Source
- URL: https://x.com/Venemo/status/1595310115401474048
- Author: Timur Kristóf (@Venemo)
- Posted: 2022-11-23 06:55:45
Branch
1/ @Venemo
@SebAaltonen
AFAIK that is only fully true on AMD GPUs, everyone else has to go through some pain to implement this extension.
2/ @SebAaltonen
@Venemo
Nvidia has also similar scalar instructions and can store descriptors in scalar registers. Their CUDA docs list the new instructions that were added in Turing. Latest 3 generations of Nvidia should be fine.