Skip to content

Port reverse from CUDA.jl - #609

Closed
christiangnrd wants to merge 7 commits into
JuliaGPU:mainfrom
christiangnrd:reverse
Closed

christiangnrd wants to merge 7 commits into
JuliaGPU:mainfrom
christiangnrd:reverse

Conversation

@christiangnrd

Copy link
Copy Markdown
Member

This may have to wait for KA 0.10 depending on how much cpu=true affects performance.

@christiangnrd

Copy link
Copy Markdown
Member Author

Seems like at least with CUDA.jl, using dynamic workgroup sizes recovers ~50% of the performance lost switching over to KernelAbstractions. Is there potentially some overhead with KA that is lesser with Dynamic workgroup sizes?

@maleadt

maleadt commented Sep 1, 2025

Copy link
Copy Markdown
Member

Is there potentially some overhead with KA that is lesser with Dynamic workgroup sizes?

cc @vchuravy

@vchuravy

vchuravy commented Sep 1, 2025

Copy link
Copy Markdown
Member

Huh, I would expect static kernel sizes to be a performance benefit or at least performance neutral.

The only thing that could happen is that suddenly we are able to unroll more or something like that.

@maleadt maleadt added the arrays The array abstraction: broadcasting, mapreduce, indexing, views, generic linear algebra. label Oct 19, 2026
@maleadt

maleadt commented Oct 19, 2026

Copy link
Copy Markdown
Member

Superseded by #790, which implements reverse/reverse! on GPU arrays with AcceleratedKernels.

@maleadt maleadt closed this Oct 19, 2026
@christiangnrd
christiangnrd deleted the reverse branch October 19, 2026 17:38
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

arrays The array abstraction: broadcasting, mapreduce, indexing, views, generic linear algebra.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants