zlacker

Not really related to golang but: AVX512 is THE sweet spot to go for that for vector instructions as 512bits=64bytes=x86_64 cache line size.

>>sylwar+(OP)
Dumb question but do you have to do anything manually to ensure it’s actually aligned to a single cache line or does it just happen because of the way things work?

>>zamada+9n
Not a dumb question. In theory, yes you want it to be aligned for performance reasons, and you can do this with one of the _aligned_malloc()/aligned_alloc() variants. In reality though, for any non-trivial algorithm, you may not be able to always enforce this.

>>stagge+Rr
Usually you alloc some slack memory to be able to re-align on a cache line. But yeah, it seems better if the allocator can cleanely deal with that.