diff --git a/paddle/fluid/operators/jit/benchmark.cc b/paddle/fluid/operators/jit/benchmark.cc
index 8dab16c284a6f483dce2d8b1dc729391e2e76b8d..b39ce280939515ec8f4fa3b443ff4332074825fd 100644
--- a/paddle/fluid/operators/jit/benchmark.cc
+++ b/paddle/fluid/operators/jit/benchmark.cc
@@ -254,6 +254,7 @@ int main(int argc, char* argv[]) {
   // xyn
   BenchXYNKernel<jit::kVRelu, T, PlaceType>();
   BenchXYNKernel<jit::kVIdentity, T, PlaceType>();
+  BenchXYNKernel<jit::kVSquare, T, PlaceType>();
   BenchXYNKernel<jit::kVExp, T, PlaceType>();
   BenchXYNKernel<jit::kVSigmoid, T, PlaceType>();
   BenchXYNKernel<jit::kVTanh, T, PlaceType>();
diff --git a/paddle/fluid/operators/jit/helper.cc b/paddle/fluid/operators/jit/helper.cc
index 2465199f430b5c13a0615b6868951be2ac5996d2..5dbe22a81b4866bdf60a03710d8ffd0b7bcb597b 100644
--- a/paddle/fluid/operators/jit/helper.cc
+++ b/paddle/fluid/operators/jit/helper.cc
@@ -36,6 +36,7 @@ const char* to_string(KernelType kt) {
     ONE_CASE(kVRelu);
     ONE_CASE(kVIdentity);
     ONE_CASE(kVExp);
+    ONE_CASE(kVSquare);
     ONE_CASE(kVSigmoid);
     ONE_CASE(kVTanh);
     ONE_CASE(kLSTMCtHt);
diff --git a/paddle/fluid/operators/jit/kernel_base.h b/paddle/fluid/operators/jit/kernel_base.h
index 69112c0ee9e2232f604e4821f3852bfe646bdac4..adb101bd5cdf231ac330dbf44beb4c24c1fcf29e 100644
--- a/paddle/fluid/operators/jit/kernel_base.h
+++ b/paddle/fluid/operators/jit/kernel_base.h
@@ -30,6 +30,7 @@ typedef enum {
   kVAddBias,
   kVRelu,
   kVIdentity,
+  kVSquare,
   kVExp,
   kVSigmoid,
   kVTanh,
diff --git a/paddle/fluid/operators/jit/more/mkl/CMakeLists.txt b/paddle/fluid/operators/jit/more/mkl/CMakeLists.txt
index 7c6a75d35f654b6c2c46b4d498e401db92237341..667c6dfad6676d00ab994564bff57c90caa0cb41 100644
--- a/paddle/fluid/operators/jit/more/mkl/CMakeLists.txt
+++ b/paddle/fluid/operators/jit/more/mkl/CMakeLists.txt
@@ -8,6 +8,7 @@ USE_JITKERNEL_MORE(kVMul, mkl)
 USE_JITKERNEL_MORE(kVAdd, mkl)
 USE_JITKERNEL_MORE(kVScal, mkl)
 USE_JITKERNEL_MORE(kVExp, mkl)
+USE_JITKERNEL_MORE(kVSquare, mkl)
 USE_JITKERNEL_MORE(kVSigmoid, mkl)
 USE_JITKERNEL_MORE(kVTanh, mkl)
 USE_JITKERNEL_MORE(kSeqPool, mkl)
diff --git a/paddle/fluid/operators/jit/more/mkl/mkl.cc b/paddle/fluid/operators/jit/more/mkl/mkl.cc
index 5b20ae4da96f262657e487cd231b5f319bf92641..fccdc68f5efa34bac6f5a34a41569d2f77416284 100644
--- a/paddle/fluid/operators/jit/more/mkl/mkl.cc
+++ b/paddle/fluid/operators/jit/more/mkl/mkl.cc
@@ -86,6 +86,16 @@ void VExp<double>(const double* x, double* y, int n) {
   platform::dynload::vdExp(n, x, y);
 }
 
+template <>
+void VSquare<float>(const float* x, float* y, int n) {
+  platform::dynload::vsSqr(n, x, y);
+}
+
+template <>
+void VSquare<double>(const double* x, double* y, int n) {
+  platform::dynload::vdSqr(n, x, y);
+}
+
 template <>
 void VCopy<float>(const float* x, float* y, int n) {
   platform::dynload::cblas_scopy(n, x, 1, y, 1);
@@ -132,6 +142,11 @@ bool VExpKernel<float>::UseMe(const int& d) const {
   return d > 7;
 }
 
+template <>
+bool VSquareKernel<float>::UseMe(const int& d) const {
+  return d > 7;
+}
+
 template <>
 bool VSigmoidKernel<float>::UseMe(const int& d) const {
   return d > 7;
@@ -165,6 +180,7 @@ AWALYS_USE_ME_WITH_DOUBLE(VScal);
 AWALYS_USE_ME_WITH_DOUBLE(VExp);
 AWALYS_USE_ME_WITH_DOUBLE(VSigmoid);
 AWALYS_USE_ME_WITH_DOUBLE(VTanh);
+AWALYS_USE_ME_WITH_DOUBLE(VSquare);
 
 #undef AWALYS_USE_ME_WITH_DOUBLE
 }  // namespace mkl
@@ -184,6 +200,7 @@ REGISTER_MKL_KERNEL(kVMul, VMul);
 REGISTER_MKL_KERNEL(kVAdd, VAdd);
 REGISTER_MKL_KERNEL(kVScal, VScal);
 REGISTER_MKL_KERNEL(kVExp, VExp);
+REGISTER_MKL_KERNEL(kVSquare, VSquare);
 REGISTER_MKL_KERNEL(kVSigmoid, VSigmoid);
 REGISTER_MKL_KERNEL(kVTanh, VTanh);
 REGISTER_MKL_KERNEL(kSeqPool, SeqPool);
diff --git a/paddle/fluid/operators/jit/more/mkl/mkl.h b/paddle/fluid/operators/jit/more/mkl/mkl.h
index 314ef73d8a581eaaf16ff3b8a58189d209b1735f..a27196fa19f1d3e9aa6c414b6b9f99a21ef49025 100644
--- a/paddle/fluid/operators/jit/more/mkl/mkl.h
+++ b/paddle/fluid/operators/jit/more/mkl/mkl.h
@@ -39,6 +39,9 @@ void VScal(const T* a, const T* x, T* y, int n);
 template <typename T>
 void VExp(const T* x, T* y, int n);
 
+template <typename T>
+void VSquare(const T* x, T* y, int n);
+
 template <typename T>
 void VCopy(const T* x, T* y, int n);
 
@@ -110,6 +113,7 @@ DECLARE_MKL_KERNEL(VScal, AXYNTuples);
 DECLARE_MKL_KERNEL(VExp, XYNTuples);
 DECLARE_MKL_KERNEL(VSigmoid, XYNTuples);
 DECLARE_MKL_KERNEL(VTanh, XYNTuples);
+DECLARE_MKL_KERNEL(VSquare, XYNTuples);
 
 DECLARE_MKL_KERNEL(SeqPool, SeqPoolTuples);
 
diff --git a/paddle/fluid/operators/jit/refer/CMakeLists.txt b/paddle/fluid/operators/jit/refer/CMakeLists.txt
index 9a7e80740faa0245332bab2e046476c9014ef992..4b9bc5e8d49c62404d5d4ef99b7c50987fcb415a 100644
--- a/paddle/fluid/operators/jit/refer/CMakeLists.txt
+++ b/paddle/fluid/operators/jit/refer/CMakeLists.txt
@@ -28,3 +28,4 @@ USE_JITKERNEL_REFER(kLayerNorm)
 USE_JITKERNEL_REFER(kNCHW16CMulNC)
 USE_JITKERNEL_REFER(kSeqPool)
 USE_JITKERNEL_REFER(kMatMul)
+USE_JITKERNEL_REFER(kVSquare)
diff --git a/paddle/fluid/operators/jit/refer/refer.cc b/paddle/fluid/operators/jit/refer/refer.cc
index 1b8dd0e31517f9e37ef3665af48b6707a56a9425..3512ad7fe7921381afb6152330fff6be34de5ad7 100644
--- a/paddle/fluid/operators/jit/refer/refer.cc
+++ b/paddle/fluid/operators/jit/refer/refer.cc
@@ -31,6 +31,7 @@ REGISTER_REFER_KERNEL(kVAddBias, VAddBias);
 
 REGISTER_REFER_KERNEL(kVRelu, VRelu);
 REGISTER_REFER_KERNEL(kVIdentity, VIdentity);
+REGISTER_REFER_KERNEL(kVSquare, VSquare);
 REGISTER_REFER_KERNEL(kVExp, VExp);
 REGISTER_REFER_KERNEL(kVSigmoid, VSigmoid);
 REGISTER_REFER_KERNEL(kVTanh, VTanh);
diff --git a/paddle/fluid/operators/jit/refer/refer.h b/paddle/fluid/operators/jit/refer/refer.h
index 225319c059e581dd83049c5bf9b93c4ed516de0a..97d029358594d757f0e1874e9c87ecb8f97c9d50 100644
--- a/paddle/fluid/operators/jit/refer/refer.h
+++ b/paddle/fluid/operators/jit/refer/refer.h
@@ -83,6 +83,13 @@ inline void VIdentity(const T* x, T* y, int n) {
   }
 }
 
+template <typename T>
+inline void VSquare(const T* x, T* y, int n) {
+  for (int i = 0; i < n; ++i) {
+    y[i] = x[i] * x[i];
+  }
+}
+
 template <typename T>
 void VExp(const T* x, T* y, int n) {
   for (int i = 0; i < n; ++i) {
@@ -394,6 +401,7 @@ DECLARE_REFER_KERNEL(VIdentity, XYNTuples);
 DECLARE_REFER_KERNEL(VExp, XYNTuples);
 DECLARE_REFER_KERNEL(VSigmoid, XYNTuples);
 DECLARE_REFER_KERNEL(VTanh, XYNTuples);
+DECLARE_REFER_KERNEL(VSquare, XYNTuples);
 
 // lstm_t*, const lstm_attr_t*
 DECLARE_REFER_KERNEL(LSTMCtHt, LSTMTuples);
diff --git a/paddle/fluid/operators/jit/test.cc b/paddle/fluid/operators/jit/test.cc
index 1246ee7c24e0de1363fb35351d15fd4bb912c9e8..f4415a54ca9678c75038a820bb5d212e61593ec7 100644
--- a/paddle/fluid/operators/jit/test.cc
+++ b/paddle/fluid/operators/jit/test.cc
@@ -604,6 +604,12 @@ TEST(JITKernel, kVIdentity) {
   TestXYNKernel<jit::kVIdentity, double, paddle::platform::CPUPlace>();
 }
 
+TEST(JITKernel, kVSquare) {
+  namespace jit = paddle::operators::jit;
+  TestXYNKernel<jit::kVSquare, float, paddle::platform::CPUPlace>();
+  TestXYNKernel<jit::kVSquare, double, paddle::platform::CPUPlace>();
+}
+
 TEST(JITKernel, kVExp) {
   namespace jit = paddle::operators::jit;
   TestXYNKernel<jit::kVExp, float, paddle::platform::CPUPlace>();