Introduction to hpxMP
summary
An OpenMP runtime built on HPX that replaces libgomp by LD_PRELOAD and runs OpenMP parallel regions as HPX threads and tasks, so OpenMP code needs no source changes. Compile with -fopenmp and launch with LD_PRELOAD=libhpxmp.so. Benchmarks on a 4‑thread vector kernel show HPXMP beating libgomp at larger problem sizes; a task‑based quicksort shows libgomp winning when cutoff makes many tiny tasks, while HPXMP is slightly faster at coarser grain. OMPT tracing is supported.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:25OpenP HPXMP
- 0:41Why we need HPXMP
- 1:20Structure of OpenP
- 1:50hpxMP
- 2:38examples
- 2:48deeper code
- 3:21performance analysis
- 3:52taskbased analysis
related talks
57mHHVM at Etsy: Harder, Better, Faster, Stronger
24mMaximizing Workload's Performance With Smarter Runtimes
34mDeclarative Device Virtualization: Orchestrating GPUs & Hardware in Cloud Native Environments
25mAllegro.pl in flames
56mExecutors in C++
09mLightning Talk: Vllm-triton-backend: How To Get State-of-the-art Performance on...
20mBringing PyTorch Monarch to AMD GPUs: Single-Controller Distributed Tra...
11mLightning Talk: Torch-Spyre: Compiling To a Multi-core Dataflow Accelerator...