llama.cpp
b1a5bd4e - CUDA: better coalesce data-access for contiguous concat (#22330)

Commit

22 days ago

CUDA: better coalesce data-access for contiguous concat (#22330) Also, distribute all elements across CTAs evenly instead of launching one CTA per dim

References

#22330 - CUDA: better coalesce data-access for contiguous concat

Author

ORippler

Parents

0c6ee1ca

llama.cpp b1a5bd4e - CUDA: better coalesce data-access for contiguous concat (#22330)

llama.cpp
b1a5bd4e - CUDA: better coalesce data-access for contiguous concat (#22330)