cuda-gdb crash when stepping into kernel in Nsight Eclipse Edition

[jhanweck]

Hi, I’m getting this error when trying to step into a kernel (mine) using Nsight. Similar problems regarding any of my kernels I try to debug. However, I can debug Sample code kernels just fine.

cuda-gdb/7.12/gdb/dwarf2-frame.c:1772: internal-error: add_cie: Assertion `n < 1 || cie_table->entries[n - 1]->cie_pointer < cie->cie_pointer’ failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
Quit this debugging session?
This is a bug, please report it.

Details:

* RedHat Linux 7.5 on AWS p3.2xlarge (x86_64 w/ V100 GPU) * Nsight Eclipse Edition v. 9.2.201806130420 (running Nsight on the server remotely from my Windows desktop via MobaXterm SSH X-windows tunneling.. Could this be the problem? ) * Running debugger via Nsight * NVIDIA-SMI 396.44 Driver Version: 396.44 | * Stepping through the program using Nsight debugger, trying to step into the kernel. (Note: this works on Sample code, e.g., NVIDIA_CUDA-9.2_Samples/6_Advanced/cdpLUDecomposition, just not on my kernel) * CUDA V9.2.148 * NVIDIA (R) CUDA Debugger 9.2 release / GNU gdb (GDB) 7.12 * Program is compiled in Nsight using Separate Compilation, Compute Capability 7.0, -G, -g, -O0. The kernel-portion is compiled into an .so, and linked into a separate main() app. It runs just fine. I simply cannot step into or break on any kernels.

Thanks,
Jerry

[kjalaludeen]

Can you try to debug your program outside of Nsight using cuda-gdb command line?

Please attach the output from cuda-gdb command line when the crash occurs.