I have an Trinity APU (A10-5800K) and would like to use both CPU and GPU in one context.
When I use the GPU in a single context, I can specify buffers with CL_MEM_READ_ONLY | CL_MEM_ALLOC_HOST_PTR and the buffer is placed in uncached host memory. This means fast memory access for the GPU (around 25 GB/s) and size constraints beyond GPU dedicated memory. For the other buffer flags with CL_MEM_ALLOC_HOST_PTR I see 6 GB/s which is typical for cached host memory. (all good so far)
When I have CPU and GPU together in one context, I cannot use the uncached memory. I see around 6 GB/s memory bandwidth for all memory flag options that use the host memory. There is no combination which gives me the fast uncached host memory.
Is there a way to use uncached host memory in a shared context or is this not meant to be?
Thanks a lot,