Acceleration of image processing algorithms using simd technology

A rational choice of computing platform and software optimization that considers the specifics of processor architecture can significantly reduce the time of complex computations, improve overall system performance, and ensure efficient scalability when processing large amounts of data.This study addresses the problem of accelerating computational image processing algorithms, specifically the Gaussian smoothing algorithm, using SIMD data-level parallelism technology.The Gaussian smoothing algorithm, which is commonly used to reduce noise and remove small image details, is characterized by high computational complexity due to the need to perform numerous arithmetic operations on each pixel.In this regard, the optimization of such algorithms is a relevant task, and SIMD technology offers the potential for parallel data processing using extended processor instructions, thereby significantly enhancing computational performance.The aim of this study is to enhance the efficiency of the Gaussian smoothing algorithm through SIMDbased computation optimization.Two variants of the software implementation of the studied algorithm have been developed: one is scalar and the other has been optimized through the use of AVX-256 instructions.The optimized version employs computation vectorization, processing eight pixels simultaneously using 256-bit registers, which theoretically allows up to an eightfold speedup.Computational experiments were conducted using images with resolutions of 1920x1080 and 2560x1440 pixels, across various kernel radius and standard deviation values.The results demonstrated that implementing SIMD instructions resulted in a speed enhancement ranging from 6.9 to 7.3 times when compared to the scalar approach.When the kernel radius increased, the acceleration remained consistently high, confirming the approach's effectiveness for more complex computations.It was confirmed that execution time is primarily influenced by the kernel radius, while the standard deviation has a lesser effect, since the radius defines the filter's area of influence.The speedup achieved in the experiments is close to the theoretical maximum, demonstrating the advantages of the optimized implementation.Future research prospects include combining SIMD optimization with multithreaded processing and studying the potential of more powerful instruction sets such as AVX-512.