Subversion Repositories Kolibri OS

Rev

Go to most recent revision | Blame | Last modification | View Log | RSS feed

  1. TGSI
  2. ====
  3.  
  4. TGSI, Tungsten Graphics Shader Infrastructure, is an intermediate language
  5. for describing shaders. Since Gallium is inherently shaderful, shaders are
  6. an important part of the API. TGSI is the only intermediate representation
  7. used by all drivers.
  8.  
  9. Basics
  10. ------
  11.  
  12. All TGSI instructions, known as *opcodes*, operate on arbitrary-precision
  13. floating-point four-component vectors. An opcode may have up to one
  14. destination register, known as *dst*, and between zero and three source
  15. registers, called *src0* through *src2*, or simply *src* if there is only
  16. one.
  17.  
  18. Some instructions, like :opcode:`I2F`, permit re-interpretation of vector
  19. components as integers. Other instructions permit using registers as
  20. two-component vectors with double precision; see :ref:`Double Opcodes`.
  21.  
  22. When an instruction has a scalar result, the result is usually copied into
  23. each of the components of *dst*. When this happens, the result is said to be
  24. *replicated* to *dst*. :opcode:`RCP` is one such instruction.
  25.  
  26. Modifiers
  27. ^^^^^^^^^^^^^^^
  28.  
  29. TGSI supports modifiers on inputs (as well as saturate modifier on instructions).
  30.  
  31. For inputs which have a floating point type, both absolute value and negation
  32. modifiers are supported (with absolute value being applied first).
  33. TGSI_OPCODE_MOV is considered to have float input type for applying modifiers.
  34.  
  35. For inputs which have signed or unsigned type only the negate modifier is
  36. supported.
  37.  
  38. Instruction Set
  39. ---------------
  40.  
  41. Core ISA
  42. ^^^^^^^^^^^^^^^^^^^^^^^^^
  43.  
  44. These opcodes are guaranteed to be available regardless of the driver being
  45. used.
  46.  
  47. .. opcode:: ARL - Address Register Load
  48.  
  49. .. math::
  50.  
  51.   dst.x = \lfloor src.x\rfloor
  52.  
  53.   dst.y = \lfloor src.y\rfloor
  54.  
  55.   dst.z = \lfloor src.z\rfloor
  56.  
  57.   dst.w = \lfloor src.w\rfloor
  58.  
  59.  
  60. .. opcode:: MOV - Move
  61.  
  62. .. math::
  63.  
  64.   dst.x = src.x
  65.  
  66.   dst.y = src.y
  67.  
  68.   dst.z = src.z
  69.  
  70.   dst.w = src.w
  71.  
  72.  
  73. .. opcode:: LIT - Light Coefficients
  74.  
  75. .. math::
  76.  
  77.   dst.x = 1
  78.  
  79.   dst.y = max(src.x, 0)
  80.  
  81.   dst.z = (src.x > 0) ? max(src.y, 0)^{clamp(src.w, -128, 128))} : 0
  82.  
  83.   dst.w = 1
  84.  
  85.  
  86. .. opcode:: RCP - Reciprocal
  87.  
  88. This instruction replicates its result.
  89.  
  90. .. math::
  91.  
  92.   dst = \frac{1}{src.x}
  93.  
  94.  
  95. .. opcode:: RSQ - Reciprocal Square Root
  96.  
  97. This instruction replicates its result. The results are undefined for src <= 0.
  98.  
  99. .. math::
  100.  
  101.   dst = \frac{1}{\sqrt{src.x}}
  102.  
  103.  
  104. .. opcode:: SQRT - Square Root
  105.  
  106. This instruction replicates its result. The results are undefined for src < 0.
  107.  
  108. .. math::
  109.  
  110.   dst = {\sqrt{src.x}}
  111.  
  112.  
  113. .. opcode:: EXP - Approximate Exponential Base 2
  114.  
  115. .. math::
  116.  
  117.   dst.x = 2^{\lfloor src.x\rfloor}
  118.  
  119.   dst.y = src.x - \lfloor src.x\rfloor
  120.  
  121.   dst.z = 2^{src.x}
  122.  
  123.   dst.w = 1
  124.  
  125.  
  126. .. opcode:: LOG - Approximate Logarithm Base 2
  127.  
  128. .. math::
  129.  
  130.   dst.x = \lfloor\log_2{|src.x|}\rfloor
  131.  
  132.   dst.y = \frac{|src.x|}{2^{\lfloor\log_2{|src.x|}\rfloor}}
  133.  
  134.   dst.z = \log_2{|src.x|}
  135.  
  136.   dst.w = 1
  137.  
  138.  
  139. .. opcode:: MUL - Multiply
  140.  
  141. .. math::
  142.  
  143.   dst.x = src0.x \times src1.x
  144.  
  145.   dst.y = src0.y \times src1.y
  146.  
  147.   dst.z = src0.z \times src1.z
  148.  
  149.   dst.w = src0.w \times src1.w
  150.  
  151.  
  152. .. opcode:: ADD - Add
  153.  
  154. .. math::
  155.  
  156.   dst.x = src0.x + src1.x
  157.  
  158.   dst.y = src0.y + src1.y
  159.  
  160.   dst.z = src0.z + src1.z
  161.  
  162.   dst.w = src0.w + src1.w
  163.  
  164.  
  165. .. opcode:: DP3 - 3-component Dot Product
  166.  
  167. This instruction replicates its result.
  168.  
  169. .. math::
  170.  
  171.   dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z
  172.  
  173.  
  174. .. opcode:: DP4 - 4-component Dot Product
  175.  
  176. This instruction replicates its result.
  177.  
  178. .. math::
  179.  
  180.   dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z + src0.w \times src1.w
  181.  
  182.  
  183. .. opcode:: DST - Distance Vector
  184.  
  185. .. math::
  186.  
  187.   dst.x = 1
  188.  
  189.   dst.y = src0.y \times src1.y
  190.  
  191.   dst.z = src0.z
  192.  
  193.   dst.w = src1.w
  194.  
  195.  
  196. .. opcode:: MIN - Minimum
  197.  
  198. .. math::
  199.  
  200.   dst.x = min(src0.x, src1.x)
  201.  
  202.   dst.y = min(src0.y, src1.y)
  203.  
  204.   dst.z = min(src0.z, src1.z)
  205.  
  206.   dst.w = min(src0.w, src1.w)
  207.  
  208.  
  209. .. opcode:: MAX - Maximum
  210.  
  211. .. math::
  212.  
  213.   dst.x = max(src0.x, src1.x)
  214.  
  215.   dst.y = max(src0.y, src1.y)
  216.  
  217.   dst.z = max(src0.z, src1.z)
  218.  
  219.   dst.w = max(src0.w, src1.w)
  220.  
  221.  
  222. .. opcode:: SLT - Set On Less Than
  223.  
  224. .. math::
  225.  
  226.   dst.x = (src0.x < src1.x) ? 1 : 0
  227.  
  228.   dst.y = (src0.y < src1.y) ? 1 : 0
  229.  
  230.   dst.z = (src0.z < src1.z) ? 1 : 0
  231.  
  232.   dst.w = (src0.w < src1.w) ? 1 : 0
  233.  
  234.  
  235. .. opcode:: SGE - Set On Greater Equal Than
  236.  
  237. .. math::
  238.  
  239.   dst.x = (src0.x >= src1.x) ? 1 : 0
  240.  
  241.   dst.y = (src0.y >= src1.y) ? 1 : 0
  242.  
  243.   dst.z = (src0.z >= src1.z) ? 1 : 0
  244.  
  245.   dst.w = (src0.w >= src1.w) ? 1 : 0
  246.  
  247.  
  248. .. opcode:: MAD - Multiply And Add
  249.  
  250. .. math::
  251.  
  252.   dst.x = src0.x \times src1.x + src2.x
  253.  
  254.   dst.y = src0.y \times src1.y + src2.y
  255.  
  256.   dst.z = src0.z \times src1.z + src2.z
  257.  
  258.   dst.w = src0.w \times src1.w + src2.w
  259.  
  260.  
  261. .. opcode:: SUB - Subtract
  262.  
  263. .. math::
  264.  
  265.   dst.x = src0.x - src1.x
  266.  
  267.   dst.y = src0.y - src1.y
  268.  
  269.   dst.z = src0.z - src1.z
  270.  
  271.   dst.w = src0.w - src1.w
  272.  
  273.  
  274. .. opcode:: LRP - Linear Interpolate
  275.  
  276. .. math::
  277.  
  278.   dst.x = src0.x \times src1.x + (1 - src0.x) \times src2.x
  279.  
  280.   dst.y = src0.y \times src1.y + (1 - src0.y) \times src2.y
  281.  
  282.   dst.z = src0.z \times src1.z + (1 - src0.z) \times src2.z
  283.  
  284.   dst.w = src0.w \times src1.w + (1 - src0.w) \times src2.w
  285.  
  286.  
  287. .. opcode:: CND - Condition
  288.  
  289. .. math::
  290.  
  291.   dst.x = (src2.x > 0.5) ? src0.x : src1.x
  292.  
  293.   dst.y = (src2.y > 0.5) ? src0.y : src1.y
  294.  
  295.   dst.z = (src2.z > 0.5) ? src0.z : src1.z
  296.  
  297.   dst.w = (src2.w > 0.5) ? src0.w : src1.w
  298.  
  299.  
  300. .. opcode:: DP2A - 2-component Dot Product And Add
  301.  
  302. .. math::
  303.  
  304.   dst.x = src0.x \times src1.x + src0.y \times src1.y + src2.x
  305.  
  306.   dst.y = src0.x \times src1.x + src0.y \times src1.y + src2.x
  307.  
  308.   dst.z = src0.x \times src1.x + src0.y \times src1.y + src2.x
  309.  
  310.   dst.w = src0.x \times src1.x + src0.y \times src1.y + src2.x
  311.  
  312.  
  313. .. opcode:: FRC - Fraction
  314.  
  315. .. math::
  316.  
  317.   dst.x = src.x - \lfloor src.x\rfloor
  318.  
  319.   dst.y = src.y - \lfloor src.y\rfloor
  320.  
  321.   dst.z = src.z - \lfloor src.z\rfloor
  322.  
  323.   dst.w = src.w - \lfloor src.w\rfloor
  324.  
  325.  
  326. .. opcode:: CLAMP - Clamp
  327.  
  328. .. math::
  329.  
  330.   dst.x = clamp(src0.x, src1.x, src2.x)
  331.  
  332.   dst.y = clamp(src0.y, src1.y, src2.y)
  333.  
  334.   dst.z = clamp(src0.z, src1.z, src2.z)
  335.  
  336.   dst.w = clamp(src0.w, src1.w, src2.w)
  337.  
  338.  
  339. .. opcode:: FLR - Floor
  340.  
  341. This is identical to :opcode:`ARL`.
  342.  
  343. .. math::
  344.  
  345.   dst.x = \lfloor src.x\rfloor
  346.  
  347.   dst.y = \lfloor src.y\rfloor
  348.  
  349.   dst.z = \lfloor src.z\rfloor
  350.  
  351.   dst.w = \lfloor src.w\rfloor
  352.  
  353.  
  354. .. opcode:: ROUND - Round
  355.  
  356. .. math::
  357.  
  358.   dst.x = round(src.x)
  359.  
  360.   dst.y = round(src.y)
  361.  
  362.   dst.z = round(src.z)
  363.  
  364.   dst.w = round(src.w)
  365.  
  366.  
  367. .. opcode:: EX2 - Exponential Base 2
  368.  
  369. This instruction replicates its result.
  370.  
