Accelerate Computing For Forming Simulation Based On Collaborative Heterogeneous Model

Multi-physics coupling, nonlinear multi-scale coupling and other factors make large amounts of data need to be calculated in forming simulation,while these jobs must also be completed within a reasonable time.This thesis mainly focus on the research of parallel computing architecture accelerate computing for forming simulation, proposed collaborative heterogeneous computing model to shorten forming simulation computing time based on CPU/GPU architecture.In this thesis focus on how to optimize the CPU/GPU collaborative heterogeneous computing model and reduce the computing time. Some algorithm such as finite difference method, matrix and vector product are often used in calculating stress filed and temperature filed,which considered multi-physics coupling and multi-scale coupling. Those algorithms have been implemented under GPU architecture. Large calculation jobs are transferred to GPU architecture,which has a higher calculation efficiency, alleviate CPU calculation pressure and shorten computing time.Greatly improves calculation efficiency.Optimize CPU/GPU collaborative heterogeneous computing model design based on six experiments of data storage optimization analysis.A variety of experiment data show that had obtained better data storage solution.There are many shortcomings in GPU architecture for scientific computing, so must be coordinated by CPU to complete.Optimize data storage architecture on the computational performance of model through bit string array used together with bitmask.Test found no optimized program can be as high as 25 times compared to the optimized one.Make good use of shared memory,reduce unnecessary data synchronization,reduce cache hit rate of decline factors through program design and convergence calculation process. Synchronized by using the branch method to deal with conditional branches, conditional branches in the same direction, the transfer of the same thread of the first implementation of the direction of branch instructions completed, and then switch to the direction of another thread and then another branch of the implementation of optimized SIMT instructions threading the calculation of performance. With CPU/GPU collaborative and heterogeneous computing prototype shows that the performance model can predict the execution time of problem sizes that are 16 times as large as the profile runs with less than 20% error, and that the predicted optimal load distribution ratios have less than 60% and the resulting performance improvement using both CPUs and GPUs can be as high as 50% compared to using either a CPU core or a GPU.Based on our results and current trends in microarchitecture,Efficient use of CPU and GPU collaborative and heterogeneous computing environment will become increasingly important to high performance computing. The study in this paper is of great significance and wide application prospect.