public void NormDiff_L1(NPPImage_8uC3 tpl, CudaDeviceVariable<double> pNormDiff, int nCOI, NPPImage_8uC1 pMask)
{
int bufferSize = NormDiffL1MaskedGetBufferHostSize();
CudaDeviceVariable<byte> buffer = new CudaDeviceVariable<byte>(bufferSize);
status = NPPNativeMethods.NPPi.NormDiff.nppiNormDiff_L1_8u_C3CMR(_devPtrRoi, _pitch, tpl.DevicePointerRoi, tpl.Pitch, pMask.DevicePointerRoi, pMask.Pitch, _sizeRoi, nCOI, pNormDiff.DevicePointer, buffer.DevicePointer);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "nppiNormDiff_L1_8u_C3CMR", status));
buffer.Dispose();
NPPException.CheckNppStatus(status, this);
}