ScaLAPACK double complex bug on Mac OS X with Intel

[magnetron]

I installed ScaLAPACK without any issues on an Mac Pro running OS X 10.7.4 with the open-mpi compilers built around Intel fortran and c compilers, and linking to the BLAS and LAPACK in the intel_mkl. I then ran the test functions in /EXAMPLES and /TESTING and noticed that the single, double and complex routines pass the tests but all the double complex (z) routines fail. For example:

/EXAMPLES/xzscaex finishes with:

||A * X - B|| / ( ||X|| * ||A|| * eps * N ) = 556537385074557.

The answer is suspicious.


And /TESTING/xzlu has a bunch of these fails:

WALL 13 13 4 9 1 1 1 0.00 0.06 0.32 FAILED
||Ax-b||/(||x||*||A||*eps*N) 173390579830764.8750000

Any thoughts on what's going on? This is my first time using ScaLAPACK. I would like to be able to use it for development and testing on my OS X desktop (for complex double EM problems) before moving them over to the long queue waits on a bigger Linux cluster.

-Kerry

[rodney]

I'm not sure what could be wrong with your installation -- everything works fine on my 10.7.4 Mac OS system. I am using OpenMPI 1.6 and the latest versions of the Intel compilers with MKL. I ran the tests and example with mpirun -np 4. Below is the SLmake.inc file I used.

--Rodney

---

FC = mpif90
CC = mpicc
NOOPT = -O0
FCFLAGS = -O3
CCFLAGS = -O3
FCLOADER = $(FC)
CCLOADER = $(CC)
FCLOADFLAGS = $(FCFLAGS)
CCLOADFLAGS = $(CCFLAGS)
ARCH = ar
ARCHFLAGS = cr
RANLIB = ranlib
SCALAPACKLIB = libscalapack.a
LIBS=-mkl=sequential

[magnetron]

Thanks for letting me know it works when the compilers are recent. I just updated to the latest Intel C and Fortran compilers (2011 update 12 or something) and then built the latest Open MPI compilers (1.6) around those and it's all hunky dory now. ( For reference, it didn't work before with the 11.1 Intel Compilers and Open MPI 1.4.1).