diplomsko delo
Filip Urh (Author), Damjan Strnad (Mentor)

Abstract

V diplomskem delu preučimo in predstavimo novo funkcionalnost arhitekture CUDA. Gre za dinamični paralelizem, ki omogoča poganjanje programskih jeder neposredno iz grafične procesne enote. V začetku podrobno predstavimo arhitekturo CUDA in algoritem CART za gradnjo odločitvenih dreves, ki smo ga uporabili za demonstracijo uporabe dinamičnega paralelizma. Algoritem smo implementirali v zaporedni različici na CPE ter v paralelnih različicah z in brez dinamičnega paralelizma na GPE. Predstavili smo primerjalne meritve časov izvajanja vseh treh implementacij in ugotovili, da uporaba dinamičnega paralelizma omogoča krajši čas izvajanja in lažjo implementacijo algoritma.

Keywords

grafična procesna enota;odločitveno drevo;CART;CUDA;dinamični paralelizem;

Data

Language: Slovenian
Year of publishing:
Typology: 2.11 - Undergraduate Thesis
Organization: UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher: F. Urh
UDC: 004.274.032.24(043.2)
COBISS: 19332630 Link will open in a new window
Views: 1204
Downloads: 75
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: DYNAMIC PARALLELISM ON A GPU
Secondary abstract: In the thesis, we investigate and introduce new functionality of CUDA architecture - the dynamic parallelism, which allows us to start new kernels directly from the GPU. In the beginning, we present in detail the CUDA architecture and algorithm CART for the construction of decision trees, which was used to demonstrate the use of dynamic parallelism. The serial algorithm was implemented on the CPU and the parallel versions with and without dynamic parallelism were implemented on the GPU. We presented a comparison of execution times for all three implementations and concluded that the usage of dynamic parallelism provides faster execution and easier implementation of the algorithm.
Secondary keywords: graphics processing;decision tree;CART;CUDA;dynamic parallelism;
URN: URN:SI:UM:
Type (COBISS): Bachelor thesis/paper
Thesis comment: Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije
Pages: IX, 39 str.
ID: 8890146