Re: memory manager interface ...

2005-10-26 Thread Benjamin Herrenschmidt
On Sun, 2005-10-23 at 11:03 +0100, Dave Airlie wrote: > > > > So, something like Get/Set/UnsetAttribute() perhaps. > > > > But also, thinking about the use of these regions for DMA, synchronization > > is > > important. I saw the SetFence() entrypoint, there's no way to retire these > > fences, e

Re: memory manager interface ...

2005-10-24 Thread Dave Airlie
> > I guess I don't see a problem with putting it in libdrm. I had just > assumed that the user-mode portion would end up living in > src/mesa/driver/dri/common or some place similar. There are some > advantages to putting it in libdrm, though (e.g., easier for EXA to use it). Well the X server

Re: memory manager interface ...

2005-10-24 Thread Ian Romanick
/common or some place similar. There are some advantages to putting it in libdrm, though (e.g., easier for EXA to use it). > So I've modified > http://dri.freedesktop.org/wiki/DriMemoryManagerDesign > and added a Memory Manager Interface section with some initial thoughts, > in th

Re: memory manager interface ...

2005-10-23 Thread Dave Airlie
> > So, something like Get/Set/UnsetAttribute() perhaps. > > But also, thinking about the use of these regions for DMA, synchronization is > important. I saw the SetFence() entrypoint, there's no way to retire these > fences, eg a FinishFence(). > > You've probably seen this one: > > http://oss.s

Re: memory manager interface ...

2005-10-23 Thread Keith Whitwell
atter if it can be hidden behind the libdrm interface.. So I've modified http://dri.freedesktop.org/wiki/DriMemoryManagerDesign and added a Memory Manager Interface section with some initial thoughts, in the hope that maybe some discussion occurs on it, in theory we should be able to implement

memory manager interface ...

2005-10-23 Thread Dave Airlie
hidden behind the libdrm interface.. So I've modified http://dri.freedesktop.org/wiki/DriMemoryManagerDesign and added a Memory Manager Interface section with some initial thoughts, in the hope that maybe some discussion occurs on it, in theory we should be able to implement the current memor