Use MemRefDescriptor in Linalg-to-LLVM conversion
authorAlex Zinenko <zinenko@google.com>
Thu, 14 Nov 2019 16:03:39 +0000 (08:03 -0800)
committerA. Unique TensorFlower <gardener@tensorflow.org>
Thu, 14 Nov 2019 16:04:10 +0000 (08:04 -0800)
commit7c28de4aef6da3ab2f53118ecf717e56c68352e7
tree97a223bbd9ce3d650f8ee17560d7693256e4eb60
parenta007d4395a36adf3aad0f4b9914dcb8756a37c7d
Use MemRefDescriptor in Linalg-to-LLVM conversion

Following up on the consolidation of MemRef descriptor conversion, update
Linalg-to-LLVM conversion to use the helper class that abstracts away the
implementation details of the MemRef descriptor. This required MemRefDescriptor
to become publicly visible. Since this conversion is heavily EDSC-based,
introduce locally an additional wrapper that uses builder and location pointed
to by the EDSC context while emitting descriptor manipulation operations.

PiperOrigin-RevId: 280429228
mlir/include/mlir/Conversion/StandardToLLVM/ConvertStandardToLLVM.h
mlir/lib/Conversion/StandardToLLVM/ConvertStandardToLLVM.cpp
mlir/lib/Dialect/Linalg/Transforms/LowerToLLVMDialect.cpp