| Soft fabric are so common in our lives,so as that it became an important class of simulation objects in virtual reality technology. As early as the 1880s, researchers began to focus on physics-based simulation of soft fabric. After 30 years of development, this technology is widely used in many fields, such as game, entertainment, film and television animation,costume design etc. Despite the soft fabric simulation related technology has been sufficiently researched, it still face enormous challenges in some respects. For example, how to improve the performance of simulation in order to give more detail about fabric folds in acceptable time is still a problem waiting for researchers to overcome. In order to not lose fidelity and significantly improve time performance in cloth simulation, more and more people began to focus on parallelized simulation which based on GPUs. This paper will be designed to parallelize the relevant algorithm based on CUDA architecture for soft fabric simulation and also optimize the system performance time. This thesis includes the following aspects:Firstly, the soft fabric simulation-related software and hardware technology as well as the basis of parallel programming are introduced, such as the hardware architectures of GPU parallel programming and parallel programming thinking and so on. It introduces the theoretical basis of parallelization for the numerical solver and collision detection algorithms.Then we divided the simulation system into several major components and describes them individually.The first step is for the physical modeling of soft fabric. Physical modeling technology was chose as the focus of the research methods after a comparative analysis of the three mainstream soft fabric modeling methods. Next, details about the method of physical modeling mass - spring systems theory was described by two aspects, building the model and stress analysis system elaborated mass - spring system. We also analyze and solve the"super-flexible" problem, which is commonly seen in mass - spring system. Then, several methods of dynamics integration solver for mass-spring model are compared, and we choose the Verlet integration and take a deep research as it is easy for the design of parallelization and also relatively more stable and efficient. Finally, a dynamic mapping method from shared memory to thread is proposed in order to ensure the load balancing in parallelism calculation.Next is the arallelized collision detection for soft fabric simulation system. At the beginning, we introduced and compared several common collision detection methods. Adding the soft fabrics characteristics, such as easy deformation into consideration, we chose AABB bounding box for our system. And then a more deep sight about this was introduced. Based on parallel consideration,we design an appropriate data structure which can store all types of information of bounding box. We also proposed a simple pre-processing algorithm for its update. Then, parallel design and optimization algorithm that based on hardware features CUDA parallel architecture are deployed on collision detection.Finally, the results of CUDA soft fabric simulation system are showed. In order to show the impact on system performance which contributed by both numerical solver and collision detection algorithms, we designed several appropriate simulation scenarios, and tested these two algorithms respectively on them,and also a serial implementation was deployed as comparison. The results of the entire simulation system were given at the end. |
PDF Full Text Request