LAV Filters - DirectShow Media Splitter and Decoders

[NikosD]

@nev

I found out a strange behavior of DXVA today.

Using this sample with MPC-HC latest nightly (LAV internal 0.61.0) in DXVA native mode with various HW decoders from UVD2.2 up to Haswell QS3, it dropped frames after seeking - back and forth - even with Haswell!

Changing mode to DXVA-CB, even with slower decoders like VP5 or Sandy QS1, it didn't drop a lot of frames or even zero!

How is this possible ?

[nevcairiel]

Seeking works in mysterious ways.

(How would I know why your system does what it does? Dropped frames means the decoder did its job, but the renderer didn't like it for reasons unknown.)

[NikosD]

I meant if you could test it yourself and find out same results, in order to exclude my system's strange behavior.

The renderer was the default EVR-CP in all times.

I always have dropped frames in slow decoder situations, when the decoder is not fast enough for the content.

Which means DXVAn slower than DXVA-CB for the same decoder, which is not possible.

[NikosD]

If you still have a Sandy CPU, I tested iGPU using DXVAn under Win 8.1 using latest drivers 3347 and found out a few clips with small artifacts happening after the first second of the clips.

Two samples with artifacts:

H.264 progressive

H.264 interlaced

[NikosD]

Originally Posted by DragonQ

Depends on your hardware. I highly doubt a 5 year old nVidia/AMD/Intel card has the same deinterlacing algorithm as a current model.

I tested my Radeon 3650 (UVD+) and it definitely looks better than YADIF in VC-1 interlaced content and very similar to Radeon 5750 (UVD2.2).

I don't know if Radeon 7000 series or Kaveri have changed their algorithms regarding DI or maybe newer drivers.

I have to admit that I had a not good experience using latest nightly of MPC-HC continuously crashing all my DTS audio samples!, so I couldn't test all of my interlaced clips.

I can't figure out why.

[mhourousha]

Originally Posted by NikosD

QS decoder nowadays has only 2 advantages.
If you have a Sandy processor, it's the only way to HW accelerate VC-1/WMV3 clips, because Intel hasn't implemented a proprietary DXVA HW decoder for those codecs in Sandy, yet.
I have reported this issue to Intel Support forums and I'm looking forward for the next driver release hoping to include it.
For Ivy/ Haswell the issue has been resolved.
The next reason that someone could use QS decoder is in case he has a dGPU as primary - for example for games - but he wants to decode/ playback movies with iGPU (Intel).
Then QS decoder is a nice solution to have both - Gaming in dGPU and playback decoding in iGPU.
For all the other cases - like for example my Haswell iGPU only configuration - go for a DXVA (native) configuration of your favorite player/ decoder.

My HD4000 has decode error in WMV3 too(DXVA2n).It due to some issues in the VC1_VLD2010 maybe,the QS use VC1_VLD_2_ClearVideo instead and everything is OK.

[NikosD]

Why don't you post a sample of the decoding error ?

[mhourousha]

Originally Posted by NikosD

Why don't you post a sample of the decoding error ?

error happened in all wmv3 samples I tested,for example,the'MP10 Digital Life 1080/24p-10Mbps'clip you provided.All wmv3 clips show a single-colored image only.
for VC-1,some color error or fliker occurred in the first few seconds,then everything is OK.
Driver version is 15.33.14.3412

[NikosD]

I have tested with signature system more than 100 samples of VC-1 (progressive & interlaced) and WMV3 clips (in .wmv and .mkv formats) and I can say for sure that the decoding is perfect.

To be honest I thought that Ivy has similar QuickSync ASIC with Haswell, I can't think of a good reason of bad decoding except faulty driver settings/ system of your side.

Unless I'm wrong and Ivy has problems with VC1_VLD2010 (I don't think so), but I don't have an Ivy to test it myself.

[nevcairiel]

If the driver exposes support for VC1_VLD2010, then it should work - or if it doesn't, complain to Intel to disable it.
I'm assuming you are using the latest version of LAV, right? A lot of stuff changed recently to fix decoding on Intel in this mode.

My laptop has a HD4000, I can probably try myself at some point.

[jkauff]

Speaking of the latest version of LAV, do the MPC-HC nightlies keep up with your changes pretty well, or should I download the latest and set it up as external?

[mhourousha]

Originally Posted by nevcairiel

If the driver exposes support for VC1_VLD2010, then it should work - or if it doesn't, complain to Intel to disable it.
I'm assuming you are using the latest version of LAV, right? A lot of stuff changed recently to fix decoding on Intel in this mode.
My laptop has a HD4000, I can probably try myself at some point.

yes, I use the latest version of LAV for testing.OS is win7
I wonder is there any chance for creating dxva device with VC1_VLD_2_ClearVideo instead of VC1_VLD2010?the DXVAChecker show that the QS use the former and everything is OK.

[turbojet]

Is it possible to have 2 separate 32 bit installs of lav video with different settings?

[nevcairiel]

Originally Posted by mhourousha

I wonder is there any chance for creating dxva device with VC1_VLD_2_ClearVideo instead of VC1_VLD2010?the DXVAChecker show that the QS use the former and everything is OK.

Its impossible, it works completely differently, and its not documented how.

Originally Posted by turbojet

Is it possible to have 2 separate 32 bit installs of lav video with different settings?

There is no support for such a thing in LAV, you can of course try to hack it together manually by re-importing settings into the registry automatically or something like that.

[turbojet]

Originally Posted by nevcairiel

Originally Posted by turbojet View Post
Is it possible to have 2 separate 32 bit installs of lav video with different settings?
There is no support for such a thing in LAV, you can of course try to hack it together manually by re-importing settings into the registry automatically or something like that.

I've tried the registry hacking but that only works if there's no decoder process. The goal is to gpu decode for encoding and cpu decode for playback in potplayer, lav video is a must for playback because it's the only decoder I know that supports repeat field first. I know I could achieve this by switching to mpc-hc but would rather not. Is there a hidden way to use a non-registered lav for playback in a player?

[nevcairiel]

Even if its not registered, it reads its settings from the registry.

[NikosD]

Originally Posted by mhourousha

yes, I use the latest version of LAV for testing.OS is win7
I wonder is there any chance for creating dxva device with VC1_VLD_2_ClearVideo instead of VC1_VLD2010?the DXVAChecker show that the QS use the former and everything is OK.

Did you try Microsoft's decoders or PotPlayer ?

[nevcairiel]

I tested LAV 0.61.1 on my Laptop with a HD4000 using the 3412 driver, and everything works perfectly.
I can only conclude that you're either using an old version of LAV, or not LAV at all, or who knows what else.

[mhourousha]

Originally Posted by nevcairiel

I tested LAV 0.61.1 on my Laptop with a HD4000 using the 3412 driver, and everything works perfectly.
I can only conclude that you're either using an old version of LAV, or not LAV at all, or who knows what else.

[NikosD]

@nev

Why do you need SSE2 as mandatory for LAV filters ?
Is this so important, or is it so much faster ?

AMD Athlons, even the most powerful ones like AthlonXP 3200+ don't have SSE2, only SSE.
AMD decided to put SSE2 in Athlon64 CPUs.

Couldn't you enable a different optimized path for SSE2, without dropping support for SSE or older ?