public void NormDiff_L2(NPPImage_8uC3 tpl, CudaDeviceVariable<double> pNormDiff)
{
int bufferSize = NormDiffL2GetBufferHostSize();
CudaDeviceVariable<byte> buffer = new CudaDeviceVariable<byte>(bufferSize);
status = NPPNativeMethods.NPPi.NormDiff.nppiNormDiff_L2_8u_C3R(_devPtrRoi, _pitch, tpl.DevicePointerRoi, tpl.Pitch, _sizeRoi, pNormDiff.DevicePointer, buffer.DevicePointer);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "nppiNormDiff_L2_8u_C3R", status));
buffer.Dispose();
NPPException.CheckNppStatus(status, this);
}