This project aims at developing exascale computing technologies for multi-scale CFD simulations on exascale platforms. For this purpose, we promote a collaborative research with respect to the following three main subjects, a)High performance sparse matrix solvers on accelerated computing platforms, b) Adaptive Mesh Refinement (AMR) methods for multi-scale turbulence problems with complicated boundaries, c)Performance portability of CFD codes based on advanced frameworks and programming models. The target CFD applications in this project are the 3D multi-phase multi-component thermal hydraulic CFD code JUPITER [Yamashita et al., Nuclear Engineering and Design 2017] and the 3D compressible hydrodynamic astrophysics CFD code ARK [Padioleau et al., Astrophysical Journal 2019]. In this project, we accelerate the development of common exascale computing technologies for AMR-CFD simulations by sharing the knowledges on high performance matrix solvers, performance portability, and AMR methods.