diff --git a/src/tim/vx/ops/max_pool_test.cc b/src/tim/vx/ops/max_pool_test.cc
index 5010ff85..2ed8a9d8 100644
--- a/src/tim/vx/ops/max_pool_test.cc
+++ b/src/tim/vx/ops/max_pool_test.cc
@@ -29,6 +29,66 @@
 #include "gtest/gtest.h"
 #include "test_utils.h"
 
+#include "third_party/half/half.hpp"
+
+TEST(MAX, shape_8_8_1_1_float16_kernel_3x3_stride_2x2) {
+  auto ctx = tim::vx::Context::Create();
+  auto graph = ctx->CreateGraph();
+  using namespace half_float::literal;
+
+  tim::vx::ShapeType input_shape({8, 8, 1, 1});   //whcn
+  tim::vx::ShapeType output_shape(
+      {3, 3, 1, 1});  //whcn
+
+  tim::vx::TensorSpec input_spec(tim::vx::DataType::FLOAT16, input_shape,
+                                 tim::vx::TensorAttribute::INPUT);
+  tim::vx::TensorSpec output_spec(tim::vx::DataType::FLOAT16, output_shape,
+                                  tim::vx::TensorAttribute::OUTPUT);
+
+  // Input data  nchw
+  std::vector<half_float::half> input_data = {
+      1.0_h, 2.0_h, 3.0_h, 4.0_h, 5.0_h, 6.0_h, 7.0_h, 8.0_h,
+      11.0_h, 12.0_h, 13.0_h, 14.0_h, 15.0_h, 16.0_h, 17.0_h, 18.0_h,
+      21.0_h, 22.0_h, 23.0_h, 24.0_h, 25.0_h, 26.0_h, 27.0_h, 28.0_h,
+      31.0_h, 32.0_h, 33.0_h, 34.0_h, 35.0_h, 36.0_h, 37.0_h, 38.0_h,
+      41.0_h, 42.0_h, 43.0_h, 44.0_h, 45.0_h, 46.0_h, 47.0_h, 48.0_h,
+      51.0_h, 52.0_h, 53.0_h, 54.0_h, 55.0_h, 56.0_h, 57.0_h, 58.0_h,
+      61.0_h, 62.0_h, 63.0_h, 64.0_h, 65.0_h, 66.0_h, 67.0_h, 68.0_h,
+      71.0_h, 72.0_h, 73.0_h, 74.0_h, 75.0_h, 76.0_h, 77.0_h, 78.0_h,
+  };
+
+  std::vector<half_float::half> golden = {
+        23.0_h, 25.0_h, 27.0_h,
+        43.0_h, 45.0_h, 47.0_h,
+        63.0_h, 65.0_h, 67.0_h,
+    };
+
+  auto input_tensor = graph->CreateTensor(input_spec);
+  auto output_tensor = graph->CreateTensor(output_spec);
+
+  std::array<uint32_t, 2> ksize = {3, 3};
+  std::array<uint32_t, 2> stride = {2, 2};
+  auto round_type = tim::vx::RoundType::FLOOR;
+
+  auto op = graph->CreateOperation<tim::vx::ops::Pool2d>(
+      tim::vx::PoolType::MAX, tim::vx::PadType::VALID, ksize, stride, round_type);
+  (*op).BindInputs({input_tensor}).BindOutputs({output_tensor});
+
+  EXPECT_TRUE(graph->Compile());
+
+  input_tensor->CopyDataToTensor(input_data.data());
+
+  EXPECT_TRUE(graph->Run());
+
+  uint32_t output_size = 1;
+  for (auto i : output_tensor->GetShape()) {
+    output_size *= i;
+  }
+  std::vector<half_float::half> output(output_size);
+  EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
+  EXPECT_TRUE(ArraysMatch(golden, output, (half_float::half)0.1));
+}
+
 TEST(MAX, shape_32_3_1_fp32_kernel_2_stride_1) {
     auto ctx = tim::vx::Context::Create();
     auto graph = ctx->CreateGraph();