cuda shared library linking: undefined reference to cudaRegisterLinkedBinary

Goal:

1. create a shared library containing my CUDA kernels that has a CUDA-free wrapper/header.
2. create a test executable for the shared library.

Problem

1. shared library MYLIB.so seems to compile fine. (no problem).
2. Error in linking:

./libMYLIB.so: undefined reference to __cudaRegisterLinkedBinary_39_tmpxft_000018cf_00000000_6_MYLIB_cpp1_ii_74c599a1

simplified makefile:

libMYlib.so :  MYLIB.o
    g++  -shared  -Wl,-soname,libMYLIB.so  -o libMYLIB.so    MYLIB.o  -L/the/cuda/lib/dir  -lcudart


MYLIB.o : MYLIB.cu   MYLIB.h
    nvcc  -m64   -arch=sm_20 -dc  -Xcompiler '-fPIC'  MYLIB.cu  -o  MYLIB.o  -L/the/cuda/lib/dir  -lcudart


test : test.cpp  libMYlib.so
        g++   test.cpp  -o test  -L.  -ldl -Wl,-rpath,.   -lMYLIB  -L/the/cuda/lib/dir  -lcudart

indeed

nm libMYLIB.so shows that all CUDA api functions are "undefined symbols":

         U __cudaRegisterFunction
         U __cudaRegisterLinkedBinary_39_tmpxft_0000598c_00000000_6_CUPA_cpp1_ii_74c599a1
         U cudaEventRecord
         U cudaFree
         U cudaGetDevice
         U cudaGetDeviceProperties
         U cudaGetErrorString
         U cudaLaunch
         U cudaMalloc
         U cudaMemcpy

So CUDA somehow did not get linked to the shared library MYLIB.so What am I missing?

CUDA did not even get linked to the object file somehow:

nm MYLIB.o

         U __cudaRegisterFunction
         U __cudaRegisterLinkedBinary_39_tmpxft_0000598c_00000000_6_CUPA_cpp1_ii_74c599a1
         U cudaEventRecord
         U cudaFree
         U cudaGetDevice
         U cudaGetDeviceProperties
         U cudaGetErrorString
         U cudaLaunch
         U cudaMalloc
         U cudaMemcpy

(same as above)