LAV Filters - DirectShow Media Splitter and Decoders

[chros]

@Nev: I have found something interesting things about the videocard power management feature (performace levels) when the player is paused or the playback is ended.

setup: MPC-HC r4040, LAVFilters 0.47, madvr 0.80, CoreAVC 3.0.1, Evr Custom, nvidia driver v295.73, geforce 9600m gt: which has 4 performace levels: p0 (max), p8, p10, p12 (min)

using a h264 sample:
- lav software decoder + evr or madvr : it can fall back to p12 (perfect :) )
- lav dxva2 native decoder or dxva2 copy back + evr : it can fall back only to p8 (but it's better than nothing)
- mpc-hc built-in dxva2 decoder + evr: it can fall back only to p8 too
- lav cuvid decoder + evr or madvr : it sticks to p0 (max) !
- coreavc + madvr : it sticks to p0 too

Can you do about it something?
Mostly, for me, the main problem is when the playback is ended cuvid decoder doesn't allow the driver to fall back into a lower power state...

Thank you!

Edit: I added coreavc and mpc internal dxva to the test.
So it seems that this behaviour is by driver design ... :( But who knows ... :)

[chros]

I have a nvidia 9600m gt in my laptop with (maybe) a VP2 VPU. It only can accelerate h264 in dxva mode.
Lately I checked the VC1 and MPEG checkboxes at the CUVID part of the video decoder, and I can see that there's VPU usage during decoding (with nvidia inspector).
The question: what kind of acceleration is given by the decoder on a VP2 VPU using VC1 or MPEG CUVID? (full, idct, etc.)
I figured out (with the help of mpc-hc evr custom renderer), that my geforce 9600m gt is doing IDCT to mpeg2 and vc-1 streams with the help of LAV CUVID. (So it has only a VP2 VPU.)

[aufkrawall]

@
Can you do about it something?
You can do something about it by yourself. :)
Or to be more precise, by using the multi display power saver (can be also used for single display) of the Nvidia Inspector:
http://www.3dcenter.org/artikel/nvidia-inspector-bringt-komfortables-multi-display-power-saving
Guide is only in German, unfortunately.

E.g. mine looks like this:
http://www.ld-host.de/uploads/thumbnails/20d4b0adacaff2e1b85eb7085213394f.png (http://www.ld-host.de/show/20d4b0adacaff2e1b85eb7085213394f.png)

It's also nice that it prevents the GPU to get into P0 state while surfing.

[CruNcher]

@Nev
nope libmpeg2 vs libavcodec cant be the cause for this strange peakings because they are also happening with Quicksync and Lav Video in this constellation it must be something else that causes the hicups (exact interval jumps in cpu utilization) deeper inside maybe the threading core.

[06_taro]

@nev

Could you please make it possible to set merit of HW accelerate settings?

e.g.,
use DXVA2(native) whenever possible,
else if decoding by DXVA2 is still possible, but renderer is not EVR, or some other filters pin between LAV Video and EVR, switch to DXVA2(copy back),
otherwise(video cannot be decoded via DXVA2 at all), switch to software decoder.

[dukey]

nevcairiel, I am getting a crash in my EVR custom presenter with DXVA and mpeg2 only. H264 appears to be fine.
The error is caused by

SafeRelease(&m_pDeviceManager);

m_pDeviceManager is of type IDirect3DDeviceManager9

My filter graph looks something like

filter name Default DirectSound Device
filter name Enhanced Video Renderer
filter name LAV Audio Decoder
filter name LAV Video Decoder

Using haali media splitter as source currently.

I've not had this problem with other codecs using DXVA eg CoreAVC or the MS digi tv codec.

