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
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
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`.
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.
29 TGSI supports modifiers on inputs (as well as saturate and precise modifier
32 For arithmetic instruction having a precise modifier certain optimizations
33 which may alter the result are disallowed. Example: *add(mul(a,b),c)* can't be
34 optimized to TGSI_OPCODE_MAD, because some hardware only supports the fused
37 For inputs which have a floating point type, both absolute value and
38 negation modifiers are supported (with absolute value being applied
39 first). The only source of TGSI_OPCODE_MOV and the second and third
40 sources of TGSI_OPCODE_UCMP are considered to have float type for
43 For inputs which have signed or unsigned type only the negate modifier is
50 ^^^^^^^^^^^^^^^^^^^^^^^^^
52 These opcodes are guaranteed to be available regardless of the driver being
55 .. opcode:: ARL - Address Register Load
59 dst.x = (int) \lfloor src.x\rfloor
61 dst.y = (int) \lfloor src.y\rfloor
63 dst.z = (int) \lfloor src.z\rfloor
65 dst.w = (int) \lfloor src.w\rfloor
68 .. opcode:: MOV - Move
81 .. opcode:: LIT - Light Coefficients
86 dst.y &= max(src.x, 0) \\
87 dst.z &= (src.x > 0) ? max(src.y, 0)^{clamp(src.w, -128, 128))} : 0 \\
91 .. opcode:: RCP - Reciprocal
93 This instruction replicates its result.
100 .. opcode:: RSQ - Reciprocal Square Root
102 This instruction replicates its result. The results are undefined for src <= 0.
106 dst = \frac{1}{\sqrt{src.x}}
109 .. opcode:: SQRT - Square Root
111 This instruction replicates its result. The results are undefined for src < 0.
118 .. opcode:: EXP - Approximate Exponential Base 2
122 dst.x &= 2^{\lfloor src.x\rfloor} \\
123 dst.y &= src.x - \lfloor src.x\rfloor \\
124 dst.z &= 2^{src.x} \\
128 .. opcode:: LOG - Approximate Logarithm Base 2
132 dst.x &= \lfloor\log_2{|src.x|}\rfloor \\
133 dst.y &= \frac{|src.x|}{2^{\lfloor\log_2{|src.x|}\rfloor}} \\
134 dst.z &= \log_2{|src.x|} \\
138 .. opcode:: MUL - Multiply
142 dst.x = src0.x \times src1.x
144 dst.y = src0.y \times src1.y
146 dst.z = src0.z \times src1.z
148 dst.w = src0.w \times src1.w
151 .. opcode:: ADD - Add
155 dst.x = src0.x + src1.x
157 dst.y = src0.y + src1.y
159 dst.z = src0.z + src1.z
161 dst.w = src0.w + src1.w
164 .. opcode:: DP3 - 3-component Dot Product
166 This instruction replicates its result.
170 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z
173 .. opcode:: DP4 - 4-component Dot Product
175 This instruction replicates its result.
179 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z + src0.w \times src1.w
182 .. opcode:: DST - Distance Vector
187 dst.y &= src0.y \times src1.y\\
192 .. opcode:: MIN - Minimum
196 dst.x = min(src0.x, src1.x)
198 dst.y = min(src0.y, src1.y)
200 dst.z = min(src0.z, src1.z)
202 dst.w = min(src0.w, src1.w)
205 .. opcode:: MAX - Maximum
209 dst.x = max(src0.x, src1.x)
211 dst.y = max(src0.y, src1.y)
213 dst.z = max(src0.z, src1.z)
215 dst.w = max(src0.w, src1.w)
218 .. opcode:: SLT - Set On Less Than
222 dst.x = (src0.x < src1.x) ? 1.0F : 0.0F
224 dst.y = (src0.y < src1.y) ? 1.0F : 0.0F
226 dst.z = (src0.z < src1.z) ? 1.0F : 0.0F
228 dst.w = (src0.w < src1.w) ? 1.0F : 0.0F
231 .. opcode:: SGE - Set On Greater Equal Than
235 dst.x = (src0.x >= src1.x) ? 1.0F : 0.0F
237 dst.y = (src0.y >= src1.y) ? 1.0F : 0.0F
239 dst.z = (src0.z >= src1.z) ? 1.0F : 0.0F
241 dst.w = (src0.w >= src1.w) ? 1.0F : 0.0F
244 .. opcode:: MAD - Multiply And Add
246 Perform a * b + c. The implementation is free to decide whether there is an
247 intermediate rounding step or not.
251 dst.x = src0.x \times src1.x + src2.x
253 dst.y = src0.y \times src1.y + src2.y
255 dst.z = src0.z \times src1.z + src2.z
257 dst.w = src0.w \times src1.w + src2.w
260 .. opcode:: LRP - Linear Interpolate
264 dst.x = src0.x \times src1.x + (1 - src0.x) \times src2.x
266 dst.y = src0.y \times src1.y + (1 - src0.y) \times src2.y
268 dst.z = src0.z \times src1.z + (1 - src0.z) \times src2.z
270 dst.w = src0.w \times src1.w + (1 - src0.w) \times src2.w
273 .. opcode:: FMA - Fused Multiply-Add
275 Perform a * b + c with no intermediate rounding step.
279 dst.x = src0.x \times src1.x + src2.x
281 dst.y = src0.y \times src1.y + src2.y
283 dst.z = src0.z \times src1.z + src2.z
285 dst.w = src0.w \times src1.w + src2.w
288 .. opcode:: FRC - Fraction
292 dst.x = src.x - \lfloor src.x\rfloor
294 dst.y = src.y - \lfloor src.y\rfloor
296 dst.z = src.z - \lfloor src.z\rfloor
298 dst.w = src.w - \lfloor src.w\rfloor
301 .. opcode:: FLR - Floor
305 dst.x = \lfloor src.x\rfloor
307 dst.y = \lfloor src.y\rfloor
309 dst.z = \lfloor src.z\rfloor
311 dst.w = \lfloor src.w\rfloor
314 .. opcode:: ROUND - Round
327 .. opcode:: EX2 - Exponential Base 2
329 This instruction replicates its result.
336 .. opcode:: LG2 - Logarithm Base 2
338 This instruction replicates its result.
345 .. opcode:: POW - Power
347 This instruction replicates its result.
351 dst = src0.x^{src1.x}
354 .. opcode:: COS - Cosine
356 This instruction replicates its result.
363 .. opcode:: DDX, DDX_FINE - Derivative Relative To X
365 The fine variant is only used when ``PIPE_CAP_TGSI_FS_FINE_DERIVATIVE`` is
366 advertised. When it is, the fine version guarantees one derivative per row
367 while DDX is allowed to be the same for the entire 2x2 quad.
371 dst.x = partialx(src.x)
373 dst.y = partialx(src.y)
375 dst.z = partialx(src.z)
377 dst.w = partialx(src.w)
380 .. opcode:: DDY, DDY_FINE - Derivative Relative To Y
382 The fine variant is only used when ``PIPE_CAP_TGSI_FS_FINE_DERIVATIVE`` is
383 advertised. When it is, the fine version guarantees one derivative per column
384 while DDY is allowed to be the same for the entire 2x2 quad.
388 dst.x = partialy(src.x)
390 dst.y = partialy(src.y)
392 dst.z = partialy(src.z)
394 dst.w = partialy(src.w)
397 .. opcode:: PK2H - Pack Two 16-bit Floats
399 This instruction replicates its result.
403 dst = f32\_to\_f16(src.x) | f32\_to\_f16(src.y) << 16
406 .. opcode:: PK2US - Pack Two Unsigned 16-bit Scalars
411 .. opcode:: PK4B - Pack Four Signed 8-bit Scalars
416 .. opcode:: PK4UB - Pack Four Unsigned 8-bit Scalars
421 .. opcode:: SEQ - Set On Equal
425 dst.x = (src0.x == src1.x) ? 1.0F : 0.0F
427 dst.y = (src0.y == src1.y) ? 1.0F : 0.0F
429 dst.z = (src0.z == src1.z) ? 1.0F : 0.0F
431 dst.w = (src0.w == src1.w) ? 1.0F : 0.0F
434 .. opcode:: SGT - Set On Greater Than
438 dst.x = (src0.x > src1.x) ? 1.0F : 0.0F
440 dst.y = (src0.y > src1.y) ? 1.0F : 0.0F
442 dst.z = (src0.z > src1.z) ? 1.0F : 0.0F
444 dst.w = (src0.w > src1.w) ? 1.0F : 0.0F
447 .. opcode:: SIN - Sine
449 This instruction replicates its result.
456 .. opcode:: SLE - Set On Less Equal Than
460 dst.x = (src0.x <= src1.x) ? 1.0F : 0.0F
462 dst.y = (src0.y <= src1.y) ? 1.0F : 0.0F
464 dst.z = (src0.z <= src1.z) ? 1.0F : 0.0F
466 dst.w = (src0.w <= src1.w) ? 1.0F : 0.0F
469 .. opcode:: SNE - Set On Not Equal
473 dst.x = (src0.x != src1.x) ? 1.0F : 0.0F
475 dst.y = (src0.y != src1.y) ? 1.0F : 0.0F
477 dst.z = (src0.z != src1.z) ? 1.0F : 0.0F
479 dst.w = (src0.w != src1.w) ? 1.0F : 0.0F
482 .. opcode:: TEX - Texture Lookup
484 for array textures src0.y contains the slice for 1D,
485 and src0.z contain the slice for 2D.
487 for shadow textures with no arrays (and not cube map),
488 src0.z contains the reference value.
490 for shadow textures with arrays, src0.z contains
491 the reference value for 1D arrays, and src0.w contains
492 the reference value for 2D arrays and cube maps.
494 for cube map array shadow textures, the reference value
495 cannot be passed in src0.w, and TEX2 must be used instead.
501 shadow_ref = src0.z or src0.w (optional)
505 dst = texture\_sample(unit, coord, shadow_ref)
508 .. opcode:: TEX2 - Texture Lookup (for shadow cube map arrays only)
510 this is the same as TEX, but uses another reg to encode the
521 dst = texture\_sample(unit, coord, shadow_ref)
526 .. opcode:: TXD - Texture Lookup with Derivatives
538 dst = texture\_sample\_deriv(unit, coord, ddx, ddy)
541 .. opcode:: TXP - Projective Texture Lookup
545 coord.x = src0.x / src0.w
547 coord.y = src0.y / src0.w
549 coord.z = src0.z / src0.w
555 dst = texture\_sample(unit, coord)
558 .. opcode:: UP2H - Unpack Two 16-Bit Floats
562 dst.x = f16\_to\_f32(src0.x \& 0xffff)
564 dst.y = f16\_to\_f32(src0.x >> 16)
566 dst.z = f16\_to\_f32(src0.x \& 0xffff)
568 dst.w = f16\_to\_f32(src0.x >> 16)
572 Considered for removal.
574 .. opcode:: UP2US - Unpack Two Unsigned 16-Bit Scalars
580 Considered for removal.
582 .. opcode:: UP4B - Unpack Four Signed 8-Bit Values
588 Considered for removal.
590 .. opcode:: UP4UB - Unpack Four Unsigned 8-Bit Scalars
596 Considered for removal.
599 .. opcode:: ARR - Address Register Load With Round
603 dst.x = (int) round(src.x)
605 dst.y = (int) round(src.y)
607 dst.z = (int) round(src.z)
609 dst.w = (int) round(src.w)
612 .. opcode:: SSG - Set Sign
616 dst.x = (src.x > 0) ? 1 : (src.x < 0) ? -1 : 0
618 dst.y = (src.y > 0) ? 1 : (src.y < 0) ? -1 : 0
620 dst.z = (src.z > 0) ? 1 : (src.z < 0) ? -1 : 0
622 dst.w = (src.w > 0) ? 1 : (src.w < 0) ? -1 : 0
625 .. opcode:: CMP - Compare
629 dst.x = (src0.x < 0) ? src1.x : src2.x
631 dst.y = (src0.y < 0) ? src1.y : src2.y
633 dst.z = (src0.z < 0) ? src1.z : src2.z
635 dst.w = (src0.w < 0) ? src1.w : src2.w
638 .. opcode:: KILL_IF - Conditional Discard
640 Conditional discard. Allowed in fragment shaders only.
644 if (src.x < 0 || src.y < 0 || src.z < 0 || src.w < 0)
649 .. opcode:: KILL - Discard
651 Unconditional discard. Allowed in fragment shaders only.
654 .. opcode:: TXB - Texture Lookup With Bias
656 for cube map array textures and shadow cube maps, the bias value
657 cannot be passed in src0.w, and TXB2 must be used instead.
