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:`Double Opcodes`.
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 modifier on instructions).
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.
35 For inputs which have signed or unsigned type only the negate modifier is
42 ^^^^^^^^^^^^^^^^^^^^^^^^^
44 These opcodes are guaranteed to be available regardless of the driver being
47 .. opcode:: ARL - Address Register Load
51 dst.x = \lfloor src.x\rfloor
53 dst.y = \lfloor src.y\rfloor
55 dst.z = \lfloor src.z\rfloor
57 dst.w = \lfloor src.w\rfloor
60 .. opcode:: MOV - Move
73 .. opcode:: LIT - Light Coefficients
81 dst.z = (src.x > 0) ? max(src.y, 0)^{clamp(src.w, -128, 128))} : 0
86 .. opcode:: RCP - Reciprocal
88 This instruction replicates its result.
95 .. opcode:: RSQ - Reciprocal Square Root
97 This instruction replicates its result. The results are undefined for src <= 0.
101 dst = \frac{1}{\sqrt{src.x}}
104 .. opcode:: SQRT - Square Root
106 This instruction replicates its result. The results are undefined for src < 0.
113 .. opcode:: EXP - Approximate Exponential Base 2
117 dst.x = 2^{\lfloor src.x\rfloor}
119 dst.y = src.x - \lfloor src.x\rfloor
126 .. opcode:: LOG - Approximate Logarithm Base 2
130 dst.x = \lfloor\log_2{|src.x|}\rfloor
132 dst.y = \frac{|src.x|}{2^{\lfloor\log_2{|src.x|}\rfloor}}
134 dst.z = \log_2{|src.x|}
139 .. opcode:: MUL - Multiply
143 dst.x = src0.x \times src1.x
145 dst.y = src0.y \times src1.y
147 dst.z = src0.z \times src1.z
149 dst.w = src0.w \times src1.w
152 .. opcode:: ADD - Add
156 dst.x = src0.x + src1.x
158 dst.y = src0.y + src1.y
160 dst.z = src0.z + src1.z
162 dst.w = src0.w + src1.w
165 .. opcode:: DP3 - 3-component Dot Product
167 This instruction replicates its result.
171 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z
174 .. opcode:: DP4 - 4-component Dot Product
176 This instruction replicates its result.
180 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z + src0.w \times src1.w
183 .. opcode:: DST - Distance Vector
189 dst.y = src0.y \times src1.y
196 .. opcode:: MIN - Minimum
200 dst.x = min(src0.x, src1.x)
202 dst.y = min(src0.y, src1.y)
204 dst.z = min(src0.z, src1.z)
206 dst.w = min(src0.w, src1.w)
209 .. opcode:: MAX - Maximum
213 dst.x = max(src0.x, src1.x)
215 dst.y = max(src0.y, src1.y)
217 dst.z = max(src0.z, src1.z)
219 dst.w = max(src0.w, src1.w)
222 .. opcode:: SLT - Set On Less Than
226 dst.x = (src0.x < src1.x) ? 1 : 0
228 dst.y = (src0.y < src1.y) ? 1 : 0
230 dst.z = (src0.z < src1.z) ? 1 : 0
232 dst.w = (src0.w < src1.w) ? 1 : 0
235 .. opcode:: SGE - Set On Greater Equal Than
239 dst.x = (src0.x >= src1.x) ? 1 : 0
241 dst.y = (src0.y >= src1.y) ? 1 : 0
243 dst.z = (src0.z >= src1.z) ? 1 : 0
245 dst.w = (src0.w >= src1.w) ? 1 : 0
248 .. opcode:: MAD - Multiply And Add
252 dst.x = src0.x \times src1.x + src2.x
254 dst.y = src0.y \times src1.y + src2.y
256 dst.z = src0.z \times src1.z + src2.z
258 dst.w = src0.w \times src1.w + src2.w
261 .. opcode:: SUB - Subtract
265 dst.x = src0.x - src1.x
267 dst.y = src0.y - src1.y
269 dst.z = src0.z - src1.z
271 dst.w = src0.w - src1.w
274 .. opcode:: LRP - Linear Interpolate
278 dst.x = src0.x \times src1.x + (1 - src0.x) \times src2.x
280 dst.y = src0.y \times src1.y + (1 - src0.y) \times src2.y
282 dst.z = src0.z \times src1.z + (1 - src0.z) \times src2.z
284 dst.w = src0.w \times src1.w + (1 - src0.w) \times src2.w
287 .. opcode:: CND - Condition
291 dst.x = (src2.x > 0.5) ? src0.x : src1.x
293 dst.y = (src2.y > 0.5) ? src0.y : src1.y
295 dst.z = (src2.z > 0.5) ? src0.z : src1.z
297 dst.w = (src2.w > 0.5) ? src0.w : src1.w
300 .. opcode:: DP2A - 2-component Dot Product And Add
304 dst.x = src0.x \times src1.x + src0.y \times src1.y + src2.x
306 dst.y = src0.x \times src1.x + src0.y \times src1.y + src2.x
308 dst.z = src0.x \times src1.x + src0.y \times src1.y + src2.x
310 dst.w = src0.x \times src1.x + src0.y \times src1.y + src2.x
313 .. opcode:: FRC - Fraction
317 dst.x = src.x - \lfloor src.x\rfloor
319 dst.y = src.y - \lfloor src.y\rfloor
321 dst.z = src.z - \lfloor src.z\rfloor
323 dst.w = src.w - \lfloor src.w\rfloor
326 .. opcode:: CLAMP - Clamp
330 dst.x = clamp(src0.x, src1.x, src2.x)
332 dst.y = clamp(src0.y, src1.y, src2.y)
334 dst.z = clamp(src0.z, src1.z, src2.z)
336 dst.w = clamp(src0.w, src1.w, src2.w)
339 .. opcode:: FLR - Floor
341 This is identical to :opcode:`ARL`.
345 dst.x = \lfloor src.x\rfloor
347 dst.y = \lfloor src.y\rfloor
349 dst.z = \lfloor src.z\rfloor
351 dst.w = \lfloor src.w\rfloor
354 .. opcode:: ROUND - Round
367 .. opcode:: EX2 - Exponential Base 2
369 This instruction replicates its result.
376 .. opcode:: LG2 - Logarithm Base 2
378 This instruction replicates its result.
385 .. opcode:: POW - Power
387 This instruction replicates its result.
391 dst = src0.x^{src1.x}
393 .. opcode:: XPD - Cross Product
397 dst.x = src0.y \times src1.z - src1.y \times src0.z
399 dst.y = src0.z \times src1.x - src1.z \times src0.x
401 dst.z = src0.x \times src1.y - src1.x \times src0.y
406 .. opcode:: ABS - Absolute
419 .. opcode:: RCC - Reciprocal Clamped
421 This instruction replicates its result.
423 XXX cleanup on aisle three
427 dst = (1 / src.x) > 0 ? clamp(1 / src.x, 5.42101e-020, 1.884467e+019) : clamp(1 / src.x, -1.884467e+019, -5.42101e-020)
430 .. opcode:: DPH - Homogeneous Dot Product
432 This instruction replicates its result.
436 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z + src1.w
439 .. opcode:: COS - Cosine
441 This instruction replicates its result.
448 .. opcode:: DDX - Derivative Relative To X
452 dst.x = partialx(src.x)
454 dst.y = partialx(src.y)
456 dst.z = partialx(src.z)
458 dst.w = partialx(src.w)
461 .. opcode:: DDY - Derivative Relative To Y
465 dst.x = partialy(src.x)
467 dst.y = partialy(src.y)
469 dst.z = partialy(src.z)
471 dst.w = partialy(src.w)
