LAV Filters - DirectShow Media Splitter and Decoders

[pururin]

If you're talking about 32-bit IEEE float, converting to floating point is much more complicated than adding zeroes. However, if done correctly, 24 bit → 32 bit FP → 24 bit is lossless (bit-perfect). Of course most software will probably dither when converting back to integer, so it won't be bit-perfect, but the difference is absolutely negligible.

Oh, that's interesting. Isn't more complicate= more lossy? I have no idea about converting to float :confused:

How can I achieve that btw? Now I connect to ffdshow processor, does it do correctly going forth and back like u said?

[nevcairiel]

Oh, that's interesting. Isn't more complicate= more lossy? I have no idea about converting to float :confused:

Float isn't all that complicated, but its more then just a shift.
16-bit Integer is defined as 32767 to -32768. Float is defined as 1.0 to -1.0

To Convert from int to float, this is done: (16-bit int as example)
float = int / 32768

16-bit and 24-bit Integer can be stored in float without losing any data.

[e-t172]

How can I achieve that btw? Now I connect to ffdshow processor, does it do correctly going forth and back like u said?

The way I do it is this: everything works in 32-bit float in the entire chain until the very last step. This means I use ffdshow/LAV in 32-bit float mode only, same for Reclock, and output everything to DirectSound as 32-bit float. Then the Windows audio engine takes over (still in 32-bit float), does the necessary mixing and other adjustments, and finally dithers to 24-bit integer and sends the result to my sound card.

Using this configuration, there is *never* any conversion of audio sample format, except the necessary integer conversion at the sound card output stage.

[nevcairiel]

Using this configuration, there is *never* any conversion of audio sample format, except the necessary integer conversion at the sound card output stage.

And the conversion to float at some point, at least for lossless codecs like TrueHD or FLAC that output integer.... :)

[e-t172]

Yeah, but it's a int → float conversion, which should be lossless. Most float → int conversions aren't due to dithering.

[pururin]

Float isn't all that complicated, but its more then just a shift.
16-bit Integer is defined as 32767 to -32768. Float is defined as 1.0 to -1.0

To Convert from int to float, this is done: (16-bit int as example)
float = int / 32768

16-bit and 24-bit Integer can be stored in float without losing any data.
I see. That's a lot easier than I thought. Thx nev!
But If you may(and boring enough atm)... What about 32-bit integer?
- - - - -

So it's *perfectly* fine to convert 16/24 bit integer to 32 float (lossless), but the downside is when down-converting back to integer, as most DACs accept 24 bit int.

What's the easy way and the hard-better way to get it back to integer?
no-dithering, some dithering, complex-dithering ?

From what I've observed, both LAV and ffdshow seem like don't consume cpu at all (sure it did, but show as 0 all the time)
when converting between formats whether float or int.

So I guess dithering isn't that hard or consumable.
Or there's no dithering apply? :eek: What it really does actually then?


The way I do it is this: everything works in 32-bit float... Then the Windows audio engine takes over... and finally dithers to 24-bit integer and sends the result to my sound card.


Thx mate :cool: This way the Windows mixer dithering it and everything include digital volume leveling.

But what about WASAPI exclusive mode like when using Reclock? Quite a number of people prefer bypassing the Windows mixer.
So in this case.. will have to trust Reclock in how it does the float → int conversion I guess.

Btw do you know whether DACs can do volume adjustment at hardware? Is there such a way?

[iron2000]

Strange, I get 'Available' instead of 'Active' when watch most of the H264 mkv files.
Is it because I'm using Haali Media Splitter?
H264 is checked for HW decoding.

[DragonQ]

Mine always says "available" in the main configuration window. If you're using MPC-HC, you have to go to Play -> Filters -> LAV Video Decoder to see the configuration window with one of the decoders "active".

[nevcairiel]

So here i was, thinking about the VC-1 troubles with hardware decoders and the complexity in remembering to enable the decoder to utilize it.
A solution that only uses Hardware and never software was the first idea, but it doesn't really work all that well, with some decoders we just don't know early enough if the hardware can decode it.

Second idea, just implement a decoder based on Microsofts WMV DMO.
You know, this was much easier then i expected.

http://files.1f0.de/lavf/LAVFilters-0.48-26-gb6411b9.zip

- VC-1 decoding is now enabled by default, and uses the MS WMV9 DMO decoder
- You can turn the DMO decoder off on the "Formats" property page in case you want the ffmpeg decoder (new option at the bottom)

I did a quick benchmark on a 1080p VC-1 file (progressive):
- ffmpeg did around 68 fps
- MS DMO does 101 fps

So, its faster, it can decode interlaced content, sounds like a win-win. :)
This is the same decoder ffdshow uses when you select "WMV9", however its a completely separate implementation. Cleaner and more complete. :)

I appreciate any testing you can do. I'm sure there is still some issues in it.

