Open Bug 1795031 Opened 3 years ago Updated 3 years ago

Strange problem with Amazon prime video. Often the first 1 min or so, the video quality is dismal Does not happen with Chrome.

Categories

(Core :: Audio/Video, enhancement)

Product:
Core
Component:
Audio/Video
Version:
Firefox 105
Platform:
x86_64
Windows 10
Type:
enhancement

Tracking

()

Status:
REOPENED

People

(Reporter: ishikawa, Unassigned, NeedInfo)

Details

Attachments

(19 files)

raw output of about:support
55.90 KB, text/plain
Details
Blurred image from Firefox of an amazon video.
2.16 MB, image/png
Details
Same video from Chrome. It is clear.
2.42 MB, image/png
Details
After fast fowarding to 10 minutes, the video is clear initially in firefox.
2.36 MB, image/png
Details
But after 10:20, the video degrades and it is now pretty much blurred at 10:36 in firefox.
2.16 MB, image/png
Details
Chrome has no problem showing the video throughout. This is after I fast foward to 10:00 and the same scence in the preivous attachment.
2.30 MB, image/png
Details
about:suipport shown in Japanese characters.
37.93 KB, text/plain
Details
about:support - raw Jason data for analysis
54.76 KB, text/plain
Details
Network speed check at "fast.com" maybe about 20-30 seconds I began observing the blur.
69.03 KB, image/png
Details
about:support, after the image starts to be blurred and still so.
37.93 KB, text/plain
Details
about:support after I observed video quality degradation, but the quality recovered.
23.73 KB, text/plain
Details
about:support when the low video rendering is observed.
38.02 KB, text/plain
Details
About support output in raw JSON form.
54.85 KB, text/plain
Details
some features of graphics rendering that are explicitly blacklisted.
86.13 KB, image/png
Details
raw JSON data from about:support when the video degradation occurred.
34.86 KB, text/plain
Details
GPU usage when FF accesses youtube video. It uses HW decoding.
84.07 KB, image/png
Details
Both Chrome and FF don't use HW decoding for Amazon Prime video on my PC.
109.11 KB, image/png
Details
Shari007 Codecs. LAV decoder setting. marked parts changed from default.
301.70 KB, image/png
Details
HW decode availabilty: Title vs. Browser
108.17 KB, image/png
Details

Windows 10 Pro 22H2 (I think I have seen this earlier, too).
Firefox 105.0.3 64-bit. But I have noticed the issue for several months now.

I watch videos from Amazon Prime.

But for sometime, I have noticed the following strange behavior.

I try to see a video from Amazon for the fist time.

1 - Then I suddenly realize the video quality is dismal. I can't see it the details at all. (Like the one attached. The overlaid character title, etc. look crisp. It is the video rendering that looks blurred.).

(Back when this started, I thought of a network connection problem. No, I verified the network is good via other means.)

2 - Surprised, I switch to Chrome. Somehow Chrome usually can show the video clearly.
I figured this out several months ago.
(A photo is attached to show chrome rendering)
The network seems to work just fine. Only the video rendering of the particular amazon prime video by firefox. So I tried Chrome. Voila Chrome could show the video clearly.
This is very strange, but it has been like this for some time.
(Lately, the network connection has been good. Maybe too good.)

What puzzled me most is actually though is the next discovery.

3 - Then after seeing Chrome can show the video clearly, I want to re-check the video replay in Firefox. To my utter surprise, Somehow the video playing becomes crystal clear in Firefox.

This behavior has perplexed me for the last few months.
(I thought caching may play a role, but see comment

Step 3 is something I only noticed for the last couple of months.

Before that, I switched to Chrome to watch lengthy video streaming of sport games such as soccer/football because of the other similar video problems. (Even crashes sometimes. I think those are gone after I increased the amount of RAM and the size of paging file accordingly.)

I am attaching the strangely blurred image of firefox video rendering.

I know there are too many variables.

a - Does Amazon Prime Japan try sending different video stream to firefox vs Chrome?
Maybe there is a strange video protocol negotiation which firefox may not handle well?
Horrible, if so. I have no idea.
But there have been sites that claim firefox is not supported for video viewing.: such as Japan's NHK, a half public broadcasting corporation. It states Firefox is not supported and blocks video viewing in its NHK+ streaming service.)

b - Why, after seeing dismal video in firefox, switching to Chrome and seeing clear video there, and then when I try seeing video with Firefox again, now I get clear video?
Does this suggest there is something with the initial caching with Firefox?
Does the cache size matter?
But given that the clear video display happens even for later segment which would not have been in the cache, I don't think so. I have verified this several times.
Come to think of it, if Amazon considers connection from a single IP address as all the same as far as video replay is concerned (and disregard the protocol negotiation done per browser, this may explain that after Chrome browser sets the protocol negotiation in a sane state, I can see clear video in Firefox. But this is very unlikely. Such negotiation should be per programs. But I have no idea what Amazon prime does.
The browser header should differentiate two accesses from Firefox and Chrome.
Amazon Prime DOES check the number of connections from a single IP and refuse to
stream video after a certain # of simultaneous connections.
I am talking about the video rendering on a same PC so IP address recognized by Amazon is the same irrespective of IPv4 (NAT) or IPv6.

c - Are there issues with available codecs on the system?
Again, I have no idea. I think we are past the age when one needed to load codecs from third party sites for seeing videos well, but I may be wrong.

d- I said, "Lately, the network connection has been good. Maybe too good."
Obviously local telco has upgraded the regional optical fibre infrastructure very well in
recent few months. I get 450-500 Mbps connection even during busy afternoon time (when people working remotely heavily used the network. I noticed the network slowdown last year, no more this year.)
Can there be any flaky logic in network video negotiation that somehow gets confused by a very fast network response (i.e., can there be a overflow of a signed entity that can result in such a strange behavior?).
You can argue that I was unlucky and when I saw blurred video, there was network issue.
But so far, the blurred video happened with Firefox only and in a 10 seconds - 20 seconds or so Chrome showed clear video. And I had a few occasions where I literally run firefox and Chrome side by side and it was only Firefox that showed blurred videos. So network is not an issue I think.

However, debugging this issue is difficult.

Since there are so many variables, even before I begin debugging the issue somewhat seriously, I would like to now if others see similar problems at all.

If not, then this problem is one of those unique problems caused by resident anti-virus program, video driver (I use Nvidia card FYI), some unintended codecs installed by 3rd party software which I am not awere, etc. which I can think of right now, very unique issue to my PC. So it certainly is a big problem for me, but maybe not that important to firefox user base.

For me the problem is so real. I just saw another symptom a few dozen minutes ago. I have got really tired of this problem and thus am writing this entry. The problem seems to persist for quite some time.

I am posting this bugzilla to see if there are users who experience similar issues.

The symptom really puzzled me.

If possible, I would like to stick to using a single browser with careful privacy policy, i.e. Firefox, , but due to the video issues for the past months, I have begun running Chrome browser quite often for viewing videos.
Chrome's, erh rather Google's position on handling ads or privacy information leave room for improvement, but the way things are going I may have to live with Chrome more often and actually, I have begun doing for the last couple of months an have become numb to it.

TIA

PS: My PC is on the resource rich side:
CPU AMD Ryzen 7 3700X 8-Core Processor 3.59 GHz
RAM 32.0 GB
Windows 10 Pro
version 21H2
(re-) installed on 2022/‎05/‎09
OS build 19044.2130
Windows Feature Experience Pack 120.2212.4180.0

I am attaching about:support in the first attachment, and then attach the video photos to show what I am talking about.

This is the video rendering from firefox.
You can see that the overlaid character title is shown clearly whereas the video is pretty much blurred.

This is the screen of the same video from Chrome.

You can see the facial feature, etc. very clearly.

I had to take the photo since the image protection does not allow me to capture the screen via software.

Another surprise.
In this particular bad video rendering, firefox video is still bad: I started re-playing the video at aronund 20 seconds into the feature but it became slightly blurred around 50 seconds, and became worse around 60 seconds and the bad rendering still continues.

Then is it possible that the particular codec that Firefox is more susceptible to random noise or something than the codec used by Chrome?

I fast forwarded to 20 MINUTES in the video since that part is unlikely to be cached by Firefox even it may prefech some dozen seconds worth. Then I see clear video. But it happened again after 30 seconds of viewing.
That is, initially firefox shows clear video after showing a fresh video segment, but after about 30 seconds the video quality degrades (!). [

I am going to attach the sequence I tried at about 10 minute into the video.
This does not happen all the time mind you. But with today's replay, I seem to be able to reproduce the behavior.

Something IS wrong with the way MY firefox interacts with Amazon Prime Japan's video feed.
I bet this is way beyond the frontline support of Amazon. They will tell me simply to use Chrome.

After fast forwarding to 10 minute scene, the video was clear in Firefox. Good.
But in the next video around 10 min 30 seconds, now you see the degradation I have explained.

Because the degradation happens after a few dozen seconds, I suspect some kind of improper data negotiaion (or codec) problem.
At this stage, I can't think of where to look into the issue.

TIA

But after 10:20, the video degrades and it is now pretty much blurred at 10:36 in firefox.

I have no idea. Maybe the first fetch of video has problem near the tail end of the data block?

This is from Chrome.
I fast fowarded to 10:00 and continue seeing the video.
This is where the video quality is already unbearable in Firefox, but Chrome show clear video.

Since this is repeatable today, I sort of suspect there is something wrong in the initial portion of video fetching done by firefox. Maybe the first video chunk is corrupted near the end or the
caching interaction is not quite right?

I am afraid this anime fan is not going to use firefox for video viewing soon :-(

Of course, there is a possibility of graphics driver problem.
I use a particular board based on Nvidia GT 1030 since it is fanless and silent. I prefer silent computer for for study/work and movie.
Since Nvidia is popular, I would think similar problem is observed by others if the problem I see is related to graphics driver.
The driver is updated regularly using Nvidia's tool.
The problem I noted has happened for the last few months at least and so is not tied to particular nvidia driver.

If anyone has suggestions for what type of information to collect (and how on windows 10 pro) for potential debugging effort, I would appreciate them.

TIA

Today it was bad in the sense I could NOT get back clearn video rendering in firefox after Chrome was tested and I came back to firefox.
I had a hunch and restarted firefox (shutdown and start again). And voila.
I seem to be able to see clear video rendering this time (!)

So when something goes bad near the beginning, that bad effect may still persists from what I observed today.

But if the first one minute is shown OK, then it does not happen later (at least today, it did not)?
Sounds to me an initialized variable or something is causing the issue.

I have noticed this so far with Amazon (and possibly with DAZN, a sport video stream provider).
The video degradation happens again in firefox and so either not robust enough codec decoder in the presence of noise, or some data corruption may occur depending on the size of data transfer(?).

If the problem is reproducible easily, things would be much easier.
But the problem happens somewhat randomly and I am not sure if the replay of the same video tomorrow evening shows the behavior. :-(

Attachment #9298415 - Attachment description: After fast fowarding to 10 minutes, the video is first clear in firefox. → After fast fowarding to 10 minutes, the video is clear initially in firefox.