474 .. opcode:: PK2H - Pack Two 16-bit Floats
479 .. opcode:: PK2US - Pack Two Unsigned 16-bit Scalars
484 .. opcode:: PK4B - Pack Four Signed 8-bit Scalars
489 .. opcode:: PK4UB - Pack Four Unsigned 8-bit Scalars
494 .. opcode:: RFL - Reflection Vector
498 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
500 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
502 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
508 Considered for removal.
511 .. opcode:: SEQ - Set On Equal
515 dst.x = (src0.x == src1.x) ? 1.0F : 0.0F
517 dst.y = (src0.y == src1.y) ? 1.0F : 0.0F
519 dst.z = (src0.z == src1.z) ? 1.0F : 0.0F
521 dst.w = (src0.w == src1.w) ? 1.0F : 0.0F
524 .. opcode:: SFL - Set On False
526 This instruction replicates its result.
534 Considered for removal.
537 .. opcode:: SGT - Set On Greater Than
541 dst.x = (src0.x > src1.x) ? 1.0F : 0.0F
543 dst.y = (src0.y > src1.y) ? 1.0F : 0.0F
545 dst.z = (src0.z > src1.z) ? 1.0F : 0.0F
547 dst.w = (src0.w > src1.w) ? 1.0F : 0.0F
550 .. opcode:: SIN - Sine
552 This instruction replicates its result.
559 .. opcode:: SLE - Set On Less Equal Than
563 dst.x = (src0.x <= src1.x) ? 1.0F : 0.0F
565 dst.y = (src0.y <= src1.y) ? 1.0F : 0.0F
567 dst.z = (src0.z <= src1.z) ? 1.0F : 0.0F
569 dst.w = (src0.w <= src1.w) ? 1.0F : 0.0F
572 .. opcode:: SNE - Set On Not Equal
576 dst.x = (src0.x != src1.x) ? 1.0F : 0.0F
578 dst.y = (src0.y != src1.y) ? 1.0F : 0.0F
580 dst.z = (src0.z != src1.z) ? 1.0F : 0.0F
582 dst.w = (src0.w != src1.w) ? 1.0F : 0.0F
585 .. opcode:: STR - Set On True
587 This instruction replicates its result.
594 .. opcode:: TEX - Texture Lookup
602 dst = texture_sample(unit, coord, bias)
604 for array textures src0.y contains the slice for 1D,
605 and src0.z contain the slice for 2D.
606 for shadow textures with no arrays, src0.z contains
608 for shadow textures with arrays, src0.z contains
609 the reference value for 1D arrays, and src0.w contains
610 the reference value for 2D arrays.
611 There is no way to pass a bias in the .w value for
612 shadow arrays, and GLSL doesn't allow this.
613 GLSL does allow cube shadows maps to take a bias value,
614 and we have to determine how this will look in TGSI.
616 .. opcode:: TXD - Texture Lookup with Derivatives
628 dst = texture_sample_deriv(unit, coord, bias, ddx, ddy)
631 .. opcode:: TXP - Projective Texture Lookup
635 coord.x = src0.x / src.w
637 coord.y = src0.y / src.w
639 coord.z = src0.z / src.w
645 dst = texture_sample(unit, coord, bias)
648 .. opcode:: UP2H - Unpack Two 16-Bit Floats
654 Considered for removal.
656 .. opcode:: UP2US - Unpack Two Unsigned 16-Bit Scalars
662 Considered for removal.
664 .. opcode:: UP4B - Unpack Four Signed 8-Bit Values
670 Considered for removal.
672 .. opcode:: UP4UB - Unpack Four Unsigned 8-Bit Scalars
678 Considered for removal.
680 .. opcode:: X2D - 2D Coordinate Transformation
684 dst.x = src0.x + src1.x \times src2.x + src1.y \times src2.y
686 dst.y = src0.y + src1.x \times src2.z + src1.y \times src2.w
688 dst.z = src0.x + src1.x \times src2.x + src1.y \times src2.y
690 dst.w = src0.y + src1.x \times src2.z + src1.y \times src2.w
694 Considered for removal.
697 .. opcode:: ARA - Address Register Add
703 Considered for removal.
705 .. opcode:: ARR - Address Register Load With Round
718 .. opcode:: SSG - Set Sign
722 dst.x = (src.x > 0) ? 1 : (src.x < 0) ? -1 : 0
724 dst.y = (src.y > 0) ? 1 : (src.y < 0) ? -1 : 0
726 dst.z = (src.z > 0) ? 1 : (src.z < 0) ? -1 : 0
728 dst.w = (src.w > 0) ? 1 : (src.w < 0) ? -1 : 0
731 .. opcode:: CMP - Compare
735 dst.x = (src0.x < 0) ? src1.x : src2.x
737 dst.y = (src0.y < 0) ? src1.y : src2.y
739 dst.z = (src0.z < 0) ? src1.z : src2.z
741 dst.w = (src0.w < 0) ? src1.w : src2.w
744 .. opcode:: KILL_IF - Conditional Discard
746 Conditional discard. Allowed in fragment shaders only.
750 if (src.x < 0 || src.y < 0 || src.z < 0 || src.w < 0)
755 .. opcode:: KILL - Discard
757 Unconditional discard. Allowed in fragment shaders only.
760 .. opcode:: SCS - Sine Cosine
773 .. opcode:: TXB - Texture Lookup With Bias
787 dst = texture_sample(unit, coord, bias)
790 .. opcode:: NRM - 3-component Vector Normalise
794 dst.x = src.x / (src.x \times src.x + src.y \times src.y + src.z \times src.z)
796 dst.y = src.y / (src.x \times src.x + src.y \times src.y + src.z \times src.z)
798 dst.z = src.z / (src.x \times src.x + src.y \times src.y + src.z \times src.z)
803 .. opcode:: DIV - Divide
807 dst.x = \frac{src0.x}{src1.x}
809 dst.y = \frac{src0.y}{src1.y}
811 dst.z = \frac{src0.z}{src1.z}
813 dst.w = \frac{src0.w}{src1.w}
816 .. opcode:: DP2 - 2-component Dot Product
818 This instruction replicates its result.
822 dst = src0.x \times src1.x + src0.y \times src1.y
825 .. opcode:: TXL - Texture Lookup With explicit LOD
839 dst = texture_sample(unit, coord, lod)
842 .. opcode:: PUSHA - Push Address Register On Stack
851 Considered for cleanup.
855 Considered for removal.
857 .. opcode:: POPA - Pop Address Register From Stack
866 Considered for cleanup.
870 Considered for removal.
873 .. opcode:: BRA - Branch
879 Considered for removal.
882 .. opcode:: CALLNZ - Subroutine Call If Not Zero
888 Considered for cleanup.
892 Considered for removal.
896 ^^^^^^^^^^^^^^^^^^^^^^^^
898 These opcodes are primarily provided for special-use computational shaders.
899 Support for these opcodes indicated by a special pipe capability bit (TBD).
901 XXX doesn't look like most of the opcodes really belong here.
903 .. opcode:: CEIL - Ceiling
907 dst.x = \lceil src.x\rceil
909 dst.y = \lceil src.y\rceil
911 dst.z = \lceil src.z\rceil
913 dst.w = \lceil src.w\rceil
916 .. opcode:: TRUNC - Truncate
929 .. opcode:: MOD - Modulus
933 dst.x = src0.x \bmod src1.x
935 dst.y = src0.y \bmod src1.y
937 dst.z = src0.z \bmod src1.z
939 dst.w = src0.w \bmod src1.w
942 .. opcode:: UARL - Integer Address Register Load
944 Moves the contents of the source register, assumed to be an integer, into the
945 destination register, which is assumed to be an address (ADDR) register.
948 .. opcode:: SAD - Sum Of Absolute Differences
952 dst.x = |src0.x - src1.x| + src2.x
954 dst.y = |src0.y - src1.y| + src2.y
956 dst.z = |src0.z - src1.z| + src2.z
958 dst.w = |src0.w - src1.w| + src2.w
961 .. opcode:: TXF - Texel Fetch (as per NV_gpu_shader4), extract a single texel
962 from a specified texture image. The source sampler may
963 not be a CUBE or SHADOW.
964 src 0 is a four-component signed integer vector used to
965 identify the single texel accessed. 3 components + level.
966 src 1 is a 3 component constant signed integer vector,
967 with each component only have a range of
968 -8..+8 (hw only seems to deal with this range, interface
969 allows for up to unsigned int).
