kit

Stephen Weeks sweeks@intertrust.com
Tue, 21 Dec 1999 13:37:42 -0800 (PST)


> It's really
> quite surprising that the latest release which does no worse (and often is
> significantly better) than the earlier releases on all the other benchmarks
> loses so badly on the kit.

We don't know this.  My guess about flattening was just wrong.  We
still need to run -flatten 1 -inline 15 -no-polyvariance in order to
have the same flags as with 1999-3-19.  If that is still 2X slower,
then I'll be very confused.