LAV Filters - DirectShow Media Splitter and Decoders

[betaking]

Just speculation. I'm not sure its really worth it.


Thats not a bug.
I just don't believe in screwing up the users DirectShow settings like Haali does, and inserting LAV Splitter as the absolute forced splitter. The player should manage which filter is used, and the user should have a choice. Haali doesn't give you a choice. If its installed, it forces every player to use it by setting special entries in the registry. There is no bug that prevents LAV from being used. If you use the LAV installer, it'll set those Source Filter registry keys for you.

In MPC-HC you can also just set Haali to blocked, which should work.


The DirectShow "way" is to actually use File Source (async) together with a splitter filter. Not a source filter that does both. This works perfectly if you don't have a power hungry source filter that steals the show.


Thats a bug in GraphEdit, nothing i can fix. Use GraphStudio.


http://code.google.com/p/lavfilters/issues/detail?id=1


http://code.google.com/p/lavfilters/issues/detail?id=3

Quote:
Originally Posted by dead_screem
Now, "LAV Splitter Source" doesn't need the "File Source (Async.)" because it doubles as one apparently, and it also has a merit of 0x00800004 which is higher than Haali but Haali still gets used for some reason, infact even when lowering Haali's merit way down "LAV Splitter Source" still doesn't get used, there is probably a bug which is preventing dshow from finding it. The only way I can get dshow to use it is by forcing it's use by setting "Source Filter"="{B98D13E7-55DB-4385-A33D-09FD1BA26338}" at HKEY_CLASSES_ROOT\Media Type\Extensions\.ts for example.

Thats not a bug.
I just don't believe in screwing up the users DirectShow settings like Haali does, and inserting LAV Splitter as the absolute forced splitter. The player should manage which filter is used, and the user should have a choice. Haali doesn't give you a choice. If its installed, it forces every player to use it by setting special entries in the registry. There is no bug that prevents LAV from being used. If you use the LAV installer, it'll set those Source Filter registry keys for you.

In MPC-HC you can also just set Haali to blocked, which should work.
YES last haali install to my system put some to my registry
like it "Source Filter"="XX" at HKEY_CLASSES_ROOT\Media Type\Extensions\.mp4
"Source Filter"="XX" at HKEY_CLASSES_ROOT\Media Type\Extensions\.264
if I use lavspiltte and lavvideo by wmp12 to decoder mpeg4 files(endecoder by X.264 or H.264)WMP12 give my error!First of all I have in HAALI's settings to disable HAALI MKV MP4 separator!
if my del "Source Filter"="XX" at HKEY_CLASSES_ROOT\Media Type\Extensions\.mp4 and "Source Filter"="XX" at HKEY_CLASSES_ROOT\Media Type\Extensions\.264
wmp12 can use lavspiltte and lavvideo to decoder mpeg4 files(endecoder by X.264 or H.264) to fine! so i Never install HAALI!!!!!And to my friends recommend LAV!
big thanks for nev!

[dead_screem]

Thats not a bug.
I just don't believe in screwing up the users DirectShow settings like Haali does, and inserting LAV Splitter as the absolute forced splitter. The player should manage which filter is used, and the user should have a choice. Haali doesn't give you a choice. If its installed, it forces every player to use it by setting special entries in the registry. There is no bug that prevents LAV from being used. If you use the LAV installer, it'll set those Source Filter registry keys for you.

I know what Haali does, my point was that the "LAV Splitter Source" version of the LAV splitter does not load in dshow even with Haali forced reg keys removed and it's merit lowered or even completely uninstalled. only the Version which requires File Source (Async.) will ever be loaded. The "LAV Splitter Source" version can only be used if its forced to by those same reg keys Haali uses.
And if like you also said, the File Source (Async.) requiring ver is the right way to do it anyway... I take it then that "LAV Splitter Source" not loading in dshow normally is by design then? then why is this version of LAV Splitter which does not need File Source (Async.) even included if it's normally never used?

In any case using the File Source (Async.) ver seems to cause slower opening times, probably mostly because File Source (Async.) tries to connect to other filters first before it tries to connect to LAv Splitter. So I'm using those same reg keys to force the use of the "LAV Splitter Source" version to avoid that nonsense. Whyever it's not used normally, I'm glad it's still included anyhow.