970 TXF(uint_vec coord, int_vec offset).
973 .. opcode:: TXQ - Texture Size Query (as per NV_gpu_program4)
974 retrieve the dimensions of the texture
975 depending on the target. For 1D (width), 2D/RECT/CUBE
976 (width, height), 3D (width, height, depth),
977 1D array (width, layers), 2D array (width, height, layers)
983 dst.x = texture_width(unit, lod)
985 dst.y = texture_height(unit, lod)
987 dst.z = texture_depth(unit, lod)
991 ^^^^^^^^^^^^^^^^^^^^^^^^
992 These opcodes are used for integer operations.
993 Support for these opcodes indicated by PIPE_SHADER_CAP_INTEGERS (all of them?)
996 .. opcode:: I2F - Signed Integer To Float
998 Rounding is unspecified (round to nearest even suggested).
1002 dst.x = (float) src.x
1004 dst.y = (float) src.y
1006 dst.z = (float) src.z
1008 dst.w = (float) src.w
1011 .. opcode:: U2F - Unsigned Integer To Float
1013 Rounding is unspecified (round to nearest even suggested).
1017 dst.x = (float) src.x
1019 dst.y = (float) src.y
1021 dst.z = (float) src.z
1023 dst.w = (float) src.w
1026 .. opcode:: F2I - Float to Signed Integer
1028 Rounding is towards zero (truncate).
1029 Values outside signed range (including NaNs) produce undefined results.
1042 .. opcode:: F2U - Float to Unsigned Integer
1044 Rounding is towards zero (truncate).
1045 Values outside unsigned range (including NaNs) produce undefined results.
1049 dst.x = (unsigned) src.x
1051 dst.y = (unsigned) src.y
1053 dst.z = (unsigned) src.z
1055 dst.w = (unsigned) src.w
1058 .. opcode:: UADD - Integer Add
1060 This instruction works the same for signed and unsigned integers.
1061 The low 32bit of the result is returned.
1065 dst.x = src0.x + src1.x
1067 dst.y = src0.y + src1.y
1069 dst.z = src0.z + src1.z
1071 dst.w = src0.w + src1.w
1074 .. opcode:: UMAD - Integer Multiply And Add
1076 This instruction works the same for signed and unsigned integers.
1077 The multiplication returns the low 32bit (as does the result itself).
1081 dst.x = src0.x \times src1.x + src2.x
1083 dst.y = src0.y \times src1.y + src2.y
1085 dst.z = src0.z \times src1.z + src2.z
1087 dst.w = src0.w \times src1.w + src2.w
1090 .. opcode:: UMUL - Integer Multiply
1092 This instruction works the same for signed and unsigned integers.
1093 The low 32bit of the result is returned.
1097 dst.x = src0.x \times src1.x
1099 dst.y = src0.y \times src1.y
1101 dst.z = src0.z \times src1.z
1103 dst.w = src0.w \times src1.w
1106 .. opcode:: IDIV - Signed Integer Division
1108 TBD: behavior for division by zero.
1112 dst.x = src0.x \ src1.x
1114 dst.y = src0.y \ src1.y
1116 dst.z = src0.z \ src1.z
1118 dst.w = src0.w \ src1.w
1121 .. opcode:: UDIV - Unsigned Integer Division
1123 For division by zero, 0xffffffff is returned.
1127 dst.x = src0.x \ src1.x
1129 dst.y = src0.y \ src1.y
1131 dst.z = src0.z \ src1.z
1133 dst.w = src0.w \ src1.w
1136 .. opcode:: UMOD - Unsigned Integer Remainder
1138 If second arg is zero, 0xffffffff is returned.
1142 dst.x = src0.x \ src1.x
1144 dst.y = src0.y \ src1.y
1146 dst.z = src0.z \ src1.z
1148 dst.w = src0.w \ src1.w
1151 .. opcode:: NOT - Bitwise Not
1164 .. opcode:: AND - Bitwise And
1168 dst.x = src0.x & src1.x
1170 dst.y = src0.y & src1.y
1172 dst.z = src0.z & src1.z
1174 dst.w = src0.w & src1.w
1177 .. opcode:: OR - Bitwise Or
1181 dst.x = src0.x | src1.x
1183 dst.y = src0.y | src1.y
1185 dst.z = src0.z | src1.z
1187 dst.w = src0.w | src1.w
1190 .. opcode:: XOR - Bitwise Xor
1194 dst.x = src0.x \oplus src1.x
1196 dst.y = src0.y \oplus src1.y
1198 dst.z = src0.z \oplus src1.z
1200 dst.w = src0.w \oplus src1.w
1203 .. opcode:: IMAX - Maximum of Signed Integers
1207 dst.x = max(src0.x, src1.x)
1209 dst.y = max(src0.y, src1.y)
1211 dst.z = max(src0.z, src1.z)
1213 dst.w = max(src0.w, src1.w)
1216 .. opcode:: UMAX - Maximum of Unsigned Integers
1220 dst.x = max(src0.x, src1.x)
1222 dst.y = max(src0.y, src1.y)
1224 dst.z = max(src0.z, src1.z)
1226 dst.w = max(src0.w, src1.w)
1229 .. opcode:: IMIN - Minimum of Signed Integers
1233 dst.x = min(src0.x, src1.x)
1235 dst.y = min(src0.y, src1.y)
1237 dst.z = min(src0.z, src1.z)
1239 dst.w = min(src0.w, src1.w)
1242 .. opcode:: UMIN - Minimum of Unsigned Integers
1246 dst.x = min(src0.x, src1.x)
1248 dst.y = min(src0.y, src1.y)
1250 dst.z = min(src0.z, src1.z)
1252 dst.w = min(src0.w, src1.w)
1255 .. opcode:: SHL - Shift Left
1257 The shift count is masked with 0x1f before the shift is applied.
1261 dst.x = src0.x << (0x1f & src1.x)
1263 dst.y = src0.y << (0x1f & src1.y)
1265 dst.z = src0.z << (0x1f & src1.z)
1267 dst.w = src0.w << (0x1f & src1.w)
1270 .. opcode:: ISHR - Arithmetic Shift Right (of Signed Integer)
1272 The shift count is masked with 0x1f before the shift is applied.
1276 dst.x = src0.x >> (0x1f & src1.x)
1278 dst.y = src0.y >> (0x1f & src1.y)
1280 dst.z = src0.z >> (0x1f & src1.z)
1282 dst.w = src0.w >> (0x1f & src1.w)
