Programming model for heterogeneous computing systems with customizable accelerators

doctoral thesis

• Title (croatian): Programski model za raznorodne računalne sustave s prilagodljivim ubrzivačima
• Author: Branimir Pervan
• Mentor: Josip Knezović (mentor)
• Committee member: Josip Knezović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; (Department of Control and Computer Engineering); Zagreb
• Defense date and country: 2022, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Universal decimal classification (UDC): 004 - Computer science and technology. Computing. Data processing

Abstract

Nowadays, heterogeneity emerged as one of the key development points for reaching the so-called exascale domain in the field of high-performance computing. Unlike fully customized accelerators, customizable accelerators in heterogeneous systems balance performance and energy efficiency, and the ability to conform to more than one application domain. However, current programming models usually rely on the imperative programming paradigm which inefficiently exploits heterogeneity and parallelism. This thesis proposes an alternative approach to heterogeneous systems programming, by exposing a model which enables the programmer to express what is being computed, rather than how computation is executed. To deliver such a model, the use of RISE language is proposed, which was appropriately extended and adapted. As a testing platform, GAP8, a heterogeneous system with a customizable accelerator and hardware convolution engine was used. The results show that it is possible to provide a simpler and more concise programming model while keeping performance and energy efficiency at least on par compared to the conventional programming model for the target platform. Finally, this thesis achieved the following scientific contribution: 1. Programming model for heterogeneous systems with customizable accelerators based on a domain-specific language; 2. Adaptive mechanism of expression customization for accelerator utilization using high-level framework with hardware-agnostic domain transformations for performance and energy efficiency optimization.

Abstract (croatian)

Ključna točka razvoja za postizanje takozvane exascale domene u području računarstva visokih performansi je heterogenost sustava. Za razliku od potpuno prilagođenih ubrzivača, prilagodljivi ubrzivači u heterogenim sustavima balansiraju između performansi i energetske efikasnosti, te mogućnosti da se prilagode za više od jednu aplikacijsku domenu. S druge strane, trenutni programski modeli obično se oslanjaju na imperativnu programsku paradigmu koja inherentno neefikasno iskorištava heterogenost i paralelizam. Ovaj rad predlaže alternativan pristup programiranju heterogenih sustava izlaganjem modela pomoću kojega programer izražava što se računa, umjesto kako se neka operacija izvodi. Za izvedbu takvog modela, predlaže se korištenje jezika RISE, koji je na prikladan način proširen i prilagođen. Kao testna platforma korišten je heterogeni sustav GAP8 s prilagodljivim ubrzivačem, te ubrzivačem za izračun operacije konvolucije. Rezultati pokazuju kako je, korištenjem jednostavnijeg i konciznijeg programskog modela, moguće zadržati performanse i energetsku efikasnost barem jednakima u usporedbi s konvencionalnim programskim modelom za ciljnu platformu. Ova disertacija postigla je sljedeći znanstveni doprinos: 1. Programski model za raznorodne sustave s prilagodljivim ubrzivačima temeljen na domenski specifičnom jeziku; 2. Adaptivni mehanizam prilagodbe izraza za iskorištavanje sklopovskog ubrzivača primjenom domenskih transformacija na višim razinama neovisnim o detaljima sklopovske platforme s ciljem optimizacije performanci i energetske učinkovitosti.

• Keywords: domain-specific languages; heterogeneous systems; parallel processing; multicore systems; RISC-V
• Keywords (croatian): domenski specifiˇcni jezici; heterogeni sustavi; paralelno procesiranje; višejezgreni sustavi; RISC-V
• Language: english
• URN:NBN: urn:nbn:hr:168:513967
• Promotion: 2022
• Study programme: Title: Computer Science; Study programme type: university; Study level: postgraduate; Academic / professional title: Doktor znanosti (Doktor znanosti)
• Type of resource: Text
• Extent: xvi, 140 str. : graf. prikazi
• File origin: Born digital
• Access conditions: Open access
• Repository: FER Repository
• Created on: 2022-05-23 13:38:24