Skip to content

Conversation

@pull
Copy link

@pull pull bot commented May 21, 2025

See Commits and Changes for more details.


Created by pull[bot] (v2.0.0-alpha.1)

Can you help keep this open source service alive? 💖 Please sponsor : )

jtuyls and others added 3 commits May 20, 2025 22:38
Fixes an issue encountered while enabling the pad based encoding:
#20835 (comment)

When no GPU::L1CacheInfo can be retrieved, the `GPUPadLayoutAttr` is
resolved to a `EncodingNopLayoutAttr`, however this results in the
following error:
```
LLVM ERROR: can't create Attribute 'mlir::iree_compiler::IREE::Codegen::EncodingNopLayoutAttr' because storage uniquer isn't initialized: the dialect was likely not loaded, or the attribute wasn't added with addAttributes<...>() in the Dialect::initialize() method.
```
While this could be resolved by adding
`ctx->getOrLoadDialect<IREE::Codegen::IREECodegenDialect>();` and
implementing the SerializableEncodingAttrInterface for
`EncodingNopLayoutAttr`, it's chosen to replace the nop encoding
attribute resolver with the identity `GPUPadLayoutAttr` resolver.

Signed-off-by: Jorn Tuyls <jorn.tuyls@gmail.com>
-- Initially, the vector distribute pipeline would check for a single
`DispatchTensorStoreOp` inside a dispatch. This patch extends to adding
vector distribution config in the case of multiple store operations.
-- Also, the workgroup tile sizes are added to every operation for
compatibility with tileUsingForall. (It updates in the case of reduction
and forwards in the case of parallel operation.)
@pull pull bot added the ⤵️ pull label May 21, 2025
@pull pull bot merged commit 6476f8b into VitalyAnkh:main May 21, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants