diff --git a/include/tim/vx/ops/simple_operations.h b/include/tim/vx/ops/simple_operations.h
index 7e11324..514ae3a 100644
--- a/include/tim/vx/ops/simple_operations.h
+++ b/include/tim/vx/ops/simple_operations.h
@@ -90,6 +90,9 @@ namespace ops {
  *
  * Change the format from input tensor to output tensor. This operation ignores
  * the scale and zeroPoint of quanized tensors.
+ * 
+ * ## Rcp
+ * Computes the reciprocal of input element-wise.
  */
 
 DECLARE_SIMPLE_OP(DataConvert)
@@ -108,6 +111,7 @@ DECLARE_SIMPLE_OP(Floor)
 DECLARE_SIMPLE_OP(Ceil)
 DECLARE_SIMPLE_OP(Round)
 DECLARE_SIMPLE_OP(Cast)
+DECLARE_SIMPLE_OP(Rcp)
 
 #undef DECLARE_SIMPLE_OP
 
diff --git a/src/tim/vx/ops/README.md b/src/tim/vx/ops/README.md
index dc29cb5..295b214 100644
--- a/src/tim/vx/ops/README.md
+++ b/src/tim/vx/ops/README.md
@@ -116,6 +116,7 @@ GroupedConv1d|GROUPED_CONV1D|Mapped|[tf.keras.layers.Conv1D](https://tensorflow.
 Mod|MOD|Mapped|[Onnx.mod](https://github.com/onnx/onnx/blob/main/docs/Operators.md#Mod)
 Selu|SELU|Mapped|[tf.keras.activations.selu](https://www.tensorflow.org/api_docs/python/tf/keras/activations/selu)
 Celu|CELU|Mapped|[Onnx.celu](https://github.com/onnx/onnx/blob/main/docs/Operators.md#Celu)
+Rcp|RCP|Mapped|[tf.math.reciprocal](https://www.tensorflow.org/api_docs/python/tf/math/reciprocal)
 |UnidirectionalSequenceRNN|UNIDIRECTIONAL_SEQUENCE_RNN|Planned 22Q3|[ANEURALNETWORKS_UNIDIRECTIONAL_SEQUENCE_RNN](https://developer.android.com/ndk/reference/group/neural-networks#group___neural_networks_1ggaabbe492c60331b13038e39d4207940e0ae11aa1d461d2abaa117f6ee2cb503dd8)
 |BidirectionalSequenceRNN|BIDIRECTIONAL_SEQUENCE_RNN|Planned 22Q3|[ANEURALNETWORKS_BIDIRECTIONAL_SEQUENCE_RNN](https://developer.android.com/ndk/reference/group/neural-networks#group___neural_networks_1ggaabbe492c60331b13038e39d4207940e0a487fc5ae247de828f13e62b99f259f3c)
 |BidirectionalSequenceLSTM|BIDIRECTIONAL_SEQUENCE_LSTM|Mapped|[ANEURALNETWORKS_BIDIRECTIONAL_SEQUENCE_LSTM](https://developer.android.com/ndk/reference/group/neural-networks#group___neural_networks_1ggaabbe492c60331b13038e39d4207940e0a492a71cb7aa50b9a1a834a3cb269d778)
diff --git a/src/tim/vx/ops/simple_operations.cc b/src/tim/vx/ops/simple_operations.cc
index 4a3b8ba..08636cf 100644
--- a/src/tim/vx/ops/simple_operations.cc
+++ b/src/tim/vx/ops/simple_operations.cc
@@ -52,6 +52,7 @@ DEFINE_SIMPLE_OP(Floor, VSI_NN_OP_FLOOR)
 DEFINE_SIMPLE_OP(Ceil, VSI_NN_OP_CEIL)
 DEFINE_SIMPLE_OP(Round, VSI_NN_OP_ROUND)
 DEFINE_SIMPLE_OP(Cast, VSI_NN_OP_CAST)
+DEFINE_SIMPLE_OP(Rcp, VSI_NN_OP_RCP)
 
 #undef DEFINE_SIMPLE_OP
 
diff --git a/src/tim/vx/ops/simple_operations_test.cc b/src/tim/vx/ops/simple_operations_test.cc
index 6c4828c..efadec6 100644
--- a/src/tim/vx/ops/simple_operations_test.cc
+++ b/src/tim/vx/ops/simple_operations_test.cc
@@ -24,6 +24,7 @@
 #include "tim/vx/context.h"
 #include "tim/vx/graph.h"
 #include "tim/vx/ops/simple_operations.h"
+#include "test_utils.h"
 
 #include "gtest/gtest.h"
 #include <cstdlib>
@@ -232,4 +233,32 @@ TEST(DataConvert, requantize_shape_2_3_asym_u8) {
     std::vector<uint8_t> output(6, 0);
     EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
     EXPECT_EQ(golden, output);
+}
+
+TEST(Rcp, shape_5_1_fp32) {
+   auto ctx = tim::vx::Context::Create();
+    auto graph = ctx->CreateGraph();
+
+    tim::vx::ShapeType io_shape({5, 1});
+    tim::vx::TensorSpec input_spec(tim::vx::DataType::FLOAT32,
+                            io_shape, tim::vx::TensorAttribute::INPUT);
+    tim::vx::TensorSpec output_spec(tim::vx::DataType::FLOAT32,
+                            io_shape, tim::vx::TensorAttribute::OUTPUT);
+
+    auto input_tensor = graph->CreateTensor(input_spec);
+    auto output_tensor = graph->CreateTensor(output_spec);
+
+    std::vector<float> in_data = { -2.5, -0.1, 0, 0.55, std::numeric_limits<float>::infinity() };
+    std::vector<float> golden = {-0.4, -10, std::numeric_limits<float>::infinity(), 1.81818, 0.};
+
+    EXPECT_TRUE(input_tensor->CopyDataToTensor(in_data.data(), in_data.size()*4));
+
+    auto add = graph->CreateOperation<tim::vx::ops::Rcp>();
+    (*add).BindInputs({input_tensor}).BindOutputs({output_tensor});
+
+    EXPECT_TRUE(graph->Compile());
+    EXPECT_TRUE(graph->Run());
+    std::vector<float> output(5, 0);
+    EXPECT_TRUE(output_tensor->CopyDataFromTensor(output.data()));
+    EXPECT_TRUE(ArraysMatch(golden, output, 1e-5f));
 }
\ No newline at end of file
diff --git a/src/tim/vx/test_utils.h b/src/tim/vx/test_utils.h
index f2493eb..a06d312 100644
--- a/src/tim/vx/test_utils.h
+++ b/src/tim/vx/test_utils.h
@@ -115,7 +115,11 @@ template<typename T>
                                        const std::vector<T>& actual,
                                        T abs_error){
     for (size_t i = 0; i < expected.size(); ++i){
-        EXPECT_NEAR(expected[i], actual[i], abs_error) << "at index:" << i;
+      // If expected and actual are INF, "==" operation will return true while EXPEC_NEAR return false
+        if (expected[i] == actual[i]) continue;
+        else {
+          EXPECT_NEAR(expected[i], actual[i], abs_error) << "at index:" << i;
+        }
     }
 
     return ::testing::AssertionSuccess();