Lightning Talk: Triton Compiler
summary
Triton is a Python DSL and compiler that generates GPU kernels shown to reach near‑CUDA performance with far less code, and it ships as a fundamental part of Torch inductor. The compiler maps tensor layouts to threads, applies coalescing and target‑specific lowering to LLVM, and supports autotuning knobs for block sizes and memory tiling. It enables more fusion, smaller memory footprints, and retuning rather than full rewrites for different GPUs.
AI-written summary · about
chapters (8)
- 0:00Triton Compiler intro
- 1:49Why Triton created
- 3:13Design philosophy
- 4:36Softmax kernel example
- 6:20Integration in ML stacks
- 8:07Compiler architecture
- 10:15Simplifying kernel dev
- 11:41Q&A hardware targets
related talks
05mLightning Talk: GPU Programming with C++ and Triton
08mGPU Programming with Triton Kernels
28mHelion: A High-level DSL for Kernel Authoring
25mLeveraging AI Ecosystem for Portable and Sustainable GPU Kernels in HPC
10mMaximizing Kernel Development Productivity Under Performance Constraints
34mBridging the Gap: Compiling and Optimizing Triton Kernels Onto RISC-V Targets Based on MLIR
10mLightning Talk: Faster Than SOTA Kernels in Torch.compile With Subgrap...
24mHelion 1.0: A High-Level DSL for Performance Portable Kernels