[cairo] [PATCH] A set of NEON blitters for Pixman.
sandmann at daimi.au.dk
Thu Jun 4 22:21:57 PDT 2009
Siarhei Siamashka <siarhei.siamashka at gmail.com> writes:
> But realistically, neon optimizations have quite a number of alternative
> code paths and branching logic. I would not be surprised if it has some other
> bugs ;-) Seems like your tests are not good enough, so they could not detect
> the problem in fill function.
> I'm repeating myself, but I still think that tests with good
> coverage for neon optimizations are needed, and they need to end up
> in pixman git repository. So that any modifications to this code
> could be tested at any time easily.
Now that we have the concept of 'implementation', it would be
relatively straightforward to add support for enabling and disabling
fast paths at runtime, which would make it possible to compare how the fast
paths behave on random compositing operations versus the general
implementation. If they disagree, at least one of them is wrong.
More information about the cairo