Success on OpenSuse 11 With a list of fixes

Hi,

I just got CUDA running on OpenSuse 11 (64 bit) - all issues had known fixes (see link, near end) that I found via Google except for one, which was that the Perl script for the SDK was buffering output intended for the screen.

Since writing that I’ve run almost all the demos with no problems.

Cheers,
Andrew

It works great! Thank you. I only had to delete the project/threadMigration folder and it compiled without any further problems on Suse11 x86_64. :cool:

what was the problem with threadMigration? worked ok for me…

qi6e release: ./threadMigration
threadMigration API test…
1 CUDA devices detected, 2 Threads to launch

Device 0: GeForce 8600 GT
sharedMemPerBlock: 16384
constantMemory : 65536
regsPerBlock : 8192
clockRate : 1188000
ThreadProc() Launched…
ThreadProc() Finished!
ThreadProc() Launched…
ThreadProc() Finished!

Test PASSED

Press ENTER to exit…

This is what gcc complains about threadMigration:

make -C projects/threadMigration/ 

make[1]: Entering directory `/home/tian/NVIDIA_CUDA_SDK/projects/threadMigration'

obj/release/threadMigration.cpp_o: In function `FinalErrorCheck(int, int, int)':

threadMigration.cpp:(.text+0x4f): undefined reference to `cuCtxDestroy'

obj/release/threadMigration.cpp_o: In function `ThreadProc(_CUDAContext_st*)':

threadMigration.cpp:(.text+0xea): undefined reference to `cuCtxPushCurrent'

threadMigration.cpp:(.text+0x1f6): undefined reference to `cuCtxPopCurrent'

obj/release/threadMigration.cpp_o: In function `InitCUDAContext(_CUDAContext_st*, int, int, char*)':

threadMigration.cpp:(.text+0x3f4): undefined reference to `cuCtxPopCurrent'

collect2: ld returned 1 exit status

make[1]: *** [../../bin/linux/release/threadMigration] Fehler 1

make[1]: Leaving directory `/home/tian/NVIDIA_CUDA_SDK/projects/threadMigration'

make: *** [projects/threadMigration/Makefile.ph_build] Fehler 2

Strange, but as I said, that’s the only problem.

The small problem with threadMigration is stille a mystery to me.

I don’t know why [font=“Courier”]cuCtxDestroy[/font], [font=“Courier”]cuCtxPushCurrent[/font], [font=“Courier”]cuCtxPopCurrent[/font] can’t be found. I grepped the sources (and objdumped the modules, which was clompletely useless), but the results were only cuda.h and some man pages. Google shows me that several people have problems linking to [font=“Courier”]cuCtxDestroy[/font], etc. The interesting thing is that Andrew had no problems with this. Maybe there are subtle differences in the configurations?

Here is mine:
NVIDIA-Linux-x86_64-173.08-pkg2
NVIDIA_CUDA_Toolkit_2.0beta2_Suse10.3_x86_64
sdk-patched :smile:
gcc 4.3.1_20080507-6.1 (x86_64)
g++ 4.3.1_20080507-6.1 (x86_64)

sorry, forgot about this thread. these are all teh nvidia and gcc packages i have installed (i assume!):

qi6e efe: rpm -qa | egrep -i “(nvidia|gcc)”
libgcc43-4.3.1_20080507-6.1
gcc43-c+±4.3.1_20080507-6.1
nvidia-gfxG01-kmp-default-173.14.09_2.6.25.5_1.1-0.1
gcc43-fortran-32bit-4.3.1_20080507-6.1
gcc-c+±4.3-39.1
gcc43-fortran-4.3.1_20080507-6.1
libgcc43-32bit-4.3.1_20080507-6.1
gcc-4.3-39.1
gcc-fortran-32bit-4.3-39.1
x11-video-nvidiaG01-173.14.09-0.1
gcc-fortran-4.3-39.1
gcc-32bit-4.3-39.1
gcc43-4.3.1_20080507-6.1
gcc43-32bit-4.3.1_20080507-6.1

…and the nvidia packages (which aren’t rpms):
NVIDIA_CUDA_sdk_2.0beta2_linux.run (this is what i patched, unless i am hopelessly confused)
NVIDIA_CUDA_Toolkit_2.0beta2_Suse10.3_x86_64.run
NVIDIA-Linux-x86_64-177.13-pkg2.run (this is the driver)

that makes sense right? my memory is terrible, so i am just listing the files on my disk…

FYI

The CUDA 2.1 beta is running on openSuSE 11.1 (x86_64), too. I got some simple errors during the installation, but you can always find fast help via google.