Introduction to C++ AMP (GPGPU Computing)
summary
Demoed C++ AMP code converts a parallel CPU loop into a GPU kernel via parallel_for_each and array_view and hits concrete speedups: a laptop N‑body goes from ~40 GFLOPS on CPU to 74 GFLOPS straightforward and 134 GFLOPS with tiling, a desktop with two AMD cards reaches 1 teraflop single‑precision, an image “cartoonizer” drops from 8.2 s on CPU to ~380 ms on GPU enabling ~30 FPS, and Visual Studio + WARP fallback are used for debugging.
AI-written summary · about
chapters (26)
- 0:00Start
- 0:59The N-Body Simulation Problem
- 3:27Parallel Implementation to C + + Amp
- 7:43Warp
- 10:00Cpu Parallelism
- 14:16The Heterogeneous Mix of Accelerators
- 15:36Problem Domains
- 18:54Concurrency Array
- 21:03Read-Only Buffer
- 21:34Right Only Buffer
- 22:18Extent Class
- 28:19Error Handling
- 28:46Synchronization Point
- 29:07Automatic Synchronization
- 29:29The Worst Case
- 30:41Tiling
- 36:57Visual Studio Integration
- 38:04Software Emulator
- 38:16Multiple Breakpoints in Cpu Code and in Gpu
- 39:33The Concurrency Visualizer
- 43:20How Well Does C + + Amp Interact with the C + + Trailing Library
- 44:49Concurrency Visualizer
- 47:36Allocates a Buffer That's Too Big To Fit on the on the Gpu
- 48:21Treatment for Sparse Matrices
- 49:54Resources
- 51:42Adding Parallelism in the Library
related talks
52mC++ Standard Parallelism
59mUsing Template Magic to Automatically Generate Hybrid CPU/GPU-Code
01h 07mGeneric Parallel Programming
01h 06mFoundation of GPU Programming
25mExploit Parallelism for AI Workloads with WASM and OpenMP
01h 06mEfficient GPU Programming with Modern C++
27mVendor-Neutral GPU Programming in Chapel
28mArray Programming on GPUs: Challenges and Opportunities