src

Build Instructions

This project only uses C, the build system is a separate C file nob.c (check the structure section).

Requirements:

• A C11 compatible compiler, tested mainly with gcc on GNU + Linux, cl (msvc, make sure you are using a 64-bit version) on windows.
• FFmpeg development libraries: avformat, avcodec, swscale, and avutil. If running on Windows the environment variable FFMPEG_DIR must be set to the directory where the FFmpeg libraries are located.

The libraries can be installed in GNU + Linux systems as follows (if I added anything wrong here please open an issue):

• Debian / Ubuntu:

sudo apt install libavformat-dev libavcodec-dev libswscale-dev libavutil-dev

• Fedora / RHEL / CentOS / Rocky Linux / AlmaLinux (FFmpeg packages are generally provided through additional repositories such as RPM Fusion):

sudo dnf install ffmpeg-devel

• Arch Linux (and adjacent like Atrix) / Manjaro / CachyOS:

sudo pacman -S ffmpeg

• openSUSE (Package names may vary with the FFmpeg version available in the configured repositories, I don't know the exact versions that will work as I don't use open SUSE nor anyone that does, but I'm using FFmpeg version 9 and from what I read the repos use this version as well):

sudo zypper install ffmpeg-9-libavformat-devel ffmpeg-9-libavcodec-devel ffmpeg-9-libswscale-devel ffmpeg-9-libavutil-devel

• Alpine Linux:

sudo apk add ffmpeg-dev

• Gentoo:

USE flags: dav1d (enabled by default)

sudo emerge media-video/ffmpeg

If you are using a different distribution, please check the package manager for the FFmpeg development libraries.

On Windows the easiest way is with win-get, just run:

winget install "FFmpeg (Shared)"

Building:

The first time you compile you will have to generate the build tool, simply run (uses gcc by default): gcc nob.c -o nob or cl nob.c

Now you can compile the project with: ./nob(.exe) -target TARGET -platform PLATFORM -build_type BUILD_TYPE

The parameters are as follows:

• TARGET: inspector | reconstructor | tests (see below)
• PLATFORM: gnu_linux_x11 | gnu_linux_wayland | windows (windows does not exist for now)
• BUILD_TYPE: debug | release

For example, to build the inspector for GNU + Linux with X11: ./nob -target inspector -platform gnu_linux_x11 -build_type release

Both GNU + Linux back-ends speak their display protocols directly, so no X11 or Wayland development packages are required.

The generated executable will be in the form ""TARGET_BUILD_TYPE_PLATFORM(.exe)"" in the build/ directory. Exemple: build/inspector_release_gnu_linux_x11 or build\inspector_debug_windows.exe

Running:

You can tun the inspector using the executable and passing a video path: build/inspector_...(.exe) video_path

Testing:

Testing can be performed with either a whitelist or a blacklist: ./build/tests_...(.exe) -enable TESTS_TO_RUN or ./build/tests_...(.exe) -disable TESTS_TO_EXCLUDE

Not providing any option will run all of the tests.

List of tests:

• dyn_arr: dynamic array tests

Structure

Common code (shared across inspector and reconstructor)

This includes the following directories:

1. platform, with the platform layers for GNU + Linux (X11 and Wayland) and Windows (unimplemented)
2. input, input system
3. renderer, the source code for the software renderer
4. ds, the data structures used in the projects
5. math, the mathematical functions and objects used in the code
6. lib, third-party code used in the project
7. test_files, files used to test the correctness of the programs. For now this is empty as I didn't write any tests that directly require video data, but I recommend checking the sources for where to get some testing data.

The following are present in lib/ (the .h files are in there):

• nob, the build system (by TSoding)
• flag, for parsing the build system arguments (by TSoding)
• astf, for automated testing (by me)

Inspector

The inspector is a tool to visualize compression data of the videos, such as color channels, chroma channels and inter prediction. See more here

Reconstructor

The reconstructor is a tool to reconstruct 3d models from video data using computer vision. See more here

Profiling

In this project we use both perf and "Intel VTune Profiler" (check sources). Profiling data is not uploaded to GitHub as it's hardware dependent and can be extracted with ease.

Information relevat to performance can be found on the TODO list