[Dri-devel] Radeon on mesa-3-5-branch

[Gareth H. <ga...@va...>]

I'm guessing the plan is to wait for the context-private SAREA before we
switch from putting the elts directly into the DMA buffer, right?  If
not, what's the plan to address this?  If you guys don't mind, I would
like to move to an interim copy solution sooner rather than later, as it
will affect the way DMA buffers are handled (specifically, vertices will
be the only things that go directly into them, and they will go from low
to high in all cases).

My tnl module is just using verts for everything at the moment, but we
can get the same kind of improvement as we see from the render-to-dma
stage for quads by using indexed primitives instead.

While I'm working with this code, I might as well do it if there are no
objections.

-- Gareth