There is a 20% execution regression on the polyhedron induct.f90 benchmark when compiled with gfortran in gcc 4.3.1 compared to gcc 4.2.4. This reduction appears to be triggered by the -fassociative-math (which appears to require -fno-signed-zeros -fno-trapping-math to actually be used by gfortran). I am attaching induct.s.good generated by...
gfortran -fno-signed-zeros -fno-trapping-math --save-temps -O3 induct.f90 which doesn't show the speed regression and induct.s.bad generated by... gfortran -fassociative-math -fno-signed-zeros -fno-trapping-math --save-temps -O3 induct.f90 The induct.f90 and induct.inp data files for compiling this benchmark can be downloaded from http://www.polyhedron.com/web_images/documents/pb05.zip. -- Summary: major execution regression for induct.f90 polyhedron benchmark in 4.3.1 on Intel Product: gcc Version: 4.3.1 Status: UNCONFIRMED Severity: normal Priority: P3 Component: fortran AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: howarth at nitro dot med dot uc dot edu GCC build triplet: i686-apple-darwin9 GCC host triplet: i686-apple-darwin9 GCC target triplet: i686-apple-darwin9 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36599
