Reproduce “Inside VOLT: Designing an Open-Source GPU Compiler” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shinnung Jeong, Chihyo Ahn, Huanzhi Pu, Jisheng Zhao, Hyesoon Kim, Blaise Tine
VOLT is an open-source, end-to-end compiler infrastructure developed for the Vortex RISC-V GPU, successfully compiling 32 OpenCL and 17 CUDA benchmarks with correctness, and demonstrating performance improvements through middle-end SIMT-aware optimizations while incurring a negligible 0.18% geomean compile-time overhead.
Key insights
- Centralizing SIMT-aware optimizations like divergence management in the middle-end significantly enhances compiler portability and reusability across different GPU architectures.
- Sophisticated uniformity analysis and control-flow structurization are critical for both correctness and efficiency when executing divergent code on Single Instruction, Multiple Threads (SIMT) architectures.
- Leveraging and extending existing open-source compiler frameworks can substantially reduce the engineering burden and enhance the sustainability of open-source GPU hardware projects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.