Implemented SSE version of ClampX_ClampY_{no}filter_affine

This is the follow up for SSE version of ClampX_ClampY_{no}filter_scale.
With this SSE patch, drawBitmap with rotation can boost about 20+%. Below is some benchmark data:
Original version:
out\Release\bench.exe -config 8888 -match bitmap -rotate -f
orceFilter 0 -repeat 20
skia bench: alpha=0xFF antialias=1 filter=0 rotate=1 scale=0 clip=0 dither=defau
lt strokeWidth=none scalar=float system=WIN32
running bench [640 480]           bitmap_8888_update  8888: cmsecs =  52.26
running bench [640 480]  bitmap_8888_update_volatile  8888: cmsecs =  49.92
running bench [640 480]                bitmap_index8  8888: cmsecs =  54.60
running bench [640 480]              bitmap_index8_A  8888: cmsecs =  67.86
running bench [640 480]                  bitmap_4444  8888: cmsecs =  66.30
running bench [640 480]                bitmap_4444_A  8888: cmsecs =  82.68
running bench [640 480]                   bitmap_565  8888: cmsecs =  72.54
running bench [640 480]                  bitmap_8888  8888: cmsecs =  50.70
running bench [640 480]                bitmap_8888_A  8888: cmsecs =  67.08

out\Release\bench.exe -config 8888 -match bitmap -rotate -f
orceFilter 1 -repeat 20
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 = 155.22
running bench [640 480]  bitmap_8888_update_volatile  8888: cmsecs = 153.66
running bench [640 480]                bitmap_index8  8888: cmsecs = 173.16
running bench [640 480]              bitmap_index8_A  8888: cmsecs = 190.32
running bench [640 480]                  bitmap_4444  8888: cmsecs = 141.18
running bench [640 480]                bitmap_4444_A  8888: cmsecs = 163.80
running bench [640 480]                   bitmap_565  8888: cmsecs = 145.86
running bench [640 480]                  bitmap_8888  8888: cmsecs = 153.66
running bench [640 480]                bitmap_8888_A  8888: cmsecs = 173.94

The SSE2 version:
out\Release\bench.exe -config 8888 -match bitmap -rotate -f
orceFilter 0 -repeat 20
skia bench: alpha=0xFF antialias=1 filter=0 rotate=1 scale=0 clip=0 dither=defau
lt strokeWidth=none scalar=float system=WIN32
running bench [640 480]           bitmap_8888_update  8888: cmsecs =  39.78
running bench [640 480]  bitmap_8888_update_volatile  8888: cmsecs =  41.34
running bench [640 480]                bitmap_index8  8888: cmsecs =  44.46
running bench [640 480]              bitmap_index8_A  8888: cmsecs =  60.06
running bench [640 480]                  bitmap_4444  8888: cmsecs =  60.06
running bench [640 480]                bitmap_4444_A  8888: cmsecs =  73.32
running bench [640 480]                   bitmap_565  8888: cmsecs =  67.08
running bench [640 480]                  bitmap_8888  8888: cmsecs =  43.68
running bench [640 480]                bitmap_8888_A  8888: cmsecs =  58.50

out\Release\bench.exe -config 8888 -match bitmap -rotate -f
orceFilter 1 -repeat 20
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 = 122.46
running bench [640 480]  bitmap_8888_update_volatile  8888: cmsecs = 121.68
running bench [640 480]                bitmap_index8  8888: cmsecs = 136.50
running bench [640 480]              bitmap_index8_A  8888: cmsecs = 156.00
running bench [640 480]                  bitmap_4444  8888: cmsecs = 113.88
running bench [640 480]                bitmap_4444_A  8888: cmsecs = 126.36
running bench [640 480]                   bitmap_565  8888: cmsecs = 113.10
running bench [640 480]                  bitmap_8888  8888: cmsecs = 118.56
running bench [640 480]                bitmap_8888_A  8888: cmsecs = 138.84

[Jin A.Yang, 2012-02-28 13:17:16]

Hi Tom,

Please help review this patch when you are available, thanks!

[TomH, 2012-02-28 15:43:59]

Thank you! I'm seeing 25-30% speedup. Committed as r3272.

Do you have a real-world use case that drives this? We don't see much rotation
of bitmaps in our clients. We're currently investigating accelerating some of
the matrix operations, or wondering if there are refinements for some of the
things we've already SSE-accelerated which continue to be bottlenecks.

[Jin A.Yang, 2012-02-29 01:21:51]

On 2012/02/28 15:43:59, TomH wrote:
> Thank you! I'm seeing 25-30% speedup. Committed as r3272.
> 
> Do you have a real-world use case that drives this? We don't see much rotation
> of bitmaps in our clients. We're currently investigating accelerating some of
> the matrix operations, or wondering if there are refinements for some of the
> things we've already SSE-accelerated which continue to be bottlenecks.

Yes, my real usage case for this is the HTML5 benchmark below:
http://demos.hacks.mozilla.org/openweb/HWACCEL/
In this benchmark, it draws a lot of the bitmaps in rotation. And we are
measuring the HTML5 2D canvas performance using this benchmark.
There is another bottleneck in this case, that is the
S32_opaque_D32_filter_DXDY.
I am working on the SSSE3 version of this now.

[TomH, 2012-02-29 15:08:32]

On 2012/02/29 01:21:51, Jin Yang wrote:
> There is another bottleneck in this case, that is the
> S32_opaque_D32_filter_DXDY.
> I am working on the SSSE3 version of this now.

Sounds great!
Please close this issue when you get a chance.