  371. .. math::
  372.  
  373.   dst = 2^{src.x}
  374.  
  375.  
  376. .. opcode:: LG2 - Logarithm Base 2
  377.  
  378. This instruction replicates its result.
  379.  
  380. .. math::
  381.  
  382.   dst = \log_2{src.x}
  383.  
  384.  
  385. .. opcode:: POW - Power
  386.  
  387. This instruction replicates its result.
  388.  
  389. .. math::
  390.  
  391.   dst = src0.x^{src1.x}
  392.  
  393. .. opcode:: XPD - Cross Product
  394.  
  395. .. math::
  396.  
  397.   dst.x = src0.y \times src1.z - src1.y \times src0.z
  398.  
  399.   dst.y = src0.z \times src1.x - src1.z \times src0.x
  400.  
  401.   dst.z = src0.x \times src1.y - src1.x \times src0.y
  402.  
  403.   dst.w = 1
  404.  
  405.  
  406. .. opcode:: ABS - Absolute
  407.  
  408. .. math::
  409.  
  410.   dst.x = |src.x|
  411.  
  412.   dst.y = |src.y|
  413.  
  414.   dst.z = |src.z|
  415.  
  416.   dst.w = |src.w|
  417.  
  418.  
  419. .. opcode:: RCC - Reciprocal Clamped
  420.  
  421. This instruction replicates its result.
  422.  
  423. XXX cleanup on aisle three
  424.  
  425. .. math::
  426.  
  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)
  428.  
  429.  
  430. .. opcode:: DPH - Homogeneous Dot Product
  431.  
  432. This instruction replicates its result.
  433.  
  434. .. math::
  435.  
  436.   dst = src0.x \times src1.x + src0.y \times src1.y + src0.z \times src1.z + src1.w
  437.  
  438.  
  439. .. opcode:: COS - Cosine
  440.  
  441. This instruction replicates its result.
  442.  
  443. .. math::
  444.  
  445.   dst = \cos{src.x}
  446.  
  447.  
  448. .. opcode:: DDX - Derivative Relative To X
  449.  
  450. .. math::
  451.  
  452.   dst.x = partialx(src.x)
  453.  
  454.   dst.y = partialx(src.y)
  455.  
  456.   dst.z = partialx(src.z)
  457.  
  458.   dst.w = partialx(src.w)
  459.  
  460.  
  461. .. opcode:: DDY - Derivative Relative To Y
  462.  
  463. .. math::
  464.  
  465.   dst.x = partialy(src.x)
  466.  
  467.   dst.y = partialy(src.y)
  468.  
  469.   dst.z = partialy(src.z)
  470.  
  471.   dst.w = partialy(src.w)
  472.  
  473.  
  474. .. opcode:: PK2H - Pack Two 16-bit Floats
  475.  
  476.   TBD
  477.  
  478.  
  479. .. opcode:: PK2US - Pack Two Unsigned 16-bit Scalars
  480.  
  481.   TBD
  482.  
  483.  
  484. .. opcode:: PK4B - Pack Four Signed 8-bit Scalars
  485.  
  486.   TBD
  487.  
  488.  
  489. .. opcode:: PK4UB - Pack Four Unsigned 8-bit Scalars
  490.  
  491.   TBD
  492.  
  493.  
  494. .. opcode:: RFL - Reflection Vector
  495.  
  496. .. math::
  497.  
  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
  499.  
  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
  501.  
  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
  503.  
  504.   dst.w = 1
  505.  
  506. .. note::
  507.  
  508.    Considered for removal.
  509.  
  510.  
  511. .. opcode:: SEQ - Set On Equal
  512.  
  513. .. math::
  514.  
  515.   dst.x = (src0.x == src1.x) ? 1 : 0
  516.  
  517.   dst.y = (src0.y == src1.y) ? 1 : 0
  518.  
  519.   dst.z = (src0.z == src1.z) ? 1 : 0
  520.  
  521.   dst.w = (src0.w == src1.w) ? 1 : 0
  522.  
  523.  
  524. .. opcode:: SFL - Set On False
  525.  
  526. This instruction replicates its result.
  527.  
  528. .. math::
  529.  
  530.   dst = 0
  531.  
  532. .. note::
  533.  
  534.    Considered for removal.
  535.  
  536.  
  537. .. opcode:: SGT - Set On Greater Than
  538.  
  539. .. math::
  540.  
  541.   dst.x = (src0.x > src1.x) ? 1 : 0
  542.  
  543.   dst.y = (src0.y > src1.y) ? 1 : 0
  544.  
  545.   dst.z = (src0.z > src1.z) ? 1 : 0
  546.  
  547.   dst.w = (src0.w > src1.w) ? 1 : 0
  548.  
  549.  
  550. .. opcode:: SIN - Sine
  551.  
  552. This instruction replicates its result.
  553.  
  554. .. math::
  555.  
  556.   dst = \sin{src.x}
  557.  
  558.  
  559. .. opcode:: SLE - Set On Less Equal Than
  560.  
  561. .. math::
  562.  
  563.   dst.x = (src0.x <= src1.x) ? 1 : 0
  564.  
  565.   dst.y = (src0.y <= src1.y) ? 1 : 0
  566.  
  567.   dst.z = (src0.z <= src1.z) ? 1 : 0
  568.  
  569.   dst.w = (src0.w <= src1.w) ? 1 : 0
  570.  
  571.  
  572. .. opcode:: SNE - Set On Not Equal
  573.  
  574. .. math::
  575.  
  576.   dst.x = (src0.x != src1.x) ? 1 : 0
  577.  
  578.   dst.y = (src0.y != src1.y) ? 1 : 0
  579.  
  580.   dst.z = (src0.z != src1.z) ? 1 : 0
  581.  
  582.   dst.w = (src0.w != src1.w) ? 1 : 0
  583.  
  584.  
  585. .. opcode:: STR - Set On True
  586.  
  587. This instruction replicates its result.
  588.  
  589. .. math::
  590.  
  591.   dst = 1
  592.  
  593.  
  594. .. opcode:: TEX - Texture Lookup
  595.  
  596. .. math::
  597.  
  598.   coord = src0
  599.  
  600.   bias = 0.0
  601.  
  602.   dst = texture_sample(unit, coord, bias)
  603.  
  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
  607.   the reference value.
  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.
  615.  
  616. .. opcode:: TXD - Texture Lookup with Derivatives
  617.  
  618. .. math::
  619.  
  620.   coord = src0
  621.  
  622.   ddx = src1
  623.  
  624.   ddy = src2
  625.  
  626.   bias = 0.0
  627.  
  628.   dst = texture_sample_deriv(unit, coord, bias, ddx, ddy)
  629.  
  630.  
  631. .. opcode:: TXP - Projective Texture Lookup
  632.  
  633. .. math::
  634.  
  635.   coord.x = src0.x / src.w
  636.  
  637.   coord.y = src0.y / src.w
  638.  
  639.   coord.z = src0.z / src.w
  640.  
  641.   coord.w = src0.w
  642.  
  643.   bias = 0.0
  644.  
  645.   dst = texture_sample(unit, coord, bias)
  646.  
  647.  
  648. .. opcode:: UP2H - Unpack Two 16-Bit Floats
  649.  
  650.   TBD
  651.  
  652. .. note::
  653.  
  654.    Considered for removal.
  655.  
  656. .. opcode:: UP2US - Unpack Two Unsigned 16-Bit Scalars
  657.  
  658.   TBD
  659.  
  660. .. note::
  661.  
  662.    Considered for removal.
  663.  
  664. .. opcode:: UP4B - Unpack Four Signed 8-Bit Values
  665.  
  666.   TBD
  667.  
  668. .. note::
  669.  
  670.    Considered for removal.
  671.  
  672. .. opcode:: UP4UB - Unpack Four Unsigned 8-Bit Scalars
  673.  
  674.   TBD
  675.  
  676. .. note::
  677.  
  678.    Considered for removal.
  679.  
  680. .. opcode:: X2D - 2D Coordinate Transformation
  681.  
  682. .. math::
  683.  
  684.   dst.x = src0.x + src1.x \times src2.x + src1.y \times src2.y
  685.  
  686.   dst.y = src0.y + src1.x \times src2.z + src1.y \times src2.w
  687.  
  688.   dst.z = src0.x + src1.x \times src2.x + src1.y \times src2.y
  689.  
  690.   dst.w = src0.y + src1.x \times src2.z + src1.y \times src2.w
  691.  
  692. .. note::
  693.  
  694.    Considered for removal.
  695.  
  696.  
  697. .. opcode:: ARA - Address Register Add
  698.  
  699.   TBD
  700.  
  701. .. note::
  702.  
  703.    Considered for removal.
  704.  
  705. .. opcode:: ARR - Address Register Load With Round
  706.  
  707. .. math::
  708.  
  709.   dst.x = round(src.x)
  710.  
  711.   dst.y = round(src.y)
  712.  
  713.   dst.z = round(src.z)
  714.  
  715.   dst.w = round(src.w)
  716.  
  717.  
  718. .. opcode:: SSG - Set Sign
  719.  
  720. .. math::
  721.  
  722.   dst.x = (src.x > 0) ? 1 : (src.x < 0) ? -1 : 0
  723.  
  724.   dst.y = (src.y > 0) ? 1 : (src.y < 0) ? -1 : 0
  725.  
  726.   dst.z = (src.z > 0) ? 1 : (src.z < 0) ? -1 : 0
  727.  
  728.   dst.w = (src.w > 0) ? 1 : (src.w < 0) ? -1 : 0
  729.  
  730.  
  731. .. opcode:: CMP - Compare
  732.  
  733. .. math::
  734.  
  735.   dst.x = (src0.x < 0) ? src1.x : src2.x
  736.  
  737.   dst.y = (src0.y < 0) ? src1.y : src2.y
  738.  
  739.   dst.z = (src0.z < 0) ? src1.z : src2.z
  740.  
  741.   dst.w = (src0.w < 0) ? src1.w : src2.w
  742.  
  743.  
  744. .. opcode:: KILL_IF - Conditional Discard
  745.  
  746.   Conditional discard.  Allowed in fragment shaders only.
  747.  
  748. .. math::
  749.  
  750.   if (src.x < 0 || src.y < 0 || src.z < 0 || src.w < 0)
  751.     discard
  752.   endif
  753.  
  754.  
  755. .. opcode:: KILL - Discard
  756.  
  757.   Unconditional discard.  Allowed in fragment shaders only.
  758.  
  759.  
  760. .. opcode:: SCS - Sine Cosine
  761.  
  762. .. math::
  763.  
  764.   dst.x = \cos{src.x}
  765.  
  766.   dst.y = \sin{src.x}
  767.  
  768.   dst.z = 0
  769.  
  770.   dst.w = 1
  771.  
  772.  
  773. .. opcode:: TXB - Texture Lookup With Bias
  774.  
  775. .. math::
  776.  
  777.   coord.x = src.x
  778.  
  779.   coord.y = src.y
  780.  
  781.   coord.z = src.z
  782.  
  783.   coord.w = 1.0
  784.  
  785.   bias = src.z
  786.  
  787.   dst = texture_sample(unit, coord, bias)
  788.  
  789.  
  790. .. opcode:: NRM - 3-component Vector Normalise
  791.  
  792. .. math::
  793.  
  794.   dst.x = src.x / (src.x \times src.x + src.y \times src.y + src.z \times src.z)
  795.  
  796.   dst.y = src.y / (src.x \times src.x + src.y \times src.y + src.z \times src.z)
  797.  
  798.   dst.z = src.z / (src.x \times src.x + src.y \times src.y + src.z \times src.z)
  799.  
