#GPU Development Documentation

For the specific GPU hardware introduction, development environment, and development software interface, please refer to the introduction in the Horizon Journey 6 System Software - J6X GPU Module Development and Debugging Guide.

#Sample

This sample shows you how to call the OpenCL API to implement a simple threshold function: The gpu sample is located in samples/ucp_tutorial/gpu_sample, and the code structure is as follows:

+--gpu_sample
├── build.sh
├── CMakeLists.txt
├── code                      # sample implementation code
│   ├── main.cc
│   ├── threshold.cl
│   ├── tool.cc
│   └── tool.h
├── README.md
└── script                    # aarch64 board production
    ├── data                  # data directory
    └── run.sh

#Principle

$\begin{aligned}dst_x=\begin{cases}maxval, src_x > threshold\\0, src_x < = threshold\end{cases}\end{aligned}$

Input Image	Parameters	Output Image
	threshold=100 maxval=200

#Thread Configuration

1. The total task calculation amount is WIDTHxHEIGHT. Each work item calculates the pixel block size of 32x16. From this, the total number of work items under this division can be obtained as follows:

(static_cast<size_t>(width) + workItemWidthSize - 1) / workItemWidthSize
(static_cast<size_t>(height) + workItemHeightSize - 1) /workItemHeightSize

2. Each thread group is configured with 4x4 work items.

#Environment Construction

#Development Board Preparation

After getting the development board, make sure to use the recommended system image version and that the local development machine and the development board can be remotely connected.

#Compile

The GPU sample only supports compiling and generating products that are executed on the board, the execution script is as follows:

cd gpu_sample/
bash build.sh

#Sample Usage

Run on the board, copy the script directory to the board, and execute the following command:

cd script
bash run.sh