这里我的程序还是g++和nvcc编译的,比其他的可能还复杂一点,大致步骤是:
(1)g++编译.cpp文件生成.o文件
(2)nvcc编译.cu文件生成.o文件
(3)所有.o文件生成动态链接库.so文件
步骤(1)Makefile为:
CPPFLAGS += -I./include
CPPFLAGS += -I/usr/local/cuda/include
CPPFLAGS += -I/home/lzc/Desktop/soft/cudnn-tensorRT/tensorRT/TensorRT-7.0.0.11/include
#opencv3
CPPFLAGS += -I/usr/local/include
CPPFLAGS += -I/usr/local/include/opencv
CPPFLAGS += -I/usr/local/include/opencv2
#opencv4
#CPPFLAGS += -I/usr/local/opencv4/include/opencv4
#CPPFLAGS += -I/usr/local/opencv4/include/opencv4/opencv2
CPP_SRC := common_para.cpp face_det.cpp FacePreprocess.cpp face_recogn.cpp
CPP_OBJS := $(CPP_SRC:.cpp=.o)
$(CPP_OBJS):$(CPP_SRC)
g++ -fPIC -std=c++11 -Wall -Wfatal-errors -Ofast -c $^ $(CPPFLAGS)
步骤(2)Makefile为:
CPPFLAGS += -I./include
CPPFLAGS += -I/usr/local/cuda/include
CPPFLAGS += -I/home/lzc/Desktop/soft/cudnn-tensorRT/tensorRT/TensorRT-7.0.0.11/include
CU_SRC := decode.cu
CU_OBJS := $(CU_SRC:.cu=.o)
$(CU_OBJS):$(CU_SRC)
nvcc -std=c++11 --compiler-options "-Wall -Wfatal-errors -Ofast -fPIC" -c $^ $(CPPFLAGS)
步骤(3)的脚本makeFinal.sh为:
g++ -shared -o libfacecpplib.so *.o -L/usr/local/cuda/lib64 -L/home/lzc/Desktop/soft/cudnn-tensorRT/tensorRT/TensorRT-7.0.0.11/lib -lcuda -lcudart -lcublas -lcurand -lcudnn -lnvinfer
rm ./*.o
这样,libfacecpplib.so就生成出来了。
现在再用.so去编译程序,在程序的CMakeList.txt中:
将原本依赖的cuda_add_library删掉,
增加target_link_libraries(faceLib -L/home/lzc/Desktop/detection/face/faceRecogn-lzc/faceLib facecpplib)
即可再次编译
编译完成后,运行可执行文件时,需要将.so拷贝致与可执行文件相同目录下。