prikaz prve stranice dokumenta Primjena ARM NEON tehnologije
Access restricted to students and staff of home institution
master's thesis
Primjena ARM NEON tehnologije
2016. urn:nbn:hr:190:848332
Čerljenko, Ricardo
University of Rijeka
Faculty of Engineering

Cite this document

Čerljenko, R. (2016). Primjena ARM NEON tehnologije (Master's thesis). Rijeka: University of Rijeka, Faculty of Engineering. Retrieved from https://urn.nsk.hr/urn:nbn:hr:190:848332

Čerljenko, Ricardo. "Primjena ARM NEON tehnologije." Master's thesis, University of Rijeka, Faculty of Engineering, 2016. https://urn.nsk.hr/urn:nbn:hr:190:848332

Čerljenko, Ricardo. "Primjena ARM NEON tehnologije." Master's thesis, University of Rijeka, Faculty of Engineering, 2016. https://urn.nsk.hr/urn:nbn:hr:190:848332

Čerljenko, R. (2016). 'Primjena ARM NEON tehnologije', Master's thesis, University of Rijeka, Faculty of Engineering, accessed 28 October 2024, https://urn.nsk.hr/urn:nbn:hr:190:848332

Čerljenko R. Primjena ARM NEON tehnologije [Master's thesis]. Rijeka: University of Rijeka, Faculty of Engineering; 2016 [cited 2024 October 28] Available at: https://urn.nsk.hr/urn:nbn:hr:190:848332

R. Čerljenko, "Primjena ARM NEON tehnologije", Master's thesis, University of Rijeka, Faculty of Engineering, Rijeka, 2016. Available at: https://urn.nsk.hr/urn:nbn:hr:190:848332

Please login to the repository to save this object to your list.
Metadata
TitlePrimjena ARM NEON tehnologije
AuthorRicardo Čerljenko
MentorKristijan Lenac (mentor)
Committee memberKristijan Lenac (predsjednik povjerenstva)
GranterUniversity of Rijeka
Faculty of Engineering
Rijeka
Defense date and country2016-11, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Mechanical Engineering
Abstract
Ovaj rad govori o pokušaju i rezultatima konverzije intrističnih SSSE3 instrukcija x86 arhitekture u ekvivalentna NEON ubrzanja ARM arhitekture. Kako bi se to postiglo, bilo je potrebno detaljno proučiti osnove tih dviju arhitektura kako bi se shvatio način rada instrukcija. Ubrzanja nije jednostavno postići potrebno je uložiti znatan trud u programiranje softvera na način da bude optimalan za ubrzanja što zahtjeva dobro shvaćanje tehnika poravnavanja i vektoriziranja grupa podataka. Upravo iz tog razloga se tijekom zadnjih par godina ulaže veliki trud u razvoj posebnih modula za moderne prevodioce koji bi bili sposobni da sami optimiziraju programski kod (automatska vektorizacija podataka). Prevođenje instrukcija testiralo se na ugradbenom računalu ARM arhitekture te su dobiveni rezultati uspoređivali s onima bez ubrzanja te s onima gdje je prevodioc koristio tehnike automatske vektorizacije. Na dobivenim rezultatima jasno je vidljivo da je automatska vektorizacija postigla najbolje performanse, iza nje slijedi transformirani ručno pisani kod te očekivano, najlošije rezultate ima kod bez ubrzanja (čisti skalarni kod). Bez obzira na rezultate, sva su tri načina implementacije brzo i stabilno odrađivala posao koji im je namijenjen, tj. pretvorba video formata u stvarnom vremenu za Intel RealSense R200 kameru.
Abstract (english)
The effort of this work is about attempt of converting intrinsic SSSE3 instructions of x86 architecture into equivalent NEON speedups of ARM architecture. In order to do that, it was necessary to cover the basics of both architectures in detail for better understanding of how the instructions are processed. Speedups are not easy to accomplish a great effort must be taken to learn how to program the software so that is in suitable form for optimizations which means very good understanding of terms like data alignment and data vectorization. For that particular reason, over the last few years many of the modern compilers are trying to develop near-perfect automatic vectorization modules so that people can program software easier. Process of converting instructions was tested on embedded computer with ARM architecture. Results were compared with the automatic vectorization results and with results of clean scalar code (no speedups). It is clear that automatic vectorization gives the best performance, manually written code is at second place and, obviously, the worst results are with clean scalar code. Nevertheless, all three modes gave a stable and satisfying performance used on video converting rutine on Intel RealSense R200 camera.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:190:848332
Study programmeTitle: Computer Engineering
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
Type of resourceText
File originBorn digital
Access conditionsAccess restricted to students and staff of home institution
Terms of useLicence In copyright icon
Created on2017-01-24 11:23:15