openAcc Error: cuda_enter.c: No such file or directory

I’m facing problem while running a code of matrix multiplication using openacc on GPU.
Following is the execution trace from cuda-gdb. I couldn’t locate cuda_enter.c on my machine. My machine configuration is : GPU= Kepler GTX690 (K10), CUDA 5.0, Ubuntu11.10, Xeon 64bit processor.


Starting program: /home/arun/SW_vaibhav/openACC/mm
[Thread debugging using libthread_db enabled]
warning: Cuda API error detected: cuCtxAttach returned (0xc9)

[New Thread 0x7ffff5ea7700 (LWP 21379)]
[Context Create of context 0x766c60 on Device 0]

Program received signal SIGSEGV, Segmentation fault.
0x0000000000405172 in __pgi_uacc_cuda_enter (rversion=0x7fffffffdda8, objinfo=0x61d720, dindex=1) at cuda_enter.c:137
137 cuda_enter.c: No such file or directory.
in cuda_enter.c
(cuda-gdb) q
A debugging session is active.

Inferior 1 [process 21373] will be killed.


Thanks in Advance
Arun

Hi Arun,

cuda_enter.c is part of the CUDA run time library so not a file you’ll be able to debug. Though, this indicates that there’s something wrong with your configuration and that the PGI run time isn’t able to create a device context.

Are you able to run a CUDA C program? What is the output from the “pgaccelinfo” utility?

  • Mat

Hi Arun,

I was wrong. What’s happening is that the compiler is failing to generate a kernel but the run time is still trying to launch it. Can you post the “-Minfo=accel” output from your compile and then add the flag “-ta=nvidia,required”?

Also, can you please send to PGI Customer Service (trs@pgroup.com) a reproducing example?

Thanks,
Mat