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 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
78 dst.y &= max(src.x, 0) \\
79 dst.z &= (src.x > 0) ? max(src.y, 0)^{clamp(src.w, -128, 128))} : 0 \\
83 .. opcode:: RCP - Reciprocal
85 This instruction replicates its result.
92 .. opcode:: RSQ - Reciprocal Square Root
94 This instruction replicates its result. The results are undefined for src <= 0.
98 dst = \frac{1}{\sqrt{src.x}}
101 .. opcode:: SQRT - Square Root
103 This instruction replicates its result. The results are undefined for src < 0.
110 .. opcode:: EXP - Approximate Exponential Base 2
114 dst.x &= 2^{\lfloor src.x\rfloor} \\
115 dst.y &= src.x - \lfloor src.x\rfloor \\
116 dst.z &= 2^{src.x} \\
120 .. opcode:: LOG - Approximate Logarithm Base 2
124 dst.x &= \lfloor\log_2{|src.x|}\rfloor \\
125 dst.y &= \frac{|src.x|}{2^{\lfloor\log_2{|src.x|}\rfloor}} \\
126 dst.z &= \log_2{|src.x|} \\
130 .. opcode:: MUL - Multiply
134 dst.x = src0.x \times src1.x
136 dst.y = src0.y \times src1.y
138 dst.z = src0.z \times src1.z
140 dst.w = src0.w \times src1.w
143 .. opcode:: ADD - Add
147 dst.x = src0.x + src1.x
149 dst.y = src0.y + src1.y
151 dst.z = src0.z + src1.z
153 dst.w = src0.w + src1.w
156 .. opcode:: DP3 - 3-component Dot Product
158 This instruction replicates its result.
162 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z
165 .. opcode:: DP4 - 4-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 + src0.w \times src1.w
174 .. opcode:: DST - Distance Vector
179 dst.y &= src0.y \times src1.y\\
184 .. opcode:: MIN - Minimum
188 dst.x = min(src0.x, src1.x)
190 dst.y = min(src0.y, src1.y)
192 dst.z = min(src0.z, src1.z)
194 dst.w = min(src0.w, src1.w)
197 .. opcode:: MAX - Maximum
201 dst.x = max(src0.x, src1.x)
203 dst.y = max(src0.y, src1.y)
205 dst.z = max(src0.z, src1.z)
207 dst.w = max(src0.w, src1.w)
210 .. opcode:: SLT - Set On Less Than
214 dst.x = (src0.x < src1.x) ? 1.0F : 0.0F
216 dst.y = (src0.y < src1.y) ? 1.0F : 0.0F
218 dst.z = (src0.z < src1.z) ? 1.0F : 0.0F
220 dst.w = (src0.w < src1.w) ? 1.0F : 0.0F
223 .. opcode:: SGE - Set On Greater Equal Than
227 dst.x = (src0.x >= src1.x) ? 1.0F : 0.0F
229 dst.y = (src0.y >= src1.y) ? 1.0F : 0.0F
231 dst.z = (src0.z >= src1.z) ? 1.0F : 0.0F
233 dst.w = (src0.w >= src1.w) ? 1.0F : 0.0F
236 .. opcode:: MAD - Multiply And Add
240 dst.x = src0.x \times src1.x + src2.x
242 dst.y = src0.y \times src1.y + src2.y
244 dst.z = src0.z \times src1.z + src2.z
246 dst.w = src0.w \times src1.w + src2.w
249 .. opcode:: SUB - Subtract
253 dst.x = src0.x - src1.x
255 dst.y = src0.y - src1.y
257 dst.z = src0.z - src1.z
259 dst.w = src0.w - src1.w
262 .. opcode:: LRP - Linear Interpolate
266 dst.x = src0.x \times src1.x + (1 - src0.x) \times src2.x
268 dst.y = src0.y \times src1.y + (1 - src0.y) \times src2.y
270 dst.z = src0.z \times src1.z + (1 - src0.z) \times src2.z
272 dst.w = src0.w \times src1.w + (1 - src0.w) \times src2.w
275 .. opcode:: CND - Condition
279 dst.x = (src2.x > 0.5) ? src0.x : src1.x
281 dst.y = (src2.y > 0.5) ? src0.y : src1.y
283 dst.z = (src2.z > 0.5) ? src0.z : src1.z
285 dst.w = (src2.w > 0.5) ? src0.w : src1.w
288 .. opcode:: DP2A - 2-component Dot Product And Add
292 dst.x = src0.x \times src1.x + src0.y \times src1.y + src2.x
294 dst.y = src0.x \times src1.x + src0.y \times src1.y + src2.x
296 dst.z = src0.x \times src1.x + src0.y \times src1.y + src2.x
298 dst.w = src0.x \times src1.x + src0.y \times src1.y + src2.x
301 .. opcode:: FRC - Fraction
305 dst.x = src.x - \lfloor src.x\rfloor
307 dst.y = src.y - \lfloor src.y\rfloor
309 dst.z = src.z - \lfloor src.z\rfloor
311 dst.w = src.w - \lfloor src.w\rfloor
314 .. opcode:: CLAMP - Clamp
318 dst.x = clamp(src0.x, src1.x, src2.x)
320 dst.y = clamp(src0.y, src1.y, src2.y)
322 dst.z = clamp(src0.z, src1.z, src2.z)
324 dst.w = clamp(src0.w, src1.w, src2.w)
327 .. opcode:: FLR - Floor
329 This is identical to :opcode:`ARL`.
333 dst.x = \lfloor src.x\rfloor
335 dst.y = \lfloor src.y\rfloor
337 dst.z = \lfloor src.z\rfloor
339 dst.w = \lfloor src.w\rfloor
342 .. opcode:: ROUND - Round
355 .. opcode:: EX2 - Exponential Base 2
357 This instruction replicates its result.
364 .. opcode:: LG2 - Logarithm Base 2
366 This instruction replicates its result.
373 .. opcode:: POW - Power
375 This instruction replicates its result.
379 dst = src0.x^{src1.x}
381 .. opcode:: XPD - Cross Product
385 dst.x = src0.y \times src1.z - src1.y \times src0.z
387 dst.y = src0.z \times src1.x - src1.z \times src0.x
389 dst.z = src0.x \times src1.y - src1.x \times src0.y
394 .. opcode:: ABS - Absolute
407 .. opcode:: RCC - Reciprocal Clamped
409 This instruction replicates its result.
411 XXX cleanup on aisle three
415 dst = (1 / src.x) > 0 ? clamp(1 / src.x, 5.42101e-020, 1.84467e+019) : clamp(1 / src.x, -1.84467e+019, -5.42101e-020)
418 .. opcode:: DPH - Homogeneous Dot Product
420 This instruction replicates its result.
424 dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z + src1.w
427 .. opcode:: COS - Cosine
429 This instruction replicates its result.
436 .. opcode:: DDX - Derivative Relative To X
440 dst.x = partialx(src.x)
442 dst.y = partialx(src.y)
444 dst.z = partialx(src.z)
446 dst.w = partialx(src.w)
449 .. opcode:: DDY - Derivative Relative To Y
453 dst.x = partialy(src.x)
455 dst.y = partialy(src.y)
457 dst.z = partialy(src.z)
459 dst.w = partialy(src.w)
462 .. opcode:: PK2H - Pack Two 16-bit Floats
467 .. opcode:: PK2US - Pack Two Unsigned 16-bit Scalars
472 .. opcode:: PK4B - Pack Four Signed 8-bit Scalars
477 .. opcode:: PK4UB - Pack Four Unsigned 8-bit Scalars
482 .. opcode:: RFL - Reflection Vector
486 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
488 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
490 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
496 Considered for removal.
499 .. opcode:: SEQ - Set On Equal
503 dst.x = (src0.x == src1.x) ? 1.0F : 0.0F
505 dst.y = (src0.y == src1.y) ? 1.0F : 0.0F
507 dst.z = (src0.z == src1.z) ? 1.0F : 0.0F
509 dst.w = (src0.w == src1.w) ? 1.0F : 0.0F
512 .. opcode:: SFL - Set On False
514 This instruction replicates its result.
522 Considered for removal.
525 .. opcode:: SGT - Set On Greater Than
529 dst.x = (src0.x > src1.x) ? 1.0F : 0.0F
531 dst.y = (src0.y > src1.y) ? 1.0F : 0.0F
533 dst.z = (src0.z > src1.z) ? 1.0F : 0.0F
535 dst.w = (src0.w > src1.w) ? 1.0F : 0.0F
538 .. opcode:: SIN - Sine
540 This instruction replicates its result.
547 .. opcode:: SLE - Set On Less Equal Than
551 dst.x = (src0.x <= src1.x) ? 1.0F : 0.0F
553 dst.y = (src0.y <= src1.y) ? 1.0F : 0.0F
555 dst.z = (src0.z <= src1.z) ? 1.0F : 0.0F
557 dst.w = (src0.w <= src1.w) ? 1.0F : 0.0F
560 .. opcode:: SNE - Set On Not Equal
564 dst.x = (src0.x != src1.x) ? 1.0F : 0.0F
566 dst.y = (src0.y != src1.y) ? 1.0F : 0.0F
568 dst.z = (src0.z != src1.z) ? 1.0F : 0.0F
570 dst.w = (src0.w != src1.w) ? 1.0F : 0.0F
573 .. opcode:: STR - Set On True
575 This instruction replicates its result.
582 .. opcode:: TEX - Texture Lookup
584 for array textures src0.y contains the slice for 1D,
585 and src0.z contain the slice for 2D.
587 for shadow textures with no arrays, src0.z contains
590 for shadow textures with arrays, src0.z contains
591 the reference value for 1D arrays, and src0.w contains
592 the reference value for 2D arrays.
594 There is no way to pass a bias in the .w value for
595 shadow arrays, and GLSL doesn't allow this.
596 GLSL does allow cube shadows maps to take a bias value,
597 and we have to determine how this will look in TGSI.
605 dst = texture\_sample(unit, coord, bias)
607 .. opcode:: TXD - Texture Lookup with Derivatives
619 dst = texture\_sample\_deriv(unit, coord, bias, ddx, ddy)
622 .. opcode:: TXP - Projective Texture Lookup
626 coord.x = src0.x / src.w
628 coord.y = src0.y / src.w
630 coord.z = src0.z / src.w
636 dst = texture\_sample(unit, coord, bias)
639 .. opcode:: UP2H - Unpack Two 16-Bit Floats
645 Considered for removal.
647 .. opcode:: UP2US - Unpack Two Unsigned 16-Bit Scalars
653 Considered for removal.
655 .. opcode:: UP4B - Unpack Four Signed 8-Bit Values
661 Considered for removal.
663 .. opcode:: UP4UB - Unpack Four Unsigned 8-Bit Scalars
669 Considered for removal.
671 .. opcode:: X2D - 2D Coordinate Transformation
675 dst.x = src0.x + src1.x \times src2.x + src1.y \times src2.y
677 dst.y = src0.y + src1.x \times src2.z + src1.y \times src2.w
679 dst.z = src0.x + src1.x \times src2.x + src1.y \times src2.y
681 dst.w = src0.y + src1.x \times src2.z + src1.y \times src2.w
685 Considered for removal.
688 .. opcode:: ARA - Address Register Add
694 Considered for removal.
696 .. opcode:: ARR - Address Register Load With Round
709 .. opcode:: SSG - Set Sign
713 dst.x = (src.x > 0) ? 1 : (src.x < 0) ? -1 : 0
715 dst.y = (src.y > 0) ? 1 : (src.y < 0) ? -1 : 0
717 dst.z = (src.z > 0) ? 1 : (src.z < 0) ? -1 : 0
719 dst.w = (src.w > 0) ? 1 : (src.w < 0) ? -1 : 0
722 .. opcode:: CMP - Compare
726 dst.x = (src0.x < 0) ? src1.x : src2.x
728 dst.y = (src0.y < 0) ? src1.y : src2.y
730 dst.z = (src0.z < 0) ? src1.z : src2.z
732 dst.w = (src0.w < 0) ? src1.w : src2.w
735 .. opcode:: KILL_IF - Conditional Discard
737 Conditional discard. Allowed in fragment shaders only.
741 if (src.x < 0 || src.y < 0 || src.z < 0 || src.w < 0)
746 .. opcode:: KILL - Discard
748 Unconditional discard. Allowed in fragment shaders only.
751 .. opcode:: SCS - Sine Cosine
764 .. opcode:: TXB - Texture Lookup With Bias
778 dst = texture\_sample(unit, coord, bias)
781 .. opcode:: NRM - 3-component Vector Normalise
785 dst.x = src.x / (src.x \times src.x + src.y \times src.y + src.z \times src.z)
787 dst.y = src.y / (src.x \times src.x + src.y \times src.y + src.z \times src.z)
789 dst.z = src.z / (src.x \times src.x + src.y \times src.y + src.z \times src.z)
794 .. opcode:: DIV - Divide
798 dst.x = \frac{src0.x}{src1.x}
800 dst.y = \frac{src0.y}{src1.y}
802 dst.z = \frac{src0.z}{src1.z}
804 dst.w = \frac{src0.w}{src1.w}
807 .. opcode:: DP2 - 2-component Dot Product
809 This instruction replicates its result.
813 dst = src0.x \times src1.x + src0.y \times src1.y
816 .. opcode:: TXL - Texture Lookup With explicit LOD
830 dst = texture\_sample(unit, coord, lod)
833 .. opcode:: PUSHA - Push Address Register On Stack
842 Considered for cleanup.
846 Considered for removal.
848 .. opcode:: POPA - Pop Address Register From Stack
857 Considered for cleanup.
861 Considered for removal.
864 .. opcode:: BRA - Branch
870 Considered for removal.
873 .. opcode:: CALLNZ - Subroutine Call If Not Zero
879 Considered for cleanup.
883 Considered for removal.
887 ^^^^^^^^^^^^^^^^^^^^^^^^
889 These opcodes are primarily provided for special-use computational shaders.
890 Support for these opcodes indicated by a special pipe capability bit (TBD).
892 XXX doesn't look like most of the opcodes really belong here.
894 .. opcode:: CEIL - Ceiling
898 dst.x = \lceil src.x\rceil
900 dst.y = \lceil src.y\rceil
902 dst.z = \lceil src.z\rceil
904 dst.w = \lceil src.w\rceil
907 .. opcode:: TRUNC - Truncate
920 .. opcode:: MOD - Modulus
924 dst.x = src0.x \bmod src1.x
926 dst.y = src0.y \bmod src1.y
928 dst.z = src0.z \bmod src1.z
930 dst.w = src0.w \bmod src1.w
933 .. opcode:: UARL - Integer Address Register Load
935 Moves the contents of the source register, assumed to be an integer, into the
936 destination register, which is assumed to be an address (ADDR) register.
939 .. opcode:: SAD - Sum Of Absolute Differences
943 dst.x = |src0.x - src1.x| + src2.x
945 dst.y = |src0.y - src1.y| + src2.y
947 dst.z = |src0.z - src1.z| + src2.z
949 dst.w = |src0.w - src1.w| + src2.w
952 .. opcode:: TXF - Texel Fetch
954 As per NV_gpu_shader4, extract a single texel from a specified texture
955 image. The source sampler may not be a CUBE or SHADOW. src 0 is a
956 four-component signed integer vector used to identify the single texel
957 accessed. 3 components + level. src 1 is a 3 component constant signed
958 integer vector, with each component only have a range of -8..+8 (hw only
959 seems to deal with this range, interface allows for up to unsigned int).
960 TXF(uint_vec coord, int_vec offset).
963 .. opcode:: TXQ - Texture Size Query
965 As per NV_gpu_program4, retrieve the dimensions of the texture depending on
966 the target. For 1D (width), 2D/RECT/CUBE (width, height), 3D (width, height,
967 depth), 1D array (width, layers), 2D array (width, height, layers)
973 dst.x = texture\_width(unit, lod)
975 dst.y = texture\_height(unit, lod)
977 dst.z = texture\_depth(unit, lod)
979 .. opcode:: TG4 - Texture Gather
981 As per ARB_texture_gather, gathers the four texels to be used in a bi-linear
982 filtering operation and packs them into a single register. Only works with
983 2D, 2D array, cubemaps, and cubemaps arrays. For 2D textures, only the
984 addressing modes of the sampler and the top level of any mip pyramid are
985 used. Set W to zero. It behaves like the TEX instruction, but a filtered
986 sample is not generated. The four samples that contribute to filtering are
987 placed into xyzw in clockwise order, starting with the (u,v) texture
988 coordinate delta at the following locations (-, +), (+, +), (+, -), (-, -),
989 where the magnitude of the deltas are half a texel.
