S3TC => r600{c,g}

[adamk]

I've only tried a few apps that use S3TC on r300... Neverwinter Nights, doom3, ut2004, and quake4 come to mind. Of those four, S3TC only works properly with NWN (last I tried).

Q4 won't even start without S3TC, and using the external lib on r300 with Q4 screws up the textures so bad the game is unplayable (as is the case with ut2004 and doom3). At least UT2004 and Doom3 will run without S3TC as it's not a hard requirement on those games.

So, here, s3tc works on r300g with one out of four apps. Not exactly impressive, and something that desperately needs to be addressed by someone.

Adam

[Nille]

in germany software patents are against the law but microsoft win on the highest Court with there FAT software patents.

But not in Germany.

do you really think there are only 2 countries?

USA and Japan.

[Nille]

Gallium can load an external s3tc library on demand if it's available.

The Mesa Classic driver do the same and if you install the Library the Driver advertise the Extension but in the Application you see only broken Textures.

I have postet an Link to the Bug Report. In this bug are some Screenshots.

[RavFX]

You don't need to do anything to decompress DXT textures as the card can handle them directly. However, the card needs to be told it is getting compressed textures. Unfortunately, the register documentation released by AMD doesn't give the format codes for DXT textures (for IP reasons). The r300 driver has support because the format codes were guessed.

Can it be the reason why it's not implemented?

I did notfind the DXT* register number in the AMD documentation. Or I may have searched wrong.

[Qaridarium]

But not in Germany.

you are kidding on me? i SPEAK for germany!

M$ won in GERMANY! in the 3. instance highest german Curt

USA and Japan.

and germany ;-)

thats because software patents are valid in EU if the patent is bound to the hardware.

and the S3TC Patent ist bound to the HARDWARE->Graphic-card!

and the FAT Patent is bound to the HARDWARE (harddrive)


in german language his is the highest normal curt in germany: http://www.heise.de/newsticker/meldu...ht-985124.html

[Nille]

you are kidding on me? i SPEAK for germany!

No, you do not speak for germany

thats because software patents are valid in EU if the patent is bound to the hardware.
and the S3TC Patent ist bound to the HARDWARE->Graphic-card!
and the FAT Patent is bound to the HARDWARE (harddrive)

And if its bound to Hardware its no longer an SW Patent in EU.

[Qaridarium]

No, you do not speak for germany

yes just my bad english i mean i speak for the sitation in germany.

the S3TC patent is valid in germany like the FAT patent in germany.

And if its bound to Hardware its no longer an SW Patent in EU.

the S3TC patent is bount to hardware (graphic card) means its a valid software patent in germany and EU.

if you emulate S3TC in the cpu or in 'shader' units without any special gpu features in use this kind of hack is legal in the EU.

but this slow's the usage of S3TC down very hard.

[whizse]

I've only tried a few apps that use S3TC on r300... Neverwinter Nights, doom3, ut2004, and quake4 come to mind. Of those four, S3TC only works properly with NWN (last I tried).

Q4 won't even start without S3TC, and using the external lib on r300 with Q4 screws up the textures so bad the game is unplayable (as is the case with ut2004 and doom3). At least UT2004 and Doom3 will run without S3TC as it's not a hard requirement on those games.

So, here, s3tc works on r300g with one out of four apps. Not exactly impressive, and something that desperately needs to be addressed by someone.

Adam

Interesting, except for NWN I have played all those games with S3TC enabled on r300g, and I never noticed anything screwed up.

Is there a bug report open for this?

[adamk]

Honestly, I'm not aware of a bug report specific to s3tc on r300g. I mostly use FreeBSD, but next time I'm in Fedora, I'll give it a shot again and see if anything has improved.

[whizse]

Can it be the reason why it's not implemented?

I did notfind the DXT* register number in the AMD documentation. Or I may have searched wrong.

Maybe you can ping bridgman about this?

There was also some discussion and code written earlier this year, which sort of language lawyered around part of the problem, but I don't think it was ever merged? Not sure if more work was required or somebody silently vetoed the whole idea:
http://sourceforge.net/mailarchive/m...mail.gmail.com