bug fix: BlockwiseGenericTensorSliceCopy_v2::MoveDstSlicingWindow

composable_kernel/include/kernel_algorithm/gridwise_convolution_implicit_gemm_v1r3_chwn_cyxk_khwn_padded.hpp

@@ -125,8 +125,8 @@ struct GridwiseConvolutionImplicitGemm_v1r3_chwn_cyxk_khwn_padded
         constexpr auto out_k_h_w_n_thread_desc = make_ConstantTensorDescriptor_packed(
             Sequence<KPerThread, HoPerThread, WoPerThread, NPerThread>{});
 
-        // blockwise copy
-        // input: format is [C, Hi, Wi, N]
+        // blockwise input copy
+        //   format is [C, Hi, Wi, N]
         auto blockwise_in_copy =
             BlockwiseGenericTensorSliceCopy_v2<BlockSize,
                                                decltype(in_c_h_w_n_global_desc),

composable_kernel/include/tensor_operation/blockwise_generic_tensor_slice_copy.hpp

@@ -417,6 +417,8 @@ struct BlockwiseGenericTensorSliceCopy_v1
     }
 };
 
+// Slice a (normal or merged) tensor, and copy it into another (normal or merged) tensor
+// memory layout (ordering of dimensions) can be different between src and dst.
 template <index_t BlockSize,
           class SrcDesc,
           class DstDesc,
@@ -510,7 +512,7 @@ struct BlockwiseGenericTensorSliceCopy_v2
     MoveDstSlicingWindow(T step_sizes,
                          integral_constant<bool, PositiveDirection> positive_direction)
     {
-        mThreadwiseLoad.MoveDstSlicingWindow(step_sizes, positive_direction);
+        mThreadwiseStore.MoveDstSlicingWindow(step_sizes, positive_direction);
     }
 
     private: