Add void linux to README, update pkill command

Fixes portal capture failing on some systems

.gitignore

@@ -4,14 +4,15 @@ compile_commands.json
 tests/sibs-build/
 tests/compile_commands.json
 
-external/wlr-export-dmabuf-unstable-v1-client-protocol.h
-external/wlr-export-dmabuf-unstable-v1-protocol.c
+**/xdg-output-unstable-v1-client-protocol.h
+**/xdg-output-unstable-v1-protocol.c
 
 .clangd/
 .cache/
 .vscode/
 
 build/
+debug-build/
 
 *.o
 gpu-screen-recorder

README.md

@@ -14,7 +14,7 @@ This is a cli-only tool, if you want an UI for this check out [GPU Screen Record
 Supported video codecs:
 * H264 (default)
 * HEVC (Optionally with HDR)
-* AV1 (Optionally with HDR. Not currently supported on NVIDIA if you use GPU Screen Recorder flatpak)
+* AV1 (Optionally with HDR. Not currently supported on NVIDIA in the flatpak version of GPU Screen Recorder)
 * VP8
 * VP9
 
@@ -26,50 +26,29 @@ Supported image formats:
 * JPEG
 * PNG
 
-This software works on X11 and Wayland on AMD, Intel and NVIDIA. Replay data is stored in RAM, not disk.
-### TEMPORARY ISSUES
-1) Videos are in variable framerate format. Use MPV to play such videos, otherwise you might experience stuttering in the video if you are using a buggy video player. You can try saving the video into a .mkv file instead as some software may have better support for .mkv files (such as kdenlive). You can use the "-fm cfr" option to to use constant framerate mode.
-2) FLAC audio codec is disabled at the moment because of temporary issues.
-### AMD/Intel/Wayland root permission
-When recording a window or when using the `-w portal` option no special user permission is required,
-however when recording a monitor the program needs root permission (to access KMS).\
-This is safe in GPU Screen Recorder as the part that needs root access has been moved to its own small program that only does one thing.\
-For you as a user this only means that if you installed GPU Screen Recorder as a flatpak then a prompt asking for root password will show up once when you start recording.
-# Performance
-On a system with a i5 4690k CPU and a GTX 1080 GPU:\
-When recording Legend of Zelda Breath of the Wild at 4k, fps drops from 30 to 7 when using OBS Studio + nvenc, however when using this screen recorder the fps remains at 30.\
-When recording GTA V at 4k on highest settings, fps drops from 60 to 23 when using obs-nvfbc + nvenc, however when using this screen recorder the fps only drops to 58.\
-GPU Screen Recorder also produces much smoother videos than OBS when GPU utilization is close to 100%, see comparison here: [https://www.youtube.com/watch?v=zfj4sNVLLLg](https://www.youtube.com/watch?v=zfj4sNVLLLg).\
-GPU Screen Recorder has much better performance than OBS Studio even with version 30.2 that does "zero-copy" recording and encoding, see: [https://www.youtube.com/watch?v=jdroRjibsDw](https://www.youtube.com/watch?v=jdroRjibsDw).\
-It is recommended to save the video to a SSD because of the large file size, which a slow HDD might not be fast enough to handle. Using variable framerate mode (-fm vfr) which is the default is also recommended as this reduces encoding load. Ultra quality is also overkill most of the time, very high (the default) or lower quality is usually enough.\
-Note that recording on AMD can have some performance issues on Wayland in the recording itself when recording without desktop portal unless your mesa version is 25.0.0 or greater.
-## Note about optimal performance on NVIDIA
-NVIDIA driver has a "feature" (read: bug) where it will downclock memory transfer rate when a program uses cuda (or nvenc, which uses cuda), such as GPU Screen Recorder. To work around this bug, GPU Screen Recorder can overclock your GPU memory transfer rate to it's normal optimal level.\
-To enable overclocking for optimal performance use the `-oc` option when running GPU Screen Recorder. You also need to have "Coolbits" NVIDIA X setting set to "12" to enable overclocking. You can automatically add this option if you run `sudo nvidia-xconfig --cool-bits=12` and then reboot your computer.\
-Note that this only works when Xorg server is running as root, and using this option will only give you a performance boost if the game you are recording is bottlenecked by your GPU.\
-Note! use at your own risk!
-# VRR/G-SYNC
-This should work fine on AMD/Intel X11 or Wayland. On Nvidia X11 G-SYNC only works with the -w screen-direct option, but because of bugs in the Nvidia driver this option is not always recommended.
-For example it can cause your computer to freeze when recording certain games.
+This software works on X11 and Wayland on AMD, Intel and NVIDIA.
 
 # Installation
-If you are running an Arch Linux based distro then you can find gpu screen recorder on aur under the name gpu-screen-recorder (`yay -S gpu-screen-recorder`).\
+If you are running an Arch Linux based distro then you can find gpu screen recorder in the official repositories under the name gpu-screen-recorder (`sudo pacman -S gpu-screen-recorder`).\
 If you are running another distro then you can run `sudo ./install.sh`, but you need to manually install the dependencies, as described below.\
-You can also install gpu screen recorder ([the gtk gui version](https://git.dec05eba.com/gpu-screen-recorder-gtk/)) from [flathub](https://flathub.org/apps/details/com.dec05eba.gpu_screen_recorder), which is the easiest method
+You can also install gpu screen recorder ([the ui version](https://git.dec05eba.com/gpu-screen-recorder-gtk/)) from [flathub](https://flathub.org/apps/details/com.dec05eba.gpu_screen_recorder), which is the easiest method
 to install GPU Screen Recorder on non-arch based distros.\
-If you install GPU Screen Recorder flatpak, which is the gtk gui version then you can still run GPU Screen Recorder command line by using the flatpak command option, for example `flatpak run --command=gpu-screen-recorder com.dec05eba.gpu_screen_recorder -w screen -f 60 -o video.mp4`. Note that if you want to record your monitor on AMD/Intel then you need to install the flatpak system-wide (like so: `flatpak install flathub --system com.dec05eba.gpu_screen_recorder`).
+If you install GPU Screen Recorder flatpak, which is the gtk gui version then you can still run GPU Screen Recorder command line by using the flatpak command option, for example `flatpak run --command=gpu-screen-recorder com.dec05eba.gpu_screen_recorder -w screen -f 60 -o video.mp4`. Note that if you want to record your monitor on AMD/Intel then you need to install the flatpak system-wide (like so: `flatpak install --system com.dec05eba.gpu_screen_recorder`).
 
 ## Unofficial install methods
-The only official ways to install GPU Screen Recorder is either from source, AUR or flathub. Other sources may be out of date and missing features or may not work correctly.\
+The only official ways to install GPU Screen Recorder is either from source, arch linux extra repository or flathub. Other sources may be out of date and missing features or may not work correctly.\
 If you install GPU Screen Recorder from somewhere else and have an issue then try installing it from one of the official sources before reporting it as an issue.\
 If you still prefer to install GPU Screen Recorder with a package manager instead of from source or as a flatpak then you may be able to find a package for your distro.\
 Here are some known unofficial packages:
 * Debian/Ubuntu: [Pacstall](https://pacstall.dev/packages/gpu-screen-recorder)
 * Nix: [NixOS wiki](https://wiki.nixos.org/wiki/Gpu-screen-recorder)
 * openSUSE: [openSUSE software repository](https://software.opensuse.org/package/gpu-screen-recorder)
-* Fedora: [Copr](https://copr.fedorainfracloud.org/coprs/brycensranch/gpu-screen-recorder-git/)
+* Fedora, CentOS: [Copr](https://copr.fedorainfracloud.org/coprs/brycensranch/gpu-screen-recorder-git/)
 * OpenMandriva: [gpu-screen-recorder](https://github.com/OpenMandrivaAssociation/gpu-screen-recorder)
 * Solus: [gpu-screen-recorder](https://github.com/getsolus/packages/tree/main/packages/g/gpu-screen-recorder)
+* Nobara: [Nobara wiki](https://wiki.nobaraproject.org/en/general-usage/additional-software/GPU-Screen-Recorder)
+* AppImage [AppImage GitHub releases](https://github.com/pkgforge-dev/gpu-screen-recorder-AppImage/releases)
+* Void Linux: [gpu-screen-recorder](https://voidlinux.org/packages/?arch=x86_64&q=gpu-screen-recorder) (Make sure to read the README in the package)
 
 # Dependencies
 GPU Screen Recorder uses meson build system so you need to install `meson` to build GPU Screen Recorder.
@@ -77,18 +56,25 @@ GPU Screen Recorder uses meson build system so you need to install `meson` to bu
 ## Build dependencies
 These are the dependencies needed to build GPU Screen Recorder:
 
-* libglvnd (which provides libgl, libglx and libegl)
-* vulkan-headers
-* ffmpeg (libavcodec, libavformat, libavutil, libswresample, libavfilter)
 * x11 (libx11, libxcomposite, libxrandr, libxfixes, libxdamage)
-* libpulse
+* wayland (wayland-client, wayland-egl, wayland-scanner)
+* ffmpeg (libavcodec, libavformat, libavutil, libswresample, libavfilter)
 * libva (and libva-drm)
+* libpulse
 * libdrm
 * libcap
-* wayland-client
-* wayland-egl
+* vulkan-headers
+* linux-api-headers
+
+## Optional dependencies
+When building GPU Screen Recorder with portal support (`-Dportal=true` meson option, which is enabled by default) these dependencies are also needed:
+* libdbus
+* libpipewire (and libspa which is usually part of libpipewire)
 
 ## Runtime dependencies
+* libglvnd (which provides libgl, libglx and libegl) is needed. Your system needs to support at least OpenGL ES 3.0 (released in 2012)
+* libturbojpeg (aka libjpeg-turbo) is needed when capturing camera with mjpeg pixel format option
+
 There are also additional dependencies needed at runtime depending on your GPU vendor:
 
 ### AMD
@@ -97,7 +83,7 @@ There are also additional dependencies needed at runtime depending on your GPU v
 
 ### Intel
 * mesa
-* vaapi (intel-media-driver/libva-intel-driver/linux-firmware, depending on which intel iGPU you have)
+* vaapi (intel-media-driver/libva-intel-driver/linux-firmware-intel, depending on which intel GPU you have)
 
 ### NVIDIA
 * cuda runtime (libcuda.so.1) (libnvidia-compute)
@@ -105,53 +91,110 @@ There are also additional dependencies needed at runtime depending on your GPU v
 * nvfbc (libnvidia-fbc1, when recording the screen on x11)
 * xnvctrl (libxnvctrl0, when using the `-oc` option)
 
-## Optional dependencies
-When compiling GPU Screen Recorder with portal support (`-Dportal=true`, which is enabled by default) these dependencies are also needed:
-* libdbus
-* libpipewire (and libspa which is usually part of libpipewire)
-
 # How to use
 Run `gpu-screen-recorder --help` to see all options and also examples.\
 There is also a gui for the gpu screen recorder called [GPU Screen Recorder GTK](https://git.dec05eba.com/gpu-screen-recorder-gtk/).\
 There is also a new alternative UI for GPU Screen Recorder in the style of ShadowPlay called [GPU Screen Recorder UI](https://git.dec05eba.com/gpu-screen-recorder-ui/).
 ## Recording
 Here is an example of how to record your monitor and the default audio output: `gpu-screen-recorder -w screen -f 60 -a default_output -o ~/Videos/test_video.mp4`.
-Yyou can stop and save the recording with `Ctrl+C` or by running `killall -SIGINT gpu-screen-recorder`.
+Yyou can stop and save the recording with `Ctrl+C` or by running `pkill -SIGINT -f "^gpu-screen-recorder"`.
 You can see a list of capture options to record if you run `gpu-screen-recorder --list-capture-options`. This will list possible capture options and monitor names, for example:\
 ```
   window
   DP-1|1920x1080
 ```
-in this case you could record a window or a monitor with the name `DP-1`.
-## Streaming
-Streaming works the same as recording, but the `-o` argument should be path to the live streaming service you want to use (including your live streaming key). Take a look at `scripts/twitch-stream.sh` to see an example of how to stream to twitch.
+in this case you could record a window or a monitor with the name `DP-1`.\
+To list available audio devices that you can use you can run `gpu-screen-recorder --list-audio-devices` and the name to use is on the left size of the `|`.\
+To list available audio application names that you can use you can run `gpu-screen-recorder --list-application-audio`.\
+You can run `gpu-screen-recorder --info` to list more information about the system, such as the device that is used for capture and video encoding and supported codecs. These commands can be parsed by scripts/programs.
 ## Replay mode
 Run `gpu-screen-recorder` with the `-c mp4` and `-r` option, for example: `gpu-screen-recorder -w screen -f 60 -r 30 -c mp4 -o ~/Videos`. Note that in this case, `-o` should point to a directory.\
 If `-df yes` is set, replays are save in folders based on the date.
 The file path to the saved replay is output to stdout. All other output from GPU Screen Recorder are output to stderr.
 You can also use the `-sc` option to specify a script that should be run (asynchronously) when the video has been saved and the script will have access to the location of the saved file as its first argument.
-This can be used for example to show a notification when a replay has been saved, to rename the video with a title that matches the game played (see `scripts/record-save-application-name.sh` as an example on how to do this on X11) or to re-encode the video.\
-The replay buffer is stored in ram (as encoded video), so don't use a too large replay time and/or video quality unless you have enough ram to store it.
+This can be used for example to show a notification when a replay has been saved, to rename the video with a title that matches the game played (see `scripts/record-save-application-name.sh` as an example on how to do this on X11) or to re-encode the video.
+
+The replay buffer is stored in ram (as encoded video) by default, so don't use a too large replay time and/or video quality unless you have enough ram to store it.\
+You can use the `-replay-storage disk` option to store the replay buffer on disk instead of ram (in the same location as the output video).\
+By default videos are recorded with constant quality, but with replay mode you might want to record in constant bitrate mode instead for consistent ram/disk usage in high motion scenes. You can do that by using the `-bm cbr` option (along with `-q` option, for example `-bm cbr -q 20000`).
+## Streaming
+Streaming works the same way as recording, but the `-o` argument should be path to the live streaming service you want to use (including your live streaming key). Take a look at `scripts/twitch-stream.sh` to see an example of how to stream to twitch.\
+GPU Screen Recorder uses Ffmpeg so GPU Screen Recorder supports all protocols that Ffmpeg supports.\
+If you want to reduce latency one thing you can do is to use the `-keyint` option, for example `-keyint 0.5`. Lower value means lower latency at the cost of increased bitrate/decreased quality.
+## Recording while using replay/streaming
+You can record a regular video while using replay/streaming by launching GPU Screen Recorder with the `-ro` option to specify a directory where to save the recording (for example: `gpu-screen-recorder -w screen -c mp4 -r 60 -o "$HOME/Videos/replays" -ro "$HOME/Videos/recordings"`).\
+To start/stop (and save) recording use the SIGRTMIN signal, for example `pkill -SIGRTMIN -f "^gpu-screen-recorder"`. The path to the video will be displayed in stdout when saving the video.\
+This way of recording while using replay/streaming is more efficient than running GPU Screen Recorder multiple times since this way it only records the screen and encodes the video once.
 ## Controlling GPU Screen Recorder remotely
-To save a video in replay mode, you need to send signal SIGUSR1 to gpu screen recorder. You can do this by running `killall -SIGUSR1 gpu-screen-recorder`.\
-To stop recording send SIGINT to gpu screen recorder. You can do this by running `killall -SIGINT gpu-screen-recorder` or pressing `Ctrl-C` in the terminal that runs gpu screen recorder. When recording a regular non-replay video this will also save the video.\
-To pause/unpause recording send SIGUSR2 to gpu screen recorder. You can do this by running `killall -SIGUSR2 gpu-screen-recorder`. This is only applicable and useful when recording (not streaming nor replay).\
+To save a video in replay mode, you need to send signal SIGUSR1 to gpu screen recorder. You can do this by running `pkill -SIGUSR1 -f "^gpu-screen-recorder"`.\
+To stop recording send SIGINT to gpu screen recorder. You can do this by running `pkill -SIGINT -f "^gpu-screen-recorder"` or pressing `Ctrl-C` in the terminal that runs gpu screen recorder. When recording a regular non-replay video this will also save the video.\
+To pause/unpause recording send SIGUSR2 to gpu screen recorder. You can do this by running `pkill -SIGUSR2 -f "^gpu-screen-recorder"`. This is only applicable and useful when recording (not streaming nor replay).\
+There are more signals to control GPU Screen Recorder. Run `gpu-screen-recorder --help` to list them all (under `NOTES` section).
 ## Simple way to run replay without gui
 Run the script `scripts/start-replay.sh` to start replay and then `scripts/save-replay.sh` to save a replay and `scripts/stop-replay.sh` to stop the replay. The videos are saved to `$HOME/Videos`.
 You can use these scripts to start replay at system startup if you add `scripts/start-replay.sh` to startup (this can be done differently depending on your desktop environment / window manager) and then go into
 hotkey settings on your system and choose a hotkey to run the script `scripts/save-replay.sh`. Modify `scripts/start-replay.sh` if you want to use other replay options.
 ## Run replay on system startup
 If you installed GPU Screen Recorder from AUR or from source and you are running a distro that uses systemd then you will have a systemd service installed that can be started with `systemctl enable --now --user gpu-screen-recorder`. This systemd service runs GPU Screen Recorder on system startup.\
-It's configured with `$HOME/.config/gpu-screen-recorder.env` (create it if it doesn't exist). You can look at [extra/gpu-screen-recorder.env](https://git.dec05eba.com/gpu-screen-recorder/plain/extra/gpu-screen-recorder.env) to see an example.
+It's configured with `$HOME/.config/gpu-screen-recorder/gpu-screen-recorder.env` (create it if it doesn't exist). You can look at [extra/gpu-screen-recorder.env](https://git.dec05eba.com/gpu-screen-recorder/plain/extra/gpu-screen-recorder.env) to see an example.
 You can see which variables that you can use in the `gpu-screen-recorder.env` file by looking at the `extra/gpu-screen-recorder.service` file. Note that all of the variables are optional, you only have to set the ones that are you interested in.
 You can use the `scripts/save-replay.sh` script to save a replay and by default the systemd service saves videos in `$HOME/Videos`.
+## Run a script when a video is saved
+Run `gpu-screen-recorder` with the `-sc` option to specify a script that should be run when a recording/replay a saved, for example `gpu-screen-recorder -w screen -sc ./script.sh -o video.mp4`.\
+The first argument to the script is the file path to the saved video. The second argument is either "regular" for regular recordings, "replay" for replays or "screenshot" for screenshots.\
+This can be used to for example showing a notification with the name of video or moving a video to a folder based on the name of the game that was recorded.
+## Plugins
+GPU Screen Recorder supports plugins for rendering additional graphics on top of the monitor/window capture. The plugin interface is defined in `plugin/plugin.h` and it gets installed to `gsr/plugin.h` in the systems include directory (usually `/usr/include`).
+An example plugin can be found at `plugin/examples/hello_triangle`.\
+Run `gpu-screen-recorder` with the `-p` option to specify a plugin to load, for example `gpu-screen-recorder -w screen -p ./triangle.so -o video.mp4`.
+`-p` can be specified multiple times to load multiple plugins.\
+Build GPU Screen Recorder with the `-Dplugin_examples=true` meson option to build plugin examples.
+## Smoother recording
+If you record at your monitors refresh rate and enabled vsync in a game then there might be a desync between the game updating a frame and GPU Screen Recorder capturing a frame.
+This is an issue in some games.
+If you experience this issue then you might want to either disable vsync in the game or use the `-fm content` option to sync capture to the content on the screen. For example: `gpu-screen-recorder -w screen -fm content -o video.mp4`.\
+Note that this option is currently only available on X11, or with desktop portal capture on Wayland (`-w portal`).
+
+# Performance
+On a system with an i5 4690k CPU and a GTX 1080 GPU:\
+When recording Legend of Zelda Breath of the Wild at 4k, fps drops from 30 to 7 when using OBS Studio + nvenc, however when using this screen recorder the fps remains at 30.\
+When recording GTA V at 4k on highest settings, fps drops from 60 to 23 when using obs-nvfbc + nvenc, however when using this screen recorder the fps only drops to 58.\
+On a system with an AMD Ryzen 9 5900X CPU and an RX 7800XT GPU I don't see any fps drop at all, even when recording at 4k 60fps with AV1 codec with 10-bit colors.\
+GPU Screen Recorder also produces much smoother videos than OBS when GPU utilization is close to 100%, see comparison here: [https://www.youtube.com/watch?v=zfj4sNVLLLg](https://www.youtube.com/watch?v=zfj4sNVLLLg) and [https://www.youtube.com/watch?v=aK67RSZw2ZQ](https://www.youtube.com/watch?v=aK67RSZw2ZQ).\
+GPU Screen Recorder has much better performance than OBS Studio even with version 30.2 that does "zero-copy" recording and encoding, see: [https://www.youtube.com/watch?v=jdroRjibsDw](https://www.youtube.com/watch?v=jdroRjibsDw).\
+It is recommended to save the video to a SSD because of the large file size, which a slow HDD might not be fast enough to handle. Using variable framerate mode (-fm vfr) which is the default is also recommended as this reduces encoding load. Ultra quality is also overkill most of the time, very high (the default) or lower quality is usually enough.\
+Note that for best performance you should close other screen recorders such as OBS Studio when using GPU Screen Recorder even if they are not recording, since they can affect performance even when idle. This is the case with OBS Studio.
+## Note about optimal performance on NVIDIA
+NVIDIA driver has a "feature" (read: bug) where it will downclock memory transfer rate when a program uses cuda (or nvenc, which uses cuda), such as GPU Screen Recorder. To work around this bug, GPU Screen Recorder can overclock your GPU memory transfer rate to it's normal optimal level.\
+To enable overclocking for optimal performance use the `-oc` option when running GPU Screen Recorder. You also need to have "Coolbits" NVIDIA X setting set to "12" to enable overclocking. You can automatically add this option if you run `sudo nvidia-xconfig --cool-bits=12` and then reboot your computer.\
+Note that this only works when Xorg server is running as root, and using this option will only give you a performance boost if the game you are recording is bottlenecked by your GPU.\
+Note! use at your own risk!
+
 # Issues
 ## NVIDIA
 Nvidia drivers have an issue where CUDA breaks if CUDA is running when suspend/hibernation happens, and it remains broken until you reload the nvidia driver. `extra/gsr-nvidia.conf` will be installed by default when you install GPU Screen Recorder and that should fix this issue. If this doesn't fix the issue for you then your distro may use a different path for modprobe files. In that case you have to install that `extra/gsr-nvidia.conf` yourself into that location.
 You have to reboot your computer after installing GPU Screen Recorder for the first time for the fix to have any effect.
+
+## TEMPORARY ISSUES
+1) Videos are in variable framerate format. Use MPV to play such videos, otherwise you might experience stuttering in the video if you are using a buggy video player. You can try saving the video into a .mkv file instead as some software may have better support for .mkv files (such as kdenlive). You can use the "-fm cfr" option to to use constant framerate mode.
+2) FLAC audio codec is disabled at the moment because of temporary issues.
+
 # Examples
 Look at the [scripts](https://git.dec05eba.com/gpu-screen-recorder/tree/scripts) directory for script examples. For example if you want to automatically save a recording/replay into a folder with the same name as the game you are recording.
 
+# AMD/Intel/Wayland root permission
+When recording a window or when using the `-w portal` option no special user permission is required,
+however when recording a monitor the program needs root permission (to access KMS).\
+This is safe in GPU Screen Recorder as the part that needs root access has been moved to its own small program that only does one thing.\
+For you as a user this only means that if you installed GPU Screen Recorder as a flatpak then a prompt asking for root password will show up once when you start recording.
+
+# VRR/G-SYNC
+This should work fine on AMD/Intel X11 or Wayland. On Nvidia X11 G-SYNC only works with the -w screen-direct option, but because of bugs in the Nvidia driver this option is not always recommended.
+For example it can cause your computer to freeze when recording certain games.
+
+# License
+This software is licensed under GPL-3.0-only, see the LICENSE file for more information.
+
 # Reporting bugs, contributing patches, questions or donation
 See [https://git.dec05eba.com/?p=about](https://git.dec05eba.com/?p=about).
 
@@ -161,17 +204,19 @@ See [https://git.dec05eba.com/?p=about](https://git.dec05eba.com/?p=about).
 # FAQ
 ## It tells me that my AMD/Intel GPU is not supported or that my GPU doesn't support h264/hevc, but that's not true!
 Some linux distros (such as manjaro and fedora) disable hardware accelerated h264/hevc on AMD/Intel because of "patent license issues". If you are using an arch-based distro then you can install mesa-git instead of mesa and if you are using another distro then you may have to switch to a better distro. On fedora based distros you can follow this: [Hardware Accelerated Codec](https://rpmfusion.org/Howto/Multimedia).\
-If you installed GPU Screen Recorder flatpak then you can try installing mesa-extra freedesktop runtime by running this command: `flatpak install --system org.freedesktop.Platform.GL.default//23.08-extra`
+You can alternatively install the flatpak version of GPU Screen Recorder from [flathub](https://flathub.org/apps/details/com.dec05eba.gpu_screen_recorder) which doesn't have this issue on any distro.
 ## I have an old nvidia GPU that supports nvenc but I get a cuda error when trying to record
 Newer ffmpeg versions don't support older nvidia cards. Try installing GPU Screen Recorder flatpak from [flathub](https://flathub.org/apps/details/com.dec05eba.gpu_screen_recorder) instead. It comes with an older ffmpeg version which might work for your GPU.
 ## I get a black screen/glitches while live streaming
 It seems like ffmpeg earlier than version 6.1 has some type of bug. Install ffmpeg version 6.1 or later and then reinstall GPU Screen Recorder to fix this issue. The flatpak version of GPU Screen Recorder comes with a newer version of ffmpeg so no extra steps are needed.
 ## I can't play the video in my browser directly or in discord
-Browsers and discord don't support hevc video codec at the moment. Choose h264 video codec instead with the -k h264 option.
+Browsers and discord don't support hevc video codec at the moment. You can instead choose h264 video codec with the -k h264 option or av1 video codec with the -k av1 option.
 Note that websites such as youtube support hevc so there is no need to choose h264 video codec if you intend to upload the video to youtube or if you want to play the video locally or if you intend to
 edit the video with a video editor. Hevc allows for better video quality (especially at lower file sizes) so hevc (or av1) is recommended for source videos.
 ## I get a black bar/distorted colors on the sides in the video
-This is mostly an issue on AMD. For av1 it's a hardware issue, see: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9185. For hevc it's a software issue in the AMD driver that hasn't been fixed yet. This issue happens at certain video resolutions. If you get this issue then a workaround is to record with h264 video codec instead (using the -k h264 option).
+This is mostly an issue on AMD. For av1 it's a hardware issue, see: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9185. For hevc it's a software issue in ffmpeg that was fixed in ffmpeg version 8.\
+If your ffmpeg version is older than 8 then you can use the flatpak version of GPU Screen Recorder which comes with ffmpeg version >= 8.\
+Alternatively you can record with h264 codec (-k h264, which is also the default codec) to workaround this issue.
 ## The video doesn't display or has a green/yellow overlay
 This can happen if your video player is missing the H264/HEVC video codecs. Either install the codecs or use mpv.
 ## I get stutter in the video
@@ -181,7 +226,41 @@ You have to either record in hdr mode (-k `hevc_hdr` or -k `av1_hdr` option) to
 ## GPU Screen Recorder records night light when recording in HDR mode
 You can record with desktop portal option (`-w portal`) instead which ignores night light, if you are ok with recording without HDR.
 ## Kdenlive says that the video is not usable for editing because it has variable frame rate
-To fix this you can either record the video in .mkv format or constant frame rate (-fm cfr).
+To fix this you can either just press cancel, which will allow you to continue or record the video in .mkv format or constant frame rate (-fm cfr). I recommend recording the video in .mkv format and variable frame rate (-fm vfr).
 ## Colors look incorrect when recording HDR (with hevc_hdr/av1_hdr) or using an ICC profile
 KDE Plasma version 6.2 broke HDR and ICC profiles for screen recorders. This was changed in KDE plasma version 6.3 and recording HDR works now, as long as you set HDR brightness to 100% (which means setting "Maximum SDR Brightness" in KDE plasma display settings to 203) and set color accuracy to "Prefer color accuracy". If you want to convert HDR to SDR then record with desktop portal option (`-w portal`) instead.
 I don't know how well recording HDR works in wayland compositors other than KDE plasma.
+## GPU Screen Recorder starts lagging after 30-40 minutes when launching GPU Screen Recorder from steam command launcher
+This is a [steam issue](https://github.com/ValveSoftware/steam-for-linux/issues/11446). Prepend the gpu-screen-recorder command with `LD_PREFIX=""`, for example `LD_PREFIX="" gpu-screen-recorder -w screen -o video.mp4`.
+## How do I apply audio effects, such as noise suppression?
+You have to use external software for that, such as Easy Effects or NoiseTorch.
+## How do I choose which GPU to record/encode with?
+It's not really possible except in some cases. You can only record with the GPU that is displaying the graphics on your monitor.\
+Some laptops have display adapters that connect external monitors directly to the external GPU (if you have one)
+and on Wayland the external GPU will display the graphics for that monitor.
+In that case you can record the monitor with the external GPU by launching GPU Screen Recorder with [prime-run or by setting the DRI_PRIME environment variable](https://wiki.archlinux.org/title/PRIME) depending on your GPU brand.
+Alternatively you can capture with the desktop portal option (`-w portal`), which should allow you to capture any monitor.\
+However if you really want to change which GPU you want to record and encode with with then you can instead configure your display server (Xorg or Wayland compositor) to run with that GPU,
+then GPU Screen Recorder will automatically use that same GPU for recording and encoding.
+## The rotation of the video is incorrect when the monitor is rotated when using desktop portal capture
+This is a bug in kde plasma wayland. When using desktop portal capture and the monitor is rotated and a window is made fullscreen kde plasma wayland will give incorrect rotation to GPU Screen Recorder.
+This also affects other screen recording software such as obs studio.\
+Capture a monitor directly instead to workaround this issue until kde plasma devs fix it, or use another wayland compositor that doesn't have this issue.
+## System notifications get disabled when recording with desktop portal option
+Some desktop environments such as KDE Plasma turn off notifications when you record the screen with the desktop portal option. You can disable this by going into KDE Plasma settings -> search for notifications and then under "Do Not Disturb mode" untick "During screen sharing".
+## The recorded video lags or I get dropped frames in the video
+This is likely not an issue in the recorded video itself, but the video player you use. GPU Screen Recorder doesn't record by dropping frames. Some video players don't play videos with hardware acceleration by default,
+especially if you record with HEVC/AV1 video codec. In such cases it's recommended to play the video with mpv instead with hardware acceleration enabled (for example: `mpv --vo=gpu --hwdec=auto video.mp4`).
+Some corporate distros such as Fedora (or some Fedora based distros) also disable hardware accelerated video codecs on AMD/Intel GPUs, so you might need to install mpv (or another video player) with flathub instead, which bypasses this restriction.
+## My cursor is flickering in the recorded video
+This is likely an AMD gpu driver issue. It only happens to certain generations of AMD GPUs. On Wayland you can record with the desktop portal option (`-w portal`) to workaround this issue.
+This issue hasn't been observed on X11 yet, but if you do observe it you can either record a window (`-w $(xdotool selectwindow)`) or change your xorg config to use software cursor instead (Add `Option "SWcursor" "true"` under modesetting "Device" section in your xorg config file).
+## Password prompt shows up when I try to record my screen
+If GPU Screen Recorder is installed with -Dcapabilities=true (which is the default option) then `gsr-kms-server` is installed with admin capabilities.
+This removes a password prompt when recording a monitor with the `-w monitor` option (for example `-w screen`). However if the root user is disabled on the system then the password prompt will show up anyways.
+If the root user is disabled on your system then you can instead record with `-w focused` or `-w window_id` on X11 or `-w portal` on Wayland.
+## GPU usage is high on my laptop
+GPU usage on battery powered devices is misleading. For example Intel iGPUs has multiple performance levels and the GPU usage reported on the system is the GPU usage at the current performance level.
+The performance level changes depending on the GPU load, so it may say that GPU usage is 80%, but the actual GPU usage may be 5%.
+## The video is too dark when capturing full-range video or 10-bit video
+This is an issue in some broken video players such as vlc. Play the video with a video player such as mpv (or a mpv frontend such as celluloid) or a browser instead.

TODO

@@ -4,13 +4,8 @@ See https://trac.ffmpeg.org/wiki/EncodingForStreamingSites for optimizing stream
 Look at VK_EXT_external_memory_dma_buf.
 Use mov+faststart.
 Allow recording all monitors/selected monitor without nvfbc by recording the compositor proxy window and only recording the part that matches the monitor(s).
-Allow recording a region by recording the compositor proxy window / nvfbc window and copying part of it.
-Support amf and qsv.
 Disable flipping on nvidia? this might fix some stuttering issues on some setups. See NvCtrlGetAttribute/NvCtrlSetAttributeAndGetStatus NV_CTRL_SYNC_TO_VBLANK https://github.com/NVIDIA/nvidia-settings/blob/d5f022976368cbceb2f20b838ddb0bf992f0cfb9/src/gtk%2B-2.x/ctkopengl.c.
-Replays seem to have some issues with audio/video. Why?
 Cleanup unused gl/egl functions, macro, etc.
-Set audio track name to audio device name (if not merge of multiple audio devices).
-Add support for webcam, but only really for amd/intel because amd/intel can get drm fd access to webcam, nvidia cant. This allows us to create an opengl texture directly from the webcam fd for optimal performance.
 Reverse engineer nvapi so we can disable "force p2 state" on linux too (nvapi profile api with the settings id 0x50166c5e).
 Support yuv444p on amd/intel.
 fix yuv444 for hevc.
@@ -31,9 +26,6 @@ Window capture doesn't work properly in _control_ game after going from pause me
 Monitor capture on steam deck is slightly below the game fps, but only when capturing on the steam deck screen. If capturing on another monitor, there is no issue.
     Is this related to the dma buf rotation issue? different modifier being slow? does this always happen?
 
-Fallback to vaapi copy in kms if opengl version fails. This can happen on steam deck for some reason (driver bug?). Also vaapi copy uses less gpu since it uses video codec unit to copy.
-Test if vaapi copy version uses less memory than opengl version.
-
 Intel is a bit weird with monitor capture and multiple monitors. If one of the monitors is rotated then all the kms will be rotated as well.
     Is that only the case when the primary monitor is rotated? Also the primary monitor becomes position 0, 0 so crtc (x11 randr) position doesn't match the drm pos. Maybe get monitor position and size from drm instead.
     How about if multiple monitors are rotated?
@@ -65,38 +57,24 @@ Remove follow focused option.
 
 Exit if X11/Wayland killed (if drm plane dead or something?)
 
-Use SRC_W and SRC_H for screen plane instead of crtc_w and crtc_h.
-
 Test if p2 state can be worked around by using pure nvenc api and overwriting cuInit/cuCtxCreate* to not do anything. Cuda might be loaded when using nvenc but it might not be used, with certain record options? (such as h264 p5).
     nvenc uses cuda when using b frames and rgb->yuv conversion, so convert the image ourselves instead.-
 
 Drop frames if live streaming cant keep up with target fps, or dynamically change resolution/quality.
 
-Support low power option.
-
 Instead of sending a big list of drm data back to kms client, send the monitor we want to record to kms server and the server should respond with only the matching monitor, and cursor.
 
 Tonemap hdr to sdr when hdr is enabled and when hevc_hdr/av1_hdr is not used.
 
-Add 10 bit record option, h264_10bit, hevc_10bit and av1_10bit.
-
 Rotate cursor texture properly (around top left origin).
 
-Setup hardware video context so we can query constraints and capabilities for better default and better error messages.
-
-Use CAP_SYS_NICE in flatpak too on the main gpu screen recorder binary. It makes recording smoother, especially with constant framerate.
-
 Modify ffmpeg to accept opengl texture for nvenc encoding. Removes extra buffers and copies.
 
 When vulkan encode is added, mention minimum nvidia driver required. (550.54.14?).
 
-Support drm plane rotation. Neither X11 nor any Wayland compositor currently rotates drm planes so this might not be needed.
-
 Investigate if there is a way to do gpu->gpu copy directly without touching system ram to enable video encoding on a different gpu. On nvidia this is possible with cudaMemcpyPeer, but how about from an intel/amd gpu to an nvidia gpu or the other way around or any combination of iGPU and dedicated GPU?
     Maybe something with clEnqueueMigrateMemObjects? on AMD something with DirectGMA maybe?
 
-Go back to using pure vaapi without opengl for video encoding? rotation (transpose) can be done if its done after (rgb to yuv) color conversion.
-
 Use lanczos resampling for better scaling quality. Lanczos resampling can also be used for YUV chroma for better color quality on small text.
 
 Flac is disabled because the frame sizes are too large which causes big audio/video desync.
@@ -105,11 +83,14 @@ Enable b-frames.
 
 Support vfr matching games exact fps all the time. On x11 use damage tracking, on wayland? maybe there is drm plane damage tracking. But that may not be accurate as the compositor may update it every monitor hz anyways. On wayland maybe only support it for desktop portal + pipewire capture.
     Another method to track damage that works regardless of the display server would be to do a diff between frames with a shader.
+    A 1x1 texture could be created and then write to the texture with imageStore in glsl.
+    Multiple textures aren't needed for diff, the diff between the color conversion output can be done by using it as an input
+    as well, which would diff it against the previous frame.
 
 Support selecting which gpu to use. This can be done in egl with eglQueryDevicesEXT and then eglGetPlatformDisplayEXT. This will automatically work on AMD and Intel as vaapi uses the same device. On nvidia we need to use eglQueryDeviceAttribEXT with EGL_CUDA_DEVICE_NV.
     Maybe on glx (nvidia x11 nvfbc) we need to use __NV_PRIME_RENDER_OFFLOAD, __NV_PRIME_RENDER_OFFLOAD_PROVIDER, __GLX_VENDOR_LIBRARY_NAME, __VK_LAYER_NV_optimus, VK_ICD_FILENAMES instead. Just look at prime-run /usr/bin/prime-run.
 
-When adding support for steam deck, add option to send video to another computer.
+Add option to send video to another computer.
 New gpu screen recorder gui should have the option to cut the video directly, maybe running an ffmpeg command or implementing that ourselves. Only support gpu screen recorder video files.
 
 Check if is software renderer by using eglQueryDisplayAttribEXT(egl_display, EGL_DEVICE_EXT..) eglQueryDeviceStringEXT(egl_device, EGL_EXTENSIONS) and check for "EGL_MESA_device_software".
@@ -146,8 +127,6 @@ Enable 2-pass encoding.
 
 Restart replay/update video resolution if monitor resolution changes.
 
-Fix pure vaapi copy on intel.
-
 Use nvidia low latency options for better encoding times.
 
 Test ideal async_depth value. Increasing async_depth also increased gpu memory usage a lot (from 100mb to 500mb when moving from async_depth 2 to 16) at 4k resolution. Setting it to 8 increases it by 200mb which might be ok.
@@ -178,9 +157,7 @@ Test if `xrandr --output DP-1 --scale 1.5` captures correct size on nvidia.
 
 Fix cursor position and scale when scaling x11 display.
 
-Support surround audio in application audio recording. Right now only stereo and mono sound is supported.
-
-Support application audio recording without pulseaudio combined sink.
+Support application audio recording without pipewire combined sink.
 
 Support transposing (rotating) with vaapi. This isn't supported on many devices with rgb buffer, but its supported with nv12 buffer (on intel at least).
 
@@ -199,13 +176,6 @@ Use wayland color management protocol when it's available: https://gitlab.freede
 Use different exit codes for different errors. Use one for invalid -w option, another one for invalid -a option for audio devices, etc. This is to make UI error reporting better.
     Document these exit codes in an exit code .md file, or finally create a manpage where this can be documented.
 
-Ffmpeg fixed black bars in videos on amd when using hevc and when recording at some resolutions, such as 1080p:
-    https://github.com/FFmpeg/FFmpeg/commit/bcfbf2bac8f9eeeedc407b40596f5c7aaa0d5b47
-    https://github.com/FFmpeg/FFmpeg/commit/d0facac679faf45d3356dff2e2cb382580d7a521
-    Disable gpu screen recorder black bar handling when using hevc on amd when the libavcodec version is the one that comes after those commits.
-    Also consider the mesa version, to see if the gpu supports this.
-    The version is libavcodec >= 61.28.100
-
 Use opengl compute shader instead of graphics shader. This might allow for better performance when games are using 100% of graphics unit which might fix issue with 100% gpu usage causing gpu screen recorder to run slow when not using vaapi to convert rgb to nv12(?).
 
 Always disable prime run/dri prime and list all monitors to record from from all cards.
@@ -219,8 +189,6 @@ Always disable prime run/dri prime and list all monitors to record from from all
 Allow flv av1 if recent ffmpeg version and streaming to youtube (and twitch?) and for custom services.
 Use explicit sync in pipewire video code: https://docs.pipewire.org/page_dma_buf.html.
 
-Support vaapi rotation. Support for it is added in mesa here: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32919.
-
 Replay (and recording?) fails to save properly sometimes (especially for long videos). This is noticable with mp4 files since they get corrupt and become unplayable.
     The entire video does seem to get saved (it's a large video file) and it seems to have the correct headers but it's not playable.
 
@@ -249,6 +217,191 @@ Support high quality scaling with -s by using lanczos.
 
 Support spanning multiple monitors with region capture. This would also allow the user to record multiple monitors at the same time, the same way screen-direct works on nvidia x11.
 
-When webcam support is added also support v4l2loopback? this is done by using avdevice_register_all(); and -c v4l2 -o /dev/video0; but it needs to output raw data as well instead of h264 and possibly yuv420p. Maybe add a -k yuv420p option to do that.
+When webcam support is added also support v4l2loopback? this is done by using avdevice_register_all(); and -c v4l2 -o /dev/video0; but it needs to output raw data as well instead of h264 and possibly yuv420p. Maybe add a -k yuv420p option to do that or -k rgb.
+    This would be implemented by outputting the raw data directly into the output file, without using the video encoder.
 
-Do proper exit, to call gsr_capture_destroy which will properly stop gsr-kms-server. Otherwise there can be zombie gsr-kms-server on error.
+Do proper exit, to call gsr_capture_destroy which will properly stop gsr-kms-server. Otherwise there can be zombie gsr-kms-server on error.
+
+Cursor position might be slightly wrong on rotated monitor.
+
+External texture doesn't work on nvidia x11, probably because of glx context (requires gles es). External texture is not used on nvidia x11 right now so it's not an issue.
+
+Add option to save replay buffer on disk instead of ram.
+
+nvfbc capture cursor with cursor.h instead and composite that on top. This allows us to also always get a cursor in direct capture mode. This could possible give better performance as well.
+
+Maybe remove external shader code and make a simple external to internal texture converter (compute shader), to reduce texture sampling. Maybe this is faster?
+
+Fix opengl context broken after suspend on nvidia by using this: https://registry.khronos.org/OpenGL/extensions/NV/NV_robustness_video_memory_purge.txt requires glx context creation flags and GetGraphicsResetStatusARB() == PURGED_CONTEXT_RESET_NV check to recreate all graphics.
+
+HDR looks incorrect, brightest point gets cut off.
+
+Make "screen" capture the preferred monitor.
+
+When webcam support is added add the option to add it as a second video track, to make it easier to edit in video editors.
+
+Fix constant framerate not working properly on amd/intel because capture framerate gets locked to the same framerate as
+    game framerate, which doesn't work well when you need to encode multiple duplicate frames (AMD/Intel is slow at encoding!).
+    It also appears to skip audio frames on nvidia wayland? why? that should be fine, but it causes video stuttering because of audio/video sync.
+
+Add option to pass a fd (from socketpair) to use for rpc. In the rpc have a common header, with protocol version, data type and data in an enum.
+
+Add the option to set audio track name, for example with -a "track-name:blabla|device:default_output|app:firefox"
+
+Maybe disable qp/vbr for replay. In that case we can preallocate all replay data (for both ram and disk) and write to that directly when receiving packet (dont do that when also recording at the same time).
+    That could improve performance/disk write optimization and maybe even reduce ram usage because of less blocks/fragmentation.
+
+When rpc is added add the option to add/remove audio devices/app audio and also overlays (from new capture sources). It should also be possible to save a replay of any duration (below duration set with -r).
+    Have an rpc info/status command that just returns the status. This is to check if gpu screen recorder is running.
+
+Support hdr screenshot.
+
+Recreate opengl context on loss. This can happen if there is a gpu driver bug, causing context to need to be recreated. This is a nice improvement to not break recording even with buggy driver.
+
+Support saving video with surround sound. Surround sound audio capture does work, but it gets downmixed to stereo.
+
+Add (render) plugin support. To simplify it (and possibly best performance) create one rgba texture (with the size of the output video) that is used across all plugins.
+    Create a framebuffer and set this texture and the target and set the framebuffer as active before calling the plugins.
+    Then the plugins can render simply by doing simple opengl draw functions.
+    Maybe send some metadata to the plugin, such as video (and framebuffer) size. Although this data can be retrieved from the active framebuffer.
+
+Either support webcam support with raw yuyv, mapping the buffer directly to opengl. Or use mjpeg, mapping the buffer directly to vaapi jpeg decoder and then get then map the decoded buffer to opengl.
+    Some webcams dont support raw yuyv and many webcams support higher framerates for mjpeg.
+
+Allow medium, high, very_high and ultra quality for -bm cbr. If that is used then it will automatically estimate the best bitrate for that quality based on resolution and fps.
+    Maybe do this in the ui instead (or both?), to show estimated file size.
+
+Maybe remove shader compute code. It doesn't seem necessary anymore now that glSwapBuffer/glFinish isn't used. dbus server isn't needed anymore either, the code can be moved back to the gpu screen recorder process.
+
+Add proper check if opengl functions are supported. dlsym for the symbol will return a no-op function if it's not supported, so it silently fails if used.
+
+Colors are offset to bottom left by 1 pixel or so on steam deck in landscape mode.
+
+When constant framerate is used (and for audio) multiple frames need to be encoded after resuming from suspend.
+    The clock jumps forward by around 2-3 seconds (on my machine). Is there a way to make sure the clock doesn't jump forward?
+
+Colors are correct, but they look incorrect for thin elements, such as colored text. This can be improved by sampling neighbor pixels for color average.
+
+Record first video/audio frame immediately.
+
+Disable GL_DEPTH_TEST, GL_CULL_FACE.
+
+kde plasma portal capture for screenshot doesn't work well because the portal ui is still visible when taking a screenshot because of its animation.
+
+We can use dri2connect/dri3open to get the /dev/dri/card device. Note that this doesn't work on nvidia x11.
+
+Add support for QVBR (QP with target bitrate). Maybe use VBR instead, since nvidia doesn't support QVBR and neither does vulkan.
+
+KDE Plasma Wayland seems to use overlay planes now in non-fullscreen mode(limited to 1 overlay plane per gpu). Check if this is the case in the latest kde on arch linux.
+    If it is, then support it in kms capture.
+
+Check if pipewire audio link-factory is available before attempting to use app audio or merging audio with pipewire.
+    Also do the same in supports_app_audio check in gpu-screen-recorder --info output.
+
+Move region capture to an option in the color conversion instead of having the region code in kms capture code. This would make it cleaner and make it work with all capture methods.
+    -w region would just be an option for selecting the monitor and -region would work with all capture methods (-w).
+
+Set top level window argument for portal capture. Same for gpu-screen-recorder-gtk global shortcuts.
+
+Remove unix domain socket code from kms-client/server and use socketpair directly. To make this possible always execute the kms server permission setup in flatpak, before starting recording (in gpu-screen-recorder-gtk).
+
+Add -k best/best_hdr/best_10bit option, to automatically choose the best codec (prefer av1, then hevc and then h264. For webm files choose vp9 and then vp8).
+
+Check if region capture works properly with fractional scaling on wayland.
+
+Add option to specify medium/high/very high/ultra for -bm cbr as well, which should automatically pick bitrate based on resolution and framerate.
+    This should also be reflected in gsr ui.
+
+After adding rpc, making recording while in replay/streaming work differently. Start recording should take audio as an argument, to optionally specify different audio for recording than replay/stream.
+
+After adding rpc, make it possible to add/remove audio and video. The same number of audio tracks should remain, but the audio devices/app should be possible to configure. You should be able to configure the capture sources however you want.
+
+It takes a while to shutdown gpu screen recorder when using replay with a long replay time (when the replay buffer is for example 6gb). Maybe this has to do with cleaning up so much memory. Try if _exit(0) is faster, ignoring cleanup of anything (assuming the gpu driver is happy with that, nvidia doesn't like such things. GPU drivers are often buggy when it comes to automatic cleanup).
+
+Support desktop portal hdr capture when this gets merged: https://invent.kde.org/plasma/kwin/-/merge_requests/8293 note that it only works with pipewire >= 1.5.81. If codec is set to non-hdr codec when use desktop portal without hdr (convert to sdr, remove the code that forces non-hdr equivalent of the codec). Update gsr-gtk and gsr-ui accordingly.
+
+Add option to save screenshot as .qoi. Use that then in gsr-ui for the background for the window (capture the monitor that the ui is going to show on). Do that on cosmic and hyprland, which are unable to display things behind the fullscreen window.
+
+Support pausing recording when recording while replay/streaming.
+
+Maybe use VK_VALVE_video_encode_rgb_conversion with vulkan encoding for shader-less rgb to yuv conversion. That would allow screen capture with no gpu processing.
+
+Cursor sometimes doesn't have color when capturing region scaled (on kde plasma wayland at least).
+
+Remove drm_monitor_get_display_server_data and do that work in the drm monitor query.
+
+In gpu screen recorder --info output codec max resolutions. This allows for better error messages in ui frontends and easier and better error handling.
+
+Set minimum fps for live stream or piping or always.
+
+Support youtube sso.
+
+Remove -fm content (support it but remove it from documentation and output deprecation notice when its used) and use it when using -fm vbr (which is the default option).
+    But first -fm content needs to be support on wayland as well, by checking if there is a difference between frames (checksum the frame content).
+    -fm content also needs to have a minimum fps to prevent live stream from timing out when nothing changes on the screen.
+
+There is a leak in nvfbc. When a monitor is turned off and then on there will be an x11 display leak inside nvfbc. This seems to be a bug in nvfbc.
+    Right now a mitigation has been added to not try to recreate the nvfbc session if the capture target (monitor) isn't connected (predict if nvfbc session create will fail).
+    One possible reason this happens is because bExternallyManagedContext is set to true.
+    This also means that nvfbc leaks connection when destroying nvfbc, even if the monitor is connected (this is not an issue right now because exit is done, but if gsr was turned into a library it would be).
+
+Add option to set audio source volume, maybe by doing for example: -a "default_input|app:firefox;volume=50"
+
+Optimize v4l2 mjpeg by decompressing to yuv (tjDecompressToYUV) instead of rgb. This would allow removing the yuv to rgb conversion on cpu step (and rgb to yuv on the gpu as well) as well as reducing the cpu->gpu image data bandwidth as yuv is compressed.
+
+Do jpeg decoding on the gpu for v4l2 mjpeg capture (see https://github.com/negge/jpeg_gpu). Or use jpeg decoding from vaapi/nvdec.
+
+Support other v4l2 pixel formats, such as h264 and rgb/bgr. High-end cameras use h264 for high resolution high framerate option (4k 60fps or higher).
+
+AV1 medium quality on nvidia seems to be very high quality. The quality needs to be rescaled. Need to test on nvidia av1 gpu, but I dont have any such gpu.
+
+Support multiple -w, to record different sources to different video tracks. For example gameplay to video track 1 and webcam to video track 2.
+
+Update man page with info about v4l2 and new region capture format. Mention that -region is deprecated.
+
+Make multiple -w portal work with -restore-portal-session.
+
+Play with DRM_FORMAT_MOD_LINEAR or other linear formats to avoid cuda copy. If it's already in linear format then instead check the format of the target texture and use the same format.
+
+Give early error when using invalid v4l2 path like with monitor. Use gsr_capture_v4l2_list_devices for query. Or maybe remove the early error for monitor to simplify the code.
+
+Support camera capture with pipewire to support multiple applications recording camera at the same time. Might not be as efficient as V4L2.
+
+Taking a screenshot of camera with mjpeg doesn't work correctly because it updates buffer asynchronously. v4l2 capture should only return 0 if the copy of the data to the texture has finished.
+
+Make capture from multiple sources work on nvidia x11 when capturing monitor + window. It doesn't work right now because monitor requires glx (nvfbc) while window requires egl.
+
+Support v4l2 mplane on devices where it's supported (where it's more efficient). My camera doesn't support mplane.
+
+Implement v4l2 yuyv nvidia capture by capturing rg88 yuyv to rgb as is done now for screenshot, but also do that for video by creating an intermediate rgb texture for the camera. Then render that rgb texture to the video texture.
+    This is needed to properly scale the yuyv texture without messing it up (the texture indexing).
+
+Set v4l2 camera fps to video output fps (if lower than the camera fps) and the same for resolution.
+
+Support camera controls, such as white balance. Otherwise tell user to use cameractrl software.
+
+Camera capture doesn't work perfectly. The image gets glitched, need to properly wait for image to be done.
+
+Use one pipewire connection (pipewire video) instead of multiple ones when recording with portal multiple times (multiple sources).
+
+Close pipewire links or maybe there are file descriptor leaks?
+
+Make multiple capture sources work properly in regards to size. The size of the video should be the region size of each capture source.
+
+Support hdr camera capture.
+
+Return the max resolution of each codec in --info to display an error in the UI before capture starts. Right now its fine since the UI will report bad resolution after capture starts and fails but it doesn't say what the max resolution is.
+
+Should -low-power option also use vaapi/vulkan low power, if available?
+
+Should capture option x=bla;y=bla be scaled by -s (output resolution scale)? width and height is.
+
+Add option to capture all monitors automatically.
+
+Make -w optional, to only capture audio.
+
+Use GL_RGBA16F or GL_RGBA32F for hdr, that allows color values to be outside the 0.0 to 1.0 range.
+
+https://registry.khronos.org/EGL/extensions/EXT/EGL_EXT_surface_SMPTE2086_metadata.txt
+
+Doesn't work: sibs run --args -w "/dev/video0;camera_width=800;camera_height=600;pixfmt=yuyv" -fm content -o video.mp4

extra/gpu-screen-recorder.service

@@ -3,9 +3,10 @@ Description=GPU Screen Recorder Service
 
 [Service]
 EnvironmentFile=-%h/.config/gpu-screen-recorder.env
+EnvironmentFile=-%h/.config/gpu-screen-recorder/gpu-screen-recorder.env
 Environment=WINDOW=screen
 Environment=CONTAINER=mp4
-Environment=QUALITY=50000
+Environment=QUALITY=40000
 Environment=BITRATE_MODE=cbr
 Environment=CODEC=auto
 Environment=AUDIO_CODEC=opus
@@ -27,4 +28,4 @@ Restart=on-failure
 RestartSec=5s
 