Thats a bug in GraphEdit, nothing i can fix. Use GraphStudio.
Oh, well ok then.

http://code.google.com/p/lavfilters/issues/detail?id=1


http://code.google.com/p/lavfilters/issues/detail?id=3cool. I take it these are low priority then?

and what about LAV Splitter TS AAC not connecting to MPC-HC or CoreAAC? Is this not a bug as well? What's the reason?

[nevcairiel]

The Source variant is also required for the Blu-ray support, because it needs to open more then one file.
Anyhow, like i said, if you use the Installer, it'll actually create those "Source Filter" entries. Using the Installer is the recommended way, unless you really know what you're doing with the .zip files.

Regarding AAC, all AAC codecs based on libfaad2 don't work if they don't get AAC init data. Its a serious limitation in the codecs, so meh. I can probably fix that somehow by artificially creating that init data.

[dead_screem]

The Source variant is also required for the Blu-ray support, because it needs to open more then one file.
Anyhow, like i said, if you use the Installer, it'll actually create those "Source Filter" entries. Using the Installer is the recommended way, unless you really know what you're doing with the .zip files.ok... gotcha.

Regarding AAC, all AAC codecs based on libfaad2 don't work if they don't get AAC init data. Its a serious limitation in the codecs, so meh. I can probably fix that somehow by artificially creating that init data.
So then faad2 and CoreAAC and similar decoders are bugged?
Does that mean MPC-HC mpeg ts splitter and Haali ts support are bugged too since they work with MPC-HC/faad2 and CoreAAC? Maybe then MPC-HC guys should fix their splitter and their internal faad2 AAC decoder. Getting Core to fix CoreAAC would be pretty unlikely, as I've been waiting for them to fix the stupid channel order bugs since CoreAAC 2.0 was released..

[nevcairiel]

Its something special about AAC wrapped in ADTS, i haven't really bothered to check into it much.
Both ffdshow and my own LAV Audio offer libavcodec based AAC decoders, which work just fine and support the same features.

Edit:
I just fixed it, adding proper parsing for ADTS.

[nevcairiel]

You say GPU<->CPU data transfer is fast on nVidia. How? CUDA only? Just locking DXVA2 NV12 1080p surface for readout gives me spikes up to 60ms on G92 card.

Thats pretty much what i do.

I dunno about older cards, but on my GTX 570, it tops out at around 200 fps of 1080p NV12 material (with 250 fps from the decoder itself) - that is surface upload, DXVA2-VP processing, surface download.
The process isn't even streamlined yet or anything, i just slaped some test code together to see if its working at all.

I should publish some test version some day for speed tests on different hardware, especially ATI.

[SEt]

Test version would be nice - I'm not sure is I am doing something wrong or it's hardware/driver to blame.

[nevcairiel]

I'm currently re-writing it from DXVA2-VP to use DXVA-HD instead, once its at least working somewhat, i can post something to test.

[fastplayer]

Is DxDiag a reliable indicator of DXVA-HD support? It says "not supported" for my HD4670... :(

[dead_screem]

Its something special about AAC wrapped in ADTS, i haven't really bothered to check into it much.
Both ffdshow and my own LAV Audio offer libavcodec based AAC decoders, which work just fine and support the same features.

Edit:
I just fixed it, adding proper parsing for ADTS.

Cool. What about AAC in LATM for MPEG-TS? Is it the same thing?

[nevcairiel]

Cool. What about AAC in LATM for MPEG-TS? Is it the same thing?

LATM requires decoder support. Use LAV Audio for the best results. ;)


Is DxDiag a reliable indicator of DXVA-HD support? It says "not supported" for my HD4670... :(
No idea. NVIDIA supports it, and its working marvelous. Much better then my previous attempt with DXVA2-VP.

Of course, DXVA-HD is limited to Windows 7 (not even Vista supports it), but its so easy to use, i can't help it. :p

[nevcairiel]

WARNING: Do not use this version without reading this post thoroughly, and understand what it means. Its a TEST version for a new feature, its not usable otherwise!

http://files.1f0.de/lavf/LAVVideo-dxvahd-test.zip

This version will try to send every video frame through a DXVA-HD Video Processor. The idea is to use the GPUs deinterlacer without DXVA or EVR, but its nowhere near finished.
Why am i posting this in its unfinished, barely working state? Mostly, i'm interested if ATI supports this at all. And if they do, how fast it'll be. Another point is speed on other NVIDIA hardware.

How to test? Drop this into your LAV Filters install.
IMPORTANT: Configure LAV Video to only output NV12, disable all other pixel formats, or stuff WILL break. Its not finished yet!

You'll notice if its working, i left a small "feature" in there which should create some visible chroma error in the video, so its quite obvious that its active.

Like i said, i'm mostly interested if it works for you at all, and if it does, how fast.
You can Benchmark with GraphStudios (http://blog.monogram.sk/janos/tools/monogram-graphstudio/) Decoder Performance Test (in the "View" menu)

Here is a short H264 1080i sample i used to test:
http://files.1f0.de/samples/Test_clip_avc.1080i59.94.ac3.5.1.mkv

For reference, without the feature i get around 250 fps, with it enabled i get 200 on that sample file.

It only works on Windows 7! DXVA HD is not supported on previous versions of Windows.

This is a debug build that'll leave a log file on your Desktop. If it doesn't work, it might give some clues why. It also dumps the capabilities of your hardware in the log. The DXVA-HD related parts should be right on top of the log (everything up to "CTransformInputPin::CTransformInputPin")

[SamuriHL]

Thanks, Nev. If I get a chance tomorrow I'll try it out on my 450 and throw my AMD 5870 at it and see what we get for results.

[pankov]

Nev,
this dxvahd test version of LAVVideo.ax doesn't load on my system. It also can't be registered using regsvr. Is this normal? Am I the only one having a problem?

[Superb]

http://i.minus.com/iwJKiTJ6xOGfS.png

I simply replaced the LAVVideo.ax inside C:\Program Files\LAV Filters\x86 w/ the ax file inside the zip.

EDIT: btw, does the time measure filter (which GraphStudio uses) distorts the image on your machine as well? It doesn't affect the test, but I still wonder...

[Xaurus]

Nev,
this dxvahd test version of LAVVideo.ax doesn't load on my system. It also can't be registered using regsvr. Is this normal? Am I the only one having a problem?
Same here. Can't register it and it's not usable by simply replacing the "real" LAVVideo.ax file.

[PrayForMojo]

Is there a way to output to three front speakers with the Lavaudio decoder like you can with the AC3filter?

[clsid]

That debug build probably needs the debug c++ runtime.

[Superb]

That debug build probably needs the debug c++ runtime.Then it would make sense it worked on my machine. I have Visual Studio (which includes VC++) 2010 installed.

[Superb]

Is there a way to output to three front speakers with the Lavaudio decoder like you can with the AC3filter?You mean expand stereo to center? Nope.
LAV Audio doesn't have advanced mixing capabilities yet (only some basic ones, like "mono to stereo").

[nevcairiel]

I uploaded a new build, using static linking against the debug runtime. I usually have dynamic linking on for debug, because it makes certain debug tasks easier.

Same URL
http://files.1f0.de/lavf/LAVVideo-dxvahd-test.zip

For everyone that didn't read the old post:
READ BEFORE DOWNLOAD! http://forum.doom9.org/showthread.php?p=1525394#post1525394

Also, please mention some system specs if you post some results. :)

[Superb]

Also, please mention some system specs if you post some results. :)Mine are written in my signature.

[nevcairiel]

Mine are written in my signature.

I saw, just as a reminder for everyone else. ;)

[VipZ]

Doesn't seem DXVA-HD is supported on the ATI 5xxx series :(

LAVVideo.ax(tid 1e0c) 554 : -> DXVAHD_CreateDevice failed (hr: 80004002)

[nevcairiel]

Doesn't seem DXVA-HD is supported on the ATI 5xxx series :(

LAVVideo.ax(tid 1e0c) 554 : -> DXVAHD_CreateDevice failed (hr: 80004002)

Thats sad. Its such a nice API. DXVA2-VP does not allow me to use the full-quality deinterlacing on NVIDIA, but DXVA-HD does (at least as far as i can see)

Oh well. I still have the code from DXVA2-VP, i should put that up as well for performance testing.

