GMA500 BSOD after calling to IAMVIdeoAccelerator::BeginFrame

[tal_aloni]

Hello All,

Myname is Tal Aloni, I'm a member of ffdshow tryouts development team (open source video decoder).

we implemented aDXVA decoder that has been tested on dozens of different hardware video accelerators (including Intel G45), but our userswho own GMA500 and use Windows XP report that they get a BSOD when trying to utilize DXVA.

I've been working with one of the GMA500 owners (Asus 1201HA withZ520 and GMA500), he has a dual boot Windows XP / Windows 7 setup, he can utilize DXVA under Windows 7 successfully, but gets BSOD under XP.

he testeddifferent driver versions, including the latestOEM drivers and intel driver 3.3.0 dated 12/18/2009.

debugging revealed that the call that triggers the crash is IAMVIdeoAccelerator::BeginFrame (the first call to this method, previous initialization calls to IAMVIdeoAccelerator succeeded)

(which obviously only executes under Windows XP)

here is the minidump:

http://iknowu.net/files/public/ffdshow/GMA500/Mini022810-05.dmp

*** STOP: 0x0000008E (0xC0000005,0xBF047911,0xBA235BF4,0x00000000)

*** igxpdd32.dll - Address BF047911 base at BF01E000, DateStamp 4a438636

you can recreate this BSOD by trying to playH.264 L4.1 content with ffdshow DXVA decoder (the latest is rev. 3295 at this point)

source code and builds is available at sourceforge ( http://sourceforge.net/projects/ffdshow-tryout/ ).

I would be happy to provide more information if necessary,

Tal Aloni

[Sankar_V]

Hi Tal,

Can you please point us to the specific clip that you've tried? Or is it happening with any H.264 L4.1 clip? Which demuxer/parser does the player (which player) use when it crashes?

Please let us know these details.

Regards,

Sankaranarayanan V

[tal_aloni]

Hi Sankaranarayanan, Thanks for replying.

1. it is happening with every H.264 L4.1 clip we tested.

if you need a sample, here is one:

http://iknowu.net/files/Andy.Barker.Sample.mkv

2. demuxers tested are latest haali media splitter and MPC-HC matroska splitter.

3. players tested are MPC-HC, GraphEdit and GraphStudio.

we even tested video only streams.

it's not related to a specific player / demuxer / H.264 file.

Tal

[jhb50]

My thanks too Sankaranarayanan V.

I am the originator of the bug as reported as #233 on Sourceforge MPC-HC project and confirmed by another GMA500 XP user. I have run the tests under Tal's guidance on this hardware to trace the point of failure which has been consistent.

I am available to run and report any tests on this hardware thatyou may require.

[tal_aloni]

as you may know, it's possible to get DXVA 2.0 working under Win XP for VC-1 decoding. (.Net Framework 3 installs the necessary components)

at this mode of operation, DXVA 2.0 calls are translated by the microsoft components to DXVA 1.0 driver calls.

since we know our DXVA 2.0 code works well under Win 7, we assumed that if that issue was a result of DXVA 1.0 bug in our code, the translated driver calls (which are not done by us) would result in perfect playback, just like on windows 7.

however, we received error 0x80004005 (unexpected error) when calling IDirectXVideoDecoder::BeginFrame for the decoding of the first frame.

I think it confirms that this is an issue with the current drivers.

Tal

[Sankar_V]

Hi Tal and jhb50,

I am able to reproduce the H.264 issue. I will analyze and let you know.

Sankar

[Sankar_V]

Tal,

Is this a separate issue? Or, are you reporting another issue with VC-1?

BTW, the clip that you shared is a H.264 L5.1 and not L4.1.

Thanks.

Sankar

[tal_aloni]

Sankar,

I'm glad you're able to reproduce the issue.

the H.264 sample indeed have L5.1 header, but it complies with all L4.1 specs (3 ref frames are used, you can examine the encoding parameters with MediaInfo).

the VC-1 issue is not a seperate issue, it's the same driver issue with windows XP.

Thanks,

Tal

[Sankar_V]

So, basically what you are saying is that if you use any DXVA2 codec in Windows XP in GMA500, it crashes??