 [Install]
-WantedBy=default.target
+WantedBy=default.target

extra/meson_post_install.sh

@@ -3,15 +3,3 @@
 # Needed to remove password prompt when recording a monitor (without desktop portal option) on amd/intel or nvidia wayland
 /usr/sbin/setcap cap_sys_admin+ep ${MESON_INSTALL_DESTDIR_PREFIX}/bin/gsr-kms-server \
     || echo "\n!!! Please re-run install as root\n"
-
-# Cant do this because it breaks desktop portal (create session)!!!.
-# For some reason the desktop portal tries to access /proc/gpu-screen-recorder-pid/root from the portal process
-# which doesn't work because for some reason CAP_SYS_NICE on a program makes /proc/self/root not readable by other processes.
-# The reason portal reads that file might be because portal seems to have a security feature where its able to identify the
-# process and if the session token is stolen by another application then it will ignore the session token as it wasn't that
-# application that created the session token.
-# ---
-# This is needed (for EGL_CONTEXT_PRIORITY_HIGH_IMG) to allow gpu screen recorder to run faster than the heaviest application on AMD.
-# For example when trying to record a game at 60 fps and the game drops to 45 fps in some place that would also make gpu screen recorder
-# drop to 45 fps unless this setcap is used.
-#/usr/sbin/setcap cap_sys_nice+ep ${MESON_INSTALL_DESTDIR_PREFIX}/bin/gpu-screen-recorder

gpu-screen-recorder.1

@@ -0,0 +1,621 @@
+.TH GPU-SCREEN-RECORDER 1 "2025-12-22" "5.10.2" "GPU Screen Recorder Manual"
+.SH NAME
+gpu-screen-recorder \- The fastest screen recording tool for Linux
+.SH SYNOPSIS
+.B gpu-screen-recorder
+.B \-w
+.I window_id|monitor|focused|portal|region|v4l2_device_path
+.RI [ options ]
+.B \-o
+.I output_file
+.PP
+.B gpu-screen-recorder
+.B \-\-help
+|
+.B \-\-version
+|
+.B \-\-info
+|
+.B \-\-list\-capture\-options
+|
+.B \-\-list\-monitors
+|
+.B \-\-list\-v4l2\-devices
+|
+.B \-\-list\-audio\-devices
+|
+.B \-\-list\-application\-audio
+.SH DESCRIPTION
+.B gpu-screen-recorder
+is the fastest screen recording tool for Linux. It uses the GPU
+only to record the screen to minimize system performance impact. It supports recording,
+live streaming, instant replay (similar to NVIDIA ShadowPlay), and
+screenshot capture. The tool works on both X11 and Wayland with AMD,
+Intel, and NVIDIA graphics cards.
+.PP
+Key features include:
+.RS
+.IP \(bu 3
+GPU-accelerated encoding using H264, HEVC, AV1, VP8, or VP9 codecs
+.IP \(bu 3
+Support for HDR recording (HEVC HDR and AV1 HDR)
+.IP \(bu 3
+Instant replay buffer (last N seconds recording)
+.IP \(bu 3
+Live streaming capabilities
+.IP \(bu 3
+Audio recording from devices and applications (PipeWire required for app audio)
+.IP \(bu 3
+Plugin system for custom graphics overlay
+.IP \(bu 3
+Minimal performance impact compared to traditional screen recorders
+.RE
+.SH OPTIONS
+.SS Capture Options
+.TP
+.BI \-w " window_id|monitor|focused|portal|region|v4l2_device_path"
+Specify what to record.
+Formats:
+.RS
+.IP \(bu 3
+.B window
+- Record a specific window (use window ID) (X11 only)
+.IP \(bu 3
+.B screen
+- Record the first monitor found
+.IP \(bu 3
+.B screen\-direct
+- NVIDIA X11 only, for VRR/G-SYNC support (not recommended due to driver issues)
+.IP \(bu 3
+.B focused
+- Record the currently focused window (X11 only) (use with -s option)
+.IP \(bu 3
+.B portal
+- Use xdg-desktop-portal with PipeWire (Wayland only)
+.IP \(bu 3
+.B region
+- Record a specific region (use with
+.B \-region
+option)
+.IP \(bu 3
+Monitor name (e.g.,
+.BR DP\-1 )
+- Record specific monitor
+.IP \(bu 3
+V4L2 device path (e.g.,
+.BR /dev/video0 )
+- Record camera device (V4L2).
+
+Other applications can't use the camera when GPU Screen Recorder is using the camera and GPU Screen Recorder may not be able to use the camera if another application is already using the camera
+.IP \(bu 3
+Combine sources with | (e.g.,
+.BR "monitor:screen|v4l2:/dev/video0" )
+.RE
+.PP
+Run
+.B \-\-list\-capture\-options
+to list available capture sources.
+.PP
+Additional options can be passed to each capture source by splitting capture source with
+.B ;
+for example
+.BR "screen;x=50;y=50".
+.br
+These are the available options for all capture sources (optional):
+.RS
+.IP \(bu 3
+.B x
+- The X position in pixels. If the number ends with % then this sets the X position relative to the video width (integer percentage where 100 = 100%)
+.IP \(bu 3
+.B y
+- The Y position in pixels. If the number ends with % then this sets the Y position relative to the video height (integer percentage where 100 = 100%)
+.IP \(bu 3
+.B width
+- The width in pixels. If the number ends with % then this sets the width relative to the video width (integer percentage where 100 = 100%).
+
+A value of 0 means to not scale the capture source and instead use the original width.
+.IP \(bu 3
+.B height
+- The height in pixels. If the number ends with % then this sets the height relative to the video height (integer percentage where 100 = 100%).
+
+A value of 0 means to not scale the capture source and instead use the original height.
+.IP \(bu 3
+.B halign
+- The horizontal alignment, should be either
+.BR "start",
+.B center
+or
+.BR "end".
+
+Set to
+.B center
+by default when using one capture source, otherwise it's set to
+.B start
+by default.
+.IP \(bu 3
+.B valign
+- The vertical alignment, should be either
+.BR "start",
+.B center
+or
+.BR "end".
+
+Set to
+.B center
+by default when using one capture source, otherwise it's set to
+.B start
+by default.
+.IP \(bu 3
+.B hflip
+- If the source should be flipped horizontally, should be either
+.B "true"
+or
+.BR "false".
+Set to
+.B false
+by default
+.IP \(bu 3
+.B vflip
+- If the source should be flipped vertically, should be either
+.B "true"
+or
+.BR "false".
+Set to
+.B false
+by default
+.RE
+.PP
+These are the additional options available for camera (V4L2) sources (optional):
+.RS
+.IP \(bu 3
+.B pixfmt
+- The pixel format, should be either
+.BR "auto",
+.B yuyv
+or
+.BR "mjpeg".
+Set to
+.B auto
+by default
+.IP \(bu 3
+.B camera_fps
+- The camera fps. Has to match a camera fps returned in
+.B --info
+or
+.B --list-v4l2-devices
+for the device.
+
+A value of 0 means to use the best option available.
+.IP \(bu 3
+.B camera_width
+- The camera width in pixels. Has to match a camera width returned in
+.B --info
+or
+.B --list-v4l2-devices
+for the device.
+
+A value of 0 means to use the best option available.
+.IP \(bu 3
+.B camera_height
+- The camera height in pixels. Has to match a camera height returned in
+.B --info
+or
+.B --list-v4l2-devices
+for the device.
+
+A value of 0 means to use the best option available.
+.RE
+.TP
+.BI \-region " WxH+X+Y"
+Specify region to capture when using
+.BR \-w " region."
+Format is width x height + X offset + Y offset. Set width and height to 0 to capture the whole monitor that contains the position.
+
+It is compatible with tools such as slop (X11) and slurp (Wayland).
+.TP
+.BI \-c " container_format"
+Container format (mp4, mkv, flv, webm). Defaults to extension from
+.B \-o
+filename.
+.TP
+.BI \-s " WxH"
+Output resolution limit (e.g., 1920x1080). Use 0x0 for original resolution.
+Required for
+.BR \-w " focused."
+.TP
+.BI \-f " fps"
+Frame rate (default: 60). For variable frame rate mode, this is maximum FPS.
+.TP
+.BI \-cursor " yes|no"
+Record cursor (default: yes).
+.TP
+.BI \-restore\-portal\-session " yes|no"
+Restore previous portal session (used with -w portal option, default: no).
+.SS Audio Options
+.TP
+.BI \-a " audio_source"
+Audio device or application to record. Can be specified multiple times.
+Formats:
+.RS
+.IP \(bu 3
+.B default_output
+- Default output device
+.IP \(bu 3
+.B default_input
+- Default input device
+.IP \(bu 3
+.B device:name
+- Specific device
+.IP \(bu 3
+.B app:name
+- Application audio (case-insensitive)
+.IP \(bu 3
+.B app\-inverse:name
+- All applications except specified
+.IP \(bu 3
+Combine sources with | (e.g.,
+.BR "default_output|app:firefox" )
+.RE
+.PP
+Run
+.B \-\-list\-audio\-devices
+to list available audio devices.
+.br
+Run
+.B \-\-list\-application\-audio
+to list available applications for audio capture.
+.TP
+.BI \-ac " aac|opus|flac"
+Audio codec (default: opus for mp4/mkv, aac otherwise). FLAC temporarily disabled.
+.TP
+.BI \-ab " bitrate"
+Audio bitrate in kbps (default: 128 for opus/flac, 160 for aac). 0 = automatic.
+.SS Video Quality Options
+.TP
+.BI \-k " codec"
+Video codec:
+.BR auto ", " h264 ", " hevc ", " av1 ", " vp8 ", " vp9 ", " hevc_hdr ", " av1_hdr ", " hevc_10bit ", " av1_10bit
+(default: auto → h264). HDR options not available on X11 or portal capture.
+
+10-bit capture reduces banding but may not be supported properly by all video players.
+.TP
+.BI \-q " quality"
+Quality preset (medium, high, very_high, ultra) for QP/VBR mode, or bitrate (kbps) for CBR mode (default: very_high).
+.TP
+.BI \-bm " auto|qp|vbr|cbr"
+Bitrate mode (default: auto → qp). CBR recommended for replay buffer and live streaming.
+
+QP means to capture with constant quality, even in motion, while VBR and CBR means to capture with constant size.
+.TP
+.BI \-fm " cfr|vfr|content"
+Frame rate mode: cfr (constant), vfr (variable), or content (match content) (default: vfr). Content mode is only available on X11 or portal.
+
+Content mode syncs video to the captured content and is recommended for smoothest video when the game is running
+at the same frame rate or lower than what you are trying to record at.
+.TP
+.BI \-cr " limited|full"
+Color range (default: limited). Full range may cause issues with some video players/websites.
+.TP
+.BI \-tune " performance|quality"
+Encoding tune (default: performance). Currently NVIDIA only.
+.TP
+.BI \-keyint " seconds"
+Keyframe interval in seconds (default: 2.0). Affects seeking precision.
+.TP
+.BI \-encoder " gpu|cpu"
+Encoding device (default: gpu). CPU encoding only for H264.
+.TP
+.BI \-fallback\-cpu\-encoding " yes|no"
+Fallback to CPU if GPU encoding unavailable (default: no).
+.SS Replay Buffer Options
+.TP
+.BI \-r " seconds"
+Replay buffer duration (2-86400 seconds). When set, only saves on command.
+.TP
+.BI \-replay\-storage " ram|disk"
+Buffer storage location (default: ram). Disk mode may reduce SSD lifespan.
+.TP
+.BI \-restart\-replay\-on\-save " yes|no"
+Clear buffer after saving replay (default: no).
+.TP
+.BI \-df " yes|no"
+Organize replays in date-based folders (default: no).
+.SS Advanced Options
+.TP
+.BI \-oc " yes|no"
+Overclock NVIDIA memory (X11 only). Requires Coolbits=12. Use at own risk.
+.TP
+.BI \-p " plugin.so"
+Load plugin (.so file). Can be specified multiple times.
+.TP
+.BI \-sc " script_path"
+Script to run after saving video. Receives filepath and type ("regular", "replay", "screenshot").
+.TP
+.BI \-portal\-session\-token\-filepath " path"
+Portal session token file (default: ~/.config/gpu-screen-recorder/restore_token).
+.TP
+.BI \-ffmpeg-opts " options"
+Additional arguments to pass to FFmpeg for the file in a list of key-values pairs in the format "key=value;key=value",
+.br
+for example: -ffmpeg-opts "hls_list_size=3;hls_time=1;hls_flags=delete_segments".
+.br
+Note: this overwrites options set by GPU Screen Recorder with the same name.
+.TP
+.BI \-ffmpeg-video-opts " options"
+Additional arguments to pass to FFmpeg for the video in a list of key-values pairs in the format "key=value;key=value",
+.br
+for example: -ffmpeg-video-opts "codec=cabac;rc_mode=CQP;qp=16".
+.br
+Note: this overwrites options set by GPU Screen Recorder with the same name.
+.TP
+.BI \-ffmpeg-audio-opts " options"
+Additional arguments to pass to FFmpeg for the audio in a list of key-values pairs in the format "key=value;key=value",
+.br
+for example: -ffmpeg-audio-opts "aac_coder=fast;aac_pce=true".
+.br
+Note: this overwrites options set by GPU Screen Recorder with the same name.
+.TP
+.BI \-gl\-debug " yes|no"
+OpenGL debug output (default: no).
+.TP
+.BI \-v " yes|no"
+Print FPS and damage info (default: yes).
+.TP
+.BI \-low-power " yes|no"
+Run in low power mode. This currently has only an affect on AMD (as it's only an issue on AMD) and allows the GPU to go into a lower power mode when recording (default: no).
+.br
+Setting this to
+.B yes
+might not always be ideal because of AMD driver issues where after playing a video with VAAPI on the system the video encoding performance
+also reduces, which affects GPU Screen Recorder.
+.br
+It's recommended to also use the option
+.B -fm content
+when this is set to
+.B yes
+to only encode frames when the screen content updates to lower GPU and video encoding usage when the system is idle.
+.TP
+.BI \-write\-first\-frame\-ts " yes|no"
+When enabled, writes a timestamp file with extra extension \fI.ts\fR next to the output video containing:
+.nf
+monotonic_microsec	realtime_microsec
+<monotonic_microsec>	<realtime_microsec>
+.fi
+(default: no). Ignored for live streaming and when output is piped.
+.SS Output Options
+.TP
+.BI \-o " output"
+Output file path (or directory for replay mode). Required except when outputting to stdout.
+.TP
+.BI \-ro " directory"
+Output directory for regular recordings during replay/streaming mode.
+.SS Information Commands
+.TP
+.B \-\-help
+Show help message.
+.TP
+.B \-\-version
+Show version (5.10.2).
+.TP
+.B \-\-info
+Show system info (codecs, capture options).
+.TP
+.B \-\-list\-capture\-options
+List available capture sources (window, monitors, portal, v4l2 device path).
+.TP
+.B \-\-list\-monitors
+List available monitors.
+.TP
+.B \-\-list\-v4l2\-devices
+List available cameras devices (V4L2).
+.TP
+.B \-\-list\-audio\-devices
+List available audio devices.
+.TP
+.B \-\-list\-application\-audio
+List available applications for audio capture.
+.SH SIGNALS
+GPU Screen Recorder can be controlled via signals:
+.TP
+.B SIGINT (Ctrl+C)
+Stop and save recording (stop without save in replay mode).
+.TP
+.B SIGUSR1
+Save replay (replay mode only).
+.TP
+.B SIGUSR2
+Pause/unpause recording (not for streaming/replay).
+.TP
+.B SIGRTMIN
+Start/stop regular recording during replay/streaming.
+.TP
+.B SIGRTMIN+1
+Save last 10 seconds (replay mode).
+.TP
+.B SIGRTMIN+2
+Save last 30 seconds (replay mode).
+.TP
+.B SIGRTMIN+3
+Save last 60 seconds (replay mode).
+.TP
+.B SIGRTMIN+4
+Save last 5 minutes (replay mode).
+.TP
+.B SIGRTMIN+5
+Save last 10 minutes (replay mode).
+.TP
+.B SIGRTMIN+6
+Save last 30 minutes (replay mode).
+.PP
+Use
+.B pkill
+to send signals (e.g.,
+.BR "pkill -SIGUSR1 -f ""^gpu-screen-recorder""" ).
+.SH EXAMPLES
+Record monitor at 60 FPS with desktop audio:
+.PP
+.nf
+.RS
+gpu-screen-recorder -w screen -f 60 -a default_output -o video.mp4
+.RE
+.fi
+.PP
+Record monitor at 60 FPS with desktop audio and microphone:
+.PP
+.nf
+.RS
+gpu-screen-recorder -w screen -f 60 -a "default_output|default_input" -o video.mp4
+.RE
+.fi
+.PP
+Record specific window (X11 only):
+.PP
+.nf
+.RS
+gpu-screen-recorder -w $(xdotool selectwindow) -f 60 -o video.mp4
+.RE
+.fi
+.PP
+Instant replay (last 60 seconds):
+.PP
+.nf
+.RS
+gpu-screen-recorder -w screen -c mkv -r 60 -o ~/Videos
+.RE
+.fi
+.PP
+Record region using slop:
+.PP
+.nf
+.RS
+gpu-screen-recorder -w $(slop) -o video.mp4
+.RE
+.fi
+.PP
+Record region using slurp:
+.PP
+.nf
+.RS
+gpu-screen-recorder -w $(slurp -f "%wx%h+%x+%y") -o video.mp4
+.RE
+.fi
+.PP
+Instant replay and launch a script when saving replay:
+.PP
+.nf
+.RS
+gpu-screen-recorder -w screen -c mkv -r 60 -sc ./script.sh -o ~/Videos
+.RE
+.fi
+.PP
+Stream to Twitch with constant bitrate mode:
+.PP
+.nf
+.RS
+gpu-screen-recorder -w screen -c flv -a default_output -bm cbr -q 8000 -o "rtmp://live.twitch.tv/app/stream_key"
+.RE
+.fi
+.PP
+Take screenshot:
+.PP
+.nf
+.RS
+gpu-screen-recorder -w screen -o screenshot.jpg
+.PP
+.RE
+.fi
+Record screen and camera:
+.PP
+.nf
+.RS
+gpu-screen-recorder -w "screen|/dev/video0" -o video.mp4
+.PP
+.RE
+.fi
+Record screen and camera. The camera is located at the bottom right and flipped horizontally:
+.PP
+.nf
+.RS
+gpu-screen-recorder -w "screen|/dev/video0;halign=end;valign=end;hflip=true;width=30%;height=30%" -o video.mp4
+.PP
+.RE
+.fi
+Record two monitors, side by side (assuming the first monitor has a resolution of 1920x1080)
+.PP
+.nf
+.RS
+gpu-screen-recorder -w "DP-1|DP-2;x=1920" -o video.mp4
+.RE
+.fi
+.SH FILES
+.TP
+.I ~/.config/gpu-screen-recorder/gpu-screen-recorder.env
+Environment variables for systemd service (optional).
+.TP
+.I /usr/lib/modprobe.d/gsr-nvidia.conf
+NVIDIA suspend/resume fix.
+.TP
+.I ~/.config/gpu-screen-recorder/restore_token
+Portal session token storage (default location).
+.SH TIPS
+.IP \(bu 3
+Some video editors don't support videos in variable frame rate mode.
+
+Change video container to .mkv or capture in constant frame rate mode instead to workaround this issue
+.IP \(bu 3
+Variable frame rate may cause issues in some video players (use MPV)
+.IP \(bu 3
+H.265/HEVC not supported in browsers/Discord (use H264 for sharing)
+.IP \(bu 3
+Some distros disable hardware codecs (use Flatpak version)
+.IP \(bu 3
+Use CBR mode (
+.B \-bm cbr
+) for replay buffer to control RAM usage
+.SH PERFORMANCE TIPS
+.IP \(bu 3
+Save to SSD (HDD may be too slow)
+.IP \(bu 3
+Close other screen recorders (including idle OBS)
+.SH KNOWN ISSUES
+.IP \(bu 3
+NVIDIA: CUDA breaks after suspend (install gsr-nvidia.conf fix)
+.IP \(bu 3
+AMD: Possible black bars colors with HEVC/AV1 (use H264 or FFmpeg >=8)
+.SH SEE ALSO
+.UR https://git.dec05eba.com/gpu-screen-recorder
+Project homepage
+.UE
+.br
+.UR https://git.dec05eba.com/gpu-screen-recorder/about
+Project readme, with more information
+.UE
+.br
+.UR https://git.dec05eba.com/gpu-screen-recorder-gtk
+GTK GUI version (deprecated)
+.UE
+.br
+.UR https://git.dec05eba.com/gpu-screen-recorder-ui
+ShadowPlay-style UI
+.UE
+.br
+.BR xdotool (1),
+.BR slop (1),
+.BR slurp (1),
+.SH AUTHORS
+Developed by dec05eba and contributors.
+.SH COPYRIGHT
+Copyright © dec05eba. Licensed under GPL3-only.
+.SH REPORTING BUGS
+Report bugs at
+.UR mailto:dec05eba@protonmail.com
+dec05eba@protonmail.com
+.UE .
+.UR https://git.dec05eba.com/?p=about
+See more information about reporting bugs at the gpu-screen-recorder website
+.UE .
+.br
+.UR https://git.dec05eba.com/gpu-screen-recorder/about/
+Before reporting a bug or an issue, please take a look at FAQ part of the README
+.UE .
+The bug or issue may have been previously reported or may not be related to gpu-screen-recorder.

gsr-kms-server.1

@@ -0,0 +1,65 @@
+.TH GSR\-KMS\-SERVER 1 "2025-12-22" "5.10.2" "GPU Screen Recorder KMS Server Manual"
+.SH NAME
+gsr\-kms\-server \- KMS (Kernel Mode Setting) server for GPU Screen Recorder
+.SH SYNOPSIS
+.B gsr\-kms\-server
+<\fIdomain_socket_path\fR> <\fIcard_path\fR>
+.SH DESCRIPTION
+\fBgsr\-kms\-server\fR is a companion daemon for GPU Screen Recorder that provides
+Kernel Mode Setting (KMS) access for screen capture in environments where
+direct DRM access is required. It acts as a local server that communicates
+via a Unix domain socket, allowing GPU Screen Recorder to capture screen
+content through the KMS interface.
+.PP
+This component is typically used when GPU Screen Recorder needs to access
+display hardware directly, bypassing the windowing system for more efficient
+or lower-level capture. It runs as a separate process and communicates with
+the main GPU Screen Recorder instance.
+.SH ARGUMENTS
+.TP
+.B \fIdomain_socket_path\fR
+Path to the Unix domain socket that the server will create and listen on.
+This socket is used for communication between GPU Screen Recorder and the
+KMS server. The path should be in a location where both processes have
+appropriate permissions to read and write.
+.TP
+.B \fIcard_path\fR
+Path to the DRM device file (typically \fI/dev/dri/cardX\fR) that the server
+will use for KMS operations. This corresponds to a specific graphics card
+in the system.
+.SH EXAMPLES
+Start the KMS server using card0 and create socket at /tmp/gsr-kms.sock:
+.RS
+.B gsr\-kms\-server /tmp/gsr\-kms.sock /dev/dri/card0
+.RE
+.PP
+Start the KMS server using card1 and create socket in a user-specific location:
+.RS
+.B gsr\-kms\-server ~/.cache/gsr\-kms.sock /dev/dri/card1
+.RE
+.SH NOTES
+.IP \(bu 3
+The \fBgsr-kms-server\fR typically runs with elevated permissions to access
+DRM devices, which requires root permission.
+.IP \(bu 3
+Only one instance should run per DRM card at a time.
+.IP \(bu 3
+The server will exit when the client disconnects or when terminated by a signal.
+.IP \(bu 3
+The socket file is created by the server and removed when the server exits.
+.SH FILES
+.TP
+.B /dev/dri/card*
+DRM device files for graphics cards
+.TP
+.B /tmp/gsr-kms.sock
+Typical location for the domain socket (user-configurable)
+.SH SEE ALSO
+.BR gpu\-screen\-recorder (1),
+.BR drm (4)
+.PP
+Project homepage: <https://git.dec05eba.com/gpu-screen-recorder/about/>
+.SH BUGS
+Report bugs at the project homepage.
+.SH AUTHORS
+gsr\-kms\-server was written by the GPU Screen Recorder contributors.

include/args_parser.h

@@ -0,0 +1,124 @@
+#ifndef GSR_ARGS_PARSER_H
+#define GSR_ARGS_PARSER_H
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "defs.h"
+#include "vec2.h"
+
+typedef struct gsr_egl gsr_egl;
+
+#define NUM_ARGS 37
+
+typedef enum {
+    GSR_CAPTURE_SOURCE_TYPE_WINDOW,
+    GSR_CAPTURE_SOURCE_TYPE_FOCUSED_WINDOW,
+    GSR_CAPTURE_SOURCE_TYPE_MONITOR,
+    GSR_CAPTURE_SOURCE_TYPE_REGION,
+    GSR_CAPTURE_SOURCE_TYPE_PORTAL,
+    GSR_CAPTURE_SOURCE_TYPE_V4L2
+} CaptureSourceType;
+
+typedef enum {
+    ARG_TYPE_STRING,
+    ARG_TYPE_BOOLEAN,
+    ARG_TYPE_ENUM,
+    ARG_TYPE_I64,
+    ARG_TYPE_DOUBLE,
+} ArgType;
+
+typedef struct {
+    const char *name;
+    int value;
+} ArgEnum;
+
+typedef struct {
+    ArgType type;
+    const char **values;
+    int capacity_num_values;
+    int num_values;
+
+    const char *key;
+    bool optional;
+    bool list;
+
+    const ArgEnum *enum_values;
+    int num_enum_values;
+
+    int64_t integer_value_min;
+    int64_t integer_value_max;
+
+    union {
+        bool boolean;
+        int enum_value;
+        int64_t i64_value;
+        double d_value;
+    } typed_value;
+} Arg;
+
+typedef struct {
+    void (*version)(void *userdata);
+    void (*info)(void *userdata);
+    void (*list_audio_devices)(void *userdata);
+    void (*list_application_audio)(void *userdata);
+    void (*list_v4l2_devices)(void *userdata);
+    void (*list_capture_options)(const char *card_path, void *userdata);
+    void (*list_monitors)(void *userdata);
+} args_handlers;
+
+typedef struct {
+    Arg args[NUM_ARGS];
+
+    gsr_video_encoder_hardware video_encoder;
+    gsr_pixel_format pixel_format;
+    gsr_framerate_mode framerate_mode;
+    gsr_color_range color_range;
+    gsr_tune tune;
+    gsr_video_codec video_codec;
+    gsr_audio_codec audio_codec;
+    gsr_bitrate_mode bitrate_mode;
+    gsr_video_quality video_quality;
+    gsr_replay_storage replay_storage;
+
+    const char *capture_source;
+    const char *container_format;
+    const char *filename;
+    const char *replay_recording_directory;
+    const char *portal_session_token_filepath;
+    const char *recording_saved_script;
+    const char *ffmpeg_opts;
+    const char *ffmpeg_video_opts;
+    const char *ffmpeg_audio_opts;
+    bool verbose;
+    bool gl_debug;
+    bool fallback_cpu_encoding;
+    bool low_power;
+    bool record_cursor;
+    bool date_folders;
+    bool restore_portal_session;
+    bool restart_replay_on_save;
+    bool overclock;
+    bool write_first_frame_ts;
+    bool is_livestream;
+    bool is_output_piped;
+    bool low_latency_recording;
+    bool very_old_gpu;
+    int64_t video_bitrate;
+    int64_t audio_bitrate;
+    int64_t fps;
+    int64_t replay_buffer_size_secs;
+    double keyint;
+    vec2i output_resolution;
+    vec2i region_size;
+    vec2i region_position;
+} args_parser;
+
+/* |argv| is stored as a reference */
+bool args_parser_parse(args_parser *self, int argc, char **argv, const args_handlers *args_handlers, void *userdata);
+void args_parser_deinit(args_parser *self);
+
+bool args_parser_validate_with_gl_info(args_parser *self, gsr_egl *egl);
+void args_parser_print_usage(void);
+Arg* args_parser_get_arg(args_parser *self, const char *arg_name);
+
+#endif /* GSR_ARGS_PARSER_H */

include/capture/capture.h

@@ -12,22 +12,35 @@ typedef struct AVFrame AVFrame;
 typedef struct AVMasteringDisplayMetadata AVMasteringDisplayMetadata;
 typedef struct AVContentLightMetadata AVContentLightMetadata;
 typedef struct gsr_capture gsr_capture;
+typedef struct gsr_capture_metadata gsr_capture_metadata;
 
-typedef struct {
-    int width;
-    int height;
+typedef enum {
+    GSR_CAPTURE_ALIGN_START,
+    GSR_CAPTURE_ALIGN_CENTER,
+    GSR_CAPTURE_ALIGN_END
+} gsr_capture_alignment;
+
+struct gsr_capture_metadata {
+    // Size of the video
+    vec2i video_size;
+    // The captured output gets scaled to this size. By default this will be the same size as the captured target
+    vec2i recording_size;
+    vec2i position;
     int fps;
-    AVCodecContext *video_codec_context; /* can be NULL */
-    AVFrame *frame; /* can be NULL, but will never be NULL if |video_codec_context| is set */
-} gsr_capture_metadata;
+    gsr_capture_alignment halign;
+    gsr_capture_alignment valign;
+    gsr_flip flip;
+};
 
 struct gsr_capture {
-    /* These methods should not be called manually. Call gsr_capture_* instead */
+    /* These methods should not be called manually. Call gsr_capture_* instead. |capture_metadata->video_size| should be set by this function */
     int (*start)(gsr_capture *cap, gsr_capture_metadata *capture_metadata);
     void (*on_event)(gsr_capture *cap, gsr_egl *egl); /* can be NULL */
     void (*tick)(gsr_capture *cap); /* can be NULL. If there is an event then |on_event| is called before this */
     bool (*should_stop)(gsr_capture *cap, bool *err); /* can be NULL. If NULL, return false */
-    int (*capture)(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion);
+    bool (*capture_has_synchronous_task)(gsr_capture *cap); /* can be NULL. If this returns true then the time spent in |capture| is ignored for video/audio (capture is paused while the synchronous task happens) */
+    void (*pre_capture)(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion); /* can be NULL */
+    int (*capture)(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion); /* Return 0 if the frame was captured */
     bool (*uses_external_image)(gsr_capture *cap); /* can be NULL. If NULL, return false */
     bool (*set_hdr_metadata)(gsr_capture *cap, AVMasteringDisplayMetadata *mastering_display_metadata, AVContentLightMetadata *light_metadata); /* can be NULL. If NULL, return false */
     uint64_t (*get_window_id)(gsr_capture *cap); /* can be NULL. Returns 0 if unknown */

include/capture/kms.h

@@ -2,9 +2,13 @@
 #define GSR_CAPTURE_KMS_H
 
 #include "capture.h"
+#include "../cursor.h"
+#include "../../kms/kms_shared.h"
 
 typedef struct {
     gsr_egl *egl;
+    gsr_cursor *x11_cursor;
+    gsr_kms_response *kms_response;
     const char *display_to_capture; /* A copy is made of this */
     bool hdr;
     bool record_cursor;

include/capture/v4l2.h

@@ -0,0 +1,44 @@
+#ifndef GSR_CAPTURE_V4L2_H
+#define GSR_CAPTURE_V4L2_H
+
+#include "capture.h"
+
+typedef enum {
+    GSR_CAPTURE_V4L2_PIXFMT_AUTO,
+    GSR_CAPTURE_V4L2_PIXFMT_YUYV,
+    GSR_CAPTURE_V4L2_PIXFMT_MJPEG
+} gsr_capture_v4l2_pixfmt;
+
+typedef struct {
+    uint32_t width;
+    uint32_t height;
+} gsr_capture_v4l2_resolution;
+
+typedef struct {
+    uint32_t denominator;
+    uint32_t numerator;
+} gsr_capture_v4l2_framerate;
+
+typedef struct {
+    gsr_capture_v4l2_pixfmt pixfmt;
+    gsr_capture_v4l2_resolution resolution;
+    gsr_capture_v4l2_framerate framerate;
+} gsr_capture_v4l2_supported_setup;
+
+typedef struct {
+    gsr_egl *egl;
+    vec2i output_resolution;
+    const char *device_path;
+    gsr_capture_v4l2_pixfmt pixfmt;
+    uint32_t camera_fps; /* Set to 0 if the best option should be chosen */
+    gsr_capture_v4l2_resolution camera_resolution; /* Set to 0, 0 if the best option should be chosen */
+} gsr_capture_v4l2_params;
+
+gsr_capture* gsr_capture_v4l2_create(const gsr_capture_v4l2_params *params);
+
+const char* gsr_capture_v4l2_pixfmt_to_string(gsr_capture_v4l2_pixfmt pixfmt);
+uint32_t gsr_capture_v4l2_framerate_to_number(gsr_capture_v4l2_framerate framerate);
+typedef void (*v4l2_devices_query_callback)(const char *path, const gsr_capture_v4l2_supported_setup *supported_setup, void *userdata);
+void gsr_capture_v4l2_list_devices(v4l2_devices_query_callback callback, void *userdata);
+
+#endif /* GSR_CAPTURE_V4L2_H */

include/capture/xcomposite.h

@@ -3,9 +3,11 @@
 
 #include "capture.h"
 #include "../vec2.h"
+#include "../cursor.h"
 
 typedef struct {
     gsr_egl *egl;
+    gsr_cursor *cursor;
     unsigned long window;
     bool follow_focused; /* If this is set then |window| is ignored */
     bool record_cursor;

include/capture/ximage.h

@@ -3,9 +3,11 @@
 
 #include "capture.h"
 #include "../vec2.h"
+#include "../cursor.h"
 
 typedef struct {
     gsr_egl *egl;
+    gsr_cursor *cursor;
     const char *display_to_capture; /* A copy is made of this */
     bool record_cursor;
     vec2i output_resolution;

include/codec_query/codec_query.h

@@ -2,10 +2,12 @@
 #define GSR_CODEC_QUERY_H
 
 #include <stdbool.h>
+#include "../vec2.h"
 
 typedef struct {
     bool supported;
     bool low_power;
+    vec2i max_resolution;
 } gsr_supported_video_codec;
 
 typedef struct {

include/color_conversion.h

@@ -2,22 +2,17 @@
 #define GSR_COLOR_CONVERSION_H
 
 #include "shader.h"
+#include "defs.h"
 #include "vec2.h"
 #include <stdbool.h>
 
-typedef enum {
-    GSR_COLOR_RANGE_LIMITED,
-    GSR_COLOR_RANGE_FULL
-} gsr_color_range;
-
-typedef enum {
-    GSR_COLOR_DEPTH_8_BITS,
-    GSR_COLOR_DEPTH_10_BITS
-} gsr_color_depth;
+#define GSR_COLOR_CONVERSION_MAX_GRAPHICS_SHADERS 12
+#define GSR_COLOR_CONVERSION_MAX_FRAMEBUFFERS 2
 
 typedef enum {
     GSR_SOURCE_COLOR_RGB,
-    GSR_SOURCE_COLOR_BGR
+    GSR_SOURCE_COLOR_BGR,
+    GSR_SOURCE_COLOR_YUYV
 } gsr_source_color;
 
 typedef enum {
@@ -26,10 +21,23 @@ typedef enum {
     GSR_DESTINATION_COLOR_RGB8
 } gsr_destination_color;
 
+typedef enum {
+    GSR_ROT_0,
+    GSR_ROT_90,
+    GSR_ROT_180,
+    GSR_ROT_270
+} gsr_rotation;
+
+typedef enum {
+    GSR_FLIP_NONE       = 0,
+    GSR_FLIP_HORIZONTAL = (1 << 0),
+    GSR_FLIP_VERTICAL   = (1 << 1)
+} gsr_flip;
+
 typedef struct {
+    int rotation_matrix;
     int offset;
-    int rotation;
-} gsr_color_uniforms;
+} gsr_color_graphics_uniforms;
 
 typedef struct {
     gsr_egl *egl;
@@ -37,6 +45,7 @@ typedef struct {
     gsr_destination_color destination_color;
 
     unsigned int destination_textures[2];
+    vec2i destination_textures_size[2];
     int num_destination_textures;
 
     gsr_color_range color_range;
@@ -45,19 +54,25 @@ typedef struct {
 
 typedef struct {
     gsr_color_conversion_params params;
-    gsr_color_uniforms uniforms[4];
-    gsr_shader shaders[4];
 
-    unsigned int framebuffers[2];
+    gsr_color_graphics_uniforms graphics_uniforms[GSR_COLOR_CONVERSION_MAX_GRAPHICS_SHADERS];
+    gsr_shader graphics_shaders[GSR_COLOR_CONVERSION_MAX_GRAPHICS_SHADERS];
+
+    unsigned int framebuffers[GSR_COLOR_CONVERSION_MAX_FRAMEBUFFERS];
 
     unsigned int vertex_array_object_id;
     unsigned int vertex_buffer_object_id;
+
+    bool schedule_clear;
 } gsr_color_conversion;
 
 int gsr_color_conversion_init(gsr_color_conversion *self, const gsr_color_conversion_params *params);
 void gsr_color_conversion_deinit(gsr_color_conversion *self);
 
-void gsr_color_conversion_draw(gsr_color_conversion *self, unsigned int texture_id, vec2i source_pos, vec2i source_size, vec2i texture_pos, vec2i texture_size, float rotation, bool external_texture, gsr_source_color source_color);
+void gsr_color_conversion_draw(gsr_color_conversion *self, unsigned int texture_id, vec2i destination_pos, vec2i destination_size, vec2i source_pos, vec2i source_size, vec2i texture_size, gsr_rotation rotation, gsr_flip flip, gsr_source_color source_color, bool external_texture);
 void gsr_color_conversion_clear(gsr_color_conversion *self);
+void gsr_color_conversion_read_destination_texture(gsr_color_conversion *self, int destination_texture_index, int x, int y, int width, int height, unsigned int color_format, unsigned int data_format, void *pixels);
+
+gsr_rotation gsr_monitor_rotation_to_rotation(gsr_monitor_rotation monitor_rotation);
 
 #endif /* GSR_COLOR_CONVERSION_H */

include/damage.h

@@ -9,40 +9,62 @@
 typedef struct _XDisplay Display;
 typedef union _XEvent XEvent;
 
-typedef enum {
-    GSR_DAMAGE_TRACK_NONE,
-    GSR_DAMAGE_TRACK_WINDOW,
-    GSR_DAMAGE_TRACK_MONITOR
-} gsr_damage_track_type;
+#define GSR_DAMAGE_MAX_MONITORS 32
+#define GSR_DAMAGE_MAX_TRACKED_TARGETS 12
+
+typedef struct {
+    int64_t window_id;
+    vec2i window_pos;
+    vec2i window_size;
+    uint64_t damage;
+    int refcount;
+} gsr_damage_window;
+
+typedef struct {
+    char *monitor_name;
+    gsr_monitor *monitor;
+    int refcount;
+} gsr_damage_monitor;
 
 typedef struct {
     gsr_egl *egl;
     Display *display;
     bool track_cursor;
-    gsr_damage_track_type track_type;
 
     int damage_event;
     int damage_error;
-    uint64_t damage;
     bool damaged;
 
+    uint64_t monitor_damage;
+
     int randr_event;
     int randr_error;
 
-    uint64_t window;
-    //vec2i window_pos;
-    vec2i window_size;
+    gsr_cursor *cursor;
+    gsr_monitor monitors[GSR_DAMAGE_MAX_MONITORS];
+    int num_monitors;
 
-    gsr_cursor cursor; /* Relative to |window| */
-    gsr_monitor monitor;
-    char monitor_name[32];
+    gsr_damage_window windows_tracked[GSR_DAMAGE_MAX_TRACKED_TARGETS];
+    int num_windows_tracked;
+
+    gsr_damage_monitor monitors_tracked[GSR_DAMAGE_MAX_TRACKED_TARGETS];
+    int num_monitors_tracked;
+
+    int all_monitors_tracked_refcount;
+    vec2i cursor_pos;
 } gsr_damage;
 
-bool gsr_damage_init(gsr_damage *self, gsr_egl *egl, bool track_cursor);
+bool gsr_damage_init(gsr_damage *self, gsr_egl *egl, gsr_cursor *cursor, bool track_cursor);
 void gsr_damage_deinit(gsr_damage *self);
 
-bool gsr_damage_set_target_window(gsr_damage *self, uint64_t window);
-bool gsr_damage_set_target_monitor(gsr_damage *self, const char *monitor_name);
+/* This is reference counted */
+bool gsr_damage_start_tracking_window(gsr_damage *self, int64_t window);
+void gsr_damage_stop_tracking_window(gsr_damage *self, int64_t window);
+
+/* This is reference counted. If |monitor_name| is NULL then all monitors are tracked */
+bool gsr_damage_start_tracking_monitor(gsr_damage *self, const char *monitor_name);
+void gsr_damage_stop_tracking_monitor(gsr_damage *self, const char *monitor_name);
+
 void gsr_damage_on_event(gsr_damage *self, XEvent *xev);
 void gsr_damage_tick(gsr_damage *self);
 /* Also returns true if damage tracking is not available */

include/dbus.h

@@ -7,16 +7,6 @@
 
 #define DBUS_RANDOM_STR_SIZE 16
 
-typedef struct {
-    DBusConnection *con;
-    DBusError err;
-    char random_str[DBUS_RANDOM_STR_SIZE + 1];
-    unsigned int handle_counter;
-    bool desktop_portal_rule_added;
-    uint32_t screencast_version;
-    char *screencast_restore_token;
-} gsr_dbus;
-
 typedef enum {
     GSR_PORTAL_CAPTURE_TYPE_MONITOR = 1 << 0,
     GSR_PORTAL_CAPTURE_TYPE_WINDOW  = 1 << 1,
@@ -30,6 +20,16 @@ typedef enum {
     GSR_PORTAL_CURSOR_MODE_METADATA = 1 << 2
 } gsr_portal_cursor_mode;
 
+typedef struct {
+    DBusConnection *con;
+    DBusError err;
+    char random_str[DBUS_RANDOM_STR_SIZE + 1];
+    unsigned int handle_counter;
+    bool desktop_portal_rule_added;
+    uint32_t screencast_version;
+    char *screencast_restore_token;
+} gsr_dbus;
+
 /* Blocking. TODO: Make non-blocking */
 bool gsr_dbus_init(gsr_dbus *self, const char *screencast_restore_token);
 void gsr_dbus_deinit(gsr_dbus *self);

include/defs.h

@@ -3,34 +3,110 @@
 
 #include <stdbool.h>
 
+#define GSR_VIDEO_CODEC_AUTO -1
+#define GSR_BITRATE_MODE_AUTO -1
+
 typedef enum {
     GSR_GPU_VENDOR_AMD,
     GSR_GPU_VENDOR_INTEL,
-    GSR_GPU_VENDOR_NVIDIA
+    GSR_GPU_VENDOR_NVIDIA,
+    GSR_GPU_VENDOR_BROADCOM,
 } gsr_gpu_vendor;
 
 typedef struct {
     gsr_gpu_vendor vendor;
     int gpu_version; /* 0 if unknown */
     bool is_steam_deck;
-
-    /* Only currently set for Mesa. 0 if unknown format */
-    int driver_major;
-    int driver_minor;
-    int driver_patch;
 } gsr_gpu_info;
 
 typedef enum {
     GSR_MONITOR_ROT_0,
     GSR_MONITOR_ROT_90,
     GSR_MONITOR_ROT_180,
-    GSR_MONITOR_ROT_270
+    GSR_MONITOR_ROT_270,
 } gsr_monitor_rotation;
 
 typedef enum {
     GSR_CONNECTION_X11,
     GSR_CONNECTION_WAYLAND,
-    GSR_CONNECTION_DRM
+    GSR_CONNECTION_DRM,
 } gsr_connection_type;
 
+typedef enum {
+    GSR_VIDEO_QUALITY_MEDIUM,
+    GSR_VIDEO_QUALITY_HIGH,
+    GSR_VIDEO_QUALITY_VERY_HIGH,
+    GSR_VIDEO_QUALITY_ULTRA,
+} gsr_video_quality;
+
+typedef enum {
+    GSR_VIDEO_CODEC_H264,
+    GSR_VIDEO_CODEC_HEVC,
+    GSR_VIDEO_CODEC_HEVC_HDR,
+    GSR_VIDEO_CODEC_HEVC_10BIT,
+    GSR_VIDEO_CODEC_AV1,
+    GSR_VIDEO_CODEC_AV1_HDR,
+    GSR_VIDEO_CODEC_AV1_10BIT,
+    GSR_VIDEO_CODEC_VP8,
+    GSR_VIDEO_CODEC_VP9,
+    GSR_VIDEO_CODEC_H264_VULKAN,
+    GSR_VIDEO_CODEC_HEVC_VULKAN,
+} gsr_video_codec;
+
+typedef enum {
+    GSR_AUDIO_CODEC_AAC,
+    GSR_AUDIO_CODEC_OPUS,
+    GSR_AUDIO_CODEC_FLAC,
+} gsr_audio_codec;
+
+typedef enum {
+    GSR_PIXEL_FORMAT_YUV420,
+    GSR_PIXEL_FORMAT_YUV444,
+} gsr_pixel_format;
+
+typedef enum {
+    GSR_FRAMERATE_MODE_CONSTANT,
+    GSR_FRAMERATE_MODE_VARIABLE,
+    GSR_FRAMERATE_MODE_CONTENT,
+} gsr_framerate_mode;
+
+typedef enum {
+    GSR_BITRATE_MODE_QP,
+    GSR_BITRATE_MODE_VBR,
+    GSR_BITRATE_MODE_CBR,
+} gsr_bitrate_mode;
+
+typedef enum {
+    GSR_TUNE_PERFORMANCE,
+    GSR_TUNE_QUALITY,
+} gsr_tune;
+
+typedef enum {
+    GSR_VIDEO_ENCODER_HW_GPU,
+    GSR_VIDEO_ENCODER_HW_CPU,
+} gsr_video_encoder_hardware;
+
+typedef enum {
+    GSR_COLOR_RANGE_LIMITED,
+    GSR_COLOR_RANGE_FULL,
+} gsr_color_range;
+
+typedef enum {
+    GSR_COLOR_DEPTH_8_BITS,
+    GSR_COLOR_DEPTH_10_BITS,
+} gsr_color_depth;
+
+typedef enum {
+    GSR_REPLAY_STORAGE_RAM,
+    GSR_REPLAY_STORAGE_DISK,
+} gsr_replay_storage;
+
+bool video_codec_is_hdr(gsr_video_codec video_codec);
+gsr_video_codec hdr_video_codec_to_sdr_video_codec(gsr_video_codec video_codec);
+gsr_color_depth video_codec_to_bit_depth(gsr_video_codec video_codec);
+const char* video_codec_to_string(gsr_video_codec video_codec);
+bool video_codec_is_av1(gsr_video_codec video_codec);
+bool video_codec_is_vulkan(gsr_video_codec video_codec);
+const char* audio_codec_get_name(gsr_audio_codec audio_codec);
+
 #endif /* GSR_DEFS_H */

include/egl.h

@@ -48,8 +48,11 @@ typedef void(*__GLXextFuncPtr)(void);
 #define EGL_OPENGL_ES_API                       0x30A0
 #define EGL_OPENGL_BIT                          0x0008
 #define EGL_OPENGL_ES_BIT                       0x0001
+#define EGL_OPENGL_ES2_BIT                      0x0004
+#define EGL_OPENGL_ES3_BIT                      0x00000040
 #define EGL_NONE                                0x3038
 #define EGL_CONTEXT_CLIENT_VERSION              0x3098
+#define EGL_CONTEXT_OPENGL_DEBUG                0x31B0
 #define EGL_BACK_BUFFER                         0x3084
 #define EGL_GL_TEXTURE_2D                       0x30B1
 #define EGL_TRUE                                1
@@ -83,10 +86,6 @@ typedef void(*__GLXextFuncPtr)(void);
 #define EGL_ALPHA_SIZE                          0x3021
 #define EGL_BLUE_SIZE                           0x3022
 #define EGL_GREEN_SIZE                          0x3023
-#define EGL_CONTEXT_PRIORITY_LEVEL_IMG          0x3100
-#define EGL_CONTEXT_PRIORITY_HIGH_IMG           0x3101
-#define EGL_CONTEXT_PRIORITY_MEDIUM_IMG         0x3102
-#define EGL_CONTEXT_PRIORITY_LOW_IMG            0x3103
 #define EGL_DEVICE_EXT                          0x322C
 #define EGL_DRM_DEVICE_FILE_EXT                 0x3233
 
@@ -98,7 +97,7 @@ typedef void(*__GLXextFuncPtr)(void);
 #define GL_TEXTURE_EXTERNAL_OES                 0x8D65
 #define GL_RED                                  0x1903
 #define GL_GREEN                                0x1904
-#define GL_BLUE                                    0x1905
+#define GL_BLUE                                 0x1905
 #define GL_ALPHA                                0x1906
 #define GL_TEXTURE_SWIZZLE_RGBA                 0x8E46
 #define GL_RG                                   0x8227
@@ -106,19 +105,19 @@ typedef void(*__GLXextFuncPtr)(void);
 #define GL_RGBA                                 0x1908
 #define GL_RGB8                                 0x8051
 #define GL_RGBA8                                0x8058
+#define GL_RGBA16                               0x805B
 #define GL_R8                                   0x8229
 #define GL_RG8                                  0x822B
 #define GL_R16                                  0x822A
 #define GL_RG16                                 0x822C
 #define GL_RGB16                                0x8054
+#define GL_RGBA32F                              0x8814
 #define GL_UNSIGNED_BYTE                        0x1401
 #define GL_COLOR_BUFFER_BIT                     0x00004000
 #define GL_TEXTURE_WRAP_S                       0x2802
 #define GL_TEXTURE_WRAP_T                       0x2803
 #define GL_TEXTURE_MAG_FILTER                   0x2800
 #define GL_TEXTURE_MIN_FILTER                   0x2801
-#define GL_TEXTURE_WIDTH                        0x1000
-#define GL_TEXTURE_HEIGHT                       0x1001
 #define GL_NEAREST                              0x2600
 #define GL_CLAMP_TO_EDGE                        0x812F
 #define GL_LINEAR                               0x2601
@@ -130,10 +129,16 @@ typedef void(*__GLXextFuncPtr)(void);
 #define GL_BLEND                                0x0BE2
 #define GL_SRC_ALPHA                            0x0302
 #define GL_ONE_MINUS_SRC_ALPHA                  0x0303
-#define GL_DEBUG_OUTPUT                         0x92E0
+#define GL_DEBUG_OUTPUT_SYNCHRONOUS             0x8242
 #define GL_SCISSOR_TEST                         0x0C11
 #define GL_PACK_ALIGNMENT                       0x0D05
 #define GL_UNPACK_ALIGNMENT                     0x0CF5
+#define GL_TEXTURE0                             0x84C0
+#define GL_TEXTURE1                             0x84C1
+#define GL_SHADER_IMAGE_ACCESS_BARRIER_BIT      0x00000020
+#define GL_ALL_BARRIER_BITS                     0xFFFFFFFF
+#define GL_PIXEL_UNPACK_BUFFER                  0x88EC
+#define GL_MAP_WRITE_BIT                        0x0002
 
 #define GL_VENDOR                               0x1F00
 #define GL_RENDERER                             0x1F01
@@ -157,6 +162,13 @@ typedef void (*GLDEBUGPROC)(unsigned int source, unsigned int type, unsigned int
 typedef int (*FUNC_eglQueryDisplayAttribEXT)(EGLDisplay dpy, int32_t attribute, intptr_t *value);
 typedef const char* (*FUNC_eglQueryDeviceStringEXT)(void *device, int32_t name);
 typedef int (*FUNC_eglQueryDmaBufModifiersEXT)(EGLDisplay dpy, int32_t format, int32_t max_modifiers, uint64_t *modifiers, int *external_only, int32_t *num_modifiers);
+typedef void (*FUNC_glCreateMemoryObjectsEXT)(int n, unsigned int *memoryObjects);
+typedef void (*FUNC_glImportMemoryFdEXT)(unsigned int memory, uint64_t size, unsigned int handleType, int fd);
+typedef unsigned char (*FUNC_glIsMemoryObjectEXT)(unsigned int memoryObject);
+typedef void (*FUNC_glTexStorageMem2DEXT)(unsigned int target, int levels, unsigned int internalFormat, int width, int height, unsigned int memory, uint64_t offset);
+typedef void (*FUNC_glBufferStorageMemEXT)(unsigned int target, ssize_t size, unsigned int memory, uint64_t offset);
+typedef void (*FUNC_glNamedBufferStorageMemEXT)(unsigned int buffer, ssize_t size, unsigned int memory, uint64_t offset);
+typedef void (*FUNC_glMemoryObjectParameterivEXT)(unsigned int memoryObject, unsigned int pname, const int *params);
 
 typedef enum {
     GSR_GL_CONTEXT_TYPE_EGL,
@@ -207,6 +219,13 @@ struct gsr_egl {
     FUNC_eglQueryDisplayAttribEXT eglQueryDisplayAttribEXT;
     FUNC_eglQueryDeviceStringEXT eglQueryDeviceStringEXT;
     FUNC_eglQueryDmaBufModifiersEXT eglQueryDmaBufModifiersEXT;
+    FUNC_glCreateMemoryObjectsEXT glCreateMemoryObjectsEXT;
+    FUNC_glImportMemoryFdEXT glImportMemoryFdEXT;
+    FUNC_glIsMemoryObjectEXT glIsMemoryObjectEXT;
+    FUNC_glTexStorageMem2DEXT glTexStorageMem2DEXT;
+    FUNC_glBufferStorageMemEXT glBufferStorageMemEXT;
+    FUNC_glNamedBufferStorageMemEXT glNamedBufferStorageMemEXT;
+    FUNC_glMemoryObjectParameterivEXT glMemoryObjectParameterivEXT;
 
     __GLXextFuncPtr (*glXGetProcAddress)(const unsigned char *procName);
     GLXFBConfig* (*glXChooseFBConfig)(Display *dpy, int screen, const int *attribList, int *nitems);
@@ -230,16 +249,21 @@ struct gsr_egl {
     void (*glClearColor)(float red, float green, float blue, float alpha);
     void (*glGenTextures)(int n, unsigned int *textures);
     void (*glDeleteTextures)(int n, const unsigned int *texture);
+    void (*glActiveTexture)(unsigned int texture);
     void (*glBindTexture)(unsigned int target, unsigned int texture);
+    void (*glBindImageTexture)(unsigned int unit, unsigned int texture, int level, unsigned char layered, int layer, unsigned int access, unsigned int format);
     void (*glTexParameteri)(unsigned int target, unsigned int pname, int param);
     void (*glTexParameteriv)(unsigned int target, unsigned int pname, const int *params);
-    void (*glGetTexLevelParameteriv)(unsigned int target, int level, unsigned int pname, int *params);
+    void (*glTexParameterfv)(unsigned int target, unsigned int pname, const float *params);
     void (*glTexImage2D)(unsigned int target, int level, int internalFormat, int width, int height, int border, unsigned int format, unsigned int type, const void *pixels);
     void (*glTexSubImage2D)(unsigned int target, int level, int xoffset, int yoffset, int width, int height, unsigned format, unsigned type, const void *pixels);
+    void (*glTexStorage2D)(unsigned int target, int levels, unsigned int internalformat, int width, int height);
     void (*glGetTexImage)(unsigned int target, int level, unsigned int format, unsigned int type, void *pixels);
     void (*glGenFramebuffers)(int n, unsigned int *framebuffers);
     void (*glBindFramebuffer)(unsigned int target, unsigned int framebuffer);
     void (*glDeleteFramebuffers)(int n, const unsigned int *framebuffers);
+    /* TODO: Note: this is only available since OpenGL ES 3.1 */
+    void (*glMemoryBarrier)(unsigned int barriers);
     void (*glViewport)(int x, int y, int width, int height);
     void (*glFramebufferTexture2D)(unsigned int target, unsigned int attachment, unsigned int textarget, unsigned int texture, int level);
     void (*glDrawBuffers)(int n, const unsigned int *bufs);
@@ -276,11 +300,16 @@ struct gsr_egl {
     int (*glGetUniformLocation)(unsigned int program, const char *name);
     void (*glUniform1f)(int location, float v0);
     void (*glUniform2f)(int location, float v0, float v1);
+    void (*glUniform1i)(int location, int v0);
+    void (*glUniform2i)(int location, int v0, int v1);
+    void (*glUniformMatrix2fv)(int location, int count, unsigned char transpose, const float *value);
     void (*glDebugMessageCallback)(GLDEBUGPROC callback, const void *userParam);
     void (*glScissor)(int x, int y, int width, int height);
+    void (*glCreateBuffers)(int n, unsigned int *buffers);
     void (*glReadPixels)(int x, int y, int width, int height, unsigned int format, unsigned int type, void *pixels);
-    void* (*glMapBuffer)(unsigned int target, unsigned int access);
+    void* (*glMapBufferRange)(unsigned int target, intptr_t offset, ssize_t length, unsigned int access);
     unsigned char (*glUnmapBuffer)(unsigned int target);
+    void (*glGetIntegerv)(unsigned int pname, int *params);
 };
 
 bool gsr_egl_load(gsr_egl *self, gsr_window *window, bool is_monitor_capture, bool enable_debug);

include/encoder/encoder.h

@@ -0,0 +1,50 @@
+#ifndef GSR_ENCODER_H
+#define GSR_ENCODER_H
+
+#include "../replay_buffer/replay_buffer.h"
+#include <stdbool.h>
+#include <stdint.h>
+#include <stddef.h>
+#include <pthread.h>
+
+#define GSR_MAX_RECORDING_DESTINATIONS 128
+
+typedef struct AVCodecContext AVCodecContext;
+typedef struct AVFormatContext AVFormatContext;
+typedef struct AVStream AVStream;
+
+typedef struct {
+    size_t id;
+    AVCodecContext *codec_context;
+    AVFormatContext *format_context;
+    AVStream *stream;
+    int64_t start_pts;
+    bool has_received_keyframe;
+    char *first_frame_ts_filepath;
+    bool first_frame_ts_written;
+} gsr_encoder_recording_destination;
+
+typedef struct {
+    gsr_replay_buffer *replay_buffer;
+
+    pthread_mutex_t file_write_mutex;
+    bool file_write_mutex_created;
+
+    pthread_mutex_t replay_mutex;
+    bool replay_mutex_created;
+
+    gsr_encoder_recording_destination recording_destinations[GSR_MAX_RECORDING_DESTINATIONS];
+    size_t num_recording_destinations;
+    size_t recording_destination_id_counter;
+} gsr_encoder;
+
+bool gsr_encoder_init(gsr_encoder *self, gsr_replay_storage replay_storage, size_t replay_buffer_num_packets, double replay_buffer_time, const char *replay_directory);
+void gsr_encoder_deinit(gsr_encoder *self);
+
+void gsr_encoder_receive_packets(gsr_encoder *self, AVCodecContext *codec_context, int64_t pts, int stream_index);
+/* Returns the id to the recording destination, or -1 on error */
+size_t gsr_encoder_add_recording_destination(gsr_encoder *self, AVCodecContext *codec_context, AVFormatContext *format_context, AVStream *stream, int64_t start_pts);
+bool gsr_encoder_remove_recording_destination(gsr_encoder *self, size_t id);
+bool gsr_encoder_set_recording_destination_first_frame_ts_filepath(gsr_encoder *self, size_t id, const char *filepath);
+
+#endif /* GSR_ENCODER_H */

include/encoder/video/video.h

@@ -4,24 +4,27 @@
 #include "../../color_conversion.h"
 #include <stdbool.h>
 
+#define GSR_MAX_RECORDING_DESTINATIONS 128
+
 typedef struct gsr_video_encoder gsr_video_encoder;
 typedef struct AVCodecContext AVCodecContext;
 typedef struct AVFrame AVFrame;
 
 struct gsr_video_encoder {
     bool (*start)(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame);
+    void (*destroy)(gsr_video_encoder *encoder, AVCodecContext *video_codec_context);
     void (*copy_textures_to_frame)(gsr_video_encoder *encoder, AVFrame *frame, gsr_color_conversion *color_conversion); /* Can be NULL */
     /* |textures| should be able to fit 2 elements */
-    void (*get_textures)(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color);
-    void (*destroy)(gsr_video_encoder *encoder, AVCodecContext *video_codec_context);
+    void (*get_textures)(gsr_video_encoder *encoder, unsigned int *textures, vec2i *texture_sizes, int *num_textures, gsr_destination_color *destination_color);
 
     void *priv;
     bool started;
 };
 
+/* Set |replay_buffer_time_seconds| and |fps| to 0 to disable replay buffer */
 bool gsr_video_encoder_start(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame);
-void gsr_video_encoder_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame, gsr_color_conversion *color_conversion);
-void gsr_video_encoder_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color);
 void gsr_video_encoder_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context);
+void gsr_video_encoder_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame, gsr_color_conversion *color_conversion);
+void gsr_video_encoder_get_textures(gsr_video_encoder *encoder, unsigned int *textures, vec2i *texture_sizes, int *num_textures, gsr_destination_color *destination_color);
 
 #endif /* GSR_ENCODER_VIDEO_H */

include/image_writer.h

@@ -10,23 +10,14 @@ typedef enum {
     GSR_IMAGE_FORMAT_PNG
 } gsr_image_format;
 
-typedef enum {
-    GSR_IMAGE_WRITER_SOURCE_OPENGL,
-    GSR_IMAGE_WRITER_SOURCE_MEMORY
-} gsr_image_writer_source;
-
 typedef struct {
-    gsr_image_writer_source source;
     gsr_egl *egl;
     int width;
     int height;
     unsigned int texture;
-    const void *memory; /* Reference */
 } gsr_image_writer;
 
 bool gsr_image_writer_init_opengl(gsr_image_writer *self, gsr_egl *egl, int width, int height);
-/* |memory| is taken as a reference */
-bool gsr_image_writer_init_memory(gsr_image_writer *self, const void *memory, int width, int height);
 void gsr_image_writer_deinit(gsr_image_writer *self);
 
 /* Quality is between 1 and 100 where 100 is the max quality. Quality doesn't apply to lossless formats */

include/pipewire_audio.h

@@ -8,12 +8,6 @@
 
 #include <stdbool.h>
 
-#define GSR_PIPEWIRE_AUDIO_MAX_STREAM_NODES 128
-#define GSR_PIPEWIRE_AUDIO_MAX_PORTS 256
-#define GSR_PIPEWIRE_AUDIO_MAX_LINKS 256
-#define GSR_PIPEWIRE_AUDIO_MAX_REQUESTED_LINKS 32
-#define GSR_PIPEWIRE_AUDIO_MAX_VIRTUAL_SINKS 32
-
 typedef enum {
     GSR_PIPEWIRE_AUDIO_NODE_TYPE_STREAM_OUTPUT, /* Application audio */
     GSR_PIPEWIRE_AUDIO_NODE_TYPE_STREAM_INPUT,  /* Audio recording input */
@@ -84,27 +78,28 @@ typedef struct {
     char default_output_device_name[128];
     char default_input_device_name[128];
 
-    gsr_pipewire_audio_node stream_nodes[GSR_PIPEWIRE_AUDIO_MAX_STREAM_NODES];
-    int num_stream_nodes;
+    gsr_pipewire_audio_node *stream_nodes;
+    size_t num_stream_nodes;
+    size_t stream_nodes_capacity_items;
 