  800.   dst.w = 1
  801.  
  802.  
  803. .. opcode:: DIV - Divide
  804.  
  805. .. math::
  806.  
  807.   dst.x = \frac{src0.x}{src1.x}
  808.  
  809.   dst.y = \frac{src0.y}{src1.y}
  810.  
  811.   dst.z = \frac{src0.z}{src1.z}
  812.  
  813.   dst.w = \frac{src0.w}{src1.w}
  814.  
  815.  
  816. .. opcode:: DP2 - 2-component Dot Product
  817.  
  818. This instruction replicates its result.
  819.  
  820. .. math::
  821.  
  822.   dst = src0.x \times src1.x + src0.y \times src1.y
  823.  
  824.  
  825. .. opcode:: TXL - Texture Lookup With explicit LOD
  826.  
  827. .. math::
  828.  
  829.   coord.x = src0.x
  830.  
  831.   coord.y = src0.y
  832.  
  833.   coord.z = src0.z
  834.  
  835.   coord.w = 1.0
  836.  
  837.   lod = src0.w
  838.  
  839.   dst = texture_sample(unit, coord, lod)
  840.  
  841.  
  842. .. opcode:: PUSHA - Push Address Register On Stack
  843.  
  844.   push(src.x)
  845.   push(src.y)
  846.   push(src.z)
  847.   push(src.w)
  848.  
  849. .. note::
  850.  
  851.    Considered for cleanup.
  852.  
  853. .. note::
  854.  
  855.    Considered for removal.
  856.  
  857. .. opcode:: POPA - Pop Address Register From Stack
  858.  
  859.   dst.w = pop()
  860.   dst.z = pop()
  861.   dst.y = pop()
  862.   dst.x = pop()
  863.  
  864. .. note::
  865.  
  866.    Considered for cleanup.
  867.  
  868. .. note::
  869.  
  870.    Considered for removal.
  871.  
  872.  
  873. .. opcode:: BRA - Branch
  874.  
  875.   pc = target
  876.  
  877. .. note::
  878.  
  879.    Considered for removal.
  880.  
  881.  
  882. .. opcode:: CALLNZ - Subroutine Call If Not Zero
  883.  
  884.    TBD
  885.  
  886. .. note::
  887.  
  888.    Considered for cleanup.
  889.  
  890. .. note::
  891.  
  892.    Considered for removal.
  893.  
  894.  
  895. Compute ISA
  896. ^^^^^^^^^^^^^^^^^^^^^^^^
  897.  
  898. These opcodes are primarily provided for special-use computational shaders.
  899. Support for these opcodes indicated by a special pipe capability bit (TBD).
  900.  
  901. XXX doesn't look like most of the opcodes really belong here.
  902.  
  903. .. opcode:: CEIL - Ceiling
  904.  
  905. .. math::
  906.  
  907.   dst.x = \lceil src.x\rceil
  908.  
  909.   dst.y = \lceil src.y\rceil
  910.  
  911.   dst.z = \lceil src.z\rceil
  912.  
  913.   dst.w = \lceil src.w\rceil
  914.  
  915.  
  916. .. opcode:: TRUNC - Truncate
  917.  
  918. .. math::
  919.  
  920.   dst.x = trunc(src.x)
  921.  
  922.   dst.y = trunc(src.y)
  923.  
  924.   dst.z = trunc(src.z)
  925.  
  926.   dst.w = trunc(src.w)
  927.  
  928.  
  929. .. opcode:: MOD - Modulus
  930.  
  931. .. math::
  932.  
  933.   dst.x = src0.x \bmod src1.x
  934.  
  935.   dst.y = src0.y \bmod src1.y
  936.  
  937.   dst.z = src0.z \bmod src1.z
  938.  
  939.   dst.w = src0.w \bmod src1.w
  940.  
  941.  
  942. .. opcode:: UARL - Integer Address Register Load
  943.  
  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.
  946.  
  947.  
  948. .. opcode:: SAD - Sum Of Absolute Differences
  949.  
  950. .. math::
  951.  
  952.   dst.x = |src0.x - src1.x| + src2.x
  953.  
  954.   dst.y = |src0.y - src1.y| + src2.y
  955.  
  956.   dst.z = |src0.z - src1.z| + src2.z
  957.  
  958.   dst.w = |src0.w - src1.w| + src2.w
  959.  
  960.  
  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).
  971.  
  972.  
  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)
  978.  
  979. .. math::
  980.  
  981.   lod = src0.x
  982.  
  983.   dst.x = texture_width(unit, lod)
  984.  
  985.   dst.y = texture_height(unit, lod)
  986.  
  987.   dst.z = texture_depth(unit, lod)
  988.  
  989.  
  990. Integer ISA
  991. ^^^^^^^^^^^^^^^^^^^^^^^^
  992. These opcodes are used for integer operations.
  993. Support for these opcodes indicated by PIPE_SHADER_CAP_INTEGERS (all of them?)
  994.  
  995.  
  996. .. opcode:: I2F - Signed Integer To Float
  997.  
  998.    Rounding is unspecified (round to nearest even suggested).
  999.  
  1000. .. math::
  1001.  
  1002.   dst.x = (float) src.x
  1003.  
  1004.   dst.y = (float) src.y
  1005.  
  1006.   dst.z = (float) src.z
  1007.  
  1008.   dst.w = (float) src.w
  1009.  
  1010.  
  1011. .. opcode:: U2F - Unsigned Integer To Float
  1012.  
  1013.    Rounding is unspecified (round to nearest even suggested).
  1014.  
  1015. .. math::
  1016.  
  1017.   dst.x = (float) src.x
  1018.  
  1019.   dst.y = (float) src.y
  1020.  
  1021.   dst.z = (float) src.z
  1022.  
  1023.   dst.w = (float) src.w
  1024.  
  1025.  
  1026. .. opcode:: F2I - Float to Signed Integer
  1027.  
  1028.    Rounding is towards zero (truncate).
  1029.    Values outside signed range (including NaNs) produce undefined results.
  1030.  
  1031. .. math::
  1032.  
  1033.   dst.x = (int) src.x
  1034.  
  1035.   dst.y = (int) src.y
  1036.  
  1037.   dst.z = (int) src.z
  1038.  
  1039.   dst.w = (int) src.w
  1040.  
  1041.  
  1042. .. opcode:: F2U - Float to Unsigned Integer
  1043.  
  1044.    Rounding is towards zero (truncate).
  1045.    Values outside unsigned range (including NaNs) produce undefined results.
  1046.  
  1047. .. math::
  1048.  
  1049.   dst.x = (unsigned) src.x
  1050.  
  1051.   dst.y = (unsigned) src.y
  1052.  
  1053.   dst.z = (unsigned) src.z
  1054.  
  1055.   dst.w = (unsigned) src.w
  1056.  
  1057.  
  1058. .. opcode:: UADD - Integer Add
  1059.  
  1060.    This instruction works the same for signed and unsigned integers.
  1061.    The low 32bit of the result is returned.
  1062.  
  1063. .. math::
  1064.  
  1065.   dst.x = src0.x + src1.x
  1066.  
  1067.   dst.y = src0.y + src1.y
  1068.  
  1069.   dst.z = src0.z + src1.z
  1070.  
  1071.   dst.w = src0.w + src1.w
  1072.  
  1073.  
  1074. .. opcode:: UMAD - Integer Multiply And Add
  1075.  
  1076.    This instruction works the same for signed and unsigned integers.
  1077.    The multiplication returns the low 32bit (as does the result itself).
  1078.  
  1079. .. math::
  1080.  
  1081.   dst.x = src0.x \times src1.x + src2.x
  1082.  
  1083.   dst.y = src0.y \times src1.y + src2.y
  1084.  
  1085.   dst.z = src0.z \times src1.z + src2.z
  1086.  
  1087.   dst.w = src0.w \times src1.w + src2.w
  1088.  
  1089.  
  1090. .. opcode:: UMUL - Integer Multiply
  1091.  
  1092.    This instruction works the same for signed and unsigned integers.
  1093.    The low 32bit of the result is returned.
  1094.  
  1095. .. math::
  1096.  
  1097.   dst.x = src0.x \times src1.x
  1098.  
  1099.   dst.y = src0.y \times src1.y
  1100.  
  1101.   dst.z = src0.z \times src1.z
  1102.  
  1103.   dst.w = src0.w \times src1.w
  1104.  
  1105.  
  1106. .. opcode:: IDIV - Signed Integer Division
  1107.  
  1108.    TBD: behavior for division by zero.
  1109.  
  1110. .. math::
  1111.  
  1112.   dst.x = src0.x \ src1.x
  1113.  
  1114.   dst.y = src0.y \ src1.y
  1115.  
  1116.   dst.z = src0.z \ src1.z
  1117.  
  1118.   dst.w = src0.w \ src1.w
  1119.  
  1120.  
  1121. .. opcode:: UDIV - Unsigned Integer Division
  1122.  
  1123.    For division by zero, 0xffffffff is returned.
  1124.  
  1125. .. math::
  1126.  
  1127.   dst.x = src0.x \ src1.x
  1128.  
  1129.   dst.y = src0.y \ src1.y
  1130.  
  1131.   dst.z = src0.z \ src1.z
  1132.  
  1133.   dst.w = src0.w \ src1.w
  1134.  
  1135.  
  1136. .. opcode:: UMOD - Unsigned Integer Remainder
  1137.  
  1138.    If second arg is zero, 0xffffffff is returned.
  1139.  
  1140. .. math::
  1141.  
  1142.   dst.x = src0.x \ src1.x
  1143.  
  1144.   dst.y = src0.y \ src1.y
  1145.  
  1146.   dst.z = src0.z \ src1.z
  1147.  
  1148.   dst.w = src0.w \ src1.w
  1149.  
  1150.  
  1151. .. opcode:: NOT - Bitwise Not
  1152.  
  1153. .. math::
  1154.  
  1155.   dst.x = ~src.x
  1156.  
  1157.   dst.y = ~src.y
  1158.  
  1159.   dst.z = ~src.z
  1160.  
  1161.   dst.w = ~src.w
  1162.  
  1163.  
  1164. .. opcode:: AND - Bitwise And
  1165.  
  1166. .. math::
  1167.  
  1168.   dst.x = src0.x & src1.x
  1169.  
  1170.   dst.y = src0.y & src1.y
  1171.  
  1172.   dst.z = src0.z & src1.z
  1173.  
  1174.   dst.w = src0.w & src1.w
  1175.  
  1176.  
  1177. .. opcode:: OR - Bitwise Or
  1178.  
  1179. .. math::
  1180.  
  1181.   dst.x = src0.x | src1.x
  1182.  
  1183.   dst.y = src0.y | src1.y
  1184.  
  1185.   dst.z = src0.z | src1.z
  1186.  
  1187.   dst.w = src0.w | src1.w
  1188.  
  1189.  
  1190. .. opcode:: XOR - Bitwise Xor
  1191.  
  1192. .. math::
  1193.  
  1194.   dst.x = src0.x \oplus src1.x
  1195.  
  1196.   dst.y = src0.y \oplus src1.y
  1197.  
  1198.   dst.z = src0.z \oplus src1.z
  1199.  
  1200.   dst.w = src0.w \oplus src1.w
  1201.  
  1202.  
  1203. .. opcode:: IMAX - Maximum of Signed Integers
  1204.  
  1205. .. math::
  1206.  
  1207.   dst.x = max(src0.x, src1.x)
  1208.  
  1209.   dst.y = max(src0.y, src1.y)
  1210.  
  1211.   dst.z = max(src0.z, src1.z)
  1212.  
