Tags: ashvardanian/SimSIMD
Tags
Build: Released 4.2.0 [skip ci] # [4.2.0](v4.1.1...v4.2.0) (2024-03-25) ### Add * Compile-time dispatch ([b72bf56](b72bf56)) ### Fix * Rust benchmarks ([5279de7](5279de7)) * Setting signaling NaN ([4e6e686](4e6e686)) ### Make * Build with SIMD in Rust ([391545e](391545e)) * Sync `VERSION` for NPM ([7ea0ea6](7ea0ea6))
Build: Released 4.1.0 [skip ci] # [4.1.0](v4.0.0...v4.1.0) (2024-03-23) ### Add * Bench against TF, PyTorch, JAX ([5e64152](5e64152)) * Complex dot-products for Rust ([936fe73](936fe73)) * Double-precision interfaces for JS ([07f2aca](07f2aca)) ### Docs * List all APIs ([0a987a3](0a987a3)) ### Fix * Avoid `vld2_f16` on MSVC ([ce9800e](ce9800e)) * Complex dispatch in Rust & C ([d349bcd](d349bcd)) * Missing `_mm_rsqrt14_ps` in MSVC ([21e30fe](21e30fe)) * Missing `float16_t` in MSVC Arm64 builds ([94442c3](94442c3)) ### Improve * Pragmas for MSVC compatibility ([9d8a8d0](9d8a8d0)), closes [#74](#74) * Silence TF warnings ([0cffc9c](0cffc9c)) * Type-casting ([353fe43](353fe43)) ### Make * 120 line width ([35c1de3](35c1de3)) * Enable SIMD in MSVC builds ([27f24ef](27f24ef)) * Missing Node & TS dependencies ([9983167](9983167))
Build: Released 4.0.0 [skip ci] # [4.0.0](v3.9.0...v4.0.0) (2024-03-13) ### Add * `complex32` benchmarks ([edba699](edba699)) * `dot_f64c_serial` benchmark ([7f6aabf](7f6aabf)) * API declarations ([1f05820](1f05820)) * Complex dot-products ([14c2302](14c2302)) * Complex f16 dot-products ([d92704a](d92704a)) * Dynamic dispatch for complex backends ([b6d15b1](b6d15b1)) * Haswell and BLAS baseline for dot-products ([0646e5e](0646e5e)) * SkyLake backends for complex dot-products ([2844182](2844182)) * SVE for Complex Dot ([052e40e](052e40e)) ### Break * Full library remodelling ([c3bba95](c3bba95)) * New Rust API ([21dfaf7](21dfaf7)) * Rename `OutputDistances` to `DistancesTensor` ([c09d895](c09d895)) ### Docs * Update for v4 ([97f4b4e](97f4b4e)) ### Fix * `__AVX512BITALG__` detection ([0edcdfa](0edcdfa)) * Downgrade FMLA to FP16 ([f37fc09](f37fc09)) * Duplicate `cos_i8_neon` ([2b5bcfe](2b5bcfe)) * Inner product PyTest ([a2191aa](a2191aa)) * Passing JS tests ([fbd7e02](fbd7e02)) * Skip complex tests in QEMU ([a8eaf83](a8eaf83)) * Take the size of the bigger matrix ([7c4b5df](7c4b5df)) ### Improve * `dot_f32c` on Haswell from 2.5 GB/s to 5 ([bf1940d](bf1940d)) * Fine-grained AVX-512 compilation ([c0f29f6](c0f29f6)) ### Make * Add PyPi classifiers ([31a5b09](31a5b09)) * Rename `complex` to `dot` ([ffee893](ffee893))
PreviousNext