Intel VAAPI video freeze/hang bug
#16
Then your bug is something else. I cannot reproduce the hang with swfilter turned on (respectively Prefer VAAPI Render Method turned off for later versions). Even not after 10 hours continuously watching ...
First decide what functions / features you expect from a system. Then decide for the hardware. Don't waste your money on crap.
Reply
#17
(2014-07-08, 19:30)Matt Devo Wrote:
(2014-07-07, 21:55)fritsch Wrote: No progress.

Go to Video -> Acceleration (Turn on Expert Settings view) and enable "Use SW Filter" ... that will workaround this bug.

this actually makes the issue worse on the ChromeBox on OE 4.0.6. With the software filter option deselected, the GPU will hang for a few seconds, then resume. If the option is selected, the GPU will hang indefinitely. I can reproduce this reliably, and will be testing the FernetMenta master with OE master as soon as it finishes building

Thanks for the update Matt, what do you mean by "will be testing the FernetMenta master with OE master as soon as it finishes building"? Are you testing a newer or modified build of some sort?

tret

(2014-07-08, 19:57)fritsch Wrote: Then your bug is something else. I cannot reproduce the hang with swfilter turned on (respectively Prefer VAAPI Render Method turned off for later versions). Even not after 10 hours continuously watching ...

One interesting thing to note is that the freezing I've been experiencing on the Chromebox is very random. I can watch multiple videos, spanning several hours without a freeze but then, at random, a freeze will occur.
Reply
#18
Matt, I'll enable the SW Filter option this afternoon and let you know if I get the same results. BTW, do you have a video (or videos) in particular that you can regularly reproduce the freeze with? Like I mentioned in my last comment I can go hours (and spanning multiple videos) without a freeze.
Reply
#19
I can only reproduce after some hours 720p50 upscaled via lanczos 3 optimized to 1080. That's what german TV transmits for the FTA HD channels.
First decide what functions / features you expect from a system. Then decide for the hardware. Don't waste your money on crap.
Reply
#20
Time to wake up for Intel's GPU driver developers! Or the GPU part is going to get a disaster as AMD's fglrx. After years of trying no reasonable vpp, what a shame.
Reply
#21
Yeah :-)

@all also affected by that bug, gather your crash dumps and post them into the above bugreport and subscribe yourself. The more the better. But: Don't spam(!). I don't like to read "me too" without proper log.
First decide what functions / features you expect from a system. Then decide for the hardware. Don't waste your money on crap.
Reply
#22
(2014-07-08, 19:57)fritsch Wrote: Then your bug is something else. I cannot reproduce the hang with swfilter turned on (respectively Prefer VAAPI Render Method turned off for later versions). Even not after 10 hours continuously watching ...

(2014-07-08, 22:15)tret Wrote: Matt, I'll enable the SW Filter option this afternoon and let you know if I get the same results. BTW, do you have a video (or videos) in particular that you can regularly reproduce the freeze with? Like I mentioned in my last comment I can go hours (and spanning multiple videos) without a freeze.

(2014-07-08, 23:03)FernetMenta Wrote: Time to wake up for Intel's GPU driver developers! Or the GPU part is going to get a disaster as AMD's fglrx. After years of trying no reasonable vpp, what a shame.


I'm testing with the BluRay version of Joe (2013), extracted to mkv (using MakeMKV), no recompression - simply because someone reported it as having a reproduceable lockup. At 1:16:29, it will reliably lock up completely with 'Use SW filter' enabled; with it disabled, the video will pause for ~5s, then FF to catch up. Here's the debug log (using OE 4.0.6):

http://pastebin.com/rFBUppea

a build I just compiled using OE git master + FernetMenta git master will do the ~5s pause/speedup regardless of the 'Prefer VAAPI render method' setting. It did reboot the first time I testing it, but I haven't been able to reproduce since.
Reply
#23
(2014-07-08, 23:34)Matt Devo Wrote:
(2014-07-08, 19:57)fritsch Wrote: Then your bug is something else. I cannot reproduce the hang with swfilter turned on (respectively Prefer VAAPI Render Method turned off for later versions). Even not after 10 hours continuously watching ...

(2014-07-08, 22:15)tret Wrote: Matt, I'll enable the SW Filter option this afternoon and let you know if I get the same results. BTW, do you have a video (or videos) in particular that you can regularly reproduce the freeze with? Like I mentioned in my last comment I can go hours (and spanning multiple videos) without a freeze.

