Notice: The following uses compute_30 as an example, and the same applies to others
- Reason: compute_30 has been deprecated in cuda11
Another method is to delete the corresponding code in CMakeLists or makefile, but ①makefile will be overwritten in cmake, and directly modifying makefile is not elegant; ②ElastcFusion's CMakeLists does not manually specify compute_30 but reads the code. I searched for it myself, and the solution is as follows.
-
Solution: modify
Core/src/CudaComputeTargetFlags.cmake
, add the following contentif(ARCH MATCHES "30") continue() endif()
参考内容:
nvcc fatal : Unsupported gpu architecture ‘compute_30‘
nvcc fatal : Unsupported gpu architecture ‘compute_‘