diff --git a/mmcv/ops/csrc/pytorch/mlu/ball_query_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/ball_query_mlu.cpp
index f57b0e88fb..5016e4a4e3 100644
--- a/mmcv/ops/csrc/pytorch/mlu/ball_query_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/ball_query_mlu.cpp
@@ -34,9 +34,9 @@ void ball_query_forward_mlu(int b, int n, int m, float min_radius,
   auto idx_ptr = idx_impl->cnnlMalloc();
 
   auto handle = mluOpGetCurrentHandle();
-  TORCH_MLUOP_CHECK(mluOpBallQuery(handle, new_xyz_desc.desc(), new_xyz_ptr, xyz_desc.desc(),
-                                   xyz_ptr, min_radius, max_radius, nsample, idx_desc.desc(),
-                                   idx_ptr));
+  TORCH_MLUOP_CHECK(mluOpBallQuery(
+      handle, new_xyz_desc.desc(), new_xyz_ptr, xyz_desc.desc(), xyz_ptr,
+      min_radius, max_radius, nsample, idx_desc.desc(), idx_ptr));
 }
 
 void ball_query_forward_impl(int b, int n, int m, float min_radius,
diff --git a/mmcv/ops/csrc/pytorch/mlu/box_iou_rotated.cpp b/mmcv/ops/csrc/pytorch/mlu/box_iou_rotated.cpp
index 15fd47ace8..b04f51cb4b 100644
--- a/mmcv/ops/csrc/pytorch/mlu/box_iou_rotated.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/box_iou_rotated.cpp
@@ -38,9 +38,9 @@ void BoxIouRotatedMLUKernelLauncher(const Tensor boxes1, const Tensor boxes2,
   auto ious_ptr = ious_impl->cnnlMalloc();
 
   CNLOG(INFO) << "Call mluOpBoxIouRotated().";
-  TORCH_MLUOP_CHECK(mluOpBoxIouRotated(handle, mode_flag, aligned, boxes1_desc.desc(), boxes1_ptr,
-                                       boxes2_desc.desc(), boxes2_ptr, ious_desc.desc(),
-                                       ious_ptr));
+  TORCH_MLUOP_CHECK(mluOpBoxIouRotated(
+      handle, mode_flag, aligned, boxes1_desc.desc(), boxes1_ptr,
+      boxes2_desc.desc(), boxes2_ptr, ious_desc.desc(), ious_ptr));
 }
 
 void box_iou_rotated_mlu(const Tensor boxes1, const Tensor boxes2, Tensor ious,
diff --git a/mmcv/ops/csrc/pytorch/mlu/carafe_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/carafe_mlu.cpp
index a78c7cfda3..5d6e7e2092 100644
--- a/mmcv/ops/csrc/pytorch/mlu/carafe_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/carafe_mlu.cpp
@@ -72,12 +72,12 @@ void CARAFEForwardMLUKernelLauncher(const Tensor input, const Tensor mask,
   auto handle = mluOpGetCurrentHandle();
   mluOpCarafeDescriptor_t carafe_desc;
   TORCH_MLUOP_CHECK(mluOpCreateCarafeDescriptor(&carafe_desc));
-  TORCH_MLUOP_CHECK(mluOpSetCarafeDescriptor(carafe_desc, input.dim(), kernel_size, group_size,
-                                             scale_factor));
+  TORCH_MLUOP_CHECK(mluOpSetCarafeDescriptor(
+      carafe_desc, input.dim(), kernel_size, group_size, scale_factor));
   // launch kernel
-  TORCH_MLUOP_CHECK(mluOpCarafeForward(handle, carafe_desc, input_desc.desc(), input_ptr,
-                                       mask_desc.desc(), mask_ptr, output_desc.desc(),
-                                       output_ptr));
+  TORCH_MLUOP_CHECK(mluOpCarafeForward(handle, carafe_desc, input_desc.desc(),
+                                       input_ptr, mask_desc.desc(), mask_ptr,
+                                       output_desc.desc(), output_ptr));
   // destroy op descriptor
   TORCH_MLUOP_CHECK(mluOpDestroyCarafeDescriptor(carafe_desc));
 
@@ -160,13 +160,14 @@ void CARAFEBackwardMLUKernelLauncher(
   auto handle = mluOpGetCurrentHandle();
   mluOpCarafeDescriptor_t carafe_desc;
   TORCH_MLUOP_CHECK(mluOpCreateCarafeDescriptor(&carafe_desc));
-  TORCH_MLUOP_CHECK(mluOpSetCarafeDescriptor(carafe_desc, grad_output.dim(), kernel_size,
-                                             group_size, scale_factor));
+  TORCH_MLUOP_CHECK(mluOpSetCarafeDescriptor(
+      carafe_desc, grad_output.dim(), kernel_size, group_size, scale_factor));
   // launch kernel
-  TORCH_MLUOP_CHECK(mluOpCarafeBackward(handle, carafe_desc, input_desc.desc(), input_ptr,
-                                        mask_desc.desc(), mask_ptr, grad_output_desc.desc(),
-                                        grad_output_ptr, grad_input_desc.desc(), grad_input_ptr,
-                                        grad_mask_desc.desc(), grad_mask_ptr));
+  TORCH_MLUOP_CHECK(mluOpCarafeBackward(
+      handle, carafe_desc, input_desc.desc(), input_ptr, mask_desc.desc(),
+      mask_ptr, grad_output_desc.desc(), grad_output_ptr,
+      grad_input_desc.desc(), grad_input_ptr, grad_mask_desc.desc(),
+      grad_mask_ptr));
   // destroy op descriptor
   TORCH_MLUOP_CHECK(mluOpDestroyCarafeDescriptor(carafe_desc));
 
diff --git a/mmcv/ops/csrc/pytorch/mlu/deform_roi_pool_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/deform_roi_pool_mlu.cpp
index dd6925e168..26705d9f2e 100644
--- a/mmcv/ops/csrc/pytorch/mlu/deform_roi_pool_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/deform_roi_pool_mlu.cpp
@@ -51,9 +51,9 @@ void DeformRoIPoolForwardMLUKernelLauncher(Tensor input, Tensor rois,
   // get compute handle
   auto handle = mluOpGetCurrentHandle();
   TORCH_MLUOP_CHECK(mluOpDeformRoiPoolForward(
-                        handle, input_desc.desc(), input_ptr, rois_desc.desc(), rois_ptr,
-                        offset_real_desc, offset_ptr, pooled_height, pooled_width, spatial_scale,
-                        sampling_ratio, gamma, output_desc.desc(), output_ptr));
+      handle, input_desc.desc(), input_ptr, rois_desc.desc(), rois_ptr,
+      offset_real_desc, offset_ptr, pooled_height, pooled_width, spatial_scale,
+      sampling_ratio, gamma, output_desc.desc(), output_ptr));
 
   output.copy_(output_contiguous);
 }
@@ -114,11 +114,11 @@ void DeformRoIPoolBackwardMLUKernelLauncher(
   // get compute handle
   auto handle = mluOpGetCurrentHandle();
   TORCH_MLUOP_CHECK(mluOpDeformRoiPoolBackward(
-                        handle, grad_output_desc.desc(), grad_output_ptr, input_desc.desc(),
-                        input_ptr, rois_desc.desc(), rois_ptr, offset_real_desc, offset_ptr,
-                        pooled_height, pooled_width, spatial_scale, sampling_ratio, gamma,
-                        grad_input_desc.desc(), grad_input_ptr, grad_offset_real_desc,
-                        grad_offset_ptr));
+      handle, grad_output_desc.desc(), grad_output_ptr, input_desc.desc(),
+      input_ptr, rois_desc.desc(), rois_ptr, offset_real_desc, offset_ptr,
+      pooled_height, pooled_width, spatial_scale, sampling_ratio, gamma,
+      grad_input_desc.desc(), grad_input_ptr, grad_offset_real_desc,
+      grad_offset_ptr));
   grad_input.copy_(grad_input_);
 }
 
diff --git a/mmcv/ops/csrc/pytorch/mlu/diff_iou_rotated_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/diff_iou_rotated_mlu.cpp
index 10267b8bdd..046f589575 100644
--- a/mmcv/ops/csrc/pytorch/mlu/diff_iou_rotated_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/diff_iou_rotated_mlu.cpp
@@ -43,8 +43,8 @@ Tensor diff_iou_rotated_sort_vertices_forward_mlu(Tensor vertices, Tensor mask,
 
   // launch kernel
   TORCH_MLUOP_CHECK(mluOpDiffIouRotatedSortVerticesForward(
-                        handle, vertices_desc.desc(), vertices_ptr, mask_desc.desc(), mask_ptr,
-                        num_valid_desc.desc(), num_valid_ptr, idx_desc.desc(), idx_ptr));
+      handle, vertices_desc.desc(), vertices_ptr, mask_desc.desc(), mask_ptr,
+      num_valid_desc.desc(), num_valid_ptr, idx_desc.desc(), idx_ptr));
   return idx;
 }
 
diff --git a/mmcv/ops/csrc/pytorch/mlu/iou3d_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/iou3d_mlu.cpp
index 43abebc5bb..58bf6fc4bf 100644
--- a/mmcv/ops/csrc/pytorch/mlu/iou3d_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/iou3d_mlu.cpp
@@ -30,7 +30,8 @@ void IoU3DNMS3DMLUKernelLauncher(Tensor boxes, Tensor &keep, Tensor &keep_num,
   // workspace
   size_t workspace_size = 0;
   auto handle = mluOpGetCurrentHandle();
-  TORCH_MLUOP_CHECK(mluOpGetNmsWorkspaceSize(handle, boxes_desc.desc(), NULL, &workspace_size));
+  TORCH_MLUOP_CHECK(mluOpGetNmsWorkspaceSize(handle, boxes_desc.desc(), NULL,
+                                             &workspace_size));
   auto workspace = at::empty(workspace_size, boxes.options().dtype(at::kByte));
 
   // get compute queue
@@ -57,14 +58,14 @@ void IoU3DNMS3DMLUKernelLauncher(Tensor boxes, Tensor &keep, Tensor &keep_num,
   const float offset = 0.0;
 
   TORCH_MLUOP_CHECK(mluOpCreateNmsDescriptor(&nms_desc));
-  TORCH_MLUOP_CHECK(mluOpSetNmsDescriptor(nms_desc, box_mode, output_mode, algo, method_mode,
-                                          iou_threshold, soft_nms_sigma, max_output_size,
-                                          confidence_threshold, offset, input_layout,
-                                          pad_to_max_output_size));
+  TORCH_MLUOP_CHECK(mluOpSetNmsDescriptor(
+      nms_desc, box_mode, output_mode, algo, method_mode, iou_threshold,
+      soft_nms_sigma, max_output_size, confidence_threshold, offset,
+      input_layout, pad_to_max_output_size));
 
