Dynamic Partitioning-based JPEG Decompression on Heterogeneous Multicore Architectures

Abstract

With the emergence of social networks and improvements in computational photography, billions of JPEG images are shared and viewed on a daily basis. Desktops, tablets and smartphones constitute the vast majority of hardware platforms used for displaying JPEG images. Despite the fact that these platforms are heterogeneous multicores, no approach exists yet that is capable of joining forces of a system's CPU and GPU for JPEG decoding.In this paper we introduce a novel JPEG decoding scheme for heterogeneous architectures consisting of a CPU and an OpenCLprogrammable GPU. We employ an offline profiling step to determine the performance of a system's CPU and GPU with respect to JPEG decoding. For a given JPEG image, our performance model uses (1) the CPU and GPU performance characteristics, (2) the image entropy and (3) the width and height of the image to balance the JPEG decoding workload on the underlying hardware. Our runtime partitioning and scheduling scheme exploits task, data and pipeline parallelism by scheduling the non-parallelizable entropy decoding task on the CPU, whereas inverse cosine transformations (IDCTs), color conversions and upsampling are conducted on both the CPU and the GPU. Our kernels have been optimized for GPU memory hierarchies.We have implemented the proposed method in the context of the libjpeg-turbo library, which is an industrial-strength JPEG encoding and decoding engine. Libjpeg-turbo's hand-optimized SIMD routines for ARM and x86 constitute a competitive yardstick for the comparison to the proposed approach. Retro-fitting our method with libjpeg-turbo provides insights on the software-engineering aspects of re-engineering legacy code for heterogeneous multicores.We have evaluated our approach for a total of 7194 JPEG images across three high-and middle-end CPU-GPU combinations. We achieve speedups of up to 4.2x over the SIMD-version of libjpeg-turbo, and speedups of up to 8.5x over its sequential code. Taking into account the non-parallelizable JPEG entropy decoding part, our approach achieves up to 95% of the theoretically attainable maximal speedup, with an average of 88%.

CPU-GPU hybrid computing for feature extraction from video stream

“…Therefore, it is important to find an effective method to make full use of all the available computational resources of both the CPU and GPU. Recently, some approaches [3,4,5,6,7] have been developed to perform a specific task using both multi-core CPU and GPU simultaneously, instead of the CPU or GPU alone. In this paper, we present a way to distribute the workload into both the CPU and GPU, with a performance prediction model (i.e., a static strategy) including characteristics of feature extraction from the video stream data.…”

PeaPaw

“…Workload partitioning aims to partition a given workload (measured by flops for computation and by bytes for memory access) via dividing code or dataset among the processors of a given multiprocessor platform (e.g., Tang et al [2013], Ma et al [2012], and Sodsong et al [2014]). For homogeneous multiprocessor platforms, the general criterion [Huang and Feng 2009] of workload partitioning is to evenly partition the input dataset among the processors since dataset partitioning (DP) can achieve load balancing easily and each processor behaves identically for a particular algorithm.…”

“…A heterogeneous platform has different processors that deliver distinct performance/energy characteristics, and this processor heterogeneity increases the design space for partitioning the workload between different processors [Angiolini et al 2006;Chung et al 2010;Koufaty et al 2010]. For example, some studies have shown that partitioning the code [Daga and Nutter 2012;Sodsong et al 2014] instead of the input dataset [Zhong et al 2012;Grasso et al 2013] can better exploit the performance/energy potential of heterogeneous platforms. Such consideration is not necessary for uniprocessor or homogeneous multiprocessor platforms.…”

JParEnt: Parallel entropy decoding for JPEG decompression on heterogeneous multicore architectures

“…In previous work, we have devised a JPEG decoder capable of accelerating inverse discrete cosine transformations (IDCT), image upsampling and YCbCr to RGB color space conversions on systems consisting of a CPU and a GPU. Entropy decoding is the only remaining sequential task with this decoding scheme and thus stands in the way to further speedups.…”

“…Data transfers between CPU and GPU via the PCI bus can generate significant overhead . In our previous work, entropy decoding was conducted on the CPU, which necessitated the transfer of the inflated DCT coefficients to the GPU, where they eventually became the IDCT input. To reduce transfer costs, a run‐length encoded format was employed.…”