-    gsr_pipewire_audio_port ports[GSR_PIPEWIRE_AUDIO_MAX_PORTS];
-    int num_ports;
+    gsr_pipewire_audio_port *ports;
+    size_t num_ports;
+    size_t ports_capacity_items;
 
-    gsr_pipewire_audio_link links[GSR_PIPEWIRE_AUDIO_MAX_LINKS];
-    int num_links;
+    gsr_pipewire_audio_link *links;
+    size_t num_links;
+    size_t links_capacity_items;
 
-    gsr_pipewire_audio_requested_link requested_links[GSR_PIPEWIRE_AUDIO_MAX_REQUESTED_LINKS];
-    int num_requested_links;
+    gsr_pipewire_audio_requested_link *requested_links;
+    size_t num_requested_links;
+    size_t requested_links_capacity_items;
 
-    struct pw_proxy *virtual_sink_proxies[GSR_PIPEWIRE_AUDIO_MAX_VIRTUAL_SINKS];
-    int num_virtual_sink_proxies;
+    bool running;
 } gsr_pipewire_audio;
 
 bool gsr_pipewire_audio_init(gsr_pipewire_audio *self);
 void gsr_pipewire_audio_deinit(gsr_pipewire_audio *self);
 
-bool gsr_pipewire_audio_create_virtual_sink(gsr_pipewire_audio *self, const char *name);
-
 /*
     This function links audio source outputs from applications that match the name |app_names| to the input
     that matches the name |stream_name_input|.
@@ -139,6 +134,17 @@ bool gsr_pipewire_audio_add_link_from_apps_to_sink(gsr_pipewire_audio *self, con
 */
 bool gsr_pipewire_audio_add_link_from_apps_to_sink_inverted(gsr_pipewire_audio *self, const char **app_names, int num_app_names, const char *sink_name_input);
 
+/*
+    This function links audio source outputs from devices that match the name |source_names| to the input
+    that matches the name |stream_name_input|.
+    If a device or a new device starts outputting audio after this function is called and the device name matches
+    then it will automatically link the audio sources.
+    |source_names| and |stream_name_input| are case-insensitive matches.
+    |source_names| can include "default_output" or "default_input" to use the default output/input
+    and it will automatically switch when the default output/input is changed in system audio settings.
+*/
+bool gsr_pipewire_audio_add_link_from_sources_to_stream(gsr_pipewire_audio *self, const char **source_names, int num_source_names, const char *stream_name_input);
+
 /*
     This function links audio source outputs from devices that match the name |source_names| to the input
     that matches the name |sink_name_input|.

include/pipewire_video.h

@@ -1,6 +1,7 @@
 #ifndef GSR_PIPEWIRE_VIDEO_H
 #define GSR_PIPEWIRE_VIDEO_H
 
+#include "defs.h"
 #include <stdbool.h>
 #include <stdint.h>
 #include <pthread.h>
@@ -9,7 +10,7 @@
 #include <spa/param/video/format.h>
 
 #define GSR_PIPEWIRE_VIDEO_MAX_MODIFIERS 1024
-#define GSR_PIPEWIRE_VIDEO_MAX_VIDEO_FORMATS 12
+#define GSR_PIPEWIRE_VIDEO_MAX_VIDEO_FORMATS 10
 #define GSR_PIPEWIRE_VIDEO_DMABUF_MAX_PLANES 4
 
 typedef struct gsr_egl gsr_egl;
@@ -48,6 +49,19 @@ typedef struct {
     unsigned int cursor_texture_id;
 } gsr_texture_map;
 
+typedef struct {
+    gsr_pipewire_video_region region;
+    gsr_pipewire_video_region cursor_region;
+    gsr_pipewire_video_dmabuf_data dmabuf_data[GSR_PIPEWIRE_VIDEO_DMABUF_MAX_PLANES];
+    int num_dmabuf_data;
+    uint32_t fourcc;
+    uint64_t modifiers;
+    bool using_external_image;
+    gsr_monitor_rotation rotation;
+    int texture_width;
+    int texture_height;
+} gsr_map_texture_output;
+
 typedef struct {
     gsr_egl *egl;
     int fd;
@@ -66,6 +80,7 @@ typedef struct {
     int server_version_sync;
     bool negotiated;
     bool damaged;
+    bool has_modifier;
 
     struct {
         bool visible;
@@ -94,6 +109,12 @@ typedef struct {
 
     uint64_t modifiers[GSR_PIPEWIRE_VIDEO_MAX_MODIFIERS];
     size_t num_modifiers;
+
+    bool paused;
+    double paused_start_secs;
+    bool streaming;
+
+    gsr_monitor_rotation rotation;
 } gsr_pipewire_video;
 
 /*
@@ -104,9 +125,9 @@ typedef struct {
 bool gsr_pipewire_video_init(gsr_pipewire_video *self, int pipewire_fd, uint32_t pipewire_node, int fps, bool capture_cursor, gsr_egl *egl);
 void gsr_pipewire_video_deinit(gsr_pipewire_video *self);
 
-/* |dmabuf_data| should be at least GSR_PIPEWIRE_VIDEO_DMABUF_MAX_PLANES in size */
-bool gsr_pipewire_video_map_texture(gsr_pipewire_video *self, gsr_texture_map texture_map, gsr_pipewire_video_region *region, gsr_pipewire_video_region *cursor_region, gsr_pipewire_video_dmabuf_data *dmabuf_data, int *num_dmabuf_data, uint32_t *fourcc, uint64_t *modifiers, bool *using_external_image);
+bool gsr_pipewire_video_map_texture(gsr_pipewire_video *self, gsr_texture_map texture_map, gsr_map_texture_output *output);
 bool gsr_pipewire_video_is_damaged(gsr_pipewire_video *self);
 void gsr_pipewire_video_clear_damage(gsr_pipewire_video *self);
+bool gsr_pipewire_video_should_restart(gsr_pipewire_video *self);
 
 #endif /* GSR_PIPEWIRE_VIDEO_H */

include/plugins.h

@@ -0,0 +1,38 @@
+#ifndef GSR_PLUGINS_H
+#define GSR_PLUGINS_H
+
+#include "../plugin/plugin.h"
+#include <stdbool.h>
+#include "color_conversion.h"
+
+#define GSR_MAX_PLUGINS 128
+
+typedef bool (*gsr_plugin_init_func)(const gsr_plugin_init_params *params, gsr_plugin_init_return *ret);
+typedef void (*gsr_plugin_deinit_func)(void *userdata);
+
+typedef struct {
+    gsr_plugin_init_return data;
+    void *lib;
+    gsr_plugin_init_func gsr_plugin_init;
+    gsr_plugin_deinit_func gsr_plugin_deinit;
+} gsr_plugin;
+
+typedef struct {
+    gsr_plugin plugins[GSR_MAX_PLUGINS];
+    int num_plugins;
+
+    gsr_plugin_init_params init_params;
+    gsr_egl *egl;
+
+    unsigned int texture;
+    gsr_color_conversion color_conversion;
+} gsr_plugins;
+
+bool gsr_plugins_init(gsr_plugins *self, gsr_plugin_init_params init_params, gsr_egl *egl);
+/* Plugins are unloaded in reverse order */
+void gsr_plugins_deinit(gsr_plugins *self);
+
+bool gsr_plugins_load_plugin(gsr_plugins *self, const char *plugin_filepath);
+void gsr_plugins_draw(gsr_plugins *self);
+
+#endif /* GSR_PLUGINS_H */

include/replay_buffer/replay_buffer.h

@@ -0,0 +1,47 @@
+#ifndef GSR_REPLAY_BUFFER_H
+#define GSR_REPLAY_BUFFER_H
+
+#include "../defs.h"
+#include <stdbool.h>
+#include <libavcodec/packet.h>
+
+typedef struct gsr_replay_buffer gsr_replay_buffer;
+
+typedef struct {
+    size_t packet_index;
+    size_t file_index;
+} gsr_replay_buffer_iterator;
+
+struct gsr_replay_buffer {
+    void (*destroy)(gsr_replay_buffer *self);
+    bool (*append)(gsr_replay_buffer *self, const AVPacket *av_packet, double timestamp);
+    void (*clear)(gsr_replay_buffer *self);
+    AVPacket* (*iterator_get_packet)(gsr_replay_buffer *self, gsr_replay_buffer_iterator iterator);
+    /* The returned data should be free'd with free */
+    uint8_t* (*iterator_get_packet_data)(gsr_replay_buffer *self, gsr_replay_buffer_iterator iterator);
+    /* The clone has to be destroyed before the replay buffer it clones is destroyed */
+    gsr_replay_buffer* (*clone)(gsr_replay_buffer *self);
+    /* Returns {0, 0} if replay buffer is empty */
+    gsr_replay_buffer_iterator (*find_packet_index_by_time_passed)(gsr_replay_buffer *self, int seconds);
+    /* Returns {-1, 0} if not found */
+    gsr_replay_buffer_iterator (*find_keyframe)(gsr_replay_buffer *self, gsr_replay_buffer_iterator start_iterator, int stream_index, bool invert_stream_index);
+    bool (*iterator_next)(gsr_replay_buffer *self, gsr_replay_buffer_iterator *iterator);
+};
+
+gsr_replay_buffer* gsr_replay_buffer_create(gsr_replay_storage replay_storage, const char *replay_directory, double replay_buffer_time, size_t replay_buffer_num_packets);
+void gsr_replay_buffer_destroy(gsr_replay_buffer *self);
+
+bool gsr_replay_buffer_append(gsr_replay_buffer *self, const AVPacket *av_packet, double timestamp);
+void gsr_replay_buffer_clear(gsr_replay_buffer *self);
+AVPacket* gsr_replay_buffer_iterator_get_packet(gsr_replay_buffer *self, gsr_replay_buffer_iterator iterator);
+/* The returned data should be free'd with free */
+uint8_t* gsr_replay_buffer_iterator_get_packet_data(gsr_replay_buffer *self, gsr_replay_buffer_iterator iterator);
+/* The clone has to be destroyed before the replay buffer it clones is destroyed */
+gsr_replay_buffer* gsr_replay_buffer_clone(gsr_replay_buffer *self);
+/* Returns {0, 0} if replay buffer is empty */
+gsr_replay_buffer_iterator gsr_replay_buffer_find_packet_index_by_time_passed(gsr_replay_buffer *self, int seconds);
+/* Returns {-1, 0} if not found */
+gsr_replay_buffer_iterator gsr_replay_buffer_find_keyframe(gsr_replay_buffer *self, gsr_replay_buffer_iterator start_iterator, int stream_index, bool invert_stream_index);
+bool gsr_replay_buffer_iterator_next(gsr_replay_buffer *self, gsr_replay_buffer_iterator *iterator);
+
+#endif /* GSR_REPLAY_BUFFER_H */

include/replay_buffer/replay_buffer_disk.h

@@ -0,0 +1,44 @@
+#ifndef GSR_REPLAY_BUFFER_DISK_H
+#define GSR_REPLAY_BUFFER_DISK_H
+
+#include "replay_buffer.h"
+#include <limits.h>
+
+#define GSR_REPLAY_BUFFER_CAPACITY_NUM_FILES 1024
+
+typedef struct {
+    AVPacket packet;
+    size_t data_index;
+    double timestamp;
+} gsr_av_packet_disk;
+
+typedef struct {
+    size_t id;
+    double start_timestamp;
+    double end_timestamp;
+    int ref_counter;
+    int fd;
+    
+    gsr_av_packet_disk *packets;
+    size_t capacity_num_packets;
+    size_t num_packets;
+} gsr_replay_buffer_file;
+
+typedef struct {
+    gsr_replay_buffer replay_buffer;
+    double replay_buffer_time;
+
+    size_t storage_counter;
+    size_t storage_num_bytes_written;
+    int storage_fd;
+    gsr_replay_buffer_file *files[GSR_REPLAY_BUFFER_CAPACITY_NUM_FILES]; // GSR_REPLAY_BUFFER_CAPACITY_NUM_FILES * REPLAY_BUFFER_FILE_SIZE_BYTES = 256gb, should be enough for everybody
+    size_t num_files;
+
+    char replay_directory[PATH_MAX];
+
+    bool owns_directory;
+} gsr_replay_buffer_disk;
+
+gsr_replay_buffer* gsr_replay_buffer_disk_create(const char *replay_directory, double replay_buffer_time);
+
+#endif /* GSR_REPLAY_BUFFER_DISK_H */

include/replay_buffer/replay_buffer_ram.h

@@ -0,0 +1,22 @@
+#ifndef GSR_REPLAY_BUFFER_RAM_H
+#define GSR_REPLAY_BUFFER_RAM_H
+
+#include "replay_buffer.h"
+
+typedef struct {
+    AVPacket packet;
+    int ref_counter;
+    double timestamp;
+} gsr_av_packet_ram;
+
+typedef struct {
+    gsr_replay_buffer replay_buffer;
+    gsr_av_packet_ram **packets;
+    size_t capacity_num_packets;
+    size_t num_packets;
+    size_t index;
+} gsr_replay_buffer_ram;
+
+gsr_replay_buffer* gsr_replay_buffer_ram_create(size_t replay_buffer_num_packets);
+
+#endif /* GSR_REPLAY_BUFFER_RAM_H */

include/shader.h

@@ -1,6 +1,8 @@
 #ifndef GSR_SHADER_H
 #define GSR_SHADER_H
 
+#include <stdbool.h>
+
 typedef struct gsr_egl gsr_egl;
 
 typedef struct {
@@ -16,4 +18,6 @@ int gsr_shader_bind_attribute_location(gsr_shader *self, const char *attribute,
 void gsr_shader_use(gsr_shader *self);
 void gsr_shader_use_none(gsr_shader *self);
 
+void gsr_shader_enable_debug_output(bool enable);
+
 #endif /* GSR_SHADER_H */

include/sound.hpp

@@ -61,12 +61,12 @@ typedef enum {
 
 /*
     Get a sound device by name, returning the device into the |device| parameter.
-    |device_name| can be a device name or "default_output" or "default_input".
+    |device_name| can be a device name or "default_output", "default_input" or "" to not connect to any device (used for app audio for example).
     If the device name is "default_output" or "default_input" then it will automatically switch which
     device is records from when the default output/input is changed in the system audio settings.
     Returns 0 on success, or a negative value on failure.
 */
-int sound_device_get_by_name(SoundDevice *device, const char *device_name, const char *description, unsigned int num_channels, unsigned int period_frame_size, AudioFormat audio_format);
+int sound_device_get_by_name(SoundDevice *device, const char *node_name, const char *device_name, const char *description, unsigned int num_channels, unsigned int period_frame_size, AudioFormat audio_format);
 
 void sound_device_close(SoundDevice *device);
 

include/utils.h

@@ -7,16 +7,17 @@
 #include <stdbool.h>
 #include <stdint.h>
 
-#define CONNECTOR_TYPE_COUNTS 32
-
 typedef struct AVCodecContext AVCodecContext;
 typedef struct AVFrame AVFrame;
+typedef struct gsr_capture_metadata gsr_capture_metadata;
 
 typedef struct {
     const char *name;
     int name_len;
     vec2i pos; /* This is 0, 0 on wayland. Use |drm_monitor_get_display_server_data| to get the position */
     vec2i size;
+    vec2i logical_pos;
+    vec2i logical_size;
     uint32_t connector_id; /* Only on x11 and drm */
     gsr_monitor_rotation rotation; /* Only on x11 and wayland */
     uint32_t monitor_identifier; /* On x11 this is the crtc id */
@@ -29,12 +30,6 @@ typedef struct {
     bool found_monitor;
 } get_monitor_by_name_userdata;
 
-typedef struct {
-    int type;
-    int count;
-    int count_active;
-} drm_connector_type_count;
-
 double clock_get_monotonic_seconds(void);
 bool generate_random_characters(char *buffer, int buffer_size, const char *alphabet, size_t alphabet_size);
 bool generate_random_characters_standard_alphabet(char *buffer, int buffer_size);
@@ -46,12 +41,10 @@ bool get_monitor_by_name(const gsr_egl *egl, gsr_connection_type connection_type
 bool drm_monitor_get_display_server_data(const gsr_window *window, const gsr_monitor *monitor, gsr_monitor_rotation *monitor_rotation, vec2i *monitor_position);
 
 int get_connector_type_by_name(const char *name);
-drm_connector_type_count* drm_connector_types_get_index(drm_connector_type_count *type_counts, int *num_type_counts, int connector_type);
+int get_connector_type_id_by_name(const char *name);
 uint32_t monitor_identifier_from_type_and_count(int monitor_type_index, int monitor_type_count);
 
 bool gl_get_gpu_info(gsr_egl *egl, gsr_gpu_info *info);
-bool version_greater_than(int major, int minor, int patch, int other_major, int other_minor, int other_patch);
-bool gl_driver_version_greater_than(const gsr_gpu_info *gpu_info, int major, int minor, int patch);
 
 bool try_card_has_valid_plane(const char *card_path);
 /* |output| should be at least 128 bytes in size */
@@ -63,12 +56,12 @@ int create_directory_recursive(char *path);
 
 /* |img_attr| needs to be at least 44 in size */
 void setup_dma_buf_attrs(intptr_t *img_attr, uint32_t format, uint32_t width, uint32_t height, const int *fds, const uint32_t *offsets, const uint32_t *pitches, const uint64_t *modifiers, int num_planes, bool use_modifier);
-bool video_codec_context_is_vaapi(AVCodecContext *video_codec_context);
-bool vaapi_copy_drm_planes_to_video_surface(AVCodecContext *video_codec_context, AVFrame *video_frame, vec2i source_pos, vec2i source_size, vec2i dest_pos, vec2i dest_size, uint32_t format, vec2i size, const int *fds, const uint32_t *offsets, const uint32_t *pitches, const uint64_t *modifiers, int num_planes);
-bool vaapi_copy_egl_image_to_video_surface(gsr_egl *egl, EGLImage image, vec2i source_pos, vec2i source_size, vec2i dest_pos, vec2i dest_size, AVCodecContext *video_codec_context, AVFrame *video_frame);
 
 vec2i scale_keep_aspect_ratio(vec2i from, vec2i to);
+vec2i gsr_capture_get_target_position(vec2i output_size, gsr_capture_metadata *capture_metadata);
 
 unsigned int gl_create_texture(gsr_egl *egl, int width, int height, int internal_format, unsigned int format, int filter);
 
+bool get_nvidia_driver_version(int *major, int *minor);
+
 #endif /* GSR_UTILS_H */

include/window_texture.h

@@ -11,6 +11,8 @@ typedef struct {
     unsigned int texture_id;
     int redirected;
     gsr_egl *egl;
+    unsigned int window_width;
+    unsigned int window_height;
 } WindowTexture;
 
 /* Returns 0 on success */

kms/client/kms_client.c

@@ -13,7 +13,12 @@
 #include <sys/wait.h>
 #include <poll.h>
 #include <sys/stat.h>
+#ifdef __linux__
 #include <sys/capability.h>
+#endif
+#ifdef __FreeBSD__
+#include <sys/sysctl.h>
+#endif
 
 #define GSR_SOCKET_PAIR_LOCAL  0
 #define GSR_SOCKET_PAIR_REMOTE 1
@@ -119,8 +124,11 @@ static int recv_msg_from_server(int server_pid, int server_fd, gsr_kms_response
 
 /* We have to use $HOME because in flatpak there is no simple path that is accessible, read and write, that multiple flatpak instances can access */
 static bool create_socket_path(char *output_path, size_t output_path_size) {
+    const bool inside_flatpak = getenv("FLATPAK_ID") != NULL;
     const char *home = getenv("HOME");
-    if(!home)
+    // Portable home with AppImage can cause the socket path to be longer than 108 characters (unix domain socket path max length).
+    // Using gsr-kms-socket in $HOME is only needed in flatpak, so use /tmp everywhere else instead.
+    if(!home || !inside_flatpak)
         home = "/tmp";
 
     char random_characters[11];
@@ -132,6 +140,7 @@ static bool create_socket_path(char *output_path, size_t output_path_size) {
     return true;
 }
 
+#ifdef __linux__
 static bool readlink_realpath(const char *filepath, char *buffer) {
     char symlinked_path[PATH_MAX];
     ssize_t bytes_written = readlink(filepath, symlinked_path, sizeof(symlinked_path) - 1);
@@ -149,6 +158,7 @@ static bool readlink_realpath(const char *filepath, char *buffer) {
 
     return true;
 }
+#endif
 
 static bool strcat_safe(char *str, int size, const char *str_to_add) {
     const int str_len = strlen(str);
@@ -220,10 +230,24 @@ int gsr_kms_client_init(gsr_kms_client *self, const char *card_path) {
     }
 
     char server_filepath[PATH_MAX];
+#ifdef __linux__
     if(!readlink_realpath("/proc/self/exe", server_filepath)) {
         fprintf(stderr, "gsr error: gsr_kms_client_init: failed to resolve /proc/self/exe\n");
         return -1;
     }
+
+#elif defined(__FreeBSD__)
+    int mib[4] = { CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, getpid() };
+    size_t size = PATH_MAX;
+
+    if (sysctl(mib, 4, server_filepath, &size, NULL, 0) != 0) {
+        fprintf(stderr, "gsr error: gsr_kms_client_init: failed to resolve pathname using sysctl\n");
+        return -1;
+    }
+
+#else
+#error "Implement it by yourself"
+#endif
     file_get_directory(server_filepath);
 
     if(!strcat_safe(server_filepath, sizeof(server_filepath), "/gsr-kms-server")) {
@@ -250,6 +274,7 @@ int gsr_kms_client_init(gsr_kms_client *self, const char *card_path) {
     if(geteuid() == 0) {
         has_perm = true;
     } else {
+#ifdef __linux__
         cap_t kms_server_cap = cap_get_file(server_filepath);
         if(kms_server_cap) {
             cap_flag_value_t res = CAP_CLEAR;
@@ -267,6 +292,9 @@ int gsr_kms_client_init(gsr_kms_client *self, const char *card_path) {
             else
                 fprintf(stderr, "gsr info: gsr_kms_client_init: failed to get cap\n");
         }
+#else
+        fprintf(stderr, "gsr info: gsr_kms_client_init: platform doesn't support cap\n");
+#endif
     }
 
     if(socketpair(AF_UNIX, SOCK_STREAM, 0, self->socket_pair) == -1) {
@@ -312,12 +340,14 @@ int gsr_kms_client_init(gsr_kms_client *self, const char *card_path) {
             const char *args[] = { "pkexec", server_filepath, self->initial_socket_path, card_path, NULL };
             execvp(args[0], (char *const*)args);
         }
-        fprintf(stderr, "gsr error: gsr_kms_client_init: execvp failed, error: %s\n", strerror(errno));
+        fprintf(stderr, "gsr error: gsr_kms_client_init: failed to launch \"gsr-kms-server\", error: %s\n", strerror(errno));
         _exit(127);
     } else { /* parent */
         self->kms_server_pid = pid;
     }
 
+    // We need this dumb-shit retardation with unix domain socket and then replace it with socketpair because
+    // pkexec doesn't work with socketpair................
     fprintf(stderr, "gsr info: gsr_kms_client_init: waiting for server to connect\n");
     struct pollfd poll_fd = {
         .fd = self->initial_socket_fd,
@@ -348,6 +378,7 @@ int gsr_kms_client_init(gsr_kms_client *self, const char *card_path) {
                 goto err;
             }
         }
+        poll_fd.revents = 0;
     }
     fprintf(stderr, "gsr info: gsr_kms_client_init: server connected\n");
 
@@ -387,6 +418,7 @@ void cleanup_socket(gsr_kms_client *self, bool kill_server) {
 
     if(kill_server && self->kms_server_pid > 0) {
         kill(self->kms_server_pid, SIGKILL);
+        // TODO:
         //int status;
         //waitpid(self->kms_server_pid, &status, 0);
         self->kms_server_pid = -1;
@@ -439,6 +471,7 @@ int gsr_kms_client_get_kms(gsr_kms_client *self, gsr_kms_response *response) {
     response->version = 0;
     response->result = KMS_RESULT_FAILED_TO_SEND;
     response->err_msg[0] = '\0';
+    response->num_items = 0;
 
     gsr_kms_request request;
     request.version = GSR_KMS_PROTOCOL_VERSION;

kms/kms_shared.h

@@ -5,7 +5,7 @@
 #include <stdbool.h>
 #include <drm_mode.h>
 
-#define GSR_KMS_PROTOCOL_VERSION 4
+#define GSR_KMS_PROTOCOL_VERSION 5
 
 #define GSR_KMS_MAX_ITEMS 8
 #define GSR_KMS_MAX_DMA_BUFS 4
@@ -39,6 +39,13 @@ struct gsr_kms_response_dma_buf {
     uint32_t offset;
 };
 
+typedef enum {
+    KMS_ROT_0,
+    KMS_ROT_90,
+    KMS_ROT_180,
+    KMS_ROT_270
+} gsr_kms_rotation;
+
 struct gsr_kms_response_item {
     gsr_kms_response_dma_buf dma_buf[GSR_KMS_MAX_DMA_BUFS];
     int num_dma_bufs;
@@ -49,6 +56,7 @@ struct gsr_kms_response_item {
     uint32_t connector_id; /* 0 if unknown */
     bool is_cursor;
     bool has_hdr_metadata;
+    gsr_kms_rotation rotation;
     int x;
     int y;
     int src_w;

kms/server/kms_server.c

@@ -7,7 +7,7 @@
 #include <stdio.h>
 #include <string.h>
 #include <errno.h>
-#include <stdlib.h>
+#include <locale.h>
 
 #include <unistd.h>
 #include <limits.h>
@@ -144,16 +144,18 @@ typedef enum {
     PLANE_PROPERTY_SRC_H      = 1 << 5,
     PLANE_PROPERTY_IS_CURSOR  = 1 << 6,
     PLANE_PROPERTY_IS_PRIMARY = 1 << 7,
+    PLANE_PROPERTY_ROTATION   = 1 << 8,
 } plane_property_mask;
 
 /* Returns plane_property_mask */
-static uint32_t plane_get_properties(int drmfd, uint32_t plane_id, int *x, int *y, int *src_x, int *src_y, int *src_w, int *src_h) {
+static uint32_t plane_get_properties(int drmfd, uint32_t plane_id, int *x, int *y, int *src_x, int *src_y, int *src_w, int *src_h, gsr_kms_rotation *rotation) {
     *x = 0;
     *y = 0;
     *src_x = 0;
     *src_y = 0;
     *src_w = 0;
     *src_h = 0;
+    *rotation = KMS_ROT_0;
 
     plane_property_mask property_mask = 0;
 
@@ -198,6 +200,15 @@ static uint32_t plane_get_properties(int drmfd, uint32_t plane_id, int *x, int *
                     break;
                 }
             }
+        } else if((type & DRM_MODE_PROP_BITMASK) && strcmp(prop->name, "rotation") == 0) {
+            const uint64_t rotation_bitmask = props->prop_values[i];
+            *rotation = KMS_ROT_0;
+            if(rotation_bitmask & 2)
+                *rotation = (*rotation + KMS_ROT_90) % 4;
+            if(rotation_bitmask & 4)
+                *rotation = (*rotation + KMS_ROT_180) % 4;
+            if(rotation_bitmask & 8)
+                *rotation = (*rotation + KMS_ROT_270) % 4;
         }
 
         drmModeFreeProperty(prop);
@@ -207,7 +218,7 @@ static uint32_t plane_get_properties(int drmfd, uint32_t plane_id, int *x, int *
     return property_mask;
 }
 
-/* Returns 0 if not found */
+/* Returns NULL if not found */
 static const connector_crtc_pair* get_connector_pair_by_crtc_id(const connector_to_crtc_map *c2crtc_map, uint32_t crtc_id) {
     for(int i = 0; i < c2crtc_map->num_maps; ++i) {
         if(c2crtc_map->maps[i].crtc_id == crtc_id)
@@ -340,7 +351,8 @@ static int kms_get_fb(gsr_drm *drm, gsr_kms_response *response) {
         // TODO: Support other plane formats than rgb (with multiple planes, such as direct YUV420 on wayland).
 
         int x = 0, y = 0, src_x = 0, src_y = 0, src_w = 0, src_h = 0;
-        plane_property_mask property_mask = plane_get_properties(drm->drmfd, plane->plane_id, &x, &y, &src_x, &src_y, &src_w, &src_h);
+        gsr_kms_rotation rotation = KMS_ROT_0;
+        const uint32_t property_mask = plane_get_properties(drm->drmfd, plane->plane_id, &x, &y, &src_x, &src_y, &src_w, &src_h, &rotation);
         if(!(property_mask & PLANE_PROPERTY_IS_PRIMARY) && !(property_mask & PLANE_PROPERTY_IS_CURSOR))
             continue;
 
@@ -374,6 +386,7 @@ static int kms_get_fb(gsr_drm *drm, gsr_kms_response *response) {
         response->items[item_index].pixel_format = drmfb->pixel_format;
         response->items[item_index].modifier = drmfb->flags & DRM_MODE_FB_MODIFIERS ? drmfb->modifier : DRM_FORMAT_MOD_INVALID;
         response->items[item_index].connector_id = crtc_pair ? crtc_pair->connector_id : 0;
+        response->items[item_index].rotation = rotation;
         response->items[item_index].is_cursor = property_mask & PLANE_PROPERTY_IS_CURSOR;
         if(property_mask & PLANE_PROPERTY_IS_CURSOR) {
             response->items[item_index].x = x;
@@ -433,82 +446,9 @@ static double clock_get_monotonic_seconds(void) {
     return (double)ts.tv_sec + (double)ts.tv_nsec * 0.000000001;
 }
 
-// static bool readlink_realpath(const char *filepath, char *buffer) {
-//     char symlinked_path[PATH_MAX];
-//     ssize_t bytes_written = readlink(filepath, symlinked_path, sizeof(symlinked_path) - 1);
-//     if(bytes_written == -1 && errno == EINVAL) {
-//         /* Not a symlink */
-//         snprintf(symlinked_path, sizeof(symlinked_path), "%s", filepath);
-//     } else if(bytes_written == -1) {
-//         return false;
-//     } else {
-//         symlinked_path[bytes_written] = '\0';
-//     }
-
-//     if(!realpath(symlinked_path, buffer))
-//         return false;
-
-//     return true;
-// }
-
-// static void file_get_directory(char *filepath) {
-//     char *end = strrchr(filepath, '/');
-//     if(end == NULL)
-//         filepath[0] = '\0';
-//     else
-//         *end = '\0';
-// }
-
-// static bool string_ends_with(const char *str, const char *ends_with) {
-//     const int len = strlen(str);
-//     const int ends_with_len = strlen(ends_with);
-//     return len >= ends_with_len && memcmp(str + len - ends_with_len, ends_with, ends_with_len) == 0;
-// }
-
-// This is not foolproof, but the assumption is that gsr-kms-server and gpu-screen-recorder are installed in the same directory
-// in a location that only the root user can write to (usually /usr/bin or /usr/local/bin) and if the client runs from that location
-// and is called gpu-screen-recorder then gsr-kms-server can only be used by a malicious program if the malicious program
-// had root access, to modify that program install directory.
-// static bool is_remote_peer_program_gpu_screen_recorder(int socket_fd) {
-//     // TODO: Use SO_PEERPIDFD on kernel >= 6.5 to avoid a race condition in the /proc/<pid> check
-//     struct ucred cred;
-//     socklen_t ucred_len = sizeof(cred);
-//     if(getsockopt(socket_fd, SOL_SOCKET, SO_PEERCRED, &cred, &ucred_len) == -1) {
-//         fprintf(stderr, "kms server error: failed to get peer credentials, error: %s\n", strerror(errno));
-//         return false;
-//     }
-
-//     char self_directory[PATH_MAX];
-//     if(!readlink_realpath("/proc/self/exe", self_directory)) {
-//         fprintf(stderr, "kms server error: failed to resolve /proc/self/exe\n");
-//         return false;
-//     }
-//     file_get_directory(self_directory);
-
-//     char peer_directory[PATH_MAX];
-//     char peer_exe_path[PATH_MAX];
-//     snprintf(peer_exe_path, sizeof(peer_exe_path), "/proc/%d/exe", (int)cred.pid);
-//     if(!readlink_realpath(peer_exe_path, peer_directory)) {
-//         fprintf(stderr, "kms server error: failed to resolve /proc/self/exe\n");
-//         return false;
-//     }
-
-//     if(!string_ends_with(peer_directory, "/gpu-screen-recorder")) {
-//         fprintf(stderr, "kms server error: only gpu-screen-recorder can use gsr-kms-server. client program location is %s\n", peer_directory);
-//         return false;
-//     }
-
-//     file_get_directory(peer_directory);
-
-//     if(strcmp(self_directory, peer_directory) != 0) {
-//         fprintf(stderr, "kms server error: the client program is in directory %s but only programs in %s can run gsr-kms-server\n", peer_directory, self_directory);
-//         return false;
-//     }
-
-//     return true;
-// }
-
 int main(int argc, char **argv) {
+    setlocale(LC_ALL, "C"); // Sigh... stupid C
+
     int res = 0;
     int socket_fd = 0;
     gsr_drm drm;
@@ -579,11 +519,6 @@ int main(int argc, char **argv) {
         goto done;
     }
 
-    // if(!is_remote_peer_program_gpu_screen_recorder(socket_fd)) {
-    //     res = 3;
-    //     goto done;
-    // }
-
     for(;;) {
         gsr_kms_request request;
         request.version = 0;

meson.build

@@ -1,4 +1,4 @@
-project('gpu-screen-recorder', ['c', 'cpp'], version : '5.3.3', default_options : ['warning_level=2'])
+project('gpu-screen-recorder', ['c', 'cpp'], version : '5.12.5', default_options : ['warning_level=2'])
 
 add_project_arguments('-Wshadow', language : ['c', 'cpp'])
 if get_option('buildtype') == 'debug'
@@ -14,6 +14,8 @@ src = [
     'src/capture/xcomposite.c',
     'src/capture/ximage.c',
     'src/capture/kms.c',
+    'src/capture/v4l2.c',
+    'src/encoder/encoder.c',
     'src/encoder/video/video.c',
     'src/encoder/video/nvenc.c',
     'src/encoder/video/vaapi.c',
@@ -25,6 +27,9 @@ src = [
     'src/window/window.c',
     'src/window/x11.c',
     'src/window/wayland.c',
+    'src/replay_buffer/replay_buffer.c',
+    'src/replay_buffer/replay_buffer_ram.c',
+    'src/replay_buffer/replay_buffer_disk.c',
     'src/egl.c',
     'src/cuda.c',
     'src/xnvctrl.c',
@@ -37,12 +42,22 @@ src = [
     'src/cursor.c',
     'src/damage.c',
     'src/image_writer.c',
+    'src/args_parser.c',
+    'src/defs.c',
+    'src/plugins.c',
     'src/sound.cpp',
     'src/main.cpp',
 ]
 
+subdir('protocol')
+src += protocol_src
+
+cc = meson.get_compiler('c')
+m_dep = cc.find_library('m', required : true)
+
 dep = [
     dependency('threads'),
+    m_dep,
     dependency('libavcodec'),
     dependency('libavformat'),
     dependency('libavutil'),
@@ -56,12 +71,17 @@ dep = [
     dependency('libavfilter'),
     dependency('libva'),
     dependency('libva-drm'),
-    dependency('libcap'),
     dependency('libdrm'),
     dependency('wayland-egl'),
     dependency('wayland-client'),
 ]
 
+if build_machine.system() == 'linux'
+    dep += [
+        dependency('libcap'),
+    ]
+endif
+
 uses_pipewire = false
 
 if get_option('portal') == true
@@ -70,7 +90,6 @@ if get_option('portal') == true
         'src/dbus.c',
         'src/pipewire_video.c',
     ]
-    dep += dependency('dbus-1')
     add_project_arguments('-DGSR_PORTAL', language : ['c', 'cpp'])
     uses_pipewire = true
 endif
@@ -87,6 +106,7 @@ if uses_pipewire == true
     dep += [
         dependency('libpipewire-0.3'),
         dependency('libspa-0.2'),
+        dependency('dbus-1'),
     ]
 endif
 
@@ -95,6 +115,10 @@ add_project_arguments('-DGSR_VERSION="' + meson.project_version() + '"', languag
 executable('gsr-kms-server', 'kms/server/kms_server.c', dependencies : dependency('libdrm'), c_args : '-fstack-protector-all', install : true)
 executable('gpu-screen-recorder', src, dependencies : dep, install : true)
 
+install_headers('plugin/plugin.h', install_dir : 'include/gsr')
+install_man('gpu-screen-recorder.1', 'gsr-kms-server.1')
+install_subdir('scripts', install_dir: 'share/gpu-screen-recorder')
+
 if get_option('systemd') == true
     install_data(files('extra/gpu-screen-recorder.service'), install_dir : 'lib/systemd/user')
 endif
@@ -106,3 +130,10 @@ endif
 if get_option('nvidia_suspend_fix') == true
     install_data(files('extra/gsr-nvidia.conf'), install_dir : 'lib/modprobe.d')
 endif
+
+if get_option('plugin_examples') == true
+    shared_library('triangle', 'plugin/examples/hello_triangle/triangle.c',
+        dependencies: [dependency('gl'), m_dep],
+        name_prefix : '',
+        name_suffix: 'so')
+endif

meson_options.txt

@@ -1,5 +1,6 @@
 option('systemd', type : 'boolean', value : true, description : 'Install systemd service file')
 option('capabilities', type : 'boolean', value : true, description : 'Set binary admin capability on gsr-kms-server binary to remove password prompt when recording monitor (without desktop portal option) on amd/intel or nvidia wayland')
 option('nvidia_suspend_fix', type : 'boolean', value : true, description : 'Install nvidia modprobe config file to tell nvidia driver to preserve video memory on suspend. This is a workaround for an nvidia driver bug that breaks cuda (and gpu screen recorder) on suspend')
-option('portal', type : 'boolean', value : true, description : 'Build with support for xdg desktop portal ScreenCast capture (wayland only) (-w portal option)')
-option('app_audio', type : 'boolean', value : true, description : 'Build with support for recording a single audio source (-aa option). Requires pipewire')
+option('portal', type : 'boolean', value : true, description : 'Build with support for xdg desktop portal ScreenCast capture (wayland only) (-w portal option). Requires pipewire')
+option('app_audio', type : 'boolean', value : true, description : 'Build with support for recording a single audio source (-a app: option). Requires pipewire')
+option('plugin_examples', type : 'boolean', value : false, description : 'Build plugin examples')

plugin/examples/hello_triangle/triangle.c

@@ -0,0 +1,108 @@
+#include <gsr/plugin.h>
+#include <stddef.h>
+#include <stdlib.h>
+#include <math.h>
+#define GL_GLEXT_PROTOTYPES
+#include <GL/gl.h>
+
+const char vertex_shader_source[] =
+    "attribute vec4 vertex_pos;   \n"
+    "void main() {                \n"
+    "   gl_Position = vertex_pos; \n"
+    "}";
+
+const char fragment_shader_source[] =
+    "precision mediump float;                    \n"
+    "uniform vec3 color;                         \n"
+    "void main() {                               \n"
+    "   gl_FragColor = vec4(color, 1.0); \n"
+    "}";
+
+typedef struct {
+    GLuint shader_program;
+    GLuint vao;
+    GLuint vbo;
+    GLint color_uniform;
+    unsigned int counter;
+} Triangle;
+
+static GLuint load_shader(const char *shaderSrc, GLenum type) {
+    GLuint shader = glCreateShader(type);
+    glShaderSource(shader, 1, &shaderSrc, NULL);
+    glCompileShader(shader);
+    return shader;
+}
+
+static void draw(const gsr_plugin_draw_params *params, void *userdata) {
+    Triangle *triangle = userdata;
+
+    GLfloat glverts[6];
+
+    glverts[0] = -0.5f;
+    glverts[1] = -0.5f;
+
+    glverts[2] = 0.5f;
+    glverts[3] = -0.5f;
+
+    glverts[4] = 0.0f;
+    glverts[5] = 0.5f;
+
+    glBindVertexArray(triangle->vao);
+    glBindBuffer(GL_ARRAY_BUFFER, triangle->vbo);
+    glBufferSubData(GL_ARRAY_BUFFER, 0, 6 * sizeof(float), glverts);
+
+    glUseProgram(triangle->shader_program);
+    const double pp = triangle->counter * 0.05;
+    glUniform3f(triangle->color_uniform, 0.5 + sin(pp)*0.5, 0.5 + cos(pp)*0.5, 0.5 + sin(0.2 + pp)*0.5);
+    glDrawArrays(GL_TRIANGLES, 0, 3);
+
+    glBindVertexArray(0);
+    glUseProgram(0);
+
+    ++triangle->counter;
+}
+
+bool gsr_plugin_init(const gsr_plugin_init_params *params, gsr_plugin_init_return *ret) {
+    Triangle *triangle = calloc(1, sizeof(Triangle));
+    if(!triangle)
+        return false;
+
+    triangle->shader_program = glCreateProgram();
+
+    const GLuint vertex_shader = load_shader(vertex_shader_source, GL_VERTEX_SHADER);
+    const GLuint fragment_shader = load_shader(fragment_shader_source, GL_FRAGMENT_SHADER);
+
+    glAttachShader(triangle->shader_program, vertex_shader);
+    glAttachShader(triangle->shader_program, fragment_shader);
+    glBindAttribLocation(triangle->shader_program, 0, "vertex_pos");
+    glLinkProgram(triangle->shader_program);
+
+    glGenVertexArrays(1, &triangle->vao);
+    glBindVertexArray(triangle->vao);
+
+    glGenBuffers(1, &triangle->vbo);
+    glBindBuffer(GL_ARRAY_BUFFER, triangle->vbo);
+    glBufferData(GL_ARRAY_BUFFER, 6 * sizeof(float), NULL, GL_DYNAMIC_DRAW);
+
+    glEnableVertexAttribArray(0);
+    glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 2 * sizeof(float), (void *)0);
+
+    glBindVertexArray(0);
+
+    glDeleteShader(vertex_shader);
+    glDeleteShader(fragment_shader);
+
+    triangle->color_uniform = glGetUniformLocation(triangle->shader_program, "color");
+
+    ret->name = "hello_triangle";
+    ret->version = 1;
+    ret->userdata = triangle;
+    ret->draw = draw;
+    return true;
+}
+
+void gsr_plugin_deinit(void *userdata) {
+    Triangle *triangle = userdata;
+    glDeleteProgram(triangle->shader_program);
+    free(triangle);
+}

plugin/plugin.h

@@ -0,0 +1,56 @@
+#ifndef GSR_PLUGIN_H
+#define GSR_PLUGIN_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define GSR_PLUGIN_INTERFACE_MAJOR_VERSION 0
+#define GSR_PLUGIN_INTERFACE_MINOR_VERSION 1
+
+#define GSR_PLUGIN_INTERFACE_MAKE_VERSION(major, minor) (((major) << 16) | (minor))
+#define GSR_PLUGIN_INTERFACE_VERSION GSR_PLUGIN_INTERFACE_MAKE_VERSION(GSR_PLUGIN_INTERFACE_MAJOR_VERSION, GSR_PLUGIN_INTERFACE_MINOR_VERSION)
+
+#include <stdbool.h>
+
+typedef enum {
+    GSR_PLUGIN_GRAPHICS_API_EGL_ES,
+    GSR_PLUGIN_GRAPHICS_API_GLX,
+} gsr_plugin_graphics_api;
+
+typedef enum {
+    GSR_PLUGIN_COLOR_DEPTH_8_BITS,
+    GSR_PLUGIN_COLOR_DEPTH_10_BITS,
+} gsr_plugin_color_depth;
+
+typedef struct {
+    unsigned int width;
+    unsigned int height;
+} gsr_plugin_draw_params;
+
+typedef struct {
+    unsigned int width;
+    unsigned int height;
+    unsigned int fps;
+    gsr_plugin_color_depth color_depth;
+    gsr_plugin_graphics_api graphics_api;
+} gsr_plugin_init_params;
+
+typedef struct {
+    const char *name; /* Mandatory */
+    unsigned int version; /* Mandatory, can't be 0 */
+    void *userdata; /* Optional */
+
+    /* Optional, called when the plugin is expected to draw something to the current framebuffer */
+    void (*draw)(const gsr_plugin_draw_params *params, void *userdata);
+} gsr_plugin_init_return;
+
+/* The plugin is expected to implement these functions and export them: */
+bool gsr_plugin_init(const gsr_plugin_init_params *params, gsr_plugin_init_return *ret);
+void gsr_plugin_deinit(void *userdata);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* GSR_PLUGIN_H */

project.conf

@@ -1,11 +1,11 @@
 [package]
 name = "gpu-screen-recorder"
 type = "executable"
-version = "5.3.3"
+version = "5.12.5"
 platforms = ["posix"]
 
 [config]
-ignore_dirs = ["kms/server", "build", "debug-build"]
+ignore_dirs = ["kms/server", "build", "debug-build", "plugin/examples"]
 #error_on_warning = "true"
 
 [define]
@@ -33,4 +33,4 @@ wayland-client = ">=1"
 dbus-1 = ">=1"
 libpipewire-0.3 = ">=1"
 libspa-0.2 = ">=0"
-#vulkan = ">=1"
+vulkan = ">=1"

protocol/meson.build

@@ -0,0 +1,25 @@
+wayland_scanner = dependency('wayland-scanner', native: true)
+wayland_scanner_path = wayland_scanner.get_variable(pkgconfig: 'wayland_scanner')
+wayland_scanner_prog = find_program(wayland_scanner_path, native: true)
+
+wayland_scanner_code = generator(
+	wayland_scanner_prog,
+	output: '@BASENAME@-protocol.c',
+	arguments: ['private-code', '@INPUT@', '@OUTPUT@'],
+)
+
+wayland_scanner_client = generator(
+	wayland_scanner_prog,
+	output: '@BASENAME@-client-protocol.h',
+	arguments: ['client-header', '@INPUT@', '@OUTPUT@'],
+)
+
+protocols = [
+	'xdg-output-unstable-v1.xml',
+]
+
+protocol_src = []
+foreach xml : protocols
+	protocol_src += wayland_scanner_code.process(xml)
+	protocol_src += wayland_scanner_client.process(xml)
+endforeach

protocol/xdg-output-unstable-v1.xml

@@ -0,0 +1,222 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<protocol name="xdg_output_unstable_v1">
+
+  <copyright>
+    Copyright © 2017 Red Hat Inc.
+
+    Permission is hereby granted, free of charge, to any person obtaining a
+    copy of this software and associated documentation files (the "Software"),
+    to deal in the Software without restriction, including without limitation
+    the rights to use, copy, modify, merge, publish, distribute, sublicense,
+    and/or sell copies of the Software, and to permit persons to whom the
+    Software is furnished to do so, subject to the following conditions:
+
+    The above copyright notice and this permission notice (including the next
+    paragraph) shall be included in all copies or substantial portions of the
+    Software.
+
+    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+    THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+    FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+    DEALINGS IN THE SOFTWARE.
+  </copyright>
+
+  <description summary="Protocol to describe output regions">
+    This protocol aims at describing outputs in a way which is more in line
+    with the concept of an output on desktop oriented systems.
+
+    Some information are more specific to the concept of an output for
+    a desktop oriented system and may not make sense in other applications,
+    such as IVI systems for example.
+
+    Typically, the global compositor space on a desktop system is made of
+    a contiguous or overlapping set of rectangular regions.
+
+    The logical_position and logical_size events defined in this protocol
+    might provide information identical to their counterparts already
+    available from wl_output, in which case the information provided by this
+    protocol should be preferred to their equivalent in wl_output. The goal is
+    to move the desktop specific concepts (such as output location within the
+    global compositor space, etc.) out of the core wl_output protocol.
+
+    Warning! The protocol described in this file is experimental and
+    backward incompatible changes may be made. Backward compatible
+    changes may be added together with the corresponding interface
+    version bump.
+    Backward incompatible changes are done by bumping the version
+    number in the protocol and interface names and resetting the
+    interface version. Once the protocol is to be declared stable,
+    the 'z' prefix and the version number in the protocol and
+    interface names are removed and the interface version number is
+    reset.
+  </description>
+
+  <interface name="zxdg_output_manager_v1" version="3">
+    <description summary="manage xdg_output objects">
+      A global factory interface for xdg_output objects.
+    </description>
+
+    <request name="destroy" type="destructor">
+      <description summary="destroy the xdg_output_manager object">
+	Using this request a client can tell the server that it is not
+	going to use the xdg_output_manager object anymore.
+
+	Any objects already created through this instance are not affected.
+      </description>
+    </request>
+
+    <request name="get_xdg_output">
+      <description summary="create an xdg output from a wl_output">
+	This creates a new xdg_output object for the given wl_output.
+      </description>
+      <arg name="id" type="new_id" interface="zxdg_output_v1"/>
+      <arg name="output" type="object" interface="wl_output"/>
+    </request>
+  </interface>
+
+  <interface name="zxdg_output_v1" version="3">
+    <description summary="compositor logical output region">
+      An xdg_output describes part of the compositor geometry.
+
+      This typically corresponds to a monitor that displays part of the
+      compositor space.
+
+      For objects version 3 onwards, after all xdg_output properties have been
+      sent (when the object is created and when properties are updated), a
+      wl_output.done event is sent. This allows changes to the output
+      properties to be seen as atomic, even if they happen via multiple events.
+    </description>
+
+    <request name="destroy" type="destructor">
+      <description summary="destroy the xdg_output object">
+	Using this request a client can tell the server that it is not
+	going to use the xdg_output object anymore.
+      </description>
+    </request>
+
+    <event name="logical_position">
+      <description summary="position of the output within the global compositor space">
+	The position event describes the location of the wl_output within
+	the global compositor space.
+
+	The logical_position event is sent after creating an xdg_output
+	(see xdg_output_manager.get_xdg_output) and whenever the location
+	of the output changes within the global compositor space.
+      </description>
+      <arg name="x" type="int"
+	   summary="x position within the global compositor space"/>
+      <arg name="y" type="int"
+	   summary="y position within the global compositor space"/>
+    </event>
+
+    <event name="logical_size">
+      <description summary="size of the output in the global compositor space">
+	The logical_size event describes the size of the output in the
+	global compositor space.
+
+	Most regular Wayland clients should not pay attention to the
+	logical size and would rather rely on xdg_shell interfaces.
+
+	Some clients such as Xwayland, however, need this to configure
+	their surfaces in the global compositor space as the compositor
+	may apply a different scale from what is advertised by the output
+	scaling property (to achieve fractional scaling, for example).
+
+	For example, for a wl_output mode 3840×2160 and a scale factor 2:
+
+	- A compositor not scaling the monitor viewport in its compositing space
+	  will advertise a logical size of 3840×2160,
+
+	- A compositor scaling the monitor viewport with scale factor 2 will
+	  advertise a logical size of 1920×1080,
+
+	- A compositor scaling the monitor viewport using a fractional scale of
+	  1.5 will advertise a logical size of 2560×1440.
+
+	For example, for a wl_output mode 1920×1080 and a 90 degree rotation,
+	the compositor will advertise a logical size of 1080x1920.
+
+	The logical_size event is sent after creating an xdg_output
+	(see xdg_output_manager.get_xdg_output) and whenever the logical
+	size of the output changes, either as a result of a change in the
+	applied scale or because of a change in the corresponding output
+	mode(see wl_output.mode) or transform (see wl_output.transform).
+      </description>
+      <arg name="width" type="int"
+	   summary="width in global compositor space"/>
+      <arg name="height" type="int"
+	   summary="height in global compositor space"/>
+    </event>
+
+    <event name="done">
+      <description summary="all information about the output have been sent">
+	This event is sent after all other properties of an xdg_output
+	have been sent.
+
+	This allows changes to the xdg_output properties to be seen as
+	atomic, even if they happen via multiple events.
+
+	For objects version 3 onwards, this event is deprecated. Compositors
+	are not required to send it anymore and must send wl_output.done
+	instead.
+      </description>
+    </event>
+
+    <!-- Version 2 additions -->
+
+    <event name="name" since="2">
+      <description summary="name of this output">
+	Many compositors will assign names to their outputs, show them to the
+	user, allow them to be configured by name, etc. The client may wish to
+	know this name as well to offer the user similar behaviors.
+
+	The naming convention is compositor defined, but limited to
+	alphanumeric characters and dashes (-). Each name is unique among all
+	wl_output globals, but if a wl_output global is destroyed the same name
+	may be reused later. The names will also remain consistent across
+	sessions with the same hardware and software configuration.
+
+	Examples of names include 'HDMI-A-1', 'WL-1', 'X11-1', etc. However, do
+	not assume that the name is a reflection of an underlying DRM
+	connector, X11 connection, etc.
+
+	The name event is sent after creating an xdg_output (see
+	xdg_output_manager.get_xdg_output). This event is only sent once per
+	xdg_output, and the name does not change over the lifetime of the
+	wl_output global.
+
+        This event is deprecated, instead clients should use wl_output.name.
+        Compositors must still support this event.
+      </description>
+      <arg name="name" type="string" summary="output name"/>
+    </event>
+
+    <event name="description" since="2">
+      <description summary="human-readable description of this output">
+	Many compositors can produce human-readable descriptions of their
+	outputs.  The client may wish to know this description as well, to
+	communicate the user for various purposes.
+
+	The description is a UTF-8 string with no convention defined for its
+	contents. Examples might include 'Foocorp 11" Display' or 'Virtual X11
+	output via :1'.
+
+	The description event is sent after creating an xdg_output (see
+	xdg_output_manager.get_xdg_output) and whenever the description
+	changes. The description is optional, and may not be sent at all.
+
+	For objects of version 2 and lower, this event is only sent once per
+	xdg_output, and the description does not change over the lifetime of
+	the wl_output global.
+
+	This event is deprecated, instead clients should use
+	wl_output.description. Compositors must still support this event.
+      </description>
+      <arg name="description" type="string" summary="output description"/>
+    </event>
+
+  </interface>
+</protocol>

scripts/replay-application-name.sh

@@ -3,4 +3,4 @@
 window=$(xdotool selectwindow)
 window_name=$(xdotool getwindowname "$window" || xdotool getwindowclassname "$window" || echo "Game")
 window_name="$(echo "$window_name" | tr '/\\' '_')"
-gpu-screen-recorder -w "$window" -f 60 -c mkv -a default_output -bm cbr -q 45000 -r 60 -o "$HOME/Videos/Replays/$window_name"
+gpu-screen-recorder -w "$window" -f 60 -c mkv -a default_output -bm cbr -q 40000 -r 60 -o "$HOME/Videos/Replays/$window_name"

scripts/start-replay.sh

@@ -3,4 +3,4 @@
 pidof -q gpu-screen-recorder && exit 0
 video_path="$HOME/Videos"
 mkdir -p "$video_path"
-gpu-screen-recorder -w screen -f 60 -a default_output -c mkv -bm cbr -q 45000 -r 30 -o "$video_path"
+gpu-screen-recorder -w screen -f 60 -a default_output -c mkv -bm cbr -q 40000 -r 30 -o "$video_path"

scripts/toggle-recording.sh

@@ -0,0 +1,6 @@
+#!/bin/sh -e
+
+killall -SIGINT gpu-screen-recorder && sleep 0.5 && notify-send -t 1500 -u low 'GPU Screen Recorder' 'Stopped recording' && exit 0;
+video="$HOME/Videos/$(date +"Video_%Y-%m-%d_%H-%M-%S.mp4")"
+notify-send -t 1500 -u low 'GPU Screen Recorder' "Started recording video to $video"
+gpu-screen-recorder -w screen -f 60 -a "default_output" -o "$video"

scripts/twitch-stream-local-copy.sh

@@ -4,4 +4,4 @@
 
 [ "$#" -ne 4 ] && echo "usage: twitch-stream-local-copy.sh <window_id> <fps> <livestream_key> <local_file>" && exit 1
 active_sink=default_output
-gpu-screen-recorder -w "$1" -c flv -f "$2" -q high -a "$active_sink" | tee -- "$4" | ffmpeg -i pipe:0 -c copy -f flv -- "rtmp://live.twitch.tv/app/$3"
+gpu-screen-recorder -w "$1" -c flv -f "$2" -bm cbr -q 8000 -a "$active_sink" | tee -- "$4" | ffmpeg -i pipe:0 -c copy -f flv -- "rtmp://live.twitch.tv/app/$3"

scripts/twitch-stream.sh

@@ -2,4 +2,4 @@
 
 [ "$#" -ne 3 ] && echo "usage: twitch-stream.sh <window_id> <fps> <livestream_key>" && exit 1
 active_sink=default_output
-gpu-screen-recorder -w "$1" -c flv -f "$2" -q high -a "$active_sink" -o "rtmp://live.twitch.tv/app/$3"
+gpu-screen-recorder -w "$1" -c flv -f "$2" -bm cbr -q 8000 -a "$active_sink" -o "rtmp://live.twitch.tv/app/$3"

scripts/youtube-hls-stream.sh

@@ -2,4 +2,4 @@
 
 [ "$#" -ne 3 ] && echo "usage: youtube-hls-stream.sh <window_id> <fps> <livestream_key>" && exit 1
 active_sink=default_output
-gpu-screen-recorder -w "$1" -c hls -f "$2" -q high -a "$active_sink" -ac aac -o "https://a.upload.youtube.com/http_upload_hls?cid=$3&copy=0&file=stream.m3u8"
+gpu-screen-recorder -w "$1" -c hls -f "$2" -bm cbr -q 8000 -a "$active_sink" -ac aac -o "https://a.upload.youtube.com/http_upload_hls?cid=$3&copy=0&file=stream.m3u8"