The next question is, how do i enable it by default for everyone, considering the registry already contains the "off" setting from previous installs <.<

[DragonQ]

Surely the installer can set the default to on for anyone using the installer to upgrade. Can't really help with testing this as I only have a few VC-1 files and none are interlaced.

[fastplayer]

The next question is, how do i enable it by default for everyone, considering the registry already contains the "off" setting from previous installs <.<
This only applies to people that have already made changes in LAV, right? I'm pretty sure they can figure it out for themselves.

[kalston]

But what about WASAPI exclusive mode like when using Reclock? Quite a number of people prefer bypassing the Windows mixer.
So in this case.. will have to trust Reclock in how it does the float → int conversion I guess.


Yes, nothing wrong with having Reclock do the float to int conversion.

[nevcairiel]

This only applies to people that have already made changes in LAV, right? I'm pretty sure they can figure it out for themselves.

Anyone that ever did any change in the config, yes. Thats when it writes the settings.

[e-t172]

So it's *perfectly* fine to convert 16/24 bit integer to 32 float (lossless), but the downside is when down-converting back to integer, as most DACs accept 24 bit int.

What's the easy way and the hard-better way to get it back to integer?
no-dithering, some dithering, complex-dithering ?

From what I've observed, both LAV and ffdshow seem like don't consume cpu at all (sure it did, but show as 0 all the time)
when converting between formats whether float or int.

So I guess dithering isn't that hard or consumable.
Or there's no dithering apply? :eek: What it really does actually then?

If by "hard" you mean "CPU-hungry", then you're right: nowadays, we have CPUs which can do that (and plenty other audio effects) in their sleep. Dithering a simple audio stream can be done so fast that it will probably not even be visible on the task manager. It's basically a non-issue.

[madshi]

Second idea, just implement a decoder based on Microsofts WMV DMO.
You know, this was much easier then i expected.
Nice! May I suggest adding the following code (copy & paste from madVR):

// tell the MS VC-1 decoder to *not* deinterlace
HKEY hk1;
if (SUCCEEDED(RegOpenKeyExW(HKEY_CURRENT_USER, L"Software\\Microsoft\\Scrunch", 0, KEY_ALL_ACCESS, &hk1)))
{
if (RegQueryValueExW(hk1, L"Deinterlace.old", NULL, NULL, NULL, NULL) != ERROR_SUCCESS)
{
DWORD value = 1;
DWORD size = 4;
if (RegQueryValueExW(hk1, L"Deinterlace", NULL, NULL, (LPBYTE) &value, &size) == ERROR_SUCCESS)
RegSetValueExW(hk1, L"Deinterlace.old", 0, REG_DWORD, (LPBYTE) &value, 4);
value = 0;
RegSetValueExW(hk1, L"Deinterlace", 0, REG_DWORD, (LPBYTE) &value, 4);
}
RegCloseKey(hk1);
}
It's needed (at least on XP, not sure about win7) to make sure that the MS VC-1 decoder does not apply some stupid butt ugly and very slow field blended "deinterlacing" for interlaced material.

[cengizhan]

nevcairiel,

You could make an exception for next version and reset wmv decoding to on with 0.49.

[nevcairiel]

It's needed (at least on XP, not sure about win7) to make sure that the MS VC-1 decoder does not apply some stupid butt ugly and very slow field blended "deinterlacing" for interlaced material.

Are you sure that this doesn't only affect the DMO Wrapper filter that gets loaded by DirectShow?
At least on my Win7, the DMO that i load directly doesn't do any deinterlacing, and outputs interlaced flags and everything.

The DMO even has a property to programmatically enable/disable deinterlacing. I can try to enable that and see if something happens, and force that value to off.
( http://msdn.microsoft.com/en-us/library/windows/desktop/ff819311(v=vs.85).aspx )

Maybe you can test it on XP? :) I have no such system anymore.

Edit:
I enabled the property on the DMO, and boy is that ugly. :D
I would've noticed that before. ;)

I kept the code and force the setting to 0, but thats what it seemed to default to for me too.

[madshi]

Are you sure that this doesn't only affect the DMO Wrapper filter that gets loaded by DirectShow?
At least on my Win7, the DMO that i load directly doesn't do any deinterlacing, and outputs interlaced flags and everything.

The DMO even has a property to programmatically enable/disable deinterlacing. I can try to enable that and see if something happens, and force that value to off.
( http://msdn.microsoft.com/en-us/library/windows/desktop/ff819311(v=vs.85).aspx )

Maybe you can test it on XP? :) I have no such system anymore.