Hi, Ishikawa,

Would you mind help me use Firefox profiler to capture the detail while the issue is happening?

The steps will be like this

  1. Go to Firefox profiler and enable this feature (then you should be able to see a small button appearing on your browser toolbar.)
  2. Go to about:networking, select the tab Logging on the left
  3. In the box Current Log Modules, enter one of the following strings,

timestamp,nsMediaElement:5,nsMediaElementEvents:5,MediaDecoder:5,MediaSource:5,MediaSourceSamples:5

  1. Start profiling with the Media preset
  2. Go to the website where the issue happens, and start reproducing the issue
  3. Stop profiling
  4. Upload the profile, keeping all hidden threads

There is an example video showing how to do that step by step (but it uses the different log modules than the step3), thanks.

Flags: needinfo?(ishikawa)

Sorry, I got tied up with other things this week and could not pay attention to this, but I will follow your suggestions.

I don't want to be a conspiracy theorist, but when I realized that only Firefox allows the capture of streamed video screen, I had to think hard what Amazon might be doing.
Chrome, Edge, etc. blanks the window where the video is streamed, when a simple capture action such as ALT+PrtScreen,
or MS Windows Snipping Tool is invoked.
I now understand why NHK, Japan's public broadcasting corporation similar to BBC in UK, or PBS in USA, basically excludes Firefox from accessing its video streaming service since spring this year. Content holder may have a big say on this decision behind doors.

At the same time, I still have a pet detective theory that Firefox may not be recovering from minor noise or
may not somehow interact nicely when Amazon thinks the network is in bad condition and reduces the quality of video. There ought to be a some probing of network condition and firefox may not tell Amazon that the network condition has become good.
Since I posted the original buzgilla comment 0, I have noticed Google Chrome sometimes DO have such moments, but much, much rarer. I think Chrome-Amazon combination recovers from low quality video condition much quicker and thus I don't get annoyed very muc at all.

Anyway, I will post the results and hope they can give us insight about what really goes on (rather than my wild guess).

TIA

Flags: needinfo?(ishikawa)

OK, this is the uploaded profile URL: I started to watch a segment and then about 5 seconds into the video, I got blurred image.
It continued. But toward the end before I stopped capturing, the video image quality improved somehow (maybe for 10 seconds or so)..

https://share.firefox.dev/3sb2XkO

I captured profile during a clear video streaming for comparison. I failed to record the shorthand URL.
I believe the following is the URL for the profile.
https://profiler.firefox.com/from-browser/calltree/?globalTrackOrder=0wg&hiddenGlobalTracks=1wcf&hiddenLocalTracksByPid=16196-0w24wacwx1~25016-0~34424-01~26608-0w2~38836-0w2~23844-0w2~6636-0w2~22260-0w2~24072-0w2~18648-0w4~16880-0w2~34148-0w2~38792-0w2~28288-0w6~6548-0w36d&thread=yg&v=7

Flags: needinfo?(alwu)

I don't know why, but when I ran the video streaming of a series of anime videos in FF,
I found that at about 50sec 1min into each episode, the screen quality degrades conspicuously.

I wonder if FF fails to negotiate some resolution settings requested by Amazon host.

The degradation has occurred rather constantly for each episode and I am very surprised.

For those who have Amazon Japan's Prime membership, this is the video URL.
https://www.amazon.co.jp/gp/video/detail/0SQVZTZEFQB7DBOF29U5P7B2PB/ref=pe_10728342_466825422_SH_6_2_JP_ja_JP

I simply ran the episodes from the first one and as soon as the episode is switched, I observed very bad quaility video at aroudn 50sec - 1min.

BTW, the degradation happened in safe mode.
The captures were recorded in safe mode.
As a matter of fact, some of my add-ons seem to interfere with profileer function and thus I was forced to record the profile in safe mode.

And I see the degradation without using a local proxy, too.

So it IS something in FF and Amazon video streaming server interaction that causes this strange video degradation.

I have not seen such strange quality degradation so often with youtube even if I choose so called 4K video on youtube.

I think Amazon Prime video may occasionally require the client to perform negotiation of the least expected video quality which Firefox may not perform as expected (which Chrome performs well).
The above is a pure guess. But I can only think of either a non-robust error recovery from the slightest of transmission error or such protocol level issue to explain the video degradation between Amazon prime video and Firefox.
As I mentioned I have hardly seen such degradation with Chrome when I see Amazon prime video.
And Youtube performs well with Firefox. (I don't know if the video format used by youtube and Amazon prime is different or not. If it is the same, then it can mean either youtube/google streaming infrastructure is less susceptible to random noise, or youtube does not expect its client to negotiate video quality.)

I know there are so many factors that are not under the control of user and thus debugging is difficult.

If someone else observes such degradation with Amazon Prime Japan, it would at least add a bit more urgency to this problem.

Sorry for my late reply, from the profiled result you captured in the comment 13, I can see the video quality dropped to low resolution for most of the time and recovered to the original resolution in the end of the capturing.

the degradation happened in safe mode.
Does the issue only happen on the safe mode? or it happens on the normal mode as well? If you start Firefox with a new profile, would issue still happen?

How is your average internet speed while you're watching videos on Amazon Prime? Does that happen on every videos, or only on some specific videos? Would you mind to provide your about:support as well?

Thank you.

Flags: needinfo?(alwu) → needinfo?(ishikawa)

(In reply to Alastor Wu [:alwu] from comment #17)

Sorry for my late reply, from the profiled result you captured in the comment 13, I can see the video quality dropped to low resolution for most of the time and recovered to the original resolution in the end of the capturing.