src/args_parser.c

@@ -0,0 +1,722 @@
+#include "../include/args_parser.h"
+#include "../include/defs.h"
+#include "../include/egl.h"
+#include "../include/window/window.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <inttypes.h>
+#include <limits.h>
+#include <assert.h>
+#include <libgen.h>
+#include <sys/stat.h>
+
+#ifndef GSR_VERSION
+#define GSR_VERSION "unknown"
+#endif
+
+static const ArgEnum video_codec_enums[] = {
+    { .name = "auto",        .value = GSR_VIDEO_CODEC_AUTO       },
+    { .name = "h264",        .value = GSR_VIDEO_CODEC_H264       },
+    { .name = "h265",        .value = GSR_VIDEO_CODEC_HEVC       },
+    { .name = "hevc",        .value = GSR_VIDEO_CODEC_HEVC       },
+    { .name = "hevc_hdr",    .value = GSR_VIDEO_CODEC_HEVC_HDR   },
+    { .name = "hevc_10bit",  .value = GSR_VIDEO_CODEC_HEVC_10BIT },
+    { .name = "av1",         .value = GSR_VIDEO_CODEC_AV1        },
+    { .name = "av1_hdr",     .value = GSR_VIDEO_CODEC_AV1_HDR    },
+    { .name = "av1_10bit",   .value = GSR_VIDEO_CODEC_AV1_10BIT  },
+    { .name = "vp8",         .value = GSR_VIDEO_CODEC_VP8        },
+    { .name = "vp9",         .value = GSR_VIDEO_CODEC_VP9        },
+};
+
+static const ArgEnum audio_codec_enums[] = {
+    { .name = "opus", .value = GSR_AUDIO_CODEC_OPUS },
+    { .name = "aac",  .value = GSR_AUDIO_CODEC_AAC  },
+    { .name = "flac", .value = GSR_AUDIO_CODEC_FLAC },
+};
+
+static const ArgEnum video_encoder_enums[] = {
+    { .name = "gpu", .value = GSR_VIDEO_ENCODER_HW_GPU },
+    { .name = "cpu", .value = GSR_VIDEO_ENCODER_HW_CPU },
+};
+
+static const ArgEnum pixel_format_enums[] = {
+    { .name = "yuv420", .value = GSR_PIXEL_FORMAT_YUV420 },
+    { .name = "yuv444", .value = GSR_PIXEL_FORMAT_YUV444 },
+};
+
+static const ArgEnum framerate_mode_enums[] = {
+    { .name = "vfr",     .value = GSR_FRAMERATE_MODE_VARIABLE },
+    { .name = "cfr",     .value = GSR_FRAMERATE_MODE_CONSTANT },
+    { .name = "content", .value = GSR_FRAMERATE_MODE_CONTENT  },
+};
+
+static const ArgEnum bitrate_mode_enums[] = {
+    { .name = "auto", .value = GSR_BITRATE_MODE_AUTO },
+    { .name = "qp",   .value = GSR_BITRATE_MODE_QP   },
+    { .name = "cbr",  .value = GSR_BITRATE_MODE_CBR  },
+    { .name = "vbr",  .value = GSR_BITRATE_MODE_VBR  },
+};
+
+static const ArgEnum color_range_enums[] = {
+    { .name = "limited", .value = GSR_COLOR_RANGE_LIMITED },
+    { .name = "full",    .value = GSR_COLOR_RANGE_FULL    },
+};
+
+static const ArgEnum tune_enums[] = {
+    { .name = "performance", .value = GSR_TUNE_PERFORMANCE },
+    { .name = "quality",     .value = GSR_TUNE_QUALITY     },
+};
+
+static const ArgEnum replay_storage_enums[] = {
+    { .name = "ram",  .value = GSR_REPLAY_STORAGE_RAM  },
+    { .name = "disk", .value = GSR_REPLAY_STORAGE_DISK },
+};
+
+static void arg_deinit(Arg *arg) {
+    if(arg->values) {
+        free(arg->values);
+        arg->values = NULL;
+    }
+}
+
+static bool arg_append_value(Arg *arg, const char *value) {
+    if(arg->num_values + 1 >= arg->capacity_num_values) {
+        const int new_capacity_num_values = arg->capacity_num_values == 0 ? 4 : arg->capacity_num_values*2;
+        void *new_data = realloc(arg->values, new_capacity_num_values * sizeof(const char*));
+        if(!new_data)
+            return false;
+
+        arg->values = new_data;
+        arg->capacity_num_values = new_capacity_num_values;
+    }
+
+    arg->values[arg->num_values] = value;
+    ++arg->num_values;
+    return true;
+}
+
+static bool arg_get_enum_value_by_name(const Arg *arg, const char *name, int *enum_value) {
+    assert(arg->type == ARG_TYPE_ENUM);
+    assert(arg->enum_values);
+    for(int i = 0; i < arg->num_enum_values; ++i) {
+        if(strcmp(arg->enum_values[i].name, name) == 0) {
+            *enum_value = arg->enum_values[i].value;
+            return true;
+        }
+    }
+    return false;
+}
+
+static void arg_print_expected_enum_names(const Arg *arg) {
+    assert(arg->type == ARG_TYPE_ENUM);
+    assert(arg->enum_values);
+    for(int i = 0; i < arg->num_enum_values; ++i) {
+        if(i > 0) {
+            if(i == arg->num_enum_values -1)
+                fprintf(stderr, " or ");
+            else
+                fprintf(stderr, ", ");
+        }
+        fprintf(stderr, "'%s'", arg->enum_values[i].name);
+    }
+}
+
+static Arg* args_get_by_key(Arg *args, int num_args, const char *key) {
+    for(int i = 0; i < num_args; ++i) {
+        if(strcmp(args[i].key, key) == 0)
+            return &args[i];
+    }
+    return NULL;
+}
+
+static const char* args_get_value_by_key(Arg *args, int num_args, const char *key) {
+    for(int i = 0; i < num_args; ++i) {
+        if(strcmp(args[i].key, key) == 0) {
+            if(args[i].num_values == 0)
+                return NULL;
+            else
+                return args[i].values[0];
+        }
+    }
+    return NULL;
+}
+
+static bool args_get_boolean_by_key(Arg *args, int num_args, const char *key, bool default_value) {
+    Arg *arg = args_get_by_key(args, num_args, key);
+    assert(arg);
+    if(arg->num_values == 0) {
+        return default_value;
+    } else {
+        assert(arg->type == ARG_TYPE_BOOLEAN);
+        return arg->typed_value.boolean;
+    }
+}
+
+static int args_get_enum_by_key(Arg *args, int num_args, const char *key, int default_value) {
+    Arg *arg = args_get_by_key(args, num_args, key);
+    assert(arg);
+    if(arg->num_values == 0) {
+        return default_value;
+    } else {
+        assert(arg->type == ARG_TYPE_ENUM);
+        return arg->typed_value.enum_value;
+    }
+}
+
+static int64_t args_get_i64_by_key(Arg *args, int num_args, const char *key, int64_t default_value) {
+    Arg *arg = args_get_by_key(args, num_args, key);
+    assert(arg);
+    if(arg->num_values == 0) {
+        return default_value;
+    } else {
+        assert(arg->type == ARG_TYPE_I64);
+        return arg->typed_value.i64_value;
+    }
+}
+
+static double args_get_double_by_key(Arg *args, int num_args, const char *key, double default_value) {
+    Arg *arg = args_get_by_key(args, num_args, key);
+    assert(arg);
+    if(arg->num_values == 0) {
+        return default_value;
+    } else {
+        assert(arg->type == ARG_TYPE_DOUBLE);
+        return arg->typed_value.d_value;
+    }
+}
+
+static void usage_header(void) {
+    const bool inside_flatpak = getenv("FLATPAK_ID") != NULL;
+    const char *program_name = inside_flatpak ? "flatpak run --command=gpu-screen-recorder com.dec05eba.gpu_screen_recorder" : "gpu-screen-recorder";
+    printf("usage: %s -w <window_id|monitor|focused|portal|region|v4l2_device_path> [-c <container_format>] [-s WxH] [-region WxH+X+Y] [-f <fps>] [-a <audio_input>] "
+           "[-q <quality>] [-r <replay_buffer_size_sec>] [-replay-storage ram|disk] [-restart-replay-on-save yes|no] "
+           "[-k h264|hevc|av1|vp8|vp9|hevc_hdr|av1_hdr|hevc_10bit|av1_10bit] [-ac aac|opus|flac] [-ab <bitrate>] [-oc yes|no] [-fm cfr|vfr|content] "
+           "[-bm auto|qp|vbr|cbr] [-cr limited|full] [-tune performance|quality] [-df yes|no] [-sc <script_path>] [-p <plugin_path>] "
+           "[-cursor yes|no] [-keyint <value>] [-restore-portal-session yes|no] [-portal-session-token-filepath filepath] [-encoder gpu|cpu] "
+           "[-fallback-cpu-encoding yes|no] [-o <output_file>] [-ro <output_directory>] [-ffmpeg-opts <options>] [--list-capture-options [card_path]] "
+           "[--list-monitors] [--list-audio-devices] [--list-application-audio] [--list-v4l2-devices] [-write-first-frame-ts yes|no] [-low-power yes|no] "
+           "[-v yes|no] [-gl-debug yes|no] [--version] [-h|--help]\n", program_name);
+    fflush(stdout);
+}
+
+static void usage_full(void) {
+    const bool inside_flatpak = getenv("FLATPAK_ID") != NULL;
+    usage_header();
+    printf("\n");
+    printf("NOTES:\n");
+    if(inside_flatpak)
+        printf("  Run \"man /var/lib/flatpak/app/com.dec05eba.gpu_screen_recorder/current/active/files/share/man/man1/gpu-screen-recorder.1\" to open the man page for GPU Screen Recorder to see an explanation for each option and examples\n");
+    else
+        printf("  Run \"man gpu-screen-recorder.1\" to open the man page for GPU Screen Recorder to see an explanation for each option and examples\n");
+    fflush(stdout);
+}
+
+static void usage(void) {
+    usage_header();
+}
+
+// TODO: Does this match all livestreaming cases?
+static bool is_livestream_path(const char *str) {
+    const int len = strlen(str);
+    if((len >= 7 && memcmp(str, "http://", 7) == 0) || (len >= 8 && memcmp(str, "https://", 8) == 0))
+        return true;
+    else if((len >= 7 && memcmp(str, "rtmp://", 7) == 0) || (len >= 8 && memcmp(str, "rtmps://", 8) == 0))
+        return true;
+    else if((len >= 7 && memcmp(str, "rtsp://", 7) == 0))
+        return true;
+    else if((len >= 6 && memcmp(str, "srt://", 6) == 0))
+        return true;
+    else if((len >= 6 && memcmp(str, "tcp://", 6) == 0))
+        return true;
+    else if((len >= 6 && memcmp(str, "udp://", 6) == 0))
+        return true;
+    else
+        return false;
+}
+
+static bool args_parser_set_values(args_parser *self) {
+    self->video_encoder = (gsr_video_encoder_hardware)args_get_enum_by_key(self->args, NUM_ARGS, "-encoder", GSR_VIDEO_ENCODER_HW_GPU);
+    self->pixel_format = (gsr_pixel_format)args_get_enum_by_key(self->args, NUM_ARGS, "-pixfmt", GSR_PIXEL_FORMAT_YUV420);
+    self->framerate_mode = (gsr_framerate_mode)args_get_enum_by_key(self->args, NUM_ARGS, "-fm", GSR_FRAMERATE_MODE_VARIABLE);
+    self->color_range = (gsr_color_range)args_get_enum_by_key(self->args, NUM_ARGS, "-cr", GSR_COLOR_RANGE_LIMITED);
+    self->tune = (gsr_tune)args_get_enum_by_key(self->args, NUM_ARGS, "-tune", GSR_TUNE_PERFORMANCE);
+    self->video_codec = (gsr_video_codec)args_get_enum_by_key(self->args, NUM_ARGS, "-k", GSR_VIDEO_CODEC_AUTO);
+    self->audio_codec = (gsr_audio_codec)args_get_enum_by_key(self->args, NUM_ARGS, "-ac", GSR_AUDIO_CODEC_OPUS);
+    self->bitrate_mode = (gsr_bitrate_mode)args_get_enum_by_key(self->args, NUM_ARGS, "-bm", GSR_BITRATE_MODE_AUTO);
+    self->replay_storage = (gsr_replay_storage)args_get_enum_by_key(self->args, NUM_ARGS, "-replay-storage", GSR_REPLAY_STORAGE_RAM);
+
+    self->capture_source = args_get_value_by_key(self->args, NUM_ARGS, "-w");
+    self->verbose = args_get_boolean_by_key(self->args, NUM_ARGS, "-v", true);
+    self->gl_debug = args_get_boolean_by_key(self->args, NUM_ARGS, "-gl-debug", false);
+    self->record_cursor = args_get_boolean_by_key(self->args, NUM_ARGS, "-cursor", true);
+    self->date_folders = args_get_boolean_by_key(self->args, NUM_ARGS, "-df", false);
+    self->restore_portal_session = args_get_boolean_by_key(self->args, NUM_ARGS, "-restore-portal-session", false);
+    self->restart_replay_on_save = args_get_boolean_by_key(self->args, NUM_ARGS, "-restart-replay-on-save", false);
+    self->overclock = args_get_boolean_by_key(self->args, NUM_ARGS, "-oc", false);
+    self->fallback_cpu_encoding = args_get_boolean_by_key(self->args, NUM_ARGS, "-fallback-cpu-encoding", false);
+    self->write_first_frame_ts = args_get_boolean_by_key(self->args, NUM_ARGS, "-write-first-frame-ts", false);
+    self->low_power = args_get_boolean_by_key(self->args, NUM_ARGS, "-low-power", false);
+
+    self->audio_bitrate = args_get_i64_by_key(self->args, NUM_ARGS, "-ab", 0);
+    self->audio_bitrate *= 1000LL;
+
+    self->keyint = args_get_double_by_key(self->args, NUM_ARGS, "-keyint", 2.0);
+
+    if(self->audio_codec == GSR_AUDIO_CODEC_FLAC) {
+        fprintf(stderr, "gsr warning: flac audio codec is temporary disabled, using opus audio codec instead\n");
+        self->audio_codec = GSR_AUDIO_CODEC_OPUS;
+    }
+
+    self->portal_session_token_filepath = args_get_value_by_key(self->args, NUM_ARGS, "-portal-session-token-filepath");
+    if(self->portal_session_token_filepath) {
+        int len = strlen(self->portal_session_token_filepath);
+        if(len > 0 && self->portal_session_token_filepath[len - 1] == '/') {
+            fprintf(stderr, "gsr error: -portal-session-token-filepath should be a path to a file but it ends with a /: %s\n", self->portal_session_token_filepath);
+            return false;
+        }
+    }
+
+    self->recording_saved_script = args_get_value_by_key(self->args, NUM_ARGS, "-sc");
+    if(self->recording_saved_script) {
+        struct stat buf;
+        if(stat(self->recording_saved_script, &buf) == -1 || !S_ISREG(buf.st_mode)) {
+            fprintf(stderr, "gsr error: Script \"%s\" either doesn't exist or it's not a file\n", self->recording_saved_script);
+            usage();
+            return false;
+        }
+
+        if(!(buf.st_mode & S_IXUSR)) {
+            fprintf(stderr, "gsr error: Script \"%s\" is not executable\n", self->recording_saved_script);
+            usage();
+            return false;
+        }
+    }
+
+    const char *quality_str = args_get_value_by_key(self->args, NUM_ARGS, "-q");
+    self->video_quality = GSR_VIDEO_QUALITY_VERY_HIGH;
+    self->video_bitrate = 0;
+
+    if(self->bitrate_mode == GSR_BITRATE_MODE_CBR) {
+        if(!quality_str) {
+            fprintf(stderr, "gsr error: option '-q' is required when using '-bm cbr' option\n");
+            usage();
+            return false;
+        }
+
+        if(sscanf(quality_str, "%" PRIi64, &self->video_bitrate) != 1) {
+            fprintf(stderr, "gsr error: -q argument \"%s\" is not an integer value. When using '-bm cbr' option '-q' is expected to be an integer value\n", quality_str);
+            usage();
+            return false;
+        }
+
+        if(self->video_bitrate < 0) {
+            fprintf(stderr, "gsr error: -q is expected to be 0 or larger, got %" PRIi64 "\n", self->video_bitrate);
+            usage();
+            return false;
+        }
+
+        self->video_bitrate *= 1000LL;
+    } else {
+        if(!quality_str)
+            quality_str = "very_high";
+
+        if(strcmp(quality_str, "medium") == 0) {
+            self->video_quality = GSR_VIDEO_QUALITY_MEDIUM;
+        } else if(strcmp(quality_str, "high") == 0) {
+            self->video_quality = GSR_VIDEO_QUALITY_HIGH;
+        } else if(strcmp(quality_str, "very_high") == 0) {
+            self->video_quality = GSR_VIDEO_QUALITY_VERY_HIGH;
+        } else if(strcmp(quality_str, "ultra") == 0) {
+            self->video_quality = GSR_VIDEO_QUALITY_ULTRA;
+        } else {
+            fprintf(stderr, "gsr error: -q should either be 'medium', 'high', 'very_high' or 'ultra', got: '%s'\n", quality_str);
+            usage();
+            return false;
+        }
+    }
+
+    self->output_resolution = (vec2i){0, 0};
+
+    const char *output_resolution_str = args_get_value_by_key(self->args, NUM_ARGS, "-s");
+    if(output_resolution_str) {
+        if(sscanf(output_resolution_str, "%dx%d", &self->output_resolution.x, &self->output_resolution.y) != 2) {
+            fprintf(stderr, "gsr error: invalid value for option -s '%s', expected a value in format WxH\n", output_resolution_str);
+            usage();
+            return false;
+        }
+
+        if(self->output_resolution.x < 0 || self->output_resolution.y < 0) {
+            fprintf(stderr, "gsr error: invalid value for option -s '%s', expected width and height to be greater or equal to 0\n", output_resolution_str);
+            usage();
+            return false;
+        }
+    }
+
+    self->region_size = (vec2i){0, 0};
+    self->region_position = (vec2i){0, 0};
+    const char *region_str = args_get_value_by_key(self->args, NUM_ARGS, "-region");
+    if(region_str) {
+        if(sscanf(region_str, "%dx%d+%d+%d", &self->region_size.x, &self->region_size.y, &self->region_position.x, &self->region_position.y) != 4) {
+            fprintf(stderr, "gsr error: invalid value for option -region '%s', expected a value in format WxH+X+Y\n", region_str);
+            usage();
+            return false;
+        }
+
+        if(self->region_size.x < 0 || self->region_size.y < 0) {
+            fprintf(stderr, "gsr error: invalid value for option -region '%s', expected width and height to be greater or equal to 0\n", region_str);
+            usage();
+            return false;
+        }
+    }
+
+    self->fps = args_get_i64_by_key(self->args, NUM_ARGS, "-f", 60);
+    self->replay_buffer_size_secs = args_get_i64_by_key(self->args, NUM_ARGS, "-r", -1);
+    if(self->replay_buffer_size_secs != -1)
+        self->replay_buffer_size_secs += (int64_t)(self->keyint + 0.5); // Add a few seconds to account of lost packets because of non-keyframe packets skipped
+
+    self->container_format = args_get_value_by_key(self->args, NUM_ARGS, "-c");
+    if(self->container_format && strcmp(self->container_format, "mkv") == 0)
+        self->container_format = "matroska";
+
+    const bool is_replaying = self->replay_buffer_size_secs != -1;
+    self->is_livestream = false;
+    self->filename = args_get_value_by_key(self->args, NUM_ARGS, "-o");
+    if(self->filename) {
+        self->is_livestream = is_livestream_path(self->filename);
+        if(self->is_livestream) {
+            if(is_replaying) {
+                fprintf(stderr, "gsr error: replay mode is not applicable to live streaming\n");
+                return false;
+            }
+        } else {
+            if(!is_replaying) {
+                char directory_buf[PATH_MAX];
+                snprintf(directory_buf, sizeof(directory_buf), "%s", self->filename);
+                char *directory = dirname(directory_buf);
+                if(strcmp(directory, ".") != 0 && strcmp(directory, "/") != 0) {
+                    if(create_directory_recursive(directory) != 0) {
+                        fprintf(stderr, "gsr error: failed to create directory for output file: %s\n", self->filename);
+                        return false;
+                    }
+                }
+            } else {
+                if(!self->container_format) {
+                    fprintf(stderr, "gsr error: option -c is required when using option -r\n");
+                    usage();
+                    return false;
+                }
+
+                struct stat buf;
+                if(stat(self->filename, &buf) != -1 && !S_ISDIR(buf.st_mode)) {
+                    fprintf(stderr, "gsr error: File \"%s\" exists but it's not a directory\n", self->filename);
+                    usage();
+                    return false;
+                }
+            }
+        }
+    } else {
+        if(!is_replaying) {
+            self->filename = "/dev/stdout";
+        } else {
+            fprintf(stderr, "gsr error: Option -o is required when using option -r\n");
+            usage();
+            return false;
+        }
+
+        if(!self->container_format) {
+            fprintf(stderr, "gsr error: option -c is required when not using option -o\n");
+            usage();
+            return false;
+        }
+    }
+
+    self->is_output_piped = strcmp(self->filename, "/dev/stdout") == 0;
+    self->low_latency_recording = self->is_livestream || self->is_output_piped;
+    if(self->write_first_frame_ts && (self->is_livestream || self->is_output_piped)) {
+        fprintf(stderr, "gsr warning: -write-first-frame-ts is ignored for livestreaming or when output is piped\n");
+        self->write_first_frame_ts = false;
+    }
+
+    self->replay_recording_directory = args_get_value_by_key(self->args, NUM_ARGS, "-ro");
+
+    if(self->is_livestream && self->recording_saved_script) {
+        fprintf(stderr, "gsr warning: live stream detected, -sc script is ignored\n");
+        self->recording_saved_script = NULL;
+    }
+
+    self->ffmpeg_opts = args_get_value_by_key(self->args, NUM_ARGS, "-ffmpeg-opts");
+    self->ffmpeg_video_opts = args_get_value_by_key(self->args, NUM_ARGS, "-ffmpeg-video-opts");
+    self->ffmpeg_audio_opts = args_get_value_by_key(self->args, NUM_ARGS, "-ffmpeg-audio-opts");
+
+    return true;
+}
+
+bool args_parser_parse(args_parser *self, int argc, char **argv, const args_handlers *arg_handlers, void *userdata) {
+    assert(arg_handlers);
+    memset(self, 0, sizeof(*self));
+
+    if(argc <= 1) {
+        usage_full();
+        return false;
+    }
+
+    if(argc == 2 && (strcmp(argv[1], "-h") == 0 || strcmp(argv[1], "--help") == 0)) {
+        usage_full();
+        return false;
+    }
+
+    if(argc == 2 && strcmp(argv[1], "--info") == 0) {
+        arg_handlers->info(userdata);
+        return true;
+    }
+
+    if(argc == 2 && strcmp(argv[1], "--list-audio-devices") == 0) {
+        arg_handlers->list_audio_devices(userdata);
+        return true;
+    }
+
+    if(argc == 2 && strcmp(argv[1], "--list-application-audio") == 0) {
+        arg_handlers->list_application_audio(userdata);
+        return true;
+    }
+
+    if(argc == 2 && strcmp(argv[1], "--list-v4l2-devices") == 0) {
+        arg_handlers->list_v4l2_devices(userdata);
+        return true;
+    }
+
+    if(strcmp(argv[1], "--list-capture-options") == 0) {
+        if(argc == 2) {
+            arg_handlers->list_capture_options(NULL, userdata);
+            return true;
+        } else if(argc == 3 || argc == 4) {
+            const char *card_path = argv[2];
+            arg_handlers->list_capture_options(card_path, userdata);
+            return true;
+        } else {
+            fprintf(stderr, "gsr error: expected --list-capture-options to be called with either no extra arguments or 1 extra argument (card path)\n");
+            return false;
+        }
+    }
+
+    if(strcmp(argv[1], "--list-monitors") == 0) {
+        arg_handlers->list_monitors(userdata);
+        return true;
+    }
+
+    if(argc == 2 && strcmp(argv[1], "--version") == 0) {
+        arg_handlers->version(userdata);
+        return true;
+    }
+
+    int arg_index = 0;
+    self->args[arg_index++] = (Arg){ .key = "-w",                             .optional = false, .list = false, .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-c",                             .optional = true,  .list = false, .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-f",                             .optional = true,  .list = false, .type = ARG_TYPE_I64, .integer_value_min = 1, .integer_value_max = 1000 };
+    self->args[arg_index++] = (Arg){ .key = "-s",                             .optional = true,  .list = false, .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-region",                        .optional = true,  .list = false, .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-a",                             .optional = true,  .list = true,  .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-q",                             .optional = true,  .list = false, .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-o",                             .optional = true,  .list = false, .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-ro",                            .optional = true,  .list = false, .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-r",                             .optional = true,  .list = false, .type = ARG_TYPE_I64, .integer_value_min = 2, .integer_value_max = 86400 };
+    self->args[arg_index++] = (Arg){ .key = "-restart-replay-on-save",        .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    self->args[arg_index++] = (Arg){ .key = "-k",                             .optional = true,  .list = false, .type = ARG_TYPE_ENUM, .enum_values = video_codec_enums, .num_enum_values = sizeof(video_codec_enums)/sizeof(ArgEnum) };
+    self->args[arg_index++] = (Arg){ .key = "-ac",                            .optional = true,  .list = false, .type = ARG_TYPE_ENUM, .enum_values = audio_codec_enums, .num_enum_values = sizeof(audio_codec_enums)/sizeof(ArgEnum) };
+    self->args[arg_index++] = (Arg){ .key = "-ab",                            .optional = true,  .list = false, .type = ARG_TYPE_I64, .integer_value_min = 0, .integer_value_max = 50000 };
+    self->args[arg_index++] = (Arg){ .key = "-oc",                            .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    self->args[arg_index++] = (Arg){ .key = "-fm",                            .optional = true,  .list = false, .type = ARG_TYPE_ENUM, .enum_values = framerate_mode_enums, .num_enum_values = sizeof(framerate_mode_enums)/sizeof(ArgEnum) };
+    self->args[arg_index++] = (Arg){ .key = "-bm",                            .optional = true,  .list = false, .type = ARG_TYPE_ENUM, .enum_values = bitrate_mode_enums, .num_enum_values = sizeof(bitrate_mode_enums)/sizeof(ArgEnum) };
+    self->args[arg_index++] = (Arg){ .key = "-pixfmt",                        .optional = true,  .list = false, .type = ARG_TYPE_ENUM, .enum_values = pixel_format_enums, .num_enum_values = sizeof(pixel_format_enums)/sizeof(ArgEnum) };
+    self->args[arg_index++] = (Arg){ .key = "-v",                             .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    self->args[arg_index++] = (Arg){ .key = "-gl-debug",                      .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    self->args[arg_index++] = (Arg){ .key = "-df",                            .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    self->args[arg_index++] = (Arg){ .key = "-sc",                            .optional = true,  .list = false, .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-cr",                            .optional = true,  .list = false, .type = ARG_TYPE_ENUM, .enum_values = color_range_enums, .num_enum_values = sizeof(color_range_enums)/sizeof(ArgEnum) };
+    self->args[arg_index++] = (Arg){ .key = "-tune",                          .optional = true,  .list = false, .type = ARG_TYPE_ENUM, .enum_values = tune_enums, .num_enum_values = sizeof(tune_enums)/sizeof(ArgEnum) };
+    self->args[arg_index++] = (Arg){ .key = "-cursor",                        .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    self->args[arg_index++] = (Arg){ .key = "-keyint",                        .optional = true,  .list = false, .type = ARG_TYPE_DOUBLE, .integer_value_min = 0, .integer_value_max = 500 };
+    self->args[arg_index++] = (Arg){ .key = "-restore-portal-session",        .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    self->args[arg_index++] = (Arg){ .key = "-portal-session-token-filepath", .optional = true,  .list = false, .type = ARG_TYPE_STRING  };
+    self->args[arg_index++] = (Arg){ .key = "-encoder",                       .optional = true,  .list = false, .type = ARG_TYPE_ENUM, .enum_values = video_encoder_enums, .num_enum_values = sizeof(video_encoder_enums)/sizeof(ArgEnum) };
+    self->args[arg_index++] = (Arg){ .key = "-fallback-cpu-encoding",         .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    self->args[arg_index++] = (Arg){ .key = "-replay-storage",                .optional = true,  .list = false, .type = ARG_TYPE_ENUM, .enum_values = replay_storage_enums, .num_enum_values = sizeof(replay_storage_enums)/sizeof(ArgEnum) };
+    self->args[arg_index++] = (Arg){ .key = "-p",                             .optional = true,  .list = true,  .type = ARG_TYPE_STRING };
+    self->args[arg_index++] = (Arg){ .key = "-ffmpeg-opts",                   .optional = true,  .list = false, .type = ARG_TYPE_STRING };
+    self->args[arg_index++] = (Arg){ .key = "-ffmpeg-video-opts",             .optional = true,  .list = false, .type = ARG_TYPE_STRING };
+    self->args[arg_index++] = (Arg){ .key = "-ffmpeg-audio-opts",             .optional = true,  .list = false, .type = ARG_TYPE_STRING };
+    self->args[arg_index++] = (Arg){ .key = "-write-first-frame-ts",          .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    self->args[arg_index++] = (Arg){ .key = "-low-power",                     .optional = true,  .list = false, .type = ARG_TYPE_BOOLEAN };
+    assert(arg_index == NUM_ARGS);
+
+    for(int i = 1; i < argc; i += 2) {
+        const char *arg_name = argv[i];
+        Arg *arg = args_get_by_key(self->args, NUM_ARGS, arg_name);
+        if(!arg) {
+            fprintf(stderr, "gsr error: invalid argument '%s'\n", arg_name);
+            usage();
+            return false;
+        }
+
+        if(arg->num_values > 0 && !arg->list) {
+            fprintf(stderr, "gsr error: expected argument '%s' to only be specified once\n", arg_name);
+            usage();
+            return false;
+        }
+
+        if(i + 1 >= argc) {
+            fprintf(stderr, "gsr error: missing value for argument '%s'\n", arg_name);
+            usage();
+            return false;
+        }
+
+        const char *arg_value = argv[i + 1];
+        switch(arg->type) {
+            case ARG_TYPE_STRING: {
+                break;
+            }
+            case ARG_TYPE_BOOLEAN: {
+                if(strcmp(arg_value, "yes") == 0) {
+                    arg->typed_value.boolean = true;
+                } else if(strcmp(arg_value, "no") == 0) {
+                    arg->typed_value.boolean = false;
+                } else {
+                    fprintf(stderr, "gsr error: %s should either be 'yes' or 'no', got: '%s'\n", arg_name, arg_value);
+                    usage();
+                    return false;
+                }
+                break;
+            }
+            case ARG_TYPE_ENUM: {
+                if(!arg_get_enum_value_by_name(arg, arg_value, &arg->typed_value.enum_value)) {
+                    fprintf(stderr, "gsr error: %s should either be ", arg_name);
+                    arg_print_expected_enum_names(arg);
+                    fprintf(stderr, ", got: '%s'\n", arg_value);
+                    usage();
+                    return false;
+                }
+                break;
+            }
+            case ARG_TYPE_I64: {
+                if(sscanf(arg_value, "%" PRIi64, &arg->typed_value.i64_value) != 1) {
+                    fprintf(stderr, "gsr error: %s argument \"%s\" is not an integer\n", arg_name, arg_value);
+                    usage();
+                    return false;
+                }
+
+                if(arg->typed_value.i64_value < arg->integer_value_min) {
+                    fprintf(stderr, "gsr error: %s argument is expected to be larger than %" PRIi64 ", got %" PRIi64 "\n", arg_name, arg->integer_value_min, arg->typed_value.i64_value);
+                    usage();
+                    return false;
+                }
+
+                if(arg->typed_value.i64_value > arg->integer_value_max) {
+                    fprintf(stderr, "gsr error: %s argument is expected to be less than %" PRIi64 ", got %" PRIi64 "\n", arg_name, arg->integer_value_max, arg->typed_value.i64_value);
+                    usage();
+                    return false;
+                }
+                break;
+            }
+            case ARG_TYPE_DOUBLE: {
+                if(sscanf(arg_value, "%lf", &arg->typed_value.d_value) != 1) {
+                    fprintf(stderr, "gsr error: %s argument \"%s\" is not an floating-point number\n", arg_name, arg_value);
+                    usage();
+                    return false;
+                }
+
+                if(arg->typed_value.d_value < arg->integer_value_min) {
+                    fprintf(stderr, "gsr error: %s argument is expected to be larger than %" PRIi64 ", got %lf\n", arg_name, arg->integer_value_min, arg->typed_value.d_value);
+                    usage();
+                    return false;
+                }
+
+                if(arg->typed_value.d_value > arg->integer_value_max) {
+                    fprintf(stderr, "gsr error: %s argument is expected to be less than %" PRIi64 ", got %lf\n", arg_name, arg->integer_value_max, arg->typed_value.d_value);
+                    usage();
+                    return false;
+                }
+                break;
+            }
+        }
+
+        if(!arg_append_value(arg, arg_value)) {
+            fprintf(stderr, "gsr error: failed to append argument, out of memory\n");
+            return false;
+        }
+    }
+
+    for(int i = 0; i < NUM_ARGS; ++i) {
+        const Arg *arg = &self->args[i];
+        if(!arg->optional && arg->num_values == 0) {
+            fprintf(stderr, "gsr error: missing argument '%s'\n", arg->key);
+            usage();
+            return false;
+        }
+    }
+
+    return args_parser_set_values(self);
+}
+
+void args_parser_deinit(args_parser *self) {
+    for(int i = 0; i < NUM_ARGS; ++i) {
+        arg_deinit(&self->args[i]);
+    }
+}
+
+bool args_parser_validate_with_gl_info(args_parser *self, gsr_egl *egl) {
+    const bool wayland = gsr_window_get_display_server(egl->window) == GSR_DISPLAY_SERVER_WAYLAND;
+
+    if(self->bitrate_mode == (gsr_bitrate_mode)GSR_BITRATE_MODE_AUTO) {
+        // QP is broken on steam deck, see https://github.com/ValveSoftware/SteamOS/issues/1609
+        self->bitrate_mode = egl->gpu_info.is_steam_deck ? GSR_BITRATE_MODE_VBR : GSR_BITRATE_MODE_QP;
+    }
+
+    if(egl->gpu_info.is_steam_deck && self->bitrate_mode == GSR_BITRATE_MODE_QP) {
+        fprintf(stderr, "gsr warning: qp bitrate mode is not supported on Steam Deck because of Steam Deck driver bugs. Using vbr instead\n");
+        self->bitrate_mode = GSR_BITRATE_MODE_VBR;
+    }
+
+    if(self->video_encoder == GSR_VIDEO_ENCODER_HW_CPU && self->bitrate_mode == GSR_BITRATE_MODE_VBR) {
+        fprintf(stderr, "gsr warning: bitrate mode has been forcefully set to qp because software encoding option doesn't support vbr option\n");
+        self->bitrate_mode = GSR_BITRATE_MODE_QP;
+    }
+
+    if(egl->gpu_info.vendor != GSR_GPU_VENDOR_NVIDIA && self->overclock) {
+        fprintf(stderr, "gsr info: overclock option has no effect on amd/intel, ignoring option\n");
+        self->overclock = false;
+    }
+
+    if(egl->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA && self->overclock && wayland) {
+        fprintf(stderr, "gsr info: overclocking is not possible on nvidia on wayland, ignoring option\n");
+        self->overclock = false;
+    }
+
+    if(egl->gpu_info.is_steam_deck) {
+        fprintf(stderr, "gsr warning: steam deck has multiple driver issues. One of them has been reported here: https://github.com/ValveSoftware/SteamOS/issues/1609\n"
+            "If you have issues with GPU Screen Recorder on steam deck that you don't have on a desktop computer then report the issue to Valve and/or AMD.\n");
+    }
+
+    self->very_old_gpu = false;
+    if(egl->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA && egl->gpu_info.gpu_version != 0 && egl->gpu_info.gpu_version < 900) {
+        fprintf(stderr, "gsr info: your gpu appears to be very old (older than maxwell architecture). Switching to lower preset\n");
+        self->very_old_gpu = true;
+    }
+
+    if(video_codec_is_hdr(self->video_codec) && !wayland) {
+        fprintf(stderr, "gsr error: hdr video codec option %s is not available on X11\n", video_codec_to_string(self->video_codec));
+        usage();
+        return false;
+    }
+
+    return true;
+}
+
+void args_parser_print_usage(void) {
+    usage();
+}
+
+Arg* args_parser_get_arg(args_parser *self, const char *arg_name) {
+    return args_get_by_key(self->args, NUM_ARGS, arg_name);
+}

src/capture/kms.c

@@ -3,7 +3,6 @@
 #include "../../include/color_conversion.h"
 #include "../../include/cursor.h"
 #include "../../include/window/window.h"
-#include "../../kms/client/kms_client.h"
 
 #include <stdlib.h>
 #include <string.h>
@@ -12,7 +11,7 @@
 #include <fcntl.h>
 
 #include <xf86drm.h>
-#include <libdrm/drm_fourcc.h>
+#include <drm_fourcc.h>
 
 #include <libavutil/mastering_display_metadata.h>
 
@@ -30,15 +29,13 @@ typedef struct {
 
 typedef struct {
     gsr_capture_kms_params params;
-    
-    gsr_kms_client kms_client;
-    gsr_kms_response kms_response;
 
     vec2i capture_pos;
     vec2i capture_size;
     MonitorId monitor_id;
 
-    gsr_monitor_rotation monitor_rotation;
+    gsr_monitor_rotation display_server_monitor_rotation;
+    gsr_monitor_rotation final_monitor_rotation;
 
     unsigned int input_texture_id;
     unsigned int external_input_texture_id;
@@ -51,11 +48,6 @@ typedef struct {
     bool hdr_metadata_set;
 
     bool is_x11;
-    gsr_cursor x11_cursor;
-
-    bool performance_error_shown;
-    bool fast_path_failed;
-    bool mesa_supports_compute_only_vaapi_copy;
 
     //int drm_fd;
     //uint64_t prev_sequence;
@@ -65,21 +57,12 @@ typedef struct {
     vec2i prev_plane_size;
 
     double last_time_monitor_check;
-} gsr_capture_kms;
 
-static void gsr_capture_kms_cleanup_kms_fds(gsr_capture_kms *self) {
-    for(int i = 0; i < self->kms_response.num_items; ++i) {
-        for(int j = 0; j < self->kms_response.items[i].num_dma_bufs; ++j) {
-            gsr_kms_response_dma_buf *dma_buf = &self->kms_response.items[i].dma_buf[j];
-            if(dma_buf->fd > 0) {
-                close(dma_buf->fd);
-                dma_buf->fd = -1;
-            }
-        }
-        self->kms_response.items[i].num_dma_bufs = 0;
-    }
-    self->kms_response.num_items = 0;
-}
+    bool capture_is_combined_plane;
+    gsr_kms_response_item *drm_fd;
+    vec2i output_size;
+    vec2i target_pos;
+} gsr_capture_kms;
 
 static void gsr_capture_kms_stop(gsr_capture_kms *self) {
     if(self->input_texture_id) {
@@ -101,10 +84,6 @@ static void gsr_capture_kms_stop(gsr_capture_kms *self) {
     //     close(self->drm_fd);
     //     self->drm_fd = -1;
     // }
-
-    gsr_capture_kms_cleanup_kms_fds(self);
-    gsr_kms_client_deinit(&self->kms_client);
-    gsr_cursor_deinit(&self->x11_cursor);
 }
 
 static int max_int(int a, int b) {
@@ -114,16 +93,12 @@ static int max_int(int a, int b) {
 static void gsr_capture_kms_create_input_texture_ids(gsr_capture_kms *self) {
     self->params.egl->glGenTextures(1, &self->input_texture_id);
     self->params.egl->glBindTexture(GL_TEXTURE_2D, self->input_texture_id);
-    self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
     self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
 
     self->params.egl->glGenTextures(1, &self->external_input_texture_id);
     self->params.egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, self->external_input_texture_id);
-    self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
     self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     self->params.egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, 0);
@@ -133,8 +108,6 @@ static void gsr_capture_kms_create_input_texture_ids(gsr_capture_kms *self) {
 
     self->params.egl->glGenTextures(1, &self->cursor_texture_id);
     self->params.egl->glBindTexture(cursor_texture_id_target, self->cursor_texture_id);
-    self->params.egl->glTexParameteri(cursor_texture_id_target, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    self->params.egl->glTexParameteri(cursor_texture_id_target, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     self->params.egl->glTexParameteri(cursor_texture_id_target, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
     self->params.egl->glTexParameteri(cursor_texture_id_target, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     self->params.egl->glBindTexture(cursor_texture_id_target, 0);
@@ -165,8 +138,8 @@ static void monitor_callback(const gsr_monitor *monitor, void *userdata) {
         fprintf(stderr, "gsr warning: reached max connector ids\n");
 }
 
-static vec2i rotate_capture_size_if_rotated(gsr_capture_kms *self, vec2i capture_size) {
-    if(self->monitor_rotation == GSR_MONITOR_ROT_90 || self->monitor_rotation == GSR_MONITOR_ROT_270) {
+static vec2i rotate_capture_size_if_rotated(gsr_capture_kms *self, vec2i capture_size, gsr_monitor_rotation rotation) {
+    if(rotation == GSR_MONITOR_ROT_90 || rotation == GSR_MONITOR_ROT_270) {
         int tmp_x = capture_size.x;
         capture_size.x = capture_size.y;
         capture_size.y = tmp_x;
@@ -182,16 +155,8 @@ static int gsr_capture_kms_start(gsr_capture *cap, gsr_capture_metadata *capture
     gsr_monitor monitor;
     self->monitor_id.num_connector_ids = 0;
 
-    int kms_init_res = gsr_kms_client_init(&self->kms_client, self->params.egl->card_path);
-    if(kms_init_res != 0)
-        return kms_init_res;
-
     self->is_x11 = gsr_window_get_display_server(self->params.egl->window) == GSR_DISPLAY_SERVER_X11;
     const gsr_connection_type connection_type = self->is_x11 ? GSR_CONNECTION_X11 : GSR_CONNECTION_DRM;
-    if(self->is_x11) {
-        Display *display = gsr_window_get_display(self->params.egl->window);
-        gsr_cursor_init(&self->x11_cursor, self->params.egl, display);
-    }
 
     MonitorCallbackUserdata monitor_callback_userdata = {
         &self->monitor_id,
@@ -208,51 +173,28 @@ static int gsr_capture_kms_start(gsr_capture *cap, gsr_capture_metadata *capture
 
     monitor.name = self->params.display_to_capture;
     vec2i monitor_position = {0, 0};
-    drm_monitor_get_display_server_data(self->params.egl->window, &monitor, &self->monitor_rotation, &monitor_position);
+    drm_monitor_get_display_server_data(self->params.egl->window, &monitor, &self->display_server_monitor_rotation, &monitor_position);
 
     self->capture_pos = monitor.pos;
     /* Monitor size is already rotated on x11 when the monitor is rotated, no need to apply it ourselves */
     if(self->is_x11)
         self->capture_size = monitor.size;
     else
-        self->capture_size = rotate_capture_size_if_rotated(self, monitor.size);
+        self->capture_size = rotate_capture_size_if_rotated(self, monitor.size, self->display_server_monitor_rotation);
 
     if(self->params.output_resolution.x > 0 && self->params.output_resolution.y > 0) {
         self->params.output_resolution = scale_keep_aspect_ratio(self->capture_size, self->params.output_resolution);
-        capture_metadata->width = self->params.output_resolution.x;
-        capture_metadata->height = self->params.output_resolution.y;
+        capture_metadata->video_size = self->params.output_resolution;
     } else if(self->params.region_size.x > 0 && self->params.region_size.y > 0) {
-        capture_metadata->width = self->params.region_size.x;
-        capture_metadata->height = self->params.region_size.y;
+        capture_metadata->video_size = self->params.region_size;
     } else {
-        capture_metadata->width = self->capture_size.x;
-        capture_metadata->height = self->capture_size.y;
+        capture_metadata->video_size = self->capture_size;
     }
 
-    self->fast_path_failed = self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && !gl_driver_version_greater_than(&self->params.egl->gpu_info, 24, 0, 9);
-    if(self->fast_path_failed)
-        fprintf(stderr, "gsr warning: gsr_capture_kms_start: your amd driver (mesa) version is known to be buggy (<= version 24.0.9), falling back to opengl copy\n");
-
-    //if(self->params.hdr) {
-    //    self->fast_path_failed = true;
-    //    fprintf(stderr, "gsr warning: gsr_capture_kms_start: recording with hdr requires shader color conversion which might be slow. If this is an issue record with -w portal instead (which converts HDR to SDR)\n");
-    //}
-
-    self->mesa_supports_compute_only_vaapi_copy = self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && gl_driver_version_greater_than(&self->params.egl->gpu_info, 24, 3, 6);
-
     self->last_time_monitor_check = clock_get_monotonic_seconds();
     return 0;
 }
 
-static void gsr_capture_kms_on_event(gsr_capture *cap, gsr_egl *egl) {
-    gsr_capture_kms *self = cap->priv;
-    if(!self->is_x11)
-        return;
-
-    XEvent *xev = gsr_window_get_event_data(egl->window);
-    gsr_cursor_on_event(&self->x11_cursor, xev);
-}
-
 // TODO: This is disabled for now because we want to be able to record at a framerate higher than the monitor framerate
 // static void gsr_capture_kms_tick(gsr_capture *cap) {
 //     gsr_capture_kms *self = cap->priv;
@@ -274,16 +216,6 @@ static void gsr_capture_kms_on_event(gsr_capture *cap, gsr_egl *egl) {
 //     }
 // }
 
-static float monitor_rotation_to_radians(gsr_monitor_rotation rot) {
-    switch(rot) {
-        case GSR_MONITOR_ROT_0:   return 0.0f;
-        case GSR_MONITOR_ROT_90:  return M_PI_2;
-        case GSR_MONITOR_ROT_180: return M_PI;
-        case GSR_MONITOR_ROT_270: return M_PI + M_PI_2;
-    }
-    return 0.0f;
-}
-
 static gsr_kms_response_item* find_drm_by_connector_id(gsr_kms_response *kms_response, uint32_t connector_id) {
     for(int i = 0; i < kms_response->num_items; ++i) {
         if(kms_response->items[i].connector_id == connector_id && !kms_response->items[i].is_cursor)
@@ -428,14 +360,14 @@ static gsr_kms_response_item* find_monitor_drm(gsr_capture_kms *self, bool *capt
     gsr_kms_response_item *drm_fd = NULL;
 
     for(int i = 0; i < self->monitor_id.num_connector_ids; ++i) {
-        drm_fd = find_drm_by_connector_id(&self->kms_response, self->monitor_id.connector_ids[i]);
+        drm_fd = find_drm_by_connector_id(self->params.kms_response, self->monitor_id.connector_ids[i]);
         if(drm_fd)
             break;
     }
 
     // Will never happen on wayland unless the target monitor has been disconnected
     if(!drm_fd && self->is_x11) {
-        drm_fd = find_largest_drm(&self->kms_response);
+        drm_fd = find_largest_drm(self->params.kms_response);
         *capture_is_combined_plane = true;
     }
 
@@ -443,13 +375,26 @@ static gsr_kms_response_item* find_monitor_drm(gsr_capture_kms *self, bool *capt
 }
 
 static gsr_kms_response_item* find_cursor_drm_if_on_monitor(gsr_capture_kms *self, uint32_t monitor_connector_id, bool capture_is_combined_plane) {
-    gsr_kms_response_item *cursor_drm_fd = find_cursor_drm(&self->kms_response, monitor_connector_id);
+    gsr_kms_response_item *cursor_drm_fd = find_cursor_drm(self->params.kms_response, monitor_connector_id);
     if(!capture_is_combined_plane && cursor_drm_fd && cursor_drm_fd->connector_id != monitor_connector_id)
         cursor_drm_fd = NULL;
     return cursor_drm_fd;
 }
 
-static void render_drm_cursor(gsr_capture_kms *self, gsr_color_conversion *color_conversion, const gsr_kms_response_item *cursor_drm_fd, vec2i target_pos, float texture_rotation, vec2i output_size, vec2i framebuffer_size) {
+static gsr_monitor_rotation kms_rotation_to_gsr_monitor_rotation(gsr_kms_rotation rotation) {
+    // Right now both enums have the same values
+    return (gsr_monitor_rotation)rotation;
+}
+
+static int remainder_int(int a, int b) {
+    return a - (a / b) * b;
+}
+
+static gsr_monitor_rotation sub_rotations(gsr_monitor_rotation rot1, gsr_monitor_rotation rot2) {
+    return remainder_int(rot1 - rot2, 4);
+}
+
+static void render_drm_cursor(gsr_capture_kms *self, gsr_color_conversion *color_conversion, gsr_capture_metadata *capture_metadata, const gsr_kms_response_item *cursor_drm_fd, vec2i target_pos, vec2i output_size, vec2i framebuffer_size) {
     const vec2d scale = {
         self->capture_size.x == 0 ? 0 : (double)output_size.x / (double)self->capture_size.x,
         self->capture_size.y == 0 ? 0 : (double)output_size.y / (double)self->capture_size.y
@@ -458,8 +403,11 @@ static void render_drm_cursor(gsr_capture_kms *self, gsr_color_conversion *color
     const bool cursor_texture_id_is_external = self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA;
     const vec2i cursor_size = {cursor_drm_fd->width, cursor_drm_fd->height};
 
+    const gsr_monitor_rotation cursor_plane_rotation = kms_rotation_to_gsr_monitor_rotation(cursor_drm_fd->rotation);
+    const gsr_monitor_rotation rotation = sub_rotations(self->display_server_monitor_rotation, cursor_plane_rotation);
+
     vec2i cursor_pos = {cursor_drm_fd->x, cursor_drm_fd->y};
-    switch(self->monitor_rotation) {
+    switch(rotation) {
         case GSR_MONITOR_ROT_0:
             break;
         case GSR_MONITOR_ROT_90:
@@ -522,14 +470,14 @@ static void render_drm_cursor(gsr_capture_kms *self, gsr_color_conversion *color
 
     gsr_color_conversion_draw(color_conversion, self->cursor_texture_id,
         cursor_pos, (vec2i){cursor_size.x * scale.x, cursor_size.y * scale.y},
-        (vec2i){0, 0}, cursor_size,
-        texture_rotation, cursor_texture_id_is_external, GSR_SOURCE_COLOR_RGB);
+        (vec2i){0, 0}, cursor_size, cursor_size,
+        gsr_monitor_rotation_to_rotation(rotation), capture_metadata->flip, GSR_SOURCE_COLOR_RGB, cursor_texture_id_is_external);
 
     self->params.egl->glDisable(GL_SCISSOR_TEST);
 }
 
-static void render_x11_cursor(gsr_capture_kms *self, gsr_color_conversion *color_conversion, vec2i capture_pos, vec2i target_pos, vec2i output_size) {
-    if(!self->x11_cursor.visible)
+static void render_x11_cursor(gsr_capture_kms *self, gsr_color_conversion *color_conversion, gsr_capture_metadata *capture_metadata, vec2i capture_pos, vec2i target_pos, vec2i output_size) {
+    if(!self->params.x11_cursor->visible)
         return;
 
     const vec2d scale = {
@@ -537,21 +485,18 @@ static void render_x11_cursor(gsr_capture_kms *self, gsr_color_conversion *color
         self->capture_size.y == 0 ? 0 : (double)output_size.y / (double)self->capture_size.y
     };
 
-    Display *display = gsr_window_get_display(self->params.egl->window);
-    gsr_cursor_tick(&self->x11_cursor, DefaultRootWindow(display));
-
     const vec2i cursor_pos = {
-        target_pos.x + (self->x11_cursor.position.x - self->x11_cursor.hotspot.x - capture_pos.x) * scale.x,
-        target_pos.y + (self->x11_cursor.position.y - self->x11_cursor.hotspot.y - capture_pos.y) * scale.y
+        target_pos.x + (self->params.x11_cursor->position.x - self->params.x11_cursor->hotspot.x - capture_pos.x) * scale.x,
+        target_pos.y + (self->params.x11_cursor->position.y - self->params.x11_cursor->hotspot.y - capture_pos.y) * scale.y
     };
 
     self->params.egl->glEnable(GL_SCISSOR_TEST);
     self->params.egl->glScissor(target_pos.x, target_pos.y, output_size.x, output_size.y);
 
-    gsr_color_conversion_draw(color_conversion, self->x11_cursor.texture_id,
-        cursor_pos, (vec2i){self->x11_cursor.size.x * scale.x, self->x11_cursor.size.y * scale.y},
-        (vec2i){0, 0}, self->x11_cursor.size,
-        0.0f, false, GSR_SOURCE_COLOR_RGB);
+    gsr_color_conversion_draw(color_conversion, self->params.x11_cursor->texture_id,
+        cursor_pos, (vec2i){self->params.x11_cursor->size.x * scale.x, self->params.x11_cursor->size.y * scale.y},
+        (vec2i){0, 0}, self->params.x11_cursor->size, self->params.x11_cursor->size,
+        GSR_ROT_0, capture_metadata->flip, GSR_SOURCE_COLOR_RGB, false);
 
     self->params.egl->glDisable(GL_SCISSOR_TEST);
 }
@@ -560,7 +505,7 @@ static void gsr_capture_kms_update_capture_size_change(gsr_capture_kms *self, gs
     if(target_pos.x != self->prev_target_pos.x || target_pos.y != self->prev_target_pos.y || drm_fd->src_w != self->prev_plane_size.x || drm_fd->src_h != self->prev_plane_size.y) {
         self->prev_target_pos = target_pos;
         self->prev_plane_size = self->capture_size;
-        gsr_color_conversion_clear(color_conversion);
+        color_conversion->schedule_clear = true;
     }
 }
 
@@ -594,123 +539,81 @@ static void gsr_capture_kms_update_connector_ids(gsr_capture_kms *self) {
 
     monitor.name = self->params.display_to_capture;
     vec2i monitor_position = {0, 0};
-    drm_monitor_get_display_server_data(self->params.egl->window, &monitor, &self->monitor_rotation, &monitor_position);
+    // TODO: This is cached. We need it updated.
+    drm_monitor_get_display_server_data(self->params.egl->window, &monitor, &self->display_server_monitor_rotation, &monitor_position);
 
     self->capture_pos = monitor.pos;
     /* Monitor size is already rotated on x11 when the monitor is rotated, no need to apply it ourselves */
     if(self->is_x11)
         self->capture_size = monitor.size;
     else
-        self->capture_size = rotate_capture_size_if_rotated(self, monitor.size);
+        self->capture_size = rotate_capture_size_if_rotated(self, monitor.size, self->display_server_monitor_rotation);
 }
 
-static void gsr_capture_kms_fail_fast_path_if_not_fast(gsr_capture_kms *self, uint32_t pixel_format) {
-    const uint8_t pixel_format_color_depth_1 = (pixel_format >> 16) & 0xFF;
-    if(!self->fast_path_failed && self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && !self->mesa_supports_compute_only_vaapi_copy && (pixel_format_color_depth_1 == '3' || pixel_format_color_depth_1 == '4')) {
-        self->fast_path_failed = true;
-        fprintf(stderr, "gsr warning: gsr_capture_kms_capture: the monitor you are recording is in 10/12-bit color format and your mesa version is <= 24.3.6, composition will be used."
-            " If you experience performance problems in the video then record on a single window on X11 or use portal capture option instead or disable 10/12-bit color option in your desktop environment settings,"
-            " or try to record the monitor on X11 instead (if you aren't already doing that) or update your mesa version.\n");
-    }
-}
-
-static int gsr_capture_kms_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
+static void gsr_capture_kms_pre_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
     gsr_capture_kms *self = cap->priv;
 
-    gsr_capture_kms_cleanup_kms_fds(self);
-
-    if(gsr_kms_client_get_kms(&self->kms_client, &self->kms_response) != 0) {
-        fprintf(stderr, "gsr error: gsr_capture_kms_capture: failed to get kms, error: %d (%s)\n", self->kms_response.result, self->kms_response.err_msg);
-        return -1;
-    }
-
-    if(self->kms_response.num_items == 0) {
+    if(self->params.kms_response->num_items == 0) {
         static bool error_shown = false;
         if(!error_shown) {
             error_shown = true;
-            fprintf(stderr, "gsr error: no drm found, capture will fail\n");
+            fprintf(stderr, "gsr error: gsr_capture_kms_pre_capture: no drm found, capture will fail\n");
         }
-        return -1;
+        return;
     }
 
     gsr_capture_kms_update_connector_ids(self);
 
-    bool capture_is_combined_plane = false;
-    const gsr_kms_response_item *drm_fd = find_monitor_drm(self, &capture_is_combined_plane);
-    if(!drm_fd) {
-        gsr_capture_kms_cleanup_kms_fds(self);
-        return -1;
-    }
+    self->capture_is_combined_plane = false;
+    self->drm_fd = find_monitor_drm(self, &self->capture_is_combined_plane);
+    if(!self->drm_fd)
+        return;
 
-    if(drm_fd->has_hdr_metadata && self->params.hdr && hdr_metadata_is_supported_format(&drm_fd->hdr_metadata))
-        gsr_kms_set_hdr_metadata(self, drm_fd);
+    if(self->drm_fd->has_hdr_metadata && self->params.hdr && hdr_metadata_is_supported_format(&self->drm_fd->hdr_metadata))
+        gsr_kms_set_hdr_metadata(self, self->drm_fd);
 
-    if(!self->performance_error_shown && self->monitor_rotation != GSR_MONITOR_ROT_0 && video_codec_context_is_vaapi(capture_metadata->video_codec_context) && self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD) {
-        self->performance_error_shown = true;
-        self->fast_path_failed = true;
-        fprintf(stderr, "gsr warning: gsr_capture_kms_capture: the monitor you are recording is rotated, composition will have to be used."
-            " If you experience performance problems in the video then record a single window on X11 or use portal capture option instead\n");
-    }
+    const gsr_monitor_rotation plane_rotation = kms_rotation_to_gsr_monitor_rotation(self->drm_fd->rotation);
+    self->final_monitor_rotation = self->capture_is_combined_plane ? GSR_MONITOR_ROT_0 : sub_rotations(self->display_server_monitor_rotation, plane_rotation);
 
-    gsr_capture_kms_fail_fast_path_if_not_fast(self, drm_fd->pixel_format);
-
-    self->capture_size = rotate_capture_size_if_rotated(self, (vec2i){ drm_fd->src_w, drm_fd->src_h });
+    self->capture_size = rotate_capture_size_if_rotated(self, (vec2i){ self->drm_fd->src_w, self->drm_fd->src_h }, self->final_monitor_rotation);
     if(self->params.region_size.x > 0 && self->params.region_size.y > 0)
         self->capture_size = self->params.region_size;
 
-    const bool is_scaled = self->params.output_resolution.x > 0 && self->params.output_resolution.y > 0;
-    vec2i output_size = is_scaled ? self->params.output_resolution : self->capture_size;
-    output_size = scale_keep_aspect_ratio(self->capture_size, output_size);
+    self->output_size = scale_keep_aspect_ratio(self->capture_size, capture_metadata->recording_size);
+    self->target_pos = gsr_capture_get_target_position(self->output_size, capture_metadata);
+    gsr_capture_kms_update_capture_size_change(self, color_conversion, self->target_pos, self->drm_fd);
+}
 
-    const float texture_rotation = monitor_rotation_to_radians(self->monitor_rotation);
-    const vec2i target_pos = { max_int(0, capture_metadata->width / 2 - output_size.x / 2), max_int(0, capture_metadata->height / 2 - output_size.y / 2) };
-    gsr_capture_kms_update_capture_size_change(self, color_conversion, target_pos, drm_fd);
+static int gsr_capture_kms_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
+    (void)capture_metadata;
+    gsr_capture_kms *self = cap->priv;
+
+    if(self->params.kms_response->num_items == 0)
+        return -1;
 
     vec2i capture_pos = self->capture_pos;
-    if(!capture_is_combined_plane)
-        capture_pos = (vec2i){drm_fd->x, drm_fd->y};
+    if(!self->capture_is_combined_plane)
+        capture_pos = (vec2i){self->drm_fd->x, self->drm_fd->y};
 
     capture_pos.x += self->params.region_position.x;
     capture_pos.y += self->params.region_position.y;
 
-    self->params.egl->glFlush();
-    self->params.egl->glFinish();
+    //self->params.egl->glFlush();
+    //self->params.egl->glFinish();
 
-    /* Fast opengl free path */
-    if(!self->fast_path_failed && self->monitor_rotation == GSR_MONITOR_ROT_0 && video_codec_context_is_vaapi(capture_metadata->video_codec_context) && self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD) {
-        int fds[4];
-        uint32_t offsets[4];
-        uint32_t pitches[4];
-        uint64_t modifiers[4];
-        for(int i = 0; i < drm_fd->num_dma_bufs; ++i) {
-            fds[i] = drm_fd->dma_buf[i].fd;
-            offsets[i] = drm_fd->dma_buf[i].offset;
-            pitches[i] = drm_fd->dma_buf[i].pitch;
-            modifiers[i] = drm_fd->modifier;
-        }
-        if(!vaapi_copy_drm_planes_to_video_surface(capture_metadata->video_codec_context, capture_metadata->frame, (vec2i){capture_pos.x, capture_pos.y}, self->capture_size, target_pos, output_size, drm_fd->pixel_format, (vec2i){drm_fd->width, drm_fd->height}, fds, offsets, pitches, modifiers, drm_fd->num_dma_bufs)) {
-            fprintf(stderr, "gsr error: gsr_capture_kms_capture: vaapi_copy_drm_planes_to_video_surface failed, falling back to opengl copy. Please report this as an issue at https://github.com/dec05eba/gpu-screen-recorder-issues\n");
-            self->fast_path_failed = true;
-        }
-    } else {
-        self->fast_path_failed = true;
+    EGLImage image = gsr_capture_kms_create_egl_image_with_fallback(self, self->drm_fd);
+    if(image) {
+        gsr_capture_kms_bind_image_to_input_texture_with_fallback(self, image);
+        self->params.egl->eglDestroyImage(self->params.egl->egl_display, image);
     }
 
-    if(self->fast_path_failed) {
-        EGLImage image = gsr_capture_kms_create_egl_image_with_fallback(self, drm_fd);
-        if(image) {
-            gsr_capture_kms_bind_image_to_input_texture_with_fallback(self, image);
-            self->params.egl->eglDestroyImage(self->params.egl->egl_display, image);
-        }
-
-        gsr_color_conversion_draw(color_conversion, self->external_texture_fallback ? self->external_input_texture_id : self->input_texture_id,
-            target_pos, output_size,
-            capture_pos, self->capture_size,
-            texture_rotation, self->external_texture_fallback, GSR_SOURCE_COLOR_RGB);
-    }
+    gsr_color_conversion_draw(color_conversion, self->external_texture_fallback ? self->external_input_texture_id : self->input_texture_id,
+        self->target_pos, self->output_size,
+        capture_pos, self->capture_size, (vec2i){ self->drm_fd->width, self->drm_fd->height },
+        gsr_monitor_rotation_to_rotation(self->final_monitor_rotation), capture_metadata->flip, GSR_SOURCE_COLOR_RGB, self->external_texture_fallback);
 
     if(self->params.record_cursor) {
-        gsr_kms_response_item *cursor_drm_fd = find_cursor_drm_if_on_monitor(self, drm_fd->connector_id, capture_is_combined_plane);
+        gsr_kms_response_item *cursor_drm_fd = find_cursor_drm_if_on_monitor(self, self->drm_fd->connector_id, self->capture_is_combined_plane);
         // The cursor is handled by x11 on x11 instead of using the cursor drm plane because on prime systems with a dedicated nvidia gpu
         // the cursor plane is not available when the cursor is on the monitor controlled by the nvidia device.
         // TODO: This doesn't work properly with software cursor on x11 since it will draw the x11 cursor on top of the cursor already in the framebuffer.
@@ -719,17 +622,15 @@ static int gsr_capture_kms_capture(gsr_capture *cap, gsr_capture_metadata *captu
             vec2i cursor_monitor_offset = self->capture_pos;
             cursor_monitor_offset.x += self->params.region_position.x;
             cursor_monitor_offset.y += self->params.region_position.y;
-            render_x11_cursor(self, color_conversion, cursor_monitor_offset, target_pos, output_size);
+            render_x11_cursor(self, color_conversion, capture_metadata, cursor_monitor_offset, self->target_pos, self->output_size);
         } else if(cursor_drm_fd) {
-            const vec2i framebuffer_size = rotate_capture_size_if_rotated(self, (vec2i){ drm_fd->src_w, drm_fd->src_h });
-            render_drm_cursor(self, color_conversion, cursor_drm_fd, target_pos, texture_rotation, output_size, framebuffer_size);
+            const vec2i framebuffer_size = rotate_capture_size_if_rotated(self, (vec2i){ self->drm_fd->src_w, self->drm_fd->src_h }, self->final_monitor_rotation);
+            render_drm_cursor(self, color_conversion, capture_metadata, cursor_drm_fd, self->target_pos, self->output_size, framebuffer_size);
         }
     }
 
-    self->params.egl->glFlush();
-    self->params.egl->glFinish();
-
-    gsr_capture_kms_cleanup_kms_fds(self);
+    //self->params.egl->glFlush();
+    //self->params.egl->glFinish();
 
     return 0;
 }
@@ -822,9 +723,9 @@ gsr_capture* gsr_capture_kms_create(const gsr_capture_kms_params *params) {
     
     *cap = (gsr_capture) {
         .start = gsr_capture_kms_start,
-        .on_event = gsr_capture_kms_on_event,
         //.tick = gsr_capture_kms_tick,
         .should_stop = gsr_capture_kms_should_stop,
+        .pre_capture = gsr_capture_kms_pre_capture,
         .capture = gsr_capture_kms_capture,
         .uses_external_image = gsr_capture_kms_uses_external_image,
         .set_hdr_metadata = gsr_capture_kms_set_hdr_metadata,

src/capture/nvfbc.c

@@ -13,6 +13,7 @@
 #include <assert.h>
 
 #include <X11/Xlib.h>
+#include <X11/extensions/Xrandr.h>
 
 typedef struct {
     gsr_capture_nvfbc_params params;
@@ -55,40 +56,6 @@ static uint32_t get_output_id_from_display_name(NVFBC_RANDR_OUTPUT_INFO *outputs
     return 0;
 }
 
-/* TODO: Test with optimus and open kernel modules */
-static bool get_driver_version(int *major, int *minor) {
-    *major = 0;
-    *minor = 0;
-
-    FILE *f = fopen("/proc/driver/nvidia/version", "rb");
-    if(!f) {
-        fprintf(stderr, "gsr warning: failed to get nvidia driver version (failed to read /proc/driver/nvidia/version)\n");
-        return false;
-    }
-
-    char buffer[2048];
-    size_t bytes_read = fread(buffer, 1, sizeof(buffer) - 1, f);
-    buffer[bytes_read] = '\0';
-
-    bool success = false;
-    const char *p = strstr(buffer, "Kernel Module");
-    if(p) {
-        p += 13;
-        int driver_major_version = 0, driver_minor_version = 0;
-        if(sscanf(p, "%d.%d", &driver_major_version, &driver_minor_version) == 2) {
-            *major = driver_major_version;
-            *minor = driver_minor_version;
-            success = true;
-        }
-    }
-
-    if(!success)
-        fprintf(stderr, "gsr warning: failed to get nvidia driver version\n");
-
-    fclose(f);
-    return success;
-}
-
 static bool version_at_least(int major, int minor, int expected_major, int expected_minor) {
     return major > expected_major || (major == expected_major && minor >= expected_minor);
 }
@@ -288,13 +255,13 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, gsr_capture_metadata *captu
     self->supports_direct_cursor = false;
     int driver_major_version = 0;
     int driver_minor_version = 0;
-    if(self->params.direct_capture && get_driver_version(&driver_major_version, &driver_minor_version)) {
-        fprintf(stderr, "Info: detected nvidia version: %d.%d\n", driver_major_version, driver_minor_version);
+    if(self->params.direct_capture && get_nvidia_driver_version(&driver_major_version, &driver_minor_version)) {
+        fprintf(stderr, "gsr info: detected nvidia version: %d.%d\n", driver_major_version, driver_minor_version);
 
         // TODO:
         if(version_at_least(driver_major_version, driver_minor_version, 515, 57) && version_less_than(driver_major_version, driver_minor_version, 520, 56)) {
             self->params.direct_capture = false;
-            fprintf(stderr, "Warning: \"screen-direct\" has temporary been disabled as it causes stuttering with driver versions >= 515.57 and < 520.56. Please update your driver if possible. Capturing \"screen\" instead.\n");
+            fprintf(stderr, "gsr warning: \"screen-direct\" has temporary been disabled as it causes stuttering with driver versions >= 515.57 and < 520.56. Please update your driver if possible. Capturing \"screen\" instead.\n");
         }
 
         // TODO:
@@ -304,7 +271,7 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, gsr_capture_metadata *captu
             if(version_at_least(driver_major_version, driver_minor_version, 515, 57))
                 self->supports_direct_cursor = true;
             else
-                fprintf(stderr, "Info: capturing \"screen-direct\" but driver version appears to be less than 515.57. Disabling capture of cursor. Please update your driver if you want to capture your cursor or record \"screen\" instead.\n");
+                fprintf(stderr, "gsr info: capturing \"screen-direct\" but driver version appears to be less than 515.57. Disabling capture of cursor. Please update your driver if you want to capture your cursor or record \"screen\" instead.\n");
         }
         */
     }
@@ -317,16 +284,14 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, gsr_capture_metadata *captu
         goto error_cleanup;
     }
 
-    capture_metadata->width = self->tracking_width;
-    capture_metadata->height = self->tracking_height;
+    capture_metadata->video_size.x = self->tracking_width;
+    capture_metadata->video_size.y = self->tracking_height;
 
     if(self->params.output_resolution.x > 0 && self->params.output_resolution.y > 0) {
-        self->params.output_resolution = scale_keep_aspect_ratio((vec2i){capture_metadata->width, capture_metadata->height}, self->params.output_resolution);
-        capture_metadata->width = self->params.output_resolution.x;
-        capture_metadata->height = self->params.output_resolution.y;
+        self->params.output_resolution = scale_keep_aspect_ratio(capture_metadata->video_size, self->params.output_resolution);
+        capture_metadata->video_size = self->params.output_resolution;
     } else if(self->params.region_size.x > 0 && self->params.region_size.y > 0) {
-        capture_metadata->width = self->params.region_size.x;
-        capture_metadata->height = self->params.region_size.y;
+        capture_metadata->video_size = self->params.region_size;
     }
 
     return 0;
@@ -336,6 +301,35 @@ static int gsr_capture_nvfbc_start(gsr_capture *cap, gsr_capture_metadata *captu
     return -1;
 }
 
+static bool gsr_capture_nvfbc_is_capture_monitor_connected(gsr_capture_nvfbc *self) {
+    Display *dpy = gsr_window_get_display(self->params.egl->window);
+    int num_monitors = 0;
+    XRRMonitorInfo *monitors = XRRGetMonitors(dpy, DefaultRootWindow(dpy), True, &num_monitors);
+    if(!monitors)
+        return false;
+
+    bool capture_monitor_connected = false;
+    if(strcmp(self->params.display_to_capture, "screen") == 0) {
+        capture_monitor_connected = num_monitors > 0;
+    } else {
+        for(int i = 0; i < num_monitors; ++i) {
+            char *monitor_name = XGetAtomName(dpy, monitors[i].name);
+            if(!monitor_name)
+                continue;
+
+            if(strcmp(monitor_name, self->params.display_to_capture) == 0) {
+                capture_monitor_connected = true;
+                XFree(monitor_name);
+                break;
+            }
+            XFree(monitor_name);
+        }
+    }
+
+    XRRFreeMonitors(monitors);
+    return capture_monitor_connected;
+}
+
 static int gsr_capture_nvfbc_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
     gsr_capture_nvfbc *self = cap->priv;
 
@@ -344,6 +338,13 @@ static int gsr_capture_nvfbc_capture(gsr_capture *cap, gsr_capture_metadata *cap
         const double now = clock_get_monotonic_seconds();
         if(now - self->nvfbc_dead_start >= nvfbc_recreate_retry_time_seconds) {
             self->nvfbc_dead_start = now;
+            /*
+                Do not attempt to recreate the nvfbc session if the monitor isn't turned on/connected.
+                This is to predict if the nvfbc session create below will fail since if it fails it leaks an x11 display (a bug in the nvidia driver).
+            */
+            if(!gsr_capture_nvfbc_is_capture_monitor_connected(self))
+                return 0;
+
             gsr_capture_nvfbc_destroy_session_and_handle(self);
 
             if(gsr_capture_nvfbc_setup_handle(self) != 0) {
@@ -356,6 +357,7 @@ static int gsr_capture_nvfbc_capture(gsr_capture *cap, gsr_capture_metadata *cap
                 return -1;
             }
 
+            fprintf(stderr, "gsr info: gsr_capture_nvfbc_capture: recreated nvfbc session after modeset recovery\n");
             self->nvfbc_needs_recreate = false;
         } else {
             return 0;
@@ -363,14 +365,12 @@ static int gsr_capture_nvfbc_capture(gsr_capture *cap, gsr_capture_metadata *cap
     }
 
     vec2i frame_size = (vec2i){self->width, self->height};
+    const vec2i original_frame_size = frame_size;
     if(self->params.region_size.x > 0 && self->params.region_size.y > 0)
         frame_size = self->params.region_size;
 
-    const bool is_scaled = self->params.output_resolution.x > 0 && self->params.output_resolution.y > 0;
-    vec2i output_size = is_scaled ? self->params.output_resolution : frame_size;
-    output_size = scale_keep_aspect_ratio(frame_size, output_size);
-
-    const vec2i target_pos = { max_int(0, capture_metadata->width / 2 - output_size.x / 2), max_int(0, capture_metadata->height / 2 - output_size.y / 2) };
+    const vec2i output_size = scale_keep_aspect_ratio(frame_size, capture_metadata->recording_size);
+    const vec2i target_pos = gsr_capture_get_target_position(output_size, capture_metadata);
 
     NVFBC_FRAME_GRAB_INFO frame_info;
     memset(&frame_info, 0, sizeof(frame_info));
@@ -390,16 +390,16 @@ static int gsr_capture_nvfbc_capture(gsr_capture *cap, gsr_capture_metadata *cap
         return 0;
     }
 
-    self->params.egl->glFlush();
-    self->params.egl->glFinish();
+    //self->params.egl->glFlush();
+    //self->params.egl->glFinish();
 
     gsr_color_conversion_draw(color_conversion, self->setup_params.dwTextures[grab_params.dwTextureIndex],
         target_pos, (vec2i){output_size.x, output_size.y},
-        self->params.region_position, frame_size,
-        0.0f, false, GSR_SOURCE_COLOR_BGR);
+        self->params.region_position, frame_size, original_frame_size,
+        GSR_ROT_0, capture_metadata->flip, GSR_SOURCE_COLOR_BGR, false);
 
-    self->params.egl->glFlush();
-    self->params.egl->glFinish();
+    //self->params.egl->glFlush();
+    //self->params.egl->glFinish();
 
     return 0;
 }

src/capture/portal.c

@@ -11,6 +11,15 @@
 #include <limits.h>
 #include <assert.h>
 
+#define PORTAL_CAPTURE_CANCELED_BY_USER_EXIT_CODE 60
+
+typedef enum {
+    PORTAL_CAPTURE_SETUP_IDLE,
+    PORTAL_CAPTURE_SETUP_IN_PROGRESS,
+    PORTAL_CAPTURE_SETUP_FINISHED,
+    PORTAL_CAPTURE_SETUP_FAILED
+} gsr_portal_capture_setup_state;
+
 typedef struct {
     gsr_capture_portal_params params;
 
@@ -21,21 +30,22 @@ typedef struct {
 
     gsr_pipewire_video pipewire;
     vec2i capture_size;
-    gsr_pipewire_video_dmabuf_data dmabuf_data[GSR_PIPEWIRE_VIDEO_DMABUF_MAX_PLANES];
-    int num_dmabuf_data;
 
-    bool fast_path_failed;
-    bool mesa_supports_compute_only_vaapi_copy;
+    gsr_map_texture_output pipewire_data;
+
+    bool should_stop;
+    bool stop_is_error;
+    bool do_capture;
 } gsr_capture_portal;
 
 static void gsr_capture_portal_cleanup_plane_fds(gsr_capture_portal *self) {
-    for(int i = 0; i < self->num_dmabuf_data; ++i) {
-        if(self->dmabuf_data[i].fd > 0) {
-            close(self->dmabuf_data[i].fd);
-            self->dmabuf_data[i].fd = 0;
+    for(int i = 0; i < self->pipewire_data.num_dmabuf_data; ++i) {
+        if(self->pipewire_data.dmabuf_data[i].fd > 0) {
+            close(self->pipewire_data.dmabuf_data[i].fd);
+            self->pipewire_data.dmabuf_data[i].fd = 0;
         }
     }
-    self->num_dmabuf_data = 0;
+    self->pipewire_data.num_dmabuf_data = 0;
 }
 
 static void gsr_capture_portal_stop(gsr_capture_portal *self) {
@@ -55,38 +65,25 @@ static void gsr_capture_portal_stop(gsr_capture_portal *self) {
     }
 
     gsr_capture_portal_cleanup_plane_fds(self);
-
     gsr_pipewire_video_deinit(&self->pipewire);
-
-    if(self->session_handle) {
-        free(self->session_handle);
-        self->session_handle = NULL;
-    }
-
     gsr_dbus_deinit(&self->dbus);
 }
 
 static void gsr_capture_portal_create_input_textures(gsr_capture_portal *self) {
     self->params.egl->glGenTextures(1, &self->texture_map.texture_id);
     self->params.egl->glBindTexture(GL_TEXTURE_2D, self->texture_map.texture_id);
-    self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
     self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
 
     self->params.egl->glGenTextures(1, &self->texture_map.external_texture_id);
     self->params.egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, self->texture_map.external_texture_id);
-    self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
     self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     self->params.egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, 0);
 
     self->params.egl->glGenTextures(1, &self->texture_map.cursor_texture_id);
     self->params.egl->glBindTexture(GL_TEXTURE_2D, self->texture_map.cursor_texture_id);
-    self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
     self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
@@ -232,19 +229,13 @@ static int gsr_capture_portal_setup_dbus(gsr_capture_portal *self, int *pipewire
 }
 
 static bool gsr_capture_portal_get_frame_dimensions(gsr_capture_portal *self) {
-    gsr_pipewire_video_region region = {0, 0, 0, 0};
-    gsr_pipewire_video_region cursor_region = {0, 0, 0, 0};
     fprintf(stderr, "gsr info: gsr_capture_portal_start: waiting for pipewire negotiation\n");
 
     const double start_time = clock_get_monotonic_seconds();
     while(clock_get_monotonic_seconds() - start_time < 5.0) {
-        bool uses_external_image = false;
-        uint32_t fourcc = 0;
-        uint64_t modifiers = 0;
-        if(gsr_pipewire_video_map_texture(&self->pipewire, self->texture_map, &region, &cursor_region, self->dmabuf_data, &self->num_dmabuf_data, &fourcc, &modifiers, &uses_external_image)) {
-            gsr_capture_portal_cleanup_plane_fds(self);
-            self->capture_size.x = region.width;
-            self->capture_size.y = region.height;
+        if(gsr_pipewire_video_map_texture(&self->pipewire, self->texture_map, &self->pipewire_data)) {
+            self->capture_size.x = self->pipewire_data.region.width;
+            self->capture_size.y = self->pipewire_data.region.height;
             fprintf(stderr, "gsr info: gsr_capture_portal_start: pipewire negotiation finished\n");
             return true;
         }
@@ -255,62 +246,60 @@ static bool gsr_capture_portal_get_frame_dimensions(gsr_capture_portal *self) {
     return false;
 }
 
-static int gsr_capture_portal_start(gsr_capture *cap, gsr_capture_metadata *capture_metadata) {
-    gsr_capture_portal *self = cap->priv;
-
+static int gsr_capture_portal_setup(gsr_capture_portal *self, int fps) {
     gsr_capture_portal_create_input_textures(self);
 
     int pipewire_fd = 0;
     uint32_t pipewire_node = 0;
     const int response_status = gsr_capture_portal_setup_dbus(self, &pipewire_fd, &pipewire_node);
     if(response_status != 0) {
-        gsr_capture_portal_stop(self);
         // Response status values:
         // 0: Success, the request is carried out
         // 1: The user cancelled the interaction
         // 2: The user interaction was ended in some other way
         // Response status value 2 happens usually if there was some kind of error in the desktop portal on the system
         if(response_status == 2) {
-            fprintf(stderr, "gsr error: gsr_capture_portal_start: desktop portal capture failed. Either you Wayland compositor doesn't support desktop portal capture or it's incorrectly setup on your system\n");
+            fprintf(stderr, "gsr error: gsr_capture_portal_setup: desktop portal capture failed. Either you Wayland compositor doesn't support desktop portal capture or it's incorrectly setup on your system\n");
             return 50;
         } else if(response_status == 1) {
-            fprintf(stderr, "gsr error: gsr_capture_portal_start: desktop portal capture failed. It seems like desktop portal capture was canceled by the user.\n");
-            return 60;
+            fprintf(stderr, "gsr error: gsr_capture_portal_setup: desktop portal capture failed. It seems like desktop portal capture was canceled by the user.\n");
+            return PORTAL_CAPTURE_CANCELED_BY_USER_EXIT_CODE;
         } else {
             return -1;
         }
     }
 
-    fprintf(stderr, "gsr info: gsr_capture_portal_start: setting up pipewire\n");
+    fprintf(stderr, "gsr info: gsr_capture_portal_setup: setting up pipewire\n");
     /* TODO: support hdr when pipewire supports it */
     /* gsr_pipewire closes the pipewire fd, even on failure */
-    if(!gsr_pipewire_video_init(&self->pipewire, pipewire_fd, pipewire_node, capture_metadata->fps, self->params.record_cursor, self->params.egl)) {
-        fprintf(stderr, "gsr error: gsr_capture_portal_start: failed to setup pipewire with fd: %d, node: %" PRIu32 "\n", pipewire_fd, pipewire_node);
-        gsr_capture_portal_stop(self);
+    if(!gsr_pipewire_video_init(&self->pipewire, pipewire_fd, pipewire_node, fps, self->params.record_cursor, self->params.egl)) {
+        fprintf(stderr, "gsr error: gsr_capture_portal_setup: failed to setup pipewire with fd: %d, node: %" PRIu32 "\n", pipewire_fd, pipewire_node);
         return -1;
     }
-    fprintf(stderr, "gsr info: gsr_capture_portal_start: pipewire setup finished\n");
+    fprintf(stderr, "gsr info: gsr_capture_portal_setup: pipewire setup finished\n");
 
-    if(!gsr_capture_portal_get_frame_dimensions(self)) {
-        gsr_capture_portal_stop(self);
+    if(!gsr_capture_portal_get_frame_dimensions(self))
         return -1;
+
+    return 0;
+}
+
+static int gsr_capture_portal_start(gsr_capture *cap, gsr_capture_metadata *capture_metadata) {
+    gsr_capture_portal *self = cap->priv;
+
+    const int result = gsr_capture_portal_setup(self, capture_metadata->fps);
+    if(result != 0) {
+        gsr_capture_portal_stop(self);
+        return result;
     }
 
     if(self->params.output_resolution.x == 0 && self->params.output_resolution.y == 0) {
-        capture_metadata->width = self->capture_size.x;
-        capture_metadata->height = self->capture_size.y;
+        capture_metadata->video_size = self->capture_size;
     } else {
         self->params.output_resolution = scale_keep_aspect_ratio(self->capture_size, self->params.output_resolution);
-        capture_metadata->width = self->params.output_resolution.x;
-        capture_metadata->height = self->params.output_resolution.y;
+        capture_metadata->video_size = self->params.output_resolution;
     }
 
-    self->fast_path_failed = self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && !gl_driver_version_greater_than(&self->params.egl->gpu_info, 24, 0, 9);
-    if(self->fast_path_failed)
-        fprintf(stderr, "gsr warning: gsr_capture_kms_start: your amd driver (mesa) version is known to be buggy (<= version 24.0.9), falling back to opengl copy\n");
-
-    self->mesa_supports_compute_only_vaapi_copy = self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && gl_driver_version_greater_than(&self->params.egl->gpu_info, 24, 3, 6);
-
     return 0;
 }
 
@@ -318,98 +307,107 @@ static int max_int(int a, int b) {
     return a > b ? a : b;
 }
 
-static void gsr_capture_portal_fail_fast_path_if_not_fast(gsr_capture_portal *self, uint32_t pixel_format) {
-    const uint8_t pixel_format_color_depth_1 = (pixel_format >> 16) & 0xFF;
-    if(!self->fast_path_failed && self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && !self->mesa_supports_compute_only_vaapi_copy && (pixel_format_color_depth_1 == '3' || pixel_format_color_depth_1 == '4')) {
-        self->fast_path_failed = true;
-        fprintf(stderr, "gsr warning: gsr_capture_kms_capture: the monitor you are recording is in 10/12-bit color format and your mesa version is <= 24.3.6, composition will be used."
-            " If you experience performance problems in the video then record on a single window on X11 instead or disable 10/12-bit color option in your desktop environment settings,"
-            " or try to record the monitor on X11 instead (if you aren't already doing that) or update your mesa version.\n");
+static bool gsr_capture_portal_capture_has_synchronous_task(gsr_capture *cap) {
+    gsr_capture_portal *self = cap->priv;
+    return gsr_pipewire_video_should_restart(&self->pipewire);
+}
+
+static bool fourcc_has_alpha(uint32_t fourcc) {
+    const uint8_t *p = (const uint8_t*)&fourcc;
+    for(int i = 0; i < 4; ++i) {
+        if(p[i] == 'A')
+            return true;
     }
+    return false;
+}
+
+static void gsr_capture_portal_pre_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
+    gsr_capture_portal *self = cap->priv;
+    self->do_capture = false;
+
+    if(self->should_stop)
+        return;
+
+    if(gsr_pipewire_video_should_restart(&self->pipewire)) {
+        fprintf(stderr, "gsr info: gsr_capture_portal_pre_capture: pipewire capture was paused, trying to start capture again\n");
+        gsr_capture_portal_stop(self);
+        const int result = gsr_capture_portal_setup(self, capture_metadata->fps);
+        if(result != 0) {
+            self->stop_is_error = result != PORTAL_CAPTURE_CANCELED_BY_USER_EXIT_CODE;
+            self->should_stop = true;
+        }
+        return;
+    }
+
+    /* TODO: Handle formats other than RGB(A) */
+    if(self->pipewire_data.num_dmabuf_data == 0) {
+        if(gsr_pipewire_video_map_texture(&self->pipewire, self->texture_map, &self->pipewire_data)) {
+            if(self->pipewire_data.region.width != self->capture_size.x || self->pipewire_data.region.height != self->capture_size.y) {
+                self->capture_size.x = self->pipewire_data.region.width;
+                self->capture_size.y = self->pipewire_data.region.height;
+                color_conversion->schedule_clear = true;
+            }
+        } else {
+            return;
+        }
+    }
+
+    const bool fourcc_alpha = fourcc_has_alpha(self->pipewire_data.fourcc);
+    if(fourcc_alpha)
+        color_conversion->schedule_clear = true;
+
+    self->do_capture = true;
 }
 
 static int gsr_capture_portal_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
     (void)color_conversion;
     gsr_capture_portal *self = cap->priv;
 
-    /* TODO: Handle formats other than RGB(a) */
-    gsr_pipewire_video_region region = {0, 0, 0, 0};
-    gsr_pipewire_video_region cursor_region = {0, 0, 0, 0};
-    uint32_t pipewire_fourcc = 0;
-    uint64_t pipewire_modifiers = 0;
-    bool using_external_image = false;
-    if(gsr_pipewire_video_map_texture(&self->pipewire, self->texture_map, &region, &cursor_region, self->dmabuf_data, &self->num_dmabuf_data, &pipewire_fourcc, &pipewire_modifiers, &using_external_image)) {
-        if(region.width != self->capture_size.x || region.height != self->capture_size.y) {
-            self->capture_size.x = region.width;
-            self->capture_size.y = region.height;
-            gsr_color_conversion_clear(color_conversion);
-        }
-    } else {
-        return 0;
-    }
+    if(self->should_stop || !self->do_capture)
+        return -1;
 
-    gsr_capture_portal_fail_fast_path_if_not_fast(self, pipewire_fourcc);
+    const vec2i output_size = scale_keep_aspect_ratio(self->capture_size, capture_metadata->recording_size);
+    const vec2i target_pos = gsr_capture_get_target_position(output_size, capture_metadata);
 
-    const bool is_scaled = self->params.output_resolution.x > 0 && self->params.output_resolution.y > 0;
-    vec2i output_size = is_scaled ? self->params.output_resolution : self->capture_size;
-    output_size = scale_keep_aspect_ratio(self->capture_size, output_size);
-    
-    const vec2i target_pos = { max_int(0, capture_metadata->width / 2 - output_size.x / 2), max_int(0, capture_metadata->height / 2 - output_size.y / 2) };
+    const vec2i actual_texture_size = {self->pipewire_data.texture_width, self->pipewire_data.texture_height};
 
-    self->params.egl->glFlush();
-    self->params.egl->glFinish();
+    //self->params.egl->glFlush();
+    //self->params.egl->glFinish();
 
     // TODO: Handle region crop
 
-    /* Fast opengl free path */
-    if(!self->fast_path_failed && video_codec_context_is_vaapi(capture_metadata->video_codec_context) && self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD) {
-        int fds[4];
-        uint32_t offsets[4];
-        uint32_t pitches[4];
-        uint64_t modifiers[4];
-        for(int i = 0; i < self->num_dmabuf_data; ++i) {
-            fds[i] = self->dmabuf_data[i].fd;
-            offsets[i] = self->dmabuf_data[i].offset;
-            pitches[i] = self->dmabuf_data[i].stride;
-            modifiers[i] = pipewire_modifiers;
-        }
-        if(!vaapi_copy_drm_planes_to_video_surface(capture_metadata->video_codec_context, capture_metadata->frame, (vec2i){region.x, region.y}, self->capture_size, target_pos, output_size, pipewire_fourcc, self->capture_size, fds, offsets, pitches, modifiers, self->num_dmabuf_data)) {
-            fprintf(stderr, "gsr error: gsr_capture_portal_capture: vaapi_copy_drm_planes_to_video_surface failed, falling back to opengl copy. Please report this as an issue at https://github.com/dec05eba/gpu-screen-recorder-issues\n");
-            self->fast_path_failed = true;
-        }
-    } else {
-        self->fast_path_failed = true;
-    }
+    gsr_color_conversion_draw(color_conversion, self->pipewire_data.using_external_image ? self->texture_map.external_texture_id : self->texture_map.texture_id,
+       target_pos, output_size,
+       (vec2i){self->pipewire_data.region.x, self->pipewire_data.region.y}, (vec2i){self->pipewire_data.region.width, self->pipewire_data.region.height}, actual_texture_size,
+       gsr_monitor_rotation_to_rotation(self->pipewire_data.rotation), capture_metadata->flip, GSR_SOURCE_COLOR_RGB, self->pipewire_data.using_external_image);
 
-    if(self->fast_path_failed) {
-        gsr_color_conversion_draw(color_conversion, using_external_image ? self->texture_map.external_texture_id : self->texture_map.texture_id,
-            target_pos, output_size,
-            (vec2i){region.x, region.y}, self->capture_size,
-            0.0f, using_external_image, GSR_SOURCE_COLOR_RGB);
-    }
-
-    if(self->params.record_cursor && self->texture_map.cursor_texture_id > 0 && cursor_region.width > 0) {
+    if(self->params.record_cursor && self->texture_map.cursor_texture_id > 0 && self->pipewire_data.cursor_region.width > 0) {
         const vec2d scale = {
             self->capture_size.x == 0 ? 0 : (double)output_size.x / (double)self->capture_size.x,
             self->capture_size.y == 0 ? 0 : (double)output_size.y / (double)self->capture_size.y
         };
 
         const vec2i cursor_pos = {
-            target_pos.x + (cursor_region.x * scale.x),
-            target_pos.y + (cursor_region.y * scale.y)
+            target_pos.x + (self->pipewire_data.cursor_region.x * scale.x),
+            target_pos.y + (self->pipewire_data.cursor_region.y * scale.y)
         };
 
         self->params.egl->glEnable(GL_SCISSOR_TEST);
         self->params.egl->glScissor(target_pos.x, target_pos.y, output_size.x, output_size.y);
+
         gsr_color_conversion_draw(color_conversion, self->texture_map.cursor_texture_id,
-            (vec2i){cursor_pos.x, cursor_pos.y}, (vec2i){cursor_region.width * scale.x, cursor_region.height * scale.y},
-            (vec2i){0, 0}, (vec2i){cursor_region.width, cursor_region.height},
-            0.0f, false, GSR_SOURCE_COLOR_RGB);
+            (vec2i){cursor_pos.x, cursor_pos.y},
+            (vec2i){self->pipewire_data.cursor_region.width * scale.x, self->pipewire_data.cursor_region.height * scale.y},
+            (vec2i){0, 0},
+            (vec2i){self->pipewire_data.cursor_region.width, self->pipewire_data.cursor_region.height},
+            (vec2i){self->pipewire_data.cursor_region.width, self->pipewire_data.cursor_region.height},
+            gsr_monitor_rotation_to_rotation(self->pipewire_data.rotation), capture_metadata->flip, GSR_SOURCE_COLOR_RGB, false);
+
         self->params.egl->glDisable(GL_SCISSOR_TEST);
     }
 
-    self->params.egl->glFlush();
-    self->params.egl->glFinish();
+    //self->params.egl->glFlush();
+    //self->params.egl->glFinish();
 
     gsr_capture_portal_cleanup_plane_fds(self);
 
@@ -421,6 +419,13 @@ static bool gsr_capture_portal_uses_external_image(gsr_capture *cap) {
     return true;
 }
 
+static bool gsr_capture_portal_should_stop(gsr_capture *cap, bool *err) {
+    gsr_capture_portal *self = cap->priv;
+    if(err)
+        *err = self->stop_is_error;
+    return self->should_stop;
+}
+
 static bool gsr_capture_portal_is_damaged(gsr_capture *cap) {
     gsr_capture_portal *self = cap->priv;
     return gsr_pipewire_video_is_damaged(&self->pipewire);
@@ -462,7 +467,9 @@ gsr_capture* gsr_capture_portal_create(const gsr_capture_portal_params *params)
     *cap = (gsr_capture) {
         .start = gsr_capture_portal_start,
         .tick = NULL,
-        .should_stop = NULL,
+        .should_stop = gsr_capture_portal_should_stop,
+        .capture_has_synchronous_task = gsr_capture_portal_capture_has_synchronous_task,
+        .pre_capture = gsr_capture_portal_pre_capture,
         .capture = gsr_capture_portal_capture,
         .uses_external_image = gsr_capture_portal_uses_external_image,
         .is_damaged = gsr_capture_portal_is_damaged,

src/capture/v4l2.c

@@ -0,0 +1,919 @@
+#include "../../include/capture/v4l2.h"
+#include "../../include/color_conversion.h"
+#include "../../include/egl.h"
+#include "../../include/utils.h"
+
+#include <dlfcn.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <poll.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <linux/videodev2.h>
+#include <linux/dma-buf.h>
+#include <drm_fourcc.h>
+
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <errno.h>
+#include <assert.h>
+
+#define TJPF_RGB 0
+#define TJPF_RGBA 7
+#define TJFLAG_FASTDCT  2048
+
+#define NUM_BUFFERS 2
+#define NUM_PBOS 2
+
+typedef void* tjhandle;
+typedef tjhandle (*FUNC_tjInitDecompress)(void);
+typedef int (*FUNC_tjDestroy)(tjhandle handle);
+typedef int (*FUNC_tjDecompressHeader2)(tjhandle handle,
+                                        unsigned char *jpegBuf, unsigned long jpegSize,
+                                        int *width, int *height, int *jpegSubsamp);
+typedef int (*FUNC_tjDecompress2)(tjhandle handle, const unsigned char *jpegBuf,
+                                  unsigned long jpegSize, unsigned char *dstBuf,
+                                  int width, int pitch, int height, int pixelFormat,
+                                  int flags);
+typedef char* (*FUNC_tjGetErrorStr2)(tjhandle handle);
+
+typedef enum {
+    V4L2_BUFFER_TYPE_DMABUF,
+    V4L2_BUFFER_TYPE_MMAP
+} v4l2_buffer_type;
+
+typedef struct {
+    bool yuyv;
+    bool mjpeg;
+} gsr_capture_v4l2_supported_pixfmts;
+
+typedef struct {
+    gsr_capture_v4l2_params params;
+    vec2i capture_size;
+
+    bool should_stop;
+    bool stop_is_error;
+
+    int fd;
+    int dmabuf_fd[NUM_BUFFERS];
+    EGLImage dma_image[NUM_BUFFERS];
+    unsigned int texture_id[NUM_BUFFERS];
+    unsigned int prev_texture_index;
+    bool got_first_frame;
+
+    void *dmabuf_map[NUM_BUFFERS];
+    size_t dmabuf_size[NUM_BUFFERS];
+    unsigned int pbos[NUM_PBOS];
+    unsigned int pbo_index;
+    
+    v4l2_buffer_type buffer_type;
+
+    void *libturbojpeg_lib;
+    FUNC_tjInitDecompress tjInitDecompress;
+    FUNC_tjDestroy tjDestroy;
+    FUNC_tjDecompressHeader2 tjDecompressHeader2;
+    FUNC_tjDecompress2 tjDecompress2;
+    FUNC_tjGetErrorStr2 tjGetErrorStr2;
+    tjhandle jpeg_decompressor;
+
+    double capture_start_time;
+
+    bool yuyv_conversion_fallback;
+} gsr_capture_v4l2;
+
+static int xioctl(int fd, unsigned long request, void *arg) {
+    int r;
+
+    do {
+        r = ioctl(fd, request, arg);
+    } while (-1 == r && EINTR == errno);
+
+    return r;
+}
+
+static void gsr_capture_v4l2_stop(gsr_capture_v4l2 *self) {
+    self->params.egl->glDeleteBuffers(NUM_PBOS, self->pbos);
+    for(int i = 0; i < NUM_PBOS; ++i) {
+        self->pbos[i] = 0;
+    }
+
+    self->params.egl->glDeleteTextures(NUM_BUFFERS, self->texture_id);
+    for(int i = 0; i < NUM_BUFFERS; ++i) {
+        self->texture_id[i] = 0;
+    }
+
+    for(int i = 0; i < NUM_BUFFERS; ++i) {
+        if(self->dmabuf_map[i]) {
+            munmap(self->dmabuf_map[i], self->dmabuf_size[i]);
+            self->dmabuf_map[i] = NULL;
+        }
+
+        if(self->dma_image[i]) {
+            self->params.egl->eglDestroyImage(self->params.egl->egl_display, self->dma_image[i]);
+            self->dma_image[i] = NULL;
+        }
+
+        if(self->dmabuf_fd[i] > 0) {
+            close(self->dmabuf_fd[i]);
+            self->dmabuf_fd[i] = 0;
+        }
+    }
+
+    if(self->fd > 0) {
+        xioctl(self->fd, VIDIOC_STREAMOFF, &(enum v4l2_buf_type){V4L2_BUF_TYPE_VIDEO_CAPTURE});
+        close(self->fd);
+        self->fd = 0;
+    }
+
+    if(self->jpeg_decompressor) {
+        self->tjDestroy(self->jpeg_decompressor);
+        self->jpeg_decompressor = NULL;
+    }
+
+    if(self->libturbojpeg_lib) {
+        dlclose(self->libturbojpeg_lib);
+        self->libturbojpeg_lib = NULL;
+    }
+}
+
+static void gsr_capture_v4l2_reset_cropping(gsr_capture_v4l2 *self) {
+    struct v4l2_cropcap cropcap = {
+        .type = V4L2_BUF_TYPE_VIDEO_CAPTURE
+    };
+    if(xioctl(self->fd, VIDIOC_CROPCAP, &cropcap) == 0) {
+        struct v4l2_crop crop = {
+            .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+            .c = cropcap.defrect /* reset to default */
+        };
+
+        if(xioctl(self->fd, VIDIOC_S_CROP, &crop) == -1) {
+            switch (errno) {
+            case EINVAL:
+                /* Cropping not supported. */
+                break;
+            default:
+                /* Errors ignored. */
+                break;
+            }
+        }
+    } else {
+        /* Errors ignored. */
+    }
+}
+
+static uint32_t gsr_pixfmt_to_v4l2_pixfmt(gsr_capture_v4l2_pixfmt pixfmt) {
+    switch(pixfmt) {
+        case GSR_CAPTURE_V4L2_PIXFMT_AUTO:
+            assert(false);
+            break;
+        case GSR_CAPTURE_V4L2_PIXFMT_YUYV:
+            return V4L2_PIX_FMT_YUYV;
+        case GSR_CAPTURE_V4L2_PIXFMT_MJPEG:
+            return V4L2_PIX_FMT_MJPEG;
+    }
+    assert(false);
+    return V4L2_PIX_FMT_YUYV;
+}
+
+const char* gsr_capture_v4l2_pixfmt_to_string(gsr_capture_v4l2_pixfmt pixfmt) {
+    switch(pixfmt) {
+        case GSR_CAPTURE_V4L2_PIXFMT_AUTO:
+            assert(false);
+            break;
+        case GSR_CAPTURE_V4L2_PIXFMT_YUYV:
+            return "yuyv";
+        case GSR_CAPTURE_V4L2_PIXFMT_MJPEG:
+            return "mjpeg";
+    }
+    assert(false);
+    return "";
+}
+
+static gsr_capture_v4l2_supported_pixfmts gsr_capture_v4l2_get_supported_pixfmts(int fd) {
+    gsr_capture_v4l2_supported_pixfmts result = {0};
+
+    struct v4l2_fmtdesc fmt = {
+        .type = V4L2_BUF_TYPE_VIDEO_CAPTURE
+    };
+    while(xioctl(fd, VIDIOC_ENUM_FMT, &fmt) == 0) {
+        //fprintf(stderr, "fmt: %d, desc: %s, flags: %d\n", fmt.pixelformat, fmt.description, fmt.flags);
+        switch(fmt.pixelformat) {
+            case V4L2_PIX_FMT_YUYV:
+                result.yuyv = true;
+                break;
+            case V4L2_PIX_FMT_MJPEG:
+                result.mjpeg = true;
+                break;
+        }
+        ++fmt.index;
+    }
+
+    return result;
+}
+
+/* Returns the number of resolutions added */
+static size_t gsr_capture_v4l2_get_supported_resolutions(int fd, gsr_capture_v4l2_pixfmt pixfmt, gsr_capture_v4l2_resolution *resolutions, size_t max_resolutions) {
+    size_t resolution_index = 0;
+    struct v4l2_frmsizeenum fmt = {
+        .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+        .pixel_format = gsr_pixfmt_to_v4l2_pixfmt(pixfmt),
+    };
+
+    while(xioctl(fd, VIDIOC_ENUM_FRAMESIZES, &fmt) == 0) {
+        if(fmt.type == V4L2_FRMSIZE_TYPE_DISCRETE && resolution_index < max_resolutions) {
+            resolutions[resolution_index] = (gsr_capture_v4l2_resolution){
+                .width = fmt.discrete.width,
+                .height = fmt.discrete.height,
+            };
+            ++resolution_index;
+        }
+        ++fmt.index;
+    }
+
+    return resolution_index;
+}
+
+/* Returns the number of framerates added */
+static size_t gsr_capture_v4l2_get_supported_framerates(int fd, gsr_capture_v4l2_pixfmt pixfmt, gsr_capture_v4l2_resolution resolution, gsr_capture_v4l2_framerate *framerates, size_t max_framerates) {
+    size_t framerate_index = 0;
+    struct v4l2_frmivalenum fmt = {
+        .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+        .pixel_format = gsr_pixfmt_to_v4l2_pixfmt(pixfmt),
+        .width = resolution.width,
+        .height = resolution.height,
+    };
+
+    while(xioctl(fd, VIDIOC_ENUM_FRAMEINTERVALS, &fmt) == 0) {
+        if(fmt.type == V4L2_FRMIVAL_TYPE_DISCRETE && fmt.discrete.denominator > 0 && fmt.discrete.numerator > 0 && framerate_index < max_framerates) {
+            framerates[framerate_index] = (gsr_capture_v4l2_framerate){
+                .denominator = fmt.discrete.denominator,
+                .numerator = fmt.discrete.numerator,
+            };
+            ++framerate_index;
+        }
+        ++fmt.index;
+    }
+
+    return framerate_index;
+}
+
+/* Returns the number of setups added */
+static size_t gsr_capture_v4l2_get_supported_setups(int fd, gsr_capture_v4l2_supported_setup *supported_setups, size_t max_supported_setups, bool has_libturbojpeg_lib) {
+    const gsr_capture_v4l2_supported_pixfmts supported_pixfmts = gsr_capture_v4l2_get_supported_pixfmts(fd);
+
+    size_t num_pixfmts = 0;
+    gsr_capture_v4l2_pixfmt pixfmts[2];
+
+    if(supported_pixfmts.yuyv)
+        pixfmts[num_pixfmts++] = GSR_CAPTURE_V4L2_PIXFMT_YUYV;
+
+    if(supported_pixfmts.mjpeg && has_libturbojpeg_lib)
+        pixfmts[num_pixfmts++] = GSR_CAPTURE_V4L2_PIXFMT_MJPEG;
+
+    gsr_capture_v4l2_resolution resolutions[32];
+    gsr_capture_v4l2_framerate framerates[32];
+    size_t supported_setup_index = 0;
+
+    for(size_t pixfmt_index = 0; pixfmt_index < num_pixfmts; ++pixfmt_index) {
+        const gsr_capture_v4l2_pixfmt pixfmt = pixfmts[pixfmt_index];
+        const size_t num_resolutions = gsr_capture_v4l2_get_supported_resolutions(fd, pixfmt, resolutions, 32);
+
+        for(size_t resolution_index = 0; resolution_index < num_resolutions; ++resolution_index) {
+            const gsr_capture_v4l2_resolution resolution = resolutions[resolution_index];
+            const size_t num_framerates = gsr_capture_v4l2_get_supported_framerates(fd, pixfmt, resolution, framerates, 32);
+
+            for(size_t framerate_index = 0; framerate_index < num_framerates; ++framerate_index) {
+                const gsr_capture_v4l2_framerate framerate = framerates[framerate_index];
+
+                if(supported_setup_index < max_supported_setups) {
+                    supported_setups[supported_setup_index] = (gsr_capture_v4l2_supported_setup){
+                        .pixfmt = pixfmt,
+                        .resolution = resolution,
+                        .framerate = framerate,
+                    };
+                    ++supported_setup_index;
+                }
+            }
+        }
+    }
+
+    return supported_setup_index;
+}
+
+uint32_t gsr_capture_v4l2_framerate_to_number(gsr_capture_v4l2_framerate framerate) {
+    return (uint32_t)((double)framerate.denominator / (double)framerate.numerator);
+}
+
+static bool gsr_capture_v4l2_get_best_matching_setup(
+    const gsr_capture_v4l2_supported_setup *supported_setups,
+    size_t num_supported_setups,
+    gsr_capture_v4l2_pixfmt pixfmt,
+    uint32_t camera_fps,
+    gsr_capture_v4l2_resolution camera_resolution,
+    gsr_capture_v4l2_supported_setup *best_supported_setup)
+{
+    memset(best_supported_setup, 0, sizeof(*best_supported_setup));
+
+    int best_match_index = -1;
+    uint64_t best_match_score = 0;
+
+    for(size_t i = 0; i < num_supported_setups; ++i) {
+        const gsr_capture_v4l2_supported_setup *setup = &supported_setups[i];
+        if(pixfmt != GSR_CAPTURE_V4L2_PIXFMT_AUTO && pixfmt != setup->pixfmt)
+            continue;
+
+        uint64_t setup_resolution_width = (uint64_t)setup->resolution.width;
+        uint64_t setup_resolution_height = (uint64_t)setup->resolution.height;
+        uint64_t setup_framerate = gsr_capture_v4l2_framerate_to_number(setup->framerate);
+
+        if(setup_resolution_width == camera_resolution.width && setup_resolution_height == camera_resolution.height) {
+            setup_resolution_width = 50000;
+            setup_resolution_height = 50000;
+        }
+
+        if(setup_framerate == camera_fps) {
+            setup_framerate = 50000;
+        }
+
+        const uint64_t match_score = setup_resolution_width * setup_resolution_height * setup_framerate + (pixfmt == GSR_CAPTURE_V4L2_PIXFMT_YUYV ? 5 : 0);
+        if(match_score > best_match_score) {
+            best_match_score = match_score;
+            best_match_index = i;
+        }
+
+        //fprintf(stderr, "supported setup[%d]: pixfmt: %d, size: %ux%u, fps: %u/%u\n", (int)i, setup->pixfmt, setup->resolution.width, setup->resolution.height, setup->framerate.denominator, setup->framerate.numerator);
+    }
+
+    if(best_match_index == -1)
+        return false;
+
+    //fprintf(stderr, "best match index: %d\n", best_match_index);
+    *best_supported_setup = supported_setups[best_match_index];
+    return true;
+}
+
+/* Seems like some cameras need this? */
+static void gsr_capture_v4l2_update_params(int fd) {
+    struct v4l2_streamparm streamparm = {
+        .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+    };
+    if(xioctl(fd, VIDIOC_G_PARM, &streamparm) == -1) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_set_framerate: VIDIOC_G_PARM failed, error: %s\n", strerror(errno));
+        return;
+    }
+
+    if(xioctl(fd, VIDIOC_S_PARM, &streamparm) == -1) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_set_framerate: VIDIOC_S_PARM failed, error: %s\n", strerror(errno));
+        return;
+    }
+}
+
+static void gsr_capture_v4l2_set_framerate(int fd, gsr_capture_v4l2_framerate framerate) {
+    struct v4l2_streamparm streamparm = {
+        .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+    };
+    if(xioctl(fd, VIDIOC_G_PARM, &streamparm) == -1) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_set_framerate: VIDIOC_G_PARM failed, error: %s\n", strerror(errno));
+        return;
+    }
+
+    streamparm.parm.capture.timeperframe.denominator = framerate.denominator;
+    streamparm.parm.capture.timeperframe.numerator = framerate.numerator;
+    if(xioctl(fd, VIDIOC_S_PARM, &streamparm) == -1) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_set_framerate: VIDIOC_S_PARM failed, error: %s\n", strerror(errno));
+        return;
+    }
+
+    if(streamparm.parm.capture.timeperframe.denominator == 0 || streamparm.parm.capture.timeperframe.numerator == 0) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_set_framerate: VIDIOC_S_PARM failed, error: invalid framerate: %u/%u\n", framerate.denominator, framerate.numerator);;
+        return;
+    }
+}
+
+static bool gsr_capture_v4l2_validate_pixfmt(const gsr_capture_v4l2 *self, const gsr_capture_v4l2_supported_pixfmts supported_pixfmts) {
+    switch(self->params.pixfmt) {
+        case GSR_CAPTURE_V4L2_PIXFMT_AUTO: {
+            if(!supported_pixfmts.yuyv && !supported_pixfmts.mjpeg) {
+                fprintf(stderr, "gsr error: gsr_capture_v4l2_create: %s doesn't support yuyv nor mjpeg. GPU Screen Recorder supports only yuyv and mjpeg at the moment. Report this as an issue, see: https://git.dec05eba.com/?p=about\n", self->params.device_path);
+                return false;
+            }
+            break;
+        }
+        case GSR_CAPTURE_V4L2_PIXFMT_YUYV: {
+            if(!supported_pixfmts.yuyv) {
+                fprintf(stderr, "gsr error: gsr_capture_v4l2_create: %s doesn't support yuyv. Try recording with pixfmt=mjpeg or pixfmt=auto instead\n", self->params.device_path);
+                return false;
+            }
+            break;
+        }
+        case GSR_CAPTURE_V4L2_PIXFMT_MJPEG: {
+            if(!supported_pixfmts.mjpeg) {
+                fprintf(stderr, "gsr error: gsr_capture_v4l2_create: %s doesn't support mjpeg. Try recording with pixfmt=yuyv or pixfmt=auto instead\n", self->params.device_path);
+                return false;
+            }
+            break;
+        }
+    }
+    return true;
+}
+
+static bool gsr_capture_v4l2_create_pbos(gsr_capture_v4l2 *self, int width, int height) {
+    self->pbo_index = 0;
+
+    self->params.egl->glGenBuffers(NUM_PBOS, self->pbos);
+    for(int i = 0; i < NUM_PBOS; ++i) {
+        if(self->pbos[i] == 0) {
+            fprintf(stderr, "gsr error: gsr_capture_v4l2_create_pbos: failed to create pixel buffer objects\n");
+            return false;
+        }
+
+        self->params.egl->glBindBuffer(GL_PIXEL_UNPACK_BUFFER, self->pbos[i]);
+        self->params.egl->glBufferData(GL_PIXEL_UNPACK_BUFFER, width * height * 4, 0, GL_DYNAMIC_DRAW);
+    }
+
+    self->params.egl->glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
+    return true;
+}
+
+static bool gsr_capture_v4l2_map_buffer(gsr_capture_v4l2 *self, const struct v4l2_format *fmt) {
+    switch(self->params.pixfmt) {
+        case GSR_CAPTURE_V4L2_PIXFMT_AUTO: {
+            assert(false);
+            return false;
+        }
+        case GSR_CAPTURE_V4L2_PIXFMT_YUYV: {
+            self->params.egl->glGenTextures(NUM_BUFFERS, self->texture_id);
+
+            for(int i = 0; i < NUM_BUFFERS; ++i) {
+                self->dma_image[i] = self->params.egl->eglCreateImage(self->params.egl->egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, (intptr_t[]) {
+                    EGL_WIDTH, fmt->fmt.pix.width,
+                    EGL_HEIGHT, fmt->fmt.pix.height,
+                    EGL_LINUX_DRM_FOURCC_EXT, DRM_FORMAT_YUYV,
+                    EGL_DMA_BUF_PLANE0_FD_EXT, self->dmabuf_fd[i],
+                    EGL_DMA_BUF_PLANE0_OFFSET_EXT, 0,
+                    EGL_DMA_BUF_PLANE0_PITCH_EXT, fmt->fmt.pix.bytesperline,
+                    EGL_NONE
+                });
+
+                if(!self->dma_image[i]) {
+                    self->yuyv_conversion_fallback = true;
+                    // This doesn't work for everybody on nvidia. On pop os 24.04 when egl_display is NULL (when capturing monitor on x11) then this fails with EGL_BAD_DISPLAY
+                    self->dma_image[i] = self->params.egl->eglCreateImage(self->params.egl->egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, (intptr_t[]) {
+                        EGL_WIDTH, fmt->fmt.pix.width,
+                        EGL_HEIGHT, fmt->fmt.pix.height,
+                        EGL_LINUX_DRM_FOURCC_EXT, DRM_FORMAT_RG88,
+                        EGL_DMA_BUF_PLANE0_FD_EXT, self->dmabuf_fd[i],
+                        EGL_DMA_BUF_PLANE0_OFFSET_EXT, 0,
+                        EGL_DMA_BUF_PLANE0_PITCH_EXT, fmt->fmt.pix.bytesperline,
+                        EGL_NONE
+                    });
+
+                    if(!self->dma_image[i]) {
+                        fprintf(stderr, "gsr error: gsr_capture_v4l2_map_buffer: eglCreateImage failed, error: %d\n", self->params.egl->eglGetError());
+                        return false;
+                    }
+                }
+
+                self->params.egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, self->texture_id[i]);
+                self->params.egl->glEGLImageTargetTexture2DOES(GL_TEXTURE_EXTERNAL_OES, self->dma_image[i]);
+                self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+                self->params.egl->glTexParameteri(GL_TEXTURE_EXTERNAL_OES, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
+                self->params.egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, 0);
+                if(self->texture_id[i] == 0) {
+                    fprintf(stderr, "gsr error: gsr_capture_v4l2_map_buffer: failed to create texture\n");
+                    return false;
+                }
+            }
+
+            self->buffer_type = V4L2_BUFFER_TYPE_DMABUF;
+            break;
+        }
+        case GSR_CAPTURE_V4L2_PIXFMT_MJPEG: {
+            for(int i = 0; i < NUM_BUFFERS; ++i) {
+                self->dmabuf_size[i] = fmt->fmt.pix.sizeimage;
+                self->dmabuf_map[i] = mmap(NULL, fmt->fmt.pix.sizeimage, PROT_READ, MAP_SHARED, self->dmabuf_fd[i], 0);
+                if(self->dmabuf_map[i] == MAP_FAILED) {
+                    fprintf(stderr, "gsr error: gsr_capture_v4l2_map_buffer: mmap failed, error: %s\n", strerror(errno));
+                    return false;
+                }
+
+                // GL_RGBA is intentionally used here instead of GL_RGB, because the performance is much better when using glTexSubImage2D (22% cpu usage compared to 38% cpu usage)
+                self->texture_id[i] = gl_create_texture(self->params.egl, fmt->fmt.pix.width, fmt->fmt.pix.height, GL_RGBA8, GL_RGBA, GL_LINEAR);
+                if(self->texture_id[i] == 0) {
+                    fprintf(stderr, "gsr error: gsr_capture_v4l2_map_buffer: failed to create texture\n");
+                    return false;
+                }
+            }
+
+            if(!gsr_capture_v4l2_create_pbos(self, fmt->fmt.pix.width, fmt->fmt.pix.height))
+                return false;
+
+            self->buffer_type = V4L2_BUFFER_TYPE_MMAP;
+            break;
+        }
+    }
+    return true;
+}
+
+static bool is_libturbojpeg_library_available(void) {
+    void *libturbojpeg_lib = dlopen("libturbojpeg.so.0", RTLD_LAZY);
+    const bool has_libturbojpeg_lib = libturbojpeg_lib != NULL;
+    if(libturbojpeg_lib)
+        dlclose(libturbojpeg_lib);
+    return has_libturbojpeg_lib;
+}
+
+static int gsr_capture_v4l2_setup(gsr_capture_v4l2 *self) {
+    self->fd = open(self->params.device_path, O_RDWR | O_NONBLOCK);
+    if(self->fd < 0) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_create: failed to open %s, error: %s\n", self->params.device_path, strerror(errno));
+        return -1;
+    }
+
+    struct v4l2_capability cap = {0};
+    if(xioctl(self->fd, VIDIOC_QUERYCAP, &cap) == -1) {
+        if(EINVAL == errno) {
+            fprintf(stderr, "gsr error: gsr_capture_v4l2_create: %s isn't a v4l2 device\n", self->params.device_path);
+            return -1;
+        } else {
+            fprintf(stderr, "gsr error: gsr_capture_v4l2_create: VIDIOC_QUERYCAP failed, error: %s\n", strerror(errno));
+            return -1;
+        }
+    }
+
+    if(!(cap.capabilities & V4L2_CAP_VIDEO_CAPTURE)) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_create: %s isn't a video capture device\n", self->params.device_path);
+        return -1;
+    }
+
+    if(!(cap.capabilities & V4L2_CAP_STREAMING)) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_create: %s doesn't support streaming i/o\n", self->params.device_path);
+        return -1;
+    }
+
+    gsr_capture_v4l2_reset_cropping(self);
+
+    const bool has_libturbojpeg_lib = is_libturbojpeg_library_available();
+    if(!has_libturbojpeg_lib && self->params.pixfmt == GSR_CAPTURE_V4L2_PIXFMT_AUTO) {
+        fprintf(stderr, "gsr warning: gsr_capture_v4l2_create: libturbojpeg.so.0 isn't available on the system, yuyv camera capture will be used\n");
+        self->params.pixfmt = GSR_CAPTURE_V4L2_PIXFMT_YUYV;
+    }
+
+    const gsr_capture_v4l2_supported_pixfmts supported_pixfmts = gsr_capture_v4l2_get_supported_pixfmts(self->fd);
+    if(!gsr_capture_v4l2_validate_pixfmt(self, supported_pixfmts))
+        return -1;
+
+    gsr_capture_v4l2_supported_setup supported_setups[128];
+    const size_t num_supported_setups = gsr_capture_v4l2_get_supported_setups(self->fd, supported_setups, 128, has_libturbojpeg_lib);
+
+    gsr_capture_v4l2_supported_setup best_supported_setup = {0};
+    if(!gsr_capture_v4l2_get_best_matching_setup(supported_setups, num_supported_setups, self->params.pixfmt, self->params.camera_fps, self->params.camera_resolution, &best_supported_setup)) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_create: %s doesn't report any frame resolutions and framerates\n", self->params.device_path);
+        return -1;
+    }
+
+    fprintf(stderr, "gsr info: gsr_capture_v4l2_create: capturing %s at %ux%u@%dhz, pixfmt: %s\n",
+        self->params.device_path,
+        best_supported_setup.resolution.width,
+        best_supported_setup.resolution.height,
+        gsr_capture_v4l2_framerate_to_number(best_supported_setup.framerate),
+        gsr_capture_v4l2_pixfmt_to_string(best_supported_setup.pixfmt));
+
+    gsr_capture_v4l2_update_params(self->fd);
+    self->params.pixfmt = best_supported_setup.pixfmt;
+
+    if(self->params.pixfmt == GSR_CAPTURE_V4L2_PIXFMT_MJPEG) {
+        dlerror(); /* clear */
+        self->libturbojpeg_lib = dlopen("libturbojpeg.so.0", RTLD_LAZY);
+        if(!self->libturbojpeg_lib) {
+            fprintf(stderr, "gsr error: gsr_capture_v4l2_create: failed to load libturbojpeg.so.0 which is required for camera mjpeg capture, error: %s\n", dlerror());
+            return -1;
+        }
+
+        self->tjInitDecompress = (FUNC_tjInitDecompress)dlsym(self->libturbojpeg_lib, "tjInitDecompress");
+        self->tjDestroy = (FUNC_tjDestroy)dlsym(self->libturbojpeg_lib, "tjDestroy");
+        self->tjDecompressHeader2 = (FUNC_tjDecompressHeader2)dlsym(self->libturbojpeg_lib, "tjDecompressHeader2");
+        self->tjDecompress2 = (FUNC_tjDecompress2)dlsym(self->libturbojpeg_lib, "tjDecompress2");
+        self->tjGetErrorStr2 = (FUNC_tjGetErrorStr2)dlsym(self->libturbojpeg_lib, "tjGetErrorStr2");
+
+        if(!self->tjInitDecompress || !self->tjDestroy || !self->tjDecompressHeader2 || !self->tjDecompress2 || !self->tjGetErrorStr2) {
+            fprintf(stderr, "gsr error: gsr_capture_v4l2_create: libturbojpeg.so.0 is missing functions. The libturbojpeg version installed on your system might be outdated\n");
+            return -1;
+        }
+
+        self->jpeg_decompressor = self->tjInitDecompress();
+        if(!self->jpeg_decompressor) {
+            fprintf(stderr, "gsr error: gsr_capture_v4l2_create: failed to create jpeg decompressor\n");
+            return -1;
+        }
+    }
+
+    const uint32_t v4l2_pixfmt = gsr_pixfmt_to_v4l2_pixfmt(self->params.pixfmt);
+    struct v4l2_format fmt = {
+        .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+        .fmt.pix.pixelformat = v4l2_pixfmt,
+        .fmt.pix.width = best_supported_setup.resolution.width,
+        .fmt.pix.height = best_supported_setup.resolution.height,
+    };
+    if(xioctl(self->fd, VIDIOC_S_FMT, &fmt) == -1) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_create: VIDIOC_S_FMT failed, error: %s\n", strerror(errno));
+        return -1;
+    }
+
+    if(fmt.fmt.pix.pixelformat != v4l2_pixfmt) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_create: pixel format isn't as requested (got pixel format: %u, requested: %u), error: %s\n", fmt.fmt.pix.pixelformat, v4l2_pixfmt, strerror(errno));
+        return -1;
+    }
+
+    self->capture_size.x = fmt.fmt.pix.width;
+    self->capture_size.y = fmt.fmt.pix.height;
+
+    gsr_capture_v4l2_set_framerate(self->fd, best_supported_setup.framerate);
+
+    struct v4l2_requestbuffers reqbuf = {
+        .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+        .memory = V4L2_MEMORY_MMAP,
+        .count = NUM_BUFFERS
+    };
+    if(xioctl(self->fd, VIDIOC_REQBUFS, &reqbuf) == -1) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_create: VIDIOC_REQBUFS failed, error: %s\n", strerror(errno));
+        return -1;
+    }
+
+    for(int i = 0; i < NUM_BUFFERS; ++i) {
+        struct v4l2_exportbuffer expbuf = {
+            .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+            .index = i,
+            .flags = O_RDONLY
+        };
+        if(xioctl(self->fd, VIDIOC_EXPBUF, &expbuf) == -1) {
+            fprintf(stderr, "gsr error: gsr_capture_v4l2_create: VIDIOC_EXPBUF failed, error: %s\n", strerror(errno));
+            return -1;
+        }
+        self->dmabuf_fd[i] = expbuf.fd;
+    }
+
+    if(!gsr_capture_v4l2_map_buffer(self, &fmt))
+        return -1;
+
+    for(int i = 0; i < NUM_BUFFERS; ++i) {
+        struct v4l2_buffer buf = {
+            .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+            .index = i,
+            .memory = V4L2_MEMORY_MMAP
+        };
+        xioctl(self->fd, VIDIOC_QBUF, &buf);
+    }
+
+    if(xioctl(self->fd, VIDIOC_STREAMON, &(enum v4l2_buf_type){V4L2_BUF_TYPE_VIDEO_CAPTURE})) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_create: VIDIOC_STREAMON failed, error: %s\n", strerror(errno));
+        return -1;
+    }
+
+    fprintf(stderr, "gsr info: gsr_capture_v4l2_create: waiting for camera %s to be ready\n", self->params.device_path);
+    return 0;
+}
+
+static int gsr_capture_v4l2_start(gsr_capture *cap, gsr_capture_metadata *capture_metadata) {
+    gsr_capture_v4l2 *self = cap->priv;
+
+    const int result = gsr_capture_v4l2_setup(self);
+    if(result != 0) {
+        gsr_capture_v4l2_stop(self);
+        return result;
+    }
+
+    if(self->params.output_resolution.x == 0 && self->params.output_resolution.y == 0) {
+        capture_metadata->video_size = self->capture_size;
+    } else {
+        self->params.output_resolution = scale_keep_aspect_ratio(self->capture_size, self->params.output_resolution);
+        capture_metadata->video_size = self->params.output_resolution;
+    }
+
+    self->capture_start_time = clock_get_monotonic_seconds();
+    return 0;
+}
+
+static void gsr_capture_v4l2_tick(gsr_capture *cap) {
+    gsr_capture_v4l2 *self = cap->priv;
+    if(!self->got_first_frame && !self->should_stop) {
+        const double timeout_sec = 5.0;
+        if(clock_get_monotonic_seconds() - self->capture_start_time >= timeout_sec) {
+            fprintf(stderr, "gsr error: gsr_capture_v4l2_capture: didn't receive camera data in %f seconds\n", timeout_sec);
+            self->should_stop = true;
+            self->stop_is_error = true;
+        }
+    }
+}
+
+static void gsr_capture_v4l2_decode_jpeg_to_texture(gsr_capture_v4l2 *self, const struct v4l2_buffer *buf) {
+    int jpeg_subsamp = 0;
+    int jpeg_width = 0;
+    int jpeg_height = 0;
+    if(self->tjDecompressHeader2(self->jpeg_decompressor, self->dmabuf_map[buf->index], buf->bytesused, &jpeg_width, &jpeg_height, &jpeg_subsamp) != 0) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_capture: failed to decompress camera jpeg header data, error: %s\n", self->tjGetErrorStr2(self->jpeg_decompressor));
+        return;
+    }
+
+    if(jpeg_width != self->capture_size.x || jpeg_height != self->capture_size.y) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_capture: got jpeg data of incorrect dimensions. Expected %dx%d, got %dx%d\n", self->capture_size.x, self->capture_size.y, jpeg_width, jpeg_height);
+        return;
+    }
+
+    self->params.egl->glBindTexture(GL_TEXTURE_2D, self->texture_id[buf->index]);
+
+    self->pbo_index = (self->pbo_index + 1) % NUM_PBOS;
+    const unsigned int next_pbo_index = (self->pbo_index + 1) % NUM_PBOS;
+    
+    self->params.egl->glBindBuffer(GL_PIXEL_UNPACK_BUFFER, self->pbos[self->pbo_index]);
+    self->params.egl->glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, self->capture_size.x, self->capture_size.y, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+
+    self->params.egl->glBindBuffer(GL_PIXEL_UNPACK_BUFFER, self->pbos[next_pbo_index]);
+    self->params.egl->glBufferData(GL_PIXEL_UNPACK_BUFFER, self->capture_size.x * self->capture_size.y * 4, 0, GL_DYNAMIC_DRAW);
+
+    void *mapped_buffer = self->params.egl->glMapBufferRange(GL_PIXEL_UNPACK_BUFFER, 0, self->capture_size.x * self->capture_size.y * 4, GL_MAP_WRITE_BIT);
+    if(mapped_buffer) {
+        if(self->tjDecompress2(self->jpeg_decompressor, self->dmabuf_map[buf->index], buf->bytesused, mapped_buffer, jpeg_width, 0, jpeg_height, TJPF_RGBA, TJFLAG_FASTDCT) != 0)
+            fprintf(stderr, "gsr error: gsr_capture_v4l2_capture: failed to decompress camera jpeg data, error: %s\n", self->tjGetErrorStr2(self->jpeg_decompressor));
+        self->params.egl->glUnmapBuffer(GL_PIXEL_UNPACK_BUFFER);
+    }
+
+    self->params.egl->glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
+    self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
+}
+
+static int gsr_capture_v4l2_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
+    (void)color_conversion;
+    gsr_capture_v4l2 *self = cap->priv;
+
+    struct v4l2_buffer buf = {
+        .type = V4L2_BUF_TYPE_VIDEO_CAPTURE,
+        .memory = V4L2_MEMORY_MMAP
+    };
+
+    xioctl(self->fd, VIDIOC_DQBUF, &buf);
+    unsigned int texture_index = buf.index;
+
+    if(buf.bytesused > 0 && !(buf.flags & V4L2_BUF_FLAG_ERROR)) {
+        if(!self->got_first_frame)
+            fprintf(stderr, "gsr info: gsr_capture_v4l2_capture: camera %s is now ready\n", self->params.device_path);
+        self->got_first_frame = true;
+
+        switch(self->buffer_type) {
+            case V4L2_BUFFER_TYPE_DMABUF: {
+                //self->params.egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, self->texture_id);
+                //self->params.egl->glEGLImageTargetTexture2DOES(GL_TEXTURE_EXTERNAL_OES, self->dma_image[buf.index]);
+                //self->params.egl->glBindTexture(GL_TEXTURE_EXTERNAL_OES, 0);
+                break;
+            }
+            case V4L2_BUFFER_TYPE_MMAP: {
+                //xioctl(self->dmabuf_fd[buf.index], DMA_BUF_IOCTL_SYNC, &(struct dma_buf_sync){ .flags = DMA_BUF_SYNC_START });
+                gsr_capture_v4l2_decode_jpeg_to_texture(self, &buf);
+                //xioctl(self->dmabuf_fd[buf.index], DMA_BUF_IOCTL_SYNC, &(struct dma_buf_sync){ .flags = DMA_BUF_SYNC_END });
+                break;
+            }
+        }
+
+        self->prev_texture_index = buf.index;
+    } else {
+        texture_index = self->prev_texture_index;
+    }
+    xioctl(self->fd, VIDIOC_QBUF, &buf);
+
+    const vec2i output_size = scale_keep_aspect_ratio(self->capture_size, capture_metadata->recording_size);
+    const vec2i target_pos = gsr_capture_get_target_position(output_size, capture_metadata);
+
+    self->params.egl->glFlush();
+    // TODO: Use the minimal barrier required
+    self->params.egl->glMemoryBarrier(GL_ALL_BARRIER_BITS);
+    // TODO: Remove this?
+    if(self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA)
+        self->params.egl->glFinish();
+
+    if(self->buffer_type == V4L2_BUFFER_TYPE_DMABUF) {
+        gsr_color_conversion_draw(color_conversion, self->texture_id[texture_index],
+            target_pos, output_size,
+            (vec2i){0, 0}, self->capture_size, self->capture_size,
+            GSR_ROT_0, capture_metadata->flip, self->yuyv_conversion_fallback ? GSR_SOURCE_COLOR_YUYV : GSR_SOURCE_COLOR_RGB, true);
+    } else {
+        gsr_color_conversion_draw(color_conversion, self->texture_id[texture_index],
+            target_pos, output_size,
+            (vec2i){0, 0}, self->capture_size, self->capture_size,
+            GSR_ROT_0, capture_metadata->flip, GSR_SOURCE_COLOR_RGB, false);
+    }
+
+    return self->got_first_frame ? 0 : -1;
+}
+
+static bool gsr_capture_v4l2_uses_external_image(gsr_capture *cap) {
+    (void)cap;
+    return true;
+}
+
+static bool gsr_capture_v4l2_should_stop(gsr_capture *cap, bool *err) {
+    gsr_capture_v4l2 *self = cap->priv;
+    if(err)
+        *err = self->stop_is_error;
+    return self->should_stop;
+}
+
+static bool gsr_capture_v4l2_is_damaged(gsr_capture *cap) {
+    gsr_capture_v4l2 *self = cap->priv;
+    struct pollfd poll_data = {
+        .fd = self->fd,
+        .events = POLLIN,
+        .revents = 0
+    };
+    return poll(&poll_data, 1, 0) > 0 && (poll_data.revents & POLLIN);
+}
+
+static void gsr_capture_v4l2_clear_damage(gsr_capture *cap) {
+    gsr_capture_v4l2 *self = cap->priv;
+    (void)self;
+}
+
+static void gsr_capture_v4l2_destroy(gsr_capture *cap) {
+    gsr_capture_v4l2 *self = cap->priv;
+    if(cap->priv) {
+        gsr_capture_v4l2_stop(self);
+        free(cap->priv);
+        cap->priv = NULL;
+    }
+    free(cap);
+}
+
+gsr_capture* gsr_capture_v4l2_create(const gsr_capture_v4l2_params *params) {
+    if(!params) {
+        fprintf(stderr, "gsr error: gsr_capture_v4l2_create params is NULL\n");
+        return NULL;
+    }
+
+    gsr_capture *cap = calloc(1, sizeof(gsr_capture));
+    if(!cap)
+        return NULL;
+
+    gsr_capture_v4l2 *cap_camera = calloc(1, sizeof(gsr_capture_v4l2));
+    if(!cap_camera) {
+        free(cap);
+        return NULL;
+    }
+
+    cap_camera->params = *params;
+
+    *cap = (gsr_capture) {
+        .start = gsr_capture_v4l2_start,
+        .tick = gsr_capture_v4l2_tick,
+        .should_stop = gsr_capture_v4l2_should_stop,
+        .capture = gsr_capture_v4l2_capture,
+        .uses_external_image = gsr_capture_v4l2_uses_external_image,
+        .is_damaged = gsr_capture_v4l2_is_damaged,
+        .clear_damage = gsr_capture_v4l2_clear_damage,
+        .destroy = gsr_capture_v4l2_destroy,
+        .priv = cap_camera
+    };
+
+    return cap;
+}
+
+void gsr_capture_v4l2_list_devices(v4l2_devices_query_callback callback, void *userdata) {
+    const bool has_libturbojpeg_lib = is_libturbojpeg_library_available();
+    char v4l2_device_path[128];
+    gsr_capture_v4l2_supported_setup supported_setups[128];
+
+    for(int i = 0; i < 8; ++i) {
+        snprintf(v4l2_device_path, sizeof(v4l2_device_path), "/dev/video%d", i);
+
+        const int fd = open(v4l2_device_path, O_RDWR | O_NONBLOCK);
+        if(fd < 0)
+            continue;
+
+        struct v4l2_capability cap = {0};
+        if(xioctl(fd, VIDIOC_QUERYCAP, &cap) == -1)
+            goto next;
+
+        if(!(cap.capabilities & V4L2_CAP_VIDEO_CAPTURE))
+            goto next;
+
+        if(!(cap.capabilities & V4L2_CAP_STREAMING))
+            goto next;
+
+        struct v4l2_format fmt = {
+            .type = V4L2_BUF_TYPE_VIDEO_CAPTURE
+        };
+        if(xioctl(fd, VIDIOC_G_FMT, &fmt) == -1)
+            goto next;
+
+        const size_t num_supported_setups = gsr_capture_v4l2_get_supported_setups(fd, supported_setups, 128, has_libturbojpeg_lib);
+        if(num_supported_setups == 0)
+            continue;
+
+        for(size_t j = 0; j < num_supported_setups; ++j) {
+            const gsr_capture_v4l2_supported_setup *setup = &supported_setups[j];
+            callback(v4l2_device_path, setup, userdata);
+        }
+
+        next:
+        close(fd);
+    }
+}

