Merge remote-tracking branch 'upstream/develop' into bert

change to lib

CMakeLists.txt

@@ -75,6 +75,7 @@ include(generic)
 include(flags)
 
 if (NOT CLIENT_ONLY)
+include(external/cudnn)
 include(paddlepaddle)
 include(external/opencv)
 endif()

cmake/external/cudnn.cmake

+if(NOT WITH_GPU)
+    return()
+endif()
+
+if(WIN32)
+    set(CUDNN_ROOT ${CUDA_TOOLKIT_ROOT_DIR})
+else(WIN32)
+    set(CUDNN_ROOT "/usr" CACHE PATH "CUDNN ROOT")
+endif(WIN32)
+
+find_path(CUDNN_INCLUDE_DIR cudnn.h
+    PATHS ${CUDNN_ROOT} ${CUDNN_ROOT}/include
+    $ENV{CUDNN_ROOT} $ENV{CUDNN_ROOT}/include ${CUDA_TOOLKIT_INCLUDE}
+    NO_DEFAULT_PATH
+)
+
+get_filename_component(__libpath_hist ${CUDA_CUDART_LIBRARY} PATH)
+
+set(TARGET_ARCH "x86_64")
+if(NOT ${CMAKE_SYSTEM_PROCESSOR})
+    set(TARGET_ARCH ${CMAKE_SYSTEM_PROCESSOR})
+endif()
+
+list(APPEND CUDNN_CHECK_LIBRARY_DIRS
+    ${CUDNN_ROOT}
+    ${CUDNN_ROOT}/lib64
+    ${CUDNN_ROOT}/lib
+    ${CUDNN_ROOT}/lib/${TARGET_ARCH}-linux-gnu
+    ${CUDNN_ROOT}/local/cuda-${CUDA_VERSION}/targets/${TARGET_ARCH}-linux/lib/
+    $ENV{CUDNN_ROOT}
+    $ENV{CUDNN_ROOT}/lib64
+    $ENV{CUDNN_ROOT}/lib
+    /usr/lib
+	${CUDA_TOOLKIT_ROOT_DIR}
+	${CUDA_TOOLKIT_ROOT_DIR}/lib/x64
+	)
+set(CUDNN_LIB_NAME "")
+if (LINUX)
+set(CUDNN_LIB_NAME "libcudnn.so")
+endif(LINUX)
+
+if(WIN32)
+# only support cudnn7
+set(CUDNN_LIB_NAME "cudnn.lib" "cudnn64_7.dll")
+endif(WIN32)
+
+if(APPLE)
+set(CUDNN_LIB_NAME "libcudnn.dylib" "libcudnn.so")
+endif(APPLE)
+
+find_library(CUDNN_LIBRARY NAMES ${CUDNN_LIB_NAME} # libcudnn_static.a
+    PATHS ${CUDNN_CHECK_LIBRARY_DIRS} ${CUDNN_INCLUDE_DIR} ${__libpath_hist}
+          NO_DEFAULT_PATH
+    DOC "Path to cuDNN library.")
+
+
+if(CUDNN_INCLUDE_DIR AND CUDNN_LIBRARY)
+    set(CUDNN_FOUND ON)
+else()
+    set(CUDNN_FOUND OFF)
+endif()
+
+if(CUDNN_FOUND)
+    file(READ ${CUDNN_INCLUDE_DIR}/cudnn.h CUDNN_VERSION_FILE_CONTENTS)
+
+    get_filename_component(CUDNN_LIB_PATH ${CUDNN_LIBRARY} DIRECTORY)
+
+    string(REGEX MATCH "define CUDNN_VERSION +([0-9]+)"
+        CUDNN_VERSION "${CUDNN_VERSION_FILE_CONTENTS}")
+    string(REGEX REPLACE "define CUDNN_VERSION +([0-9]+)" "\\1"
+        CUDNN_VERSION "${CUDNN_VERSION}")
+
+    if("${CUDNN_VERSION}" STREQUAL "2000")
+        message(STATUS "Current cuDNN version is v2. ")
+    else()
+        string(REGEX MATCH "define CUDNN_MAJOR +([0-9]+)" CUDNN_MAJOR_VERSION
+            "${CUDNN_VERSION_FILE_CONTENTS}")
+        string(REGEX REPLACE "define CUDNN_MAJOR +([0-9]+)" "\\1"
+            CUDNN_MAJOR_VERSION "${CUDNN_MAJOR_VERSION}")
+        string(REGEX MATCH "define CUDNN_MINOR +([0-9]+)" CUDNN_MINOR_VERSION
+            "${CUDNN_VERSION_FILE_CONTENTS}")
+        string(REGEX REPLACE "define CUDNN_MINOR +([0-9]+)" "\\1"
+            CUDNN_MINOR_VERSION "${CUDNN_MINOR_VERSION}")
+        string(REGEX MATCH "define CUDNN_PATCHLEVEL +([0-9]+)"
+            CUDNN_PATCHLEVEL_VERSION "${CUDNN_VERSION_FILE_CONTENTS}")
+        string(REGEX REPLACE "define CUDNN_PATCHLEVEL +([0-9]+)" "\\1"
+            CUDNN_PATCHLEVEL_VERSION "${CUDNN_PATCHLEVEL_VERSION}")
+
+        if(NOT CUDNN_MAJOR_VERSION)
+            set(CUDNN_VERSION "???")
+        else()
+            add_definitions("-DPADDLE_CUDNN_BINVER=\"${CUDNN_MAJOR_VERSION}\"")
+            math(EXPR CUDNN_VERSION
+                "${CUDNN_MAJOR_VERSION} * 1000 +
+                 ${CUDNN_MINOR_VERSION} * 100 + ${CUDNN_PATCHLEVEL_VERSION}")
+        endif()
+
+        message(STATUS "Current cuDNN header is ${CUDNN_INCLUDE_DIR}/cudnn.h. "
+            "Current cuDNN version is v${CUDNN_MAJOR_VERSION}.${CUDNN_MINOR_VERSION}. ")
+
+    endif()
+endif()

