PadeLibs

PadeLibs is a computational framework for high-order and high-resolution simulations of compressible turbulent flows. The software design specifically considers the needs of scientific research in computational fluid dynamics of compressible flows in many different aspects. PadeLibs is implemented in modern C++ with an object-oriented programming manner. It is desired to be used as a library. The provided interfaces on different levels of the code permit easy implementations for investigations of flow physics, numerical methods, as well as physical and closure models. Further implementation including modifying built-in schemes or hooking extra equations or variables in a flow solver does not need to change the source code in PadeLibs. A consistent set of numerical schemes used for a simulation can be also used for data post-processing or on-the-fly data processing.

PadeLibs integrates various types of high-order numerical schemes and robust high-resolution simulation techniques including the most recent achievements from the research team. It supports simulations on both uniform Cartesian meshes or curvilinear meshes. The spatial discretization is primarily based on the six-order compact finite difference methods. For the presence of eddy shocklets in highly compressible turbulence or strong shocks in high speed flows, artificial diffusivities (without solution filtering) and nonlinear shock capturing schemes combined with approximated Riemann solver are available. Some subgrid scale models are also provided for large-eddy simulations.

PadeLibs supports both CPU-only and GPU-accelerated computing on distributed architectures with portable performance benefited from Kokkos programming model. The parallelism among distributed memory chunks are managed by the message passing interface (MPI). The code has optimized for operations on both shared and distributed memory.

PadeLibs is highly scalable on the modern high performance computing platforms. It maintains excellent parallel efficiency at scale with the application of the compact numerical schemes. The parallel performance is fundamentally boosted by the parallel algorithm developed from the research team. The parallel performance of PadeLibs used for a simulation of compressible Taylor-Green vortex problem has been measured on the Summit supercomputer from the Oak Ridge National Laboratory (the largest supercomputer in the world until 2020). The weak scaling measurement shows that the computation scales up to 24,576 GPUs with more than 50% parallel efficiency preserved.

Code Development Team:  Hang Song,  Kristen V. Matsuno,  Jacob R. West,  Aditya S. Ghate,  Akshay Subramaniam