[Feature] Support RoIAwarePool3d with cambricon MLU backend (#2359)

* [Feature] Support RoiawarePool3d with cambricon MLU backend.

* [Feature] Support RoiawarePool3d with cambricon MLU backend.

docs/en/understand_mmcv/ops.md

@@ -47,7 +47,7 @@ We implement common ops used in detection, segmentation, etc.
 | RoIAlignRotated              | √   | √    | √   |     |
 | RiRoIAlignRotated            |     | √    |     |     |
 | RoIAlign                     | √   | √    | √   |     |
-| RoIAwarePool3d               |     | √    |     |     |
+| RoIAwarePool3d               |     | √    | √   |     |
 | SAConv2d                     |     | √    |     |     |
 | SigmoidFocalLoss             |     | √    | √   |     |
 | SoftmaxFocalLoss             |     | √    |     |     |

docs/zh_cn/understand_mmcv/ops.md

@@ -47,7 +47,7 @@ MMCV 提供了检测、分割等任务中常用的算子
 | RoIAlignRotated              | √   | √    | √   |     |
 | RiRoIAlignRotated            |     | √    |     |     |
 | RoIAlign                     | √   | √    | √   |     |
-| RoIAwarePool3d               |     | √    |     |     |
+| RoIAwarePool3d               |     | √    | √   |     |
 | SAConv2d                     |     | √    |     |     |
 | SigmoidFocalLoss             |     | √    | √   |     |
 | SoftmaxFocalLoss             |     | √    |     |     |

tests/test_ops/test_roiaware_pool3d.py

@@ -5,11 +5,27 @@ import torch
 
 from mmcv.ops import (RoIAwarePool3d, points_in_boxes_all, points_in_boxes_cpu,
                       points_in_boxes_part)
-
-
-@pytest.mark.skipif(
-    not torch.cuda.is_available(), reason='requires CUDA support')
-def test_RoIAwarePool3d():
+from mmcv.utils import IS_CUDA_AVAILABLE, IS_MLU_AVAILABLE
+
+
+@pytest.mark.parametrize('device', [
+    pytest.param(
+        'cuda',
+        marks=pytest.mark.skipif(
+            not IS_CUDA_AVAILABLE, reason='requires CUDA support')),
+    pytest.param(
+        'mlu',
+        marks=pytest.mark.skipif(
+            not IS_MLU_AVAILABLE, reason='requires MLU support'))
+])
+@pytest.mark.parametrize('dtype', [
+    torch.float, torch.half,
+    pytest.param(
+        torch.double,
+        marks=pytest.mark.skipif(
+            IS_MLU_AVAILABLE, reason='MLU does not support for double'))
+])
+def test_RoIAwarePool3d(device, dtype):
     roiaware_pool3d_max = RoIAwarePool3d(
         out_size=4, max_pts_per_voxel=128, mode='max')
     roiaware_pool3d_avg = RoIAwarePool3d(
@@ -17,27 +33,27 @@ def test_RoIAwarePool3d():
     rois = torch.tensor(
         [[1.0, 2.0, 3.0, 5.0, 4.0, 6.0, -0.3 - np.pi / 2],
          [-10.0, 23.0, 16.0, 20.0, 10.0, 20.0, -0.5 - np.pi / 2]],
-        dtype=torch.float32).cuda(
-        )  # boxes (m, 7) with bottom center in lidar coordinate
+        dtype=dtype).to(device)
+    # boxes (m, 7) with bottom center in lidar coordinate
     pts = torch.tensor(
         [[1, 2, 3.3], [1.2, 2.5, 3.0], [0.8, 2.1, 3.5], [1.6, 2.6, 3.6],
          [0.8, 1.2, 3.9], [-9.2, 21.0, 18.2], [3.8, 7.9, 6.3],
          [4.7, 3.5, -12.2], [3.8, 7.6, -2], [-10.6, -12.9, -20], [-16, -18, 9],
          [-21.3, -52, -5], [0, 0, 0], [6, 7, 8], [-2, -3, -4]],
-        dtype=torch.float32).cuda()  # points (n, 3) in lidar coordinate
+        dtype=dtype).to(device)  # points (n, 3) in lidar coordinate
     pts_feature = pts.clone()
 
     pooled_features_max = roiaware_pool3d_max(
         rois=rois, pts=pts, pts_feature=pts_feature)
     assert pooled_features_max.shape == torch.Size([2, 4, 4, 4, 3])
     assert torch.allclose(pooled_features_max.sum(),
-                          torch.tensor(51.100).cuda(), 1e-3)
+                          torch.tensor(51.100, dtype=dtype).to(device), 1e-3)
 
     pooled_features_avg = roiaware_pool3d_avg(
         rois=rois, pts=pts, pts_feature=pts_feature)
     assert pooled_features_avg.shape == torch.Size([2, 4, 4, 4, 3])
     assert torch.allclose(pooled_features_avg.sum(),
-                          torch.tensor(49.750).cuda(), 1e-3)
+                          torch.tensor(49.750, dtype=dtype).to(device), 1e-3)
 
 
 @pytest.mark.skipif(