src/capture/xcomposite.c

@@ -23,6 +23,7 @@ typedef struct {
     bool init_new_window;
 
     Window window;
+    vec2i window_pos;
     vec2i window_size;
     vec2i texture_size;
     double window_resize_timer;
@@ -31,15 +32,11 @@ typedef struct {
 
     Atom net_active_window_atom;
 
-    gsr_cursor cursor;
-
     bool clear_background;
-    bool fast_path_failed;
 } gsr_capture_xcomposite;
 
 static void gsr_capture_xcomposite_stop(gsr_capture_xcomposite *self) {
     window_texture_deinit(&self->window_texture);
-    gsr_cursor_deinit(&self->cursor);
 }
 
 static int max_int(int a, int b) {
@@ -82,6 +79,9 @@ static int gsr_capture_xcomposite_start(gsr_capture *cap, gsr_capture_metadata *
         return -1;
     }
 
+    self->window_pos.x = attr.x;
+    self->window_pos.y = attr.y;
+
     self->window_size.x = max_int(attr.width, 0);
     self->window_size.y = max_int(attr.height, 0);
 
@@ -96,31 +96,15 @@ static int gsr_capture_xcomposite_start(gsr_capture *cap, gsr_capture_metadata *
         return -1;
     }
 
-    if(gsr_cursor_init(&self->cursor, self->params.egl, self->display) != 0) {
-        gsr_capture_xcomposite_stop(self);
-        return -1;
-    }
-
-    self->texture_size.x = 0;
-    self->texture_size.y = 0;
-
-    self->params.egl->glBindTexture(GL_TEXTURE_2D, window_texture_get_opengl_texture_id(&self->window_texture));
-    self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_WIDTH, &self->texture_size.x);
-    self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_HEIGHT, &self->texture_size.y);
-    self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
+    self->texture_size.x = self->window_texture.window_width;
+    self->texture_size.y = self->window_texture.window_height;
 
     if(self->params.output_resolution.x == 0 && self->params.output_resolution.y == 0) {
-        capture_metadata->width = self->texture_size.x;
-        capture_metadata->height = self->texture_size.y;
+        capture_metadata->video_size = self->texture_size;
     } else {
-        capture_metadata->width = self->params.output_resolution.x;
-        capture_metadata->height = self->params.output_resolution.y;
+        capture_metadata->video_size = self->params.output_resolution;
     }
 
-    self->fast_path_failed = self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && !gl_driver_version_greater_than(&self->params.egl->gpu_info, 24, 0, 9);
-    if(self->fast_path_failed)
-        fprintf(stderr, "gsr warning: gsr_capture_kms_start: your amd driver (mesa) version is known to be buggy (<= version 24.0.9), falling back to opengl copy\n");
-
     self->window_resize_timer = clock_get_monotonic_seconds();
     return 0;
 }
@@ -147,19 +131,17 @@ static void gsr_capture_xcomposite_tick(gsr_capture *cap) {
             if(!XGetWindowAttributes(self->display, self->window, &attr))
                 fprintf(stderr, "gsr error: gsr_capture_xcomposite_tick failed: invalid window id: %lu\n", self->window);
 
+            self->window_pos.x = attr.x;
+            self->window_pos.y = attr.y;
+
             self->window_size.x = max_int(attr.width, 0);
             self->window_size.y = max_int(attr.height, 0);
 
             window_texture_deinit(&self->window_texture);
             window_texture_init(&self->window_texture, self->display, self->window, self->params.egl); // TODO: Do not do the below window_texture_on_resize after this
 
-            self->texture_size.x = 0;
-            self->texture_size.y = 0;
-
-            self->params.egl->glBindTexture(GL_TEXTURE_2D, window_texture_get_opengl_texture_id(&self->window_texture));
-            self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_WIDTH, &self->texture_size.x);
-            self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_HEIGHT, &self->texture_size.y);
-            self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
+            self->texture_size.x = self->window_texture.window_width;
+            self->texture_size.y = self->window_texture.window_height;
 
             self->window_resized = false;
             self->clear_background = true;
@@ -177,13 +159,8 @@ static void gsr_capture_xcomposite_tick(gsr_capture *cap) {
             return;
         }
 
-        self->texture_size.x = 0;
-        self->texture_size.y = 0;
-
-        self->params.egl->glBindTexture(GL_TEXTURE_2D, window_texture_get_opengl_texture_id(&self->window_texture));
-        self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_WIDTH, &self->texture_size.x);
-        self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_HEIGHT, &self->texture_size.y);
-        self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
+        self->texture_size.x = self->window_texture.window_width;
+        self->texture_size.y = self->window_texture.window_height;
 
         self->clear_background = true;
     }
@@ -210,6 +187,10 @@ static void gsr_capture_xcomposite_on_event(gsr_capture *cap, gsr_egl *egl) {
             break;
         }
         case ConfigureNotify: {
+            // TODO: Use PresentConfigureNotify instead
+            self->window_pos.x = xev->xconfigure.x;
+            self->window_pos.y = xev->xconfigure.y;
+
             /* Window resized */
             if(xev->xconfigure.window == self->window && (xev->xconfigure.width != self->window_size.x || xev->xconfigure.height != self->window_size.y)) {
                 self->window_size.x = max_int(xev->xconfigure.width, 0);
@@ -227,8 +208,6 @@ static void gsr_capture_xcomposite_on_event(gsr_capture *cap, gsr_egl *egl) {
             break;
         }
     }
-
-    gsr_cursor_on_event(&self->cursor, xev);
 }
 
 static bool gsr_capture_xcomposite_should_stop(gsr_capture *cap, bool *err) {
@@ -244,66 +223,54 @@ static bool gsr_capture_xcomposite_should_stop(gsr_capture *cap, bool *err) {
     return false;
 }
 
-static int gsr_capture_xcomposite_capture(gsr_capture *cap, gsr_capture_metadata *capture_metdata, gsr_color_conversion *color_conversion) {
+static void gsr_capture_xcomposite_pre_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
+    (void)capture_metadata;
     gsr_capture_xcomposite *self = cap->priv;
 
     if(self->clear_background) {
         self->clear_background = false;
-        gsr_color_conversion_clear(color_conversion);
+        color_conversion->schedule_clear = true;
     }
+}
 
-    const bool is_scaled = self->params.output_resolution.x > 0 && self->params.output_resolution.y > 0;
-    vec2i output_size = is_scaled ? self->params.output_resolution : self->texture_size;
-    output_size = scale_keep_aspect_ratio(self->texture_size, output_size);
+static int gsr_capture_xcomposite_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
+    gsr_capture_xcomposite *self = cap->priv;
 
-    const vec2i target_pos = { max_int(0, capture_metdata->width / 2 - output_size.x / 2), max_int(0, capture_metdata->height / 2 - output_size.y / 2) };
+    const vec2i output_size = scale_keep_aspect_ratio(self->texture_size, capture_metadata->recording_size);
+    const vec2i target_pos = gsr_capture_get_target_position(output_size, capture_metadata);
 
-    self->params.egl->glFlush();
-    self->params.egl->glFinish();
+    //self->params.egl->glFlush();
+    //self->params.egl->glFinish();
 
-    /* Fast opengl free path */
-    if(!self->fast_path_failed && video_codec_context_is_vaapi(capture_metdata->video_codec_context) && self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD) {
-        if(!vaapi_copy_egl_image_to_video_surface(self->params.egl, self->window_texture.image, (vec2i){0, 0}, self->texture_size, target_pos, output_size, capture_metdata->video_codec_context, capture_metdata->frame)) {
-            fprintf(stderr, "gsr error: gsr_capture_xcomposite_capture: vaapi_copy_egl_image_to_video_surface failed, falling back to opengl copy. Please report this as an issue at https://github.com/dec05eba/gpu-screen-recorder-issues\n");
-            self->fast_path_failed = true;
-        }
-    } else {
-        self->fast_path_failed = true;
-    }
+    gsr_color_conversion_draw(color_conversion, window_texture_get_opengl_texture_id(&self->window_texture),
+        target_pos, output_size,
+        (vec2i){0, 0}, self->texture_size, self->texture_size,
+        GSR_ROT_0, capture_metadata->flip, GSR_SOURCE_COLOR_RGB, false);
 
-    if(self->fast_path_failed) {
-        gsr_color_conversion_draw(color_conversion, window_texture_get_opengl_texture_id(&self->window_texture),
-            target_pos, output_size,
-            (vec2i){0, 0}, self->texture_size,
-            0.0f, false, GSR_SOURCE_COLOR_RGB);
-    }
-
-    if(self->params.record_cursor && self->cursor.visible) {
+    if(self->params.record_cursor && self->params.cursor->visible) {
         const vec2d scale = {
             self->texture_size.x == 0 ? 0 : (double)output_size.x / (double)self->texture_size.x,
             self->texture_size.y == 0 ? 0 : (double)output_size.y / (double)self->texture_size.y
         };
 
-        gsr_cursor_tick(&self->cursor, self->window);
-
         const vec2i cursor_pos = {
-            target_pos.x + (self->cursor.position.x - self->cursor.hotspot.x) * scale.x,
-            target_pos.y + (self->cursor.position.y - self->cursor.hotspot.y) * scale.y
+            target_pos.x + (self->params.cursor->position.x - self->params.cursor->hotspot.x - self->window_pos.x) * scale.x,
+            target_pos.y + (self->params.cursor->position.y - self->params.cursor->hotspot.y - self->window_pos.y) * scale.y
         };
 
         self->params.egl->glEnable(GL_SCISSOR_TEST);
         self->params.egl->glScissor(target_pos.x, target_pos.y, output_size.x, output_size.y);
 
-        gsr_color_conversion_draw(color_conversion, self->cursor.texture_id,
-            cursor_pos, (vec2i){self->cursor.size.x * scale.x, self->cursor.size.y * scale.y},
-            (vec2i){0, 0}, self->cursor.size,
-            0.0f, false, GSR_SOURCE_COLOR_RGB);
+        gsr_color_conversion_draw(color_conversion, self->params.cursor->texture_id,
+            cursor_pos, (vec2i){self->params.cursor->size.x * scale.x, self->params.cursor->size.y * scale.y},
+            (vec2i){0, 0}, self->params.cursor->size, self->params.cursor->size,
+            GSR_ROT_0, capture_metadata->flip, GSR_SOURCE_COLOR_RGB, false);
 
         self->params.egl->glDisable(GL_SCISSOR_TEST);
     }
 
-    self->params.egl->glFlush();
-    self->params.egl->glFinish();
+    //self->params.egl->glFlush();
+    //self->params.egl->glFinish();
 
     return 0;
 }
@@ -346,6 +313,7 @@ gsr_capture* gsr_capture_xcomposite_create(const gsr_capture_xcomposite_params *
         .on_event = gsr_capture_xcomposite_on_event,
         .tick = gsr_capture_xcomposite_tick,
         .should_stop = gsr_capture_xcomposite_should_stop,
+        .pre_capture = gsr_capture_xcomposite_pre_capture,
         .capture = gsr_capture_xcomposite_capture,
         .uses_external_image = NULL,
         .get_window_id = gsr_capture_xcomposite_get_window_id,

src/capture/ximage.c

@@ -16,7 +16,6 @@
 typedef struct {
     gsr_capture_ximage_params params;
     Display *display;
-    gsr_cursor cursor;
     gsr_monitor monitor;
     vec2i capture_pos;
     vec2i capture_size;
@@ -25,7 +24,6 @@ typedef struct {
 } gsr_capture_ximage;
 
 static void gsr_capture_ximage_stop(gsr_capture_ximage *self) {
-    gsr_cursor_deinit(&self->cursor);
     if(self->texture_id) {
         self->params.egl->glDeleteTextures(1, &self->texture_id);
         self->texture_id = 0;
@@ -40,11 +38,6 @@ static int gsr_capture_ximage_start(gsr_capture *cap, gsr_capture_metadata *capt
     gsr_capture_ximage *self = cap->priv;
     self->root_window = DefaultRootWindow(self->display);
 
-    if(gsr_cursor_init(&self->cursor, self->params.egl, self->display) != 0) {
-        gsr_capture_ximage_stop(self);
-        return -1;
-    }
-
     if(!get_monitor_by_name(self->params.egl, GSR_CONNECTION_X11, self->params.display_to_capture, &self->monitor)) {
         fprintf(stderr, "gsr error: gsr_capture_ximage_start: failed to find monitor by name \"%s\"\n", self->params.display_to_capture);
         gsr_capture_ximage_stop(self);
@@ -59,14 +52,11 @@ static int gsr_capture_ximage_start(gsr_capture *cap, gsr_capture_metadata *capt
 
     if(self->params.output_resolution.x > 0 && self->params.output_resolution.y > 0) {
         self->params.output_resolution = scale_keep_aspect_ratio(self->capture_size, self->params.output_resolution);
-        capture_metadata->width = self->params.output_resolution.x;
-        capture_metadata->height = self->params.output_resolution.y;
+        capture_metadata->video_size = self->params.output_resolution;
     } else if(self->params.region_size.x > 0 && self->params.region_size.y > 0) {
-        capture_metadata->width = self->params.region_size.x;
-        capture_metadata->height = self->params.region_size.y;
+        capture_metadata->video_size = self->params.region_size;
     } else {
-        capture_metadata->width = self->capture_size.x;
-        capture_metadata->height = self->capture_size.y;
+        capture_metadata->video_size = self->capture_size;
     }
 
     self->texture_id = gl_create_texture(self->params.egl, self->capture_size.x, self->capture_size.y, GL_RGB8, GL_RGB, GL_LINEAR);
@@ -79,12 +69,6 @@ static int gsr_capture_ximage_start(gsr_capture *cap, gsr_capture_metadata *capt
     return 0;
 }
 
-static void gsr_capture_ximage_on_event(gsr_capture *cap, gsr_egl *egl) {
-    gsr_capture_ximage *self = cap->priv;
-    XEvent *xev = gsr_window_get_event_data(egl->window);
-    gsr_cursor_on_event(&self->cursor, xev);
-}
-
 static bool gsr_capture_ximage_upload_to_texture(gsr_capture_ximage *self, int x, int y, int width, int height) {
     const int max_width = XWidthOfScreen(DefaultScreenOfDisplay(self->display));
     const int max_height = XHeightOfScreen(DefaultScreenOfDisplay(self->display));
@@ -137,6 +121,7 @@ static bool gsr_capture_ximage_upload_to_texture(gsr_capture_ximage *self, int x
     }
 
     self->params.egl->glBindTexture(GL_TEXTURE_2D, self->texture_id);
+    // TODO: Change to GL_RGBA for better performance? image_data needs alpha then as well
     self->params.egl->glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, image->width, image->height, GL_RGB, GL_UNSIGNED_BYTE, image_data);
     self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
     success = true;
@@ -147,41 +132,36 @@ static bool gsr_capture_ximage_upload_to_texture(gsr_capture_ximage *self, int x
     return success;
 }
 
-static int gsr_capture_ximage_capture(gsr_capture *cap, gsr_capture_metadata *capture_metdata, gsr_color_conversion *color_conversion) {
+static int gsr_capture_ximage_capture(gsr_capture *cap, gsr_capture_metadata *capture_metadata, gsr_color_conversion *color_conversion) {
     gsr_capture_ximage *self = cap->priv;
 
-    const bool is_scaled = self->params.output_resolution.x > 0 && self->params.output_resolution.y > 0;
-    vec2i output_size = is_scaled ? self->params.output_resolution : self->capture_size;
-    output_size = scale_keep_aspect_ratio(self->capture_size, output_size);
-
-    const vec2i target_pos = { max_int(0, capture_metdata->width / 2 - output_size.x / 2), max_int(0, capture_metdata->height / 2 - output_size.y / 2) };
+    const vec2i output_size = scale_keep_aspect_ratio(self->capture_size, capture_metadata->recording_size);
+    const vec2i target_pos = gsr_capture_get_target_position(output_size, capture_metadata);
     gsr_capture_ximage_upload_to_texture(self, self->capture_pos.x + self->params.region_position.x, self->capture_pos.y + self->params.region_position.y, self->capture_size.x, self->capture_size.y);
 
     gsr_color_conversion_draw(color_conversion, self->texture_id,
         target_pos, output_size,
-        (vec2i){0, 0}, self->capture_size,
-        0.0f, false, GSR_SOURCE_COLOR_RGB);
+        (vec2i){0, 0}, self->capture_size, self->capture_size,
+        GSR_ROT_0, capture_metadata->flip, GSR_SOURCE_COLOR_RGB, false);
 
-    if(self->params.record_cursor && self->cursor.visible) {
+    if(self->params.record_cursor && self->params.cursor->visible) {
         const vec2d scale = {
             self->capture_size.x == 0 ? 0 : (double)output_size.x / (double)self->capture_size.x,
             self->capture_size.y == 0 ? 0 : (double)output_size.y / (double)self->capture_size.y
         };
 
-        gsr_cursor_tick(&self->cursor, self->root_window);
-
         const vec2i cursor_pos = {
-            target_pos.x + (self->cursor.position.x - self->cursor.hotspot.x) * scale.x - self->capture_pos.x - self->params.region_position.x,
-            target_pos.y + (self->cursor.position.y - self->cursor.hotspot.y) * scale.y - self->capture_pos.y - self->params.region_position.y
+            target_pos.x + (self->params.cursor->position.x - self->params.cursor->hotspot.x) * scale.x - self->capture_pos.x - self->params.region_position.x,
+            target_pos.y + (self->params.cursor->position.y - self->params.cursor->hotspot.y) * scale.y - self->capture_pos.y - self->params.region_position.y
         };
 
         self->params.egl->glEnable(GL_SCISSOR_TEST);
         self->params.egl->glScissor(target_pos.x, target_pos.y, output_size.x, output_size.y);
 
-        gsr_color_conversion_draw(color_conversion, self->cursor.texture_id,
-            cursor_pos, (vec2i){self->cursor.size.x * scale.x, self->cursor.size.y * scale.y},
-            (vec2i){0, 0}, self->cursor.size,
-            0.0f, false, GSR_SOURCE_COLOR_RGB);
+        gsr_color_conversion_draw(color_conversion, self->params.cursor->texture_id,
+            cursor_pos, (vec2i){self->params.cursor->size.x * scale.x, self->params.cursor->size.y * scale.y},
+            (vec2i){0, 0}, self->params.cursor->size, self->params.cursor->size,
+            GSR_ROT_0, capture_metadata->flip, GSR_SOURCE_COLOR_RGB, false);
 
         self->params.egl->glDisable(GL_SCISSOR_TEST);
     }
@@ -233,7 +213,6 @@ gsr_capture* gsr_capture_ximage_create(const gsr_capture_ximage_params *params)
     
     *cap = (gsr_capture) {
         .start = gsr_capture_ximage_start,
-        .on_event = gsr_capture_ximage_on_event,
         .tick = NULL,
         .should_stop = NULL,
         .capture = gsr_capture_ximage_capture,

src/codec_query/nvenc.c

@@ -6,6 +6,11 @@
 #include <stdio.h>
 #include <string.h>
 
+#define NVENCAPI_MAJOR_VERSION_470 11
+#define NVENCAPI_MINOR_VERSION_470 1
+#define NVENCAPI_VERSION_470 (NVENCAPI_MAJOR_VERSION_470 | (NVENCAPI_MINOR_VERSION_470 << 24))
+#define NVENCAPI_STRUCT_VERSION_CUSTOM(nvenc_api_version, struct_version)  ((uint32_t)(nvenc_api_version) | ((struct_version)<<16) | (0x7 << 28))
+
 static void* open_nvenc_library(void) {
     dlerror(); /* clear */
     void *lib = dlopen("libnvidia-encode.so.1", RTLD_LAZY);
@@ -75,7 +80,28 @@ static bool profile_is_av1(const GUID *profile_guid) {
     return false;
 }
 
-static bool encoder_get_supported_profiles(const NV_ENCODE_API_FUNCTION_LIST *function_list, void *nvenc_encoder, const GUID *encoder_guid, gsr_supported_video_codecs *supported_video_codecs) {
+/* Returns 0 on error */
+static int nvenc_get_encoding_capability(const NV_ENCODE_API_FUNCTION_LIST *function_list, void *nvenc_encoder, const GUID *encode_guid, uint32_t nvenc_api_version, NV_ENC_CAPS cap) {
+    NV_ENC_CAPS_PARAM param = {
+        .version = NVENCAPI_STRUCT_VERSION_CUSTOM(nvenc_api_version, 1),
+        .capsToQuery = cap
+    };
+
+    int value = 0;
+    if(function_list->nvEncGetEncodeCaps(nvenc_encoder, *encode_guid, &param, &value) != NV_ENC_SUCCESS)
+        return 0;
+
+    return value;
+}
+
+static vec2i encoder_get_max_resolution(const NV_ENCODE_API_FUNCTION_LIST *function_list, void *nvenc_encoder, const GUID *encode_guid, uint32_t nvenc_api_version) {
+    return (vec2i){
+        .x = nvenc_get_encoding_capability(function_list, nvenc_encoder, encode_guid, nvenc_api_version, NV_ENC_CAPS_WIDTH_MAX),
+        .y = nvenc_get_encoding_capability(function_list, nvenc_encoder, encode_guid, nvenc_api_version, NV_ENC_CAPS_HEIGHT_MAX),
+    };
+}
+
+static bool encoder_get_supported_profiles(const NV_ENCODE_API_FUNCTION_LIST *function_list, void *nvenc_encoder, const GUID *encoder_guid, gsr_supported_video_codecs *supported_video_codecs, uint32_t nvenc_api_version) {
     bool success = false;
     GUID *profile_guids = NULL;
 
@@ -99,18 +125,19 @@ static bool encoder_get_supported_profiles(const NV_ENCODE_API_FUNCTION_LIST *fu
         goto fail;
     }
 
+    const vec2i max_resolution = encoder_get_max_resolution(function_list, nvenc_encoder, encoder_guid, nvenc_api_version);
     for(uint32_t i = 0; i < profile_guid_count; ++i) {
         if(profile_is_h264(&profile_guids[i])) {
-            supported_video_codecs->h264 = (gsr_supported_video_codec){ true, false };
+            supported_video_codecs->h264 = (gsr_supported_video_codec){ true, false, max_resolution };
         } else if(profile_is_hevc(&profile_guids[i])) {
-            supported_video_codecs->hevc = (gsr_supported_video_codec){ true, false };
+            supported_video_codecs->hevc = (gsr_supported_video_codec){ true, false, max_resolution };
         } else if(profile_is_hevc_10bit(&profile_guids[i])) {
-            supported_video_codecs->hevc_hdr = (gsr_supported_video_codec){ true, false };
-            supported_video_codecs->hevc_10bit = (gsr_supported_video_codec){ true, false };
+            supported_video_codecs->hevc_hdr = (gsr_supported_video_codec){ true, false, max_resolution };
+            supported_video_codecs->hevc_10bit = (gsr_supported_video_codec){ true, false, max_resolution };
         } else if(profile_is_av1(&profile_guids[i])) {
-            supported_video_codecs->av1 = (gsr_supported_video_codec){ true, false };
-            supported_video_codecs->av1_hdr = (gsr_supported_video_codec){ true, false };
-            supported_video_codecs->av1_10bit = (gsr_supported_video_codec){ true, false };
+            supported_video_codecs->av1 = (gsr_supported_video_codec){ true, false, max_resolution };
+            supported_video_codecs->av1_hdr = (gsr_supported_video_codec){ true, false, max_resolution };
+            supported_video_codecs->av1_10bit = (gsr_supported_video_codec){ true, false, max_resolution };
         }
     }
 
@@ -123,7 +150,7 @@ static bool encoder_get_supported_profiles(const NV_ENCODE_API_FUNCTION_LIST *fu
     return success;
 }
 
-static bool get_supported_video_codecs(const NV_ENCODE_API_FUNCTION_LIST *function_list, void *nvenc_encoder, gsr_supported_video_codecs *supported_video_codecs) {
+static bool get_supported_video_codecs(const NV_ENCODE_API_FUNCTION_LIST *function_list, void *nvenc_encoder, gsr_supported_video_codecs *supported_video_codecs, uint32_t nvenc_api_version) {
     bool success = false;
     GUID *encoder_guids = NULL;
     *supported_video_codecs = (gsr_supported_video_codecs){0};
@@ -149,7 +176,7 @@ static bool get_supported_video_codecs(const NV_ENCODE_API_FUNCTION_LIST *functi
     }
 
     for(uint32_t i = 0; i < encode_guid_count; ++i) {
-        encoder_get_supported_profiles(function_list, nvenc_encoder, &encoder_guids[i], supported_video_codecs);
+        encoder_get_supported_profiles(function_list, nvenc_encoder, &encoder_guids[i], supported_video_codecs, nvenc_api_version);
     }
 
     success = true;
@@ -161,9 +188,6 @@ static bool get_supported_video_codecs(const NV_ENCODE_API_FUNCTION_LIST *functi
     return success;
 }
 
-#define NVENCAPI_VERSION_470 (11 | (1 << 24))
-#define NVENCAPI_STRUCT_VERSION_470(ver) ((uint32_t)NVENCAPI_VERSION_470 | ((ver)<<16) | (0x7 << 28))
-
 bool gsr_get_supported_video_codecs_nvenc(gsr_supported_video_codecs *video_codecs, bool cleanup) {
     memset(video_codecs, 0, sizeof(*video_codecs));
 
@@ -206,13 +230,13 @@ bool gsr_get_supported_video_codecs_nvenc(gsr_supported_video_codecs *video_code
     if(function_list.nvEncOpenEncodeSessionEx(&params, &nvenc_encoder) != NV_ENC_SUCCESS) {
         // Old nvidia gpus dont support the new nvenc api (which is required for av1).
         // In such cases fallback to old api version if possible and try again.
-        function_list.version = NVENCAPI_STRUCT_VERSION_470(2);
+        function_list.version = NVENCAPI_STRUCT_VERSION_CUSTOM(NVENCAPI_VERSION_470, 2);
         if(nvEncodeAPICreateInstance(&function_list) != NV_ENC_SUCCESS) {
             fprintf(stderr, "gsr error: gsr_get_supported_video_codecs_nvenc: nvEncodeAPICreateInstance (retry) failed\n");
             goto done;
         }
 
-        params.version = NVENCAPI_STRUCT_VERSION_470(1);
+        params.version = NVENCAPI_STRUCT_VERSION_CUSTOM(NVENCAPI_VERSION_470, 1);
         params.apiVersion = NVENCAPI_VERSION_470;
         if(function_list.nvEncOpenEncodeSessionEx(&params, &nvenc_encoder) != NV_ENC_SUCCESS) {
             fprintf(stderr, "gsr error: gsr_get_supported_video_codecs_nvenc: nvEncOpenEncodeSessionEx (retry) failed\n");
@@ -220,7 +244,7 @@ bool gsr_get_supported_video_codecs_nvenc(gsr_supported_video_codecs *video_code
         }
     }
 
-    success = get_supported_video_codecs(&function_list, nvenc_encoder, video_codecs);
+    success = get_supported_video_codecs(&function_list, nvenc_encoder, video_codecs, params.apiVersion);
 
     done:
     if(cleanup) {

src/codec_query/vaapi.c

@@ -77,32 +77,51 @@ static bool profile_is_vp9(VAProfile profile) {
     }
 }
 
-static bool profile_supports_video_encoding(VADisplay va_dpy, VAProfile profile, bool *low_power) {
-    *low_power = false;
+/* Returns 0, 0 on error */
+static vec2i profile_entrypoint_get_max_resolution(VADisplay va_dpy, VAProfile profile, VAEntrypoint entrypoint) {
+    VAConfigAttrib attribs[2] = {
+        {
+            .type = VAConfigAttribMaxPictureWidth,
+        },
+        {
+            .type = VAConfigAttribMaxPictureHeight,
+        }
+    };
+    if(vaGetConfigAttributes(va_dpy, profile, entrypoint, attribs, 2) != VA_STATUS_SUCCESS || (attribs[0].value & VA_ATTRIB_NOT_SUPPORTED) || (attribs[1].value & VA_ATTRIB_NOT_SUPPORTED))
+        return (vec2i){0, 0};
+
+    return (vec2i){ attribs[0].value, attribs[1].value };
+}
+
+/* Returns 0 on error or if none is supported */
+static VAEntrypoint profile_get_video_encoding_entrypoint(VADisplay va_dpy, VAProfile profile) {
     int num_entrypoints = vaMaxNumEntrypoints(va_dpy);
     if(num_entrypoints <= 0)
-        return false;
+        return 0;
 
     VAEntrypoint *entrypoint_list = calloc(num_entrypoints, sizeof(VAEntrypoint));
     if(!entrypoint_list)
-        return false;
+        return 0;
 
-    bool supports_encoding = false;
-    bool supports_low_power_encoding = false;
+    int encoding_entrypoint_index = -1;
+    int lower_power_entrypoint_index = -1;
     if(vaQueryConfigEntrypoints(va_dpy, profile, entrypoint_list, &num_entrypoints) == VA_STATUS_SUCCESS) {
         for(int i = 0; i < num_entrypoints; ++i) {
             if(entrypoint_list[i] == VAEntrypointEncSlice)
-                supports_encoding = true;
+                encoding_entrypoint_index = i;
             else if(entrypoint_list[i] == VAEntrypointEncSliceLP)
-                supports_low_power_encoding = true;
+                lower_power_entrypoint_index = i;
         }
     }
 
-    if(!supports_encoding && supports_low_power_encoding)
-        *low_power = true;
+    VAEntrypoint encoding_entrypoint = 0;
+    if(encoding_entrypoint_index != -1)
+        encoding_entrypoint = entrypoint_list[encoding_entrypoint_index];
+    else if(lower_power_entrypoint_index != -1)
+        encoding_entrypoint = entrypoint_list[lower_power_entrypoint_index];
 
     free(entrypoint_list);
-    return supports_encoding || supports_low_power_encoding;
+    return encoding_entrypoint;
 }
 
 static bool get_supported_video_codecs(VADisplay va_dpy, gsr_supported_video_codecs *video_codecs, bool cleanup) {
@@ -116,7 +135,7 @@ static bool get_supported_video_codecs(VADisplay va_dpy, gsr_supported_video_cod
     int va_minor = 0;
     if(vaInitialize(va_dpy, &va_major, &va_minor) != VA_STATUS_SUCCESS) {
         fprintf(stderr, "gsr error: gsr_get_supported_video_codecs_vaapi: vaInitialize failed\n");
-        goto fail;
+        return false;
     }
 
     int num_profiles = vaMaxNumProfiles(va_dpy);
@@ -128,31 +147,45 @@ static bool get_supported_video_codecs(VADisplay va_dpy, gsr_supported_video_cod
         goto fail;
 
     for(int i = 0; i < num_profiles; ++i) {
-        bool low_power = false;
         if(profile_is_h264(profile_list[i])) {
-            if(profile_supports_video_encoding(va_dpy, profile_list[i], &low_power)) {
-                video_codecs->h264 = (gsr_supported_video_codec){ true, low_power };
+            const VAEntrypoint encoding_entrypoint = profile_get_video_encoding_entrypoint(va_dpy, profile_list[i]);
+            if(encoding_entrypoint != 0) {
+                const vec2i max_resolution = profile_entrypoint_get_max_resolution(va_dpy, profile_list[i], encoding_entrypoint);
+                video_codecs->h264 = (gsr_supported_video_codec){ true, encoding_entrypoint == VAEntrypointEncSliceLP, max_resolution };
             }
         } else if(profile_is_hevc_8bit(profile_list[i])) {
-            if(profile_supports_video_encoding(va_dpy, profile_list[i], &low_power))
-                video_codecs->hevc = (gsr_supported_video_codec){ true, low_power };
+            const VAEntrypoint encoding_entrypoint = profile_get_video_encoding_entrypoint(va_dpy, profile_list[i]);
+            if(encoding_entrypoint != 0) {
+                const vec2i max_resolution = profile_entrypoint_get_max_resolution(va_dpy, profile_list[i], encoding_entrypoint);
+                video_codecs->hevc = (gsr_supported_video_codec){ true, encoding_entrypoint == VAEntrypointEncSliceLP, max_resolution };
+            }
         } else if(profile_is_hevc_10bit(profile_list[i])) {
-            if(profile_supports_video_encoding(va_dpy, profile_list[i], &low_power)) {
-                video_codecs->hevc_hdr = (gsr_supported_video_codec){ true, low_power };
-                video_codecs->hevc_10bit = (gsr_supported_video_codec){ true, low_power };
+            const VAEntrypoint encoding_entrypoint = profile_get_video_encoding_entrypoint(va_dpy, profile_list[i]);
+            if(encoding_entrypoint != 0) {
+                const vec2i max_resolution = profile_entrypoint_get_max_resolution(va_dpy, profile_list[i], encoding_entrypoint);
+                video_codecs->hevc_hdr = (gsr_supported_video_codec){ true, encoding_entrypoint == VAEntrypointEncSliceLP, max_resolution };
+                video_codecs->hevc_10bit = (gsr_supported_video_codec){ true, encoding_entrypoint == VAEntrypointEncSliceLP, max_resolution };
             }
         } else if(profile_is_av1(profile_list[i])) {
-            if(profile_supports_video_encoding(va_dpy, profile_list[i], &low_power)) {
-                video_codecs->av1 = (gsr_supported_video_codec){ true, low_power };
-                video_codecs->av1_hdr = (gsr_supported_video_codec){ true, low_power };
-                video_codecs->av1_10bit = (gsr_supported_video_codec){ true, low_power };
+            const VAEntrypoint encoding_entrypoint = profile_get_video_encoding_entrypoint(va_dpy, profile_list[i]);
+            if(encoding_entrypoint != 0) {
+                const vec2i max_resolution = profile_entrypoint_get_max_resolution(va_dpy, profile_list[i], encoding_entrypoint);
+                video_codecs->av1 = (gsr_supported_video_codec){ true, encoding_entrypoint == VAEntrypointEncSliceLP, max_resolution };
+                video_codecs->av1_hdr = (gsr_supported_video_codec){ true, encoding_entrypoint == VAEntrypointEncSliceLP, max_resolution };
+                video_codecs->av1_10bit = (gsr_supported_video_codec){ true, encoding_entrypoint == VAEntrypointEncSliceLP, max_resolution };
             }
         } else if(profile_is_vp8(profile_list[i])) {
-            if(profile_supports_video_encoding(va_dpy, profile_list[i], &low_power))
-                video_codecs->vp8 = (gsr_supported_video_codec){ true, low_power };
+            const VAEntrypoint encoding_entrypoint = profile_get_video_encoding_entrypoint(va_dpy, profile_list[i]);
+            if(encoding_entrypoint != 0) {
+                const vec2i max_resolution = profile_entrypoint_get_max_resolution(va_dpy, profile_list[i], encoding_entrypoint);
+                video_codecs->vp8 = (gsr_supported_video_codec){ true, encoding_entrypoint == VAEntrypointEncSliceLP, max_resolution };
+            }
         } else if(profile_is_vp9(profile_list[i])) {
-            if(profile_supports_video_encoding(va_dpy, profile_list[i], &low_power))
-                video_codecs->vp9 = (gsr_supported_video_codec){ true, low_power };
+            const VAEntrypoint encoding_entrypoint = profile_get_video_encoding_entrypoint(va_dpy, profile_list[i]);
+            if(encoding_entrypoint != 0) {
+                const vec2i max_resolution = profile_entrypoint_get_max_resolution(va_dpy, profile_list[i], encoding_entrypoint);
+                video_codecs->vp9 = (gsr_supported_video_codec){ true, encoding_entrypoint == VAEntrypointEncSliceLP, max_resolution };
+            }
         }
     }
 

src/color_conversion.c

@@ -2,25 +2,20 @@
 #include "../include/egl.h"
 #include <stdio.h>
 #include <string.h>
-#include <math.h>
 #include <assert.h>
 
-/* TODO: highp instead of mediump? */
-
-#define MAX_SHADERS 4
-#define MAX_FRAMEBUFFERS 2
-#define EXTERNAL_TEXTURE_SHADER_OFFSET 2
-
-static float abs_f(float v) {
-    return v >= 0.0f ? v : -v;
-}
-
-#define ROTATE_Z   "mat4 rotate_z(in float angle) {\n"                        \
-                   "    return mat4(cos(angle), -sin(angle), 0.0, 0.0,\n"     \
-                   "                sin(angle),  cos(angle), 0.0, 0.0,\n"     \
-                   "                0.0,           0.0,      1.0, 0.0,\n"     \
-                   "                0.0,           0.0,      0.0, 1.0);\n"    \
-                   "}\n"
+#define GRAPHICS_SHADER_INDEX_Y                    0
+#define GRAPHICS_SHADER_INDEX_UV                   1
+#define GRAPHICS_SHADER_INDEX_Y_EXTERNAL           2
+#define GRAPHICS_SHADER_INDEX_UV_EXTERNAL          3
+#define GRAPHICS_SHADER_INDEX_RGB                  4
+#define GRAPHICS_SHADER_INDEX_RGB_EXTERNAL         5
+#define GRAPHICS_SHADER_INDEX_YUYV_TO_Y            6
+#define GRAPHICS_SHADER_INDEX_YUYV_TO_UV           7
+#define GRAPHICS_SHADER_INDEX_YUYV_TO_Y_EXTERNAL   8
+#define GRAPHICS_SHADER_INDEX_YUYV_TO_UV_EXTERNAL  9
+#define GRAPHICS_SHADER_INDEX_YUYV_TO_RGB          10
+#define GRAPHICS_SHADER_INDEX_YUYV_TO_RGB_EXTERNAL 11
 
 /* https://en.wikipedia.org/wiki/YCbCr, see study/color_space_transform_matrix.png */
 
@@ -78,7 +73,7 @@ static const char* color_format_range_get_transform_matrix(gsr_destination_color
     return NULL;
 }
 
-static int load_shader_y(gsr_shader *shader, gsr_egl *egl, gsr_color_uniforms *uniforms, gsr_destination_color color_format, gsr_color_range color_range, bool external_texture) {
+static int load_graphics_shader_y(gsr_shader *shader, gsr_egl *egl, gsr_color_graphics_uniforms *uniforms, gsr_destination_color color_format, gsr_color_range color_range, bool external_texture) {
     const char *color_transform_matrix = color_format_range_get_transform_matrix(color_format, color_range);
 
     char vertex_shader[2048];
@@ -89,15 +84,14 @@ static int load_shader_y(gsr_shader *shader, gsr_egl *egl, gsr_color_uniforms *u
         "out vec2 texcoords_out;                           \n"
         "uniform vec2 offset;                              \n"
         "uniform float rotation;                           \n"
-        ROTATE_Z
+        "uniform mat2 rotation_matrix;                     \n"
         "void main()                                       \n"
         "{                                                 \n"
-        "  texcoords_out = (vec4(texcoords.x - 0.5, texcoords.y - 0.5, 0.0, 0.0) * rotate_z(rotation)).xy + vec2(0.5, 0.5);  \n"
+        "  texcoords_out = vec2(texcoords.x - 0.5, texcoords.y - 0.5) * rotation_matrix + vec2(0.5, 0.5);  \n"
         "  gl_Position = vec4(offset.x, offset.y, 0.0, 0.0) + vec4(pos.x, pos.y, 0.0, 1.0);    \n"
         "}                                                 \n");
 
     const char *main_code =
-        main_code =
             "  vec4 pixel = texture(tex1, texcoords_out);                                    \n"
             "  FragColor.x = (RGBtoYUV * vec4(pixel.rgb, 1.0)).x;                            \n"
             "  FragColor.w = pixel.a;                                                        \n";
@@ -108,7 +102,7 @@ static int load_shader_y(gsr_shader *shader, gsr_egl *egl, gsr_color_uniforms *u
             "#version 300 es                                                                 \n"
             "#extension GL_OES_EGL_image_external : enable                                   \n"
             "#extension GL_OES_EGL_image_external_essl3 : require                            \n"
-            "precision mediump float;                                                        \n"
+            "precision highp float;                                                        \n"
             "in vec2 texcoords_out;                                                          \n"
             "uniform samplerExternalOES tex1;                                                \n"
             "out vec4 FragColor;                                                             \n"
@@ -120,7 +114,7 @@ static int load_shader_y(gsr_shader *shader, gsr_egl *egl, gsr_color_uniforms *u
     } else {
         snprintf(fragment_shader, sizeof(fragment_shader),
             "#version 300 es                                                                 \n"
-            "precision mediump float;                                                        \n"
+            "precision highp float;                                                        \n"
             "in vec2 texcoords_out;                                                          \n"
             "uniform sampler2D tex1;                                                         \n"
             "out vec4 FragColor;                                                             \n"
@@ -137,11 +131,11 @@ static int load_shader_y(gsr_shader *shader, gsr_egl *egl, gsr_color_uniforms *u
     gsr_shader_bind_attribute_location(shader, "pos", 0);
     gsr_shader_bind_attribute_location(shader, "texcoords", 1);
     uniforms->offset = egl->glGetUniformLocation(shader->program_id, "offset");
-    uniforms->rotation = egl->glGetUniformLocation(shader->program_id, "rotation");
+    uniforms->rotation_matrix = egl->glGetUniformLocation(shader->program_id, "rotation_matrix");
     return 0;
 }
 
-static unsigned int load_shader_uv(gsr_shader *shader, gsr_egl *egl, gsr_color_uniforms *uniforms, gsr_destination_color color_format, gsr_color_range color_range, bool external_texture) {
+static unsigned int load_graphics_shader_uv(gsr_shader *shader, gsr_egl *egl, gsr_color_graphics_uniforms *uniforms, gsr_destination_color color_format, gsr_color_range color_range, bool external_texture) {
     const char *color_transform_matrix = color_format_range_get_transform_matrix(color_format, color_range);
 
     char vertex_shader[2048];
@@ -152,15 +146,14 @@ static unsigned int load_shader_uv(gsr_shader *shader, gsr_egl *egl, gsr_color_u
         "out vec2 texcoords_out;                         \n"
         "uniform vec2 offset;                            \n"
         "uniform float rotation;                         \n"
-        ROTATE_Z
+        "uniform mat2 rotation_matrix;                   \n"
         "void main()                                     \n"
         "{                                               \n"
-        "  texcoords_out = (vec4(texcoords.x - 0.5, texcoords.y - 0.5, 0.0, 0.0) * rotate_z(rotation)).xy + vec2(0.5, 0.5);                      \n"
+        "  texcoords_out = vec2(texcoords.x - 0.5, texcoords.y - 0.5) * rotation_matrix + vec2(0.5, 0.5);                      \n"
         "  gl_Position = (vec4(offset.x, offset.y, 0.0, 0.0) + vec4(pos.x, pos.y, 0.0, 1.0)) * vec4(0.5, 0.5, 1.0, 1.0) - vec4(0.5, 0.5, 0.0, 0.0);   \n"
         "}                                               \n");
 
     const char *main_code =
-        main_code =
             "  vec4 pixel = texture(tex1, texcoords_out);                                          \n"
             "  FragColor.xy = (RGBtoYUV * vec4(pixel.rgb, 1.0)).yz;                                \n"
             "  FragColor.w = pixel.a;                                                              \n";
@@ -171,7 +164,7 @@ static unsigned int load_shader_uv(gsr_shader *shader, gsr_egl *egl, gsr_color_u
             "#version 300 es                                                                       \n"
             "#extension GL_OES_EGL_image_external : enable                                         \n"
             "#extension GL_OES_EGL_image_external_essl3 : require                                  \n"
-            "precision mediump float;                                                              \n"
+            "precision highp float;                                                              \n"
             "in vec2 texcoords_out;                                                                \n"
             "uniform samplerExternalOES tex1;                                                      \n"
             "out vec4 FragColor;                                                                   \n"
@@ -183,7 +176,7 @@ static unsigned int load_shader_uv(gsr_shader *shader, gsr_egl *egl, gsr_color_u
     } else {
         snprintf(fragment_shader, sizeof(fragment_shader),
             "#version 300 es                                                                       \n"
-            "precision mediump float;                                                              \n"
+            "precision highp float;                                                              \n"
             "in vec2 texcoords_out;                                                                \n"
             "uniform sampler2D tex1;                                                               \n"
             "out vec4 FragColor;                                                                   \n"
@@ -200,11 +193,11 @@ static unsigned int load_shader_uv(gsr_shader *shader, gsr_egl *egl, gsr_color_u
     gsr_shader_bind_attribute_location(shader, "pos", 0);
     gsr_shader_bind_attribute_location(shader, "texcoords", 1);
     uniforms->offset = egl->glGetUniformLocation(shader->program_id, "offset");
-    uniforms->rotation = egl->glGetUniformLocation(shader->program_id, "rotation");
+    uniforms->rotation_matrix = egl->glGetUniformLocation(shader->program_id, "rotation_matrix");
     return 0;
 }
 
-static unsigned int load_shader_rgb(gsr_shader *shader, gsr_egl *egl, gsr_color_uniforms *uniforms, bool external_texture) {
+static unsigned int load_graphics_shader_rgb(gsr_shader *shader, gsr_egl *egl, gsr_color_graphics_uniforms *uniforms, bool external_texture) {
     char vertex_shader[2048];
     snprintf(vertex_shader, sizeof(vertex_shader),
         "#version 300 es                                   \n"
@@ -213,15 +206,14 @@ static unsigned int load_shader_rgb(gsr_shader *shader, gsr_egl *egl, gsr_color_
         "out vec2 texcoords_out;                           \n"
         "uniform vec2 offset;                              \n"
         "uniform float rotation;                           \n"
-        ROTATE_Z
+        "uniform mat2 rotation_matrix;                     \n"
         "void main()                                       \n"
         "{                                                 \n"
-        "  texcoords_out = (vec4(texcoords.x - 0.5, texcoords.y - 0.5, 0.0, 0.0) * rotate_z(rotation)).xy + vec2(0.5, 0.5);  \n"
+        "  texcoords_out = vec2(texcoords.x - 0.5, texcoords.y - 0.5) * rotation_matrix + vec2(0.5, 0.5);  \n"
         "  gl_Position = vec4(offset.x, offset.y, 0.0, 0.0) + vec4(pos.x, pos.y, 0.0, 1.0);    \n"
         "}                                                 \n");
 
     const char *main_code =
-        main_code =
             "  vec4 pixel = texture(tex1, texcoords_out);                                          \n"
             "  FragColor = pixel;                                                                  \n";
 
@@ -231,7 +223,7 @@ static unsigned int load_shader_rgb(gsr_shader *shader, gsr_egl *egl, gsr_color_
             "#version 300 es                                                                       \n"
             "#extension GL_OES_EGL_image_external : enable                                         \n"
             "#extension GL_OES_EGL_image_external_essl3 : require                                  \n"
-            "precision mediump float;                                                              \n"
+            "precision highp float;                                                              \n"
             "in vec2 texcoords_out;                                                                \n"
             "uniform samplerExternalOES tex1;                                                      \n"
             "out vec4 FragColor;                                                                   \n"
@@ -242,7 +234,7 @@ static unsigned int load_shader_rgb(gsr_shader *shader, gsr_egl *egl, gsr_color_
     } else {
         snprintf(fragment_shader, sizeof(fragment_shader),
             "#version 300 es                                                                       \n"
-            "precision mediump float;                                                              \n"
+            "precision highp float;                                                              \n"
             "in vec2 texcoords_out;                                                                \n"
             "uniform sampler2D tex1;                                                               \n"
             "out vec4 FragColor;                                                                   \n"
@@ -258,14 +250,220 @@ static unsigned int load_shader_rgb(gsr_shader *shader, gsr_egl *egl, gsr_color_
     gsr_shader_bind_attribute_location(shader, "pos", 0);
     gsr_shader_bind_attribute_location(shader, "texcoords", 1);
     uniforms->offset = egl->glGetUniformLocation(shader->program_id, "offset");
-    uniforms->rotation = egl->glGetUniformLocation(shader->program_id, "rotation");
+    uniforms->rotation_matrix = egl->glGetUniformLocation(shader->program_id, "rotation_matrix");
+    return 0;
+}
+
+static int load_graphics_shader_yuyv_to_y(gsr_shader *shader, gsr_egl *egl, gsr_color_graphics_uniforms *uniforms, bool external_texture) {
+    char vertex_shader[2048];
+    snprintf(vertex_shader, sizeof(vertex_shader),
+        "#version 300 es                                   \n"
+        "in vec2 pos;                                      \n"
+        "in vec2 texcoords;                                \n"
+        "out vec2 texcoords_out;                           \n"
+        "uniform vec2 offset;                              \n"
+        "uniform float rotation;                           \n"
+        "uniform mat2 rotation_matrix;                     \n"
+        "void main()                                       \n"
+        "{                                                 \n"
+        "  texcoords_out = vec2(texcoords.x - 0.5, texcoords.y - 0.5) * rotation_matrix + vec2(0.5, 0.5);  \n"
+        "  gl_Position = vec4(offset.x, offset.y, 0.0, 0.0) + vec4(pos.x, pos.y, 0.0, 1.0);    \n"
+        "}                                                 \n");
+
+    const char *main_code =
+            "  vec4 pixel = texture(tex1, texcoords_out);                                    \n"
+            "  FragColor.x = pixel.r;                                                        \n"
+            "  FragColor.w = 1.0;                                                            \n";
+
+    char fragment_shader[2048];
+    if(external_texture) {
+        snprintf(fragment_shader, sizeof(fragment_shader),
+            "#version 300 es                                                                 \n"
+            "#extension GL_OES_EGL_image_external : enable                                   \n"
+            "#extension GL_OES_EGL_image_external_essl3 : require                            \n"
+            "precision highp float;                                                          \n"
+            "in vec2 texcoords_out;                                                          \n"
+            "uniform samplerExternalOES tex1;                                                \n"
+            "out vec4 FragColor;                                                             \n"
+            "void main()                                                                     \n"
+            "{                                                                               \n"
+            "%s"
+            "}                                                                               \n", main_code);
+    } else {
+        snprintf(fragment_shader, sizeof(fragment_shader),
+            "#version 300 es                                                                 \n"
+            "precision highp float;                                                          \n"
+            "in vec2 texcoords_out;                                                          \n"
+            "uniform sampler2D tex1;                                                         \n"
+            "out vec4 FragColor;                                                             \n"
+            "void main()                                                                     \n"
+            "{                                                                               \n"
+            "%s"
+            "}                                                                               \n", main_code);
+    }
+
+    if(gsr_shader_init(shader, egl, vertex_shader, fragment_shader) != 0)
+        return -1;
+
+    gsr_shader_bind_attribute_location(shader, "pos", 0);
+    gsr_shader_bind_attribute_location(shader, "texcoords", 1);
+    uniforms->offset = egl->glGetUniformLocation(shader->program_id, "offset");
+    uniforms->rotation_matrix = egl->glGetUniformLocation(shader->program_id, "rotation_matrix");
+    return 0;
+}
+
+static unsigned int load_graphics_shader_yuyv_to_uv(gsr_shader *shader, gsr_egl *egl, gsr_color_graphics_uniforms *uniforms, bool external_texture) {
+    char vertex_shader[2048];
+    snprintf(vertex_shader, sizeof(vertex_shader),
+        "#version 300 es                                 \n"
+        "in vec2 pos;                                    \n"
+        "in vec2 texcoords;                              \n"
+        "out vec2 texcoords_out;                         \n"
+        "uniform vec2 offset;                            \n"
+        "uniform float rotation;                         \n"
+        "uniform mat2 rotation_matrix;                   \n"
+        "void main()                                     \n"
+        "{                                               \n"
+        "  texcoords_out = vec2(texcoords.x - 0.5, texcoords.y - 0.5) * rotation_matrix + vec2(0.5, 0.5);                      \n"
+        "  gl_Position = (vec4(offset.x, offset.y, 0.0, 0.0) + vec4(pos.x, pos.y, 0.0, 1.0)) * vec4(0.5, 0.5, 1.0, 1.0) - vec4(0.5, 0.5, 0.0, 0.0);   \n"
+        "}                                               \n");
+
+    const char *main_code =
+            "    vec2 resolution = vec2(textureSize(tex1, 0));\n"
+            "    ivec2 uv = ivec2(texcoords_out * resolution);\n"
+            "    float u = 0.0;\n"
+            "    float v = 0.0;\n"
+            "    vec4 this_color = texelFetch(tex1, uv, 0);\n"
+            "    if((uv.x & 1) == 0) {\n"
+            "        vec2 next_color = texelFetch(tex1, uv + ivec2(1, 0), 0).rg;\n"
+            "        u = this_color.g;\n"
+            "        v = next_color.g;\n"
+            "    } else {\n"
+            "        vec2 prev_color = texelFetch(tex1, uv - ivec2(1, 0), 0).rg;\n"
+            "        u = prev_color.g;\n"
+            "        v = this_color.g;\n"
+            "    }\n"
+            "    FragColor.rg = vec2(u, v);\n"
+            "    FragColor.w = 1.0;\n";
+
+    char fragment_shader[2048];
+    if(external_texture) {
+        snprintf(fragment_shader, sizeof(fragment_shader),
+            "#version 300 es                                                                       \n"
+            "#extension GL_OES_EGL_image_external : enable                                         \n"
+            "#extension GL_OES_EGL_image_external_essl3 : require                                  \n"
+            "precision highp float;                                                                \n"
+            "in vec2 texcoords_out;                                                                \n"
+            "uniform samplerExternalOES tex1;                                                      \n"
+            "out vec4 FragColor;                                                                   \n"
+            "void main()                                                                           \n"
+            "{                                                                                     \n"
+            "%s"
+            "}                                                                                     \n", main_code);
+    } else {
+        snprintf(fragment_shader, sizeof(fragment_shader),
+            "#version 300 es                                                                       \n"
+            "precision highp float;                                                                \n"
+            "in vec2 texcoords_out;                                                                \n"
+            "uniform sampler2D tex1;                                                               \n"
+            "out vec4 FragColor;                                                                   \n"
+            "void main()                                                                           \n"
+            "{                                                                                     \n"
+            "%s"
+            "}                                                                                     \n", main_code);
+    }
+
+    if(gsr_shader_init(shader, egl, vertex_shader, fragment_shader) != 0)
+        return -1;
+
+    gsr_shader_bind_attribute_location(shader, "pos", 0);
+    gsr_shader_bind_attribute_location(shader, "texcoords", 1);
+    uniforms->offset = egl->glGetUniformLocation(shader->program_id, "offset");
+    uniforms->rotation_matrix = egl->glGetUniformLocation(shader->program_id, "rotation_matrix");
+    return 0;
+}
+
+static unsigned int load_graphics_shader_yuyv_to_rgb(gsr_shader *shader, gsr_egl *egl, gsr_color_graphics_uniforms *uniforms, bool external_texture) {
+    char vertex_shader[2048];
+    snprintf(vertex_shader, sizeof(vertex_shader),
+        "#version 300 es                                   \n"
+        "in vec2 pos;                                      \n"
+        "in vec2 texcoords;                                \n"
+        "out vec2 texcoords_out;                           \n"
+        "uniform vec2 offset;                              \n"
+        "uniform float rotation;                           \n"
+        "uniform mat2 rotation_matrix;                     \n"
+        "void main()                                       \n"
+        "{                                                 \n"
+        "  texcoords_out = vec2(texcoords.x - 0.5, texcoords.y - 0.5) * rotation_matrix + vec2(0.5, 0.5);  \n"
+        "  gl_Position = vec4(offset.x, offset.y, 0.0, 0.0) + vec4(pos.x, pos.y, 0.0, 1.0);    \n"
+        "}                                                 \n");
+
+    const char *main_code =
+        "    vec2 resolution = vec2(textureSize(tex1, 0));\n"
+        "    ivec2 uv = ivec2(texcoords_out * resolution);\n"
+        "    float y = 0.0;\n"
+        "    float u = 0.0;\n"
+        "    float v = 0.0;\n"
+        "    vec4 this_color = texelFetch(tex1, uv, 0);\n"
+        "    if((uv.x & 1) == 0) {\n"
+        "        vec2 next_color = texelFetch(tex1, uv + ivec2(1, 0), 0).rg;\n"
+        "        y = this_color.r;\n"
+        "        u = this_color.g;\n"
+        "        v = next_color.g;\n"
+        "    } else {\n"
+        "        vec2 prev_color = texelFetch(tex1, uv - ivec2(1, 0), 0).rg;\n"
+        "        y = this_color.r;\n"
+        "        u = prev_color.g;\n"
+        "        v = this_color.g;\n"
+        "    }\n"
+        "    FragColor = vec4(\n"
+        "        y + 1.4065 * (v - 0.5),\n"
+        "        y - 0.3455 * (u - 0.5) - 0.7169 * (v - 0.5),\n"
+        "        y + 1.1790 * (u - 0.5),\n"
+        "        1.0);\n";
+
+    char fragment_shader[4096];
+    if(external_texture) {
+        snprintf(fragment_shader, sizeof(fragment_shader),
+            "#version 300 es                                                                       \n"
+            "#extension GL_OES_EGL_image_external : enable                                         \n"
+            "#extension GL_OES_EGL_image_external_essl3 : require                                  \n"
+            "precision highp float;                                                              \n"
+            "in vec2 texcoords_out;                                                                \n"
+            "uniform samplerExternalOES tex1;                                                      \n"
+            "out vec4 FragColor;                                                                   \n"
+            "void main()                                                                           \n"
+            "{                                                                                     \n"
+            "%s"
+            "}                                                                                     \n", main_code);
+    } else {
+        snprintf(fragment_shader, sizeof(fragment_shader),
+            "#version 300 es                                                                       \n"
+            "precision highp float;                                                              \n"
+            "in vec2 texcoords_out;                                                                \n"
+            "uniform sampler2D tex1;                                                               \n"
+            "out vec4 FragColor;                                                                   \n"
+            "void main()                                                                           \n"
+            "{                                                                                     \n"
+            "%s"
+            "}                                                                                     \n", main_code);
+    }
+
+    if(gsr_shader_init(shader, egl, vertex_shader, fragment_shader) != 0)
+        return -1;
+
+    gsr_shader_bind_attribute_location(shader, "pos", 0);
+    gsr_shader_bind_attribute_location(shader, "texcoords", 1);
+    uniforms->offset = egl->glGetUniformLocation(shader->program_id, "offset");
+    uniforms->rotation_matrix = egl->glGetUniformLocation(shader->program_id, "rotation_matrix");
     return 0;
 }
 
 static int load_framebuffers(gsr_color_conversion *self) {
     /* TODO: Only generate the necessary amount of framebuffers (self->params.num_destination_textures) */
     const unsigned int draw_buffer = GL_COLOR_ATTACHMENT0;
-    self->params.egl->glGenFramebuffers(MAX_FRAMEBUFFERS, self->framebuffers);
+    self->params.egl->glGenFramebuffers(GSR_COLOR_CONVERSION_MAX_FRAMEBUFFERS, self->framebuffers);
 
     self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[0]);
     self->params.egl->glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, self->params.destination_textures[0], 0);
@@ -311,6 +509,88 @@ static int create_vertices(gsr_color_conversion *self) {
     return 0;
 }
 
+static bool gsr_color_conversion_load_graphics_shaders(gsr_color_conversion *self) {
+    switch(self->params.destination_color) {
+        case GSR_DESTINATION_COLOR_NV12:
+        case GSR_DESTINATION_COLOR_P010: {
+            if(load_graphics_shader_y(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_Y], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_Y], self->params.destination_color, self->params.color_range, false) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load Y graphics shader\n");
+                return false;
+            }
+
+            if(load_graphics_shader_uv(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_UV], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_UV], self->params.destination_color, self->params.color_range, false) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load UV graphics shader\n");
+                return false;
+            }
+
+            if(load_graphics_shader_yuyv_to_y(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_YUYV_TO_Y], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_YUYV_TO_Y], false) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load YUYV to Y graphics shader\n");
+                return false;
+            }
+
+            if(load_graphics_shader_yuyv_to_uv(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_YUYV_TO_UV], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_YUYV_TO_UV], false) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load YUYV to UV graphics shader\n");
+                return false;
+            }
+            break;
+        }
+        case GSR_DESTINATION_COLOR_RGB8: {
+            if(load_graphics_shader_rgb(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_RGB], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_RGB], false) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load RGB graphics shader\n");
+                return false;
+            }
+
+            if(load_graphics_shader_yuyv_to_rgb(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_YUYV_TO_RGB], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_YUYV_TO_RGB], false) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load YUYV to RGB graphics shader\n");
+                return false;
+            }
+            break;
+        }
+    }
+    return true;
+}
+
+static bool gsr_color_conversion_load_external_graphics_shaders(gsr_color_conversion *self) {
+    switch(self->params.destination_color) {
+        case GSR_DESTINATION_COLOR_NV12:
+        case GSR_DESTINATION_COLOR_P010: {
+            if(load_graphics_shader_y(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_Y_EXTERNAL], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_Y_EXTERNAL], self->params.destination_color, self->params.color_range, true) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load Y graphics shader (external)\n");
+                return false;
+            }
+
+            if(load_graphics_shader_uv(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_UV_EXTERNAL], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_UV_EXTERNAL], self->params.destination_color, self->params.color_range, true) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load UV graphics shader (external)\n");
+                return false;
+            }
+
+            if(load_graphics_shader_yuyv_to_y(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_YUYV_TO_Y_EXTERNAL], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_YUYV_TO_Y_EXTERNAL], true) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load YUYV to Y graphics shader (external)\n");
+                return false;
+            }
+
+            if(load_graphics_shader_yuyv_to_uv(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_YUYV_TO_UV_EXTERNAL], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_YUYV_TO_UV_EXTERNAL], true) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load YUYV to UV graphics shader (external)\n");
+                return false;
+            }
+            break;
+        }
+        case GSR_DESTINATION_COLOR_RGB8: {
+            if(load_graphics_shader_rgb(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_RGB_EXTERNAL], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_RGB_EXTERNAL], true) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load RGB graphics shader (external)\n");
+                return false;
+            }
+
+            if(load_graphics_shader_yuyv_to_rgb(&self->graphics_shaders[GRAPHICS_SHADER_INDEX_YUYV_TO_RGB_EXTERNAL], self->params.egl, &self->graphics_uniforms[GRAPHICS_SHADER_INDEX_YUYV_TO_RGB_EXTERNAL], true) != 0) {
+                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load YUYV to RGB graphics shader (external)\n");
+                return false;
+            }
+            break;
+        }
+    }
+    return true;
+}
+
 int gsr_color_conversion_init(gsr_color_conversion *self, const gsr_color_conversion_params *params) {
     assert(params);
     assert(params->egl);
@@ -318,58 +598,32 @@ int gsr_color_conversion_init(gsr_color_conversion *self, const gsr_color_conver
     self->params.egl = params->egl;
     self->params = *params;
 
-    switch(params->destination_color) {
+    switch(self->params.destination_color) {
         case GSR_DESTINATION_COLOR_NV12:
         case GSR_DESTINATION_COLOR_P010: {
             if(self->params.num_destination_textures != 2) {
                 fprintf(stderr, "gsr error: gsr_color_conversion_init: expected 2 destination textures for destination color NV12/P010, got %d destination texture(s)\n", self->params.num_destination_textures);
-                return -1;
-            }
-
-            if(load_shader_y(&self->shaders[0], self->params.egl, &self->uniforms[0], params->destination_color, params->color_range, false) != 0) {
-                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load Y shader\n");
                 goto err;
             }
-
-            if(load_shader_uv(&self->shaders[1], self->params.egl, &self->uniforms[1], params->destination_color, params->color_range, false) != 0) {
-                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load UV shader\n");
-                goto err;
-            }
-
-            if(self->params.load_external_image_shader) {
-                if(load_shader_y(&self->shaders[EXTERNAL_TEXTURE_SHADER_OFFSET], self->params.egl, &self->uniforms[EXTERNAL_TEXTURE_SHADER_OFFSET], params->destination_color, params->color_range, true) != 0) {
-                    fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load Y shader\n");
-                    goto err;
-                }
-
-                if(load_shader_uv(&self->shaders[EXTERNAL_TEXTURE_SHADER_OFFSET + 1], self->params.egl, &self->uniforms[EXTERNAL_TEXTURE_SHADER_OFFSET + 1], params->destination_color, params->color_range, true) != 0) {
-                    fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load UV shader\n");
-                    goto err;
-                }
-            }
             break;
         }
         case GSR_DESTINATION_COLOR_RGB8: {
             if(self->params.num_destination_textures != 1) {
                 fprintf(stderr, "gsr error: gsr_color_conversion_init: expected 1 destination textures for destination color RGB8, got %d destination texture(s)\n", self->params.num_destination_textures);
-                return -1;
-            }
-
-            if(load_shader_rgb(&self->shaders[0], self->params.egl, &self->uniforms[0], false) != 0) {
-                fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load Y shader\n");
                 goto err;
             }
-
-            if(self->params.load_external_image_shader) {
-                if(load_shader_rgb(&self->shaders[EXTERNAL_TEXTURE_SHADER_OFFSET], self->params.egl, &self->uniforms[EXTERNAL_TEXTURE_SHADER_OFFSET], true) != 0) {
-                    fprintf(stderr, "gsr error: gsr_color_conversion_init: failed to load Y shader\n");
-                    goto err;
-                }
-            }
             break;
         }
     }
 
+    if(!gsr_color_conversion_load_graphics_shaders(self))
+        goto err;
+
+    if(self->params.load_external_image_shader) {
+        if(!gsr_color_conversion_load_external_graphics_shaders(self))
+            goto err;
+    }
+
     if(load_framebuffers(self) != 0)
         goto err;
 
@@ -397,86 +651,106 @@ void gsr_color_conversion_deinit(gsr_color_conversion *self) {
         self->vertex_array_object_id = 0;
     }
 
-    self->params.egl->glDeleteFramebuffers(MAX_FRAMEBUFFERS, self->framebuffers);
-    for(int i = 0; i < MAX_FRAMEBUFFERS; ++i) {
+    self->params.egl->glDeleteFramebuffers(GSR_COLOR_CONVERSION_MAX_FRAMEBUFFERS, self->framebuffers);
+    for(int i = 0; i < GSR_COLOR_CONVERSION_MAX_FRAMEBUFFERS; ++i) {
         self->framebuffers[i] = 0;
     }
 
-    for(int i = 0; i < MAX_SHADERS; ++i) {
-        gsr_shader_deinit(&self->shaders[i]);
+    for(int i = 0; i < GSR_COLOR_CONVERSION_MAX_GRAPHICS_SHADERS; ++i) {
+        gsr_shader_deinit(&self->graphics_shaders[i]);
     }
 
     self->params.egl = NULL;
 }
 
-static void gsr_color_conversion_swizzle_texture_source(gsr_color_conversion *self, gsr_source_color source_color) {
+static void gsr_color_conversion_apply_rotation(gsr_rotation rotation, float rotation_matrix[2][2]) {
+    /*
+    rotation_matrix[0][0] =  cos(angle);
+    rotation_matrix[0][1] = -sin(angle);
+    rotation_matrix[1][0] =  sin(angle);
+    rotation_matrix[1][1] =  cos(angle);
+    The manual matrix code below is the same as this code above, but without floating-point errors.
+    This is done to remove any blurring caused by these floating-point errors.
+    */
+    switch(rotation) {
+        case GSR_ROT_0:
+            rotation_matrix[0][0] = 1.0f;
+            rotation_matrix[0][1] = 0.0f;
+            rotation_matrix[1][0] = 0.0f;
+            rotation_matrix[1][1] = 1.0f;
+            break;
+        case GSR_ROT_90:
+            rotation_matrix[0][0] =  0.0f;
+            rotation_matrix[0][1] = -1.0f;
+            rotation_matrix[1][0] =  1.0f;
+            rotation_matrix[1][1] =  0.0f;
+            break;
+        case GSR_ROT_180:
+            rotation_matrix[0][0] = -1.0f;
+            rotation_matrix[0][1] =  0.0f;
+            rotation_matrix[1][0] =  0.0f;
+            rotation_matrix[1][1] = -1.0f;
+            break;
+        case GSR_ROT_270:
+            rotation_matrix[0][0] =  0.0f;
+            rotation_matrix[0][1] =  1.0f;
+            rotation_matrix[1][0] = -1.0f;
+            rotation_matrix[1][1] =  0.0f;
+            break;
+    }
+}
+
+static void gsr_color_conversion_swizzle_texture_source(gsr_color_conversion *self, unsigned int texture_target, gsr_source_color source_color) {
     if(source_color == GSR_SOURCE_COLOR_BGR) {
         const int swizzle_mask[] = { GL_BLUE, GL_GREEN, GL_RED, 1 };
-        self->params.egl->glTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_RGBA, swizzle_mask);
+        self->params.egl->glTexParameteriv(texture_target, GL_TEXTURE_SWIZZLE_RGBA, swizzle_mask);
     }
 }
 
-static void gsr_color_conversion_swizzle_reset(gsr_color_conversion *self, gsr_source_color source_color) {
+static void gsr_color_conversion_swizzle_reset(gsr_color_conversion *self, unsigned int texture_target, gsr_source_color source_color) {
     if(source_color == GSR_SOURCE_COLOR_BGR) {
         const int swizzle_mask[] = { GL_RED, GL_GREEN, GL_BLUE, GL_ALPHA };
-        self->params.egl->glTexParameteriv(GL_TEXTURE_2D, GL_TEXTURE_SWIZZLE_RGBA, swizzle_mask);
+        self->params.egl->glTexParameteriv(texture_target, GL_TEXTURE_SWIZZLE_RGBA, swizzle_mask);
     }
 }
 
-/* |source_pos| is in pixel coordinates and |source_size|  */
-void gsr_color_conversion_draw(gsr_color_conversion *self, unsigned int texture_id, vec2i source_pos, vec2i source_size, vec2i texture_pos, vec2i texture_size, float rotation, bool external_texture, gsr_source_color source_color) {
-    // TODO: Remove this crap
-    rotation = M_PI*2.0f - rotation;
+static void gsr_color_conversion_draw_graphics(gsr_color_conversion *self, unsigned int texture_id, bool external_texture, gsr_rotation rotation, gsr_flip flip, float rotation_matrix[2][2], vec2i source_position, vec2i source_size, vec2i destination_pos, vec2i texture_size, vec2f scale, gsr_source_color source_color) {
+    if(source_size.x == 0 || source_size.y == 0)
+        return;
 
-    /* TODO: Do not call this every frame? */
-    vec2i dest_texture_size = {0, 0};
-    self->params.egl->glBindTexture(GL_TEXTURE_2D, self->params.destination_textures[0]);
-    self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_WIDTH, &dest_texture_size.x);
-    self->params.egl->glGetTexLevelParameteriv(GL_TEXTURE_2D, 0, GL_TEXTURE_HEIGHT, &dest_texture_size.y);
-    self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
+    const vec2i dest_texture_size = self->params.destination_textures_size[0];
+    const unsigned int texture_target = external_texture ? GL_TEXTURE_EXTERNAL_OES : GL_TEXTURE_2D;
 
-    const int texture_target = external_texture ? GL_TEXTURE_EXTERNAL_OES : GL_TEXTURE_2D;
+    if(rotation == GSR_ROT_90 || rotation == GSR_ROT_270) {
+        const float tmp = texture_size.x;
+        texture_size.x = texture_size.y;
+        texture_size.y = tmp;
+    }
 
     self->params.egl->glBindTexture(texture_target, texture_id);
-
-    vec2i source_texture_size = {0, 0};
-    if(external_texture) {
-        assert(self->params.load_external_image_shader);
-        source_texture_size = source_size;
-    } else {
-        /* TODO: Do not call this every frame? */
-        self->params.egl->glGetTexLevelParameteriv(texture_target, 0, GL_TEXTURE_WIDTH, &source_texture_size.x);
-        self->params.egl->glGetTexLevelParameteriv(texture_target, 0, GL_TEXTURE_HEIGHT, &source_texture_size.y);
-    }
-
-    // TODO: Remove this crap
-    if(abs_f(M_PI * 0.5f - rotation) <= 0.001f || abs_f(M_PI * 1.5f - rotation) <= 0.001f) {
-        float tmp = source_texture_size.x;
-        source_texture_size.x = source_texture_size.y;
-        source_texture_size.y = tmp;
-    }
+    gsr_color_conversion_swizzle_texture_source(self, texture_target, source_color);
 
     const vec2f pos_norm = {
-        ((float)source_pos.x / (dest_texture_size.x == 0 ? 1.0f : (float)dest_texture_size.x)) * 2.0f,
-        ((float)source_pos.y / (dest_texture_size.y == 0 ? 1.0f : (float)dest_texture_size.y)) * 2.0f,
+        ((float)destination_pos.x / (dest_texture_size.x == 0 ? 1.0f : (float)dest_texture_size.x)) * 2.0f,
+        ((float)destination_pos.y / (dest_texture_size.y == 0 ? 1.0f : (float)dest_texture_size.y)) * 2.0f,
     };
 
     const vec2f size_norm = {
-        ((float)source_size.x / (dest_texture_size.x == 0 ? 1.0f : (float)dest_texture_size.x)) * 2.0f,
-        ((float)source_size.y / (dest_texture_size.y == 0 ? 1.0f : (float)dest_texture_size.y)) * 2.0f,
+        ((float)source_size.x / (dest_texture_size.x == 0 ? 1.0f : (float)dest_texture_size.x)) * 2.0f * scale.x,
+        ((float)source_size.y / (dest_texture_size.y == 0 ? 1.0f : (float)dest_texture_size.y)) * 2.0f * scale.y,
     };
 
     const vec2f texture_pos_norm = {
-        (float)texture_pos.x / (source_texture_size.x == 0 ? 1.0f : (float)source_texture_size.x),
-        (float)texture_pos.y / (source_texture_size.y == 0 ? 1.0f : (float)source_texture_size.y),
+        (float)source_position.x / (texture_size.x == 0 ? 1.0f : (float)texture_size.x),
+        (float)source_position.y / (texture_size.y == 0 ? 1.0f : (float)texture_size.y),
     };
 
     const vec2f texture_size_norm = {
-        (float)texture_size.x / (source_texture_size.x == 0 ? 1.0f : (float)source_texture_size.x),
-        (float)texture_size.y / (source_texture_size.y == 0 ? 1.0f : (float)source_texture_size.y),
+        (float)source_size.x / (texture_size.x == 0 ? 1.0f : (float)texture_size.x),
+        (float)source_size.y / (texture_size.y == 0 ? 1.0f : (float)texture_size.y),
     };
 
-    const float vertices[] = {
+    float vertices[] = {
         -1.0f + 0.0f,               -1.0f + 0.0f + size_norm.y, texture_pos_norm.x,                       texture_pos_norm.y + texture_size_norm.y,
         -1.0f + 0.0f,               -1.0f + 0.0f,               texture_pos_norm.x,                       texture_pos_norm.y,
         -1.0f + 0.0f + size_norm.x, -1.0f + 0.0f,               texture_pos_norm.x + texture_size_norm.x, texture_pos_norm.y,
@@ -486,43 +760,144 @@ void gsr_color_conversion_draw(gsr_color_conversion *self, unsigned int texture_
         -1.0f + 0.0f + size_norm.x, -1.0f + 0.0f + size_norm.y, texture_pos_norm.x + texture_size_norm.x, texture_pos_norm.y + texture_size_norm.y
     };
 
-    gsr_color_conversion_swizzle_texture_source(self, source_color);
+    if(flip & GSR_FLIP_HORIZONTAL) {
+        for(int i = 0; i < 6; ++i) {
+            const float prev_x = vertices[i*4 + 2];
+            vertices[i*4 + 2] = texture_pos_norm.x + texture_size_norm.x - prev_x;
+        }
+    }
+
+    if(flip & GSR_FLIP_VERTICAL) {
+        for(int i = 0; i < 6; ++i) {
+            const float prev_y = vertices[i*4 + 3];
+            vertices[i*4 + 3] = texture_pos_norm.y + texture_size_norm.y - prev_y;
+        }
+    }
 
     self->params.egl->glBindVertexArray(self->vertex_array_object_id);
     self->params.egl->glViewport(0, 0, dest_texture_size.x, dest_texture_size.y);
 
     /* TODO: this, also cleanup */
-    //self->params.egl->glBindBuffer(GL_ARRAY_BUFFER, self->vertex_buffer_object_id);
+    self->params.egl->glBindBuffer(GL_ARRAY_BUFFER, self->vertex_buffer_object_id);
     self->params.egl->glBufferSubData(GL_ARRAY_BUFFER, 0, 24 * sizeof(float), vertices);
 
-    {
-        self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[0]);
-        //cap_xcomp->params.egl->glClear(GL_COLOR_BUFFER_BIT); // TODO: Do this in a separate clear_ function. We want to do that when using multiple drm to create the final image (multiple monitors for example)
+    switch(source_color) {
+        case GSR_SOURCE_COLOR_RGB:
+        case GSR_SOURCE_COLOR_BGR: {
+            switch(self->params.destination_color) {
+                case GSR_DESTINATION_COLOR_NV12:
+                case GSR_DESTINATION_COLOR_P010: {
+                    self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[0]);
+                    //cap_xcomp->params.egl->glClear(GL_COLOR_BUFFER_BIT); // TODO: Do this in a separate clear_ function. We want to do that when using multiple drm to create the final image (multiple monitors for example)
 
-        const int shader_index = external_texture ? EXTERNAL_TEXTURE_SHADER_OFFSET : 0;
-        gsr_shader_use(&self->shaders[shader_index]);
-        self->params.egl->glUniform1f(self->uniforms[shader_index].rotation, rotation);
-        self->params.egl->glUniform2f(self->uniforms[shader_index].offset, pos_norm.x, pos_norm.y);
-        self->params.egl->glDrawArrays(GL_TRIANGLES, 0, 6);
-    }
+                    int shader_index = external_texture ? GRAPHICS_SHADER_INDEX_Y_EXTERNAL : GRAPHICS_SHADER_INDEX_Y;
+                    gsr_shader_use(&self->graphics_shaders[shader_index]);
+                    self->params.egl->glUniformMatrix2fv(self->graphics_uniforms[shader_index].rotation_matrix, 1, GL_TRUE, (const float*)rotation_matrix);
+                    self->params.egl->glUniform2f(self->graphics_uniforms[shader_index].offset, pos_norm.x, pos_norm.y);
+                    self->params.egl->glDrawArrays(GL_TRIANGLES, 0, 6);
 
-    if(self->params.num_destination_textures > 1) {
-        self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[1]);
-        //cap_xcomp->params.egl->glClear(GL_COLOR_BUFFER_BIT);
+                    if(self->params.num_destination_textures > 1) {
+                        self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[1]);
+                        //cap_xcomp->params.egl->glClear(GL_COLOR_BUFFER_BIT);
 
-        const int shader_index = external_texture ? EXTERNAL_TEXTURE_SHADER_OFFSET + 1 : 1;
-        gsr_shader_use(&self->shaders[shader_index]);
-        self->params.egl->glUniform1f(self->uniforms[shader_index].rotation, rotation);
-        self->params.egl->glUniform2f(self->uniforms[shader_index].offset, pos_norm.x, pos_norm.y);
-        self->params.egl->glDrawArrays(GL_TRIANGLES, 0, 6);
+                        shader_index = external_texture ? GRAPHICS_SHADER_INDEX_UV_EXTERNAL : GRAPHICS_SHADER_INDEX_UV;
+                        gsr_shader_use(&self->graphics_shaders[shader_index]);
+                        self->params.egl->glUniformMatrix2fv(self->graphics_uniforms[shader_index].rotation_matrix, 1, GL_TRUE, (const float*)rotation_matrix);
+                        self->params.egl->glUniform2f(self->graphics_uniforms[shader_index].offset, pos_norm.x, pos_norm.y);
+                        self->params.egl->glDrawArrays(GL_TRIANGLES, 0, 6);
+                    }
+                    break;
+                }
+                case GSR_DESTINATION_COLOR_RGB8: {
+                    self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[0]);
+                    //cap_xcomp->params.egl->glClear(GL_COLOR_BUFFER_BIT); // TODO: Do this in a separate clear_ function. We want to do that when using multiple drm to create the final image (multiple monitors for example)
+
+                    const int shader_index = external_texture ? GRAPHICS_SHADER_INDEX_RGB_EXTERNAL : GRAPHICS_SHADER_INDEX_RGB;
+                    gsr_shader_use(&self->graphics_shaders[shader_index]);
+                    self->params.egl->glUniformMatrix2fv(self->graphics_uniforms[shader_index].rotation_matrix, 1, GL_TRUE, (const float*)rotation_matrix);
+                    self->params.egl->glUniform2f(self->graphics_uniforms[shader_index].offset, pos_norm.x, pos_norm.y);
+                    self->params.egl->glDrawArrays(GL_TRIANGLES, 0, 6);
+                    break;
+                }
+            }
+            break;
+        }
+        case GSR_SOURCE_COLOR_YUYV: {
+            switch(self->params.destination_color) {
+                case GSR_DESTINATION_COLOR_NV12:
+                case GSR_DESTINATION_COLOR_P010: {
+                    self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[0]);
+                    //cap_xcomp->params.egl->glClear(GL_COLOR_BUFFER_BIT); // TODO: Do this in a separate clear_ function. We want to do that when using multiple drm to create the final image (multiple monitors for example)
+
+                    int shader_index = external_texture ? GRAPHICS_SHADER_INDEX_YUYV_TO_Y_EXTERNAL : GRAPHICS_SHADER_INDEX_YUYV_TO_Y;
+                    gsr_shader_use(&self->graphics_shaders[shader_index]);
+                    self->params.egl->glUniformMatrix2fv(self->graphics_uniforms[shader_index].rotation_matrix, 1, GL_TRUE, (const float*)rotation_matrix);
+                    self->params.egl->glUniform2f(self->graphics_uniforms[shader_index].offset, pos_norm.x, pos_norm.y);
+                    self->params.egl->glDrawArrays(GL_TRIANGLES, 0, 6);
+
+                    if(self->params.num_destination_textures > 1) {
+                        self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[1]);
+                        //cap_xcomp->params.egl->glClear(GL_COLOR_BUFFER_BIT);
+
+                        shader_index = external_texture ? GRAPHICS_SHADER_INDEX_YUYV_TO_UV_EXTERNAL : GRAPHICS_SHADER_INDEX_YUYV_TO_UV;
+                        gsr_shader_use(&self->graphics_shaders[shader_index]);
+                        self->params.egl->glUniformMatrix2fv(self->graphics_uniforms[shader_index].rotation_matrix, 1, GL_TRUE, (const float*)rotation_matrix);
+                        self->params.egl->glUniform2f(self->graphics_uniforms[shader_index].offset, pos_norm.x, pos_norm.y);
+                        self->params.egl->glDrawArrays(GL_TRIANGLES, 0, 6);
+                    }
+                    break;
+                }
+                case GSR_DESTINATION_COLOR_RGB8: {
+                    self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[0]);
+                    //cap_xcomp->params.egl->glClear(GL_COLOR_BUFFER_BIT); // TODO: Do this in a separate clear_ function. We want to do that when using multiple drm to create the final image (multiple monitors for example)
+
+                    const int shader_index = external_texture ? GRAPHICS_SHADER_INDEX_YUYV_TO_RGB_EXTERNAL : GRAPHICS_SHADER_INDEX_YUYV_TO_RGB;
+                    gsr_shader_use(&self->graphics_shaders[shader_index]);
+                    self->params.egl->glUniformMatrix2fv(self->graphics_uniforms[shader_index].rotation_matrix, 1, GL_TRUE, (const float*)rotation_matrix);
+                    self->params.egl->glUniform2f(self->graphics_uniforms[shader_index].offset, pos_norm.x, pos_norm.y);
+                    self->params.egl->glDrawArrays(GL_TRIANGLES, 0, 6);
+                    break;
+                }
+            }
+            break;
+        }
     }
 
     self->params.egl->glBindVertexArray(0);
-    gsr_shader_use_none(&self->shaders[0]);
+    self->params.egl->glUseProgram(0);
+    gsr_color_conversion_swizzle_reset(self, texture_target, source_color);
     self->params.egl->glBindTexture(texture_target, 0);
     self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, 0);
+}
 
-    gsr_color_conversion_swizzle_reset(self, source_color);
+void gsr_color_conversion_draw(gsr_color_conversion *self, unsigned int texture_id, vec2i destination_pos, vec2i destination_size, vec2i source_pos, vec2i source_size, vec2i texture_size, gsr_rotation rotation, gsr_flip flip, gsr_source_color source_color, bool external_texture) {
+    assert(!external_texture || self->params.load_external_image_shader);
+    if(external_texture && !self->params.load_external_image_shader) {
+        fprintf(stderr, "gsr error: gsr_color_conversion_draw: external texture not loaded\n");
+        return;
+    }
+
+    vec2f scale = {0.0f, 0.0f};
+    if(source_size.x > 0 && source_size.y > 0)
+        scale = (vec2f){ (double)destination_size.x/(double)source_size.x, (double)destination_size.y/(double)source_size.y };
+
+    vec2i source_position = {0, 0};
+    float rotation_matrix[2][2] = {{0, 0}, {0, 0}};
+    gsr_color_conversion_apply_rotation(rotation, rotation_matrix);
+
+    const int texture_target = external_texture ? GL_TEXTURE_EXTERNAL_OES : GL_TEXTURE_2D;
+    self->params.egl->glBindTexture(texture_target, texture_id);
+
+    source_position.x += source_pos.x;
+    source_position.y += source_pos.y;
+    gsr_color_conversion_draw_graphics(self, texture_id, external_texture, rotation, flip, rotation_matrix, source_position, source_size, destination_pos, texture_size, scale, source_color);
+
+    self->params.egl->glFlush();
+    // TODO: Use the minimal barrier required
+    self->params.egl->glMemoryBarrier(GL_ALL_BARRIER_BITS); // GL_SHADER_IMAGE_ACCESS_BARRIER_BIT
+    self->params.egl->glUseProgram(0);
+
+    self->params.egl->glBindTexture(texture_target, 0);
 }
 
 void gsr_color_conversion_clear(gsr_color_conversion *self) {
@@ -559,3 +934,14 @@ void gsr_color_conversion_clear(gsr_color_conversion *self) {
 
     self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, 0);
 }
+
+void gsr_color_conversion_read_destination_texture(gsr_color_conversion *self, int destination_texture_index, int x, int y, int width, int height, unsigned int color_format, unsigned int data_format, void *pixels) {
+    assert(destination_texture_index >= 0 && destination_texture_index < self->params.num_destination_textures);
+    self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, self->framebuffers[destination_texture_index]);
+    self->params.egl->glReadPixels(x, y, width, height, color_format, data_format, pixels);
+    self->params.egl->glBindFramebuffer(GL_FRAMEBUFFER, 0);
+}
+
+gsr_rotation gsr_monitor_rotation_to_rotation(gsr_monitor_rotation monitor_rotation) {
+    return (gsr_rotation)monitor_rotation;
+}

src/cuda.c

@@ -65,6 +65,7 @@ bool gsr_cuda_load(gsr_cuda *self, Display *display, bool do_overclock) {
         goto fail;
     }
 
+    // TODO: Use the device associated with the opengl graphics context
     CUdevice cu_dev;
     res = self->cuDeviceGet(&cu_dev, 0);
     if(res != CUDA_SUCCESS) {

src/cursor.c

@@ -56,8 +56,6 @@ static bool gsr_cursor_set_from_x11_cursor_image(gsr_cursor *self, XFixesCursorI
     self->egl->glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, self->size.x, self->size.y, 0, GL_RGBA, GL_UNSIGNED_BYTE, cursor_data);
     free(cursor_data);
 
-    self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
 
@@ -98,7 +96,7 @@ int gsr_cursor_init(gsr_cursor *self, gsr_egl *egl, Display *display) {
 }
 
 void gsr_cursor_deinit(gsr_cursor *self) {
-    if(!self->egl)
+    if(!self->display)
         return;
 
     if(self->texture_id) {
@@ -114,6 +112,9 @@ void gsr_cursor_deinit(gsr_cursor *self) {
 }
 
 bool gsr_cursor_on_event(gsr_cursor *self, XEvent *xev) {
+    if(!self->display)
+        return false;
+
     bool updated = false;
 
     if(xev->type == self->x_fixes_event_base + XFixesCursorNotify) {
@@ -133,6 +134,9 @@ bool gsr_cursor_on_event(gsr_cursor *self, XEvent *xev) {
 }
 
 void gsr_cursor_tick(gsr_cursor *self, Window relative_to) {
+    if(!self->display)
+        return;
+
     Window dummy_window;
     int dummy_i;
     unsigned int dummy_u;

src/damage.c

@@ -4,6 +4,7 @@
 
 #include <stdio.h>
 #include <string.h>
+#include <stdlib.h>
 #include <X11/extensions/Xdamage.h>
 #include <X11/extensions/Xrandr.h>
 
@@ -26,51 +27,115 @@ static bool xrandr_is_supported(Display *display) {
     return major_version > 1 || (major_version == 1 && minor_version >= 2);
 }
 
-bool gsr_damage_init(gsr_damage *self, gsr_egl *egl, bool track_cursor) {
+static int gsr_damage_get_tracked_monitor_index(const gsr_damage *self, const char *monitor_name) {
+    for(int i = 0; i < self->num_monitors_tracked; ++i) {
+        if(strcmp(self->monitors_tracked[i].monitor_name, monitor_name) == 0)
+            return i;
+    }
+    return -1;
+}
+
+static void add_monitor_callback(const gsr_monitor *monitor, void *userdata) {
+    gsr_damage *self = userdata;
+
+    const int damage_monitor_index = gsr_damage_get_tracked_monitor_index(self, monitor->name);
+    gsr_damage_monitor *damage_monitor = NULL;
+    if(damage_monitor_index != -1) {
+        damage_monitor = &self->monitors_tracked[damage_monitor_index];
+        damage_monitor->monitor = NULL;
+    }
+
+    if(self->num_monitors + 1 > GSR_DAMAGE_MAX_MONITORS) {
+        fprintf(stderr, "gsr error: gsr_damage_on_output_change: max monitors reached\n");
+        return;
+    }
+
+    char *monitor_name_copy = strdup(monitor->name);
+    if(!monitor_name_copy) {
+        fprintf(stderr, "gsr error: gsr_damage_on_output_change: strdup failed for monitor: %s\n", monitor->name);
+        return;
+    }
+
+    self->monitors[self->num_monitors] = *monitor;
+    self->monitors[self->num_monitors].name = monitor_name_copy;
+    ++self->num_monitors;
+
+    if(damage_monitor)
+        damage_monitor->monitor = &self->monitors[self->num_monitors - 1];
+}
+
+bool gsr_damage_init(gsr_damage *self, gsr_egl *egl, gsr_cursor *cursor, bool track_cursor) {
     memset(self, 0, sizeof(*self));
     self->egl = egl;
     self->track_cursor = track_cursor;
+    self->cursor = cursor;
 
     if(gsr_window_get_display_server(egl->window) != GSR_DISPLAY_SERVER_X11) {
-        fprintf(stderr, "gsr warning: gsr_damage_init: damage tracking is not supported on wayland\n");
+        fprintf(stderr, "gsr error: gsr_damage_init: damage tracking is not supported on wayland\n");
         return false;
     }
     self->display = gsr_window_get_display(egl->window);
 
     if(!XDamageQueryExtension(self->display, &self->damage_event, &self->damage_error)) {
-        fprintf(stderr, "gsr warning: gsr_damage_init: XDamage is not supported by your X11 server\n");
+        fprintf(stderr, "gsr error: gsr_damage_init: XDamage is not supported by your X11 server\n");
         gsr_damage_deinit(self);
         return false;
     }
 
     if(!XRRQueryExtension(self->display, &self->randr_event, &self->randr_error)) {
-        fprintf(stderr, "gsr warning: gsr_damage_init: XRandr is not supported by your X11 server\n");
+        fprintf(stderr, "gsr error: gsr_damage_init: XRandr is not supported by your X11 server\n");
         gsr_damage_deinit(self);
         return false;
     }
 
     if(!xrandr_is_supported(self->display)) {
-        fprintf(stderr, "gsr warning: gsr_damage_init: your X11 randr version is too old\n");
+        fprintf(stderr, "gsr error: gsr_damage_init: your X11 randr version is too old\n");
         gsr_damage_deinit(self);
         return false;
     }
 
-    if(self->track_cursor)
-        self->track_cursor = gsr_cursor_init(&self->cursor, self->egl, self->display) == 0;
-
     XRRSelectInput(self->display, DefaultRootWindow(self->display), RRScreenChangeNotifyMask | RRCrtcChangeNotifyMask | RROutputChangeNotifyMask);
 
+    self->monitor_damage = XDamageCreate(self->display, DefaultRootWindow(self->display), XDamageReportNonEmpty);
+    if(!self->monitor_damage) {
+        fprintf(stderr, "gsr error: gsr_damage_init: XDamageCreate failed\n");
+        gsr_damage_deinit(self);
+        return false;
+    }
+    XDamageSubtract(self->display, self->monitor_damage, None, None);
+
+    for_each_active_monitor_output_x11_not_cached(self->display, add_monitor_callback, self);
+
     self->damaged = true;
     return true;
 }
 
+static void gsr_damage_deinit_monitors(gsr_damage *self) {
+    for(int i = 0; i < self->num_monitors; ++i) {
+        free((char*)self->monitors[i].name);
+    }
+    self->num_monitors = 0;
+}
+
 void gsr_damage_deinit(gsr_damage *self) {
-    if(self->damage) {
-        XDamageDestroy(self->display, self->damage);
-        self->damage = None;
+    if(self->monitor_damage) {
+        XDamageDestroy(self->display, self->monitor_damage);
+        self->monitor_damage = None;
     }
 
-    gsr_cursor_deinit(&self->cursor);
+    for(int i = 0; i < self->num_monitors_tracked; ++i) {
+        free(self->monitors_tracked[i].monitor_name);
+    }
+    self->num_monitors_tracked = 0;
+
+    for(int i = 0; i < self->num_windows_tracked; ++i) {
+        XSelectInput(self->display, self->windows_tracked[i].window_id, 0);
+        XDamageDestroy(self->display, self->windows_tracked[i].damage);
+    }
+    self->num_windows_tracked = 0;
+
+    self->all_monitors_tracked_refcount = 0;
+    gsr_damage_deinit_monitors(self);
 
     self->damage_event = 0;
     self->damage_error = 0;
@@ -79,139 +144,191 @@ void gsr_damage_deinit(gsr_damage *self) {
     self->randr_error = 0;
 }
 
-bool gsr_damage_set_target_window(gsr_damage *self, uint64_t window) {
-    if(self->damage_event == 0)
+static int gsr_damage_get_tracked_window_index(const gsr_damage *self, int64_t window) {
+    for(int i = 0; i < self->num_windows_tracked; ++i) {
+        if(self->windows_tracked[i].window_id == window)
+            return i;
+    }
+    return -1;
+}
+
+bool gsr_damage_start_tracking_window(gsr_damage *self, int64_t window) {
+    if(self->damage_event == 0 || window == None)
         return false;
 
-    if(window == self->window)
+    const int damage_window_index = gsr_damage_get_tracked_window_index(self, window);
+    if(damage_window_index != -1) {
+        ++self->windows_tracked[damage_window_index].refcount;
         return true;
-
-    if(self->damage) {
-        XDamageDestroy(self->display, self->damage);
-        self->damage = None;
     }
 
-    if(self->window)
-        XSelectInput(self->display, self->window, 0);
-
-    self->window = window;
-    XSelectInput(self->display, self->window, StructureNotifyMask | ExposureMask);
+    if(self->num_windows_tracked + 1 > GSR_DAMAGE_MAX_TRACKED_TARGETS) {
+        fprintf(stderr, "gsr error: gsr_damage_start_tracking_window: max window targets reached\n");
+        return false;
+    }
 
     XWindowAttributes win_attr;
     win_attr.x = 0;
     win_attr.y = 0;
     win_attr.width = 0;
     win_attr.height = 0;
-    if(!XGetWindowAttributes(self->display, self->window, &win_attr))
-        fprintf(stderr, "gsr warning: gsr_damage_set_target_window failed: failed to get window attributes: %ld\n", (long)self->window);
+    if(!XGetWindowAttributes(self->display, window, &win_attr))
+        fprintf(stderr, "gsr warning: gsr_damage_start_tracking_window failed: failed to get window attributes: %ld\n", (long)window);
 
-    //self->window_pos.x = win_attr.x;
-    //self->window_pos.y = win_attr.y;
-
-    self->window_size.x = win_attr.width;
-    self->window_size.y = win_attr.height;
-
-    self->damage = XDamageCreate(self->display, window, XDamageReportNonEmpty);
-    if(self->damage) {
-        XDamageSubtract(self->display, self->damage, None, None);
-        self->damaged = true;
-        self->track_type = GSR_DAMAGE_TRACK_WINDOW;
-        return true;
-    } else {
-        fprintf(stderr, "gsr warning: gsr_damage_set_target_window: XDamageCreate failed\n");
-        self->track_type = GSR_DAMAGE_TRACK_NONE;
+    const Damage damage = XDamageCreate(self->display, window, XDamageReportNonEmpty);
+    if(!damage) {
+        fprintf(stderr, "gsr error: gsr_damage_start_tracking_window: XDamageCreate failed\n");
         return false;
     }
+    XDamageSubtract(self->display, damage, None, None);
+
+    XSelectInput(self->display, window, StructureNotifyMask | ExposureMask);
+
+    gsr_damage_window *damage_window = &self->windows_tracked[self->num_windows_tracked];
+    ++self->num_windows_tracked;
+
+    damage_window->window_id = window;
+    damage_window->window_pos.x = win_attr.x;
+    damage_window->window_pos.y = win_attr.y;
+    damage_window->window_size.x = win_attr.width;
+    damage_window->window_size.y = win_attr.height;
+    damage_window->damage = damage;
+    damage_window->refcount = 1;
+    return true;
+}
+
+void gsr_damage_stop_tracking_window(gsr_damage *self, int64_t window) {
+    if(window == None)
+        return;
+
+    const int damage_window_index = gsr_damage_get_tracked_window_index(self, window);
+    if(damage_window_index == -1)
+        return;
+
+    gsr_damage_window *damage_window = &self->windows_tracked[damage_window_index];
+    --damage_window->refcount;
+    if(damage_window->refcount <= 0) {
+        XSelectInput(self->display, damage_window->window_id, 0);
+        XDamageDestroy(self->display, damage_window->damage);
+        self->windows_tracked[damage_window_index] = self->windows_tracked[self->num_windows_tracked - 1];
+        --self->num_windows_tracked;
+    }
 }
 
-bool gsr_damage_set_target_monitor(gsr_damage *self, const char *monitor_name) {
+static gsr_monitor* gsr_damage_get_monitor_by_id(gsr_damage *self, RRCrtc id) {
+    for(int i = 0; i < self->num_monitors; ++i) {
+        if(self->monitors[i].monitor_identifier == id)
+            return &self->monitors[i];
+    }
+    return NULL;
+}
+
+static gsr_monitor* gsr_damage_get_monitor_by_name(gsr_damage *self, const char *name) {
+    for(int i = 0; i < self->num_monitors; ++i) {
+        if(strcmp(self->monitors[i].name, name) == 0)
+            return &self->monitors[i];
+    }
+    return NULL;
+}
+
+bool gsr_damage_start_tracking_monitor(gsr_damage *self, const char *monitor_name) {
     if(self->damage_event == 0)
         return false;
 
-    if(strcmp(self->monitor_name, monitor_name) == 0)
-        return true;
+    if(strcmp(monitor_name, "screen-direct") == 0 || strcmp(monitor_name, "screen-direct-force") == 0)
+        monitor_name = NULL;
 
-    if(self->damage) {
-        XDamageDestroy(self->display, self->damage);
-        self->damage = None;
+    if(!monitor_name) {
+        ++self->all_monitors_tracked_refcount;
+        return true;
     }
 
-    memset(&self->monitor, 0, sizeof(self->monitor));
-    if(strcmp(monitor_name, "screen-direct") != 0 && strcmp(monitor_name, "screen-direct-force") != 0) {
-        if(!get_monitor_by_name(self->egl, GSR_CONNECTION_X11, monitor_name, &self->monitor))
-            fprintf(stderr, "gsr warning: gsr_damage_set_target_monitor: failed to find monitor: %s\n", monitor_name);
+    const int damage_monitor_index = gsr_damage_get_tracked_monitor_index(self, monitor_name);
+    if(damage_monitor_index != -1) {
+        ++self->monitors_tracked[damage_monitor_index].refcount;
+        return true;
     }
 
