[all-commits] [llvm/llvm-project] fc1260: [mlir][sparse] fuse collapse_shape on sparse tenso...

[PeimingLiu via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: fc126022e884da273d6796d97741fc15ce50fd54
      https://github.com/llvm/llvm-project/commit/fc126022e884da273d6796d97741fc15ce50fd54
  Author: Peiming Liu <peiming at google.com>
  Date:   2023-03-01 (Wed, 01 Mar 2023)

  Changed paths:
    M mlir/include/mlir/Dialect/SparseTensor/IR/SparseTensor.h
    M mlir/lib/Dialect/SparseTensor/IR/SparseTensorDialect.cpp
    M mlir/lib/Dialect/SparseTensor/Transforms/LoopEmitter.cpp
    M mlir/lib/Dialect/SparseTensor/Transforms/LoopEmitter.h
    A mlir/test/Dialect/SparseTensor/sparse_reshape_dot.mlir
    A mlir/test/Integration/Dialect/SparseTensor/CPU/reshape_dot.mlir

  Log Message:
  -----------
  [mlir][sparse] fuse collapse_shape on sparse tensor with GenericOp.

Instead of always materializing a new sparse tensor after reshape, this patch tries to fuses the reshape (currently only on COO) with GenericOp and coiterates with the reshaped tensors without allocating a new sparse tensor.

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D145016