Thank you for your comment.
I am glad that profile I uploaded enabled you to analyze the low video quality issue.
I am now wondering if it is Amazoneserver that proactively sent me low-quality video?

the degradation happened in safe mode.
Does the issue only happen on the safe mode? or it happens on the normal mode as well?
It happens in normal mode as well. I tested in safe-mode to eliminate the possibility of plugins doing something nasty.
But since safe-mode saw the degradation, the plug-ins don't matter.
Except that my plugins interfered with profiling to my surprise. I have not investigated which ones interfered.
I could not obtain profiling in the normal mode. Only in safe mode, the profiling worked.

If you start Firefox with a new profile, would issue still happen?
I will try, but given that on another windows 10 PC with a different CPU and graphics board, I saw degradation at least for the last few months, I suspect it won't matter, but please wait for my report.

How is your average internet speed while you're watching videos on Amazon Prime?

You may be surprised that my video connection is rather good these days.
I checked with "fast.com". I get 500Mbps -780Mbps from "fast.com" speed test.
That makes me think the following very strange coding issue. : there may be a code that assumes some timing assumptions that no longer hold true in a really fast setting.
I think there is something wrong either with the server or the firefox cache/javascript implementation.
I get numbers which were more than 1Gbps a few times in the last couple of weeks. They can't be true, or can they?

Anyway, when I realized that the video streaming can be quite fast, I realized that there may be some timing assumptions that are not true any more and the Firefox code may get confused. (For example, are the some types of quality negotiations from time to time?)
Yes, the degradation happens when the speed is good, I think.
To top it off, Chrome running in parallel does not experience such degradation. That is very mysterious.

Does that happen on every videos, or only on some specific videos?
It is very difficult to explain the patterns.
It does not happen every day on every video.

But once it happens, this is what I observe often.
I run videos consecutively. Autoplay when an episode of a series ends.
Now the initial 30-40 seconds look OK.
But somehow around 40-1 minute range, the video quality starts to degrade and continue for a while in low quality.
When the low quality continues for such a long time and I can't stand it, I run Chrome for comparison. Chrome does not see any such low video
quality play in comparison.

Would you mind to provide your about:support as well?
I am afraid that it is described with Japanese heading titles, but I will upload it in another message.

Thank you.
You are welcome and thank you very much for your attention on this matter.
I searched and found a few blogs that mention such degradation over the last few years, but I am not entirely sure if we have the same cause.

This is the textual display of about:support in Japanese characters.
I am using Japanese locale under windows 10.

I am uploading the raw Jason data in the next message so that it may be easier to analyze.

Here is the raw Jason data from about: support.

I did not realize that I have changed so many configuration parameters over the years.
I looked at them but I don't see much problem in them.
Maybe I should move the cache directory to SSD? (M:\temp is on a rotating disk).

Flags: needinfo?(ishikawa)

I am testing the video play in a new profile.
I will report if the problem appears in there, too.

It happens in the new profile.
In the first episode play, I did not see the blur at the beginning. And the 2nd episode, too.
So I fast forwarded to a new position.
About 50 seconds later, I get the blurred video (!).
It has something to do with the downloading of new video segment(s) and after about 40 - 50 seconds of the play, this blur is likely to occur.
This is the often observed pattern. New video play (or play from a new position) and then I see a blur at 1 min mark. It takes me 5-10 seconds during normal viewing to realize that the image is degrading to the point I notice it easily.

40-50 seconds number may vary depending on the network speed (and the amount of cache?)

I get the blur both in
a) - FF executing with newly created profile. All I did was to log into Japan Amazon to see a video.
b) - FF executing in the old profile (with plugins and all that).

I wised up and checked the speed.
It is very fast.

I am investigating a bit.
I do not see tell tail sign in error console.
I wonder if the memory pressure issue may be at play although again, Chrome browser runs OK. That is a mystery to me.

That is, after I saw the blur both in a), b) above,
I stopped FF with new profile. (a) and instead ran Chrome to see the same episode.(c)

I fast forwarded the video in b) and sure enough after about 30-40 seconds it got blurred (but this time it became clear after another minute or so. It may be that we have a cache problem?)

The chrome video replay (c) did not get degraded at all during all this time.

Something is different between the video play of Chrome and FF against this Japan Amazon prime video server.
(I don't think I see this somewhat repeatable blur with youtube video at all.
I say repeatable, but this may not happen on a certain day. But today it seems to be very often repeatable.)

I thought the problem occurs less often with newly created profile, but it does happen.
And as I have been working on a draft in one screen, I just noticed the low-quality video in FF with new profile on another screen.
I checked and I found that it happened at 1 minute into the new episode.
There is something going on when a new video starts and gets played. Something happens around 40-1 minute mark.
That is where I often see the low quality video.
(But not always. Difficult to debug.

I tried to show the blurring by recording what is shown on the display via my smartphone camera.
It turns out youtube is so clever that the content is protected by copyright and blocked the sharing of the video.
Oh well.

OK, with a different title, and focusing only on a part of the screen, youtube allowed me to share the following.

The beginning of a video in chrome. No blur.
https://www.youtube.com/watch?v=ZMFSEE5_1l4

Same beginning of a video in FF. About 30-40 seconds, the video starts to get blurred.
It is difficult to figure out since it is a video taken by smartphone's camera.
But you can see the characters on the screen is barely readable.
https://www.youtube.com/watch?v=zc31f2DDkrs

Is there any chance that FF gets confused that I use two screens with different resolution (under windows). One enlarges the characters on the screen by 200%. The other uses the default setup. Just a remote possibility but maybe worth mentioning here.

