gallium: Drop unused X2D opcode.
[mesa.git] / src / gallium / docs / source / tgsi.rst
1 TGSI
2 ====
3
4 TGSI, Tungsten Graphics Shader Infrastructure, is an intermediate language
5 for describing shaders. Since Gallium is inherently shaderful, shaders are
6 an important part of the API. TGSI is the only intermediate representation
7 used by all drivers.
8
9 Basics
10 ------
11
12 All TGSI instructions, known as *opcodes*, operate on arbitrary-precision
13 floating-point four-component vectors. An opcode may have up to one
14 destination register, known as *dst*, and between zero and three source
15 registers, called *src0* through *src2*, or simply *src* if there is only
16 one.
17
18 Some instructions, like :opcode:`I2F`, permit re-interpretation of vector
19 components as integers. Other instructions permit using registers as
20 two-component vectors with double precision; see :ref:`doubleopcodes`.
21
22 When an instruction has a scalar result, the result is usually copied into
23 each of the components of *dst*. When this happens, the result is said to be
24 *replicated* to *dst*. :opcode:`RCP` is one such instruction.
25
26 Modifiers
27 ^^^^^^^^^^^^^^^
28
29 TGSI supports modifiers on inputs (as well as saturate modifier on instructions).
30
31 For inputs which have a floating point type, both absolute value and negation
32 modifiers are supported (with absolute value being applied first).
33 TGSI_OPCODE_MOV is considered to have float input type for applying modifiers.
34
35 For inputs which have signed or unsigned type only the negate modifier is
36 supported.
37
38 Instruction Set
39 ---------------
40
41 Core ISA
42 ^^^^^^^^^^^^^^^^^^^^^^^^^
43
44 These opcodes are guaranteed to be available regardless of the driver being
45 used.
46
47 .. opcode:: ARL - Address Register Load
48
49 .. math::
50
51 dst.x = \lfloor src.x\rfloor
52
53 dst.y = \lfloor src.y\rfloor
54
55 dst.z = \lfloor src.z\rfloor
56
57 dst.w = \lfloor src.w\rfloor
58
59
60 .. opcode:: MOV - Move
61
62 .. math::
63
64 dst.x = src.x
65
66 dst.y = src.y
67
68 dst.z = src.z
69
70 dst.w = src.w
71
72
73 .. opcode:: LIT - Light Coefficients
74
75 .. math::
76
77 dst.x &= 1 \\
78 dst.y &= max(src.x, 0) \\
79 dst.z &= (src.x > 0) ? max(src.y, 0)^{clamp(src.w, -128, 128))} : 0 \\
80 dst.w &= 1
81
82
83 .. opcode:: RCP - Reciprocal
84
85 This instruction replicates its result.
86
87 .. math::
88
89 dst = \frac{1}{src.x}
90
91
92 .. opcode:: RSQ - Reciprocal Square Root
93
94 This instruction replicates its result. The results are undefined for src <= 0.
95
96 .. math::
97
98 dst = \frac{1}{\sqrt{src.x}}
99
100
101 .. opcode:: SQRT - Square Root
102
103 This instruction replicates its result. The results are undefined for src < 0.
104
105 .. math::
106
107 dst = {\sqrt{src.x}}
108
109
110 .. opcode:: EXP - Approximate Exponential Base 2
111
112 .. math::
113
114 dst.x &= 2^{\lfloor src.x\rfloor} \\
115 dst.y &= src.x - \lfloor src.x\rfloor \\
116 dst.z &= 2^{src.x} \\
117 dst.w &= 1
118
119
120 .. opcode:: LOG - Approximate Logarithm Base 2
121
122 .. math::
123
124 dst.x &= \lfloor\log_2{|src.x|}\rfloor \\
125 dst.y &= \frac{|src.x|}{2^{\lfloor\log_2{|src.x|}\rfloor}} \\
126 dst.z &= \log_2{|src.x|} \\
127 dst.w &= 1
128
129
130 .. opcode:: MUL - Multiply
131
132 .. math::
133
134 dst.x = src0.x \times src1.x
135
136 dst.y = src0.y \times src1.y
137
138 dst.z = src0.z \times src1.z
139
140 dst.w = src0.w \times src1.w
141
142
143 .. opcode:: ADD - Add
144
145 .. math::
146
147 dst.x = src0.x + src1.x
148
149 dst.y = src0.y + src1.y
150
151 dst.z = src0.z + src1.z
152
153 dst.w = src0.w + src1.w
154
155
156 .. opcode:: DP3 - 3-component Dot Product
157
158 This instruction replicates its result.
159
160 .. math::
161
162 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z
163
164
165 .. opcode:: DP4 - 4-component Dot Product
166
167 This instruction replicates its result.
168
169 .. math::
170
171 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z + src0.w \times src1.w
172
173
174 .. opcode:: DST - Distance Vector
175
176 .. math::
177
178 dst.x &= 1\\
179 dst.y &= src0.y \times src1.y\\
180 dst.z &= src0.z\\
181 dst.w &= src1.w
182
183
184 .. opcode:: MIN - Minimum
185
186 .. math::
187
188 dst.x = min(src0.x, src1.x)
189
190 dst.y = min(src0.y, src1.y)
191
192 dst.z = min(src0.z, src1.z)
193
194 dst.w = min(src0.w, src1.w)
195
196
197 .. opcode:: MAX - Maximum
198
199 .. math::
200
201 dst.x = max(src0.x, src1.x)
202
203 dst.y = max(src0.y, src1.y)
204
205 dst.z = max(src0.z, src1.z)
206
207 dst.w = max(src0.w, src1.w)
208
209
210 .. opcode:: SLT - Set On Less Than
211
212 .. math::
213
214 dst.x = (src0.x < src1.x) ? 1.0F : 0.0F
215
216 dst.y = (src0.y < src1.y) ? 1.0F : 0.0F
217
218 dst.z = (src0.z < src1.z) ? 1.0F : 0.0F
219
220 dst.w = (src0.w < src1.w) ? 1.0F : 0.0F
221
222
223 .. opcode:: SGE - Set On Greater Equal Than
224
225 .. math::
226
227 dst.x = (src0.x >= src1.x) ? 1.0F : 0.0F
228
229 dst.y = (src0.y >= src1.y) ? 1.0F : 0.0F
230
231 dst.z = (src0.z >= src1.z) ? 1.0F : 0.0F
232
233 dst.w = (src0.w >= src1.w) ? 1.0F : 0.0F
234
235
236 .. opcode:: MAD - Multiply And Add
237
238 .. math::
239
240 dst.x = src0.x \times src1.x + src2.x
241
242 dst.y = src0.y \times src1.y + src2.y
243
244 dst.z = src0.z \times src1.z + src2.z
245
246 dst.w = src0.w \times src1.w + src2.w
247
248
249 .. opcode:: SUB - Subtract
250
251 .. math::
252
253 dst.x = src0.x - src1.x
254
255 dst.y = src0.y - src1.y
256
257 dst.z = src0.z - src1.z
258
259 dst.w = src0.w - src1.w
260
261
262 .. opcode:: LRP - Linear Interpolate
263
264 .. math::
265
266 dst.x = src0.x \times src1.x + (1 - src0.x) \times src2.x
267
268 dst.y = src0.y \times src1.y + (1 - src0.y) \times src2.y
269
270 dst.z = src0.z \times src1.z + (1 - src0.z) \times src2.z
271
272 dst.w = src0.w \times src1.w + (1 - src0.w) \times src2.w
273
274
275 .. opcode:: CND - Condition
276
277 .. math::
278
279 dst.x = (src2.x > 0.5) ? src0.x : src1.x
280
281 dst.y = (src2.y > 0.5) ? src0.y : src1.y
282
283 dst.z = (src2.z > 0.5) ? src0.z : src1.z
284
285 dst.w = (src2.w > 0.5) ? src0.w : src1.w
286
287
288 .. opcode:: DP2A - 2-component Dot Product And Add
289
290 .. math::
291
292 dst.x = src0.x \times src1.x + src0.y \times src1.y + src2.x
293
294 dst.y = src0.x \times src1.x + src0.y \times src1.y + src2.x
295
296 dst.z = src0.x \times src1.x + src0.y \times src1.y + src2.x
297
298 dst.w = src0.x \times src1.x + src0.y \times src1.y + src2.x
299
300
301 .. opcode:: FRC - Fraction
302
303 .. math::
304
305 dst.x = src.x - \lfloor src.x\rfloor
306
307 dst.y = src.y - \lfloor src.y\rfloor
308
309 dst.z = src.z - \lfloor src.z\rfloor
310
311 dst.w = src.w - \lfloor src.w\rfloor
312
313
314 .. opcode:: CLAMP - Clamp
315
316 .. math::
317
318 dst.x = clamp(src0.x, src1.x, src2.x)
319
320 dst.y = clamp(src0.y, src1.y, src2.y)
321
322 dst.z = clamp(src0.z, src1.z, src2.z)
323
324 dst.w = clamp(src0.w, src1.w, src2.w)
325
326
327 .. opcode:: FLR - Floor
328
329 This is identical to :opcode:`ARL`.
330
331 .. math::
332
333 dst.x = \lfloor src.x\rfloor
334
335 dst.y = \lfloor src.y\rfloor
336
337 dst.z = \lfloor src.z\rfloor
338
339 dst.w = \lfloor src.w\rfloor
340
341
342 .. opcode:: ROUND - Round
343
344 .. math::
345
346 dst.x = round(src.x)
347
348 dst.y = round(src.y)
349
350 dst.z = round(src.z)
351
352 dst.w = round(src.w)
353
354
355 .. opcode:: EX2 - Exponential Base 2
356
357 This instruction replicates its result.
358
359 .. math::
360
361 dst = 2^{src.x}
362
363
364 .. opcode:: LG2 - Logarithm Base 2
365
366 This instruction replicates its result.
367
368 .. math::
369
370 dst = \log_2{src.x}
371
372
373 .. opcode:: POW - Power
374
375 This instruction replicates its result.
376
377 .. math::
378
379 dst = src0.x^{src1.x}
380
381 .. opcode:: XPD - Cross Product
382
383 .. math::
384
385 dst.x = src0.y \times src1.z - src1.y \times src0.z
386
387 dst.y = src0.z \times src1.x - src1.z \times src0.x
388
389 dst.z = src0.x \times src1.y - src1.x \times src0.y
390
391 dst.w = 1
392
393
394 .. opcode:: ABS - Absolute
395
396 .. math::
397
398 dst.x = |src.x|
399
400 dst.y = |src.y|
401
402 dst.z = |src.z|
403
404 dst.w = |src.w|
405
406
407 .. opcode:: DPH - Homogeneous Dot Product
408
409 This instruction replicates its result.
410
411 .. math::
412
413 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z + src1.w
414
415
416 .. opcode:: COS - Cosine
417
418 This instruction replicates its result.
419
420 .. math::
421
422 dst = \cos{src.x}
423
424
425 .. opcode:: DDX, DDX_FINE - Derivative Relative To X
426
427 The fine variant is only used when ``PIPE_CAP_TGSI_FS_FINE_DERIVATIVE`` is
428 advertised. When it is, the fine version guarantees one derivative per row
429 while DDX is allowed to be the same for the entire 2x2 quad.
430
431 .. math::
432
433 dst.x = partialx(src.x)
434
435 dst.y = partialx(src.y)
436
437 dst.z = partialx(src.z)
438
439 dst.w = partialx(src.w)
440
441
442 .. opcode:: DDY, DDY_FINE - Derivative Relative To Y
443
444 The fine variant is only used when ``PIPE_CAP_TGSI_FS_FINE_DERIVATIVE`` is
445 advertised. When it is, the fine version guarantees one derivative per column
446 while DDY is allowed to be the same for the entire 2x2 quad.