[GollyJer]

how the hell do you bitstream pcm? and what would be the point anyways?
There is no such thing as "PCM bitstreaming".
If you mean what i think you mean, its also nothing an audio decoder can do, and needs to be done in the audio renderer - hence ReClock.
give lav audio a dts-hd ma encoder and have all output as dts-hd ma? That would allow you to bitstream pcm...in a sense.
OK, I'm obvously a little confused here. I was under the impression that "bitstreaming" means passing the audio from the PC to the AV receiver "untouched". I wrote this post on getting bitstreaming audio to work in Windows 7 (http://www.gollyjer.com/2011/02/quick-guide-how-to-bitstream-audio-in.html) that many have found useful but it appears I lucked into PCM working correctly.

Here is what I understand about the source HD audio formats found on bluray movies.
DTS-HD MA (Compressed, lossless)
Dolby TrueHD (Compressed, lossless)
PCM (raw, unadulterated, lossless-because-there-ain't-nothin'-to-lose, uncompressed bits)

I want my PC to pass all audio through untouched to be completely managed by the receiver.
From what I understand, after further researching because of your comments, given a bit-exact copy of the audio stream the AV receiver will uncompress and then play DTS-HD and Dolby TrueHD audio, and simply play PCM audio.

After some testing tonight it dawned on my that WASAPI Exclusive is what causes PCM to truly passthrough to the receiver. I just tested the movie Kill Bill for example. Without WASAPI Exclusive the receiver only gets 2 Channel PCM and outputs Dolby PL2 Cinema. Using WASAPI Exclusive via ReClock it gets 3/2/.1 channel PCM and outputs multi-channel audio.

Here's what I don't get. Why does the audio decoder, i.e. - LAV, play a role in bitstreamimg compressed lossless formats but can't help with PCM passthrough? Why does the PC audio decoder come into play at all if the audio renderer can pass the audio through untouched?

[mindbomb]

directsound - multiple audio sources can be played at the same time, only one user defined channel count/bit depth/frequency allowed for all of them
wasapi - only one audio source can be played on a given sound device at a time, any channel count/bit depth/frequency allowed, as long as the hardware supports it.

the reason wasapi did that for you is because you had directsound set to stereo. It is configurable if you go to control panel, sound, select your sound device, right click it and go to properties, and then go to advanced.

Also, I'm not 100% clear on this, since whenever I tell people this, they are incredulous, but bitstreaming is the transport of a compressed stream to an external decoder. So that is why it is impossible, since it is uncompressed.

[RBG]

Can you upload a 2 minutes sample somewhere, which is working?
Thanks

Here you are.

This one is not working. (http://www.mediafire.com/?1yltwkyniual60v)

This one playbacks just fine on every nvidia driver (http://www.mediafire.com/?hk2n89anr4yke41)


I wrote "CUVID DXVA". ;)

I understand that DXVA is often used by many people as common noun for video acceleration, but still it sounds awkward as DXVA and CUVID are two different API.

[nevcairiel]

Here's what I don't get. Why does the audio decoder, i.e. - LAV, play a role in bitstreamimg compressed lossless formats but can't help with PCM passthrough? Why does the PC audio decoder come into play at all if the audio renderer can pass the audio through untouched?

"Bitstreaming" usually refers to passing the compressed formats to an external decoder. Its also called SPDIF Passthrough (or maybe HDMI Passthrough).

Anyway, to do this, the audio needs to be in a certain format, specifically packed in a IEC 61937 format.
This is a special format with a few headers that identify which format is being send, and all that stuff. Someone has to take the compressed audio and format it in this manner - that is usually the audio decoder. The audio itself remains untouched, but the data stream needs to be modified to be able to bitstream it.

