Error while compiling Cuda Accelerated Linpack hpl_2.0_FERMI

Question

I use

• Ubuntu 11.04 x86_64
• CUDA 4.0
• OpenMpi 1.4stable
• MKL

When I compile, I get this error:

ar r -L/home/limksadmin/hpl-2.0_FERMI_v13/lib/CUDA/libhpl.a  HPL_dlacpy.o           HPL_dlatcpy.o          HPL_fprintf.o HPL_warn.o             HPL_abort.o            HPL_dlaprnt.o HPL_dlange.o HPL_dlamch.o
ar: -L/home/limksadmin/hpl-2.0_FERMI_v13/lib/CUDA/libhpl.a: No such file or directory
make[2]: *** [lib.grd] Error 9
make[2]: Leaving directory `/home/limksadmin/hpl-2.0_FERMI_v13/src/auxil/CUDA'
make[1]: *** [build_src] Error 2
make[1]: Leaving directory `/home/limksadmin/hpl-2.0_FERMI_v13'
make: *** [build] Error 2

Make.CUDA:

LAdir        = /opt/intel/mkl/lib/intel64
LAlib        = -L $(TOPdir)/src/cuda  -ldgemm -L/usr/local/cuda/lib64 -lcuda -lcudart -lcublas -L$(LAdir) -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -liomp5
MPdir        = /usr/local/mpi/openmpi
MPinc        = -I$(MPdir)/include
MPlib        = -L$(MPdir)/lib/libmpi.so

CC      = /usr/local/mpi/openmpi/bin/mpicc

What could be the problem?

Answer 1

ar: -L/home/limksadmin/hpl-2.0_FERMI_v13/lib/CUDA/libhpl.a: No such file or directory

This line is stating that it can't find the proper HPL static library file for CUDA. One thing you could try to do (if you know that you are 100% positive everything is set correctly in your Makefile) is to add that exact path, without the file name, to your LD_LIBRARY_PATH environment variable:

export LD_LIBRARY_PATH=/home/limksadmin/hpl-2.0_FERMI_v13/lib/CUDA/:$LD_LIBRARY_PATH

Answer 2

Try replacing -openmp with -fopenmp in CCFLAGS variable.

It's specified like this for use with Intel compiler, but such notion confuses gcc (which is base of default ubuntu OpenMPI installation), making it think that penmp is desired output filename (which is not actually true).