Dear All
I have done some matrix calculations in the VS that I want to re utilize in the PS.
The problem I am having is that the slots of my PS buffer are already taken.
Any thoughts on how can I read those computations from the PS previous done in the VS?
Thanks!
Passing Computations from VS to PS
The only way to pass data from the VS to the PS is through interpolants. Is this what you mean when you say that "slots of my PS buffer"?
On certain GPU's having too many interpolants can be bad for performance, so I wouldn't recommend trying to use a lot of them just to save a few computations in the pixel shader. How much data are you trying to pass from the VS to the PS?
On certain GPU's having too many interpolants can be bad for performance, so I wouldn't recommend trying to use a lot of them just to save a few computations in the pixel shader. How much data are you trying to pass from the VS to the PS?
Author
Thanks MJP.
The VS_OUTPUT struct of the vertex shader already uses most of the available TEXCOORD variables and I am looking to pass two matrices over to the pixel shader (i.e. two float4x4 variables).
As I am doing this via shader injection at D3DCompile time, I don't have access to the C++ side which would enable me to possibly pass the variables to the pixel shader otherwise - it has to be done via a vertex-to-pixel shader technique. However, I cannot think of one, apart from using VS_OUTPUT TEXCOORD variables, but in this case they're not enough (two float4x4s require 8 TEXCOORDs and there are 10 used by the vertex shader already).
Any further ideas?
The VS_OUTPUT struct of the vertex shader already uses most of the available TEXCOORD variables and I am looking to pass two matrices over to the pixel shader (i.e. two float4x4 variables).
As I am doing this via shader injection at D3DCompile time, I don't have access to the C++ side which would enable me to possibly pass the variables to the pixel shader otherwise - it has to be done via a vertex-to-pixel shader technique. However, I cannot think of one, apart from using VS_OUTPUT TEXCOORD variables, but in this case they're not enough (two float4x4s require 8 TEXCOORDs and there are 10 used by the vertex shader already).
Any further ideas?
Could you post the matrix computation code from your vertex shader? Then we could get an idea of how to approach your problem :)
It's not possible to use shader constants?
Author
The matrix computation code in the vertex shader (not my code) is quite complex and has many branches depending on material flags etc.
When you say "use shader constants" - I tried to do that:
1) I enabled D3DCOMPILE_ENABLE_BACKWARDS_COMPATIBILITY
2) I placed two global variables "matrix_1", "matrix_2" in the shader code, whereby inside the Vertex Shader, I populate them (right before the return statement) like so:
matrix matrix_1;
matrix matrix_2;
VS_OUTPUT VS( VS_INPUT In )
{
VS_OUTPUT Out = (VS_OUTPUT) 0;
... (other code calculating various things) ...
matrix_1 = calculatedMatrix1;
matrix_2 = calculatedMatrix2;
return Out;
}
float4 PS(VS_OUTPUT In) : SV_TARGET
{
// check variable contents of matrix_1 and matrix_2 - doesn't work...
};
Is that what you had in mind with 'shader constants'?
I also tried to add them to constant buffers used (with the same compatibility flag), populate them in the vertex shader so that the pixel shader will receive values, that didn't work either.
As I said earlier in my posts, TEXCOORD0 to TEXCOORD11 are taken already, so I can't use the 8 TEXCOORDs required for storing 2 float4x4 variables...
Thanks for your help!
Semantics-wise, since D3D10 (SM4) you don't have the restriction to use TEXCOORD for "custom" interpolators anymore. Also you can use almost arbitrary indices, e.g. you can use MYVERYOWNSEMANTIC1024 (Edit: As long as they don't collide with system value semantics).
There's a limit on the total count though. Can't find a doc entry, but I checked with the command line compiler:
For SM4 : 16 interpolators (4 channels)
For SM5 : 32 interpolators (4 channels)
Check if you really need that a lot. E.g. for affine transformations you should only need float3x4. You could also reduce the number if you don't need interpolants (HLSL nointerpolation) and e.g. pack two 16 bit integers into a 32 bit integers and unpack in the PS.
There's a limit on the total count though. Can't find a doc entry, but I checked with the command line compiler:
For SM4 : 16 interpolators (4 channels)
For SM5 : 32 interpolators (4 channels)
Check if you really need that a lot. E.g. for affine transformations you should only need float3x4. You could also reduce the number if you don't need interpolants (HLSL nointerpolation) and e.g. pack two 16 bit integers into a 32 bit integers and unpack in the PS.
Author
Thanks,
that worked just fine - I defined my own interpolators as you said - I was quite unaware that you can do that in D3D10 (and later).
your help was much appreciated!
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
