Merge pull request #12049 from luotao1/demo

add independent inference demo on teamcity

paddle/contrib/inference/demo/CMakeLists.txt

@@ -13,8 +13,6 @@
 # limitations under the License.
 #
 
-inference_api_test(simple_on_word2vec ARGS test_word2vec)
-
 option(WITH_INFERENCE_DEMO "Compile with Inference demo" OFF)
 if(NOT WITH_INFERENCE_DEMO)
   return()

paddle/contrib/inference/demo_ci/CMakeLists.txt

+cmake_minimum_required(VERSION 3.0)
+
+project(cpp_inference_demo CXX C)
+
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+
+if(NOT DEFINED PADDLE_LIB)
+  message(FATAL_ERROR "please set PADDLE_LIB with -DPADDLE_LIB=/path/paddle/lib")
+endif()
+if(NOT DEFINED DEMO_NAME)
+  message(FATAL_ERROR "please set DEMO_NAME with -DDEMO_NAME=demo_name")
+endif()
+
+option(WITH_MKL        "Compile demo with MKL/OpenBlas support, default use MKL."       ON)
+option(WITH_GPU        "Compile demo with GPU/CPU, default use CPU."                    OFF)
+option(WITH_STATIC_LIB "Compile demo with static/shared library, default use static."   ON)
+
+if(WITH_GPU)
+  set(CUDA_LIB "/usr/local/cuda/lib64/" CACHE STRING "CUDA Library")
+endif()
+
+include_directories("${PADDLE_LIB}")
+include_directories("${PADDLE_LIB}/third_party/install/protobuf/include")
+include_directories("${PADDLE_LIB}/third_party/install/glog/include")
+include_directories("${PADDLE_LIB}/third_party/install/gflags/include")
+include_directories("${PADDLE_LIB}/third_party/install/snappy/include")
+include_directories("${PADDLE_LIB}/third_party/install/snappystream/include")
+include_directories("${PADDLE_LIB}/third_party/install/zlib/include")
+
+include_directories("${PADDLE_LIB}/third_party/boost")
+include_directories("${PADDLE_LIB}/third_party/eigen3")
+
+link_directories("${PADDLE_LIB}/third_party/install/snappy/lib")
+link_directories("${PADDLE_LIB}/third_party/install/snappystream/lib")
+link_directories("${PADDLE_LIB}/third_party/install/protobuf/lib")
+link_directories("${PADDLE_LIB}/third_party/install/glog/lib")
+link_directories("${PADDLE_LIB}/third_party/install/gflags/lib")
+link_directories("${PADDLE_LIB}/third_party/install/zlib/lib")
+
+add_executable(${DEMO_NAME} ${DEMO_NAME}.cc)
+
+if(WITH_MKL)
+  include_directories("${PADDLE_LIB}/third_party/install/mklml/include")
+  set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/libmklml_intel.so 
+               ${PADDLE_LIB}/third_party/install/mklml/lib/libiomp5.so)
+  set(MKLDNN_PATH "${PADDLE_LIB}/third_party/install/mkldnn")
+  if(EXISTS ${MKLDNN_PATH})
+    include_directories("${MKLDNN_PATH}/include")
+    set(MKLDNN_LIB ${MKLDNN_PATH}/lib/libmkldnn.so.0)
+  endif()
+else()
+  set(MATH_LIB ${PADDLE_LIB}/third_party/install/openblas/lib/libopenblas.a)
+endif()
+
+if(WITH_STATIC_LIB)
+  set(DEPS
+      "-Wl,--whole-archive"
+      ${PADDLE_LIB}/paddle/fluid/inference/libpaddle_fluid.a
+      "-Wl,--no-whole-archive"
+      ${PADDLE_LIB}/contrib/inference/libpaddle_inference_api.a)
+else()
+  # Note: libpaddle_inference_api.so must put before libpaddle_fluid.so
+  set(DEPS
+      ${PADDLE_LIB}/contrib/inference/libpaddle_inference_api.so
+      ${PADDLE_LIB}/paddle/fluid/inference/libpaddle_fluid.so)
+endif()
+set(EXTERNAL_LIB "-lrt -ldl -lpthread")
+
+set(DEPS ${DEPS}
+    ${MATH_LIB} ${MKLDNN_LIB}
+    glog gflags protobuf snappystream snappy z
+    ${EXTERNAL_LIB})
+if(WITH_GPU)
+  set(DEPS ${DEPS} ${CUDA_LIB}/libcudart.so)
+endif()
+
+target_link_libraries(${DEMO_NAME} ${DEPS})

