Development Verification

Set the Paddle Inference product path.

export PADDLE_INFERENCE_DIR=$(pwd)/paddle_inference_install_dir 
export LD_LIBRARY_PATH=${PADDLE_INFERENCE_DIR}/paddle/lib:${CUDA_HOME}/lib64:${LD_LIBRARY_PATH:-}

Check the product and version information.

test -f ${PADDLE_INFERENCE_DIR}/paddle/include/paddle_inference_api.h 
test -f ${PADDLE_INFERENCE_DIR}/paddle/lib/libpaddle_inference.so 
find ${PADDLE_INFERENCE_DIR}/paddle/lib -maxdepth 1 -type f \
  -name 'libpaddle_inference*' | sort
cat ${PADDLE_INFERENCE_DIR}/version.txt

The expected output is as follows.

libpaddle_inference.a 
libpaddle_inference.so 
WITH_GPU: ON 
CUDA version: 13.0 
CUDNN version: v9.21

Compile the minimum C++ GPU configuration verification program.

cat > paddle_infer_min.cc <<'CPP'
#include <iostream>
#include "paddle_inference_api.h"

int main() {
  paddle_infer::Config config;
  config.EnableUseGpu(100, 0);
  std::cout << "paddle_inference_gpu_config=ok" << std::endl;
  return 0;
} 
CPP

g++ -std=c++17 paddle_infer_min.cc \
  -I${PADDLE_INFERENCE_DIR}/paddle/include \
  -L${PADDLE_INFERENCE_DIR}/paddle/lib \
  -lpaddle_inference \
  -Wl,-rpath,${PADDLE_INFERENCE_DIR}/paddle/lib \
  -o paddle_infer_min

./paddle_infer_min

The expected output is as follows.

paddle_inference_gpu_config=ok

The pass criteria are as follows.

The inference_lib_dist is successfully built.
The paddle_inference_install_dir and paddle_inference_c_install_dir are successfully generated.
The version.txt file displays WITH_GPU: ON, CUDA version: 13.0, and CUDNN version: v9.21.
The minimum C++ program is successfully compiled and the output is paddle_inference_gpu_config=ok.

Parent topic: Developer Guide