The unstructured grids are widely used in the processing and interpretation of geophysical data with terrain due to their excellent ability to simulate shape. Among them, the efficiency of the curved hexahedral grid in its gravity forward modeling based on the isoparametric finite-element method is poor due to the complex transformations involving numerous morphological nodes, which limits its application to large-scale data. For this reason, combining with the deep learning technology, the letter proposes a fast forward method of 3D gravity data for the curved hexahedral grid based on the back-propagation (BP) neural network. In the training phase, the method learns the complex mapping of curved hexahedral elements to their gravity sensitivities through the neural network, thereby achieving fast forward modeling during the prediction phase. Numerical examples show that the new method has good simulation accuracy and generalization ability. Under the premise that the training phase can be completed upfront with its cost excluded, its forward efficiency is tens of times higher than that of the isoparametric finite-element method. The successful application of the new method in the actual terrain model of Mount Taishan area in China further proves its practicality.

With the rapid development of artificial intelligence (AI), convolutional neural networks (CNNs) have been widely applied in fields like computer vision and recommendation systems. This growth has intensified the demand for hardware acceleration of CNNs. Existing accelerators are either designed as co-processors or improve performance through extended instructions. While these methods can significantly improve performance, they often result in limited programming and execution flexibility. In this paper, we design custom RISC-V instructions specifically for CNNs to maximize data reuse and exploit parallelism. Then, to efficiently execute CNNs instructions, we extend a Pipelined Vector Computing Unit (PPVCU). Finally, we incorporate Pattern Detection Logic (PDL) to identify common data dependence patterns in CNNs, enabling the Data Dependence Computing Unit (DDCU) to process instructions within each pattern in parallel. Experimental results show that our approach achieves, on average, 9.54× performance improvement and 6.7× energy efficiency improvement compared to our baseline, 8.34× performance improvement and 3.1× energy efficiency improvement compared to state-of-the-art designs.