paddle/contrib/inference/demo_ci/run.sh

+set -x
+PADDLE_ROOT=$1
+WITH_MKL=$2
+WITH_GPU=$3
+if [ $3 == "ON" ]; then
+  use_gpu_list='true false'
+else    
+  use_gpu_list='false'
+fi
+
+mkdir -p build
+cd build
+
+for WITH_STATIC_LIB in false; do
+  rm -rf *
+  cmake .. -DPADDLE_LIB=${PADDLE_ROOT}/build/fluid_install_dir/ \
+    -DWITH_MKL=$WITH_MKL \
+    -DDEMO_NAME=simple_on_word2vec \
+    -DWITH_GPU=$WITH_GPU \
+    -DWITH_STATIC_LIB=$WITH_STATIC_LIB
+  make
+  for use_gpu in $use_gpu_list; do
+    ./simple_on_word2vec \
+      --dirname=${PADDLE_ROOT}/build/python/paddle/fluid/tests/book/word2vec.inference.model \
+      --use_gpu=$use_gpu
+  done
+done
+if [ $? -eq 0 ]; then
+  exit 0
+else
+  echo "inference demo runs fail."
+  exit 1
+fi
+set +x

paddle/contrib/inference/demo/simple_on_word2vec.cc → paddle/contrib/inference/demo_ci/simple_on_word2vec.cc

@@ -16,21 +16,27 @@ limitations under the License. */
  * This file contains a simple demo for how to take a model for inference.
  */
 
+#include <gflags/gflags.h>
 #include <glog/logging.h>
-#include <gtest/gtest.h>
 #include <memory>
 #include <thread>
-#include "paddle/contrib/inference/paddle_inference_api.h"
+#include "contrib/inference/paddle_inference_api.h"
+#include "paddle/fluid/platform/enforce.h"
+
+DEFINE_string(dirname, "", "Directory of the inference model.");
+DEFINE_bool(use_gpu, false, "Whether use gpu.");
 
 namespace paddle {
 namespace demo {
 
-DEFINE_string(dirname, "", "Directory of the inference model.");
-
 void Main(bool use_gpu) {
   //# 1. Create PaddlePredictor with a config.
   NativeConfig config;
-  config.model_dir = FLAGS_dirname + "word2vec.inference.model";
+  if (FLAGS_dirname.empty()) {
+    LOG(INFO) << "Usage: ./simple_on_word2vec --dirname=path/to/your/model";
+    exit(1);
+  }
+  config.model_dir = FLAGS_dirname;
   config.use_gpu = use_gpu;
   config.fraction_of_gpu_memory = 0.15;
   config.device = 0;
@@ -54,12 +60,16 @@ void Main(bool use_gpu) {
     CHECK(predictor->Run(slots, &outputs));
 
     //# 4. Get output.
-    ASSERT_EQ(outputs.size(), 1UL);
-    LOG(INFO) << "output buffer size: " << outputs.front().data.length();
+    PADDLE_ENFORCE(outputs.size(), 1UL);
+    // Check the output buffer size and result of each tid.
+    PADDLE_ENFORCE(outputs.front().data.length(), 33168UL);
+    float result[5] = {
+        0.00129761, 0.00151112, 0.000423564, 0.00108815, 0.000932706};
     const size_t num_elements = outputs.front().data.length() / sizeof(float);
     // The outputs' buffers are in CPU memory.
     for (size_t i = 0; i < std::min(5UL, num_elements); i++) {
-      LOG(INFO) << static_cast<float*>(outputs.front().data.data())[i];
+      PADDLE_ENFORCE(static_cast<float*>(outputs.front().data.data())[i],
+                     result[i]);
     }
   }
 }
@@ -68,7 +78,7 @@ void MainThreads(int num_threads, bool use_gpu) {
   // Multi-threads only support on CPU
   // 0. Create PaddlePredictor with a config.
   NativeConfig config;
-  config.model_dir = FLAGS_dirname + "word2vec.inference.model";
+  config.model_dir = FLAGS_dirname;
   config.use_gpu = use_gpu;
   config.fraction_of_gpu_memory = 0.15;
   config.device = 0;
@@ -94,14 +104,17 @@ void MainThreads(int num_threads, bool use_gpu) {
         CHECK(predictor->Run(inputs, &outputs));
 
