Compiling Lapack

Compiling Lapack

tonio's picture


Compiling (i.e. compiling, testing and timing) lapack with ifcas well asifort is ok with -O0, but always fails when the very fisrt optimizations are used.
It seems an infinite loop occurs at the "Testing double precision lapack linear equation routines" step.

This was tested on a mdk 9.1 and9.2, both with pentium M and amd barton and the results are the same.

I emailed directly to the lapack web site, but theanswer was not helpfull at all.

After a lot of googling, I'm hopeless. Is itdefinitely impossible to have lapack working with intel compiler orsome hidden voodoo trick is necessary ?

4 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.
Tim Prince's picture

Useful information on this topic has been posted many times. I don't understand why you didn't learn more via Google. Check the comp.lang.fortran archives, as well as the history of this forum. Consider replacing the loops which attempt to discover the numerical properties of your Fortran implementation with standard Fortran intrinsics like TINY, EPSILON, and the like. You may be able to look up specific recommendations.At least, separate those out and compile with -mp. Intel's solution is to sell MKL. I personally believe that the compiler should at least issue a warning when a loop is optimized in such a way that there is no exit, but I'm certainly a minority opinion.

Steve Lionel (Intel)'s picture

I don't think the compiler can tell in this case.

Steve
tonio's picture


Thx for reply.

Adding the -mp flag solved my problems, and this was also explained on comp.lang.fortran, which I didn't see in my googling.

I just have a few failed tests ( it was also the case with g77), butthings seem to work fine.

Thx again. :-)

Message Edited by Tonio on 01-08-2004 05:00 AM

Login to leave a comment.