We Have Poulsbo Hardware, But No Driver

[Kano]

Did somebody notice this:

http://www.fit-pc2.com/forum/viewtopic.php?f=43&t=1150

It seems the kernel is patched with a new iegd_mod module there, so only the driver needs to be installed. mplayer vaapi seems to be the state from the last official snapshot - i would compile it with my script however for better m2ts support.

[Svartalf]

didn't they state that this PowerVR crapstickle will only have a Closed Source 3D driver, ever?

We shouldn't waste time on this piece of junk, it's even slower than GMA950 anyway.

How would you know? We've no drivers for us- and it's been known that the drivers are marginal at best on the Windows side for the GPU.

If it were slower than the GMA950, Caster3D for the iPhone wouldn't have happened. Right now, it's possible on it and hopefully shortly on other ARM Linux based platforms using the same class of GPU on a similar screen size. In comparison, the GMA950 can't QUITE keep up with the game's rendering requests on my eeePC 701- which is roughly in the same CPU class as my beagleboard and the iPhone actually has less oomph.

[AdamW]

kano: that sounds like they've just built the IEGD driver, which you can get at http://edc.intel.com/Software/Downloads/IEGD/#download . I don't use that one because it only supports one of the Poulsbo PCI IDs, the other one isn't supported by that driver.

[gbeauche]

We shouldn't waste time on this piece of junk, it's even slower than GMA950 anyway.

In my experience, GMA500 performed better than GMA950 on OpenGL. Michael would probably want to kick some PTS at it comparing GMA950 and GMA500 with PSB and IEGD drivers?

kano: that sounds like they've just built the IEGD driver, which you can get at http://edc.intel.com/Software/Downloads/IEGD/#download . I don't use that one because it only supports one of the Poulsbo PCI IDs, the other one isn't supported by that driver.

US15L is indeed not supported by the IEGD driver but who is insane enough to use that chip in a product as it has the same TDP as an US15W an lower video decoding capabilities?

[Kano]

http://ark.intel.com/Compare.aspx?ids=35447,35444,

Well it seems the L allows HDMI + VGA output but the W does not? Also usually it is not that critical to add some extra pci ids to a generic driver. No idea if the IEGD driver is generic.

@gbeauche

Could you compare both drivers for vaapi support? Do they render correctly - even the problematic files for xvba? Need/Can use gl(:reflect) output? Whats the output of vainfo? Some infos about stability for 2d, 3d and video.

[gbeauche]

[url]Could you compare both drivers for vaapi support? Do they render correctly - even the problematic files for xvba? Need/Can use gl(:reflect) output? Whats the output of vainfo? Some infos about stability for 2d, 3d and video.

They all work correctly on the Poulsbo. It does also decode MPEG-4 Part-2, though only up to 720p contents. The rest is 1080p capable, including Blu-Ray dumps and L5.1 clips.

However, IEGD does not implement subpictures yet (OSD) and PSB driver has bugs for rendering RGBA and IA44 formats. IEGD driver does not support rendering to a pixmap, thus GLX texture-from-pixmap extension can't be used to render to an OpenGL texture. However, the Moorestown driver can but the relevant code was not backported to the Poulsbo branch yet, and I doubt it will.

The Poulsbo is a very good chip for video decoding. The VXD370 ASIC is below 200 mW. Sure, there is no Open Source driver, but the closed source one works well.

VAProfileMPEG2Main : VAEntrypointVLD
VAProfileMPEG2Main : VAEntrypointMoComp
VAProfileMPEG4Simple : VAEntrypointVLD
VAProfileMPEG4AdvancedSimple : VAEntrypointVLD
VAProfileH264Baseline : VAEntrypointVLD
VAProfileH264Main : VAEntrypointVLD
VAProfileH264High : VAEntrypointVLD
VAProfileVC1Simple : VAEntrypointVLD
VAProfileVC1Main : VAEntrypointVLD
VAProfileVC1Advanced : VAEntrypointVLD

That's MPEG-2, MPEG-4, H.263, H.264, VC-1 (including 'WMV3'). I have very limited access to H.263 contents so the MPlayer support probably still contains bugs.

BTW, a friend of mine reported 2x OpenGL performance with his program on Ubuntu+IEGD than F11+PSB.

[Kano]

H263 is very easy to get. At least the low res examples That's the video part of the 3gp files which many mobile phones record - otherwise ffmpeg is your friend. Working OSD is high requirement for many people however, i hope the driver gets fixed. At least vainfo shows VLD for everything - xvba does not even have got VLD for mpeg2.

[AdamW]

gbeauche: blino has a system with the unsupported chip in it (which is how he found this problem). So I dunno why, but someone does use it . The problem with putting support for it into the driver is it appears to be in the proprietary bit of the driver, so no (easy) way to do it (guess it may be possible to hexedit it in there).

[jbernardo]

Any news on this? Or has Intel abandoned us gma500 notebook owners?
The least they could do would be to act like nvidia, release binary drivers, so we could still use our netbooks with the latest distros.

[jonwil]

ok, question: Why hasn't someone attempted to reverse engineer the binary drivers and the hardware ala the NOVEAU project for NVIDIA?
Coming up with reverse engineered specs and code for PowerVR chips would help out with a LOT of different gadgets (depending on how similar the different PowerVR cores actually are)

Has anyone ever attempted such a thing? Has PowerVR ever said something about it (e.g. legal threats?)