On Friday 18 March 2005 05:29, Attila Kinali wrote: > On Fri, 18 Mar 2005 10:29:02 +0100 (CET) > > [EMAIL PROTECTED] wrote: > > You can't do that. It could means : software filtering that is slow, > > and/or kernel system call that is also slow. > > > > Why do you need that ? If you provided hardware access to a safe > > interface of the card, there is no problem. (this interface context > > should be saved and restored by the kernel when a task switch occur, > > the card could also manage "few" context (3/4, the opengl programme and > > the windows manager at least)) > > You don't want to access the graphics card on every task switch. > On a current linux system you have something between 100 and 2000 > context switches per second. If you do something like this, then > the driver has to intercept every access and check whether > the registers need to be swapped.
Actually, this is something I'd like to think about for the future. I would like to see the GPU supported by the task switch just like a numeric co-processor. This is a lazy switch: a task that isn't using the co-processor doesn't pay any penalty. There are significant synchronization advantages to doing things that way. On the other hand, it is a fairly deep kernel hack, and we have other things to worry about. Plus, we really ought to have some functioning hardware first. Regards, Daniel _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