[fairchild]

my results for the Test_clip_avc.1080i59.94.ac3.5.1.mkv clip are:

avg around 141-142 fps with the latest stable LavVideo

avg around 128-129 fps with the LAVVideo-dxvahd-test

Here is the top of the log:

LAVVideo.ax(tid 1344) 0 : DXVAVideoProcessor::InitializeD3D9(): Initializing D3DEx Interfaces
LAVVideo.ax(tid 1344) 20 : -> Creating D3D9Ex Device
LAVVideo.ax(tid 1344) 60 : DXVAVideoProcessor::InitializeDXVA2(): Initializing DXVA-HD Interfaces
LAVVideo.ax(tid 1344) 60 : -> Creating DXVAHD_Device
LAVVideo.ax(tid 1344) 60 : -> DXVAHD_CreateDevice failed (hr: 80004002)
LAVVideo.ax(tid 1344) 60 : CTransformInputPin::CTransformInputPin

Also it always crashes after you close the media file which seems to play fine throughout, not sure if this is normal. System in sig. I also don't notice a difference with it on or off, not sure what I'm looking for that you said is easy to notice if it's working. I even tried another interlaced video that I had and even though it plays fine, deinterlacing is not happening. (again not sure if this is normal as this is a test file and maybe the actual deinterlacing is not active yet)

[nevcairiel]

It worked on your ATI card? Did you see the chroma errors?

The crash is from a bug in madVR, which causes a debug thing to catch a breakpoint.

[Xaurus]

Nev,

Here's the results:
test build (http://home.halden.net/mordor/lav1.png)
normal build (http://home.halden.net/mordor/lav2.png)

All specs & results in the screenshot.

[mindbomb]

19.6 FPS stable
17.9 FPS test

amd e-350

log here
http://pastebin.com/pt5WCxiG

[fairchild]

It worked on your ATI card? Did you see the chroma errors?

The crash is from a bug in madVR, which causes a debug thing to catch a breakpoint.

Yes it runs fine on my HD5830 with 11.8 drivers + 11.8 CAP3. I don't see any difference between the two versions. What should I be noticing as far as chroma errors in that test video?

[SEt]

Without: 263, with deinterlacing on G92: 133

I can't say the quality is good though.

[noee]

On my Toshiba laptop, A6-3400M HD6520, Win7 x64 Home

~54fps and the cores are all at 98% and I can even see the turbo core come on the one core. Hard for me to tell, is the chroma "bug" in the transitions?

I'll try the 6570 HTPC next...

[nevcairiel]

It probably does nothing for ATI/AMD cards. DXVA-HD seems not supported. You would notice the chroma problem immediately.

Here is another version, this one uses DXVA2 Video Processing, instead of DXVA-HD.

http://files.1f0.de/lavf/LAVVideo-dxva2vp-test.zip

This should probably work on ATI/AMD.
Same rules and restrictions apply.

BTW, its not tuned for quality or anything at all yet. Its just to test device capabilities and speed.
Its also a debug build, which is generally slower then a release build. ;)

[fairchild]

Yep, I just looked at the log. Even though it runs and whatnot, I guess it's not doing anything per this line:

LAVVideo.ax(tid 1344) 60 : -> DXVAHD_CreateDevice failed (hr: 80004002)

[noee]

On my 6520, the DXVA2 version gets to the end of the first pass and then just stops. Here is the log:

