[KinoSearch] improving vbyte decode performance

[Nathan Kurz]

Here's a patch that should improve decode performance about 15%.  I've
tested it only a little, but it passes the Build tests.  I've done
testing with 'static inline', and at least with GCC 3.4 it seems like
using the macro directly might shave off another 15%.

I managed to get a branchless version working, but didn't see the
speedup I was hoping for. It might come eventually, but I'm not as
optimistic as I was before.    I also found a few slightly faster
versions, but these were more susceptible to compiler optimization
flags.

I can take another pass at it later, but this seemed good to get in
now in case that's delayed.

Nathan Kurz
nate at verse.com

ps. I tried committing directly to SVN, but got a 500 error.
pps.  cc'd to Marvin as I haven't seen replies to my posts lately.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: decode.patch
Type: application/octet-stream
Size: 1799 bytes
Desc: not available
Url : http://rectangular.com/pipermail/kinosearch/attachments/20090105/ba3bb930/attachment-0002.obj