert_at_head(&lp->setup_variants_list, &variant->list_item_global);
         lp->nr_setup_variants++;
      }
   }

   lp_setup_set_setup_variant(lp->setup, variant);
}

void
lp_delete_setup_variants(struct llvmpipe_context *lp)
{
   struct lp_setup_variant_list_item *li;
   li = first_elem(&lp->setup_variants_list);
   while(!at_end(&lp->setup_variants_list, li)) {
      struct lp_setup_variant_list_item *next = next_elem(li);
      remove_setup_variant(lp, li->base);
      li = next;
   }
}

void
lp_dump_setup_coef(const struct lp_setup_variant_key *key,
                   const float (*sa0)[4],
                   const float (*sdadx)[4],
                   const float (*sdady)[4])
{
   int i, slot;

   for (i = 0; i < TGSI_NUM_CHANNELS; i++) {
      float a0   = sa0  [0][i];
      float dadx = sdadx[0][i];
      float dady = sdady[0][i];

      debug_printf("POS.%c: a0 = %f, dadx = %f, dady = %f\n",
                   "xyzw"[i], a0, dadx, dady);
   }

   for (slot = 0; slot < key->num_inputs; slot++) {
      unsigned usage_mask = key->inputs[slot].usage_mask;
      for (i = 0; i < TGSI_NUM_CHANNELS; i++) {
         if (usage_mask & (1 << i)) {
            float a0   = sa0  [1 + slot][i];
            float dadx = sdadx[1 + slot][i];
            float dady = sdady[1 + slot][i];

            debug_printf("IN[%u].%c: a0 = %f, dadx = %f, dady = %f\n",
                         slot, "xyzw"[i], a0, dadx, dady);
         }
      }
   }
}