447
448 .. math::
449
450 dst.x = partialy(src.x)
451
452 dst.y = partialy(src.y)
453
454 dst.z = partialy(src.z)
455
456 dst.w = partialy(src.w)
457
458
459 .. opcode:: PK2H - Pack Two 16-bit Floats
460
461 TBD
462
463
464 .. opcode:: PK2US - Pack Two Unsigned 16-bit Scalars
465
466 TBD
467
468
469 .. opcode:: PK4B - Pack Four Signed 8-bit Scalars
470
471 TBD
472
473
474 .. opcode:: PK4UB - Pack Four Unsigned 8-bit Scalars
475
476 TBD
477
478
479 .. opcode:: RFL - Reflection Vector
480
481 .. math::
482
483 dst.x = 2 \times (src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z) / (src0.x \times src0.x + src0.y \times src0.y + src0.z \times src0.z) \times src0.x - src1.x
484
485 dst.y = 2 \times (src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z) / (src0.x \times src0.x + src0.y \times src0.y + src0.z \times src0.z) \times src0.y - src1.y
486
487 dst.z = 2 \times (src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z) / (src0.x \times src0.x + src0.y \times src0.y + src0.z \times src0.z) \times src0.z - src1.z
488
489 dst.w = 1
490
491 .. note::
492
493 Considered for removal.
494
495
496 .. opcode:: SEQ - Set On Equal
497
498 .. math::
499
500 dst.x = (src0.x == src1.x) ? 1.0F : 0.0F
501
502 dst.y = (src0.y == src1.y) ? 1.0F : 0.0F
503
504 dst.z = (src0.z == src1.z) ? 1.0F : 0.0F
505
506 dst.w = (src0.w == src1.w) ? 1.0F : 0.0F
507
508
509 .. opcode:: SFL - Set On False
510
511 This instruction replicates its result.
512
513 .. math::
514
515 dst = 0.0F
516
517 .. note::
518
519 Considered for removal.
520
521
522 .. opcode:: SGT - Set On Greater Than
523
524 .. math::
525
526 dst.x = (src0.x > src1.x) ? 1.0F : 0.0F
527
528 dst.y = (src0.y > src1.y) ? 1.0F : 0.0F
529
530 dst.z = (src0.z > src1.z) ? 1.0F : 0.0F
531
532 dst.w = (src0.w > src1.w) ? 1.0F : 0.0F
533
534
535 .. opcode:: SIN - Sine
536
537 This instruction replicates its result.
538
539 .. math::
540
541 dst = \sin{src.x}
542
543
544 .. opcode:: SLE - Set On Less Equal Than
545
546 .. math::
547
548 dst.x = (src0.x <= src1.x) ? 1.0F : 0.0F
549
550 dst.y = (src0.y <= src1.y) ? 1.0F : 0.0F
551
552 dst.z = (src0.z <= src1.z) ? 1.0F : 0.0F
553
554 dst.w = (src0.w <= src1.w) ? 1.0F : 0.0F
555
556
557 .. opcode:: SNE - Set On Not Equal
558
559 .. math::
560
561 dst.x = (src0.x != src1.x) ? 1.0F : 0.0F
562
563 dst.y = (src0.y != src1.y) ? 1.0F : 0.0F
564
565 dst.z = (src0.z != src1.z) ? 1.0F : 0.0F
566
567 dst.w = (src0.w != src1.w) ? 1.0F : 0.0F
568
569
570 .. opcode:: STR - Set On True
571
572 This instruction replicates its result.
573
574 .. math::
575
576 dst = 1.0F
577
578
579 .. opcode:: TEX - Texture Lookup
580
581 for array textures src0.y contains the slice for 1D,
582 and src0.z contain the slice for 2D.
583
584 for shadow textures with no arrays (and not cube map),
585 src0.z contains the reference value.
586
587 for shadow textures with arrays, src0.z contains
588 the reference value for 1D arrays, and src0.w contains
589 the reference value for 2D arrays and cube maps.
590
591 for cube map array shadow textures, the reference value
592 cannot be passed in src0.w, and TEX2 must be used instead.
593
594 .. math::
595
596 coord = src0
597
598 shadow_ref = src0.z or src0.w (optional)
599
600 unit = src1
601
602 dst = texture\_sample(unit, coord, shadow_ref)
603
604
605 .. opcode:: TEX2 - Texture Lookup (for shadow cube map arrays only)
606
607 this is the same as TEX, but uses another reg to encode the
608 reference value.
609
610 .. math::
611
612 coord = src0
613
614 shadow_ref = src1.x
615
616 unit = src2
617
618 dst = texture\_sample(unit, coord, shadow_ref)
619
620
621
622
623 .. opcode:: TXD - Texture Lookup with Derivatives
624
625 .. math::
626
627 coord = src0
628
629 ddx = src1
630
631 ddy = src2
632
633 unit = src3
634
635 dst = texture\_sample\_deriv(unit, coord, ddx, ddy)
636
637
638 .. opcode:: TXP - Projective Texture Lookup
639
640 .. math::
641
642 coord.x = src0.x / src0.w
643
644 coord.y = src0.y / src0.w
645
646 coord.z = src0.z / src0.w
647
648 coord.w = src0.w
649
650 unit = src1
651
652 dst = texture\_sample(unit, coord)
653
654
655 .. opcode:: UP2H - Unpack Two 16-Bit Floats
656
657 TBD
658
659 .. note::
660
661 Considered for removal.
662
663 .. opcode:: UP2US - Unpack Two Unsigned 16-Bit Scalars
664
665 TBD
666
667 .. note::
668
669 Considered for removal.
670
671 .. opcode:: UP4B - Unpack Four Signed 8-Bit Values
672
673 TBD
674
675 .. note::
676
677 Considered for removal.
678
679 .. opcode:: UP4UB - Unpack Four Unsigned 8-Bit Scalars
680
681 TBD
682
683 .. note::
684
685 Considered for removal.
686
687
688 .. opcode:: ARR - Address Register Load With Round
689
690 .. math::
691
692 dst.x = round(src.x)
693
694 dst.y = round(src.y)
695
696 dst.z = round(src.z)
697
698 dst.w = round(src.w)
699
700
701 .. opcode:: SSG - Set Sign
702
703 .. math::
704
705 dst.x = (src.x > 0) ? 1 : (src.x < 0) ? -1 : 0
706
707 dst.y = (src.y > 0) ? 1 : (src.y < 0) ? -1 : 0
708
709 dst.z = (src.z > 0) ? 1 : (src.z < 0) ? -1 : 0
710
711 dst.w = (src.w > 0) ? 1 : (src.w < 0) ? -1 : 0
712
713
714 .. opcode:: CMP - Compare
715
716 .. math::
717
718 dst.x = (src0.x < 0) ? src1.x : src2.x
719
720 dst.y = (src0.y < 0) ? src1.y : src2.y
721
722 dst.z = (src0.z < 0) ? src1.z : src2.z
723
724 dst.w = (src0.w < 0) ? src1.w : src2.w
725
726
727 .. opcode:: KILL_IF - Conditional Discard
728
729 Conditional discard. Allowed in fragment shaders only.
730
731 .. math::
732
733 if (src.x < 0 || src.y < 0 || src.z < 0 || src.w < 0)
734 discard
735 endif
736
737
738 .. opcode:: KILL - Discard
739
740 Unconditional discard. Allowed in fragment shaders only.
741
742
743 .. opcode:: SCS - Sine Cosine
744
745 .. math::
746
747 dst.x = \cos{src.x}
748
749 dst.y = \sin{src.x}
750
751 dst.z = 0
752
753 dst.w = 1
754
755
756 .. opcode:: TXB - Texture Lookup With Bias
757
758 for cube map array textures and shadow cube maps, the bias value
759 cannot be passed in src0.w, and TXB2 must be used instead.
760
761 if the target is a shadow texture, the reference value is always
762 in src.z (this prevents shadow 3d and shadow 2d arrays from
763 using this instruction, but this is not needed).
764
765 .. math::
766
767 coord.x = src0.x
768
769 coord.y = src0.y
770
771 coord.z = src0.z
772
773 coord.w = none
774
775 bias = src0.w
776
777 unit = src1
778
779 dst = texture\_sample(unit, coord, bias)
780
781
782 .. opcode:: TXB2 - Texture Lookup With Bias (some cube maps only)
783
784 this is the same as TXB, but uses another reg to encode the
785 lod bias value for cube map arrays and shadow cube maps.
786 Presumably shadow 2d arrays and shadow 3d targets could use
787 this encoding too, but this is not legal.
788
789 shadow cube map arrays are neither possible nor required.
790
791 .. math::
792
793 coord = src0
794
795 bias = src1.x
796
797 unit = src2
798
799 dst = texture\_sample(unit, coord, bias)
800
801
802 .. opcode:: DIV - Divide
803
804 .. math::
805
806 dst.x = \frac{src0.x}{src1.x}
807
808 dst.y = \frac{src0.y}{src1.y}
809
810 dst.z = \frac{src0.z}{src1.z}
811
812 dst.w = \frac{src0.w}{src1.w}
813
814
815 .. opcode:: DP2 - 2-component Dot Product
816
817 This instruction replicates its result.
818
819 .. math::
820
821 dst = src0.x \times src1.x + src0.y \times src1.y
822
823
824 .. opcode:: TXL - Texture Lookup With explicit LOD
825
826 for cube map array textures, the explicit lod value
827 cannot be passed in src0.w, and TXL2 must be used instead.
828
829 if the target is a shadow texture, the reference value is always
830 in src.z (this prevents shadow 3d / 2d array / cube targets from
831 using this instruction, but this is not needed).
832
833 .. math::
834
835 coord.x = src0.x
836
837 coord.y = src0.y
838
839 coord.z = src0.z
840
841 coord.w = none
842
843 lod = src0.w
844
845 unit = src1
846
847 dst = texture\_sample(unit, coord, lod)
848
849
850 .. opcode:: TXL2 - Texture Lookup With explicit LOD (for cube map arrays only)
851
852 this is the same as TXL, but uses another reg to encode the
853 explicit lod value.
854 Presumably shadow 3d / 2d array / cube targets could use
855 this encoding too, but this is not legal.
856
857 shadow cube map arrays are neither possible nor required.
858
859 .. math::
860
861 coord = src0
862
863 lod = src1.x
864
865 unit = src2
866
867 dst = texture\_sample(unit, coord, lod)
868
869
870 .. opcode:: PUSHA - Push Address Register On Stack
871
872 push(src.x)
873 push(src.y)
874 push(src.z)
875 push(src.w)
876
877 .. note::
878
879 Considered for cleanup.
880
881 .. note::
882
883 Considered for removal.
884
885 .. opcode:: POPA - Pop Address Register From Stack
886
887 dst.w = pop()
888 dst.z = pop()
889 dst.y = pop()
890 dst.x = pop()
891
892 .. note::
893
894 Considered for cleanup.
895
896 .. note::
897
898 Considered for removal.
899
900
901 .. opcode:: BRA - Branch
902
903 pc = target
904
905 .. note::
906
907 Considered for removal.
908
909
910 .. opcode:: CALLNZ - Subroutine Call If Not Zero
911
912 TBD
913
914 .. note::
915
916 Considered for cleanup.
917
918 .. note::
919
920 Considered for removal.
921
922
923 Compute ISA
924 ^^^^^^^^^^^^^^^^^^^^^^^^
925
926 These opcodes are primarily provided for special-use computational shaders.
927 Support for these opcodes indicated by a special pipe capability bit (TBD).
928
929 XXX doesn't look like most of the opcodes really belong here.
930
931 .. opcode:: CEIL - Ceiling
932
933 .. math::
934
935 dst.x = \lceil src.x\rceil
936
937 dst.y = \lceil src.y\rceil
938
939 dst.z = \lceil src.z\rceil
940
941 dst.w = \lceil src.w\rceil
942
943
944 .. opcode:: TRUNC - Truncate
945
946 .. math::
947
948 dst.x = trunc(src.x)
949
950 dst.y = trunc(src.y)
951
952 dst.z = trunc(src.z)
953
954 dst.w = trunc(src.w)
955
956
957 .. opcode:: MOD - Modulus
958
959 .. math::
960
961 dst.x = src0.x \bmod src1.x
962
963 dst.y = src0.y \bmod src1.y
964
965 dst.z = src0.z \bmod src1.z
966
967 dst.w = src0.w \bmod src1.w
968
969
970 .. opcode:: UARL - Integer Address Register Load
971
972 Moves the contents of the source register, assumed to be an integer, into the
973 destination register, which is assumed to be an address (ADDR) register.
