WIP: Cycles: Use popcount intrinsic on Windows
AbandonedPublic
Actions

Authored by Thomas Dinges (dingto) on Dec 10 2022, 7:12 PM.

Details

Reviewers

Brecht Van Lommel (brecht)

Summary

According to Stackoverflow, it doesn't matter if __popcnt() or _mm_popcnt_u32() is used, they should be the same.
Popcount isn't part of SSE4.2, but was released around that. To be on the safe side, I only added it for the AVX kernel and above, to avoid some systems without support for it fallback to SSE3.
Some quick benchmarks on Windows with bistro.blend show no difference to the previous code, but since this solved a todo it might be still worth adding. GCC should already use the intrinsic via __builtin_popcount() when compiled with sse4 but worth double checking.