Intel Compiler (original post in the Mac OS X section)

Can anyone give me an idea of why NVIDIA doesn’t currently support the Intel Compiler for CUDA? It seems apparent that there should be some compiler settings that would enable the Intel code to work correctly, presuming that GCC does.

If anyone knows how to get this compiler to support CUDA, your help would be greatly appreciated.

One way that I’ve tried to work around this is to create a dynamic library (that performs all of the CUDA operations, compiled with gcc) and then link against the rest of my code (which is compiled with ICC). However, I’m getting linking errors when I try to use functions from the CUDA dynlib, and I haven’t had much luck resolving those.

I’m very new to using the Intel Compiler, so perhaps my settings are off.

Any help is appreciated.

I have the project that uses the Intel Compiler for almost everything. The only thing that gets compiled with CL(Windows), and GCC(Linux) is the host wrapper in the .cu file.

Basically, put as little code in as possible in the .cu, use nvcc to genarate an .obj or .o file, compile your .c and .cpp files with Intel’s compiler, then link everything together using Intel’s linker. Intel did a smart move and made its Windows compiler pint by pint compatible with Visual Studio, and its Linux compiler pint by pint compatible with GCC.
Check the link in my signature for more details.

Have you tried the CUDA CMake script? Just run ccmake and change the compiler – this works fine on my system.