summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/llvmpipe/lp_setup_tri.c
blob: 4925da98bdb0aab54e667275f7e49ad50f9132e9 (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
/**************************************************************************
 *
 * Copyright 2007 VMware, Inc.
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 **************************************************************************/

/*
 * Binning code for triangles
 */

#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/u_rect.h"
#include "util/u_sse.h"
#include "lp_perf.h"
#include "lp_setup_context.h"
#include "lp_rast.h"
#include "lp_state_fs.h"
#include "lp_state_setup.h"
#include "lp_context.h"

#include <inttypes.h>

#define NUM_CHANNELS 4

#if defined(PIPE_ARCH_SSE)
#include <emmintrin.h>
#elif defined(_ARCH_PWR8) && defined(PIPE_ARCH_LITTLE_ENDIAN)
#include <altivec.h>
#include "util/u_pwr8.h"
#endif

static inline int
subpixel_snap(float a)
{
   return util_iround(FIXED_ONE * a);
}

static inline float
fixed_to_float(int a)
{
   return a * (1.0f / FIXED_ONE);
}


/* Position and area in fixed point coordinates */
struct fixed_position {
   int32_t x[4];
   int32_t y[4];
   int32_t dx01;
   int32_t dy01;
   int32_t dx20;
   int32_t dy20;
   int64_t area;
};


/**
 * Alloc space for a new triangle plus the input.a0/dadx/dady arrays
 * immediately after it.
 * The memory is allocated from the per-scene pool, not per-tile.
 * \param tri_size  returns number of bytes allocated
 * \param num_inputs  number of fragment shader inputs
 * \return pointer to triangle space
 */
struct lp_rast_triangle *
lp_setup_alloc_triangle(struct lp_scene *scene,
                        unsigned nr_inputs,
                        unsigned nr_planes,
                        unsigned *tri_size)
{
   unsigned input_array_sz = NUM_CHANNELS * (nr_inputs + 1) * sizeof(float);
   unsigned plane_sz = nr_planes * sizeof(struct lp_rast_plane);
   struct lp_rast_triangle *tri;

   *tri_size = (sizeof(struct lp_rast_triangle) +
                3 * input_array_sz +
                plane_sz);

   tri = lp_scene_alloc_aligned( scene, *tri_size, 16 );
   if (!tri)
      return NULL;

   tri->inputs.stride = input_array_sz;

   {
      char *a = (char *)tri;
      char *b = (char *)&GET_PLANES(tri)[nr_planes];
      assert(b - a == *tri_size);
   }

   return tri;
}

void
lp_setup_print_vertex(struct lp_setup_context *setup,
                      const char *name,
                      const float (*v)[4])
{
   const struct lp_setup_variant_key *key = &setup->setup.variant->key;
   int i, j;

   debug_printf("   wpos (%s[0]) xyzw %f %f %f %f\n",
                name,
                v[0][0], v[0][1], v[0][2], v[0][3]);

   for (i = 0; i < key->num_inputs; i++) {
      const float *in = v[key->inputs[i].src_index];

      debug_printf("  in[%d] (%s[%d]) %s%s%s%s ",
                   i, 
                   name, key->inputs[i].src_index,
                   (key->inputs[i].usage_mask & 0x1) ? "x" : " ",
                   (key->inputs[i].usage_mask & 0x2) ? "y" : " ",
                   (key->inputs[i].usage_mask & 0x4) ? "z" : " ",
                   (key->inputs[i].usage_mask & 0x8) ? "w" : " ");

      for (j = 0; j < 4; j++)
         if (key->inputs[i].usage_mask & (1<<j))
            debug_printf("%.5f ", in[j]);

      debug_printf("\n");
   }
}


/**
 * Print triangle vertex attribs (for debug).
 */
void
lp_setup_print_triangle(struct lp_setup_context *setup,
                        const float (*v0)[4],
                        const float (*v1)[4],
                        const float (*v2)[4])
{
   debug_printf("triangle\n");

   {
      const float ex = v0[0][0] - v2[0][0];
      const float ey = v0[0][1] - v2[0][1];
      const float fx = v1[0][0] - v2[0][0];
      const float fy = v1[0][1] - v2[0][1];

      /* det = cross(e,f).z */
      const float det = ex * fy - ey * fx;
      if (det < 0.0f) 
         debug_printf("   - ccw\n");
      else if (det > 0.0f)
         debug_printf("   - cw\n");
      else
         debug_printf("   - zero area\n");
   }

   lp_setup_print_vertex(setup, "v0", v0);
   lp_setup_print_vertex(setup, "v1", v1);
   lp_setup_print_vertex(setup, "v2", v2);
}


#define MAX_PLANES 8
static unsigned
lp_rast_tri_tab[MAX_PLANES+1] = {
   0,               /* should be impossible */
   LP_RAST_OP_TRIANGLE_1,
   LP_RAST_OP_TRIANGLE_2,
   LP_RAST_OP_TRIANGLE_3,
   LP_RAST_OP_TRIANGLE_4,
   LP_RAST_OP_TRIANGLE_5,
   LP_RAST_OP_TRIANGLE_6,
   LP_RAST_OP_TRIANGLE_7,
   LP_RAST_OP_TRIANGLE_8
};

static unsigned
lp_rast_32_tri_tab[MAX_PLANES+1] = {
   0,               /* should be impossible */
   LP_RAST_OP_TRIANGLE_32_1,
   LP_RAST_OP_TRIANGLE_32_2,
   LP_RAST_OP_TRIANGLE_32_3,
   LP_RAST_OP_TRIANGLE_32_4,
   LP_RAST_OP_TRIANGLE_32_5,
   LP_RAST_OP_TRIANGLE_32_6,
   LP_RAST_OP_TRIANGLE_32_7,
   LP_RAST_OP_TRIANGLE_32_8
};



