I am working with a large set of code with mixed c / c++ / cuda source files that I am trying to build and am getting an error on the cuda front end (cudafe++) step. The error is in the include file opt_random.h where it gives a number of “error identifier” messages for Intel intrinsics (_mm_set_epi64x, _mm_and_si128, mm_mul_pd, etc).
As background, I have a custom set of build scripts to allow compiling as CPU only or with CUDA GPU support. When compiling without CUDA, the build system ignores .cu files and uses the Intel compiler (parallel studio XE). When compiling with GPU enabled, it uses nvcc and passes the Intel compiler with the ccbin option. Because I have some code that I need to compile as host device, I put those in .c files so they build for the host in either CPU/GPU build. To get this to work, I include the -x cu option to nvcc to compile all code as if it were cuda file. The error only occurs when compiling with nvcc. I used the -dryrun nvcc command to isolate the build steps and most of them are successful.
I know this is not a lot to go on, but wanted to see if anyone has encountered a similar issue.