659 if the target is a shadow texture, the reference value is always
660 in src.z (this prevents shadow 3d and shadow 2d arrays from
661 using this instruction, but this is not needed).
677 dst = texture\_sample(unit, coord, bias)
680 .. opcode:: TXB2 - Texture Lookup With Bias (some cube maps only)
682 this is the same as TXB, but uses another reg to encode the
683 lod bias value for cube map arrays and shadow cube maps.
684 Presumably shadow 2d arrays and shadow 3d targets could use
685 this encoding too, but this is not legal.
687 shadow cube map arrays are neither possible nor required.
697 dst = texture\_sample(unit, coord, bias)
700 .. opcode:: DIV - Divide
704 dst.x = \frac{src0.x}{src1.x}
706 dst.y = \frac{src0.y}{src1.y}
708 dst.z = \frac{src0.z}{src1.z}
710 dst.w = \frac{src0.w}{src1.w}
713 .. opcode:: DP2 - 2-component Dot Product
715 This instruction replicates its result.
719 dst = src0.x \times src1.x + src0.y \times src1.y
722 .. opcode:: TEX_LZ - Texture Lookup With LOD = 0
724 This is the same as TXL with LOD = 0. Like every texture opcode, it obeys
725 pipe_sampler_view::u.tex.first_level and pipe_sampler_state::min_lod.
726 There is no way to override those two in shaders.
742 dst = texture\_sample(unit, coord, lod)
745 .. opcode:: TXL - Texture Lookup With explicit LOD
747 for cube map array textures, the explicit lod value
748 cannot be passed in src0.w, and TXL2 must be used instead.
750 if the target is a shadow texture, the reference value is always
751 in src.z (this prevents shadow 3d / 2d array / cube targets from
752 using this instruction, but this is not needed).
768 dst = texture\_sample(unit, coord, lod)
771 .. opcode:: TXL2 - Texture Lookup With explicit LOD (for cube map arrays only)
773 this is the same as TXL, but uses another reg to encode the
775 Presumably shadow 3d / 2d array / cube targets could use
776 this encoding too, but this is not legal.
778 shadow cube map arrays are neither possible nor required.
788 dst = texture\_sample(unit, coord, lod)
792 ^^^^^^^^^^^^^^^^^^^^^^^^
794 These opcodes are primarily provided for special-use computational shaders.
795 Support for these opcodes indicated by a special pipe capability bit (TBD).
797 XXX doesn't look like most of the opcodes really belong here.
799 .. opcode:: CEIL - Ceiling
803 dst.x = \lceil src.x\rceil
805 dst.y = \lceil src.y\rceil
807 dst.z = \lceil src.z\rceil
809 dst.w = \lceil src.w\rceil
812 .. opcode:: TRUNC - Truncate
825 .. opcode:: MOD - Modulus
829 dst.x = src0.x \bmod src1.x
831 dst.y = src0.y \bmod src1.y
833 dst.z = src0.z \bmod src1.z
835 dst.w = src0.w \bmod src1.w
838 .. opcode:: UARL - Integer Address Register Load
840 Moves the contents of the source register, assumed to be an integer, into the
841 destination register, which is assumed to be an address (ADDR) register.
844 .. opcode:: TXF - Texel Fetch
846 As per NV_gpu_shader4, extract a single texel from a specified texture
847 image or PIPE_BUFFER resource. The source sampler may not be a CUBE or
849 four-component signed integer vector used to identify the single texel
850 accessed. 3 components + level. If the texture is multisampled, then
851 the fourth component indicates the sample, not the mipmap level.
852 Just like texture instructions, an optional
853 offset vector is provided, which is subject to various driver restrictions
854 (regarding range, source of offsets). This instruction ignores the sampler
857 TXF(uint_vec coord, int_vec offset).
860 .. opcode:: TXQ - Texture Size Query
862 As per NV_gpu_program4, retrieve the dimensions of the texture depending on
863 the target. For 1D (width), 2D/RECT/CUBE (width, height), 3D (width, height,
864 depth), 1D array (width, layers), 2D array (width, height, layers).
865 Also return the number of accessible levels (last_level - first_level + 1)
868 For components which don't return a resource dimension, their value
875 dst.x = texture\_width(unit, lod)
877 dst.y = texture\_height(unit, lod)
879 dst.z = texture\_depth(unit, lod)
881 dst.w = texture\_levels(unit)
884 .. opcode:: TXQS - Texture Samples Query
886 This retrieves the number of samples in the texture, and stores it
887 into the x component as an unsigned integer. The other components are
888 undefined. If the texture is not multisampled, this function returns
889 (1, undef, undef, undef).
893 dst.x = texture\_samples(unit)
896 .. opcode:: TG4 - Texture Gather
898 As per ARB_texture_gather, gathers the four texels to be used in a bi-linear
899 filtering operation and packs them into a single register. Only works with
900 2D, 2D array, cubemaps, and cubemaps arrays. For 2D textures, only the
901 addressing modes of the sampler and the top level of any mip pyramid are
902 used. Set W to zero. It behaves like the TEX instruction, but a filtered
903 sample is not generated. The four samples that contribute to filtering are
904 placed into xyzw in clockwise order, starting with the (u,v) texture
905 coordinate delta at the following locations (-, +), (+, +), (+, -), (-, -),
906 where the magnitude of the deltas are half a texel.
908 PIPE_CAP_TEXTURE_SM5 enhances this instruction to support shadow per-sample
909 depth compares, single component selection, and a non-constant offset. It
910 doesn't allow support for the GL independent offset to get i0,j0. This would
911 require another CAP is hw can do it natively. For now we lower that before
920 dst = texture\_gather4 (unit, coord, component)
922 (with SM5 - cube array shadow)
930 dst = texture\_gather (uint, coord, compare)
932 .. opcode:: LODQ - level of detail query
934 Compute the LOD information that the texture pipe would use to access the
935 texture. The Y component contains the computed LOD lambda_prime. The X
936 component contains the LOD that will be accessed, based on min/max lod's
943 dst.xy = lodq(uint, coord);
945 .. opcode:: CLOCK - retrieve the current shader time
947 Invoking this instruction multiple times in the same shader should
948 cause monotonically increasing values to be returned. The values
949 are implicitly 64-bit, so if fewer than 64 bits of precision are
950 available, to provide expected wraparound semantics, the value
951 should be shifted up so that the most significant bit of the time
952 is the most significant bit of the 64-bit value.
960 ^^^^^^^^^^^^^^^^^^^^^^^^
961 These opcodes are used for integer operations.
962 Support for these opcodes indicated by PIPE_SHADER_CAP_INTEGERS (all of them?)
965 .. opcode:: I2F - Signed Integer To Float
967 Rounding is unspecified (round to nearest even suggested).
971 dst.x = (float) src.x
973 dst.y = (float) src.y
975 dst.z = (float) src.z
977 dst.w = (float) src.w
980 .. opcode:: U2F - Unsigned Integer To Float
982 Rounding is unspecified (round to nearest even suggested).
986 dst.x = (float) src.x
988 dst.y = (float) src.y
990 dst.z = (float) src.z
992 dst.w = (float) src.w
995 .. opcode:: F2I - Float to Signed Integer
997 Rounding is towards zero (truncate).
998 Values outside signed range (including NaNs) produce undefined results.
1011 .. opcode:: F2U - Float to Unsigned Integer
1013 Rounding is towards zero (truncate).
1014 Values outside unsigned range (including NaNs) produce undefined results.
1018 dst.x = (unsigned) src.x
1020 dst.y = (unsigned) src.y
1022 dst.z = (unsigned) src.z
1024 dst.w = (unsigned) src.w
1027 .. opcode:: UADD - Integer Add
1029 This instruction works the same for signed and unsigned integers.
1030 The low 32bit of the result is returned.
1034 dst.x = src0.x + src1.x
1036 dst.y = src0.y + src1.y
1038 dst.z = src0.z + src1.z
1040 dst.w = src0.w + src1.w
1043 .. opcode:: UMAD - Integer Multiply And Add
1045 This instruction works the same for signed and unsigned integers.
1046 The multiplication returns the low 32bit (as does the result itself).
1050 dst.x = src0.x \times src1.x + src2.x
1052 dst.y = src0.y \times src1.y + src2.y
1054 dst.z = src0.z \times src1.z + src2.z
1056 dst.w = src0.w \times src1.w + src2.w
1059 .. opcode:: UMUL - Integer Multiply
1061 This instruction works the same for signed and unsigned integers.
1062 The low 32bit of the result is returned.
1066 dst.x = src0.x \times src1.x
1068 dst.y = src0.y \times src1.y
1070 dst.z = src0.z \times src1.z
1072 dst.w = src0.w \times src1.w
1075 .. opcode:: IMUL_HI - Signed Integer Multiply High Bits
1077 The high 32bits of the multiplication of 2 signed integers are returned.
1081 dst.x = (src0.x \times src1.x) >> 32
1083 dst.y = (src0.y \times src1.y) >> 32
1085 dst.z = (src0.z \times src1.z) >> 32
1087 dst.w = (src0.w \times src1.w) >> 32
1090 .. opcode:: UMUL_HI - Unsigned Integer Multiply High Bits
1092 The high 32bits of the multiplication of 2 unsigned integers are returned.
1096 dst.x = (src0.x \times src1.x) >> 32
1098 dst.y = (src0.y \times src1.y) >> 32
1100 dst.z = (src0.z \times src1.z) >> 32
1102 dst.w = (src0.w \times src1.w) >> 32
1105 .. opcode:: IDIV - Signed Integer Division
1107 TBD: behavior for division by zero.
1111 dst.x = \frac{src0.x}{src1.x}
1113 dst.y = \frac{src0.y}{src1.y}
1115 dst.z = \frac{src0.z}{src1.z}
1117 dst.w = \frac{src0.w}{src1.w}
1120 .. opcode:: UDIV - Unsigned Integer Division
1122 For division by zero, 0xffffffff is returned.
1126 dst.x = \frac{src0.x}{src1.x}
1128 dst.y = \frac{src0.y}{src1.y}
1130 dst.z = \frac{src0.z}{src1.z}
1132 dst.w = \frac{src0.w}{src1.w}
1135 .. opcode:: UMOD - Unsigned Integer Remainder
1137 If second arg is zero, 0xffffffff is returned.
1141 dst.x = src0.x \bmod src1.x
1143 dst.y = src0.y \bmod src1.y
1145 dst.z = src0.z \bmod src1.z
1147 dst.w = src0.w \bmod src1.w
1150 .. opcode:: NOT - Bitwise Not
1163 .. opcode:: AND - Bitwise And
1167 dst.x = src0.x \& src1.x
1169 dst.y = src0.y \& src1.y
1171 dst.z = src0.z \& src1.z
1173 dst.w = src0.w \& src1.w
1176 .. opcode:: OR - Bitwise Or
1180 dst.x = src0.x | src1.x
1182 dst.y = src0.y | src1.y
1184 dst.z = src0.z | src1.z
1186 dst.w = src0.w | src1.w
1189 .. opcode:: XOR - Bitwise Xor
1193 dst.x = src0.x \oplus src1.x
1195 dst.y = src0.y \oplus src1.y
1197 dst.z = src0.z \oplus src1.z
1199 dst.w = src0.w \oplus src1.w
1202 .. opcode:: IMAX - Maximum of Signed Integers
1206 dst.x = max(src0.x, src1.x)
1208 dst.y = max(src0.y, src1.y)
1210 dst.z = max(src0.z, src1.z)
1212 dst.w = max(src0.w, src1.w)
1215 .. opcode:: UMAX - Maximum of Unsigned Integers
1219 dst.x = max(src0.x, src1.x)
1221 dst.y = max(src0.y, src1.y)
1223 dst.z = max(src0.z, src1.z)
1225 dst.w = max(src0.w, src1.w)
1228 .. opcode:: IMIN - Minimum of Signed Integers
1232 dst.x = min(src0.x, src1.x)
1234 dst.y = min(src0.y, src1.y)
1236 dst.z = min(src0.z, src1.z)
1238 dst.w = min(src0.w, src1.w)
1241 .. opcode:: UMIN - Minimum of Unsigned Integers
1245 dst.x = min(src0.x, src1.x)
1247 dst.y = min(src0.y, src1.y)
1249 dst.z = min(src0.z, src1.z)
1251 dst.w = min(src0.w, src1.w)
1254 .. opcode:: SHL - Shift Left
1256 The shift count is masked with 0x1f before the shift is applied.
1260 dst.x = src0.x << (0x1f \& src1.x)
1262 dst.y = src0.y << (0x1f \& src1.y)
1264 dst.z = src0.z << (0x1f \& src1.z)
1266 dst.w = src0.w << (0x1f \& src1.w)
1269 .. opcode:: ISHR - Arithmetic Shift Right (of Signed Integer)
1271 The shift count is masked with 0x1f before the shift is applied.
1275 dst.x = src0.x >> (0x1f \& src1.x)
1277 dst.y = src0.y >> (0x1f \& src1.y)
1279 dst.z = src0.z >> (0x1f \& src1.z)
1281 dst.w = src0.w >> (0x1f \& src1.w)
1284 .. opcode:: USHR - Logical Shift Right
1286 The shift count is masked with 0x1f before the shift is applied.
1290 dst.x = src0.x >> (unsigned) (0x1f \& src1.x)
1292 dst.y = src0.y >> (unsigned) (0x1f \& src1.y)
1294 dst.z = src0.z >> (unsigned) (0x1f \& src1.z)
1296 dst.w = src0.w >> (unsigned) (0x1f \& src1.w)
1299 .. opcode:: UCMP - Integer Conditional Move
1303 dst.x = src0.x ? src1.x : src2.x
1305 dst.y = src0.y ? src1.y : src2.y
1307 dst.z = src0.z ? src1.z : src2.z
1309 dst.w = src0.w ? src1.w : src2.w
1313 .. opcode:: ISSG - Integer Set Sign
1317 dst.x = (src0.x < 0) ? -1 : (src0.x > 0) ? 1 : 0
1319 dst.y = (src0.y < 0) ? -1 : (src0.y > 0) ? 1 : 0
1321 dst.z = (src0.z < 0) ? -1 : (src0.z > 0) ? 1 : 0
1323 dst.w = (src0.w < 0) ? -1 : (src0.w > 0) ? 1 : 0
1327 .. opcode:: FSLT - Float Set On Less Than (ordered)
1329 Same comparison as SLT but returns integer instead of 1.0/0.0 float
1333 dst.x = (src0.x < src1.x) ? \sim 0 : 0
1335 dst.y = (src0.y < src1.y) ? \sim 0 : 0
1337 dst.z = (src0.z < src1.z) ? \sim 0 : 0
1339 dst.w = (src0.w < src1.w) ? \sim 0 : 0
1342 .. opcode:: ISLT - Signed Integer Set On Less Than
1346 dst.x = (src0.x < src1.x) ? \sim 0 : 0
1348 dst.y = (src0.y < src1.y) ? \sim 0 : 0
1350 dst.z = (src0.z < src1.z) ? \sim 0 : 0
1352 dst.w = (src0.w < src1.w) ? \sim 0 : 0
1355 .. opcode:: USLT - Unsigned Integer Set On Less Than
1359 dst.x = (src0.x < src1.x) ? \sim 0 : 0
1361 dst.y = (src0.y < src1.y) ? \sim 0 : 0
1363 dst.z = (src0.z < src1.z) ? \sim 0 : 0
1365 dst.w = (src0.w < src1.w) ? \sim 0 : 0
1368 .. opcode:: FSGE - Float Set On Greater Equal Than (ordered)
1370 Same comparison as SGE but returns integer instead of 1.0/0.0 float
1374 dst.x = (src0.x >= src1.x) ? \sim 0 : 0
1376 dst.y = (src0.y >= src1.y) ? \sim 0 : 0
1378 dst.z = (src0.z >= src1.z) ? \sim 0 : 0
1380 dst.w = (src0.w >= src1.w) ? \sim 0 : 0
1383 .. opcode:: ISGE - Signed Integer Set On Greater Equal Than
1387 dst.x = (src0.x >= src1.x) ? \sim 0 : 0
1389 dst.y = (src0.y >= src1.y) ? \sim 0 : 0
1391 dst.z = (src0.z >= src1.z) ? \sim 0 : 0
1393 dst.w = (src0.w >= src1.w) ? \sim 0 : 0
1396 .. opcode:: USGE - Unsigned Integer Set On Greater Equal Than
1400 dst.x = (src0.x >= src1.x) ? \sim 0 : 0
1402 dst.y = (src0.y >= src1.y) ? \sim 0 : 0
1404 dst.z = (src0.z >= src1.z) ? \sim 0 : 0
1406 dst.w = (src0.w >= src1.w) ? \sim 0 : 0
1409 .. opcode:: FSEQ - Float Set On Equal (ordered)
1411 Same comparison as SEQ but returns integer instead of 1.0/0.0 float
1415 dst.x = (src0.x == src1.x) ? \sim 0 : 0
1417 dst.y = (src0.y == src1.y) ? \sim 0 : 0
1419 dst.z = (src0.z == src1.z) ? \sim 0 : 0
1421 dst.w = (src0.w == src1.w) ? \sim 0 : 0
1424 .. opcode:: USEQ - Integer Set On Equal
1428 dst.x = (src0.x == src1.x) ? \sim 0 : 0
1430 dst.y = (src0.y == src1.y) ? \sim 0 : 0
1432 dst.z = (src0.z == src1.z) ? \sim 0 : 0
1434 dst.w = (src0.w == src1.w) ? \sim 0 : 0
1437 .. opcode:: FSNE - Float Set On Not Equal (unordered)
1439 Same comparison as SNE but returns integer instead of 1.0/0.0 float
1443 dst.x = (src0.x != src1.x) ? \sim 0 : 0
1445 dst.y = (src0.y != src1.y) ? \sim 0 : 0
1447 dst.z = (src0.z != src1.z) ? \sim 0 : 0
1449 dst.w = (src0.w != src1.w) ? \sim 0 : 0
1452 .. opcode:: USNE - Integer Set On Not Equal
1456 dst.x = (src0.x != src1.x) ? \sim 0 : 0
1458 dst.y = (src0.y != src1.y) ? \sim 0 : 0
1460 dst.z = (src0.z != src1.z) ? \sim 0 : 0
1462 dst.w = (src0.w != src1.w) ? \sim 0 : 0
1465 .. opcode:: INEG - Integer Negate
1480 .. opcode:: IABS - Integer Absolute Value
1494 These opcodes are used for bit-level manipulation of integers.
1496 .. opcode:: IBFE - Signed Bitfield Extract
1498 Like GLSL bitfieldExtract. Extracts a set of bits from the input, and
1499 sign-extends them if the high bit of the extracted window is set.
1503 def ibfe(value, offset, bits):
1504 if offset < 0 or bits < 0 or offset + bits > 32:
1506 if bits == 0: return 0
1507 # Note: >> sign-extends
1508 return (value << (32 - offset - bits)) >> (32 - bits)
1510 .. opcode:: UBFE - Unsigned Bitfield Extract
1512 Like GLSL bitfieldExtract. Extracts a set of bits from the input, without
1517 def ubfe(value, offset, bits):
1518 if offset < 0 or bits < 0 or offset + bits > 32:
1520 if bits == 0: return 0
1521 # Note: >> does not sign-extend
1522 return (value << (32 - offset - bits)) >> (32 - bits)
1524 .. opcode:: BFI - Bitfield Insert
1526 Like GLSL bitfieldInsert. Replaces a bit region of 'base' with the low bits
1531 def bfi(base, insert, offset, bits):
1532 if offset < 0 or bits < 0 or offset + bits > 32:
1534 # << defined such that mask == ~0 when bits == 32, offset == 0
1535 mask = ((1 << bits) - 1) << offset
1536 return ((insert << offset) & mask) | (base & ~mask)
1538 .. opcode:: BREV - Bitfield Reverse
1540 See SM5 instruction BFREV. Reverses the bits of the argument.
1542 .. opcode:: POPC - Population Count
1544 See SM5 instruction COUNTBITS. Counts the number of set bits in the argument.
1546 .. opcode:: LSB - Index of lowest set bit
1548 See SM5 instruction FIRSTBIT_LO. Computes the 0-based index of the first set
1549 bit of the argument. Returns -1 if none are set.
1551 .. opcode:: IMSB - Index of highest non-sign bit
1553 See SM5 instruction FIRSTBIT_SHI. Computes the 0-based index of the highest
1554 non-sign bit of the argument (i.e. highest 0 bit for negative numbers,
1555 highest 1 bit for positive numbers). Returns -1 if all bits are the same
1556 (i.e. for inputs 0 and -1).
1558 .. opcode:: UMSB - Index of highest set bit
1560 See SM5 instruction FIRSTBIT_HI. Computes the 0-based index of the highest
1561 set bit of the argument. Returns -1 if none are set.
1564 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1566 These opcodes are only supported in geometry shaders; they have no meaning
1567 in any other type of shader.
1569 .. opcode:: EMIT - Emit
1571 Generate a new vertex for the current primitive into the specified vertex
1572 stream using the values in the output registers.
1575 .. opcode:: ENDPRIM - End Primitive
1577 Complete the current primitive in the specified vertex stream (consisting of
1578 the emitted vertices), and start a new one.
1584 These opcodes are part of :term:`GLSL`'s opcode set. Support for these
1585 opcodes is determined by a special capability bit, ``GLSL``.
1586 Some require glsl version 1.30 (UIF/SWITCH/CASE/DEFAULT/ENDSWITCH).
1588 .. opcode:: CAL - Subroutine Call
1594 .. opcode:: RET - Subroutine Call Return
1599 .. opcode:: CONT - Continue
1601 Unconditionally moves the point of execution to the instruction after the
1602 last bgnloop. The instruction must appear within a bgnloop/endloop.
1606 Support for CONT is determined by a special capability bit,
1607 ``TGSI_CONT_SUPPORTED``. See :ref:`Screen` for more information.
1610 .. opcode:: BGNLOOP - Begin a Loop
1612 Start a loop. Must have a matching endloop.
1615 .. opcode:: BGNSUB - Begin Subroutine
1617 Starts definition of a subroutine. Must have a matching endsub.
1620 .. opcode:: ENDLOOP - End a Loop
1622 End a loop started with bgnloop.
1625 .. opcode:: ENDSUB - End Subroutine
1627 Ends definition of a subroutine.
1630 .. opcode:: NOP - No Operation
1635 .. opcode:: BRK - Break
1637 Unconditionally moves the point of execution to the instruction after the
1638 next endloop or endswitch. The instruction must appear within a loop/endloop
1639 or switch/endswitch.
1642 .. opcode:: IF - Float If
1644 Start an IF ... ELSE .. ENDIF block. Condition evaluates to true if
1648 where src0.x is interpreted as a floating point register.
1651 .. opcode:: UIF - Bitwise If
1653 Start an UIF ... ELSE .. ENDIF block. Condition evaluates to true if
1657 where src0.x is interpreted as an integer register.
1660 .. opcode:: ELSE - Else
1662 Starts an else block, after an IF or UIF statement.
1665 .. opcode:: ENDIF - End If
1667 Ends an IF or UIF block.
1670 .. opcode:: SWITCH - Switch
1672 Starts a C-style switch expression. The switch consists of one or multiple
1673 CASE statements, and at most one DEFAULT statement. Execution of a statement
1674 ends when a BRK is hit, but just like in C falling through to other cases
1675 without a break is allowed. Similarly, DEFAULT label is allowed anywhere not
1676 just as last statement, and fallthrough is allowed into/from it.
1677 CASE src arguments are evaluated at bit level against the SWITCH src argument.
1683 (some instructions here)
1686 (some instructions here)
1689 (some instructions here)
1694 .. opcode:: CASE - Switch case
1696 This represents a switch case label. The src arg must be an integer immediate.
1699 .. opcode:: DEFAULT - Switch default
1701 This represents the default case in the switch, which is taken if no other
1705 .. opcode:: ENDSWITCH - End of switch
1707 Ends a switch expression.
1713 The interpolation instructions allow an input to be interpolated in a
1714 different way than its declaration. This corresponds to the GLSL 4.00
1715 interpolateAt* functions. The first argument of each of these must come from
1716 ``TGSI_FILE_INPUT``.
1718 .. opcode:: INTERP_CENTROID - Interpolate at the centroid
1720 Interpolates the varying specified by src0 at the centroid
1722 .. opcode:: INTERP_SAMPLE - Interpolate at the specified sample
1724 Interpolates the varying specified by src0 at the sample id specified by
1725 src1.x (interpreted as an integer)
1727 .. opcode:: INTERP_OFFSET - Interpolate at the specified offset
1729 Interpolates the varying specified by src0 at the offset src1.xy from the
1730 pixel center (interpreted as floats)
1738 The double-precision opcodes reinterpret four-component vectors into
1739 two-component vectors with doubled precision in each component.
