Hi all,
I have two minor grumps with the 3.2 SDK:
- Experimenting with the 3.2 version of cudaopenmp, I tried setting /openmp as an extra C++ option in the host tab of the Runtime API options.
This did not work correctly, because /openmp was added to the options of NVCC as well as to the options of CL.exe (visible in the commandline, after -Xcompiler).
NVCC then produces an error that it needs a single target (if I remember correctly), anyhow, it does not compile.
I made a work-around by putting the kernel in a separate .cu file and adding a wrapper, which gets called from cudaOpenMP.cpp (rename).
That allows to set the openmp option in the VS C++ configuration properties.
AFAIK this host-option problem exists also in earlier toolkit versions.
- After installing (the 64 bit installer) of the SDK, VFlocking.h was not present, so I couldn’t build this interesting example.
Hi all,
I have two minor grumps with the 3.2 SDK:
- Experimenting with the 3.2 version of cudaopenmp, I tried setting /openmp as an extra C++ option in the host tab of the Runtime API options.
This did not work correctly, because /openmp was added to the options of NVCC as well as to the options of CL.exe (visible in the commandline, after -Xcompiler).
NVCC then produces an error that it needs a single target (if I remember correctly), anyhow, it does not compile.
I made a work-around by putting the kernel in a separate .cu file and adding a wrapper, which gets called from cudaOpenMP.cpp (rename).
That allows to set the openmp option in the VS C++ configuration properties.
AFAIK this host-option problem exists also in earlier toolkit versions.
- After installing (the 64 bit installer) of the SDK, VFlocking.h was not present, so I couldn’t build this interesting example.
to 1.:
Solution without separate .cu file