topic Re: compiling errors using dcsrilu0() in Intel® oneAPI Math Kernel Library

compiling errors using dcsrilu0()

mhuang05 — Mon, 23 Nov 2009 19:54:49 GMT

I use the ILU-preconditional BiCGStab to solve a sparse nonsymmetric linear system on a single core. I employ the funciton dcsrilu0() to create a preconditioner. But I got the following error message when I complied my code:

undefined reference to `dcsrilu0_'

I guess I must miss some links. I write my makefile like

*************************************************

CC = mpicc
CFLAGS = -Wall -g
Includes =-I/opt/intel/mkl/10.2.1.017/include
LFLAGS = -L/opt/intel/mkl/10.2.1.017/lib/em64t
LIBS =-lmkl_scalapack_lp64 -lmkl_blacs_openmpi_lp64 -lmkl_solver_lp64 -lmkl_intel_lp64 -Wl,--start-group -lmkl_intel_thread -lmkl_core -Wl,--end-group -liomp5 -lpthread -lm -lmpi

srcs1 = BiCGStab_OneCore.c mmio.c
objs1 = $(srcs1: .c=.o)

prog1 = bicgstabOneCore

srcs = $(srcs1)
objs = $(objs1)
progs = $(prog1)

all : $(progs)

$(prog1): $(objs1)
$(CC) $(CFLAGS) $(Includes) $(LFLAGS) $(LIBS) -o $@ $^

.c.o:
$(CC) $(CFLAGS) $(Includes) -c $<

clean:
$(RM) -rf *.o $(progs)

*******************************************************************************

I am just wondering what's wrong with this makefile. Thanks.

Min

Re: compiling errors using dcsrilu0()

TimP — Mon, 23 Nov 2009 21:48:45 GMT

2nd attempt to post reply:
Are you mixing up the advice for static and dynamic linking?
If you static link, but don't put all the necessary libraries inside the begin ... end group, and specify full paths, you will see such errors.
If you link dynamic libraries, you should require the begin...end qualifiers.
Please try adhering more closely to the advice in the link advisor, on the right column of forum header page.
If you are running on a single core, there's no point in linking the thread and OpenMP libraries, rather than sequential.

Re: compiling errors using dcsrilu0()

mhuang05 — Mon, 23 Nov 2009 23:42:56 GMT

Quoting - tim18

Thanks for your reply.

I tried the MKL link line advisor and was given the following suggested link line:

-L$MKLPATH $MKLPATH/libmkl_solver_lp64_sequential.a -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group -lpthread

So I modified my makefile as follows:

LIBS = /opt/intel/mkl/10.2.1.017/lib/em64t/libmkl_solver_lp64_sequential.a -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group -lpthread

But, still I got the same error message. Also, if I use static link, it's even worse.

Btw, the reanson I linked the thread and OpenMP is that I previously compiled my sequential programs and parallel programs simultaneously and the makefile worked fine. At that time, I called functions in lapack, scalapack, blacs
and there was no problem.

Re: compiling errors using dcsrilu0()

Vladimir_Koldakov__I — Tue, 24 Nov 2009 04:30:12 GMT

Hi,

Try this line - prerequisites before libs:

$(CC) $(CFLAGS) $(Includes) $(LFLAGS) $^ $(LIBS) -o $@

Thanks,
Vladimir

Re: compiling errors using dcsrilu0()

mhuang05 — Wed, 25 Nov 2009 03:45:05 GMT

Quoting - Vladimir Koldakov (Intel)

Hi,

Try this line - prerequisites before libs:

$(CC) $(CFLAGS) $(Includes) $(LFLAGS) $^ $(LIBS) -o $@

Thanks,
Vladimir

Hi Vladimir,

I just modified my makefile according to your suggestion. It works now!!!

Thank you so much.

Min