[Intel-gfx] [RFC] execbuf2 support to avoid fence register allocation

Chris Wilson chris at chris-wilson.co.uk
Thu Jun 18 10:04:26 CEST 2009

On Wed, 2009-06-17 at 22:33 +0100, Chris Wilson wrote:
> (poppler-alt-20090608)
>      drm-rgba-fenced	 764675.49:  1.31x ▎
>      drm-rgba-no-fences	 994886.76:  1.00x 
>      drm-rgba-no-fences2 998345.74:  1.00x

Hmm, I've just had a run (under sysprof) of 776.001s for execbuffer2
(i.e the same speed as execbuffer1). Comparing a few profiles the
difference would appear to be time spent within slob_alloc(). So at
first glance not directly related to the i915 driver -- but it may be
something that we can mitigate by managing our batches and buffers more

/me continues to pine for better performance analysis tools.

More information about the Intel-gfx mailing list