Explicit vectorized loads/stores
Open
cuda kernels
enhancement
- Dominant language
- Julia
- Stars
- 1.4k
- Forks
- 281
- Avg merge
- 1d 7h
- Merged PRs (30d)
- 30
Description
In CUDA C you can explicitly request vectorized loads/stores using the special vector types (`float2`, `float4`). Sometimes I found those useful to squeeze out the last bit of performance. This definitely isn't high priority, but I was wondering how hard would be to add something similar to `CUDAnative`.
JuliaGPU/CUDAnative.jl#174 is related, but maybe some of the problems have been solved ?
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.