[Pixman] [PATCH 2/4] Added fast path for "pad" type repeats

Ben Avison bavison at riscosopen.org
Wed Feb 13 06:01:43 PST 2013

On Wed, 13 Feb 2013 07:37:06 -0000, Søren Sandmann <sandmann at cs.au.dk> wrote:
> For this one, did you try comparing Chris' patch to your on ARMv6?

Not previously, but I just checked. These are the times for
t-firefox-chalkboard on ARMv6:

Current HEAD:  35.5 s
Chris's patch:  9.7 s
My patch:       6.6 s

They're not necessarily mutually exclusive, as mine won't accelerate
operations where the source and mask need to be padded differently.

Since there haven't been any comments on the way in which I implemented
fallback to general_composite_rect(), there's just the minor issue of
updating .gitignore before you do push it - I'll reissue this patch with
that in.

> Also, did we ever find out whether it was
> a bug in Firefox. I'm still somewhat skeptical that it's intended for a
> PAD image to be accessed so far out of bounds.

No idea, though I find it hard to imagine that the single-pixel-wide
images weren't deliberately being extended a long way, as they're not
very useful otherwise. Obviously they could use tile repeat for such
images and it would have the same result, though it would boil down to
src_8888_8888 calls rather than src_n_8888 and so I expect that would
be slower.


More information about the Pixman mailing list