Java lookup error

Hi, all!
I used jcuda ver 0.3.0? for long time. Now I decide update to 0.3.2a, but get runtime error:

**/usr/lib/jvm/sun-jdk-1.6/bin/java: symbol lookup error: libJCudaDriver-linux-x86_64.so: undefined symbol: cuInit.

**The reason of this error is “JCudaDriver.cuInit(0)” in my source code.
I need in big memory alocation for data arrays (more 2Gb for data array) and I need in new jcuda version.

My system is Gentoo 2.6.37 x86_64 Intel

Help me, please.

Hello,

Can I assume that you have installed the appropriate CUDA Driver and Toolkit, namely version 3.2 (build 16), and that there are no old installations which might interfere with the new one?

In any case, I contacted the contributor who created these binaries, maybe he has an idea what might be wrong there.

bye
Marco

Hello!

I use

  1. x11-drivers/nvidia-drivers-260.19.36
  2. dev-util/nvidia-cuda-toolkit-3.2

I have no another verison of the nvidia driver or toolkit.

Also I try used jcuda-0.3.2 drivers (not 0.3.2a) and my program work fine!

Bye.

[QUOTE=Unregistered]Also I try used jcuda-0.3.2 drivers (not 0.3.2a) and my program work fine!
[/QUOTE]

So this might at least be a preliminary solution, athough the Bugfix that was added between 0.3.2 and 0.3.2a only referred to the aspect that you mentioned: Proper handling of large memory blocks. This bug was reported in this thread about cuDeviceGetTotalMem . Does cuDeviceGetTotalMem return the proper values with the 0.3.2 (NOT ‘a’) version on Linux64?

bye

With jcuda-0.3.2:

   
   public static long getDeviceTotalMem(final CUdevice dev) {         
     final long[] bytes = new long[1];
     CudaUtils.checkCudaError(JCudaDriver.cuDeviceTotalMem(bytes, dev));
     return bytes[0];     
   }

result:

name=Tesla C1060
maxDeviceTotalMem=4294770688
name=Tesla C2070
maxDeviceTotalMem=6441992192

[QUOTE=Unregistered]With jcuda-0.3.2:

name=Tesla C1060
maxDeviceTotalMem=4294770688
name=Tesla C2070
maxDeviceTotalMem=6441992192[/QUOTE]

I have 4Gb on C1060 and 6 on C2070

That’s strange - maybe the bug was specific for Win64 then…? Or the original contributor of version 0.3.2 fixed it silently for the compilation :wink: So you may stick to 0.3.2, and I’ll try to find out what’s wrong with 0.3.2a.

This thread http://forum.byte-welt.de/showthread.php?t=4038 probably might provide a solution to this problem.