1741 .. opcode:: DABS - Absolute
1749 .. opcode:: DADD - Add
1753 dst.xy = src0.xy + src1.xy
1755 dst.zw = src0.zw + src1.zw
1757 .. opcode:: DSEQ - Set on Equal
1761 dst.x = src0.xy == src1.xy ? \sim 0 : 0
1763 dst.z = src0.zw == src1.zw ? \sim 0 : 0
1765 .. opcode:: DSNE - Set on Not Equal
1769 dst.x = src0.xy != src1.xy ? \sim 0 : 0
1771 dst.z = src0.zw != src1.zw ? \sim 0 : 0
1773 .. opcode:: DSLT - Set on Less than
1777 dst.x = src0.xy < src1.xy ? \sim 0 : 0
1779 dst.z = src0.zw < src1.zw ? \sim 0 : 0
1781 .. opcode:: DSGE - Set on Greater equal
1785 dst.x = src0.xy >= src1.xy ? \sim 0 : 0
1787 dst.z = src0.zw >= src1.zw ? \sim 0 : 0
1789 .. opcode:: DFRAC - Fraction
1793 dst.xy = src.xy - \lfloor src.xy\rfloor
1795 dst.zw = src.zw - \lfloor src.zw\rfloor
1797 .. opcode:: DTRUNC - Truncate
1801 dst.xy = trunc(src.xy)
1803 dst.zw = trunc(src.zw)
1805 .. opcode:: DCEIL - Ceiling
1809 dst.xy = \lceil src.xy\rceil
1811 dst.zw = \lceil src.zw\rceil
1813 .. opcode:: DFLR - Floor
1817 dst.xy = \lfloor src.xy\rfloor
1819 dst.zw = \lfloor src.zw\rfloor
1821 .. opcode:: DROUND - Fraction
1825 dst.xy = round(src.xy)
1827 dst.zw = round(src.zw)
1829 .. opcode:: DSSG - Set Sign
1833 dst.xy = (src.xy > 0) ? 1.0 : (src.xy < 0) ? -1.0 : 0.0
1835 dst.zw = (src.zw > 0) ? 1.0 : (src.zw < 0) ? -1.0 : 0.0
1837 .. opcode:: DFRACEXP - Convert Number to Fractional and Integral Components
1839 Like the ``frexp()`` routine in many math libraries, this opcode stores the
1840 exponent of its source to ``dst0``, and the significand to ``dst1``, such that
1841 :math:`dst1 \times 2^{dst0} = src` . The results are replicated across
1846 dst0.xy = dst.zw = frac(src.xy)
1851 .. opcode:: DLDEXP - Multiply Number by Integral Power of 2
1853 This opcode is the inverse of :opcode:`DFRACEXP`. The second
1854 source is an integer.
1858 dst.xy = src0.xy \times 2^{src1.x}
1860 dst.zw = src0.zw \times 2^{src1.z}
1862 .. opcode:: DMIN - Minimum
1866 dst.xy = min(src0.xy, src1.xy)
1868 dst.zw = min(src0.zw, src1.zw)
1870 .. opcode:: DMAX - Maximum
1874 dst.xy = max(src0.xy, src1.xy)
1876 dst.zw = max(src0.zw, src1.zw)
1878 .. opcode:: DMUL - Multiply
1882 dst.xy = src0.xy \times src1.xy
1884 dst.zw = src0.zw \times src1.zw
1887 .. opcode:: DMAD - Multiply And Add
1891 dst.xy = src0.xy \times src1.xy + src2.xy
1893 dst.zw = src0.zw \times src1.zw + src2.zw
1896 .. opcode:: DFMA - Fused Multiply-Add
1898 Perform a * b + c with no intermediate rounding step.
1902 dst.xy = src0.xy \times src1.xy + src2.xy
1904 dst.zw = src0.zw \times src1.zw + src2.zw
1907 .. opcode:: DDIV - Divide
1911 dst.xy = \frac{src0.xy}{src1.xy}
1913 dst.zw = \frac{src0.zw}{src1.zw}
1916 .. opcode:: DRCP - Reciprocal
1920 dst.xy = \frac{1}{src.xy}
1922 dst.zw = \frac{1}{src.zw}
1924 .. opcode:: DSQRT - Square Root
1928 dst.xy = \sqrt{src.xy}
1930 dst.zw = \sqrt{src.zw}
1932 .. opcode:: DRSQ - Reciprocal Square Root
1936 dst.xy = \frac{1}{\sqrt{src.xy}}
1938 dst.zw = \frac{1}{\sqrt{src.zw}}
1940 .. opcode:: F2D - Float to Double
1944 dst.xy = double(src0.x)
1946 dst.zw = double(src0.y)
1948 .. opcode:: D2F - Double to Float
1952 dst.x = float(src0.xy)
1954 dst.y = float(src0.zw)
1956 .. opcode:: I2D - Int to Double
1960 dst.xy = double(src0.x)
1962 dst.zw = double(src0.y)
1964 .. opcode:: D2I - Double to Int
1968 dst.x = int(src0.xy)
1970 dst.y = int(src0.zw)
1972 .. opcode:: U2D - Unsigned Int to Double
1976 dst.xy = double(src0.x)
1978 dst.zw = double(src0.y)
1980 .. opcode:: D2U - Double to Unsigned Int
1984 dst.x = unsigned(src0.xy)
1986 dst.y = unsigned(src0.zw)
1991 The 64-bit integer opcodes reinterpret four-component vectors into
1992 two-component vectors with 64-bits in each component.
1994 .. opcode:: I64ABS - 64-bit Integer Absolute Value
2002 .. opcode:: I64NEG - 64-bit Integer Negate
2012 .. opcode:: I64SSG - 64-bit Integer Set Sign
2016 dst.xy = (src0.xy < 0) ? -1 : (src0.xy > 0) ? 1 : 0
2018 dst.zw = (src0.zw < 0) ? -1 : (src0.zw > 0) ? 1 : 0
2020 .. opcode:: U64ADD - 64-bit Integer Add
2024 dst.xy = src0.xy + src1.xy
2026 dst.zw = src0.zw + src1.zw
2028 .. opcode:: U64MUL - 64-bit Integer Multiply
2032 dst.xy = src0.xy * src1.xy
2034 dst.zw = src0.zw * src1.zw
2036 .. opcode:: U64SEQ - 64-bit Integer Set on Equal
2040 dst.x = src0.xy == src1.xy ? \sim 0 : 0
2042 dst.z = src0.zw == src1.zw ? \sim 0 : 0
2044 .. opcode:: U64SNE - 64-bit Integer Set on Not Equal
2048 dst.x = src0.xy != src1.xy ? \sim 0 : 0
2050 dst.z = src0.zw != src1.zw ? \sim 0 : 0
2052 .. opcode:: U64SLT - 64-bit Unsigned Integer Set on Less Than
2056 dst.x = src0.xy < src1.xy ? \sim 0 : 0
2058 dst.z = src0.zw < src1.zw ? \sim 0 : 0
2060 .. opcode:: U64SGE - 64-bit Unsigned Integer Set on Greater Equal
2064 dst.x = src0.xy >= src1.xy ? \sim 0 : 0
2066 dst.z = src0.zw >= src1.zw ? \sim 0 : 0
2068 .. opcode:: I64SLT - 64-bit Signed Integer Set on Less Than
2072 dst.x = src0.xy < src1.xy ? \sim 0 : 0
2074 dst.z = src0.zw < src1.zw ? \sim 0 : 0
2076 .. opcode:: I64SGE - 64-bit Signed Integer Set on Greater Equal
2080 dst.x = src0.xy >= src1.xy ? \sim 0 : 0
2082 dst.z = src0.zw >= src1.zw ? \sim 0 : 0
2084 .. opcode:: I64MIN - Minimum of 64-bit Signed Integers
2088 dst.xy = min(src0.xy, src1.xy)
2090 dst.zw = min(src0.zw, src1.zw)
2092 .. opcode:: U64MIN - Minimum of 64-bit Unsigned Integers
2096 dst.xy = min(src0.xy, src1.xy)
2098 dst.zw = min(src0.zw, src1.zw)
2100 .. opcode:: I64MAX - Maximum of 64-bit Signed Integers
2104 dst.xy = max(src0.xy, src1.xy)
2106 dst.zw = max(src0.zw, src1.zw)
2108 .. opcode:: U64MAX - Maximum of 64-bit Unsigned Integers
2112 dst.xy = max(src0.xy, src1.xy)
2114 dst.zw = max(src0.zw, src1.zw)
2116 .. opcode:: U64SHL - Shift Left 64-bit Unsigned Integer
2118 The shift count is masked with 0x3f before the shift is applied.
2122 dst.xy = src0.xy << (0x3f \& src1.x)
2124 dst.zw = src0.zw << (0x3f \& src1.y)
2126 .. opcode:: I64SHR - Arithmetic Shift Right (of 64-bit Signed Integer)
2128 The shift count is masked with 0x3f before the shift is applied.
2132 dst.xy = src0.xy >> (0x3f \& src1.x)
2134 dst.zw = src0.zw >> (0x3f \& src1.y)
2136 .. opcode:: U64SHR - Logical Shift Right (of 64-bit Unsigned Integer)
2138 The shift count is masked with 0x3f before the shift is applied.
2142 dst.xy = src0.xy >> (unsigned) (0x3f \& src1.x)
2144 dst.zw = src0.zw >> (unsigned) (0x3f \& src1.y)
2146 .. opcode:: I64DIV - 64-bit Signed Integer Division
2150 dst.xy = \frac{src0.xy}{src1.xy}
2152 dst.zw = \frac{src0.zw}{src1.zw}
2154 .. opcode:: U64DIV - 64-bit Unsigned Integer Division
2158 dst.xy = \frac{src0.xy}{src1.xy}
2160 dst.zw = \frac{src0.zw}{src1.zw}
2162 .. opcode:: U64MOD - 64-bit Unsigned Integer Remainder
2166 dst.xy = src0.xy \bmod src1.xy
2168 dst.zw = src0.zw \bmod src1.zw
2170 .. opcode:: I64MOD - 64-bit Signed Integer Remainder
2174 dst.xy = src0.xy \bmod src1.xy
2176 dst.zw = src0.zw \bmod src1.zw
2178 .. opcode:: F2U64 - Float to 64-bit Unsigned Int
2182 dst.xy = (uint64_t) src0.x
2184 dst.zw = (uint64_t) src0.y
2186 .. opcode:: F2I64 - Float to 64-bit Int
2190 dst.xy = (int64_t) src0.x
2192 dst.zw = (int64_t) src0.y
2194 .. opcode:: U2I64 - Unsigned Integer to 64-bit Integer
2196 This is a zero extension.
2200 dst.xy = (int64_t) src0.x
2202 dst.zw = (int64_t) src0.y
2204 .. opcode:: I2I64 - Signed Integer to 64-bit Integer
2206 This is a sign extension.
2210 dst.xy = (int64_t) src0.x
2212 dst.zw = (int64_t) src0.y
2214 .. opcode:: D2U64 - Double to 64-bit Unsigned Int
2218 dst.xy = (uint64_t) src0.xy
2220 dst.zw = (uint64_t) src0.zw
2222 .. opcode:: D2I64 - Double to 64-bit Int
2226 dst.xy = (int64_t) src0.xy
2228 dst.zw = (int64_t) src0.zw
2230 .. opcode:: U642F - 64-bit unsigned integer to float
2234 dst.x = (float) src0.xy
2236 dst.y = (float) src0.zw
2238 .. opcode:: I642F - 64-bit Int to Float
2242 dst.x = (float) src0.xy
2244 dst.y = (float) src0.zw
2246 .. opcode:: U642D - 64-bit unsigned integer to double
2250 dst.xy = (double) src0.xy
2252 dst.zw = (double) src0.zw
2254 .. opcode:: I642D - 64-bit Int to double
2258 dst.xy = (double) src0.xy
2260 dst.zw = (double) src0.zw
2262 .. _samplingopcodes:
2264 Resource Sampling Opcodes
2265 ^^^^^^^^^^^^^^^^^^^^^^^^^
2267 Those opcodes follow very closely semantics of the respective Direct3D
2268 instructions. If in doubt double check Direct3D documentation.
2269 Note that the swizzle on SVIEW (src1) determines texel swizzling
2274 Using provided address, sample data from the specified texture using the
2275 filtering mode identified by the given sampler. The source data may come from
2276 any resource type other than buffers.