PCM on the other hand cannot be "bitstreamed" like that. IEC 61937 is only for compressed formats like AC3, DTS, DTS-HD and TrueHD.
PCM is PCM is PCM. All the decoder does is decode the audio to PCM, and send it to the renderer. There is nothing else it can possibly do.
The renderer then has to decide how it sends the audio to the hardware. Usually there are two options - shared or exclusive mode. In shared mode, it sends it through the Windows Mixer, mixed with all other audio sources, and then send to the hardware. This is not an "untouched" process (it may however be lossless, depending on the source and the "shared" output configuration). In exclusive mode, the audio renderer blocks the audio device, so that only it can send audio, and circumvents the Windows Mixer, so you can get "untouched" audio (assuming the audio renderer itself doesn't modify it)

Sadly the only usable Exclusive Mode renderer is right now ReClock, so you don't get around using it (unless you use a player which ships its own renderer)
MPC-HCs audio renderer supports exclusive mode, however the renderer lacks quite a lot of features and i would never recommend using it to anyone. It'll most likely cause audio glitches in many cases.

[nevcairiel]

Can you do about it something?
Mostly, for me, the main problem is when the playback is ended cuvid decoder doesn't allow the driver to fall back into a lower power state...

Thank you!

Edit: I added coreavc and mpc internal dxva to the test.
So it seems that this behaviour is by driver design ... :( But who knows ... :)

Like you figured out by yourself, the driver controls this.
There are tools to override this, someone linked it above (NVIDIA Inspector with its Multi-Display Power Saver), but there is nothing that can be done automatically.

[jmone]

"Bitstreaming" usually refers to Sadly the only usable Exclusive Mode renderer is right now ReClock, so you don't get around using it (unless you use a player which ships its own renderer)
MPC-HCs audio renderer supports exclusive mode, however the renderer lacks quite a lot of features and i would never recommend using it to anyone. It'll most likely cause audio glitches in many cases.

You also have heaps of options with JR's Media Center own Exclusive Mode renderer .... but you have ot use JR Media Center as the player.

[NikosD]

I have updated my exchaustive survey comparing ALL LAV Video modes (DXVA2 copy-back, DXVA2 native, QuickSync) with ATI and Intel video HW processors, on my list of 10 H.264 clips, 1 VC-1 clip and 1 WMV3 clip.

Also I used other decoders for comparing different software DXVA implementations, like MS DS/MFT, CoreAVC, FFDShow.

LAV CB for Intel SHOULD BE IMMEDIATELY disabled, it makes QuickSync HW look slower than VP5!

Take a look at the benchmark results here:
http://forum.doom9.org/showthread.php?t=163110

I will update Nvidia results soon.

[aufkrawall]

I understand that DXVA is often used by many people as common noun for video acceleration, but still it sounds awkward as DXVA and CUVID are two different API.
So, what would be the best naming then? :)
VPU-CUVID?

Has CUDA APIs to access the VPU or how does it work (not in detail ;) )?
Then just CUVID would be enough, at least if the card has a fitting VPU version.

[bjd]

Sadly the only usable Exclusive Mode renderer is right now ReClock

Yes, but at least providing you set ReClock to "Slave reference clock to audio" and Media Adaptation is locked to the original speed and Mpc-Hc volume control is set at 100%, ReClock will display two magic words next to the Audio Renderer info - "(Bit Exact)".

To me it is as flexible as most people need it to be.

[BetA13]

i have a big problem since 0.47....
CUVID was working perfect before with zero glitches..now its allmost unwatchable :(

I have a GTX460, q9550 @ 3.8Ghz and the 295.51 Drivers...

Here is a screen on what i mean..

CUVID:
http://www.abload.de/img/1vw64s.jpg

DXVA (native or CopyBack | same )
http://www.abload.de/img/4325ryvpy.jpg

i have no clue why this is now..i didnt change anything in my settings...I wanna use CUVID...Did something changed there?

greetz Beta

[madshi]

FWIW, BetA13's decoder queue is too empty when using CUVID, so it seems the decoder isn't delivering frames fast enough, for whatever reason.

[BetA13]

ahh ok...What does that mean madshi? is it my fault? do i have to change something or is something wrong with CUVID? sorry for the noob question but this happened to me now just when i thought i had it all figured out, hehe..

