Feature: Windows build is not compiled with SimSIMD, f16 support for cosine distance #325
Closed
3 tasks done
Labels
Comments
|
Hi, @monatis! Indeed, that issue exists. I use GCC attributes to annotate function-level compilation settings in SimSIMD. If you can find a way to make them work with MSVC - we will be able to accelerate the Windows build 🤗 |
|
Hi @ashvardanian thanks for the quick response! Yes, I can work on it this weekend. My use case is semantic search on consumer-grade hardware, so every bit of optimization counts. |
|
Hi @monatis! How is it going? Would be happy to merge if its ready 🤗 |
ashvardanian
added a commit
that referenced
this issue
Mar 20, 2024
ashvardanian
pushed a commit
that referenced
this issue
Mar 31, 2024
# [2.10.0](v2.9.2...v2.10.0) (2024-03-31) ### Add * `Index.__repr_pretty__` ([7087138](7087138)) * Introspect Rust builds ([16fa90f](16fa90f)) * Standalone SQLite binary ([227083b](227083b)), closes [#358](#358) ### Chore * 120-char lines in Python ([92cf01e](92cf01e)) ### Docs * Cover cross-compilation ([de1a8e9](de1a8e9)) * Typo - "desnse" -> "dense" (#370) ([719cf7a](719cf7a)), closes [#370](#370) ### Fix * `ascii` name collision in Py ([4ac3509](4ac3509)) * Allow missing metric state ([4b8e1a8](4b8e1a8)) * Avoid dynamic arrays for MSVC ([89d70cd](89d70cd)) * Exceptions for missing distances ([349c02d](349c02d)) * JIT compilation in Python ([a3287f1](a3287f1)) * JS compilation ([cea50ba](cea50ba)) * Missing `download_binary` arg ([64d211a](64d211a)) * Missing include ([6454aae](6454aae)) * mixed-precision tolerance in JS tests ([b72f658](b72f658)) * No `-Wno-vla` MSVC flag ([84e8e1e](84e8e1e)) * No extensions in some SQLite builds ([39264d0](39264d0)) * OpenMP max threads issue ([6fd4a2d](6fd4a2d)), closes [ClickHouse/ClickHouse#61780](ClickHouse/ClickHouse#61780) * Silence false warnings ([ed3a23b](ed3a23b)) ### Improve * bit-casting for negative flaots ([1d03a9d](1d03a9d)) * Clean Python w. Ruff ([87e200b](87e200b)) * Concrete `scalar_kind_t` values ([27d911a](27d911a)) * Handle failing downloads ([8649815](8649815)) * Normalize Pearson as distance ([8a697f7](8a697f7)) * Pass `vectors` to `self_recall` ([565625e](565625e)) * Python evals for exact search ([36f6c5e](36f6c5e)) * Silence warnings ([3d229e0](3d229e0)) ### Make * `CMP0135` CMake policy ([d335df7](d335df7)) * `pypy` and `musl` wheels ([31665d5](31665d5)) * Add SimSIMD dynamic dispatch ([8b8d2f5](8b8d2f5)) * Avoid SimSIMD inner dispatch ([28b0850](28b0850)) * Build SQLite from source ([0dc59a8](0dc59a8)) * CMake pre 3.22 support ([145a4f4](145a4f4)) * Cross-compilation deps ([1de8d7a](1de8d7a)) * Cross-compile with Clang ([94f97cb](94f97cb)), closes [#322](#322) * Disable SQLite extension for Windows ([c612cc2](c612cc2)) * Fetch local or remote SQLite builds ([f6b75e0](f6b75e0)) * Format Py lines to 120 columns ([5b5973c](5b5973c)) * Lantern build issues ([a400742](a400742)), closes [/github.com/unum-cloud/usearch/blob/4747ef42f4140a1fde16118f25f079f9af79649e/include/usearch/index_plugins.hpp#L43-L45](https://github.com//github.com/unum-cloud/usearch/blob/4747ef42f4140a1fde16118f25f079f9af79649e/include/usearch/index_plugins.hpp/issues/L43-L45) [/github.com/unum-cloud/usearch/blob/ce54b814a8a10f4c0c32fee7aad9451231b63f75/include/usearch/index_plugins.hpp#L50](https://github.com//github.com/unum-cloud/usearch/blob/ce54b814a8a10f4c0c32fee7aad9451231b63f75/include/usearch/index_plugins.hpp/issues/L50) * Prebuild JS binaries (#342) ([267fbb5](267fbb5)), closes [#342](#342) [#336](#336) * Pull submodules before cross-comp ([24b3529](24b3529)) * Read UTF8 README on Windows ([a49d17e](a49d17e)) * shared libs on MacOS are `.dylib` ([fd94aa8](fd94aa8)) * SimSIMD in Windows Python wheels ([ecb4948](ecb4948)), closes [#325](#325) * SQlite as shared library ([3db3d33](3db3d33)) * Take SQLite source for cross-compilation ([95b260f](95b260f)) * Upgrade SimSIMD ([908df0d](908df0d)) * Upgrade SimSIMD to 4.2.2 ([9f53f15](9f53f15)) * Use AVX in Windows Python images ([c694880](c694880)), closes [#325](#325)
|
🎉 This issue has been resolved in version 2.10.0 🎉 The release is available on GitHub release Your semantic-release bot 📦🚀 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Describe what you are looking for
The pip-installable package for Windows seems to be not compiled with SimSIMD (can be seen in
index.specs), which makes it ~3-10x slower than the Linux build.Additionally, It seems that cosine distance supports AVX512 with
dtype='f32', but it fallbacks to AVX2 withdtype='f16'. So we need to make a choice between optimization for speed or optimization for storage. It would be great to support AVX512 with float16 as well and of course, to have SimSIMD support in pip-installable package for Windows.I can contribute to the implementation if there is no known limitation for these.
Can you contribute to the implementation?
Is your feature request specific to a certain interface?
It applies to everything
Contact Details
[email protected]
Is there an existing issue for this?
Code of Conduct
The text was updated successfully, but these errors were encountered: