kaivalnp commented on PR #14178: URL: https://github.com/apache/lucene/pull/14178#issuecomment-2835496582
> One big part of the complexity of building these packages is the low-level optimizations based on which "bare metal" you are running on > but how do I know whichever I pick is optimized properly for my particular env? These artifacts are created by the scripts in [this folder](https://github.com/facebookresearch/faiss/tree/main/conda/faiss) which are separate based on CPU architecture (like OSX, Windows, Arm, Intel, etc. -- see [these files](https://anaconda.org/pytorch/faiss-cpu/files) for different artifacts in Conda) All builds enable standard [release flags](https://github.com/facebookresearch/faiss/blob/ea1cdf0fb28f31ab36ea95d8d65abfcc184d1b8f/conda/faiss/build-lib.sh#L20), and produce libraries with optimized instructions like AVX2, AVX512, etc. as well (in this PR we're not using the optimized libraries yet, but I've added a TODO -- it's a matter of picking the library with the highest optimization level available at runtime) > maybe we can help debug those builds to nudge things forwards! Would be great to have everything on `conda-forge`! As for the build issues there, looks like some of [these patches](https://github.com/conda-forge/faiss-split-feedstock/tree/main/recipe/patches) are outdated -- I could try to modify those for the latest version but I'm not sure if some of those aren't required / some others are missing (not super familiar with build intricacies of Faiss / `conda-forge`) Taking a step back, I wonder if having everything on `conda-forge` is good-to-have, but the libraries on `pytorch` are still usable in the short term? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
