diff --git a/paddle/phi/kernels/cpu/flip_kernel.cc b/paddle/phi/kernels/cpu/flip_kernel.cc
index d53ffaa3df439b..d6c8637399b2de 100644
--- a/paddle/phi/kernels/cpu/flip_kernel.cc
+++ b/paddle/phi/kernels/cpu/flip_kernel.cc
@@ -42,6 +42,9 @@ void FlipKernel(const Context& dev_ctx,
   auto numel = x.numel();
   const T* x_data = x.data<T>();
   T* out_data = dev_ctx.template Alloc<T>(out);
+  if (out->numel() == 0) {
+    return;
+  }
 #ifdef PADDLE_WITH_MKLML
 #pragma omp parallel for
 #endif
diff --git a/paddle/phi/kernels/gpu/flip_kernel.cu b/paddle/phi/kernels/gpu/flip_kernel.cu
index 4b73fd48d95985..22fb297d0a5f98 100644
--- a/paddle/phi/kernels/gpu/flip_kernel.cu
+++ b/paddle/phi/kernels/gpu/flip_kernel.cu
@@ -68,6 +68,9 @@ void FlipKernel(const Context& dev_ctx,
                 DenseTensor* out) {
   auto* in_data = x.data<T>();
   auto* out_data = dev_ctx.template Alloc<T>(out);
+  if (out->numel() == 0) {
+    return;
+  }
 
   auto x_dims = x.dims();
   const int rank = x_dims.size();
diff --git a/test/legacy_test/test_reverse_op.py b/test/legacy_test/test_reverse_op.py
index ded51a261721e0..3cab066fde9155 100644
--- a/test/legacy_test/test_reverse_op.py
+++ b/test/legacy_test/test_reverse_op.py
@@ -91,6 +91,26 @@ def initTestCase(self):
         self.axis = [-1, -2]
 
 
+class TestCase4(TestReverseOp):
+    def initTestCase(self):
+        self.x = np.random.random((3, 0, 10)).astype('float64')
+        self.axis = [1, 2]
+
+
+class TestCase5(TestReverseOp):
+    def initTestCase(self):
+        self.x = np.random.random((2, 0, 3)).astype('float32')
+        self.axis = [0]
+
+
+class TestCase6(TestReverseOp):
+    def initTestCase(self):
+        self.x = np.random.random((2, 0, 0)).astype('float32')
+        self.axis = [
+            0,
+        ]
+
+
 if __name__ == '__main__':
     paddle.enable_static()
     unittest.main()