-  TORCH_MLUOP_CHECK(mluOpNms(handle, nms_desc, boxes_desc.desc(), boxes_ptr, NULL, NULL,
-                    workspace_ptr, workspace_size, output_desc.desc(), output_ptr,
-                    output_size_ptr));
+  TORCH_MLUOP_CHECK(mluOpNms(handle, nms_desc, boxes_desc.desc(), boxes_ptr,
+                             NULL, NULL, workspace_ptr, workspace_size,
+                             output_desc.desc(), output_ptr, output_size_ptr));
   TORCH_MLUOP_CHECK(mluOpDestroyNmsDescriptor(nms_desc));
 }
 
diff --git a/mmcv/ops/csrc/pytorch/mlu/mlu_common_helper.cpp b/mmcv/ops/csrc/pytorch/mlu/mlu_common_helper.cpp
index 35395c5ba9..9f0b50ae89 100644
--- a/mmcv/ops/csrc/pytorch/mlu/mlu_common_helper.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/mlu_common_helper.cpp
@@ -123,7 +123,8 @@ void MluOpTensorDescriptor::set_desc(const at::Tensor& t,
                                      mluOpDataType_t dtype,
                                      std::vector<int>& dims) {
   int dimNb = dims.size();
-  TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(desc_, layout, dtype, dimNb, dims.data()));
+  TORCH_MLUOP_CHECK(
+      mluOpSetTensorDescriptor(desc_, layout, dtype, dimNb, dims.data()));
 }
 
 // Handles
diff --git a/mmcv/ops/csrc/pytorch/mlu/mlu_common_helper.h b/mmcv/ops/csrc/pytorch/mlu/mlu_common_helper.h
index 806cc2acc8..362ea33564 100644
--- a/mmcv/ops/csrc/pytorch/mlu/mlu_common_helper.h
+++ b/mmcv/ops/csrc/pytorch/mlu/mlu_common_helper.h
@@ -35,13 +35,13 @@
   auto NAME##_ptr = NAME##_impl->cnnlMalloc();
 
 #ifndef TORCH_MLUOP_CHECK
-#define TORCH_MLUOP_CHECK(EXPR)                                                \
-  do {                                                                         \
-    mluOpStatus_t status = EXPR;                                               \
-      if (status != MLUOP_STATUS_SUCCESS) {                                    \
-        CNLOG(ERROR) << "";                                                    \
-        TORCH_CHECK(false, "MLUOPS error: ", mluOpGetErrorString(status));     \
-      }                                                                        \
+#define TORCH_MLUOP_CHECK(EXPR)                                          \
+  do {                                                                   \
+    mluOpStatus_t status = EXPR;                                         \
+    if (status != MLUOP_STATUS_SUCCESS) {                                \
+      CNLOG(ERROR) << "";                                                \
+      TORCH_CHECK(false, "MLUOPS error: ", mluOpGetErrorString(status)); \
+    }                                                                    \
   } while (0);
 #endif
 
@@ -65,8 +65,12 @@ mluOpReduceMode_t getMluOpReduceMode(const reduce_t reduce_type);
 
 class MluOpTensorDescriptor {
  public:
-  MluOpTensorDescriptor() { TORCH_MLUOP_CHECK(mluOpCreateTensorDescriptor(&desc_)); };
-  ~MluOpTensorDescriptor() { TORCH_MLUOP_CHECK(mluOpDestroyTensorDescriptor(desc_)); }
+  MluOpTensorDescriptor() {
+    TORCH_MLUOP_CHECK(mluOpCreateTensorDescriptor(&desc_));
+  };
+  ~MluOpTensorDescriptor() {
+    TORCH_MLUOP_CHECK(mluOpDestroyTensorDescriptor(desc_));
+  }
 
   void set(at::Tensor);
   void set_with_layout(at::Tensor, mluOpTensorLayout_t layout);
@@ -89,7 +93,9 @@ class MluOpHandle {
       handle = nullptr;
     }
   }
-  void setQueue(cnrtQueue_t queue) { TORCH_MLUOP_CHECK(mluOpSetQueue(handle, queue)); }
+  void setQueue(cnrtQueue_t queue) {
+    TORCH_MLUOP_CHECK(mluOpSetQueue(handle, queue));
+  }
   mluOpHandle_t handle;
 };
 
diff --git a/mmcv/ops/csrc/pytorch/mlu/ms_deform_attn_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/ms_deform_attn_mlu.cpp
index 2b46abfdd3..59192063e4 100644
--- a/mmcv/ops/csrc/pytorch/mlu/ms_deform_attn_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/ms_deform_attn_mlu.cpp
@@ -36,11 +36,11 @@ Tensor MsDeformAttnForwardLauncher(const Tensor& value,
   INITIAL_MLU_PARAM_WITH_TENSOR(attn_weight);
 
   TORCH_MLUOP_CHECK(mluOpMsDeformAttnForward(
-                        handle, value_desc.desc(), value_ptr, spatial_shapes_int_desc.desc(),
-                        spatial_shapes_int_ptr, level_start_index_int_desc.desc(),
-                        level_start_index_int_ptr, sampling_loc_desc.desc(), sampling_loc_ptr,
-                        attn_weight_desc.desc(), attn_weight_ptr, im2col_step, output_desc.desc(),
-                        output_ptr));
+      handle, value_desc.desc(), value_ptr, spatial_shapes_int_desc.desc(),
+      spatial_shapes_int_ptr, level_start_index_int_desc.desc(),
+      level_start_index_int_ptr, sampling_loc_desc.desc(), sampling_loc_ptr,
+      attn_weight_desc.desc(), attn_weight_ptr, im2col_step, output_desc.desc(),
+      output_ptr));
 
   output = output.view({batch_size, num_queries, num_heads * channels});
   return output;
diff --git a/mmcv/ops/csrc/pytorch/mlu/nms_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/nms_mlu.cpp
index 32353fa07a..675c11e597 100644
--- a/mmcv/ops/csrc/pytorch/mlu/nms_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/nms_mlu.cpp
@@ -34,8 +34,8 @@ Tensor NMSMLUKernelLauncher(Tensor boxes, Tensor scores, float iou_threshold,
   // workspace
   size_t workspace_size = 0;
   auto handle = mluOpGetCurrentHandle();
-  TORCH_MLUOP_CHECK(mluOpGetNmsWorkspaceSize(handle, boxes_desc.desc(), scores_desc.desc(),
-                                             &workspace_size));
+  TORCH_MLUOP_CHECK(mluOpGetNmsWorkspaceSize(
+      handle, boxes_desc.desc(), scores_desc.desc(), &workspace_size));
   auto workspace = at::empty(workspace_size, boxes.options().dtype(at::kByte));
 
   // get compute queue
@@ -63,14 +63,15 @@ Tensor NMSMLUKernelLauncher(Tensor boxes, Tensor scores, float iou_threshold,
   const int max_output_size = max_output_boxes;
 
   TORCH_MLUOP_CHECK(mluOpCreateNmsDescriptor(&nms_desc));
-  TORCH_MLUOP_CHECK(mluOpSetNmsDescriptor(nms_desc, box_mode, output_mode, algo, method_mode,
-                                          iou_threshold, soft_nms_sigma, max_output_size,
-                                          confidence_threshold, (float)offset, input_layout,
-                                          pad_to_max_output_size));
+  TORCH_MLUOP_CHECK(mluOpSetNmsDescriptor(
+      nms_desc, box_mode, output_mode, algo, method_mode, iou_threshold,
+      soft_nms_sigma, max_output_size, confidence_threshold, (float)offset,
+      input_layout, pad_to_max_output_size));
 
-  TORCH_MLUOP_CHECK(mluOpNms(handle, nms_desc, boxes_desc.desc(), boxes_ptr, scores_desc.desc(),
-                             scores_ptr, workspace_ptr, workspace_size, output_desc.desc(),
-                             output_ptr, output_size_ptr));
+  TORCH_MLUOP_CHECK(mluOpNms(handle, nms_desc, boxes_desc.desc(), boxes_ptr,
+                             scores_desc.desc(), scores_ptr, workspace_ptr,
+                             workspace_size, output_desc.desc(), output_ptr,
+                             output_size_ptr));
   TORCH_MLUOP_CHECK(mluOpDestroyNmsDescriptor(nms_desc));
   int output_num = *static_cast<int *>(output_size.cpu().data_ptr());
   auto ret = output.to(boxes.options().dtype(at::kLong));
diff --git a/mmcv/ops/csrc/pytorch/mlu/nms_rotated_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/nms_rotated_mlu.cpp
index 22b4f5f0f8..f3b6b333ee 100644
--- a/mmcv/ops/csrc/pytorch/mlu/nms_rotated_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/nms_rotated_mlu.cpp
@@ -30,7 +30,8 @@ Tensor nms_rotated_mlu(Tensor boxes, Tensor scores, float iou_threshold) {
   // workspace
   size_t workspace_size = 0;
   auto handle = mluOpGetCurrentHandle();
-  TORCH_MLUOP_CHECK(mluOpGetNmsRotatedWorkspaceSize(handle, boxes_desc.desc(), &workspace_size));
+  TORCH_MLUOP_CHECK(mluOpGetNmsRotatedWorkspaceSize(handle, boxes_desc.desc(),
+                                                    &workspace_size));
   auto workspace = at::empty(workspace_size, boxes.options().dtype(at::kByte));
 
   auto boxes_impl = torch_mlu::getMluTensorImpl(boxes_);
@@ -44,9 +45,10 @@ Tensor nms_rotated_mlu(Tensor boxes, Tensor scores, float iou_threshold) {
   auto output_size_impl = torch_mlu::getMluTensorImpl(output_size);
   auto output_size_ptr = output_size_impl->cnnlMalloc();
 
-  TORCH_MLUOP_CHECK(mluOpNmsRotated(handle, iou_threshold, boxes_desc.desc(), boxes_ptr,
-                                    scores_desc.desc(), scores_ptr, workspace_ptr, workspace_size,
-                                    output_desc.desc(), output_ptr, (int *)output_size_ptr));
+  TORCH_MLUOP_CHECK(mluOpNmsRotated(
+      handle, iou_threshold, boxes_desc.desc(), boxes_ptr, scores_desc.desc(),
+      scores_ptr, workspace_ptr, workspace_size, output_desc.desc(), output_ptr,
+      (int *)output_size_ptr));
   int output_num = *static_cast<int *>(output_size.cpu().data_ptr());
   auto ret = output.to(boxes.options().dtype(at::kLong));
   return ret.slice(0, 0, output_num);
diff --git a/mmcv/ops/csrc/pytorch/mlu/psamask_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/psamask_mlu.cpp
index 6c36bea598..1e43787a7a 100644
--- a/mmcv/ops/csrc/pytorch/mlu/psamask_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/psamask_mlu.cpp
@@ -35,8 +35,8 @@ void PSAMaskForwardMLUKernelLauncher(const int psa_type, const Tensor x,
   auto y_impl = torch_mlu::getMluTensorImpl(y_tmp);
   auto y_ptr = y_impl->cnnlMalloc();
 
