RealFeedApp Futhark speeds up byte copying by permuting iteration spaces
The Futhark compiler now optimizes byte copying on GPUs by leveraging runtime information about array layouts. While the impact on specific programs is small, the technique is particularly interesting for handling large operations with fewer of them.
- Field
- programming
- What they did
- Developers demonstrated how to speed up data copying on graphics processing units (GPUs) by using information about array layouts available only at runtime.
- Why it matters
- This could improve performance for programs working with large arrays, although the method is described as more interesting than practically useful.
#futhark#gpu#optimization#array layout#compiler
