Merge pull request #547 from codeWorm2015/develop

fix #546 fix docker complile error 

CMakeLists.txt

@@ -10,20 +10,56 @@ option(CPU "armv7 with neon" ON)
 option(MALI_GPU "mali gpu" OFF)
 option(FPGA "fpga" OFF)
 
-if (ARM_LINUX)
-include("${CMAKE_CURRENT_LIST_DIR}/tools/arm-platform.cmake")
-endif ()
-
 file(GLOB_RECURSE PADDLE_MOBILE_CC src/*.cc src/*.cpp src/*.c src/*.mm)
 file(GLOB_RECURSE PADDLE_MOBILE_H src/*.h)
+include_directories(src/)
+
+set(CMAKE_CXX_FLAGS "-std=c++14 -O3 -s ${CMAKE_CXX_FLAGS}")
+if (DEBUGING)
+    message(STATUS "debug")
+    set(CMAKE_BUILD_TYPE Debug)
+    set(CMAKE_CXX_FLAGS_DEBUG "-g -DNDEBUG")
+    add_definitions(-DPADDLE_MOBILE_DEBUG)
+else ()
+    set(CMAKE_BUILD_TYPE Release)
+    set(CMAKE_CXX_FLAGS_RELEASE "-DNDEBUG")
+    add_definitions(-fvisibility=hidden -fvisibility-inlines-hidden)
+endif ()
+
+if (USE_EXCEPTION)
+    message(STATUS "use exception")
+    add_definitions(-DENABLE_EXCEPTION)
+    add_definitions(-fexceptions)
+else()
+    add_definitions(-fno-exceptions)
+endif ()
+
+if (LOG_PROFILE)
+    add_definitions(-DPADDLE_MOBILE_PROFILE)
+endif()
+
+if(USE_OPENMP)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
+    add_definitions(-DPADDLE_MOBILE_USE_OPENMP)
+endif()
+
+# platform control
+if (ARM_LINUX)
+    include("${CMAKE_CURRENT_LIST_DIR}/tools/arm-platform.cmake")
+endif ()
 
 if (CPU)
   add_definitions(-DPADDLE_MOBILE_CPU)
 else()
-    list(REMOVE_ITEM PADDLE_MOBILE_CC ./src/operators/kernel/arm/*.h)
-    list(REMOVE_ITEM PADDLE_MOBILE_CC ./src/operators/kernel/arm/*.cc)
-    list(REMOVE_ITEM PADDLE_MOBILE_CC ./src/operators/kernel/arm/*.cpp)
+    file(GLOB_RECURSE _tmp_list src/operators/kernel/arm/*.cpp src/operators/kernel/arm/*.cc)
+    foreach(f ${_tmp_list})
+        list(REMOVE_ITEM PADDLE_MOBILE_CC ${f})
+    endforeach()
 
+    file(GLOB_RECURSE _tmp_list_h src/operators/kernel/arm/*.h)
+    foreach(f ${_tmp_list_h})
+        list(REMOVE_ITEM PADDLE_MOBILE_H ${f})
+    endforeach()
 endif()
 
 if (MALI_GPU)
@@ -54,55 +90,33 @@ endif()
 if(FPGA)
     add_definitions(-DPADDLE_MOBILE_FPGA)
 else()
-    list(REMOVE_ITEM PADDLE_MOBILE_CC ${CMAKE_CURRENT_SOURCE_DIR}/src/operators/kernel/fpga/*.h)
-    list(REMOVE_ITEM PADDLE_MOBILE_CC ${CMAKE_CURRENT_SOURCE_DIR}/src/operators/kernel/fpga/*.cc)
-    list(REMOVE_ITEM PADDLE_MOBILE_CC ${CMAKE_CURRENT_SOURCE_DIR}/src/operators/kernel/fpga/*.cpp)
+    file(GLOB_RECURSE _tmp_list src/operators/kernel/fpga/*.cpp src/operators/kernel/fpga/*.cc)
+    foreach(f ${_tmp_list})
+        list(REMOVE_ITEM PADDLE_MOBILE_CC ${f})
+    endforeach()
+
+    file(GLOB_RECURSE _tmp_list_h src/operators/kernel/fpga/*.h)
+    foreach(f ${_tmp_list_h})
+        list(REMOVE_ITEM PADDLE_MOBILE_H ${f})
+    endforeach()
 endif()
 
 
-set(CMAKE_CXX_FLAGS "-std=c++14 -O3 -s ${CMAKE_CXX_FLAGS}")
-if (DEBUGING)
-  message(STATUS "debug")
-  set(CMAKE_BUILD_TYPE Debug)
-  set(CMAKE_CXX_FLAGS_DEBUG "-g -DNDEBUG")
-  add_definitions(-DPADDLE_MOBILE_DEBUG)
-  if (ANDROID_NDK_TOOLCHAIN_INCLUDED)
-    add_definitions(-DARMV7)
+if (ANDROID_NDK_TOOLCHAIN_INCLUDED)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -llog")
-  endif ()
-else ()
-  set(CMAKE_BUILD_TYPE Release)
-  set(CMAKE_CXX_FLAGS_RELEASE "-DNDEBUG")
-  add_definitions(-fvisibility=hidden -fvisibility-inlines-hidden)
-endif ()
-
-if (USE_EXCEPTION)
-    message(STATUS "use exception")
-    add_definitions(-DENABLE_EXCEPTION)
-    add_definitions(-fexceptions)
+    add_definitions(-DARMV7)
 else()
-    add_definitions(-fno-exceptions)
-endif ()
-
-if (LOG_PROFILE)
-    add_definitions(-DPADDLE_MOBILE_PROFILE)
-endif()
-
-if(USE_OPENMP)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
-    add_definitions(-DPADDLE_MOBILE_USE_OPENMP)
-endif()
-
-
-
-
-if (NOT ANDROID_NDK_TOOLCHAIN_INCLUDED)
     list(REMOVE_ITEM PADDLE_MOBILE_H ${CMAKE_CURRENT_SOURCE_DIR}/src/jni/paddle_mobile_jni.h)
     list(REMOVE_ITEM PADDLE_MOBILE_CC ${CMAKE_CURRENT_SOURCE_DIR}/src/jni/paddle_mobile_jni.cpp)
     list(REMOVE_ITEM PADDLE_MOBILE_H ${CMAKE_CURRENT_SOURCE_DIR}/src/operators/math/math_func_neon.h)
 endif ()
 
-include_directories(src/)
+if (IS_IOS)
+else()
+    list(REMOVE_ITEM PADDLE_MOBILE_H ${CMAKE_CURRENT_SOURCE_DIR}/src/ios_io/PaddleMobile.h)
+    list(REMOVE_ITEM PADDLE_MOBILE_CC ${CMAKE_CURRENT_SOURCE_DIR}/src/ios_io/PaddleMobile.mm)
+    list(REMOVE_ITEM PADDLE_MOBILE_H ${CMAKE_CURRENT_SOURCE_DIR}/src/ios_io/op_symbols.h)
+endif ()
 
 set(CMAKE_VERBOSE_MAKEFILE ON)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
@@ -113,25 +127,20 @@ set(CMAKE_RUNTIME_OUTPUT_DIRECTORY build)
 # NET default
 set(NET "defult" CACHE STRING "select net type")
 set_property(CACHE NET PROPERTY STRINGS "defult" "googlenet" "mobilenet" "yolo" "squeezenet")
-
 include("${CMAKE_CURRENT_LIST_DIR}/tools/op.cmake")
 
-if (IS_IOS)
-    add_library(paddle-mobile STATIC ${PADDLE_MOBILE_CC} ${PADDLE_MOBILE_H})
-else()
-    list(REMOVE_ITEM PADDLE_MOBILE_H ${CMAKE_CURRENT_SOURCE_DIR}/src/ios_io/PaddleMobile.h)
-    list(REMOVE_ITEM PADDLE_MOBILE_CC ${CMAKE_CURRENT_SOURCE_DIR}/src/ios_io/PaddleMobile.mm)
-    list(REMOVE_ITEM PADDLE_MOBILE_H ${CMAKE_CURRENT_SOURCE_DIR}/src/ios_io/op_symbols.h)
-endif ()
 
+# build library
 if (ANDROID_NDK_TOOLCHAIN_INCLUDED)
     list(REMOVE_DUPLICATES CMAKE_CXX_FLAGS)
     add_library(paddle-mobile SHARED ${PADDLE_MOBILE_CC} ${PADDLE_MOBILE_H})
 elseif(IS_IOS)
+    add_library(paddle-mobile STATIC ${PADDLE_MOBILE_CC} ${PADDLE_MOBILE_H})
 else ()
     add_library(paddle-mobile SHARED ${PADDLE_MOBILE_CC} ${PADDLE_MOBILE_H})
 endif ()
 
+# unit test
 if(DEBUGING)
     if(IS_IOS)
     else()

README.md

@@ -91,8 +91,8 @@ ONNX全称为“Open Neural Network Exchange”，即“开放的神经网络切
 
 ![](http://7xop3k.com1.z0.glb.clouddn.com/15311951836000.jpg)
 
-### 4. 部分测试模型下载
-[下载链接](https://mms-mis.cdn.bcebos.com/paddle-mobile/models.zip)
+### 4. 部分测试模型和测试图片下载
+[下载链接](http://mms-graph.bj.bcebos.com/paddle-mobile%2FmodelsAndImages.zip)
 
 ## 问题解决
 

src/common/variant.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. */
 
+#include <cstdlib>
+
 #include "common/enforce.h"
 #include "common/log.h"
 

src/framework/attribute.h

@@ -14,6 +14,7 @@ limitations under the License. */
 
 #pragma once
 
+#include <cstdlib>
 #include <string>
 #include <typeinfo>
 #include <unordered_map>

src/framework/data_layout.h

@@ -15,6 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include <cctype>
+#include <cstdlib>
 #include <string>
 
 namespace paddle_mobile {

src/framework/ddim.h

@@ -14,9 +14,11 @@ limitations under the License. */
 
 #pragma once
 
+#include <cstdlib>
 #include <initializer_list>
 #include <typeinfo>
 #include <vector>
+
 #include "common/enforce.h"
 #include "common/variant.h"
 #include "dim.h"

src/framework/dim.h

@@ -14,6 +14,7 @@ limitations under the License. */
 
 #pragma once
 
+#include <cstdlib>
 #include "common/enforce.h"
 namespace paddle_mobile {
 namespace framework {

src/framework/tensor.h

@@ -152,7 +152,7 @@ class Tensor {
     if (holder_ != nullptr) {
       holder_->set_type(type);
     }
-    PADDLE_MOBILE_ENFORCE(numel() >= 0, "the Tensor'snumel must >=0.")
+    PADDLE_MOBILE_ENFORCE(numel() >= 0, "the Tensor's numel must >=0.")
     int64_t size = numel() * SizeOfType(type);
     if (holder_ == nullptr || holder_->size() < size + offset_) {
       holder_.reset(new PlaceholderImpl(size, type));

src/operators/kernel/arm/sigmoid_kernel.cpp

@@ -16,7 +16,7 @@ limitations under the License. */
 
 #include "../sigmoid_kernel.h"
 #include "../central-arm-func/sigmoid_arm_func.h"
-#if __ARM_NEON
+#ifdef __ARM_NEON
 #include "../../math/math_func_neon.h"
 #endif
 #include <cmath>

src/operators/kernel/central-arm-func/sigmoid_arm_func.h

@@ -14,8 +14,10 @@ limitations under the License. */
 #ifdef SIGMOID_OP
 #pragma once
 
+#include <cmath>
+
 #include "operators/op_param.h"
-#if __ARM_NEON
+#ifdef __ARM_NEON
 #include <arm_neon.h>
 #include "operators/math/math_func_neon.h"
 #endif
@@ -24,7 +26,7 @@ namespace paddle_mobile {
 namespace operators {
 using framework::DDim;
 void sigmoid(const Tensor *X, Tensor *Y) {
-#if __ARM_NEON
+#ifdef __ARM_NEON
   const float *input = X->data<float>();
   float *output = Y->mutable_data<float>();
   const DDim &dDim = X->dims();

src/operators/kernel/conv_add_kernel.h

@@ -17,7 +17,7 @@ limitations under the License. */
 #pragma once
 
 #include <vector>
-#if __ARM_NEON
+#ifdef __ARM_NEON
 #include <arm_neon.h>
 #endif
 #include "common/common.h"

src/operators/math/conv_func.h

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #pragma once
 
-#if __ARM_NEON
+#ifdef __ARM_NEON
 #include <arm_neon.h>
 #endif
 
@@ -49,7 +49,7 @@ inline void expand_bias(Tensor &bias, int axis, const DDim &dDim) {
   auto new_ptr = bias.mutable_data<float>();
   int axis_size = dDim[axis];
 
-#if __ARM_NEON
+#ifdef __ARM_NEON
   for (int i = 0; i < outer_size; ++i) {
     int inner_num = inner_size >> 4;
     int remain = inner_size - (inner_num << 4);

src/operators/math/depthwise_conv_3x3.cpp

@@ -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. */
 #include "operators/math/depthwise_conv_3x3.h"
+#ifdef __ARM_NEON
 #include <arm_neon.h>
+#endif
 #include <vector>
 
 namespace paddle_mobile {
@@ -21,7 +23,7 @@ namespace math {
 void DepthwiseConv3x3(const Tensor *input, vector<int> strides,
                       vector<int> paddings, const Tensor *filter, Tensor *bias,
                       Tensor *output, bool if_bias) {
-#if __ARM_NEON
+#ifdef __ARM_NEON
   const int batch_size = input->dims()[0];
 
   const int input_height = input->dims()[2];
@@ -242,6 +244,7 @@ void DepthwiseConv3x3(const Tensor *input, vector<int> strides,
 
 void DepthwiseConv3x3s1p1(const Tensor *input, const Tensor *filter,
                           Tensor *output, Tensor *bias, bool if_bias) {
+#ifdef __ARM_NEON
   const float *input_data = input->data<float>();
   const float *filter_data = filter->data<float>();
   float *output_data = output->data<float>();
@@ -511,11 +514,13 @@ void DepthwiseConv3x3s1p1(const Tensor *input, const Tensor *filter,
       filter_data_tmp += 9;
     }
   }
+#endif
 }
 
 void DepthwiseConvAddBNRelu3x3s1p1(const Tensor *input, const Tensor *filter,
                                    Tensor *output, const Tensor *new_scale,
                                    const Tensor *new_bias, bool if_relu) {
+#ifdef __ARM_NEON
   const float *input_data = input->data<float>();
   const float *filter_data = filter->data<float>();
   float *output_data = output->data<float>();
@@ -813,11 +818,14 @@ void DepthwiseConvAddBNRelu3x3s1p1(const Tensor *input, const Tensor *filter,
       filter_data_tmp += 9;
     }
   }
+#endif
 }
 
 void DepthwiseConvAddBNRelu3x3s2p1(const Tensor *input, const Tensor *filter,
                                    Tensor *output, const Tensor *new_scale,
                                    const Tensor *new_bias, bool if_relu) {
+#ifdef __ARM_NEON
+
   const int batch_size = input->dims()[0];
 
   const int input_height = input->dims()[2];
@@ -1009,10 +1017,12 @@ void DepthwiseConvAddBNRelu3x3s2p1(const Tensor *input, const Tensor *filter,
     input_data += input_batch_stride;
     output_data += output_batch_stride;
   }
+#endif
 }
 
 void DepthwiseConv3x3s2p1v2(const Tensor *input, const Tensor *filter,
                             Tensor *output, Tensor bias, bool if_bias) {
+#ifdef __ARM_NEON
   const float *input_data = input->data<float>();
   const float *filter_data = filter->data<float>();
   float *output_data = output->data<float>();
@@ -1209,11 +1219,13 @@ void DepthwiseConv3x3s2p1v2(const Tensor *input, const Tensor *filter,
     input_data += inhxw * c;
     output_data += outhxw * c;
   }
+#endif
 }
 
 void DepthwiseConvAddBNRelu3x3s2p1v2(const Tensor *input, const Tensor *filter,
                                      Tensor *output, const Tensor *new_scale,
                                      const Tensor *new_bias, bool if_relu) {
+#ifdef __ARM_NEON
   const float *input_data = input->data<float>();
   const float *filter_data = filter->data<float>();
   float *output_data = output->data<float>();
@@ -1444,6 +1456,7 @@ void DepthwiseConvAddBNRelu3x3s2p1v2(const Tensor *input, const Tensor *filter,
     input_data += inhxw * c;
     output_data += outhxw * c;
   }
+#endif
 }
 
 }  // namespace math

src/operators/math/pool_2x2.cpp

@@ -21,7 +21,7 @@ namespace math {
 
 void Pool2x2Max(vector<int> strides, vector<int> paddings, const Tensor *input,
                 Tensor *output) {
-#if __ARM_NEON
+#ifdef __ARM_NEON
 
 #ifdef ARMV7
 
@@ -99,7 +99,7 @@ void Pool2x2Max(vector<int> strides, vector<int> paddings, const Tensor *input,
 
 void Pool2x2Avg(vector<int> strides, vector<int> paddings, const Tensor *input,
                 Tensor *output) {
-#if __ARM_NEON
+#ifdef __ARM_NEON
 
 #ifdef ARMV7
   const int batch_size = input->dims()[0];

src/operators/math/pool_2x2.h

@@ -17,7 +17,7 @@ limitations under the License. */
 #pragma once
 
 #include "framework/tensor.h"
-#if __ARM_NEON
+#ifdef __ARM_NEON
 #include <arm_neon.h>
 #endif  // __ARM_NEON
 namespace paddle_mobile {

src/operators/math/pool_3x3.cpp

@@ -13,13 +13,12 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #ifdef POOL_OP
-#define __ARM_NEON true
 #ifdef _OPENMP
 #include <omp.h>
 #endif
 #include "framework/tensor.h"
 #include "pool_3x3.h"
-#if __ARM_NEON
+#ifdef __ARM_NEON
 #include <arm_neon.h>
 #endif  // __ARM_NEON
 #include <climits>
@@ -31,7 +30,7 @@ using std::max;
 using std::min;
 using std::vector;
 void Pool3x3Avgs1p1(const Tensor *input, Tensor *output) {
-#if __ARM_NEON
+#ifdef __ARM_NEON
   const int batch_size = input->dims()[0];
 
   const int h_in = input->dims()[2];
@@ -281,7 +280,7 @@ void Pool3x3Avgs1p1(const Tensor *input, Tensor *output) {
 }
 
 void Pool3x3Maxs1p1(const Tensor *input, Tensor *output) {
-#if __ARM_NEON
+#ifdef __ARM_NEON
   const int batch_size = input->dims()[0];
 
   const int h_in = input->dims()[2];
@@ -524,7 +523,7 @@ void Pool3x3Maxs1p1(const Tensor *input, Tensor *output) {
 
 void Pool3x3Max(vector<int> strides, vector<int> paddings, const Tensor *input,
                 Tensor *output) {
-#if __ARM_NEON
+#ifdef __ARM_NEON
   const int batch_size = input->dims()[0];
 
   const int input_height = input->dims()[2];
@@ -583,7 +582,7 @@ void Pool3x3Max(vector<int> strides, vector<int> paddings, const Tensor *input,
             }
             output_seg[ph * output_width + pw] = max_value;
           } else {
-#if defined(ARMV7)
+#ifdef ARMV7
             asm volatile(
                 "vld1.32  {q1}, [%[pos1]]        \n\t"
                 "vld1.32  {q2}, [%[pos2]]        \n\t"
@@ -623,7 +622,7 @@ void Pool3x3Max(vector<int> strides, vector<int> paddings, const Tensor *input,
 
 void Pool3x3Avg(vector<int> strides, vector<int> paddings, const Tensor *input,
                 Tensor *output) {
-#if __ARM_NEON
+#ifdef __ARM_NEON
   const int batch_size = input->dims()[0];
 
   const int input_height = input->dims()[2];
@@ -677,7 +676,7 @@ void Pool3x3Avg(vector<int> strides, vector<int> paddings, const Tensor *input,
             }
             output_seg[ph * output_width + pw] = sum / 9.0;
           } else {
-#if defined(ARMV7)
+#ifdef ARMV7
 
             asm volatile(
                 "vld1.32  {q1}, [%[pos1]]        \n\t"

src/operators/math/pool_3x3.h

@@ -21,7 +21,7 @@ limitations under the License. */
 #include <algorithm>
 #include <vector>
 #include "framework/tensor.h"
-#if __ARM_NEON
+#ifdef __ARM_NEON
 #include <arm_neon.h>
 #endif  // __ARM_NEON
 

src/operators/math/softmax.cpp

@@ -16,7 +16,7 @@ limitations under the License. */
 
 #include "operators/math/softmax.h"
 #include "common/types.h"
-#if __ARM_NEON
+#ifdef __ARM_NEON
 #include <math.h>
 #include <algorithm>
 #include "operators/math/math_func_neon.h"
@@ -29,7 +29,7 @@ using framework::DDim;
 using framework::Tensor;
 template <typename T>
 class SoftmaxFuntor<CPU, T> {
-#if __ARM_NEON
+#ifdef __ARM_NEON
   void sum(float *input, float *sumptr, int inner_size, int outter_size) {
     float32x4_t acc = vdupq_n_f32(0);
     float sum_ = 0;
@@ -144,7 +144,7 @@ class SoftmaxFuntor<CPU, T> {
       framework::Tensor sub_X = X->Slice(i, i + 1);
       framework::Tensor sub_Y = Y->Slice(i, i + 1);
 
-#if __ARM_NEON
+#ifdef __ARM_NEON
       SoftmaxCacl(&sub_X, &sub_Y);
 #endif
     }