  1213.   dst.w = max(src0.w, src1.w)
  1214.  
  1215.  
  1216. .. opcode:: UMAX - Maximum of Unsigned Integers
  1217.  
  1218. .. math::
  1219.  
  1220.   dst.x = max(src0.x, src1.x)
  1221.  
  1222.   dst.y = max(src0.y, src1.y)
  1223.  
  1224.   dst.z = max(src0.z, src1.z)
  1225.  
  1226.   dst.w = max(src0.w, src1.w)
  1227.  
  1228.  
  1229. .. opcode:: IMIN - Minimum of Signed Integers
  1230.  
  1231. .. math::
  1232.  
  1233.   dst.x = min(src0.x, src1.x)
  1234.  
  1235.   dst.y = min(src0.y, src1.y)
  1236.  
  1237.   dst.z = min(src0.z, src1.z)
  1238.  
  1239.   dst.w = min(src0.w, src1.w)
  1240.  
  1241.  
  1242. .. opcode:: UMIN - Minimum of Unsigned Integers
  1243.  
  1244. .. math::
  1245.  
  1246.   dst.x = min(src0.x, src1.x)
  1247.  
  1248.   dst.y = min(src0.y, src1.y)
  1249.  
  1250.   dst.z = min(src0.z, src1.z)
  1251.  
  1252.   dst.w = min(src0.w, src1.w)
  1253.  
  1254.  
  1255. .. opcode:: SHL - Shift Left
  1256.  
  1257. .. math::
  1258.  
  1259.   dst.x = src0.x << src1.x
  1260.  
  1261.   dst.y = src0.y << src1.x
  1262.  
  1263.   dst.z = src0.z << src1.x
  1264.  
  1265.   dst.w = src0.w << src1.x
  1266.  
  1267.  
  1268. .. opcode:: ISHR - Arithmetic Shift Right (of Signed Integer)
  1269.  
  1270. .. math::
  1271.  
  1272.   dst.x = src0.x >> src1.x
  1273.  
  1274.   dst.y = src0.y >> src1.x
  1275.  
  1276.   dst.z = src0.z >> src1.x
  1277.  
  1278.   dst.w = src0.w >> src1.x
  1279.  
  1280.  
  1281. .. opcode:: USHR - Logical Shift Right
  1282.  
  1283. .. math::
  1284.  
  1285.   dst.x = src0.x >> (unsigned) src1.x
  1286.  
  1287.   dst.y = src0.y >> (unsigned) src1.x
  1288.  
  1289.   dst.z = src0.z >> (unsigned) src1.x
  1290.  
  1291.   dst.w = src0.w >> (unsigned) src1.x
  1292.  
  1293.  
  1294. .. opcode:: UCMP - Integer Conditional Move
  1295.  
  1296. .. math::
  1297.  
  1298.   dst.x = src0.x ? src1.x : src2.x
  1299.  
  1300.   dst.y = src0.y ? src1.y : src2.y
  1301.  
  1302.   dst.z = src0.z ? src1.z : src2.z
  1303.  
  1304.   dst.w = src0.w ? src1.w : src2.w
  1305.  
  1306.  
  1307.  
  1308. .. opcode:: ISSG - Integer Set Sign
  1309.  
  1310. .. math::
  1311.  
  1312.   dst.x = (src0.x < 0) ? -1 : (src0.x > 0) ? 1 : 0
  1313.  
  1314.   dst.y = (src0.y < 0) ? -1 : (src0.y > 0) ? 1 : 0
  1315.  
  1316.   dst.z = (src0.z < 0) ? -1 : (src0.z > 0) ? 1 : 0
  1317.  
  1318.   dst.w = (src0.w < 0) ? -1 : (src0.w > 0) ? 1 : 0
  1319.  
  1320.  
  1321.  
  1322. .. opcode:: ISLT - Signed Integer Set On Less Than
  1323.  
  1324. .. math::
  1325.  
  1326.   dst.x = (src0.x < src1.x) ? ~0 : 0
  1327.  
  1328.   dst.y = (src0.y < src1.y) ? ~0 : 0
  1329.  
  1330.   dst.z = (src0.z < src1.z) ? ~0 : 0
  1331.  
  1332.   dst.w = (src0.w < src1.w) ? ~0 : 0
  1333.  
  1334.  
  1335. .. opcode:: USLT - Unsigned Integer Set On Less Than
  1336.  
  1337. .. math::
  1338.  
  1339.   dst.x = (src0.x < src1.x) ? ~0 : 0
  1340.  
  1341.   dst.y = (src0.y < src1.y) ? ~0 : 0
  1342.  
  1343.   dst.z = (src0.z < src1.z) ? ~0 : 0
  1344.  
  1345.   dst.w = (src0.w < src1.w) ? ~0 : 0
  1346.  
  1347.  
  1348. .. opcode:: ISGE - Signed Integer Set On Greater Equal Than
  1349.  
  1350. .. math::
  1351.  
  1352.   dst.x = (src0.x >= src1.x) ? ~0 : 0
  1353.  
  1354.   dst.y = (src0.y >= src1.y) ? ~0 : 0
  1355.  
  1356.   dst.z = (src0.z >= src1.z) ? ~0 : 0
  1357.  
  1358.   dst.w = (src0.w >= src1.w) ? ~0 : 0
  1359.  
  1360.  
  1361. .. opcode:: USGE - Unsigned Integer Set On Greater Equal Than
  1362.  
  1363. .. math::
  1364.  
  1365.   dst.x = (src0.x >= src1.x) ? ~0 : 0
  1366.  
  1367.   dst.y = (src0.y >= src1.y) ? ~0 : 0
  1368.  
  1369.   dst.z = (src0.z >= src1.z) ? ~0 : 0
  1370.  
  1371.   dst.w = (src0.w >= src1.w) ? ~0 : 0
  1372.  
  1373.  
  1374. .. opcode:: USEQ - Integer Set On Equal
  1375.  
  1376. .. math::
  1377.  
  1378.   dst.x = (src0.x == src1.x) ? ~0 : 0
  1379.  
  1380.   dst.y = (src0.y == src1.y) ? ~0 : 0
  1381.  
  1382.   dst.z = (src0.z == src1.z) ? ~0 : 0
  1383.  
  1384.   dst.w = (src0.w == src1.w) ? ~0 : 0
  1385.  
  1386.  
  1387. .. opcode:: USNE - Integer Set On Not Equal
  1388.  
  1389. .. math::
  1390.  
  1391.   dst.x = (src0.x != src1.x) ? ~0 : 0
  1392.  
  1393.   dst.y = (src0.y != src1.y) ? ~0 : 0
  1394.  
  1395.   dst.z = (src0.z != src1.z) ? ~0 : 0
  1396.  
  1397.   dst.w = (src0.w != src1.w) ? ~0 : 0
  1398.  
  1399.  
  1400. .. opcode:: INEG - Integer Negate
  1401.  
  1402.   Two's complement.
  1403.  
  1404. .. math::
  1405.  
  1406.   dst.x = -src.x
  1407.  
  1408.   dst.y = -src.y
  1409.  
  1410.   dst.z = -src.z
  1411.  
  1412.   dst.w = -src.w
  1413.  
  1414.  
  1415. .. opcode:: IABS - Integer Absolute Value
  1416.  
  1417. .. math::
  1418.  
  1419.   dst.x = |src.x|
  1420.  
  1421.   dst.y = |src.y|
  1422.  
  1423.   dst.z = |src.z|
  1424.  
  1425.   dst.w = |src.w|
  1426.  
  1427.  
  1428. Geometry ISA
  1429. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  1430.  
  1431. These opcodes are only supported in geometry shaders; they have no meaning
  1432. in any other type of shader.
  1433.  
  1434. .. opcode:: EMIT - Emit
  1435.  
  1436.   Generate a new vertex for the current primitive using the values in the
  1437.   output registers.
  1438.  
  1439.  
  1440. .. opcode:: ENDPRIM - End Primitive
  1441.  
  1442.   Complete the current primitive (consisting of the emitted vertices),
  1443.   and start a new one.
  1444.  
  1445.  
  1446. GLSL ISA
  1447. ^^^^^^^^^^
  1448.  
  1449. These opcodes are part of :term:`GLSL`'s opcode set. Support for these
  1450. opcodes is determined by a special capability bit, ``GLSL``.
  1451. Some require glsl version 1.30 (UIF/BREAKC/SWITCH/CASE/DEFAULT/ENDSWITCH).
  1452.  
  1453. .. opcode:: CAL - Subroutine Call
  1454.  
  1455.   push(pc)
  1456.   pc = target
  1457.  
  1458.  
  1459. .. opcode:: RET - Subroutine Call Return
  1460.  
  1461.   pc = pop()
  1462.  
  1463.  
  1464. .. opcode:: CONT - Continue
  1465.  
  1466.   Unconditionally moves the point of execution to the instruction after the
  1467.   last bgnloop. The instruction must appear within a bgnloop/endloop.
  1468.  
  1469. .. note::
  1470.  
  1471.    Support for CONT is determined by a special capability bit,
  1472.    ``TGSI_CONT_SUPPORTED``. See :ref:`Screen` for more information.
  1473.  
  1474.  
  1475. .. opcode:: BGNLOOP - Begin a Loop
  1476.  
  1477.   Start a loop. Must have a matching endloop.
  1478.  
  1479.  
  1480. .. opcode:: BGNSUB - Begin Subroutine
  1481.  
  1482.   Starts definition of a subroutine. Must have a matching endsub.
  1483.  
  1484.  
  1485. .. opcode:: ENDLOOP - End a Loop
  1486.  
  1487.   End a loop started with bgnloop.
  1488.  
  1489.  
  1490. .. opcode:: ENDSUB - End Subroutine
  1491.  
  1492.   Ends definition of a subroutine.
  1493.  
  1494.  
  1495. .. opcode:: NOP - No Operation
  1496.  
  1497.   Do nothing.
  1498.  
  1499.  
  1500. .. opcode:: BRK - Break
  1501.  
  1502.   Unconditionally moves the point of execution to the instruction after the
  1503.   next endloop or endswitch. The instruction must appear within a loop/endloop
  1504.   or switch/endswitch.
  1505.  
  1506.  
  1507. .. opcode:: BREAKC - Break Conditional
  1508.  
  1509.   Conditionally moves the point of execution to the instruction after the
  1510.   next endloop or endswitch. The instruction must appear within a loop/endloop
  1511.   or switch/endswitch.
  1512.   Condition evaluates to true if src0.x != 0 where src0.x is interpreted
  1513.   as an integer register.
  1514.  
  1515. .. note::
  1516.  
  1517.    Considered for removal as it's quite inconsistent wrt other opcodes
  1518.    (could emulate with UIF/BRK/ENDIF).
  1519.  
  1520.  
  1521. .. opcode:: IF - Float If
  1522.  
  1523.   Start an IF ... ELSE .. ENDIF block.  Condition evaluates to true if
  1524.  
  1525.     src0.x != 0.0
  1526.  
  1527.   where src0.x is interpreted as a floating point register.
  1528.  
  1529.  
  1530. .. opcode:: UIF - Bitwise If
  1531.  
  1532.   Start an UIF ... ELSE .. ENDIF block. Condition evaluates to true if
  1533.  
  1534.     src0.x != 0
  1535.  
  1536.   where src0.x is interpreted as an integer register.
  1537.  
  1538.  
  1539. .. opcode:: ELSE - Else
  1540.  
  1541.   Starts an else block, after an IF or UIF statement.
  1542.  
  1543.  
  1544. .. opcode:: ENDIF - End If
  1545.  
