Modified Incomplete Cholesky Preconditioned Conjugate Gradient Algorithm on GPU for the 3D Parabolic Equation

In this study, for solving the three-dimensional partial differential equation ut = uxx + uyy + uzz, an efficient parallel method based on the modified incomplete Cholesky preconditioned conjugate gradient algorithm (MICPCGA) on the GPU is presented. In our proposed method, for this case, we overcome the drawbacks that the MIC preconditioner is generally difficult to be parallelized on the GPU due to the forward/backward substitutions, and thus present an efficient parallel implementation method on the GPU. Moreover, a vector kernel for the sparse matrix-vector multiplication, and optimization of vector operations by grouping several vector operations into a single kernel are adopted. Numerical results show that our proposed forward/backward substitutions and MICPCGA on the GPU both can achieve a significant speedup, and compared to an approximate inverse SSOR preconditioned conjugate gradient algorithm (SSORPCGA), our proposed MICPCGA obtains a bigger speedup, and outperforms it in solving the three-dimensional partial differential equation.