Visual Studio 2008 - debug release different exact same code, different result

Hi everyone,

I have stumbled upon a strange behavior with my CUDA code: the results are different for debug and release mode of Visual Studio 2008. The emulate mode is not used in both cases.

I have only made the changes described in this page:…t-cuda-program/

Did anyone have this problem for debug/release differences? What other parameters could influence results or is the difference normal (I hope not… I’d like to have something valid running on GPU for my release version)?

Now investigating the custom build rules of my visual studio 2008.

I was editing the nvcc command line by hand for each .cu files . But since this problem in release mode, I tried the cuda sdk programs in release mode and there’s no problem. The difference is they have Cuda Build Rules v2.1.0 in their properties list for their .cu files. How to have the same in visual studio 2008? There are no helpfull explanations by nvidia and I didn’t find any forums that explain it either.

Thanks for any help!!

Found what was keeping CUDA from giving the same results in release mode: the use of OpenMP. They are incompatible, I had to disble OpenMP in Visual Studio for the CUDA code to work properly.