/**
 * The primitive covers the whole tile- shade whole tile.
 *
 * \param tx, ty  the tile position in tiles, not pixels
 */
static boolean
lp_setup_whole_tile(struct lp_setup_context *setup,
                    const struct lp_rast_shader_inputs *inputs,
                    int tx, int ty)
{
   struct lp_scene *scene = setup->scene;

   LP_COUNT(nr_fully_covered_64);

   /* if variant is opaque and scissor doesn't effect the tile */
   if (inputs->opaque) {
      /* Several things prevent this optimization from working:
       * - For layered rendering we can't determine if this covers the same layer
       * as previous rendering (or in case of clears those actually always cover
       * all layers so optimization is impossible). Need to use fb_max_layer and
       * not setup->layer_slot to determine this since even if there's currently
       * no slot assigned previous rendering could have used one.
       * - If there were any Begin/End query commands in the scene then those
       * would get removed which would be very wrong. Furthermore, if queries
       * were just active we also can't do the optimization since to get
       * accurate query results we unfortunately need to execute the rendering
       * commands.
       */
      if (!scene->fb.zsbuf && scene->fb_max_layer == 0 && !scene->had_queries) {
         /*
          * All previous rendering will be overwritten so reset the bin.
          */
         lp_scene_bin_reset( scene, tx, ty );
      }

      LP_COUNT(nr_shade_opaque_64);
      return lp_scene_bin_cmd_with_state( scene, tx, ty,
                                          setup->fs.stored,
                                          LP_RAST_OP_SHADE_TILE_OPAQUE,
                                          lp_rast_arg_inputs(inputs) );
   } else {
      LP_COUNT(nr_shade_64);
      return lp_scene_bin_cmd_with_state( scene, tx, ty,
                                          setup->fs.stored, 
                                          LP_RAST_OP_SHADE_TILE,
                                          lp_rast_arg_inputs(inputs) );
   }
}


/**
 * Do basic setup for triangle rasterization and determine which
 * framebuffer tiles are touched.  Put the triangle in the scene's
 * bins for the tiles which we overlap.
 */
static boolean
do_triangle_ccw(struct lp_setup_context *setup,
                struct fixed_position* position,
                const float (*v0)[4],
                const float (*v1)[4],
                const float (*v2)[4],
                boolean frontfacing )
{
   struct lp_scene *scene = setup->scene;
   const struct lp_setup_variant_key *key = &setup->setup.variant->key;
   struct lp_rast_triangle *tri;
   struct lp_rast_plane *plane;
   struct u_rect bbox;
   unsigned tri_bytes;
   int nr_planes = 3;
   unsigned viewport_index = 0;
   unsigned layer = 0;
   const float (*pv)[4];

   /* Area should always be positive here */
   assert(position->area > 0);

   if (0)
      lp_setup_print_triangle(setup, v0, v1, v2);

   if (setup->flatshade_first) {
      pv = v0;
   }
   else {
      pv = v2;
   }
   if (setup->viewport_index_slot > 0) {
      unsigned *udata = (unsigned*)pv[setup->viewport_index_slot];
      viewport_index = lp_clamp_viewport_idx(*udata);
   }
   if (setup->layer_slot > 0) {
      layer = *(unsigned*)pv[setup->layer_slot];
      layer = MIN2(layer, scene->fb_max_layer);
   }

   if (setup->scissor_test) {
      nr_planes = 7;
   }
   else {
      nr_planes = 3;
   }

   /* Bounding rectangle (in pixels) */
   {
      /* Yes this is necessary to accurately calculate bounding boxes
       * with the two fill-conventions we support.  GL (normally) ends
       * up needing a bottom-left fill convention, which requires
       * slightly different rounding.
       */
      int adj = (setup->bottom_edge_rule != 0) ? 1 : 0;

      /* Inclusive x0, exclusive x1 */
      bbox.x0 =  MIN3(position->x[0], position->x[1], position->x[2]) >> FIXED_ORDER;
      bbox.x1 = (MAX3(position->x[0], position->x[1], position->x[2]) - 1) >> FIXED_ORDER;

      /* Inclusive / exclusive depending upon adj (bottom-left or top-right) */
      bbox.y0 = (MIN3(position->y[0], position->y[1], position->y[2]) + adj) >> FIXED_ORDER;
      bbox.y1 = (MAX3(position->y[0], position->y[1], position->y[2]) - 1 + adj) >> FIXED_ORDER;
   }

   if (bbox.x1 < bbox.x0 ||
       bbox.y1 < bbox.y0) {
      if (0) debug_printf("empty bounding box\n");
      LP_COUNT(nr_culled_tris);
      return TRUE;
   }

   if (!u_rect_test_intersection(&setup->draw_regions[viewport_index], &bbox)) {
      if (0) debug_printf("offscreen\n");
      LP_COUNT(nr_culled_tris);
      return TRUE;
   }