THX for help and answer madshi...Oh and by the way...Good job on MadVR, keep it up mate..i love it :)
Also Props to nevcairiel and thank u so much for the LAV Filters...its a great combo and really nice..MADVR + LAV = best quality ever :) keep it up guys...

greetz Beta

[wanezhiling]

Hi nev, http://www.gokuai.com/f/78Jy98y7s07e0pF9
LAV Splitter 0.47 failed to get video parts.

The file is odd, mediainfo says its progressive but a interlaced one, and seems that only Haali could deal with well.

[madshi]

ahh ok...What does that mean madshi?
It means CUVID is not decoding fast enough in your situation. I don't know the cause/reason.

[BetA13]

It means CUVID is not decoding fast enough in your situation. I don't know the cause/reason.

hmmm, ok..
well gettin back to LAV video 0.46 fixed it again..so it must be something that changed in 0.47 i guess...

[jmonier]

hmmm, ok..
well gettin back to LAV video 0.46 fixed it again..so it must be something that changed in 0.47 i guess...

FWIW, I've always (in any version) seen a problem like this in Windowed mode. It works fine for me in Exclusive mode.

Have you tried Exclusive mode? Is it in Windowed mode when you're using 0.46?

[nevcairiel]

hmmm, ok..
well gettin back to LAV video 0.46 fixed it again..so it must be something that changed in 0.47 i guess...

Try another driver, there are known issues with the 295 driver.
For the record, nothing really changed in 0.47 with CUVID decoding.

[BetA13]

Try another driver, there are known issues with the 295 driver.
For the record, nothing really changed in 0.47 with CUVID decoding.

i used exclusive mode also..

anyway..THX to nevcairiel for the tip..Dankeschön :)
Updated to the newest Driver (295.73 WHQL) and look at that..works like a charm again..Thanks nev ;)

:D

CUVID:
http://www.abload.de/img/sdvlkke6.jpg

decoder queue is also working fine again lol.
didnt know there was such an change in the drivers...

THX for the help guys...

:p

[SamuriHL]

Nev, are there any issues with the released 295 driver?

[nevcairiel]

Nev, are there any issues with the released 295 driver?

Seems to be fixed in the WHQL, as far as i can tell.

[SamuriHL]

Seems to be fixed in the WHQL, as far as i can tell.

Alright, sweet. I'll probably update to it then. I had gone back a few versions but I'll be getting a game in a couple weeks that will perform much better with the WHQL driver.

[Xaurus]

I have also been testing 295.73 and I am really happy with them, they did not have the memory leak from the previous beta and I am certain that there were fewer presentation glitches in the clips I tried.

[SamuriHL]

That's great to hear. Thanks for the feedback!

[Mosley]

I have also been testing 295.73 and I am really happy with them, they did not have the memory leak from the previous beta and I am certain that there were fewer presentation glitches in the clips I tried.

I confirm that also with ION GPU 295.75 work definitely better then previous official version. Moreover I'm very sadisfied with new LAV filters (0.47) working in DXVA mode. ;)

[NikosD]

I've finished my benchmark survey.

All platforms included (ATI, Nvidia, Intel) and the main decoders:

Microsoft DirectShow,
Microsoft MediaFoundation,
LAV Video all modes (NVCUVID, QuickSync decoder, DXVA2 Copy-back, DXVA2 Native),
CoreAVC (NVCUVID, DXVA2),
FFDShow QuickSync decoder.

Results here:
http://forum.doom9.org/showthread.php?t=163110

[Gleb Egorych]

Slow motion problem (http://forum.doom9.org/showthread.php?p=1534900#post1534900) was finally fixed in 295.73.

[n3w813]

I'm seeing an issue bitstreaming DTSHD-MA audio from both MKV and full BR folder rip. Win7-32, Zoom Player, Nvidia HDMI, Elite SC-57 receiver.

The issue I encounter is when bitstreaming DTSHD-MA 5.1 track, my receiver detects the signal as DTS-HD MA but with only 5 channels, no subwoofer channel. All other formats bitstream perfectly fine with all channels (DTS 5.1, DD 5.1, TrueHD 5.1/7.1). I thought it was an issue with the latest 0.47 I installed but I retro installed the last 4 versions and all shows the same. Could it be something wrong with my receiver?

