Skip to content

[Codegen] Handle SwizzleHintOp in shared memory pipeliner#23945

Draft
Yu-Zhewen wants to merge 1 commit intoiree-org:mainfrom
Yu-Zhewen:pipeliner-swizzle-hint
Draft

[Codegen] Handle SwizzleHintOp in shared memory pipeliner#23945
Yu-Zhewen wants to merge 1 commit intoiree-org:mainfrom
Yu-Zhewen:pipeliner-swizzle-hint

Conversation

@Yu-Zhewen
Copy link
Copy Markdown
Contributor

Fixes: #23919

Assisted-by: Cursor (Claude)

Signed-off-by: Yu-Zhewen <zhewenyu@amd.com>
Made-with: Cursor
@Yu-Zhewen Yu-Zhewen force-pushed the pipeliner-swizzle-hint branch from 7a56bc8 to 35de084 Compare March 27, 2026 12:18
@Yu-Zhewen Yu-Zhewen marked this pull request as ready for review March 27, 2026 13:50
@Yu-Zhewen Yu-Zhewen requested review from jerryyin and lialan March 27, 2026 13:50
@@ -289,7 +399,7 @@ static LogicalResult multiBufferLDSAllocations(scf::ForOp forOp,

// First, clone view ops inside the loop for each allocation
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

nit: no more view.

@Yu-Zhewen
Copy link
Copy Markdown
Contributor Author

Yu-Zhewen commented Mar 27, 2026

After discussing with @jerryyin, we feel the current approach is quite brittle. I'm going to explore alternative options. One possibility is to have SwizzleHintOp handle multi-buffering directly (currently it only handles flattening), and leave it outside the pipelining loop.

Converting back to draft while I investigate.

@Yu-Zhewen Yu-Zhewen marked this pull request as draft March 27, 2026 15:42
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.

SwizzleHintOp blocks double-buffering in ROCDLPrefetchSharedMemoryPass

2 participants