Hello there,
I have been struggling with the vectorAddDrv example from CUDA sdk version 3.1. The example code works perfactly with the way it is setup by default to use <data/vectorAdd.ptx> module. But I wanted to check the code with .cubin file and it fails with segmentation fault.
result with ptx file:
[tvsingh@fuji working]$ /usr/local/cuda/sdk/C/bin/linux/release/vectorAddDrv
Vector Addition (Driver API)
findModulePath found file at </usr/local/cuda/sdk/C/bin/linux/release/…/…/…/src/vectorAddDrv/data/vectorAdd.ptx>
initCUDA loading module: </usr/local/cuda/sdk/C/bin/linux/release/…/…/…/src/vectorAddDrv/data/vectorAdd.ptx>
PTX JIT log:
: Considering profile ‘compute_20’ for gpu=‘sm_20’ in ‘cuModuleLoadDataEx_5’
PASSED
Press ENTER to exit…
result with cubin file without any other change:[tvsingh@fuji working]$ ./vectorAddDrv
Vector Addition (Driver API)
Segmentation fault
[tvsingh@fuji working]$ ls data/
vectorAdd.cubin
[tvsingh@fuji working]$ ls -l data/
total 4
-rw-r–r-- 1 tvsingh staff 1152 Sep 16 14:39 vectorAdd.cubin
I started looking at this example because of the similar behavior with other cubin file, that I wanted to load as module but was failing with same situation. The configuration of the system is as follows:
CUDA 3.1
Nvidia Driver: 258.19
Operating system: Linux-x86_64
Any help will be appreciated.
Thanks in advance,
TV