  1546.   Ends an IF or UIF block.
  1547.  
  1548.  
  1549. .. opcode:: SWITCH - Switch
  1550.  
  1551.    Starts a C-style switch expression. The switch consists of one or multiple
  1552.    CASE statements, and at most one DEFAULT statement. Execution of a statement
  1553.    ends when a BRK is hit, but just like in C falling through to other cases
  1554.    without a break is allowed. Similarly, DEFAULT label is allowed anywhere not
  1555.    just as last statement, and fallthrough is allowed into/from it.
  1556.    CASE src arguments are evaluated at bit level against the SWITCH src argument.
  1557.  
  1558.    Example:
  1559.    SWITCH src[0].x
  1560.    CASE src[0].x
  1561.    (some instructions here)
  1562.    (optional BRK here)
  1563.    DEFAULT
  1564.    (some instructions here)
  1565.    (optional BRK here)
  1566.    CASE src[0].x
  1567.    (some instructions here)
  1568.    (optional BRK here)
  1569.    ENDSWITCH
  1570.  
  1571.  
  1572. .. opcode:: CASE - Switch case
  1573.  
  1574.    This represents a switch case label. The src arg must be an integer immediate.
  1575.  
  1576.  
  1577. .. opcode:: DEFAULT - Switch default
  1578.  
  1579.    This represents the default case in the switch, which is taken if no other
  1580.    case matches.
  1581.  
  1582.  
  1583. .. opcode:: ENDSWITCH - End of switch
  1584.  
  1585.    Ends a switch expression.
  1586.  
  1587.  
  1588. .. opcode:: NRM4 - 4-component Vector Normalise
  1589.  
  1590. This instruction replicates its result.
  1591.  
  1592. .. math::
  1593.  
  1594.   dst = \frac{src.x}{src.x \times src.x + src.y \times src.y + src.z \times src.z + src.w \times src.w}
  1595.  
  1596.  
  1597. .. _doubleopcodes:
  1598.  
  1599. Double ISA
  1600. ^^^^^^^^^^^^^^^
  1601.  
  1602. The double-precision opcodes reinterpret four-component vectors into
  1603. two-component vectors with doubled precision in each component.
  1604.  
  1605. Support for these opcodes is XXX undecided. :T
  1606.  
  1607. .. opcode:: DADD - Add
  1608.  
  1609. .. math::
  1610.  
  1611.   dst.xy = src0.xy + src1.xy
  1612.  
  1613.   dst.zw = src0.zw + src1.zw
  1614.  
  1615.  
  1616. .. opcode:: DDIV - Divide
  1617.  
  1618. .. math::
  1619.  
  1620.   dst.xy = src0.xy / src1.xy
  1621.  
  1622.   dst.zw = src0.zw / src1.zw
  1623.  
  1624. .. opcode:: DSEQ - Set on Equal
  1625.  
  1626. .. math::
  1627.  
  1628.   dst.xy = src0.xy == src1.xy ? 1.0F : 0.0F
  1629.  
  1630.   dst.zw = src0.zw == src1.zw ? 1.0F : 0.0F
  1631.  
  1632. .. opcode:: DSLT - Set on Less than
  1633.  
  1634. .. math::
  1635.  
  1636.   dst.xy = src0.xy < src1.xy ? 1.0F : 0.0F
  1637.  
  1638.   dst.zw = src0.zw < src1.zw ? 1.0F : 0.0F
  1639.  
  1640. .. opcode:: DFRAC - Fraction
  1641.  
  1642. .. math::
  1643.  
  1644.   dst.xy = src.xy - \lfloor src.xy\rfloor
  1645.  
  1646.   dst.zw = src.zw - \lfloor src.zw\rfloor
  1647.  
  1648.  
  1649. .. opcode:: DFRACEXP - Convert Number to Fractional and Integral Components
  1650.  
  1651. Like the ``frexp()`` routine in many math libraries, this opcode stores the
  1652. exponent of its source to ``dst0``, and the significand to ``dst1``, such that
  1653. :math:`dst1 \times 2^{dst0} = src` .
  1654.  
  1655. .. math::
  1656.  
  1657.   dst0.xy = exp(src.xy)
  1658.  
  1659.   dst1.xy = frac(src.xy)
  1660.  
  1661.   dst0.zw = exp(src.zw)
  1662.  
  1663.   dst1.zw = frac(src.zw)
  1664.  
  1665. .. opcode:: DLDEXP - Multiply Number by Integral Power of 2
  1666.  
  1667. This opcode is the inverse of :opcode:`DFRACEXP`.
  1668.  
  1669. .. math::
  1670.  
  1671.   dst.xy = src0.xy \times 2^{src1.xy}
  1672.  
  1673.   dst.zw = src0.zw \times 2^{src1.zw}
  1674.  
  1675. .. opcode:: DMIN - Minimum
  1676.  
  1677. .. math::
  1678.  
  1679.   dst.xy = min(src0.xy, src1.xy)
  1680.  
  1681.   dst.zw = min(src0.zw, src1.zw)
  1682.  
  1683. .. opcode:: DMAX - Maximum
  1684.  
  1685. .. math::
  1686.  
  1687.   dst.xy = max(src0.xy, src1.xy)
  1688.  
  1689.   dst.zw = max(src0.zw, src1.zw)
  1690.  
  1691. .. opcode:: DMUL - Multiply
  1692.  
  1693. .. math::
  1694.  
  1695.   dst.xy = src0.xy \times src1.xy
  1696.  
  1697.   dst.zw = src0.zw \times src1.zw
  1698.  
  1699.  
  1700. .. opcode:: DMAD - Multiply And Add
  1701.  
  1702. .. math::
  1703.  
  1704.   dst.xy = src0.xy \times src1.xy + src2.xy
  1705.  
  1706.   dst.zw = src0.zw \times src1.zw + src2.zw
  1707.  
  1708.  
  1709. .. opcode:: DRCP - Reciprocal
  1710.  
  1711. .. math::
  1712.  
  1713.    dst.xy = \frac{1}{src.xy}
  1714.  
  1715.    dst.zw = \frac{1}{src.zw}
  1716.  
  1717. .. opcode:: DSQRT - Square Root
  1718.  
  1719. .. math::
  1720.  
  1721.    dst.xy = \sqrt{src.xy}
  1722.  
  1723.    dst.zw = \sqrt{src.zw}
  1724.  
  1725.  
  1726. .. _samplingopcodes:
  1727.  
  1728. Resource Sampling Opcodes
  1729. ^^^^^^^^^^^^^^^^^^^^^^^^^
  1730.  
  1731. Those opcodes follow very closely semantics of the respective Direct3D
  1732. instructions. If in doubt double check Direct3D documentation.
  1733.  
  1734. .. opcode:: SAMPLE - Using provided address, sample data from the
  1735.                specified texture using the filtering mode identified
  1736.                by the gven sampler. The source data may come from
  1737.                any resource type other than buffers.
  1738.                SAMPLE dst, address, sampler_view, sampler
  1739.                e.g.
  1740.                SAMPLE TEMP[0], TEMP[1], SVIEW[0], SAMP[0]
  1741.  
  1742. .. opcode:: SAMPLE_I - Simplified alternative to the SAMPLE instruction.
  1743.                Using the provided integer address, SAMPLE_I fetches data
  1744.                from the specified sampler view without any filtering.
  1745.                The source data may come from any resource type other
  1746.                than CUBE.
  1747.                SAMPLE_I dst, address, sampler_view
  1748.                e.g.
  1749.                SAMPLE_I TEMP[0], TEMP[1], SVIEW[0]
  1750.                The 'address' is specified as unsigned integers. If the
  1751.                'address' is out of range [0...(# texels - 1)] the
  1752.                result of the fetch is always 0 in all components.
  1753.                As such the instruction doesn't honor address wrap
  1754.                modes, in cases where that behavior is desirable
  1755.                'SAMPLE' instruction should be used.
  1756.                address.w always provides an unsigned integer mipmap
  1757.                level. If the value is out of the range then the
  1758.                instruction always returns 0 in all components.
  1759.                address.yz are ignored for buffers and 1d textures.
  1760.                address.z is ignored for 1d texture arrays and 2d
  1761.                textures.
  1762.                For 1D texture arrays address.y provides the array
  1763.                index (also as unsigned integer). If the value is
  1764.                out of the range of available array indices
  1765.                [0... (array size - 1)] then the opcode always returns
  1766.                0 in all components.
  1767.                For 2D texture arrays address.z provides the array
  1768.                index, otherwise it exhibits the same behavior as in
  1769.                the case for 1D texture arrays.
  1770.                The exact semantics of the source address are presented
  1771.                in the table below:
  1772.                resource type         X     Y     Z       W
  1773.                -------------         ------------------------
  1774.                PIPE_BUFFER           x                ignored
  1775.                PIPE_TEXTURE_1D       x                  mpl
  1776.                PIPE_TEXTURE_2D       x     y            mpl
  1777.                PIPE_TEXTURE_3D       x     y     z      mpl
  1778.                PIPE_TEXTURE_RECT     x     y            mpl
  1779.                PIPE_TEXTURE_CUBE     not allowed as source
  1780.                PIPE_TEXTURE_1D_ARRAY x    idx           mpl
  1781.                PIPE_TEXTURE_2D_ARRAY x     y    idx     mpl
  1782.  
  1783.                Where 'mpl' is a mipmap level and 'idx' is the
  1784.                array index.
  1785.  
  1786. .. opcode:: SAMPLE_I_MS - Just like SAMPLE_I but allows fetch data from
  1787.                multi-sampled surfaces.
  1788.                SAMPLE_I_MS dst, address, sampler_view, sample
  1789.  
  1790. .. opcode:: SAMPLE_B - Just like the SAMPLE instruction with the
  1791.                exception that an additional bias is applied to the
  1792.                level of detail computed as part of the instruction
  1793.                execution.
  1794.                SAMPLE_B dst, address, sampler_view, sampler, lod_bias
  1795.                e.g.
  1796.                SAMPLE_B TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x
  1797.  
  1798. .. opcode:: SAMPLE_C - Similar to the SAMPLE instruction but it
  1799.                performs a comparison filter. The operands to SAMPLE_C
  1800.                are identical to SAMPLE, except that there is an additional
  1801.                float32 operand, reference value, which must be a register
  1802.                with single-component, or a scalar literal.
  1803.                SAMPLE_C makes the hardware use the current samplers
  1804.                compare_func (in pipe_sampler_state) to compare
  1805.                reference value against the red component value for the
  1806.                surce resource at each texel that the currently configured
  1807.                texture filter covers based on the provided coordinates.
  1808.                SAMPLE_C dst, address, sampler_view.r, sampler, ref_value
  1809.                e.g.
  1810.                SAMPLE_C TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x
  1811.  
  1812. .. opcode:: SAMPLE_C_LZ - Same as SAMPLE_C, but LOD is 0 and derivatives
  1813.                are ignored. The LZ stands for level-zero.
  1814.                SAMPLE_C_LZ dst, address, sampler_view.r, sampler, ref_value
  1815.                e.g.
  1816.                SAMPLE_C_LZ TEMP[0], TEMP[1], SVIEW[0].r, SAMP[0], TEMP[2].x
  1817.  
  1818.  
  1819. .. opcode:: SAMPLE_D - SAMPLE_D is identical to the SAMPLE opcode except
  1820.                that the derivatives for the source address in the x
  1821.                direction and the y direction are provided by extra
  1822.                parameters.
  1823.                SAMPLE_D dst, address, sampler_view, sampler, der_x, der_y
  1824.                e.g.
