x264 development: reference list reordering testing

[Dark Shikari]

http://mirror05.x264.nl/Dark/test_old.mkv

http://mirror05.x264.nl/Dark/test_new.mkv

Before I commit a change I have locally, I'd like to make sure no decoders barf on it. In short, is there a decoder out there that plays one of these without problems, but not the other?

libavcodec, JM, and Elecard Streameye have no issues. CMA1820 whines and whines but decodes it correctly anyways.

Challenge: find one that does have issues.

Bonus points: decode to raw YUV and diff against a correct decode (e.g. from ffmpeg).

[Chengbin]

Can you make a 30fps version for portable devices? The Archos 5 does not take 60fps well (slight choppiness, which means a lot for a 60fps video)

[Boolsheet]

No problems with the DivX H.264 Decoder, output is identical to FFmpeg.

[RunningSkittle]

NMT plays them both

[Guest]

OK with Nvidia CUVID.

[Dark Shikari]

Can you make a 30fps version for portable devices? The Archos 5 does not take 60fps well (slight choppiness, which means a lot for a 60fps video)Just remux them.

[modlinsky]

Both plays ok on macosx 10.5.8 (ffmpeg decoder)

[roozhou]

OK with CoreAVC 2.0 software decoder, identical to ffmpeg

[CruNcher]

Arcsoft Video Decoder 2.27.319.108 DXVA1/Software (JAN 20 2010) no issues
CoreAVC 2.0 (CUDA) Forceware 197.45 (G92) 1 frame @ the 4 second is lost for both playback continues normally after that black frame for both, you can see a short black screen flash @ that time
CoreAVC 2.0 (Software) no issues

[Fr4nz]

I've encountered a strange problem so I think it's useful to report here. When playing back both "old" and "new" through DXVA (media player is "media player classic (latest rev.)", videocard is an ATI 5770 and drivers are Catalyst 10.3a) this is what I see:

http://img265.imageshack.us/img265/2084/macrosr.th.png (http://img265.imageshack.us/i/macrosr.png/)

As you can see, there are noticeable macroblocks on the rays, and few other macroblocks around in the picture. These macroblocks, during the playback, goes away for an instant when there is an I-frame, but then they come back.

Maybe it's non-DXVA compliant? Or is it a DVXA-ATI bug?

[Dark Shikari]

I've encountered a strange problem so I think it's useful to report here. When playing back both "old" and "new" through DXVA (media player is "media player classic (latest rev.)", videocard is an ATI 5770 and drivers are Catalyst 10.3a) this is what I see:

http://img265.imageshack.us/img265/2084/macrosr.th.png (http://img265.imageshack.us/i/macrosr.png/)

As you can see, there are noticeable macroblocks on the rays, and few other macroblocks around in the picture. These macroblocks, during the playback, goes away for an instant when there is an I-frame, but then they come back.

Maybe it's non-DXVA compliant? Or is it a DVXA-ATI bug?It should be DXVA compliant. It's 8 reference frames at 640x480, which should fit L4.1 requirements.

Sounds like an ATI bug...

[wlee15]

I've encountered a strange problem so I think it's useful to report here. When playing back both "old" and "new" through DXVA (media player is "media player classic (latest rev.)", videocard is an ATI 5770 and drivers are Catalyst 10.3a) this is what I see:

http://img265.imageshack.us/img265/2084/macrosr.th.png (http://img265.imageshack.us/i/macrosr.png/)

As you can see, there are noticeable macroblocks on the rays, and few other macroblocks around in the picture. These macroblocks, during the playback, goes away for an instant when there is an I-frame, but then they come back.

Maybe it's non-DXVA compliant? Or is it a DVXA-ATI bug?

It doesn't happen here on HD 4870 with the same drivers using both MPC HC and Windows 7 media player.

[julius666]

It doesn't happen here on HD 4870 with the same drivers using both MPC HC and Windows 7 media player.

It's fine here too (HD4830). Even with DivX in DXVA mode.

[Fr4nz]

Maybe it's a bug which affects only 5xxx ATI videocards? Is it possible?

PS: My OS is Win7 64-bit.

[nurbs]

Both clips work fine on my WD TV.

[vucloutr]

MPC-HC r1788 x86 (EVR Custom, optimal renderer settings, internal matroska splitter)

- internal ffmpeg & DXVA work perfectly, no errors or dropped frames on both clips
- CoreAVC 2.0 software & CUDA drop 1 frame in the very beginning and 1 more frame at second 9 with CUDA, same on both clips

[creamyhorror]

Both seem to work fine on (S)MPlayer git20100415 using VDPAU on nVidia driver 195.36.15 and a GeForce 8400GS (G98 chip).

I don't know how to check for dropped frames, though. Any tips?

[Atak_Snajpera]

R4850 win7 x64 ffdshow r3355 with DXVA enabled = all ok

[cogman]

I haven't downloaded it yet, what is this patch supposed to accomplish? (IE, what are its benefits.)

[LoRd_MuldeR]

I haven't downloaded it yet, what is this patch supposed to accomplish? (IE, what are its benefits.)

I could imagine that re-ordering the references in a way that a "usefull" ref-frame is kept longer in DPB while a "less usefull" one is dropped earlier, may be helpful in certain situations...

[dstln]

I've encountered a strange problem so I think it's useful to report here. When playing back both "old" and "new" through DXVA (media player is "media player classic (latest rev.)", videocard is an ATI 5770 and drivers are Catalyst 10.3a) this is what I see:

http://img265.imageshack.us/img265/2084/macrosr.th.png (http://img265.imageshack.us/i/macrosr.png/)

As you can see, there are noticeable macroblocks on the rays, and few other macroblocks around in the picture. These macroblocks, during the playback, goes away for an instant when there is an I-frame, but then they come back.

Maybe it's non-DXVA compliant? Or is it a DVXA-ATI bug?

I have the same card, same drivers and have had the same problems. They broke something and you need to go down to 3.0 levels for it to work properly... :rolleyes:

Extremely annoying.

[burfadel]

Try the 10.4 preview driver (its an official unofficial driver, that is, from AMD/ATI but not WHQL or final 10.4). There are also one or two beta drivers floating around which may react differently. The drivers below are all straight from ATI, and are likely more stable than the leaked drivers, some of which are apparently modified.

There is the normal 10.3 driver, which is the current mainstream driver available here (http://game.amd.com/us-en/drivers_catalyst.aspx?p=win7/windows-7-64bit), the 10.3 OpenGL driver (http://support.amd.com/us/kbarticles/Pages/Catalyst-OpenGL-preview-driver.aspx) and the latest preview/hotfix driver, released 15 April: (http://support.amd.com/us/kbarticles/Pages/ATI-Catalyst-10-4-Hotfix.aspx)

Despite it claiming it has only one fix, it is likely to contain many more fixes but is released just to provide a fix for the above issue. Try it out the 10.4 preview driver and see if the problem still occurs :)

[Fr4nz]

Hi burfadel, thank you for your suggestion.
I've tried catalyst 10.4 preview drivers, but the macroblock-bug remains :(

So, maybe it is a bug that affects 5xxx serie...

[Revgen]

Only problem I found was with an old version of VLC Player. Which I only use since it's capturing function isn't broken like newer versions. I usually use newer Mplayers anyway.

Other than that, everything works fine.

[Warpman]

Nvidia 8800gt dxva ->no issues
ffmpeg-mt -> no issues
microsoft DTV decoder-> no issues

(win7 64bit)

[dstln]

Hi burfadel, thank you for your suggestion.
I've tried catalyst 10.4 preview drivers, but the macroblock-bug remains :(

So, maybe it is a bug that affects 5xxx serie...

Used to work perfect to 4.1 (aside from an odd issue limited it to 11 refs but I think that's a dxva global issue tbh). Something got broken :P

Honestly I'd send them an e-mail because they may not even know about it. In fact I may also later.

[DeeGee]

I'd guess Terry Makedon aka Catalystmaker would be the person to ask.
His Twitter account http://twitter.com/catalystmaker

[ffmpeg]

why do we need the feature of reference list reordering?

[schweinsz]

The DiAVC produce identical output with JM bit by bit for the two bitstreams.

[z0rc]

I've encountered a strange problem so I think it's useful to report here. When playing back both "old" and "new" through DXVA (media player is "media player classic (latest rev.)", videocard is an ATI 5770 and drivers are Catalyst 10.3a) this is what I see:

http://img265.imageshack.us/img265/2084/macrosr.th.png (http://img265.imageshack.us/i/macrosr.png/)

As you can see, there are noticeable macroblocks on the rays, and few other macroblocks around in the picture. These macroblocks, during the playback, goes away for an instant when there is an I-frame, but then they come back.

Maybe it's non-DXVA compliant? Or is it a DVXA-ATI bug?

Same issues on DXVA with ATI Radeon HD4650, Catalyst 10.3, MPC HC r1788 both x86 and x64 builds. It looks like an ATI bug. Also I have another one video with same problem, encoded with x264 r1510 and proper DXVA restrictions.

[anyone23]

Both files play fine:

1:
Mobility Radeon HD 2400 XT
Catalyst 10.3
Windows XP SP3
MPC-HC

2.
Radeon HD 4870
Catalyst 10.3b Preview
Windows Vista SP2
MPC-HC

But with Catalyst 10.4 Preview I have the same issues on PC 2.:rolleyes: (Didn't test PC 1 with 10.4 Preview).

[G_M_C]

why do we need the feature of reference list reordering?

I'm curious too. I've searched around on Google, found some explanations referring to better motion vectors because you can reference "better frames", but i also found some articles speaking of multi-view encoding.

[Dark Shikari]

I'm curious too. I've searched around on Google, found some explanations referring to better motion vectors because you can reference "better frames", but i also found some articles speaking of multi-view encoding.It's required to make basically any multi-ref H.264 encoder work correctly. This test was just to see if a particularly obnoxious corner case (that we didn't create previously) caused any problems in any decoders.

[G_M_C]

It's required to make basically any multi-ref H.264 encoder work correctly. This test was just to see if a particularly obnoxious corner case (that we didn't create previously) caused any problems in any decoders.

Ahh, i see. Thx for the info :)
I'll let up to someone else to go into subject of the meaning of your "obnoxious corner case" ;)

[Midzuki]

I'll let up to someone else to go into subject of the meaning of your "obnoxious corner case" ;)

Count me out! ^_^

[radiant chains]

Just a few more confirmations of working decoders - both files play fine with all of the following:

Intel 4500mhd DXVA
WDTV v2
Microsoft Xbox 360

[G_M_C]

Looks like all decoders work fine with this :)

The error displayed by the The Ati, must have been the decoders fault, cause it seems to be the only decoder that displayed errors. Also there are different posts in the MPC-HT thread mentioning errors too. So it might not even be related.

Personally: I've just ordered a HD5770, so i do hope they fix the problem :p