Xfermode: SSE2 implementation of overlay_modeproc

Xfermode: SSE2 implementation of overlay_modeproc

With SSE2 optimization, performance of Xfermode_Overlay will improve
about 35% on desktop i7-3770. Here are the data:
before:
Xfermode_Overlay   8888:  cmsecs =     44.17   565:  cmsecs =     59.27
after:
Xfermode_Overlay   8888:  cmsecs =     28.30   565:  cmsecs =     35.84

BUG=skia:

Committed: http://code.google.com/p/skia/source/detail?r=14370

[qiankun, 2014-04-10 09:23:49]

PTAL.

[qiankun, 2014-04-15 08:07:29]

On 2014/04/10 09:23:49, qiankun wrote:
> PTAL.

rebase code. PTAL.

[mtklein, 2014-04-22 15:18:20]

Just wanted to let you know that I do still to intend to review and land all
your SSE transfermodes.  I've been a bit busy with other projects and also want
to spend some time carefully reviewing the code for my own edification.  I don't
anticipate any major changes; this is mostly about me finding the time to catch
up to you.  I'll try to get through them all by Thursday.

This one LGTM, but if you could add some comments where I marked I'd be very
grateful.

https://codereview.chromium.org/232783002/diff/20001/src/opts/SkColor_opts_SS...
File src/opts/SkColor_opts_SSE2.h (right):

https://codereview.chromium.org/232783002/diff/20001/src/opts/SkColor_opts_SS...
src/opts/SkColor_opts_SSE2.h:15: __m128i r2 = _mm_mul_epu32(_mm_srli_si128(a,
4), _mm_srli_si128(b, 4));
Can you narrate this method a bit more, at least with a comment at the top about
what math it does exactly on a and b?  I only ask because Multiply32_SSE2 is a
pretty generic name, and I can't find any corresponding non-SIMD code to compare
it with.

https://codereview.chromium.org/232783002/diff/20001/src/opts/SkXfermode_opts...
File src/opts/SkXfermode_opts_SSE2.cpp (right):

https://codereview.chromium.org/232783002/diff/20001/src/opts/SkXfermode_opts...
src/opts/SkXfermode_opts_SSE2.cpp:232: static inline __m128i
overlay_byte_SSE2(const __m128i& sc, const __m128i& dc,
A brief description of this one would also help.

[qiankun, 2014-04-23 07:40:04]

I am very appreciated for finding time to review the patch.

https://codereview.chromium.org/232783002/diff/20001/src/opts/SkColor_opts_SS...
File src/opts/SkColor_opts_SSE2.h (right):

https://codereview.chromium.org/232783002/diff/20001/src/opts/SkColor_opts_SS...
src/opts/SkColor_opts_SSE2.h:15: __m128i r2 = _mm_mul_epu32(_mm_srli_si128(a,
4), _mm_srli_si128(b, 4));
On 2014/04/22 15:18:21, mtklein wrote:
> Can you narrate this method a bit more, at least with a comment at the top
about
> what math it does exactly on a and b?  I only ask because Multiply32_SSE2 is a
> pretty generic name, and I can't find any corresponding non-SIMD code to
compare
> it with.

Done.

https://codereview.chromium.org/232783002/diff/20001/src/opts/SkXfermode_opts...
File src/opts/SkXfermode_opts_SSE2.cpp (right):

https://codereview.chromium.org/232783002/diff/20001/src/opts/SkXfermode_opts...
src/opts/SkXfermode_opts_SSE2.cpp:232: static inline __m128i
overlay_byte_SSE2(const __m128i& sc, const __m128i& dc,
On 2014/04/22 15:18:21, mtklein wrote:
> A brief description of this one would also help.

This SSE2 function is translated from overlay_byte() in SkXfermode.cpp. You can
refer it for details. Just add a simple description about this.

[qiankun, 2014-04-25 01:46:13]

The CQ bit was checked by qiankun.miao@intel.com

[commit-bot: I haz the power, 2014-04-25 01:52:45]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/qiankun.miao@intel.com/232783002/40001

[commit-bot: I haz the power, 2014-04-25 02:08:38]

Change committed as 14370