GPU Programming with Triton Kernels
summary
Practical primer on writing Triton GPU kernels with concrete results: shows how kernels use tl.load/tl.store and block-offset indexing for vector add, how a multi-kernel structure handles global reduction, and claims a 70% throughput gain for a GEMV kernel over a PyTorch AMD benchmark. Includes an educational repo of pitfalls from development and experiments using AI-assisted kernel tools like kernelloom and mako for generation, benchmarking and validation.
AI-written summary · about
chapters (5)
- 0:00Introduction to GPU kernels
- 1:38Introduction to Triton
- 2:21Triton code walkthrough
- 5:11Red Hat experience and tools
- 7:12Q&A and closing
related talks
05mLightning Talk: GPU Programming with C++ and Triton
10mMaximizing Kernel Development Productivity Under Performance Constraints
25mLeveraging AI Ecosystem for Portable and Sustainable GPU Kernels in HPC
16mLightning Talk: Triton Compiler
34mBridging the Gap: Compiling and Optimizing Triton Kernels Onto RISC-V Targets Based on MLIR
20mUnlocking Performance: Harnessing LLMs To Streamline GPU Kernel Development in...
22mEnabling VLLM V1 on AMD GPUs With Triton
09mLightning Talk: Vllm-triton-backend: How To Get State-of-the-art Performance on...