?url_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Adc&rft.title=Dise%C3%B1o+de+t%C3%A9cnicas+para+la+paralelizaci%C3%B3n+de+algoritmos+de+descodificaci%C3%B3n+de+video+digital+sobre+la+plataforma+NVIDIA+JETSON+TX2&rft.creator=Garc%C3%ADa+An%C3%A9s%2C+Daniel&rft.contributor=Pescador+del+Oso%2C+Fernando&rft.contributor=Chavarr%C3%ADas+Lapastora%2C+Miguel&rft.subject=Telecommunications&rft.description=Este+proyecto+est%C3%A1+enmarcado+dentro+de+la+l%C3%ADnea+de+investigaci%C3%B3n+relacionada+con+la+modificaci%C3%B3n%2Fdescodificaci%C3%B3n+de+v%C3%ADdeo+digital+del+grupo+de+investigaci%C3%B3n+GDEM+viene+trabajando+desde+hace+ya+varios+a%C3%B1os.+A+lo+largo+del+documento+se+desarrolla+una+metodolog%C3%ADa+para+la+paralelizaci%C3%B3n+de+uno+de+los+algoritmos+de+filtrado+de+un+descodificador+de+v%C3%ADdeo+digital+que+puede+ser+generalizable+a+otros+algoritmos+de+descodificaci%C3%B3n.+Esta+metodolog%C3%ADa+permite+llevar+a+cabo+una+implementaci%C3%B3n+de+un+descodificador+de+v%C3%ADdeo%2C+basado+en+el+est%C3%A1ndar+HEVC%2C+en+la+plataforma+heterog%C3%A9nea+NVIDIA+Jetson+TX2.+En+esta+implementaci%C3%B3n%2C+el+algoritmo+de+filtrado+ser%C3%A1+ejecutado+por+la+GPU%2C+de+forma+que+se+pueda+aprovechar+la+gran+capacidad+de+c%C3%B3mputo+en+paralelo+que+ofrece+este+procesador%2C+y+el+resto+de+los+algoritmos+ser%C3%A1n+ejecutados+por+la+CPU+de+dicha+plataforma.+Tras+esta+primera+implementaci%C3%B3n%2C+se+procede+a+realizar+varias+versiones+de+%C3%A9sta+usando+distintos+tipos+de+procedimientos+en+la+parte+de+sincronizaci%C3%B3n+entre+los+procesadores%3A+transferencias+de+memoria+usando+memoria+unificada+o++memcpy%2C+al+igual+que+distintos+tipos+de+t%C3%A9cnicas+de+paralelizaci%C3%B3n+en+la+CPU%3A+pthreads+y+OpenMP.+Una+vez+que+el+desarrollo+de+la+metodolog%C3%ADa+y+la+implementaci%C3%B3n+han+finalizado%2C+se+realiza+un+banco+de+pruebas+empleando+ocho+secuencias+de+v%C3%ADdeo%3A+cuatro+con+resoluci%C3%B3n+en+p%C3%ADxeles+de+1920x1080+y+otras+cuatro+con+resoluci%C3%B3n+en+p%C3%ADxeles+de+2560x1600.+Estas+pruebas+se+llevan+a+cabo+con+las+distintas+versiones+del+programa.+En+primer+lugar%2C+la+versi%C3%B3n+original.+En+segundo+lugar%2C+la+versi%C3%B3n+en+la+cual+se+usa+pthreads+para+la+creaci%C3%B3n+y+sincronizaci%C3%B3n+entre+los+hilos+de+la+CPU+y+el+filtrado+se+lleva+a+cabo+en+la+GPU.+Por+%C3%BAltimo%2C+la+versi%C3%B3n+en+la+que+se+emplea+OpenMP+para+la+creaci%C3%B3n+y+sincronizaci%C3%B3n+de+hilos+en+la+CPU+y+el+filtrado+se+realiza+en+la+GPU.+A+partir+de+los+resultados+es+posible+apreciar+las+ventajas+e+inconvenientes+que+el+uso+de+una+plataforma+heterog%C3%A9nea+ofrece.+Esta+implementaci%C3%B3n+abre+la+puerta+a+futuras+mejoras+sobre+dicho+trabajo+en+las+distintas+implementaciones+de+descodificadores+de+v%C3%ADdeo+digital+sobre+plataformas+heterog%C3%A9neas.