This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: Speed comparison with Intel compiler

From: Tobias Schlüter <tobias dot schlueter at physik dot uni-muenchen dot de>
To: Steve Kargl <sgk at troutmask dot apl dot washington dot edu>
Cc: Ignacio Fern?ndez Galv?n <jellby at yahoo dot com>, fortran at gcc dot gnu dot org
Date: Thu, 05 Apr 2007 17:54:25 +0200
Subject: Re: Speed comparison with Intel compiler
References: <915890.4000.qm@web33013.mail.mud.yahoo.com> <20070405152602.GA55858@troutmask.apl.washington.edu>

Steve Kargl wrote:

How could I (easily) tell what is
taking so long time (I guess I could try to prepare some simpler
program which shows the same effect)?

man gprof

You can also use an optimize blas instead of MATMUL.  See the
gfortran docs.

Profiling is the only way to really tell. I presume your rotation matrices are 2x2 complex matrices, where using BLAS won't buy you anything, but which OTOH means that gcc's safe complex multiplication could be the culprit. I seem to recall that ifc uses the naive algorithm for complex multiplication, foregoing numerical stability. Gfortran OTOH only uses it when -ffast-math is given, and the library is not compiled with -ffast-math.

- Tobi

References:
- Speed comparison with Intel compiler
  - From: Ignacio Fernández Galván
- Re: Speed comparison with Intel compiler
  - From: Steve Kargl

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]