GPU Memory Access Optimization for 2D Electrical Wave Propagation Through Cardiac Tissue and Karma Model Using Time and Space Blocking

Abstract

Abnormal heart function is one of the largest causes of death in the world. Understanding and simulating its contraction and relaxation movements may reduce this mortality and improve specific treatments. Due the high processing time to compute propagation of an electrical wave through cardiac tissue, interactive and real time simulations are a challenge, making unable corrective treatments on the fly. In this paper we propose an approach to reduce the electrophysiology models simulation computational time using the finite difference method to solve the laplacian operator, required by the karma model. Our solution optimizes accesses to global memory by storing more than one instant of time in shared memory within each GPU kernel call. We present tests that validate our proposal and show an increase performance close to when compared with available GPU implementations.