[plt-scheme] Noticeable slowdown after 369.4

From: Matthew Flatt (mflatt at cs.utah.edu)
Date: Sat Jan 13 20:56:53 EST 2007

At Sat, 13 Jan 2007 16:58:59 -0800, Eric Hanchrow wrote:
> Attached is a program of mine that is notably slower with 369.4 than
> it was with 301.4 -- and I'm pretty sure it's also slower than it
> would have been just a few days ago, before the precise GC became the
> default (I'd be happy to check if you want).
> 
> To run the program, just type "make check" -- assuming that you've got
> an English dictionary in /usr/share/dict/words.
> 
> On my machine, it took 40 seconds under 301.4, but now takes about
> twice that.

What platform is your machine?

For x86[_64] and PPC machines, the main difference between v301.4 and
v369.4 should be the JIT, not the GC.

On x86 Mac OS X with v369.4, I get

 3m, JIT:
   real    1m18.784s
   user    1m9.848s
   sys     0m4.111s

 CGC, JIT:
   real    1m14.912s
   user    1m13.427s
   sys     0m0.680s

 3m, JIT:
   real    1m58.384s
   user    1m54.474s
   sys     0m3.621s

 CGC, no JIT:
   real    1m52.333s
   user    1m51.132s
   sys     0m0.799s

The CGC JIT number is about the same for v360 on my machine.

I can imagine the non-JIT 3m vs. CGC difference might be bigger for
Sparc, where there's still no JIT (due only to lack of demand so far).

Matthew



Posted on the users mailing list.