2278 Syntax: ``SAMPLE dst, address, sampler_view, sampler``
2280 Example: ``SAMPLE TEMP[0], TEMP[1], SVIEW[0], SAMP[0]``
2282 .. opcode:: SAMPLE_I
2284 Simplified alternative to the SAMPLE instruction. Using the provided
2285 integer address, SAMPLE_I fetches data from the specified sampler view
2286 without any filtering. The source data may come from any resource type
2289 Syntax: ``SAMPLE_I dst, address, sampler_view``
2291 Example: ``SAMPLE_I TEMP[0], TEMP[1], SVIEW[0]``
2293 The 'address' is specified as unsigned integers. If the 'address' is out of
2294 range [0...(# texels - 1)] the result of the fetch is always 0 in all
2295 components. As such the instruction doesn't honor address wrap modes, in
2296 cases where that behavior is desirable 'SAMPLE' instruction should be used.
2297 address.w always provides an unsigned integer mipmap level. If the value is
2298 out of the range then the instruction always returns 0 in all components.
2299 address.yz are ignored for buffers and 1d textures. address.z is ignored
2300 for 1d texture arrays and 2d textures.
2302 For 1D texture arrays address.y provides the array index (also as unsigned
2303 integer). If the value is out of the range of available array indices
2304 [0... (array size - 1)] then the opcode always returns 0 in all components.
2305 For 2D texture arrays address.z provides the array index, otherwise it
2306 exhibits the same behavior as in the case for 1D texture arrays. The exact
2307 semantics of the source address are presented in the table below:
2309 +---------------------------+----+-----+-----+---------+
2310 | resource type | X | Y | Z | W |
2311 +===========================+====+=====+=====+=========+
2312 | ``PIPE_BUFFER`` | x | | | ignored |
2313 +---------------------------+----+-----+-----+---------+
2314 | ``PIPE_TEXTURE_1D`` | x | | | mpl |
2315 +---------------------------+----+-----+-----+---------+
2316 | ``PIPE_TEXTURE_2D`` | x | y | | mpl |
2317 +---------------------------+----+-----+-----+---------+
2318 | ``PIPE_TEXTURE_3D`` | x | y | z | mpl |
2319 +---------------------------+----+-----+-----+---------+
2320 | ``PIPE_TEXTURE_RECT`` | x | y | | mpl |
2321 +---------------------------+----+-----+-----+---------+
2322 | ``PIPE_TEXTURE_CUBE`` | not allowed as source |
2323 +---------------------------+----+-----+-----+---------+
2324 | ``PIPE_TEXTURE_1D_ARRAY`` | x | idx | | mpl |
2325 +---------------------------+----+-----+-----+---------+
2326 | ``PIPE_TEXTURE_2D_ARRAY`` | x | y | idx | mpl |
2327 +---------------------------+----+-----+-----+---------+
2329 Where 'mpl' is a mipmap level and 'idx' is the array index.
2331 .. opcode:: SAMPLE_I_MS
2333 Just like SAMPLE_I but allows fetch data from multi-sampled surfaces.
2335 Syntax: ``SAMPLE_I_MS dst, address, sampler_view, sample``
2337 .. opcode:: SAMPLE_B
2339 Just like the SAMPLE instruction with the exception that an additional bias
2340 is applied to the level of detail computed as part of the instruction
2343 Syntax: ``SAMPLE_B dst, address, sampler_view, sampler, lod_bias``
2345 Example: ``SAMPLE_B TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x``
2347 .. opcode:: SAMPLE_C
2349 Similar to the SAMPLE instruction but it performs a comparison filter. The
2350 operands to SAMPLE_C are identical to SAMPLE, except that there is an
2351 additional float32 operand, reference value, which must be a register with
2352 single-component, or a scalar literal. SAMPLE_C makes the hardware use the
2353 current samplers compare_func (in pipe_sampler_state) to compare reference
2354 value against the red component value for the surce resource at each texel
2355 that the currently configured texture filter covers based on the provided
2358 Syntax: ``SAMPLE_C dst, address, sampler_view.r, sampler, ref_value``
2360 Example: ``SAMPLE_C TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x``
2362 .. opcode:: SAMPLE_C_LZ
2364 Same as SAMPLE_C, but LOD is 0 and derivatives are ignored. The LZ stands
2367 Syntax: ``SAMPLE_C_LZ dst, address, sampler_view.r, sampler, ref_value``
2369 Example: ``SAMPLE_C_LZ TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x``
2372 .. opcode:: SAMPLE_D
2374 SAMPLE_D is identical to the SAMPLE opcode except that the derivatives for
2375 the source address in the x direction and the y direction are provided by
2378 Syntax: ``SAMPLE_D dst, address, sampler_view, sampler, der_x, der_y``
2380 Example: ``SAMPLE_D TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2], TEMP[3]``
2382 .. opcode:: SAMPLE_L
2384 SAMPLE_L is identical to the SAMPLE opcode except that the LOD is provided
2385 directly as a scalar value, representing no anisotropy.
2387 Syntax: ``SAMPLE_L dst, address, sampler_view, sampler, explicit_lod``
2389 Example: ``SAMPLE_L TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x``
2393 Gathers the four texels to be used in a bi-linear filtering operation and
2394 packs them into a single register. Only works with 2D, 2D array, cubemaps,
2395 and cubemaps arrays. For 2D textures, only the addressing modes of the
2396 sampler and the top level of any mip pyramid are used. Set W to zero. It
2397 behaves like the SAMPLE instruction, but a filtered sample is not
2398 generated. The four samples that contribute to filtering are placed into
2399 xyzw in counter-clockwise order, starting with the (u,v) texture coordinate
2400 delta at the following locations (-, +), (+, +), (+, -), (-, -), where the
2401 magnitude of the deltas are half a texel.
2404 .. opcode:: SVIEWINFO
2406 Query the dimensions of a given sampler view. dst receives width, height,
2407 depth or array size and number of mipmap levels as int4. The dst can have a
2408 writemask which will specify what info is the caller interested in.
2410 Syntax: ``SVIEWINFO dst, src_mip_level, sampler_view``
2412 Example: ``SVIEWINFO TEMP[0], TEMP[1].x, SVIEW[0]``
2414 src_mip_level is an unsigned integer scalar. If it's out of range then
2415 returns 0 for width, height and depth/array size but the total number of
2416 mipmap is still returned correctly for the given sampler view. The returned
2417 width, height and depth values are for the mipmap level selected by the
2418 src_mip_level and are in the number of texels. For 1d texture array width
2419 is in dst.x, array size is in dst.y and dst.z is 0. The number of mipmaps is
2420 still in dst.w. In contrast to d3d10 resinfo, there's no way in the tgsi
2421 instruction encoding to specify the return type (float/rcpfloat/uint), hence
2422 always using uint. Also, unlike the SAMPLE instructions, the swizzle on src1
2423 resinfo allowing swizzling dst values is ignored (due to the interaction
2424 with rcpfloat modifier which requires some swizzle handling in the state
2427 .. opcode:: SAMPLE_POS
2429 Query the position of a sample in the given resource or render target
2430 when per-sample fragment shading is in effect.
2432 Syntax: ``SAMPLE_POS dst, source, sample_index``
2434 dst receives float4 (x, y, undef, undef) indicated where the sample is
2435 located. Sample locations are in the range [0, 1] where 0.5 is the center
2438 source is either a sampler view (to indicate a shader resource) or temp
2439 register (to indicate the render target). The source register may have
2440 an optional swizzle to apply to the returned result
2442 sample_index is an integer scalar indicating which sample position is to
2445 If per-sample shading is not in effect or the source resource or render
2446 target is not multisampled, the result is (0.5, 0.5, undef, undef).
2448 NOTE: no driver has implemented this opcode yet (and no state tracker
2449 emits it). This information is subject to change.
2451 .. opcode:: SAMPLE_INFO
2453 Query the number of samples in a multisampled resource or render target.
2455 Syntax: ``SAMPLE_INFO dst, source``
2457 dst receives int4 (n, 0, 0, 0) where n is the number of samples in a
2458 resource or the render target.
2460 source is either a sampler view (to indicate a shader resource) or temp
2461 register (to indicate the render target). The source register may have
2462 an optional swizzle to apply to the returned result
2464 If per-sample shading is not in effect or the source resource or render
2465 target is not multisampled, the result is (1, 0, 0, 0).
2467 NOTE: no driver has implemented this opcode yet (and no state tracker
2468 emits it). This information is subject to change.
2470 .. _resourceopcodes:
2472 Resource Access Opcodes
2473 ^^^^^^^^^^^^^^^^^^^^^^^
2475 For these opcodes, the resource can be a BUFFER, IMAGE, or MEMORY.
2477 .. opcode:: LOAD - Fetch data from a shader buffer or image
2479 Syntax: ``LOAD dst, resource, address``
2481 Example: ``LOAD TEMP[0], BUFFER[0], TEMP[1]``
2483 Using the provided integer address, LOAD fetches data
2484 from the specified buffer or texture without any
2487 The 'address' is specified as a vector of unsigned
2488 integers. If the 'address' is out of range the result
2491 Only the first mipmap level of a resource can be read
2492 from using this instruction.
2494 For 1D or 2D texture arrays, the array index is
2495 provided as an unsigned integer in address.y or
2496 address.z, respectively. address.yz are ignored for
2497 buffers and 1D textures. address.z is ignored for 1D
2498 texture arrays and 2D textures. address.w is always
2501 A swizzle suffix may be added to the resource argument
2502 this will cause the resource data to be swizzled accordingly.
2504 .. opcode:: STORE - Write data to a shader resource
2506 Syntax: ``STORE resource, address, src``
2508 Example: ``STORE BUFFER[0], TEMP[0], TEMP[1]``
2510 Using the provided integer address, STORE writes data
2511 to the specified buffer or texture.
2513 The 'address' is specified as a vector of unsigned
2514 integers. If the 'address' is out of range the result
2517 Only the first mipmap level of a resource can be
2518 written to using this instruction.
2520 For 1D or 2D texture arrays, the array index is
2521 provided as an unsigned integer in address.y or
2522 address.z, respectively. address.yz are ignored for
2523 buffers and 1D textures. address.z is ignored for 1D
2524 texture arrays and 2D textures. address.w is always
2527 .. opcode:: RESQ - Query information about a resource
2529 Syntax: ``RESQ dst, resource``
2531 Example: ``RESQ TEMP[0], BUFFER[0]``
2533 Returns information about the buffer or image resource. For buffer
2534 resources, the size (in bytes) is returned in the x component. For
2535 image resources, .xyz will contain the width/height/layers of the
2536 image, while .w will contain the number of samples for multi-sampled
2539 .. opcode:: FBFETCH - Load data from framebuffer
2541 Syntax: ``FBFETCH dst, output``
2543 Example: ``FBFETCH TEMP[0], OUT[0]``
2545 This is only valid on ``COLOR`` semantic outputs. Returns the color
2546 of the current position in the framebuffer from before this fragment
2547 shader invocation. May return the same value from multiple calls for
2548 a particular output within a single invocation. Note that result may
2549 be undefined if a fragment is drawn multiple times without a blend
2553 .. _threadsyncopcodes:
2555 Inter-thread synchronization opcodes
2556 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2558 These opcodes are intended for communication between threads running
2559 within the same compute grid. For now they're only valid in compute
2562 .. opcode:: BARRIER - Thread group barrier
2566 This opcode suspends the execution of the current thread until all
2567 the remaining threads in the working group reach the same point of
2568 the program. Results are unspecified if any of the remaining
2569 threads terminates or never reaches an executed BARRIER instruction.
2571 .. opcode:: MEMBAR - Memory barrier
2575 This opcode waits for the completion of all memory accesses based on
2576 the type passed in. The type is an immediate bitfield with the following
2579 Bit 0: Shader storage buffers
2580 Bit 1: Atomic buffers
2582 Bit 3: Shared memory
2585 These may be passed in in any combination. An implementation is free to not
2586 distinguish between these as it sees fit. However these map to all the
2587 possibilities made available by GLSL.
2594 These opcodes provide atomic variants of some common arithmetic and
2595 logical operations. In this context atomicity means that another
2596 concurrent memory access operation that affects the same memory
2597 location is guaranteed to be performed strictly before or after the
2598 entire execution of the atomic operation. The resource may be a BUFFER,
2599 IMAGE, or MEMORY. In the case of an image, the offset works the same as for
2600 ``LOAD`` and ``STORE``, specified above. These atomic operations may
2601 only be used with 32-bit integer image formats.
