AGAL (Adobe Graphics Assembly Language) 记录

《Adobe图形汇编语言》
来自官方参考文档，记录一下，方便查看..
———————
Program bytecode can be created using the Pixel Bender 3D offline tools. It can also be created dynamically. This specification describes the raw bytecode format.

The input ByteArrays are in Endian.LITTLE_ENDIAN format.

TypeDescription
int88 bit little endian integer
int1616 bit little endian integer
int3232 bit little endian integer
Header: Bytecode always begins with this 7 byte header:

Offset(Bytes)Size(Bytes)NameDescription
01magicmust be 0xa0
14versionmust be 1
51shadertypeidmust be 0xa1
61shadertypeeither 0 for vertex or 1 for fragment programs

Tokens: The header is immediately followed by any number of tokens. Every
token is 192bits (24 bytes) in size and always has the format:

[opcode][destination][source1][source2 or sampler]
[操作码] [目标] [source1] [source2或采样器]
Unused fields must be set to 0.

[opcode]
The [opcode] field is 32 bits in size and can take one of these values:

Name  Value  Description
名称    值       说明
mov   0x00   move           move data from source1 to destination, componentwise
add   0x01   add            destination = source1 + source2, componentwise
sub   0x02   subtract       destination = source1 - source2, componentwise
mul   0x03   multiply       destination = source1 * source2, componentwise
div   0x04   divide         destination = source1 / source2, componentwise
rcp   0x05   reciprocal     destination = 1/source1, componentwise
min   0x06   minimum        destination = minimum(source1,source2), componentwise
max   0x07   maximum        destination = maximum(source1,source2), componentwise
frc   0x08   fractional     destination = source1 - (float)floor(source1), componentwise
sqt   0x09   square root    destination = sqrt(source1), componentwise
rsq   0x0a   recip. root    destination = 1/sqrt(source1), componentwise
pow   0x0b   power          destination = pow(source1,source2), componentwise
log   0x0c   logarithm      destination = log_2(source1), componentwise
exp   0x0d   exponential    destination = 2^source1, componentwise
nrm   0x0e   normalize      destination = normalize(source1), componentwise
sin   0x0f   sine           destination = sin(source1), componentwise
cos   0x10   cosine         destination = cos(source1), componentwise
crs   0x11   cross product  destination.x = source1.y * source2.z - source1.z * source2.y
destination.y = source1.z * source2.x - source1.x * source2.z
destination.z = source1.x * source2.y - source1.y * source2.x
dp3   0x12   dot product    destination = source1.x*source2.x + source1.y*source2.y + source1.z*source2.z
dp4   0x13   dot product    destination = source1.x*source2.x + source1.y*source2.y + source1.z*source2.z + source1.w*source2.w
abs   0x14   absolute       destination = abs(source1), componentwise
neg   0x15   negate         destination = -source1, componentwise
sat   0x16   saturate       destination = maximum(minimum(source1,1),0), componentwise
m33   0x17   multiply       destination.x = (source1.x * source2[0].x) + (source1.y * source2[0].y) + (source1.z * source2[0].z)
matrix 3x3     destination.y = (source1.x * source2[1].x) + (source1.y * source2[1].y) + (source1.z * source2[1].z)
destination.z = (source1.x * source2[2].x) + (source1.y * source2[2].y) + (source1.z * source2[2].z)
m44   0x18   multiply       destination.x = (source1.x * source2[0].x) + (source1.y * source2[0].y) + (source1.z * source2[0].z) + (source1.w * source2[0].w)
matrix 4x4     destination.y = (source1.x * source2[1].x) + (source1.y * source2[1].y) + (source1.z * source2[1].z) + (source1.w * source2[1].w)
destination.z = (source1.x * source2[2].x) + (source1.y * source2[2].y) + (source1.z * source2[2].z) + (source1.w * source2[2].w)
destination.w = (source1.x * source2[3].x) + (source1.y * source2[3].y) + (source1.z * source2[3].z) + (source1.w * source2[3].w)
m34   0x19   multiply       destination.x = (source1.x * source2[0].x) + (source1.y * source2[0].y) + (source1.z * source2[0].z) + (source1.w * source2[0].w)
matrix 3x4     destination.y = (source1.x * source2[1].x) + (source1.y * source2[1].y) + (source1.z * source2[1].z) + (source1.w * source2[1].w)
destination.z = (source1.x * source2[2].x) + (source1.y * source2[2].y) + (source1.z * source2[2].z) + (source1.w * source2[2].w)
kil   0x27   kill / discard (fragment shader only)
If single scalar source component is less than zero, fragment is discarded and not drawn to the frame buffer.
The destination register must be all 0.
tex   0x28   texture sample (fragment shader only)
destination = load from texture source2 at coordinates source1. In this source2 must be in sampler format.
sge   0x29   set-if-greater-equal
destination = source1 >= source2 ? 1 : 0, componentwise
slt   0x2a   set-if-less-than


destination = source1 < source2 ? 1 : 0, componentwise

[destination]

The [destination] field is 32 bits in size:
31………………………..0
—-TTTT—-MMMMNNNNNNNNNNNNNNNN
T = Register type (4 bits)
M = Write mask (4 bits)
N = Register number (16 bits)
- = undefined, must be 0

[source]

The [source] field is 64 bits in size:
63…………………………………………………….0
D————-QQ—-IIII—-TTTTSSSSSSSSOOOOOOOONNNNNNNNNNNNNNNN
D = Direct=0/Indirect=1 for direct Q and I are ignored, 1bit (currently only direct allowed)
Q = Index register component select (2 bits)
I = Index register type (4 bits)
T = Register type (4 bits)
S = Swizzle (8bits, 2bits per component)
O = Indirect offset (8bits)
N = Register number (16 bits)
- = undefined, must be 0

[sampler]

The second source of the tex opcode must be in [sampler] format:
63…………………………………………………….0
FFFFMMMMWWWWSSSSDDDD——–TTTT—————-NNNNNNNNNNNNNNNN
N = Sampler index (16 bits)
T = Register type, must be 5, Sampler
F = Filter (0=nearest,1=linear) (4bits)
M = Mipmap (0=disable,1=nearest,2=linear)
W = Wrapping (0=clamp,1=repeat)
S = Special flag bits (must be 0)
D = Dimension (0=2D,1=Cube)

The following register types are available:

Name      Value   Count/Fragment  Count/Vertex    Usage
Attribute 0       n/a             8               input using Context3D::setVertexBufferAt
Constant  1       28              128             input using Context3D::setProgramConstants family of functions
Temporary 2       8               8               temporary for computation, not visible outside program
Output    3       1               1               vertex: clipspace position, fragment: color
Varying   4       8               8               output from vertex, input to fragment, transfer interpolated data
Sampler   5       8               n/a             read texture, set using Context3D::setTextureAt