Compile CUDA C++ to execute on manycore host

I am interested in compiling existing CUDA C++ code to execute on a manycore host (Xeon Phi 7210 and 7250).

I see in the documentation that there are means to do this for CUDA Fortran, but no mention of C++. Does this functionality exist, and if not, is there any alternative workaround?

This is inherited code, and there’s no real option to run on anything besides KNL. To make matters worse, I’m not comfortable with Fortran, so translating isn’t a possibility.

Thanks in advance

Hi kachind,

CUDA use to support emulation mode (circa 2011) but that was dropped after CUDA 3.1. CUDA Fortran use to as well, but was dropped not long after CUDA C stopped supporting it.

For this scenario, you’re better off using OpenACC which does allow for targeting both NVIDA GPUs as well as Multicore CPU devices.