This project contains scripts to setup, build, and profile AMD's ROCm MIVisionX. The open-source GitHub project can be found at MIVisionX

For convenience of the developer, we here provide the scripts which will install all the dependencies required by this project and clone the project from GitHub to build and install MIVisionX on your system.

• Linux distribution
  • Ubuntu - 18.04 / 20.04
  • CentOS - 7 / 8
• ROCm supported hardware
• ROCm

This folder has the following python scripts

MIVisionX-setup.py builds all the prerequisites required by MIVisionX. The setup script creates a deps folder and installs all the prerequisites, this script only needs to be executed once. If directory option is not given, the script will install deps folder in the home directory(~/) by default, else in the user specified location.

usage:

python MIVisionX-setup.py --directory [setup directory - optional (default:~/)]
                          --opencv    [OpenCV Version - optional (default:4.5.5)]
                          --protobuf  [ProtoBuf Version - optional (default:3.12.0)]
                          --rpp       [RPP Version - optional (default:0.91)]
                          --ffmpeg    [FFMPEG V4.0.4 Installation - optional (default:no) [options:yes/no]]
                          --rocal     [MIVisionX rocAL Dependency Install - optional (default:yes) [options:yes/no]]
                          --neural_net[MIVisionX Neural Net Dependency Install - optional (default:yes) [options:yes/no]]
                          --reinstall [Remove previous setup and reinstall (default:no)[options:yes/no]]
                          --backend   [MIVisionX Dependency Backend - optional (default:HIP) [options:OCL/HIP]]
                          --rocm_path [ROCm Installation Path - optional (default:/opt/rocm) - ROCm Installation Required]

Note:

• ROCm upgrade with sudo apt upgrade requires the setup script rerun.
• use X Window / X11 for remote GUI app control

MIVisionX-build.py - This script clones the latest MIVisionX from GitHub, builds and installs the project. If the -d build directory is not given the script creates a MIVisionX folder in the home/'~/' directory by default, else in the user-specified folder.

usage:

python MIVisionX-build.py --directory [Build directory - optional (default:~/)]
                          --cmake     [Linux cmake - optional (default:cmake) [options:Ubuntu - cmake; CentOS - cmake3]]  

Note: use --cmake cmake3 for CentOS

MIVisionX-generatePlatformReport.py - This Scripts generates the platfrom report for the system.

usage:

python scripts/MIVision-generatePlatformReport.py

MIVisionX-profile.py - This script downloads the caffe .models & .prototxt from a remote file server and runs every model with different batch sizes and dumps an output.log file, profile.csv & profile.txt. The build directory should be the same director passed to the MIVision-build.py script. If no directory was given, pass '~/' for the directory option.

usage:

python scripts/MIVision-profile.py  -d [build directory - required (directory used in MIVisionX-build.py)]
                                    -l [profile level - optional (level 1-8, default:7)]
                                    -m [profile mode - optional (level 1-6, default:2)]
                                    -f [MIOPEN_FIND_ENFORCE mode - optional (level 1-5, default:1)]

MIVisionX-generateFullReport.py - This Scripts generates the platform report & benchmark reports for the system.

usage:

python scripts/MIVision-generateFullReport.py --directory [Directory used in MIVisionX-profile.py - optional (default:~/)]
                                              --mode      [Mode used in MIVisionX-profile.py      - optional (default: 1))]

• The MIVision-profile.py will generate profile.txt and profile.csv.
• The MIVision-generatePlatformReport.py will generate platform report .md file.
• The MIVision-generateFullReport.py will generate platform and benchmark report .md file. A Diff report will be generated if the report was previously run on the same machine.