Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Video Encoding > VP9 and AV1

Reply
 
Thread Tools Search this Thread Display Modes
Old 3rd December 2019, 22:47   #41  |  Link
utack
Registered User
 
Join Date: Apr 2018
Posts: 63
Beating a dead horse at this point maybe, but dav1d having a milestone for better PPC support and no word about adding basic 10bit support seems extremely odd
Netflix has signalled they are only interested in 10bit content, Youtube started encoding 10bit for their new higher resolution videos as well.
It should clearly be a priority over armv7 and PPC assembly, and imho all the "make it fast" milestones are not reached yet.
utack is offline   Reply With Quote
Old 4th December 2019, 02:29   #42  |  Link
Blue_MiSfit
Derek Prestegard IRL
 
Blue_MiSfit's Avatar
 
Join Date: Nov 2003
Location: Los Angeles
Posts: 5,988
I also hope that the dav1d team focuses more on 10 bit soon. I think social media / user generated content is a huge use case for AV1, and most of that content is 8 bit for now.
Blue_MiSfit is offline   Reply With Quote
Old 4th December 2019, 23:06   #43  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by Blue_MiSfit View Post
I also hope that the dav1d team focuses more on 10 bit soon. I think social media / user generated content is a huge use case for AV1, and most of that content is 8 bit for now.
Do we have any evidence that 8-bit sources encode better in 10-bit than 8-bit in AV1? While that was true for H.264, it was much less so for HEVC, and I don't see why AV1 would have any regressions versus HEVC in that regard.

SW encoding and decoding of >8-bit content is always at least 25% slower, and can be more depending on the bottlenecks. And there's really not much point in doing >8-bit unless the source or display controller can do more than that. Most social media is consumed on phones and computers, for which very few end-to-end >8-bit pipelines exist. And with really high ppi, dithering is nigh invisible.

10-bit is much more valuable on living room screens, which are much larger and have native >8-bit support.

I think everything is going to go half float linear light for internal processing next decade, to make tone mapping, particularly of mixed color space content, way easier and better.

Sent from my SM-T837V using Tapatalk
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 5th December 2019, 01:56   #44  |  Link
hajj_3
Registered User
 
Join Date: Mar 2004
Posts: 1,120
Dav1d v0.5.2 'Asiatic Cheetah' changelog:

ARM32 optimizations for loopfilter, ipred_dc|h|v
Add section-5 raw OBU demuxer
Improve the speed by reducing the L2 cache collisions
Fix minor issues, including compilation on some OSes
hajj_3 is offline   Reply With Quote
Old 5th December 2019, 12:58   #45  |  Link
soresu
Registered User
 
Join Date: May 2005
Location: Swansea, Wales, UK
Posts: 196
Quote:
Originally Posted by Blue_MiSfit View Post
I also hope that the dav1d team focuses more on 10 bit soon. I think social media / user generated content is a huge use case for AV1, and most of that content is 8 bit for now.
It's a shame Google pursued gAV1 rather than shunting engineer time to dav1d, they could work much faster on all fronts with some serious money behind them.

I get the whole competition is good angle, but it doesn't seem to have made much of a difference, other than prompting dav1d to shore up their ARM32 priorities - from which it seems dav1d are soundly ahead on all fronts again.
soresu is offline   Reply With Quote
Old 5th December 2019, 13:00   #46  |  Link
soresu
Registered User
 
Join Date: May 2005
Location: Swansea, Wales, UK
Posts: 196
Quote:
Originally Posted by utack View Post
Beating a dead horse at this point maybe, but dav1d having a milestone for better PPC support and no word about adding basic 10bit support seems extremely odd
Netflix has signalled they are only interested in 10bit content, Youtube started encoding 10bit for their new higher resolution videos as well.
It should clearly be a priority over armv7 and PPC assembly, and imho all the "make it fast" milestones are not reached yet.
They have basic 10 bit support I think, just very little SIMD asm to accelerate it beyond C code.
soresu is offline   Reply With Quote
Old 7th February 2020, 09:02   #47  |  Link
dapperdan
Registered User
 
Join Date: Aug 2009
Posts: 201
Quote:
Originally Posted by birdie View Post
This announcement makes no sense: I don't know a single mobile SoC which supports AV1 HW decoding acceleration right now which means the poor users who will watch AV1 content will decimate their battery life.
They note that they're using Dav1d and supporting development of 10-bit assembly speedups so that's what wil be used until hardware arrives.

