uncurry

Suresh Jagannathan suresh@research.nj.nec.com
Thu, 16 Dec 1999 11:42:51 -0500


   From: "Stephen Weeks" <sweeks@intertrust.com>
   Content-Type: text/plain; charset=us-ascii
   Date: Wed, 15 Dec 1999 15:14:47 -0800 (PST)
   X-UIDL: dcd6d5c70f86581758a96bf668d851bf


   > I added some more tweaks to the uncurrier.  For the kit, the
   > uncurry time improves a bit to 72.73; without uncurrying,
   > it's 77.45.  So, we get roughly a 6.5% speedup.  

   Cool.  I think that's more than negligible.  What about code size?

   > Still quite a bit slower than NJ, though.

   ??  I thought we were running at about 90% the speed of NJ on this
   benchmark, before the 6.5% speedup.  So now, we should be at 95%.

Unfortunately, on the latest benchmark run, NJ ran in 59.35 seconds.
So, it appears we're ~20% slower.  I've got a couple of more ideas
I'm going to pursue wrt uncurrier, but I think this is roughly about
the best we can expect. I didn't measure code size -- I'll do that in
the next run.

    -- Suresh