seg fault under Cuda 2.3 My code worked for 6 months under 2.2!

Accelerated Computing CUDA CUDA Programming and Performance
1

I made some changes to my computer, and now a piece of code that I know is correct immediately returns a segmentation fault. Has anybody encountered this problem? Here are the changes I made:

The first change I made was that I upgraded from Ubuntu 8.04 to 9.04, and consequently updgraded from Cuda 2.2 to 2.3.

I upgraded my graphics display card from an X1650 to a GT 9400. Could my code be trying to run on my display card? My Tesla card is device0, which is the default right? How would I go about manually specifying which card I want my code to run on?

deviceQuery and bandwidthTest run perfectly. My PATH and LD_LIBRARY_PATH is correct (I set these through bash.bashrc)

Thanks in advance!

2

Calling cudaSetDevice() before any other cuda function will let you pick the device number. What order does deviceQuery list your CUDA devices in?

3

I’m not at my computer now, but as I recall, deviceQuery says the Tesla is device0. Since my post, I found out about the --device= flag, so I used this, and it didn’t solve the problem.

Just to clarify, I should say that all of the SDK samples I’ve tried work, but all of my codes (there are several) that worked in 2.2 do not work in 2.3.

4

I’ve fixed the problem - it was a stupid coding mistake :">

The first line of my code opens a file that I write data to, and the folder containing that file no longer exists because of my OS upgrade, hence the segmentation fault.