Archives Discussions

[macdonc]

I have a  simple copy kernel which copies from array a to b. When using a pixel shader of uint4 types, I divide the domain width by 4. It validates correctly.

How do I properly do this in a compute shader. The domain size is too large. I am getting 15 GB/sec for the copy instead of 60 GB/sec (in pixel shader).

[gaurav_garg]

I am not sure if I understand your question correctly. Are you using domainOffset and domainSize methods to change the domain? Or, you have just reduced the stream size becuase you are using uint4 instead of uint.

You can use both the methods in compute Shader too.

[macdonc]

I just reduced the stream size because I am using uint4.

E.g.

a = NxN (in uint4)

b = NxN (out uint4)

domain width = N/4 domain height = N