974
975
976 .. opcode:: SAD - Sum Of Absolute Differences
977
978 .. math::
979
980 dst.x = |src0.x - src1.x| + src2.x
981
982 dst.y = |src0.y - src1.y| + src2.y
983
984 dst.z = |src0.z - src1.z| + src2.z
985
986 dst.w = |src0.w - src1.w| + src2.w
987
988
989 .. opcode:: TXF - Texel Fetch
990
991 As per NV_gpu_shader4, extract a single texel from a specified texture
992 image. The source sampler may not be a CUBE or SHADOW. src 0 is a
993 four-component signed integer vector used to identify the single texel
994 accessed. 3 components + level. Just like texture instructions, an optional
995 offset vector is provided, which is subject to various driver restrictions
996 (regarding range, source of offsets).
997 TXF(uint_vec coord, int_vec offset).
998
999
1000 .. opcode:: TXQ - Texture Size Query
1001
1002 As per NV_gpu_program4, retrieve the dimensions of the texture depending on
1003 the target. For 1D (width), 2D/RECT/CUBE (width, height), 3D (width, height,
1004 depth), 1D array (width, layers), 2D array (width, height, layers).
1005 Also return the number of accessible levels (last_level - first_level + 1)
1006 in W.
1007
1008 For components which don't return a resource dimension, their value
1009 is undefined.
1010
1011
1012 .. math::
1013
1014 lod = src0.x
1015
1016 dst.x = texture\_width(unit, lod)
1017
1018 dst.y = texture\_height(unit, lod)
1019
1020 dst.z = texture\_depth(unit, lod)
1021
1022 dst.w = texture\_levels(unit)
1023
1024 .. opcode:: TG4 - Texture Gather
1025
1026 As per ARB_texture_gather, gathers the four texels to be used in a bi-linear
1027 filtering operation and packs them into a single register. Only works with
1028 2D, 2D array, cubemaps, and cubemaps arrays. For 2D textures, only the
1029 addressing modes of the sampler and the top level of any mip pyramid are
1030 used. Set W to zero. It behaves like the TEX instruction, but a filtered
1031 sample is not generated. The four samples that contribute to filtering are
1032 placed into xyzw in clockwise order, starting with the (u,v) texture
1033 coordinate delta at the following locations (-, +), (+, +), (+, -), (-, -),
1034 where the magnitude of the deltas are half a texel.
1035
1036 PIPE_CAP_TEXTURE_SM5 enhances this instruction to support shadow per-sample
1037 depth compares, single component selection, and a non-constant offset. It
1038 doesn't allow support for the GL independent offset to get i0,j0. This would
1039 require another CAP is hw can do it natively. For now we lower that before
1040 TGSI.
1041
1042 .. math::
1043
1044 coord = src0
1045
1046 component = src1
1047
1048 dst = texture\_gather4 (unit, coord, component)
1049
1050 (with SM5 - cube array shadow)
1051
1052 .. math::
1053
1054 coord = src0
1055
1056 compare = src1
1057
1058 dst = texture\_gather (uint, coord, compare)
1059
1060 .. opcode:: LODQ - level of detail query
1061
1062 Compute the LOD information that the texture pipe would use to access the
1063 texture. The Y component contains the computed LOD lambda_prime. The X
1064 component contains the LOD that will be accessed, based on min/max lod's
1065 and mipmap filters.
1066
1067 .. math::
1068
1069 coord = src0
1070
1071 dst.xy = lodq(uint, coord);
1072
1073 Integer ISA
1074 ^^^^^^^^^^^^^^^^^^^^^^^^
1075 These opcodes are used for integer operations.
1076 Support for these opcodes indicated by PIPE_SHADER_CAP_INTEGERS (all of them?)
1077
1078
1079 .. opcode:: I2F - Signed Integer To Float
1080
1081 Rounding is unspecified (round to nearest even suggested).
1082
1083 .. math::
1084
1085 dst.x = (float) src.x
1086
1087 dst.y = (float) src.y
1088
1089 dst.z = (float) src.z
1090
1091 dst.w = (float) src.w
1092
1093
1094 .. opcode:: U2F - Unsigned Integer To Float
1095
1096 Rounding is unspecified (round to nearest even suggested).
1097
1098 .. math::
1099
1100 dst.x = (float) src.x
1101
1102 dst.y = (float) src.y
1103
1104 dst.z = (float) src.z
1105
1106 dst.w = (float) src.w
1107
1108
1109 .. opcode:: F2I - Float to Signed Integer
1110
1111 Rounding is towards zero (truncate).
1112 Values outside signed range (including NaNs) produce undefined results.
1113
1114 .. math::
1115
1116 dst.x = (int) src.x
1117
1118 dst.y = (int) src.y
1119
1120 dst.z = (int) src.z
1121
1122 dst.w = (int) src.w
1123
1124
1125 .. opcode:: F2U - Float to Unsigned Integer
1126
1127 Rounding is towards zero (truncate).
1128 Values outside unsigned range (including NaNs) produce undefined results.
1129
1130 .. math::
1131
1132 dst.x = (unsigned) src.x
1133
1134 dst.y = (unsigned) src.y
1135
1136 dst.z = (unsigned) src.z
1137
1138 dst.w = (unsigned) src.w
1139
1140
1141 .. opcode:: UADD - Integer Add
1142
1143 This instruction works the same for signed and unsigned integers.
1144 The low 32bit of the result is returned.
1145
1146 .. math::
1147
1148 dst.x = src0.x + src1.x
1149
1150 dst.y = src0.y + src1.y
1151
1152 dst.z = src0.z + src1.z
1153
1154 dst.w = src0.w + src1.w
1155
1156
1157 .. opcode:: UMAD - Integer Multiply And Add
1158
1159 This instruction works the same for signed and unsigned integers.
1160 The multiplication returns the low 32bit (as does the result itself).
1161
1162 .. math::
1163
1164 dst.x = src0.x \times src1.x + src2.x
1165
1166 dst.y = src0.y \times src1.y + src2.y
1167
1168 dst.z = src0.z \times src1.z + src2.z
1169
1170 dst.w = src0.w \times src1.w + src2.w
1171
1172
1173 .. opcode:: UMUL - Integer Multiply
1174
1175 This instruction works the same for signed and unsigned integers.
1176 The low 32bit of the result is returned.
1177
1178 .. math::
1179
1180 dst.x = src0.x \times src1.x
1181
1182 dst.y = src0.y \times src1.y
1183
1184 dst.z = src0.z \times src1.z
1185
1186 dst.w = src0.w \times src1.w
1187
1188
1189 .. opcode:: IMUL_HI - Signed Integer Multiply High Bits
1190
1191 The high 32bits of the multiplication of 2 signed integers are returned.
1192
1193 .. math::
1194
1195 dst.x = (src0.x \times src1.x) >> 32
1196
1197 dst.y = (src0.y \times src1.y) >> 32
1198
1199 dst.z = (src0.z \times src1.z) >> 32
1200
1201 dst.w = (src0.w \times src1.w) >> 32
1202
1203
1204 .. opcode:: UMUL_HI - Unsigned Integer Multiply High Bits
1205
1206 The high 32bits of the multiplication of 2 unsigned integers are returned.
1207
1208 .. math::
1209
1210 dst.x = (src0.x \times src1.x) >> 32
1211
1212 dst.y = (src0.y \times src1.y) >> 32
1213
1214 dst.z = (src0.z \times src1.z) >> 32
1215
1216 dst.w = (src0.w \times src1.w) >> 32
1217
1218
1219 .. opcode:: IDIV - Signed Integer Division
1220
1221 TBD: behavior for division by zero.
1222
1223 .. math::
1224
1225 dst.x = src0.x \ src1.x
1226
1227 dst.y = src0.y \ src1.y
1228
1229 dst.z = src0.z \ src1.z
1230
1231 dst.w = src0.w \ src1.w
1232
1233
1234 .. opcode:: UDIV - Unsigned Integer Division
1235
1236 For division by zero, 0xffffffff is returned.
1237
1238 .. math::
1239
1240 dst.x = src0.x \ src1.x
1241
1242 dst.y = src0.y \ src1.y
1243
1244 dst.z = src0.z \ src1.z
1245
1246 dst.w = src0.w \ src1.w
1247
1248
1249 .. opcode:: UMOD - Unsigned Integer Remainder
1250
1251 If second arg is zero, 0xffffffff is returned.
1252
1253 .. math::
1254
1255 dst.x = src0.x \ src1.x
1256
1257 dst.y = src0.y \ src1.y
1258
1259 dst.z = src0.z \ src1.z
1260
1261 dst.w = src0.w \ src1.w
1262
1263
1264 .. opcode:: NOT - Bitwise Not
1265
1266 .. math::
1267
1268 dst.x = \sim src.x
1269
1270 dst.y = \sim src.y
1271
1272 dst.z = \sim src.z
1273
1274 dst.w = \sim src.w
1275
1276
1277 .. opcode:: AND - Bitwise And
1278
1279 .. math::
1280
1281 dst.x = src0.x \& src1.x
1282
1283 dst.y = src0.y \& src1.y
1284
1285 dst.z = src0.z \& src1.z
1286
1287 dst.w = src0.w \& src1.w
1288
1289
1290 .. opcode:: OR - Bitwise Or
1291
1292 .. math::
1293
1294 dst.x = src0.x | src1.x
1295
1296 dst.y = src0.y | src1.y
1297
1298 dst.z = src0.z | src1.z
1299
1300 dst.w = src0.w | src1.w
1301
1302
1303 .. opcode:: XOR - Bitwise Xor
1304
1305 .. math::
1306
1307 dst.x = src0.x \oplus src1.x
1308
1309 dst.y = src0.y \oplus src1.y
1310
1311 dst.z = src0.z \oplus src1.z
1312
1313 dst.w = src0.w \oplus src1.w
1314
1315
1316 .. opcode:: IMAX - Maximum of Signed Integers
1317
1318 .. math::
1319
1320 dst.x = max(src0.x, src1.x)
1321
1322 dst.y = max(src0.y, src1.y)
1323
1324 dst.z = max(src0.z, src1.z)
1325
1326 dst.w = max(src0.w, src1.w)
1327
1328
1329 .. opcode:: UMAX - Maximum of Unsigned Integers
1330
1331 .. math::
1332
1333 dst.x = max(src0.x, src1.x)
1334
1335 dst.y = max(src0.y, src1.y)
1336
1337 dst.z = max(src0.z, src1.z)
1338
1339 dst.w = max(src0.w, src1.w)
1340
1341
1342 .. opcode:: IMIN - Minimum of Signed Integers
1343
1344 .. math::
1345
1346 dst.x = min(src0.x, src1.x)
1347
1348 dst.y = min(src0.y, src1.y)
1349
1350 dst.z = min(src0.z, src1.z)
1351
1352 dst.w = min(src0.w, src1.w)
1353
1354
1355 .. opcode:: UMIN - Minimum of Unsigned Integers
1356
1357 .. math::
1358
1359 dst.x = min(src0.x, src1.x)
1360
1361 dst.y = min(src0.y, src1.y)
1362
1363 dst.z = min(src0.z, src1.z)
1364
1365 dst.w = min(src0.w, src1.w)
