As a SIMD noob, one thing that wasn't obvious to me is that SIMD can also speed you up if your mem throughput is underutilized by having the CPU load more data per instruction. It isn't just about compute speedups, which is typically what it's advertised for. Using perf on Linux has been very educational for me to get an intuition for modern CPU performance.
Cloudef•16m ago
The way simd speeds up "compute" is indeed mainly the reason that you operate on multiple pieces of data at once.
grg0•8m ago
Yeah, and in the blog post he mentions that he had to transform the data to SoA. If he had done that alone, he might already have seen a speedup from better cache utilization.
Also, I see no mention of alignment in the post. I understand x86/AVX2 likes your load/stores to be aligned, even if it technically allows unaligned access.
brcmthrowaway•1h ago
tom_•53m ago