use_gpu_acceleration.md

Using GPU Acceleration

By migrating point cloud parsing tasks to the GPU, not only can CPU computing resources be freed up, but the speed and efficiency of point cloud data parsing can also be effectively improved to a certain extent.

Preparation

Using GPU for point cloud parsing requires proper installation of graphics drivers and CUDA environment.

1. Enter nvidia-smi command in terminal to see current GPU information. If not installed, refer to: NVIDIA-DRIVERS
2. Enter nvcc -V command in terminal to see current CUDA version. If not installed, refer to: CUDA-DOWNLOADS

1 SDK Configuration

Please refer to the operations in Compile Macro Control to configure the macro FIND_CUDA to take effect, and add the cuda parameter when running.

2 Parsing Configuration

Refer to How to Parse Lidar Data Online and How to Parse PCAP File Data Offline

Steps

1 Compilation

In the HesaiLidar_SDK_2.0 folder, open a terminal and execute the following commands:

cd HesaiLidar_SDK_2.0
mkdir -p build 
cd build
cmake .. -DFIND_CUDA=true
make

2 Run

After successful compilation, run the generated executable file in the build folder. To use GPU parsing, please add the cuda parameter:

./sample cuda

Additional References

1 If multiple CUDA versions are installed, you can specify the CUDA version path in CMakeLists.txt:

 # if install different cuda version, set the cuda path, like cuda-11.4
 # set(CUDA_TOOLKIT_ROOT_DIR /usr/local/cuda-11.4/)

2 Control parameter for enabling GPU parsing in code

Actually use use_gpu in DriverParam to control whether GPU parsing is enabled. In the main function of test.cc, control whether GPU parsing is enabled by assigning param.use_gpu