search

Tony-X / search-benchmark-game

https://github.com/quickwit-oss/search-benchmark-game
https://tantivy-search.github.io/bench/
MIT License
10 stars 6 forks source link

Does Tantivy also specialize SIMD for aarch64? #17

Open mikemccand opened 1 year ago

mikemccand commented 1 year ago

Arm (aarch64) cpus are becoming more popular/powerful (e.g. AWS Graviton instances) ... does Tantivy also specialize SIMD for Arm CPUs?

mikemccand commented 1 year ago

We could also simply run the current benchmarks on a Graviton2 instance?

fulmicoton commented 1 year ago

We do not use much handrolled simd stuff these days. I assume you mean for decompression? Some contributors added a NEON implementation for bitpacking, although it is not that beneficial.

I'd love to see have graviton figures

Tony-X commented 1 year ago

does Tantivy also specialize SIMD for Arm CPUs?

I don't think so. However, I think we can have some hope for LLVM to figure out using SIMD wherever it finds suitable

fulmicoton commented 1 year ago

@Tony-X see my reply above. https://github.com/quickwit-oss/bitpacking/blob/master/src/bitpacker4x.rs#L71-L120

Tony-X commented 1 year ago

Ah! I missed it. @fulmicoton thanks for the pointer