CUDA 1.1 SDK on openSUSE 10.2 Compilation problems

Accelerated Computing CUDA CUDA Programming and Performance
1

I could install CUDA on my opensuse 10.2 box without problems. The nvidia driver correctly determines that I’m running a Tesla/8800. The problem is when I try to compile the SDK. I get a lot of errors in compilation which look like the following
( and are spit out by nvcc ):

(I am listing the last 3)

“/usr/include/c++/4.2.3/cstdlib”, line 217: error: name followed by “::” must
be a class or namespace name
using ::__gnu_cxx::strtoll;
^

“/usr/include/c++/4.2.3/cstdlib”, line 218: error: name followed by “::” must
be a class or namespace name
using ::__gnu_cxx::strtoull;

“/usr/include/c++/4.2.3/cstdlib”, line 219: error: name followed by “::” must
be a class or namespace name
using ::__gnu_cxx::strtold;
^

86 errors detected in the compilation of “/tmp/tmpxft_000011f7_00000000-2.ii”.
make[1]: *** [obj/release/imageDenoising.cu_o] Error 255
make[1]: Leaving directory `/root/NVIDIA_CUDA_SDK/projects/imageDenoising’
make: *** [projects/imageDenoising/Makefile.ph_build] Error 2

Anyone has seen this before? How can I fix this?

Thanks in advance for the help.

2

Please ignore my last post. I fixed it.

It was caused because CUDA does not compile with gcc 4.2. I downgraded to gcc 4.1

and it compiled…I would urge nvidia to make their code 4.2 compliant…:)

3

As far as I can tell, SUSE-10.2 didn’t ship with gcc-4.2, and none of the Linux distributions qualified against CUDA_1.1 shipped with gcc-4.2. Where did you obtain the gcc-4.2 binaries that you’re using?

The next CUDA release will include distributions which shipped with gcc-4.2.