the method of compiling pointops_cuda(cuda==11.1 GPU==RTX3090)

[Roger-Ni]

when compile pointops_cuda，I have the error: nvcc fatal : Unsupported gpu architecture 'compute_86'
solution：modify the setup.py as is shown in the below image（of course，you should change the py file according to your cuda version. As my cuda==11.1，GPU==RTX 3090, solutions on the web suggested that I should add command "export TORCH_CUDA_ARCH_LIST=“8.0”" into ～/.bashrc. However, it does not work）

as is known from the web, RTX 3090's cuda frame is sm_86 (or compute_86), so I changed the setup.py as is shown in the below image(add -arch=’sm_86’ or -arch=’compute_86’), but I always have the error abot 'nvcc compute_86'. As is suggested by GPT，I add the command -arch=’compute_75’ which is more common. Then the compilation works(python3 setup.py install）.