[nevcairiel]

Must be something in your receiver. If the bitstreaming signal reaches your receiver and gets decoded, LAVs job is done. It has no influence on the contents of the signal or the decoding behaviour on the receiver.

[nevcairiel]

Has CUDA APIs to access the VPU or how does it work (not in detail ;) )?
Then just CUVID would be enough, at least if the card has a fitting VPU version.

CUVID is the official name of the CUDA APIs to access the video decoder (all the APIs are named cuvidXXXX), so yes, just CUVID is enough.

@nev

Could you please make it possible to set merit of HW accelerate settings?

e.g.,
use DXVA2(native) whenever possible,
else if decoding by DXVA2 is still possible, but renderer is not EVR, or some other filters pin between LAV Video and EVR, switch to DXVA2(copy back),
otherwise(video cannot be decoded via DXVA2 at all), switch to software decoder.

This may be a possibility for the future, but not soon.

[aufkrawall]

CUVID is the official name of the CUDA APIs to access the video decoder (all the APIs are named cuvidXXXX), so yes, just CUVID is enough.

Thanks for confirmation.

[Redemption80]

Anyone tried the Quadro/Tesla 295.73 drivers, i used to find them noticeably better for HTPC applications.

New in Release 295.73:

•ODE Driver
◦The R295 drivers are the second ‘Optimal Drivers for Enterprise’, a release dedicated to relatively long term stability for ISV certification, OEMs, and Enterprise customers.
•Added Support for OpenGL 4.2 on the Quadro series cards and higher.•Added Support for CUDA 4.1
•Added support for the Open Computing Language (OpenCL) 1.1 in Quadro FX Series x700 and newer as well as the FX4600 and FX5600.•NVIDIA Mosaic
◦Added support for 8-display topologies that use higher resolution displays, such as 2560x1600
•NVIDIA WMI 2.4
◦Added functions to configure Mosaic topologies
◦Added function to control Vsync
◦Added function to set Edid
◦Added property to query GPU available memory
◦Added function to enable/disable the nView Desktop Manager, and property to get the current nView state.
•NVIDIA Control Panel Manage 3D Settings Page
◦Added Preferences tab which lets the user control the application profile notification feature.
◦Upon launch of a windowed 3D program, this feature causes a popup to appear notifying the user which workstation application profile was applied.
•Added GPUDirect for Video to support professional video applications.•256GB Virtual Space
◦OpenGL and CUDA users can now take full advantage of applications that utilize 256GB of system memory
•Digital Video Pipeline
◦Added new API’s for supporting low latency access to SDI video hardware
•Workstation application compatibility fixes. Please read the release notes for more information on product support, feature limitations, driver fixes and known compatibility issues.
•If you would like to be notified of upcoming drivers for Windows, please subscribe to the newsletter.
Existing Support:

http://www.nvidia.com/object/quadro-tesla-win7-winvista-64bit-295.73-whql-driver.html

Not being lazy, at work for another 10 hours

[Xaurus]

Anyone tried the Quadro/Tesla 295.73 drivers, i used to find them noticeably better for HTPC applications.
Interesting.

It doesn't say that the regular 400-series are supported though. How does that work?

[Redemption80]

Sorry, you will need to use a modified inf file with strings for other GPU's.

Extract the driver, then overwrite the existing inf with the new one, then run the installer.

http://www.mediafire.com/?7j1wcck1febbsj0

http://forums.guru3d.com/showthread.php?p=4252344#post4252344

I personally found they tend to be geared more towards stability and image quality, rather than speed and game optimisations.

Could be a placebo effect though lol.

[aufkrawall]

I don't think that in this particular case there would be a huge difference since both drivers have exactly the same version -> the very same driver?

