Add support for halft in reduction op Add.

include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_tile_loop.hpp

@@ -131,8 +131,6 @@ __global__ void
         const auto StrideB = gemm_desc_ptr[group_id].StrideB;
 
         auto gridwise_gemm = GridwiseGemm();
-        auto& results_buffer = gridwise_gemm.GetCThreadBuffer();
-
         b2c_tile_map.CalculateBottomIndex(work_scheduler.tile_id_ - offset);
 
         // Iterate over K dimension for this [M,N] tile

include/ck/utility/reduction_operator.hpp

@@ -52,7 +52,7 @@ struct Add
     __host__ __device__ inline constexpr void operator()(T& a, T b) const
     {
         static_assert(is_same<T, float>::value || is_same<T, double>::value ||
-                          is_same<T, int32_t>::value,
+                          is_same<T, half_t>::value || is_same<T, int32_t>::value,
                       "The data type is not supported by the Add accumulator!");
 
         a = a + b;