991 PIPE_CAP_TEXTURE_SM5 enhances this instruction to support shadow per-sample
992 depth compares, single component selection, and a non-constant offset. It
993 doesn't allow support for the GL independent offset to get i0,j0. This would
994 require another CAP is hw can do it natively. For now we lower that before
1003 dst = texture\_gather4 (unit, coord, component)
1005 (with SM5 - cube array shadow)
1013 dst = texture\_gather (uint, coord, compare)
1015 .. opcode:: LODQ - level of detail query
1017 Compute the LOD information that the texture pipe would use to access the
1018 texture. The Y component contains the computed LOD lambda_prime. The X
1019 component contains the LOD that will be accessed, based on min/max lod's
1026 dst.xy = lodq(uint, coord);
1029 ^^^^^^^^^^^^^^^^^^^^^^^^
1030 These opcodes are used for integer operations.
1031 Support for these opcodes indicated by PIPE_SHADER_CAP_INTEGERS (all of them?)
1034 .. opcode:: I2F - Signed Integer To Float
1036 Rounding is unspecified (round to nearest even suggested).
1040 dst.x = (float) src.x
1042 dst.y = (float) src.y
1044 dst.z = (float) src.z
1046 dst.w = (float) src.w
1049 .. opcode:: U2F - Unsigned Integer To Float
1051 Rounding is unspecified (round to nearest even suggested).
1055 dst.x = (float) src.x
1057 dst.y = (float) src.y
1059 dst.z = (float) src.z
1061 dst.w = (float) src.w
1064 .. opcode:: F2I - Float to Signed Integer
1066 Rounding is towards zero (truncate).
1067 Values outside signed range (including NaNs) produce undefined results.
1080 .. opcode:: F2U - Float to Unsigned Integer
1082 Rounding is towards zero (truncate).
1083 Values outside unsigned range (including NaNs) produce undefined results.
1087 dst.x = (unsigned) src.x
1089 dst.y = (unsigned) src.y
1091 dst.z = (unsigned) src.z
1093 dst.w = (unsigned) src.w
1096 .. opcode:: UADD - Integer Add
1098 This instruction works the same for signed and unsigned integers.
1099 The low 32bit of the result is returned.
1103 dst.x = src0.x + src1.x
1105 dst.y = src0.y + src1.y
1107 dst.z = src0.z + src1.z
1109 dst.w = src0.w + src1.w
1112 .. opcode:: UMAD - Integer Multiply And Add
1114 This instruction works the same for signed and unsigned integers.
1115 The multiplication returns the low 32bit (as does the result itself).
1119 dst.x = src0.x \times src1.x + src2.x
1121 dst.y = src0.y \times src1.y + src2.y
1123 dst.z = src0.z \times src1.z + src2.z
1125 dst.w = src0.w \times src1.w + src2.w
1128 .. opcode:: UMUL - Integer Multiply
1130 This instruction works the same for signed and unsigned integers.
1131 The low 32bit of the result is returned.
1135 dst.x = src0.x \times src1.x
1137 dst.y = src0.y \times src1.y
1139 dst.z = src0.z \times src1.z
1141 dst.w = src0.w \times src1.w
1144 .. opcode:: IMUL_HI - Signed Integer Multiply High Bits
1146 The high 32bits of the multiplication of 2 signed integers are returned.
1150 dst.x = (src0.x \times src1.x) >> 32
1152 dst.y = (src0.y \times src1.y) >> 32
1154 dst.z = (src0.z \times src1.z) >> 32
1156 dst.w = (src0.w \times src1.w) >> 32
1159 .. opcode:: UMUL_HI - Unsigned Integer Multiply High Bits
1161 The high 32bits of the multiplication of 2 unsigned integers are returned.
1165 dst.x = (src0.x \times src1.x) >> 32
1167 dst.y = (src0.y \times src1.y) >> 32
1169 dst.z = (src0.z \times src1.z) >> 32
1171 dst.w = (src0.w \times src1.w) >> 32
1174 .. opcode:: IDIV - Signed Integer Division
1176 TBD: behavior for division by zero.
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:: UDIV - Unsigned Integer Division
1191 For division by zero, 0xffffffff is returned.
1195 dst.x = src0.x \ src1.x
1197 dst.y = src0.y \ src1.y
1199 dst.z = src0.z \ src1.z
1201 dst.w = src0.w \ src1.w
1204 .. opcode:: UMOD - Unsigned Integer Remainder
1206 If second arg is zero, 0xffffffff is returned.
1210 dst.x = src0.x \ src1.x
1212 dst.y = src0.y \ src1.y
1214 dst.z = src0.z \ src1.z
1216 dst.w = src0.w \ src1.w
1219 .. opcode:: NOT - Bitwise Not
1232 .. opcode:: AND - Bitwise And
1236 dst.x = src0.x \& src1.x
1238 dst.y = src0.y \& src1.y
1240 dst.z = src0.z \& src1.z
1242 dst.w = src0.w \& src1.w
1245 .. opcode:: OR - Bitwise Or
1249 dst.x = src0.x | src1.x
1251 dst.y = src0.y | src1.y
1253 dst.z = src0.z | src1.z
1255 dst.w = src0.w | src1.w
1258 .. opcode:: XOR - Bitwise Xor
1262 dst.x = src0.x \oplus src1.x
1264 dst.y = src0.y \oplus src1.y
1266 dst.z = src0.z \oplus src1.z
1268 dst.w = src0.w \oplus src1.w
1271 .. opcode:: IMAX - Maximum of Signed Integers
1275 dst.x = max(src0.x, src1.x)
1277 dst.y = max(src0.y, src1.y)
1279 dst.z = max(src0.z, src1.z)
1281 dst.w = max(src0.w, src1.w)
1284 .. opcode:: UMAX - Maximum of Unsigned Integers
1288 dst.x = max(src0.x, src1.x)
1290 dst.y = max(src0.y, src1.y)
1292 dst.z = max(src0.z, src1.z)
1294 dst.w = max(src0.w, src1.w)
1297 .. opcode:: IMIN - Minimum of Signed Integers
1301 dst.x = min(src0.x, src1.x)
1303 dst.y = min(src0.y, src1.y)
1305 dst.z = min(src0.z, src1.z)
1307 dst.w = min(src0.w, src1.w)
1310 .. opcode:: UMIN - Minimum of Unsigned Integers
1314 dst.x = min(src0.x, src1.x)
1316 dst.y = min(src0.y, src1.y)
1318 dst.z = min(src0.z, src1.z)
1320 dst.w = min(src0.w, src1.w)
1323 .. opcode:: SHL - Shift Left
1325 The shift count is masked with 0x1f before the shift is applied.
1329 dst.x = src0.x << (0x1f \& src1.x)
1331 dst.y = src0.y << (0x1f \& src1.y)
1333 dst.z = src0.z << (0x1f \& src1.z)
1335 dst.w = src0.w << (0x1f \& src1.w)
1338 .. opcode:: ISHR - Arithmetic Shift Right (of Signed Integer)
1340 The shift count is masked with 0x1f before the shift is applied.
1344 dst.x = src0.x >> (0x1f \& src1.x)
1346 dst.y = src0.y >> (0x1f \& src1.y)
1348 dst.z = src0.z >> (0x1f \& src1.z)
1350 dst.w = src0.w >> (0x1f \& src1.w)
