The system has to run at around 50 fps, so it should not be overly costly in processing power. Here is a short description of what I’m trying to do using CUDA and directX.
- Allocate 2 buffers (pGpu1 and pGpu2) using cudaMalloc
- Load data from system memory to pGpu1
- Process the data, using pGpu1 as input and pGpu2 as output. The processing is defined by a normal CUDA kernel.
- Use the processed data available at pGpu2 as the texture for a directX10 shader resource.
As the size of the da might be changing, I would like to avoid having to create new textures all the time, and also reduce complexity. I can’t figure out if it is possible to simply set the pointer of a 2D texture to the CUDA allocated memory, define the pitch etc. and then simply use it in DX10.
Any help and hints are welcome