Edit:
I enabled the property on the DMO, and boy is that ugly. :D
I would've noticed that before. ;)

I kept the code and force the setting to 0, but thats what it seemed to default to for me too.
I've just tested your beta build on XP and the registry value does affect your decoder implemention, too. If you want, you can upload another beta build with that DMO deinterlacing property set "off". If that works we can forget about the registry hack.

Edit: And in XP the ugly deinterlacing in *ON* by default (!!).

[DragonQ]

With YADIF enabled in LAV you should see "50.000 fps (says source filter)" [in MadVR]. If you don't then you need to talk to nevcairiel.
I see 25.000 fps. :p

Minor cosmetic bug since it's definitely 50 fps.

[nevcairiel]

I see 25.000 fps. :p

Minor cosmetic bug since it's definitely 50 fps.

Since most screens run at 50Hz and not 25Hz anyway, thats really just cosmetic, and fixing it would be somewhat annoying.

[nevcairiel]

I've just tested your beta build on XP and the registry value does affect your decoder implemention, too. If you want, you can upload another beta build with that DMO deinterlacing property set "off". If that works we can forget about the registry hack.

Edit: And in XP the ugly deinterlacing in *ON* by default (!!).

Try this:
http://files.1f0.de/lavf/LAVVideo-0.48-28-g2b854b9.zip

Put ontop of the previous test version.

[King Kong]

I've just a little question out of curiosity: if I mostly handle around with 1080p MKVs (AVC & VC-1) in conjunction w/ i7-2600K & AMD Radeon HD 6950 2GB should I regularly use the Software decoding mode still on or would it be better to switch to DXVA?

Regarding to a quick stress test with "Avatar" (by James Cameron) I hardly recognize any substantial increase neither in Software decoding mode nor in DXVA mode - it's always fluent in both so I'm thinking to leave it as it is (Software decoing mode) or will my electricity bill decrease if I decide to DXVA?

Well, the only drawback I've encountered so far: a test file of "Avatar" and "Devil May Cry" w/ 60 FPS plays smooth with Software decoding mode and stutters/unwatchable with DXVA.

What do you suggest to be best for my rig?
Thanks in advance!

[madshi]

Since most screens run at 50Hz and not 25Hz anyway, thats really just cosmetic, and fixing it would be somewhat annoying.
It's not just cosmetic. There are some screens which support 25Hz and which may do intermediate frame interpolation to improve motion sharpness and smoothness. Also, madVR takes the movie framerate into account when deciding whether deinterlacing should be enabled or not. For 50p content deinterlacing is definitely disabled, for 25p/50i content maybe not.

Try this:
http://files.1f0.de/lavf/LAVVideo-0.48-28-g2b854b9.zip

Put ontop of the previous test version.
Sadly, it doesn't work, the registry value wins... :(

[nevcairiel]

Sadly, it doesn't work, the registry value wins... :(

Stupid XP. On 7, even explicitly setting that value to 1 has no effect when i use the programmatic interface.
Guess i'll overwrite that setting as well now. We'll be doing the world a favor by disabling that in the registry for everyone. ;)

[fairchild]

Quick test for you on my sig rig:

Lav old VC-1 95fps
Lav Microsoft WMV decoder VC-1 120fps
Lav DXVA2 CB 65fps

Test using null renderer + graphstudionext + twinpeak.1080.wmv (the 12.5Mbps version)

[VipZ]

Awesome work on the addition of the WMV9 DMO, all my samples work perfectly with it including VC1 within EVO and that halo sample.


@King Kong, I would recommend software decoding, the 2600k is very efficient at it.

[madshi]

Stupid XP. On 7, even explicitly setting that value to 1 has no effect when i use the programmatic interface.
Guess i'll overwrite that setting as well now. We'll be doing the world a favor by disabling that in the registry for everyone. ;)
I definitely agree. I don't see any situation where anybody could possibly want that ugly (and slow!) deinterlacing.

(It would be nice if you could use the same "Deinterlace.old" logic. Doing so would mean that only one of us will ever change that registry value on an XP installation, and afterwards it will never be changed (by us) again. So if the user insists on changing the registry value back manually, we won't override it for him again.)

[aufkrawall]

Hm, I'm using Windows 7 x64 but somehow the VC-1i sample is always deinterlaced, even if deinterlacing is disabled in LAV and madVR.
How to disable, if possible?

