A Pipeline-Based ODE Solving Framework

The traditional parallel solving methods of ordinary differential equations (ODE) are mainly classified into task-parallelism, data-parallelism, and instruction-level parallelism. Based on the RIDC (revisionist integral deferred correction) algorithm, a hybrid solver dispatched on both CPU and GPU is proposed, which realizes computing in a pipeline form and a remarkable parallelism is obtained both inside a single equation and among many different equations. The proposed framework can make full use of the multi-core advantage of GPU, which is conducive to load balancing within computing nodes. The efficiency and accuracy of the framework are verified in several experiments.