Enable SSE2 compilation for ChromeOS....

Enable SSE2 compilation for ChromeOS.
In practice the <video> tag needs this for faster YUV conversion/filtering.
BUG=19113
TEST=Test youtube html5 with chromeos (not chromiumos).  It should not crash, and it should be faster.


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=44740

[fbarchard, 2010-04-15 23:01:18]

chromiumos already has sse2, but like chrome, chromeos had no options specified
at all.
chrome now specifies mmx, because chrome is intended to run on machines without
sse.  But chromeos targets pinetrail and better CPU's.

I've added -msse2 to allow YUV code to improve performance.
mmx = 6.02 ms /frame
sse2 = 4.81 ms / frame

[piman, 2010-04-15 23:03:54]

LGTM

[fbarchard, 2010-04-15 23:26:14]

Note that ChromiumOS uses
                  '-march=pentium4',
                  '-msse2',
                  '-mfpmath=sse',
sse2 is to allow skia to pass layout tests, and not really desirable for
chromium, which should run on as many CPU's as possible, or be compatible with
chrome.
pentium4 for chromium is because pentium4 really is the target CPU we want to
work well... at this time.

ChromeOS should be reviewed in the future
Shortterm, I'd suggest
-march=core, -O3, -mfpmath=sse and -mssse3
And when we get a gcc 4.5 or patched 4.4, use -march=atom
I prefer O2 for performance for Core and Pentium, but Atom benefits from O3.
GCC 32 bit has a bug with mmx.
http://gcc.gnu.org/PR43743
Most issues go away with 64 bit.

[fbarchard, 2010-04-15 23:28:59]

Confirmed build uses the new option with make V=1
make -j6 -r media_unittests BUILDTYPE=Release V=1
 g++  -Werror -pthread -m32 -mmmx -msse2 
objdump -D out/Release/obj.target/media/media/base/yuv_convert.o

50:   f3 0f 7e 09             movq   (%ecx),%xmm1
54:   f3 0f 7e 06             movq   (%esi),%xmm0
58:   66 0f 60 ca             punpcklbw %xmm2,%xmm1
5c:   66 0f 60 c2             punpcklbw %xmm2,%xmm0
60:   66 0f d5 cc             pmullw %xmm4,%xmm1
64:   66 0f d5 c3             pmullw %xmm3,%xmm0
68:   66 0f fd c1             paddw  %xmm1,%xmm0
6c:   83 c6 08                add    $0x8,%esi
6f:   66 0f 71 d0 08          psrlw  $0x8,%xmm0
74:   83 c1 08                add    $0x8,%ecx
77:   66 0f 67 c0             packuswb %xmm0,%xmm0
7b:   66 0f d6 03             movq   %xmm0,(%ebx)
7f:   83 c3 08                add    $0x8,%ebx
82:   39 d8                   cmp    %ebx,%eax
84:   77 ca                   ja     50 <_ZN5media10FilterRowsEPhPKhS2_ii+0x50>
Mmmm good code.  Bon Appetite