1285 .. opcode:: USHR - Logical Shift Right
1287 The shift count is masked with 0x1f before the shift is applied.
1291 dst.x = src0.x >> (unsigned) (0x1f & src1.x)
1293 dst.y = src0.y >> (unsigned) (0x1f & src1.y)
1295 dst.z = src0.z >> (unsigned) (0x1f & src1.z)
1297 dst.w = src0.w >> (unsigned) (0x1f & src1.w)
1300 .. opcode:: UCMP - Integer Conditional Move
1304 dst.x = src0.x ? src1.x : src2.x
1306 dst.y = src0.y ? src1.y : src2.y
1308 dst.z = src0.z ? src1.z : src2.z
1310 dst.w = src0.w ? src1.w : src2.w
1314 .. opcode:: ISSG - Integer Set Sign
1318 dst.x = (src0.x < 0) ? -1 : (src0.x > 0) ? 1 : 0
1320 dst.y = (src0.y < 0) ? -1 : (src0.y > 0) ? 1 : 0
1322 dst.z = (src0.z < 0) ? -1 : (src0.z > 0) ? 1 : 0
1324 dst.w = (src0.w < 0) ? -1 : (src0.w > 0) ? 1 : 0
1328 .. opcode:: FSLT - Float Set On Less Than (ordered)
1330 Same comparison as SLT but returns integer instead of 1.0/0.0 float
1334 dst.x = (src0.x < src1.x) ? ~0 : 0
1336 dst.y = (src0.y < src1.y) ? ~0 : 0
1338 dst.z = (src0.z < src1.z) ? ~0 : 0
1340 dst.w = (src0.w < src1.w) ? ~0 : 0
1343 .. opcode:: ISLT - Signed Integer Set On Less Than
1347 dst.x = (src0.x < src1.x) ? ~0 : 0
1349 dst.y = (src0.y < src1.y) ? ~0 : 0
1351 dst.z = (src0.z < src1.z) ? ~0 : 0
1353 dst.w = (src0.w < src1.w) ? ~0 : 0
1356 .. opcode:: USLT - Unsigned Integer Set On Less Than
1360 dst.x = (src0.x < src1.x) ? ~0 : 0
1362 dst.y = (src0.y < src1.y) ? ~0 : 0
1364 dst.z = (src0.z < src1.z) ? ~0 : 0
1366 dst.w = (src0.w < src1.w) ? ~0 : 0
1369 .. opcode:: FSGE - Float Set On Greater Equal Than (ordered)
1371 Same comparison as SGE but returns integer instead of 1.0/0.0 float
1375 dst.x = (src0.x >= src1.x) ? ~0 : 0
1377 dst.y = (src0.y >= src1.y) ? ~0 : 0
1379 dst.z = (src0.z >= src1.z) ? ~0 : 0
1381 dst.w = (src0.w >= src1.w) ? ~0 : 0
1384 .. opcode:: ISGE - Signed Integer Set On Greater Equal Than
1388 dst.x = (src0.x >= src1.x) ? ~0 : 0
1390 dst.y = (src0.y >= src1.y) ? ~0 : 0
1392 dst.z = (src0.z >= src1.z) ? ~0 : 0
1394 dst.w = (src0.w >= src1.w) ? ~0 : 0
1397 .. opcode:: USGE - Unsigned Integer Set On Greater Equal Than
1401 dst.x = (src0.x >= src1.x) ? ~0 : 0
1403 dst.y = (src0.y >= src1.y) ? ~0 : 0
1405 dst.z = (src0.z >= src1.z) ? ~0 : 0
1407 dst.w = (src0.w >= src1.w) ? ~0 : 0
1410 .. opcode:: FSEQ - Float Set On Equal (ordered)
1412 Same comparison as SEQ but returns integer instead of 1.0/0.0 float
1416 dst.x = (src0.x == src1.x) ? ~0 : 0
1418 dst.y = (src0.y == src1.y) ? ~0 : 0
1420 dst.z = (src0.z == src1.z) ? ~0 : 0
1422 dst.w = (src0.w == src1.w) ? ~0 : 0
1425 .. opcode:: USEQ - Integer Set On Equal
1429 dst.x = (src0.x == src1.x) ? ~0 : 0
1431 dst.y = (src0.y == src1.y) ? ~0 : 0
1433 dst.z = (src0.z == src1.z) ? ~0 : 0
1435 dst.w = (src0.w == src1.w) ? ~0 : 0
1438 .. opcode:: FSNE - Float Set On Not Equal (unordered)
1440 Same comparison as SNE but returns integer instead of 1.0/0.0 float
1444 dst.x = (src0.x != src1.x) ? ~0 : 0
1446 dst.y = (src0.y != src1.y) ? ~0 : 0
1448 dst.z = (src0.z != src1.z) ? ~0 : 0
1450 dst.w = (src0.w != src1.w) ? ~0 : 0
1453 .. opcode:: USNE - Integer Set On Not Equal
1457 dst.x = (src0.x != src1.x) ? ~0 : 0
1459 dst.y = (src0.y != src1.y) ? ~0 : 0
1461 dst.z = (src0.z != src1.z) ? ~0 : 0
1463 dst.w = (src0.w != src1.w) ? ~0 : 0
1466 .. opcode:: INEG - Integer Negate
1481 .. opcode:: IABS - Integer Absolute Value
1495 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1497 These opcodes are only supported in geometry shaders; they have no meaning
1498 in any other type of shader.
1500 .. opcode:: EMIT - Emit
1502 Generate a new vertex for the current primitive using the values in the
1506 .. opcode:: ENDPRIM - End Primitive
1508 Complete the current primitive (consisting of the emitted vertices),
1509 and start a new one.
1515 These opcodes are part of :term:`GLSL`'s opcode set. Support for these
1516 opcodes is determined by a special capability bit, ``GLSL``.
1517 Some require glsl version 1.30 (UIF/BREAKC/SWITCH/CASE/DEFAULT/ENDSWITCH).
1519 .. opcode:: CAL - Subroutine Call
1525 .. opcode:: RET - Subroutine Call Return
1530 .. opcode:: CONT - Continue
1532 Unconditionally moves the point of execution to the instruction after the
1533 last bgnloop. The instruction must appear within a bgnloop/endloop.
1537 Support for CONT is determined by a special capability bit,
1538 ``TGSI_CONT_SUPPORTED``. See :ref:`Screen` for more information.
1541 .. opcode:: BGNLOOP - Begin a Loop
1543 Start a loop. Must have a matching endloop.
1546 .. opcode:: BGNSUB - Begin Subroutine
1548 Starts definition of a subroutine. Must have a matching endsub.
1551 .. opcode:: ENDLOOP - End a Loop
1553 End a loop started with bgnloop.
1556 .. opcode:: ENDSUB - End Subroutine
1558 Ends definition of a subroutine.
1561 .. opcode:: NOP - No Operation
1566 .. opcode:: BRK - Break
1568 Unconditionally moves the point of execution to the instruction after the
1569 next endloop or endswitch. The instruction must appear within a loop/endloop
1570 or switch/endswitch.
1573 .. opcode:: BREAKC - Break Conditional
1575 Conditionally moves the point of execution to the instruction after the
1576 next endloop or endswitch. The instruction must appear within a loop/endloop
1577 or switch/endswitch.
1578 Condition evaluates to true if src0.x != 0 where src0.x is interpreted
1579 as an integer register.
1583 Considered for removal as it's quite inconsistent wrt other opcodes
1584 (could emulate with UIF/BRK/ENDIF).
1587 .. opcode:: IF - Float If
1589 Start an IF ... ELSE .. ENDIF block. Condition evaluates to true if
1593 where src0.x is interpreted as a floating point register.
1596 .. opcode:: UIF - Bitwise If
1598 Start an UIF ... ELSE .. ENDIF block. Condition evaluates to true if
1602 where src0.x is interpreted as an integer register.
1605 .. opcode:: ELSE - Else
1607 Starts an else block, after an IF or UIF statement.
1610 .. opcode:: ENDIF - End If
1612 Ends an IF or UIF block.
1615 .. opcode:: SWITCH - Switch
1617 Starts a C-style switch expression. The switch consists of one or multiple
1618 CASE statements, and at most one DEFAULT statement. Execution of a statement
1619 ends when a BRK is hit, but just like in C falling through to other cases
1620 without a break is allowed. Similarly, DEFAULT label is allowed anywhere not
1621 just as last statement, and fallthrough is allowed into/from it.
1622 CASE src arguments are evaluated at bit level against the SWITCH src argument.
1627 (some instructions here)
1630 (some instructions here)
1633 (some instructions here)
1638 .. opcode:: CASE - Switch case
1640 This represents a switch case label. The src arg must be an integer immediate.
1643 .. opcode:: DEFAULT - Switch default
1645 This represents the default case in the switch, which is taken if no other
1649 .. opcode:: ENDSWITCH - End of switch
1651 Ends a switch expression.
1654 .. opcode:: NRM4 - 4-component Vector Normalise
1656 This instruction replicates its result.
1660 dst = \frac{src.x}{src.x \times src.x + src.y \times src.y + src.z \times src.z + src.w \times src.w}
1668 The double-precision opcodes reinterpret four-component vectors into
1669 two-component vectors with doubled precision in each component.
1671 Support for these opcodes is XXX undecided. :T
1673 .. opcode:: DADD - Add
1677 dst.xy = src0.xy + src1.xy
1679 dst.zw = src0.zw + src1.zw
1682 .. opcode:: DDIV - Divide
1686 dst.xy = src0.xy / src1.xy
1688 dst.zw = src0.zw / src1.zw
1690 .. opcode:: DSEQ - Set on Equal
1694 dst.xy = src0.xy == src1.xy ? 1.0F : 0.0F
1696 dst.zw = src0.zw == src1.zw ? 1.0F : 0.0F
1698 .. opcode:: DSLT - Set on Less than
1702 dst.xy = src0.xy < src1.xy ? 1.0F : 0.0F
1704 dst.zw = src0.zw < src1.zw ? 1.0F : 0.0F
1706 .. opcode:: DFRAC - Fraction
1710 dst.xy = src.xy - \lfloor src.xy\rfloor
1712 dst.zw = src.zw - \lfloor src.zw\rfloor
1715 .. opcode:: DFRACEXP - Convert Number to Fractional and Integral Components
1717 Like the ``frexp()`` routine in many math libraries, this opcode stores the
1718 exponent of its source to ``dst0``, and the significand to ``dst1``, such that
1719 :math:`dst1 \times 2^{dst0} = src` .