LAVVideo.ax(tid 670) 57908 : DXVAVideoProcessor::InitializeD3D9(): Initializing D3DEx Interfaces
LAVVideo.ax(tid 670) 57986 : -> Creating D3D9Ex Device
LAVVideo.ax(tid 670) 58074 : DXVAVideoProcessor::InitializeDXVA2(): Initializing DXVA2 Interfaces
LAVVideo.ax(tid 670) 58074 : -> Creating IDirectXVideoProcessorService
LAVVideo.ax(tid 670) 58074 : -> Querying Processor Device GUIDs
LAVVideo.ax(tid 670) 58075 : -> Processor 0 with GUID {3C5323C1-6FB7-44F5-9081-056BF2EE449D}
LAVVideo.ax(tid 670) 58075 : -> Target 0: 22
LAVVideo.ax(tid 670) 58075 : -> Target 1: NV12
LAVVideo.ax(tid 670) 58075 : -> Processor does support NV12
LAVVideo.ax(tid 670) 58075 : -> Querying Processor capabilities
LAVVideo.ax(tid 670) 58075 : -> DeviceCaps: 1
LAVVideo.ax(tid 670) 58075 : -> InputPool: 4
LAVVideo.ax(tid 670) 58075 : -> NumForwardRefSamples: 0
LAVVideo.ax(tid 670) 58075 : -> NumBackwardRefSamples: 2
LAVVideo.ax(tid 670) 58075 : -> DeinterlaceTechnology: 64
LAVVideo.ax(tid 670) 58075 : -> Querying ProcAmp values
LAVVideo.ax(tid 670) 58075 : -> Querying Filter values
LAVVideo.ax(tid 670) 58075 : -> Creating surfaces
LAVVideo.ax(tid 670) 58076 : -> Creating Video Processor
LAVVideo.ax(tid 670) 58076 : -> Creating Processor and Surfaces successfull
LAVVideo.ax(tid 670) 58076 : CTransformInputPin::CTransformInputPin
LAVVideo.ax(tid 670) 58076 : CTransformOutputPin::CTransformOutputPin
LAVVideo.ax(tid 670) 58076 : SetMediaType -- in
LAVVideo.ax(tid 670) 58076 : Shutting down ffmpeg...
LAVVideo.ax(tid 670) 58076 : Initializing ffmpeg for codec 28
LAVVideo.ax(tid 670) 58076 : -> File extension: .mkv
LAVVideo.ax(tid 670) 58077 : -> Processing extradata of 27 bytes
LAVVideo.ax(tid 670) 58079 : -> ffmpeg codec opened successfully (ret: 0)
LAVVideo.ax(tid 670) 58080 : Trying to connect Pins :
LAVVideo.ax(tid 670) 58080 : <XForm Out>
LAVVideo.ax(tid 670) 58080 : <Input>
LAVVideo.ax(tid 670) 58080 : ::GetMediaType(): position: 0
LAVVideo.ax(tid 670) 58080 : Trying media type:
LAVVideo.ax(tid 670) 58080 : major type: MEDIATYPE_Video
LAVVideo.ax(tid 670) 58080 : sub type : MEDIASUBTYPE_NV12
LAVVideo.ax(tid 670) 58080 : ::CheckTransform()
LAVVideo.ax(tid 670) 58080 : ::CheckTransform()
LAVVideo.ax(tid 670) 58080 : SetMediaType -- out
LAVVideo.ax(tid 670) 58081 : ::DecideBufferSize()
LAVVideo.ax(tid 670) 58081 : Connection succeeded
LAVVideo.ax(tid 670) 58081 : ::GetMediaType(): position: 0
LAVVideo.ax(tid 670) 58081 : ::CheckTransform()
LAVVideo.ax(tid 670) 58081 : ::BreakConnect
LAVVideo.ax(tid 670) 58081 : Trying to connect Pins :
LAVVideo.ax(tid 670) 58081 : <XForm Out>
LAVVideo.ax(tid 670) 58081 : <Input>
LAVVideo.ax(tid 670) 58081 : Trying media type:
LAVVideo.ax(tid 670) 58081 : major type: MEDIATYPE_Video
LAVVideo.ax(tid 670) 58081 : sub type : MEDIASUBTYPE_NV12
LAVVideo.ax(tid 670) 58081 : ::CheckTransform()
LAVVideo.ax(tid 670) 58081 : ::CheckTransform()
LAVVideo.ax(tid 670) 58081 : SetMediaType -- out
LAVVideo.ax(tid 670) 58082 : ::DecideBufferSize()
LAVVideo.ax(tid 670) 58082 : Connection succeeded
LAVVideo.ax(tid 2cc) 58086 : ::NewSegment - 0 / 0
LAVVideo.ax(tid b60) 58087 : h264RandomAccess::parseForRecoveryPoint(): Found IDR slice
LAVVideo.ax(tid b60) 58150 : ::ReconnectOutput(): Performing reconnect
LAVVideo.ax(tid 670) 103342 : ::BeginFlush
LAVVideo.ax(tid 670) 103343 : ::EndFlush
LAVVideo.ax(tid 670) 103344 : ::BreakConnect
LAVVideo.ax(tid 670) 103344 : Shutting down ffmpeg...
LAVVideo.ax(tid 670) 103384 : ::BreakConnect
LAVVideo.ax(tid 670) 103386 : Shutting down ffmpeg...
LAVVideo.ax(tid 670) 103410 : CTransformOutputPin::~CTransformOutputPin

