Removed trivial copy constructors on parameter classes to enable devi… (#366)

* Removed trivial copy constructors on parameter classes to enable device-side launch of CUTLASS kernels

* Added SFINAE to the `TensorRef(NonConstTensorRef const&)` constructor to avoid making it a copy-constructor for device code

* std => platform

* fix affine2

* really fix affine2

Co-authored-by: Haicheng Wu <haichengw@nvidia.com>

include/cutlass/tensor_ref.h

@@ -219,9 +219,12 @@ class TensorRef {
   }
 
   /// Converting constructor from TensorRef to non-constant data.
+  template<typename _Magic = int>
   CUTLASS_HOST_DEVICE
   TensorRef(
-    NonConstTensorRef const &ref              ///< TensorRef to non-const data
+    NonConstTensorRef const &ref,             ///< TensorRef to non-const data
+    ///SFINAE trick to avoid creating a copy-constructor when Element_ is already non-const
+    _Magic magic = (typename platform::enable_if< ! platform::is_same<NonConstTensorRef, TensorRef<Element_, Layout_> >::value, _Magic>::type)0
   ):
     ptr_(ref.data()), layout_(ref.layout()) { }