merge develop

include/ck/tensor_operation/gpu/element/element_wise_operation.hpp

@@ -159,7 +159,7 @@ struct Normalize
         using ck::math::sqrt;
 
         float variance = mean_square - (mean * mean);
-        y = ((x - mean) / sqrt(variance + static_cast<float>(epsilon_))) * gamma + beta;
+        y = ((x - mean) / sqrt(variance + type_convert<float>(epsilon_))) * gamma + beta;
     };
 
     template <>

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

@@ -30,6 +30,8 @@ struct ThreadwiseReduction
 
     static_assert(src_length_m == dst_length_m, "lengths of source and dst buffer must match!");
 
+    using Op = OpReduce;
+
     template <typename SrcBufferType, typename DstBufferType>
     __device__ static void Reduce(const SrcBufferType& src_buf, DstBufferType& dst_buf)
     {