[nevcairiel]

Are you sure it just stops, and doesn't just take *very long*? (which could happen if the process is just very slow)

[fairchild]

DXVA2 version for me is similair to previous poster, it loads runs like a slideshow on the test file. I close the file as it is probably going to take a loooooong time to complete. It won't complete graphstudio, i mean it probably does but I just stopped it manually as well.

LAVVideo.ax(tid c98) 1 : DXVAVideoProcessor::InitializeD3D9(): Initializing D3DEx Interfaces
LAVVideo.ax(tid c98) 18 : -> Creating D3D9Ex Device
LAVVideo.ax(tid c98) 54 : DXVAVideoProcessor::InitializeDXVA2(): Initializing DXVA2 Interfaces
LAVVideo.ax(tid c98) 54 : -> Creating IDirectXVideoProcessorService
LAVVideo.ax(tid c98) 54 : -> Querying Processor Device GUIDs
LAVVideo.ax(tid c98) 54 : -> Processor 0 with GUID {3C5323C1-6FB7-44F5-9081-056BF2EE449D}
LAVVideo.ax(tid c98) 54 : -> Target 0: 22
LAVVideo.ax(tid c98) 54 : -> Target 1: NV12
LAVVideo.ax(tid c98) 54 : -> Processor does support NV12
LAVVideo.ax(tid c98) 54 : -> Querying Processor capabilities
LAVVideo.ax(tid c98) 54 : -> DeviceCaps: 1
LAVVideo.ax(tid c98) 54 : -> InputPool: 4
LAVVideo.ax(tid c98) 54 : -> NumForwardRefSamples: 0
LAVVideo.ax(tid c98) 54 : -> NumBackwardRefSamples: 2
LAVVideo.ax(tid c98) 54 : -> DeinterlaceTechnology: 64
LAVVideo.ax(tid c98) 55 : -> Querying ProcAmp values
LAVVideo.ax(tid c98) 55 : -> Querying Filter values
LAVVideo.ax(tid c98) 55 : -> Creating surfaces
LAVVideo.ax(tid c98) 55 : -> Creating Video Processor
LAVVideo.ax(tid c98) 55 : -> Creating Processor and Surfaces successfull
LAVVideo.ax(tid 1080) 55 : CTransformInputPin::CTransformInputPin
LAVVideo.ax(tid 1080) 55 : CTransformOutputPin::CTransformOutputPin
LAVVideo.ax(tid 1080) 55 : SetMediaType -- in
LAVVideo.ax(tid 1080) 55 : Shutting down ffmpeg...
LAVVideo.ax(tid 1080) 55 : Initializing ffmpeg for codec 28
LAVVideo.ax(tid 1080) 55 : -> File extension: .mkv
LAVVideo.ax(tid 1080) 55 : -> Processing extradata of 27 bytes
LAVVideo.ax(tid 1080) 57 : -> ffmpeg codec opened successfully (ret: 0)
LAVVideo.ax(tid 1080) 57 : ::BreakConnect
LAVVideo.ax(tid 1080) 57 : Shutting down ffmpeg...
LAVVideo.ax(tid c98) 62 : Shutting down ffmpeg...
LAVVideo.ax(tid c98) 64 : CTransformOutputPin::~CTransformOutputPin

[nevcairiel]

I guess the rumors are true, and ATI is really abysmally slow on memory copys.

Fun fact: Intels IGP supports DXVA-HD, at least partly. It supports NV12 input, but only YUY2 or RGB output.
Its somewhat slow, but still watchable.

