I have a program which results in an “Invalid global read” when I run it with standalone cuda-memcheck from the command line. If I run the program from cuda-gdb and “set cuda memcheck on” the same program runs fine and does not show any errors.
Thus, I wonder if there are any differences between standalone and integrated mode of memcheck? I did compile the code with “-g -G” parameters. Is there anything else I need to do to get the same behavior?
Providing a minimal working example is a bit tricky, but if this turns out to be unintended behavior I will try to come up with one.
PS. I searched the forum, but did not find any similar questions. If this a duplicate question, sorry for that.