thu-ml / thu-ml/TurboDiffusion
AMD/ROCm Support Request - CUTLASS dependency blocks HIP compilation
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 3.7k
- Forks
- 277
- Avg merge
- 2h 57m
- Merged PRs (30d)
- 2
Description
Environment
- ROCm 7.11
- PyTorch 2.9.1+rocm7.11
- AMD Strix Halo (gfx1151)
- Ubuntu 25.10
What we did
Patched setup.py for HIP compilation:
- Changed nvcc flags to hipcc-compatible flags
- Removed -gencode, --expt-*, --threads
- Added --offload-arch=gfx1151
- Changed libraries=["cuda"] to libraries=["amdhip64"]
- Added /opt/rocm/include for thrust headers
Result
Compilation proceeds until CUTLASS, which requires cuda/std/* headers from libcu++ (NVIDIA-only):
equest
Consider hipBLASLt or AMD composable_kernel as GEMM alternative to enable AMD/ROCm support.
10x Wan acceleration on AMD hardware would be huge for the community.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with setup.py and trace the CUTLASS dependency that requires cuda/std/* headers. Review the existing HIP compilation changes and evaluate the issue's proposed hipBLASLt or AMD composable_kernel alternatives. Done means establishing a supported AMD/ROCm compilation path for gfx1151 without the NVIDIA-only CUTLASS dependency.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- ai, performance
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100