-  TORCH_MLUOP_CHECK(mluOpPsamaskForward(handle, psa_type, x_desc.desc(), x_ptr, h_mask, w_mask,
-                                        y_desc.desc(), y_ptr));
+  TORCH_MLUOP_CHECK(mluOpPsamaskForward(handle, psa_type, x_desc.desc(), x_ptr,
+                                        h_mask, w_mask, y_desc.desc(), y_ptr));
 
   y.copy_(y_tmp);
 }
@@ -67,7 +67,8 @@ void PSAMaskBackwardMLUKernelLauncher(const int psa_type, const Tensor dy,
   auto dy_impl = torch_mlu::getMluTensorImpl(dy_tensor);
   auto dy_ptr = dy_impl->cnnlMalloc();
 
-  TORCH_MLUOP_CHECK(mluOpPsamaskBackward(handle, psa_type, dy_desc.desc(), dy_ptr, h_mask, w_mask,
+  TORCH_MLUOP_CHECK(mluOpPsamaskBackward(handle, psa_type, dy_desc.desc(),
+                                         dy_ptr, h_mask, w_mask,
                                          dx_tmp_desc.desc(), dx_ptr));
 
   dx.copy_(dx_tmp);
diff --git a/mmcv/ops/csrc/pytorch/mlu/roi_align_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/roi_align_mlu.cpp
index 9c4b9a36c8..41307fd2ff 100644
--- a/mmcv/ops/csrc/pytorch/mlu/roi_align_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/roi_align_mlu.cpp
@@ -49,9 +49,9 @@ void ROIAlignForwardMLUKernelLauncher(Tensor input, Tensor rois, Tensor output,
 
   mluOpRoiAlignForwardDescriptor_t roialign_desc;
   TORCH_MLUOP_CHECK(mluOpCreateRoiAlignForwardDescriptor(&roialign_desc));
-  TORCH_MLUOP_CHECK(mluOpSetRoiAlignForwardDescriptor_v2(roialign_desc, aligned_height,
-                                                         aligned_width, sampling_ratio,
-                                                         spatial_scale, pool_mode, aligned));
+  TORCH_MLUOP_CHECK(mluOpSetRoiAlignForwardDescriptor_v2(
+      roialign_desc, aligned_height, aligned_width, sampling_ratio,
+      spatial_scale, pool_mode, aligned));
 
   auto handle = mluOpGetCurrentHandle();
   if (pool_mode == 0) {
@@ -65,16 +65,16 @@ void ROIAlignForwardMLUKernelLauncher(Tensor input, Tensor rois, Tensor output,
     auto argmax_y_ptr = argmax_y_impl->cnnlMalloc();
     argmax_y_desc.set_with_layout(argmax_x_contiguous, MLUOP_LAYOUT_NHWC);
     argmax_x_desc.set_with_layout(argmax_x_contiguous, MLUOP_LAYOUT_NHWC);
-    TORCH_MLUOP_CHECK(mluOpRoiAlignForward_v2(handle, roialign_desc, input_desc.desc(), self_ptr,
-                                              rois_desc.desc(), rois_ptr, output_desc.desc(),
-                                              output_ptr, argmax_x_desc.desc(), argmax_x_ptr,
-                                              argmax_y_desc.desc(), argmax_y_ptr);
+    TORCH_MLUOP_CHECK(mluOpRoiAlignForward_v2(
+        handle, roialign_desc, input_desc.desc(), self_ptr, rois_desc.desc(),
+        rois_ptr, output_desc.desc(), output_ptr, argmax_x_desc.desc(),
+        argmax_x_ptr, argmax_y_desc.desc(), argmax_y_ptr));
     argmax_x.copy_(argmax_x_contiguous);
     argmax_y.copy_(argmax_y_contiguous);
   } else {
-    TORCH_MLUOP_CHECK(mluOpRoiAlignForward_v2(handle, roialign_desc, input_desc.desc(), self_ptr,
-                                              rois_desc.desc(), rois_ptr, output_desc.desc(),
-                                              output_ptr, NULL, NULL, NULL, NULL);
+    TORCH_MLUOP_CHECK(mluOpRoiAlignForward_v2(
+        handle, roialign_desc, input_desc.desc(), self_ptr, rois_desc.desc(),
+        rois_ptr, output_desc.desc(), output_ptr, NULL, NULL, NULL, NULL));
   }
   TORCH_MLUOP_CHECK(mluOpDestroyRoiAlignForwardDescriptor(roialign_desc));
   output.copy_(output_contiguous);
@@ -136,16 +136,16 @@ void ROIAlignBackwardMLUKernelLauncher(Tensor grad, Tensor rois,
     auto argmax_y_ptr = argmax_y_impl->cnnlMalloc();
     argmax_y_desc.set_with_layout(argmax_x_contiguous, MLUOP_LAYOUT_NHWC);
     argmax_x_desc.set_with_layout(argmax_x_contiguous, MLUOP_LAYOUT_NHWC);
-    TORCH_MLUOP_CHECK(mluOpRoiAlignBackward_v2(handle, grads_desc.desc(), grad_ptr,
-                          rois_desc.desc(), rois_ptr, argmax_y_desc.desc(),
-                          argmax_x_ptr, argmax_y_desc.desc(), argmax_y_ptr,
-                          spatial_scale, sampling_ratio, aligned, pool_mode,
-                          grad_input_desc.desc(), grad_input_ptr));
+    TORCH_MLUOP_CHECK(mluOpRoiAlignBackward_v2(
+        handle, grads_desc.desc(), grad_ptr, rois_desc.desc(), rois_ptr,
+        argmax_y_desc.desc(), argmax_x_ptr, argmax_y_desc.desc(), argmax_y_ptr,
+        spatial_scale, sampling_ratio, aligned, pool_mode,
+        grad_input_desc.desc(), grad_input_ptr));
   } else {
-    TORCH_MLUOP_CHECK(mluOpRoiAlignBackward_v2(handle, grads_desc.desc(), grad_ptr,
-                          rois_desc.desc(), rois_ptr, NULL, NULL, NULL, NULL,
-                          spatial_scale, sampling_ratio, aligned, pool_mode,
-                          grad_input_desc.desc(), grad_input_ptr));
+    TORCH_MLUOP_CHECK(mluOpRoiAlignBackward_v2(
+        handle, grads_desc.desc(), grad_ptr, rois_desc.desc(), rois_ptr, NULL,
+        NULL, NULL, NULL, spatial_scale, sampling_ratio, aligned, pool_mode,
+        grad_input_desc.desc(), grad_input_ptr));
   }
   grad_input.copy_(grad_input_);
 }
diff --git a/mmcv/ops/csrc/pytorch/mlu/roi_align_rotated_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/roi_align_rotated_mlu.cpp
index fc05283c97..10a8b68363 100644
--- a/mmcv/ops/csrc/pytorch/mlu/roi_align_rotated_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/roi_align_rotated_mlu.cpp
@@ -41,9 +41,9 @@ void ROIAlignRotatedForwardMLUKernelLauncher(Tensor input, Tensor rois,
   // get compute handle
   auto handle = mluOpGetCurrentHandle();
   TORCH_MLUOP_CHECK(mluOpRoiAlignRotatedForward(
-                        handle, input_desc.desc(), input_ptr, rois_desc.desc(), rois_ptr,
-                        pooled_height, pooled_width, sampling_ratio, spatial_scale, aligned,
-                        clockwise, output_desc.desc(), output_ptr);
+      handle, input_desc.desc(), input_ptr, rois_desc.desc(), rois_ptr,
+      pooled_height, pooled_width, sampling_ratio, spatial_scale, aligned,
+      clockwise, output_desc.desc(), output_ptr));
 
   output.copy_(output_contiguous);
 }
@@ -77,9 +77,9 @@ void ROIAlignRotatedBackwardMLUKernelLauncher(
   // get compute handle
   auto handle = mluOpGetCurrentHandle();
   TORCH_MLUOP_CHECK(mluOpRoiAlignRotatedBackward(
-                        handle, top_grad_desc.desc(), top_grad_ptr, rois_desc.desc(), rois_ptr,
-                        pooled_height, pooled_width, sampling_ratio, spatial_scale, aligned,
-                        clockwise, bottom_grad_desc.desc(), bottom_grad_ptr);
+      handle, top_grad_desc.desc(), top_grad_ptr, rois_desc.desc(), rois_ptr,
+      pooled_height, pooled_width, sampling_ratio, spatial_scale, aligned,
+      clockwise, bottom_grad_desc.desc(), bottom_grad_ptr));
   bottom_grad.copy_(bottom_grad_);
 }
 
diff --git a/mmcv/ops/csrc/pytorch/mlu/roiaware_pool3d_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/roiaware_pool3d_mlu.cpp
index 058be6b27b..ad01a36bd3 100644
--- a/mmcv/ops/csrc/pytorch/mlu/roiaware_pool3d_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/roiaware_pool3d_mlu.cpp
@@ -45,8 +45,8 @@ void RoiawarePool3dForwardMLUKernelLauncher(
   // allocate extra space for workspace
   size_t workspace_size = 0;
   TORCH_MLUOP_CHECK(mluOpGetRoiawarePool3dForwardWorkspaceSize(
-                        handle, rois_desc.desc(), pts_desc.desc(), pts_feature_desc.desc(),
-                        &workspace_size));
+      handle, rois_desc.desc(), pts_desc.desc(), pts_feature_desc.desc(),
+      &workspace_size));
 
   auto workspace = at::empty(workspace_size, rois.options().dtype(at::kByte));
   auto workspace_impl = torch_mlu::getMluTensorImpl(workspace);
diff --git a/mmcv/ops/csrc/pytorch/mlu/rotated_feature_align_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/rotated_feature_align_mlu.cpp
index 9891123f6c..46f93facda 100644
--- a/mmcv/ops/csrc/pytorch/mlu/rotated_feature_align_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/rotated_feature_align_mlu.cpp
@@ -76,10 +76,10 @@ void RotatedFeatureAlignBackwardMLUKernelLauncher(const Tensor top_grad,
 
