OpenMP-RT: Native Pragma Support for Real-Time Tasks and Synchronization with LLVM(…)
summary
Implements a new pragma-based real-time task construct in the LLVM C compiler under OpenMP 5.1 for Linux (uses PREEMPT_RT scheduler), providing a OpenMP-RT runtime that maps periodic RT tasks to dedicated threads, a per-priority threadpool to avoid fine-grain contention, and two lock-free sync schemes (retry and double buffer). Evaluated with 100 synthetic applications built from BOTS/EPCC tasks, reaches about 0.93 per-core utilization with no deadline misses and shows double-buffer gives the lowest consumer response times.
AI-written summary · about
chapters (11)
- 0:00OpenMP-RT intro and RT constraints
- 1:27Project goals and RT task framework
- 3:13RT task construct and thread sync
- 6:48Config file and core partitioning
- 8:33Intertask sync methods overview
- 12:31Experimental setup and benchmarks
- 14:18Utilization stress test results
- 15:00Sync methods performance comparison
- 16:49Retry method behavior analysis
- 17:55Summary and conclusions
- 18:15Q&A on lock-free implementation
related talks
01h 13mTalk by I-Ting Angelina Lee (Washington University in St. Louis)
17mEvaluation of Speedup & Energy with Multigrain Parallelizing Compiler
17mUnlocking 15% More Performance: A Case Study in LLVM Optimization for RISC-V
01h 03mReal-time Safety — Guaranteed by the Compiler!
16mResponsive Parallelism with Dynamic and First-Class Priorities
14mPerformance Interfaces for Hardware Accelerators
20mStreamAlloc: A Framework for Analyzing and Transforming CUDA Code to Enable(…)
20mThe status of parallel rustc