Memory-accelerated parallel method for multidimensional fast fourier implementation on GPU

Hu, Yichang; Lu, Lu; Li, Cuixu

doi:10.1007/s11227-022-04570-9

Cited by 4 publications

(3 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Parallelization strategies FHT have also evolved, with recent works exploring advanced techniques such as dynamic parallelism [39], shared memory utilization [25], and warplevel programming [40]. These approaches aim to reduce the computational overhead and improve the scalability of the FHT on GPUs, enabling its application in more complex and data-intensive scenarios.…”

Section: Related Workmentioning

confidence: 99%

Accelerating the Fast Hadamard Single-Pixel Imaging (FHSI) on GPUs: Techniques and Optimizations

Quero,

Durini,

Rangel-Magdaleno

et al. 2024

Preprint

View full text Add to dashboard Cite

The recent advancements in edge computing power are primarily attributable to technological innovations enabling accelerators with extensive hardware parallelism. One practical application is in computer imaging (CI), where GPU acceleration is pivotal, especially in reconstructing 2D images through techniques like Single-Pixel Imaging (SPI). In SPI, compressive sensing (CS) algorithms, deep learning, and Fourier transformation are essential for 2D image reconstruction. These algorithms derive substantial performance enhancements through parallelism, thereby reducing processing times. These techniques fully utilize the potential of the GPU by implementing several strategies. These include optimizing memory accessed, expanding loops for efficiency, designing effective computational kernels to reduce the number of operations, using asynchronous operations for better performance, and increasing the number of actively running threads and warps. In lab scenarios, integrating embedded GPUs becomes essential for algorithmic optimization on SoC-GPUs. This study focuses on quickly improving the Fast Hadamard Single-Pixel Imaging (FHSI) for 2D image reconstruction on Nvidia's Xavier platform. By implementing various parallel computing techniques in PyCUDA, we managed to speed up the process by approximately 10 times, significantly reducing processing times to nearly real-time levels.

show abstract

Section: Related Workmentioning

confidence: 99%

Accelerating the Fast Hadamard Single-Pixel Imaging (FHSI) on GPUs: Techniques and Optimizations

Quero,

Durini,

Rangel-Magdaleno

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…AMD also released the rocFFT library that runs on the Radeon Open Computing Platform (ROCm) [18]. Hu and others [19] optimized the memory-access pattern for FFT computations on ROCm. The proposed implementation achieved speed enhancements ranging from 25% to 250% compared with the speed of rocFFT.…”

Section: Related Workmentioning

confidence: 99%

Large‐scale 3D fast Fourier transform computation on a GPU

Lee

Kim

2023

ETRI Journal

View full text Add to dashboard Cite

We propose a novel graphics processing unit (GPU) algorithm that can handle a large‐scale 3D fast Fourier transform (i.e., 3D‐FFT) problem whose data size is larger than the GPU's memory. A 1D FFT‐based 3D‐FFT computational approach is used to solve the limited device memory issue. Moreover, to reduce the communication overhead between the CPU and GPU, we propose a 3D data‐transposition method that converts the target 1D vector into a contiguous memory layout and improves data transfer efficiency. The transposed data are communicated between the host and device memories efficiently through the pinned buffer and multiple streams. We apply our method to various large‐scale benchmarks and compare its performance with the state‐of‐the‐art multicore CPU FFT library (i.e., fastest Fourier transform in the West [FFTW]) and a prior GPU‐based 3D‐FFT algorithm. Our method achieves a higher performance (up to 2.89 times) than FFTW; it yields more performance gaps as the data size increases. The performance of the prior GPU algorithm decreases considerably in massive‐scale problems, whereas our method's performance is stable.

show abstract

“…Migrating existing FT algorithms directly to NPUs is infeasible due to the distinctive hardware architecture of NPUs. This is because the matrix and vector units of NPUs process data blocks in a serial manner, while GPUs and FPGAs process bytes with multiple threads in parallel [15]. Consequently, the conventional fast FT algorithm may not be suitable for NPUs.…”

Section: Introductionmentioning

confidence: 99%

Research on High-Performance Fourier Transform Algorithms Based on the NPU

Li,

Zuo,

Feng

et al. 2024

Applied Sciences

View full text Add to dashboard Cite

Backpack computers require powerful, intelligent computing capabilities for field wearables while taking energy consumption into careful consideration. A recommended solution for this demand is the CPU + NPU-based SoC. In many wearable intelligence applications, the Fourier Transform is an essential, computationally intensive preprocessing task. However, due to the unique structure of the NPU, the conventional Fourier Transform algorithms cannot be applied directly to it. This paper proposes two NPU-accelerated Fourier Transform algorithms that leverage the unique hardware structure of the NPU and provides three implementations of those algorithms, namely MM-2DFT, MV-2FFTm, and MV-2FFTv. Then, we benchmarked the speed and energy efficiency of our algorithms for the gray image edge filtering task on the Huawei Atlas200I-DK-A2 development kits against the Cooley-Tukey algorithm running on CPU and GPU platforms. The experiment results reveal MM-2DFT outperforms OpenCL-based FFT on NVIDIA Tegra X2 GPU for small input sizes, with a 4- to 8-time speedup. As the input image resolution exceeds 2048, MV-2FFTv approaches GPU computation speed. Additionally, two scenarios were tested and analyzed for energy efficiency, revealing that cube units of the NPU are more energy efficient. The vector and CPU units are better suited for sparse matrix multiplication and small-scale inputs, respectively.

show abstract

Memory-accelerated parallel method for multidimensional fast fourier implementation on GPU

Cited by 4 publications

References 34 publications

Accelerating the Fast Hadamard Single-Pixel Imaging (FHSI) on GPUs: Techniques and Optimizations

Accelerating the Fast Hadamard Single-Pixel Imaging (FHSI) on GPUs: Techniques and Optimizations

Large‐scale 3D fast Fourier transform computation on a GPU

Research on High-Performance Fourier Transform Algorithms Based on the NPU

Contact Info

Product

Resources

About