diff --git a/src/operator/subgraph/mkldnn/mkldnn_fc-inl.h b/src/operator/subgraph/mkldnn/mkldnn_fc-inl.h
index 9a09d91ae5d0..438495870166 100644
--- a/src/operator/subgraph/mkldnn/mkldnn_fc-inl.h
+++ b/src/operator/subgraph/mkldnn/mkldnn_fc-inl.h
@@ -36,7 +36,8 @@ static inline bool SupportMKLDNNFCEltwiseFusion(const std::string op_name) {
       op_name == "sqrt" ||
       op_name == "exp" ||
       op_name == "abs" ||
-      op_name == "clip") {
+      op_name == "clip" ||
+      op_name == "LeakyReLU") {
     return true;
   } else {
     return false;
diff --git a/src/operator/subgraph/mkldnn/mkldnn_fc.cc b/src/operator/subgraph/mkldnn/mkldnn_fc.cc
index e2b1807b6559..fd1e156fcd5e 100644
--- a/src/operator/subgraph/mkldnn/mkldnn_fc.cc
+++ b/src/operator/subgraph/mkldnn/mkldnn_fc.cc
@@ -286,8 +286,14 @@ void SgMKLDNNFCOp::Forward(const OpContext &ctx,
       if (fuse_requantize || mkldnn_param.enable_float_output) {
         float tmp_scale_ = 1.0f;
         if (fuse_requantize) {
-          tmp_scale_ =
-            GetQuantizeScale(output.dtype(), cached_min_output_, cached_max_output_) / data_scale_;
+          if (mkldnn_param.with_eltwise) {
+            tmp_scale_ = 1.0 / data_scale_;
+            full_param_.eltwise_param.scale =
+              GetQuantizeScale(output.dtype(), cached_min_output_, cached_max_output_);
+          } else {
+            tmp_scale_ =
+              GetQuantizeScale(output.dtype(), cached_min_output_, cached_max_output_) / data_scale_;
+          }
         } else {
           tmp_scale_ = 1.0 / data_scale_;
         }
@@ -405,6 +411,10 @@ static void SgMKLDNNFCParamParser(nnvm::NodeAttrs *attrs) {
       if (op_name == "Activation") {
         const ActivationParam act_param = nnvm::get<ActivationParam>(node->attrs.parsed);
         full_param.eltwise_param.alg = GetMKLDNNActAlgo(act_param);
+      } else if (op_name == "LeakyReLU") {
+        const auto act_param = nnvm::get<LeakyReLUParam>(node->attrs.parsed);
+        full_param.eltwise_param.alpha = act_param.slope;
+        full_param.eltwise_param.alg = GetMKLDNNActAlgo(act_param);
       } else if (op_name == "clip") {
         const ClipParam clip_param = nnvm::get<ClipParam>(node->attrs.parsed);
         full_param.eltwise_param.alg = mkldnn::algorithm::eltwise_bounded_relu;
diff --git a/src/operator/subgraph/mkldnn/mkldnn_fc_property.h b/src/operator/subgraph/mkldnn/mkldnn_fc_property.h
index aecb3a7a8477..432772d36298 100644
--- a/src/operator/subgraph/mkldnn/mkldnn_fc_property.h
+++ b/src/operator/subgraph/mkldnn/mkldnn_fc_property.h
@@ -102,6 +102,16 @@ class SgMKLDNNFCSelector : public SubgraphSelector {
             return true;
           }
         }
+        if (new_node.op() == Op::Get("LeakyReLU")) {
+          const LeakyReLUParam &param = nnvm::get<LeakyReLUParam>(new_node.attrs.parsed);
+          if (param.act_type == leakyrelu::kLeakyReLU ||
+              param.act_type == leakyrelu::kELU ||
+              param.act_type == leakyrelu::kGELU) {
+            matched_list_.push_back(&new_node);
+            status_ = kSuccess;
+            return true;
+          }
+        }
         if (!quantized_ && (new_node.op() == Op::Get("square") ||
             new_node.op() == Op::Get("sqrt") ||
             new_node.op() == Op::Get("exp"))) {
diff --git a/tests/python/mkl/test_subgraph.py b/tests/python/mkl/test_subgraph.py
index 65b73e438ea6..f4d421c1df39 100644
--- a/tests/python/mkl/test_subgraph.py
+++ b/tests/python/mkl/test_subgraph.py
@@ -49,7 +49,7 @@
 }
 
 DATA_SHAPE=[(64, 4, 10, 10), (4, 3, 24, 24), (1, 16, 32, 32)]
-fc_post_ops_list=['relu', 'sigmoid', 'tanh', 'softrelu',
+fc_post_ops_list=['relu', 'sigmoid', 'tanh', 'softrelu', 'gelu',
                   'square', 'square_root', 'abs', 'exp', 'bounded_relu']
 
 def check_qsym_calibrated(qsym, out_type, name='conv'):
@@ -654,6 +654,17 @@ def single_fc(no_bias, data_shape, flatten=True):
                                 no_bias=no_bias, flatten=flatten)
   return fc, attr
 
+def fc_gelu(no_bias, data_shape, flatten):
+  attrs = {'fc': {'with_eltwise': 'true'}}
+  data, weight = head_symbol(data_shape)
+  weight_2 = mx.symbol.Variable('2nd_weight', dtype='float32')
+  fc = mx.symbol.FullyConnected(name='fc', data=data, weight=weight, num_hidden=6,
+                                no_bias=no_bias, flatten=flatten)
+  sym = mx.symbol.LeakyReLU(data=fc, act_type='gelu')
+  sym = mx.symbol.FullyConnected(name='fc2', data=sym, weight=weight_2, num_hidden=6,
+                                no_bias=no_bias, flatten=flatten)
+  return sym, attrs
+
 # fc + eltwise fusion case
 def fc_eltwise(no_bias, data_shape, flatten=True, alg='relu'):
   assert alg in fc_post_ops_list
@@ -664,6 +675,8 @@ def fc_eltwise(no_bias, data_shape, flatten=True, alg='relu'):
                                 no_bias=no_bias, flatten=flatten)
   if alg in ['relu', 'sigmoid', 'tanh', 'softrelu']:
     sym = mx.symbol.Activation(data=fc, name='act', act_type=alg)
+  elif alg == "gelu":
+    sym = mx.symbol.LeakyReLU(data=fc, act_type='gelu')
   elif alg == 'square':
     sym = mx.symbol.square(data=fc, name='square')
   elif alg == 'square_root':
@@ -865,6 +878,12 @@ def test_fc_eltwise():
     else:
       check_fusion(syms, dshape, attrs, check_quantization=False)
 
+@with_seed()
+def test_fc_gelu():
+  for dshape, no_bias, flatten in itertools.product(DATA_SHAPE, [True, False], [True, False]):
+    sym, attrs = fc_gelu(no_bias, dshape, flatten)
+    check_fusion(sym, dshape, attrs, check_quantization=True)
+
 @with_seed()
 def test_neg_fc_relu():
   for dshape, no_bias, flatten in itertools.product(DATA_SHAPE, [True, False], [True, False]):