We investigate the utilization of vector registers (VRs) on reducing memory references for single instruction multiple data fast Fourier transform calculation. We propose to group the butterfly computations in several consecutive stages to maximize utilization of the available VRs and take the advantage of the symmetries in twiddle factors. All the butterflies sharing identical twiddle factors are clustered and computed together to further improve performance. The relationship between the number of fused stages and the number of available VRs is then examined. Experimental results on different platforms show that the proposed method is effective.
Feng YU
Zhejiang University
Ruifeng GE
Zhejiang University
Zeke WANG
Zhejiang University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copy
Feng YU, Ruifeng GE, Zeke WANG, "Efficient Utilization of Vector Registers to Improve FFT Performance on SIMD Microprocessors" in IEICE TRANSACTIONS on Fundamentals,
vol. E96-A, no. 7, pp. 1637-1641, July 2013, doi: 10.1587/transfun.E96.A.1637.
Abstract: We investigate the utilization of vector registers (VRs) on reducing memory references for single instruction multiple data fast Fourier transform calculation. We propose to group the butterfly computations in several consecutive stages to maximize utilization of the available VRs and take the advantage of the symmetries in twiddle factors. All the butterflies sharing identical twiddle factors are clustered and computed together to further improve performance. The relationship between the number of fused stages and the number of available VRs is then examined. Experimental results on different platforms show that the proposed method is effective.
URL: https://global.ieice.org/en_transactions/fundamentals/10.1587/transfun.E96.A.1637/_p
Copy
@ARTICLE{e96-a_7_1637,
author={Feng YU, Ruifeng GE, Zeke WANG, },
journal={IEICE TRANSACTIONS on Fundamentals},
title={Efficient Utilization of Vector Registers to Improve FFT Performance on SIMD Microprocessors},
year={2013},
volume={E96-A},
number={7},
pages={1637-1641},
abstract={We investigate the utilization of vector registers (VRs) on reducing memory references for single instruction multiple data fast Fourier transform calculation. We propose to group the butterfly computations in several consecutive stages to maximize utilization of the available VRs and take the advantage of the symmetries in twiddle factors. All the butterflies sharing identical twiddle factors are clustered and computed together to further improve performance. The relationship between the number of fused stages and the number of available VRs is then examined. Experimental results on different platforms show that the proposed method is effective.},
keywords={},
doi={10.1587/transfun.E96.A.1637},
ISSN={1745-1337},
month={July},}
Copy
TY - JOUR
TI - Efficient Utilization of Vector Registers to Improve FFT Performance on SIMD Microprocessors
T2 - IEICE TRANSACTIONS on Fundamentals
SP - 1637
EP - 1641
AU - Feng YU
AU - Ruifeng GE
AU - Zeke WANG
PY - 2013
DO - 10.1587/transfun.E96.A.1637
JO - IEICE TRANSACTIONS on Fundamentals
SN - 1745-1337
VL - E96-A
IS - 7
JA - IEICE TRANSACTIONS on Fundamentals
Y1 - July 2013
AB - We investigate the utilization of vector registers (VRs) on reducing memory references for single instruction multiple data fast Fourier transform calculation. We propose to group the butterfly computations in several consecutive stages to maximize utilization of the available VRs and take the advantage of the symmetries in twiddle factors. All the butterflies sharing identical twiddle factors are clustered and computed together to further improve performance. The relationship between the number of fused stages and the number of available VRs is then examined. Experimental results on different platforms show that the proposed method is effective.
ER -