memmov perf bug

[dmpots]

I am looking at the AOCL memcpy/memmov functions and I think I found a perf bug in the memmov function.
In the memmove function there is a check for aligned src and dst so we can use the vmovaps for aligned load/store. But the check is comparing the whole src pointer with the low bits of dst.

uint32_t dst_align = ((size_t)dst & (YMM_SZ - 1));
...
if (!((uintptr_t)src ^ dst_align))

Compare that to the check in memcpy that looks at the low bits of both.

if ((((size_t)src & (YMM_SZ - 1)) | dst_align) == 0)

I think this is going to prevent the aligned path from being taken in memmov. I tried changing it to match the check in memcpy, but I did not see much difference in perf in the micro-benchmarks. Probably worth fixing still though.