[Bug fortran/48937] Discrepancy in computation between 32 and 64-bit builds

[kargl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48937

kargl at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |kargl at gcc dot gnu.org

--- Comment #3 from kargl at gcc dot gnu.org 2011-05-09 18:05:04 UTC ---
Have you read Goldberg's paper on floating point computations?

On x86_64-*-freebsd, I get the following results 

fc4x ${OPT} -o z sip.f95 sip_test.f95

OPT=''
 1000 0.695040E-11
Error: 0.20815460466394597E-10

OPT='-O'
 1000 0.646004E-11
Error: 0.20326407224047216E-10

OPT='-O2'
 1000 0.694955E-11
Error: 0.21079360479347997E-10

OPT='-O3'
 1000 0.504164E-11
Error: 0.10621281631983948E-10

OPT='-O3 -funroll-loops'
 1000 0.502817E-11
Error: 0.11637135699515966E-10

OPT='-O3 -funroll-loops -ftree-vectorize'
 1000 0.669018E-11
Error: 0.18756551867227245E-10