stepanov_v1p2 benchmark

[Daniel Berlin]

Mike Stump <mrs@windriver.com> writes:

> > From: larsbj@lyx.org (Lars Gullik Bjønnes)
> > Date: 12 Oct 2000 01:58:22 +0200
> 
> > I saw that this should be used for a measurement of code quality. I
> > just run it [ ... ]
> 
> > Results for 2.95.2:
> > test      absolute   additions      ratio with
> > number    time       per second     test0
> >  2        1.71sec    29.24M         0.90
> >  
> > Results for cvs gcc:
> > test      absolute   additions      ratio with
> > number    time       per second     test0
> >  2        4.58sec    10.92M         2.57
> 
> Now, just binary search the cvs archive for when test 2 started working bad.
> 
> When you get it down to about 1 day, stop, and then do a diff on
> gcc/{cp/,}Changelog between the two, and then send that in, with the
> above results.
> 
> One day the dejagnu tests will include PERF: %f name of testcase, and cmp_logs will
> issue good diagnostics for these things, until then...


No, we already know what caused this, IIRC.
It's the new tree inliner, as far as anyone can tell.
We already had this discussion about this very benchmark before, if
you search the archives.
I reported almost these exact same results months ago.
Mark said he'd make sure it got taken care of by release, and that was
the end of the discussion.

So in other words, don't waste your time.
--Dan