Mailinglist Archive: opensuse-factory (850 mails)
| < Previous | Next > |
Re: [opensuse-factory] Dist Meeting 2006-09-14: Proposed Agenda
- From: Tobias Burnus <burnus@xxxxxxxx>
- Date: Wed, 13 Sep 2006 23:35:07 +0200
- Message-id: <4508798B.4070906@xxxxxxxx>
Hello,
Marc Collin schrieb:
> -ffast-math
>
Note that this requires a carefully written program otherwise it may not
work reliably. From reading *@gcc.gnu.org and from the manpage ("This
option should never be turned on by any -O option since it can result in
incorrect output for programs which depend on an exact implementation of
IEEE or ISO rules/specifications for math functions.") I think one
should be really careful with this option.
For my calculations, I don't dare it to use it - better 1% slower and
correct than (potentially) 100% wrong and 1% faster.*
For non-number crunching programs I would even be more careful.
Tobias
*In similar regard: Using GCC gfortran compiled Netlib LAPACK [SUSE's
lapack.rpm] with the Intel Fortran Compiler (ifort) versus using the
Intel Math Kernel Library mit ifort: With one of my programs the speedup
was around one percent. [Ok, with one program: ifort compiled is 20%
faster than gfortran compiled.]
---------------------------------------------------------------------
To unsubscribe, e-mail: opensuse-factory+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: opensuse-factory+help@xxxxxxxxxxxx
Marc Collin schrieb:
> -ffast-math
>
Note that this requires a carefully written program otherwise it may not
work reliably. From reading *@gcc.gnu.org and from the manpage ("This
option should never be turned on by any -O option since it can result in
incorrect output for programs which depend on an exact implementation of
IEEE or ISO rules/specifications for math functions.") I think one
should be really careful with this option.
For my calculations, I don't dare it to use it - better 1% slower and
correct than (potentially) 100% wrong and 1% faster.*
For non-number crunching programs I would even be more careful.
Tobias
*In similar regard: Using GCC gfortran compiled Netlib LAPACK [SUSE's
lapack.rpm] with the Intel Fortran Compiler (ifort) versus using the
Intel Math Kernel Library mit ifort: With one of my programs the speedup
was around one percent. [Ok, with one program: ifort compiled is 20%
faster than gfortran compiled.]
---------------------------------------------------------------------
To unsubscribe, e-mail: opensuse-factory+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: opensuse-factory+help@xxxxxxxxxxxx
| < Previous | Next > |