[all-commits] [llvm/llvm-project] 96ff02: [mlir] cleanup of structured.tile* transform ops (...

[ftynse via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 96ff0255f2ecd0bd5d1aba650a99e046a4bd5ec6
      https://github.com/llvm/llvm-project/commit/96ff0255f2ecd0bd5d1aba650a99e046a4bd5ec6
  Author: Oleksandr "Alex" Zinenko <zinenko at google.com>
  Date:   2023-09-26 (Tue, 26 Sep 2023)

  Changed paths:
    M mlir/docs/Tutorials/transform/Ch1.md
    M mlir/docs/Tutorials/transform/Ch2.md
    M mlir/docs/Tutorials/transform/ChH.md
    M mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td
    M mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
    M mlir/python/mlir/dialects/_structured_transform_ops_ext.py
    M mlir/test/Dialect/Affine/loop-tiling-parametric.mlir
    M mlir/test/Dialect/Affine/loop-tiling.mlir
    M mlir/test/Dialect/GPU/transform-gpu-failing.mlir
    M mlir/test/Dialect/LLVM/transform-e2e.mlir
    M mlir/test/Dialect/Linalg/generalize-tensor-pack-tile.mlir
    M mlir/test/Dialect/Linalg/generalize-tensor-unpack-tile.mlir
    M mlir/test/Dialect/Linalg/matmul-shared-memory-padding.mlir
    M mlir/test/Dialect/Linalg/multisize-tiling-full.mlir
    M mlir/test/Dialect/Linalg/promote.mlir
    M mlir/test/Dialect/Linalg/promotion_options.mlir
    M mlir/test/Dialect/Linalg/tile-conv.mlir
    M mlir/test/Dialect/Linalg/tile-indexed.mlir
    M mlir/test/Dialect/Linalg/tile-softmax.mlir
    M mlir/test/Dialect/Linalg/tile-tensors.mlir
    M mlir/test/Dialect/Linalg/tile-to-forall.mlir
    M mlir/test/Dialect/Linalg/transform-op-compose-masked-vectorize-and-cleanups.mlir
    M mlir/test/Dialect/Linalg/transform-op-fuse-into-containing.mlir
    M mlir/test/Dialect/Linalg/transform-op-fuse.mlir
    M mlir/test/Dialect/Linalg/transform-op-hoist-pad-build-packing-loop-nest.mlir
    M mlir/test/Dialect/Linalg/transform-op-hoist-pad.mlir
    M mlir/test/Dialect/Linalg/transform-op-scalarize.mlir
    M mlir/test/Dialect/Linalg/transform-op-tile.mlir
    M mlir/test/Dialect/Linalg/transform-ops.mlir
    M mlir/test/Dialect/Linalg/transform-patterns.mlir
    M mlir/test/Dialect/Linalg/transform-tile-and-fuse.mlir
    M mlir/test/Dialect/Linalg/transform-tile-reduction.mlir
    M mlir/test/Dialect/SCF/transform-loop-fuse-sibling.mlir
    M mlir/test/Dialect/SCF/transform-op-take-assumed-branch.mlir
    M mlir/test/Dialect/Tensor/tiling.mlir
    M mlir/test/Dialect/Transform/ops.mlir
    M mlir/test/Dialect/Transform/selective-targeting.mlir
    M mlir/test/Dialect/Vector/transform-vector.mlir
    M mlir/test/Examples/transform/Ch1/invalidation-1.mlir
    M mlir/test/Examples/transform/Ch1/invalidation-2.mlir
    M mlir/test/Examples/transform/Ch1/sequence.mlir
    M mlir/test/Examples/transform/Ch2/sequence.mlir
    M mlir/test/Examples/transform/Ch3/sequence.mlir
    M mlir/test/Examples/transform/ChH/full.mlir
    M mlir/test/Integration/Dialect/Linalg/CPU/test-conv-1d-call.mlir
    M mlir/test/Integration/Dialect/Linalg/CPU/test-conv-1d-nwc-wcf-call.mlir
    M mlir/test/Integration/Dialect/Linalg/CPU/test-conv-2d-call.mlir
    M mlir/test/Integration/Dialect/Linalg/CPU/test-conv-2d-nhwc-hwcf-call.mlir
    M mlir/test/Integration/Dialect/Linalg/CPU/test-conv-3d-call.mlir
    M mlir/test/Integration/Dialect/Linalg/CPU/test-conv-3d-ndhwc-dhwcf-call.mlir
    M mlir/test/Integration/Dialect/Linalg/CPU/test-tensor-matmul.mlir
    M mlir/test/python/dialects/transform_structured_ext.py

  Log Message:
  -----------
  [mlir] cleanup of structured.tile* transform ops (#67320)

Rename and restructure tiling-related transform ops from the structured
extension to be more homogeneous. In particular, all ops now follow a
consistent naming scheme:

 - `transform.structured.tile_using_for`;
 - `transform.structured.tile_using_forall`;
 - `transform.structured.tile_reduction_using_for`;
 - `transform.structured.tile_reduction_using_forall`.

This drops the "_op" naming artifact from `tile_to_forall_op` that
shouldn't have been included in the first place, consistently specifies
the name of the control flow op to be produced for loops (instead of
`tile_reduction_using_scf` since `scf.forall` also belongs to `scf`),
and opts for the `using` connector to avoid ambiguity.

The loops produced by tiling are now systematically placed as *trailing*
results of the transform op. While this required changing 3 out of 4 ops
(except for `tile_using_for`), this is the only choice that makes sense
when producing multiple `scf.for` ops that can be associated with a
variadic number of handles. This choice is also most consistent with
*other* transform ops from the structured extension, in particular with
fusion ops, that produce the structured op as the leading result and the
loop as the trailing result.