Automatic Instrumentation and PGO Optimization of HPC Compute Dwarfs
Tuesday, June 10, 2025 3:00 PM to Thursday, June 12, 2025 4:00 PM · 2 days 1 hr. (Europe/Berlin)
Foyer D-G - 2nd floor
Women in HPC Poster
Compiler and Tools for Parallel ProgrammingHeterogeneous System ArchitecturesPerformance MeasurementPerformance Tools and Simulators
Information
Poster is on display and will be presented at the poster pitch session.
Profile-guided optimization (PGO) is an effective technique for enhancing program performance by utilizing runtime profiling data to guide optimizations. This poster explores the integration of PGO with the LLVM compiler infrastructure, focusing on the use of LLVM and Clang for source code instrumentation and optimization. Through source-level instrumentation, we capture detailed profiling information that is later used to optimize execution paths, enhance branch prediction, and refine function inlining decisions. The process begins with the compilation of the source code using Clang with appropriate instrumentation flags, which produce profiling data during program execution. This data is subsequently fed back into the LLVM compilation pipeline for further optimization during subsequent builds. We discuss the steps involved in this workflow, challenges encountered, and the impact of various optimizations on application performance. The study demonstrates the potential of LLVM’s toolchain in efficiently applying profile-guided transformations, as well as the advantages of fine-tuned performance through data-driven decisions enabled by Clang's instrumentation capabilities.
Contributors:
Profile-guided optimization (PGO) is an effective technique for enhancing program performance by utilizing runtime profiling data to guide optimizations. This poster explores the integration of PGO with the LLVM compiler infrastructure, focusing on the use of LLVM and Clang for source code instrumentation and optimization. Through source-level instrumentation, we capture detailed profiling information that is later used to optimize execution paths, enhance branch prediction, and refine function inlining decisions. The process begins with the compilation of the source code using Clang with appropriate instrumentation flags, which produce profiling data during program execution. This data is subsequently fed back into the LLVM compilation pipeline for further optimization during subsequent builds. We discuss the steps involved in this workflow, challenges encountered, and the impact of various optimizations on application performance. The study demonstrates the potential of LLVM’s toolchain in efficiently applying profile-guided transformations, as well as the advantages of fine-tuned performance through data-driven decisions enabled by Clang's instrumentation capabilities.
Contributors:
Format
On DemandOn Site
Speakers
Anja Gerbes
Community ManagerGWDG
