Kévin Petit

Results 219 issues of Kévin Petit

clvk is currently designed with Unified Memory Architecture systems in mind. All the device memory allocations are currently `HOST_VISIBLE ` which is inefficient on systems with a discrete GPU. Instead...

Should probably be preferred to `HOST_COHERENT` uncached memory. Will need cache maintenance in a few places.

Infinite loop in clspv.

Basic support is there but missing - [ ] saturating conversions - [ ] explicit rounding modes