03-14-2009, 01:48 PM
I read in this forum:
and now I've a few more questions:
1. What cards will support OpenCL?
2. What will happen to brook+/cal?
3. How will OpenCL be implemented / will it be based uppon LLVM?
Thanks a lot,
03-14-2009, 02:01 PM
I think that's just a consequence of moving from ATI to AMD web site tools, but I can ask about it.
Is there an actual fetch restriction (ie even when you register you can't get the tools for some other reason) or is it just that you don't want to register ?
03-14-2009, 04:24 PM
03-14-2009, 04:46 PM
OK, thanks; I'll pass that on.
03-16-2009, 04:55 PM
Originally Posted by bridgman
Just for the record: setting the kernel parameter "nomodeset" has solved a number of issues for me:
Google-Earth doesn't work with the open source driver. => solved!
When DPMS mode is enabled and the screen goes to sleep it will not wake up from it anymore. => opengl screensaver issue with kms enabled => solved!
Pretty much everything else I do with this driver is al least 50% slower then with the fglrx driver. => solved!
But the compiz alpha blur thing still isn't working as aspected,
But without the *blur* effects it is much more useable now. Xvideo plays fine now (also windowed) and everything is a lot more "snappier".
Also, I read on the fedoraproject wiki that the radeon open source driver will support kms and dri2 in the upcoming Fedora 11 release for the older r100, r200 and r300 cards.
So, it seems there still is some future for my *old* hardware here.
03-16-2009, 05:02 PM
Originally Posted by mr_ed
this also is solved for me setting the *nomodeset* parameter.
03-17-2009, 02:04 PM
There is microcode (a "binary blob") in the Radeon driver in the Linux kernel. Because of this, according to the Free Software Definition, the open source Radeon drivers are non-free software.
Has the information about what this microcode does been released by ATI? Is there anything stopping ATI from releasing it?
The microcode appears to only be used to initialize the Radeon command processor. With proper documentation for the microcode, the open source Radeon drivers could be 100% free software. This would allow free software users the opportunity to use many powerful Radeon video cards for 3D graphics in GNU/Linux.
Thank you so much!
03-17-2009, 03:21 PM
This is one of the few places where I suspect the Free Software Definition is being misapplied, or at least being used in a way which brings unintended consequences. I can't believe that the definition was written to deliberately penalize hardware vendors who load microcode in the driver rather than autoloading it from ROM or permanently storing the microcode on-chip, but that is what is happening right now.
We are documenting all of the programming information required to write 100% free, highly functional and performant drivers. We are not documenting the internal functioning of the hardware, and the microcode images are part of that internal hardware function. No GPU vendor allows modification of the internal microcode, whether it be stored permanently on the chip or loaded at power-up. If I was being told that all microcode was evil and that you were not going to use any chips which run internal microcode unless you had source for that microcode I could understand, but that is not the case here.
The argument, as I understand it, is "since your microcode image is loaded at power-up by the driver then the microcode needs to meet all the rules we apply to the rest of the driver code, even though it is not driver code, is never executed by the CPU, and once it is loaded it is NO DIFFERENT from microcode which had been permanently stored in the chip or autoloaded from ROM".
Honestly I don't see why a microcode image loaded at power-up is any less free than the same microcode stored permanently in the chip. Neither one can be read or understood, neither one allows learning or changing or redistributing the changes, yet apparently one is Good and one is Evil. I can't believe that was Mr. Stallman's intention.
AFAIK most of the older Radeon GPUs can use the 3D engine *without* the command processor, albeit more slowly, since there is a command FIFO which can accept command packets without requiring that the microcode images be loaded. I have mentioned this to a couple of developers but so far there hasn't been much interest. Only the 6xx-and-up parts absolutely require the command processor microcde for 3D.
Last edited by bridgman; 03-17-2009 at 04:31 PM.
03-17-2009, 04:23 PM
also - gpu's aren't the only microcode users. If you are anal about that, there are a lot more drivers you couldn't use. SCSI, gb-ethernet, 10gb-ethernet and a couple of others ...
03-17-2009, 04:42 PM
Bridgman, when ATI says that it is dropping support for R300-R500 support in Catalyst does that include the windows side of the driver?