cudaMalloc returns unknown error

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

Hi,
I have an application that uses multiple shared libraries. The libraries contain some custom code as well as code from other static libraries. To avoid conflicts, I want to hide the symbols of theses static ibraries using the “-Wl,–exclude-libs,ALL” linker flags. Unfortunately, when I use these flags, the first call of cudaMalloc returns an unknown error. Without these flags, everything is working fine. I tried it also on a PC and there it works with and without the flags. cuda-memcheck and cuda-gdb also give no additional information. Any idea what could be the reason for this?

2

Hi,

Maybe the config causes some link error in CUDA.
Is it possible to list the library name rather than use “-Wl,–exclude-libs,ALL”?

Thanks.

3

Hi AastaLLL,
yes it works if I list the library names. Unfortunately I have to list a huge number of libs, therefore I would prefer to use “-Wl,–exclude-libs,ALL”. It is weird that it works on the PC. The only difference is, that I use libargus on the TX2. Could it be that this influences somehow the CUDA memory management?

4

Hi,

Do you mind to write a simple example for reproducing this issue?
We want to deliver it to our internal team to get some advice.

Thanks.