CUDA Deformers for Model Reduction

The finite element method (FEM) with model reduction is widely used to simulate deformable objects. Improving the performance and rendering quality of deformable objects simulated using FEM with model reduction has been a critical problem for several years. As computer technologies rapidly evolve, GPUs have seen significant improvement. It is clear that modern graphics processing units (GPUs) play an important role in real-time rendering and parallel computing. The existing FEM model reduction rendering systems, however, utilize a large amount of CPU resources but relatively little GPU resources. Our experiments show that vertex position computations and mesh rendering account for a large percentage of the CPU’s overhead in the existing system. Therefore, this work investigates how to efficiently employ GPUs for vertex position computation (Uq computation) and rendering of deformable objects simulated using model reduction. We first survey the related GPU technologies. Then we give a CUDA-based algorithm for Uq computation and vertex global position computation, which we demonstrate to remarkably speed up the Uq computation. We then give a new system architecture to reduce the communication between the CPU and GPU. Meanwhile, the system performance is further improved by making the CPU and GPU work in parallel. Additionally, a new rendering system based on modern OpenGL features is developed. It not only enhances the rendering performance but also improves the rendering quality. Finally, our experiments show that the proposed system outperforms the previous systems in terms of the computational cost as well as the rendering quality.