ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-13 17:55:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor

Browse Source
This commit is contained in:
Chao Liu
2019-09-17 11:19:15 -05:00
parent f7be86b9e4
commit e1a67b693e
5 changed files with 17 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
composable_kernel/include/tensor_operation/blockwise_generic_tensor_slice_copy.hpp
View File

@@ -684,12 +684,10 @@ template <index_t BlockSize,
struct BlockwiseGenericTensorSliceCopy_v4
{
static constexpr index_t nDim = SrcDesc::GetNumOfDimension();
using Index = MultiIndex<nDim>;
using SrcCoord = typename TensorCoordinate_v2<SrcDesc>::type;
using DstCoord = typename TensorCoordinate_v2<DstDesc>::type;
__device__ constexpr BlockwiseGenericTensorSliceCopy_v4(SrcCoord src_block_slice_origin,
DstCoord dst_block_slice_origin)
__device__ constexpr BlockwiseGenericTensorSliceCopy_v4(const Index& src_block_slice_origin,
const Index& dst_block_slice_origin)
{
static_assert(nDim == SrcDesc::GetNumOfDimension() &&
nDim == DstDesc::GetNumOfDimension() && nDim == SliceLengths::Size() &&

4
composable_kernel/include/tensor_operation/threadwise_generic_tensor_slice_copy.hpp
View File

@@ -966,8 +966,8 @@ struct ThreadwiseGenericTensorSliceCopy_v4r2
using SrcCoord = typename TensorCoordinate_v2<SrcDesc>::type;
using DstCoord = typename TensorCoordinate_v2<DstDesc>::type;
__device__ constexpr ThreadwiseGenericTensorSliceCopy_v4r2(SrcCoord src_slice_origin,
DstCoord dst_slice_origin)
__device__ constexpr ThreadwiseGenericTensorSliceCopy_v4r2(const Index& src_slice_origin,
const Index& dst_slice_origin)
: mSrcSliceOrigin(src_slice_origin), mDstSliceOrigin(dst_slice_origin)
{
static_assert(nDim == SrcDesc::GetNumOfDimension() &&