Reputation: 11916
(Assuming there are many avx instructions before and after movd)If I use vmovd to move data between general purpose registers and ymm registers, does it get slower because of using only 1 float value of ymm?
Upvotes: 0
Views: 273
Reputation: 93082
The avx-sse transition penalty only applies when you execute non-VEX encoded SSE instructions and SSE-encoded instructions. As long as you use vmovd
instead of, say, movd
, everything should be fine.
Upvotes: 2