[Glamor] glamor: Changes to 'master'

[Chris Wilson]

On Thu, 12 Jul 2012 11:45:03 +0200, Michel Dänzer <michel at daenzer.net> wrote:
> From my experience with EXA drivers, a good way to achieve that is to
> allocate a VBO large enough to keep the VBO allocation overhead low and
> to keep appending vertices to it until they no longer fit, at which
> point the next VBO needs to be allocated.

Agreed, that's how vertex data should be streamed to the GPU (and what I
patched dynamic draw on intel to do). That the unsynchronized write
buffer doesn't perform as well on mesa/i965 indicates a problem in the
driver, not in glamor.

DYNAMIC_DRAW 4.25 Mglyphs/s
DYNAMIC_DRAW (patched) 4.62 Mglyphs/s
STREAM_DRAW 4.21 Mglyphs/s

> The patch below implements something like that. The GL API usage and VBO
> size could probably use some more tuning, but this already performs
> better for me than reverting commit
> 593bd206bf2794dc1450e8e7d20142b8d0ca074b or disabling VBO usage with
> your patch. How does it work for you?

Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre