develooper Front page | perl.perl5.porters | Postings from January 2001

Re: Tweaked config for Solaris/sparc == 30% perfomance boost

Thread Previous | Thread Next
Alan Burlison
January 10, 2001 15:07
Re: Tweaked config for Solaris/sparc == 30% perfomance boost
Message ID:
Nick Ing-Simmons wrote:

> How do I get a measure of what sigsetjmp() is costing.
> A very simplistic test gave
> With sigsetjmp():
> All tests successful.
> u=0.9  s=0.85  cu=72.37  cs=34.45  scripts=275  tests=21137
> With setjmp():
> All tests successful.
> u=1  s=0.75  cu=74.35  cs=35.23  scripts=275  tests=21137

Anything that causes JMPENV_PUSH to be called, eg eval.  Ilya did some
work on this the back end of last year, and his tests indicated a
significant speedup in object creation by switching from sigsetjmp to
setjmp - see,
there is some test code in there.

Here is another noddy test:

$ time /perl -e 'for ($i=100000;$i;$i--) { eval "die" }'

With sigsetjmp
real    0m28.70s
user    0m25.81s
sys     0m2.76s

With setjmp
real    0m24.20s
user    0m23.93s
sys     0m0.02s

A 15% difference.  With XS code you can push a bit harder, so the
difference is more marked.  My app does lots and lots of XS-based object
creation/destruction, so I see about 18%.

Alan Burlison

Thread Previous | Thread Next Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at | Group listing | About