PROJECT TITLE :

Designing Efficient Index-Digit Algorithms for CUDA GPU Architectures

ABSTRACT:

Trendy graphics processing units (GPUs) provide very high computing power at comparatively low cost. Nevertheless, designing economical algorithms for the GPUs normally needs extra effort and time, even for knowledgeable programmers. In this work we have a tendency to gift a tuning methodology that permits the design for CUDA-enabled GPU architectures of index-digit algorithms, that's, algorithms where the information movement can be described as the permutations of the digits comprising the indices of the information components. This methodology, based mostly on 2-stages identified as GPU resource analysis and operators string manipulation, is applied to FFT and tridiagonal systems solver algorithms, analyzing the performance options and the most adequate solutions. The ensuing implementation is compact and outperforms alternative well-known and commonly used state-of-the-art libraries, with an improvement of up to 19.a pair of p.c over NVIDIA's complicated CUFFT , and more than 300zero percent over the NVIDIA's CUDPP for real information tridiagonal systems.