提交 f5690950 编写于 作者: N nhzlx

add tensorrt api lib to paddle_fluid

上级 a46e30aa
......@@ -19,9 +19,19 @@ cc_library(paddle_fluid_origin DEPS ${fluid_modules} paddle_fluid_api)
add_subdirectory(api)
set(STATIC_INFERENCE_APIS paddle_fluid_api paddle_inference_api analysis_predictor)
set(SHARED_INFERENCE_SRCS
io.cc ${CMAKE_CURRENT_SOURCE_DIR}/api/api.cc ${CMAKE_CURRENT_SOURCE_DIR}/api/api_impl.cc
${CMAKE_CURRENT_SOURCE_DIR}/api/analysis_predictor.cc
${CMAKE_CURRENT_SOURCE_DIR}/api/details/zero_copy_tensor.cc)
if (WITH_GPU AND TENSORRT_FOUND)
set(STATIC_INFERENCE_APIS ${STATIC_INFERENCE_APIS} paddle_inference_tensorrt_subgraph_engine)
set(SHARED_INFERENCE_SRCS ${SHARED_INFERENCE_SRCS} ${CMAKE_CURRENT_SOURCE_DIR}/api/api_tensorrt_subgraph_engine.cc)
endif()
# Create static library
cc_library(paddle_fluid DEPS ${fluid_modules} paddle_fluid_api paddle_inference_api
analysis_predictor zero_copy_tensor)
cc_library(paddle_fluid DEPS ${fluid_modules} ${STATIC_INFERENCE_APIS} zero_copy_tensor)
if(NOT APPLE)
# TODO(liuyiqu: Temporarily disable the link flag because it is not support on Mac.
set(LINK_FLAGS "-Wl,--retain-symbols-file ${CMAKE_CURRENT_SOURCE_DIR}/paddle_fluid.sym")
......@@ -29,10 +39,7 @@ if(NOT APPLE)
endif()
# Create shared library
cc_library(paddle_fluid_shared SHARED
SRCS io.cc ${CMAKE_CURRENT_SOURCE_DIR}/api/api.cc ${CMAKE_CURRENT_SOURCE_DIR}/api/api_impl.cc
${CMAKE_CURRENT_SOURCE_DIR}/api/analysis_predictor.cc
${CMAKE_CURRENT_SOURCE_DIR}/api/details/zero_copy_tensor.cc
cc_library(paddle_fluid_shared SHARED SRCS ${SHARED_INFERENCE_SRCS}
DEPS ${fluid_modules} paddle_fluid_api)
set_target_properties(paddle_fluid_shared PROPERTIES OUTPUT_NAME paddle_fluid)
......
......@@ -3,6 +3,7 @@ project(cpp_inference_demo CXX C)
option(WITH_MKL "Compile demo with MKL/OpenBlas support, default use MKL." ON)
option(WITH_GPU "Compile demo with GPU/CPU, default use CPU." OFF)
option(WITH_STATIC_LIB "Compile demo with static/shared library, default use static." ON)
option(USE_TENSORRT "Compile demo with TensorRT." OFF)
macro(safe_set_static_flag)
foreach(flag_var
......@@ -60,6 +61,13 @@ endif(NOT WIN32)
include_directories("${PADDLE_LIB}/third_party/boost")
include_directories("${PADDLE_LIB}/third_party/eigen3")
if (NOT WIN32)
if (USE_TENSORRT AND WITH_GPU)
include_directories("${TENSORRT_INCLUDE_DIR}")
link_directories("${TENSORRT_LIB_DIR}")
endif()
endif(NOT WIN32)
if (NOT WIN32)
link_directories("${PADDLE_LIB}/third_party/install/snappy/lib")
link_directories("${PADDLE_LIB}/third_party/install/snappystream/lib")
......@@ -112,6 +120,10 @@ endif(NOT WIN32)
if(WITH_GPU)
if(NOT WIN32)
if (USE_TENSORRT)
set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/libnvinfer${CMAKE_STATIC_LIBRARY_SUFFIX})
set(DEPS ${DEPS} ${TENSORRT_LIB_DIR}/libnvinfer_plugin${CMAKE_STATIC_LIBRARY_SUFFIX})
endif()
set(DEPS ${DEPS} ${CUDA_LIB}/libcudart${CMAKE_SHARED_LIBRARY_SUFFIX})
else()
set(DEPS ${DEPS} ${CUDA_LIB}/cudart${CMAKE_STATIC_LIBRARY_SUFFIX} )
......
