[Previous replay get sent prematurely] > > I guess you could probably invent a simple DMA interface and feed ogsim > > commands over that. I wouldn't like to say whether that'll end up > > anything like the final hardware, but it might give you enough to take a > > stab at writing the rest of the 3D driver. > > I intend to give the 3D engine a PCI BAR, where every engine register > has an address. The engine will be complex, and so will DMA, so it > will be important for us to be able to eliminate variables whe > debugging. Thus, the first driver should be all PIO. Indeed, > anything that isn't performance critical that needs to use the GPU > would do well to use the PIO interface, rather than going through the > trouble to allocate DMA'able memory space, set up the DMA controller, > etc., etc.
PIO regs are nice for debugging and poking at the card state from within a running machine. However I'm not so sure about actually using them in anything approaching a real driver. I guess with appropriate abstraction the driver may be able to use either. In practice it may be easier to do everything with a fire and forget DMA command queue. Any time you use PIO you're probably going to have to make sure the DMA command queue is flushed and the card quiescent before you start poking at state. For linux at least you've already got the DRM/DRI/Mesa kernel/userspace split, so I wouldn't expect PIO to gain you anything by way of simplicitly. If you want to do anything interesting you're already going to have the DMA queues configured. Paul _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
