[Feature] Support Voxelization with cambricon MLU device

[ZShaopeng]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.
 

Motivation

 
The motivation of the PR is to support running hard_voxelize with Cambricon MLU backend.
 
It includes three parts:
 
    1. Add voxelization_mlu_kernel.mlu src code as MLU kernel.
    2. Add voxelization_mlu.cpp to support launching kernel in PyTorch.
    3. Refactor test_voxelization.py to support testing hard_voxelization with MLU backend.
 

Modification

 

• MLU src code
    Add MLU src code of hard_voxelize in directory mmcv/ops/csrc/common/mlu/voxelization_mlu_kernel.mlu.
• PyTorch adaptation
    Adapt hard_voxelize for PyTorch in mmcv/ops/csrc/pytorch/mlu/voxelization_mlu.cpp.
• Unit test
    Support test hard_voxelize with various backends in tests/test_ops/test_voxelization.py.
   

BC-breaking (Optional)

 
Does the modification introduce changes that break the backward-compatibility of the downstream repositories?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.
 

Use cases (Optional)

 
If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.
 

Checklist

 
Before PR:
 

• I have read and followed the workflow indicated in the CONTRIBUTING.md to create this PR.
• Pre-commit or linting tools indicated in CONTRIBUTING.md are used to fix the potential lint issues.
• Bug fixes are covered by unit tests, the case that causes the bug should be added in the unit tests.
• New functionalities are covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, including docstring or example tutorials.
   
  After PR:
   
• If the modification has potential influence on downstream or other related projects, this PR should be tested with some of those projects, like MMDet or MMCls.
• CLA has been signed and all committers have signed the CLA in this PR.

[grimoire]

Unit test failed with log:

            ext_module.hard_voxelize_forward(
                points,
                torch.tensor(voxel_size, dtype=torch.float),
                torch.tensor(coors_range, dtype=torch.float),
                voxels,
                coors,
                num_points_per_voxel,
                voxel_num,
                max_points=max_points,
                max_voxels=max_voxels,
                NDim=3,
>               deterministic=deterministic)
E           RuntimeError: MLU neg don't support tensor dtype Int.

mmcv/ops/voxelize.py:87: RuntimeError
-------------------------------------------------------------------- Captured stderr call ---------------------------------------------------------------------
[WARNING][/torch/catch/torch_mlu/csrc/aten/util/version.cpp:131][operator()][process:2491][thread:139776105277248]: Cambricon NEUWARE minimum version requirements not met! Require DRIVER minimum verion is 4.20.15-1, but current version is 4.20.11-1

Is it caused by the neuware version mismatch?

[grimoire]

LGTM