2. Change set(CMAKE_CUDA_ARCHITECTURES 89) in CMakeLists.txt
3. Run cmake from cmd(Not powershell! Powershell does not work for `set(CMAKE_CUDA_ARCHITECTURES 89).`)
4. We need 7 dlls in `cpp\mgcg_cuda\lib`:
1. cublas64_12.dll
2. cublasLt64_12.dll
3. cusparse64_12.dll
4. nvJitLink_120_0.dll
5. fast-vcycle-gpu.dll
6. fast-vcycle-gpu.exp
7. fast-vcycle-gpu.lib
First 4 files are from `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.5\bin`. They are used by cublas and cusparse. The last 3 files are our own dlls. They are generated by cmake.
Extra Notes:
- Either copy the four cuda dlls to `cpp\mgcg_cuda\lib` or use `os.add_dll_directory("C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.5/bin")`.
- In Windows, use dumpbin /dependents to check the dependencies of a dll. For example, `dumpbin /dependents fast-vcycle-gpu.dll` will show the dependencies of fast-vcycle-gpu.dll.
Download(600MB):
https://bhpan.buaa.edu.cn/link/AA1B784714F4D846E59851D1B12E83196D
or
https://www.dropbox.com/scl/fi/tvpr3g3btjca0maaz6gfl/dlls.zip?rlkey=68u687oghe5i7tddvedggcy7l&st=eo8dg8cd&dl=0
AMGCG
run script/whole_amg_solver.py for separate solving Ax=b
run engin/cloth/cloth3d.py by python run.py -case 7 see a cloth simulation