   /* Can safely discard negative regions, but need to keep hold of
    * information about when the triangle extends past screen
    * boundaries.  See trimmed_box in lp_setup_bin_triangle().
    */
   bbox.x0 = MAX2(bbox.x0, 0);
   bbox.y0 = MAX2(bbox.y0, 0);

   tri = lp_setup_alloc_triangle(scene,
                                 key->num_inputs,
                                 nr_planes,
                                 &tri_bytes);
   if (!tri)
      return FALSE;

#if 0
   tri->v[0][0] = v0[0][0];
   tri->v[1][0] = v1[0][0];
   tri->v[2][0] = v2[0][0];
   tri->v[0][1] = v0[0][1];
   tri->v[1][1] = v1[0][1];
   tri->v[2][1] = v2[0][1];
#endif

   LP_COUNT(nr_tris);

   /* Setup parameter interpolants:
    */
   setup->setup.variant->jit_function( v0,
				       v1,
				       v2,
				       frontfacing,
				       GET_A0(&tri->inputs),
				       GET_DADX(&tri->inputs),
				       GET_DADY(&tri->inputs) );

   tri->inputs.frontfacing = frontfacing;
   tri->inputs.disable = FALSE;
   tri->inputs.opaque = setup->fs.current.variant->opaque;
   tri->inputs.layer = layer;
   tri->inputs.viewport_index = viewport_index;

   if (0)
      lp_dump_setup_coef(&setup->setup.variant->key,
			 (const float (*)[4])GET_A0(&tri->inputs),
			 (const float (*)[4])GET_DADX(&tri->inputs),
			 (const float (*)[4])GET_DADY(&tri->inputs));

   plane = GET_PLANES(tri);

#if defined(PIPE_ARCH_SSE)
   if (1) {
      __m128i vertx, verty;
      __m128i shufx, shufy;
      __m128i dcdx, dcdy;
      __m128i cdx02, cdx13, cdy02, cdy13, c02, c13;
      __m128i c01, c23, unused;
      __m128i dcdx_neg_mask;
      __m128i dcdy_neg_mask;
      __m128i dcdx_zero_mask;
      __m128i top_left_flag, c_dec;
      __m128i eo, p0, p1, p2;
      __m128i zero = _mm_setzero_si128();

      vertx = _mm_load_si128((__m128i *)position->x); /* vertex x coords */
      verty = _mm_load_si128((__m128i *)position->y); /* vertex y coords */

      shufx = _mm_shuffle_epi32(vertx, _MM_SHUFFLE(3,0,2,1));
      shufy = _mm_shuffle_epi32(verty, _MM_SHUFFLE(3,0,2,1));

      dcdx = _mm_sub_epi32(verty, shufy);
      dcdy = _mm_sub_epi32(vertx, shufx);

      dcdx_neg_mask = _mm_srai_epi32(dcdx, 31);
      dcdx_zero_mask = _mm_cmpeq_epi32(dcdx, zero);
      dcdy_neg_mask = _mm_srai_epi32(dcdy, 31);

      top_left_flag = _mm_set1_epi32((setup->bottom_edge_rule == 0) ? ~0 : 0);

      c_dec = _mm_or_si128(dcdx_neg_mask,
                           _mm_and_si128(dcdx_zero_mask,
                                         _mm_xor_si128(dcdy_neg_mask,
                                                       top_left_flag)));

      /*
       * 64 bit arithmetic.
       * Note we need _signed_ mul (_mm_mul_epi32) which we emulate.
       */
      cdx02 = mm_mullohi_epi32(dcdx, vertx, &cdx13);
      cdy02 = mm_mullohi_epi32(dcdy, verty, &cdy13);
      c02 = _mm_sub_epi64(cdx02, cdy02);
      c13 = _mm_sub_epi64(cdx13, cdy13);
      c02 = _mm_sub_epi64(c02, _mm_shuffle_epi32(c_dec,
                                                 _MM_SHUFFLE(2,2,0,0)));
      c13 = _mm_sub_epi64(c13, _mm_shuffle_epi32(c_dec,
                                                 _MM_SHUFFLE(3,3,1,1)));

      /*
       * Useful for very small fbs/tris (or fewer subpixel bits) only:
       * c = _mm_sub_epi32(mm_mullo_epi32(dcdx, vertx),
       *                   mm_mullo_epi32(dcdy, verty));
       *
       * c = _mm_sub_epi32(c, c_dec);
       */

      /* Scale up to match c:
       */
      dcdx = _mm_slli_epi32(dcdx, FIXED_ORDER);
      dcdy = _mm_slli_epi32(dcdy, FIXED_ORDER);

      /*
       * Calculate trivial reject values:
       * Note eo cannot overflow even if dcdx/dcdy would already have
       * 31 bits (which they shouldn't have). This is because eo
       * is never negative (albeit if we rely on that need to be careful...)
       */
      eo = _mm_sub_epi32(_mm_andnot_si128(dcdy_neg_mask, dcdy),
                         _mm_and_si128(dcdx_neg_mask, dcdx));

