Skip to content

ARCH/X86: Move AVX2 NT code and extract ERMS memcpy helper - #11878

Merged
tvegas1 merged 2 commits into
openucx:masterfrom
arun-chandran-edarath:ntbt_avx2_move
Sep 3, 2026
Merged

tvegas1 merged 2 commits into
openucx:masterfrom
arun-chandran-edarath:ntbt_avx2_move

Conversation

@arun-chandran-edarath

Copy link
Copy Markdown
Contributor

What?

Refactor the code

Why?

Feature: splitting as per comments from #11696

How?

#11696 will be broken into logical PRs, this is the first in the series

Move the existing AVX2 non-temporal buffer-transfer helpers from cpu.c
to an ISA-specific inline fragment without changing their implementation
or dispatcher behavior.

Signed-off-by: Arun Chandran <Arun.Chandran@amd.com>
Move the existing rep movsb sequence into ucs_x86_memcpy_erms() and call
it from the same built-in memcpy size window. Keep the compile guard and
return semantics unchanged.

Signed-off-by: Arun Chandran <Arun.Chandran@amd.com>
@svc-nvidia-pr-review

Copy link
Copy Markdown

🤖 Starting review — findings will be posted here when done.

@arun-chandran-edarath

Copy link
Copy Markdown
Contributor Author

@tvegas1 @roiedanino please check

Comment thread src/ucs/arch/x86_64/cpu_nt_avx2.inl
@tvegas1 tvegas1 changed the title move avx2 nontemporal functions to separate file ARCH/X86: Move AVX2 NT code and extract ERMS memcpy helper Sep 2, 2026
@arun-chandran-edarath

Copy link
Copy Markdown
Contributor Author

Is this ready to merge?

@tvegas1
tvegas1 merged commit aa0b65d into openucx:master Sep 3, 2026
162 checks passed
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.

4 participants