[GollyJer]

I'm seeing an issue bitstreaming DTSHD-MA audio from both MKV and full BR folder rip. Win7-32, Zoom Player, Nvidia HDMI, Elite SC-57 receiver.

The issue I encounter is when bitstreaming DTSHD-MA 5.1 track, my receiver detects the signal as DTS-HD MA but with only 5 channels, no subwoofer channel. All other formats bitstream perfectly fine with all channels (DTS 5.1, DD 5.1, TrueHD 5.1/7.1). I thought it was an issue with the latest 0.47 I installed but I retro installed the last 4 versions and all shows the same. Could it be something wrong with my receiver?This may seem like a stupid question but have you tried powering off (completely unplugging) the receiver and powering it back on. I had the same problem a couple months ago and after pulling my hair out for a while that ended up being the solution. :rolleyes:

[n3w813]

This may seem like a stupid question but have you tried powering off (completely unplugging) the receiver and powering it back on. I had the same problem a couple months ago and after pulling my hair out for a while that ended up being the solution. :rolleyes:

Not a stupid question at all...if it works when I try tonight. ;)

[fairchild]

Hey Nev, I just came across a bug of some sort when using Lav Splitter with PotPlayer. Basically when Lav Splitter is doing the splitting, it doesn't send the proper identification flag to PotPlayer's skin system so that it can display the audio codec used. It only is affecting AAC it seems. You can refer to this post:

http://forum.doom9.org/showthread.php?t=164222

Not a major thing since I primarily use MPC-HC, but thought I'd point it out in case you want to fix it.

[nevcairiel]

Hey Nev, I just came across a bug of some sort when using Lav Splitter with PotPlayer. Basically when Lav Splitter is doing the splitting, it doesn't send the proper identification flag to PotPlayer's skin system so that it can display the audio codec used. It only is affecting AAC it seems. You can refer to this post:

http://forum.doom9.org/showthread.php?t=164222

Not a major thing since I primarily use MPC-HC, but thought I'd point it out in case you want to fix it.

Since playback of AAC has always just worked perfectly, i'll put that down as a PotPlayer issue.
Actually, the media type for AAC is defined by Microsoft, so there really isn't any room to go wrong here.

Edit:
I just remembered something. If you use it in combination with LAV Audio, it might use another media type (for good reasons, though)... :)

[fairchild]

Since playback of AAC has always just worked perfectly, i'll put that down as a PotPlayer issue.
Actually, the media type for AAC is defined by Microsoft, so there really isn't any room to go wrong here.

Edit:
I just remembered something. If you use it in combination with LAV Audio, it might use another media type (for good reasons, though)... :)

Yeah I dunno who's at fault, but it's definitely weird and only happens when Lav Splitter Source is doing the splitting. If I change it to the built-in MP4 splitter it passes through the audio codec information. I also tried different audio decoders with Lav Splitter (ffdshow audio, built-in), and always the information being passed is:

0x4441

When I use the built-in MP4 splitter it passes:

AAC(0xaac0)

[RBG]

I personally found they tend to be geared more towards stability and image quality, rather than speed and game optimisations.

Could be a placebo effect though lol.

Trust me, that is just a common myth, that quadro drivers are stable because they are provided for professional usage. I myself have encountered several much more serious problems(power saving issues) with two different quadro cards than I have ever with GeForce series.

[Mosu]

Hey,

I'm trying to fix an error/improve mkvmerge's handling of COOK audio. Unfortunately I haven't found a way to play back such files. Here's how it is:

A user has uploaded a sample RMVB file with RV40 video and COOK audio codecs. That file plays nicely in MPC-HC and LAVFilter (version from today... that should be 0.47). Everything is A-OK.

Remuxing said file to Matroska with mkvmerge's old way doesn't produce a playable file, but that's what I'm trying to fix. Ignoring this.

