On Sunday 30 January 2005 07:10, you wrote:
> Mind blowing stuff Nicolai!  I'll try it out hopefully tomorrow.
> 
> On Saturday 29 January 2005 18:20, Nicolai Haehnle wrote:
> > 4. Textures should work, but there's no way to upload them and I
> > haven't tried render to texture yet. Shared memory support for the
> > IPC connection has a high priority on my todo list.
> 
> I'd just write them over another socket to simulate DMA, it's more like 
> the actual hardware anyway (unless I missed something and the board is 
> going to map host memory now).

Maybe I'm mistaken about the way that DMA works, but my rationale was this: 
The CPU tells the graphics card to do a DMA transfer, but it's the graphics 
card that actually initiates the DMA by asking "send me data from this 
address" over the bus. Now simulating this over another socket would 
involve sending packets from the simulator to the client, with the client 
having to respond. This complicates the client design - in fact, the client 
would probably have to be multithreaded in order to model the asynchronous 
nature of DMA transfers.
Using shared memory simplifies all this, and the fact that the graphics card 
can't access host memory directly can be modelled using accessor functions. 
In fact, I've already changed the software model to use calls like 
READ_DWORD(addr) and so on to access video memory (and only video memory).

Another factor is that the CPU is able to map video memory, so the simulator 
has to use shared memory in at least one direction anyway.

cu,
Nicolai

> Regards,
> 
> Daniel

Attachment: pgptHM65amo7S.pgp
Description: PGP signature

_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to