2603 .. opcode:: ATOMUADD - Atomic integer addition
2605 Syntax: ``ATOMUADD dst, resource, offset, src``
2607 Example: ``ATOMUADD TEMP[0], BUFFER[0], TEMP[1], TEMP[2]``
2609 The following operation is performed atomically:
2613 dst_x = resource[offset]
2615 resource[offset] = dst_x + src_x
2618 .. opcode:: ATOMXCHG - Atomic exchange
2620 Syntax: ``ATOMXCHG dst, resource, offset, src``
2622 Example: ``ATOMXCHG TEMP[0], BUFFER[0], TEMP[1], TEMP[2]``
2624 The following operation is performed atomically:
2628 dst_x = resource[offset]
2630 resource[offset] = src_x
2633 .. opcode:: ATOMCAS - Atomic compare-and-exchange
2635 Syntax: ``ATOMCAS dst, resource, offset, cmp, src``
2637 Example: ``ATOMCAS TEMP[0], BUFFER[0], TEMP[1], TEMP[2], TEMP[3]``
2639 The following operation is performed atomically:
2643 dst_x = resource[offset]
2645 resource[offset] = (dst_x == cmp_x ? src_x : dst_x)
2648 .. opcode:: ATOMAND - Atomic bitwise And
2650 Syntax: ``ATOMAND dst, resource, offset, src``
2652 Example: ``ATOMAND TEMP[0], BUFFER[0], TEMP[1], TEMP[2]``
2654 The following operation is performed atomically:
2658 dst_x = resource[offset]
2660 resource[offset] = dst_x \& src_x
2663 .. opcode:: ATOMOR - Atomic bitwise Or
2665 Syntax: ``ATOMOR dst, resource, offset, src``
2667 Example: ``ATOMOR TEMP[0], BUFFER[0], TEMP[1], TEMP[2]``
2669 The following operation is performed atomically:
2673 dst_x = resource[offset]
2675 resource[offset] = dst_x | src_x
2678 .. opcode:: ATOMXOR - Atomic bitwise Xor
2680 Syntax: ``ATOMXOR dst, resource, offset, src``
2682 Example: ``ATOMXOR TEMP[0], BUFFER[0], TEMP[1], TEMP[2]``
2684 The following operation is performed atomically:
2688 dst_x = resource[offset]
2690 resource[offset] = dst_x \oplus src_x
2693 .. opcode:: ATOMUMIN - Atomic unsigned minimum
2695 Syntax: ``ATOMUMIN dst, resource, offset, src``
2697 Example: ``ATOMUMIN TEMP[0], BUFFER[0], TEMP[1], TEMP[2]``
2699 The following operation is performed atomically:
2703 dst_x = resource[offset]
2705 resource[offset] = (dst_x < src_x ? dst_x : src_x)
2708 .. opcode:: ATOMUMAX - Atomic unsigned maximum
2710 Syntax: ``ATOMUMAX dst, resource, offset, src``
2712 Example: ``ATOMUMAX TEMP[0], BUFFER[0], TEMP[1], TEMP[2]``
2714 The following operation is performed atomically:
2718 dst_x = resource[offset]
2720 resource[offset] = (dst_x > src_x ? dst_x : src_x)
2723 .. opcode:: ATOMIMIN - Atomic signed minimum
2725 Syntax: ``ATOMIMIN dst, resource, offset, src``
2727 Example: ``ATOMIMIN TEMP[0], BUFFER[0], TEMP[1], TEMP[2]``
2729 The following operation is performed atomically:
2733 dst_x = resource[offset]
2735 resource[offset] = (dst_x < src_x ? dst_x : src_x)
2738 .. opcode:: ATOMIMAX - Atomic signed maximum
2740 Syntax: ``ATOMIMAX dst, resource, offset, src``
2742 Example: ``ATOMIMAX TEMP[0], BUFFER[0], TEMP[1], TEMP[2]``
2744 The following operation is performed atomically:
2748 dst_x = resource[offset]
2750 resource[offset] = (dst_x > src_x ? dst_x : src_x)
2753 .. _interlaneopcodes:
2758 These opcodes reduce the given value across the shader invocations
2759 running in the current SIMD group. Every thread in the subgroup will receive
2760 the same result. The BALLOT operations accept a single-channel argument that
2761 is treated as a boolean and produce a 64-bit value.
2763 .. opcode:: VOTE_ANY - Value is set in any of the active invocations
2765 Syntax: ``VOTE_ANY dst, value``
2767 Example: ``VOTE_ANY TEMP[0].x, TEMP[1].x``
2770 .. opcode:: VOTE_ALL - Value is set in all of the active invocations
2772 Syntax: ``VOTE_ALL dst, value``
2774 Example: ``VOTE_ALL TEMP[0].x, TEMP[1].x``
2777 .. opcode:: VOTE_EQ - Value is the same in all of the active invocations
2779 Syntax: ``VOTE_EQ dst, value``
2781 Example: ``VOTE_EQ TEMP[0].x, TEMP[1].x``
2784 .. opcode:: BALLOT - Lanemask of whether the value is set in each active
2787 Syntax: ``BALLOT dst, value``
2789 Example: ``BALLOT TEMP[0].xy, TEMP[1].x``
2791 When the argument is a constant true, this produces a bitmask of active
2792 invocations. In fragment shaders, this can include helper invocations
2793 (invocations whose outputs and writes to memory are discarded, but which
2794 are used to compute derivatives).
2797 .. opcode:: READ_FIRST - Broadcast the value from the first active
2798 invocation to all active lanes
2800 Syntax: ``READ_FIRST dst, value``
2802 Example: ``READ_FIRST TEMP[0], TEMP[1]``
2805 .. opcode:: READ_INVOC - Retrieve the value from the given invocation
2806 (need not be uniform)
2808 Syntax: ``READ_INVOC dst, value, invocation``
2810 Example: ``READ_INVOC TEMP[0].xy, TEMP[1].xy, TEMP[2].x``
2812 invocation.x controls the invocation number to read from for all channels.
2813 The invocation number must be the same across all active invocations in a
2814 sub-group; otherwise, the results are undefined.
2817 Explanation of symbols used
2818 ------------------------------
2825 :math:`|x|` Absolute value of `x`.
2827 :math:`\lceil x \rceil` Ceiling of `x`.
2829 clamp(x,y,z) Clamp x between y and z.
2830 (x < y) ? y : (x > z) ? z : x
2832 :math:`\lfloor x\rfloor` Floor of `x`.
2834 :math:`\log_2{x}` Logarithm of `x`, base 2.
2836 max(x,y) Maximum of x and y.
2839 min(x,y) Minimum of x and y.
2842 partialx(x) Derivative of x relative to fragment's X.
2844 partialy(x) Derivative of x relative to fragment's Y.
2846 pop() Pop from stack.
2848 :math:`x^y` `x` to the power `y`.
2850 push(x) Push x on stack.
2854 trunc(x) Truncate x, i.e. drop the fraction bits.
2861 discard Discard fragment.
2865 target Label of target instruction.
2876 Declares a register that is will be referenced as an operand in Instruction
2879 File field contains register file that is being declared and is one
2882 UsageMask field specifies which of the register components can be accessed
2883 and is one of TGSI_WRITEMASK.
2885 The Local flag specifies that a given value isn't intended for
2886 subroutine parameter passing and, as a result, the implementation
2887 isn't required to give any guarantees of it being preserved across
2888 subroutine boundaries. As it's merely a compiler hint, the
2889 implementation is free to ignore it.
2891 If Dimension flag is set to 1, a Declaration Dimension token follows.
2893 If Semantic flag is set to 1, a Declaration Semantic token follows.
2895 If Interpolate flag is set to 1, a Declaration Interpolate token follows.
2897 If file is TGSI_FILE_RESOURCE, a Declaration Resource token follows.
2899 If Array flag is set to 1, a Declaration Array token follows.
2902 ^^^^^^^^^^^^^^^^^^^^^^^^
2904 Declarations can optional have an ArrayID attribute which can be referred by
2905 indirect addressing operands. An ArrayID of zero is reserved and treated as
2906 if no ArrayID is specified.
2908 If an indirect addressing operand refers to a specific declaration by using
2909 an ArrayID only the registers in this declaration are guaranteed to be
2910 accessed, accessing any register outside this declaration results in undefined
2911 behavior. Note that for compatibility the effective index is zero-based and
2912 not relative to the specified declaration
2914 If no ArrayID is specified with an indirect addressing operand the whole
2915 register file might be accessed by this operand. This is strongly discouraged
2916 and will prevent packing of scalar/vec2 arrays and effective alias analysis.
2917 This is only legal for TEMP and CONST register files.
2919 Declaration Semantic
2920 ^^^^^^^^^^^^^^^^^^^^^^^^
2922 Vertex and fragment shader input and output registers may be labeled
2923 with semantic information consisting of a name and index.
2925 Follows Declaration token if Semantic bit is set.
2927 Since its purpose is to link a shader with other stages of the pipeline,
2928 it is valid to follow only those Declaration tokens that declare a register
2929 either in INPUT or OUTPUT file.
2931 SemanticName field contains the semantic name of the register being declared.
2932 There is no default value.
2934 SemanticIndex is an optional subscript that can be used to distinguish
2935 different register declarations with the same semantic name. The default value
2938 The meanings of the individual semantic names are explained in the following
2941 TGSI_SEMANTIC_POSITION
2942 """"""""""""""""""""""
2944 For vertex shaders, TGSI_SEMANTIC_POSITION indicates the vertex shader
2945 output register which contains the homogeneous vertex position in the clip
2946 space coordinate system. After clipping, the X, Y and Z components of the
2947 vertex will be divided by the W value to get normalized device coordinates.
2949 For fragment shaders, TGSI_SEMANTIC_POSITION is used to indicate that
2950 fragment shader input (or system value, depending on which one is
2951 supported by the driver) contains the fragment's window position. The X
2952 component starts at zero and always increases from left to right.
2953 The Y component starts at zero and always increases but Y=0 may either
2954 indicate the top of the window or the bottom depending on the fragment
2955 coordinate origin convention (see TGSI_PROPERTY_FS_COORD_ORIGIN).
2956 The Z coordinate ranges from 0 to 1 to represent depth from the front
2957 to the back of the Z buffer. The W component contains the interpolated
2958 reciprocal of the vertex position W component (corresponding to gl_Fragcoord,
2959 but unlike d3d10 which interpolates the same 1/w but then gives back
2960 the reciprocal of the interpolated value).
2962 Fragment shaders may also declare an output register with
2963 TGSI_SEMANTIC_POSITION. Only the Z component is writable. This allows
2964 the fragment shader to change the fragment's Z position.
2971 For vertex shader outputs or fragment shader inputs/outputs, this
2972 label indicates that the register contains an R,G,B,A color.
2974 Several shader inputs/outputs may contain colors so the semantic index
2975 is used to distinguish them. For example, color[0] may be the diffuse
2976 color while color[1] may be the specular color.
2978 This label is needed so that the flat/smooth shading can be applied
2979 to the right interpolants during rasterization.
2983 TGSI_SEMANTIC_BCOLOR
2984 """"""""""""""""""""
2986 Back-facing colors are only used for back-facing polygons, and are only valid
2987 in vertex shader outputs. After rasterization, all polygons are front-facing
2988 and COLOR and BCOLOR end up occupying the same slots in the fragment shader,
2989 so all BCOLORs effectively become regular COLORs in the fragment shader.
2995 Vertex shader inputs and outputs and fragment shader inputs may be
2996 labeled with TGSI_SEMANTIC_FOG to indicate that the register contains
2997 a fog coordinate. Typically, the fragment shader will use the fog coordinate
2998 to compute a fog blend factor which is used to blend the normal fragment color
2999 with a constant fog color. But fog coord really is just an ordinary vec4
3000 register like regular semantics.
3006 Vertex shader input and output registers may be labeled with
3007 TGIS_SEMANTIC_PSIZE to indicate that the register contains a point size
3008 in the form (S, 0, 0, 1). The point size controls the width or diameter
3009 of points for rasterization. This label cannot be used in fragment
3012 When using this semantic, be sure to set the appropriate state in the
3013 :ref:`rasterizer` first.
3016 TGSI_SEMANTIC_TEXCOORD
3017 """"""""""""""""""""""
3019 Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
3021 Vertex shader outputs and fragment shader inputs may be labeled with
3022 this semantic to make them replaceable by sprite coordinates via the
3023 sprite_coord_enable state in the :ref:`rasterizer`.
3024 The semantic index permitted with this semantic is limited to <= 7.
3026 If the driver does not support TEXCOORD, sprite coordinate replacement
3027 applies to inputs with the GENERIC semantic instead.
