Lightning Talk: GPU Programming with C++ and Triton
summary
Proposes using Triton Python DSL to write GPU kernels that compile to a CUDA binary loaded from C++, showing a position-update kernel for agent based simulations and how the Triton compiler applies automatic memory coalescing plus shared memory, prefetch and tensor core use so developers avoid manual CUDA tuning. Demonstrates kernels debugged on CPU, runtime kernel launches from C++ via CUDA API, and reports measurable speedups once model logic grows nontrivial.
AI-written summary · about
related talks
16mLightning Talk: Triton Compiler
08mGPU Programming with Triton Kernels
25mLeveraging AI Ecosystem for Portable and Sustainable GPU Kernels in HPC
10mMaximizing Kernel Development Productivity Under Performance Constraints
09mLightning Talk: Harnessing NVIDIA Tensor Cores: An Exploration of CUTLASS & OpenAI..
01h 26mGPU Algorithm Authoring with CUDA Tile
28mHelion: A High-level DSL for Kernel Authoring
48mGPU Programming in Pure Python