[mythtv-users] A test for anyone using vdpau on 32-bit x86, especially with 30.0

[Tom Dexter]

On 5/23/19, Peter Bennett <pb.mythtv at gmail.com> wrote:
>
>
> On 5/23/19 8:57 AM, Tom Dexter wrote:
>> On 5/23/19, beppo <beppo75 at web.de> wrote:
>>> On 5/19/19, Tom Dexter <digitalaudiorock at gmail.com[1]> wrote:
>>>> / I'm trying to narrow down some vdpau decode issues with some (but
>>>> not/
>>>> / all) h264 videos as discussed here:/
>>>>
>>>> / https://lists.gt.net/mythtv/users/625246[2]/
>>>>
>>>> / I could use a favor in the form of a quick test from anyone here is/
>>>> / using vdpau on 32 bit x86...especially anyone running version 30.0.
>>>> If/
>>>> / you could try playing this clip:/
>>>>
>>>> / http://jell.yfish.us/media/jellyfish-60-mbps-hd-h264.mkv[3]/
>>>>
>>>> / ...and see if it falls back to software decoding or if it uses
>>>> vdpau./
>>>> / For example from the playback menu Playback -> Playback Data, and
>>>> see/
>>>> / if it shows "H264 vdpau" or "H264 ffmpeg". That would be a huge
>>>> help./
>>>>
>>>> / The failure of the vdpau decode for me seems to be related to the/
>>>> / change in 30.0 to replace the ffmpeg avcodec_decode_video2 call with
>>>> 2/
>>>> / calls...one of which (avcodec_send_packet) is returning an/
>>>> / AVERROR_INVALIDDATA error causing the fallback to software. I'm only/
>>>> / guessing that it might be a 32 bit specific issue, because Peter
>>>> (see/
>>>> / the related bug) couldn't reproduce the failures./
>>>>
>>>> / The videos where I have issues decode and play fine with xine using/
>>>> / vdpau, and I just verified that the version of xine-libs I'm using
>>>> is/
>>>> / still using the old avcodec_decode_video2 as was MythTV 29.1./
>>>>
>>>> / Thanks in advance!/
>>>> / Tom/
>>>>
>>>> I've added a few notes regarding tests I've done to the related bug:
>>>>
>>>> https://code.mythtv.org/trac/ticket/13223[4]
>>> _>_
>>>> As noted there I've also reverted to 29.1 and can confirm that the
>>>> problem videos play with vdpau decoding just fine. Again, I have no
>>>> idea as to whether this is possibly a 32-bit only issue or not, but
>>>> I'm hoping someone else might be able to duplicate the issue. Thanks!
>>>> Tom
>>> Hi Tom,
>>> I have two ASRock ION 330HT
>>> https://www.asrock.com/nettop/NVIDIA/ION%20330HT/
>>> and I can reproduce the fallback to software decoded video, or better an
>>> flip-book @ 10
>>> frames per sec.
>>> 1. ION is 18.04 Ubuntu 32bit
>>> 2. ION is 16.04 Ubuntu 32bit
>>> both maschines are on MythTV 30.0/fixes
>>>
>>> How can I help?
>>>
>>> Beppo
>>>
>> Thanks for the reply! That's significant because it supports my theory
>> that the issue may be specific to 32bit. I'm not sure there's anything
>> you can do to help offhand, but just confirming that you can reproduce
>> this is great. I'll post a comment to the bug indicating this.
>>
>> Thanks again
>> Tom
>> _______________________________________________
> Is x86 still a viable platform? I see from that web page that ION is a
> 64-bit box. Ubuntu only supports 64-bit now, although Xubuntu is still
> available in 32-bit version. I recommend installing a 64-bit version on
> that ION.
>
> I can look into this problem if there is a consensus that i386 still
> needs to be supported.
>
> Peter

Thanks Peter. I'm certainly still using a 32bit (i686) frontend...an
old small form factor system. Since I'm using Gentoo it's very much
still supported. I'm running a 4.19.27 kernel actually. The only real
catch right now is that the nVidia drivers past the 390 version are
now 64bit only...but they still support the 390.xx as legacy drivers.

I know x86 is getting old in the tooth and all, and sure I could build
a new 64bit frontend...but here's the main reason I've yet to do that:
I've been able to get redundant hardware on the cheap used, so my
frontend can die and I can be back in business within about an hour by
swapping out a hard drive. That's really hard to give up, and I'd
basically have to spring for two new systems to have that.

It's certainly a perplexing issue especially given the various ffmpeg
tests I've noted in that bug. Certainly no rush for me on any of that,
but if you get a chance to look into it it'd be greatly appreciated.

Tom