<p>In this paper, a batch 3D FFT algorithm for the Sunway many-core processor SW26010P is proposed. By studying the hardware characteristics of the processor, an optimization scheme suitable for batch 3D FFT computation is designed. The algorithm divides the problem into three types according to the size of the problem and the size of the LDM space. Different computation strategies are used to reduce the number of memory visits, and the algorithm is optimized in many aspects by reusing twiddle factors, vectorizing computations, optimizating task partition, and implementing double buffering. The experimental results show that the algorithm consistently outperforms the existing xMath-FFT implementation across a diverse range of FFT sizes and precision.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Implementation and optimization of batch 3D FFT for sunway many-core processor

  • Mian Huo,
  • Yuwen Zhao,
  • Fangfang Liu

摘要

In this paper, a batch 3D FFT algorithm for the Sunway many-core processor SW26010P is proposed. By studying the hardware characteristics of the processor, an optimization scheme suitable for batch 3D FFT computation is designed. The algorithm divides the problem into three types according to the size of the problem and the size of the LDM space. Different computation strategies are used to reduce the number of memory visits, and the algorithm is optimized in many aspects by reusing twiddle factors, vectorizing computations, optimizating task partition, and implementing double buffering. The experimental results show that the algorithm consistently outperforms the existing xMath-FFT implementation across a diverse range of FFT sizes and precision.