diff --git a/test/legacy_test/CMakeLists.txt b/test/legacy_test/CMakeLists.txt
index d666116b6ad27907ffedb5fcddc76519ccbc1683..437f83a900b22bc3f9f78a4b8a5880855bc1877b 100644
--- a/test/legacy_test/CMakeLists.txt
+++ b/test/legacy_test/CMakeLists.txt
@@ -1195,7 +1195,9 @@ set(TEST_CINN_OPS
     test_flip
     test_triangular_solve_op
     test_scatter_nd_op
-    test_strided_slice_op)
+    test_strided_slice_op
+    test_instance_norm_op
+    test_cumsum_op)
 
 foreach(TEST_CINN_OPS ${TEST_CINN_OPS})
   if(WITH_CINN)
diff --git a/test/legacy_test/test_cumsum_op.py b/test/legacy_test/test_cumsum_op.py
index b2031a792bd0ec2a4b4e4ee3f09245ca6fe39bea..60d5855fb0ad2b63bc3661e958a04abedab773c6 100644
--- a/test/legacy_test/test_cumsum_op.py
+++ b/test/legacy_test/test_cumsum_op.py
@@ -314,7 +314,8 @@ class TestSumOpExclusiveFP16(OpTest):
         self.python_api = cumsum_wrapper
         self.public_python_api = paddle.cumsum
         self.init_dtype()
-        self.enable_cinn = True
+        # TODO(thisjiang): set `True` after reduce+cast at shape=[4, 5, 20, 20], dim=[2]'s fusion bug has fixed
+        self.enable_cinn = False
         self.attrs = {'axis': 2, "exclusive": True}
         self.x = np.random.random((4, 5, 20)).astype(self.dtype)
         self.out = np.concatenate(
@@ -389,7 +390,8 @@ def create_test_fp16_class(parent, max_relative_error=1e-2):
             self.dtype = self.dtype_ = np.float16
 
         def set_enable_cinn(self):
-            self.enable_cinn = True
+            # TODO(thisjiang): set `pass` after reduce+cast at shape=[4, 5, 20, 20], dim=[2]'s fusion bug has fixed
+            self.enable_cinn = False
 
         def test_check_output(self):
             self.check_output()
diff --git a/test/legacy_test/test_instance_norm_op.py b/test/legacy_test/test_instance_norm_op.py
index 84905484d07f2fc2020c850bf00526a577ba0be7..fe8ed7bf150c1424dc74f1ef60fc0eb0550586ec 100644
--- a/test/legacy_test/test_instance_norm_op.py
+++ b/test/legacy_test/test_instance_norm_op.py
@@ -103,6 +103,8 @@ class TestInstanceNormOp(OpTest):
         self.fw_comp_atol = 1e-6
         self.rev_comp_rtol = 1e-4
         self.rev_comp_atol = 1e-4
+        self.cinn_rtol = 1e-4
+        self.cinn_atol = 1e-4
         self.init_test_case()
         ref_y_np, ref_mean_np, ref_var_np_tmp = _reference_instance_norm_naive(
             self.x_np,
diff --git a/test/legacy_test/test_mean_op.py b/test/legacy_test/test_mean_op.py
index 78bcc7a46bc9b8e59fea593055a311e1fce4f7f4..9e5e7fc17c949826614fc92b9ea776a28dcd364c 100644
--- a/test/legacy_test/test_mean_op.py
+++ b/test/legacy_test/test_mean_op.py
@@ -284,9 +284,6 @@ class TestReduceMeanOpShape6D(TestReduceMeanOp):
     def set_attrs(self):
         self.shape = [2, 3, 4, 5, 6, 7]
 
-    def if_enable_cinn(self):
-        self.enable_cinn = False
-
 
 class TestReduceMeanOpShape6DBF16(TestReduceMeanBF16Op):
     def set_attrs(self):
@@ -298,9 +295,6 @@ class TestReduceMeanOpShape6DFP16(TestReduceMeanOp):
         self.shape = [2, 3, 4, 5, 6, 7]
         self.dtype = 'float16'
 
-    def if_enable_cinn(self):
-        self.enable_cinn = False
-
 
 class TestReduceMeanOpAxisAll(TestReduceMeanOp):
     def set_attrs(self):