Yuchun Liu developed custom device support for CSR sum and mean operations in the PaddlePaddle/PaddleCFD repository, enabling these computations to run on both standard GPUs and custom hardware such as metax-gpu. By updating select.cpp to recognize is_custom_device(), Yuchun broadened the hardware compatibility of PaddleCFD, aligning with the project’s roadmap for heterogeneous hardware acceleration. This work required a strong command of C++ and device abstraction, as well as an understanding of GPU computing. The feature expanded deployment options for CFD workloads, demonstrating depth in integrating custom-device checks into core operations without introducing regressions or bugs.