[RESOLVED] Bug report: VC1 decoding 480i

[Patrick65]

Using DGDecNV as frame server for BD Rebuilder I just found out that decoding VC1 streams in 480i format (1080p works fine) results in corrupt frames.
This is a problem in DGDecNV since other frame servers have no problems (but only deliver half the frame rate of DGDecNV).
Mediainfo reports the streams as

Video
ID : 4113 (0x1011)
Menu ID : 1 (0x1)
Format : VC-1
Format profile : Advanced@L2
Codec ID : 234
Duration : 58min
Bit rate : 5 764 Kbps
Width : 720 Pixel
Height : 480 Pixel
Display aspect ratio : 4:3
Frame rate : 29,970 FPS
Standard : NTSC
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Interlaced
Scan order : upper field first
Bits/(Pixel*Frame) : 0.556
Stream size : 2,34 GiB (71%)

This can directly be seen when opening the file in the DGIndexNV GUI: the display is garbled (only every second field seems to be correct).

GTX 970 with driver 359.06 (latest driver not stable so I am not using it).

[admin]

Please provide a link to a sample that I can use to duplicate your issue. You can use DGSplit to cut off the first 50MB of your stream.

[Patrick65]

Link to sample file posted via PM

[admin]

No problem here with latest 2051 build. Every frame is just fine and dandy. nVidia driver 361.75.

I don't know what you mean when you say other players deliver only half the frame rate. The stream is 29.97 fps and that is what DGIndexNV delivers, unless of course you enable the double-rate deinterlacing.

I tried no deinterlacing, single-rate deinterlacing, and double-rate deinterlacing. All are fine.

Can you give me your INI file? Also, a screenshot of the garbling would be useful.

[admin]

Tests described above were done on my 620. I tried it on my 950 and I do indeed see the garbling there. I have contacted nVidia about this.

[admin]

OK, here's the deal. Maxwell is broken for interlaced VC1, as you have discovered. My awesome contact at nVidia (we are so lucky to have him) has now filed an internal bug report about it. Depending on what they find there may or may not be a workaround for CUVID. However, if you have Win 7/8/8.1 you should be able to set DXVA mode for VC1 in your INI file as a workaround:

UseD3D = 1
Decode_modes=0,0,0,2 (for 2051) or 0,0,2 for (2050)

SW decoders will not have this problem. Other decoders that use the GPU set DXVA mode so if you have good results with them it suggests you are not on Win 10 (see below).

For Win 10, however, DX9 interop is not currently supported. The good news is that it
may be resolved relatively quickly (some good people are on it), hopefully within a
week and it can make it to the next driver release.

If you have Win 7/8/8.1 please try the workaround and report back.

Thank you, Patrick65, for bringing this to light and providing the sample stream. Neither nVidia nor I were aware of this.

[Patrick65]

Wow, you guys are amazing! The way such issues are addressed here is exemplary...glad I could be of assistence!

Using Windows 7 (never cared about the Windows 10 giveaway) I will try the selective SW-switch for VC-1. This will enable me to use DGDecNV throughout without checking the source material for VC-1 content.
Having said this, I use BDRebuilder quite regularly, in fact, this was the main reason for me to get an NVIDIA card and DGDecNV
The frame server plays an important role during the transcoding process, and this is what I meant when mentioning the speed. During transcoding, the frame rate of the material is not an issue but the speed (i.e. processed frames per second). And this is where I noted the huge difference (DGDecNV being twice as fast on VC-1 480i material. Hopefully, this will remain this fast after they fixed it
Anyway, thank you for providing us with such a great tool - no wonder NVIDIA should be as co-operative as they are: this is truly an asset to their products!

[Patrick65]

OK, here's the deal. Maxwell is broken for interlaced VC1, as you have discovered. My awesome contact at nVidia (we are so lucky to have him) has now filed an internal bug report about it. Depending on what they find there may or may not be a workaround for CUVID. However, if you have Win 7/8/8.1 you should be able to set DXVA mode for VC1 in your INI file as a workaround:

UseD3D = 1
Decode_modes=0,0,2,0 (for 2051) or 0,0,2 for (2050)

If you have Win 7/8/8.1 please try the workaround and report back.

The workaround works fine - except I had to use the setting

Decode_modes=0,0,0,2 (for 2051)

to actually modify the behaviour for VC-1 instead of MPEG

[admin]

Good to hear and thank you for the correction. I haven't yet revised the docs for 2051 and so misled myself by referring to the 2050 docs. I have corrected my post above.

And thank you for your kind words. I fully agree about the excellent support for CUVID/CUDA. The graphics driver support in general is a bit spotty right now (especially for gaming) but lots of things are changing (Maxwell, CUDA architecture, DX12 support, Win 10 support, etc.) and the developers are very busy with that. I expect things to stabilize in that regard.

[admin]

An update...

nVidia informs me that the bug has been fixed but it didn't make the 361.xx boat. Look for it in the next major number bump. It will remove the need to run in DXVA mode.