[cairo] Perf impact of MMS compilation under win32

[Frédéric Plourde]

I've applied those 2 patches :

pixman_MMX_win32_makefile.patch  and
cairo_MMX_win32_makefile.patch 
over :
cairo commit     #40f4750f5ec53ca9ca6e3d621fefb2bd1687909e   and
pixman commit #77c43c6c49511b203fb098a78d38bb4cef7c2cfd

Only for the PAINT target and  solid and image  sources.

The results are at : http://pastebin.mozilla.org/345765

These are very preliminary results, but for now, indicate that :
--------------------------------------------------------------
1) MMX (throught mmx_fast_path) is up and running under win32
2) we have an average of  3.5 speedup for image and paint targets 
without alpha
3) we find some interesting cases of 16X and 23X speedup, which I'll 
invertigate further
4) ** all the tests with alpha compositing result in SLOWDOWNS... I'll 
investigate, and correct. **

Maybe it has something to do with context switches (EMMS functions) not 
done properly
or, it may lead us to some code paths we'll have to transfer to SSE or 
SSE2 instead. I'll  check.

if anyone has any hint about those alpha-related-slowdowns before I 
attack the problem, feel free to tell ;-)

thanx
-fred-