[FlashGordon]

The way I do it is this: everything works in 32-bit float in the entire chain until the very last step. This means I use ffdshow/LAV in 32-bit float mode only, same for Reclock, and output everything to DirectSound as 32-bit float. Then the Windows audio engine takes over (still in 32-bit float), does the necessary mixing and other adjustments, and finally dithers to 24-bit integer and sends the result to my sound card.

Using this configuration, there is *never* any conversion of audio sample format, except the necessary integer conversion at the sound card output stage.

The way I currently have it is: All output formats checked in LAV Audio. Everything except Dithering and Noise Shaping checked in Processing for ffdshow, and all output formats checked in Output for ffdshow. Then 24 int padded to 32 for Reclock using WASAPI. I notice that ffdshow will output Float when I play DVDs, but for some Blu-rays it will output Integer.

My question is, is it better to just uncheck everything BUT 32 Float in LAV Audio and ffdshow and have ReClock convert the float to 24 int padded at the end (similar to the way you're doing it), or is my current set-up fine? Like Nev said, some lossless codecs output integer and so there must be a conversion to float. So, I'm thinking that in my current set-up, LAV outputs Integer for blu-rays with lossless and then that is converted internally by ffdshow to 32float for downmixing and processing and downconverted back to integer for output? If not, then ffdshow is processing in integer, which is less precise, right?

I know Nev recommends having all output formats checked, but since I'm always going to be resampling the audio with ReClock and downmixing with ffdshow, I'm leaning towards your method and just using 32 float all the way to the end...

[Pat357]

Nev,

Can you have a look at this ?

http://www.mediafire.com/imgbnc.php/4ddea0ff3b24eeb744f86b42b5e58ce5d8e2ecf081c86e99b303a5645aee8d636g.jpg

It was posted before, but it's still there in the latest LAV git build be067e7.
I get this when I open my WMP in a certain directory : WMP crashes/closes itself because of this.
Does have something to do with building thumbs using LAV i suppose...

The strange thing is that every file in that specific directory plays fine ... maybe I get this because one or more files are not-seekable although playable ?


For some reason, I could not upload the .jpg image to attach it here : I always got the "This is not a valid image file.".
I tried .jpg , .png but no avail.

[ryrynz]

Nev, is there any chance of stereo to 6/7 expansion in LAV Audio?

[nevcairiel]

Nev,

Can you have a look at this ?

http://www.mediafire.com/imgbnc.php/4ddea0ff3b24eeb744f86b42b5e58ce5d8e2ecf081c86e99b303a5645aee8d636g.jpg

It was posted before, but it's still there in the latest LAV git build be067e7.

I removed the error message now, but if you can figure out which file causes it, i would still love to get my hands on a sample to try to fix seeking properly.

Nev, is there any chance of stereo to 6/7 expansion in LAV Audio?

If you're happy with just copying the audio to all the extra channels, sure.
Personally, for movies i think that sounds like crap and you should use something like DTS:NeoX or Dolby Pro Logic IIx (which LAV will not offer). For music it might be OK.

[ryrynz]

That would be great, I find stereo x7 mode on my receiver to generally sound better than any surround mode it has and having this option would mean I wouldn't need to determine if the video has stereo or 6 channel sound.
I can just keep it in surround mode and never bother with stereo x7 for videos :D

[red5goahead]

It would be awesome if "hardware decoder to use:" had a new option "automatic" that choose Dvxa (copy-back) as prefer option if avaiable for Ati Card and perhaps Cuda for the nvidia.
"automatic" could replace none option that is the default at this moment.

:thanks:

[ryrynz]

Standard software decoding is default when set to 'none' it's probably best that people who know what they want set that option accordingly.

[red5goahead]

Standard software decoding is default when set to 'none' it's probably best that people who know what they want set that option accordingly.

Using it as myself into an other free software an automatic option to set the best choice is better.

[iron2000]

Strange, I get 'Available' instead of 'Active' when watch most of the H264 mkv files.
Is it because I'm using Haali Media Splitter?
H264 is checked for HW decoding.

Ok, on further reading of the imouto.my guide it seems like only software decoding mode works for Hi10p videos.

[sneaker_ger]

Would it be possible for LAV to report the whole stream as interlaced to madVR if LAV is in software mode (with YADIF disabled) and the "Force Deinterlacing" option is ticked? Currently madVR says "deinterlacing off (says bitstream)".

[nevcairiel]

Would it be possible for LAV to report the whole stream as interlaced to madVR if LAV is in software mode (with YADIF disabled)? Currently madVR says "deinterlacing off (says bitstream)".

"says bitstream" means that madVR determined this on its own, "says upstream" means the decoder is reporting something.
For the record, LAV will always report a stream as interlaced unless it has absolute proof that its fully progressive (h264, mpeg2 and vc1 have such information in the header, for example)

[sneaker_ger]

Oh, I thought madVR would rely on info by LAV. Will bug madshi, then.

