Re: GNAT compiler switches and optimization

["Alinabi" <alexander.the.average@gmail.com>]

I ran your test programs compiled with gcc 4.0.3 and the following
optimizations:
COMMON_FLAGS=-g -march=opteron -mtune=opteron -mfpmath=sse
-fomit-frame-pointer -O2 -fdump-tree-optimized
and I cannot reproduce the large differences in performance everyone
else talks about. Here are the times I get:

N        Ada            Fortran
====================
64      0.002029   0.000000
128    0.016321   0.016000
256    0.214143   0.204013
512    3.125888   3.124195
800    6.374982   5.864366
1024  34.10479   35.22620
2048  277.3071   283.2417

Jeffrey Creem wrote:
> Jeffrey Creem wrote:
> > Pascal Obry wrote:
> >
> >> Jeffrey Creem a écrit :
> >>
> >>
> >>> I'd certainly be willing to run a few benchmarks but the important thing
> >>> here is that rather innocent looking code is running 2-4x slower than it
> >>> "should".
> >>
> >>
> >>
> >> The first thing is to be sure that we are running the same program.
> >>
> >> Running your program with the following changes (as done in Fortran):
> >>
> >> 1. Using Sum tmp var for the computation
> >>
> >>    for I in A'range (1) loop
> >>       for J in A'range (2) loop
> >>          Sum := 0.0;
> >>          for R in A'range (2) loop
> >>             Sum := Sum + A (I, R) * B (R, J);
> >>          end loop;
> >>          C (I, J) := Sum;
> >>       end loop;
> >>    end loop;
> >>
> >> 2. Using Long_Float instead of Float (I think Fortran float is a
> >> Long_Float, to be checked).
> >>
> >> I went from 7.8s to 4.8s (with 1) and to 4.2s (with 2).
> >>
> >> Pascal.
> >>
> >
> > Actually, the original poster's Ada program had the temp var and all of
> > my comparisons of programs that I have asserted were "the same" used the
> > temporary.
> >
> > As for Long_Float v.s. Short_Float, gfortran is using 32 bit floats (as
> > verified by dumping the tree representation and assembly language).
> >
> > Since we are all getting a bit confused by specific versions and numbers
> > I thought I'd post a summary and create a way of tracking more global
> > performance issues at the gnuada.sf.net wiki.
> >
> > The direct link to these test results are here:
> >
> > http://gnuada.sourceforge.net/pmwiki.php/Main/Oct2006CLAGFORTRANComparison
> >
> >
> >
>
>
>
> Actually, as a result of this, I submitted a bug report to the GCC
> bugzilla list. You can follow progress on it here:
>
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29543
>
> Interesting initial feedback is that
> 1) Not an Ada bug.
> 2) Is a FORTRAN bug
> 3) Is a backend limitation of the optimizer.
>
> Of course, the FORTRAN one still runs correctly so I don't think most
> users will care that it is because of a bug :)