ATI IMAGEN (w100) Specifications

[mteira]

Hello.
One of the reasons that avoids me to upgrade to OpenZaurus my C760, is the lack of a driver for the 2D/Video features of the graphic chipset on our Zaurus machines.
ATI is not releasing specifications for that chip, and the only QPE implementation that is using the 2D hardware features (bitblt, fillrect,...) of the chip is the libqpe.so bundled with the Sharp ROMS (anyway, it's a buggy implementation, and crashes using some qte operations).

I think that those specifications are a very important step to make Openzaurus really usable (at least for people running emulators and games). I also think that perhaps was a good idea that all the OZ community represented by a leader could make a petition for that information to be released (perhaps ATI gives more credibility to such a petition that one from a single person).

My english (as you can read) is pathetic. So, we'll need that someone with more english skills (that's an easy to reach requirement) will write a petition, informing about the advantages for ATI if the sources were released (better driver implementation, good publicity about their chip,...)

Then, after the letter was approved, someone could send it to ATI.

What do you think?

[Mickeyl]

Agreed. Opie, GPE, Evas, and friends will forever stay slow as molasses on the clamshell models until we 1) get the specs and 2) write accellerated drivers.

If the community is able to compile a convincing document stating our needs and the potential benefits for both sides, then I will sign this too and send it to the relevant people @ ATI.

[_Psycho]

The page on the reverse engineering of the chip:

http://www.mnementh.co.uk/eseries/docs/imageon/

[lardman]

So what happens at the moment?

Is the existing module (from the Sharp ROM) used and linked against (their) libqte? Not pretty and it taints, but presumably it works (this is what happens with pdaXrom I think: https://www.oesf.org/forums/inde...showtopic=6350).

According to this post: http://forums.projectmayo.com/viewtopic.ph...c=6562&forum=23 (linked from the post I've linked above) - once upon a time ATi had an SDK for the chip in question. It may therefore be possible to obtain some specs - but through an NDA from the looks of it. On the other hand this post also talks about the chip being out-dated now, so ATi may be more receptive to the idea of releasing something with a less restrictive disclosure policy.


Si

[Mickeyl]

Using the accellerated qte from the SharpRom works for pdaXrom - although this library itself is buggy (search for more comments from mteira).

For us though, it wouldn't even work if we would consider the option to include this hack, because we are building everything with soft-float which means no chance to link against this library.

[mteira]

AFAIK that page that Psycho linked is not based in reverse engineering, but in observation of the kernel driver for the w100 and other ATI cards with programming similities. Unfortunately, it's far from complete (no bitblt examples nor specifications for example) and the author is missing (I asked for it to the page maintainer and he didn't hear nothing from him for a while).

On the other side, we have to difference two things.
1.-The real access to the card, or the real API that ATI provided (with that SDK I suppose) is a bunch of C functions embedded in the libqte shipped with the Sharp ROM. Those starting with AtiCore (you can see them using   objdump -T libqte.so | grep Ati  )
2.-What I found to be buggy is the libqte.so itself. I don't know if it's a bad usage of the AtiCore API  or another mistake. Perhaps we will never know. The problem (there's an old post about it) is trying to render QPixmaps generated from a QImage tranformed using a QWMatrix.

What people from pdaxrom make is linking this libqte against their Xserver (and using directly AtiCore based on tracing the use of the functions on the original libqte). I think that Sashz was trying to use a dll with the Ati Core (with a wine dll loader, I think) to avoid loading all the libqte stuff.

Could the AtiCore bunch of functions or one of this dll be used on the Openzaurus stuff? Or will it be not legal?
I don't understand that technical limitation of soft-float linking, Mickeyl?

[Mickeyl]

Could the AtiCore bunch of functions or one of this dll be used on the Openzaurus stuff? Or will it be not legal?
I don't understand that technical limitation of soft-float linking, Mickeyl?

Whether it is legal or not is nothing to worry about, because it doesn't work either way. Because of the way the parameters are passed when calling functions, soft-float compiled programs can't link to hard-float compiled libraries.

All further OpenZaurus and familiar linux distributions will be built using soft-float, because it gives us a speedup of up to 300% when using floating point operations.

[Laze]

We have contacted ATI many times but no reply - the next version of pdaXrom will have further support for ATI hardware accel. - but basically its not "good" because of the hacking done to get it working. :-) We are hoping that ATI will get back anytime soon (we have phoned, e-mailed and faxed).

[raduga]

- the next version of pdaXrom will have further support for ATI hardware accel. - but basically its not "good" because of the hacking done to get it working. :-)

Hm....
do you mean, the next version after 1.1.0?  
Or 1.1.0 final release?
Or do the Release Candidates for 1.1 have the ATI code already?

anyway thanks

But beware of visiting the USA (remember Sklyarov...)

[mteira]

He said "further support". Nowadays they have support for bitblt and some other operations. Anyway, with the specifications, I think that the way to go is upgrading the kernel w100 driver, perhaps with some IOCTL operations to perform all the required primitives. This way, the driver would be usable for every ROM/distribution/flavor of X/QT/text/whatever system we want to develop.

So, anybody with good english skills could be so kind to redact a preliminary version for our "letter to ATI". Once this draft is made, we could contribute with ideas, for example, benefits for ATI if they open the code (a better driver, for sure).

[sashz]

we use hacked aticore in X server, which moved from libqte to shared library (disassembled->restored as ASM sources->compiled as shared/static library), and have size only 96KB (if use it with libqte the size is much more).

[Mickeyl]

That sounds interesting. As always, well done sashz! 96K is something we can cope with. Please forward your work to me when you have a chance to.