1. 在程序中调用gpu加速，有两种方法

1.1方法1 终端运行程序

程序要想调用gpu推理，依靠的是CUDA和cudnn做桥梁，所以程序运行的环境必须要能够索引到cuda和cudnn的路径。一般我们在安装CUDA和cudnn的时候，都会在.bashrc里面加上这么一组路径：

# for cuda
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
export PATH=$PATH:/usr/local/cuda/bin
export CUDA_HOME=$CUDA_HOME:/usr/local/cuda

这就帮助我们在终端中运行py程序时，就可以自动索引cuda路径，从而调用gpu。

1.2方法2 在pycharm中运行程序

pycharm大家安装方式各有不同，如果在终端中使用pycharm-2021.3/bin/pycharm.sh 打开，那么也会自动加载cuda路径，但是如果是用别的方式打开，可能就会无法加载。此时可以改为使用pycharm-2021.3/bin/pycharm.sh 打开，为了方便期间，可以创建一个pycharm.desktop文件，文件内容为（需要注意，Exec和Icon需要改成自己pycharm的对应路径）：

[Desktop Entry]
Name=pycharm
Comment=Code Editing. Redefined.
GenericName=Text Editor
Exec=/opt/pycharm-2021.3/bin/pycharm.sh
Icon=/opt/pycharm-2021.3/bin/pycharm.png
Type=Application
StartupNotify=false
StartupWMClass=Code
Categories=TextEditor;Development;IDE;
MimeType=text/plain;inode/directory;application/x-code-workspace;
Actions=new-empty-window;
Keywords=pycharm;

[Desktop Action new-empty-window]
Name=New Empty Window
Exec=/opt/pycharm-2021.3/bin/pycharm.sh --new-window %F
Icon=/opt/pycharm-2021.3/bin/pycharm.png

保存后，修改权限后复制到/usr/share/applications/目录下。

sudo chmod 777 pycharm.desktop
sudo cp pycharm.desktop /usr/share/applications/

然后打开开始菜单就会出现pycharm的图标，打开后可以添加到dock栏。

1.3 gpu测试程序

安装了tensorflow2.2.0：

import os
from tensorflow.python.client import device_lib
import tensorflow as tf
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "99"

if __name__ == "__main__":
    print(device_lib.list_local_devices())
    flag = tf.test.is_gpu_available()
    if flag:
        print('GPU is available~~')
    else:
        print("no GPU!!!")

运行后会显示GPU的信息，说明读到了GPU信息。

[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 5098004230979266591
, name: "/device:XLA_CPU:0"
device_type: "XLA_CPU"
memory_limit: 17179869184
locality {
}
incarnation: 3107575638754036874
physical_device_desc: "device: XLA_CPU device"
, name: "/device:XLA_GPU:0"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 14659935295890188920
physical_device_desc: "device: XLA_GPU device"
, name: "/device:GPU:0"
device_type: "GPU"
memory_limit: 5689053024
locality {
  bus_id: 1
  links {
  }
}
incarnation: 16846892936777864981
physical_device_desc: "device: 0, name: NVIDIA GeForce GTX 1660 Ti, pci bus id: 0000:01:00.0, compute capability: 7.5"
]
GPU is available~~

Process finished with exit code 0

2. 使用tensorflow调用gpu遇到过的问题

2.1 错误1

运行一个推理程序后，报错：

tensorflow.python.framework.errors_impl.UnknownError:  Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
	 [[node TrajectoryNet18/navigation_net/conv2d/Conv2D (defined at /study/Lidar_Based_Navigation/lidar_trajectory/modules/navigation_net.py:31) ]] [Op:__inference_predict_4030]

Errors may have originated from an input operation.
Input Source operations connected to node TrajectoryNet18/navigation_net/conv2d/Conv2D:
 maps (defined at /study/Lidar_Based_Navigation/lidar_trajectory/scripts/ros_test_polar.py:701)

Function call stack:
predict

报错原因：
显存吃满了，需要加一个语句：

import tensorflow as tf
gpus= tf.config.experimental.list_physical_devices('GPU') # tf2.1版本该函数不再是experimental
print(gpus) # 前面限定了只使用GPU1(索引是从0开始的)
tf.config.experimental.set_memory_growth(gpus[0], True) # 其实gpus本身就只有一个元素