ARM Skia NEON patches - 11 - Blitter_RGB16

ARM Skia NEON patches - 11 - Blitter_RGB16





Blitter_RGB16: fixes and improvements

- fix alpha calculation: it was still using the old version of
  SkAlpha255To256. 11 more tests pass in gm.

- clean a lot the code: the existing code was "a bit" messy with
  a lot of duplicated hardcoded constants, got rid of all this.

- improve speed a little: part of it as a side-effect of the change
  in the way alpha is calculated but also by grouping loads and stores.

One "issue" was present and still remains: the NEON code doesn't give
the same result as the black blitter on black. It accounts for dozens
of mismatches in gm. Is this considered "not too bad"? Would you be
interested in a NEON version of the black blitter? The current
comments seem to indicate that the black blitter is here only to give
a performance boost when NEON is not presents so I didn't write a NEON
version.

BUG=

Committed: http://code.google.com/p/skia/source/detail?r=10635

[kevin.petit.not.used.account, 2013-07-25 18:18:37]

ping

[tomhudson, 2013-07-25 20:56:47]

Sorry for the lack of response, if I'm going to do more than rubberstamp this
I'm going to have to give it a heapload of quality time, which has been pretty
sparse for me recently.

[reed1, 2013-07-25 21:22:56]

General answer about the black blitter:

1. black text (which uses the mask blits) is very common, so we want to be sure
that we draw it fast
2. assembly procs are allowed to differ from the C variants (by little bits) if
that makes them faster, provided they adhere to basic invariants:
  - 0xFF mask values don't affect the src
  - 0x00 mask values don't affect the dst

[kevin.petit.not.used.account, 2013-08-07 10:35:10]

On 2013/07/25 21:22:56, reed1 wrote:
> General answer about the black blitter: [...]

OK, I'll add the black blitter to my list of Skia things to do. For now, I'll
just ping on this one.

[reed1, 2013-08-07 12:34:41]

If this code maintains the invariants I listed earlier (#1 and #2), then lgtm

[kevin.petit.not.used.account, 2013-08-08 10:43:32]

I've just done the math: they are maintained.

[commit-bot: I haz the power, 2013-08-08 10:44:09]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/kevin.petit.arm@gmail.com/18666005/1

[commit-bot: I haz the power, 2013-08-08 10:51:46]

Change committed as 10635