1366
1367
1368 .. opcode:: SHL - Shift Left
1369
1370 The shift count is masked with 0x1f before the shift is applied.
1371
1372 .. math::
1373
1374 dst.x = src0.x << (0x1f \& src1.x)
1375
1376 dst.y = src0.y << (0x1f \& src1.y)
1377
1378 dst.z = src0.z << (0x1f \& src1.z)
1379
1380 dst.w = src0.w << (0x1f \& src1.w)
1381
1382
1383 .. opcode:: ISHR - Arithmetic Shift Right (of Signed Integer)
1384
1385 The shift count is masked with 0x1f before the shift is applied.
1386
1387 .. math::
1388
1389 dst.x = src0.x >> (0x1f \& src1.x)
1390
1391 dst.y = src0.y >> (0x1f \& src1.y)
1392
1393 dst.z = src0.z >> (0x1f \& src1.z)
1394
1395 dst.w = src0.w >> (0x1f \& src1.w)
1396
1397
1398 .. opcode:: USHR - Logical Shift Right
1399
1400 The shift count is masked with 0x1f before the shift is applied.
1401
1402 .. math::
1403
1404 dst.x = src0.x >> (unsigned) (0x1f \& src1.x)
1405
1406 dst.y = src0.y >> (unsigned) (0x1f \& src1.y)
1407
1408 dst.z = src0.z >> (unsigned) (0x1f \& src1.z)
1409
1410 dst.w = src0.w >> (unsigned) (0x1f \& src1.w)
1411
1412
1413 .. opcode:: UCMP - Integer Conditional Move
1414
1415 .. math::
1416
1417 dst.x = src0.x ? src1.x : src2.x
1418
1419 dst.y = src0.y ? src1.y : src2.y
1420
1421 dst.z = src0.z ? src1.z : src2.z
1422
1423 dst.w = src0.w ? src1.w : src2.w
1424
1425
1426
1427 .. opcode:: ISSG - Integer Set Sign
1428
1429 .. math::
1430
1431 dst.x = (src0.x < 0) ? -1 : (src0.x > 0) ? 1 : 0
1432
1433 dst.y = (src0.y < 0) ? -1 : (src0.y > 0) ? 1 : 0
1434
1435 dst.z = (src0.z < 0) ? -1 : (src0.z > 0) ? 1 : 0
1436
1437 dst.w = (src0.w < 0) ? -1 : (src0.w > 0) ? 1 : 0
1438
1439
1440
1441 .. opcode:: FSLT - Float Set On Less Than (ordered)
1442
1443 Same comparison as SLT but returns integer instead of 1.0/0.0 float
1444
1445 .. math::
1446
1447 dst.x = (src0.x < src1.x) ? \sim 0 : 0
1448
1449 dst.y = (src0.y < src1.y) ? \sim 0 : 0
1450
1451 dst.z = (src0.z < src1.z) ? \sim 0 : 0
1452
1453 dst.w = (src0.w < src1.w) ? \sim 0 : 0
1454
1455
1456 .. opcode:: ISLT - Signed Integer Set On Less Than
1457
1458 .. math::
1459
1460 dst.x = (src0.x < src1.x) ? \sim 0 : 0
1461
1462 dst.y = (src0.y < src1.y) ? \sim 0 : 0
1463
1464 dst.z = (src0.z < src1.z) ? \sim 0 : 0
1465
1466 dst.w = (src0.w < src1.w) ? \sim 0 : 0
1467
1468
1469 .. opcode:: USLT - Unsigned Integer Set On Less Than
1470
1471 .. math::
1472
1473 dst.x = (src0.x < src1.x) ? \sim 0 : 0
1474
1475 dst.y = (src0.y < src1.y) ? \sim 0 : 0
1476
1477 dst.z = (src0.z < src1.z) ? \sim 0 : 0
1478
1479 dst.w = (src0.w < src1.w) ? \sim 0 : 0
1480
1481
1482 .. opcode:: FSGE - Float Set On Greater Equal Than (ordered)
1483
1484 Same comparison as SGE but returns integer instead of 1.0/0.0 float
1485
1486 .. math::
1487
1488 dst.x = (src0.x >= src1.x) ? \sim 0 : 0
1489
1490 dst.y = (src0.y >= src1.y) ? \sim 0 : 0
1491
1492 dst.z = (src0.z >= src1.z) ? \sim 0 : 0
1493
1494 dst.w = (src0.w >= src1.w) ? \sim 0 : 0
1495
1496
1497 .. opcode:: ISGE - Signed Integer Set On Greater Equal Than
1498
1499 .. math::
1500
1501 dst.x = (src0.x >= src1.x) ? \sim 0 : 0
1502
1503 dst.y = (src0.y >= src1.y) ? \sim 0 : 0
1504
1505 dst.z = (src0.z >= src1.z) ? \sim 0 : 0
1506
1507 dst.w = (src0.w >= src1.w) ? \sim 0 : 0
1508
1509
1510 .. opcode:: USGE - Unsigned Integer Set On Greater Equal Than
1511
1512 .. math::
1513
1514 dst.x = (src0.x >= src1.x) ? \sim 0 : 0
1515
1516 dst.y = (src0.y >= src1.y) ? \sim 0 : 0
1517
1518 dst.z = (src0.z >= src1.z) ? \sim 0 : 0
1519
1520 dst.w = (src0.w >= src1.w) ? \sim 0 : 0
1521
1522
1523 .. opcode:: FSEQ - Float Set On Equal (ordered)
1524
1525 Same comparison as SEQ but returns integer instead of 1.0/0.0 float
1526
1527 .. math::
1528
1529 dst.x = (src0.x == src1.x) ? \sim 0 : 0
1530
1531 dst.y = (src0.y == src1.y) ? \sim 0 : 0
1532
1533 dst.z = (src0.z == src1.z) ? \sim 0 : 0
1534
1535 dst.w = (src0.w == src1.w) ? \sim 0 : 0
1536
1537
1538 .. opcode:: USEQ - Integer Set On Equal
1539
1540 .. math::
1541
1542 dst.x = (src0.x == src1.x) ? \sim 0 : 0
1543
1544 dst.y = (src0.y == src1.y) ? \sim 0 : 0
1545
1546 dst.z = (src0.z == src1.z) ? \sim 0 : 0
1547
1548 dst.w = (src0.w == src1.w) ? \sim 0 : 0
1549
1550
1551 .. opcode:: FSNE - Float Set On Not Equal (unordered)
1552
1553 Same comparison as SNE but returns integer instead of 1.0/0.0 float
1554
1555 .. math::
1556
1557 dst.x = (src0.x != src1.x) ? \sim 0 : 0
1558
1559 dst.y = (src0.y != src1.y) ? \sim 0 : 0
1560
1561 dst.z = (src0.z != src1.z) ? \sim 0 : 0
1562
1563 dst.w = (src0.w != src1.w) ? \sim 0 : 0
1564
1565
1566 .. opcode:: USNE - Integer Set On Not Equal
1567
1568 .. math::
1569
1570 dst.x = (src0.x != src1.x) ? \sim 0 : 0
1571
1572 dst.y = (src0.y != src1.y) ? \sim 0 : 0
1573
1574 dst.z = (src0.z != src1.z) ? \sim 0 : 0
1575
1576 dst.w = (src0.w != src1.w) ? \sim 0 : 0
1577
1578
1579 .. opcode:: INEG - Integer Negate
1580
1581 Two's complement.
1582
1583 .. math::
1584
1585 dst.x = -src.x
1586
1587 dst.y = -src.y
1588
1589 dst.z = -src.z
1590
1591 dst.w = -src.w
1592
1593
1594 .. opcode:: IABS - Integer Absolute Value
1595
1596 .. math::
1597
1598 dst.x = |src.x|
1599
1600 dst.y = |src.y|
1601
1602 dst.z = |src.z|
1603
1604 dst.w = |src.w|
1605
1606 Bitwise ISA
1607 ^^^^^^^^^^^
1608 These opcodes are used for bit-level manipulation of integers.
1609
1610 .. opcode:: IBFE - Signed Bitfield Extract
1611
1612 See SM5 instruction of the same name. Extracts a set of bits from the input,
1613 and sign-extends them if the high bit of the extracted window is set.
1614
1615 Pseudocode::
1616
1617 def ibfe(value, offset, bits):
1618 offset = offset & 0x1f
1619 bits = bits & 0x1f
1620 if bits == 0: return 0
1621 # Note: >> sign-extends
1622 if width + offset < 32:
1623 return (value << (32 - offset - bits)) >> (32 - bits)
1624 else:
1625 return value >> offset
1626
1627 .. opcode:: UBFE - Unsigned Bitfield Extract
1628
1629 See SM5 instruction of the same name. Extracts a set of bits from the input,
1630 without any sign-extension.
1631
1632 Pseudocode::
1633
1634 def ubfe(value, offset, bits):
1635 offset = offset & 0x1f
1636 bits = bits & 0x1f
1637 if bits == 0: return 0
1638 # Note: >> does not sign-extend
1639 if width + offset < 32:
1640 return (value << (32 - offset - bits)) >> (32 - bits)
1641 else:
1642 return value >> offset
1643
1644 .. opcode:: BFI - Bitfield Insert
1645
1646 See SM5 instruction of the same name. Replaces a bit region of 'base' with
1647 the low bits of 'insert'.
1648
1649 Pseudocode::
1650
1651 def bfi(base, insert, offset, bits):
1652 offset = offset & 0x1f
1653 bits = bits & 0x1f
1654 mask = ((1 << bits) - 1) << offset
1655 return ((insert << offset) & mask) | (base & ~mask)
1656
1657 .. opcode:: BREV - Bitfield Reverse
1658
1659 See SM5 instruction BFREV. Reverses the bits of the argument.
1660
1661 .. opcode:: POPC - Population Count
1662
1663 See SM5 instruction COUNTBITS. Counts the number of set bits in the argument.
1664
1665 .. opcode:: LSB - Index of lowest set bit
1666
1667 See SM5 instruction FIRSTBIT_LO. Computes the 0-based index of the first set
1668 bit of the argument. Returns -1 if none are set.
1669
1670 .. opcode:: IMSB - Index of highest non-sign bit
1671
1672 See SM5 instruction FIRSTBIT_SHI. Computes the 0-based index of the highest
1673 non-sign bit of the argument (i.e. highest 0 bit for negative numbers,
1674 highest 1 bit for positive numbers). Returns -1 if all bits are the same
1675 (i.e. for inputs 0 and -1).
1676
1677 .. opcode:: UMSB - Index of highest set bit
1678
1679 See SM5 instruction FIRSTBIT_HI. Computes the 0-based index of the highest
1680 set bit of the argument. Returns -1 if none are set.
1681
1682 Geometry ISA
1683 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1684
1685 These opcodes are only supported in geometry shaders; they have no meaning
1686 in any other type of shader.
1687
1688 .. opcode:: EMIT - Emit
1689
1690 Generate a new vertex for the current primitive into the specified vertex
1691 stream using the values in the output registers.
1692
1693
1694 .. opcode:: ENDPRIM - End Primitive
1695
1696 Complete the current primitive in the specified vertex stream (consisting of
1697 the emitted vertices), and start a new one.
1698
1699
1700 GLSL ISA
1701 ^^^^^^^^^^
1702
1703 These opcodes are part of :term:`GLSL`'s opcode set. Support for these
1704 opcodes is determined by a special capability bit, ``GLSL``.