   // get compute handle
   auto handle = mluOpGetCurrentHandle();
-  TORCH_MLUOP_CHECK(mluOpRotatedFeatureAlignBackward(handle, top_grad_desc.desc(), top_grad_ptr,
-                                                     best_bboxes_desc.desc(), best_bboxes_ptr,
-                                                     spatial_scale, points,
-                                                     bottom_grad_desc.desc(), bottom_grad_ptr));
+  TORCH_MLUOP_CHECK(mluOpRotatedFeatureAlignBackward(
+      handle, top_grad_desc.desc(), top_grad_ptr, best_bboxes_desc.desc(),
+      best_bboxes_ptr, spatial_scale, points, bottom_grad_desc.desc(),
+      bottom_grad_ptr));
   bottom_grad.copy_(bottom_grad_);
 }
 
diff --git a/mmcv/ops/csrc/pytorch/mlu/sparse_conv_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/sparse_conv_mlu.cpp
index 72ff94d24b..03acd84abc 100644
--- a/mmcv/ops/csrc/pytorch/mlu/sparse_conv_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/sparse_conv_mlu.cpp
@@ -86,17 +86,17 @@ std::vector<torch::Tensor> GetIndicePairsForwardMLUKernelLauncher(
     mluOpDataType_t dtype = MLUOP_DTYPE_INT32;
     std::vector<int> dims;
     dims = {numAct, coorDim + 1};
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(indices_desc.desc(), layout, dtype, dims.size(),
-                          dims.data()));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        indices_desc.desc(), layout, dtype, dims.size(), dims.data()));
     dims = {kernelVolume, 2, numAct};
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(indicePairs_desc.desc(), layout, dtype,
-                          dims.size(), dims.data()));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        indicePairs_desc.desc(), layout, dtype, dims.size(), dims.data()));
     dims = {kernelVolume};
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(indiceNum_desc.desc(), layout, dtype, dims.size(),
-                          dims.data()));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        indiceNum_desc.desc(), layout, dtype, dims.size(), dims.data()));
     dims = {out_size, coorDim + 1};
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(out_indices_desc.desc(), layout, dtype,
-                          dims.size(), dims.data()));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        out_indices_desc.desc(), layout, dtype, dims.size(), dims.data()));
   }
 
   mluOpSparseConvolutionDescriptor_t sparse_conv_desc;
@@ -127,13 +127,14 @@ std::vector<torch::Tensor> GetIndicePairsForwardMLUKernelLauncher(
   auto indiceNum_ptr = indiceNum_impl->cnnlMalloc();
   auto indice_workspace_ptr = indice_workspace_impl->cnnlMalloc();
 
-  TORCH_MLUOP_CHECK(mluOpGetIndicePairs(handle, sparse_conv_desc, indices_desc.desc(),
-                                        indices_ptr, indice_workspace_ptr, workspace_size,
-                                        indicePairs_desc.desc(), indicePairs_ptr,
-                                        out_indices_desc.desc(), out_indices_ptr,
-                                        indiceNum_desc.desc(), indiceNum_ptr));
+  TORCH_MLUOP_CHECK(mluOpGetIndicePairs(
+      handle, sparse_conv_desc, indices_desc.desc(), indices_ptr,
+      indice_workspace_ptr, workspace_size, indicePairs_desc.desc(),
+      indicePairs_ptr, out_indices_desc.desc(), out_indices_ptr,
+      indiceNum_desc.desc(), indiceNum_ptr));
   int num_act_out = 0;
-  TORCH_MLUOP_CHECK(mluOpGetSparseConvolutionNumActOut(sparse_conv_desc, &num_act_out));
+  TORCH_MLUOP_CHECK(
+      mluOpGetSparseConvolutionNumActOut(sparse_conv_desc, &num_act_out));
   TORCH_MLUOP_CHECK(mluOpDestroySparseConvolutionDescriptor(sparse_conv_desc));
   if (!sub_m) {
     return {out_indices.slice(0, 0, num_act_out), indicePairs, indiceNum};
@@ -179,25 +180,28 @@ torch::Tensor IndiceConvForwardMLUKernelLauncher(
     int dims[8];
 
     // features_desc
-    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(features_desc.desc(), &layout, &dtype, &dim, dims));
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(features_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype,
-                          dim, dims));
+    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(features_desc.desc(), &layout,
+                                               &dtype, &dim, dims));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        features_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim, dims));
 
     // filters_desc
-    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(filters_desc.desc(), &layout, &dtype, &dim, dims));
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(filters_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype,
-                          dim, dims));
+    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(filters_desc.desc(), &layout,
+                                               &dtype, &dim, dims));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        filters_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim, dims));
 
     // indice_pairs_desc
-    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(indice_pairs_desc.desc(), &layout, &dtype, &dim,
-                          dims));
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(indice_pairs_desc.desc(), MLUOP_LAYOUT_ARRAY,
-                          dtype, dim, dims));
+    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(indice_pairs_desc.desc(),
+                                               &layout, &dtype, &dim, dims));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        indice_pairs_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim, dims));
 
     // output_desc
-    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(output_desc.desc(), &layout, &dtype, &dim, dims));
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(output_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim,
-                          dims));
+    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(output_desc.desc(), &layout,
+                                               &dtype, &dim, dims));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        output_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim, dims));
   }
 
   auto handle = mluOpGetCurrentHandle();
@@ -290,37 +294,39 @@ std::vector<torch::Tensor> IndiceConvBackwardMLUKernelLauncher(
     int dims[8];
 
     // features_desc
-    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(features_desc.desc(), &layout, &dtype, &dim, dims));
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(features_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype,
-                          dim, dims));
+    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(features_desc.desc(), &layout,
+                                               &dtype, &dim, dims));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        features_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim, dims));
 
     // filters_desc
-    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(filters_desc.desc(), &layout, &dtype, &dim, dims));
+    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(filters_desc.desc(), &layout,
+                                               &dtype, &dim, dims));
     if (dim == 4) {
-      TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(filters_desc.desc(), MLUOP_LAYOUT_HWCN, dtype,
-                            dim, dims));
+      TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+          filters_desc.desc(), MLUOP_LAYOUT_HWCN, dtype, dim, dims));
     } else {
-      TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(filters_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype,
-                            dim, dims));
+      TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+          filters_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim, dims));
     }
 
     // output_grad_desc
-    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(output_grad_desc.desc(), &layout, &dtype, &dim,
-                          dims));
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(output_grad_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype,
-                          dim, dims));
+    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(output_grad_desc.desc(), &layout,
+                                               &dtype, &dim, dims));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        output_grad_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim, dims));
 
     // indice_pairs_desc
-    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(indice_pairs_desc.desc(), &layout, &dtype, &dim,
-                          dims));
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(indice_pairs_desc.desc(), MLUOP_LAYOUT_ARRAY,
-                          dtype, dim, dims));
+    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(indice_pairs_desc.desc(),
+                                               &layout, &dtype, &dim, dims));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        indice_pairs_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim, dims));
 
     // input_grad_desc
-    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(input_grad_desc.desc(), &layout, &dtype, &dim,
-                          dims));
-    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(input_grad_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype,
-                          dim, dims));
+    TORCH_MLUOP_CHECK(mluOpGetTensorDescriptor(input_grad_desc.desc(), &layout,
+                                               &dtype, &dim, dims));
+    TORCH_MLUOP_CHECK(mluOpSetTensorDescriptor(
+        input_grad_desc.desc(), MLUOP_LAYOUT_ARRAY, dtype, dim, dims));
   }
 
   auto handle = mluOpGetCurrentHandle();
diff --git a/mmcv/ops/csrc/pytorch/mlu/three_nn_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/three_nn_mlu.cpp
index 0394bdc11a..9aa57472d9 100644
--- a/mmcv/ops/csrc/pytorch/mlu/three_nn_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/three_nn_mlu.cpp
@@ -30,8 +30,8 @@ void ThreeNNMLUKernelLauncher(int b, int n, int m, const Tensor unknown,
 
   auto handle = mluOpGetCurrentHandle();
   size_t workspace_size = 0;
-  TORCH_MLUOP_CHECK(mluOpGetThreeNNForwardWorkspaceSize(handle, known_desc.desc(),
-                                                        &workspace_size));
+  TORCH_MLUOP_CHECK(mluOpGetThreeNNForwardWorkspaceSize(
+      handle, known_desc.desc(), &workspace_size));
   auto known_workspace =
       at::empty(workspace_size, known.options().dtype(at::kByte));
 
@@ -46,10 +46,10 @@ void ThreeNNMLUKernelLauncher(int b, int n, int m, const Tensor unknown,
   auto idx_ptr = idx_impl->cnnlMalloc();
   auto workspace_ptr = workspace_impl->cnnlMalloc();
 
-  TORCH_MLUOP_CHECK(mluOpThreeNNForward(handle, unknown_desc.desc(), unknown_ptr,
-                                        known_desc.desc(), known_ptr, workspace_ptr,
-                                        workspace_size, dist2_desc.desc(), dist2_ptr,
-                                        idx_desc.desc(), idx_ptr));
+  TORCH_MLUOP_CHECK(mluOpThreeNNForward(
+      handle, unknown_desc.desc(), unknown_ptr, known_desc.desc(), known_ptr,
+      workspace_ptr, workspace_size, dist2_desc.desc(), dist2_ptr,
+      idx_desc.desc(), idx_ptr));
 }
 
 void three_nn_forward_mlu(int b, int n, int m, const Tensor unknown,
diff --git a/mmcv/ops/csrc/pytorch/mlu/voxelization_mlu.cpp b/mmcv/ops/csrc/pytorch/mlu/voxelization_mlu.cpp
index 24226261b8..e6d4ff6d2b 100644
--- a/mmcv/ops/csrc/pytorch/mlu/voxelization_mlu.cpp
+++ b/mmcv/ops/csrc/pytorch/mlu/voxelization_mlu.cpp
@@ -62,14 +62,14 @@ int HardVoxelizeForwardMLUKernelLauncher(
       at::empty(workspace_size, points.options().dtype(at::kByte));
   INITIAL_MLU_PARAM_WITH_TENSOR(workspace_tensor);
 
-  TORCH_MLUOP_CHECK(mluOpVoxelization(handle, points_desc.desc(), points_ptr,
-                    voxel_size_tensor_desc.desc(), voxel_size_tensor_ptr,
-                    coors_range_tensor_desc.desc(), coors_range_tensor_ptr,
-                    max_points, max_voxels, NDim, true, workspace_tensor_ptr,
-                    workspace_size, voxels_desc.desc(), voxels_ptr,
-                    coors_desc.desc(), coors_ptr,
-                    num_points_per_voxel_desc.desc(), num_points_per_voxel_ptr,
-                    voxel_num_tensor_desc.desc(), voxel_num_tensor_ptr));
+  TORCH_MLUOP_CHECK(mluOpVoxelization(
+      handle, points_desc.desc(), points_ptr, voxel_size_tensor_desc.desc(),
+      voxel_size_tensor_ptr, coors_range_tensor_desc.desc(),
+      coors_range_tensor_ptr, max_points, max_voxels, NDim, true,
+      workspace_tensor_ptr, workspace_size, voxels_desc.desc(), voxels_ptr,
+      coors_desc.desc(), coors_ptr, num_points_per_voxel_desc.desc(),
+      num_points_per_voxel_ptr, voxel_num_tensor_desc.desc(),
+      voxel_num_tensor_ptr));
   auto voxel_num_cpu = voxel_num_tensor.to(at::kCPU);
   int voxel_num_int = voxel_num_cpu.data_ptr<int>()[0];
   return voxel_num_int;