Han-Chung Wang
Han-Chung Wang
This unpack is valid because there is extract_slice semantic in unpack ops. You can think that it is an inverse operation of pack op. The pack op has padding semantics,...
Yes, it is intended. It is not a bug. I guess the behavior is triggered by the tiling implementation: https://github.com/llvm/llvm-project/blob/fc4b1a303b296d02f6243a083510c4ee7f290ab0/mlir/lib/Dialect/Tensor/IR/TensorTilingInterfaceImpl.cpp#L561-L588 Looking at the implementation, I think the issue is that...
> This is just merely based on the observation of this IR. Since we are adding zeros, can't we DCE the linalg.generic ? Good question. I honestly don't know where...
> I haven't reviewed the details yet, because I need to review the upstream one first. > > High-level comment: it is better to split it into two PRs: >...
The PR depends on - https://github.com/iree-org/iree/pull/22886 - https://github.com/iree-org/iree/pull/22891
Below is the output w/o the PR. I think what Ben said makes sense: they are just bytes, and you can interpret the data with tensor types with an optional...
IIUC, the bitcast op is lowered to `stream.tensor.clone` op, and the original failure is in verifier. Now we have encodings in the source and the result; they are "acceptable" because...
Reminder: please add the message to the release tracker: https://github.com/iree-org/iree/issues/22307
Thanks @jtuyls for the repro! I can reproduce it without running the test. I'll share the IR and steps. I recently learned that the unexpected memory allocation issue sometimes can...
Coming from https://github.com/iree-org/iree/pull/22118#discussion_r2457572997, and I'm mirroring the context from @benvanik, so I won't miss it in the future. I was asking question if we should do it in CombineInitilaizers pass,...