GPU Calc: use parallel reduction to implement sum

Use reduction kernel when given a large DoubleVectorRef

Change-Id: Ifd4977b81be64274733909e43f0e5ef161bb455e
Signed-off-by: I-Jui (Ray) Sung <ray@multicorewareinc.com>

sc/source/core/opencl/opbase.hxx

@@ -92,7 +92,7 @@ public:
     virtual void GenDeclRef(std::stringstream &ss) const;
 
     /// Create buffer and pass the buffer to a given kernel
-    virtual size_t Marshal(cl_kernel, int, int);
+    virtual size_t Marshal(cl_kernel, int, int, cl_program);
 
     virtual ~DynamicKernelArgument();