Use SSE2 to swizzle RGBA to BGRA

Use SSE2 to swizzle RGBA to BGRA

Using SSE2 can drastically reduce the amount of time it takes to do glTexImage2D. I've also added a plain-C path that's much faster than the one that was there before.

BUG=151
TEST=

Committed: http://code.google.com/p/angleproject/source/detail?r=649

[nicolas, 2011-05-06 07:09:57]

You could also use a rotate instruction:

abgr = (_rotl(argb, 16) & 0x00FF00FF) | (argb & 0xFF00FF00)

The _rotl intrinsic is available on all platforms.

[John Bauman, 2011-05-06 09:26:25]

On 2011/05/06 07:09:57, nicolas wrote:
> You could also use a rotate instruction:
> 
> abgr = (_rotl(argb, 16) & 0x00FF00FF) | (argb & 0xFF00FF00)
> 
> The _rotl intrinsic is available on all platforms.

Yeah, I tried it, but it didn't seem to help the performance at all, so I
figured I might as well not bother with it.

[nicolas, 2011-05-09 13:33:00]

Ok, the patch looks fine to me.

[John Bauman, 2011-05-10 19:37:59]

Actually, I changed my mind about using _rotl (might as well reduce cycles as
much as possible).

[nicolas, 2011-05-11 13:41:19]

Could you move supportsSSE2() to mathutil.h? It can become useful outside of
Texture.cpp as well...

Looks great to me otherwise.

[John Bauman, 2011-05-12 00:47:24]

Ok, moved the function around. Although, hopefully we won't have to add too much
more SSE2.

[dgkoch, 2011-05-12 18:54:40]

LGTM