mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-17 06:00:59 +08:00

Files

seyosum df8dd3e3ac 【Hackthon_4th 180】Support HORIZON BPU Backend for FastDeploy (#1822 )

* add horizon backend and PPYOLOE examples

* 更改horizon头文件编码规范

* 更改horizon头文件编码规范

* 更改horizon头文件编码规范

* 增加horizon packages下载及自动安装

* Add UseHorizonNPUBackend Method

* 删除编译FD SDK后多余的头文件,同时更改部分规范

* Update horizon.md

* Update horizon.md

---------

Co-authored-by: DefTruth <31974251+DefTruth@users.noreply.github.com>

2023-05-06 16:10:37 +08:00

3.7 KiB

Raw Blame History

English | 简体中文

How to Build Horizon Deployment Environment

Horizon refers to the BPU of the Sunrise X3 series chips launched by Horizon. Currently, FastDeploy has initially supported deploying models using Horizon. If you encounter any problems during use, please provide your operating environment and feedback in the Issues section.

Introduction

If you want to use the Horizon inference engine in FastDeploy, you need to configure the following environments:

Tools	Yes/No	Platform	Description
Paddle2ONNX	Yes	PC	Used to convert PaddleInference models to ONNX models
Horizon Environments Docker	Yes	PC	Used to convert ONNX models to Horizon models
Horizon XJ3 OpenExplorer	Yes	PC	header files and dynamic libraries

Model Conversion Environment

Horizon provides a complete model transformation environment (XJ3 chip toolchain image), and FastDeploy adopts the image version of 2.5.2, You can obtain it through the Horizon developer platform.

Software Package

Horizon also provides a complete toolkit (Horizon XJ3 OpenExplorer) , The development package version used by FastDeploy is 2.5.2, You can obtain it through the Horizon developer platform.

Due to the weak performance of the board CPU, it is recommended to perform cross compilation on a PC. The following tutorial is completed in the Docker environment provided by Horizon.

Start Docker Environment

After downloading the Horizon XJ3 chip toolchain image locally, execute the following command to import the image package into the Docker environment:

docker load < docker_openexplorer_ubuntu_20_xj3_gpu_v2.5.2_py38.tar.gz

After downloading the dependent software packages to the local machine, unzip them:

tar -xvf horizon_xj3_openexplorer_v2.5.2_py38_20230331.tar.gz

After the unzipping is complete, cd to that directory:

cd horizon_xj3_open_explorer_v2.5.2-py38_20230331/

Under the root directory, there is a script to run Docker. Run the following command:

sh run_docker.sh /home gpu

The first directory is the directory to be mounted on the container, and the latter parameter is to enable GPU acceleration for the Docker.

At this point, the preparation of the required environment for compilation is complete.

How to Build and Install C++ SDK

Download the cross-compilation tool, gcc_linaro_6.5.0_2018.12_x86_64_aarch64_linux_gnu, and it is recommended to extract it to the /opt directory.

git clone https://github.com/PaddlePaddle/FastDeploy.git
cd FastDeploy

git checkout develop

mkdir build && cd build
cmake ..  -DCMAKE_C_COMPILER=/opt/gcc_linaro_6.5.0_2018.12_x86_64_aarch64_linux_gnu/gcc-linaro-6.5.0-2018.12-x86_64_aarch64-linux-gnu/bin/aarch64-linux-gnu-gcc \
          -DCMAKE_CXX_COMPILER=/opt/gcc_linaro_6.5.0_2018.12_x86_64_aarch64_linux_gnu/gcc-linaro-6.5.0-2018.12-x86_64_aarch64-linux-gnu/bin/aarch64-linux-gnu-g++ \
          -DCMAKE_TOOLCHAIN_FILE=./../cmake/toolchain.cmake \
          -DTARGET_ABI=arm64 \
          -WITH_TIMVX=ON \
          -DENABLE_HORIZON_BACKEND=ON \
          -DENABLE_VISION=ON \
          -DCMAKE_INSTALL_PREFIX=${PWD}/fastdeploy-0.0.0 \
          -Wno-dev ..
make -j16
make install

3.7 KiB Raw Blame History