c++ - Error linking with CUDA code: multiple definitions of `__cudaRegisterLinkedBinary_ -

i have cuda code i'm compiling .a library, , (cuda-related) regular-c++ code app uses it. undergoing intermediate linking.

now, on 1 machine (with cuda 8.0 rc) build succeeds, on machine (with maxwell rather kepler card, in case matters) get:

/tmp/tmpxft_00001796_00000000-2_ktkernels_intermediate_link.reg.c:25: multiple definition of `__cudaregisterlinkedbinary_66_tmpxft_00007a5f_00000000_16_cuda_device_runtime_compute_52_cpp1_ii_8b1a5d37' cmakefiles/tester.dir/tester_intermediate_link.o:/tmp/tmpxft_0000180b_00000000-2_tester_intermediate_link.reg.c:4: first defined here collect2: error: ld returned 1 exit status cmakefiles/tester.dir/build.make:1766: recipe target 'bin/tester' failed make[2]: *** [bin/tester] error 1 

i started removing files what's compiled binary makes calls library code - , if remove of them linking succeed.

my questions:

• under circumstances possible such inconsistent behavior occur?
• can possibly result of "second linking" both library , binary?
• what can determine in conflict (e.g. symbols in )?
• if nothing conflicting, should avoid this?

notes:

• on 1 machine i'm using cuda 7.5, on other machine it's cuda 8.0 rc.

• under circumstances possible such inconsistent behavior occur?

if attempt multiple device linkages within single application.

• can possibly result of "second linking" library , binary?

almost no doubt.

• what can determine in conflict (e.g. symbols in )?

the conflict multiple definitions of boilerplate runtime generated during device link phases , used load device code context runtime api.

• if nothing conflicting, should avoid this?

the conflicts real. , avoidance involves linking separately complied device code. beyond can't tell how fix because have chosen not tell doing.