FIL

Responsible Partners: GUF

FIL, like BHAC, is a multidimensional GRMHD code. FIL is different from BHAC in the sense that it has been designed to solve the GRMHD equations in dynamical spacetimes. This allows FIL to simulate the maelstrom that are neutron star collisions. Whilst there is a greater computational cost for a full GRMHD code, there is a greater generality to the problems FIL can be applied to. FIL is a state-of-the-art numerical relativity code; it uses fourth-order finite difference schemes to provide highly accurate simulations. FIL leverages the Einstein Toolkit (ET) which provides the infrastructure to enable numerical relativity simulations. FIL provides the computational modules, whereas the ET provides the AMR grid and memory management that FIL uses via the Carpet module. A new version of the Carpet module, CarpetX has recently been released, based on the AMReX library. CarpetX is compatible with CUDA and OpenACC and is designed for exascale computing. Therefore, to accelerate and improve FIL’s scaling, a large part of our work will involve replacing Carpet with CarpetX. We can then focus on accelerating the computational modules in FIL using GPUs and optimizing them. This is the general outline of the planned work. For further information on FIL, a high-level description of the code, and the main algorithm, please refer to §6 of the deliverable D1.2 (see deliverable section).

The public code repository is for the Einstein Toolkit, as follows:
https://github.com/einsteintoolkit