Accelerating advanced preconditioning methods on hybrid architectures

Many problems, in diverse areas of science and engineering, involve the solution largescale sparse systems linear equations. In most these scenarios, they are also a computational bottleneck, therefore their efficient on parallel architectureshas motivated tremendous volume research.This dissertation targets use GPUs to enhance performance using iterative methods complemented with state-of-the-art preconditioned techniques. particular, we study ILUPACK, package for via Krylov subspace that relies modern inverse-based multilevel ILU (incomplete LU) preconditioning technique.We present new data-parallel versions preconditioner important solvers contained significantly improve its without affecting accuracy. Additionally existing task-parallel ILUPACK shared- distributed-memory inclusion GPU acceleration. The results obtained show sensible reduction runtime methods, as well possibility addressing large-scale problems efficiently.