Sakin ɗakin karatu na SVT-AV1 2.0 (Scalable Video Technology AV1) an buga shi tare da aiwatar da na'urar rikodin rikodin bidiyo na AV1, don haɓakawa wanda ake amfani da hanyoyin lissafin daidaitattun kayan aikin da ke cikin Intel CPUs na zamani. Intel ne ya kirkiro aikin tare da haɗin gwiwa tare da Netflix tare da burin cimma matakin aiki wanda ya dace da yin rikodin bidiyo akan-da-tashi da kuma amfani da sabis na bidiyo akan buƙatun (VOD). A halin yanzu, ana aiwatar da ci gaba a ƙarƙashin kulawar Open Media Alliance (AOMedia), wanda ke kula da haɓaka tsarin ɓoye bidiyo na AV1. A baya can, an ƙaddamar da aikin a cikin tsarin aikin OpenVisualCloud, wanda kuma ya haɓaka SVT-HEVC da SVT-VP9 encoders. Ana rarraba lambar a ƙarƙashin lasisin BSD.
Ana iya haɗa SVT-AV1 don tsarin dangane da kowane gine-gine wanda akwai mai tarawa wanda ke goyan bayan ma'aunin C99, amma ana samun mafi kyawun aiki akan tsarin x86_64, wanda ake amfani da haɓaka haɓakawa dangane da umarnin SIMD (yana da kyawawa a samu. Goyan bayan AVX2 a cikin CPU, amma kamar yadda mafi ƙarancin ya isa kuma SS2). Yin amfani da ƙwaƙwalwar ajiya ya dogara ne da adadin na'urorin sarrafawa da aka yi amfani da su yayin ɓoyewa, wanda zaɓin "-lp" ya tsara. Saboda rikitattun algorithms da aka yi amfani da su a cikin AV1, sanya wannan tsari yana buƙatar ƙarin albarkatu fiye da sauran tsare-tsare, wanda baya ba da damar yin amfani da madaidaicin madaidaicin AV1 don canza rikodin lokaci na gaske. Misali, mai rikodin hannun jari daga aikin AV1 yana buƙatar ƙarin lissafin 5721, 5869 da 658 idan aka kwatanta da x264 (babban bayanin martaba), x264 (“high” profile) da libvpx-vp9 encoders.
Daga cikin canje-canje a cikin sabon sakin SVT-AV1:
- An yi sauyi zuwa sabon nau'in lamba, bisa ga abin da lambar farko a cikin sigar za ta canza tare da kowane canjin API/ABI.
- An yi canje-canje ga API da ke da alaƙa da sauyawa zuwa ƙarshen nunin rafi (EOS - Ƙarshen Rafi) a cikin firam na ƙarshe maimakon yin amfani da firam mara kyau, wanda ya kawar da jinkirin jiran ƙarin firam. Canjin API ɗin ya riga ya bayyana a cikin FFmpeg codebase.
- An cire yanayin bitrate mai wucewa uku (3-pass VBR), kuma yanzu an maye gurbin shi da tsarin VBR mai wucewa da yawa. Yanayin wucewa da yawa na VBR an rage zuwa wucewa biyu don tabbatar da haɗin kai tare da FFmpeg.
- An ƙara ingantawa zuwa mai rikodin, sakamakon abin da ƙarfin matsawa na saiti na M9-M13 ya karu da 1-4%, kuma yawan ƙwaƙwalwar ajiyar saiti na M5 ya ragu da 20-35% a cikin yanayin LP 8 kuma ta 1-5% a cikin wasu hanyoyin. Amfanin ƙwaƙwalwar ajiya a wasu saitattun saiti ya ragu da 1-5%.
- An aiwatar da haɓaka haɓakar inganci / saurin daidaitawa don saitattun abubuwan da suka saita babban matakin inganci. An haɓaka ingantaccen saiti na MR da 100%.
- An ƙara takamaiman haɓakawa na ARM zuwa ayyukan C-kawai.
Bugu da ƙari, za mu iya lura da sakin aikin dav1d 1.4.1, wanda a cikinsa al'ummomin VideoLAN da FFmpeg ke haɓaka ɗakin karatu tare da aiwatar da wani zaɓi na kyauta na kyauta don tsarin rikodin bidiyo na AV1. Laburaren dav1d yana goyan bayan duk fasalulluka na AV1, gami da ci-gaban nau'ikan samfura da duk ma'aunin sarrafa zurfin launi da aka bayyana a cikin ƙayyadaddun bayanai (8, 10 da 12 bits). Muhimmin fasalin dav1d shine mayar da hankali ga cimma mafi girman aiki mai yuwuwar yankewa da kuma tabbatar da ingantaccen aiki a yanayin zaren da yawa. An rubuta lambar aikin a cikin C (C99) tare da abubuwan da ake sakawa (NASM/GAS) kuma an rarraba a ƙarƙashin lasisin BSD. Taimako don gine-ginen x86, x86_64, ARMv7 da ARMv8, da tsarin aiki FreeBSD, Linux, Windows, macOS, Android da iOS an aiwatar da su.
Siffar dav1d 1.4 sananne ne don tallafawa sabbin gine-ginen Loongarch da RISC-V, da kuma gabatar da ƙarin haɓakawa dangane da umarnin AVX-512, haɓaka aikin tacewa ta 6 tap akan tsarin ARM, haɓaka ingantaccen aiki mai zare da yawa da ragewa. Girman bayanan binary akan ARM64, ARM32 da RISC-tsarin V. Kafaffen raunin CVE-2024-1580, wanda ya haifar da rubutaccen iyaka saboda yawan adadin lamba yayin sarrafa manyan girman firam.
source: budenet.ru