[Pixman] [PATCH 0/3] ARM: NEON optimizations for bilinear scaling with A8 mask for operator src, over, add

[Taekyun Kim]

Hi,

I send new patch set of bilinear scaling scanline functions.

> 1. License/copyright notice header is needed for the newly added file.
> It can be taken from the older NEON assembly file and updated to have
> both Nokia and your (Samsung?) copyright for year 2011.
> 2. Fix indentation problems according to:
> http://cgit.freedesktop.org/pixman/tree/CODING_STYLE?id=pixman-0.21.6#n22
> 3. More descriptive commit messages would be welcome for patches. And
> commit summaries are better to start with 'ARM: '.
>
> It's a bit regrettable that there are actually no fast paths for solid
> mask and r5g6b5 color format. I think at least 'src_8888_8_0565' could
> work fine and justify the existence of 565 related macros in the code.
>
> And one more nitpick is about this part:
> + mov ip, sp
> + push {r4, r5, r6, r7, r8, r9, r10}
>
> It's usually a good idea to save even number of registers on function
> entry to keep the stack 8 bytes aligned. The 8 bytes alignment is
> required by EABI and is quite critical if you for example call some
> external functions or callbacks from your code or perform 64-bit
> aligned memory accesses with stack. Even though this is not strictly
> needed here, it's safer just in case. On ARM Cortex-A8 this has no
> impact on performance.


I modified my current snapshot based on above siarhei's comment, and made
complete new patch set.
Forget about the previous patches.

Following scanline functions are implemented.

over_8888_8888
add_8888_8888
src_8888_8_8888
src_8888_8_0565
src_0565_8_x888
src_0565_8_0565
over_8888_8_8888
add_8888_8_8888

I've got success on pixman scaling test.
Review or comment would be appreciated.

-- 
Best Regards,
Taekyun Kim
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/pixman/attachments/20110413/6c3dd7b4/attachment-0001.html>