Capture CUDA/OpenCL memory transfert to account them

[svalat]

trafficstars

When profiling an appliacation using CUDA and OpenCL we might be interesting to also account the memory transfert NUMA effects.

Might be easy just by capturing the CUDA/OpenCL memory transfert calls (at least as a first step).

[svalat]

Ok, this is a more vast question than what I originally thought because the transfert will be done by kernel. But this is a really interesting question.