[CruNcher]

@Nev
Quicksync overrides the Use WMV9 Decoder DMO for WMV3 and VC-1 selection ?

also im really wondering why i get no DXVA for this VC-1 AP@L4 in .WMV

Video: WVC1 1920x1080 59.94fps 0kbps [Video]

Video
ID : 2
Format : VC-1
Codec ID : WVC1
Codec ID/Hint : Microsoft
Duration : 44mn 3s
Bit rate : 1 bps
Width : 1 920 pixels
Height : 1 080 pixels
Display aspect ratio : 16:9
Frame rate : 59.940 fps
Bit depth : 16 bits
Compression mode : Lossy
Bits/(Pixel*Frame) : 0.000
Stream size : 330 Bytes (0%)

though not your fault it doesn't work with Microsofts Decoder also only Arcsoft and Cyberlinks VC-1 Decoder (Native Player) do it and only with AV Splitter or Potplayer with the VC-1 Adapter and other GUID ;)

Also your DXVA2 Native fails falling back to Libavcodec

Quicksync is the only way to Playback such .wmv files Accelerated somewhat ;)

ehh wait

Bit depth : 16 bits (that must be wrong)

[nevcairiel]

@Nev
Quicksync overrides the Use WMV9 Decoder DMO for WMV3 and VC-1 selection ?

Of course it does, QuickSync is a hardware decoder.

All Hardware decoders override the Software decoders.
You can turn off VC-1 under the Hardware Decoder configuration, and it'll use the software decoder then.

PS:
When reporting issues, sample file or i'll ignore you. :)

[CruNcher]

Of course it does, QuickSync is a hardware decoder.

All Hardware decoders override the Software decoders.
You can turn off VC-1 under the Hardware Decoder configuration, and it'll use the software decoder then.

PS:
When reporting issues, sample file or i'll ignore you. :)

Ah sorry yeah

http://www.mediafire.com/?ivjjwzmwztq

I created this with Expression Encoder i get it accelerated with Quicksync but DXVA fails only Arcsoft and Cyberlink DXVA it in their Players or Potplayer with their internal DXVA (IDCT mode, though not fully VLD) or tricking Arcsoft and Cyberlink with the VC-1 Adapter then it plays full acceleration VLD but currently it doesn't with the new Driver.

[nevcairiel]

Ah sorry yeah

http://www.mediafire.com/?ivjjwzmwztq

I created this with Expression Encoder i get it accelerated with Quicksync but DXVA fails only Arcsoft and Cyberlink DXVA it in their Players or Potplayer with their internal DXVA (IDCT mode)

Works just fine with LAV DXVA2 on my NVIDIA here.
Of course that doesn't work on Intel, their VC1 is just different, and how it works is not publicly available.

[CruNcher]

yup also this trick http://forum.doom9.org/showthread.php?p=1551696#post1551696 doesn't work currently @ least with the new 2639 driver :(

PS: Yep Arcsoft stopped working with the new Driver Cyberlink still works http://img535.imageshack.us/img535/6205/cyberlinkfullintelwmvvc.png
Though still no go in MPC-HC (won't connect http://img62.imageshack.us/img62/4116/failsmpchc.png , must be a bug) :(

Though converting it to .mkv and then using cyberlink works also nicely in MPC-HC see the ultrafast seeking example (wont work with lav/ffdshow quicksync/wmv decoder/libavcodec) :(

[undo]

I tried using LAV filters on my AMD E-350 (Radeon 6310) machine using mindbomb's setup guide. Everything's ok, except all the hardware decoding options are "not available". Am I doing something wrong or is this GPU not supported?

[nevcairiel]

I tried using LAV filters on my AMD E-350 (Radeon 6310) machine using mindbomb's setup guide. Everything's ok, except all the hardware decoding options are "not available". Am I doing something wrong or is this GPU not supported?

DXVA2 is only supported on Windows Vista/7, if you run XP, it'll show as not available.

[madshi]

FWIW, DXVAChecker claims that on my XPSP3 installation (Ati 3850) VC-1 DXVA2 decoding is supported, but h264 decoding is not.

[RobertinoM]

Hi there,
DXVA Checker claims that on my AMD Radeon HD 6310 nearly all DXVA1 and DXVA2 formats are supported .. but LAV 0.48 tells me on all options "not available"?????

I have XP SP3 all actual drivers... any hints?????

Regs Robertino

[DragonQ]

DXVA2 isn't available on Windows XP, as Nev said two posts above yours...