1353 .. opcode:: USHR - Logical Shift Right
1355 The shift count is masked with 0x1f before the shift is applied.
1359 dst.x = src0.x >> (unsigned) (0x1f \& src1.x)
1361 dst.y = src0.y >> (unsigned) (0x1f \& src1.y)
1363 dst.z = src0.z >> (unsigned) (0x1f \& src1.z)
1365 dst.w = src0.w >> (unsigned) (0x1f \& src1.w)
1368 .. opcode:: UCMP - Integer Conditional Move
1372 dst.x = src0.x ? src1.x : src2.x
1374 dst.y = src0.y ? src1.y : src2.y
1376 dst.z = src0.z ? src1.z : src2.z
1378 dst.w = src0.w ? src1.w : src2.w
1382 .. opcode:: ISSG - Integer Set Sign
1386 dst.x = (src0.x < 0) ? -1 : (src0.x > 0) ? 1 : 0
1388 dst.y = (src0.y < 0) ? -1 : (src0.y > 0) ? 1 : 0
1390 dst.z = (src0.z < 0) ? -1 : (src0.z > 0) ? 1 : 0
1392 dst.w = (src0.w < 0) ? -1 : (src0.w > 0) ? 1 : 0
1396 .. opcode:: FSLT - Float Set On Less Than (ordered)
1398 Same comparison as SLT but returns integer instead of 1.0/0.0 float
1402 dst.x = (src0.x < src1.x) ? \sim 0 : 0
1404 dst.y = (src0.y < src1.y) ? \sim 0 : 0
1406 dst.z = (src0.z < src1.z) ? \sim 0 : 0
1408 dst.w = (src0.w < src1.w) ? \sim 0 : 0
1411 .. opcode:: ISLT - Signed Integer Set On Less Than
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:: USLT - Unsigned Integer Set On Less Than
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:: FSGE - Float Set On Greater Equal Than (ordered)
1439 Same comparison as SGE 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:: ISGE - Signed Integer Set On Greater Equal Than
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:: USGE - Unsigned Integer Set On Greater Equal Than
1469 dst.x = (src0.x >= src1.x) ? \sim 0 : 0
1471 dst.y = (src0.y >= src1.y) ? \sim 0 : 0
1473 dst.z = (src0.z >= src1.z) ? \sim 0 : 0
1475 dst.w = (src0.w >= src1.w) ? \sim 0 : 0
1478 .. opcode:: FSEQ - Float Set On Equal (ordered)
1480 Same comparison as SEQ but returns integer instead of 1.0/0.0 float
1484 dst.x = (src0.x == src1.x) ? \sim 0 : 0
1486 dst.y = (src0.y == src1.y) ? \sim 0 : 0
1488 dst.z = (src0.z == src1.z) ? \sim 0 : 0
1490 dst.w = (src0.w == src1.w) ? \sim 0 : 0
1493 .. opcode:: USEQ - Integer Set On Equal
1497 dst.x = (src0.x == src1.x) ? \sim 0 : 0
1499 dst.y = (src0.y == src1.y) ? \sim 0 : 0
1501 dst.z = (src0.z == src1.z) ? \sim 0 : 0
1503 dst.w = (src0.w == src1.w) ? \sim 0 : 0
1506 .. opcode:: FSNE - Float Set On Not Equal (unordered)
1508 Same comparison as SNE but returns integer instead of 1.0/0.0 float
1512 dst.x = (src0.x != src1.x) ? \sim 0 : 0
1514 dst.y = (src0.y != src1.y) ? \sim 0 : 0
1516 dst.z = (src0.z != src1.z) ? \sim 0 : 0
1518 dst.w = (src0.w != src1.w) ? \sim 0 : 0
1521 .. opcode:: USNE - Integer Set On Not Equal
1525 dst.x = (src0.x != src1.x) ? \sim 0 : 0
1527 dst.y = (src0.y != src1.y) ? \sim 0 : 0
1529 dst.z = (src0.z != src1.z) ? \sim 0 : 0
1531 dst.w = (src0.w != src1.w) ? \sim 0 : 0
1534 .. opcode:: INEG - Integer Negate
1549 .. opcode:: IABS - Integer Absolute Value
1563 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1565 These opcodes are only supported in geometry shaders; they have no meaning
1566 in any other type of shader.
1568 .. opcode:: EMIT - Emit
1570 Generate a new vertex for the current primitive using the values in the
1574 .. opcode:: ENDPRIM - End Primitive
1576 Complete the current primitive (consisting of the emitted vertices),
1577 and start a new one.
1583 These opcodes are part of :term:`GLSL`'s opcode set. Support for these
1584 opcodes is determined by a special capability bit, ``GLSL``.
1585 Some require glsl version 1.30 (UIF/BREAKC/SWITCH/CASE/DEFAULT/ENDSWITCH).
1587 .. opcode:: CAL - Subroutine Call
1593 .. opcode:: RET - Subroutine Call Return
1598 .. opcode:: CONT - Continue
1600 Unconditionally moves the point of execution to the instruction after the
1601 last bgnloop. The instruction must appear within a bgnloop/endloop.
1605 Support for CONT is determined by a special capability bit,
1606 ``TGSI_CONT_SUPPORTED``. See :ref:`Screen` for more information.
1609 .. opcode:: BGNLOOP - Begin a Loop
1611 Start a loop. Must have a matching endloop.
1614 .. opcode:: BGNSUB - Begin Subroutine
1616 Starts definition of a subroutine. Must have a matching endsub.
1619 .. opcode:: ENDLOOP - End a Loop
1621 End a loop started with bgnloop.
1624 .. opcode:: ENDSUB - End Subroutine
1626 Ends definition of a subroutine.
1629 .. opcode:: NOP - No Operation
1634 .. opcode:: BRK - Break
1636 Unconditionally moves the point of execution to the instruction after the
1637 next endloop or endswitch. The instruction must appear within a loop/endloop
1638 or switch/endswitch.
1641 .. opcode:: BREAKC - Break Conditional
1643 Conditionally moves the point of execution to the instruction after the
1644 next endloop or endswitch. The instruction must appear within a loop/endloop
1645 or switch/endswitch.
1646 Condition evaluates to true if src0.x != 0 where src0.x is interpreted
1647 as an integer register.
1651 Considered for removal as it's quite inconsistent wrt other opcodes
1652 (could emulate with UIF/BRK/ENDIF).
1655 .. opcode:: IF - Float If
1657 Start an IF ... ELSE .. ENDIF block. Condition evaluates to true if
1661 where src0.x is interpreted as a floating point register.
1664 .. opcode:: UIF - Bitwise If
1666 Start an UIF ... ELSE .. ENDIF block. Condition evaluates to true if
1670 where src0.x is interpreted as an integer register.
1673 .. opcode:: ELSE - Else
1675 Starts an else block, after an IF or UIF statement.
1678 .. opcode:: ENDIF - End If
1680 Ends an IF or UIF block.
1683 .. opcode:: SWITCH - Switch
1685 Starts a C-style switch expression. The switch consists of one or multiple
1686 CASE statements, and at most one DEFAULT statement. Execution of a statement
1687 ends when a BRK is hit, but just like in C falling through to other cases
1688 without a break is allowed. Similarly, DEFAULT label is allowed anywhere not
1689 just as last statement, and fallthrough is allowed into/from it.
1690 CASE src arguments are evaluated at bit level against the SWITCH src argument.
1696 (some instructions here)
1699 (some instructions here)
1702 (some instructions here)
1707 .. opcode:: CASE - Switch case
1709 This represents a switch case label. The src arg must be an integer immediate.
1712 .. opcode:: DEFAULT - Switch default
1714 This represents the default case in the switch, which is taken if no other
1718 .. opcode:: ENDSWITCH - End of switch
1720 Ends a switch expression.
1723 .. opcode:: NRM4 - 4-component Vector Normalise
1725 This instruction replicates its result.
1729 dst = \frac{src.x}{src.x \times src.x + src.y \times src.y + src.z \times src.z + src.w \times src.w}
1737 The double-precision opcodes reinterpret four-component vectors into
1738 two-component vectors with doubled precision in each component.
1740 Support for these opcodes is XXX undecided. :T
1742 .. opcode:: DADD - Add
1746 dst.xy = src0.xy + src1.xy
1748 dst.zw = src0.zw + src1.zw
1751 .. opcode:: DDIV - Divide
1755 dst.xy = src0.xy / src1.xy
1757 dst.zw = src0.zw / src1.zw
1759 .. opcode:: DSEQ - Set on Equal
1763 dst.xy = src0.xy == src1.xy ? 1.0F : 0.0F
1765 dst.zw = src0.zw == src1.zw ? 1.0F : 0.0F
1767 .. opcode:: DSLT - Set on Less than
1771 dst.xy = src0.xy < src1.xy ? 1.0F : 0.0F
1773 dst.zw = src0.zw < src1.zw ? 1.0F : 0.0F
1775 .. opcode:: DFRAC - Fraction
1779 dst.xy = src.xy - \lfloor src.xy\rfloor
1781 dst.zw = src.zw - \lfloor src.zw\rfloor
1784 .. opcode:: DFRACEXP - Convert Number to Fractional and Integral Components
1786 Like the ``frexp()`` routine in many math libraries, this opcode stores the
1787 exponent of its source to ``dst0``, and the significand to ``dst1``, such that
1788 :math:`dst1 \times 2^{dst0} = src` .
1792 dst0.xy = exp(src.xy)
1794 dst1.xy = frac(src.xy)
1796 dst0.zw = exp(src.zw)
1798 dst1.zw = frac(src.zw)
1800 .. opcode:: DLDEXP - Multiply Number by Integral Power of 2
1802 This opcode is the inverse of :opcode:`DFRACEXP`.
1806 dst.xy = src0.xy \times 2^{src1.xy}
1808 dst.zw = src0.zw \times 2^{src1.zw}
1810 .. opcode:: DMIN - Minimum
1814 dst.xy = min(src0.xy, src1.xy)
1816 dst.zw = min(src0.zw, src1.zw)
1818 .. opcode:: DMAX - Maximum
1822 dst.xy = max(src0.xy, src1.xy)
1824 dst.zw = max(src0.zw, src1.zw)
1826 .. opcode:: DMUL - Multiply
1830 dst.xy = src0.xy \times src1.xy
1832 dst.zw = src0.zw \times src1.zw
1835 .. opcode:: DMAD - Multiply And Add
1839 dst.xy = src0.xy \times src1.xy + src2.xy
1841 dst.zw = src0.zw \times src1.zw + src2.zw
1844 .. opcode:: DRCP - Reciprocal
1848 dst.xy = \frac{1}{src.xy}
1850 dst.zw = \frac{1}{src.zw}
1852 .. opcode:: DSQRT - Square Root
1856 dst.xy = \sqrt{src.xy}
1858 dst.zw = \sqrt{src.zw}
1861 .. _samplingopcodes:
1863 Resource Sampling Opcodes
1864 ^^^^^^^^^^^^^^^^^^^^^^^^^
1866 Those opcodes follow very closely semantics of the respective Direct3D
1867 instructions. If in doubt double check Direct3D documentation.
1868 Note that the swizzle on SVIEW (src1) determines texel swizzling
1873 Using provided address, sample data from the specified texture using the
1874 filtering mode identified by the gven sampler. The source data may come from
1875 any resource type other than buffers.
1877 Syntax: ``SAMPLE dst, address, sampler_view, sampler``
1879 Example: ``SAMPLE TEMP[0], TEMP[1], SVIEW[0], SAMP[0]``
1881 .. opcode:: SAMPLE_I
1883 Simplified alternative to the SAMPLE instruction. Using the provided
1884 integer address, SAMPLE_I fetches data from the specified sampler view
1885 without any filtering. The source data may come from any resource type
1888 Syntax: ``SAMPLE_I dst, address, sampler_view``
1890 Example: ``SAMPLE_I TEMP[0], TEMP[1], SVIEW[0]``
1892 The 'address' is specified as unsigned integers. If the 'address' is out of
1893 range [0...(# texels - 1)] the result of the fetch is always 0 in all
1894 components. As such the instruction doesn't honor address wrap modes, in
1895 cases where that behavior is desirable 'SAMPLE' instruction should be used.
1896 address.w always provides an unsigned integer mipmap level. If the value is
1897 out of the range then the instruction always returns 0 in all components.
1898 address.yz are ignored for buffers and 1d textures. address.z is ignored
1899 for 1d texture arrays and 2d textures.
1901 For 1D texture arrays address.y provides the array index (also as unsigned
1902 integer). If the value is out of the range of available array indices
1903 [0... (array size - 1)] then the opcode always returns 0 in all components.
1904 For 2D texture arrays address.z provides the array index, otherwise it
1905 exhibits the same behavior as in the case for 1D texture arrays. The exact
1906 semantics of the source address are presented in the table below:
1908 +---------------------------+----+-----+-----+---------+
1909 | resource type | X | Y | Z | W |
1910 +===========================+====+=====+=====+=========+
1911 | ``PIPE_BUFFER`` | x | | | ignored |
1912 +---------------------------+----+-----+-----+---------+
1913 | ``PIPE_TEXTURE_1D`` | x | | | mpl |
1914 +---------------------------+----+-----+-----+---------+
1915 | ``PIPE_TEXTURE_2D`` | x | y | | mpl |
1916 +---------------------------+----+-----+-----+---------+
1917 | ``PIPE_TEXTURE_3D`` | x | y | z | mpl |
1918 +---------------------------+----+-----+-----+---------+
1919 | ``PIPE_TEXTURE_RECT`` | x | y | | mpl |
1920 +---------------------------+----+-----+-----+---------+
1921 | ``PIPE_TEXTURE_CUBE`` | not allowed as source |
1922 +---------------------------+----+-----+-----+---------+
1923 | ``PIPE_TEXTURE_1D_ARRAY`` | x | idx | | mpl |
1924 +---------------------------+----+-----+-----+---------+
1925 | ``PIPE_TEXTURE_2D_ARRAY`` | x | y | idx | mpl |
1926 +---------------------------+----+-----+-----+---------+
1928 Where 'mpl' is a mipmap level and 'idx' is the array index.
1930 .. opcode:: SAMPLE_I_MS
1932 Just like SAMPLE_I but allows fetch data from multi-sampled surfaces.
1934 Syntax: ``SAMPLE_I_MS dst, address, sampler_view, sample``
1936 .. opcode:: SAMPLE_B
1938 Just like the SAMPLE instruction with the exception that an additional bias
1939 is applied to the level of detail computed as part of the instruction
1942 Syntax: ``SAMPLE_B dst, address, sampler_view, sampler, lod_bias``
1944 Example: ``SAMPLE_B TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x``
1946 .. opcode:: SAMPLE_C
1948 Similar to the SAMPLE instruction but it performs a comparison filter. The
1949 operands to SAMPLE_C are identical to SAMPLE, except that there is an
1950 additional float32 operand, reference value, which must be a register with
1951 single-component, or a scalar literal. SAMPLE_C makes the hardware use the
1952 current samplers compare_func (in pipe_sampler_state) to compare reference
1953 value against the red component value for the surce resource at each texel
1954 that the currently configured texture filter covers based on the provided
1957 Syntax: ``SAMPLE_C dst, address, sampler_view.r, sampler, ref_value``
1959 Example: ``SAMPLE_C TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x``
1961 .. opcode:: SAMPLE_C_LZ
1963 Same as SAMPLE_C, but LOD is 0 and derivatives are ignored. The LZ stands
1966 Syntax: ``SAMPLE_C_LZ dst, address, sampler_view.r, sampler, ref_value``
1968 Example: ``SAMPLE_C_LZ TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x``
1971 .. opcode:: SAMPLE_D
1973 SAMPLE_D is identical to the SAMPLE opcode except that the derivatives for
1974 the source address in the x direction and the y direction are provided by
1977 Syntax: ``SAMPLE_D dst, address, sampler_view, sampler, der_x, der_y``
1979 Example: ``SAMPLE_D TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2], TEMP[3]``
1981 .. opcode:: SAMPLE_L
1983 SAMPLE_L is identical to the SAMPLE opcode except that the LOD is provided
1984 directly as a scalar value, representing no anisotropy.
1986 Syntax: ``SAMPLE_L dst, address, sampler_view, sampler, explicit_lod``
1988 Example: ``SAMPLE_L TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x``
1992 Gathers the four texels to be used in a bi-linear filtering operation and
1993 packs them into a single register. Only works with 2D, 2D array, cubemaps,
1994 and cubemaps arrays. For 2D textures, only the addressing modes of the
1995 sampler and the top level of any mip pyramid are used. Set W to zero. It
1996 behaves like the SAMPLE instruction, but a filtered sample is not
1997 generated. The four samples that contribute to filtering are placed into
1998 xyzw in counter-clockwise order, starting with the (u,v) texture coordinate
1999 delta at the following locations (-, +), (+, +), (+, -), (-, -), where the
2000 magnitude of the deltas are half a texel.
2003 .. opcode:: SVIEWINFO
2005 Query the dimensions of a given sampler view. dst receives width, height,
2006 depth or array size and number of mipmap levels as int4. The dst can have a
2007 writemask which will specify what info is the caller interested in.
2009 Syntax: ``SVIEWINFO dst, src_mip_level, sampler_view``
2011 Example: ``SVIEWINFO TEMP[0], TEMP[1].x, SVIEW[0]``
2013 src_mip_level is an unsigned integer scalar. If it's out of range then
2014 returns 0 for width, height and depth/array size but the total number of
2015 mipmap is still returned correctly for the given sampler view. The returned
2016 width, height and depth values are for the mipmap level selected by the
2017 src_mip_level and are in the number of texels. For 1d texture array width
2018 is in dst.x, array size is in dst.y and dst.z is 0. The number of mipmaps is
2019 still in dst.w. In contrast to d3d10 resinfo, there's no way in the tgsi
2020 instruction encoding to specify the return type (float/rcpfloat/uint), hence
2021 always using uint. Also, unlike the SAMPLE instructions, the swizzle on src1
2022 resinfo allowing swizzling dst values is ignored (due to the interaction
2023 with rcpfloat modifier which requires some swizzle handling in the state
2026 .. opcode:: SAMPLE_POS
2028 Query the position of a given sample. dst receives float4 (x, y, 0, 0)
2029 indicated where the sample is located. If the resource is not a multi-sample
2030 resource and not a render target, the result is 0.
2032 .. opcode:: SAMPLE_INFO
2034 dst receives number of samples in x. If the resource is not a multi-sample
2035 resource and not a render target, the result is 0.