This I say because I just noticed when I moved FF window that visits flightradar24 flight route display (on HP Z24n G2 display, 1920 x 1200, font is at 100% size) to to a higher resolution DELL display (Dell S2817Q, 3840x 2160, font size at 200%), the screen image quality obviously gets quite low for the flightradar24 display. It may be that FF picks up the display info of WRONG display to adjust its display when it was moved from a display to the other and resized, etc. I had to close the window and revisit the site to restore the expected image quality.

Something similar may be happening with the video rendering even though such replacement of window does not happen???

Jeff, Sotaro, would y'all mind to help me answer the question from comment 27?

I noticed that the reporter's graphic card is NVIDIA GeForce GT 1030, which should be able to do h264 hw. But in his profiled result, it was using h264 sw decoding. I've searched our gfx blocklist, where I couldn't find a rule explicitly disabling that graphic card/driver version. Did we incorrectly block that graphic card?

In addition, I already reported this bug to Amazon as well, to see if they have any thought why the video would get downgraded.

Flags: needinfo?(sotaro.ikeda.g)
Flags: needinfo?(jmuizelaar)

(In reply to ISHIKAWA, Chiaki from comment #13)

OK, this is the uploaded profile URL: I started to watch a segment and then about 5 seconds into the video, I got blurred image.
It continued. But toward the end before I stopped capturing, the video image quality improved somehow (maybe for 10 seconds or so)..

https://share.firefox.dev/3sb2XkO

about:support says that WebRender(hardware accelerated) is used. But from the above profile, in Renderer thread, RenderCompositorSWGL function calls exist. It uses software WebRender. From it, WebRender seemed to fall back to software WebRender.

Flags: needinfo?(sotaro.ikeda.g)

(In reply to Sotaro Ikeda [:sotaro] from comment #29)

(In reply to ISHIKAWA, Chiaki from comment #13)

OK, this is the uploaded profile URL: I started to watch a segment and then about 5 seconds into the video, I got blurred image.
It continued. But toward the end before I stopped capturing, the video image quality improved somehow (maybe for 10 seconds or so)..

https://share.firefox.dev/3sb2XkO

about:support says that WebRender(hardware accelerated) is used. But from the above profile, in Renderer thread, RenderCompositorSWGL function calls exist. It uses software WebRender. From it, WebRender seemed to fall back to software WebRender.

This is a surprise.

I have a suspicion that this does not happen always. (This is because I don't see the degradation on a particular day, OR remember that I often see the degradation into about 40 seconds - 1 minutes of a new video I watch, but often times later for quite a long interval, once the video quality returns to normal, I often see this good quality to continue for minutes.

So maybe this is a driver issue, but this board ought to be used by many...
It is one of the few fan-less graphics boards in the market that can do 4K graphics.

Is there any way I can force the board + FF to use the hardware rendering?
Oh wait, there is a hardware acceleration setting in FF, correct? I wonder what my set up is.
I just checked. I am using the recommended setting for performance setting, which means, the decision to use hardware acceleration is
left to FireFox.

I am not sure if I should uncheck the recommended performance setting, and choose to use hardware acceleration IF there is a reason this board is not enabled for hardware video rendering by default for a reason.

I unchecked the recommended performance setting and set use hardware acceleration (if possible).
Still I get the blurred image about 40 second -1min into the new video.
So FF thinks this board can't do acceleration?

I wonder what are the conditions for FF to decide to fall back to software rendering.

Hi ISHIKAWA-san, Software WebRender fallback happens in the following situations. Can you attach about:support after the problem happens, gfx section might have error log. From about:support, by default hardware WebRender was used on the PC.

-[1] Device reset in GPU process exceeds a limit

-[2] GPU process crash exceeds a limit

Flags: needinfo?(jmuizelaar) → needinfo?(ishikawa)

I am attaching about:support TEXT extract.

It was captured after the image started to blur. It still is.

I suspect this part is crucial:

決定ログ
HW_COMPOSITING	
available by default
D3D11_COMPOSITING	
available by default
DIRECT2D	
available by default
D3D11_HW_ANGLE	
available by default
GPU_PROCESS	
available by default
WEBRENDER	
available by default
WEBRENDER_QUALIFIED	
available by default
WEBRENDER_COMPOSITOR	
available by default
WEBRENDER_PARTIAL	
available by default
WEBRENDER_SHADER_CACHE	
available by default
WEBRENDER_OPTIMIZED_SHADERS	
available by default
WEBRENDER_ANGLE	
available by default
WEBRENDER_DCOMP_PRESENT	
available by default
WEBRENDER_SOFTWARE	
available by default
WEBGPU	
disabled by default: Disabled by default
blocked by runtime: WebGPU cannot be enabled in release or beta
WINDOW_OCCLUSION	
available by default
VIDEO_OVERLAY	
blocked by default: Blocklisted by gfxInfo
HW_DECODED_VIDEO_ZERO_COPY	
blocked by default: Blocklisted by gfxInfo
VP8_HW_DECODE	
available by default
VP9_HW_DECODE	
available by default
REUSE_DECODER_DEVICE	
blocked by default: Blocklisted by gfxInfo
BACKDROP_FILTER	
available by default

I am curious about this part.

REUSE_DECODER_DEVICE	
blocked by default: Blocklisted by gfxInfo

Does this mean decoding is not done IF I have more than one video being decoded at the same time?

Come to think of it, I have tab visiting youtube, but the video is not replayed at this moment although it was watched for a while and stopped for now.

