matrixmultiply

General matrix multiplication for f32, f64 matrices.

Allows arbitrary row, column strided matrices.

Uses the same microkernel algorithm as BLIS, but in a much simpler and less featureful implementation. See their multithreading page for a very good diagram over how the algorithm partitions the matrix (Note: this crate does not implement multithreading).

Please read the API documentation here

Blog posts about this crate:

• A Gemmed Rabbit Hole

NOTE: Compile this crate using RUSTFLAGS="-C target-cpu=native" so that the compiler can produce the best output.

Recent Changes

• 0.1.14
  • Avoid an unused code warning
• 0.1.13
  • Pick 8x8 sgemm (f32) kernel when AVX target feature is enabled (with Rust 1.14 or later, no effect otherwise).
  • Use rawpointer, a µcrate with raw pointer methods taken from this project.
• 0.1.12
  • Internal cleanup with retained performance
• 0.1.11
  • Adjust sgemm (f32) kernel to optimize better on recent Rust.
• 0.1.10
  • Update doc links to docs.rs
• 0.1.9
  • Workaround optimization regression in rust nightly (1.12-ish) (#9)
• 0.1.8
  • Improved docs
• 0.1.7
  • Reduce overhead slightly for small matrix multiplication problems by using only one allocation call for both packing buffers.
• 0.1.6
  • Disable manual loop unrolling in debug mode (quicker debug builds)
• 0.1.5
  • Update sgemm to use a 4x8 microkernel (“still in simplistic rust”), which improves throughput by 10%.
• 0.1.4
  • Prepare support for aligned packed buffers
  • Update dgemm to use a 8x4 microkernel, still in simplistic rust, which improves throughput by 10-20% when using AVX.
• 0.1.3
  • Silence some debug prints
• 0.1.2
  • Major performance improvement for sgemm and dgemm (20-30% when using AVX). Since it all depends on what the optimizer does, I'd love to get issue reports that report good or bad performance.
  • Made the kernel masking generic, which is a cleaner design
• 0.1.1
  • Minor improvement in the kernel