cublas-rs: A BLAS Library Built on cuda-oxide

[zTgx]

Hi everyone! I'd like to share cublas-rs, a Rust BLAS implementation built on top of cuda-oxide.

Overview

• Repository: https://github.com/zTgx/cublas-rs

Project Layout

Crates:

• cublas-core: Host-side types and configuration
• cublas-l1: BLAS Level 1 ops (saxpy, scal, copy, dot, nrm2, …)
• cublas-l2: BLAS Level 2 ops (gemv, trsv, symv)
• cublas-l3: BLAS Level 3 ops (gemm, batched gemm)
• cublas-rs: Public API surface (Handle)

Examples:

• hello: cuda-oxide smoke test
• saxpy: L1 end-to-end demo
• sgemm: performance comparison (SGEMM / DGEMM / batched / half-precision)
• mlp: inference workflow
• linreg: linear regression solver
• gemm_bench: cross-GPU benchmark

Status

This library is still under active development. It works for learning and experiments.

If you're interested, feel free to:

• Browse the code and examples
• Try it out, or use it as a learning reference
• Share feedback and suggestions

Looking forward to your thoughts!

[nihalpasham]

Awesome work and this was quick! 🚀

I haven't had the chance to take a closer look at the repo but just a couple of things I'd love to hear more about:

• How has the experience been overall?
• Have you put any of this head-to-head with NV cuBLAS?

Again, thanks for putting this out in the open. Looking forward to following along.

[zTgx]

Honestly, I'm still a beginner with CUDA. I'm building cublas-rs while learning (cuda-oxide / cuBLAS) at the same time, and I'll keep updating it — as my understanding deepens, cublas-rs will keep getting more complete.

That said, cuda-oxide has been a real pleasure to work with.