      /* ei = _mm_sub_epi32(_mm_sub_epi32(dcdy, dcdx), eo); */

      /*
       * Pointless transpose which gets undone immediately in
       * rasterization.
       * It is actually difficult to do away with it - would essentially
       * need GET_PLANES_DX, GET_PLANES_DY etc., but the calculations
       * for this then would need to depend on the number of planes.
       * The transpose is quite special here due to c being 64bit...
       * The store has to be unaligned (unless we'd make the plane size
       * a multiple of 128), and of course storing eo separately...
       */
      c01 = _mm_unpacklo_epi64(c02, c13);
      c23 = _mm_unpackhi_epi64(c02, c13);
      transpose2_64_2_32(&c01, &c23, &dcdx, &dcdy,
                         &p0, &p1, &p2, &unused);
      _mm_storeu_si128((__m128i *)&plane[0], p0);
      plane[0].eo = (uint32_t)_mm_cvtsi128_si32(eo);
      _mm_storeu_si128((__m128i *)&plane[1], p1);
      eo = _mm_shuffle_epi32(eo, _MM_SHUFFLE(3,2,0,1));
      plane[1].eo = (uint32_t)_mm_cvtsi128_si32(eo);
      _mm_storeu_si128((__m128i *)&plane[2], p2);
      eo = _mm_shuffle_epi32(eo, _MM_SHUFFLE(0,0,0,2));
      plane[2].eo = (uint32_t)_mm_cvtsi128_si32(eo);
   } else
#elif defined(_ARCH_PWR8) && defined(PIPE_ARCH_LITTLE_ENDIAN)
   /*
    * XXX this code is effectively disabled for all practical purposes,
    * as the allowed fb size is tiny if FIXED_ORDER is 8.
    */
   if (setup->fb.width <= MAX_FIXED_LENGTH32 &&
       setup->fb.height <= MAX_FIXED_LENGTH32 &&
       (bbox.x1 - bbox.x0) <= MAX_FIXED_LENGTH32 &&
       (bbox.y1 - bbox.y0) <= MAX_FIXED_LENGTH32) {
      unsigned int bottom_edge;
      __m128i vertx, verty;
      __m128i shufx, shufy;
      __m128i dcdx, dcdy, c;
      __m128i unused;
      __m128i dcdx_neg_mask;
      __m128i dcdy_neg_mask;
      __m128i dcdx_zero_mask;
      __m128i top_left_flag;
      __m128i c_inc_mask, c_inc;
      __m128i eo, p0, p1, p2;
      __m128i_union vshuf_mask;
      __m128i zero = vec_splats((unsigned char) 0);
      PIPE_ALIGN_VAR(16) int32_t temp_vec[4];

#ifdef PIPE_ARCH_LITTLE_ENDIAN
      vshuf_mask.i[0] = 0x07060504;
      vshuf_mask.i[1] = 0x0B0A0908;
      vshuf_mask.i[2] = 0x03020100;
      vshuf_mask.i[3] = 0x0F0E0D0C;
#else
      vshuf_mask.i[0] = 0x00010203;
      vshuf_mask.i[1] = 0x0C0D0E0F;
      vshuf_mask.i[2] = 0x04050607;
      vshuf_mask.i[3] = 0x08090A0B;
#endif

      /* vertex x coords */
      vertx = vec_load_si128((const uint32_t *) position->x);
      /* vertex y coords */
      verty = vec_load_si128((const uint32_t *) position->y);

      shufx = vec_perm (vertx, vertx, vshuf_mask.m128i);
      shufy = vec_perm (verty, verty, vshuf_mask.m128i);

      dcdx = vec_sub_epi32(verty, shufy);
      dcdy = vec_sub_epi32(vertx, shufx);

      dcdx_neg_mask = vec_srai_epi32(dcdx, 31);
      dcdx_zero_mask = vec_cmpeq_epi32(dcdx, zero);
      dcdy_neg_mask = vec_srai_epi32(dcdy, 31);

      bottom_edge = (setup->bottom_edge_rule == 0) ? ~0 : 0;
      top_left_flag = (__m128i) vec_splats(bottom_edge);

      c_inc_mask = vec_or(dcdx_neg_mask,
                                vec_and(dcdx_zero_mask,
                                              vec_xor(dcdy_neg_mask,
                                                            top_left_flag)));

      c_inc = vec_srli_epi32(c_inc_mask, 31);

      c = vec_sub_epi32(vec_mullo_epi32(dcdx, vertx),
                        vec_mullo_epi32(dcdy, verty));

      c = vec_add_epi32(c, c_inc);

      /* Scale up to match c:
       */
      dcdx = vec_slli_epi32(dcdx, FIXED_ORDER);
      dcdy = vec_slli_epi32(dcdy, FIXED_ORDER);

      /* Calculate trivial reject values:
       */
      eo = vec_sub_epi32(vec_andc(dcdy_neg_mask, dcdy),
                         vec_and(dcdx_neg_mask, dcdx));

      /* ei = _mm_sub_epi32(_mm_sub_epi32(dcdy, dcdx), eo); */

      /* Pointless transpose which gets undone immediately in
       * rasterization:
       */
      transpose4_epi32(&c, &dcdx, &dcdy, &eo,
                       &p0, &p1, &p2, &unused);

#define STORE_PLANE(plane, vec) do {                  \
         vec_store_si128((uint32_t *)&temp_vec, vec); \
         plane.c    = (int64_t)temp_vec[0];           \
         plane.dcdx = temp_vec[1];                    \
         plane.dcdy = temp_vec[2];                    \
         plane.eo   = temp_vec[3];                    \
      } while(0)

      STORE_PLANE(plane[0], p0);
      STORE_PLANE(plane[1], p1);
      STORE_PLANE(plane[2], p2);
#undef STORE_PLANE
   } else
#endif
   {
      int i;
      plane[0].dcdy = position->dx01;
      plane[1].dcdy = position->x[1] - position->x[2];
      plane[2].dcdy = position->dx20;
      plane[0].dcdx = position->dy01;
      plane[1].dcdx = position->y[1] - position->y[2];
      plane[2].dcdx = position->dy20;
  
      for (i = 0; i < 3; i++) {
         /* half-edge constants, will be iterated over the whole render
          * target.
          */
         plane[i].c = IMUL64(plane[i].dcdx, position->x[i]) -
                      IMUL64(plane[i].dcdy, position->y[i]);

         /* correct for top-left vs. bottom-left fill convention.
          */
         if (plane[i].dcdx < 0) {
            /* both fill conventions want this - adjust for left edges */
            plane[i].c++;
         }
         else if (plane[i].dcdx == 0) {
            if (setup->bottom_edge_rule == 0){
               /* correct for top-left fill convention:
                */
               if (plane[i].dcdy > 0) plane[i].c++;
            }
            else {
               /* correct for bottom-left fill convention:
                */
               if (plane[i].dcdy < 0) plane[i].c++;
            }
         }

         /* Scale up to match c:
          */
         assert((plane[i].dcdx << FIXED_ORDER) >> FIXED_ORDER == plane[i].dcdx);
         assert((plane[i].dcdy << FIXED_ORDER) >> FIXED_ORDER == plane[i].dcdy);
         plane[i].dcdx <<= FIXED_ORDER;
         plane[i].dcdy <<= FIXED_ORDER;

         /* find trivial reject offsets for each edge for a single-pixel
          * sized block.  These will be scaled up at each recursive level to
          * match the active blocksize.  Scaling in this way works best if
          * the blocks are square.
          */
         plane[i].eo = 0;
         if (plane[i].dcdx < 0) plane[i].eo -= plane[i].dcdx;
         if (plane[i].dcdy > 0) plane[i].eo += plane[i].dcdy;
      }
   }

   if (0) {
      debug_printf("p0: %"PRIx64"/%08x/%08x/%08x\n",
                   plane[0].c,
                   plane[0].dcdx,
                   plane[0].dcdy,
                   plane[0].eo);

      debug_printf("p1: %"PRIx64"/%08x/%08x/%08x\n",
                   plane[1].c,
                   plane[1].dcdx,
                   plane[1].dcdy,
                   plane[1].eo);

      debug_printf("p2: %"PRIx64"/%08x/%08x/%08x\n",
                   plane[2].c,
                   plane[2].dcdx,
                   plane[2].dcdy,
                   plane[2].eo);
   }


   /* 
    * When rasterizing scissored tris, use the intersection of the
    * triangle bounding box and the scissor rect to generate the
    * scissor planes.
    *
    * This permits us to cut off the triangle "tails" that are present
    * in the intermediate recursive levels caused when two of the
    * triangles edges don't diverge quickly enough to trivially reject
    * exterior blocks from the triangle.
    *
    * It's not really clear if it's worth worrying about these tails,
    * but since we generate the planes for each scissored tri, it's
    * free to trim them in this case.
    * 
    * Note that otherwise, the scissor planes only vary in 'C' value,
    * and even then only on state-changes.  Could alternatively store
    * these planes elsewhere.
    */
   if (nr_planes == 7) {
      const struct u_rect *scissor = &setup->scissors[viewport_index];

      plane[3].dcdx = -1 << 8;
      plane[3].dcdy = 0;
      plane[3].c = (1-scissor->x0) << 8;
      plane[3].eo = 1 << 8;

      plane[4].dcdx = 1 << 8;
      plane[4].dcdy = 0;
      plane[4].c = (scissor->x1+1) << 8;
      plane[4].eo = 0;

      plane[5].dcdx = 0;
      plane[5].dcdy = 1 << 8;
      plane[5].c = (1-scissor->y0) << 8;
      plane[5].eo = 1 << 8;

      plane[6].dcdx = 0;
      plane[6].dcdy = -1 << 8;
      plane[6].c = (scissor->y1+1) << 8;
      plane[6].eo = 0;
   }

   return lp_setup_bin_triangle(setup, tri, &bbox, nr_planes, viewport_index);
}

/*
 * Round to nearest less or equal power of two of the input.
 *
 * Undefined if no bit set exists, so code should check against 0 first.
 */
static inline uint32_t 
floor_pot(uint32_t n)
{
#if defined(PIPE_CC_GCC) && (defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64))
   if (n == 0)
      return 0;

   __asm__("bsr %1,%0"
          : "=r" (n)
          : "rm" (n));
   return 1 << n;
#else
   n |= (n >>  1);
   n |= (n >>  2);
   n |= (n >>  4);
   n |= (n >>  8);
   n |= (n >> 16);
   return n - (n >> 1);
#endif
}


boolean
lp_setup_bin_triangle( struct lp_setup_context *setup,
                       struct lp_rast_triangle *tri,
                       const struct u_rect *bbox,
                       int nr_planes,
                       unsigned viewport_index )
{
   struct lp_scene *scene = setup->scene;
   struct u_rect trimmed_box = *bbox;   
   int i;
   /* What is the largest power-of-two boundary this triangle crosses:
    */
   int dx = floor_pot((bbox->x0 ^ bbox->x1) |
		      (bbox->y0 ^ bbox->y1));

   /* The largest dimension of the rasterized area of the triangle
    * (aligned to a 4x4 grid), rounded down to the nearest power of two:
    */
   int max_sz = ((bbox->x1 - (bbox->x0 & ~3)) |
                 (bbox->y1 - (bbox->y0 & ~3)));
   int sz = floor_pot(max_sz);
   boolean use_32bits = max_sz <= MAX_FIXED_LENGTH32;

   /* Now apply scissor, etc to the bounding box.  Could do this
    * earlier, but it confuses the logic for tri-16 and would force
    * the rasterizer to also respect scissor, etc, just for the rare
    * cases where a small triangle extends beyond the scissor.
    */
   u_rect_find_intersection(&setup->draw_regions[viewport_index],
                            &trimmed_box);

   /* Determine which tile(s) intersect the triangle's bounding box
    */
   if (dx < TILE_SIZE)
   {
      int ix0 = bbox->x0 / TILE_SIZE;
      int iy0 = bbox->y0 / TILE_SIZE;
      unsigned px = bbox->x0 & 63 & ~3;
      unsigned py = bbox->y0 & 63 & ~3;

      assert(iy0 == bbox->y1 / TILE_SIZE &&
	     ix0 == bbox->x1 / TILE_SIZE);

      if (nr_planes == 3) {
         if (sz < 4)
         {
            /* Triangle is contained in a single 4x4 stamp:
             */
            assert(px + 4 <= TILE_SIZE);
            assert(py + 4 <= TILE_SIZE);
            return lp_scene_bin_cmd_with_state( scene, ix0, iy0,
                                                setup->fs.stored,
                                                use_32bits ?
                                                LP_RAST_OP_TRIANGLE_32_3_4 :
                                                LP_RAST_OP_TRIANGLE_3_4,
                                                lp_rast_arg_triangle_contained(tri, px, py) );
         }

         if (sz < 16)
         {
            /* Triangle is contained in a single 16x16 block:
             */

            /*
             * The 16x16 block is only 4x4 aligned, and can exceed the tile
             * dimensions if the triangle is 16 pixels in one dimension but 4
             * in the other. So budge the 16x16 back inside the tile.
             */
            px = MIN2(px, TILE_SIZE - 16);
            py = MIN2(py, TILE_SIZE - 16);

            assert(px + 16 <= TILE_SIZE);
            assert(py + 16 <= TILE_SIZE);

            return lp_scene_bin_cmd_with_state( scene, ix0, iy0,
                                                setup->fs.stored,
                                                use_32bits ?
                                                LP_RAST_OP_TRIANGLE_32_3_16 :
                                                LP_RAST_OP_TRIANGLE_3_16,
                                                lp_rast_arg_triangle_contained(tri, px, py) );
         }
      }
      else if (nr_planes == 4 && sz < 16) 
      {
         px = MIN2(px, TILE_SIZE - 16);
         py = MIN2(py, TILE_SIZE - 16);

         assert(px + 16 <= TILE_SIZE);
         assert(py + 16 <= TILE_SIZE);

         return lp_scene_bin_cmd_with_state(scene, ix0, iy0,
                                            setup->fs.stored,
                                            use_32bits ?
                                            LP_RAST_OP_TRIANGLE_32_4_16 :
                                            LP_RAST_OP_TRIANGLE_4_16,
                                            lp_rast_arg_triangle_contained(tri, px, py));
      }


      /* Triangle is contained in a single tile:
       */
      return lp_scene_bin_cmd_with_state(
         scene, ix0, iy0, setup->fs.stored,
         use_32bits ? lp_rast_32_tri_tab[nr_planes] : lp_rast_tri_tab[nr_planes],
         lp_rast_arg_triangle(tri, (1<<nr_planes)-1));
   }
   else
   {
      struct lp_rast_plane *plane = GET_PLANES(tri);
      int64_t c[MAX_PLANES];
      int64_t ei[MAX_PLANES];

      int64_t eo[MAX_PLANES];
      int64_t xstep[MAX_PLANES];
      int64_t ystep[MAX_PLANES];
      int x, y;

      int ix0 = trimmed_box.x0 / TILE_SIZE;
      int iy0 = trimmed_box.y0 / TILE_SIZE;
      int ix1 = trimmed_box.x1 / TILE_SIZE;
      int iy1 = trimmed_box.y1 / TILE_SIZE;
      
      for (i = 0; i < nr_planes; i++) {
         c[i] = (plane[i].c + 
                 IMUL64(plane[i].dcdy, iy0) * TILE_SIZE -
                 IMUL64(plane[i].dcdx, ix0) * TILE_SIZE);

         ei[i] = (plane[i].dcdy - 
                  plane[i].dcdx - 
                  (int64_t)plane[i].eo) << TILE_ORDER;

         eo[i] = (int64_t)plane[i].eo << TILE_ORDER;
         xstep[i] = -(((int64_t)plane[i].dcdx) << TILE_ORDER);
         ystep[i] = ((int64_t)plane[i].dcdy) << TILE_ORDER;
      }



