Cuda Version discrepancy

[samridhim]

When I run make on libs/smat/src I get the following error -

./smat_cuda/launch_util.h(67): error: function "atomicAdd(double *, double)" has already been defined

1 error detected in the compilation of "/tmp/tmpxft_00000e6b_00000000-4_reduce.cpp4.ii". makefile:230: recipe for target '../build/release/obj/SMAT_CUDA/reduce.o' failed make: *** [../build/release/obj/SMAT_CUDA/reduce.o] Error 2

I think it is happening because I am using CUDA 9.2 and the code doesn't support it. The CUDA samples from NVIDIA were running perfectly. Please look into it.

Thanks

[Lesley96]

I am puzzled by the same problem, have you solved it by changing the version of CUDA? Thanks!

[samridhim]

Nope. I tried running it with cuda 8. It gave me an "ld" linking error on the smat cu file

[samridhim]

maybe the code isn't compatible for higher cuda versions.

[Lesley96]

I tried running makefile with cuda7.0 on ubuntu16.04, then the new error came out:

/usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const void*, size_t)’: /usr/include/string.h:653:42: error: ‘memcpy’ was not declared in this scope return (char *) memcpy (__dest, __src, __n) + __n; ^ makefile:230: recipe for target '../build/release/obj/SMAT_CUDA/reduce.o' failed make: *** [../build/release/obj/SMAT_CUDA/reduce.o] Error 1

I think it's also due to the version of cuda.

[ljyang17]

Doing something similar to this link solved the issue for me. I'm compiling with CUDA 9.2 as well. However, still haven't found a solution of the error from kangaroo_smat.

[jkobject]

Hello ljyang.

Could you explain what exactly you did to get it to work?

Best,