diff --git a/mmcv/ops/csrc/common/pytorch_cpp_helper.hpp b/mmcv/ops/csrc/common/pytorch_cpp_helper.hpp
index 10c2794e5b..45e668440f 100644
--- a/mmcv/ops/csrc/common/pytorch_cpp_helper.hpp
+++ b/mmcv/ops/csrc/common/pytorch_cpp_helper.hpp
@@ -22,6 +22,9 @@ using namespace at;
 #define CHECK_MLU_INPUT(x) \
   CHECK_MLU(x);            \
   CHECK_CONTIGUOUS(x)
+#define CHECK_MUSA_INPUT(x) \
+  CHECK_MUSA(x);            \
+  CHECK_CONTIGUOUS(x)
 #define CHECK_CPU_INPUT(x) \
   CHECK_CPU(x);            \
   CHECK_CONTIGUOUS(x)
diff --git a/mmcv/ops/csrc/pytorch/ball_query.cpp b/mmcv/ops/csrc/pytorch/ball_query.cpp
index 7b56568338..cff552b8ea 100644
--- a/mmcv/ops/csrc/pytorch/ball_query.cpp
+++ b/mmcv/ops/csrc/pytorch/ball_query.cpp
@@ -16,7 +16,6 @@ void ball_query_forward_impl(int b, int n, int m, float min_radius,
 void ball_query_forward(Tensor new_xyz_tensor, Tensor xyz_tensor,
                         Tensor idx_tensor, int b, int n, int m,
                         float min_radius, float max_radius, int nsample) {
-    std::cout<<"ball_query_forward"<<std::endl;
   ball_query_forward_impl(b, n, m, min_radius, max_radius, nsample,
                           new_xyz_tensor, xyz_tensor, idx_tensor);
 }
diff --git a/mmcv/ops/csrc/pytorch/nms_rotated.cpp b/mmcv/ops/csrc/pytorch/nms_rotated.cpp
index a24a194392..3f8223009a 100644
--- a/mmcv/ops/csrc/pytorch/nms_rotated.cpp
+++ b/mmcv/ops/csrc/pytorch/nms_rotated.cpp
@@ -36,18 +36,12 @@ Tensor nms_rotated(const Tensor dets, const Tensor scores, const Tensor order,
                    const Tensor dets_sorted, const Tensor labels,
                    const float iou_threshold, const int multi_label) {
 
-  std::cout<<"nms_rotated"<<std::endl;
-  std::cout<<dets<<std::endl;
-  std::cout<<dets.device()<<std::endl;
-  std::cout<<dets.is_cuda()<<std::endl;
-  std::cout<<dets.is_privateuseone()<<std::endl;
   assert(dets.is_cuda() == scores.is_cuda());
   if (dets.device().is_cuda()) {
 #ifdef MMCV_WITH_CUDA
     return nms_rotated_cuda(dets, scores, order, dets_sorted.contiguous(),
                             iou_threshold, multi_label);
 #else
-  std::cout<<"nms_rotated in cuda"<<std::endl;
     AT_ERROR("Not compiled with GPU support");
 #endif
 #ifdef MMCV_WITH_XLA
@@ -64,7 +58,6 @@ Tensor nms_rotated(const Tensor dets, const Tensor scores, const Tensor order,
 #endif
 #ifdef MMCV_WITH_MUSA
   } else if (dets.device().type() == ::at::kPrivateUse1) {
-     std::cout<<"privateuse1"<<std::endl;
     return nms_rotated_musa(dets, scores, order, dets_sorted.contiguous(),
                             iou_threshold, multi_label);
 #endif
diff --git a/mmcv/ops/csrc/pytorch/spconv_ops.cpp b/mmcv/ops/csrc/pytorch/spconv_ops.cpp
index 55f59ef8f6..9658e3eab5 100644
--- a/mmcv/ops/csrc/pytorch/spconv_ops.cpp
+++ b/mmcv/ops/csrc/pytorch/spconv_ops.cpp
@@ -35,6 +35,8 @@ std::vector<torch::Tensor> get_indice_pairs_forward_cuda(
       padding, dilation, outPadding, _subM, _transpose);
 };
 
+
+
 template <unsigned NDim>
 std::vector<torch::Tensor> GetIndicePairsForwardMUSAKernelLauncher(
     torch::Tensor indices, int64_t batchSize,
@@ -97,6 +99,28 @@ std::vector<torch::Tensor> get_indice_pairs_backward_cuda(
       stride, padding, dilation, outPadding, _subM, _transpose);
 };
 
+#ifdef MMCV_WITH_MUSA
+template <unsigned NDim>
+std::vector<torch::Tensor> GetIndicePairsBackwardMUSAKernelLauncher(
+    torch::Tensor indices, torch::Tensor gridOut, int64_t batchSize,
+    std::vector<int64_t> outSpatialShape, std::vector<int64_t> spatialShape,
+    std::vector<int64_t> kernelSize, std::vector<int64_t> stride,
+    std::vector<int64_t> padding, std::vector<int64_t> dilation,
+    std::vector<int64_t> outPadding, int64_t _subM, int64_t _transpose);
+
+template <unsigned NDim>
+std::vector<torch::Tensor> get_indice_pairs_backward_musa(
+    torch::Tensor indices, torch::Tensor gridOut, int64_t batchSize,
+    std::vector<int64_t> outSpatialShape, std::vector<int64_t> spatialShape,
+    std::vector<int64_t> kernelSize, std::vector<int64_t> stride,
+    std::vector<int64_t> padding, std::vector<int64_t> dilation,
+    std::vector<int64_t> outPadding, int64_t _subM, int64_t _transpose) {
+  return GetIndicePairsBackwardMUSAKernelLauncher<NDim>(
+      indices, gridOut, batchSize, outSpatialShape, spatialShape, kernelSize,
+      stride, padding, dilation, outPadding, _subM, _transpose);
+};
+#endif
+
 template <unsigned NDim>
 std::vector<torch::Tensor> get_indice_pairs_forward(
     torch::Tensor indices, int64_t batchSize,
@@ -150,6 +174,15 @@ std::vector<torch::Tensor> get_indice_pairs_backward(
     AT_ERROR("get_indice_pairs is not compiled with GPU support");
 #endif
   } else {
+#ifdef MMCV_WITH_MUSA
+    if (indices.device().type() == at::kMUSA) {
+      CHECK_MUSA_INPUT(indices);
+      CHECK_MUSA_INPUT(gridOut);
+    return get_indice_pairs_backward_musa<NDim>(
+        indices, gridOut, batchSize, outSpatialShape, spatialShape, kernelSize,
+        stride, padding, dilation, outPadding, _subM, _transpose);
+    }
+#endif
     AT_ERROR("get_indice_pairs is not implemented on CPU");
   }
 }
diff --git a/tests/test_ops/test_chamfer_distance.py b/tests/test_ops/test_chamfer_distance.py
index 99f4627166..418294fcb5 100644
--- a/tests/test_ops/test_chamfer_distance.py
+++ b/tests/test_ops/test_chamfer_distance.py
@@ -57,7 +57,14 @@ def torch_to_np_type(dtype):
         marks=pytest.mark.skipif(
             not IS_MUSA_AVAILABLE, reason='requires MUSA support'))
 ])
-@pytest.mark.parametrize('dtype', [torch.half, torch.float32])
+@pytest.mark.parametrize('dtype', [
+    pytest.param(
+        torch.half,
+        marks=pytest.mark.skipif(
+            IS_MUSA_AVAILABLE,
+            reason='TODO haowen.han@mthreads.com: not supported yet')),
+    torch.float32
+])
 @pytest.mark.parametrize('shape', [(2, 600, 2), (2, 600, 2)])
 def test_chamfer_distance_npu_dynamic_shape(dtype, device, shape):
     if device == 'musa':