1723 dst0.xy = exp(src.xy)
1725 dst1.xy = frac(src.xy)
1727 dst0.zw = exp(src.zw)
1729 dst1.zw = frac(src.zw)
1731 .. opcode:: DLDEXP - Multiply Number by Integral Power of 2
1733 This opcode is the inverse of :opcode:`DFRACEXP`.
1737 dst.xy = src0.xy \times 2^{src1.xy}
1739 dst.zw = src0.zw \times 2^{src1.zw}
1741 .. opcode:: DMIN - Minimum
1745 dst.xy = min(src0.xy, src1.xy)
1747 dst.zw = min(src0.zw, src1.zw)
1749 .. opcode:: DMAX - Maximum
1753 dst.xy = max(src0.xy, src1.xy)
1755 dst.zw = max(src0.zw, src1.zw)
1757 .. opcode:: DMUL - Multiply
1761 dst.xy = src0.xy \times src1.xy
1763 dst.zw = src0.zw \times src1.zw
1766 .. opcode:: DMAD - Multiply And Add
1770 dst.xy = src0.xy \times src1.xy + src2.xy
1772 dst.zw = src0.zw \times src1.zw + src2.zw
1775 .. opcode:: DRCP - Reciprocal
1779 dst.xy = \frac{1}{src.xy}
1781 dst.zw = \frac{1}{src.zw}
1783 .. opcode:: DSQRT - Square Root
1787 dst.xy = \sqrt{src.xy}
1789 dst.zw = \sqrt{src.zw}
1792 .. _samplingopcodes:
1794 Resource Sampling Opcodes
1795 ^^^^^^^^^^^^^^^^^^^^^^^^^
1797 Those opcodes follow very closely semantics of the respective Direct3D
1798 instructions. If in doubt double check Direct3D documentation.
1799 Note that the swizzle on SVIEW (src1) determines texel swizzling
1802 .. opcode:: SAMPLE - Using provided address, sample data from the
1803 specified texture using the filtering mode identified
1804 by the gven sampler. The source data may come from
1805 any resource type other than buffers.
1806 SAMPLE dst, address, sampler_view, sampler
1808 SAMPLE TEMP[0], TEMP[1], SVIEW[0], SAMP[0]
1810 .. opcode:: SAMPLE_I - Simplified alternative to the SAMPLE instruction.
1811 Using the provided integer address, SAMPLE_I fetches data
1812 from the specified sampler view without any filtering.
1813 The source data may come from any resource type other
1815 SAMPLE_I dst, address, sampler_view
1817 SAMPLE_I TEMP[0], TEMP[1], SVIEW[0]
1818 The 'address' is specified as unsigned integers. If the
1819 'address' is out of range [0...(# texels - 1)] the
1820 result of the fetch is always 0 in all components.
1821 As such the instruction doesn't honor address wrap
1822 modes, in cases where that behavior is desirable
1823 'SAMPLE' instruction should be used.
1824 address.w always provides an unsigned integer mipmap
1825 level. If the value is out of the range then the
1826 instruction always returns 0 in all components.
1827 address.yz are ignored for buffers and 1d textures.
1828 address.z is ignored for 1d texture arrays and 2d
1830 For 1D texture arrays address.y provides the array
1831 index (also as unsigned integer). If the value is
1832 out of the range of available array indices
1833 [0... (array size - 1)] then the opcode always returns
1834 0 in all components.
1835 For 2D texture arrays address.z provides the array
1836 index, otherwise it exhibits the same behavior as in
1837 the case for 1D texture arrays.
1838 The exact semantics of the source address are presented
1840 resource type X Y Z W
1841 ------------- ------------------------
1842 PIPE_BUFFER x ignored
1843 PIPE_TEXTURE_1D x mpl
1844 PIPE_TEXTURE_2D x y mpl
1845 PIPE_TEXTURE_3D x y z mpl
1846 PIPE_TEXTURE_RECT x y mpl
1847 PIPE_TEXTURE_CUBE not allowed as source
1848 PIPE_TEXTURE_1D_ARRAY x idx mpl
1849 PIPE_TEXTURE_2D_ARRAY x y idx mpl
1851 Where 'mpl' is a mipmap level and 'idx' is the
1854 .. opcode:: SAMPLE_I_MS - Just like SAMPLE_I but allows fetch data from
1855 multi-sampled surfaces.
1856 SAMPLE_I_MS dst, address, sampler_view, sample
1858 .. opcode:: SAMPLE_B - Just like the SAMPLE instruction with the
1859 exception that an additional bias is applied to the
1860 level of detail computed as part of the instruction
1862 SAMPLE_B dst, address, sampler_view, sampler, lod_bias
1864 SAMPLE_B TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x
1866 .. opcode:: SAMPLE_C - Similar to the SAMPLE instruction but it
1867 performs a comparison filter. The operands to SAMPLE_C
1868 are identical to SAMPLE, except that there is an additional
1869 float32 operand, reference value, which must be a register
1870 with single-component, or a scalar literal.
1871 SAMPLE_C makes the hardware use the current samplers
1872 compare_func (in pipe_sampler_state) to compare
1873 reference value against the red component value for the
1874 surce resource at each texel that the currently configured
1875 texture filter covers based on the provided coordinates.
1876 SAMPLE_C dst, address, sampler_view.r, sampler, ref_value
1878 SAMPLE_C TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x
1880 .. opcode:: SAMPLE_C_LZ - Same as SAMPLE_C, but LOD is 0 and derivatives
1881 are ignored. The LZ stands for level-zero.
1882 SAMPLE_C_LZ dst, address, sampler_view.r, sampler, ref_value
1884 SAMPLE_C_LZ TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x
1887 .. opcode:: SAMPLE_D - SAMPLE_D is identical to the SAMPLE opcode except
1888 that the derivatives for the source address in the x
1889 direction and the y direction are provided by extra
1891 SAMPLE_D dst, address, sampler_view, sampler, der_x, der_y
1893 SAMPLE_D TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2], TEMP[3]
1895 .. opcode:: SAMPLE_L - SAMPLE_L is identical to the SAMPLE opcode except
1896 that the LOD is provided directly as a scalar value,
1897 representing no anisotropy.
1898 SAMPLE_L dst, address, sampler_view, sampler, explicit_lod
1900 SAMPLE_L TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x
1902 .. opcode:: GATHER4 - Gathers the four texels to be used in a bi-linear
1903 filtering operation and packs them into a single register.
1904 Only works with 2D, 2D array, cubemaps, and cubemaps arrays.
1905 For 2D textures, only the addressing modes of the sampler and
1906 the top level of any mip pyramid are used. Set W to zero.
1907 It behaves like the SAMPLE instruction, but a filtered
1908 sample is not generated. The four samples that contribute
1909 to filtering are placed into xyzw in counter-clockwise order,
1910 starting with the (u,v) texture coordinate delta at the
1911 following locations (-, +), (+, +), (+, -), (-, -), where
1912 the magnitude of the deltas are half a texel.
1915 .. opcode:: SVIEWINFO - query the dimensions of a given sampler view.
1916 dst receives width, height, depth or array size and
1917 number of mipmap levels as int4. The dst can have a writemask
1918 which will specify what info is the caller interested
1920 SVIEWINFO dst, src_mip_level, sampler_view
1922 SVIEWINFO TEMP[0], TEMP[1].x, SVIEW[0]
1923 src_mip_level is an unsigned integer scalar. If it's
1924 out of range then returns 0 for width, height and
1925 depth/array size but the total number of mipmap is
1926 still returned correctly for the given sampler view.
1927 The returned width, height and depth values are for
1928 the mipmap level selected by the src_mip_level and
1929 are in the number of texels.
1930 For 1d texture array width is in dst.x, array size
1931 is in dst.y and dst.z is 0. The number of mipmaps
1933 In contrast to d3d10 resinfo, there's no way in the
1934 tgsi instruction encoding to specify the return type
1935 (float/rcpfloat/uint), hence always using uint. Also,
1936 unlike the SAMPLE instructions, the swizzle on src1
1937 resinfo allowing swizzling dst values is ignored (due
1938 to the interaction with rcpfloat modifier which requires
1939 some swizzle handling in the state tracker anyway).
1941 .. opcode:: SAMPLE_POS - query the position of a given sample.
1942 dst receives float4 (x, y, 0, 0) indicated where the
1943 sample is located. If the resource is not a multi-sample
1944 resource and not a render target, the result is 0.
1946 .. opcode:: SAMPLE_INFO - dst receives number of samples in x.
1947 If the resource is not a multi-sample resource and
1948 not a render target, the result is 0.
1951 .. _resourceopcodes:
1953 Resource Access Opcodes
1954 ^^^^^^^^^^^^^^^^^^^^^^^
1956 .. opcode:: LOAD - Fetch data from a shader resource
1958 Syntax: ``LOAD dst, resource, address``
1960 Example: ``LOAD TEMP[0], RES[0], TEMP[1]``
1962 Using the provided integer address, LOAD fetches data
1963 from the specified buffer or texture without any
1966 The 'address' is specified as a vector of unsigned
1967 integers. If the 'address' is out of range the result
1970 Only the first mipmap level of a resource can be read
1971 from using this instruction.
1973 For 1D or 2D texture arrays, the array index is
1974 provided as an unsigned integer in address.y or
1975 address.z, respectively. address.yz are ignored for
1976 buffers and 1D textures. address.z is ignored for 1D
1977 texture arrays and 2D textures. address.w is always
1980 .. opcode:: STORE - Write data to a shader resource
1982 Syntax: ``STORE resource, address, src``
1984 Example: ``STORE RES[0], TEMP[0], TEMP[1]``
1986 Using the provided integer address, STORE writes data
1987 to the specified buffer or texture.
1989 The 'address' is specified as a vector of unsigned
1990 integers. If the 'address' is out of range the result
1993 Only the first mipmap level of a resource can be
1994 written to using this instruction.
1996 For 1D or 2D texture arrays, the array index is
1997 provided as an unsigned integer in address.y or
1998 address.z, respectively. address.yz are ignored for
1999 buffers and 1D textures. address.z is ignored for 1D
2000 texture arrays and 2D textures. address.w is always
2004 .. _threadsyncopcodes:
2006 Inter-thread synchronization opcodes
2007 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2009 These opcodes are intended for communication between threads running
2010 within the same compute grid. For now they're only valid in compute
2013 .. opcode:: MFENCE - Memory fence
2015 Syntax: ``MFENCE resource``
2017 Example: ``MFENCE RES[0]``
2019 This opcode forces strong ordering between any memory access
2020 operations that affect the specified resource. This means that
2021 previous loads and stores (and only those) will be performed and
2022 visible to other threads before the program execution continues.
2025 .. opcode:: LFENCE - Load memory fence
2027 Syntax: ``LFENCE resource``
2029 Example: ``LFENCE RES[0]``
2031 Similar to MFENCE, but it only affects the ordering of memory loads.
2034 .. opcode:: SFENCE - Store memory fence
2036 Syntax: ``SFENCE resource``
2038 Example: ``SFENCE RES[0]``
2040 Similar to MFENCE, but it only affects the ordering of memory stores.
2043 .. opcode:: BARRIER - Thread group barrier
2047 This opcode suspends the execution of the current thread until all
2048 the remaining threads in the working group reach the same point of
2049 the program. Results are unspecified if any of the remaining
2050 threads terminates or never reaches an executed BARRIER instruction.
2058 These opcodes provide atomic variants of some common arithmetic and
2059 logical operations. In this context atomicity means that another
2060 concurrent memory access operation that affects the same memory
2061 location is guaranteed to be performed strictly before or after the
2062 entire execution of the atomic operation.
2064 For the moment they're only valid in compute programs.
2066 .. opcode:: ATOMUADD - Atomic integer addition
2068 Syntax: ``ATOMUADD dst, resource, offset, src``
2070 Example: ``ATOMUADD TEMP[0], RES[0], TEMP[1], TEMP[2]``
2072 The following operation is performed atomically on each component:
2076 dst_i = resource[offset]_i
2078 resource[offset]_i = dst_i + src_i
2081 .. opcode:: ATOMXCHG - Atomic exchange
2083 Syntax: ``ATOMXCHG dst, resource, offset, src``
2085 Example: ``ATOMXCHG TEMP[0], RES[0], TEMP[1], TEMP[2]``
2087 The following operation is performed atomically on each component:
2091 dst_i = resource[offset]_i
2093 resource[offset]_i = src_i
2096 .. opcode:: ATOMCAS - Atomic compare-and-exchange
2098 Syntax: ``ATOMCAS dst, resource, offset, cmp, src``
2100 Example: ``ATOMCAS TEMP[0], RES[0], TEMP[1], TEMP[2], TEMP[3]``
2102 The following operation is performed atomically on each component:
2106 dst_i = resource[offset]_i
2108 resource[offset]_i = (dst_i == cmp_i ? src_i : dst_i)
2111 .. opcode:: ATOMAND - Atomic bitwise And
2113 Syntax: ``ATOMAND dst, resource, offset, src``
2115 Example: ``ATOMAND TEMP[0], RES[0], TEMP[1], TEMP[2]``
2117 The following operation is performed atomically on each component:
2121 dst_i = resource[offset]_i
2123 resource[offset]_i = dst_i \& src_i
2126 .. opcode:: ATOMOR - Atomic bitwise Or
2128 Syntax: ``ATOMOR dst, resource, offset, src``
2130 Example: ``ATOMOR TEMP[0], RES[0], TEMP[1], TEMP[2]``
2132 The following operation is performed atomically on each component:
2136 dst_i = resource[offset]_i
2138 resource[offset]_i = dst_i | src_i
2141 .. opcode:: ATOMXOR - Atomic bitwise Xor
2143 Syntax: ``ATOMXOR dst, resource, offset, src``
2145 Example: ``ATOMXOR TEMP[0], RES[0], TEMP[1], TEMP[2]``
2147 The following operation is performed atomically on each component:
2151 dst_i = resource[offset]_i
2153 resource[offset]_i = dst_i \oplus src_i
2156 .. opcode:: ATOMUMIN - Atomic unsigned minimum
2158 Syntax: ``ATOMUMIN dst, resource, offset, src``
2160 Example: ``ATOMUMIN TEMP[0], RES[0], TEMP[1], TEMP[2]``
2162 The following operation is performed atomically on each component:
2166 dst_i = resource[offset]_i
2168 resource[offset]_i = (dst_i < src_i ? dst_i : src_i)
2171 .. opcode:: ATOMUMAX - Atomic unsigned maximum
2173 Syntax: ``ATOMUMAX dst, resource, offset, src``
2175 Example: ``ATOMUMAX TEMP[0], RES[0], TEMP[1], TEMP[2]``
2177 The following operation is performed atomically on each component:
2181 dst_i = resource[offset]_i
2183 resource[offset]_i = (dst_i > src_i ? dst_i : src_i)
2186 .. opcode:: ATOMIMIN - Atomic signed minimum
2188 Syntax: ``ATOMIMIN dst, resource, offset, src``
2190 Example: ``ATOMIMIN TEMP[0], RES[0], TEMP[1], TEMP[2]``
2192 The following operation is performed atomically on each component:
2196 dst_i = resource[offset]_i
2198 resource[offset]_i = (dst_i < src_i ? dst_i : src_i)
2201 .. opcode:: ATOMIMAX - Atomic signed maximum
2203 Syntax: ``ATOMIMAX dst, resource, offset, src``
2205 Example: ``ATOMIMAX TEMP[0], RES[0], TEMP[1], TEMP[2]``
2207 The following operation is performed atomically on each component:
2211 dst_i = resource[offset]_i
2213 resource[offset]_i = (dst_i > src_i ? dst_i : src_i)
2217 Explanation of symbols used
2218 ------------------------------
2225 :math:`|x|` Absolute value of `x`.
2227 :math:`\lceil x \rceil` Ceiling of `x`.
2229 clamp(x,y,z) Clamp x between y and z.
2230 (x < y) ? y : (x > z) ? z : x
2232 :math:`\lfloor x\rfloor` Floor of `x`.
2234 :math:`\log_2{x}` Logarithm of `x`, base 2.
2236 max(x,y) Maximum of x and y.
2239 min(x,y) Minimum of x and y.
2242 partialx(x) Derivative of x relative to fragment's X.
2244 partialy(x) Derivative of x relative to fragment's Y.
2246 pop() Pop from stack.
2248 :math:`x^y` `x` to the power `y`.
2250 push(x) Push x on stack.
2254 trunc(x) Truncate x, i.e. drop the fraction bits.
2261 discard Discard fragment.
2265 target Label of target instruction.
2276 Declares a register that is will be referenced as an operand in Instruction
2279 File field contains register file that is being declared and is one
2282 UsageMask field specifies which of the register components can be accessed
2283 and is one of TGSI_WRITEMASK.
2285 The Local flag specifies that a given value isn't intended for
2286 subroutine parameter passing and, as a result, the implementation
2287 isn't required to give any guarantees of it being preserved across
2288 subroutine boundaries. As it's merely a compiler hint, the
2289 implementation is free to ignore it.
2291 If Dimension flag is set to 1, a Declaration Dimension token follows.
2293 If Semantic flag is set to 1, a Declaration Semantic token follows.
2295 If Interpolate flag is set to 1, a Declaration Interpolate token follows.
2297 If file is TGSI_FILE_RESOURCE, a Declaration Resource token follows.
2299 If Array flag is set to 1, a Declaration Array token follows.
2302 ^^^^^^^^^^^^^^^^^^^^^^^^
2304 Declarations can optional have an ArrayID attribute which can be referred by
2305 indirect addressing operands. An ArrayID of zero is reserved and treaded as
2306 if no ArrayID is specified.
2308 If an indirect addressing operand refers to a specific declaration by using
2309 an ArrayID only the registers in this declaration are guaranteed to be
2310 accessed, accessing any register outside this declaration results in undefined
2311 behavior. Note that for compatibility the effective index is zero-based and
2312 not relative to the specified declaration
2314 If no ArrayID is specified with an indirect addressing operand the whole
2315 register file might be accessed by this operand. This is strongly discouraged
2316 and will prevent packing of scalar/vec2 arrays and effective alias analysis.
2318 Declaration Semantic
2319 ^^^^^^^^^^^^^^^^^^^^^^^^
2321 Vertex and fragment shader input and output registers may be labeled
2322 with semantic information consisting of a name and index.
2324 Follows Declaration token if Semantic bit is set.
2326 Since its purpose is to link a shader with other stages of the pipeline,
2327 it is valid to follow only those Declaration tokens that declare a register
2328 either in INPUT or OUTPUT file.
2330 SemanticName field contains the semantic name of the register being declared.
2331 There is no default value.
2333 SemanticIndex is an optional subscript that can be used to distinguish
2334 different register declarations with the same semantic name. The default value
2337 The meanings of the individual semantic names are explained in the following
2340 TGSI_SEMANTIC_POSITION
2341 """"""""""""""""""""""
2343 For vertex shaders, TGSI_SEMANTIC_POSITION indicates the vertex shader
2344 output register which contains the homogeneous vertex position in the clip
2345 space coordinate system. After clipping, the X, Y and Z components of the
2346 vertex will be divided by the W value to get normalized device coordinates.
2348 For fragment shaders, TGSI_SEMANTIC_POSITION is used to indicate that
2349 fragment shader input contains the fragment's window position. The X
2350 component starts at zero and always increases from left to right.
2351 The Y component starts at zero and always increases but Y=0 may either
2352 indicate the top of the window or the bottom depending on the fragment
2353 coordinate origin convention (see TGSI_PROPERTY_FS_COORD_ORIGIN).
2354 The Z coordinate ranges from 0 to 1 to represent depth from the front
2355 to the back of the Z buffer. The W component contains the reciprocol
2356 of the interpolated vertex position W component.
2358 Fragment shaders may also declare an output register with
2359 TGSI_SEMANTIC_POSITION. Only the Z component is writable. This allows
2360 the fragment shader to change the fragment's Z position.
2367 For vertex shader outputs or fragment shader inputs/outputs, this
2368 label indicates that the resister contains an R,G,B,A color.
2370 Several shader inputs/outputs may contain colors so the semantic index
2371 is used to distinguish them. For example, color[0] may be the diffuse
2372 color while color[1] may be the specular color.
2374 This label is needed so that the flat/smooth shading can be applied
2375 to the right interpolants during rasterization.
2379 TGSI_SEMANTIC_BCOLOR
2380 """"""""""""""""""""
2382 Back-facing colors are only used for back-facing polygons, and are only valid
2383 in vertex shader outputs. After rasterization, all polygons are front-facing
2384 and COLOR and BCOLOR end up occupying the same slots in the fragment shader,
2385 so all BCOLORs effectively become regular COLORs in the fragment shader.
2391 Vertex shader inputs and outputs and fragment shader inputs may be
2392 labeled with TGSI_SEMANTIC_FOG to indicate that the register contains
2393 a fog coordinate in the form (F, 0, 0, 1). Typically, the fragment
2394 shader will use the fog coordinate to compute a fog blend factor which
2395 is used to blend the normal fragment color with a constant fog color.
2397 Only the first component matters when writing from the vertex shader;
2398 the driver will ensure that the coordinate is in this format when used
2399 as a fragment shader input.
2405 Vertex shader input and output registers may be labeled with
2406 TGIS_SEMANTIC_PSIZE to indicate that the register contains a point size
2407 in the form (S, 0, 0, 1). The point size controls the width or diameter
2408 of points for rasterization. This label cannot be used in fragment
2411 When using this semantic, be sure to set the appropriate state in the
2412 :ref:`rasterizer` first.
2415 TGSI_SEMANTIC_TEXCOORD
2416 """"""""""""""""""""""
2418 Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
2420 Vertex shader outputs and fragment shader inputs may be labeled with
2421 this semantic to make them replaceable by sprite coordinates via the
2422 sprite_coord_enable state in the :ref:`rasterizer`.
2423 The semantic index permitted with this semantic is limited to <= 7.
2425 If the driver does not support TEXCOORD, sprite coordinate replacement
2426 applies to inputs with the GENERIC semantic instead.
2428 The intended use case for this semantic is gl_TexCoord.
2431 TGSI_SEMANTIC_PCOORD
2432 """"""""""""""""""""
2434 Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
2436 Fragment shader inputs may be labeled with TGSI_SEMANTIC_PCOORD to indicate
2437 that the register contains sprite coordinates in the form (x, y, 0, 1), if
2438 the current primitive is a point and point sprites are enabled. Otherwise,
2439 the contents of the register are undefined.
2441 The intended use case for this semantic is gl_PointCoord.
2444 TGSI_SEMANTIC_GENERIC
2445 """""""""""""""""""""
2447 All vertex/fragment shader inputs/outputs not labeled with any other
2448 semantic label can be considered to be generic attributes. Typical
2449 uses of generic inputs/outputs are texcoords and user-defined values.
2452 TGSI_SEMANTIC_NORMAL
2453 """"""""""""""""""""
2455 Indicates that a vertex shader input is a normal vector. This is
2456 typically only used for legacy graphics APIs.
2462 This label applies to fragment shader inputs only and indicates that
2463 the register contains front/back-face information of the form (F, 0,
2464 0, 1). The first component will be positive when the fragment belongs
2465 to a front-facing polygon, and negative when the fragment belongs to a
2466 back-facing polygon.
2469 TGSI_SEMANTIC_EDGEFLAG
2470 """"""""""""""""""""""
2472 For vertex shaders, this sematic label indicates that an input or
2473 output is a boolean edge flag. The register layout is [F, x, x, x]
2474 where F is 0.0 or 1.0 and x = don't care. Normally, the vertex shader
2475 simply copies the edge flag input to the edgeflag output.
2477 Edge flags are used to control which lines or points are actually
2478 drawn when the polygon mode converts triangles/quads/polygons into
2482 TGSI_SEMANTIC_STENCIL
2483 """""""""""""""""""""
2485 For fragment shaders, this semantic label indicates that an output
2486 is a writable stencil reference value. Only the Y component is writable.
2487 This allows the fragment shader to change the fragments stencilref value.
2490 TGSI_SEMANTIC_VIEWPORT_INDEX
2491 """"""""""""""""""""""""""""
2493 For geometry shaders, this semantic label indicates that an output
2494 contains the index of the viewport (and scissor) to use.
2495 Only the X value is used.
2501 For geometry shaders, this semantic label indicates that an output
2502 contains the layer value to use for the color and depth/stencil surfaces.
