Multi-physics Multi-scale HPC Simulations of Skeletal Muscles

We present a highly scalable framework for the simulation of skeletal muscles as a neuromuscular system. Our work is based on previous implementations of a complex model coupling different physical phenomena on different temporal and spatial scales, in particular bio-chemical processes on the cellular level (as ordinary differential equations), electrical signal propagation along muscle fibers (as many one-dimensional diffusion equations), and the EMG signal in the three-dimensional muscle at organ scale. Our contribution is a new software toolbox that allows to generate simulation code suited for the execution on a supercomputer from the commonly used XML-based model-description used in the respective community. We present different variants of this code generation for CPUs, specific optimizations for our muscle model, and our in situ visualization approach that allows us to minimize data transfer between simulation, and the visualization front-end (such as the Powerwall at VISUS (https://www.visus.uni-stuttgart.de/)) along with results for numerical experiments on up to approximately seven thousand cores of the CRAY XC40 system Hazel Hen (https://www.hlrs.de/systems/cray-xc40-hazel-hen/) for more than 180,000 muscle fibers. The original implementation was limited to 4 cores. Compared to our previous work in [4], we further enhanced scalability, but in particular also node-level performance.