3029 The intended use case for this semantic is gl_TexCoord.
3032 TGSI_SEMANTIC_PCOORD
3033 """"""""""""""""""""
3035 Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
3037 Fragment shader inputs may be labeled with TGSI_SEMANTIC_PCOORD to indicate
3038 that the register contains sprite coordinates in the form (x, y, 0, 1), if
3039 the current primitive is a point and point sprites are enabled. Otherwise,
3040 the contents of the register are undefined.
3042 The intended use case for this semantic is gl_PointCoord.
3045 TGSI_SEMANTIC_GENERIC
3046 """""""""""""""""""""
3048 All vertex/fragment shader inputs/outputs not labeled with any other
3049 semantic label can be considered to be generic attributes. Typical
3050 uses of generic inputs/outputs are texcoords and user-defined values.
3053 TGSI_SEMANTIC_NORMAL
3054 """"""""""""""""""""
3056 Indicates that a vertex shader input is a normal vector. This is
3057 typically only used for legacy graphics APIs.
3063 This label applies to fragment shader inputs (or system values,
3064 depending on which one is supported by the driver) and indicates that
3065 the register contains front/back-face information.
3067 If it is an input, it will be a floating-point vector in the form (F, 0, 0, 1),
3068 where F will be positive when the fragment belongs to a front-facing polygon,
3069 and negative when the fragment belongs to a back-facing polygon.
3071 If it is a system value, it will be an integer vector in the form (F, 0, 0, 1),
3072 where F is 0xffffffff when the fragment belongs to a front-facing polygon and
3073 0 when the fragment belongs to a back-facing polygon.
3076 TGSI_SEMANTIC_EDGEFLAG
3077 """"""""""""""""""""""
3079 For vertex shaders, this sematic label indicates that an input or
3080 output is a boolean edge flag. The register layout is [F, x, x, x]
3081 where F is 0.0 or 1.0 and x = don't care. Normally, the vertex shader
3082 simply copies the edge flag input to the edgeflag output.
3084 Edge flags are used to control which lines or points are actually
3085 drawn when the polygon mode converts triangles/quads/polygons into
3089 TGSI_SEMANTIC_STENCIL
3090 """""""""""""""""""""
3092 For fragment shaders, this semantic label indicates that an output
3093 is a writable stencil reference value. Only the Y component is writable.
3094 This allows the fragment shader to change the fragments stencilref value.
3097 TGSI_SEMANTIC_VIEWPORT_INDEX
3098 """"""""""""""""""""""""""""
3100 For geometry shaders, this semantic label indicates that an output
3101 contains the index of the viewport (and scissor) to use.
3102 This is an integer value, and only the X component is used.
3104 If PIPE_CAP_TGSI_VS_LAYER_VIEWPORT or PIPE_CAP_TGSI_TES_LAYER_VIEWPORT is
3105 supported, then this semantic label can also be used in vertex or
3106 tessellation evaluation shaders, respectively. Only the value written in the
3107 last vertex processing stage is used.
3113 For geometry shaders, this semantic label indicates that an output
3114 contains the layer value to use for the color and depth/stencil surfaces.
3115 This is an integer value, and only the X component is used.
3116 (Also known as rendertarget array index.)
3118 If PIPE_CAP_TGSI_VS_LAYER_VIEWPORT or PIPE_CAP_TGSI_TES_LAYER_VIEWPORT is
3119 supported, then this semantic label can also be used in vertex or
3120 tessellation evaluation shaders, respectively. Only the value written in the
3121 last vertex processing stage is used.
3124 TGSI_SEMANTIC_CULLDIST
3125 """"""""""""""""""""""
3127 Used as distance to plane for performing application-defined culling
3128 of individual primitives against a plane. When components of vertex
3129 elements are given this label, these values are assumed to be a
3130 float32 signed distance to a plane. Primitives will be completely
3131 discarded if the plane distance for all of the vertices in the
3132 primitive are < 0. If a vertex has a cull distance of NaN, that
3133 vertex counts as "out" (as if its < 0);
3134 The limits on both clip and cull distances are bound
3135 by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
3136 the maximum number of components that can be used to hold the
3137 distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
3138 which specifies the maximum number of registers which can be
3139 annotated with those semantics.
3142 TGSI_SEMANTIC_CLIPDIST
3143 """"""""""""""""""""""
3145 Note this covers clipping and culling distances.
3147 When components of vertex elements are identified this way, these
3148 values are each assumed to be a float32 signed distance to a plane.
3151 Primitive setup only invokes rasterization on pixels for which
3152 the interpolated plane distances are >= 0.
3155 Primitives will be completely discarded if the plane distance
3156 for all of the vertices in the primitive are < 0.
3157 If a vertex has a cull distance of NaN, that vertex counts as "out"
3160 Multiple clip/cull planes can be implemented simultaneously, by
3161 annotating multiple components of one or more vertex elements with
3162 the above specified semantic.
3163 The limits on both clip and cull distances are bound
3164 by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
3165 the maximum number of components that can be used to hold the
3166 distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
3167 which specifies the maximum number of registers which can be
3168 annotated with those semantics.
3169 The properties NUM_CLIPDIST_ENABLED and NUM_CULLDIST_ENABLED
3170 are used to divide up the 2 x vec4 space between clipping and culling.
3172 TGSI_SEMANTIC_SAMPLEID
3173 """"""""""""""""""""""
3175 For fragment shaders, this semantic label indicates that a system value
3176 contains the current sample id (i.e. gl_SampleID) as an unsigned int.
3177 Only the X component is used. If per-sample shading is not enabled,
3178 the result is (0, undef, undef, undef).
3180 Note that if the fragment shader uses this system value, the fragment
3181 shader is automatically executed at per sample frequency.
3183 TGSI_SEMANTIC_SAMPLEPOS
3184 """""""""""""""""""""""
3186 For fragment shaders, this semantic label indicates that a system
3187 value contains the current sample's position as float4(x, y, undef, undef)
3188 in the render target (i.e. gl_SamplePosition) when per-fragment shading
3189 is in effect. Position values are in the range [0, 1] where 0.5 is
3190 the center of the fragment.
3192 Note that if the fragment shader uses this system value, the fragment
3193 shader is automatically executed at per sample frequency.
3195 TGSI_SEMANTIC_SAMPLEMASK
3196 """"""""""""""""""""""""
3198 For fragment shaders, this semantic label can be applied to either a
3199 shader system value input or output.
3201 For a system value, the sample mask indicates the set of samples covered by
3202 the current primitive. If MSAA is not enabled, the value is (1, 0, 0, 0).
3204 For an output, the sample mask is used to disable further sample processing.
3206 For both, the register type is uint[4] but only the X component is used
3207 (i.e. gl_SampleMask[0]). Each bit corresponds to one sample position (up
3208 to 32x MSAA is supported).
3210 TGSI_SEMANTIC_INVOCATIONID
3211 """"""""""""""""""""""""""
3213 For geometry shaders, this semantic label indicates that a system value
3214 contains the current invocation id (i.e. gl_InvocationID).
3215 This is an integer value, and only the X component is used.
3217 TGSI_SEMANTIC_INSTANCEID
3218 """"""""""""""""""""""""
3220 For vertex shaders, this semantic label indicates that a system value contains
3221 the current instance id (i.e. gl_InstanceID). It does not include the base
3222 instance. This is an integer value, and only the X component is used.
3224 TGSI_SEMANTIC_VERTEXID
3225 """"""""""""""""""""""
3227 For vertex shaders, this semantic label indicates that a system value contains
3228 the current vertex id (i.e. gl_VertexID). It does (unlike in d3d10) include the
3229 base vertex. This is an integer value, and only the X component is used.
3231 TGSI_SEMANTIC_VERTEXID_NOBASE
3232 """""""""""""""""""""""""""""""
3234 For vertex shaders, this semantic label indicates that a system value contains
3235 the current vertex id without including the base vertex (this corresponds to
3236 d3d10 vertex id, so TGSI_SEMANTIC_VERTEXID_NOBASE + TGSI_SEMANTIC_BASEVERTEX
3237 == TGSI_SEMANTIC_VERTEXID). This is an integer value, and only the X component
3240 TGSI_SEMANTIC_BASEVERTEX
3241 """"""""""""""""""""""""
3243 For vertex shaders, this semantic label indicates that a system value contains
3244 the base vertex (i.e. gl_BaseVertex). Note that for non-indexed draw calls,
3245 this contains the first (or start) value instead.
3246 This is an integer value, and only the X component is used.
3248 TGSI_SEMANTIC_PRIMID
3249 """"""""""""""""""""
3251 For geometry and fragment shaders, this semantic label indicates the value
3252 contains the primitive id (i.e. gl_PrimitiveID). This is an integer value,
3253 and only the X component is used.
3254 FIXME: This right now can be either a ordinary input or a system value...
3260 For tessellation evaluation/control shaders, this semantic label indicates a
3261 generic per-patch attribute. Such semantics will not implicitly be per-vertex
3264 TGSI_SEMANTIC_TESSCOORD
3265 """""""""""""""""""""""
3267 For tessellation evaluation shaders, this semantic label indicates the
3268 coordinates of the vertex being processed. This is available in XYZ; W is
3271 TGSI_SEMANTIC_TESSOUTER
3272 """""""""""""""""""""""
3274 For tessellation evaluation/control shaders, this semantic label indicates the
3275 outer tessellation levels of the patch. Isoline tessellation will only have XY
3276 defined, triangle will have XYZ and quads will have XYZW defined. This
3277 corresponds to gl_TessLevelOuter.
3279 TGSI_SEMANTIC_TESSINNER
3280 """""""""""""""""""""""
3282 For tessellation evaluation/control shaders, this semantic label indicates the
3283 inner tessellation levels of the patch. The X value is only defined for
3284 triangle tessellation, while quads will have XY defined. This is entirely
3285 undefined for isoline tessellation.
3287 TGSI_SEMANTIC_VERTICESIN
3288 """"""""""""""""""""""""
3290 For tessellation evaluation/control shaders, this semantic label indicates the
3291 number of vertices provided in the input patch. Only the X value is defined.
3293 TGSI_SEMANTIC_HELPER_INVOCATION
3294 """""""""""""""""""""""""""""""
3296 For fragment shaders, this semantic indicates whether the current
3297 invocation is covered or not. Helper invocations are created in order
3298 to properly compute derivatives, however it may be desirable to skip
3299 some of the logic in those cases. See ``gl_HelperInvocation`` documentation.
3301 TGSI_SEMANTIC_BASEINSTANCE
3302 """"""""""""""""""""""""""
3304 For vertex shaders, the base instance argument supplied for this
3305 draw. This is an integer value, and only the X component is used.
3307 TGSI_SEMANTIC_DRAWID
3308 """"""""""""""""""""
3310 For vertex shaders, the zero-based index of the current draw in a
3311 ``glMultiDraw*`` invocation. This is an integer value, and only the X
3315 TGSI_SEMANTIC_WORK_DIM
3316 """"""""""""""""""""""
3318 For compute shaders started via opencl this retrieves the work_dim
3319 parameter to the clEnqueueNDRangeKernel call with which the shader
3323 TGSI_SEMANTIC_GRID_SIZE
3324 """""""""""""""""""""""
3326 For compute shaders, this semantic indicates the maximum (x, y, z) dimensions
3327 of a grid of thread blocks.
3330 TGSI_SEMANTIC_BLOCK_ID
3331 """"""""""""""""""""""
3333 For compute shaders, this semantic indicates the (x, y, z) coordinates of the
3334 current block inside of the grid.
3337 TGSI_SEMANTIC_BLOCK_SIZE
3338 """"""""""""""""""""""""
3340 For compute shaders, this semantic indicates the maximum (x, y, z) dimensions
3341 of a block in threads.
3344 TGSI_SEMANTIC_THREAD_ID
3345 """""""""""""""""""""""
3347 For compute shaders, this semantic indicates the (x, y, z) coordinates of the
3348 current thread inside of the block.
3351 TGSI_SEMANTIC_SUBGROUP_SIZE
3352 """""""""""""""""""""""""""
3354 This semantic indicates the subgroup size for the current invocation. This is
3355 an integer of at most 64, as it indicates the width of lanemasks. It does not
3356 depend on the number of invocations that are active.
3359 TGSI_SEMANTIC_SUBGROUP_INVOCATION
3360 """""""""""""""""""""""""""""""""
3362 The index of the current invocation within its subgroup.
3365 TGSI_SEMANTIC_SUBGROUP_EQ_MASK
3366 """"""""""""""""""""""""""""""
3368 A bit mask of ``bit index == TGSI_SEMANTIC_SUBGROUP_INVOCATION``, i.e.
3369 ``1 << subgroup_invocation`` in arbitrary precision arithmetic.
