Skip to content

Avoid copies in the c4v sign flip calculation - #488

Merged
kshyatt merged 5 commits into
mainfrom
ksh/c4v_sign_flip
Nov 8, 2026
Merged

kshyatt merged 5 commits into
mainfrom
ksh/c4v_sign_flip

Conversation

@kshyatt

@kshyatt kshyatt commented Nov 6, 2026

Copy link
Copy Markdown
Member

By doing it this way we avoid some copies back and forth from the GPU that synchronize the device.

@kshyatt
kshyatt requested a review from leburgel November 6, 2026 10:47
Comment thread src/algorithms/ctmrg/c4v.jl Resolved
@kshyatt

kshyatt commented Nov 6, 2026

Copy link
Copy Markdown
Member Author

I found a way to do this in one kernel launch as opposed to n_blocks. For 32 blocks it's about 10x faster and still doesn't sync.

Comment thread ext/PEPSKitGPUArraysExt.jl Resolved
Comment thread ext/PEPSKitGPUArraysExt.jl Resolved
@kshyatt
kshyatt merged commit 2c0cf7e into main Nov 8, 2026
30 of 32 checks passed
@kshyatt
kshyatt deleted the ksh/c4v_sign_flip branch November 8, 2026 17:12
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants