linking question [SEC=UNCLASSIFIED]

[DOHERTY, Greg]

I am trying to produce a version of the matlab routine GPSR_BB that octave can call, eventually, on a Jetson board with an nvidia gpu. At the moment I am testing on an x86_64 server.

I have a working version of GPSR_BB.oct produced with mkoctfile from  a C++ version of GPSR_BB that I translated, which runs well.

I have edited the C++ version to include the openACC directives required to use the gpu. I also want to be able to call Sgemv from cublas.

I have a linking problem that you can see below. Any suggestions on how to fix it would be very welcome.

Greg Doherty

 

$ cat Makefile

CXX1=pgc++

CXX1FLAGS=-fast -ta=tesla:lineinfo -Minfo=all,ccff -fPIC -shared -acc -Mcuda

LDFLAGS= ${CXX1FLAGS}

 

MKOCT=/data1/packages/octave/4.0.3/bin/mkoctfile

 

libgdz.a: gdz.o

        ${CXX1} ${CXX1FLAGS}  gdz.o  -lcublas_static -o libgdz.a

 

gdz.o:

        ${CXX1} ${LDFLAGS} -c gdz.cc  > compiler.out 2>&1

 

GPSR_BB.oct:  GPSR_BB.cc libgdz.a

        $(MKOCT) -g -L/home/gdz/DAVID/BLAS -lgdz -Mcuda -fPIC  GPSR_BB.cc

 

clean:

        rm -Rf  GPSR_BB.oct  pgprof* core *.o *.so *.a

 

 

$ make clean

rm -Rf  GPSR_BB.oct  pgprof* core *.o *.so *.a

$ make GPSR_BB.oct > ld.out 2>&1

$ cat ld.out

pgc++ -fast -ta=tesla:lineinfo -Minfo=all,ccff -fPIC -shared -acc -Mcuda -c gdz.cc  > compiler.out 2>&1

pgc++ -fast -ta=tesla:lineinfo -Minfo=all,ccff -fPIC -shared -acc -Mcuda  gdz.o  -lcublas_static -o libgdz.a

/usr/bin/ld: /tmp/pgcudaregxXgfvNbp-fw.o: relocation R_X86_64_PC32 against undefined symbol `__pgi_cuda_register_fat_binary' can not be used when making a shared object; recompile with -fPIC

/usr/bin/ld: final link failed: Bad value

pgacclnk: child process exit status 1: /usr/bin/ld

make: *** [libgdz.a] Error 2

$