Razvoj programske podrške za integraciju akceleratora za duboko učenje NVDLA u heterogenu platformu PULP zasnovanu na procesoru arhitekture RISC-V

master's thesis

• Title (english): Software Support for Integration of NVDLA Deep Learning Accelerator into the Heterogeneous PULP Platform Based on the RISC-V ISA
• Author: Martin Šeler
• Mentor: Josip Knezović (mentor)
• Committee member: Josip Knezović (predsjednik povjerenstva)
• Committee member: Hrvoje Mlinarić (član povjerenstva)
• Committee member: Daniel Hofman (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-17, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Učinkovito ubrzavanje operacija dubokog učenja u fazi prepoznavanja, pogotovo za energetski ograničene uređaje interneta stvari, zahtijeva primjenu heterogenih računalnih sustava s akceleratorima posebne namjene. Za korištenje takvih akceleratora potrebni su programski modeli kojima se akceleratorima zadaje posao te šalju i primaju ulazni/izlazni podaci. Ovaj rad uspoređuje programske modele sustava ubrzanim akceleratorom NVIDIA Deep Learning Accelerator (NVDLA) ili Parallel Ultra Low Power (PULP) grozdom. U radu je predložena prilagodba stoga programske podrške NVDLA s cilje prilagodbe interakcije uređaja s domaćinom baziranim na PULP platformi.

Abstract (english)

Efficient execution of deep learning operations in the inference phase, especially on power-constrained IoT devices, requires the use of heterogeneous computer systems with custom accelerators. In order to leverage such accelerators, there are programming models which are used to communicate work definitions to the accelerators and send or receive input/output data. This work compares programming models for systems accelerated by NVIDIA Deep Learning Accelerator (NVDLA) or Parallel Ultra Low Power (PULP) cluster. This work also proposes modification of the NVDLA software stack with the goal of supporting its integration with a PULP platform-based host.

• Keywords: Dubokom učenje; Zaključivanje; Heterogeno računarstvo; RISC-V; NVDLA; PULP platforma
• Keywords (english): Deep learning; Inference; Heterogeneous computing; RISC-V; NVDLA; PULP Platform
• Language: english
• URN:NBN: urn:nbn:hr:168:158231
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 13:54:59