[all-commits] [llvm/llvm-project] 8cc616: [mlir] Clamp UnPackOp tiling sizes from operand ti...

[Max191 via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 8cc616bc71dfe0648de3843a006ac8827c5fe59d
      https://github.com/llvm/llvm-project/commit/8cc616bc71dfe0648de3843a006ac8827c5fe59d
  Author: Max191 <44243577+Max191 at users.noreply.github.com>
  Date:   2024-11-13 (Wed, 13 Nov 2024)

  Changed paths:
    M mlir/lib/Dialect/SCF/Transforms/TileUsingInterface.cpp
    M mlir/lib/Dialect/Tensor/IR/TensorTilingInterfaceImpl.cpp
    M mlir/test/Interfaces/TilingInterface/tile-and-fuse-consumer.mlir

  Log Message:
  -----------
  [mlir] Clamp UnPackOp tiling sizes from operand tile (#112429)

The `getIterationDomainTileFromOperandTile` implementation for
tensor.unpack did not clamp sizes when the unpack op had extract_slice
semantics. This PR fixes the bug.

The PR also makes a minor change to `tileAndFuseConsumerOfSlice`. When
replacing DPS inits, the iteration domain is needed, and it is computed
from the tiled version of the operation after the initial tiling
transformation. This can result in some extra indexing computation, so
the PR changes it to use the original full sized cloned consumer op.

---------

Signed-off-by: Max Dawkins <max.dawkins at gmail.com>



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications