error: no instance of overloaded function "atomicAdd" matches the argument list
yukaizhou opened this issue · 5 comments
Hi，Encountered the following questions when configuring the environment, and look forward to your answers.
-D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_52,code=sm_52 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=sm_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_86,code=sm_86 -std=c++14
/dfs/data/code_python/detection_3d/BEVDepth_sec/bevdepth/ops/voxel_pooling_train/src/voxel_pooling_train_forward_cuda.cu(60): error: no instance of overloaded function "atomicAdd" matches the argument list
argument types are: (half *, const half)
detected during instantiation of "void voxel_pooling_train_forward_kernel(int, int, int, int, int, int, const int *, const T *, T *, int *) [with T=half]"
(99): here
1 error detected in the compilation of "/dfs/data/code_python/detection_3d/BEVDepth_sec/bevdepth/ops/voxel_pooling_train/src/voxel_pooling_train_forward_cuda.cu".
Hello, have you solved this problem?
Did anyone solve this problem? I try all versions of Torch but still have this problem.
same error. any suggestion?
One suggestion, use bevdet-depth config in bevdet repo.
I sloved this by setting export TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 8.6+PTX" before use python setup.py develop
I think it is because the "atomicAdd" with Half type only support the GPU with 7.x+ Compute Capability