Mixed Precision Method for GPU-based FFT

Author

Qi, Shuhan ; Wang, Xuan ; Shi, Shaohuai

Author_Institution

Comput. Applic. Res. Center, Harbin Inst. of Technol., Shenzhen, China

fYear

2011

fDate

24-26 Aug. 2011

Firstpage

580

Lastpage

586

Abstract

In order to solve the low accuracy problem of GPU-based FFT, a mixed precision method is employed in this paper. A "precision cache" method is proposed as the supplement of the mixed precision method to calculate of twiddle factor. A work group split method is used to reduce the latency of access global memory frequently. The mixed precision FFT achieves 3 times accuracy improvement compared to CUFFT3.2 and 4 times peak performance to MKL FFT. The experiment shows that mixed precision method on CPU-GPU heterogeneous platform achieves high accuracy and efficient Fast Fourier Transform.

Keywords

cache storage; coprocessors; fast Fourier transforms; CPU-GPU heterogeneous platform; CUFFT3.2; GPU-based FFT; MKL FFT; access global memory; accuracy improvement; fast Fourier transform; low accuracy problem; mixed precision FFT; mixed precision method; peak performance; precision cache method; twiddle factor; work group split method; Accuracy; Algorithm design and analysis; Computer architecture; Discrete Fourier transforms; Graphics processing unit; Instruction sets; Signal processing algorithms; FFT; GPU; OpenCL; heterogeneous; mixed precision method;

fLanguage

English

Publisher

ieee

Conference_Titel

Computational Science and Engineering (CSE), 2011 IEEE 14th International Conference on

Conference_Location

Dalian, Liaoning

Print_ISBN

978-1-4577-0974-6

Type

conf

DOI

10.1109/CSE.2011.103

Filename

6062934