kit times

Stephen Weeks sweeks@intertrust.com
Tue, 21 Dec 1999 11:00:44 -0800 (PST)


> I'm attaching the log file and the run-benchmarks code used
> to run the kit.  I'm very confused about the results.  With
> polyvariance turned on, and with inline ratio set to 30, we're
> still twice as slow as nj (with and without uncurrying). Let me know 
> if there's  something obviously wrong in my reading of the results.

If I may summarize:

time	version		flatten	inline	polyvariance
 67	1999-3-19	1	15	no
141	1999-7-12	2	30	yes
133	1999-11-21	2	30	yes

The NJ runtime has remained unchanged at 59.

Thus, it looks like either some strangeness was introduced into the
compiler between 1999-3-19 and 1999-7-12 or one of the flags is
killing performance.  I doubt the former since most stuff going on
since 1999-3-19 has been functionality improvements, so I believe it's
the latter.  If I had to guess, I'd say it's mostly due to the extra
round of flattening, but I don't really have much idea.