The Landscape of Parallelism
summary
Claims that C++ can span HPC and commercial worlds with concrete features and gaps: shows openmp 4.5 device/target model for GPUs and hierarchical memory, describes published TSs (parallelism, concurrency, transactional memory) and their concrete APIs, reports a deadline-driven push toward exascale and real implementations in GCC/Clang, explains parallelism TS execution policies extended to bulk agent creation and demos an absence of standardized executors for massive accelerator dispatch and practical workarounds that do compile and run.
AI-written summary · about
chapters (14)
- 0:00Start
- 4:14Why is GPU important now?
- 5:26Top500 contenders
- 6:39Internet of Things
- 8:16What now?
- 10:32OpenMP 4.0 Device Constructs
- 11:49SAXPY: Serial (host)
- 20:39WOMB The Parallel and concurrency planets
- 22:192015: SG5 TM Language in a nutshell
- 24:23Atomic & synchronized transactions
- 26:51SG1: C++ Std Parallelism TS
- 33:19Composition of Long-Running Work
- 53:04How to write for each parallelized
- 53:16How to write for each vectorized
related talks
01h 02mOverview of Parallel Programming in C++
01h 07mAnother fundamental shift in Parallelism Paradigm?
01h 16mPaying for Lunch: C++ in the ManyCore Age
37mC++ for multi-accelerator, multi-vendor systems
44mStories from a parallel universe
01h 06mAsynchronous Computation in C++
01h 02mC++ Performance Portability - A Decade of Lessons Learned
01h 22mCostless Software Abstractions for Parallel Architectures