Introducing Chapel: A Programming Language for Productive Parallel Computing...
summary
A parallel programming language that compiles to native code, is interoperable with C/Fortran/Python, and uses GASNet and standard job launchers to run from laptops to supercomputers. Microbenchmarks (stream triad) match MPI/OpenMP scaling to thousands of cores; GPU paths target NVIDIA and AMD and approach CUDA/HIP performance for large problems. Real projects include an 85,000-line CFD code that competes with MPI/C++ and a larcuda system (30k Chapel + 25k Python) that performs a 72 TB sort in ~2.5 minutes achieving 480–1200 GB/s.
AI-written summary · about
chapters (8)
- 0:00Chapel introduction and goals
- 7:20Chapel benchmarks and apps
- 11:54Locales and locality in Chapel
- 15:00Parallelism with coforall
- 22:25Stream Triad benchmark
- 27:58Targeting GPUs with Chapel
- 33:32Champs and Arkouda applications
- 39:09Resources and community
related talks
54mThe Audacity of Chapel: Scalable Parallel Programming Done Right
57mHigh-Performance Logic Programming on Servers, Clusters, and GPUs
01h 00mBringing Clang and C++ to GPUs: An Open-Source, CUDA-Compatible GPU C++ Compiler
16mIntrepydd: Performance, Productivity, and Portability for Data Science Application Kernels
28mCode Generation in Linnea
35mA parallel programming language extension for Java
16mPricing Python Parallelism: A Dynamic Language Cost Model for Heterogeneous Platforms
01h 13mTalk by I-Ting Angelina Lee (Washington University in St. Louis)