......@@ -3,6 +3,9 @@ PADDLE_ROOT=$1
TURN_ON_MKL=$2 # use MKL or Openblas
TEST_GPU_CPU=$3 # test both GPU/CPU mode or only CPU mode
DATA_DIR=$4 # dataset
TENSORRT_INCLUDE_DIR=$5 # TensorRT header file dir, defalut to /usr/local/TensorRT/include
TENSORRT_LIB_DIR=$6 # TensorRT lib file dir, default to /usr/local/TensorRT/lib
cd `dirname $0`
current_dir=`pwd`
if [ $2 == ON ]; then
......@@ -16,6 +19,11 @@ else
use_gpu_list='false'
fi
USE_TENSORRT=OFF
if [ [-d"$TENSORRT_INCLUDE_DIR"] -a [-d"$TENSORRT_LIB_DIR"] ]; then
USE_TENSORRT=ON
fi
PREFIX=inference-vis-demos%2F
URL_ROOT=http://paddlemodels.cdn.bcebos.com/${PREFIX}
......@@ -86,5 +94,25 @@ for WITH_STATIC_LIB in ON OFF; do
fi
done
done
# --------tensorrt mobilenet------
if [ $USE_TENSORRT == ON -a $TEST_GPU_CPU == ON ]; then
rm -rf *
cmake .. -DPADDLE_LIB=${PADDLE_ROOT}/build/fluid_install_dir/ \
-DWITH_MKL=$TURN_ON_MKL \
-DDEMO_NAME=vis_demo \
-DWITH_GPU=$TEST_GPU_CPU \
-DWITH_STATIC_LIB=$WITH_STATIC_LIB \
-DUSE_TENSORRT=$USE_TENSORRT \
-DTENSORRT_INCLUDE_DIR=$TENSORRT_INCLUDE_DIR \
-DTENSORRT_LIB_DIR=$TENSORRT_LIB_DIR
make -j
./vis_demo \
--modeldir=$DATA_DIR/mobilenet/model \
--data=$DATA_DIR/mobilenet/data.txt \
--refer=$DATA_DIR/mobilenet/result.txt \
--use_gpu=true \
--use_trt=true
fi
done
set +x
......@@ -34,6 +34,7 @@ DEFINE_string(
"path of data; each line is a record, format is "
"'<space splitted floats as data>\t<space splitted ints as shape'");
DEFINE_bool(use_gpu, false, "Whether use gpu.");
DEFINE_bool(use_trt, false, "Whether use trt.");
namespace paddle {
namespace demo {
......@@ -100,20 +101,34 @@ void CheckOutput(const std::string& referfile, const PaddleTensor& output) {
/*
* Use the native fluid engine to inference the demo.
*/
void Main(bool use_gpu) {
NativeConfig config;
config.param_file = FLAGS_modeldir + "/__params__";
config.prog_file = FLAGS_modeldir + "/__model__";
config.use_gpu = use_gpu;
config.device = 0;
if (FLAGS_use_gpu) {
void Main(bool use_gpu, bool use_trt) {
std::unique_ptr<PaddlePredictor> predictor;
if (!use_trt) {
NativeConfig config;
config.param_file = FLAGS_modeldir + "/__params__";
config.prog_file = FLAGS_modeldir + "/__model__";
config.use_gpu = use_gpu;
config.device = 0;
if (FLAGS_use_gpu) {
config.fraction_of_gpu_memory = 0.1; // set by yourself
}
VLOG(3) << "init predictor";
predictor =
CreatePaddlePredictor<NativeConfig, PaddleEngineKind::kNative>(config);
} else {
paddle::contrib::MixedRTConfig config;
config.param_file = FLAGS_modeldir + "/__params__";
config.prog_file = FLAGS_modeldir + "/__model__";
config.use_gpu = true;
config.device = 0;
config.max_batch_size = 1;
config.fraction_of_gpu_memory = 0.1; // set by yourself
predictor =
CreatePaddlePredictor<paddle::contrib::MixedRTConfig,
PaddleEngineKind::kAutoMixedTensorRT>(config);
}
VLOG(3) << "init predictor";
auto predictor =
CreatePaddlePredictor<NativeConfig, PaddleEngineKind::kNative>(config);
VLOG(3) << "begin to process data";
// Just a single batch of data.
std::string line;
......@@ -131,7 +146,7 @@ void Main(bool use_gpu) {
VLOG(3) << "run executor";
std::vector<PaddleTensor> output;
predictor->Run({input}, &output);
predictor->Run({input}, &output, 1);
VLOG(3) << "output.size " << output.size();
auto& tensor = output.front();
......@@ -146,9 +161,12 @@ void Main(bool use_gpu) {
int main(int argc, char** argv) {
google::ParseCommandLineFlags(&argc, &argv, true);
paddle::demo::Main(false /* use_gpu*/);
if (FLAGS_use_gpu) {
paddle::demo::Main(true /*use_gpu*/);
if (FLAGS_use_gpu && FLAGS_use_trt) {
paddle::demo::Main(true /*use_gpu*/, true);
} else if (FLAGS_use_gpu) {
paddle::demo::Main(true /*use_gpu*/, false);
} else {
paddle::demo::Main(false /* use_gpu*/, false /*use_tensorrt*/);
}
return 0;
}
......@@ -683,7 +683,7 @@ function test_fluid_inference_lib() {
========================================
EOF
cd ${PADDLE_ROOT}/paddle/fluid/inference/api/demo_ci
./run.sh ${PADDLE_ROOT} ${WITH_MKL:-ON} ${WITH_GPU:-OFF} ${INFERENCE_DEMO_INSTALL_DIR}
./run.sh ${PADDLE_ROOT} ${WITH_MKL:-ON} ${WITH_GPU:-OFF} ${INFERENCE_DEMO_INSTALL_DIR} ${TENSORRT_INCLUDE_DIR:-/usr/local/TensorRT/include} ${TENSORRT_LIB_DIR:-/usr/local/TensorRT/lib}
./clean.sh
fi
}
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册