stable-diffusion-webui-rocm
stable-diffusion-webui-rocm copied to clipboard
segfault on RDNA3
using this as an eval of RDNA3 on ROCm 5.4.3. other users report RDNA3 is not well supported, so sharing here.
Describe the bug stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 exited with code 139
To Reproduce
- install RDNA3 GPU
- build.sh
- docker-compose up
Expected behavior UI launches
Container Output
[jem@pabroni stable-diffusion-webui-rocm]$ docker-compose up
[+] Running 1/0
✔ Container stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 Created 0.0s
Attaching to stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 | Python 3.10.6 (main, May 29 2023, 11:10:38) [GCC 11.3.0]
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 | Commit hash: d0994407921d80d6531526cce545afbc204dba80
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 | Installing requirements
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 | Launching Web UI with arguments: --port 7860
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 | No module 'xformers'. Proceeding without it.
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 | Loading weights [6ce0161689] from /sd/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 | Creating model from config: /sd/configs/v1-inference.yaml
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 | LatentDiffusion: Running in eps-prediction mode
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 | DiffusionWrapper has 859.52 M params.
stable-diffusion-webui-rocm-stable-diffusion-webui-rocm-1 exited with code 139
dmesg logs
[ 251.924718] amdgpu: Failed to pin bo. ret -1
[ 251.924725] amdgpu: Failed to map wptr bo to GART
[ 251.935580] amdgpu: Failed to pin bo. ret -1
[ 251.935585] amdgpu: Failed to map wptr bo to GART
[ 251.940877] amdgpu: Failed to pin bo. ret -1
[ 251.940880] amdgpu: Failed to map wptr bo to GART
[ 251.946127] amdgpu: Failed to pin bo. ret -1
[ 251.946131] amdgpu: Failed to map wptr bo to GART
[ 251.951397] amdgpu: Failed to pin bo. ret -1
[ 251.951401] amdgpu: Failed to map wptr bo to GART
[ 251.956657] amdgpu: Failed to pin bo. ret -1
[ 251.956661] amdgpu: Failed to map wptr bo to GART
[ 251.961917] amdgpu: Failed to pin bo. ret -1
[ 251.961921] amdgpu: Failed to map wptr bo to GART
[ 251.967154] amdgpu: Failed to pin bo. ret -1
[ 251.967157] amdgpu: Failed to map wptr bo to GART
[ 251.972390] amdgpu: Failed to pin bo. ret -1
[ 251.972393] amdgpu: Failed to map wptr bo to GART
[ 251.973594] python[2037]: segfault at 100 ip 00007f9130f2d282 sp 00007ffed8d06f00 error 4 in libamdhip64.so[7f9130e00000+3f3000] likely on CPU 27 (core 27, socket 0)
[ 251.973608] Code: 00 00 00 00 e8 9f 95 ff ff 48 8d 74 24 20 31 d2 4c 89 f7 48 89 c5 48 c7 44 24 20 00 00 00 00 e8 84 95 ff ff 48 85 c0 48 89 c3 <49> 8b 94 24 00 01 00 00 0f 95 c0 48 85 ed 0f 85 a2 01 00 00 84 c0
Desktop (please complete the following information):
- EPYC 7551P 128G RAM
- AMD GPU & VRAM: 7900 XTX 24GB
- OS + Distro and Version: arch 6.3.9-arch1-1
- Host ROCm Version: rocm 5.4.3-1
- Version ac42b362803a9e3dd09d03cc501932fa968570fa