Add missing constructor.

include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp

@@ -334,7 +334,7 @@ struct ThreadwiseTensorSliceTransfer_v3r1
             const bool is_src_valid = src_oob_thread_scratch_tuple_(thread_scratch_id)
                                           .template GetAsType<bool>(src_data_idx_seq);
 
-            auto op_r_v = is_src_valid ? op_r : vector_t(0);
+            auto op_r_v = is_src_valid ? op_r : vector_t{0};
 
             src_thread_scratch_tuple_(thread_scratch_id)
                 .template SetAsType<vector_t>(src_data_idx_seq, op_r_v);

include/ck/utility/amd_ck_fp8.hpp

@@ -406,12 +406,12 @@ struct non_native_vector_base<f8_ocp_t, 2>
 {
     using data_t = f8_ocp_t::data_type;
     using type   = non_native_vector_base<f8_ocp_t, 2>;
-
-    __host__ __device__ non_native_vector_base() = default;
-
     using data_v = fp8_impl::fp8x2_storage_t; // type of storage vector
     data_v d;                                 // storage vector
 
+    __host__ __device__ non_native_vector_base() = default;
+    __host__ __device__ non_native_vector_base(data_t a) : d{a} {}
+
     using float2_t = fp8_impl::float2_t;
 
 #if CK_USE_OCP_FP8