2503 Only the X value is used. (Also known as rendertarget array index.)
2506 TGSI_SEMANTIC_CULLDIST
2507 """"""""""""""""""""""
2509 Used as distance to plane for performing application-defined culling
2510 of individual primitives against a plane. When components of vertex
2511 elements are given this label, these values are assumed to be a
2512 float32 signed distance to a plane. Primitives will be completely
2513 discarded if the plane distance for all of the vertices in the
2514 primitive are < 0. If a vertex has a cull distance of NaN, that
2515 vertex counts as "out" (as if its < 0);
2516 The limits on both clip and cull distances are bound
2517 by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
2518 the maximum number of components that can be used to hold the
2519 distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
2520 which specifies the maximum number of registers which can be
2521 annotated with those semantics.
2524 TGSI_SEMANTIC_CLIPDIST
2525 """"""""""""""""""""""
2527 When components of vertex elements are identified this way, these
2528 values are each assumed to be a float32 signed distance to a plane.
2529 Primitive setup only invokes rasterization on pixels for which
2530 the interpolated plane distances are >= 0. Multiple clip planes
2531 can be implemented simultaneously, by annotating multiple
2532 components of one or more vertex elements with the above specified
2533 semantic. The limits on both clip and cull distances are bound
2534 by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
2535 the maximum number of components that can be used to hold the
2536 distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
2537 which specifies the maximum number of registers which can be
2538 annotated with those semantics.
2541 Declaration Interpolate
2542 ^^^^^^^^^^^^^^^^^^^^^^^
2544 This token is only valid for fragment shader INPUT declarations.
2546 The Interpolate field specifes the way input is being interpolated by
2547 the rasteriser and is one of TGSI_INTERPOLATE_*.
2549 The CylindricalWrap bitfield specifies which register components
2550 should be subject to cylindrical wrapping when interpolating by the
2551 rasteriser. If TGSI_CYLINDRICAL_WRAP_X is set to 1, the X component
2552 should be interpolated according to cylindrical wrapping rules.
2555 Declaration Sampler View
2556 ^^^^^^^^^^^^^^^^^^^^^^^^
2558 Follows Declaration token if file is TGSI_FILE_SAMPLER_VIEW.
2560 DCL SVIEW[#], resource, type(s)
2562 Declares a shader input sampler view and assigns it to a SVIEW[#]
2565 resource can be one of BUFFER, 1D, 2D, 3D, 1DArray and 2DArray.
2567 type must be 1 or 4 entries (if specifying on a per-component
2568 level) out of UNORM, SNORM, SINT, UINT and FLOAT.
2571 Declaration Resource
2572 ^^^^^^^^^^^^^^^^^^^^
2574 Follows Declaration token if file is TGSI_FILE_RESOURCE.
2576 DCL RES[#], resource [, WR] [, RAW]
2578 Declares a shader input resource and assigns it to a RES[#]
2581 resource can be one of BUFFER, 1D, 2D, 3D, CUBE, 1DArray and
2584 If the RAW keyword is not specified, the texture data will be
2585 subject to conversion, swizzling and scaling as required to yield
2586 the specified data type from the physical data format of the bound
2589 If the RAW keyword is specified, no channel conversion will be
2590 performed: the values read for each of the channels (X,Y,Z,W) will
2591 correspond to consecutive words in the same order and format
2592 they're found in memory. No element-to-address conversion will be
2593 performed either: the value of the provided X coordinate will be
2594 interpreted in byte units instead of texel units. The result of
2595 accessing a misaligned address is undefined.
2597 Usage of the STORE opcode is only allowed if the WR (writable) flag
2602 ^^^^^^^^^^^^^^^^^^^^^^^^
2605 Properties are general directives that apply to the whole TGSI program.
2610 Specifies the fragment shader TGSI_SEMANTIC_POSITION coordinate origin.
2611 The default value is UPPER_LEFT.
2613 If UPPER_LEFT, the position will be (0,0) at the upper left corner and
2614 increase downward and rightward.
2615 If LOWER_LEFT, the position will be (0,0) at the lower left corner and
2616 increase upward and rightward.
2618 OpenGL defaults to LOWER_LEFT, and is configurable with the
2619 GL_ARB_fragment_coord_conventions extension.
2621 DirectX 9/10 use UPPER_LEFT.
2623 FS_COORD_PIXEL_CENTER
2624 """""""""""""""""""""
2626 Specifies the fragment shader TGSI_SEMANTIC_POSITION pixel center convention.
2627 The default value is HALF_INTEGER.
2629 If HALF_INTEGER, the fractionary part of the position will be 0.5
2630 If INTEGER, the fractionary part of the position will be 0.0
2632 Note that this does not affect the set of fragments generated by
2633 rasterization, which is instead controlled by half_pixel_center in the
2636 OpenGL defaults to HALF_INTEGER, and is configurable with the
2637 GL_ARB_fragment_coord_conventions extension.
2639 DirectX 9 uses INTEGER.
2640 DirectX 10 uses HALF_INTEGER.
2642 FS_COLOR0_WRITES_ALL_CBUFS
2643 """"""""""""""""""""""""""
2644 Specifies that writes to the fragment shader color 0 are replicated to all
2645 bound cbufs. This facilitates OpenGL's fragColor output vs fragData[0] where
2646 fragData is directed to a single color buffer, but fragColor is broadcast.
2649 """"""""""""""""""""""""""
2650 If this property is set on the program bound to the shader stage before the
2651 fragment shader, user clip planes should have no effect (be disabled) even if
2652 that shader does not write to any clip distance outputs and the rasterizer's
2653 clip_plane_enable is non-zero.
2654 This property is only supported by drivers that also support shader clip
2656 This is useful for APIs that don't have UCPs and where clip distances written
2657 by a shader cannot be disabled.
2660 Texture Sampling and Texture Formats
2661 ------------------------------------
2663 This table shows how texture image components are returned as (x,y,z,w) tuples
2664 by TGSI texture instructions, such as :opcode:`TEX`, :opcode:`TXD`, and
2665 :opcode:`TXP`. For reference, OpenGL and Direct3D conventions are shown as
2668 +--------------------+--------------+--------------------+--------------+
2669 | Texture Components | Gallium | OpenGL | Direct3D 9 |
2670 +====================+==============+====================+==============+
2671 | R | (r, 0, 0, 1) | (r, 0, 0, 1) | (r, 1, 1, 1) |
2672 +--------------------+--------------+--------------------+--------------+
2673 | RG | (r, g, 0, 1) | (r, g, 0, 1) | (r, g, 1, 1) |
2674 +--------------------+--------------+--------------------+--------------+
2675 | RGB | (r, g, b, 1) | (r, g, b, 1) | (r, g, b, 1) |
2676 +--------------------+--------------+--------------------+--------------+
2677 | RGBA | (r, g, b, a) | (r, g, b, a) | (r, g, b, a) |
2678 +--------------------+--------------+--------------------+--------------+
2679 | A | (0, 0, 0, a) | (0, 0, 0, a) | (0, 0, 0, a) |
2680 +--------------------+--------------+--------------------+--------------+
2681 | L | (l, l, l, 1) | (l, l, l, 1) | (l, l, l, 1) |
2682 +--------------------+--------------+--------------------+--------------+
2683 | LA | (l, l, l, a) | (l, l, l, a) | (l, l, l, a) |
2684 +--------------------+--------------+--------------------+--------------+
2685 | I | (i, i, i, i) | (i, i, i, i) | N/A |
2686 +--------------------+--------------+--------------------+--------------+
2687 | UV | XXX TBD | (0, 0, 0, 1) | (u, v, 1, 1) |
2688 | | | [#envmap-bumpmap]_ | |
2689 +--------------------+--------------+--------------------+--------------+
2690 | Z | XXX TBD | (z, z, z, 1) | (0, z, 0, 1) |
2691 | | | [#depth-tex-mode]_ | |
2692 +--------------------+--------------+--------------------+--------------+
2693 | S | (s, s, s, s) | unknown | unknown |
2694 +--------------------+--------------+--------------------+--------------+
2696 .. [#envmap-bumpmap] http://www.opengl.org/registry/specs/ATI/envmap_bumpmap.txt
2697 .. [#depth-tex-mode] the default is (z, z, z, 1) but may also be (0, 0, 0, z)
2698 or (z, z, z, z) depending on the value of GL_DEPTH_TEXTURE_MODE.