Asynchronous Computation in C++
summary
Claims a C++ runtime called HPX implements futures/async/dataflow and actually runs: measured ~25 million lightweight threads per second, supports distributed execution and CUDA/OpenCL backends, and auto-builds dependency graphs to expose fine‑grain parallelism. Examples include a 10‑second thought experiment on thread overhead, an asynchronous heat‑diffusion implementation over 10 million points, and a partitioning result that finds ~15k element chunks give best throughput, showing roughly 2× speedup vs naive OpenMP.
AI-written summary · about
chapters (10)
- 0:00Intro
- 1:44WHAT IS ASYNCHRONOUS COMPUTINGP
- 8:07WHY IS STD THREAD TOO SLOWP
- 15:13WHAT IS A THE FUTURE
- 28:46COMPOSITIONAL FACILITIES
- 34:45THREAD OVERHEADS
- 35:53HPX - THE API
- 41:23EXTENDING PARALLEL ALGORITHMS
- 49:491D HEAT EQUATION, FUTURIZED
- 57:06CONCLUSIONS
related talks
59mC++ on its way to exascale and beyond
01h 02mHPX - A C++ Library for Parallelism and Concurrency
01h 03mThe Landscape of Parallelism
05mFast Parallel Programming using Modern C++
57m[filtered] Asynchronous C++ Programming
01h 04mThe C++20 Synchronization Library
37mC++ for multi-accelerator, multi-vendor systems
31mGPU Performance Portability Using Standard C++ with SYCL