Crate oxicuda_blas 

OxiCUDA BLAS — GPU-Accelerated BLAS Operations

This crate provides GPU-accelerated Basic Linear Algebra Subprograms (BLAS), serving as a pure Rust equivalent to cuBLAS.

Quick start

use std::sync::Arc;
use oxicuda_driver::Context;
use oxicuda_blas::handle::BlasHandle;

let handle = BlasHandle::new(&ctx)?;
// ... call BLAS routines via the handle ...

Re-exports

pub use algorithm_selection::AlgorithmConfig;

pub use algorithm_selection::AlgorithmHeuristic;

pub use algorithm_selection::AlgorithmId;

pub use algorithm_selection::AlgorithmSelector;

pub use algorithm_selection::EpiloguePreference;

pub use algorithm_selection::SwizzleMode;

pub use error::BlasError;

pub use error::BlasResult;

pub use handle::BlasHandle;

pub use types::DiagType;

pub use types::E4M3;

pub use types::E5M2;

pub use types::FillMode;

pub use types::GpuFloat;

pub use types::Layout;

pub use types::MathMode;

pub use types::MatrixDesc;

pub use types::MatrixDescMut;

pub use types::PointerMode;

pub use types::Side;

pub use types::Transpose;

pub use types::VectorDesc;

Modules

algorithm_selection

cuBLASLt-style algorithm selection API for GEMM kernel dispatch.

batched

Batched GEMM operations.

complex_gemm

Complex-number GEMM and GEMV (CGEMM / ZGEMM / CGEMV / ZGEMV).

elementwise

Elementwise GPU operations for OxiCUDA BLAS.

error

Error types for OxiCUDA BLAS operations.

handle

BLAS handle management.

level1

BLAS Level 1 — vector-vector operations.

level2

BLAS Level 2 – matrix-vector operations.

level3

BLAS Level 3 — matrix-matrix operations.

precision

Precision-specific GEMM optimizations and configuration.

prelude

Convenience re-exports for common BLAS usage.

reduction

Reduction operations for OxiCUDA BLAS.

types

Common types shared across BLAS operations.