A high-performance BLAS (Basic Linear Algebra Subprograms) implementation written in Mojo.

mojoBLAS is a pure-Mojo BLAS implementation focused on performance. It currently includes:

• Level 1 BLAS: vector-vector operations such as dot, axpy, nrm2, scal, and more.
• Level 2 BLAS: matrix-vector operations such as gemv, ger, triangular and packed matrix-vector routines.
• Level 3 BLAS: matrix-matrix operations such as gemm, syrk, syr2k, symm, trmm, and trsm.
• Benchmarking suite: comparison against reference/system BLAS implementations.

The codebase is currently optimized for real scalar data types through Mojo DType support.

• Pixi
• Mojo >=1.0.0b1,<2

mojoBLAS is available in the modular-community https://repo.prefix.dev/modular-community package repository. Add the following to your channels list in your pixi.toml file:

channels = ["https://conda.modular.com/max", "https://repo.prefix.dev/modular-community", "conda-forge"]

Then, you can install mojoBLAS using any of these methods:

1. From the pixi CLI, run the command pixi add mojoblas.

2. In the pixi.toml file of your project, add the following dependency:

   mojoblas = "==0.1.0"

Then run pixi install to download and install the package.

Add the repository to your pixi.toml:

[workspace]
preview = ["pixi-build"]

[dependencies]
mojo = ">=1.0.0b1,<2"
mojoblas = { git = "https://github.com/shivasankarka/mojoBLAS.git", branch = "main" }

Then run:

pixi install

git clone https://github.com/shivasankarka/mojoBLAS.git
cd mojoBLAS
pixi install

from mojoblas.src.level1 import dot, axpy, nrm2

fn main():
    var x = alloc[Float32](3)
    var y = alloc[Float32](3)

    x[0] = 1.0
    x[1] = 2.0
    x[2] = 3.0
    y[0] = 4.0
    y[1] = 5.0
    y[2] = 6.0

    print(dot(3, x, 1, y, 1))
    axpy(3, 2.0, x, 1, y, 1)
    print(y[0], y[1], y[2])
    print(nrm2(3, x, 1))

    x.free()
    y.free()

• Level 1: asum, axpy, copy, dot, iamax, nrm2, rot, rotg, rotm, rotmg, scal, swap
• Level 2: gbmv, gemv, ger, sbmv, spmv, spr, spr2, symv, syr, syr2, tbmv, tbsv, tpmv, tpsv, trmv, trsv
• Level 3: gemm, symm, syrk, syr2k, trmm, trsm

Run the test suites with Pixi:

pixi run test_level1
pixi run test_level2
pixi run test_level3

The repository includes benchmark scripts. This benchmark compares mojoblas against general openblas and Accelerate (on Apple M chips) routines. To run the full benchmarks and generate plots, run the following command

pixi run -e bench bench_all

• benchmarks/bench_plot_level1.png
• benchmarks/bench_plot_level2.png
• benchmarks/bench_plot_level3.png

• src/ - Mojo source for BLAS implementations
• tests/ - Mojo tests and reference data
• benchmarks/ - benchmark scripts and plots
• docs/ - Reference documentation.

• Level 1 BLAS
• Level 2 BLAS
• Level 3 BLAS
• Benchmarking suite

• Optimize current algorithms (Goal: openblas, accelerate performance and more :))
• Complex number support
• GPU acceleration

Contributions are welcome. If you find a bug or performance issue, please open an issue or submit a pull request.

This project is licensed under the MIT License. See LICENSE for details.

This project is inspired by the Netlib BLAS reference implementation:

http://www.netlib.org/blas/

Special thanks to the Mojo and BLAS communities for the tools and ideas that made this project possible.