[Qaq]

I wonder if DXVA-HD check with DXVAChecker is valid. Can't check myself cause I stick with XP these days.

[nevcairiel]

I wonder if DXVA-HD check with DXVAChecker is valid. Can't check myself cause I stick with XP these days.

I think it actually is.

[noee]

Athlon II @ 2.2Ghz w/HD6570, Win7 x64 Ult

HD: 77fps
DXVA2: stops at the end of the first pass, no stats, same as the failure I listed above for the 6520.

Yeah, I'm pretty sure it just stops, I waited a couple of minutes, it says :18/:18 and then it just sits there, GraphStudio is responsive at that point. Not sure what to make of that.

[fairchild]

Does this mean there is no hope for ATI HW deinterlacing with MadVR? For now the workaround for me is to use EVR.CP when I want to watch an interlaced video (which thankfully for me is a minor thing). I could alternatively run ffdshow raw video filter + yadif SW deinterlacing but I'd rather keep my filter chain as direct as possible. (LAV splitter - Lav Video - MadVR/EVR)

[nevcairiel]

Does this mean there is no hope for ATI HW deinterlacing with MadVR?

I dunno. Inside a renderer, there are other possibilities. AFAIK, only the download of the image is so super slow, which could be avoided if you jsut use that surface to render it.

Maybe there also are some tricks to make it faster, but i don't have any ATI hardware to test things on.

The odd thing is, somehow VLCs DXVA works on ATI/AMD as well, and they copy the image back from a D3D surface as well, so there must be some trick.

[VipZ]

Same for me, with DXVA2 its pretty much a slide show.

Just hope the AMD HD 7xxx will solve this.

[SEt]

I heard you can StretchRect to RGB render target and then GetRenderTargetData on ATI, but having to convert to RGB doesn't impress me.

[ForceX]

Not that I care much for interlaced content, or think the aged hardware can handle it, but both my laptop's integrated Intel and nVidia graphics are failing with same kind of error:

Intel GMA HD http://tinypaste.com/32bd1
nVidia 310M http://tinypaste.com/59667

[nevcairiel]

Not that I care much for interlaced content, or think the aged hardware can handle it, but both my laptop's integrated Intel and nVidia graphics are failing with same kind of error:

Intel GMA HD http://tinypaste.com/32bd1
nVidia 310M http://tinypaste.com/59667

Sounds like whatever renderer you're using doesnt support NV12. :)

Anyway, the main point of info i got today was that ATI/AMD is still too slow in this area, and sadly doesn't even support DXVA-HD at all.
Personally, i use NVIDIA in all my systems, so i don't care that much of a great deal. That Intel supports DXVA-HD is nice, although its also too slow (NVIDIA 200 fps, Intel 30 fps) - but it leaves me hopeful for future Intel drivers/IGPs

Since i want to move away from hardware decoding due to some limitations and annoyances of the black-box decoders, i'll continue this idea, and see where it brings me. LAV Video + DXVA-HD processing to deinterlace .. sounds like a good plan!

[Xaurus]

Since i want to move away from hardware decoding due to some limitations and annoyances of the black-box decoders, i'll continue this idea, and see where it brings me. LAV Video + DXVA-HD processing to deinterlace .. sounds like a good plan!
Nev, do you have any idea why I only got ~93 fps with the test build? I have the same gfx card as you do.

[Superb]

Nev, do you have any idea why I only got ~93 fps with the test build? I have the same gfx card as you do.Because LAV Video uses the CPU when decoding. The GPU only affects how many fps you loose when using the DXVA-HD API...
You probably have a slower CPU.

EDIT: nvmind. missed your CPU image. nevcairiel's comment cleared it.

[nevcairiel]

Because LAV Video uses the CPU when decoding. The GPU only affects how many fps you loose when using the DXVA-HD API...
You probably have a slower CPU.

Thats not it, his CPU image also shows ~240 fps.

Its probably some setting in the NVIDIA control panel, like forcing a high level of AA system-wide, or enabling things like noise reduction/edge enhancement.
Its also running in the "Playback" preset, which means it doesn't go up into the next performance mode as long as its fast enough for the FPS i gave it.