compress lib for specific net

CMakeLists.txt

 cmake_minimum_required(VERSION 3.0)
 project(paddle-mobile)
-#add_definitions(-DPADDLE_MOBILE_DEBUG)
+add_definitions(-DPADDLE_MOBILE_DEBUG)
 add_definitions(-DENABLE_EXCEPTION)
 
+#add_definitions(-DARMV7)
+#add_definitions(-DARMV8)
+#add_definitions(-DIOS)
+add_definitions(-DX86)
+
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
 set(CMAKE_BUILD_TYPE RelWithDebInfo)
 set(CMAKE_VERBOSE_MAKEFILE ON)
@@ -16,6 +21,75 @@ file(GLOB_RECURSE PADDLE_MOBILE_H src/*.h)
 
 include_directories(src/)
 
+if (googlenet)
+    add_definitions(-DCONCAT_OP)
+    add_definitions(-DCONV_OP)
+    add_definitions(-DLRN_OP)
+    add_definitions(-DMUL_OP)
+    add_definitions(-DELEMENTWISEADD_OP)
+    add_definitions(-DFUSION_FC_OP)
+    add_definitions(-DPOOL_OP)
+    add_definitions(-DRELU_OP)
+elseif (mobilenet)
+    add_definitions(-DCONV_OP)
+    add_definitions(-DELEMENTWISEADD_OP)
+    add_definitions(-DRELU_OP)
+    add_definitions(-DSOFTMAX_OP)
+    add_definitions(-DSOFTMAX_OP)
+    add_definitions(-DDEPTHWISECONV_OP)
+    add_definitions(-DBATCHNORM_OP)
+    add_definitions(-DPOOL_OP)
+    add_definitions(-DRESHAPE_OP)
+elseif (yolo)
+    add_definitions(-DBATCHNORM_OP)
+    add_definitions(-DCONV_OP)
+    add_definitions(-DRELU_OP)
+    add_definitions(-DELEMENTWISEADD_OP)
+elseif (squeezenet)
+    add_definitions(-DCONCAT_OP)
+    add_definitions(-DCONV_OP)
+    add_definitions(-DRELU_OP)
+    add_definitions(-DELEMENTWISEADD_OP)
+    add_definitions(-DPOOL_OP)
+    add_definitions(-DRESHAPE_OP)
+    add_definitions(-DSOFTMAX_OP)
+elseif(resnet)
+    add_definitions(-DCONV_OP)
+    add_definitions(-DBATCHNORM_OP)
+    add_definitions(-DELEMENTWISEADD_OP)
+    add_definitions(-DSOFTMAX_OP)
+    add_definitions(-DMUL_OP)
+    add_definitions(-DPOOL_OP)
+    add_definitions(-DRELU_OP)
+else ()
+    add_definitions(-DBATCHNORM_OP)
+    add_definitions(-DBOXCODER_OP)
+    add_definitions(-DCONCAT_OP)
+    add_definitions(-DCONV_OP)
+    add_definitions(-DDEPTHWISECONV_OP)
+    add_definitions(-DELEMENTWISEADD_OP)
+    add_definitions(-DFUSIONCONVADD_OP)
+    add_definitions(-DCONVADDRELU_OP)
+    add_definitions(-DFUSION_FC_OP)
+    add_definitions(-DLRN_OP)
+    add_definitions(-DMUL_OP)
+    add_definitions(-DMULTICLASSNMS_OP)
+    add_definitions(-DPOOL_OP)
+    add_definitions(-DPRIORBOX_OP)
+    add_definitions(-DRELU_OP)
+    add_definitions(-DRESHAPE_OP)
+    add_definitions(-DSIGMOID_OP)
+    add_definitions(-DSOFTMAX_OP)
+    add_definitions(-DTRANSPOSE_OP)
+endif()
+
 add_library(paddle-mobile SHARED ${PADDLE_MOBILE_CC} ${PADDLE_MOBILE_H})
 
+if (googlenet)
+elseif (mobilenet)
+elseif (yolo)
+elseif (squeezenet)
+elseif(resnet)
+else ()
+endif()
 add_subdirectory(test)

build.sh

@@ -33,6 +33,8 @@ build_for_mac() {
 }
 
 build_for_android() {
+
+
     if [ -z "${ANDROID_NDK}" ]; then
         echo "ANDROID_NDK not found!"
         exit -1
@@ -55,11 +57,13 @@ build_for_android() {
         exit -1
     fi
 
+
     MODE="Release"
     ANDROID_PLATFORM_VERSION="android-15"
     TOOLCHAIN_FILE="./tools/android-cmake/android.toolchain.cmake"
     ANDROID_ARM_MODE="arm"
-
+    if [ $# -eq 1 ]; then
+    NET=$1
     cmake . \
         -B"build/release/${PLATFORM}" \
         -DANDROID_ABI="${ABI}" \
@@ -69,10 +73,24 @@ build_for_android() {
         -DCMAKE_CXX_FLAGS="${CXX_FLAGS}" \
         -DANDROID_STL=c++_static \
         -DANDROID=true \
+        -D"${NET}=true" \
         -D"${ARM_PLATFORM}"=true
+    else
 
+    cmake . \
+        -B"build/release/${PLATFORM}" \
+        -DANDROID_ABI="${ABI}" \
+        -DCMAKE_BUILD_TYPE="${MODE}" \
+        -DCMAKE_TOOLCHAIN_FILE="${TOOLCHAIN_FILE}" \
+        -DANDROID_PLATFORM="${ANDROID_PLATFORM_VERSION}" \
+        -DCMAKE_CXX_FLAGS="${CXX_FLAGS}" \
+        -DANDROID_STL=c++_static \
+        -DANDROID=true \
+        -D"${ARM_PLATFORM}"=true
+    fi
     cd "./build/release/${PLATFORM}"
     make -j 8
+
 }
 
 build_for_ios() {
@@ -106,15 +124,44 @@ if [ $# -lt 1 ]; then
     echo "available targets: mac|linux|ios|android"
     echo "sample usage: ./build.sh mac"
 else
-	if [ $1 = "mac" ]; then
-		build_for_mac
-	elif [ $1 = "linux" ]; then
-		build_for_linux
-	elif [ $1 = "android" ]; then
-		build_for_android
-	elif [ $1 = "ios" ]; then
-		build_for_ios
-	else
-		build_error
+    if [ $# -eq 2 ]; then
+    
+        if [[$2 != "googlenet"]] -a [[$2 != "mobilenet"]] -a [[$2 != "yolo"]] -a [[$2 != "squeezenet"]] -a [[$2 != "resnet"]]; then
+            if [ $1 = "mac" ]; then
+		        build_for_mac
+	        elif [ $1 = "linux" ]; then
+		        build_for_linux
+	        elif [ $1 = "android" ]; then
+		        build_for_android
+	        elif [ $1 = "ios" ]; then
+		        build_for_ios
+	        else
+		        build_error
+	        fi
+        else
+            if [ $1 = "mac" ]; then
+		        build_for_mac $2
+	        elif [ $1 = "linux" ]; then
+		        build_for_linux $2
+	        elif [ $1 = "android" ]; then
+		        build_for_android $2
+	        elif [ $1 = "ios" ]; then
+		        build_for_ios $2
+	        else
+		        build_error
+	        fi
+        fi
+    else
+        if [ $1 = "mac" ]; then
+		    build_for_mac
+	    elif [ $1 = "linux" ]; then
+		    build_for_linux
+	    elif [ $1 = "android" ]; then
+		    build_for_android
+	    elif [ $1 = "ios" ]; then
+		    build_for_ios
+	    else
+		    build_error
+	    fi
 	fi
 fi

src/common/enforce.h

@@ -18,7 +18,6 @@ limitations under the License. */
 #include <stdio.h>
 #include <exception>
 #include <sstream>
-#include <stdexcept>
 #include <string>
 
 #endif

src/framework/attribute.cpp

@@ -17,14 +17,8 @@ limitations under the License. */
 namespace paddle_mobile {
 namespace framework {
 
