[all-commits] [llvm/llvm-project] a232b5: [mlir][shard, mpi] Adding Shard/MPI reduce_scatter...

[Frank Schlimbach via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: a232b5b96f67b395f52d44cf0e02c52ec658ff40
      https://github.com/llvm/llvm-project/commit/a232b5b96f67b395f52d44cf0e02c52ec658ff40
  Author: Frank Schlimbach <frank.schlimbach at intel.com>
  Date:   2026-03-03 (Tue, 03 Mar 2026)

  Changed paths:
    M mlir/include/mlir/Dialect/MPI/IR/MPIOps.td
    M mlir/include/mlir/Dialect/Shard/IR/ShardOps.td
    M mlir/include/mlir/Dialect/Shard/Transforms/Partition.h
    M mlir/include/mlir/Dialect/Shard/Transforms/Passes.td
    R mlir/include/mlir/Dialect/Shard/Transforms/Simplifications.h
    A mlir/include/mlir/Dialect/Shard/Transforms/Simplify.h
    M mlir/lib/Conversion/MPIToLLVM/MPIToLLVM.cpp
    M mlir/lib/Conversion/ShardToMPI/ShardToMPI.cpp
    M mlir/lib/Dialect/MPI/IR/MPIOps.cpp
    M mlir/lib/Dialect/Shard/IR/ShardOps.cpp
    M mlir/lib/Dialect/Shard/Transforms/CMakeLists.txt
    R mlir/lib/Dialect/Shard/Transforms/Simplifications.cpp
    A mlir/lib/Dialect/Shard/Transforms/Simplify.cpp
    M mlir/test/Conversion/MPIToLLVM/mpitollvm.mlir
    M mlir/test/Conversion/ShardToMPI/convert-shard-to-mpi.mlir
    M mlir/test/Dialect/MPI/mpiops.mlir
    M mlir/test/Dialect/Shard/all-scatter-op-lowering.mlir
    M mlir/test/Dialect/Shard/canonicalization.mlir
    M mlir/test/Dialect/Shard/folding.mlir
    M mlir/test/Dialect/Shard/invalid.mlir
    M mlir/test/Dialect/Shard/ops.mlir
    R mlir/test/Dialect/Shard/simplifications.mlir
    A mlir/test/Dialect/Shard/simplify.mlir
    M mlir/test/lib/Dialect/Shard/CMakeLists.txt
    M mlir/test/lib/Dialect/Shard/TestReshardingPartition.cpp
    R mlir/test/lib/Dialect/Shard/TestSimplifications.cpp
    M mlir/tools/mlir-opt/mlir-opt.cpp

  Log Message:
  -----------
  [mlir][shard, mpi] Adding Shard/MPI reduce_scatter and simplification (#184189)

- introduces a simplify pass, which finds such patterns and replaces it
with the equivalent `reduce-scatter`
- promotes the test-pass `test-shard-optimizations` to a proper pass and adds
  - folding allgather+allslice into reduce_scatter
- sanitizes the `shard.reduce_scatter` op
- adds a new `mpi.reduce_scatter_block` op
- lowers `shard.reduce_scatter` to MPI
- lowers `mpi-reduce_scatter_block` to llvm

---------

Co-authored-by: Copilot <175728472+Copilot at users.noreply.github.com>



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