2038 .. _resourceopcodes:
2040 Resource Access Opcodes
2041 ^^^^^^^^^^^^^^^^^^^^^^^
2043 .. opcode:: LOAD - Fetch data from a shader resource
2045 Syntax: ``LOAD dst, resource, address``
2047 Example: ``LOAD TEMP[0], RES[0], TEMP[1]``
2049 Using the provided integer address, LOAD fetches data
2050 from the specified buffer or texture without any
2053 The 'address' is specified as a vector of unsigned
2054 integers. If the 'address' is out of range the result
2057 Only the first mipmap level of a resource can be read
2058 from using this instruction.
2060 For 1D or 2D texture arrays, the array index is
2061 provided as an unsigned integer in address.y or
2062 address.z, respectively. address.yz are ignored for
2063 buffers and 1D textures. address.z is ignored for 1D
2064 texture arrays and 2D textures. address.w is always
2067 .. opcode:: STORE - Write data to a shader resource
2069 Syntax: ``STORE resource, address, src``
2071 Example: ``STORE RES[0], TEMP[0], TEMP[1]``
2073 Using the provided integer address, STORE writes data
2074 to the specified buffer or texture.
2076 The 'address' is specified as a vector of unsigned
2077 integers. If the 'address' is out of range the result
2080 Only the first mipmap level of a resource can be
2081 written to using this instruction.
2083 For 1D or 2D texture arrays, the array index is
2084 provided as an unsigned integer in address.y or
2085 address.z, respectively. address.yz are ignored for
2086 buffers and 1D textures. address.z is ignored for 1D
2087 texture arrays and 2D textures. address.w is always
2091 .. _threadsyncopcodes:
2093 Inter-thread synchronization opcodes
2094 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2096 These opcodes are intended for communication between threads running
2097 within the same compute grid. For now they're only valid in compute
2100 .. opcode:: MFENCE - Memory fence
2102 Syntax: ``MFENCE resource``
2104 Example: ``MFENCE RES[0]``
2106 This opcode forces strong ordering between any memory access
2107 operations that affect the specified resource. This means that
2108 previous loads and stores (and only those) will be performed and
2109 visible to other threads before the program execution continues.
2112 .. opcode:: LFENCE - Load memory fence
2114 Syntax: ``LFENCE resource``
2116 Example: ``LFENCE RES[0]``
2118 Similar to MFENCE, but it only affects the ordering of memory loads.
2121 .. opcode:: SFENCE - Store memory fence
2123 Syntax: ``SFENCE resource``
2125 Example: ``SFENCE RES[0]``
2127 Similar to MFENCE, but it only affects the ordering of memory stores.
2130 .. opcode:: BARRIER - Thread group barrier
2134 This opcode suspends the execution of the current thread until all
2135 the remaining threads in the working group reach the same point of
2136 the program. Results are unspecified if any of the remaining
2137 threads terminates or never reaches an executed BARRIER instruction.
2145 These opcodes provide atomic variants of some common arithmetic and
2146 logical operations. In this context atomicity means that another
2147 concurrent memory access operation that affects the same memory
2148 location is guaranteed to be performed strictly before or after the
2149 entire execution of the atomic operation.
2151 For the moment they're only valid in compute programs.
2153 .. opcode:: ATOMUADD - Atomic integer addition
2155 Syntax: ``ATOMUADD dst, resource, offset, src``
2157 Example: ``ATOMUADD TEMP[0], RES[0], TEMP[1], TEMP[2]``
2159 The following operation is performed atomically on each component:
2163 dst_i = resource[offset]_i
2165 resource[offset]_i = dst_i + src_i
2168 .. opcode:: ATOMXCHG - Atomic exchange
2170 Syntax: ``ATOMXCHG dst, resource, offset, src``
2172 Example: ``ATOMXCHG TEMP[0], RES[0], TEMP[1], TEMP[2]``
2174 The following operation is performed atomically on each component:
2178 dst_i = resource[offset]_i
2180 resource[offset]_i = src_i
2183 .. opcode:: ATOMCAS - Atomic compare-and-exchange
2185 Syntax: ``ATOMCAS dst, resource, offset, cmp, src``
2187 Example: ``ATOMCAS TEMP[0], RES[0], TEMP[1], TEMP[2], TEMP[3]``
2189 The following operation is performed atomically on each component:
2193 dst_i = resource[offset]_i
2195 resource[offset]_i = (dst_i == cmp_i ? src_i : dst_i)
2198 .. opcode:: ATOMAND - Atomic bitwise And
2200 Syntax: ``ATOMAND dst, resource, offset, src``
2202 Example: ``ATOMAND TEMP[0], RES[0], TEMP[1], TEMP[2]``
2204 The following operation is performed atomically on each component:
2208 dst_i = resource[offset]_i
2210 resource[offset]_i = dst_i \& src_i
2213 .. opcode:: ATOMOR - Atomic bitwise Or
2215 Syntax: ``ATOMOR dst, resource, offset, src``
2217 Example: ``ATOMOR TEMP[0], RES[0], TEMP[1], TEMP[2]``
2219 The following operation is performed atomically on each component:
2223 dst_i = resource[offset]_i
2225 resource[offset]_i = dst_i | src_i
2228 .. opcode:: ATOMXOR - Atomic bitwise Xor
2230 Syntax: ``ATOMXOR dst, resource, offset, src``
2232 Example: ``ATOMXOR TEMP[0], RES[0], TEMP[1], TEMP[2]``
2234 The following operation is performed atomically on each component:
2238 dst_i = resource[offset]_i
2240 resource[offset]_i = dst_i \oplus src_i
2243 .. opcode:: ATOMUMIN - Atomic unsigned minimum
2245 Syntax: ``ATOMUMIN dst, resource, offset, src``
2247 Example: ``ATOMUMIN TEMP[0], RES[0], TEMP[1], TEMP[2]``
2249 The following operation is performed atomically on each component:
2253 dst_i = resource[offset]_i
2255 resource[offset]_i = (dst_i < src_i ? dst_i : src_i)
2258 .. opcode:: ATOMUMAX - Atomic unsigned maximum
2260 Syntax: ``ATOMUMAX dst, resource, offset, src``
2262 Example: ``ATOMUMAX TEMP[0], RES[0], TEMP[1], TEMP[2]``
2264 The following operation is performed atomically on each component:
2268 dst_i = resource[offset]_i
2270 resource[offset]_i = (dst_i > src_i ? dst_i : src_i)
2273 .. opcode:: ATOMIMIN - Atomic signed minimum
2275 Syntax: ``ATOMIMIN dst, resource, offset, src``
2277 Example: ``ATOMIMIN TEMP[0], RES[0], TEMP[1], TEMP[2]``
2279 The following operation is performed atomically on each component:
2283 dst_i = resource[offset]_i
2285 resource[offset]_i = (dst_i < src_i ? dst_i : src_i)
2288 .. opcode:: ATOMIMAX - Atomic signed maximum
2290 Syntax: ``ATOMIMAX dst, resource, offset, src``
2292 Example: ``ATOMIMAX TEMP[0], RES[0], TEMP[1], TEMP[2]``
2294 The following operation is performed atomically on each component:
2298 dst_i = resource[offset]_i
2300 resource[offset]_i = (dst_i > src_i ? dst_i : src_i)
2304 Explanation of symbols used
2305 ------------------------------
2312 :math:`|x|` Absolute value of `x`.
2314 :math:`\lceil x \rceil` Ceiling of `x`.
2316 clamp(x,y,z) Clamp x between y and z.
2317 (x < y) ? y : (x > z) ? z : x
2319 :math:`\lfloor x\rfloor` Floor of `x`.
2321 :math:`\log_2{x}` Logarithm of `x`, base 2.
2323 max(x,y) Maximum of x and y.
2326 min(x,y) Minimum of x and y.
2329 partialx(x) Derivative of x relative to fragment's X.
2331 partialy(x) Derivative of x relative to fragment's Y.
2333 pop() Pop from stack.
2335 :math:`x^y` `x` to the power `y`.
2337 push(x) Push x on stack.
2341 trunc(x) Truncate x, i.e. drop the fraction bits.
2348 discard Discard fragment.
2352 target Label of target instruction.
2363 Declares a register that is will be referenced as an operand in Instruction
2366 File field contains register file that is being declared and is one
2369 UsageMask field specifies which of the register components can be accessed
2370 and is one of TGSI_WRITEMASK.
2372 The Local flag specifies that a given value isn't intended for
2373 subroutine parameter passing and, as a result, the implementation
2374 isn't required to give any guarantees of it being preserved across
2375 subroutine boundaries. As it's merely a compiler hint, the
2376 implementation is free to ignore it.