1705 Some require glsl version 1.30 (UIF/BREAKC/SWITCH/CASE/DEFAULT/ENDSWITCH).
1706
1707 .. opcode:: CAL - Subroutine Call
1708
1709 push(pc)
1710 pc = target
1711
1712
1713 .. opcode:: RET - Subroutine Call Return
1714
1715 pc = pop()
1716
1717
1718 .. opcode:: CONT - Continue
1719
1720 Unconditionally moves the point of execution to the instruction after the
1721 last bgnloop. The instruction must appear within a bgnloop/endloop.
1722
1723 .. note::
1724
1725 Support for CONT is determined by a special capability bit,
1726 ``TGSI_CONT_SUPPORTED``. See :ref:`Screen` for more information.
1727
1728
1729 .. opcode:: BGNLOOP - Begin a Loop
1730
1731 Start a loop. Must have a matching endloop.
1732
1733
1734 .. opcode:: BGNSUB - Begin Subroutine
1735
1736 Starts definition of a subroutine. Must have a matching endsub.
1737
1738
1739 .. opcode:: ENDLOOP - End a Loop
1740
1741 End a loop started with bgnloop.
1742
1743
1744 .. opcode:: ENDSUB - End Subroutine
1745
1746 Ends definition of a subroutine.
1747
1748
1749 .. opcode:: NOP - No Operation
1750
1751 Do nothing.
1752
1753
1754 .. opcode:: BRK - Break
1755
1756 Unconditionally moves the point of execution to the instruction after the
1757 next endloop or endswitch. The instruction must appear within a loop/endloop
1758 or switch/endswitch.
1759
1760
1761 .. opcode:: BREAKC - Break Conditional
1762
1763 Conditionally moves the point of execution to the instruction after the
1764 next endloop or endswitch. The instruction must appear within a loop/endloop
1765 or switch/endswitch.
1766 Condition evaluates to true if src0.x != 0 where src0.x is interpreted
1767 as an integer register.
1768
1769 .. note::
1770
1771 Considered for removal as it's quite inconsistent wrt other opcodes
1772 (could emulate with UIF/BRK/ENDIF).
1773
1774
1775 .. opcode:: IF - Float If
1776
1777 Start an IF ... ELSE .. ENDIF block. Condition evaluates to true if
1778
1779 src0.x != 0.0
1780
1781 where src0.x is interpreted as a floating point register.
1782
1783
1784 .. opcode:: UIF - Bitwise If
1785
1786 Start an UIF ... ELSE .. ENDIF block. Condition evaluates to true if
1787
1788 src0.x != 0
1789
1790 where src0.x is interpreted as an integer register.
1791
1792
1793 .. opcode:: ELSE - Else
1794
1795 Starts an else block, after an IF or UIF statement.
1796
1797
1798 .. opcode:: ENDIF - End If
1799
1800 Ends an IF or UIF block.
1801
1802
1803 .. opcode:: SWITCH - Switch
1804
1805 Starts a C-style switch expression. The switch consists of one or multiple
1806 CASE statements, and at most one DEFAULT statement. Execution of a statement
1807 ends when a BRK is hit, but just like in C falling through to other cases
1808 without a break is allowed. Similarly, DEFAULT label is allowed anywhere not
1809 just as last statement, and fallthrough is allowed into/from it.
1810 CASE src arguments are evaluated at bit level against the SWITCH src argument.
1811
1812 Example::
1813
1814 SWITCH src[0].x
1815 CASE src[0].x
1816 (some instructions here)
1817 (optional BRK here)
1818 DEFAULT
1819 (some instructions here)
1820 (optional BRK here)
1821 CASE src[0].x
1822 (some instructions here)
1823 (optional BRK here)
1824 ENDSWITCH
1825
1826
1827 .. opcode:: CASE - Switch case
1828
1829 This represents a switch case label. The src arg must be an integer immediate.
1830
1831
1832 .. opcode:: DEFAULT - Switch default
1833
1834 This represents the default case in the switch, which is taken if no other
1835 case matches.
1836
1837
1838 .. opcode:: ENDSWITCH - End of switch
1839
1840 Ends a switch expression.
1841
1842
1843 Interpolation ISA
1844 ^^^^^^^^^^^^^^^^^
1845
1846 The interpolation instructions allow an input to be interpolated in a
1847 different way than its declaration. This corresponds to the GLSL 4.00
1848 interpolateAt* functions. The first argument of each of these must come from
1849 ``TGSI_FILE_INPUT``.
1850
1851 .. opcode:: INTERP_CENTROID - Interpolate at the centroid
1852
1853 Interpolates the varying specified by src0 at the centroid
1854
1855 .. opcode:: INTERP_SAMPLE - Interpolate at the specified sample
1856
1857 Interpolates the varying specified by src0 at the sample id specified by
1858 src1.x (interpreted as an integer)
1859
1860 .. opcode:: INTERP_OFFSET - Interpolate at the specified offset
1861
1862 Interpolates the varying specified by src0 at the offset src1.xy from the
1863 pixel center (interpreted as floats)
1864
1865
1866 .. _doubleopcodes:
1867
1868 Double ISA
1869 ^^^^^^^^^^^^^^^
1870
1871 The double-precision opcodes reinterpret four-component vectors into
1872 two-component vectors with doubled precision in each component.
1873
1874 Support for these opcodes is XXX undecided. :T
1875
1876 .. opcode:: DADD - Add
1877
1878 .. math::
1879
1880 dst.xy = src0.xy + src1.xy
1881
1882 dst.zw = src0.zw + src1.zw
1883
1884
1885 .. opcode:: DDIV - Divide
1886
1887 .. math::
1888
1889 dst.xy = src0.xy / src1.xy
1890
1891 dst.zw = src0.zw / src1.zw
1892
1893 .. opcode:: DSEQ - Set on Equal
1894
1895 .. math::
1896
1897 dst.xy = src0.xy == src1.xy ? 1.0F : 0.0F
1898
1899 dst.zw = src0.zw == src1.zw ? 1.0F : 0.0F
1900
1901 .. opcode:: DSLT - Set on Less than
1902
1903 .. math::
1904
1905 dst.xy = src0.xy < src1.xy ? 1.0F : 0.0F
1906
1907 dst.zw = src0.zw < src1.zw ? 1.0F : 0.0F
1908
1909 .. opcode:: DFRAC - Fraction
1910
1911 .. math::
1912
1913 dst.xy = src.xy - \lfloor src.xy\rfloor
1914
1915 dst.zw = src.zw - \lfloor src.zw\rfloor
1916
1917
1918 .. opcode:: DFRACEXP - Convert Number to Fractional and Integral Components
1919
1920 Like the ``frexp()`` routine in many math libraries, this opcode stores the
1921 exponent of its source to ``dst0``, and the significand to ``dst1``, such that
1922 :math:`dst1 \times 2^{dst0} = src` .
1923
1924 .. math::
1925
1926 dst0.xy = exp(src.xy)
1927
1928 dst1.xy = frac(src.xy)
1929
1930 dst0.zw = exp(src.zw)
1931
1932 dst1.zw = frac(src.zw)
1933
1934 .. opcode:: DLDEXP - Multiply Number by Integral Power of 2
1935
1936 This opcode is the inverse of :opcode:`DFRACEXP`.
1937
1938 .. math::
1939
1940 dst.xy = src0.xy \times 2^{src1.xy}
1941
1942 dst.zw = src0.zw \times 2^{src1.zw}
1943
1944 .. opcode:: DMIN - Minimum
1945
1946 .. math::
1947
1948 dst.xy = min(src0.xy, src1.xy)
1949
1950 dst.zw = min(src0.zw, src1.zw)
1951
1952 .. opcode:: DMAX - Maximum
1953
1954 .. math::
1955
1956 dst.xy = max(src0.xy, src1.xy)
1957
1958 dst.zw = max(src0.zw, src1.zw)
1959
1960 .. opcode:: DMUL - Multiply
1961
1962 .. math::
1963
1964 dst.xy = src0.xy \times src1.xy
1965
1966 dst.zw = src0.zw \times src1.zw
1967
1968
1969 .. opcode:: DMAD - Multiply And Add
1970
1971 .. math::
1972
1973 dst.xy = src0.xy \times src1.xy + src2.xy
1974
1975 dst.zw = src0.zw \times src1.zw + src2.zw
1976
1977
1978 .. opcode:: DRCP - Reciprocal
1979
1980 .. math::
1981
1982 dst.xy = \frac{1}{src.xy}
1983
1984 dst.zw = \frac{1}{src.zw}
1985
1986 .. opcode:: DSQRT - Square Root
1987
1988 .. math::
1989
1990 dst.xy = \sqrt{src.xy}
1991
1992 dst.zw = \sqrt{src.zw}
1993
1994
1995 .. _samplingopcodes:
1996
1997 Resource Sampling Opcodes
1998 ^^^^^^^^^^^^^^^^^^^^^^^^^
1999
2000 Those opcodes follow very closely semantics of the respective Direct3D
2001 instructions. If in doubt double check Direct3D documentation.
2002 Note that the swizzle on SVIEW (src1) determines texel swizzling
2003 after lookup.
2004
2005 .. opcode:: SAMPLE
2006
2007 Using provided address, sample data from the specified texture using the
2008 filtering mode identified by the gven sampler. The source data may come from
2009 any resource type other than buffers.
2010
2011 Syntax: ``SAMPLE dst, address, sampler_view, sampler``
2012
2013 Example: ``SAMPLE TEMP[0], TEMP[1], SVIEW[0], SAMP[0]``
2014
2015 .. opcode:: SAMPLE_I
2016
2017 Simplified alternative to the SAMPLE instruction. Using the provided
2018 integer address, SAMPLE_I fetches data from the specified sampler view
2019 without any filtering. The source data may come from any resource type
2020 other than CUBE.
2021
2022 Syntax: ``SAMPLE_I dst, address, sampler_view``
2023
2024 Example: ``SAMPLE_I TEMP[0], TEMP[1], SVIEW[0]``
2025
2026 The 'address' is specified as unsigned integers. If the 'address' is out of
2027 range [0...(# texels - 1)] the result of the fetch is always 0 in all
2028 components. As such the instruction doesn't honor address wrap modes, in
2029 cases where that behavior is desirable 'SAMPLE' instruction should be used.
2030 address.w always provides an unsigned integer mipmap level. If the value is
2031 out of the range then the instruction always returns 0 in all components.
2032 address.yz are ignored for buffers and 1d textures. address.z is ignored
2033 for 1d texture arrays and 2d textures.
2034
2035 For 1D texture arrays address.y provides the array index (also as unsigned
2036 integer). If the value is out of the range of available array indices
2037 [0... (array size - 1)] then the opcode always returns 0 in all components.
2038 For 2D texture arrays address.z provides the array index, otherwise it
2039 exhibits the same behavior as in the case for 1D texture arrays. The exact
2040 semantics of the source address are presented in the table below:
2041
2042 +---------------------------+----+-----+-----+---------+
2043 | resource type | X | Y | Z | W |
2044 +===========================+====+=====+=====+=========+
2045 | ``PIPE_BUFFER`` | x | | | ignored |
2046 +---------------------------+----+-----+-----+---------+
2047 | ``PIPE_TEXTURE_1D`` | x | | | mpl |
2048 +---------------------------+----+-----+-----+---------+
2049 | ``PIPE_TEXTURE_2D`` | x | y | | mpl |
2050 +---------------------------+----+-----+-----+---------+
2051 | ``PIPE_TEXTURE_3D`` | x | y | z | mpl |
2052 +---------------------------+----+-----+-----+---------+
2053 | ``PIPE_TEXTURE_RECT`` | x | y | | mpl |
2054 +---------------------------+----+-----+-----+---------+
2055 | ``PIPE_TEXTURE_CUBE`` | not allowed as source |
2056 +---------------------------+----+-----+-----+---------+
2057 | ``PIPE_TEXTURE_1D_ARRAY`` | x | idx | | mpl |
2058 +---------------------------+----+-----+-----+---------+
2059 | ``PIPE_TEXTURE_2D_ARRAY`` | x | y | idx | mpl |
2060 +---------------------------+----+-----+-----+---------+
2061
2062 Where 'mpl' is a mipmap level and 'idx' is the array index.
2063
2064 .. opcode:: SAMPLE_I_MS
2065
2066 Just like SAMPLE_I but allows fetch data from multi-sampled surfaces.
2067
2068 Syntax: ``SAMPLE_I_MS dst, address, sampler_view, sample``
2069
2070 .. opcode:: SAMPLE_B
2071
2072 Just like the SAMPLE instruction with the exception that an additional bias
2073 is applied to the level of detail computed as part of the instruction
2074 execution.
2075
2076 Syntax: ``SAMPLE_B dst, address, sampler_view, sampler, lod_bias``
2077
2078 Example: ``SAMPLE_B TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x``
2079
2080 .. opcode:: SAMPLE_C
2081
2082 Similar to the SAMPLE instruction but it performs a comparison filter. The
2083 operands to SAMPLE_C are identical to SAMPLE, except that there is an
2084 additional float32 operand, reference value, which must be a register with
2085 single-component, or a scalar literal. SAMPLE_C makes the hardware use the
2086 current samplers compare_func (in pipe_sampler_state) to compare reference
2087 value against the red component value for the surce resource at each texel
2088 that the currently configured texture filter covers based on the provided
2089 coordinates.
2090
2091 Syntax: ``SAMPLE_C dst, address, sampler_view.r, sampler, ref_value``
2092
2093 Example: ``SAMPLE_C TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x``
2094
2095 .. opcode:: SAMPLE_C_LZ
2096
2097 Same as SAMPLE_C, but LOD is 0 and derivatives are ignored. The LZ stands
2098 for level-zero.
2099
2100 Syntax: ``SAMPLE_C_LZ dst, address, sampler_view.r, sampler, ref_value``
2101
2102 Example: ``SAMPLE_C_LZ TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x``
2103
2104
2105 .. opcode:: SAMPLE_D
2106
2107 SAMPLE_D is identical to the SAMPLE opcode except that the derivatives for
2108 the source address in the x direction and the y direction are provided by
2109 extra parameters.
2110
2111 Syntax: ``SAMPLE_D dst, address, sampler_view, sampler, der_x, der_y``
2112
2113 Example: ``SAMPLE_D TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2], TEMP[3]``
2114
2115 .. opcode:: SAMPLE_L
2116
2117 SAMPLE_L is identical to the SAMPLE opcode except that the LOD is provided
2118 directly as a scalar value, representing no anisotropy.
2119
2120 Syntax: ``SAMPLE_L dst, address, sampler_view, sampler, explicit_lod``
2121
2122 Example: ``SAMPLE_L TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x``
2123
2124 .. opcode:: GATHER4
2125
2126 Gathers the four texels to be used in a bi-linear filtering operation and
2127 packs them into a single register. Only works with 2D, 2D array, cubemaps,
2128 and cubemaps arrays. For 2D textures, only the addressing modes of the
2129 sampler and the top level of any mip pyramid are used. Set W to zero. It
2130 behaves like the SAMPLE instruction, but a filtered sample is not
2131 generated. The four samples that contribute to filtering are placed into
2132 xyzw in counter-clockwise order, starting with the (u,v) texture coordinate
2133 delta at the following locations (-, +), (+, +), (+, -), (-, -), where the
2134 magnitude of the deltas are half a texel.
2135
2136
2137 .. opcode:: SVIEWINFO
2138
2139 Query the dimensions of a given sampler view. dst receives width, height,
2140 depth or array size and number of mipmap levels as int4. The dst can have a
2141 writemask which will specify what info is the caller interested in.
2142
2143 Syntax: ``SVIEWINFO dst, src_mip_level, sampler_view``
2144
2145 Example: ``SVIEWINFO TEMP[0], TEMP[1].x, SVIEW[0]``
2146
2147 src_mip_level is an unsigned integer scalar. If it's out of range then
2148 returns 0 for width, height and depth/array size but the total number of
2149 mipmap is still returned correctly for the given sampler view. The returned
2150 width, height and depth values are for the mipmap level selected by the
2151 src_mip_level and are in the number of texels. For 1d texture array width
2152 is in dst.x, array size is in dst.y and dst.z is 0. The number of mipmaps is
2153 still in dst.w. In contrast to d3d10 resinfo, there's no way in the tgsi
2154 instruction encoding to specify the return type (float/rcpfloat/uint), hence
2155 always using uint. Also, unlike the SAMPLE instructions, the swizzle on src1
2156 resinfo allowing swizzling dst values is ignored (due to the interaction
2157 with rcpfloat modifier which requires some swizzle handling in the state
2158 tracker anyway).
2159
2160 .. opcode:: SAMPLE_POS
2161
2162 Query the position of a given sample. dst receives float4 (x, y, 0, 0)
2163 indicated where the sample is located. If the resource is not a multi-sample
2164 resource and not a render target, the result is 0.
2165
2166 .. opcode:: SAMPLE_INFO
2167
2168 dst receives number of samples in x. If the resource is not a multi-sample
2169 resource and not a render target, the result is 0.
2170
2171
2172 .. _resourceopcodes:
2173
2174 Resource Access Opcodes
2175 ^^^^^^^^^^^^^^^^^^^^^^^
2176
2177 .. opcode:: LOAD - Fetch data from a shader resource
2178
2179 Syntax: ``LOAD dst, resource, address``
2180
2181 Example: ``LOAD TEMP[0], RES[0], TEMP[1]``
2182
2183 Using the provided integer address, LOAD fetches data
2184 from the specified buffer or texture without any
2185 filtering.
2186
2187 The 'address' is specified as a vector of unsigned
2188 integers. If the 'address' is out of range the result
2189 is unspecified.
2190
2191 Only the first mipmap level of a resource can be read
2192 from using this instruction.
2193
2194 For 1D or 2D texture arrays, the array index is
2195 provided as an unsigned integer in address.y or
2196 address.z, respectively. address.yz are ignored for
2197 buffers and 1D textures. address.z is ignored for 1D
2198 texture arrays and 2D textures. address.w is always
2199 ignored.
2200
2201 .. opcode:: STORE - Write data to a shader resource
2202
2203 Syntax: ``STORE resource, address, src``
2204
2205 Example: ``STORE RES[0], TEMP[0], TEMP[1]``
2206
2207 Using the provided integer address, STORE writes data
2208 to the specified buffer or texture.
2209
2210 The 'address' is specified as a vector of unsigned
2211 integers. If the 'address' is out of range the result
2212 is unspecified.
2213
2214 Only the first mipmap level of a resource can be
2215 written to using this instruction.
2216
2217 For 1D or 2D texture arrays, the array index is
2218 provided as an unsigned integer in address.y or
2219 address.z, respectively. address.yz are ignored for
2220 buffers and 1D textures. address.z is ignored for 1D
2221 texture arrays and 2D textures. address.w is always
2222 ignored.
2223
2224
2225 .. _threadsyncopcodes:
2226
2227 Inter-thread synchronization opcodes
2228 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2229
2230 These opcodes are intended for communication between threads running
2231 within the same compute grid. For now they're only valid in compute
2232 programs.
2233
2234 .. opcode:: MFENCE - Memory fence
2235
2236 Syntax: ``MFENCE resource``
2237
2238 Example: ``MFENCE RES[0]``
2239
2240 This opcode forces strong ordering between any memory access
2241 operations that affect the specified resource. This means that
2242 previous loads and stores (and only those) will be performed and
2243 visible to other threads before the program execution continues.
2244
2245
2246 .. opcode:: LFENCE - Load memory fence
2247
2248 Syntax: ``LFENCE resource``
2249
2250 Example: ``LFENCE RES[0]``
2251
2252 Similar to MFENCE, but it only affects the ordering of memory loads.
2253
2254
2255 .. opcode:: SFENCE - Store memory fence
2256
2257 Syntax: ``SFENCE resource``
2258
2259 Example: ``SFENCE RES[0]``
2260
2261 Similar to MFENCE, but it only affects the ordering of memory stores.
2262
2263
2264 .. opcode:: BARRIER - Thread group barrier
2265
2266 ``BARRIER``
2267
2268 This opcode suspends the execution of the current thread until all
2269 the remaining threads in the working group reach the same point of
2270 the program. Results are unspecified if any of the remaining
2271 threads terminates or never reaches an executed BARRIER instruction.
2272
2273
2274 .. _atomopcodes:
2275
2276 Atomic opcodes
2277 ^^^^^^^^^^^^^^
2278
2279 These opcodes provide atomic variants of some common arithmetic and
2280 logical operations. In this context atomicity means that another
2281 concurrent memory access operation that affects the same memory
2282 location is guaranteed to be performed strictly before or after the
2283 entire execution of the atomic operation.
2284
2285 For the moment they're only valid in compute programs.
2286
2287 .. opcode:: ATOMUADD - Atomic integer addition
2288
2289 Syntax: ``ATOMUADD dst, resource, offset, src``
2290
2291 Example: ``ATOMUADD TEMP[0], RES[0], TEMP[1], TEMP[2]``
2292
2293 The following operation is performed atomically on each component:
2294
2295 .. math::
2296
2297 dst_i = resource[offset]_i
2298
2299 resource[offset]_i = dst_i + src_i
2300
2301
2302 .. opcode:: ATOMXCHG - Atomic exchange
2303
2304 Syntax: ``ATOMXCHG dst, resource, offset, src``
2305
2306 Example: ``ATOMXCHG TEMP[0], RES[0], TEMP[1], TEMP[2]``
2307
2308 The following operation is performed atomically on each component:
2309
2310 .. math::
2311
2312 dst_i = resource[offset]_i
2313
2314 resource[offset]_i = src_i
2315
2316
2317 .. opcode:: ATOMCAS - Atomic compare-and-exchange
2318
2319 Syntax: ``ATOMCAS dst, resource, offset, cmp, src``
2320
2321 Example: ``ATOMCAS TEMP[0], RES[0], TEMP[1], TEMP[2], TEMP[3]``
2322
2323 The following operation is performed atomically on each component:
2324
2325 .. math::
2326
2327 dst_i = resource[offset]_i
2328
2329 resource[offset]_i = (dst_i == cmp_i ? src_i : dst_i)
2330
2331
2332 .. opcode:: ATOMAND - Atomic bitwise And
2333
2334 Syntax: ``ATOMAND dst, resource, offset, src``
2335
2336 Example: ``ATOMAND TEMP[0], RES[0], TEMP[1], TEMP[2]``
2337
2338 The following operation is performed atomically on each component:
2339
2340 .. math::
2341
2342 dst_i = resource[offset]_i
2343
2344 resource[offset]_i = dst_i \& src_i
2345
2346
2347 .. opcode:: ATOMOR - Atomic bitwise Or
2348
2349 Syntax: ``ATOMOR dst, resource, offset, src``
2350
2351 Example: ``ATOMOR TEMP[0], RES[0], TEMP[1], TEMP[2]``
2352
2353 The following operation is performed atomically on each component:
2354
2355 .. math::
2356
2357 dst_i = resource[offset]_i
2358
2359 resource[offset]_i = dst_i | src_i
2360
2361
2362 .. opcode:: ATOMXOR - Atomic bitwise Xor
2363
2364 Syntax: ``ATOMXOR dst, resource, offset, src``
2365
2366 Example: ``ATOMXOR TEMP[0], RES[0], TEMP[1], TEMP[2]``
2367
2368 The following operation is performed atomically on each component:
2369
2370 .. math::
2371
2372 dst_i = resource[offset]_i
2373
2374 resource[offset]_i = dst_i \oplus src_i
2375
2376
2377 .. opcode:: ATOMUMIN - Atomic unsigned minimum
2378
2379 Syntax: ``ATOMUMIN dst, resource, offset, src``
2380
2381 Example: ``ATOMUMIN TEMP[0], RES[0], TEMP[1], TEMP[2]``
2382
2383 The following operation is performed atomically on each component:
2384
2385 .. math::
2386
2387 dst_i = resource[offset]_i
2388
2389 resource[offset]_i = (dst_i < src_i ? dst_i : src_i)
2390
2391
2392 .. opcode:: ATOMUMAX - Atomic unsigned maximum
2393
2394 Syntax: ``ATOMUMAX dst, resource, offset, src``
2395
2396 Example: ``ATOMUMAX TEMP[0], RES[0], TEMP[1], TEMP[2]``
2397
2398 The following operation is performed atomically on each component:
2399
2400 .. math::
2401
2402 dst_i = resource[offset]_i
2403
2404 resource[offset]_i = (dst_i > src_i ? dst_i : src_i)
2405
2406
2407 .. opcode:: ATOMIMIN - Atomic signed minimum
2408
2409 Syntax: ``ATOMIMIN dst, resource, offset, src``
2410
2411 Example: ``ATOMIMIN TEMP[0], RES[0], TEMP[1], TEMP[2]``
2412
2413 The following operation is performed atomically on each component:
2414
2415 .. math::
2416
2417 dst_i = resource[offset]_i
2418
2419 resource[offset]_i = (dst_i < src_i ? dst_i : src_i)
2420
2421
2422 .. opcode:: ATOMIMAX - Atomic signed maximum
2423
2424 Syntax: ``ATOMIMAX dst, resource, offset, src``
2425
2426 Example: ``ATOMIMAX TEMP[0], RES[0], TEMP[1], TEMP[2]``
2427
2428 The following operation is performed atomically on each component:
2429
2430 .. math::
2431
2432 dst_i = resource[offset]_i
2433
2434 resource[offset]_i = (dst_i > src_i ? dst_i : src_i)
2435
2436
2437
2438 Explanation of symbols used
2439 ------------------------------
2440
2441
2442 Functions
2443 ^^^^^^^^^^^^^^
2444
2445
2446 :math:`|x|` Absolute value of `x`.
2447
2448 :math:`\lceil x \rceil` Ceiling of `x`.
2449
2450 clamp(x,y,z) Clamp x between y and z.
2451 (x < y) ? y : (x > z) ? z : x
2452
2453 :math:`\lfloor x\rfloor` Floor of `x`.
2454
2455 :math:`\log_2{x}` Logarithm of `x`, base 2.
2456
2457 max(x,y) Maximum of x and y.
2458 (x > y) ? x : y
2459
2460 min(x,y) Minimum of x and y.
2461 (x < y) ? x : y
2462
2463 partialx(x) Derivative of x relative to fragment's X.
2464
2465 partialy(x) Derivative of x relative to fragment's Y.
2466
2467 pop() Pop from stack.
2468
2469 :math:`x^y` `x` to the power `y`.
2470
2471 push(x) Push x on stack.
2472
2473 round(x) Round x.
2474
2475 trunc(x) Truncate x, i.e. drop the fraction bits.
2476
2477
2478 Keywords
2479 ^^^^^^^^^^^^^
2480
2481
2482 discard Discard fragment.
2483
2484 pc Program counter.
2485
2486 target Label of target instruction.
2487
2488
2489 Other tokens
2490 ---------------
2491
2492
2493 Declaration
2494 ^^^^^^^^^^^
2495
2496
2497 Declares a register that is will be referenced as an operand in Instruction
2498 tokens.
2499
2500 File field contains register file that is being declared and is one
2501 of TGSI_FILE.
2502
2503 UsageMask field specifies which of the register components can be accessed
2504 and is one of TGSI_WRITEMASK.
2505
2506 The Local flag specifies that a given value isn't intended for
2507 subroutine parameter passing and, as a result, the implementation
2508 isn't required to give any guarantees of it being preserved across
2509 subroutine boundaries. As it's merely a compiler hint, the
2510 implementation is free to ignore it.
2511
2512 If Dimension flag is set to 1, a Declaration Dimension token follows.
2513
2514 If Semantic flag is set to 1, a Declaration Semantic token follows.
2515
2516 If Interpolate flag is set to 1, a Declaration Interpolate token follows.
2517
2518 If file is TGSI_FILE_RESOURCE, a Declaration Resource token follows.
2519
2520 If Array flag is set to 1, a Declaration Array token follows.
2521
2522 Array Declaration
2523 ^^^^^^^^^^^^^^^^^^^^^^^^
2524
2525 Declarations can optional have an ArrayID attribute which can be referred by
2526 indirect addressing operands. An ArrayID of zero is reserved and treaded as
2527 if no ArrayID is specified.
2528
2529 If an indirect addressing operand refers to a specific declaration by using
2530 an ArrayID only the registers in this declaration are guaranteed to be
2531 accessed, accessing any register outside this declaration results in undefined
2532 behavior. Note that for compatibility the effective index is zero-based and
2533 not relative to the specified declaration
2534
2535 If no ArrayID is specified with an indirect addressing operand the whole
2536 register file might be accessed by this operand. This is strongly discouraged
2537 and will prevent packing of scalar/vec2 arrays and effective alias analysis.
2538
2539 Declaration Semantic
2540 ^^^^^^^^^^^^^^^^^^^^^^^^
2541
2542 Vertex and fragment shader input and output registers may be labeled
2543 with semantic information consisting of a name and index.
2544
2545 Follows Declaration token if Semantic bit is set.
2546
2547 Since its purpose is to link a shader with other stages of the pipeline,
2548 it is valid to follow only those Declaration tokens that declare a register
2549 either in INPUT or OUTPUT file.
2550
2551 SemanticName field contains the semantic name of the register being declared.
2552 There is no default value.
2553
2554 SemanticIndex is an optional subscript that can be used to distinguish
2555 different register declarations with the same semantic name. The default value
2556 is 0.
2557
2558 The meanings of the individual semantic names are explained in the following
2559 sections.
2560
2561 TGSI_SEMANTIC_POSITION
2562 """"""""""""""""""""""
2563
2564 For vertex shaders, TGSI_SEMANTIC_POSITION indicates the vertex shader
2565 output register which contains the homogeneous vertex position in the clip
2566 space coordinate system. After clipping, the X, Y and Z components of the
2567 vertex will be divided by the W value to get normalized device coordinates.
2568
2569 For fragment shaders, TGSI_SEMANTIC_POSITION is used to indicate that
2570 fragment shader input contains the fragment's window position. The X
2571 component starts at zero and always increases from left to right.
2572 The Y component starts at zero and always increases but Y=0 may either
2573 indicate the top of the window or the bottom depending on the fragment
2574 coordinate origin convention (see TGSI_PROPERTY_FS_COORD_ORIGIN).
2575 The Z coordinate ranges from 0 to 1 to represent depth from the front
2576 to the back of the Z buffer. The W component contains the reciprocol
2577 of the interpolated vertex position W component.
2578
2579 Fragment shaders may also declare an output register with
2580 TGSI_SEMANTIC_POSITION. Only the Z component is writable. This allows
2581 the fragment shader to change the fragment's Z position.
2582
2583
2584
2585 TGSI_SEMANTIC_COLOR
2586 """""""""""""""""""
2587
2588 For vertex shader outputs or fragment shader inputs/outputs, this
2589 label indicates that the resister contains an R,G,B,A color.
2590
2591 Several shader inputs/outputs may contain colors so the semantic index
2592 is used to distinguish them. For example, color[0] may be the diffuse
2593 color while color[1] may be the specular color.
2594
2595 This label is needed so that the flat/smooth shading can be applied
2596 to the right interpolants during rasterization.
2597
2598
2599
2600 TGSI_SEMANTIC_BCOLOR
2601 """"""""""""""""""""
2602
2603 Back-facing colors are only used for back-facing polygons, and are only valid
2604 in vertex shader outputs. After rasterization, all polygons are front-facing
2605 and COLOR and BCOLOR end up occupying the same slots in the fragment shader,
2606 so all BCOLORs effectively become regular COLORs in the fragment shader.
2607
2608
2609 TGSI_SEMANTIC_FOG
2610 """""""""""""""""
2611
2612 Vertex shader inputs and outputs and fragment shader inputs may be
2613 labeled with TGSI_SEMANTIC_FOG to indicate that the register contains
2614 a fog coordinate. Typically, the fragment shader will use the fog coordinate
2615 to compute a fog blend factor which is used to blend the normal fragment color
2616 with a constant fog color. But fog coord really is just an ordinary vec4
2617 register like regular semantics.
2618
2619
2620 TGSI_SEMANTIC_PSIZE
2621 """""""""""""""""""
2622
2623 Vertex shader input and output registers may be labeled with
2624 TGIS_SEMANTIC_PSIZE to indicate that the register contains a point size
2625 in the form (S, 0, 0, 1). The point size controls the width or diameter
2626 of points for rasterization. This label cannot be used in fragment
2627 shaders.
2628
2629 When using this semantic, be sure to set the appropriate state in the
2630 :ref:`rasterizer` first.
2631
2632
2633 TGSI_SEMANTIC_TEXCOORD
2634 """"""""""""""""""""""
2635
2636 Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
2637
2638 Vertex shader outputs and fragment shader inputs may be labeled with
2639 this semantic to make them replaceable by sprite coordinates via the
2640 sprite_coord_enable state in the :ref:`rasterizer`.
2641 The semantic index permitted with this semantic is limited to <= 7.
2642
2643 If the driver does not support TEXCOORD, sprite coordinate replacement
2644 applies to inputs with the GENERIC semantic instead.
2645
2646 The intended use case for this semantic is gl_TexCoord.
2647
2648
2649 TGSI_SEMANTIC_PCOORD
2650 """"""""""""""""""""
2651
2652 Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
2653
2654 Fragment shader inputs may be labeled with TGSI_SEMANTIC_PCOORD to indicate
2655 that the register contains sprite coordinates in the form (x, y, 0, 1), if
2656 the current primitive is a point and point sprites are enabled. Otherwise,
2657 the contents of the register are undefined.
2658
2659 The intended use case for this semantic is gl_PointCoord.
2660
2661
2662 TGSI_SEMANTIC_GENERIC
2663 """""""""""""""""""""
2664
2665 All vertex/fragment shader inputs/outputs not labeled with any other
2666 semantic label can be considered to be generic attributes. Typical
2667 uses of generic inputs/outputs are texcoords and user-defined values.
2668
2669
2670 TGSI_SEMANTIC_NORMAL
2671 """"""""""""""""""""
2672
2673 Indicates that a vertex shader input is a normal vector. This is
2674 typically only used for legacy graphics APIs.
2675
2676
2677 TGSI_SEMANTIC_FACE
2678 """"""""""""""""""
2679
2680 This label applies to fragment shader inputs only and indicates that
2681 the register contains front/back-face information of the form (F, 0,
2682 0, 1). The first component will be positive when the fragment belongs
2683 to a front-facing polygon, and negative when the fragment belongs to a
2684 back-facing polygon.
2685
2686
2687 TGSI_SEMANTIC_EDGEFLAG
2688 """"""""""""""""""""""
2689
2690 For vertex shaders, this sematic label indicates that an input or
2691 output is a boolean edge flag. The register layout is [F, x, x, x]
2692 where F is 0.0 or 1.0 and x = don't care. Normally, the vertex shader
2693 simply copies the edge flag input to the edgeflag output.
2694
2695 Edge flags are used to control which lines or points are actually
2696 drawn when the polygon mode converts triangles/quads/polygons into
2697 points or lines.
2698
2699
2700 TGSI_SEMANTIC_STENCIL
2701 """""""""""""""""""""
2702
2703 For fragment shaders, this semantic label indicates that an output
2704 is a writable stencil reference value. Only the Y component is writable.
2705 This allows the fragment shader to change the fragments stencilref value.
2706
2707
2708 TGSI_SEMANTIC_VIEWPORT_INDEX
2709 """"""""""""""""""""""""""""
2710
2711 For geometry shaders, this semantic label indicates that an output
2712 contains the index of the viewport (and scissor) to use.
2713 Only the X value is used.
2714
2715
2716 TGSI_SEMANTIC_LAYER
2717 """""""""""""""""""
2718
2719 For geometry shaders, this semantic label indicates that an output
2720 contains the layer value to use for the color and depth/stencil surfaces.
2721 Only the X value is used. (Also known as rendertarget array index.)
2722
2723
2724 TGSI_SEMANTIC_CULLDIST
2725 """"""""""""""""""""""
2726
2727 Used as distance to plane for performing application-defined culling
2728 of individual primitives against a plane. When components of vertex
2729 elements are given this label, these values are assumed to be a
2730 float32 signed distance to a plane. Primitives will be completely
2731 discarded if the plane distance for all of the vertices in the
2732 primitive are < 0. If a vertex has a cull distance of NaN, that
2733 vertex counts as "out" (as if its < 0);
2734 The limits on both clip and cull distances are bound
2735 by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
2736 the maximum number of components that can be used to hold the
2737 distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
2738 which specifies the maximum number of registers which can be
2739 annotated with those semantics.
2740
2741
2742 TGSI_SEMANTIC_CLIPDIST
2743 """"""""""""""""""""""
2744
2745 When components of vertex elements are identified this way, these
2746 values are each assumed to be a float32 signed distance to a plane.
2747 Primitive setup only invokes rasterization on pixels for which
2748 the interpolated plane distances are >= 0. Multiple clip planes
2749 can be implemented simultaneously, by annotating multiple
2750 components of one or more vertex elements with the above specified
2751 semantic. The limits on both clip and cull distances are bound
2752 by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
2753 the maximum number of components that can be used to hold the
2754 distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
2755 which specifies the maximum number of registers which can be
2756 annotated with those semantics.
2757
2758 TGSI_SEMANTIC_SAMPLEID
2759 """"""""""""""""""""""
2760
2761 For fragment shaders, this semantic label indicates that a system value
2762 contains the current sample id (i.e. gl_SampleID). Only the X value is used.
2763
2764 TGSI_SEMANTIC_SAMPLEPOS
2765 """""""""""""""""""""""
2766
2767 For fragment shaders, this semantic label indicates that a system value
2768 contains the current sample's position (i.e. gl_SamplePosition). Only the X
2769 and Y values are used.
2770
2771 TGSI_SEMANTIC_SAMPLEMASK
2772 """"""""""""""""""""""""
2773
2774 For fragment shaders, this semantic label indicates that an output contains
2775 the sample mask used to disable further sample processing
2776 (i.e. gl_SampleMask). Only the X value is used, up to 32x MS.
2777
2778 TGSI_SEMANTIC_INVOCATIONID
2779 """"""""""""""""""""""""""
2780
2781 For geometry shaders, this semantic label indicates that a system value
2782 contains the current invocation id (i.e. gl_InvocationID). Only the X value is
2783 used.
2784
2785 Declaration Interpolate
2786 ^^^^^^^^^^^^^^^^^^^^^^^
2787
2788 This token is only valid for fragment shader INPUT declarations.
2789
2790 The Interpolate field specifes the way input is being interpolated by
2791 the rasteriser and is one of TGSI_INTERPOLATE_*.
2792
2793 The Location field specifies the location inside the pixel that the
2794 interpolation should be done at, one of ``TGSI_INTERPOLATE_LOC_*``. Note that
2795 when per-sample shading is enabled, the implementation may choose to
2796 interpolate at the sample irrespective of the Location field.
2797
2798 The CylindricalWrap bitfield specifies which register components
2799 should be subject to cylindrical wrapping when interpolating by the
2800 rasteriser. If TGSI_CYLINDRICAL_WRAP_X is set to 1, the X component
2801 should be interpolated according to cylindrical wrapping rules.
2802
2803
2804 Declaration Sampler View
2805 ^^^^^^^^^^^^^^^^^^^^^^^^
2806
2807 Follows Declaration token if file is TGSI_FILE_SAMPLER_VIEW.
2808
2809 DCL SVIEW[#], resource, type(s)
2810
2811 Declares a shader input sampler view and assigns it to a SVIEW[#]
2812 register.
2813
2814 resource can be one of BUFFER, 1D, 2D, 3D, 1DArray and 2DArray.
2815
2816 type must be 1 or 4 entries (if specifying on a per-component
2817 level) out of UNORM, SNORM, SINT, UINT and FLOAT.
2818
2819
2820 Declaration Resource
2821 ^^^^^^^^^^^^^^^^^^^^
2822
2823 Follows Declaration token if file is TGSI_FILE_RESOURCE.
2824
2825 DCL RES[#], resource [, WR] [, RAW]
2826
2827 Declares a shader input resource and assigns it to a RES[#]
2828 register.
2829
2830 resource can be one of BUFFER, 1D, 2D, 3D, CUBE, 1DArray and
2831 2DArray.
2832
2833 If the RAW keyword is not specified, the texture data will be
2834 subject to conversion, swizzling and scaling as required to yield
2835 the specified data type from the physical data format of the bound
2836 resource.
2837
2838 If the RAW keyword is specified, no channel conversion will be
2839 performed: the values read for each of the channels (X,Y,Z,W) will
2840 correspond to consecutive words in the same order and format
2841 they're found in memory. No element-to-address conversion will be
2842 performed either: the value of the provided X coordinate will be
2843 interpreted in byte units instead of texel units. The result of
2844 accessing a misaligned address is undefined.
2845
2846 Usage of the STORE opcode is only allowed if the WR (writable) flag
2847 is set.
2848
2849
2850 Properties
2851 ^^^^^^^^^^^^^^^^^^^^^^^^
2852
2853 Properties are general directives that apply to the whole TGSI program.
2854
2855 FS_COORD_ORIGIN
2856 """""""""""""""
2857
2858 Specifies the fragment shader TGSI_SEMANTIC_POSITION coordinate origin.
2859 The default value is UPPER_LEFT.
2860
2861 If UPPER_LEFT, the position will be (0,0) at the upper left corner and
2862 increase downward and rightward.
2863 If LOWER_LEFT, the position will be (0,0) at the lower left corner and
2864 increase upward and rightward.
2865
2866 OpenGL defaults to LOWER_LEFT, and is configurable with the
2867 GL_ARB_fragment_coord_conventions extension.
2868
2869 DirectX 9/10 use UPPER_LEFT.
2870
2871 FS_COORD_PIXEL_CENTER
2872 """""""""""""""""""""
2873
2874 Specifies the fragment shader TGSI_SEMANTIC_POSITION pixel center convention.
2875 The default value is HALF_INTEGER.
2876
2877 If HALF_INTEGER, the fractionary part of the position will be 0.5
2878 If INTEGER, the fractionary part of the position will be 0.0
2879
2880 Note that this does not affect the set of fragments generated by
2881 rasterization, which is instead controlled by half_pixel_center in the
2882 rasterizer.
2883
2884 OpenGL defaults to HALF_INTEGER, and is configurable with the
2885 GL_ARB_fragment_coord_conventions extension.
2886
2887 DirectX 9 uses INTEGER.
2888 DirectX 10 uses HALF_INTEGER.
2889
2890 FS_COLOR0_WRITES_ALL_CBUFS
2891 """"""""""""""""""""""""""
2892 Specifies that writes to the fragment shader color 0 are replicated to all
2893 bound cbufs. This facilitates OpenGL's fragColor output vs fragData[0] where
2894 fragData is directed to a single color buffer, but fragColor is broadcast.
2895
2896 VS_PROHIBIT_UCPS
2897 """"""""""""""""""""""""""
2898 If this property is set on the program bound to the shader stage before the
2899 fragment shader, user clip planes should have no effect (be disabled) even if
2900 that shader does not write to any clip distance outputs and the rasterizer's
2901 clip_plane_enable is non-zero.
2902 This property is only supported by drivers that also support shader clip
2903 distance outputs.
2904 This is useful for APIs that don't have UCPs and where clip distances written
2905 by a shader cannot be disabled.
2906
2907 GS_INVOCATIONS
2908 """"""""""""""
2909
2910 Specifies the number of times a geometry shader should be executed for each
2911 input primitive. Each invocation will have a different
2912 TGSI_SEMANTIC_INVOCATIONID system value set. If not specified, assumed to
2913 be 1.
2914
2915 VS_WINDOW_SPACE_POSITION
2916 """"""""""""""""""""""""""
2917 If this property is set on the vertex shader, the TGSI_SEMANTIC_POSITION output
2918 is assumed to contain window space coordinates.
2919 Division of X,Y,Z by W and the viewport transformation are disabled, and 1/W is
2920 directly taken from the 4-th component of the shader output.
2921 Naturally, clipping is not performed on window coordinates either.
2922 The effect of this property is undefined if a geometry or tessellation shader
2923 are in use.
2924
2925 Texture Sampling and Texture Formats
2926 ------------------------------------
2927
2928 This table shows how texture image components are returned as (x,y,z,w) tuples
2929 by TGSI texture instructions, such as :opcode:`TEX`, :opcode:`TXD`, and
2930 :opcode:`TXP`. For reference, OpenGL and Direct3D conventions are shown as
2931 well.
2932
2933 +--------------------+--------------+--------------------+--------------+
2934 | Texture Components | Gallium | OpenGL | Direct3D 9 |
2935 +====================+==============+====================+==============+
2936 | R | (r, 0, 0, 1) | (r, 0, 0, 1) | (r, 1, 1, 1) |
2937 +--------------------+--------------+--------------------+--------------+
2938 | RG | (r, g, 0, 1) | (r, g, 0, 1) | (r, g, 1, 1) |
2939 +--------------------+--------------+--------------------+--------------+
2940 | RGB | (r, g, b, 1) | (r, g, b, 1) | (r, g, b, 1) |
2941 +--------------------+--------------+--------------------+--------------+
2942 | RGBA | (r, g, b, a) | (r, g, b, a) | (r, g, b, a) |
2943 +--------------------+--------------+--------------------+--------------+
2944 | A | (0, 0, 0, a) | (0, 0, 0, a) | (0, 0, 0, a) |
2945 +--------------------+--------------+--------------------+--------------+
2946 | L | (l, l, l, 1) | (l, l, l, 1) | (l, l, l, 1) |
2947 +--------------------+--------------+--------------------+--------------+
2948 | LA | (l, l, l, a) | (l, l, l, a) | (l, l, l, a) |
2949 +--------------------+--------------+--------------------+--------------+
2950 | I | (i, i, i, i) | (i, i, i, i) | N/A |
2951 +--------------------+--------------+--------------------+--------------+
2952 | UV | XXX TBD | (0, 0, 0, 1) | (u, v, 1, 1) |
2953 | | | [#envmap-bumpmap]_ | |
2954 +--------------------+--------------+--------------------+--------------+
2955 | Z | XXX TBD | (z, z, z, 1) | (0, z, 0, 1) |
2956 | | | [#depth-tex-mode]_ | |
2957 +--------------------+--------------+--------------------+--------------+
2958 | S | (s, s, s, s) | unknown | unknown |
2959 +--------------------+--------------+--------------------+--------------+
2960
2961 .. [#envmap-bumpmap] http://www.opengl.org/registry/specs/ATI/envmap_bumpmap.txt
2962 .. [#depth-tex-mode] the default is (z, z, z, 1) but may also be (0, 0, 0, z)
2963 or (z, z, z, z) depending on the value of GL_DEPTH_TEXTURE_MODE.