diff --git a/python/tvm/topi/cuda/scatter.py b/python/tvm/topi/cuda/scatter.py
index be602c8ab7a3..48da7e1cfd80 100644
--- a/python/tvm/topi/cuda/scatter.py
+++ b/python/tvm/topi/cuda/scatter.py
@@ -21,10 +21,7 @@
 from ..scatter import _verify_scatter_nd_inputs
 from .nms import atomic_add
 from .sort import stable_sort_by_key_thrust, is_thrust_available
-
-
-def ceil_div(a, b):
-    return (a + b - 1) // b
+from ..utils import ceil_div
 
 
 def gen_ir_1d(data, indices, updates, axis, out, update_func):
diff --git a/python/tvm/topi/cuda/sparse.py b/python/tvm/topi/cuda/sparse.py
index c59e6887d47e..5cc4d284d565 100644
--- a/python/tvm/topi/cuda/sparse.py
+++ b/python/tvm/topi/cuda/sparse.py
@@ -23,7 +23,7 @@
 from tvm import relay, te
 
 from .. import nn
-from ..utils import traverse_inline
+from ..utils import traverse_inline, ceil_div
 
 
 def sparse_dense(data, weight_data, weight_indices, weight_indptr):
@@ -152,10 +152,6 @@ def sparse_dense_tir(data, w_data, w_indices, w_indptr):
     with either default_function_kernel0 for the transpose or
     default_function_kernel1 for the multiply.
     """
-
-    def ceil_div(a, b):
-        return (a + (b - 1)) // b
-
     def gen_ir(data, w_data, w_indices, w_indptr, out):
         # pylint: disable=invalid-name
         # TODO(tkonolige): use tensorcores for block multiply
diff --git a/python/tvm/topi/utils.py b/python/tvm/topi/utils.py
index c3e14eff3919..769aa436fb08 100644
--- a/python/tvm/topi/utils.py
+++ b/python/tvm/topi/utils.py
@@ -487,3 +487,7 @@ def is_empty_shape(shape):
       Whether input shape is empty or has dimesion with size 0.
     """
     return cpp.utils.is_empty_shape(shape)
+
+
+def ceil_div(a, b):
+    return (a + b - 1) // b