Sankar

[tal_aloni]

decoding of any DXVA compliant content (H264 L4.1 / VC-1) using DXVA 1.0 under XP will cause a BSOD.

we utilized DXVA 2.0 under XP only to demonstrate that it's a driver issue and nothing is wrong with our DXVA 1.0 code.

[jhb50]

Yes, I confirm that my issue is the inability to utilize DXVA with the GMA500in playing H.264 videos under XP because it always BSOD's with an error message identifying the video driver at fault.

Thisleaves netbook systems, unable to decode H.264 video without artifacts due to their low cpupower, and essentially useless, since the whole purpose of including the GMA500 with these systemswas to provide DXVA support and enable these low power systems to play these videos.

[jhb50]

Sankar wrote:

"I am able to reproduce the H.264 issue. I will analyze and let you know.

Hi Sankar. How is this proceeding? Should we check daily or will this take some time?

There is a lot of interest in this XP support so thank you for your help.

John

[jhb50]

Sankar: The following will confirm that the BSOD playing any H264 video using the GMA500 XP driver is indeed a bug in the driver, and not in the calling code.

Using the IEGD 10.3 driver for the chipset, which is supported on XP, playing the same X264 videos, activates and plays these videos accelerated without any BSOD failure.

You should also be aware that the DXVA performance of the IEGD driver is 8 times! greater than that of the GMA500 XP driver. On the ATOM Z520, my test 1080P 4mbps X264 video plays at 4 frames per second but with the IEGD driver the frame rate jumps 8 times to 24 frames per second at the same CPU loading!

I would expect that Intel would improve the preformance of the GMA500 driver to be at least equivalent to that of the IEGD driver, but at a minimum the BSOD failure would be corrected, since every XP GMA500 user playing HD video with MPC-HC will experience it. Hardly a positive user experience with Intel amongst a rapidly growing netbook segment.

I trust your analysis has/will confirm these facts. Can you confirm that these problems have been reported to the GMA500 driver development team, or advise what different path we need to follow to report them?

Thank you for your help.

John

[Sankar_V]

Hi Tal,

I am sorry - I got into other issues - so I couldn't continue with my debug. Will debug if I get some time. Since, the other codecs are working fine it could be possibly related to the way the surfaces are created and managed.

I don't know the H.264 1080p clip that you are talking about. I am sure we could play it at 24fps mininum in XP also. Can you send me a link to the clip?

Regards,
Sankar

[jhb50]

Sankar, any H264 video will fail with a BSOD when decoded using ffdshow(DXVA) and the XP GMA500 driver.
Here is a link to the test file provided by Tal so that we could all test with the same file.

http://iknowu.net/files/Andy.Barker.Sample.mkv

Note my previous message, that the file will play correctly using the Intel IEGD Driver under XP, which demonstrates that the BSODis an XP GMA 500Driver failure, and not an ffdshow calling problem.

It also demonstrates how poorlythis driver supports X264 decoding compared to the IEGD driver.

Both need to be fixed.

Thanks for you help.

[jhb50]

Any further progress on this Sankar? It really does need to be fixed.

[rambalac]

Thank you for your topic.
I was wondering why i can not play h264 files in MediaPlayerClassic Homecinema, just BSOD with igxpdd32.dll on every h264 file opening on my Kohjinsha PA.
Now, understanding that GMA500 is not able to accelerate video, i just turned acceleration off in MPC options and can watch h264 video somehow.

[jhb50]

Pardon me if I am somewhat confused about the purpose and value of this forum. It is publicised as connecting to Intel developers and engineers and after posting our problem on Mar 1, the Intel person confirmed that the problem was reproduceable. That was great but 6 weeks later we have heard nothing more, despite our requests as to its disposition. There is no point in having this forum if users are just left in limbo. Seems pretty easy to respond that the problem will not be addressed, or has been added to the bug list for the product.

Again a response regarding this significant and reproducable bug would be appreciated.

[adw2ru]

It's looks-like Intel doesn't provide new versions of driver for XP any more. This is just my personal conclusion.

By the way why Intel not take IEGD code and combine it with "normal" driver?

[Doraisamy_G_Intel]

I will email internally to see if I can get a response to this. Thanks.

[vikking20]

Do you have any news on this? Cause this is an extremely annoying bug...