Bug Description
Bug Description
I am testing Genesis's RayTracer / LuisaRender CUDA rendering path on Ubuntu 24.04 with an RTX 5090. Genesis initializes correctly, the scene is created successfully, and then the process aborts during shutdown while LuisaRender is caching compiled kernels.
The failure happens after a minimal RayTracer scene is constructed with no viewer window. I am using this setup to verify that the LuisaRender backend works correctly in my Genesis environment before moving on to more complex simulation and rendering workflows.
The crash is reproducible and does not appear to be caused by my Python code or by scene construction. Instead, the program terminates inside LuisaRender's CUDA backend during cleanup with:
fmt::v11::format_error: string pointer is null
Steps to Reproduce
import genesis as gs
gs.init(backend=gs.gpu)
scene = gs.Scene(
renderer=gs.renderers.RayTracer(
env_surface=gs.surfaces.Emission(
emissive_texture=gs.textures.ColorTexture(color=(0.2, 0.3, 0.5)),
),
env_radius=20.0,
),
show_viewer=False,
)
### Expected Behavior
Genesis should shut down cleanly without throwing an exception or aborting.
### Screenshots/Videos
_No response_
### Relevant log output
```shell
[Genesis] Running on [NVIDIA GeForce RTX 5090] with backend gs.cuda. Device memory: 31.33 GB.
[Genesis] Genesis initialized.
[2026-06-30 02:28:24.964] [console] [warning] Failed to load built-in kernels: the provided PTX was compiled with an unsupported toolchain. Re-trying with patched PTX version...
[2026-06-30 02:28:24.964] [console] [warning] The PTX version is not supported by the installed CUDA driver. Trying to patch the PTX to make it compatible with the driver.
[2026-06-30 02:28:24.964] [console] [warning] Patch PTX version from 9.3 to 9.0.
[2026-06-30 02:28:25.009] [console] [warning] Found CUDA device runtime library '.../build/bin/libcudadevrt.a', but the driver does not recognize it when adding the kernel PTX. Indirect kernel dispatch will not be available.
[Genesis] Scene <...> created.
[Genesis] Exiting Genesis and caching compiled kernels...
terminate called after throwing an instance of 'fmt::v11::format_error'
what(): string pointer is null
Environment
Environment
OS: Ubuntu 24.04
GPU: NVIDIA GeForce RTX 5090
Driver: 580.159.04
nvidia-smi reports CUDA Version: 13.0
CUDA toolkit: 13.3 (nvcc --version)
PyTorch CUDA: 13.2
Release version or Commit ID
Genesis commit: 8e44876
LuisaRender submodule commit: ac034080af1f537f4243bfa4914dd27adbca3589
Additional Context
git submodule status shows:
genesis/ext/LuisaRender at ac034080af1f537f4243bfa4914dd27adbca3589
The crash happens after successful initialization and scene creation, so this looks like a shutdown/cleanup issue in the CUDA backend rather than a scene construction failure.
Bug Description
Bug Description
I am testing Genesis's RayTracer / LuisaRender CUDA rendering path on Ubuntu 24.04 with an RTX 5090. Genesis initializes correctly, the scene is created successfully, and then the process aborts during shutdown while LuisaRender is caching compiled kernels.
The failure happens after a minimal RayTracer scene is constructed with no viewer window. I am using this setup to verify that the LuisaRender backend works correctly in my Genesis environment before moving on to more complex simulation and rendering workflows.
The crash is reproducible and does not appear to be caused by my Python code or by scene construction. Instead, the program terminates inside LuisaRender's CUDA backend during cleanup with:
fmt::v11::format_error: string pointer is nullSteps to Reproduce
Environment
Environment
OS: Ubuntu 24.04
GPU: NVIDIA GeForce RTX 5090
Driver: 580.159.04
nvidia-smi reports CUDA Version: 13.0
CUDA toolkit: 13.3 (nvcc --version)
PyTorch CUDA: 13.2
Release version or Commit ID
Genesis commit: 8e44876
LuisaRender submodule commit: ac034080af1f537f4243bfa4914dd27adbca3589
Additional Context
git submodule status shows:
genesis/ext/LuisaRender at ac034080af1f537f4243bfa4914dd27adbca3589
The crash happens after successful initialization and scene creation, so this looks like a shutdown/cleanup issue in the CUDA backend rather than a scene construction failure.