  1825.                SAMPLE_D TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2], TEMP[3]
  1826.  
  1827. .. opcode:: SAMPLE_L - SAMPLE_L is identical to the SAMPLE opcode except
  1828.                that the LOD is provided directly as a scalar value,
  1829.                representing no anisotropy.
  1830.                SAMPLE_L dst, address, sampler_view, sampler, explicit_lod
  1831.                e.g.
  1832.                SAMPLE_L TEMP[0], TEMP[1], SVIEW[0], SAMP[0], TEMP[2].x
  1833.  
  1834. .. opcode:: GATHER4 - Gathers the four texels to be used in a bi-linear
  1835.                filtering operation and packs them into a single register.
  1836.                Only works with 2D, 2D array, cubemaps, and cubemaps arrays.
  1837.                For 2D textures, only the addressing modes of the sampler and
  1838.                the top level of any mip pyramid are used. Set W to zero.
  1839.                It behaves like the SAMPLE instruction, but a filtered
  1840.                sample is not generated. The four samples that contribute
  1841.                to filtering are placed into xyzw in counter-clockwise order,
  1842.                starting with the (u,v) texture coordinate delta at the
  1843.                following locations (-, +), (+, +), (+, -), (-, -), where
  1844.                the magnitude of the deltas are half a texel.
  1845.  
  1846.  
  1847. .. opcode:: SVIEWINFO - query the dimensions of a given sampler view.
  1848.                dst receives width, height, depth or array size and
  1849.                number of mipmap levels as int4. The dst can have a writemask
  1850.                which will specify what info is the caller interested
  1851.                in.
  1852.                SVIEWINFO dst, src_mip_level, sampler_view
  1853.                e.g.
  1854.                SVIEWINFO TEMP[0], TEMP[1].x, SVIEW[0]
  1855.                src_mip_level is an unsigned integer scalar. If it's
  1856.                out of range then returns 0 for width, height and
  1857.                depth/array size but the total number of mipmap is
  1858.                still returned correctly for the given sampler view.
  1859.                The returned width, height and depth values are for
  1860.                the mipmap level selected by the src_mip_level and
  1861.                are in the number of texels.
  1862.                For 1d texture array width is in dst.x, array size
  1863.                is in dst.y and dst.zw are always 0.
  1864.  
  1865. .. opcode:: SAMPLE_POS - query the position of a given sample.
  1866.                dst receives float4 (x, y, 0, 0) indicated where the
  1867.                sample is located. If the resource is not a multi-sample
  1868.                resource and not a render target, the result is 0.
  1869.  
  1870. .. opcode:: SAMPLE_INFO - dst receives number of samples in x.
  1871.                If the resource is not a multi-sample resource and
  1872.                not a render target, the result is 0.
  1873.  
  1874.  
  1875. .. _resourceopcodes:
  1876.  
  1877. Resource Access Opcodes
  1878. ^^^^^^^^^^^^^^^^^^^^^^^
  1879.  
  1880. .. opcode:: LOAD - Fetch data from a shader resource
  1881.  
  1882.                Syntax: ``LOAD dst, resource, address``
  1883.  
  1884.                Example: ``LOAD TEMP[0], RES[0], TEMP[1]``
  1885.  
  1886.                Using the provided integer address, LOAD fetches data
  1887.                from the specified buffer or texture without any
  1888.                filtering.
  1889.  
  1890.                The 'address' is specified as a vector of unsigned
  1891.                integers.  If the 'address' is out of range the result
  1892.                is unspecified.
  1893.  
  1894.                Only the first mipmap level of a resource can be read
  1895.                from using this instruction.
  1896.  
  1897.                For 1D or 2D texture arrays, the array index is
  1898.                provided as an unsigned integer in address.y or
  1899.                address.z, respectively.  address.yz are ignored for
  1900.                buffers and 1D textures.  address.z is ignored for 1D
  1901.                texture arrays and 2D textures.  address.w is always
  1902.                ignored.
  1903.  
  1904. .. opcode:: STORE - Write data to a shader resource
  1905.  
  1906.                Syntax: ``STORE resource, address, src``
  1907.  
  1908.                Example: ``STORE RES[0], TEMP[0], TEMP[1]``
  1909.  
  1910.                Using the provided integer address, STORE writes data
  1911.                to the specified buffer or texture.
  1912.  
  1913.                The 'address' is specified as a vector of unsigned
  1914.                integers.  If the 'address' is out of range the result
  1915.                is unspecified.
  1916.  
  1917.                Only the first mipmap level of a resource can be
  1918.                written to using this instruction.
  1919.  
  1920.                For 1D or 2D texture arrays, the array index is
  1921.                provided as an unsigned integer in address.y or
  1922.                address.z, respectively.  address.yz are ignored for
  1923.                buffers and 1D textures.  address.z is ignored for 1D
  1924.                texture arrays and 2D textures.  address.w is always
  1925.                ignored.
  1926.  
  1927.  
  1928. .. _threadsyncopcodes:
  1929.  
  1930. Inter-thread synchronization opcodes
  1931. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  1932.  
  1933. These opcodes are intended for communication between threads running
  1934. within the same compute grid.  For now they're only valid in compute
  1935. programs.
  1936.  
  1937. .. opcode:: MFENCE - Memory fence
  1938.  
  1939.   Syntax: ``MFENCE resource``
  1940.  
  1941.   Example: ``MFENCE RES[0]``
  1942.  
  1943.   This opcode forces strong ordering between any memory access
  1944.   operations that affect the specified resource.  This means that
  1945.   previous loads and stores (and only those) will be performed and
  1946.   visible to other threads before the program execution continues.
  1947.  
  1948.  
  1949. .. opcode:: LFENCE - Load memory fence
  1950.  
  1951.   Syntax: ``LFENCE resource``
  1952.  
  1953.   Example: ``LFENCE RES[0]``
  1954.  
  1955.   Similar to MFENCE, but it only affects the ordering of memory loads.
  1956.  
  1957.  
  1958. .. opcode:: SFENCE - Store memory fence
  1959.  
  1960.   Syntax: ``SFENCE resource``
  1961.  
  1962.   Example: ``SFENCE RES[0]``
  1963.  
  1964.   Similar to MFENCE, but it only affects the ordering of memory stores.
  1965.  
  1966.  
  1967. .. opcode:: BARRIER - Thread group barrier
  1968.  
  1969.   ``BARRIER``
  1970.  
  1971.   This opcode suspends the execution of the current thread until all
  1972.   the remaining threads in the working group reach the same point of
  1973.   the program.  Results are unspecified if any of the remaining
  1974.   threads terminates or never reaches an executed BARRIER instruction.
  1975.  
  1976.  
  1977. .. _atomopcodes:
  1978.  
  1979. Atomic opcodes
  1980. ^^^^^^^^^^^^^^
  1981.  
  1982. These opcodes provide atomic variants of some common arithmetic and
  1983. logical operations.  In this context atomicity means that another
  1984. concurrent memory access operation that affects the same memory
  1985. location is guaranteed to be performed strictly before or after the
  1986. entire execution of the atomic operation.
  1987.  
  1988. For the moment they're only valid in compute programs.
  1989.  
  1990. .. opcode:: ATOMUADD - Atomic integer addition
  1991.  
  1992.   Syntax: ``ATOMUADD dst, resource, offset, src``
  1993.  
  1994.   Example: ``ATOMUADD TEMP[0], RES[0], TEMP[1], TEMP[2]``
  1995.  
  1996.   The following operation is performed atomically on each component:
  1997.  
  1998. .. math::
  1999.  
  2000.   dst_i = resource[offset]_i
  2001.  
  2002.   resource[offset]_i = dst_i + src_i
  2003.  
  2004.  
  2005. .. opcode:: ATOMXCHG - Atomic exchange
  2006.  
  2007.   Syntax: ``ATOMXCHG dst, resource, offset, src``
  2008.  
  2009.   Example: ``ATOMXCHG TEMP[0], RES[0], TEMP[1], TEMP[2]``
  2010.  
  2011.   The following operation is performed atomically on each component:
  2012.  
  2013. .. math::
  2014.  
  2015.   dst_i = resource[offset]_i
  2016.  
  2017.   resource[offset]_i = src_i
  2018.  
  2019.  
  2020. .. opcode:: ATOMCAS - Atomic compare-and-exchange
  2021.  
  2022.   Syntax: ``ATOMCAS dst, resource, offset, cmp, src``
  2023.  
  2024.   Example: ``ATOMCAS TEMP[0], RES[0], TEMP[1], TEMP[2], TEMP[3]``
  2025.  
  2026.   The following operation is performed atomically on each component:
  2027.  
  2028. .. math::
  2029.  
  2030.   dst_i = resource[offset]_i
  2031.  
  2032.   resource[offset]_i = (dst_i == cmp_i ? src_i : dst_i)
  2033.  
  2034.  
  2035. .. opcode:: ATOMAND - Atomic bitwise And
  2036.  
  2037.   Syntax: ``ATOMAND dst, resource, offset, src``
  2038.  
  2039.   Example: ``ATOMAND TEMP[0], RES[0], TEMP[1], TEMP[2]``
  2040.  
  2041.   The following operation is performed atomically on each component:
  2042.  
  2043. .. math::
  2044.  
  2045.   dst_i = resource[offset]_i
  2046.  
  2047.   resource[offset]_i = dst_i \& src_i
  2048.  
  2049.  
  2050. .. opcode:: ATOMOR - Atomic bitwise Or
  2051.  
  2052.   Syntax: ``ATOMOR dst, resource, offset, src``
  2053.  
  2054.   Example: ``ATOMOR TEMP[0], RES[0], TEMP[1], TEMP[2]``
  2055.  
  2056.   The following operation is performed atomically on each component:
  2057.  
  2058. .. math::
  2059.  
  2060.   dst_i = resource[offset]_i
  2061.  
  2062.   resource[offset]_i = dst_i | src_i
  2063.  
  2064.  
  2065. .. opcode:: ATOMXOR - Atomic bitwise Xor
  2066.  
  2067.   Syntax: ``ATOMXOR dst, resource, offset, src``
  2068.  
  2069.   Example: ``ATOMXOR TEMP[0], RES[0], TEMP[1], TEMP[2]``
  2070.  
  2071.   The following operation is performed atomically on each component:
  2072.  
  2073. .. math::
  2074.  
  2075.   dst_i = resource[offset]_i
  2076.  
  2077.   resource[offset]_i = dst_i \oplus src_i
  2078.  
  2079.  
  2080. .. opcode:: ATOMUMIN - Atomic unsigned minimum
  2081.  
  2082.   Syntax: ``ATOMUMIN dst, resource, offset, src``
  2083.  
  2084.   Example: ``ATOMUMIN TEMP[0], RES[0], TEMP[1], TEMP[2]``
  2085.  
  2086.   The following operation is performed atomically on each component:
  2087.  
  2088. .. math::
  2089.  
  2090.   dst_i = resource[offset]_i
  2091.  
  2092.   resource[offset]_i = (dst_i < src_i ? dst_i : src_i)
  2093.  
  2094.  
  2095. .. opcode:: ATOMUMAX - Atomic unsigned maximum
  2096.  
  2097.   Syntax: ``ATOMUMAX dst, resource, offset, src``
  2098.  
  2099.   Example: ``ATOMUMAX TEMP[0], RES[0], TEMP[1], TEMP[2]``
  2100.  
  2101.   The following operation is performed atomically on each component:
  2102.  
  2103. .. math::
  2104.  
  2105.   dst_i = resource[offset]_i
  2106.  
  2107.   resource[offset]_i = (dst_i > src_i ? dst_i : src_i)
  2108.  