-/*
- * Variant<int, float, std::string, std::vector<int>, std::vector<float>,
-          std::vector<std::string>, bool, std::vector<bool>, BlockDesc *,
-          int64_t>
- * */
-
 struct PrintVistor : Vistor<Print &> {
-  PrintVistor(Print &printer) : printer_(printer) {}
+  explicit PrintVistor(Print &printer) : printer_(printer) {}
   template <typename T>
   Print &operator()(const T &value) {
     printer_ << value;

src/framework/attribute.h

@@ -14,7 +14,9 @@ limitations under the License. */
 
 #pragma once
 
+#include <string>
 #include <unordered_map>
+#include <vector>
 #include "common/enforce.h"
 #include "common/log.h"
 #include "common/variant.h"
@@ -22,28 +24,15 @@ limitations under the License. */
 
 namespace paddle_mobile {
 namespace framework {
+using std::string;
+using std::vector;
 
 class BlockDesc;
 
 class Attribute {
  public:
-  /*
-   *  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__INT = 0,
-  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__FLOAT = 1,
-  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__STRING = 2,
-  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__INTS = 3,
-  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__FLOATS = 4,
-  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__STRINGS = 5,
-  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BOOLEAN = 6,
-  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BOOLEANS = 7,
-  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BLOCK = 8,
-  PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__LONG = 9
-    PROTOBUF_C__FORCE_ENUM_TO_BE_INT_SIZE(PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE)
-   *
-   * */
   static Attribute GetAttrValue(
       PaddleMobile__Framework__Proto__OpDesc__Attr *attr_desc) {
-    //    std::cout << "begin get attr value" << std::endl;
     Attribute attr;
     switch (attr_desc->type) {
       case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BOOLEAN: {
@@ -63,35 +52,35 @@ class Attribute {
         break;
       }
       case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__BOOLEANS: {
-        std::vector<bool> val(attr_desc->n_bools);
+        vector<bool> val(attr_desc->n_bools);
         for (int i = 0; i < attr_desc->n_bools; ++i) {
           val[i] = attr_desc->bools[i];
         }
-        attr.Set<std::vector<bool>>(val);
+        attr.Set<vector<bool>>(val);
         break;
       }
       case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__INTS: {
-        std::vector<int> val(attr_desc->n_ints);
+        vector<int> val(attr_desc->n_ints);
         for (int i = 0; i < attr_desc->n_ints; ++i) {
           val[i] = attr_desc->ints[i];
         }
-        attr.Set<std::vector<int>>(val);
+        attr.Set<vector<int>>(val);
         break;
       }
       case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__FLOATS: {
-        std::vector<float> val(attr_desc->n_floats);
+        vector<float> val(attr_desc->n_floats);
         for (int i = 0; i < attr_desc->n_floats; ++i) {
           val[i] = attr_desc->floats[i];
         }
-        attr.Set<std::vector<float>>(val);
+        attr.Set<vector<float>>(val);
         break;
       }
       case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__STRINGS: {
-        std::vector<std::string> val(attr_desc->n_strings);
+        vector<string> val(attr_desc->n_strings);
         for (int i = 0; i < attr_desc->n_strings; ++i) {
           val[i] = attr_desc->strings[i];
         }
-        attr.Set<std::vector<std::string>>(val);
+        attr.Set<vector<string>>(val);
         break;
       }
       case PADDLE_MOBILE__FRAMEWORK__PROTO__ATTR_TYPE__LONG: {
@@ -122,21 +111,18 @@ class Attribute {
       return vistor(attr.variant_.Get<int>());
     } else if (attr.variant_.TypeId() == typeid(float).hash_code()) {
       return vistor(attr.variant_.Get<float>());
-    } else if (attr.variant_.TypeId() == typeid(std::string).hash_code()) {
-      return vistor(attr.variant_.Get<std::string>());
-    } else if (attr.variant_.TypeId() == typeid(std::vector<int>).hash_code()) {
-      return vistor(attr.variant_.Get<std::vector<int>>());
-    } else if (attr.variant_.TypeId() ==
-               typeid(std::vector<float>).hash_code()) {
-      return vistor(attr.variant_.Get<std::vector<float>>());
-    } else if (attr.variant_.TypeId() ==
-               typeid(std::vector<std::string>).hash_code()) {
-      return vistor(attr.variant_.Get<std::vector<std::string>>());
+    } else if (attr.variant_.TypeId() == typeid(string).hash_code()) {
+      return vistor(attr.variant_.Get<string>());
+    } else if (attr.variant_.TypeId() == typeid(vector<int>).hash_code()) {
+      return vistor(attr.variant_.Get<vector<int>>());
+    } else if (attr.variant_.TypeId() == typeid(vector<float>).hash_code()) {
+      return vistor(attr.variant_.Get<vector<float>>());
+    } else if (attr.variant_.TypeId() == typeid(vector<string>).hash_code()) {
+      return vistor(attr.variant_.Get<vector<string>>());
     } else if (attr.variant_.TypeId() == typeid(bool).hash_code()) {
       return vistor(attr.variant_.Get<bool>());
-    } else if (attr.variant_.TypeId() ==
-               typeid(std::vector<bool>).hash_code()) {
-      return vistor(attr.variant_.Get<std::vector<bool>>());
+    } else if (attr.variant_.TypeId() == typeid(vector<bool>).hash_code()) {
+      return vistor(attr.variant_.Get<vector<bool>>());
     } else if (attr.variant_.TypeId() == typeid(int64_t).hash_code()) {
       return vistor(attr.variant_.Get<int64_t>());
     } else {
@@ -145,24 +131,21 @@ class Attribute {
   }
 
  private:
-  Variant<int, float, std::string, std::vector<int>, std::vector<float>,
-          std::vector<std::string>, bool, std::vector<bool>, BlockDesc *,
-          int64_t>
+  Variant<int, float, string, vector<int>, vector<float>, vector<string>, bool,
+          vector<bool>, BlockDesc *, int64_t>
       variant_;
 };
 
-using AttributeMap = std::unordered_map<std::string, Attribute>;
+using AttributeMap = std::unordered_map<string, Attribute>;
 
 class AttrReader {
  public:
   explicit AttrReader(const AttributeMap &attrs) : attrs_(attrs) {}
 
   template <typename T>
-  inline T Get(const std::string &name) const {
-    //          PADDLE_ENFORCE(attrs_.count(name) != 0, "%s should
-    //          be in
-    //          AttributeMap",
-    //                         name);
+  inline T Get(const string &name) const {
+    PADDLE_MOBILE_ENFORCE(attrs_.count(name) != 0,
+                          "%s should  be in AttributeMap", name);
     return ((Attribute)attrs_.at(name)).Get<T>();
   }
 

src/framework/data_layout.h

@@ -54,7 +54,6 @@ inline std::string DataLayoutToString(const DataLayout &data_layout) {
       return "ANY_LAYOUT";
     default:
       break;
-      //      std::cout << "unknown DataLayou %d", data_layout;
   }
 }
 

src/framework/data_transform.cpp

-/* 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. */
-
-#pragma once
-
-#include "framework/data_transform.h"
-
-namespace paddle_mobile {
-namespace framework {
-
-static void PassTensorData(Tensor *from, Tensor *to) {
-  to->ShareDataWith(*from);
-  *from = Tensor();
-}
-
-void DataTransform(const OpKernelType &expected_kernel_type,
-                   const OpKernelType &kernel_type_for_var,
-                   const Tensor &input_tensor, Tensor *output_tensor) {
-  bool transformed = false;
-  Tensor in;
-  in.ShareDataWith(input_tensor);
-  Tensor out;
-
-  //  // do layout transform
-  //  if (NeedTransformLayout(expected_kernel_type.data_layout_,
-  //                          kernel_type_for_var.data_layout_)) {
-  //    TransDataLayout(kernel_type_for_var, expected_kernel_type, in,
-  //    &out);
-  //    transformed = true;
-  //    PassTensorData(&out, &in);
-  //  }
-  //
-  //  // do data type transform
-  //  if (expected_kernel_type.data_type_ !=
-  //  kernel_type_for_var.data_type_) {
-  //    TransDataType(kernel_type_for_var, expected_kernel_type, in,
-  //    &out);
-  //    transformed = true;
-  //    PassTensorData(&out, &in);
-  //  }
-  //
-  //  // do device transform
-  //  if (!platform::is_same_place(kernel_type_for_var.place_,
-  //                               expected_kernel_type.place_)) {
-  //    TransDataDevice(in, expected_kernel_type.place_, &out);
-  //    transformed = true;
-  //    PassTensorData(&out, &in);
-  //  }
-  //
-  //  PADDLE_ENFORCE(transformed, "No transform is applied, please
-  //  check!");
-  // get output data
-  output_tensor->ShareDataWith(in);
-}
-
-void CopyVariableWithTensor(const Variable &in_var, const Tensor &tensor,
-                            Variable *out_var) {
-  //  if (in_var.IsType<LoDTensor>()) {
-  //    auto& in_lod_tensor = in_var.Get<LoDTensor>();
-  //    auto* tran_lod_tensor = out_var.GetMutable<LoDTensor>();
-  //    tran_lod_tensor->set_lod(in_lod_tensor.lod());
-  //    tran_lod_tensor->set_layout(in_lod_tensor.layout());
-  //    tran_lod_tensor->ShareDataWith(tensor);
-  //  } else if (in_var.IsType<SelectedRows>()) {
-  //    auto& in_selected_rows = in_var.Get<SelectedRows>();
-  //    auto* trans_selected_rows =
-  //    out_var.GetMutable<SelectedRows>();
-  //    trans_selected_rows->set_height(in_selected_rows.height());
-  //    trans_selected_rows->set_rows(in_selected_rows.rows());
-  //    trans_selected_rows->mutable_value()->ShareDataWith(tensor);
-  //  } else {
-  //    PADDLE_THROW("unknown var type");
-  //  }
-}
-
-}  // namespace framework
-}  // namespace paddle_mobile

src/framework/data_type.h

-/* 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. */
-
-#pragma once
-
-namespace paddle_mobile {
-namespace framework {
-
-//    inline proto::VarType::Type ToDataType(std::type_index type) {
-//        using namespace paddle_mobile::framework::proto;
-//        if (typeid(float).hash_code() == type.hash_code()) {
-//            return proto::VarType::FP32;
-//        } else if (typeid(double).hash_code() == type.hash_code()) {
-//            return proto::VarType::FP64;
-//        } else if (typeid(int).hash_code() == type.hash_code()) {
-//            return proto::VarType::INT32;
-//        } else if (typeid(int64_t).hash_code() == type.hash_code()) {
-//            return proto::VarType::INT64;
-//        } else if (typeid(bool).hash_code() == type.hash_code()) {
-//            return proto::VarType::BOOL;
-//        } else {
-////            PADDLE_THROW("Not supported");
-//        }
-//    }
-}  // namespace framework
-}  // namespace paddle_mobile

src/framework/ddim.cpp

@@ -183,7 +183,7 @@ DDim DDim::operator*(DDim d) const {
 
 int64_t get(const DDim &ddim, int idx) { return ddim[idx]; }
 
-void set(DDim &ddim, int idx, int value) { ddim[idx] = value; }
+void set(DDim *ddim, int idx, int value) { (*ddim)[idx] = value; }
 
 /// @cond HIDDEN
 struct VectorizeVisitor : Vistor<void> {

src/framework/ddim.h

@@ -83,17 +83,6 @@ struct DDim {
 
   int64_t operator[](int idx) const;
 
-  //  template <typename Visitor>
-  //  typename Visitor::result_type apply_visitor(Visitor& visitor) {
-  //    return var.apply_visitor(visitor);
-  //  }
-  //
-  //  template <typename Visitor>
-  //  typename Visitor::result_type apply_visitor(Visitor& visitor)
-  //  const {
-  //    return var.apply_visitor(visitor);
-  //  }
-
   DDimVar getVar() { return var; }
 
   bool operator==(DDim d) const;
@@ -126,7 +115,7 @@ DDim make_ddim(std::initializer_list<int64_t> dims);
 
 int64_t get(const DDim &dim, int idx);
 
-void set(DDim &dim, int idx, int val);
+void set(DDim *dim, int idx, int val);
 
 std::vector<int64_t> vectorize(const DDim &ddim);
 

src/framework/lod_tensor.cpp

@@ -42,23 +42,10 @@ std::ostream &operator<<(std::ostream &os, const LoD &lod) {
 }
 
 std::ostream &operator<<(std::ostream &os, const LoDTensor &t) {
-  //  PADDLE_ENFORCE(t.type().hash_code() ==
-  //  typeid(float).hash_code());
-
-  //  if (!platform::is_cpu_place(t.place())) {
-  //    LoDTensor tt;
-  //    framework::TensorCopy(t, platform::CPUPlace(), &tt);
-  //    platform::DeviceContextPool &pool =
-  //    platform::DeviceContextPool::Instance(); auto &dev_ctx =
-  //    *pool.Get(t.place()); dev_ctx.Wait();
-  //
-  //    os << tt;
-  //    return os;
-  //  }
-
+  PADDLE_MOBILE_ENFORCE(t.type().hash_code() == typeid(float).hash_code(),
+                        "t.type() is not float");
   os << "dim: " << t.dims() << "\n";
   os << "lod: " << t.lod() << "\n";
-
   // only print first ten elements
   int64_t size = t.numel() < 10 ? t.numel() : 10;
   for (int64_t i = 0; i < size; ++i) {
@@ -76,9 +63,9 @@ std::string LoDToString(const LoD &lod) {
 
 LoD SliceInLevel(const LoD &in, size_t level, size_t elem_begin,
                  size_t elem_end) {
-  //  PADDLE_ENFORCE_LT(level, in.size());
-  //  PADDLE_ENFORCE_LT(elem_end, in[level].size());
-
+  PADDLE_MOBILE_ENFORCE(level < in.size(), "level should >= in.size()");
+  PADDLE_MOBILE_ENFORCE(elem_end < in[level].size(),
+                        "elem_end >= in[level].size()");
   LoD res;
   res.resize(in.size() - level);
   // copy the first level
@@ -211,8 +198,9 @@ LoDAndOffset GetSubLoDAndAbsoluteOffset(const LoD &lod, size_t start_idx,
   LoD sub_lod;
 
   for (size_t level_idx = start_level; level_idx < lod.size(); ++level_idx) {
-    //    PADDLE_ENFORCE_LE(start_idx, end_idx);
-    //    PADDLE_ENFORCE_LT(end_idx, lod[level_idx].size());
+    PADDLE_MOBILE_ENFORCE(start_idx <= end_idx, "start_idx > end_idx");
+    PADDLE_MOBILE_ENFORCE(end_idx < lod[level_idx].size(),
+                          "end_idx >= lod[level_idx].size()");
     std::vector<size_t> level_lens;
     for (size_t i = start_idx; i < end_idx; ++i) {
       level_lens.push_back(lod[level_idx][i + 1] - lod[level_idx][i]);
@@ -226,10 +214,9 @@ LoDAndOffset GetSubLoDAndAbsoluteOffset(const LoD &lod, size_t start_idx,
 }
 
 void AppendLoD(LoD *lod, const LoD &lod_length) {
-  //  PADDLE_ENFORCE(
-  //      lod->empty() || lod->size() == lod_length.size(),
-  //      "The lod_length should has the same size with the appended
-  //      lod.");
+  PADDLE_MOBILE_ENFORCE(
+      lod->empty() || lod->size() == lod_length.size(),
+      "The lod_length should has the same size with the appended lod.");
   if (lod->empty()) {
     for (size_t i = 0; i < lod_length.size(); ++i) {
       lod->emplace_back(1, 0);  // size = 1, value = 0;

src/framework/op_info.h

@@ -25,9 +25,8 @@ template <typename Dtype>
 struct OpInfo {
   OpCreator<Dtype> creator_;
   const OpCreator<Dtype> &Creator() const {
-    //    PADDLE_ENFORCE_NOT_NULL(creator_,
-    //                            "Operator Creator has not been
-    //                            registered");
+    PADDLE_MOBILE_ENFORCE(creator_ != nullptr,
+                          "Operator Creator has not been registered");
     return creator_;
   }
 };
@@ -48,17 +47,15 @@ class OpInfoMap {
   }
 
   void Insert(const std::string &type, const OpInfo<Dtype> &info) {
-    //    PADDLE_ENFORCE(!Has(type), "Operator %s has been
-    //    registered", type);
+    PADDLE_MOBILE_ENFORCE(!Has(type), "Operator %s has been registered",
+                          type.c_str());
     map_.insert({type, info});
   }
 
   const OpInfo<Dtype> &Get(const std::string &type) const {
     auto op_info_ptr = GetNullable(type);
-    //    PADDLE_ENFORCE_NOT_NULL(op_info_ptr, "Operator %s has not
-    //    been
-    //    registered",
-    //                            type);
+    PADDLE_MOBILE_ENFORCE(op_info_ptr != nullptr,
+                          "Operator %s has not been registered", type.c_str());
     return *op_info_ptr;
   }
 

src/framework/program/program-optimize/node.cpp

@@ -12,8 +12,6 @@ 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. */
 
-#include <sstream>
-
 #include "framework/operator.h"
 #include "framework/program/program-optimize/node.h"
 

src/framework/scope.cpp

@@ -76,7 +76,6 @@ void Scope::DeleteScope(Scope *scope) const {
   auto it = std::find(kids_.begin(), kids_.end(), scope);
   kids_.erase(it);
   delete scope;
-  // deferent
 }
 
 void Scope::EraseVars(const std::vector<std::string> &var_names) {
@@ -104,14 +103,6 @@ void Scope::Rename(const std::string &origin_name,
   vars_[new_name] = origin_it->second;
   vars_.erase(origin_it);
 }
-//
-//            std::string Scope::Rename(const std::string& origin_name)
-//            const {
-//                auto var_name = string::Sprintf("%p.%d", this,
-//                vars_.size());
-//                Rename(origin_name, var_name);
-//                return var_name;
-//            }
 
 Variable *Scope::FindVarLocally(const std::string &name) const {
   auto it = vars_.find(name);

src/io.h

@@ -14,10 +14,10 @@ limitations under the License. */
 
 #pragma once
 
-#include <memory.h>
 #include <map>
 #include <string>
 #include <vector>
+#include <memory>
 
 #include "common/types.h"
 #include "framework/lod_tensor.h"
@@ -27,7 +27,7 @@ limitations under the License. */
 
 namespace paddle_mobile {
 
-template <typename Dtype, Precision P = Precision::FP32>
+template <typename Dtype=CPU, Precision P = Precision::FP32>
 class Loader {
  public:
   const framework::Program<Dtype, P> Load(const std::string &dirname,
@@ -39,7 +39,7 @@ class Loader {
                const std::string &file_path);
 };
 