+%0D%0A%0D%0AAbstract%3A+%0D%0A%0D%0AThis+project+belongs+to+the+coding%2Fdecoding+research+line+of+GDEM+researchers+group%2C+where+several+scientific+contributions+have+been+made+over+the+last+years.+Throughout+this+document+a+methodology+for+the+parallelization+of+a+digital+video+decoding+filtering+algorithm+is+developed.+This+methodology+can+be+generalised+so+that+other+video+decoding+algorithms+could+be+parallelized.+Alongside+developing+the+mentioned+methodology%2C+an+implementation+of+a+video+decoder+(based+on+the+HEVC+standard)+in+the+heterogeneous+platform+NVIDIA+Jetson+TX2+is+made.+In+this+implementation%2C+the+filtering+algorithm+will+be+executed+in+the+GPU%2C+to+take+advantage+of+the+high+parallel+computing+capability+this+kinds+of+processors+have%2C+and+the+rest+of+the+decoding+algorithms+are+executed+in+the+CPU.+In+this+process%2C+several+problems+have+been+encountered+and+respectively+solved%2C+such+as+architecture+compatibility+problems+with+the+OpenHEVC+software%2C+synchronization+problems+between+the+different+processors+in+the+platform+and+concurrency+problems+in+the+GPU.+After+this+first+implementation%2C+several+versions+are+created+using+different+approaches+for+the+synchronization+between+processors%3A+using+unified+memory+or+memcpy+for+memory+transfers+and+data+management%2C+as+well+as+using+different+parallel+techniques+in+the+CPU%3A+pthreads+and+OpenMP.+Once+the+methodology+and+the+implementations+have+been+finished%2C+a+test+pool+is+made+using+eight+different+video+sequences%3A+four+of+them+have+FullHD+resolution+(1920x1080+pixels)+and+the+other+four+have+2K+resolution+(2560x1600+pixels).+This+sequences+are+tested+using+several+different+versions+of+the+decoder.+Firstly%2C+the+original+OpenHEVC+decoder+is+used.+Secondly%2C+the+version+used+is+the+one+in+which+DBF+is+executed+in+the+GPU+and+the+rest+of+the+program+is+executed+in+the+CPU+using+pthreads+to+create+and+synchronize+all+CPU+threads.+Lastly%2C+the+version+used+is+the+one+in+which+DBF+is+executed+in+the+GPU+and+the+rest+of+the+decoder+is+executed+in+the+CPU+using+OpenMP+to+create+and+manage+the+CPU+threads.+This+tests+offer+several+results+where+is+possible+to+see+and+understand+the+advantages+and+disadvantages+heterogeneous+platforms+have.+This+implementation+offers+a+new+opportunity+to+future+improvements+on+digital+video+decoder%E2%80%99s+implementation+in+heterogeneous+platforms.&rft.publisher=E.T.S.I+y+Sistemas+de+Telecomunicaci%C3%83%C2%B3n+(UPM)&rft.rights=https%3A%2F%2Fcreativecommons.org%2Flicenses%2Fby-nc-nd%2F3.0%2Fes%2F&rft.date=2019-07-17&rft.type=info%3Aeu-repo%2Fsemantics%2FbachelorThesis&rft.type=Final+Project&rft.type=PeerReviewed&rft.format=application%2Fpdf&rft.language=spa&rft.format=application%2Fzip&rft.language=spa&rft.rights=info%3Aeu-repo%2Fsemantics%2FrestrictedAccess&rft.identifier=https%3A%2F%2Foa.upm.es%2F62871%2F