[all-commits] [llvm/llvm-project] 0d6e41: [mlir][vector] Order parallel indices before trans...

[harsh-nod via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 0d6e4199e32a3a5942f920bf13c0a0ddf10d2579
      https://github.com/llvm/llvm-project/commit/0d6e4199e32a3a5942f920bf13c0a0ddf10d2579
  Author: harsh-nod <harsh at nod-labs.com>
  Date:   2021-06-28 (Mon, 28 Jun 2021)

  Changed paths:
    M mlir/lib/Dialect/Vector/VectorTransforms.cpp
    M mlir/test/Dialect/Vector/vector-multi-reduction-lowering.mlir

  Log Message:
  -----------
  [mlir][vector] Order parallel indices before transposing the input in multireductions

The current code does not preserve the order of the parallel
dimensions when doing multi-reductions and thus we can end
up in scenarios where the result shape does not match the
desired shape after reduction.

This patch fixes that by ensuring that the parallel indices
are in order and then concatenates them to the reduction dimensions
so that the reduction dimensions are innermost.

Differential Revision: https://reviews.llvm.org/D104884