  2109.  
  2110. .. opcode:: ATOMIMIN - Atomic signed minimum
  2111.  
  2112.   Syntax: ``ATOMIMIN dst, resource, offset, src``
  2113.  
  2114.   Example: ``ATOMIMIN TEMP[0], RES[0], TEMP[1], TEMP[2]``
  2115.  
  2116.   The following operation is performed atomically on each component:
  2117.  
  2118. .. math::
  2119.  
  2120.   dst_i = resource[offset]_i
  2121.  
  2122.   resource[offset]_i = (dst_i < src_i ? dst_i : src_i)
  2123.  
  2124.  
  2125. .. opcode:: ATOMIMAX - Atomic signed maximum
  2126.  
  2127.   Syntax: ``ATOMIMAX dst, resource, offset, src``
  2128.  
  2129.   Example: ``ATOMIMAX TEMP[0], RES[0], TEMP[1], TEMP[2]``
  2130.  
  2131.   The following operation is performed atomically on each component:
  2132.  
  2133. .. math::
  2134.  
  2135.   dst_i = resource[offset]_i
  2136.  
  2137.   resource[offset]_i = (dst_i > src_i ? dst_i : src_i)
  2138.  
  2139.  
  2140.  
  2141. Explanation of symbols used
  2142. ------------------------------
  2143.  
  2144.  
  2145. Functions
  2146. ^^^^^^^^^^^^^^
  2147.  
  2148.  
  2149.   :math:`|x|`       Absolute value of `x`.
  2150.  
  2151.   :math:`\lceil x \rceil` Ceiling of `x`.
  2152.  
  2153.   clamp(x,y,z)      Clamp x between y and z.
  2154.                     (x < y) ? y : (x > z) ? z : x
  2155.  
  2156.   :math:`\lfloor x\rfloor` Floor of `x`.
  2157.  
  2158.   :math:`\log_2{x}` Logarithm of `x`, base 2.
  2159.  
  2160.   max(x,y)          Maximum of x and y.
  2161.                     (x > y) ? x : y
  2162.  
  2163.   min(x,y)          Minimum of x and y.
  2164.                     (x < y) ? x : y
  2165.  
  2166.   partialx(x)       Derivative of x relative to fragment's X.
  2167.  
  2168.   partialy(x)       Derivative of x relative to fragment's Y.
  2169.  
  2170.   pop()             Pop from stack.
  2171.  
  2172.   :math:`x^y`       `x` to the power `y`.
  2173.  
  2174.   push(x)           Push x on stack.
  2175.  
  2176.   round(x)          Round x.
  2177.  
  2178.   trunc(x)          Truncate x, i.e. drop the fraction bits.
  2179.  
  2180.  
  2181. Keywords
  2182. ^^^^^^^^^^^^^
  2183.  
  2184.  
  2185.   discard           Discard fragment.
  2186.  
  2187.   pc                Program counter.
  2188.  
  2189.   target            Label of target instruction.
  2190.  
  2191.  
  2192. Other tokens
  2193. ---------------
  2194.  
  2195.  
  2196. Declaration
  2197. ^^^^^^^^^^^
  2198.  
  2199.  
  2200. Declares a register that is will be referenced as an operand in Instruction
  2201. tokens.
  2202.  
  2203. File field contains register file that is being declared and is one
  2204. of TGSI_FILE.
  2205.  
  2206. UsageMask field specifies which of the register components can be accessed
  2207. and is one of TGSI_WRITEMASK.
  2208.  
  2209. The Local flag specifies that a given value isn't intended for
  2210. subroutine parameter passing and, as a result, the implementation
  2211. isn't required to give any guarantees of it being preserved across
  2212. subroutine boundaries.  As it's merely a compiler hint, the
  2213. implementation is free to ignore it.
  2214.  
  2215. If Dimension flag is set to 1, a Declaration Dimension token follows.
  2216.  
  2217. If Semantic flag is set to 1, a Declaration Semantic token follows.
  2218.  
  2219. If Interpolate flag is set to 1, a Declaration Interpolate token follows.
  2220.  
  2221. If file is TGSI_FILE_RESOURCE, a Declaration Resource token follows.
  2222.  
  2223. If Array flag is set to 1, a Declaration Array token follows.
  2224.  
  2225. Array Declaration
  2226. ^^^^^^^^^^^^^^^^^^^^^^^^
  2227.  
  2228. Declarations can optional have an ArrayID attribute which can be referred by
  2229. indirect addressing operands. An ArrayID of zero is reserved and treaded as
  2230. if no ArrayID is specified.
  2231.  
  2232. If an indirect addressing operand refers to a specific declaration by using
  2233. an ArrayID only the registers in this declaration are guaranteed to be
  2234. accessed, accessing any register outside this declaration results in undefined
  2235. behavior. Note that for compatibility the effective index is zero-based and
  2236. not relative to the specified declaration
  2237.  
  2238. If no ArrayID is specified with an indirect addressing operand the whole
  2239. register file might be accessed by this operand. This is strongly discouraged
  2240. and will prevent packing of scalar/vec2 arrays and effective alias analysis.
  2241.  
  2242. Declaration Semantic
  2243. ^^^^^^^^^^^^^^^^^^^^^^^^
  2244.  
  2245.   Vertex and fragment shader input and output registers may be labeled
  2246.   with semantic information consisting of a name and index.
  2247.  
  2248.   Follows Declaration token if Semantic bit is set.
  2249.  
  2250.   Since its purpose is to link a shader with other stages of the pipeline,
  2251.   it is valid to follow only those Declaration tokens that declare a register
  2252.   either in INPUT or OUTPUT file.
  2253.  
  2254.   SemanticName field contains the semantic name of the register being declared.
  2255.   There is no default value.
  2256.  
  2257.   SemanticIndex is an optional subscript that can be used to distinguish
  2258.   different register declarations with the same semantic name. The default value
  2259.   is 0.
  2260.  
  2261.   The meanings of the individual semantic names are explained in the following
  2262.   sections.
  2263.  
  2264. TGSI_SEMANTIC_POSITION
  2265. """"""""""""""""""""""
  2266.  
  2267. For vertex shaders, TGSI_SEMANTIC_POSITION indicates the vertex shader
  2268. output register which contains the homogeneous vertex position in the clip
  2269. space coordinate system.  After clipping, the X, Y and Z components of the
  2270. vertex will be divided by the W value to get normalized device coordinates.
  2271.  
  2272. For fragment shaders, TGSI_SEMANTIC_POSITION is used to indicate that
  2273. fragment shader input contains the fragment's window position.  The X
  2274. component starts at zero and always increases from left to right.
  2275. The Y component starts at zero and always increases but Y=0 may either
  2276. indicate the top of the window or the bottom depending on the fragment
  2277. coordinate origin convention (see TGSI_PROPERTY_FS_COORD_ORIGIN).
  2278. The Z coordinate ranges from 0 to 1 to represent depth from the front
  2279. to the back of the Z buffer.  The W component contains the reciprocol
  2280. of the interpolated vertex position W component.
  2281.  
  2282. Fragment shaders may also declare an output register with
  2283. TGSI_SEMANTIC_POSITION.  Only the Z component is writable.  This allows
  2284. the fragment shader to change the fragment's Z position.
  2285.  
  2286.  
  2287.  
  2288. TGSI_SEMANTIC_COLOR
  2289. """""""""""""""""""
  2290.  
  2291. For vertex shader outputs or fragment shader inputs/outputs, this
  2292. label indicates that the resister contains an R,G,B,A color.
  2293.  
  2294. Several shader inputs/outputs may contain colors so the semantic index
  2295. is used to distinguish them.  For example, color[0] may be the diffuse
  2296. color while color[1] may be the specular color.
  2297.  
  2298. This label is needed so that the flat/smooth shading can be applied
  2299. to the right interpolants during rasterization.
  2300.  
  2301.  
  2302.  
  2303. TGSI_SEMANTIC_BCOLOR
  2304. """"""""""""""""""""
  2305.  
  2306. Back-facing colors are only used for back-facing polygons, and are only valid
  2307. in vertex shader outputs. After rasterization, all polygons are front-facing
  2308. and COLOR and BCOLOR end up occupying the same slots in the fragment shader,
  2309. so all BCOLORs effectively become regular COLORs in the fragment shader.
  2310.  
  2311.  
  2312. TGSI_SEMANTIC_FOG
  2313. """""""""""""""""
  2314.  
  2315. Vertex shader inputs and outputs and fragment shader inputs may be
  2316. labeled with TGSI_SEMANTIC_FOG to indicate that the register contains
  2317. a fog coordinate in the form (F, 0, 0, 1).  Typically, the fragment
  2318. shader will use the fog coordinate to compute a fog blend factor which
  2319. is used to blend the normal fragment color with a constant fog color.
  2320.  
  2321. Only the first component matters when writing from the vertex shader;
  2322. the driver will ensure that the coordinate is in this format when used
  2323. as a fragment shader input.
  2324.  
  2325.  
  2326. TGSI_SEMANTIC_PSIZE
  2327. """""""""""""""""""
  2328.  
  2329. Vertex shader input and output registers may be labeled with
  2330. TGIS_SEMANTIC_PSIZE to indicate that the register contains a point size
  2331. in the form (S, 0, 0, 1).  The point size controls the width or diameter
  2332. of points for rasterization.  This label cannot be used in fragment
  2333. shaders.
  2334.  
  2335. When using this semantic, be sure to set the appropriate state in the
  2336. :ref:`rasterizer` first.
  2337.  
  2338.  
  2339. TGSI_SEMANTIC_TEXCOORD
  2340. """"""""""""""""""""""
  2341.  
  2342. Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
  2343.  
  2344. Vertex shader outputs and fragment shader inputs may be labeled with
  2345. this semantic to make them replaceable by sprite coordinates via the
  2346. sprite_coord_enable state in the :ref:`rasterizer`.
  2347. The semantic index permitted with this semantic is limited to <= 7.
  2348.  
  2349. If the driver does not support TEXCOORD, sprite coordinate replacement
  2350. applies to inputs with the GENERIC semantic instead.
  2351.  
  2352. The intended use case for this semantic is gl_TexCoord.
  2353.  
  2354.  
  2355. TGSI_SEMANTIC_PCOORD
  2356. """"""""""""""""""""
  2357.  
  2358. Only available if PIPE_CAP_TGSI_TEXCOORD is exposed !
  2359.  
  2360. Fragment shader inputs may be labeled with TGSI_SEMANTIC_PCOORD to indicate
  2361. that the register contains sprite coordinates in the form (x, y, 0, 1), if
  2362. the current primitive is a point and point sprites are enabled. Otherwise,
  2363. the contents of the register are undefined.
  2364.  
  2365. The intended use case for this semantic is gl_PointCoord.
  2366.  
  2367.  
  2368. TGSI_SEMANTIC_GENERIC
  2369. """""""""""""""""""""
  2370.  
  2371. All vertex/fragment shader inputs/outputs not labeled with any other
  2372. semantic label can be considered to be generic attributes.  Typical
  2373. uses of generic inputs/outputs are texcoords and user-defined values.
  2374.  
  2375.  
  2376. TGSI_SEMANTIC_NORMAL
  2377. """"""""""""""""""""
  2378.  
  2379. Indicates that a vertex shader input is a normal vector.  This is
  2380. typically only used for legacy graphics APIs.
  2381.  
  2382.  
  2383. TGSI_SEMANTIC_FACE
  2384. """"""""""""""""""
  2385.  
