SkSplay and SkUnsplay work on any uint32_t, not just SkPMColor.

SkSplay and SkUnsplay work on any uint32_t, not just SkPMColor.

BUG=

Committed: http://code.google.com/p/skia/source/detail?r=12476

[mtklein, 2013-12-03 20:54:19]

On 2013/12/03 20:31:09, mtklein wrote:

New, less extreme, patch up.

[reed1, 2013-12-03 22:13:57]

lgtm

I wonder if we can rewrite some of them so that we *always* use a mask of
00FF00FF, instead of sometimes that, and sometimes FF00FF00... just makes it
slightly harder for me to read and compare the routines.

[tfarina, 2013-12-03 23:10:22]

Mike could you land this patch with CL description?

[mtklein, 2013-12-03 23:47:57]

On 2013/12/03 23:10:22, tfarina wrote:
> Mike could you land this patch with CL description?

?

[tfarina, 2013-12-04 00:01:01]

On 2013/12/03 23:47:57, mtklein wrote:
> On 2013/12/03 23:10:22, tfarina wrote:
> > Mike could you land this patch with CL description?
> 
> ?

You have a question in your CL description (which is only suitable for the
"Message" part). It does not describe what you are fixing and why.

[mtklein, 2013-12-04 00:13:34]

On 2013/12/04 00:01:01, tfarina wrote:
> On 2013/12/03 23:47:57, mtklein wrote:
> > On 2013/12/03 23:10:22, tfarina wrote:
> > > Mike could you land this patch with CL description?
> > 
> > ?
> 
> You have a question in your CL description (which is only suitable for the
> "Message" part). It does not describe what you are fixing and why.

Oooooh.  Sure.

[mtklein, 2013-12-04 00:27:22]

On 2013/12/03 22:13:57, reed1 wrote:
> lgtm
> 
> I wonder if we can rewrite some of them so that we *always* use a mask of
> 00FF00FF, instead of sometimes that, and sometimes FF00FF00... just makes it
> slightly harder for me to read and compare the routines.

If you mean putting mask back to 0x00FF00FF in SkUnsplay and then using ~mask or
(mask<<8), I'm not sure I'm on board.  I'm sure the generated code would be the
same but it seems unnecessarily roundabout.  I'd rather just write 0xFF00FF00,
particularly so because it makes for easier cross referencing with the examples
in the comments.

I have not been able to think up a version of SkUnsplay that uses 0x00FF00FF
directly that's as fast or faster than this, nor generally anything that's
faster than this, but I wouldn't rule either out.  Will keep this in the back of
my mind in case an idea pops up.

Sadly, I've just noticed that our compilers are, as usual, more clever than I
thought.  Given the constant src and dst I'm using the bench, they've optimized
it all together and hoisted the SkSplay work out of the bench loop.  SkUnsplay
is still in there, so it's not a totally bogus bench, but I do want to make sure
I've got it really solid before going on a replace-all-the-things crusade.

[commit-bot: I haz the power, 2013-12-04 00:30:21]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/mtklein@google.com/102813002/20001

[commit-bot: I haz the power, 2013-12-04 00:42:10]

Change committed as 12476