refactor

[ROCm/composable_kernel commit: 0597116330]

src/include/blockwise_tensor_op.cuh

@@ -76,6 +76,7 @@ blockwise_4d_tensor_pointwise_operation_unary(DstDesc, Float* __restrict__ p_dst
     }
 }
 
+// Function: p_dst[reorder[i0], reorder[i1], reorder[i2], reorder[i3]] = p_src[i0,i1,i2,i3]
 // TODO: in order to optimize mem access for different mem type,
 // need to write specialized version
 template <unsigned BlockSize,

src/include/common.cuh

@@ -11,3 +11,5 @@ struct is_same<T, T>
 {
     static const bool value = true;
 };
+
+__device__ unsigned get_thread_local_id() { return threadIdx.x; }