Figure 1From: Convolution of large 3D images on GPU and its decompositionConvolution in CUDA. The function called cuMemcpy provides data transfers between CPU (Host) and GPU (Device). The pwProd provides a pointwise multiplication of two signals on GPU. The norm constant normalizes the convolution if required. The result is stored in signal variable in order to save memory. The result could also be stored in another parameter in case we needed to keep the input data.Back to article page