[Bug fortran/50190] linkpk bench of polyhedron fails during validation with gcc trunk when it is compiled with -Ofast on amd64.

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50190

--- Comment #2 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2013-02-26 19:23:14 UTC ---
On a Core2Duo with '-Ofast' I get 2.51200998D+01 and with '-fprotect-parens
-Ofast' 2.50952975D+01.

I think the tolerance value for linpk does not make sense: the target is a
residual and its value should be 0.0 for a computation without rounding errors.
Hence both the target and tolerance values are arbitrary.
In top of that, residn = resid/(N*norma*normx*eps) and if resid pass its test,
residn should also pass its test. For my tests I have replaced 2.0 with 3.0 to
avoid this spurious "failure".

I suggest to close this PR as won't fix.