      /* Test tile-sized blocks against the triangle.
       * Discard blocks fully outside the tri.  If the block is fully
       * contained inside the tri, bin an lp_rast_shade_tile command.
       * Else, bin a lp_rast_triangle command.
       */
      for (y = iy0; y <= iy1; y++)
      {
         boolean in = FALSE;  /* are we inside the triangle? */
         int64_t cx[MAX_PLANES];

         for (i = 0; i < nr_planes; i++)
            cx[i] = c[i];

         for (x = ix0; x <= ix1; x++)
         {
            int out = 0;
            int partial = 0;

            for (i = 0; i < nr_planes; i++) {
               int64_t planeout = cx[i] + eo[i];
               int64_t planepartial = cx[i] + ei[i] - 1;
               out |= (int) (planeout >> 63);
               partial |= ((int) (planepartial >> 63)) & (1<<i);
            }

            if (out) {
               /* do nothing */
               if (in)
                  break;  /* exiting triangle, all done with this row */
               LP_COUNT(nr_empty_64);
            }
            else if (partial) {
               /* Not trivially accepted by at least one plane -
                * rasterize/shade partial tile
                */
               int count = util_bitcount(partial);
               in = TRUE;
               
               if (!lp_scene_bin_cmd_with_state( scene, x, y,
                                                 setup->fs.stored,
                                                 use_32bits ?
                                                 lp_rast_32_tri_tab[count] :
                                                 lp_rast_tri_tab[count],
                                                 lp_rast_arg_triangle(tri, partial) ))
                  goto fail;

               LP_COUNT(nr_partially_covered_64);
            }
            else {
               /* triangle covers the whole tile- shade whole tile */
               LP_COUNT(nr_fully_covered_64);
               in = TRUE;
               if (!lp_setup_whole_tile(setup, &tri->inputs, x, y))
                  goto fail;
            }

            /* Iterate cx values across the region: */
            for (i = 0; i < nr_planes; i++)
               cx[i] += xstep[i];
         }

         /* Iterate c values down the region: */
         for (i = 0; i < nr_planes; i++)
            c[i] += ystep[i];
      }
   }

   return TRUE;

fail:
   /* Need to disable any partially binned triangle.  This is easier
    * than trying to locate all the triangle, shade-tile, etc,
    * commands which may have been binned.
    */
   tri->inputs.disable = TRUE;
   return FALSE;
}


/**
 * Try to draw the triangle, restart the scene on failure.
 */
static void retry_triangle_ccw( struct lp_setup_context *setup,
                                struct fixed_position* position,
                                const float (*v0)[4],
                                const float (*v1)[4],
                                const float (*v2)[4],
                                boolean front)
{
   if (!do_triangle_ccw( setup, position, v0, v1, v2, front ))
   {
      if (!lp_setup_flush_and_restart(setup))
         return;

      if (!do_triangle_ccw( setup, position, v0, v1, v2, front ))
         return;
   }
}

/**
 * Calculate fixed position data for a triangle
 * It is unfortunate we need to do that here (as we need area
 * calculated in fixed point), as there's quite some code duplication
 * to what is done in the jit setup prog.
 */
static inline void
calc_fixed_position(struct lp_setup_context *setup,
                    struct fixed_position* position,
                    const float (*v0)[4],
                    const float (*v1)[4],
                    const float (*v2)[4])
{
   /*
    * The rounding may not be quite the same with PIPE_ARCH_SSE
    * (util_iround right now only does nearest/even on x87,
    * otherwise nearest/away-from-zero).
    * Both should be acceptable, I think.
    */
#if defined(PIPE_ARCH_SSE)
   __m128d v0r, v1r, v2r;
   __m128 vxy0xy2, vxy1xy0;
   __m128i vxy0xy2i, vxy1xy0i;
   __m128i dxdy0120, x0x2y0y2, x1x0y1y0, x0120, y0120;
   __m128 pix_offset = _mm_set1_ps(setup->pixel_offset);
   __m128 fixed_one = _mm_set1_ps((float)FIXED_ONE);
   v0r = _mm_load_sd((const double *)v0[0]);
   v1r = _mm_load_sd((const double *)v1[0]);
   v2r = _mm_load_sd((const double *)v2[0]);
   vxy0xy2 = (__m128)_mm_unpacklo_pd(v0r, v2r);
   vxy1xy0 = (__m128)_mm_unpacklo_pd(v1r, v0r);
   vxy0xy2 = _mm_sub_ps(vxy0xy2, pix_offset);
   vxy1xy0 = _mm_sub_ps(vxy1xy0, pix_offset);
   vxy0xy2 = _mm_mul_ps(vxy0xy2, fixed_one);
   vxy1xy0 = _mm_mul_ps(vxy1xy0, fixed_one);
   vxy0xy2i = _mm_cvtps_epi32(vxy0xy2);
   vxy1xy0i = _mm_cvtps_epi32(vxy1xy0);
   dxdy0120 = _mm_sub_epi32(vxy0xy2i, vxy1xy0i);
   _mm_store_si128((__m128i *)&position->dx01, dxdy0120);
   /*
    * For the mul, would need some more shuffles, plus emulation
    * for the signed mul (without sse41), so don't bother.
    */
   x0x2y0y2 = _mm_shuffle_epi32(vxy0xy2i, _MM_SHUFFLE(3,1,2,0));
   x1x0y1y0 = _mm_shuffle_epi32(vxy1xy0i, _MM_SHUFFLE(3,1,2,0));
   x0120 = _mm_unpacklo_epi32(x0x2y0y2, x1x0y1y0);
   y0120 = _mm_unpackhi_epi32(x0x2y0y2, x1x0y1y0);
   _mm_store_si128((__m128i *)&position->x[0], x0120);
   _mm_store_si128((__m128i *)&position->y[0], y0120);

