Skip to content

[COMGR] Add missing vectorization flags#3450

Open
MrSidims wants to merge 1 commit into
amd-stagingfrom
vectorization-flags
Open

[COMGR] Add missing vectorization flags#3450
MrSidims wants to merge 1 commit into
amd-stagingfrom
vectorization-flags

Conversation

@MrSidims

@MrSidims MrSidims commented Jul 17, 2026

Copy link
Copy Markdown

They are currently being filtered out. They shouldn't - they present in llvm.cmdline.

They are currently being filtered out. They shouldn't - they present in
llvm.cmdline.
@MrSidims
MrSidims force-pushed the vectorization-flags branch from 33e2f66 to db0e2ae Compare July 17, 2026 18:35
@MrSidims MrSidims changed the title Vectorization flags [COMGR] Add missing vectorization flags Jul 17, 2026
@github-actions github-actions Bot added the comgr Related to Code Object Manager label Jul 17, 2026
@MrSidims

Copy link
Copy Markdown
Author

Fun fact: the issue was observed while exploring why some benchmarks are having better performance when compiled to SPIR-V. Trivial sample would be rocm-examples/convolution, which runs 20% faster if compiled to SPIR-V.

This patch "fixes" it, so now AMDGPU and SPIR-V paths would be on par.

Performance will be restored (and improved) on AMDGPU when/if llvm#210399 (or similar fix) is merged.

@maarquitos14 maarquitos14 left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. I wonder if there's any test that we should update/add for this.

@idubinov idubinov left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM.
There is TODO about all -cc1 flags. Does the current list cover all flags or TODO still actual?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

comgr Related to Code Object Manager

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants