tropical-gemm

High-performance tropical matrix multiplication in Rust with SIMD and CUDA backends.

Features

• Multiple Semirings: MaxPlus, MinPlus, MaxMul
• SIMD Acceleration: AVX-512, AVX2, SSE4.1, NEON auto-detection
• CUDA Backend: GPU-accelerated kernels via NVRTC
• Argmax Tracking: For backpropagation in tropical neural networks
• Python Bindings: NumPy and PyTorch integration

Installation

[dependencies]
tropical-gemm = "0.1"
tropical-gemm-cuda = "0.1"  # Optional GPU support

Quick Start

use tropical_gemm::{Mat, MaxPlus};

let a = Mat::<MaxPlus<f32>>::from_row_major(&[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], 2, 3);
let b = Mat::<MaxPlus<f32>>::from_row_major(&[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], 3, 2);

// C[i,j] = max_k(A[i,k] + B[k,j])
let c = a.matmul(&b);
assert_eq!(c.get_value(0, 0), 8.0); // max(1+1, 2+3, 3+5) = 8

Python

import numpy as np
import tropical_gemm

a = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)
b = np.array([[1.0, 2.0], [3.0, 4.0]], dtype=np.float32)

c = tropical_gemm.maxplus_matmul(a, b)

Documentation

📖 User Guide - Installation, tutorials, examples

📚 API Reference - Rust API documentation

Semirings

Performance

License

MIT