The goal of these runs is to monitor the progress of different GCC
releases on x86.
Most widely used common options -O2
and -O3 are taken for all runs on a machine which is
supported for all releases.
Additionally we use
-mtune=pentium4 (-mcpu=pentium4 for
old releases) and -mpc64 of the three latest releases.
Without -mpc64 (using 64-bit floating point arithmetic
instead of 80-bit one), GCC versions of 4.3-4.5 can not generate
expected code for SPECFP2000.
Option -fgraphite was used additionally for graphite
GCC-4.5 run. Options -flto
and -fwhole-program were added for LTO GCC-4.5 run.
x86 for Pentium4 has been supported for a long time. So in major
degree these data reflect progress of machine-independent
optimizations. Exception is the three latest releases because,
unfortunatley for the comparison, -mpc64 significantly
improves SPECFP2000.
The used machine is 3.2Ghz Intel Pentium4 with 1GB memory
under RHEL4.
The compiler is configured with
--enable-checking=release and compiled by GCC-3.4.6.
All changes are given relative to 3.2.3 for SPECInt2000 and relative
to 4.0.4 for SPECFP2000 (4.0 is the first GCC version supporting
Fortran95).
-
Compilation Speed
Data for fortran compiler are given for releases starting with 4.0.4.
We used -O2 for galgel compilation by fortran of 4.0.4 because
galgel crashes the fortran compiler in -O3 mode.
-
Code Size
Here the code size is the corresponding text segment size.
-
Generated Code Performance
Last modified: 04/29/2010 - vmakarov at redhat dot com
Return to index page.
