Performance Improvement of Gaussian Filter using SIMD Technology

Moradifar, Maryam; Shahbahrami, Asadollah

doi:10.1109/mvip49855.2020.9116883

Cited by 9 publications

(5 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The first strategy (aka coefficient propagation) broadcasts each kernel's scalar coefficient to a separate vector variable [12] [13] [14] (each vector variable has multiple copies of just one scalar coefficient). In the second strategy, all the scalar coefficients which are located in a kernel's row are copied into a vector 1. https://github.com/kelefouras/2D Convolution variable, multiple times [12] [13]. According to [12], the first method above is faster than the second (when using 16/32-bit instructions), as it achieves fewer arithmetical instructions.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Design and Implementation of 2D Convolution on x86/x64 Processors

Kelefouras

Κεραμίδας

2022

IEEE Trans. Parallel Distrib. Syst.

View full text Add to dashboard Cite

In this paper, a new method for accelerating the 2D direct Convolution operation on x86/x64 processors is presented. It includes efficient vectorization by using SIMD intrinsics, bit-twiddling optimizations, the optimization of the division operation, multi-threading using OpenMP, register blocking and the shortest possible bit-width value of the intermediate results. The proposed method, which is provided as open-source, is general and can be applied to other processor families too, e.g., Arm. The proposed method has been evaluated on two different multi-core Intel CPUs, by using twenty different image sizes, 8-bit integer computations and the most commonly used kernel sizes (3x3, 5x5, 7x7, 9x9). It achieves from 2.8× to 40× speedup over the Intel IPP library (OpenCV GaussianBlur and Filter2D routines), from 105× to 400× speedup over the gemm-based convolution method (by using Intel MKL int8 matrix multiplication routine), and from 8.5× to 618× speedup over the vslsConvExec Intel MKL direct convolution routine. The proposed method is superior as it achieves far fewer arithmetical and load/store instructions.

show abstract

Section: Related Workmentioning

confidence: 99%

“…out_even contains 15 8-bit final results of output pixels 0, 2,4,6,8,10,12,14,1 6,18,20, 22,24,26,28 Vector Division (lines 48-50). out_even contains 15 8-bit final results of output pixels 1, 3,5,7,9,11,13,15,17,19,21,23,25,27,29 Blend the 30 8-bit final results to out_even and store into memory (lines 52-57). Fig.…”

Section: Vectorizationmentioning

confidence: 99%

Design and Implementation of 2D Convolution on x86/x64 Processors

Kelefouras

Κεραμίδας

2022

IEEE Trans. Parallel Distrib. Syst.

View full text Add to dashboard Cite

show abstract

“…A picture is convolved with a Gaussian kernel when it passes through the Gaussian filter, a type of spatial filter [7]. The two characteristics that affect the effectiveness of the Gaussian filter are the kernel's magnitude and the Gaussian distribution's standard deviation.…”

Section: Introductionmentioning

confidence: 99%

“…In recent years [12][5], instances of performing simultaneous processing using SBCs have been on the rise. SBCs are perfect for concurrent processing applications due to their cheap cost, low power consumption, and compact dimensions.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Parallel Implementation of Gaussian Filter Image Processing on a Cluster of Single Board Computer

Nurul Fauzie,

Purnomo Sakti,

Rahmadwati

2023

jeeccis

View full text Add to dashboard Cite

Gaussian filters are widely used in image processing applications, such as edge detection, segmentation, and feature extraction. However, computationally intensive computations can take a long time to process large images. Therefore, a parallel algorithm implementation is necessary to accelerate the process. The authors proposed the use of Orange Pi SBCs for parallel image processing tasks involving a Gaussian filter. This paper outlines the steps for implementing a parallel Gaussian filter on a cluster of SBCs. The performance of the parallel implementation was evaluated in terms of speedup and efficiency, which are essential parameters for measuring the effectiveness of the approach. The parallel implementation speedup is described as the ratio of the time required by the serial implementation to that required by the parallel implementation. The parallel implementation efficiency is described as the speedup ratio of the number of SBCs in a cluster. The results of the performance evaluation show that the parallel implementation of the Gaussian filter on a cluster of Orange Pi SBCs can achieve significant speedup and efficiency compared to the serial implementation. The speedup increases with the number of SBCs used in the cluster. Using four SBCs can result in a speedup of up to 2.1 times faster than serial implementation. The efficiency also increases with the number of SBCs used in the cluster. Using four SBCs could achieve an efficiency of up to 53.4%.

show abstract