diff --git a/src/operators/kernel/fpga/mul_kernel.cpp b/src/operators/kernel/fpga/mul_kernel.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..07aa4bcc43d28805ab0660bf89149c5ec5f1c732
--- /dev/null
+++ b/src/operators/kernel/fpga/mul_kernel.cpp
@@ -0,0 +1,70 @@
+/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License. */
+
+#ifdef MUL_OP
+
+#include "operators/kernel/mul_kernel.h"
+
+namespace paddle_mobile {
+namespace operators {
+
+template <>
+bool MulKernel<FPGA, float>::Init(MulParam<FPGA> *param) {
+  bool relu_enabled = false;
+  auto input_x = const_cast<LoDTensor *>(param->InputX());
+  auto filter = const_cast<LoDTensor *>(param->InputY());
+  auto out = param->Out();
+
+  PADDLE_MOBILE_ENFORCE(input_x->dims()[1] == filter->dims()[0],
+                        "Image channel should be equal to weight number");
+  int channel = (uint32_t)out->dims()[1];
+  auto bs_ptr =
+      (float *)fpga::fpga_malloc(2 * channel * sizeof(float));  // NOLINT
+  for (int i = 0; i < channel; i++) {
+    bs_ptr[i + channel] = 1;
+    bs_ptr[i] = 0;
+  }
+  int num = (uint32_t)filter->dims()[1];
+  int chw = (uint32_t)filter->dims()[0];
+  PADDLE_MOBILE_ENFORCE(
+      chw == input_x->numel(),
+      "Filter element num should be equal to IFM element num");
+  int height = (uint32_t)input_x->dims()[2];
+  int width = (uint32_t)input_x->dims()[3];
+  int filter_channel = chw / height / width;
+
+  filter->Resize(framework::make_ddim({num, filter_channel, height, width}));
+  float max_value = fpga::filter_find_max(filter);
+  fpga::format_fc_filter(filter, max_value);
+
+  int element_num_per_div = fpga::get_filter_num_per_div(filter, 1);
+  fpga::format_bias_scale_array(&bs_ptr, element_num_per_div, channel);
+  fpga::format_fp16_ofm(out);
+
+  fpga::WrapperConvArgs conv_arg = {0};
+  fpga::fill_conv_arg(&conv_arg, input_x, out, filter, relu_enabled, 1, 1, 1, 0,
+                      0, bs_ptr);
+  param->SetFpgaArgs(conv_arg);
+  return true;
+}
+
+template <>
+void MulKernel<FPGA, float>::Compute(const MulParam<FPGA> &param) const {
+  fpga::ComputeFpgaConv(param.FpgaArgs());
+}
+
+}  // namespace operators
+}  // namespace paddle_mobile
+
+#endif
diff --git a/src/operators/mul_op.cpp b/src/operators/mul_op.cpp
index a6b055b62fa25fbca2a85dfa386fa406e207b2e9..69e3bb300d741e74ab8d6eea6c62052b4d0d8f1d 100644
--- a/src/operators/mul_op.cpp
+++ b/src/operators/mul_op.cpp
@@ -61,5 +61,7 @@ REGISTER_OPERATOR_CPU(mul, ops::MulOp);
 #ifdef PADDLE_MOBILE_MALI_GPU
 REGISTER_OPERATOR_MALI_GPU(mul, ops::MulOp);
 #endif
-
+#ifdef PADDLE_MOBILE_FPGA
+REGISTER_OPERATOR_FPGA(mul, ops::MulOp);
+#endif
 #endif
diff --git a/src/operators/op_param.h b/src/operators/op_param.h
index f1c9f09b65ec35960320ab7b2aca0bc2904cc838..f50c00457fee099a5fb26415b846de945a66d73e 100644
--- a/src/operators/op_param.h
+++ b/src/operators/op_param.h
@@ -438,6 +438,15 @@ class MulParam : OpParam {
   GType *out_;
   int x_num_col_dims_;
   int y_num_col_dims_;
+#ifdef PADDLE_MOBILE_FPGA
+
+ private:
+  fpga::WrapperConvArgs fpga_conv_args;
+
+ public:
+  const fpga::WrapperConvArgs &FpgaArgs() const { return fpga_conv_args; }
+  void SetFpgaArgs(const fpga::WrapperConvArgs &args) { fpga_conv_args = args; }
+#endif
 };
 #endif
 
diff --git a/test/fpga/test_resnet50.cpp b/test/fpga/test_resnet50.cpp
index cca6793f10da5a0784cf8a3ba2d0104f3508028d..f850eb3e5ea3a03fe90d82c1eca2af6c9f8e9106 100644
--- a/test/fpga/test_resnet50.cpp
+++ b/test/fpga/test_resnet50.cpp
@@ -18,8 +18,9 @@ static const char *g_resnet_combine = "../models/resnet50";
 int main() {
   DLOG << paddle_mobile::fpga::open_device();
   paddle_mobile::PaddleMobile<paddle_mobile::FPGA> paddle_mobile;
-  if (paddle_mobile.Load(std::string(g_resnet_combine) + "/model",
-                         std::string(g_resnet_combine) + "/params", true)) {
+  //  if (paddle_mobile.Load(std::string(g_resnet_combine) + "/model",
+  //                         std::string(g_resnet_combine) + "/params", true)) {
+  if (paddle_mobile.Load(std::string(g_resnet_combine), true)) {
     std::vector<int64_t> dims{1, 3, 224, 224};
     Tensor input_tensor;
     SetupTensor<float>(&input_tensor, {1, 3, 224, 224}, static_cast<float>(0),
diff --git a/tools/op.cmake b/tools/op.cmake
index 9a6ec0a147b564296d89113a2838cc6bd73975a1..898f66a634d70a5def7c7ce328a7a291d9b55c70 100644
--- a/tools/op.cmake
+++ b/tools/op.cmake
@@ -121,6 +121,7 @@ if (CON GREATER -1)
   set(FUSION_CONVBNRELU_OP ON)
   set(FUSION_CONVBN_OP ON)
   set(FUSION_CONVADD_OP ON)
+  set(MUL_OP ON)
 
   set(FOUND_MATCH ON)
 endif()