  2386. This label applies to fragment shader inputs only and indicates that
  2387. the register contains front/back-face information of the form (F, 0,
  2388. 0, 1).  The first component will be positive when the fragment belongs
  2389. to a front-facing polygon, and negative when the fragment belongs to a
  2390. back-facing polygon.
  2391.  
  2392.  
  2393. TGSI_SEMANTIC_EDGEFLAG
  2394. """"""""""""""""""""""
  2395.  
  2396. For vertex shaders, this sematic label indicates that an input or
  2397. output is a boolean edge flag.  The register layout is [F, x, x, x]
  2398. where F is 0.0 or 1.0 and x = don't care.  Normally, the vertex shader
  2399. simply copies the edge flag input to the edgeflag output.
  2400.  
  2401. Edge flags are used to control which lines or points are actually
  2402. drawn when the polygon mode converts triangles/quads/polygons into
  2403. points or lines.
  2404.  
  2405.  
  2406. TGSI_SEMANTIC_STENCIL
  2407. """""""""""""""""""""
  2408.  
  2409. For fragment shaders, this semantic label indicates that an output
  2410. is a writable stencil reference value. Only the Y component is writable.
  2411. This allows the fragment shader to change the fragments stencilref value.
  2412.  
  2413.  
  2414. TGSI_SEMANTIC_VIEWPORT_INDEX
  2415. """"""""""""""""""""""""""""
  2416.  
  2417. For geometry shaders, this semantic label indicates that an output
  2418. contains the index of the viewport (and scissor) to use.
  2419. Only the X value is used.
  2420.  
  2421.  
  2422. TGSI_SEMANTIC_LAYER
  2423. """""""""""""""""""
  2424.  
  2425. For geometry shaders, this semantic label indicates that an output
  2426. contains the layer value to use for the color and depth/stencil surfaces.
  2427. Only the X value is used. (Also known as rendertarget array index.)
  2428.  
  2429.  
  2430. TGSI_SEMANTIC_CULLDIST
  2431. """"""""""""""""""""""
  2432.  
  2433. Used as distance to plane for performing application-defined culling
  2434. of individual primitives against a plane. When components of vertex
  2435. elements are given this label, these values are assumed to be a
  2436. float32 signed distance to a plane. Primitives will be completely
  2437. discarded if the plane distance for all of the vertices in the
  2438. primitive are < 0. If a vertex has a cull distance of NaN, that
  2439. vertex counts as "out" (as if its < 0);
  2440. The limits on both clip and cull distances are bound
  2441. by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
  2442. the maximum number of components that can be used to hold the
  2443. distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
  2444. which specifies the maximum number of registers which can be
  2445. annotated with those semantics.
  2446.  
  2447.  
  2448. TGSI_SEMANTIC_CLIPDIST
  2449. """"""""""""""""""""""
  2450.  
  2451. When components of vertex elements are identified this way, these
  2452. values are each assumed to be a float32 signed distance to a plane.
  2453. Primitive setup only invokes rasterization on pixels for which
  2454. the interpolated plane distances are >= 0. Multiple clip planes
  2455. can be implemented simultaneously, by annotating multiple
  2456. components of one or more vertex elements with the above specified
  2457. semantic. The limits on both clip and cull distances are bound
  2458. by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_COUNT define which defines
  2459. the maximum number of components that can be used to hold the
  2460. distances and by the PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT
  2461. which specifies the maximum number of registers which can be
  2462. annotated with those semantics.
  2463.  
  2464.  
  2465. Declaration Interpolate
  2466. ^^^^^^^^^^^^^^^^^^^^^^^
  2467.  
  2468. This token is only valid for fragment shader INPUT declarations.
  2469.  
  2470. The Interpolate field specifes the way input is being interpolated by
  2471. the rasteriser and is one of TGSI_INTERPOLATE_*.
  2472.  
  2473. The CylindricalWrap bitfield specifies which register components
  2474. should be subject to cylindrical wrapping when interpolating by the
  2475. rasteriser. If TGSI_CYLINDRICAL_WRAP_X is set to 1, the X component
  2476. should be interpolated according to cylindrical wrapping rules.
  2477.  
  2478.  
  2479. Declaration Sampler View
  2480. ^^^^^^^^^^^^^^^^^^^^^^^^
  2481.  
  2482.    Follows Declaration token if file is TGSI_FILE_SAMPLER_VIEW.
  2483.  
  2484.    DCL SVIEW[#], resource, type(s)
  2485.  
  2486.    Declares a shader input sampler view and assigns it to a SVIEW[#]
  2487.    register.
  2488.  
  2489.    resource can be one of BUFFER, 1D, 2D, 3D, 1DArray and 2DArray.
  2490.  
  2491.    type must be 1 or 4 entries (if specifying on a per-component
  2492.    level) out of UNORM, SNORM, SINT, UINT and FLOAT.
  2493.  
  2494.  
  2495. Declaration Resource
  2496. ^^^^^^^^^^^^^^^^^^^^
  2497.  
  2498.    Follows Declaration token if file is TGSI_FILE_RESOURCE.
  2499.  
  2500.    DCL RES[#], resource [, WR] [, RAW]
  2501.  
  2502.    Declares a shader input resource and assigns it to a RES[#]
  2503.    register.
  2504.  
  2505.    resource can be one of BUFFER, 1D, 2D, 3D, CUBE, 1DArray and
  2506.    2DArray.
  2507.  
  2508.    If the RAW keyword is not specified, the texture data will be
  2509.    subject to conversion, swizzling and scaling as required to yield
  2510.    the specified data type from the physical data format of the bound
  2511.    resource.
  2512.  
  2513.    If the RAW keyword is specified, no channel conversion will be
  2514.    performed: the values read for each of the channels (X,Y,Z,W) will
  2515.    correspond to consecutive words in the same order and format
  2516.    they're found in memory.  No element-to-address conversion will be
  2517.    performed either: the value of the provided X coordinate will be
  2518.    interpreted in byte units instead of texel units.  The result of
  2519.    accessing a misaligned address is undefined.
  2520.  
  2521.    Usage of the STORE opcode is only allowed if the WR (writable) flag
  2522.    is set.
  2523.  
  2524.  
  2525. Properties
  2526. ^^^^^^^^^^^^^^^^^^^^^^^^
  2527.  
  2528.  
  2529.   Properties are general directives that apply to the whole TGSI program.
  2530.  
  2531. FS_COORD_ORIGIN
  2532. """""""""""""""
  2533.  
  2534. Specifies the fragment shader TGSI_SEMANTIC_POSITION coordinate origin.
  2535. The default value is UPPER_LEFT.
  2536.  
  2537. If UPPER_LEFT, the position will be (0,0) at the upper left corner and
  2538. increase downward and rightward.
  2539. If LOWER_LEFT, the position will be (0,0) at the lower left corner and
  2540. increase upward and rightward.
  2541.  
  2542. OpenGL defaults to LOWER_LEFT, and is configurable with the
  2543. GL_ARB_fragment_coord_conventions extension.
  2544.  
  2545. DirectX 9/10 use UPPER_LEFT.
  2546.  
  2547. FS_COORD_PIXEL_CENTER
  2548. """""""""""""""""""""
  2549.  
  2550. Specifies the fragment shader TGSI_SEMANTIC_POSITION pixel center convention.
  2551. The default value is HALF_INTEGER.
  2552.  
  2553. If HALF_INTEGER, the fractionary part of the position will be 0.5
  2554. If INTEGER, the fractionary part of the position will be 0.0
  2555.  
  2556. Note that this does not affect the set of fragments generated by
  2557. rasterization, which is instead controlled by half_pixel_center in the
  2558. rasterizer.
  2559.  
  2560. OpenGL defaults to HALF_INTEGER, and is configurable with the
  2561. GL_ARB_fragment_coord_conventions extension.
  2562.  
  2563. DirectX 9 uses INTEGER.
  2564. DirectX 10 uses HALF_INTEGER.
  2565.  
  2566. FS_COLOR0_WRITES_ALL_CBUFS
  2567. """"""""""""""""""""""""""
  2568. Specifies that writes to the fragment shader color 0 are replicated to all
  2569. bound cbufs. This facilitates OpenGL's fragColor output vs fragData[0] where
  2570. fragData is directed to a single color buffer, but fragColor is broadcast.
  2571.  
  2572. VS_PROHIBIT_UCPS
  2573. """"""""""""""""""""""""""
  2574. If this property is set on the program bound to the shader stage before the
  2575. fragment shader, user clip planes should have no effect (be disabled) even if
  2576. that shader does not write to any clip distance outputs and the rasterizer's
  2577. clip_plane_enable is non-zero.
  2578. This property is only supported by drivers that also support shader clip
  2579. distance outputs.
  2580. This is useful for APIs that don't have UCPs and where clip distances written
  2581. by a shader cannot be disabled.
  2582.  
  2583.  
  2584. Texture Sampling and Texture Formats
  2585. ------------------------------------
  2586.  
  2587. This table shows how texture image components are returned as (x,y,z,w) tuples
  2588. by TGSI texture instructions, such as :opcode:`TEX`, :opcode:`TXD`, and
  2589. :opcode:`TXP`. For reference, OpenGL and Direct3D conventions are shown as
  2590. well.
  2591.  
  2592. +--------------------+--------------+--------------------+--------------+
  2593. | Texture Components | Gallium      | OpenGL             | Direct3D 9   |
  2594. +====================+==============+====================+==============+
  2595. | R                  | (r, 0, 0, 1) | (r, 0, 0, 1)       | (r, 1, 1, 1) |
  2596. +--------------------+--------------+--------------------+--------------+
  2597. | RG                 | (r, g, 0, 1) | (r, g, 0, 1)       | (r, g, 1, 1) |
  2598. +--------------------+--------------+--------------------+--------------+
  2599. | RGB                | (r, g, b, 1) | (r, g, b, 1)       | (r, g, b, 1) |
  2600. +--------------------+--------------+--------------------+--------------+
  2601. | RGBA               | (r, g, b, a) | (r, g, b, a)       | (r, g, b, a) |
  2602. +--------------------+--------------+--------------------+--------------+
  2603. | A                  | (0, 0, 0, a) | (0, 0, 0, a)       | (0, 0, 0, a) |
  2604. +--------------------+--------------+--------------------+--------------+
  2605. | L                  | (l, l, l, 1) | (l, l, l, 1)       | (l, l, l, 1) |
  2606. +--------------------+--------------+--------------------+--------------+
  2607. | LA                 | (l, l, l, a) | (l, l, l, a)       | (l, l, l, a) |
  2608. +--------------------+--------------+--------------------+--------------+
  2609. | I                  | (i, i, i, i) | (i, i, i, i)       | N/A          |
  2610. +--------------------+--------------+--------------------+--------------+
  2611. | UV                 | XXX TBD      | (0, 0, 0, 1)       | (u, v, 1, 1) |
  2612. |                    |              | [#envmap-bumpmap]_ |              |
  2613. +--------------------+--------------+--------------------+--------------+
  2614. | Z                  | XXX TBD      | (z, z, z, 1)       | (0, z, 0, 1) |
  2615. |                    |              | [#depth-tex-mode]_ |              |
  2616. +--------------------+--------------+--------------------+--------------+
  2617. | S                  | (s, s, s, s) | unknown            | unknown      |
  2618. +--------------------+--------------+--------------------+--------------+
  2619.  
  2620. .. [#envmap-bumpmap] http://www.opengl.org/registry/specs/ATI/envmap_bumpmap.txt
  2621. .. [#depth-tex-mode] the default is (z, z, z, 1) but may also be (0, 0, 0, z)
  2622.    or (z, z, z, z) depending on the value of GL_DEPTH_TEXTURE_MODE.
  2623.