Accelerating High Arithmetic Intensity Storm Surge Model using CUDA

Munesh Singh Chauhan; Mayyada Hammoshi; Balaqis Abdallah Salim Al-Bahri

doi:10.37591/rtpc.v3i2.700

Abstract

GPUs (Graphic Processing Units) have opened the floodgates for high-performance computing, especially for programs that involve high-level arithmetic computations. Though parallel computing capability has been around for many years, it requires large and expensive hardware resources in the form of supercomputers. Presently, GPUs are found in all computing devices due to their reasonable prices. GPUs are now being tapped for speeding up legacy codes, especially in FORTRAN that involves high arithmetic intensity. One of the applications that can be applied is a weather simulation program, the storm surge model, which is converted into CUDA C (Compute Unified Device Architecture). The various pitfalls and challenges faced in the conversion process are narrated in terms of data type equivalences, including file I/Os, multi-dimensional array handling, FORTRAN mathematical functions equivalence, and many others. The processes are categorized under two major segments: FORTRAN to C and C to CUDA C. Each complexity has been tested as a separate case study with the emphasis on speed, coupled with the accuracy of the results. Finally, note that parallelism is not the panacea in all cases. Those segments of code that have data dependences and file I/O are not suitable candidates for conversion.Keywords: CUDA C, FORTRAN, GPUs, high-level arithmetic computations, parallel computingCite this ArticleMunesh Singh Chauhan. Accelerating High Arithmetic Intensity Storm Surge Model using CUDA. Recent Trends in Parallel Computing. 2016; 3(2): 9–21p.