Coredump on CUDBG_EXCEPTION_WARP_ILLEGAL_ADDRESS dont show lane/warp info

Developer Tools CUDA Developer Tools CUDA-GDB
1

Platform: windows 11
$ nvidia-smi.exe
Wed Sep 10 11:01:46 2025
±----------------------------------------------------------------------------------------+
| NVIDIA-SMI 581.15 Driver Version: 581.15 CUDA Version: 13.0 |
±----------------------------------------±-----------------------±---------------------+
| GPU Name Driver-Model | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 3050 WDDM | 00000000:01:00.0 Off | N/A |
| 30% 30C P0 22W / 70W | 0MiB / 6144MiB | 0% Default |
| | | N/A |
±----------------------------------------±-----------------------±---------------------+

±----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| No running processes found |
±----------------------------------------------------------------------------------------+

env variables set:
CUDA_ENABLE_COREDUMP_ON_EXCEPTION=1
CUDA_COREDUMP_SHOW_PROGRESS=1

executable contains “-G -g“ flags.

with this build config, I can see lane/warp info on CUDBG_EXCEPTION_NONE coredump but cannot see any debug information on CUDBG_EXCEPTION_WARP_ILLEGAL_ADDRESS.

using visual studio 2019 with updated Nsight extensions to open coredumps.

2

Hi, @goutham24693

Can you please provide a mini repro for us to investigate this issue ?

3

Able to reproduce this issue when kernel input argument is invalid.
if input argument to kernel is allocated in host instead of device, I get CUDBG_EXCEPTION_NONE.

I dont have optimization on my build and I have added debug symbols on host and device. Still I am not able to get any warp/lane info & state from my coredump.

4

Hi @veraj still facing this issue. is there a solution for this ?

5

Can you please provide the coredump file?
Can you also check whether adding a while(1) before CUDA kernel exit obtains the required warp information ？