triton-cuda-elementwise

What problem does it solve?

This tool enables high-performance element-wise computations on CUDA by generating Triton-based kernels, optimizing vectorized operations for activations and broadcasting.

Core Features & Use Cases

• Triton-based kernel generation for elementwise operations (add, mul, relu, sigmoid, tanh, gelu, exp, log, div, sub, sqrt, pow).
• Memory layout optimizations to ensure coalesced, contiguous access and minimized stride overhead.
• Use Cases: accelerates activation functions and element-wise math across large tensors in CUDA workflows, including broadcasting scenarios.

Quick Start

Provide an input tensor and call the elementwise kernel generator to produce a fast Triton-based CUDA kernel.