cmake/paddlepaddle.cmake

@@ -15,71 +15,70 @@
 INCLUDE(ExternalProject)
 
 SET(PADDLE_SOURCES_DIR ${THIRD_PARTY_PATH}/Paddle)
+SET(PADDLE_DOWNLOAD_DIR ${PADDLE_SOURCES_DIR}/src/extern_paddle)
 SET(PADDLE_INSTALL_DIR ${THIRD_PARTY_PATH}/install/Paddle/)
 SET(PADDLE_INCLUDE_DIR "${PADDLE_INSTALL_DIR}/include" CACHE PATH "PaddlePaddle include directory." FORCE)
 SET(PADDLE_LIBRARIES "${PADDLE_INSTALL_DIR}/lib/libpaddle_fluid.a" CACHE FILEPATH "Paddle library." FORCE)
 
-INCLUDE_DIRECTORIES(${CMAKE_BINARY_DIR}/Paddle/fluid_install_dir)
 
 # Reference https://stackoverflow.com/questions/45414507/pass-a-list-of-prefix-paths-to-externalproject-add-in-cmake-args
 set(prefix_path "${THIRD_PARTY_PATH}/install/gflags|${THIRD_PARTY_PATH}/install/leveldb|${THIRD_PARTY_PATH}/install/snappy|${THIRD_PARTY_PATH}/install/gtest|${THIRD_PARTY_PATH}/install/protobuf|${THIRD_PARTY_PATH}/install/zlib|${THIRD_PARTY_PATH}/install/glog")
 
 message( "WITH_GPU = ${WITH_GPU}")
 
-# If minimal .a is need, you can set  WITH_DEBUG_SYMBOLS=OFF
+
+# Paddle Version should be one of:
+# latest: latest develop build
+# version number like 1.5.2
+SET(PADDLE_VERSION "latest")
+
+if (WITH_GPU)
+    SET(PADDLE_LIB_VERSION "${PADDLE_VERSION}-gpu-cuda${CUDA_VERSION_MAJOR}-cudnn7-avx-mkl")
+else()
+    if (AVX_FOUND)
+        if (WITH_MKLML)
+            SET(PADDLE_LIB_VERSION "${PADDLE_VERSION}-cpu-avx-mkl")
+        else()
+            SET(PADDLE_LIB_VERSION "${PADDLE_VERSION}-cpu-avx-openblas")
+        endif()
+    else()
+        SET(PADDLE_LIB_VERSION "${PADDLE_VERSION}-cpu-noavx-openblas")
+    endif()
+endif()
+
+SET(PADDLE_LIB_PATH "http://paddle-inference-lib.bj.bcebos.com/${PADDLE_LIB_VERSION}/fluid_inference.tgz")
+MESSAGE(STATUS "PADDLE_LIB_PATH=${PADDLE_LIB_PATH}")
+
 ExternalProject_Add(
-    extern_paddle
+    "extern_paddle"
     ${EXTERNAL_PROJECT_LOG_ARGS}
-    # TODO(wangguibao): change to de newst repo when they changed.
-    GIT_REPOSITORY  "https://github.com/PaddlePaddle/Paddle"
-    GIT_TAG         "v1.5.1"
-    PREFIX          ${PADDLE_SOURCES_DIR}
-    UPDATE_COMMAND  ""
-    BINARY_DIR ${CMAKE_BINARY_DIR}/Paddle
-    CMAKE_ARGS      -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
-                    -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
-                    -DCMAKE_INSTALL_PREFIX=${PADDLE_INSTALL_DIR}
-                    -DCMAKE_INSTALL_LIBDIR=${PADDLE_INSTALL_DIR}/lib
-                    -DCMAKE_POSITION_INDEPENDENT_CODE=ON
-                    -DCMAKE_BUILD_TYPE=${THIRD_PARTY_BUILD_TYPE}
-                    -DCMAKE_PREFIX_PATH=${prefix_path}
-                    -DCMAKE_BINARY_DIR=${CMAKE_CURRENT_BINARY_DIR}
-                    -DWITH_SWIG_PY=OFF
-                    -DWITH_PYTHON=OFF
-                    -DWITH_MKL=${WITH_MKL}
-                    -DWITH_AVX=${WITH_AVX}
-                    -DWITH_MKLDNN=OFF
-                    -DWITH_GPU=${WITH_GPU}
-                    -DWITH_FLUID_ONLY=ON
-                    -DWITH_TESTING=OFF
-                    -DWITH_DISTRIBUTE=OFF
-                    -DON_INFER=ON
-                    ${EXTERNAL_OPTIONAL_ARGS}
-    LIST_SEPARATOR |
-    CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${PADDLE_INSTALL_DIR}
-                     -DCMAKE_INSTALL_LIBDIR:PATH=${PADDLE_INSTALL_DIR}/lib
-                     -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON
-                     -DCMAKE_BUILD_TYPE:STRING=${THIRD_PARTY_BUILD_TYPE}
-    BUILD_COMMAND $(MAKE)
-    INSTALL_COMMAND $(MAKE) fluid_lib_dist
+    URL                 "${PADDLE_LIB_PATH}"
+    PREFIX              "${PADDLE_SOURCES_DIR}"
+    DOWNLOAD_DIR        "${PADDLE_DOWNLOAD_DIR}"
+    CONFIGURE_COMMAND   ""
+    BUILD_COMMAND       ""
+    UPDATE_COMMAND      ""
+    INSTALL_COMMAND
+        ${CMAKE_COMMAND} -E copy_directory ${PADDLE_DOWNLOAD_DIR}/paddle/include ${PADDLE_INSTALL_DIR}/include &&
+        ${CMAKE_COMMAND} -E copy_directory ${PADDLE_DOWNLOAD_DIR}/paddle/lib ${PADDLE_INSTALL_DIR}/lib &&
+        ${CMAKE_COMMAND} -E copy_directory ${PADDLE_DOWNLOAD_DIR}/third_party ${PADDLE_INSTALL_DIR}/third_party &&
+        ${CMAKE_COMMAND} -E copy ${PADDLE_INSTALL_DIR}/third_party/install/mkldnn/lib/libmkldnn.so.0 ${PADDLE_INSTALL_DIR}/third_party/install/mkldnn/lib/libmkldnn.so
 )
 
-ExternalProject_Get_Property(extern_paddle BINARY_DIR)
-SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${BINARY_DIR}/fluid_install_dir/third_party/install/mklml/lib")
-LINK_DIRECTORIES(${BINARY_DIR}/fluid_install_dir/third_party/install/mklml/lib)
-
-ADD_LIBRARY(paddle_fluid STATIC IMPORTED GLOBAL)
-SET_PROPERTY(TARGET paddle_fluid PROPERTY IMPORTED_LOCATION ${BINARY_DIR}/fluid_install_dir/paddle/fluid/inference/libpaddle_fluid.a)
+INCLUDE_DIRECTORIES(${PADDLE_INCLUDE_DIR})
+SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${PADDLE_INSTALL_DIR}/third_party/install/mklml/lib")
+LINK_DIRECTORIES(${PADDLE_INSTALL_DIR}/third_party/install/mklml/lib)
 
-LIST(APPEND external_project_dependencies paddle)
+SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_RPATH}" "${PADDLE_INSTALL_DIR}/third_party/install/mkldnn/lib")
+LINK_DIRECTORIES(${PADDLE_INSTALL_DIR}/third_party/install/mkldnn/lib)
 
-ADD_LIBRARY(snappystream STATIC IMPORTED GLOBAL)
-SET_PROPERTY(TARGET snappystream PROPERTY IMPORTED_LOCATION ${BINARY_DIR}/fluid_install_dir/third_party/install/snappystream/lib/libsnappystream.a)
+ADD_LIBRARY(paddle_fluid STATIC IMPORTED GLOBAL)
+SET_PROPERTY(TARGET paddle_fluid PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/lib/libpaddle_fluid.a)
 
 ADD_LIBRARY(xxhash STATIC IMPORTED GLOBAL)
-SET_PROPERTY(TARGET xxhash PROPERTY IMPORTED_LOCATION ${BINARY_DIR}/fluid_install_dir/third_party/install/xxhash/lib/libxxhash.a)
+SET_PROPERTY(TARGET xxhash PROPERTY IMPORTED_LOCATION ${PADDLE_INSTALL_DIR}/third_party/install/xxhash/lib/libxxhash.a)
+
+LIST(APPEND external_project_dependencies paddle)
 
 LIST(APPEND paddle_depend_libs
-        snappystream
-        snappy
         xxhash)

demo-serving/CMakeLists.txt

@@ -59,7 +59,7 @@ target_link_libraries(serving kvdb rocksdb)
 if(WITH_GPU)
     target_link_libraries(serving ${CUDA_LIBRARIES})
 endif()
-target_link_libraries(serving -liomp5 -lmklml_intel -lpthread
+target_link_libraries(serving -liomp5 -lmklml_intel -lmkldnn -lpthread
         -lcrypto -lm -lrt -lssl -ldl -lz -lbz2)
 
 install(TARGETS serving
@@ -75,7 +75,10 @@ install(FILES ${inc}
         DESTINATION ${PADDLE_SERVING_INSTALL_DIR}/include/serving)
 
 if (${WITH_MKL})
-    install(FILES ${CMAKE_BINARY_DIR}/Paddle/third_party/install/mklml/lib/libmklml_intel.so
-                ${CMAKE_BINARY_DIR}/Paddle/third_party/install/mklml/lib/libiomp5.so DESTINATION
+    install(FILES
+            ${CMAKE_BINARY_DIR}/third_party/install/Paddle/third_party/install/mklml/lib/libmklml_intel.so
+                ${CMAKE_BINARY_DIR}/third_party/install/Paddle/third_party/install/mklml/lib/libiomp5.so
+                ${CMAKE_BINARY_DIR}/third_party/install/Paddle/third_party/install/mkldnn/lib/libmkldnn.so
+                DESTINATION
                 ${PADDLE_SERVING_INSTALL_DIR}/demo/serving/bin)
 endif()

demo-serving/op/classify_op.h

@@ -21,7 +21,7 @@
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
 #endif
 #else
-#include "paddle/fluid/inference/paddle_inference_api.h"
+#include "paddle_inference_api.h"  // NOLINT
 #endif
 #include "demo-serving/image_class.pb.h"
 

demo-serving/op/ctr_prediction_op.h

@@ -21,7 +21,7 @@
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
 #endif
 #else
-#include "paddle/fluid/inference/paddle_inference_api.h"
+#include "paddle_inference_api.h"  // NOLINT
 #endif
 #include "demo-serving/ctr_prediction.pb.h"
 

demo-serving/op/reader_op.h

@@ -35,7 +35,7 @@
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
 #endif
 #else
-#include "paddle/fluid/inference/paddle_inference_api.h"
+#include "paddle_inference_api.h"  // NOLINT
 #endif
 
 namespace baidu {

demo-serving/op/text_classification_op.h

@@ -21,7 +21,7 @@
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
 #endif
 #else
-#include "paddle/fluid/inference/paddle_inference_api.h"
+#include "paddle_inference_api.h"  // NOLINT
 #endif
 #include "demo-serving/text_classification.pb.h"
 

inferencer-fluid-cpu/include/fluid_cpu_engine.h

@@ -28,7 +28,7 @@
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
 #endif
 #else
-#include "paddle/fluid/inference/paddle_inference_api.h"
+#include "paddle_inference_api.h"  // NOLINT
 #endif
 #include "predictor/framework/infer.h"
 
@@ -134,8 +134,7 @@ class FluidCpuAnalysisCore : public FluidFamilyCore {
     analysis_config.SetCpuMathLibraryNumThreads(1);
 
     if (params.enable_memory_optimization()) {
-      analysis_config.EnableMemoryOptim(params.static_optimization(),
-                                        params.force_update_static_cache());
+      analysis_config.EnableMemoryOptim();
     }
 
     analysis_config.SwitchSpecifyInputNames(true);
@@ -200,8 +199,7 @@ class FluidCpuAnalysisDirCore : public FluidFamilyCore {
     analysis_config.SetCpuMathLibraryNumThreads(1);
 
     if (params.enable_memory_optimization()) {
-      analysis_config.EnableMemoryOptim(params.static_optimization(),
-                                        params.force_update_static_cache());
+      analysis_config.EnableMemoryOptim();
     }
 
     AutoLock lock(GlobalPaddleCreateMutex::instance());
@@ -519,8 +517,7 @@ class FluidCpuAnalysisDirWithSigmoidCore : public FluidCpuWithSigmoidCore {
     analysis_config.SetCpuMathLibraryNumThreads(1);
 
     if (params.enable_memory_optimization()) {
-      analysis_config.EnableMemoryOptim(params.static_optimization(),
-                                        params.force_update_static_cache());
+      analysis_config.EnableMemoryOptim();
     }
 
     AutoLock lock(GlobalPaddleCreateMutex::instance());

inferencer-fluid-gpu/include/fluid_gpu_engine.h

@@ -28,7 +28,7 @@
 #include "paddle/fluid/inference/api/paddle_inference_api.h"
 #endif
 #else
-#include "paddle/fluid/inference/paddle_inference_api.h"
+#include "paddle_inference_api.h"  // NOLINT
 #endif
 #include "predictor/framework/infer.h"
 
@@ -136,8 +136,7 @@ class FluidGpuAnalysisCore : public FluidFamilyCore {
     analysis_config.SetCpuMathLibraryNumThreads(1);
 
     if (params.enable_memory_optimization()) {
-      analysis_config.EnableMemoryOptim(params.static_optimization(),
-                                        params.force_update_static_cache());
+      analysis_config.EnableMemoryOptim();
     }
 
     analysis_config.SwitchSpecifyInputNames(true);
@@ -203,8 +202,7 @@ class FluidGpuAnalysisDirCore : public FluidFamilyCore {
     analysis_config.SwitchIrOptim(true);
 
     if (params.enable_memory_optimization()) {
-      analysis_config.EnableMemoryOptim(params.static_optimization(),
-                                        params.force_update_static_cache());
+      analysis_config.EnableMemoryOptim();
     }
 
     AutoLock lock(GlobalPaddleCreateMutex::instance());
@@ -522,8 +520,7 @@ class FluidGpuAnalysisDirWithSigmoidCore : public FluidGpuWithSigmoidCore {
     analysis_config.SetCpuMathLibraryNumThreads(1);
 
     if (params.enable_memory_optimization()) {
-      analysis_config.EnableMemoryOptim(params.static_optimization(),
-                                        params.force_update_static_cache());
+      analysis_config.EnableMemoryOptim();
     }
 
     AutoLock lock(GlobalPaddleCreateMutex::instance());