[all-commits] [llvm/llvm-project] dc3258: [mlir][mesh] Add all-slice operation (#81218)

[Boian Petkantchin via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: dc3258c617420e83caff63c93d548e0923b10791
      https://github.com/llvm/llvm-project/commit/dc3258c617420e83caff63c93d548e0923b10791
  Author: Boian Petkantchin <boian.petkantchin at amd.com>
  Date:   2024-02-15 (Thu, 15 Feb 2024)

  Changed paths:
    M mlir/include/mlir/Dialect/Affine/Utils.h
    M mlir/include/mlir/Dialect/Arith/Utils/Utils.h
    M mlir/include/mlir/Dialect/Mesh/IR/MeshOps.h
    M mlir/include/mlir/Dialect/Mesh/IR/MeshOps.td
    M mlir/include/mlir/Dialect/Mesh/Transforms/Transforms.h
    M mlir/include/mlir/IR/Builders.h
    M mlir/lib/Dialect/Affine/Utils/Utils.cpp
    M mlir/lib/Dialect/Arith/Utils/Utils.cpp
    M mlir/lib/Dialect/Mesh/IR/MeshOps.cpp
    M mlir/lib/Dialect/Mesh/Transforms/CMakeLists.txt
    M mlir/lib/Dialect/Mesh/Transforms/Simplifications.cpp
    M mlir/lib/Dialect/Mesh/Transforms/Spmdization.cpp
    M mlir/lib/Dialect/Mesh/Transforms/Transforms.cpp
    A mlir/lib/Dialect/Mesh/Transforms/TransformsDetail.h
    M mlir/lib/IR/Builders.cpp
    A mlir/test/Dialect/Mesh/all-scatter-op-lowering.mlir
    M mlir/test/Dialect/Mesh/canonicalization.mlir
    M mlir/test/Dialect/Mesh/invalid.mlir
    M mlir/test/Dialect/Mesh/ops.mlir
    M mlir/test/Dialect/Mesh/resharding-spmdization.mlir
    M mlir/test/Dialect/Mesh/spmdization.mlir
    M mlir/test/lib/Dialect/Mesh/CMakeLists.txt
    A mlir/test/lib/Dialect/Mesh/TestOpLowering.cpp
    R mlir/test/lib/Dialect/Mesh/TestProcessMultiIndexOpLowering.cpp
    M mlir/tools/mlir-opt/mlir-opt.cpp
    M utils/bazel/llvm-project-overlay/mlir/BUILD.bazel

  Log Message:
  -----------
  [mlir][mesh] Add all-slice operation (#81218)

This op is the inverse of all-gather. It is useful to have an explicit
concise representation instead of having a blob of slicing logic.

Add lowering for the op that slices from the tensor based on the
in-group process index.

Make resharding generate an all-slice instead of inserting the slicing
logic directly.