-template <typename Dtype, Precision P = Precision::FP32>
+template <typename Dtype=CPU, Precision P = Precision::FP32>
 class Executor {
  public:
   typedef typename PrecisionTrait<P>::ptype Ptype;

src/operators/batchnorm_op.cpp

@@ -12,6 +12,8 @@ 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 BATCHNORM_OP
+
 #include "batchnorm_op.h"
 
 namespace paddle_mobile {
@@ -29,3 +31,5 @@ template class BatchNormOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(batch_norm);
 REGISTER_OPERATOR(batch_norm, ops::BatchNormOp);
+
+#endif

src/operators/batchnorm_op.h

@@ -12,6 +12,8 @@ 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 BATCHNORM_OP
+
 #pragma once
 
 #include <string>
@@ -47,3 +49,5 @@ class BatchNormOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/box_coder_op.cpp

@@ -12,6 +12,8 @@ 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 BOXCODER_OP
+
 #include "operators/box_coder_op.h"
 #include <vector>
 namespace paddle_mobile {
@@ -52,3 +54,5 @@ template class BoxCoderOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(box_coder);
 REGISTER_OPERATOR(box_coder, ops::BoxCoderOp);
+
+#endif

src/operators/box_coder_op.h

@@ -12,6 +12,8 @@ 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 BOXCODER_OP
+
 #pragma once
 
 #include <string>
@@ -50,3 +52,5 @@ class BoxCoderOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/concat_op.cpp

@@ -12,6 +12,8 @@ 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 CONCAT_OP
+
 #include "concat_op.h"
 
 namespace paddle_mobile {
@@ -62,3 +64,5 @@ template class ConcatOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(concat);
 REGISTER_OPERATOR(concat, ops::ConcatOp);
+
+#endif

src/operators/concat_op.h

@@ -12,6 +12,8 @@ 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 CONCAT_OP
+
 #pragma once
 
 #include <string>
@@ -45,3 +47,5 @@ class ConcatOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/conv_op.cpp

@@ -12,9 +12,10 @@ 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 CONV_OP
+
 #include "operators/conv_op.h"
 #include <vector>
-#include "framework/data_type.h"
 #include "framework/op_proto_maker.h"
 #include "framework/op_registry.h"
 
@@ -23,30 +24,11 @@ namespace operators {
 
 template <typename Dtype, typename T>
 void ConvOp<Dtype, T>::InferShape() const {
-  //  std::cout << " begin get dims: " << std::endl;
-
   auto in_dims = param_.Input()->dims();
-
-  //  std::cout << " end get in dims: " << std::endl;
-
-  //  std::cout << " in_dims: " << in_dims << std::endl;
-
-  //  std::cout << " begin get Filter " << std::endl;
-
   auto filter_dims = param_.Filter()->dims();
-
-  //  std::cout << " end get Filter " << std::endl;
-
-  //  std::cout << " begin get Attrs " << std::endl;
-
   const std::vector<int> &strides = param_.Strides();
-
-  //  std::cout << " end get Attrs " << strides[0] << std::endl;
-
   std::vector<int> paddings = param_.Paddings();
-
   int groups = param_.Groups();
-
   std::vector<int> dilations = param_.Dilations();
 
   PADDLE_MOBILE_ENFORCE((in_dims.size() == filter_dims.size() &&
@@ -73,3 +55,5 @@ template class ConvOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(conv2d);
 REGISTER_OPERATOR(conv2d, ops::ConvOp);
+
+#endif

src/operators/conv_op.h

@@ -12,6 +12,8 @@ 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 CONV_OP
+
 #pragma once
 
 #include <string>
@@ -53,3 +55,5 @@ inline int ConvOutputSize(int input_size, int filter_size, int dilation,
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/depthwise_conv_op.cpp

@@ -12,9 +12,10 @@ 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 DEPTHWISECONV_OP
+
 #include "operators/depthwise_conv_op.h"
 #include <vector>
-#include "framework/data_type.h"
 #include "framework/op_proto_maker.h"
 #include "framework/op_registry.h"
 #include "operators/conv_op.h"
@@ -55,3 +56,5 @@ template class DepthwiseConvOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(depthwise_conv2d);
 REGISTER_OPERATOR(depthwise_conv2d, ops::DepthwiseConvOp);
+
+#endif

src/operators/depthwise_conv_op.h

@@ -12,6 +12,8 @@ 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 DEPTHWISECONV_OP
+
 #pragma once
 
 #include <string>
@@ -47,3 +49,5 @@ class DepthwiseConvOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/elementwise_add_op.cpp

@@ -12,6 +12,8 @@ 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 ELEMENTWISEADD_OP
+
 #include "elementwise_add_op.h"
 
 namespace paddle_mobile {
@@ -29,3 +31,5 @@ template class ElementwiseAddOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(elementwise_add);
 REGISTER_OPERATOR(elementwise_add, ops::ElementwiseAddOp);
+
+#endif

src/operators/elementwise_add_op.h

@@ -12,6 +12,8 @@ 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 ELEMENTWISEADD_OP
+
 #pragma once
 
 #include <string>
@@ -46,3 +48,5 @@ class ElementwiseAddOp : public framework::OperatorWithKernel<DeviceType> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/fusion_conv_add.cpp

@@ -12,6 +12,8 @@ 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 FUSIONCONVADD_OP
+
 #include "operators/fusion_conv_add.h"
 namespace paddle_mobile {
 namespace operators {
@@ -25,3 +27,5 @@ template class FushionConvAddOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(conv_add);
 REGISTER_OPERATOR(conv_add, ops::FushionConvAddOp);
+
+#endif

src/operators/fusion_conv_add.h

@@ -12,6 +12,8 @@ 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 FUSIONCONVADD_OP
+
 #pragma once
 
 #include <string>
@@ -66,3 +68,5 @@ class FushionConvAddOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/fusion_conv_add_relu_op.cpp

@@ -12,4 +12,8 @@ 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 CONVADDRELU_OP
+
 #include "fusion_conv_add_relu_op.h"
+
+#endif

src/operators/fusion_conv_add_relu_op.h

@@ -12,6 +12,8 @@ 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 CONVADDRELU_OP
+
 #pragma once
 
 #include "framework/operator.h"
@@ -49,3 +51,5 @@ class ConvAddReluOp {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/fusion_fc_op.cpp

@@ -12,6 +12,8 @@ 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 FUSION_FC_OP
+
 #include "operators/fusion_fc_op.h"
 namespace paddle_mobile {
 namespace operators {
@@ -54,3 +56,5 @@ template class FushionFcOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(fc);
 REGISTER_OPERATOR(fc, ops::FushionFcOp);
+
+#endif

src/operators/fusion_fc_op.h

@@ -12,6 +12,8 @@ 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 FUSION_FC_OP
+
 #pragma once
 
 #include <string>
@@ -71,3 +73,5 @@ class FushionFcOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/batchnorm_kernel.cpp

@@ -12,6 +12,8 @@ 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 BATCHNORM_OP
+
 #pragma once
 
 #include "operators/kernel/batchnorm_kernel.h"
@@ -91,3 +93,5 @@ void BatchNormKernel<CPU, float>::Compute(const BatchNormParam &param) const {
 }
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/box_coder_kernel.cpp

@@ -12,7 +12,7 @@ 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. */
 
-#pragma once
+#ifdef BOXCODER_OP
 
 #include "operators/kernel/box_coder_kernel.h"
 
@@ -135,3 +135,5 @@ void BoxCoderKernel<CPU, float>::Compute(const BoxCoderParam& param) const {
 }
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/concat_kernel.cpp

@@ -12,7 +12,7 @@ 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. */
 
-#pragma once
+#ifdef CONCAT_OP
 
 #include "operators/kernel/concat_kernel.h"
 
@@ -85,3 +85,5 @@ void ConcatKernel<CPU, float>::Compute(const ConcatParam &param) const {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/conv_kernel.cpp

@@ -12,6 +12,8 @@ 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 CONV_OP
+
 #include "operators/kernel/conv_kernel.h"
 
 namespace paddle_mobile {
@@ -112,3 +114,5 @@ template class ConvKernel<CPU, float>;
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/depthwise_conv_kernel.cpp

@@ -12,6 +12,8 @@ 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 DEPTHWISECONV_OP
+
 #include "operators/kernel/depthwise_conv_kernel.h"
 #include "operators/kernel/conv_kernel.h"
 
@@ -124,3 +126,5 @@ template class DepthwiseConvKernel<CPU, float>;
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/elementwise_add_kernel.cpp

@@ -12,6 +12,8 @@ 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 ELEMENTWISEADD_OP
+
 #pragma once
 
 #include "operators/kernel/elementwise_add_kernel.h"
@@ -40,3 +42,5 @@ template class ElementwiseAddKernel<CPU, float>;
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/fushion_fc_kernel.cpp

@@ -12,6 +12,8 @@ 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 FUSION_FC_OP
+
 #pragma once
 
 #include "operators/kernel/fushion_fc_kernel.h"
@@ -65,3 +67,5 @@ void FushionFcKernel<CPU, float>::Compute(const FushionFcParam &param) const {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/lrn_kernel.cpp

@@ -12,6 +12,8 @@ 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 LRN_OP
+
 #pragma once
 
 #include "operators/kernel/lrn_kernel.h"
@@ -42,3 +44,5 @@ template class LrnKernel<CPU, float>;
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/mul_kernel.cpp

@@ -12,6 +12,8 @@ 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
+
 #pragma once
 
 #include "operators/kernel/mul_kernel.h"
@@ -48,3 +50,5 @@ template class MulKernel<CPU, float>;
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/multiclass_nms_kernel.cpp

@@ -12,6 +12,8 @@ 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 MULTICLASSNMS_OP
+
 #pragma once
 
 #include "operators/kernel/multiclass_nms_kernel.h"
@@ -273,3 +275,5 @@ void MultiClassNMSKernel<CPU, float>::Compute(
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/pool_kernel.cpp

@@ -12,6 +12,8 @@ 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 POOL_OP
+
 #include <operators/kernel/pool_kernel.h>
 #include "common/log.h"
 
@@ -73,3 +75,5 @@ void PoolKernel<CPU, float>::Compute(const PoolParam &param) const {
 }
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/prior_box_kernel.cpp

@@ -12,6 +12,8 @@ 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 PRIORBOX_OP
+
 #pragma once
 
 #include "operators/kernel/prior_box_kernel.h"
@@ -143,3 +145,5 @@ void PriorBoxKernel<CPU, float>::Compute(const PriorBoxParam &param) const {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/relu_kernel.cpp

@@ -12,6 +12,8 @@ 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 RELU_OP
+
 #pragma once
 
 #include "operators/kernel/relu_kernel.h"
@@ -45,3 +47,5 @@ void ReluKernel<CPU, float>::Compute(const ReluParam &param) const {
 }
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/reshape_kernel.cpp

@@ -12,6 +12,8 @@ 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 RESHAPE_OP
+
 #pragma once
 
 #include "operators/kernel/reshape_kernel.h"
@@ -49,3 +51,5 @@ void ReshapeKernel<CPU, float>::Compute(const ReshapeParam &param) const {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/sigmoid_kernel.cpp

@@ -12,6 +12,8 @@ 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 SIGMOID_OP
+
 #include "../sigmoid_kernel.h"
 #if __ARM_NEON
 #include "../../math/math_func_neon.h"
@@ -25,35 +27,21 @@ using framework::Tensor;
 
 void sigmoid(const Tensor *X, Tensor *Y) {
 #if __ARM_NEON
-  DLOG << "step1";
   const float *input = X->data<float>();
-  DLOG << "step11";
-
   float *output = Y->mutable_data<float>();
-  DLOG << "step2";
-
   const DDim &dDim = X->dims();
-  DLOG << "step3";
-
   int axis_index = 1;
   if (dDim.size() < 4) {
     axis_index = 0;
   }
-  DLOG << "step4";
-
   DDim outer_ddim =
       paddle_mobile::framework::slice_ddim(dDim, 0, axis_index + 1);
   DDim inner_ddim =
       paddle_mobile::framework::slice_ddim(dDim, axis_index + 1, dDim.size());
-  DLOG << "step5";
-
   int out_size = paddle_mobile::framework::product(outer_ddim);
   int inner_size = paddle_mobile::framework::product(inner_ddim);
-  DLOG << "step6";
 
 #pragma omp parallel for
-  DLOG << "outsize=" << out_size;
-  DLOG << "innersize=" << inner_size;
   for (int i = 0; i < out_size; ++i) {
     const float *input_outer_ptr = input + i * inner_size;
     float *output_outer_ptr = output + i * inner_size;
@@ -93,3 +81,5 @@ void SigmoidKernel<CPU, float>::Compute(const SigmoidParam &param) const {
 template class SigmoidKernel<CPU, float>;
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/softmax_kernel.cpp

@@ -12,6 +12,8 @@ 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 SOFTMAX_OP
+
 #include "../softmax_kernel.h"
 #include "../../math/softmax.h"
 namespace paddle_mobile {
@@ -29,3 +31,5 @@ void SoftmaxKernel<CPU, float>::Compute(const SoftmaxParam &param) const {
 template class SoftmaxKernel<CPU, float>;
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/arm/transpose_kernel.cpp

@@ -12,7 +12,7 @@ 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. */
 
-#pragma once
+#ifdef TRANSPOSE_OP
 
 #include "operators/kernel/transpose_kernel.h"
 
@@ -70,3 +70,5 @@ void TransposeKernel<CPU, float>::Compute(const TransposeParam& param) const {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/batchnorm_kernel.h

@@ -12,9 +12,12 @@ 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 BATCHNORM_OP
+
+#pragma once
+
 #include "framework/operator.h"
 #include "operators/op_param.h"
-#pragma once;
 
 namespace paddle_mobile {
 namespace operators {
@@ -30,3 +33,5 @@ class BatchNormKernel
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/box_coder_kernel.h

@@ -12,14 +12,16 @@ 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 BOXCODER_OP
+
+#pragma once
+
 #include <vector>
 
 #include "framework/operator.h"
 #include "operators/math/transform.h"
 #include "operators/op_param.h"
 
-#pragma once;
-
 namespace paddle_mobile {
 namespace operators {
 
@@ -31,3 +33,5 @@ class BoxCoderKernel
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/concat_kernel.h

@@ -12,6 +12,8 @@ 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 CONCAT_OP
+
 #pragma once
 #include "framework/operator.h"
 #include "operators/op_param.h"
@@ -29,3 +31,5 @@ class ConcatKernel : public framework::OpKernelBase<DeviceType, ConcatParam> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/conv_kernel.h

@@ -12,6 +12,10 @@ 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 CONV_OP
+
+#pragma once
+
 #include <vector>
 #include "framework/operator.h"
 #include "operators/math/im2col.h"
@@ -19,8 +23,6 @@ limitations under the License. */
 #include "operators/math/vol2col.h"
 #include "operators/op_param.h"
 
-#pragma once;
-
 namespace paddle_mobile {
 namespace operators {
 
@@ -49,3 +51,5 @@ inline bool IsExpand(const std::vector<int64_t> &filter_dim,
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/depthwise_conv_kernel.h

@@ -12,14 +12,16 @@ 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 DEPTHWISECONV_OP
+
+#pragma once
+
 #include "framework/operator.h"
 #include "operators/math/im2col.h"
 #include "operators/math/math_function.h"
 #include "operators/math/vol2col.h"
 #include "operators/op_param.h"
 
-#pragma once;
-
 namespace paddle_mobile {
 namespace operators {
 
@@ -32,3 +34,5 @@ class DepthwiseConvKernel : public OpKernelBase<DeviceType, ConvParam> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/elementwise_add_kernel.h

@@ -12,7 +12,9 @@ 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. */
 
-#pragma once;
+#ifdef ELEMENTWISEADD_OP
+
+#pragma once
 
 #include "framework/operator.h"
 #include "operators/math/elementwise_op_function.h"
@@ -31,3 +33,5 @@ class ElementwiseAddKernel
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/fpga/conv_kernel.cpp

@@ -12,6 +12,8 @@ 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 CONV_OP
+
 namespace paddle_mobile {
 namespace operators {
 
@@ -22,3 +24,5 @@ namespace operators {
 // template class ConvKernel<FPGA, float>;
 }
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/fushion_fc_kernel.h

@@ -12,11 +12,13 @@ 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 FUSION_FC_OP
+
+#pragma once
+
 #include "framework/operator.h"
-#include "operators/math/math_function.h"
 #include "operators/op_param.h"
-
-#pragma once;
+#include "operators/math/math_function.h"
 
 namespace paddle_mobile {
 namespace operators {
@@ -29,3 +31,5 @@ class FushionFcKernel
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/lrn_kernel.h

@@ -12,9 +12,12 @@ 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 LRN_OP
+
+#pragma once
+
 #include "framework/operator.h"
 #include "operators/op_param.h"
-#pragma once;
 
 namespace paddle_mobile {
 namespace operators {
@@ -70,3 +73,5 @@ class LrnKernel : public framework::OpKernelBase<DeviceType, LrnParam> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/mali/conv_kernel.cpp

@@ -12,6 +12,8 @@ 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 CONV_OP
+
 #include "operators/kernel/conv_kernel.h"
 
 namespace paddle_mobile {
@@ -23,3 +25,5 @@ void ConvKernel<GPU_MALI, float>::Compute(const ConvParam &param) const {}
 template class ConvKernel<GPU_MALI, float>;
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/mul_kernel.h

@@ -12,10 +12,13 @@ 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
+
+#pragma once
+
 #include "framework/operator.h"
-#include "operators/math/math_function.h"
 #include "operators/op_param.h"
-#pragma once;
+#include "operators/math/math_function.h"
 
 namespace paddle_mobile {
 namespace operators {
@@ -29,3 +32,5 @@ class MulKernel : public framework::OpKernelBase<DeviceType, MulParam> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/multiclass_nms_kernel.h

@@ -12,10 +12,13 @@ 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 MULTICLASSNMS_OP
+
+#pragma once
+
 #include "framework/operator.h"
-#include "operators/op_param.h"
 
-#pragma once;
+#include "operators/op_param.h"
 
 namespace paddle_mobile {
 namespace operators {
@@ -28,3 +31,5 @@ class MultiClassNMSKernel
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/pool_kernel.h

@@ -12,11 +12,13 @@ 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 POOL_OP
+
 #pragma once
 
 #include "framework/operator.h"
-#include "operators/math/pooling.h"
 #include "operators/op_param.h"
+#include "operators/math/pooling.h"
 
 namespace paddle_mobile {
 namespace operators {
@@ -29,3 +31,5 @@ class PoolKernel : public OpKernelBase<DeviceType, PoolParam> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/prior_box_kernel.h

@@ -12,13 +12,15 @@ 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 PRIORBOX_OP
+
+#pragma once
+
 #include <vector>
 
+#include "operators/op_param.h"
 #include "framework/operator.h"
 #include "operators/math/transform.h"
-#include "operators/op_param.h"
-
-#pragma once;
 
 namespace paddle_mobile {
 namespace operators {
@@ -55,3 +57,5 @@ class PriorBoxKernel
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/relu_kernel.h

@@ -12,10 +12,13 @@ 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 RELU_OP
+
+#pragma once
+
 #include "framework/operator.h"
-#include "operators/op_param.h"
 
-#pragma once;
+#include "operators/op_param.h"
 
 namespace paddle_mobile {
 namespace operators {
@@ -27,3 +30,5 @@ class ReluKernel : public framework::OpKernelBase<DeviceType, ReluParam> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/reshape_kernel.h

@@ -12,12 +12,14 @@ 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. */
 
-#include <vector>
+#ifdef RESHAPE_OP
+
+#pragma once
 
+#include <vector>
 #include "framework/operator.h"
-#include "operators/op_param.h"
 
-#pragma once;
+#include "operators/op_param.h"
 
 namespace paddle_mobile {
 namespace operators {
@@ -72,3 +74,5 @@ class ReshapeKernel : public framework::OpKernelBase<DeviceType, ReshapeParam> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/sigmoid_kernel.h

@@ -12,6 +12,8 @@ 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 SIGMOID_OP
+
 #pragma once
 
 #include "framework/operator.h"
@@ -27,3 +29,5 @@ class SigmoidKernel : public OpKernelBase<DeviceType, SigmoidParam> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/softmax_kernel.h

@@ -12,6 +12,8 @@ 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 SOFTMAX_OP
+
 #pragma once
 
 #include "framework/operator.h"
@@ -30,3 +32,5 @@ class SoftmaxKernel : public OpKernelBase<DeviceType, SoftmaxParam> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/kernel/transpose_kernel.h

@@ -12,13 +12,15 @@ 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 TRANSPOSE_OP
+
+#pragma once
+
 #include <vector>
 
 #include "framework/operator.h"
 #include "operators/op_param.h"
 
-#pragma once;
-
 namespace paddle_mobile {
 namespace operators {
 
@@ -30,3 +32,5 @@ class TransposeKernel
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/lrn_op.cpp

@@ -12,6 +12,8 @@ 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 LRN_OP
+
 #include "lrn_op.h"
 
 namespace paddle_mobile {
@@ -29,3 +31,5 @@ template class LrnOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(lrn);
 REGISTER_OPERATOR(lrn, ops::LrnOp);
+
+#endif

src/operators/lrn_op.h

@@ -11,6 +11,9 @@ 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 LRN_OP
+
 #pragma once
 
 #include <string>
@@ -45,3 +48,5 @@ class LrnOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/math/gemm.cpp

@@ -13,10 +13,14 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "operators/math/gemm.h"
+#ifndef X86
+#include <arm_neon.h>
+#endif
 
 namespace paddle_mobile {
 namespace operators {
 namespace math {
+float ab[MR * NR];
 // 将A矩阵分块复制到连续内存(ColMajor)
 void PackMatrixA(int m, int k, int paddingM, const float *A, int lda,
                  float *buffer) {
@@ -170,17 +174,197 @@ void InnerKernel(int m, int n, int k, float alpha, const float *A, int lda,
 }
 
 // 计算一个更小的 4 * 4 的 C 矩阵分块
+#if defined(IOS)
+void AddDot4x4(int k, float alpha, const float *a, int lda, const float *b,
+               int ldb, float beta, float *C, int ldc, int mc, int nc) {
+  // init C
+  float32x4_t cv0 = vdupq_n_f32(0.0);
+  float32x4_t cv1 = vdupq_n_f32(0.0);
+  float32x4_t cv2 = vdupq_n_f32(0.0);
+  float32x4_t cv3 = vdupq_n_f32(0.0);
+
+  float32x4_t av;
+  float32x4_t bv;
+
+  float32x2_t av01;
+  float32x2_t av23;
+
+  for (int p = 0; p < k; p += 1) {
+    av = vld1q_f32(a);
+    bv = vld1q_f32(b);
+
+    av01 = vget_low_f32(av);
+    cv0 = vmlaq_lane_f32(cv0, bv, av01, 0);
+    cv1 = vmlaq_lane_f32(cv1, bv, av01, 1);
+    av23 = vget_high_f32(av);
+    cv2 = vmlaq_lane_f32(cv2, bv, av23, 0);
+    cv3 = vmlaq_lane_f32(cv3, bv, av23, 1);
+
+    a += MR;
+    b += NR;
+  }
+  float32x4x4_t cv = {cv0, cv1, cv2, cv3};
+  int i, j;
+  for (i = 0; i < mc; ++i) {
+    for (j = 0; j < nc; ++j) {
+      if (beta == 0.0) {
+        C(i, j) = 0.0;
+      } else if (beta != 1.0) {
+        C(i, j) *= beta;
+      }
+      if (j == 0) {
+        C(i, j) += alpha * vgetq_lane_f32(cv.val[i], 0);
+      } else if (j == 1) {
+        C(i, j) += alpha * vgetq_lane_f32(cv.val[i], 1);
+      } else if (j == 2) {
+        C(i, j) += alpha * vgetq_lane_f32(cv.val[i], 2);
+      } else if (j == 3) {
+        C(i, j) += alpha * vgetq_lane_f32(cv.val[i], 3);
+      }
+    }
+  }
+}
+#elif defined(ARMV7)
+void AddDot4x4(int k, float alpha, const float *a, int lda, const float *b,
+               int ldb, float beta, float *C, int ldc, int mc, int nc) {
+  int kc1 = k / 2, kc2 = k % 2;
+  int bytes_ldc = 4 * ldc;
+  int flag_alpha = (alpha == 1.0) ? 1 : 2;
+  int flag_beta;
+  if (beta == 0.0) {
+    flag_beta = 0;
+  } else if (beta == 1.0) {
+    flag_beta = 1;
+  } else {
+    flag_beta = 2;
+  }
+  asm volatile(
+      "vmov.f32   q10,    #0.0        \n\t"
+      "vmov.f32   q11,    #0.0        \n\t"
+      "vmov.f32   q12,    #0.0        \n\t"
+      "vmov.f32   q13,    #0.0        \n\t"
+
+      "subs       %[kc1], %[kc1], #1  \n\t"
+      "blt        end_kc1_%=          \n\t"
+      "loop_kc1_%=:                   \n\t"
+      "vld1.32    {q0, q1}, [%[a]]!   \n\t"
+      "vld1.32    {q2, q3}, [%[b]]!   \n\t"
+      "vmla.f32   q10, q2, d0[0]      \n\t"
+      "vmla.f32   q11, q2, d0[1]      \n\t"
+      "vmla.f32   q12, q2, d1[0]      \n\t"
+      "vmla.f32   q13, q2, d1[1]      \n\t"
+      "vmla.f32   q10, q3, d2[0]      \n\t"
+      "vmla.f32   q11, q3, d2[1]      \n\t"
+      "vmla.f32   q12, q3, d3[0]      \n\t"
+      "vmla.f32   q13, q3, d3[1]      \n\t"
+      "subs       %[kc1], %[kc1], #1  \n\t"
+      "bge        loop_kc1_%=         \n\t"
+      "end_kc1_%=:                    \n\t"
+
+      "subs       %[kc2], %[kc2], #1  \n\t"
+      "blt        end_kc2_%=          \n\t"
+      "vld1.32    {q0}, [%[a]]!       \n\t"
+      "vld1.32    {q1}, [%[b]]!       \n\t"
+      "vmla.f32   q10, q1, d0[0]      \n\t"
+      "vmla.f32   q11, q1, d0[1]      \n\t"
+      "vmla.f32   q12, q1, d1[0]      \n\t"
+      "vmla.f32   q13, q1, d1[1]      \n\t"
+      "end_kc2_%=:                    \n\t"
+
+      "cmp        %[mc],      #4      \n\t"
+      "bne        temp_%=             \n\t"
+      "cmp        %[nc],      #4      \n\t"
+      "bne        temp_%=             \n\t"
+
+      "vmov.f32   d8[0],    %[alpha]  \n\t"
+      "vmov.f32   d8[1],    %[beta]   \n\t"
+
+      "cmp        %[flag_alpha],  #1  \n\t"
+      "bne        alpha_%=            \n\t"
+
+      "alpha_%=:                      \n\t"
+      "vmul.f32   q10, q10, d8[0]     \n\t"
+      "vmul.f32   q11, q11, d8[0]     \n\t"
+      "vmul.f32   q12, q12, d8[0]     \n\t"
+      "vmul.f32   q13, q13, d8[0]     \n\t"
+
+      "beta_%=:                       \n\t"
+      "cmp        %[flag_beta],   #0  \n\t"
+      "beq        memory_%=           \n\t"
+
+      "mov        r4,     %[C]        \n\t"
+      "mov        r6,     %[bytes_ldc]\n\t"
+      "vld1.32    {q0}, [r4], r6      \n\t"
+      "vld1.32    {q1}, [r4], r6      \n\t"
+      "vld1.32    {q2}, [r4], r6      \n\t"
+      "vld1.32    {q3}, [r4]          \n\t"
+      "cmp        %[flag_beta],   #1  \n\t"
+      "beq        beta_eq1_%=         \n\t"
+      "bne        beta_ne1_%=         \n\t"
+
+      "beta_eq1_%=:                   \n\t"
+      "vadd.f32   q10, q10, q0        \n\t"
+      "vadd.f32   q11, q11, q1        \n\t"
+      "vadd.f32   q12, q12, q2        \n\t"
+      "vadd.f32   q13, q13, q3        \n\t"
+      "b          memory_%=           \n\t"
+
+      "beta_ne1_%=:                   \n\t"
+      "vmla.f32   q10, q0, d8[1]      \n\t"
+      "vmla.f32   q11, q1, d8[1]      \n\t"
+      "vmla.f32   q12, q2, d8[1]      \n\t"
+      "vmla.f32   q13, q3, d8[1]      \n\t"
+
+      "memory_%=:                     \n\t"
+      "mov        r5,     %[C]        \n\t"
+      "mov        r6,     %[bytes_ldc]\n\t"
+      "vst1.32    {q10}, [r5], r6     \n\t"
+      "vst1.32    {q11}, [r5], r6     \n\t"
+      "vst1.32    {q12}, [r5], r6     \n\t"
+      "vst1.32    {q13}, [r5]         \n\t"
+      "b          end_%=              \n\t"
+
+      "temp_%=:                       \n\t"
+      "vst1.32    {q10, q11}, [%[ab]]!\n\t"
+      "vst1.32    {q12, q13}, [%[ab]] \n\t"
+      "end_%=:                        \n\t"
+      :
+      : [a] "r"(a), [b] "r"(b), [C] "r"(C), [ab] "r"(ab), [kc1] "r"(kc1),
+        [kc2] "r"(kc2), [mc] "r"(mc), [nc] "r"(nc), [alpha] "r"(alpha),
+        [beta] "r"(beta), [bytes_ldc] "r"(bytes_ldc),
+        [flag_alpha] "r"(flag_alpha), [flag_beta] "r"(flag_beta)
+      : "memory", "q0", "q1", "q2", "q3", "q4", "q10", "q11", "q12", "q13");
+
+  if (mc != MR || nc != NR) {
+    int i, j;
+    for (i = 0; i < mc; ++i) {
+      for (j = 0; j < nc; ++j) {
+        if (beta == 0.0) {
+          if (alpha != 1.0) {
+            C(i, j) = alpha * ab[i * MR + j];
+          } else {
+            C(i, j) = ab[i * MR + j];
+          }
+        } else {
+          if (beta != 1.0) {
+            C(i, j) *= beta;
+          }
+          if (alpha != 1.0) {
+            C(i, j) += alpha * ab[i * MR + j];
+          } else {
+            C(i, j) += ab[i * MR + j];
+          }
+        }
+      }
+    }
+  }
+}
+#else
 void AddDot4x4(int k, float alpha, const float *a, int lda, const float *b,
                int ldb, float beta, float *C, int ldc, int mc, int nc) {
   float c[16] = {0};
   float reg_a0, reg_a1, reg_a2, reg_a3, reg_b0, reg_b1, reg_b2, reg_b3;
 
-  // // init C
-  // float32x4_t cv0 = vdup_n_f32(0.0);
-  // float32x4_t cv1 = vdup_n_f32(0.0);
-  // float32x4_t cv2 = vdup_n_f32(0.0);
-  // float32x4_t cv3 = vdup_n_f32(0.0);
-
   for (int p = 0; p < k; p += 1) {
     reg_b0 = *b++;
     reg_b1 = *b++;
@@ -232,6 +416,7 @@ void AddDot4x4(int k, float alpha, const float *a, int lda, const float *b,
     }
   }
 }
+#endif
 
 // 32位 float 矩阵乘法
 void sgemm(int m, int n, int k, float alpha, const float *A, int lda,

src/operators/math/pool3x3.h

@@ -12,6 +12,8 @@ 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 POOL_OP
+
 #pragma once
 
 #if __ARM_NEON
@@ -25,3 +27,5 @@ static void Pool3x3Max() {
 static void Pool3x3Avg() {
   // todo impl with neon
 }
+
+#endif

src/operators/math/pool_2x2.h

@@ -12,6 +12,8 @@ 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 POOL_OP
+
 #pragma once
 
 #if __ARM_NEON
@@ -25,3 +27,5 @@ static void Pool2x2Max() {
 static void Pool2x2Avg() {
   // todo impl with neon
 }
+
+#endif

src/operators/math/pooling.cpp

@@ -12,6 +12,8 @@ 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 POOL_OP
+
 #include "pooling.h"
 #include <common/types.h>
 
@@ -91,3 +93,5 @@ template class PoolFunctor<CPU, math::MaxPool<float>, float>;
 }  // namespace math
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/math/pooling.h

@@ -12,6 +12,8 @@ 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 POOL_OP
+
 #pragma once
 
 #include "common/log.h"
@@ -64,3 +66,5 @@ class PoolFunctor {
 }
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/math/softmax.cpp

@@ -11,6 +11,9 @@ 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 SOFTMAX_OP
+
 #include "operators/math/softmax.h"
 #include "common/types.h"
 #if __ARM_NEON
@@ -153,3 +156,4 @@ template class SoftmaxFuntor<CPU, float>;
 }  // namespace math
 }  // namespace operators
 }  // namespace paddle_mobile
+#endif

src/operators/math/softmax.h

@@ -12,6 +12,7 @@ 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 SOFTMAX_OP
 #pragma once
 #include "framework/tensor.h"
 namespace paddle_mobile {
@@ -26,3 +27,4 @@ class SoftmaxFuntor {
 }  // namespace math
 }  // namespace operators
 }  // namespace paddle_mobile
+#endif

src/operators/mul_op.cpp

@@ -12,6 +12,8 @@ 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 "mul_op.h"
 
 namespace paddle_mobile {
@@ -55,3 +57,5 @@ template class MulOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(mul);
 REGISTER_OPERATOR(mul, ops::MulOp);
+
+#endif

src/operators/mul_op.h

@@ -11,6 +11,9 @@ 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
+
 #pragma once
 
 #include <string>
@@ -45,3 +48,5 @@ class MulOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/multiclass_nms_op.cpp

@@ -12,6 +12,8 @@ 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 MULTICLASSNMS_OP
+
 #include "operators/multiclass_nms_op.h"
 namespace paddle_mobile {
 namespace operators {
@@ -39,3 +41,5 @@ template class MultiClassNMSOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(multiclass_nms);
 REGISTER_OPERATOR(multiclass_nms, ops::MultiClassNMSOp);
+
+#endif

src/operators/multiclass_nms_op.h

@@ -12,6 +12,8 @@ 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 MULTICLASSNMS_OP
+
 #pragma once
 
 #include <string>
@@ -50,3 +52,5 @@ class MultiClassNMSOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/op_param.cpp

@@ -12,10 +12,12 @@ 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. */
 
-#include "op_param.h"
 
+#include "op_param.h"
 namespace paddle_mobile {
 namespace operators {
+
+#ifdef CONV_OP
 Print &operator<<(Print &printer, const ConvParam &conv_param) {
   printer << "parameter of conv: "
           << "\n";
@@ -36,5 +38,7 @@ Print &operator<<(Print &printer, const ConvParam &conv_param) {
   printer << "  output dims: " << conv_param.Output()->dims();
   return printer;
 }
+#endif
+
 }  // namespace operators
 }  // namespace paddle_mobile

src/operators/op_param.h

@@ -191,6 +191,7 @@ class OpParam {
   }
 };
 
+#ifdef CONV_OP
 class ConvParam : OpParam {
  public:
   ConvParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -230,7 +231,9 @@ class ConvParam : OpParam {
 };
 
 Print &operator<<(Print &printer, const ConvParam &conv_param);
+#endif
 
+#ifdef ELEMENTWISEADD_OP
 class ElementwiseAddParam : OpParam {
  public:
   ElementwiseAddParam(const VariableNameMap &inputs,
@@ -258,6 +261,9 @@ class ElementwiseAddParam : OpParam {
   int axis_;
 };
 
+#endif
+
+#ifdef MUL_OP
 class MulParam : OpParam {
  public:
   MulParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -287,7 +293,9 @@ class MulParam : OpParam {
   int x_num_col_dims_;
   int y_num_col_dims_;
 };
+#endif
 