Remuxing said file to Matroska with ffmpeg (compiled from today's git) with "ffmpeg -i RMVB_with_cook_audio.rvmb -acodec copy -vcodec copy ffmpeg.mkv" works, but the file cannot be played by LAVFilters. Instead, MPC-HC uses its own internal Matroska splitters. Sound seems to be off. Using "LAVSplitter" in graphedit with ffmpeg.mkv results in an unspecified error, 0x80004005.

Interesting is that even ffplay.exe fails to play COOK in ffmpeg.mkv with an error that the codec's initialization data is missing -- but ffmpeg does create a CodecPrivate element for it. Of course it's possible that ffmpeg's Matroska demuxer passes CodecPrivate for COOK differently than its Matroska muxer does.

Before I upload said sample file (it's 500 MB) I'd like to ask:

1. Is COOK from Matroska supposed to work?
2. If so do you have a sample file that's known to play so that I can compare its results?
3. Do you need my sample file?

[nevcairiel]

1. Is COOK from Matroska supposed to work?

No idea! I don't think i ever saw such a thing.


2. If so do you have a sample file that's known to play so that I can compare its results?

See 1. :)


3. Do you need my sample file?

I have some RMVBs here, i can try the ffmpeg way and see what happens.

[Mosu]

I've dug some more into ffmpeg's source code (yes ffmpeg's, not LAVFilter's, but that shouldn't matter in this case), especially into libavformat/matroskadec.c. I'm pretty sure that there are two issues. First an explanation how RealAudio in Matroska has been stored in the past:

The CodecPrivate element contains a copy of the full RMVB audio property structure (e.g. real_audio_v5_props_t, see https://github.com/mbunkus/mkvtoolnix/blob/master/lib/librmff/librmff.h ) That structure contains certain common header fields for all of RealAudio's codec types, and it can also be followed by a codec-specific initialization data set. This is also part of Matroska's CodecPrivate element.

Now to libavformat's Matroska demuxer and muxer. The first problem is on the demuxing side: libavformat's Matroska demuxer does indeed try to parse the CodecPrivate element according to what I've written above, but not to the extent that it should. It does set several parameters like codec_framesize, sub_packet_size and other stuff from the real_audio_v5_props_t thingy, but it doesn't pass the codec initialization data as "extradata" (in libav* terms) down to the codec. See libavformat/matroskadec.c around lines 1613. In 1613 the "st->codec->extradata" is initialized from the local variable "extradata", but that local variable is not set above for "codec_id == CODEC_ID_COOK".

The second problem seems to be in libavformat's Matroska muxer. When you do a streamcopy like I've hinted at ('ffmpeg -i bla.rmvb -acodec copy out.mkv") then the Matroska muxer will write the codec extradata into Matroska's CodecPrivate element -- but only that codec extradata, not the full real_audio_v5_props_t structure and then the extradata.

Well. That's as far as I've come with studying the code. I'm not even sure it's worth pursuing, but users are pestering me about it ~~ http://forum.doom9.org/showthread.php?p=1551018#post1551018

[DragonQ]

Just did a bit of a power consumption test on my HTPC using LAV in MediaPortal. Here's my results:

CUVID, 576i25: 61 W
CUVID, 1080i25: 65 W
DXVA2 Native, 576i25: 48 W (deinterlacing is poor)
DXVA2 Native, 1080i25: 54 W (playback is broken, picture freezes then catches up every second or two)

Unfortunately, I cannot force my GT 430 into the P8 power state (designed for video) using nVidia Inspector when using CUVID. If I do, 576i25 nearly plays back fine (80+% GPU usage and a few dropped frames) but it's nowhere near powerful enough for 1080i25. I'm not massively surprised by this since I assumed the P8 power state was more useful for the higher end cards, to stop them unnecessarily being in maximum power mode all the time. For a GT 430, it seems it has to be in full power mode when using CUVID.

Since DXVA2 playback is broken for me (and even if it wasn't, the deinterlacing isn't good enough), I guess I have no way of reducing the power usage :(. It's already fairly low but it would have been nice to reduce it anyway!