3372 TGSI_SEMANTIC_SUBGROUP_GE_MASK
3373 """"""""""""""""""""""""""""""
3375 A bit mask of ``bit index >= TGSI_SEMANTIC_SUBGROUP_INVOCATION``, i.e.
3376 ``((1 << (subgroup_size - subgroup_invocation)) - 1) << subgroup_invocation``
3377 in arbitrary precision arithmetic.
3380 TGSI_SEMANTIC_SUBGROUP_GT_MASK
3381 """"""""""""""""""""""""""""""
3383 A bit mask of ``bit index > TGSI_SEMANTIC_SUBGROUP_INVOCATION``, i.e.
3384 ``((1 << (subgroup_size - subgroup_invocation - 1)) - 1) << (subgroup_invocation + 1)``
3385 in arbitrary precision arithmetic.
3388 TGSI_SEMANTIC_SUBGROUP_LE_MASK
3389 """"""""""""""""""""""""""""""
3391 A bit mask of ``bit index <= TGSI_SEMANTIC_SUBGROUP_INVOCATION``, i.e.
3392 ``(1 << (subgroup_invocation + 1)) - 1`` in arbitrary precision arithmetic.
3395 TGSI_SEMANTIC_SUBGROUP_LT_MASK
3396 """"""""""""""""""""""""""""""
3398 A bit mask of ``bit index < TGSI_SEMANTIC_SUBGROUP_INVOCATION``, i.e.
3399 ``(1 << subgroup_invocation) - 1`` in arbitrary precision arithmetic.
3402 Declaration Interpolate
3403 ^^^^^^^^^^^^^^^^^^^^^^^
3405 This token is only valid for fragment shader INPUT declarations.
3407 The Interpolate field specifes the way input is being interpolated by
3408 the rasteriser and is one of TGSI_INTERPOLATE_*.
3410 The Location field specifies the location inside the pixel that the
3411 interpolation should be done at, one of ``TGSI_INTERPOLATE_LOC_*``. Note that
3412 when per-sample shading is enabled, the implementation may choose to
3413 interpolate at the sample irrespective of the Location field.
3415 The CylindricalWrap bitfield specifies which register components
3416 should be subject to cylindrical wrapping when interpolating by the
3417 rasteriser. If TGSI_CYLINDRICAL_WRAP_X is set to 1, the X component
3418 should be interpolated according to cylindrical wrapping rules.
3421 Declaration Sampler View
3422 ^^^^^^^^^^^^^^^^^^^^^^^^
3424 Follows Declaration token if file is TGSI_FILE_SAMPLER_VIEW.
3426 DCL SVIEW[#], resource, type(s)
3428 Declares a shader input sampler view and assigns it to a SVIEW[#]
3431 resource can be one of BUFFER, 1D, 2D, 3D, 1DArray and 2DArray.
3433 type must be 1 or 4 entries (if specifying on a per-component
3434 level) out of UNORM, SNORM, SINT, UINT and FLOAT.
3436 For TEX\* style texture sample opcodes (as opposed to SAMPLE\* opcodes
3437 which take an explicit SVIEW[#] source register), there may be optionally
3438 SVIEW[#] declarations. In this case, the SVIEW index is implied by the
3439 SAMP index, and there must be a corresponding SVIEW[#] declaration for
3440 each SAMP[#] declaration. Drivers are free to ignore this if they wish.
3441 But note in particular that some drivers need to know the sampler type
3442 (float/int/unsigned) in order to generate the correct code, so cases
3443 where integer textures are sampled, SVIEW[#] declarations should be
3446 NOTE: It is NOT legal to mix SAMPLE\* style opcodes and TEX\* opcodes
3449 Declaration Resource
3450 ^^^^^^^^^^^^^^^^^^^^
3452 Follows Declaration token if file is TGSI_FILE_RESOURCE.
3454 DCL RES[#], resource [, WR] [, RAW]
3456 Declares a shader input resource and assigns it to a RES[#]
3459 resource can be one of BUFFER, 1D, 2D, 3D, CUBE, 1DArray and
3462 If the RAW keyword is not specified, the texture data will be
3463 subject to conversion, swizzling and scaling as required to yield
3464 the specified data type from the physical data format of the bound
3467 If the RAW keyword is specified, no channel conversion will be
3468 performed: the values read for each of the channels (X,Y,Z,W) will
3469 correspond to consecutive words in the same order and format
3470 they're found in memory. No element-to-address conversion will be
3471 performed either: the value of the provided X coordinate will be
3472 interpreted in byte units instead of texel units. The result of
3473 accessing a misaligned address is undefined.
3475 Usage of the STORE opcode is only allowed if the WR (writable) flag
3480 ^^^^^^^^^^^^^^^^^^^^^^^^
3482 Properties are general directives that apply to the whole TGSI program.
3487 Specifies the fragment shader TGSI_SEMANTIC_POSITION coordinate origin.
3488 The default value is UPPER_LEFT.
3490 If UPPER_LEFT, the position will be (0,0) at the upper left corner and
3491 increase downward and rightward.
3492 If LOWER_LEFT, the position will be (0,0) at the lower left corner and
3493 increase upward and rightward.
3495 OpenGL defaults to LOWER_LEFT, and is configurable with the
3496 GL_ARB_fragment_coord_conventions extension.
3498 DirectX 9/10 use UPPER_LEFT.
3500 FS_COORD_PIXEL_CENTER
3501 """""""""""""""""""""
3503 Specifies the fragment shader TGSI_SEMANTIC_POSITION pixel center convention.
3504 The default value is HALF_INTEGER.
3506 If HALF_INTEGER, the fractionary part of the position will be 0.5
3507 If INTEGER, the fractionary part of the position will be 0.0
3509 Note that this does not affect the set of fragments generated by
3510 rasterization, which is instead controlled by half_pixel_center in the
3513 OpenGL defaults to HALF_INTEGER, and is configurable with the
3514 GL_ARB_fragment_coord_conventions extension.
3516 DirectX 9 uses INTEGER.
3517 DirectX 10 uses HALF_INTEGER.
3519 FS_COLOR0_WRITES_ALL_CBUFS
3520 """"""""""""""""""""""""""
3521 Specifies that writes to the fragment shader color 0 are replicated to all
3522 bound cbufs. This facilitates OpenGL's fragColor output vs fragData[0] where
3523 fragData is directed to a single color buffer, but fragColor is broadcast.
3526 """"""""""""""""""""""""""
3527 If this property is set on the program bound to the shader stage before the
3528 fragment shader, user clip planes should have no effect (be disabled) even if
3529 that shader does not write to any clip distance outputs and the rasterizer's
3530 clip_plane_enable is non-zero.
3531 This property is only supported by drivers that also support shader clip
3533 This is useful for APIs that don't have UCPs and where clip distances written
3534 by a shader cannot be disabled.
3539 Specifies the number of times a geometry shader should be executed for each
3540 input primitive. Each invocation will have a different
3541 TGSI_SEMANTIC_INVOCATIONID system value set. If not specified, assumed to
3544 VS_WINDOW_SPACE_POSITION
3545 """"""""""""""""""""""""""
3546 If this property is set on the vertex shader, the TGSI_SEMANTIC_POSITION output
3547 is assumed to contain window space coordinates.
3548 Division of X,Y,Z by W and the viewport transformation are disabled, and 1/W is
3549 directly taken from the 4-th component of the shader output.
3550 Naturally, clipping is not performed on window coordinates either.
3551 The effect of this property is undefined if a geometry or tessellation shader
3557 The number of vertices written by the tessellation control shader. This
3558 effectively defines the patch input size of the tessellation evaluation shader
3564 This sets the tessellation primitive mode, one of ``PIPE_PRIM_TRIANGLES``,
3565 ``PIPE_PRIM_QUADS``, or ``PIPE_PRIM_LINES``. (Unlike in GL, there is no
3566 separate isolines settings, the regular lines is assumed to mean isolines.)
3571 This sets the spacing mode of the tessellation generator, one of
3572 ``PIPE_TESS_SPACING_*``.
3577 This sets the vertex order to be clockwise if the value is 1, or
3578 counter-clockwise if set to 0.
3583 If set to a non-zero value, this turns on point mode for the tessellator,
3584 which means that points will be generated instead of primitives.
3586 NUM_CLIPDIST_ENABLED
3587 """"""""""""""""""""
3589 How many clip distance scalar outputs are enabled.
3591 NUM_CULLDIST_ENABLED
3592 """"""""""""""""""""
3594 How many cull distance scalar outputs are enabled.
3596 FS_EARLY_DEPTH_STENCIL
3597 """"""""""""""""""""""
3599 Whether depth test, stencil test, and occlusion query should run before
3600 the fragment shader (regardless of fragment shader side effects). Corresponds
3601 to GLSL early_fragment_tests.
3606 Which shader stage will MOST LIKELY follow after this shader when the shader
3607 is bound. This is only a hint to the driver and doesn't have to be precise.
3608 Only set for VS and TES.
3610 CS_FIXED_BLOCK_WIDTH / HEIGHT / DEPTH
3611 """""""""""""""""""""""""""""""""""""
3613 Threads per block in each dimension, if known at compile time. If the block size
3614 is known all three should be at least 1. If it is unknown they should all be set
3620 The MUL TGSI operation (FP32 multiplication) will return 0 if either
3621 of the operands are equal to 0. That means that 0 * Inf = 0. This
3622 should be set the same way for an entire pipeline. Note that this
3623 applies not only to the literal MUL TGSI opcode, but all FP32
3624 multiplications implied by other operations, such as MAD, FMA, DP2,
3625 DP3, DP4, DST, LOG, LRP, and possibly others. If there is a
3626 mismatch between shaders, then it is unspecified whether this behavior
3629 FS_POST_DEPTH_COVERAGE
3630 """"""""""""""""""""""
3632 When enabled, the input for TGSI_SEMANTIC_SAMPLEMASK will exclude samples
3633 that have failed the depth/stencil tests. This is only valid when
3634 FS_EARLY_DEPTH_STENCIL is also specified.
3637 Texture Sampling and Texture Formats
3638 ------------------------------------
3640 This table shows how texture image components are returned as (x,y,z,w) tuples
3641 by TGSI texture instructions, such as :opcode:`TEX`, :opcode:`TXD`, and
3642 :opcode:`TXP`. For reference, OpenGL and Direct3D conventions are shown as
3645 +--------------------+--------------+--------------------+--------------+
3646 | Texture Components | Gallium | OpenGL | Direct3D 9 |
3647 +====================+==============+====================+==============+
3648 | R | (r, 0, 0, 1) | (r, 0, 0, 1) | (r, 1, 1, 1) |
3649 +--------------------+--------------+--------------------+--------------+
3650 | RG | (r, g, 0, 1) | (r, g, 0, 1) | (r, g, 1, 1) |
3651 +--------------------+--------------+--------------------+--------------+
3652 | RGB | (r, g, b, 1) | (r, g, b, 1) | (r, g, b, 1) |
3653 +--------------------+--------------+--------------------+--------------+
3654 | RGBA | (r, g, b, a) | (r, g, b, a) | (r, g, b, a) |
3655 +--------------------+--------------+--------------------+--------------+
3656 | A | (0, 0, 0, a) | (0, 0, 0, a) | (0, 0, 0, a) |
3657 +--------------------+--------------+--------------------+--------------+
3658 | L | (l, l, l, 1) | (l, l, l, 1) | (l, l, l, 1) |
3659 +--------------------+--------------+--------------------+--------------+
3660 | LA | (l, l, l, a) | (l, l, l, a) | (l, l, l, a) |
3661 +--------------------+--------------+--------------------+--------------+
3662 | I | (i, i, i, i) | (i, i, i, i) | N/A |
3663 +--------------------+--------------+--------------------+--------------+
3664 | UV | XXX TBD | (0, 0, 0, 1) | (u, v, 1, 1) |
3665 | | | [#envmap-bumpmap]_ | |
3666 +--------------------+--------------+--------------------+--------------+
3667 | Z | XXX TBD | (z, z, z, 1) | (0, z, 0, 1) |
3668 | | | [#depth-tex-mode]_ | |
3669 +--------------------+--------------+--------------------+--------------+
3670 | S | (s, s, s, s) | unknown | unknown |
3671 +--------------------+--------------+--------------------+--------------+
3673 .. [#envmap-bumpmap] http://www.opengl.org/registry/specs/ATI/envmap_bumpmap.txt
3674 .. [#depth-tex-mode] the default is (z, z, z, 1) but may also be (0, 0, 0, z)
3675 or (z, z, z, z) depending on the value of GL_DEPTH_TEXTURE_MODE.