Implemented SSSE3 version of S32_{opaque,alpha}_D32_filter_DXDY

Implemented SSSE3 version of S32_{opaque,alpha}_D32_filter_DXDY based on the current SSSE3 framework.
Below are the benchmark data:
Original:
bench.exe -config 8888 -match bitmap_888 -rotate -for
ceFilter 1 -forceBlend 0 -repeat 200
skia bench: alpha=0xFF antialias=1 filter=1 rotate=1 scale=0 clip=0 dither=defau
lt strokeWidth=none scalar=float system=WIN32
running bench [640 480]           bitmap_8888_update  8888: cmsecs = 118.87
running bench [640 480]  bitmap_8888_update_volatile  8888: cmsecs = 117.94
running bench [640 480]                  bitmap_8888  8888: cmsecs = 118.48
running bench [640 480]                bitmap_8888_A  8888: cmsecs = 126.44

bench.exe -config 8888 -match bitmap_888 -rotate -for
ceFilter 1 -forceBlend 1 -repeat 200
skia bench: alpha=0x80 antialias=1 filter=1 rotate=1 scale=0 clip=0 dither=defau
lt strokeWidth=none scalar=float system=WIN32
running bench [640 480]           bitmap_8888_update  8888: cmsecs = 132.91
running bench [640 480]  bitmap_8888_update_volatile  8888: cmsecs = 130.88
running bench [640 480]                  bitmap_8888  8888: cmsecs = 129.64
running bench [640 480]                bitmap_8888_A  8888: cmsecs = 131.35

SSSE3 version:
bench.exe -config 8888 -match bitmap_888 -rotate -for
ceFilter 1 -forceBlend 0 -repeat 200
skia bench: alpha=0xFF antialias=1 filter=1 rotate=1 scale=0 clip=0 dither=defau
lt strokeWidth=none scalar=float system=WIN32
running bench [640 480]           bitmap_8888_update  8888: cmsecs =  82.91
running bench [640 480]  bitmap_8888_update_volatile  8888: cmsecs =  82.60
running bench [640 480]                  bitmap_8888  8888: cmsecs =  83.30
running bench [640 480]                bitmap_8888_A  8888: cmsecs =  90.25

bench.exe -config 8888 -match bitmap_888 -rotate -for
ceFilter 1 -forceBlend 1 -repeat 200
skia bench: alpha=0x80 antialias=1 filter=1 rotate=1 scale=0 clip=0 dither=defau
lt strokeWidth=none scalar=float system=WIN32
running bench [640 480]           bitmap_8888_update  8888: cmsecs = 109.12
running bench [640 480]  bitmap_8888_update_volatile  8888: cmsecs = 107.33
running bench [640 480]                  bitmap_8888  8888: cmsecs = 107.17
running bench [640 480]                bitmap_8888_A  8888: cmsecs = 108.11

We can observe about 30% boost for rotated image drawing.

For the HTML5 benchmark with Chromium in the below:
http://demos.hacks.mozilla.org/openweb/HWACCEL/

The FPS can boost from 69 to 91.

committed in http://code.google.com/p/skia/source/detail?r=3331

[Jin A.Yang, 2012-02-29 09:17:44]

fix the nit

[Jin A.Yang, 2012-03-06 03:30:08]

Hi Tom,

Please help this patch based on SSSE3 when you have time, thanks!

Jin

[TomH, 2012-03-06 14:59:27]

LGTM.

Awesome! Committed as r3331.

http://codereview.appspot.com/5704055/diff/2002/src/opts/SkBitmapProcState_op...
File src/opts/SkBitmapProcState_opts_SSSE3.cpp (right):

http://codereview.appspot.com/5704055/diff/2002/src/opts/SkBitmapProcState_op...
src/opts/SkBitmapProcState_opts_SSSE3.cpp:600: __m128i alpha;
OK, normally I'd worry about this being uninitialized, but if the templating
works it isn't referenced when has_alpha is false...

[TomH, 2012-03-06 15:39:39]

It looks like this patch works fine on Windows + Linux, but may be rendering
incorrectly on a MacMini with OS X Lion.

[TomH, 2012-03-06 16:08:07]

Rerunning that buildbot it seems to have succeeded; we'll leave this intact for
now, but monitor it closely. Please don't close this issue before the end of the
week.

[Jin A.Yang, 2012-03-07 02:29:14]

On 2012/03/06 16:08:07, TomH wrote:
> Rerunning that buildbot it seems to have succeeded; we'll leave this intact
for
> now, but monitor it closely. Please don't close this issue before the end of
the
> week.
Sure, thanks!

[TomH, 2012-03-15 17:41:55]

All good; please close when you get a chance.