+#ifdef CONCAT_OP
 class ConcatParam : public OpParam {
  public:
   ConcatParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -309,7 +317,9 @@ class ConcatParam : public OpParam {
   Tensor *out_;
   int axis_;
 };
+#endif
 
+#ifdef LRN_OP
 class LrnParam : public OpParam {
  public:
   LrnParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -351,6 +361,9 @@ class LrnParam : public OpParam {
   float k_;
   string data_format_;
 };
+#endif
+
+#ifdef BATCHNORM_OP
 class BatchNormParam : OpParam {
  public:
   BatchNormParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -399,6 +412,9 @@ class BatchNormParam : OpParam {
   bool is_test_;
   string data_format_;
 };
+#endif
+
+#ifdef POOL_OP
 class PoolParam : public OpParam {
  public:
   PoolParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -442,6 +458,9 @@ class PoolParam : public OpParam {
   bool gloabal_pooling_ = false;
 };
 
+#endif
+
+#ifdef PRIORBOX_OP
 class PriorBoxParam : public OpParam {
  public:
   PriorBoxParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -503,7 +522,9 @@ class PriorBoxParam : public OpParam {
   float step_h_;
   float offset_;
 };
+#endif
 
+#ifdef BOXCODER_OP
 class BoxCoderParam : public OpParam {
  public:
   BoxCoderParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -533,7 +554,9 @@ class BoxCoderParam : public OpParam {
   Tensor *output_box_;
   std::string code_type_;
 };
+#endif
 
+#ifdef SOFTMAX_OP
 class SoftmaxParam : public OpParam {
  public:
   SoftmaxParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -549,7 +572,9 @@ class SoftmaxParam : public OpParam {
   Tensor *input_x_;
   Tensor *out_;
 };
+#endif
 
+#ifdef SIGMOID_OP
 class SigmoidParam : public OpParam {
  public:
   SigmoidParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -565,6 +590,9 @@ class SigmoidParam : public OpParam {
   Tensor *input_x_;
   Tensor *out_;
 };
+#endif
+
+#ifdef MULTICLASSNMS_OP
 class MultiClassNMSParam : public OpParam {
  public:
   MultiClassNMSParam(const VariableNameMap &inputs,
@@ -610,6 +638,7 @@ class MultiClassNMSParam : public OpParam {
   float nms_eta_;
   float score_threshold_;
 };
+#endif
 
 class FeedParam : public OpParam {
  public:
@@ -646,6 +675,7 @@ class FetchParam : public OpParam {
   Tensor *out_;
 };
 
+#ifdef TRANSPOSE_OP
 class TransposeParam : public OpParam {
  public:
   TransposeParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -666,7 +696,9 @@ class TransposeParam : public OpParam {
   Tensor *out_;
   vector<int> axis_;
 };
+#endif
 
+#ifdef RESHAPE_OP
 class ReshapeParam : public OpParam {
  public:
   ReshapeParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -695,7 +727,9 @@ class ReshapeParam : public OpParam {
   vector<int> shape_;
   bool inplace_;
 };
+#endif
 
+#ifdef RELU_OP
 /*
  * @b op 层实例化好这个 param 传递给 kernel 层使用
  * */
@@ -715,7 +749,9 @@ class ReluParam : public OpParam {
   Tensor *input_x_;
   Tensor *out_;
 };
+#endif
 
+#ifdef FUSION_FC_OP
 class FushionFcParam : public OpParam {
  public:
   FushionFcParam(const VariableNameMap &inputs, const VariableNameMap &outputs,
@@ -751,6 +787,7 @@ class FushionFcParam : public OpParam {
   int y_num_col_dims_;
   int axis_;
 };
+#endif
 
 }  // namespace operators
 }  // namespace paddle_mobile

src/operators/pool_op.cpp

@@ -12,6 +12,8 @@ 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 POOL_OP
+
 #include "pool_op.h"
 
 namespace paddle_mobile {
@@ -57,3 +59,5 @@ template class PoolOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(pool2d);
 REGISTER_OPERATOR(pool2d, ops::PoolOp);
+
+#endif

src/operators/pool_op.h

@@ -12,6 +12,8 @@ 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 POOL_OP
+
 #pragma once
 
 #include <framework/operator.h>
@@ -47,3 +49,5 @@ class PoolOp : public OperatorWithKernel<DeviceType> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/prior_box_op.cpp

@@ -12,6 +12,8 @@ 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 PRIORBOX_OP
+
 #include "operators/prior_box_op.h"
 #include <vector>
 namespace paddle_mobile {
@@ -49,3 +51,5 @@ template class PriorBoxOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(prior_box);
 REGISTER_OPERATOR(prior_box, ops::PriorBoxOp);
+
+#endif

src/operators/prior_box_op.h

@@ -12,6 +12,8 @@ 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 PRIORBOX_OP
+
 #pragma once
 
 #include <string>
@@ -50,3 +52,5 @@ class PriorBoxOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/relu_op.cpp

@@ -12,6 +12,8 @@ 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 RELU_OP
+
 #include "operators/relu_op.h"
 namespace paddle_mobile {
 namespace operators {
@@ -33,3 +35,5 @@ template class ReluOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(relu);
 REGISTER_OPERATOR(relu, ops::ReluOp);
+
+#endif

src/operators/relu_op.h

@@ -12,6 +12,8 @@ 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 RELU_OP
+
 #pragma once
 
 #include <string>
@@ -59,3 +61,5 @@ class ReluOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/reshape_op.cpp

@@ -12,6 +12,8 @@ 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 RESHAPE_OP
+
 #include "operators/reshape_op.h"
 #include <vector>
 namespace paddle_mobile {
@@ -32,3 +34,5 @@ template class ReshapeOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(reshape);
 REGISTER_OPERATOR(reshape, ops::ReshapeOp);
+
+#endif

src/operators/reshape_op.h

@@ -12,6 +12,8 @@ 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 RESHAPE_OP
+
 #pragma once
 
 #include <string>
@@ -49,3 +51,5 @@ class ReshapeOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/sigmoid_op.cpp

@@ -12,6 +12,8 @@ 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 SIGMOID_OP
+
 #include "operators/sigmoid_op.h"
 
 namespace paddle_mobile {
@@ -27,3 +29,5 @@ template class SigmoidOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(sigmoid);
 REGISTER_OPERATOR(sigmoid, ops::SigmoidOp);
+
+#endif

src/operators/sigmoid_op.h

@@ -12,6 +12,8 @@ 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 SIGMOID_OP
+
 #pragma once
 
 #include <framework/operator.h>
@@ -47,3 +49,5 @@ class SigmoidOp : public framework::OperatorWithKernel<DeviceType> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/softmax_op.cpp

@@ -12,6 +12,8 @@ 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 SOFTMAX_OP
+
 #include "operators/softmax_op.h"
 
 namespace paddle_mobile {
@@ -27,3 +29,5 @@ template class SoftmaxOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(softmax);
 REGISTER_OPERATOR(softmax, ops::SoftmaxOp);
+
+#endif

src/operators/softmax_op.h

@@ -12,6 +12,8 @@ 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 SOFTMAX_OP
+
 #pragma once
 
 #include <framework/operator.h>
@@ -47,3 +49,5 @@ class SoftmaxOp : public framework::OperatorWithKernel<DeviceType> {
 };
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

src/operators/transpose_op.cpp

@@ -12,6 +12,8 @@ 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 TRANSPOSE_OP
+
 #include "operators/transpose_op.h"
 #include <common/enforce.h>
 #include <vector>
@@ -51,3 +53,5 @@ template class TransposeOp<CPU, float>;
 namespace ops = paddle_mobile::operators;
 USE_OP(transpose);
 REGISTER_OPERATOR(transpose, ops::TransposeOp);
+
+#endif

src/operators/transpose_op.h

@@ -12,6 +12,8 @@ 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 TRANSPOSE_OP
+
 #pragma once
 
 #include <string>
@@ -50,3 +52,5 @@ class TransposeOp : public framework::OperatorWithKernel<DeviceType> {
 
 }  // namespace operators
 }  // namespace paddle_mobile
+
+#endif

test/framework/test_load.cpp

@@ -19,7 +19,7 @@ int main() {
   paddle_mobile::Loader<paddle_mobile::CPU> loader;
   //  ../../../test/models/googlenet
   //  ../../../test/models/mobilenet
-  auto program = loader.Load(g_mobilenet_ssd, true);
+  auto program = loader.Load(g_resnet, true);
   program.originProgram->Description("program desc: ");
   return 0;
 }

scripts/push2android.sh → tools/scripts/push2android.sh

 #!/usr/bin/env sh
 
 push_fn () {
-MODELS_PATH="../test/models/*"
-IMAGE_PATH="../test/images/*"
-EXE_FILE="../test/build/*"
+MODELS_PATH="../../test/models/*"
+IMAGE_PATH="../../test/images/*"
+EXE_FILE="../../test/build/*"
 EXE_DIR="data/local/tmp/bin"
 MODELS_DIR="data/local/tmp/models"
 IMAGES_DIR="data/local/tmp/images"