This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: induct regression in gcc 4.3.1
On 21 Jun 2008, at 08:00, Jack Howarth wrote:
I see the same regression at -m64 in gcc 4.3.1 with the induct
polyhedron benchmark on i686-apple-darwin9...
powerpc-apple-darwin8 , compiler --enable-checking=release
polyhedron PB05:induct
======
4.2.4 release
m32 with -gdwarf-2
-O0 474.08
-O2 74.42
-O3 58.04
-ffast-math -funroll-loops -ftree-loop-linear -ftree-vectorize -O3
53.76
-O3 -ffast-math 50.86
m64 with -gdwarf-2
-O3 55.38
-O3 -ffast-math 49.57
======
4.3.1 release
m32 with -gdwarf-2
-O0 504.11
-O2 108.78
-O3 54.85
-ffast-math -funroll-loops -ftree-loop-linear -ftree-vectorize -O3
50.82
-O3 -ffast-math 50.07
m64 with -gdwarf-2
-O3 -m64 52.77
-O3 -ffast-math -m64 50.99
=====
gcc version 4.4.0 20080617 (experimental) (GCC)
m32 with -gdwarf-2
-O0 464.53
-O2 77.12
-O3 49.47
-ffast-math -funroll-loops -ftree-loop-linear -ftree-vectorize -O3 48.69
-O3 -ffast-math 47.84
m64 with -gdwarf-2
-O3 -m64 48.62
-O3 -ffast-math -m64 46.63
---
So this particular phenomenon seems absent on PPC.
However, note that there are regressions at 4.3.1 with O0, O2 on PPC ...
.... but these seem to have recovered somewhat with 4.4.0-head
[I only ran induct for this .. I'll go back and do the rest later].
Iain