Performance portability of large scale distributed Krylov solvers with OpenACC and CUDA

Idomura, Yasuhiro   ; Ali, Y.*; Onodera, Naoyuki   ; Hasegawa, Yuta   ; Ina, Takuya*

Krylov solvers can account for up to of the total computing cost in extreme scale nuclear CFD simulations. In order to accelerate such CFD codes, we ported the Preconditioned Conjugate Gradient (PCG), Preconditioned Chebyshev Basis communication-avoiding Conjugate Gradient (P-CBCG) and Communication-Avoiding Generalized Minimal RESidual (CA-GMRES) methods on to GPUs. In this talk, we will share our experiences in porting these solvers via OpenACC, CUDA, and CUDA aware MPI.