Integrating Streaming Computations for Efficient Execution on Novel Multicore Architectures

Abstract

Original scientific paperStreaming has emerged as an important model in present-day applications, ranging from multimedia to scientific computing. Moreover, the emergence of new multicore architectures has resulted with new challenges in efficient utilization of available computational resources. Streaming model offers the portability and scalability of performance with the increasing number of cores. In this paper we propose a tool which enables the implementation of the compute-intensive stream processing kernels as portable modules in general-purpose applications. Resulting modules can be efficiently reused with high degree of scalability in regard to increasing number of processing cores.Key words: Streaming model, Parallel programming, Signal processing, MulticoreIntegracija tokovnog modela za učinkovito izvoenje na višejezgrenim računalnim arhitekturama. Tokovni računalni model predstavlja zanimljivo područje istraživanja s ciljem ubrzanja kako multimedijskih, tako i znanstvenih aplikacija. Isto tako, pojava višejezgrenih računalnih arhitektura rezultirala je povećanjem zanimanja za istraživanje metoda i modela koji bi omogućili učinkovito iskorištavanje postojećih paralelnih resursa. Tokovni model omogućuje istovremeno visok stupanj apstrakcije, prenosivost i skalabinost aplikacija s obzirom na povećanje računskih jezgri. U ovom ječlanku predložen pristup koji omogućuje implementaciju računski zahtjevnih dijelova aplikacija u tokovnom modelu te njihovu integraciju u vidu prenosivih modula. Na taj način ostvareno je ubrzanje cjelokupnih aplikacija pri izvoenju na višejezgrenim procesorima.Ključne riječi: tokovni računalni model, paralelni sustavi, obrada signala, višejezgrena računala

Lossless Image Compression Exploiting Streaming Model for Efficient Execution on Multicores

“…We improve the execution performance of CBPC utilizing the streaming model of computation and expressing compute intensive part of the algorithm in the StreamIt programming language. Following the observations from our previous work on integrating streaming computations we modified our experimental tool reported in [15], which resulted in STREAMGATE interface and runtime. The main purpose of the interface is to provide existing applications the easy way to integrate high-level and optimized streaming computations as reusable kernels.…”

“…Additionally, our intention was to enable the portability among different architectures, from symmetric multicores with shared memory to heterogeneous multiprocessor architectures which rapidly emerge in the embedded computing devices. In order to end up with a scalable and portable implementation we exploit streaming compilation infrastructure that we described previously in [15], which we further extended in order to generate reusable streaming kernels in the form of static library. This approach enabled us to obtain scalable performance improvement in regard to increasing the number of processing cores without any modifications to existing source.…”

An Efficient Parallel Computing Method for the Processing of Large Sensed Data

“…Since the release of the first generation dual-core processor by IBM in 2001, Sun in 2004, and AMD in 2005, more and more cores are built into a single processor with the development of multicore technology [11]. Multi-core architectures are today's dominant computing platforms [12,13]. Parallel architectures and programming models are not independent.…”

Parallelizing MPEG Decoder with Scalable Streaming Computation Kernels

“…Our approach is based on a functional decomposition of existing (non-parallelized) application or a newly designed application, where the most compute-intensive elements are (re)designed and (re)implemented in the streaming domain, whereas non-critical elements are left/implemented in the sequential domain. The streaming domain part of the application is implemented using StreamIt language [7], while the bridge between the streaming and the sequential domain is provided and managed by STREAMGATE interface and run-time [8].…”