Unable to Run Genesis on AMD Device with GPU Backend

[Burnling-gx]

trafficstars

I attempted to run the following code:

import os
os.environ['PYOPENGL_PLATFORM'] = 'glx'
 
import genesis as gs

print(gs.get_gpu_device())
gs.init(backend=gs.gpu)

scene = gs.Scene(show_viewer=True)
plane = scene.add_entity(gs.morphs.Plane())
franka = scene.add_entity(
    gs.morphs.MJCF(file='xml/franka_emika_panda/panda.xml'),
)

scene.build()

for i in range(1000):
    scene.step()

The output I received is as follows:

(device(type='cuda'), 'AMD Radeon Graphics', 7.464061737060547)
[Genesis] [10:19:01] [INFO] ╭─────────────────────────────────────────────────────────────────────────────────────╮
[Genesis] [10:19:01] [INFO] │┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉ Genesis ┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉┈┉│
[Genesis] [10:19:01] [INFO] ╰─────────────────────────────────────────────────────────────────────────────────────╯
[Genesis] [10:19:01] [INFO] Running on [AMD Radeon Graphics] with backend gs.cuda. Device memory: 7.46 GB.
[W 12/22/24 10:19:01.532 18257] [cuda_driver.cpp:load_lib@36] libcuda.so lib not found.
[W 12/22/24 10:19:01.532 18257] [misc.py:adaptive_arch_select@758] Arch=[<Arch.cuda: 3>] is not supported, falling back to CPU
[Genesis] [10:19:01] [INFO] 🚀 Genesis initialized. 🔖 version: 0.2.0, 🌱 seed: None, 📏 precision: '32', 🐛 debug: False, 🎨 theme: 'dark'.
[Genesis] [10:19:01] [INFO] Scene <679c1fa> created.
[Genesis] [10:19:01] [INFO] Adding <gs.RigidEntity>. idx: 0, uid: <95b9c77>, morph: <gs.morphs.Plane>, material: <gs.materials.Rigid>.
[Genesis] [10:19:01] [INFO] Adding <gs.RigidEntity>. idx: 1, uid: <6e9f60f>, morph: <gs.morphs.MJCF(file='/home/burnling/anaconda3/envs/genesis/lib/python3.10/site-packages/genesis/assets/xml/franka_emika_panda/panda.xml')>, material: <gs.materials.Rigid>.
[Genesis] [10:19:02] [INFO] Building scene <679c1fa>...
Traceback (most recent call last):
  File "/home/burnling/Code/Genesis Learning/hello_genesis.py", line 15, in <module>
    scene.build()
  File "/home/burnling/anaconda3/envs/genesis/lib/python3.10/site-packages/genesis/utils/misc.py", line 38, in wrapper
    return method(self, *args, **kwargs)
  File "/home/burnling/anaconda3/envs/genesis/lib/python3.10/site-packages/genesis/engine/scene.py", line 539, in build
    self._parallelize(n_envs, env_spacing, n_envs_per_row, center_envs_at_origin)
  File "/home/burnling/anaconda3/envs/genesis/lib/python3.10/site-packages/genesis/engine/scene.py", line 576, in _parallelize
    self._envs_idx = torch.arange(self._B, dtype=gs.tc_int, device=gs.device)
RuntimeError: HIP error: invalid device function
HIP kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing AMD_SERIALIZE_KERNEL=3
Compile with `TORCH_USE_HIP_DSA` to enable device-side assertions.

[Genesis] [10:19:02] [INFO] 💤 Exiting Genesis and caching compiled kernels...

I noticed the device(type='cuda') in the output. I also tried switching the backend to gs.cuda explicitly, but the same issue occurred.

Upon reviewing the constants.py source code, I found this mapping

"Linux": {
        backend.cpu: backend.cpu,
        backend.gpu: backend.cuda,
        backend.cuda: backend.cuda,
        backend.vulkan: backend.vulkan,
    },

Based on this, it seems that backend.gpu defaults to backend.cuda, which might not be suitable for AMD GPUs.

How can I properly run Genesis on an AMD GPU? Is there a way to configure the backend for HIP or ROCm instead of CUDA?