CODEWITHSUNDEEP
cudaopencl
gpuguy
gpuguy

Reputation: 4585

Coalesced memory access to global memory; CUDA Vs Opencl

Are there different set of requirements for coalesced access for CUDA and OpenCL? I think it has to be hardware feature,so the API should really not matter. Any pointers??

Upvotes: 2

Views: 497

Answers (1)

Jonathan Dursi
Jonathan Dursi

Reputation: 50927

That's correct. CUDA iteself doesn't have any requirements for coalesced access; the hardware does, and the hardware has the same requirements regardless of whether you're programming it in CUDA, OpenCL, DirectCompute, OpenGL, or something else. On CPU hardware the same could be said about cache behaviour; cache line sizes and cache behaviour don't care if you're using C, Java, or Python.

Upvotes: 1

Related Questions