Currently it's opt in, you need to flick the pre-existing switch that says you want to save data when watching on mobile. I assume most of the time this just reduced file size and therefore quality, but now they have the extra option of switching codec as well.

It would be nice if they'd run the numbers and publish them, but against the backdrop of mobile streaming and a phone screen fully on, I'm not sure the lack of hardware decode will be that noticeable. I think Instagram were already shipping a VP9 software decoder on Android I don't think they were even using the fast ffmpeg decoder, just libaom.
dapperdan is offline   Reply With Quote
Old 7th February 2020, 17:21   #48  |  Link
soresu
Registered User
 
Join Date: May 2005
Location: Swansea, Wales, UK
Posts: 196
Quote:
Originally Posted by dapperdan View Post
I think Instagram were already shipping a VP9 software decoder on Android I don't think they were even using the fast ffmpeg decoder, just libaom.
Ouch, that's just being bloody minded to user battery life that is.

Here's the gitlab issue link for dav1d NEON, 4 opts for 16bpc already merged, 2 more lined up it seems.

Last edited by soresu; 7th February 2020 at 17:29.
soresu is offline   Reply With Quote
Old 8th February 2020, 23:39   #49  |  Link
Nintendo Maniac 64
Registered User
 
Nintendo Maniac 64's Avatar
 
Join Date: Nov 2009
Location: Northeast Ohio
Posts: 447
dav1d v0.5.0 performance on Ubuntu Linux 20.04 LTS snapshot with the new 64core/128thread Threadripper 3990X: https://www.phoronix.com/scan.php?pa...er-linux&num=2
__________________
____HTPC____  | __Desktop PC__
2.93GHz Xeon x3470 (4c/8t Nehalem) | 4.5GHz 1.24v dual-core Haswell G3258
Radeon HD5870  | Intel iGPU      
2x2GB+2x1GB DDR3-1333 | 4x4GB DDR3-1600       
Nintendo Maniac 64 is offline   Reply With Quote
Old 10th February 2020, 19:32   #50  |  Link
MoSal
Registered User
 
Join Date: Jun 2013
Posts: 95
Quote:
Originally Posted by soresu View Post
Ouch, that's just being bloody minded to user battery life that is.

Here's the gitlab issue link for dav1d NEON, 4 opts for 16bpc already merged, 2 more lined up it seems.
Quote:
Originally Posted by SmilingWolf View Post
Packaging differences. Compiling from sources and using "ninja -vC release install" puts the header(s) under <somewhere>/include/libvmaf/<headers>.h
Arch's package maintainer probably follows in the footsteps of the Debian package, that leaves the header under /usr/include/libvmaf.h

Arch's package maintainer is not following anyone's footsteps other than upstream. It''s just the Makefile from v1.3.15 predates the changes that made the new Makefile just a wrapper around meson/ninja.

Anyway, the last time (~18 months ago) I tried linking libvmaf to ffmpeg, I got runtime crashes after I managed to find a version that compiles. So I figured the library is not really reliable API/ABI wise for external linkage use-cases. So I opted to just script around the provided executable vmafossexec.
__________________
https://github.com/MoSal
MoSal is offline   Reply With Quote
Old 12th February 2020, 14:06   #51  |  Link
NikosD
Registered User
 
Join Date: Aug 2010
Location: Athens, Greece
Posts: 2,901
Netflix starts using AV1 initially on Android mobile for offline viewing, leveraging dAV1d 10 bit optimized app.

Interesting.

Quote:
-- Netflix --

Today we are excited to announce that Netflix has started streaming AV1 to our Android mobile app. AV1 is a high performance, royalty-free video codec that provides 20% improved compression efficiency over our VP9† encodes. AV1 is made possible by the wide-ranging industry commitment of expertise and intellectual property within the Alliance for Open Media (AOMedia), of which Netflix is a founding member.

Our support for AV1 represents Netflix’s continued investment in delivering the most efficient and highest quality video streams. For our mobile environment, AV1 follows on our work with VP9, which we released as part of our mobile encodes in 2016 and further optimized with shot-based encodes in 2018.

While our goal is to roll out AV1 on all of our platforms, we see a good fit for AV1’s compression efficiency in the mobile space where cellular networks can be unreliable, and our members have limited data plans. Selected titles are now available to stream in AV1 for customers who wish to reduce their cellular data usage by enabling the “Save Data” feature.

Our AV1 support on Android leverages the open-source dav1d decoder built by the VideoLAN, VLC, and FFmpeg communities and sponsored by the Alliance for Open Media. Here we have optimized dav1d so that it can play Netflix content, which is 10-bit color. In the spirit of making AV1 widely available, we are sponsoring an open-source effort to optimize 10-bit performance further and make these gains available to all.

As codec performance improves over time, we plan to expand our AV1 usage to more use cases and are now also working with device and chipset partners to extend this into hardware.
__________________
Win 10 x64 (19042.572) - Core i5-2400 - Radeon RX 470 (20.10.1)
HEVC decoding benchmarks
H.264 DXVA Benchmarks for all
NikosD is offline   Reply With Quote
Old 14th February 2020, 23:39   #52  |  Link
Nintendo Maniac 64
Registered User
 
Nintendo Maniac 64's Avatar
 
Join Date: Nov 2009
Location: Northeast Ohio
Posts: 447
Phononix has 16core verses 32core vs 48c vs 64c vs 64c+SMT scaling comparisons in dav1d v0.5.0 and SVT-AV1 v0.8 on Windows 10 Pro, Win10 Enterprise, and Clear Linux (all on a Threadripper 3990X):
https://www.phoronix.com/scan.php?pa...ws-linux&num=3


And regarding their choice of Linux distro:
Quote:
One of the interesting takeaways from my pre-launch briefing with AMD on the Ryzen Threadripper 3990X was AMD representatives actually recommending Clear Linux for use on this 64-core / 128-thread HEDT processor and the platform to which they've found the best performance. Yet, Clear Linux is an Intel open-source project.

The Clear Linux recommendation for the Threadripper 3990X was hardly a surprise to me given my experience with the platform, just a bit surprising AMD representatives acknowledging the Intel open-source software creation during a briefing. We've been benchmarking Clear Linux for years and were the ones to initially shine the public spotlight on its impressive performance capabilities -- that includes for AMD platforms too with numerous tests on different platforms we've performed the past few years.
__________________
____HTPC____  | __Desktop PC__
2.93GHz Xeon x3470 (4c/8t Nehalem) | 4.5GHz 1.24v dual-core Haswell G3258
Radeon HD5870  | Intel iGPU      
2x2GB+2x1GB DDR3-1333 | 4x4GB DDR3-1600       

Last edited by Nintendo Maniac 64; 14th February 2020 at 23:45.
Nintendo Maniac 64 is offline   Reply With Quote
Old 17th February 2020, 20:45   #53  |  Link
mzso
Registered User
 
Join Date: Oct 2009
Posts: 930
Is Dav1d likely to get significantly faster?
I tried it a bit in Firefox Nightly and there's a sizable gap to VP9 decoding in CPU usage (2x or more). I tried with 4k videos so the CPU usage would be more obvious.

Last edited by mzso; 17th February 2020 at 20:56.
mzso is offline   Reply With Quote
Old 17th February 2020, 21:59   #54  |  Link
Beelzebubu
Registered User
 
Join Date: Feb 2003
Location: New York, NY (USA)
Posts: 109
Quote:
Originally Posted by mzso View Post
Is Dav1d likely to get significantly faster?
Yes.

Quote:
I tried it a bit in Firefox Nightly and there's a sizable gap to VP9 decoding in CPU usage (2x or more). I tried with 4k videos so the CPU usage would be more obvious.
Could you elaborate on what sort of system (CPU chipset etc.), and where you got the content from? In particular, it'd be interesting to know the respective bitrates for the AV1 & VP9 files/streams, but knowing the encoder settings might also be somewhat useful.

Playback speed correlates a lot with bitrate. The 30% numbers that we've shown at conferences and in blogs are for same-quality encodes, where VP9 has a higher bitrate than AV1. If the files are same-bitrate, the performance difference goes up. On easy content, the postfilters also require a higher % of runtime (compared to e.g. inverse transform or predictors), and since AV1 has more postfilters, that means the difference will grow on low-complexity content, and will be smaller on high-complexity content. The 30% was also without film grain (since we assume the GPU will do that for free), but there is currently no browser that does that correctly yet.
Beelzebubu is offline   Reply With Quote
Old 17th February 2020, 22:23   #55  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by mzso View Post
Is Dav1d likely to get significantly faster?
I tried it a bit in Firefox Nightly and there's a sizable gap to VP9 decoding in CPU usage (2x or more). I tried with 4k videos so the CPU usage would be more obvious.
It'll get somewhat faster. But given equal levels of opimization, VP9 is going to decode a lot faster than AV1 beause AV1 is a lot more complex. This is how it always is. About every decade we get a new bitstream that offers an eventual ~50% reduction in bitrate for about a 100% increase in decoder complexity as long as you spend ~10x more on encoding to take avantage of all the new features.

