3 ms·
Generic code yields better results than SSE/AVX optimized ones. I wonder why that could be.
by lokopodium 9y ago
Generic code yields better results than SSE/AVX optimized ones. I wonder why that could be.
- eloff 9y agoIt's just replacing extra loads with a bunch of other instructions. In reality loads are cheap (and cached), it turns out to be cheaper than doing permutes to shuffle the vectors around.