Flags: needinfo?(ishikawa)

Hmm, the new about:support does not have error log. From it, Firefox seemed to use WebRender(hardware). The situation seemed different from https://share.firefox.dev/3sb2XkO. It seemed not have GPU process, but the about:support still had GPU process.

Can you take a profile after the problem again? Thank you.

Flags: needinfo?(ishikawa)

(In reply to Sotaro Ikeda [:sotaro] from comment #34)

Hmm, the new about:support does not have error log. From it, Firefox seemed to use WebRender(hardware). The situation seemed different from https://share.firefox.dev/3sb2XkO. It seemed not have GPU process, but the about:support still had GPU process.

Can you take a profile after the problem again? Thank you.

Addition: I am sorry that I changed the setting: Uncheck the suggested optimization setting, and I explicitly specified hardware acceleration (if possible) when I captured the last upload of about:support.

Now I have reverted it. I tried to recreate the issue this morning, but failed.
Although I suspect there is an issue on FB side, I wonder if the problem may be caused also with strange interaction with the server side.
That is, of course, we cannot control on our end.

Stay tuned. I will try this afternoon and evening.

https://share.firefox.dev/3G1kB2O
The performance profile was obtained whenI observed the following.
I started capturing.
about 35 seconds into the video streaming, the video quality degraded. It continued for 5-10 seconds.
But surprisingly, this time, the quality reverted back to normal.
It usually lingers for a minite or longer...

I am attaching the output of about:support, but I am afraid that I captured it after the video quality recovered and so it may not contain what may be relevant for debugging.

This evening I will try capturing about:support while the prolong video degradation is in progress.

Thank you for taking the profile again! Both about:support and the profile showed WebRender(hardware) existed in GPU process. And in the profile, RenderDXGITextureHost and RenderDXGIYCbCrTextureHost did not appear on Renderer thread. Then RenderBufferTextureHost/ShmemTextureData/SharedPlanarYCbCrImage seemed to be used for video frame.

From the profile, ChromiumCDM seemed to be used by decoding side. But I do not know about it.
:alwu, can you comment about decoding side?

Status: NEW → RESOLVED
Closed: 3 years ago
Flags: needinfo?(alwu)
Resolution: --- → WONTFIX
Status: RESOLVED → REOPENED
Resolution: WONTFIX → ---

After midnight I tried the profile capture again, and this time the video degradation happens on the first try and continued like I observed for a while.
(So this may suggest wall-clock-dependent behavior or server.)

For profiling data, I made a mistake of choosing the upload with video frame first.
https://share.firefox.dev/3G06w5u

Then I realize we can re-upload it with different choice of uploaded content.
I think this upload is with hidden thread or whatever.
https://share.firefox.dev/3DTucWv

What I observed was, after starting capturing, I visited a new video on Amazon prime video web page and watched it.
About 35 seconds into the streaming the video got blurred and continued the low-quality rendering.
That is when I showed "about:support" window. (This is what I should then. Correct? I hope it is not refreshed automatically. Otherwise, I may indeed inappropriate data.)

I then stopped profiling and captured the profiling data, and uploaded it twice as noted above.

I then visited the already opened about:support window and attaching the support data.

Again, I am attaching the content about:support as text, but I don't think I see the "log" file in there?
(Do you mean I need to upload the network logging data, too?)

I am not sure if there is anything useful in this about:config, but the uploaded profile probably re-confirms the past behavior.|

I wish I could use the hardware decoder (or whatever) if that is feasible.
Since Chrome does not have the low video quality issue, that is a mystery. They may have a relaxed condition to decide to use hardware acceleration.

I did notice very bad video degradation while the profile function "capture" the data for visualization.
I suspect that capturing and visualization uses up much CPU time, and thus software rendering could not perform the good rendering and could only show the half-cooked rendering (very low quality).

In the next message, I am attaching the JSON data from about:support for program analysis.

TIA

This is the "about:support" raw JSON output mentioned in the previous comment.

TIA

Flags: needinfo?(ishikawa)

Oh, I didn't realize I enabled the plugins this time around. Profiling still works.
(I think ENABLING profiling for the first time did not work with my plug-ins. But once enabled, plug-in does work with my plug-ins.)

Just like the other day, I could reproduce the video degradation in newly created profile without plug-in. Great. But profiling could not be uploaded.

The difficulty of debugging this problem is that the problem does not happen each time I visit a new video.
So I had to let the video run for a while on one display when I was doing something else on the other display.
I occasionally glance at the video to see if the video degraded.
When I realized that the video degraded while working in other display, I visited about:support.

Interestingly the video degradation happened after about 1 min into the video. (The title back song started in this Gundam The Witch of Mercury when I visited about:support window.)
This problem seems to happen very often after a new video is played and 35-40 seconds pass before the degradation starts.

But due to the way profiling was taken this time, I started profiling way before the degradation occurred, the profiling spanned 260 seconds including the video degradation in the latter part of the duration.
It seems the data is too big to upload. (I got upload error. "Uh oh, some error happened in profiler.firefox.com."
The detail error says:

div
div
div

And the profiling data seems to be gone. Tough.

I captured the about:support when the problem occurred.
But again, I don't think it is very different from others.

I am just attaching the part which bothered me.
It looks that some features of hardware decoding are explicitly disabled. Does the blocking cause my FF to render the video by software(?).

Is there any for me to narrow down the range of profiling data once the profiling data is captured so that uploading of profiling data can succeed?

Great.
I could reproduce the issue again at the start of next episode.
Upload with hidden thread: https://share.firefox.dev/3UzGMRA
Upload with screen images(?): https://share.firefox.dev/3A0KCLu

This is with the newly created profile without any of my plug-ins.

I am attaching the raw data of about:support.

If someone can think of any other data that may give insight into the problem, please let me know.

This degradation that happens near the beginning is really a drag.
I am not using FF for serious Amazon Japan prime video viewing any more :-(
Google chrome can show the videos without such issues.

Funny thing is that I have not noticed any such issues with youtube so far with FF.
I have been watching youtube's 4K HD (actually 2.4K+ pixels vertically, I think) but no
strange video degradation there.

Thank you again for your time to look into the issues.

Flags: needinfo?(alwu)

I noticed in my about:support output the following:

I got curious and found Bug 1254942
Is it working? I suppose it does considering many users who have graphics accelerator card.

So can it be this NVIDIA TI 1030 specific issue???

Regarding FF not having any issue on youtube is that FF DOES use hardware decoding when I look at 4K video on youtube.

For example, the attached table from task manager shows that FF uses HW decoding when I looked at the following youtube URL. (If you try, make sure you choose the highest resolution, 2160p.)

Google chrome also uses HW decoding when I look at the same youtube.

So obviously, FF can rely on HW much in this case and I have not seen any video degradation.

Regarding FF not using HW decoding, I found that to my surprise Google Chrome does not use HW decoding on my PC, too.

Attached info is for Google Chrome.
It does not use HW decoding when I looked at Amazon Prime video, specifically
The witch of Mercury (水星の魔女）

H.264 is not supported on NVIDIA TI 1030?
Strange.
I will check if there is a special toggle for the Nvidia driver on my PC.

At least the mystery is solved that I don't see video degradation when I see youtube.
Youtube video (mp3?) gets the HW decoding support while Amazon Prime video doesn't.

Then why Google Chrome does not show video degradation?

  • Maybe software decoder used by Google Chrome is more efficient than FF's.
  • There may be unknown protocol issues (for video resolution hand-shake?).
  • FF decoder may be more error prone to transmission noise and has tough time to recover after a single error?

Pure guesses.

Is there any chance that FF only checks video encoding feature is available on the graphics board and if it is not,
deems that decoding also is not available?
NVIDIA TI 1030 is asymmetric in the sense that it does NOT have encoding, but does have decoding.

But come to think of it, FF does use HW decoding for youtube mpeg file.
Hmm...

A REAL SURPISE

Short summary: Amazon Japan's Prime Video: trailer is decoded using HW, the main title is not.

Long summary: I tried to find the available codecs on my PC.
Google search results suggested that I use Windows Media Player.: Help -> About Technical Support -> look in the shown web page.
Unfortunately, windows media player 12 crashed during this operation (it created dump file.).
So I could not figure out what codecs are available.
I was desperate and so installed Shark007 Codecs. This allows me to tinker with codec settings. (I did back in XP days. I never thought I would be doing this again in Win10 days.)

Please recall I was checking the video degradation of the title I selected from Amazon Prime video.
Amazon inserts trailer of other videos before a title appears from time to time, but since I was interested in comparing the video quality around the 1st 60 seconds of the chosen title, I skipped the trailer all the time during my investigation so far.

However, after installing Shark007 codec, I was trying to see how this codec setting works and
monitored the GPU usage including this trailer section.

A REAL SURPRISE:
Then I found out that BOTH FF and Google Chrome used HW decoding with this setting of Shark007 during the trailer (!!!)

Since trailer is short, I only see the use of HW decoding briefly.
And, once the main title is on, HW decoding is gone.
And terrible video quality sometimes. Obviously software decoding kicks in.

So I think it is Amazon doing something funny for the main title.
Trailer is downloaded in a manner HW-decoding works.
But once the main title starts, HW-decoding is not used (both in FF and Google Chrome).

I don't know much about how video streaming is done, but this suggests some sort of server-side issue or maybe a protocol issue (handshake to agree on video quality or video types?)

Again, in any case, I think FF's software decoding sucks performance-wise in comparison to Google Chrome obviously. I say this because I rarely notice the video degradation using Google Chrome even if
only software decoding is used (presumably from my analysis).

Oh well, that is it for today.

I don't know if trailer was decoded with HW without Shark007 installation I wonder if some kind soul can reproduce my result using Amazon Japan Prime video...

With the Shark007 setting noted in comment 47, I found out that Microsoft Edge uses HW decoder to show the main title I selected from Amazon Japan Prime Video.

What gives?

Should I thank MS for this?

Now that the issue is quite likely peculiarity of Amazon prime video, which does not seem to be handled by FF and Google Chrome
I searched for

Amazon prime video hardware decode

using google.

I found a few issues regarding hardware decoding not working on smartphones in the last 12 months or, so what I see may be
related to these hardware decoding issues.
A few links.

Just a data point.

I wonder how MS Edge handles the stream from Amazon.

The attached table is what I have found out on a PC with Nvidia graphics adaptor.
I checked another PC with RADEON graphics board. The table is the same there.
(However, on that PC with Radeon, I see less often video degradation. Different CPU, etc., I suppose.

The title I used for creating the table.
Youtube 4K (try https://www.youtube.com/watch?v=Q29_OB4e-uk for example)

Amazon Primve Video Trailer: https://www.amazon.com/gp/video/storefront/ref=atv_dp_mv_c_9zZ8D2_1_2?filterId=OFFER_FILTER%3DTVOD
You can take a look at particular title, and often it shows a short trailer in reduced size in upper-right corner. They are somehow decoded using hardware.

Amazon Prime Video main title: You have to be a member of Amazon's Prime service.
They are not decoded using hardware decoder with Chrome and Firefox. But, MS Edge uses HW decoder to my surprise.

(In reply to ISHIKAWA, Chiaki from comment #33)

I am curious about this part.

REUSE_DECODER_DEVICE	
blocked by default: Blocklisted by gfxInfo

Does this mean decoding is not done IF I have more than one video being decoded at the same time?

It means that D3D11 device of WebRender is not re-used for hardware video decoding. When the D3D11 device is not re-used, just a new D3D11 device is created for decoding in DeviceManagerDx::CreateDecoderDevice(). But I do not know how it works with Gecko Media Plugins (GMPs).

I wonder if the problem might related to DRM video decoding. Non drm content does video decoding in GPU process like the following.

But drm video is decoded in a different way. I do not know about it at all.

:alwu, can you comment to Comment 45? Can you explain how drm video is decoded?

Flags: needinfo?(alwu)

(In reply to Sotaro Ikeda [:sotaro] from comment #53)

I wonder if the problem might related to DRM video decoding. Non drm content does video decoding in GPU process like the following.

But drm video is decoded in a different way. I do not know about it at all.

:alwu, can you comment to Comment 45? Can you explain how drm video is decoded?

Interesting. Then, MS Edge must be doing something differently as noted in my comment 51.

Unfortunately, I'm not familiar with the drm video. Bryce is the only person knowing the whole details of drm video, but he already left :(

If only Edge can do that, then I would suspect that the reason is related with CDM difference. Firefox and Chrome uses software DRM (Widevine L3), but Edge uses their PlayReady which already supports HD videos, which could also include the additional hardware decoding support. Also, from Andreas's comment if drm decoding are all performed by openH264, then we don't have a way to know if it does hardware or software decoding.

Flags: needinfo?(alwu)

John might have some thought about Comment 45, he knows DRM better than me.

Flags: needinfo?(jolin)

(In reply to Alastor Wu [:alwu] from comment #55)

Unfortunately, I'm not familiar with the drm video. Bryce is the only person knowing the whole details of drm video, but he already left :(

If only Edge can do that, then I would suspect that the reason is related with CDM difference. Firefox and Chrome uses software DRM (Widevine L3), but Edge uses their PlayReady which already supports HD videos, which could also include the additional hardware decoding support. Also, from Andreas's comment if drm decoding are all performed by openH264, then we don't have a way to know if it does hardware or software decoding.

Thank you for the explanation so far.
Considering the less stuttering of Chrome even it does not use hardware decoding (that is my observation. Chrome is used preferably for seeing Amazon Prime Video on my PC), IF WidevineL3 code is made available in source code, it is better to compile it it with the highest optimization level available for FF. Just a thought.
Or openH264 needs to be compiled with the highest optimization, whichever.
I am not familiar with FF build although, theoretically speaking, I can build it locally. But I am tied up with producing Thunderbird bug fixes myself right now. (I do have M-C/C-C source tree, and I am producing patches and test them locally.)

Although,it may not be known from within FF, at least on Windows 10, we can use task manager which shows the GPU usage to learn if hardwqare decoding is used or not.
That is how I figured Edge uses hardware decoding on this PC of mine that uses Nvidia card, and I also verified it on a different PC with Radeon GPU in it.

DRM is a mess.
I hope google is NOT optimizing WindevL3 for android devices only...

Also, I still have nagging doubt that FF may not be responding to some handshakes correctly and thus
the low-grade video might be sent to FF from the server side.

I studied a bit and in Google's scheme, there are levels of secure devices from the viewpoint of content provider.
L1 is the most secure.
L3 is the least secure.
The content provider is advised NOT to provide content for the securer devices to less secure device.

I found a Japanese blog that discussed that the writer's smartphone suddenly could not view high-quality video from NetFlix.
It seems that the device was suddenly deemed not so secure (L3-level) and thus
the server side decided NOT to deliver L1-level high quality video.
He did a lot of things, but according to him, even resetting the hardware restored the high-quality video device stgatus (L1) only for a day.
The server side somehow gathers information about the user device and decide what level the secure grade is.
In the blog writer's case, properly creating Google account with 2 factor authentication solved the issue. (!?)
So there obviously IS an interaction with the server-side (at least on the android implementation.) and the player side.
In Japanese: https://chibimegane.com/xiaomi-pad-5-widevine/

Checking along this line of events that impaired high-quality video viewing on android phones, I realized the
low video quality issues (or even right out failure to watch DRM-protected videos on android) in the last couple of years are related Google's Windev library AND/OR the SERVER side right management (!).
In one of such blog, I found the following table
in another Japanese blog. https://qiita.com/dcm_okada/items/c460321259413e2f60ec

Widevine DRM 	Google 	https://www.widevine.com/
FairPlay DRM 	Apple 	https://developer.apple.com/streaming/fps/
PlayReady DRM 	Microsoft 	https://www.microsoft.com/playready/

According to the table, Apple uses FairPlay. It would be illuminating to learn and compare how Safari fares in this area.

The degradation of video still continues, but when I read Bug 1738939 "Firefox 94 Download Speed Less Than Half Of Google Chrome", I wonder if that could be the cause of this degradation.
I mean soft decoding puts much workload on CPU, and that could make locally used relatively powerful CPU, AMD Ryzen 3700x 8C/16T running at 4 GHz to download the video stream slowly (?). As I write the last sentence, it is a bit hard to believe, but I want to resolve the video degradatoin issue somehow.. （I tinkered with nvidia video driver in many ways, but no luck so far.）
The problem IS real, and when that happens, it is very unpleasant.

You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: