Affiliation:
1. McGill University, Canada
Abstract
GPU database systems are an effective solution to query optimization, particularly with compilation and data caching. They fall short, however, in end-to-end workloads, as existing compiler toolchains are too expensive for use with short-running queries. In this work, we define and evaluate a runtime-suitable query compilation pipeline for NVIDIA GPUs that extracts high performance with only minimal optimization. In particular, our balanced approach successfully trades minor slowdowns in execution for major speedups in compilation, even as data sizes increase. We demonstrate performance benefits compared to both CPU and GPU database systems using interpreters and compilers, extending query compilation for GPUs beyond cached use cases.
Publisher
Association for Computing Machinery (ACM)
Subject
Hardware and Architecture,Information Systems,Software
Reference78 articles.
1. AMD. 2011. AMD Intermediate Language (IL). Retrieved from http://developer.amd.com/wordpress/media/2012/10/AMD_Intermediate_Language_(IL)_Specification_v2.pdf.
2. AMD. 2021. GCN Native ISA LLVM Code Generator–ROCm Documentation 1.0.0 documentation. Retrieved from https://rocmdocs.amd.com/en/latest/ROCm_Compiler_SDK/ROCm-Native-ISA.html.
3. AMD. 2022. Let’s Build Everything–GPUOpen. Retrieved from https://gpuopen.com/.
4. Taming Control Divergence in GPUs through Control Flow Linearization
5. Piotr Bialas and Adam Strzelecki. 2016. Benchmarking the cost of thread divergence in CUDA. In Parallel Processing and Applied Mathematics, Roman Wyrzykowski, Ewa Deelman, Jack Dongarra, Konrad Karczewski, Jacek Kitowski, and Kazimierz Wiatr (Eds.). Springer International, Cham, 570–579.