(2014-07-08, 23:03)FernetMenta Wrote: Time to wake up for Intel's GPU driver developers! Or the GPU part is going to get a disaster as AMD's fglrx. After years of trying no reasonable vpp, what a shame.


I'm testing with the BluRay version of Joe (2013), extracted to mkv (using MakeMKV), no recompression - simply because someone reported it as having a reproduceable lockup. At 1:16:29, it will reliably lock up completely with 'Use SW filter' enabled; with it disabled, the video will pause for ~5s, then FF to catch up. Here's the debug log (using OE 4.0.6):

http://pastebin.com/rFBUppea

a build I just compiled using OE git master + FernetMenta git master will do the ~5s pause/speedup regardless of the 'Prefer VAAPI render method' setting. It did reboot the first time I testing it, but I haven't been able to reproduce since.

Would it be possible to get a 1-2 minute sample including the lockup timestamp? I'd like to duplicate your testing and confirm the same results. Also, is there any way to analyze the video to find out what it is at that timestamp that differs from other video?
Reply
#24
(2014-07-08, 23:44)tret Wrote: Would it be possible to get a 1-2 minute sample including the lockup timestamp? I'd like to duplicate your testing and confirm the same results. Also, is there any way to analyze the video to find out what it is at that timestamp that differs from other video?

link: https://dl.dropboxusercontent.com/u/9830...sample.mkv

sample is 30s long, ~97MB. Will reliably freeze at ~14s.
Reply
#25
(2014-07-09, 00:20)Matt Devo Wrote:
(2014-07-08, 23:44)tret Wrote: Would it be possible to get a 1-2 minute sample including the lockup timestamp? I'd like to duplicate your testing and confirm the same results. Also, is there any way to analyze the video to find out what it is at that timestamp that differs from other video?

link: https://dl.dropboxusercontent.com/u/9830...sample.mkv

sample is 30s long, ~97MB. Will reliably freeze at ~14s.

Perfect, will test this in the next hour or two and report results.
Reply
#26
(2014-07-09, 00:20)Matt Devo Wrote:
(2014-07-08, 23:44)tret Wrote: Would it be possible to get a 1-2 minute sample including the lockup timestamp? I'd like to duplicate your testing and confirm the same results. Also, is there any way to analyze the video to find out what it is at that timestamp that differs from other video?

link: https://dl.dropboxusercontent.com/u/9830...sample.mkv

sample is 30s long, ~97MB. Will reliably freeze at ~14s.

Confirmed, reliably freezes at approx 13-14s. Fritsch, any way you can download the sample and give it a shot to see if (on your test system) you can reproduce the same freezing?

Note: w/ SW Filter enabled it hangs indefinitely. w/ SW Filter disabled it hangs for about 3-4 seconds then ffwds a bit, hangs briefly then resumes playing. Just as Matt stated.
Note 2: Disabling VAAPI allows playback with no freeze although with the high bitrate video it's really too choppy to watch (for me at least) which kinda makes the Chromebox meh unless a better fix/workaround is found.

tret
Reply
#27
Yes, we hit another VAAPI bug here. It hangs in vaSyncSurface. Not sure if it's worth to open another bug reports at freedesktop because those guys seem to ignore us.
Reply
#28
@tret:

Open the bug never the less! It is good to have that sample, make it available to the intel devs.

@FernetMenta: If you have additional debug logs, better you open it :-) - they will start to love us even more :-)
First decide what functions / features you expect from a system. Then decide for the hardware. Don't waste your money on crap.
Reply
#29
I ran the sample on Windows with DXVA hw acceleration (Intel platform). The decoder notifies us about a problem but it does recover. There is a little glitch, hardly noticeable.

@fritsch I won't open another bug report as long as they don't answer the other ones. Intel should follow the example of AMD: hire some top notch devs to implement vdpau, then sunset this vaapi crap.
Reply
#30
(2014-07-09, 05:26)tret Wrote: Note 2: Disabling VAAPI allows playback with no freeze although with the high bitrate video it's really too choppy to watch (for me at least) which kinda makes the Chromebox meh unless a better fix/workaround is found.

tret

you need to enable multi-threaded software decoding under video acceleration settings. It's only available when the hardware options are deselected/disabled.
Reply

Logout Mark Read Team Forum Stats Members Help
Intel VAAPI video freeze/hang bug0