2378 If Dimension flag is set to 1, a Declaration Dimension token follows.
2380 If Semantic flag is set to 1, a Declaration Semantic token follows.
2382 If Interpolate flag is set to 1, a Declaration Interpolate token follows.
2384 If file is TGSI_FILE_RESOURCE, a Declaration Resource token follows.
2386 If Array flag is set to 1, a Declaration Array token follows.
2389 ^^^^^^^^^^^^^^^^^^^^^^^^
2391 Declarations can optional have an ArrayID attribute which can be referred by
2392 indirect addressing operands. An ArrayID of zero is reserved and treaded as
2393 if no ArrayID is specified.
2395 If an indirect addressing operand refers to a specific declaration by using
2396 an ArrayID only the registers in this declaration are guaranteed to be
2397 accessed, accessing any register outside this declaration results in undefined
2398 behavior. Note that for compatibility the effective index is zero-based and
2399 not relative to the specified declaration
2401 If no ArrayID is specified with an indirect addressing operand the whole
2402 register file might be accessed by this operand. This is strongly discouraged
2403 and will prevent packing of scalar/vec2 arrays and effective alias analysis.
2405 Declaration Semantic
2406 ^^^^^^^^^^^^^^^^^^^^^^^^
2408 Vertex and fragment shader input and output registers may be labeled
2409 with semantic information consisting of a name and index.
2411 Follows Declaration token if Semantic bit is set.
2413 Since its purpose is to link a shader with other stages of the pipeline,
2414 it is valid to follow only those Declaration tokens that declare a register
2415 either in INPUT or OUTPUT file.
2417 SemanticName field contains the semantic name of the register being declared.
2418 There is no default value.
2420 SemanticIndex is an optional subscript that can be used to distinguish
2421 different register declarations with the same semantic name. The default value
2424 The meanings of the individual semantic names are explained in the following
2427 TGSI_SEMANTIC_POSITION
2428 """"""""""""""""""""""
2430 For vertex shaders, TGSI_SEMANTIC_POSITION indicates the vertex shader
2431 output register which contains the homogeneous vertex position in the clip
2432 space coordinate system. After clipping, the X, Y and Z components of the
2433 vertex will be divided by the W value to get normalized device coordinates.
2435 For fragment shaders, TGSI_SEMANTIC_POSITION is used to indicate that
2436 fragment shader input contains the fragment's window position. The X
2437 component starts at zero and always increases from left to right.
2438 The Y component starts at zero and always increases but Y=0 may either
2439 indicate the top of the window or the bottom depending on the fragment
2440 coordinate origin convention (see TGSI_PROPERTY_FS_COORD_ORIGIN).
2441 The Z coordinate ranges from 0 to 1 to represent depth from the front
2442 to the back of the Z buffer. The W component contains the reciprocol
2443 of the interpolated vertex position W component.
2445 Fragment shaders may also declare an output register with
2446 TGSI_SEMANTIC_POSITION. Only the Z component is writable. This allows
2447 the fragment shader to change the fragment's Z position.
2454 For vertex shader outputs or fragment shader inputs/outputs, this
2455 label indicates that the resister contains an R,G,B,A color.
2457 Several shader inputs/outputs may contain colors so the semantic index
2458 is used to distinguish them. For example, color[0] may be the diffuse
2459 color while color[1] may be the specular color.
2461 This label is needed so that the flat/smooth shading can be applied
2462 to the right interpolants during rasterization.
2466 TGSI_SEMANTIC_BCOLOR
2467 """"""""""""""""""""
2469 Back-facing colors are only used for back-facing polygons, and are only valid
2470 in vertex shader outputs. After rasterization, all polygons are front-facing
2471 and COLOR and BCOLOR end up occupying the same slots in the fragment shader,
2472 so all BCOLORs effectively become regular COLORs in the fragment shader.
2478 Vertex shader inputs and outputs and fragment shader inputs may be
2479 labeled with TGSI_SEMANTIC_FOG to indicate that the register contains
2480 a fog coordinate. Typically, the fragment shader will use the fog coordinate
2481 to compute a fog blend factor which is used to blend the normal fragment color
2482 with a constant fog color. But fog coord really is just an ordinary vec4
2483 register like regular semantics.
2489 Vertex shader input and output registers may be labeled with
2490 TGIS_SEMANTIC_PSIZE to indicate that the register contains a point size
2491 in the form (S, 0, 0, 1). The point size controls the width or diameter
2492 of points for rasterization. This label cannot be used in fragment
2495 When using this semantic, be sure to set the appropriate state in the
2496 :ref:`rasterizer` first.
2499 TGSI_SEMANTIC_TEXCOORD
2500 """"""""""""""""""""""
2502 Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
2504 Vertex shader outputs and fragment shader inputs may be labeled with
2505 this semantic to make them replaceable by sprite coordinates via the
2506 sprite_coord_enable state in the :ref:`rasterizer`.
2507 The semantic index permitted with this semantic is limited to <= 7.
2509 If the driver does not support TEXCOORD, sprite coordinate replacement
2510 applies to inputs with the GENERIC semantic instead.
2512 The intended use case for this semantic is gl_TexCoord.
2515 TGSI_SEMANTIC_PCOORD
2516 """"""""""""""""""""
2518 Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
2520 Fragment shader inputs may be labeled with TGSI_SEMANTIC_PCOORD to indicate
2521 that the register contains sprite coordinates in the form (x, y, 0, 1), if
2522 the current primitive is a point and point sprites are enabled. Otherwise,
2523 the contents of the register are undefined.
2525 The intended use case for this semantic is gl_PointCoord.
2528 TGSI_SEMANTIC_GENERIC
2529 """""""""""""""""""""
2531 All vertex/fragment shader inputs/outputs not labeled with any other
2532 semantic label can be considered to be generic attributes. Typical
2533 uses of generic inputs/outputs are texcoords and user-defined values.
2536 TGSI_SEMANTIC_NORMAL
2537 """"""""""""""""""""
2539 Indicates that a vertex shader input is a normal vector. This is
2540 typically only used for legacy graphics APIs.
2546 This label applies to fragment shader inputs only and indicates that
2547 the register contains front/back-face information of the form (F, 0,
2548 0, 1). The first component will be positive when the fragment belongs
2549 to a front-facing polygon, and negative when the fragment belongs to a
2550 back-facing polygon.
2553 TGSI_SEMANTIC_EDGEFLAG
2554 """"""""""""""""""""""
2556 For vertex shaders, this sematic label indicates that an input or
2557 output is a boolean edge flag. The register layout is [F, x, x, x]
2558 where F is 0.0 or 1.0 and x = don't care. Normally, the vertex shader
2559 simply copies the edge flag input to the edgeflag output.
2561 Edge flags are used to control which lines or points are actually
2562 drawn when the polygon mode converts triangles/quads/polygons into
2566 TGSI_SEMANTIC_STENCIL
2567 """""""""""""""""""""
2569 For fragment shaders, this semantic label indicates that an output
2570 is a writable stencil reference value. Only the Y component is writable.
2571 This allows the fragment shader to change the fragments stencilref value.
2574 TGSI_SEMANTIC_VIEWPORT_INDEX
2575 """"""""""""""""""""""""""""
2577 For geometry shaders, this semantic label indicates that an output
2578 contains the index of the viewport (and scissor) to use.
2579 Only the X value is used.
2585 For geometry shaders, this semantic label indicates that an output
2586 contains the layer value to use for the color and depth/stencil surfaces.
2587 Only the X value is used. (Also known as rendertarget array index.)
2590 TGSI_SEMANTIC_CULLDIST
2591 """"""""""""""""""""""
2593 Used as distance to plane for performing application-defined culling
2594 of individual primitives against a plane. When components of vertex
2595 elements are given this label, these values are assumed to be a
2596 float32 signed distance to a plane. Primitives will be completely
2597 discarded if the plane distance for all of the vertices in the
2598 primitive are < 0. If a vertex has a cull distance of NaN, that
2599 vertex counts as "out" (as if its < 0);
2600 The limits on both clip and cull distances are bound
2601 by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
2602 the maximum number of components that can be used to hold the
2603 distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
2604 which specifies the maximum number of registers which can be
2605 annotated with those semantics.
2608 TGSI_SEMANTIC_CLIPDIST
2609 """"""""""""""""""""""
2611 When components of vertex elements are identified this way, these
2612 values are each assumed to be a float32 signed distance to a plane.
2613 Primitive setup only invokes rasterization on pixels for which
2614 the interpolated plane distances are >= 0. Multiple clip planes
2615 can be implemented simultaneously, by annotating multiple
2616 components of one or more vertex elements with the above specified
2617 semantic. The limits on both clip and cull distances are bound
2618 by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
2619 the maximum number of components that can be used to hold the
2620 distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
2621 which specifies the maximum number of registers which can be
2622 annotated with those semantics.
2624 TGSI_SEMANTIC_SAMPLEID
2625 """"""""""""""""""""""
2627 For fragment shaders, this semantic label indicates that a system value
2628 contains the current sample id (i.e. gl_SampleID). Only the X value is used.
2630 TGSI_SEMANTIC_SAMPLEPOS
2631 """""""""""""""""""""""
2633 For fragment shaders, this semantic label indicates that a system value
2634 contains the current sample's position (i.e. gl_SamplePosition). Only the X
2635 and Y values are used.
2637 TGSI_SEMANTIC_SAMPLEMASK
2638 """"""""""""""""""""""""
2640 For fragment shaders, this semantic label indicates that an output contains
2641 the sample mask used to disable further sample processing
2642 (i.e. gl_SampleMask). Only the X value is used, up to 32x MS.
2644 TGSI_SEMANTIC_INVOCATIONID
2645 """"""""""""""""""""""""""
2647 For geometry shaders, this semantic label indicates that a system value
2648 contains the current invocation id (i.e. gl_InvocationID). Only the X value is
2651 Declaration Interpolate
2652 ^^^^^^^^^^^^^^^^^^^^^^^
2654 This token is only valid for fragment shader INPUT declarations.
2656 The Interpolate field specifes the way input is being interpolated by
2657 the rasteriser and is one of TGSI_INTERPOLATE_*.
2659 The CylindricalWrap bitfield specifies which register components
2660 should be subject to cylindrical wrapping when interpolating by the
2661 rasteriser. If TGSI_CYLINDRICAL_WRAP_X is set to 1, the X component
2662 should be interpolated according to cylindrical wrapping rules.
2665 Declaration Sampler View
2666 ^^^^^^^^^^^^^^^^^^^^^^^^
2668 Follows Declaration token if file is TGSI_FILE_SAMPLER_VIEW.
2670 DCL SVIEW[#], resource, type(s)
2672 Declares a shader input sampler view and assigns it to a SVIEW[#]
2675 resource can be one of BUFFER, 1D, 2D, 3D, 1DArray and 2DArray.
2677 type must be 1 or 4 entries (if specifying on a per-component
2678 level) out of UNORM, SNORM, SINT, UINT and FLOAT.
2681 Declaration Resource
2682 ^^^^^^^^^^^^^^^^^^^^
2684 Follows Declaration token if file is TGSI_FILE_RESOURCE.
2686 DCL RES[#], resource [, WR] [, RAW]
2688 Declares a shader input resource and assigns it to a RES[#]
2691 resource can be one of BUFFER, 1D, 2D, 3D, CUBE, 1DArray and
2694 If the RAW keyword is not specified, the texture data will be
2695 subject to conversion, swizzling and scaling as required to yield
2696 the specified data type from the physical data format of the bound
2699 If the RAW keyword is specified, no channel conversion will be
2700 performed: the values read for each of the channels (X,Y,Z,W) will
2701 correspond to consecutive words in the same order and format
2702 they're found in memory. No element-to-address conversion will be
2703 performed either: the value of the provided X coordinate will be
2704 interpreted in byte units instead of texel units. The result of
2705 accessing a misaligned address is undefined.
2707 Usage of the STORE opcode is only allowed if the WR (writable) flag
2712 ^^^^^^^^^^^^^^^^^^^^^^^^
2714 Properties are general directives that apply to the whole TGSI program.
2719 Specifies the fragment shader TGSI_SEMANTIC_POSITION coordinate origin.
2720 The default value is UPPER_LEFT.
2722 If UPPER_LEFT, the position will be (0,0) at the upper left corner and
2723 increase downward and rightward.
2724 If LOWER_LEFT, the position will be (0,0) at the lower left corner and
2725 increase upward and rightward.
2727 OpenGL defaults to LOWER_LEFT, and is configurable with the
2728 GL_ARB_fragment_coord_conventions extension.
2730 DirectX 9/10 use UPPER_LEFT.
2732 FS_COORD_PIXEL_CENTER
2733 """""""""""""""""""""
2735 Specifies the fragment shader TGSI_SEMANTIC_POSITION pixel center convention.
2736 The default value is HALF_INTEGER.
2738 If HALF_INTEGER, the fractionary part of the position will be 0.5
2739 If INTEGER, the fractionary part of the position will be 0.0
2741 Note that this does not affect the set of fragments generated by
2742 rasterization, which is instead controlled by half_pixel_center in the
2745 OpenGL defaults to HALF_INTEGER, and is configurable with the
2746 GL_ARB_fragment_coord_conventions extension.
2748 DirectX 9 uses INTEGER.
2749 DirectX 10 uses HALF_INTEGER.
2751 FS_COLOR0_WRITES_ALL_CBUFS
2752 """"""""""""""""""""""""""
2753 Specifies that writes to the fragment shader color 0 are replicated to all
2754 bound cbufs. This facilitates OpenGL's fragColor output vs fragData[0] where
2755 fragData is directed to a single color buffer, but fragColor is broadcast.
2758 """"""""""""""""""""""""""
2759 If this property is set on the program bound to the shader stage before the
2760 fragment shader, user clip planes should have no effect (be disabled) even if
2761 that shader does not write to any clip distance outputs and the rasterizer's
2762 clip_plane_enable is non-zero.
2763 This property is only supported by drivers that also support shader clip
2765 This is useful for APIs that don't have UCPs and where clip distances written
2766 by a shader cannot be disabled.
2771 Specifies the number of times a geometry shader should be executed for each
2772 input primitive. Each invocation will have a different
2773 TGSI_SEMANTIC_INVOCATIONID system value set. If not specified, assumed to
2777 Texture Sampling and Texture Formats
2778 ------------------------------------
2780 This table shows how texture image components are returned as (x,y,z,w) tuples
2781 by TGSI texture instructions, such as :opcode:`TEX`, :opcode:`TXD`, and
2782 :opcode:`TXP`. For reference, OpenGL and Direct3D conventions are shown as
2785 +--------------------+--------------+--------------------+--------------+
2786 | Texture Components | Gallium | OpenGL | Direct3D 9 |
2787 +====================+==============+====================+==============+
2788 | R | (r, 0, 0, 1) | (r, 0, 0, 1) | (r, 1, 1, 1) |
2789 +--------------------+--------------+--------------------+--------------+
2790 | RG | (r, g, 0, 1) | (r, g, 0, 1) | (r, g, 1, 1) |
2791 +--------------------+--------------+--------------------+--------------+
2792 | RGB | (r, g, b, 1) | (r, g, b, 1) | (r, g, b, 1) |
2793 +--------------------+--------------+--------------------+--------------+
2794 | RGBA | (r, g, b, a) | (r, g, b, a) | (r, g, b, a) |
2795 +--------------------+--------------+--------------------+--------------+
2796 | A | (0, 0, 0, a) | (0, 0, 0, a) | (0, 0, 0, a) |
2797 +--------------------+--------------+--------------------+--------------+
2798 | L | (l, l, l, 1) | (l, l, l, 1) | (l, l, l, 1) |
2799 +--------------------+--------------+--------------------+--------------+
2800 | LA | (l, l, l, a) | (l, l, l, a) | (l, l, l, a) |
2801 +--------------------+--------------+--------------------+--------------+
2802 | I | (i, i, i, i) | (i, i, i, i) | N/A |
2803 +--------------------+--------------+--------------------+--------------+
2804 | UV | XXX TBD | (0, 0, 0, 1) | (u, v, 1, 1) |
2805 | | | [#envmap-bumpmap]_ | |
2806 +--------------------+--------------+--------------------+--------------+
2807 | Z | XXX TBD | (z, z, z, 1) | (0, z, 0, 1) |
2808 | | | [#depth-tex-mode]_ | |
2809 +--------------------+--------------+--------------------+--------------+
2810 | S | (s, s, s, s) | unknown | unknown |
2811 +--------------------+--------------+--------------------+--------------+
2813 .. [#envmap-bumpmap] http://www.opengl.org/registry/specs/ATI/envmap_bumpmap.txt
2814 .. [#depth-tex-mode] the default is (z, z, z, 1) but may also be (0, 0, 0, z)
2815 or (z, z, z, z) depending on the value of GL_DEPTH_TEXTURE_MODE.