lance-format / lance-format/lance
[Python, Benchmarks] How could we ship binary wheel with different CPU targets
Open
@chebbyChefNEQ is already working on this.
Since Aug 23, 2023.
A-python
benchmark
- Dominant language
- Rust
- Stars
- 7.1k
- Forks
- 852
- Avg merge
- 3d 18h
- Merged PRs (30d)
- 272
Description
Problem Statement
The default target-cpu of our pypi release is --target-cpu=haswell to enable AVX2 and FMA code path. However, @chebbyChefNEQ has found that running --target-cpu=native on a Zen 4 (7950X?) platform leads to more than 10X speed up over vector indexing performance, and would expect non-trivial speed up in query path as well.
Desired Behavior
If this speed gap can not be bridged by feature flag (#[cfg=...]) in rust, we need to think about how we release binaries that can take advantage of the CPU users have, and dont leave performance on the table.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.