#else
   position->x[0] = subpixel_snap(v0[0][0] - setup->pixel_offset);
   position->x[1] = subpixel_snap(v1[0][0] - setup->pixel_offset);
   position->x[2] = subpixel_snap(v2[0][0] - setup->pixel_offset);
   position->x[3] = 0; // should be unused

   position->y[0] = subpixel_snap(v0[0][1] - setup->pixel_offset);
   position->y[1] = subpixel_snap(v1[0][1] - setup->pixel_offset);
   position->y[2] = subpixel_snap(v2[0][1] - setup->pixel_offset);
   position->y[3] = 0; // should be unused

   position->dx01 = position->x[0] - position->x[1];
   position->dy01 = position->y[0] - position->y[1];

   position->dx20 = position->x[2] - position->x[0];
   position->dy20 = position->y[2] - position->y[0];
#endif

   position->area = IMUL64(position->dx01, position->dy20) -
         IMUL64(position->dx20, position->dy01);
}


/**
 * Rotate a triangle, flipping its clockwise direction,
 * Swaps values for xy[0] and xy[1]
 */
static inline void
rotate_fixed_position_01( struct fixed_position* position )
{
   int x, y;

   x = position->x[1];
   y = position->y[1];
   position->x[1] = position->x[0];
   position->y[1] = position->y[0];
   position->x[0] = x;
   position->y[0] = y;

   position->dx01 = -position->dx01;
   position->dy01 = -position->dy01;
   position->dx20 = position->x[2] - position->x[0];
   position->dy20 = position->y[2] - position->y[0];

   position->area = -position->area;
}


/**
 * Rotate a triangle, flipping its clockwise direction,
 * Swaps values for xy[1] and xy[2]
 */
static inline void
rotate_fixed_position_12( struct fixed_position* position )
{
   int x, y;

   x = position->x[2];
   y = position->y[2];
   position->x[2] = position->x[1];
   position->y[2] = position->y[1];
   position->x[1] = x;
   position->y[1] = y;

   x = position->dx01;
   y = position->dy01;
   position->dx01 = -position->dx20;
   position->dy01 = -position->dy20;
   position->dx20 = -x;
   position->dy20 = -y;

   position->area = -position->area;
}


/**
 * Draw triangle if it's CW, cull otherwise.
 */
static void triangle_cw(struct lp_setup_context *setup,
                        const float (*v0)[4],
                        const float (*v1)[4],
                        const float (*v2)[4])
{
   PIPE_ALIGN_VAR(16) struct fixed_position position;

   calc_fixed_position(setup, &position, v0, v1, v2);

   if (position.area < 0) {
      if (setup->flatshade_first) {
         rotate_fixed_position_12(&position);
         retry_triangle_ccw(setup, &position, v0, v2, v1, !setup->ccw_is_frontface);
      } else {
         rotate_fixed_position_01(&position);
         retry_triangle_ccw(setup, &position, v1, v0, v2, !setup->ccw_is_frontface);
      }
   }
}


static void triangle_ccw(struct lp_setup_context *setup,
                         const float (*v0)[4],
                         const float (*v1)[4],
                         const float (*v2)[4])
{
   PIPE_ALIGN_VAR(16) struct fixed_position position;

   calc_fixed_position(setup, &position, v0, v1, v2);

   if (position.area > 0)
      retry_triangle_ccw(setup, &position, v0, v1, v2, setup->ccw_is_frontface);
}

/**
 * Draw triangle whether it's CW or CCW.
 */
static void triangle_both(struct lp_setup_context *setup,
                          const float (*v0)[4],
                          const float (*v1)[4],
                          const float (*v2)[4])
{
   PIPE_ALIGN_VAR(16) struct fixed_position position;
   struct llvmpipe_context *lp_context = (struct llvmpipe_context *)setup->pipe;

   if (lp_context->active_statistics_queries &&
       !llvmpipe_rasterization_disabled(lp_context)) {
      lp_context->pipeline_statistics.c_primitives++;
   }

   calc_fixed_position(setup, &position, v0, v1, v2);

   if (0) {
      assert(!util_is_inf_or_nan(v0[0][0]));
      assert(!util_is_inf_or_nan(v0[0][1]));
      assert(!util_is_inf_or_nan(v1[0][0]));
      assert(!util_is_inf_or_nan(v1[0][1]));
      assert(!util_is_inf_or_nan(v2[0][0]));
      assert(!util_is_inf_or_nan(v2[0][1]));
   }

   if (position.area > 0)
      retry_triangle_ccw( setup, &position, v0, v1, v2, setup->ccw_is_frontface );
   else if (position.area < 0) {
      if (setup->flatshade_first) {
         rotate_fixed_position_12( &position );
         retry_triangle_ccw( setup, &position, v0, v2, v1, !setup->ccw_is_frontface );
      } else {
         rotate_fixed_position_01( &position );
         retry_triangle_ccw( setup, &position, v1, v0, v2, !setup->ccw_is_frontface );
      }
   }
}


static void triangle_nop( struct lp_setup_context *setup,
			  const float (*v0)[4],
			  const float (*v1)[4],
			  const float (*v2)[4] )
{
}


void 
lp_setup_choose_triangle( struct lp_setup_context *setup )
{
   switch (setup->cullmode) {
   case PIPE_FACE_NONE:
      setup->triangle = triangle_both;
      break;
   case PIPE_FACE_BACK:
      setup->triangle = setup->ccw_is_frontface ? triangle_ccw : triangle_cw;
      break;
   case PIPE_FACE_FRONT:
      setup->triangle = setup->ccw_is_frontface ? triangle_cw : triangle_ccw;
      break;
   default:
      setup->triangle = triangle_nop;
      break;
   }
}