Building Netlib LAPACK with MKL

Building Netlib LAPACK with MKL

I am building Netlib LAPACK with Intel Fortran Compiler 9.0 and MKL blas 8.0.1
on a EM64T system with Linux. The libraries build fine but the testing program
xlintsts hangs in an infinite loop. I traced the execution and found that the
program hangs in the slascl routine.

Are there any issues with Netlib LAPACK and the MKL 8.0.1 . I have found that
Netlib LAPACK works fine with other BLAS implementations such as ATLAS.


3 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

If you are building lapack test subroutines with ifort, rather than linking them from the MKL lapack library, you need the -mp option, or, in the latest compilers, -fp-model precise In principle, this bug could also be fixed by upgrading the lapack source to use Fortran 90 intrinsics, rather than those loops in which ifort "optimizes" away the termination.
I don't understand your implication that this is a difference between MKL and ATLAS. It looks like the standard issue of which compiler and options you use when you compile the ancient lapack test source code.
If you still have trouble with this, state clearly which functions you are rebuilding from source, with which compiler options, and which functions you are linking from MKL.

I am having the same problem building CLAPACK and linking to the MKL cblas interface. In the xlintsts tests, it infinitely loops in the slascl routine. A number of the previous tests in xlintsts also fail to pass the threshold.
I am using MKL 6.1 with Visual Studio .NET.
There is no problem if I link to the reference BLAS.

Leave a Comment

Please sign in to add a comment. Not a member? Join today