Codec development is all about turning Moore's law improvements into better compression efficiency. There are all kinds of features that could have been used in AV1/HEVC/VVC that offer small improvements for bigger complexity requirements. And each generation it's a trade off for what a reasonable complexity cost is, and people design the most capable bitstream format within that reasonable decoder complexity envalope.

Sent from my SM-T837V using Tapatalk
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 1st March 2020, 17:45   #56  |  Link
mzso
Registered User
 
Join Date: Oct 2009
Posts: 930
Quote:
Originally Posted by Beelzebubu View Post
What browser/version, and on what platform/OS?



I think Youtube is known to do significantly higher bitrates for AV1 than for VP9, so that could be part of why...
I was testing Dav1d on Firefox Nightly. On Windows 8.1
mzso is offline   Reply With Quote
Old 6th March 2020, 01:03   #57  |  Link
hajj_3
Registered User
 
Join Date: Mar 2004
Posts: 1,120
dav1d 0.6.0 'Gyrfalcon'

0.6.0 brings major improvements in 10/12bit decoding on ARMv8 CPUs, up to 2.5 times faster than 0.5.2. It also brings new AVX-512, AVX2 and SSSE3 optimizations and improves the existing optimizations on all platforms. Finally, it also fixes some decoder mismatches and minor crashes.
hajj_3 is offline   Reply With Quote
Old 11th March 2020, 00:15   #58  |  Link
soresu
Registered User
 
Join Date: May 2005
Location: Swansea, Wales, UK
Posts: 196
Quote:
Originally Posted by benwaggoner View Post
Yeah, fast ARM 10-bit will make HDR feasible on 2020 mobile devices. For user generated content at least; premium studio content will still require HW DRM.

It's be nice to have some benchmarks with details beyond "Up to 2.5x faster" - is that only in some edge cases, or is ~2x speedup a practical expectation?
The bottom post on this gitlab issue has some benchmarks done by an AOM community member.

Link here.
soresu is offline   Reply With Quote
Old 11th March 2020, 02:49   #59  |  Link
Mr_Khyron
Member
 
Mr_Khyron's Avatar
 
Join Date: Nov 2002
Posts: 203
Dav1d 0.6 AV1 Video Decoder benchmark
https://phoronix.com/scan.php?page=n...AV1-Benchmarks
Mr_Khyron is offline   Reply With Quote
Old 3rd April 2020, 17:51   #60  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
I'm wanting to test av1 and some other new encoders, and have been using media-autobuild-suite to try and build a ffmpeg with it. It keeps failing trying to install cargo-c as part of rav1e. I've deleted the folder. I've removed rav1e from the build list, but it keeps always trying to compile it and then failing. It's been happening for a couple of weeks now, and has persisted despite several rav1e updates in that time period.

Any suggestions?

Code:
dav1d git  .................................................. [Up-to-date]
  Running git clone for rav1e...
┌ rav1e git  .......................................... [Recently updated]
├ Running submodule...
├ Running install-cargo-c...
Likely error (tail of the failed operation logfile):


error: aborting due to previous error

error: failed to compile `cargo-c v0.6.2`, intermediate artifacts can be found at `C:\Users\benwagg\AppData\Local\Temp\cargo-installyZ6Pas`

Caused by:
  could not compile `cargo-c`.

To learn more, run the command again with --verbose.
install-cargo-c failed. Check C:/Users/benwagg/Desktop/media-autobuild_suite-master/build/rav1e-git/ab-suite.install-cargo-c.log
This is required for other packages, so this script will exit.
  Creating diagnostics file...

All relevant logs have been anonymously uploaded to https://0x0.st/iuKh.zip
Copy and paste [logs.zip](https://0x0.st/iuKh.zip) in the GitHub issue.
Make sure the suite is up-to-date before reporting an issue. It might've been fixed already.
Try running the build again at a later time.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 13:54.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.