Hello,
I only have 1 card in 1 machine. How do I debug CUDA code? It doesn’t need to be fancy, as long as it works!~~!!.
Thanks
Hello,
I only have 1 card in 1 machine. How do I debug CUDA code? It doesn’t need to be fancy, as long as it works!~~!!.
Thanks
I found this article.
This may seem quaint and old-fashioned to some: I do all of my CUDA debugging by using printf() to capture relevant data in logs. This is sometimes coupled with targeted modifications to simplify input data or simplify code.
Beyond that I can only re-iterate what has been frequently stated in these forums in reply to requests for assistance in debugging CUDA code: Check the return code of every CUDA API call and every kernel launch.
Hello,
You mean printf or cuPrintf ? cuPrintf is pretty amazing…~!! Never heard of it until today
Ahh, printf is actually supported in CUDA. I have a 2.1 card, Yah!~!
Thank you so much. I should read the CUDA C_Programming Guide more carefully.
Here is another Debugging CUDA
Bill
If you happen to have on-board graphics, you can use that for the display and use Nvidia’s parallel nsight. Or you can use gdb under linux but x-server can’t be running.