Add ARM assembly optimised versions of memset{16,32}

Add versions of memset16() and memset32() in ARM assembly.

In benchmarks here on Cortex A9 processors, this code runs 25-30% faster than the C equivalent.

[Sledge, 2010-08-11 15:10:37]

As described in http://codereview.appspot.com/1157045/show , here's ARM assembly
code that works faster on ARM processors than the current C code.

Borrowed from existing code I found in the Android codebase, and very slightly
tweaked to fit here.

[agl, 2010-08-11 15:23:08]

I'll wait a day to see if anyone objects before landing.

Have you signed the CLA[1], or equivalent?


Cheers


[1] http://code.google.com/legal/individual-cla-v1.0.html

[Sledge, 2010-08-11 16:40:04]

On 2010/08/11 15:23:08, agl wrote:
> I'll wait a day to see if anyone objects before landing.
> 
> Have you signed the CLA[1], or equivalent?

I believe that ARM have an agreement in place. I'll double check...

[Sledge, 2010-08-12 09:54:02]

On 2010/08/11 16:40:04, Sledge wrote:
> I believe that ARM have an agreement in place. I'll double check...

Confirmed, yes. Contact me separately if you need and I'll give you the details.

[piman, 2010-08-12 22:39:16]

LGTM

[agl, 2010-08-13 17:05:56]

On Thu, Aug 12, 2010 at 6:39 PM,  <piman@chromium.org> wrote:
> LGTM

Landed as r594.


AGL