How to use Intel C++ Compiler for CUDA applications ?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi All,

I’m trying to speed my project up using Intel C++ Compiler instead of MSVC. There are two *.cu files in the project so they must be handled with nvcc. When compiling with MSVC both *.cu files are handled properly but it looks like nvcc is invoked with bad parameters when trying to compile the project with Intel Compiler.

It is possible to compile *.cu files using MSVC and then compile all the rest files using Intel Compiler, that’s actually what I’m currently doing. This approach lets the project to be compiled and linked but resulting application is very instable when compiling even with lightest optimization options, when optimizations are off everything works just fine.

So, this is the question: how to build the project with *.cu files correctly when using Intel C++ Compiler ? I’m using Visual Studio 2005 and 11.1 Intel C++ Compiler that is integrated into the IDE.

Thanks in advance for any suggestions!

2

Could you separate device code into .cu file(s), and all the rest in .c (or .cpp) for Intel C++?
Let nvcc process .cu files into ELF modules, which you can load and call kernels from using driver API.
A crude “solution”, but I don’t know better.

3

Could you separate device code into .cu file(s), and all the rest in .c (or .cpp) for Intel C++?
Let nvcc process .cu files into ELF modules, which you can load and call kernels from using driver API.
A crude “solution”, but I don’t know better.