Investigate removal of on-device image copies

[felixdoerre]

Currently primus-vk does a image copy on the rendering gpu and one on the displaying gpu. It needs to be checked if those copies are both principally needed or can be removed somehow.

[nariox]

In the meantime, would implementing a AVX or SSE accelerated memcpy improve performance?

I'm actually not too familiar with how memory copies work between the CPU and GPU.

[felixdoerre]

I assume we already use sse. On 64-bit automatically and on 32-bit with something like these:
#37 (comment)
It seems I missed adding the 32-bit trick to the generic Makefile.