LAV Filters - DirectShow Media Splitter and Decoders

[v0lt]

Originally Posted by nevcairiel

Please try again with the updates I pushed just now (tomorrows nightly, or your own builds).

I checked LAVFilters-0.74.1-28. I could not cause freezing after pressing the Stop button. Thank.

[tiresias]

Thanks for the clarification about the hardware decoder option.

I've been doing some more searching, and found some info specifically about the "DXVA2 native" and "DXVA2 copy-back" settings which I would like to clarify. Can I just run this by you for comments on what is right or wrong in the following statements:

If you are using the standard Windows EVR then it sounds like the best choice is probably "DXVA2 native" or "DXVA2 copy-back".

The setting "DXVA2 native" can only be used where the video decoder is connected directly to the renderer.

If there is a filter between the decoder and the renderer (or there is no renderer because we are outputting to a file) then "DXVA2 native" will drop back to software decoding.

If "DXVA2 copy-back" is selected, then you CAN use a filter between the decoder and the renderer and there is no drop back to software.

More recent GPUs are able to copy from video memory to system memory fast, so that there should not be a performance penalty in always selecting "DXVA2 copy-back" instead of "DXVA2 native" whether connecting directly to a renderer or not, or outputting to a file.

So, of the two settings, "DXVA2 copy-back" should always be selected, except that for OLDER GPUs that are slower to copy from video memory to system memory then selecting "DXVA2 native" may be more responsive when connected directly to a renderer.

Where "DXVA2 copy-back" is selected, regardless of whether the decoder is connected directly to the renderer or not, the decoder will use the first GPU that it finds to do the decoding, which might not be the same GPU that the renderer uses. In that case, is there a performance penalty or other problems where the decoder is using a different GPU to the renderer?

[chros]

Take a look at the links of this post.

[tiresias]

Thanks Chros, that helps a lot.

One more question...

It seems that the vast majority of desktops and laptops nowadays have some form of DXVA2 hardware acceleration. But the LAV Video Decoder installation default for Hardware Acceleration is "None", so inexperienced users are stuck with software decoding from the start (unless they investigate and change the setting).

Wouldn't it be better for the installation default to be DXVA2 copy-back instead? If there was any problem it would drop back to software (none) anyway.

It just seems that basic/inexperienced users are missing out on the hardware acceleration that their PC is capable of because they are set up with software decoding from the start.

Any thoughts on this?

[nevcairiel]

I don't believe "unexperienced" users should actually install LAV Filters themselves. They should just get MPC-HC or another player, which bundle everything they need, and players can also pre-configure it to how they see their user-base.

[v0lt]

Originally Posted by tiresias

Wouldn't it be better for the installation default to be DXVA2 copy-back instead?

It is a bad idea. The "copy-back" feature is redundant for most real systems (this will only increase power consumption). And on some systems (with integrated graphics) there will be problems with the frame output speed (50/60 fps will become unattainable).

[huhn]

just a small heads up that arm based windows notebooks/tables and such are now are now starting to show up.

one example is the surface pro x not a small product and important that it is from microsoft showing once more there commitment to arm.

[clsid]

Those Windows based ARM devices have x86 emulation, so MPC-HC and LAV Filters will work just fine. The chipset GPU supports hardware acceleration as well.

[huhn]

well yes but we talked about this before and nev was thinking about an arm build software decoding is important and the copy operation is as far as i know SIMD optimised.
the emulation is really slow. not a good video but better then nothing:
https://www.youtube.com/watch?v=oR0x_yx-6Rg

real world test with native x86 vs a emulated x86 is at ~11:20.
working is not always good enough...

and just to be clear not saying that this device is enough to justify an ARM build or even making it close to important but just look at the name of the processor this is mostly not the last one.

the surface 2? had a tegra too but newer devices used an intel CPU.

[littleD]

Originally Posted by huhn

real world test with native x86 vs a emulated x86 is at ~11:20.
working is not always good enough...

You mean geekbench chart, which is merely GPU measure benchmark?

I am worried more about raw CPU performance. Cinebench chart is not looking promising..

Is that really possible that emualted Windows gonna win with native arm Android system and even (nonexistent anymore) WinRT? Or other linux distro?

[huhn]

i mean the chrome test the emulated x86 on arm is not competitive at all but edge which was native arm vs native x86 was competitive.
beware these are pretty bad controlled test but they still hint that native ARM code is much faster then emulated code.

windows 10 is native ARM or with other words it has an ARM build.

[oldpainlesskodi]

Hi Guys,

Not sure if anyone can shed any light for me?

I have some videos with a aac 5.1 sound track, if I let Lav Audio handle the playback, the sound is as flat as a pancake (tried with saneer, with all the options on and off, no difference, and tried without saneer, again no difference). However, if I remove Lav audio from the chain, the sound is deep and high, and expansive. My windows audio (HDMI) device is set to 7.1 192 khz, so windows is doing the up sampling (or whatever the term is), but no matter what settings i use if I use Lav Audio, it still sounds flat (tried with mixing, and no mixing options on etc).

Maybe I have missed something obvious? I would have thought that using non-exclusive mode, they should sound the same?

[el Filou]

What is 'saneer'?
For the most accurate audio in LAV:
- enable all output formats
- disable DRC
- disable mixing (should not be needed if you use 7.1)
When you say "if I remove Lav audio from the chain", what decoder is used in that case? If the audio renderer is the same then that other decoder is probably applying processing to the audio that makes it sound different from the accurate decoding of LAV.

[huhn]

https://github.com/alexmarsev/sanear

[oldpainlesskodi]

Thanks for correcting my spelling error Huhn.

@ el Filou - yep, did all that (all the obvious stuff). I can use any other service to render (ffdshow) with all options disabled/inactive, and as I said, the sound is completely different.

Maybe Nev can shed some light.

[nevcairiel]

"flat" audio is not a technical description that allows any kind of diagnosis. LAV decodes all audio as accurately as possible.

[el Filou]

Did you try decoding a track to a WAV using something like ffmpeg or eac3to and then playing that WAV file to see if it's closer to what ffdshow or LAV outputs?
If you look on the Web for "WavDest", you can also try and get a WAV file output of both playback chains to find out what exactly is different in the sound, but you need to use GraphStudio.

LAV was mainly designed to be a simple and accurate decoder, while ffdshow is more a full-blown audio processing framework which increases the chances of 'doing something wrong' even unintentionally. If the sound is different between them, the most probable cause would be ffdshow modifying it rather than LAV not decoding it correctly.

[oldpainlesskodi]

Ok thanks both.

[nevcairiel]

The most likely case is that your AAC audio file decodes to a different channel layout then you might expect, since LAV actually tries to obey the layout information in the file, while older decoders like ffdshow just ignored such info and put it in a generic 5.1. That may be a case of a badly encoded audio stream, or just needing some tweaking in the audio renderer configuration to map it to the speakers you actually have.

Without a file at hand or some info I can actually go on (like maybe the output pin info of both LAV Audio and another decoder), its impossible to say anything else however.

[oldpainlesskodi]

ok thanks Nev, will do a bit more digging.