[mlir][sparse] Improve ConcatenateOp rewriting for annotated all dense result.
authorbixia1 <bixia@google.com>
Thu, 12 Jan 2023 00:32:42 +0000 (16:32 -0800)
committerbixia1 <bixia@google.com>
Fri, 13 Jan 2023 16:36:54 +0000 (08:36 -0800)
commit66466641545a0b392703c0a81b50b9f31c01d54e
tree94501773c907259e24a3faf12d6e724f16f4c037
parentb55f83d013eff244f7cf6bfcd06f06f7c894ff28
[mlir][sparse] Improve ConcatenateOp rewriting for annotated all dense result.

Previously, we rely on InsertOp to add values to the result, in the same way we
add values to a sparse tensor with compressed dimensions. We now direct store
values to the values buffer.

Reviewed By: Peiming

Differential Revision: https://reviews.llvm.org/D141517
mlir/lib/Dialect/SparseTensor/Transforms/CodegenUtils.cpp
mlir/lib/Dialect/SparseTensor/Transforms/CodegenUtils.h
mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorRewriting.cpp
mlir/test/Dialect/SparseTensor/sparse_concat_codegen.mlir