         // 4. Get output.
-        ASSERT_EQ(outputs.size(), 1UL);
-        LOG(INFO) << "TID: " << tid << ", "
-                  << "output buffer size: " << outputs.front().data.length();
+        PADDLE_ENFORCE(outputs.size(), 1UL);
+        // Check the output buffer size and result of each tid.
+        PADDLE_ENFORCE(outputs.front().data.length(), 33168UL);
+        float result[5] = {
+            0.00129761, 0.00151112, 0.000423564, 0.00108815, 0.000932706};
         const size_t num_elements =
             outputs.front().data.length() / sizeof(float);
         // The outputs' buffers are in CPU memory.
         for (size_t i = 0; i < std::min(5UL, num_elements); i++) {
-          LOG(INFO) << static_cast<float*>(outputs.front().data.data())[i];
+          PADDLE_ENFORCE(static_cast<float*>(outputs.front().data.data())[i],
+                         result[i]);
         }
       }
     });
@@ -111,15 +124,18 @@ void MainThreads(int num_threads, bool use_gpu) {
   }
 }
 
-TEST(demo, word2vec_cpu) { Main(false /*use_gpu*/); }
-TEST(demo_multi_threads, word2vec_cpu_1) { MainThreads(1, false /*use_gpu*/); }
-TEST(demo_multi_threads, word2vec_cpu_4) { MainThreads(4, false /*use_gpu*/); }
-
-#ifdef PADDLE_WITH_CUDA
-TEST(demo, word2vec_gpu) { Main(true /*use_gpu*/); }
-TEST(demo_multi_threads, word2vec_gpu_1) { MainThreads(1, true /*use_gpu*/); }
-TEST(demo_multi_threads, word2vec_gpu_4) { MainThreads(4, true /*use_gpu*/); }
-#endif
-
 }  // namespace demo
 }  // namespace paddle
+
+int main(int argc, char** argv) {
+  google::ParseCommandLineFlags(&argc, &argv, true);
+  paddle::demo::Main(false /* use_gpu*/);
+  paddle::demo::MainThreads(1, false /* use_gpu*/);
+  paddle::demo::MainThreads(4, false /* use_gpu*/);
+  if (FLAGS_use_gpu) {
+    paddle::demo::Main(true /*use_gpu*/);
+    paddle::demo::MainThreads(1, true /*use_gpu*/);
+    paddle::demo::MainThreads(4, true /*use_gpu*/);
+  }
+  return 0;
+}

paddle/scripts/paddle_build.sh

@@ -510,11 +510,23 @@ function gen_fluid_inference_lib() {
 EOF
         make -j `nproc` inference_lib_dist
         cd ${PADDLE_ROOT}/build
-        mv fluid_install_dir fluid
+        cp -r fluid_install_dir fluid
         tar -cf fluid.tgz fluid
       fi
 }
 
+function test_fluid_inference_lib() {
+    if [ ${WITH_C_API:-OFF} == "OFF" ] ; then
+        cat <<EOF
+    ========================================
+    Testing fluid inference library ...
+    ========================================
+EOF
+        cd ${PADDLE_ROOT}/paddle/contrib/inference/demo_ci
+        sh run.sh ${PADDLE_ROOT} ${WITH_MKL:-ON} ${WITH_GPU:-OFF}
+      fi
+}
+
 function main() {
     set -e
     local CMD=$1
@@ -568,6 +580,7 @@ function main() {
         run_test
         gen_capi_package
         gen_fluid_inference_lib
+        test_fluid_inference_lib
         ;;
       *)
         print_usage