Fast acceleration of 2D wave propagation simulations using modern computational accelerators.

Area of Science:

• Computational Science
• Scientific Computing
• Bioinformatics

Background:

• Modern computational accelerators like GPUs offer significant speed improvements for scientific applications.
• Efficient parallelization of scientific code often requires specialized expertise and lacks portability across architectures.

Purpose of the Study:

• To overcome challenges in parallelizing scientific code for computational accelerators.
• To achieve automated parallelization using OpenACC and enhanced portability using OpenCL for wave propagation simulations.

Main Methods:

• Applied parallelization schemes using OpenACC, OpenCL, and OpenMP.
• Utilized GPUs and Intel Many Integrated Core (MIC) coprocessors.
• Focused on a 2D cardiac action potential model for wave propagation simulations.

Main Results:

• OpenACC achieved over 150x speedup on GPUs with minimal code changes.
• OpenCL yielded speedups of at least 200x on GPUs compared to sequential code.
• OpenMP on Intel MIC coprocessor provided 120x speedups.

Conclusions:

• OpenACC offers an automatic, efficient, and portable method for parallelizing 2D cardiac wave simulations on GPUs.
• The presented parallelization strategies using OpenACC, OpenCL, and OpenMP are applicable to other wave propagation models.