-    if(self->window)
-        XSelectInput(self->display, self->window, 0);
-
-    self->window = DefaultRootWindow(self->display);
-    self->damage = XDamageCreate(self->display, self->window, XDamageReportNonEmpty);
-    if(self->damage) {
-        XDamageSubtract(self->display, self->damage, None, None);
-        self->damaged = true;
-        snprintf(self->monitor_name, sizeof(self->monitor_name), "%s", monitor_name);
-        self->track_type = GSR_DAMAGE_TRACK_MONITOR;
-        return true;
-    } else {
-        fprintf(stderr, "gsr warning: gsr_damage_set_target_monitor: XDamageCreate failed\n");
-        self->track_type = GSR_DAMAGE_TRACK_NONE;
+    if(self->num_monitors_tracked + 1 > GSR_DAMAGE_MAX_TRACKED_TARGETS) {
+        fprintf(stderr, "gsr error: gsr_damage_start_tracking_monitor: max monitor targets reached\n");
         return false;
     }
+
+    char *monitor_name_copy = strdup(monitor_name);
+    if(!monitor_name_copy) {
+        fprintf(stderr, "gsr error: gsr_damage_start_tracking_monitor: strdup failed for monitor: %s\n", monitor_name);
+        return false;
+    }
+
+    gsr_monitor *monitor = gsr_damage_get_monitor_by_name(self, monitor_name);
+    if(!monitor) {
+        fprintf(stderr, "gsr error: gsr_damage_start_tracking_monitor: failed to find monitor: %s\n", monitor_name);
+        free(monitor_name_copy);
+        return false;
+    }
+
+    gsr_damage_monitor *damage_monitor = &self->monitors_tracked[self->num_monitors_tracked];
+    ++self->num_monitors_tracked;
+
+    damage_monitor->monitor_name = monitor_name_copy;
+    damage_monitor->monitor = monitor;
+    damage_monitor->refcount = 1;
+    return true;
+}
+
+void gsr_damage_stop_tracking_monitor(gsr_damage *self, const char *monitor_name) {
+    if(strcmp(monitor_name, "screen-direct") == 0 || strcmp(monitor_name, "screen-direct-force") == 0)
+        monitor_name = NULL;
+
+    if(!monitor_name) {
+        --self->all_monitors_tracked_refcount;
+        if(self->all_monitors_tracked_refcount < 0)
+            self->all_monitors_tracked_refcount = 0;
+        return;
+    }
+
+    const int damage_monitor_index = gsr_damage_get_tracked_monitor_index(self, monitor_name);
+    if(damage_monitor_index == -1)
+        return;
+
+    gsr_damage_monitor *damage_monitor = &self->monitors_tracked[damage_monitor_index];
+    --damage_monitor->refcount;
+    if(damage_monitor->refcount <= 0) {
+        free(damage_monitor->monitor_name);
+        self->monitors_tracked[damage_monitor_index] = self->monitors_tracked[self->num_monitors_tracked - 1];
+        --self->num_monitors_tracked;
+    }
 }
 
 static void gsr_damage_on_crtc_change(gsr_damage *self, XEvent *xev) {
     const XRRCrtcChangeNotifyEvent *rr_crtc_change_event = (XRRCrtcChangeNotifyEvent*)xev;
-    if(rr_crtc_change_event->crtc == 0 || self->monitor.monitor_identifier == 0)
-        return;
-
-    if(rr_crtc_change_event->crtc != self->monitor.monitor_identifier)
+    if(rr_crtc_change_event->crtc == 0)
         return;
 
     if(rr_crtc_change_event->width == 0 || rr_crtc_change_event->height == 0)
         return;
 
-    if(rr_crtc_change_event->x != self->monitor.pos.x || rr_crtc_change_event->y != self->monitor.pos.y ||
-        (int)rr_crtc_change_event->width != self->monitor.size.x || (int)rr_crtc_change_event->height != self->monitor.size.y) {
-        self->monitor.pos.x = rr_crtc_change_event->x;
-        self->monitor.pos.y = rr_crtc_change_event->y;
+    gsr_monitor *monitor = gsr_damage_get_monitor_by_id(self, rr_crtc_change_event->crtc);
+    if(!monitor)
+        return;
 
-        self->monitor.size.x = rr_crtc_change_event->width;
-        self->monitor.size.y = rr_crtc_change_event->height;
+    if(rr_crtc_change_event->x != monitor->pos.x || rr_crtc_change_event->y != monitor->pos.y ||
+        (int)rr_crtc_change_event->width != monitor->size.x || (int)rr_crtc_change_event->height != monitor->size.y) {
+        monitor->pos.x = rr_crtc_change_event->x;
+        monitor->pos.y = rr_crtc_change_event->y;
+
+        monitor->size.x = rr_crtc_change_event->width;
+        monitor->size.y = rr_crtc_change_event->height;
     }
 }
 
 static void gsr_damage_on_output_change(gsr_damage *self, XEvent *xev) {
     const XRROutputChangeNotifyEvent *rr_output_change_event = (XRROutputChangeNotifyEvent*)xev;
-    if(!rr_output_change_event->output || self->monitor.monitor_identifier == 0)
+    if(!rr_output_change_event->output)
         return;
 
-    XRRScreenResources *screen_res = XRRGetScreenResources(self->display, DefaultRootWindow(self->display));
-    if(!screen_res)
-        return;
-
-    // TODO: What about scaled output? look at for_each_active_monitor_output_x11_not_cached
-    XRROutputInfo *out_info = XRRGetOutputInfo(self->display, screen_res, rr_output_change_event->output);
-    if(out_info && out_info->crtc && out_info->crtc == self->monitor.monitor_identifier) {
-        XRRCrtcInfo *crtc_info = XRRGetCrtcInfo(self->display, screen_res, out_info->crtc);
-        if(crtc_info && (crtc_info->x != self->monitor.pos.x || crtc_info->y != self->monitor.pos.y ||
-            (int)crtc_info->width != self->monitor.size.x || (int)crtc_info->height != self->monitor.size.y))
-        {
-            self->monitor.pos.x = crtc_info->x;
-            self->monitor.pos.y = crtc_info->y;
-
-            self->monitor.size.x = crtc_info->width;
-            self->monitor.size.y = crtc_info->height;
-        }
-
-        if(crtc_info)
-            XRRFreeCrtcInfo(crtc_info);
-    }
-
-    if(out_info)
-        XRRFreeOutputInfo(out_info);
-    
-    XRRFreeScreenResources(screen_res);
+    gsr_damage_deinit_monitors(self);
+    for_each_active_monitor_output_x11_not_cached(self->display, add_monitor_callback, self);
 }
 
 static void gsr_damage_on_randr_event(gsr_damage *self, XEvent *xev) {
@@ -232,19 +349,38 @@ static void gsr_damage_on_damage_event(gsr_damage *self, XEvent *xev) {
     /* Subtract all the damage, repairing the window */
     XDamageSubtract(self->display, de->damage, None, region);
 
-    if(self->track_type == GSR_DAMAGE_TRACK_WINDOW || (self->track_type == GSR_DAMAGE_TRACK_MONITOR && self->monitor.connector_id == 0)) {
+    if(self->all_monitors_tracked_refcount > 0)
         self->damaged = true;
-    } else {
+
+    if(!self->damaged) {
+        for(int i = 0; i < self->num_windows_tracked; ++i) {
+            const gsr_damage_window *damage_window = &self->windows_tracked[i];
+            if(damage_window->window_id == (int64_t)de->drawable) {
+                self->damaged = true;
+                break;
+            }
+        }
+    }
+
+    if(!self->damaged) {
         int num_rectangles = 0;
         XRectangle *rectangles = XFixesFetchRegion(self->display, region, &num_rectangles);
         if(rectangles) {
-            const gsr_rectangle monitor_region = { self->monitor.pos, self->monitor.size };
             for(int i = 0; i < num_rectangles; ++i) {
                 const gsr_rectangle damage_region = { (vec2i){rectangles[i].x, rectangles[i].y}, (vec2i){rectangles[i].width, rectangles[i].height} };
-                self->damaged = rectangles_intersect(monitor_region, damage_region);
-                if(self->damaged)
-                    break;
+                for(int j = 0; j < self->num_monitors_tracked; ++j) {
+                    const gsr_monitor *monitor = self->monitors_tracked[j].monitor;
+                    if(!monitor)
+                        continue;
+
+                    const gsr_rectangle monitor_region = { monitor->pos, monitor->size };
+                    self->damaged = rectangles_intersect(monitor_region, damage_region);
+                    if(self->damaged)
+                        goto intersection_found;
+                }
             }
+
+            intersection_found:
             XFree(rectangles);
         }
     }
@@ -254,23 +390,35 @@ static void gsr_damage_on_damage_event(gsr_damage *self, XEvent *xev) {
 }
 
 static void gsr_damage_on_tick_cursor(gsr_damage *self) {
-    vec2i prev_cursor_pos = self->cursor.position;
-    gsr_cursor_tick(&self->cursor, self->window);
-    if(self->cursor.position.x != prev_cursor_pos.x || self->cursor.position.y != prev_cursor_pos.y) {
-        const gsr_rectangle cursor_region = { self->cursor.position, self->cursor.size };
-        switch(self->track_type) {
-            case GSR_DAMAGE_TRACK_NONE: {
+    if(self->cursor->position.x == self->cursor_pos.x && self->cursor->position.y == self->cursor_pos.y)
+        return;
+
+    self->cursor_pos = self->cursor->position;
+    const gsr_rectangle cursor_region = { self->cursor->position, self->cursor->size };
+
+    if(self->all_monitors_tracked_refcount > 0)
+        self->damaged = true;
+
+    if(!self->damaged) {
+        for(int i = 0; i < self->num_windows_tracked; ++i) {
+            const gsr_damage_window *damage_window = &self->windows_tracked[i];
+            const gsr_rectangle window_region = { damage_window->window_pos, damage_window->window_size };
+            if(rectangles_intersect(window_region, cursor_region)) {
                 self->damaged = true;
                 break;
             }
-            case GSR_DAMAGE_TRACK_WINDOW: {
-                const gsr_rectangle window_region = { (vec2i){0, 0}, self->window_size };
-                self->damaged = self->window_size.x == 0 || rectangles_intersect(window_region, cursor_region);
-                break;
-            }
-            case GSR_DAMAGE_TRACK_MONITOR: {
-                const gsr_rectangle monitor_region = { self->monitor.pos, self->monitor.size };
-                self->damaged = self->monitor.monitor_identifier == 0 || rectangles_intersect(monitor_region, cursor_region);
+        }
+    }
+
+    if(!self->damaged) {
+        for(int i = 0; i < self->num_monitors_tracked; ++i) {
+            const gsr_monitor *monitor = self->monitors_tracked[i].monitor;
+            if(!monitor)
+                continue;
+
+            const gsr_rectangle monitor_region = { monitor->pos, monitor->size };
+            if(rectangles_intersect(monitor_region, cursor_region)) {
+                self->damaged = true;
                 break;
             }
         }
@@ -278,21 +426,24 @@ static void gsr_damage_on_tick_cursor(gsr_damage *self) {
 }
 
 static void gsr_damage_on_window_configure_notify(gsr_damage *self, XEvent *xev) {
-    if(xev->xconfigure.window != self->window)
-        return;
+    for(int i = 0; i < self->num_windows_tracked; ++i) {
+        gsr_damage_window *damage_window = &self->windows_tracked[i];
+        if(damage_window->window_id == (int64_t)xev->xconfigure.window) {
+            damage_window->window_pos.x = xev->xconfigure.x;
+            damage_window->window_pos.y = xev->xconfigure.y;
 
-    //self->window_pos.x = xev->xconfigure.x;
-    //self->window_pos.y = xev->xconfigure.y;
-    
-    self->window_size.x = xev->xconfigure.width;
-    self->window_size.y = xev->xconfigure.height;
+            damage_window->window_size.x = xev->xconfigure.width;
+            damage_window->window_size.y = xev->xconfigure.height;
+            break;
+        }
+    }
 }
 
 void gsr_damage_on_event(gsr_damage *self, XEvent *xev) {
-    if(self->damage_event == 0 || self->track_type == GSR_DAMAGE_TRACK_NONE)
+    if(self->damage_event == 0)
         return;
 
-    if(self->track_type == GSR_DAMAGE_TRACK_WINDOW && xev->type == ConfigureNotify)
+    if(xev->type == ConfigureNotify)
         gsr_damage_on_window_configure_notify(self, xev);
 
     if(self->randr_event) {
@@ -305,21 +456,18 @@ void gsr_damage_on_event(gsr_damage *self, XEvent *xev) {
 
     if(self->damage_event && xev->type == self->damage_event + XDamageNotify)
         gsr_damage_on_damage_event(self, xev);
-
-    if(self->track_cursor)
-        gsr_cursor_on_event(&self->cursor, xev);
 }
 
 void gsr_damage_tick(gsr_damage *self) {
-    if(self->damage_event == 0 || self->track_type == GSR_DAMAGE_TRACK_NONE)
+    if(self->damage_event == 0)
         return;
 
-    if(self->track_cursor && self->cursor.visible && !self->damaged)
+    if(self->track_cursor && self->cursor->visible && !self->damaged)
         gsr_damage_on_tick_cursor(self);
 }
 
 bool gsr_damage_is_damaged(gsr_damage *self) {
-    return self->damage_event == 0 || !self->damage || self->damaged || self->track_type == GSR_DAMAGE_TRACK_NONE;
+    return self->damage_event == 0 || self->damaged;
 }
 
 void gsr_damage_clear(gsr_damage *self) {

src/dbus.c

@@ -1,5 +1,6 @@
 #include "../include/dbus.h"
-#include "../include/utils.h"
+
+#include <sys/random.h>
 
 #include <stdio.h>
 #include <string.h>
@@ -28,6 +29,25 @@ typedef struct {
     };
 } dict_entry;
 
+static bool generate_random_characters(char *buffer, int buffer_size, const char *alphabet, size_t alphabet_size) {
+    /* TODO: Use other functions on other platforms than linux */
+    if(getrandom(buffer, buffer_size, 0) < buffer_size) {
+        fprintf(stderr, "Failed to get random bytes, error: %s\n", strerror(errno));
+        return false;
+    }
+
+    for(int i = 0; i < buffer_size; ++i) {
+        unsigned char c = *(unsigned char*)&buffer[i];
+        buffer[i] = alphabet[c % alphabet_size];
+    }
+
+    return true;
+}
+
+static bool generate_random_characters_standard_alphabet(char *buffer, int buffer_size) {
+    return generate_random_characters(buffer, buffer_size, "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789", 62);
+}
+
 static const char* dict_value_type_to_string(dict_value_type type) {
     switch(type) {
         case DICT_TYPE_STRING: return "string";
@@ -58,14 +78,6 @@ bool gsr_dbus_init(gsr_dbus *self, const char *screencast_restore_token) {
         return false;
     }
 
-    /* TODO: Check the name */
-    const int ret = dbus_bus_request_name(self->con, "com.dec05eba.gpu_screen_recorder", DBUS_NAME_FLAG_REPLACE_EXISTING, &self->err);
-    if(dbus_error_is_set(&self->err)) {
-        fprintf(stderr, "gsr error: gsr_dbus_init: dbus_bus_request_name failed with error: %s\n", self->err.message);
-        gsr_dbus_deinit(self);
-        return false;
-    }
-
     if(screencast_restore_token) {
         self->screencast_restore_token = strdup(screencast_restore_token);
         if(!self->screencast_restore_token) {
@@ -75,12 +87,6 @@ bool gsr_dbus_init(gsr_dbus *self, const char *screencast_restore_token) {
         }
     }
 
-    (void)ret;
-    // if(ret != DBUS_REQUEST_NAME_REPLY_PRIMARY_OWNER) {
-    //     fprintf(stderr, "gsr error: gsr_capture_portal_setup_dbus: dbus_bus_request_name failed to get primary owner\n");
-    //     return false;
-    // }
-
     return true;
 }
 
@@ -99,8 +105,6 @@ void gsr_dbus_deinit(gsr_dbus *self) {
     if(self->con) {
         dbus_error_free(&self->err);
 
-        dbus_bus_release_name(self->con, "com.dec05eba.gpu_screen_recorder", NULL);
-
         // Apparently shouldn't be used when a connection is setup by using dbus_bus_get
         //dbus_connection_close(self->con);
         dbus_connection_unref(self->con);

src/defs.c

@@ -0,0 +1,100 @@
+#include "../include/defs.h"
+#include <assert.h>
+
+bool video_codec_is_hdr(gsr_video_codec video_codec) {
+    // TODO: Vulkan
+    switch(video_codec) {
+        case GSR_VIDEO_CODEC_HEVC_HDR:
+        case GSR_VIDEO_CODEC_AV1_HDR:
+            return true;
+        default:
+            return false;
+    }
+}
+
+gsr_video_codec hdr_video_codec_to_sdr_video_codec(gsr_video_codec video_codec) {
+    // TODO: Vulkan
+    switch(video_codec) {
+        case GSR_VIDEO_CODEC_HEVC_HDR:
+            return GSR_VIDEO_CODEC_HEVC;
+        case GSR_VIDEO_CODEC_AV1_HDR:
+            return GSR_VIDEO_CODEC_AV1;
+        default:
+            return video_codec;
+    }
+}
+
+gsr_color_depth video_codec_to_bit_depth(gsr_video_codec video_codec) {
+    // TODO: 10-bit Vulkan
+    switch(video_codec) {
+        case GSR_VIDEO_CODEC_HEVC_HDR:
+        case GSR_VIDEO_CODEC_HEVC_10BIT:
+        case GSR_VIDEO_CODEC_AV1_HDR:
+        case GSR_VIDEO_CODEC_AV1_10BIT:
+            return GSR_COLOR_DEPTH_10_BITS;
+        default:
+            return GSR_COLOR_DEPTH_8_BITS;
+    }
+}
+
+const char* video_codec_to_string(gsr_video_codec video_codec) {
+    switch(video_codec) {
+        case GSR_VIDEO_CODEC_H264:        return "h264";
+        case GSR_VIDEO_CODEC_HEVC:        return "hevc";
+        case GSR_VIDEO_CODEC_HEVC_HDR:    return "hevc_hdr";
+        case GSR_VIDEO_CODEC_HEVC_10BIT:  return "hevc_10bit";
+        case GSR_VIDEO_CODEC_AV1:         return "av1";
+        case GSR_VIDEO_CODEC_AV1_HDR:     return "av1_hdr";
+        case GSR_VIDEO_CODEC_AV1_10BIT:   return "av1_10bit";
+        case GSR_VIDEO_CODEC_VP8:         return "vp8";
+        case GSR_VIDEO_CODEC_VP9:         return "vp9";
+        case GSR_VIDEO_CODEC_H264_VULKAN: return "h264_vulkan";
+        case GSR_VIDEO_CODEC_HEVC_VULKAN: return "hevc_vulkan";
+    }
+    return "";
+}
+
+// bool video_codec_is_hevc(gsr_video_codec video_codec) {
+//     // TODO: 10-bit vulkan
+//     switch(video_codec) {
+//         case GSR_VIDEO_CODEC_HEVC:
+//         case GSR_VIDEO_CODEC_HEVC_HDR:
+//         case GSR_VIDEO_CODEC_HEVC_10BIT:
+//         case GSR_VIDEO_CODEC_HEVC_VULKAN:
+//             return true;
+//         default:
+//             return false;
+//     }
+// }
+
+bool video_codec_is_av1(gsr_video_codec video_codec) {
+    // TODO: Vulkan
+    switch(video_codec) {
+        case GSR_VIDEO_CODEC_AV1:
+        case GSR_VIDEO_CODEC_AV1_HDR:
+        case GSR_VIDEO_CODEC_AV1_10BIT:
+            return true;
+        default:
+            return false;
+    }
+}
+
+bool video_codec_is_vulkan(gsr_video_codec video_codec) {
+    switch(video_codec) {
+        case GSR_VIDEO_CODEC_H264_VULKAN:
+        case GSR_VIDEO_CODEC_HEVC_VULKAN:
+            return true;
+        default:
+            return false;
+    }
+}
+
+const char* audio_codec_get_name(gsr_audio_codec audio_codec) {
+    switch(audio_codec) {
+        case GSR_AUDIO_CODEC_AAC:  return "aac";
+        case GSR_AUDIO_CODEC_OPUS: return "opus";
+        case GSR_AUDIO_CODEC_FLAC: return "flac";
+    }
+    assert(false);
+    return "";
+}

src/egl.c

@@ -9,7 +9,6 @@
 #include <dlfcn.h>
 #include <assert.h>
 #include <unistd.h>
-#include <sys/capability.h>
 
 // TODO: rename gsr_egl to something else since this includes both egl and glx and in the future maybe vulkan too
 
@@ -29,43 +28,28 @@
 #define GLX_DEPTH_SIZE                     12
 #define GLX_RGBA_TYPE                      0x8014
 
-#define GLX_CONTEXT_PRIORITY_LEVEL_EXT    0x3100
-#define GLX_CONTEXT_PRIORITY_HIGH_EXT     0x3101
-#define GLX_CONTEXT_PRIORITY_MEDIUM_EXT   0x3102
-#define GLX_CONTEXT_PRIORITY_LOW_EXT      0x3103
-
-static void reset_cap_nice(void) {
-    cap_t caps = cap_get_proc();
-    if(!caps)
-        return;
-
-    const cap_value_t cap_to_remove = CAP_SYS_NICE;
-    cap_set_flag(caps, CAP_EFFECTIVE, 1, &cap_to_remove, CAP_CLEAR);
-    cap_set_flag(caps, CAP_PERMITTED, 1, &cap_to_remove, CAP_CLEAR);
-    cap_set_proc(caps);
-    cap_free(caps);
-}
-
 // TODO: Create egl context without surface (in other words, x11/wayland agnostic, doesn't require x11/wayland dependency)
-static bool gsr_egl_create_window(gsr_egl *self) {
+static bool gsr_egl_create_window(gsr_egl *self, bool enable_debug) {
     EGLConfig  ecfg;
     int32_t    num_config = 0;
 
-    // TODO: Use EGL_OPENGL_ES_BIT as amd requires that for external texture, but that breaks software encoding
     const int32_t attr[] = {
         EGL_BUFFER_SIZE, 24,
-        EGL_RENDERABLE_TYPE, EGL_OPENGL_BIT,
+        EGL_RENDERABLE_TYPE, EGL_OPENGL_ES3_BIT,
         EGL_NONE, EGL_NONE
     };
 
-    const int32_t ctxattr[] = {
-        EGL_CONTEXT_CLIENT_VERSION, 2,
-        //EGL_CONTEXT_PRIORITY_LEVEL_IMG, EGL_CONTEXT_PRIORITY_HIGH_IMG, /* requires cap_sys_nice, ignored otherwise */
-        EGL_NONE, EGL_NONE
+    int32_t ctxattr[] = {
+        EGL_CONTEXT_CLIENT_VERSION, 3,
+        EGL_NONE, EGL_NONE, EGL_NONE
     };
 
-    // TODO: Use EGL_OPENGL_ES_API as amd requires that for external texture, but that breaks software encoding
-    self->eglBindAPI(EGL_OPENGL_API);
+    if(enable_debug) {
+        ctxattr[2] = EGL_CONTEXT_OPENGL_DEBUG;
+        ctxattr[3] = EGL_TRUE;
+    }
+
+    self->eglBindAPI(EGL_OPENGL_ES_API);
 
     self->egl_display = self->eglGetDisplay((EGLNativeDisplayType)gsr_window_get_display(self->window));
     if(!self->egl_display) {
@@ -100,16 +84,15 @@ static bool gsr_egl_create_window(gsr_egl *self) {
         goto fail;
     }
 
-    reset_cap_nice();
     return true;
 
     fail:
-    reset_cap_nice();
     gsr_egl_unload(self);
     return false;
 }
 
 static GLXFBConfig glx_fb_config_choose(gsr_egl *self, Display *display) {
+    // TODO: OpenGL debug context?
     const int glx_visual_attribs[] = {
         GLX_RENDER_TYPE, GLX_RGBA_BIT,
         GLX_DRAWABLE_TYPE, GLX_WINDOW_BIT,
@@ -225,6 +208,14 @@ static bool gsr_egl_proc_load_egl(gsr_egl *self) {
     self->eglQueryDeviceStringEXT = (FUNC_eglQueryDeviceStringEXT)self->eglGetProcAddress("eglQueryDeviceStringEXT");
     self->eglQueryDmaBufModifiersEXT = (FUNC_eglQueryDmaBufModifiersEXT)self->eglGetProcAddress("eglQueryDmaBufModifiersEXT");
 
+    self->glCreateMemoryObjectsEXT = (FUNC_glCreateMemoryObjectsEXT)self->eglGetProcAddress("glCreateMemoryObjectsEXT");
+    self->glImportMemoryFdEXT = (FUNC_glImportMemoryFdEXT)self->eglGetProcAddress("glImportMemoryFdEXT");
+    self->glIsMemoryObjectEXT = (FUNC_glIsMemoryObjectEXT)self->eglGetProcAddress("glIsMemoryObjectEXT");
+    self->glTexStorageMem2DEXT = (FUNC_glTexStorageMem2DEXT)self->eglGetProcAddress("glTexStorageMem2DEXT");
+    self->glBufferStorageMemEXT = (FUNC_glBufferStorageMemEXT)self->eglGetProcAddress("glBufferStorageMemEXT");
+    self->glNamedBufferStorageMemEXT = (FUNC_glNamedBufferStorageMemEXT)self->eglGetProcAddress("glNamedBufferStorageMemEXT");
+    self->glMemoryObjectParameterivEXT = (FUNC_glMemoryObjectParameterivEXT)self->eglGetProcAddress("glMemoryObjectParameterivEXT");
+
     if(!self->eglExportDMABUFImageQueryMESA) {
         fprintf(stderr, "gsr error: gsr_egl_load failed: could not find eglExportDMABUFImageQueryMESA\n");
         return false;
@@ -283,16 +274,20 @@ static bool gsr_egl_load_gl(gsr_egl *self, void *library) {
         { (void**)&self->glClearColor, "glClearColor" },
         { (void**)&self->glGenTextures, "glGenTextures" },
         { (void**)&self->glDeleteTextures, "glDeleteTextures" },
+        { (void**)&self->glActiveTexture, "glActiveTexture" },
         { (void**)&self->glBindTexture, "glBindTexture" },
+        { (void**)&self->glBindImageTexture, "glBindImageTexture" },
         { (void**)&self->glTexParameteri, "glTexParameteri" },
         { (void**)&self->glTexParameteriv, "glTexParameteriv" },
-        { (void**)&self->glGetTexLevelParameteriv, "glGetTexLevelParameteriv" },
+        { (void**)&self->glTexParameterfv, "glTexParameterfv" },
         { (void**)&self->glTexImage2D, "glTexImage2D" },
         { (void**)&self->glTexSubImage2D, "glTexSubImage2D" },
+        { (void**)&self->glTexStorage2D, "glTexStorage2D" },
         { (void**)&self->glGetTexImage, "glGetTexImage" },
         { (void**)&self->glGenFramebuffers, "glGenFramebuffers" },
         { (void**)&self->glBindFramebuffer, "glBindFramebuffer" },
         { (void**)&self->glDeleteFramebuffers, "glDeleteFramebuffers" },
+        { (void**)&self->glMemoryBarrier, "glMemoryBarrier" },
         { (void**)&self->glViewport, "glViewport" },
         { (void**)&self->glFramebufferTexture2D, "glFramebufferTexture2D" },
         { (void**)&self->glDrawBuffers, "glDrawBuffers" },
@@ -329,11 +324,15 @@ static bool gsr_egl_load_gl(gsr_egl *self, void *library) {
         { (void**)&self->glGetUniformLocation, "glGetUniformLocation" },
         { (void**)&self->glUniform1f, "glUniform1f" },
         { (void**)&self->glUniform2f, "glUniform2f" },
+        { (void**)&self->glUniform1i, "glUniform1i" },
+        { (void**)&self->glUniform2i, "glUniform2i" },
+        { (void**)&self->glUniformMatrix2fv, "glUniformMatrix2fv" },
         { (void**)&self->glDebugMessageCallback, "glDebugMessageCallback" },
         { (void**)&self->glScissor, "glScissor" },
         { (void**)&self->glReadPixels, "glReadPixels" },
-        { (void**)&self->glMapBuffer, "glMapBuffer" },
+        { (void**)&self->glMapBufferRange, "glMapBufferRange" },
         { (void**)&self->glUnmapBuffer, "glUnmapBuffer" },
+        { (void**)&self->glGetIntegerv, "glGetIntegerv" },
 
         { NULL, NULL }
     };
@@ -428,7 +427,7 @@ bool gsr_egl_load(gsr_egl *self, gsr_window *window, bool is_monitor_capture, bo
     if(!gsr_egl_proc_load_egl(self))
         goto fail;
 
-    if(!gsr_egl_create_window(self))
+    if(!gsr_egl_create_window(self, enable_debug))
         goto fail;
 
     if(!gl_get_gpu_info(self, &self->gpu_info))
@@ -448,25 +447,25 @@ bool gsr_egl_load(gsr_egl *self, gsr_window *window, bool is_monitor_capture, bo
             goto fail;
     }
 
+    if(enable_debug) {
+        self->glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS);
+        self->glDebugMessageCallback(debug_callback, NULL);
+    }
+
     self->glEnable(GL_BLEND);
     self->glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
     self->glPixelStorei(GL_PACK_ALIGNMENT, 1);
     self->glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
 
-    if(enable_debug) {
-        self->glEnable(GL_DEBUG_OUTPUT);
-        self->glDebugMessageCallback(debug_callback, NULL);
-    }
-
     gsr_egl_disable_vsync(self);
 
     if(self->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA) {
         /* This fixes nvenc codecs unable to load on openSUSE tumbleweed because of a cuda error. Don't ask me why */
         const bool inside_flatpak = getenv("FLATPAK_ID") != NULL;
         if(inside_flatpak)
-            system("flatpak-spawn --host -- nvidia-smi -f /dev/null");
+            system("flatpak-spawn --host -- sh -c 'grep -q openSUSE /etc/os-release && nvidia-smi -f /dev/null'");
         else
-            system("nvidia-smi -f /dev/null");
+            system("sh -c 'grep -q openSUSE /etc/os-release && nvidia-smi -f /dev/null'");
     }
 
     return true;
@@ -521,15 +520,10 @@ void gsr_egl_unload(gsr_egl *self) {
 }
 
 void gsr_egl_swap_buffers(gsr_egl *self) {
-    /* This uses less cpu than swap buffer on nvidia */
-    // TODO: Do these and remove swap
-    //self->glFlush();
-    //self->glFinish();
-    if(self->egl_display) {
-        self->eglSwapBuffers(self->egl_display, self->egl_surface);
-    } else if(gsr_window_get_display_server(self->window) == GSR_DISPLAY_SERVER_X11) {
-        Display *display = gsr_window_get_display(self->window);
-        const Window window = (Window)gsr_window_get_window(self->window);
-        self->glXSwapBuffers(display, window);
-    }
+    self->glFlush();
+    // TODO: Use the minimal barrier required
+    self->glMemoryBarrier(GL_ALL_BARRIER_BITS); // GL_SHADER_IMAGE_ACCESS_BARRIER_BIT
+    // TODO: This is needed on nvidia because the cursor can flicker otherwise. Find a better solution
+    if(self->gpu_info.vendor == GSR_GPU_VENDOR_NVIDIA)
+        self->glFinish();
 }

src/encoder/encoder.c

@@ -0,0 +1,242 @@
+#include "../../include/encoder/encoder.h"
+#include "../../include/utils.h"
+
+#include <string.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <inttypes.h>
+#include <time.h>
+#include <errno.h>
+
+#include <libavcodec/avcodec.h>
+#include <libavformat/avformat.h>
+
+static uint64_t clock_gettime_microseconds(clockid_t clock_id) {
+    struct timespec ts;
+    ts.tv_sec = 0;
+    ts.tv_nsec = 0;
+    clock_gettime(clock_id, &ts);
+    return (uint64_t)ts.tv_sec * 1000000ULL + (uint64_t)ts.tv_nsec / 1000ULL;
+}
+
+static void gsr_write_first_frame_timestamp_file(const char *filepath) {
+    const uint64_t evdev_compatible_ts = clock_gettime_microseconds(CLOCK_MONOTONIC);
+    const uint64_t unix_time_microsec = clock_gettime_microseconds(CLOCK_REALTIME);
+
+    FILE *file = fopen(filepath, "w");
+    if(!file) {
+        fprintf(stderr, "gsr warning: failed to open timestamp file '%s': %s\n", filepath, strerror(errno));
+        return;
+    }
+
+    fputs("monotonic_microsec\trealtime_microsec\n", file);
+    fprintf(file, "%" PRIu64 "\t%" PRIu64 "\n", evdev_compatible_ts, unix_time_microsec);
+    fclose(file);
+}
+
+bool gsr_encoder_init(gsr_encoder *self, gsr_replay_storage replay_storage, size_t replay_buffer_num_packets, double replay_buffer_time, const char *replay_directory) {
+    memset(self, 0, sizeof(*self));
+    self->num_recording_destinations = 0;
+    self->recording_destination_id_counter = 0;
+
+    if(pthread_mutex_init(&self->file_write_mutex, NULL) != 0) {
+        fprintf(stderr, "gsr error: gsr_encoder_init: failed to create mutex\n");
+        gsr_encoder_deinit(self);
+        return false;
+    }
+    self->file_write_mutex_created = true;
+
+    if(pthread_mutex_init(&self->replay_mutex, NULL) != 0) {
+        fprintf(stderr, "gsr error: gsr_encoder_init: failed to create mutex\n");
+        gsr_encoder_deinit(self);
+        return false;
+    }
+    self->replay_mutex_created = true;
+
+    if(replay_buffer_num_packets > 0) {
+        self->replay_buffer = gsr_replay_buffer_create(replay_storage, replay_directory, replay_buffer_time, replay_buffer_num_packets);
+        if(!self->replay_buffer) {
+            fprintf(stderr, "gsr error: gsr_encoder_init: failed to create replay buffer\n");
+            gsr_encoder_deinit(self);
+            return false;
+        }
+    }
+
+    return true;
+}
+
+void gsr_encoder_deinit(gsr_encoder *self)  {
+    if(self->file_write_mutex_created)
+        pthread_mutex_lock(&self->file_write_mutex);
+    for(size_t i = 0; i < self->num_recording_destinations; ++i) {
+        free(self->recording_destinations[i].first_frame_ts_filepath);
+        self->recording_destinations[i].first_frame_ts_filepath = NULL;
+        self->recording_destinations[i].first_frame_ts_written = false;
+    }
+    if(self->file_write_mutex_created)
+        pthread_mutex_unlock(&self->file_write_mutex);
+
+    if(self->replay_buffer) {
+        pthread_mutex_lock(&self->replay_mutex);
+        gsr_replay_buffer_destroy(self->replay_buffer);
+        self->replay_buffer = NULL;
+        pthread_mutex_unlock(&self->replay_mutex);
+    }
+
+    if(self->file_write_mutex_created) {
+        self->file_write_mutex_created = false;
+        pthread_mutex_destroy(&self->file_write_mutex);
+    }
+
+    if(self->replay_mutex_created) {
+        self->replay_mutex_created = false;
+        pthread_mutex_destroy(&self->replay_mutex);
+    }
+
+    self->num_recording_destinations = 0;
+    self->recording_destination_id_counter = 0;
+}
+
+void gsr_encoder_receive_packets(gsr_encoder *self, AVCodecContext *codec_context, int64_t pts, int stream_index) {
+    for(;;) {
+        AVPacket *av_packet = av_packet_alloc();
+        if(!av_packet)
+            break;
+
+        av_packet->data = NULL;
+        av_packet->size = 0;
+        int res = avcodec_receive_packet(codec_context, av_packet);
+        if(res == 0) { // we have a packet, send the packet to the muxer
+            av_packet->stream_index = stream_index;
+            av_packet->pts = pts;
+            av_packet->dts = pts;
+
+            if(self->replay_buffer) {
+                pthread_mutex_lock(&self->replay_mutex);
+                const double time_now = clock_get_monotonic_seconds();
+                if(!gsr_replay_buffer_append(self->replay_buffer, av_packet, time_now))
+                    fprintf(stderr, "gsr error: gsr_encoder_receive_packets: failed to add replay buffer data\n");
+                pthread_mutex_unlock(&self->replay_mutex);
+            }
+
+            pthread_mutex_lock(&self->file_write_mutex);
+            const bool is_keyframe = av_packet->flags & AV_PKT_FLAG_KEY;
+            for(size_t i = 0; i < self->num_recording_destinations; ++i) {
+                gsr_encoder_recording_destination *recording_destination = &self->recording_destinations[i];
+                if(recording_destination->codec_context != codec_context)
+                    continue;
+
+                if(is_keyframe)
+                    recording_destination->has_received_keyframe = true;
+                else if(!recording_destination->has_received_keyframe)
+                    continue;
+
+                if(recording_destination->first_frame_ts_filepath && !recording_destination->first_frame_ts_written) {
+                    gsr_write_first_frame_timestamp_file(recording_destination->first_frame_ts_filepath);
+                    recording_destination->first_frame_ts_written = true;
+                }
+
+                av_packet->pts = pts - recording_destination->start_pts;
+                av_packet->dts = pts - recording_destination->start_pts;
+
+                av_packet_rescale_ts(av_packet, codec_context->time_base, recording_destination->stream->time_base);
+                // TODO: Is av_interleaved_write_frame needed?. Answer: might be needed for mkv but dont use it! it causes frames to be inconsistent, skipping frames and duplicating frames.
+                // TODO: av_interleaved_write_frame might be needed for cfr, or always for flv
+                const int ret = av_write_frame(recording_destination->format_context, av_packet);
+                if(ret < 0) {
+                    char error_buffer[AV_ERROR_MAX_STRING_SIZE];
+                    if(av_strerror(ret, error_buffer, sizeof(error_buffer)) < 0)
+                        snprintf(error_buffer, sizeof(error_buffer), "Unknown error");
+                    fprintf(stderr, "gsr error: gsr_encoder_receive_packets: failed to write frame index %d to muxer, reason: %s (%d)\n", av_packet->stream_index, error_buffer, ret);
+                }
+            }
+            pthread_mutex_unlock(&self->file_write_mutex);
+
+            av_packet_free(&av_packet);
+        } else if (res == AVERROR(EAGAIN)) { // we have no packet
+                                             // fprintf(stderr, "No packet!\n");
+            av_packet_free(&av_packet);
+            break;
+        } else if (res == AVERROR_EOF) { // this is the end of the stream
+            av_packet_free(&av_packet);
+            fprintf(stderr, "End of stream!\n");
+            break;
+        } else {
+            av_packet_free(&av_packet);
+            fprintf(stderr, "Unexpected error: %d\n", res);
+            break;
+        }
+    }
+}
+
+size_t gsr_encoder_add_recording_destination(gsr_encoder *self, AVCodecContext *codec_context, AVFormatContext *format_context, AVStream *stream, int64_t start_pts) {
+    if(self->num_recording_destinations >= GSR_MAX_RECORDING_DESTINATIONS) {
+        fprintf(stderr, "gsr error: gsr_encoder_add_recording_destination: failed to add destination, reached the max amount of recording destinations (%d)\n", GSR_MAX_RECORDING_DESTINATIONS);
+        return (size_t)-1;
+    }
+
+    for(size_t i = 0; i < self->num_recording_destinations; ++i) {
+        if(self->recording_destinations[i].stream == stream) {
+            fprintf(stderr, "gsr error: gsr_encoder_add_recording_destination: failed to add destination, the stream %p already exists as an output\n", (void*)stream);
+            return (size_t)-1;
+        }
+    }
+
+    pthread_mutex_lock(&self->file_write_mutex);
+    gsr_encoder_recording_destination *recording_destination = &self->recording_destinations[self->num_recording_destinations];
+    recording_destination->id = self->recording_destination_id_counter;
+    recording_destination->codec_context = codec_context;
+    recording_destination->format_context = format_context;
+    recording_destination->stream = stream;
+    recording_destination->start_pts = start_pts;
+    recording_destination->has_received_keyframe = false;
+    recording_destination->first_frame_ts_filepath = NULL;
+    recording_destination->first_frame_ts_written = false;
+
+    ++self->recording_destination_id_counter;
+    ++self->num_recording_destinations;
+    pthread_mutex_unlock(&self->file_write_mutex);
+
+    return recording_destination->id;
+}
+
+bool gsr_encoder_remove_recording_destination(gsr_encoder *self, size_t id) {
+    bool found = false;
+    pthread_mutex_lock(&self->file_write_mutex);
+    for(size_t i = 0; i < self->num_recording_destinations; ++i) {
+        if(self->recording_destinations[i].id == id) {
+            free(self->recording_destinations[i].first_frame_ts_filepath);
+            self->recording_destinations[i].first_frame_ts_filepath = NULL;
+            self->recording_destinations[i].first_frame_ts_written = false;
+            self->recording_destinations[i] = self->recording_destinations[self->num_recording_destinations - 1];
+            --self->num_recording_destinations;
+            found = true;
+            break;
+        }
+    }
+    pthread_mutex_unlock(&self->file_write_mutex);
+    return found;
+}
+
+bool gsr_encoder_set_recording_destination_first_frame_ts_filepath(gsr_encoder *self, size_t id, const char *filepath) {
+    if(!filepath)
+        return false;
+
+    bool found = false;
+    pthread_mutex_lock(&self->file_write_mutex);
+    for(size_t i = 0; i < self->num_recording_destinations; ++i) {
+        if(self->recording_destinations[i].id == id) {
+            char *filepath_copy = strdup(filepath);
+            if(!filepath_copy)
+                break;
+
+            free(self->recording_destinations[i].first_frame_ts_filepath);
+            self->recording_destinations[i].first_frame_ts_filepath = filepath_copy;
+            self->recording_destinations[i].first_frame_ts_written = false;
+            found = true;
+            break;
+        }
+    }
+    pthread_mutex_unlock(&self->file_write_mutex);
+    return found;
+}

src/encoder/video/nvenc.c

@@ -12,6 +12,7 @@ typedef struct {
     gsr_video_encoder_nvenc_params params;
 
     unsigned int target_textures[2];
+    vec2i target_texture_size[2];
 
     AVBufferRef *device_ctx;
 
@@ -95,7 +96,8 @@ static bool gsr_video_encoder_nvenc_setup_textures(gsr_video_encoder_nvenc *self
     const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size
 
     for(int i = 0; i < 2; ++i) {
-        self->target_textures[i] = gl_create_texture(self->params.egl, video_codec_context->width / div[i], video_codec_context->height / div[i], self->params.color_depth == GSR_COLOR_DEPTH_8_BITS ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i], GL_NEAREST);
+        self->target_texture_size[i] = (vec2i){ video_codec_context->width / div[i], video_codec_context->height / div[i] };
+        self->target_textures[i] = gl_create_texture(self->params.egl, self->target_texture_size[i].x, self->target_texture_size[i].y, self->params.color_depth == GSR_COLOR_DEPTH_8_BITS ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i], GL_NEAREST);
         if(self->target_textures[i] == 0) {
             fprintf(stderr, "gsr error: gsr_video_encoder_nvenc_setup_textures: failed to create opengl texture\n");
             return false;
@@ -198,10 +200,12 @@ static void gsr_video_encoder_nvenc_copy_textures_to_frame(gsr_video_encoder *en
     self->cuda.cuStreamSynchronize(self->cuda_stream);
 }
 
-static void gsr_video_encoder_nvenc_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) {
+static void gsr_video_encoder_nvenc_get_textures(gsr_video_encoder *encoder, unsigned int *textures, vec2i *texture_sizes, int *num_textures, gsr_destination_color *destination_color) {
     gsr_video_encoder_nvenc *self = encoder->priv;
     textures[0] = self->target_textures[0];
     textures[1] = self->target_textures[1];
+    texture_sizes[0] = self->target_texture_size[0];
+    texture_sizes[1] = self->target_texture_size[1];
     *num_textures = 2;
     *destination_color = self->params.color_depth == GSR_COLOR_DEPTH_10_BITS ? GSR_DESTINATION_COLOR_P010 : GSR_DESTINATION_COLOR_NV12;
 }

src/encoder/video/software.c

@@ -13,6 +13,7 @@ typedef struct {
     gsr_video_encoder_software_params params;
 
     unsigned int target_textures[2];
+    vec2i texture_sizes[2];
 } gsr_video_encoder_software;
 
 static bool gsr_video_encoder_software_setup_textures(gsr_video_encoder_software *self, AVCodecContext *video_codec_context, AVFrame *frame) {
@@ -34,7 +35,8 @@ static bool gsr_video_encoder_software_setup_textures(gsr_video_encoder_software
     const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size
 
     for(int i = 0; i < 2; ++i) {
-        self->target_textures[i] = gl_create_texture(self->params.egl, video_codec_context->width / div[i], video_codec_context->height / div[i], self->params.color_depth == GSR_COLOR_DEPTH_8_BITS ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i], GL_NEAREST);
+        self->texture_sizes[i] = (vec2i){ video_codec_context->width / div[i], video_codec_context->height / div[i] };
+        self->target_textures[i] = gl_create_texture(self->params.egl, self->texture_sizes[i].x, self->texture_sizes[i].y, self->params.color_depth == GSR_COLOR_DEPTH_8_BITS ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i], GL_NEAREST);
         if(self->target_textures[i] == 0) {
             fprintf(stderr, "gsr error: gsr_capture_kms_setup_cuda_textures: failed to create opengl texture\n");
             return false;
@@ -71,26 +73,27 @@ void gsr_video_encoder_software_stop(gsr_video_encoder_software *self, AVCodecCo
 }
 
 static void gsr_video_encoder_software_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame, gsr_color_conversion *color_conversion) {
-    gsr_video_encoder_software *self = encoder->priv;
+    (void)encoder;
+    //gsr_video_encoder_software *self = encoder->priv;
     // TODO: hdr support
     const unsigned int formats[2] = { GL_RED, GL_RG };
+    const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size
     for(int i = 0; i < 2; ++i) {
-        self->params.egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[i]);
-        // We could use glGetTexSubImage and then we wouldn't have to use a specific linesize (LINESIZE_ALIGNMENT) that adds padding,
-        // but glGetTexSubImage is only available starting from opengl 4.5.
-        self->params.egl->glGetTexImage(GL_TEXTURE_2D, 0, formats[i], GL_UNSIGNED_BYTE, frame->data[i]);
+        // TODO: Use glPixelStore?
+        gsr_color_conversion_read_destination_texture(color_conversion, i, 0, 0, frame->width / div[i], frame->height / div[i], formats[i], GL_UNSIGNED_BYTE, frame->data[i]);
     }
-    self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
     // cap_kms->kms.base.egl->eglSwapBuffers(cap_kms->kms.base.egl->egl_display, cap_kms->kms.base.egl->egl_surface);
 
-    self->params.egl->glFlush();
-    self->params.egl->glFinish();
+    //self->params.egl->glFlush();
+    //self->params.egl->glFinish();
 }
 
-static void gsr_video_encoder_software_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) {
+static void gsr_video_encoder_software_get_textures(gsr_video_encoder *encoder, unsigned int *textures, vec2i *texture_sizes, int *num_textures, gsr_destination_color *destination_color) {
     gsr_video_encoder_software *self = encoder->priv;
     textures[0] = self->target_textures[0];
     textures[1] = self->target_textures[1];
+    texture_sizes[0] = self->texture_sizes[0];
+    texture_sizes[1] = self->texture_sizes[1];
     *num_textures = 2;
     *destination_color = self->params.color_depth == GSR_COLOR_DEPTH_10_BITS ? GSR_DESTINATION_COLOR_P010 : GSR_DESTINATION_COLOR_NV12;
 }

src/encoder/video/vaapi.c

@@ -6,6 +6,8 @@
 #include <libavutil/hwcontext_vaapi.h>
 #include <libavutil/intreadwrite.h>
 
+#include <va/va.h>
+#include <va/va_drm.h>
 #include <va/va_drmcommon.h>
 
 #include <stdlib.h>
@@ -16,6 +18,7 @@ typedef struct {
     gsr_video_encoder_vaapi_params params;
 
     unsigned int target_textures[2];
+    vec2i texture_sizes[2];
 
     AVBufferRef *device_ctx;
     VADisplay va_dpy;
@@ -112,6 +115,7 @@ static bool gsr_video_encoder_vaapi_setup_textures(gsr_video_encoder_vaapi *self
             intptr_t img_attr[44];
             setup_dma_buf_attrs(img_attr, formats[i], self->prime.width / div[i], self->prime.height / div[i],
                 fds, offsets, pitches, modifiers, self->prime.layers[layer].num_planes, true);
+            self->texture_sizes[i] = (vec2i){ self->prime.width / div[i], self->prime.height / div[i] };
 
             while(self->params.egl->eglGetError() != EGL_SUCCESS){}
             EGLImage image = self->params.egl->eglCreateImage(self->params.egl->egl_display, 0, EGL_LINUX_DMA_BUF_EXT, NULL, img_attr);
@@ -121,10 +125,8 @@ static bool gsr_video_encoder_vaapi_setup_textures(gsr_video_encoder_vaapi *self
             }
 
             self->params.egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[i]);
-            self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-            self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
-            self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
-            self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
+            self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+            self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
 
             while(self->params.egl->glGetError()) {}
             while(self->params.egl->eglGetError() != EGL_SUCCESS){}
@@ -150,13 +152,102 @@ static bool gsr_video_encoder_vaapi_setup_textures(gsr_video_encoder_vaapi *self
 
 static void gsr_video_encoder_vaapi_stop(gsr_video_encoder_vaapi *self, AVCodecContext *video_codec_context);
 
+static bool supports_hevc_without_padding(const char *card_path) {
+#if LIBAVCODEC_VERSION_INT >= AV_VERSION_INT(61, 28, 100) && VA_CHECK_VERSION(1, 21, 0)
+    VAStatus va_status;
+    VAConfigID va_config = 0;
+    unsigned int num_surface_attr = 0;
+    VASurfaceAttrib *surface_attr_list = NULL;
+    bool supports_surface_attrib_alignment_size = false;
+    int va_major = 0;
+    int va_minor = 0;
+    bool initialized = false;
+
+    char render_path[128];
+    if(!gsr_card_path_get_render_path(card_path, render_path)) {
+        fprintf(stderr, "gsr error: supports_hevc_without_padding: failed to get /dev/dri/renderDXXX file from %s\n", card_path);
+        return false;
+    }
+
+    const int drm_fd = open(render_path, O_RDWR);
+    if(drm_fd == -1) {
+        fprintf(stderr, "gsr error: supports_hevc_without_padding: failed to open device %s\n", render_path);
+        return false;
+    }
+
+    const VADisplay va_dpy = vaGetDisplayDRM(drm_fd);
+    if(!va_dpy) {
+        fprintf(stderr, "gsr error: supports_hevc_without_padding: failed to get vaapi display for device %s\n", render_path);
+        goto done;
+    }
+
+    vaSetInfoCallback(va_dpy, NULL, NULL);
+
+    if(vaInitialize(va_dpy, &va_major, &va_minor) != VA_STATUS_SUCCESS) {
+        fprintf(stderr, "gsr error: supports_hevc_without_padding: vaInitialize failed\n");
+        goto done;
+    }
+    initialized = true;
+
+    va_status = vaCreateConfig(va_dpy, VAProfileHEVCMain, VAEntrypointEncSlice, NULL, 0, &va_config);
+    if(va_status != VA_STATUS_SUCCESS) {
+        va_status = vaCreateConfig(va_dpy, VAProfileHEVCMain, VAEntrypointEncSliceLP, NULL, 0, &va_config);
+        if(va_status != VA_STATUS_SUCCESS) {
+            fprintf(stderr, "gsr error: supports_hevc_without_padding: failed to create hevc vaapi config, error: %s (%d)\n", vaErrorStr(va_status), va_status);
+            return false;
+        }
+    }
+
+    va_status = vaQuerySurfaceAttributes(va_dpy, va_config, 0, &num_surface_attr);
+    if(va_status != VA_STATUS_SUCCESS) {
+        fprintf(stderr, "gsr error: supports_hevc_without_padding: failed to query vaapi surface attributes size, error: %s (%d)\n", vaErrorStr(va_status), va_status);
+        goto done;
+    }
+
+    surface_attr_list = malloc(num_surface_attr * sizeof(VASurfaceAttrib));
+    if(!surface_attr_list) {
+        fprintf(stderr, "gsr error: supports_hevc_without_padding: failed to allocate memory for %u vaapi surface attributes, error: %s (%d)\n", num_surface_attr, vaErrorStr(va_status), va_status);
+        goto done;
+    }
+
+    va_status = vaQuerySurfaceAttributes(va_dpy, va_config, surface_attr_list, &num_surface_attr);
+    if(va_status != VA_STATUS_SUCCESS) {
+        fprintf(stderr, "gsr error: supports_hevc_without_padding: failed to query vaapi surface attributes data, error: %s (%d)\n", vaErrorStr(va_status), va_status);
+        goto done;
+    }
+
+    for(unsigned int i = 0; i < num_surface_attr; ++i) {
+        if(surface_attr_list[i].type == VASurfaceAttribAlignmentSize) {
+            supports_surface_attrib_alignment_size = true;
+            break;
+        }
+    }
+
+    done:
+    free(surface_attr_list);
+    if(va_config > 0)
+        vaDestroyConfig(va_dpy, va_config);
+    if(initialized)
+        vaTerminate(va_dpy);
+    if(drm_fd > 0)
+        close(drm_fd);
+    return supports_surface_attrib_alignment_size;
+#else
+    return false;
+#endif
+}
+
 static bool gsr_video_encoder_vaapi_start(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame) {
     gsr_video_encoder_vaapi *self = encoder->priv;
 
     if(self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && video_codec_context->codec_id == AV_CODEC_ID_HEVC) {
-        // TODO: dont do this if using ffmpeg reports that this is not needed (AMD driver bug that was fixed recently)
-        video_codec_context->width = FFALIGN(video_codec_context->width, 64);
-        video_codec_context->height = FFALIGN(video_codec_context->height, 16);
+        if(supports_hevc_without_padding(self->params.egl->card_path)) {
+            video_codec_context->width = FFALIGN(video_codec_context->width, 2);
+            video_codec_context->height = FFALIGN(video_codec_context->height, 2);
+        } else {
+            video_codec_context->width = FFALIGN(video_codec_context->width, 64);
+            video_codec_context->height = FFALIGN(video_codec_context->height, 16);
+        }
     } else if(self->params.egl->gpu_info.vendor == GSR_GPU_VENDOR_AMD && video_codec_context->codec_id == AV_CODEC_ID_AV1) {
         // TODO: Dont do this for VCN 5 and forward which should fix this hardware bug
         video_codec_context->width = FFALIGN(video_codec_context->width, 64);
@@ -216,10 +307,12 @@ void gsr_video_encoder_vaapi_stop(gsr_video_encoder_vaapi *self, AVCodecContext
     }
 }
 
-static void gsr_video_encoder_vaapi_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) {
+static void gsr_video_encoder_vaapi_get_textures(gsr_video_encoder *encoder, unsigned int *textures, vec2i *texture_sizes, int *num_textures, gsr_destination_color *destination_color) {
     gsr_video_encoder_vaapi *self = encoder->priv;
     textures[0] = self->target_textures[0];
     textures[1] = self->target_textures[1];
+    texture_sizes[0] = self->texture_sizes[0];
+    texture_sizes[1] = self->texture_sizes[1];
     *num_textures = 2;
     *destination_color = self->params.color_depth == GSR_COLOR_DEPTH_10_BITS ? GSR_DESTINATION_COLOR_P010 : GSR_DESTINATION_COLOR_NV12;
 }

src/encoder/video/video.c

@@ -1,4 +1,5 @@
 #include "../../../include/encoder/video/video.h"
+
 #include <assert.h>
 
 bool gsr_video_encoder_start(gsr_video_encoder *encoder, AVCodecContext *video_codec_context, AVFrame *frame) {
@@ -9,18 +10,19 @@ bool gsr_video_encoder_start(gsr_video_encoder *encoder, AVCodecContext *video_c
     return res;
 }
 
+void gsr_video_encoder_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context) {
+    assert(encoder->started);
+    encoder->started = false;
+    encoder->destroy(encoder, video_codec_context);
+}
+
 void gsr_video_encoder_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame, gsr_color_conversion *color_conversion) {
     assert(encoder->started);
     if(encoder->copy_textures_to_frame)
         encoder->copy_textures_to_frame(encoder, frame, color_conversion);
 }
 
-void gsr_video_encoder_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) {
+void gsr_video_encoder_get_textures(gsr_video_encoder *encoder, unsigned int *textures, vec2i *texture_sizes, int *num_textures, gsr_destination_color *destination_color) {
     assert(encoder->started);
-    encoder->get_textures(encoder, textures, num_textures, destination_color);
-}
-
-void gsr_video_encoder_destroy(gsr_video_encoder *encoder, AVCodecContext *video_codec_context) {
-    assert(encoder->started);
-    encoder->destroy(encoder, video_codec_context);
+    encoder->get_textures(encoder, textures, texture_sizes, num_textures, destination_color);
 }

src/encoder/video/vulkan.c

@@ -4,37 +4,27 @@
 
 #include <libavcodec/avcodec.h>
 #define VK_NO_PROTOTYPES
-#include <libavutil/hwcontext_vulkan.h>
+//#include <libavutil/hwcontext_vulkan.h>
 
 //#include <vulkan/vulkan_core.h>
 
+#define GL_HANDLE_TYPE_OPAQUE_FD_EXT      0x9586
 #define GL_TEXTURE_TILING_EXT             0x9580
 #define GL_OPTIMAL_TILING_EXT             0x9584
 #define GL_LINEAR_TILING_EXT              0x9585
 
-#define GL_PIXEL_PACK_BUFFER              0x88EB
-#define GL_PIXEL_UNPACK_BUFFER            0x88EC
-#define GL_STREAM_READ                    0x88E1
-#define GL_STREAM_DRAW                    0x88E0
-#define GL_READ_ONLY                      0x88B8
-#define GL_WRITE_ONLY                     0x88B9
-#define GL_READ_FRAMEBUFFER               0x8CA8
-
 typedef struct {
     gsr_video_encoder_vulkan_params params;
     unsigned int target_textures[2];
+    vec2i texture_sizes[2];
     AVBufferRef *device_ctx;
-    AVVulkanDeviceContext* vv;
-    unsigned int pbo_y[2];
-    unsigned int pbo_uv[2];
-    AVFrame *sw_frame;
 } gsr_video_encoder_vulkan;
 
 static bool gsr_video_encoder_vulkan_setup_context(gsr_video_encoder_vulkan *self, AVCodecContext *video_codec_context) {
     AVDictionary *options = NULL;
     //av_dict_set(&options, "linear_images", "1", 0);
     //av_dict_set(&options, "disable_multiplane", "1", 0);
-
+#if 0
     // TODO: Use correct device
     if(av_hwdevice_ctx_create(&self->device_ctx, AV_HWDEVICE_TYPE_VULKAN, NULL, options, 0) < 0) {
         fprintf(stderr, "gsr error: gsr_video_encoder_vulkan_setup_context: failed to create hardware device context\n");
@@ -68,9 +58,10 @@ static bool gsr_video_encoder_vulkan_setup_context(gsr_video_encoder_vulkan *sel
 
     video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context);
     av_buffer_unref(&frame_context);
+#endif
     return true;
 }
-
+#if 0
 static AVVulkanDeviceContext* video_codec_context_get_vulkan_data(AVCodecContext *video_codec_context) {
     AVBufferRef *hw_frames_ctx = video_codec_context->hw_frames_ctx;
     if(!hw_frames_ctx)
@@ -84,6 +75,24 @@ static AVVulkanDeviceContext* video_codec_context_get_vulkan_data(AVCodecContext
     return (AVVulkanDeviceContext*)device_context->hwctx;
 }
 
+static uint32_t get_memory_type_idx(VkPhysicalDevice pdev, const VkMemoryRequirements *mem_reqs, VkMemoryPropertyFlagBits prop_flags, PFN_vkGetPhysicalDeviceMemoryProperties vkGetPhysicalDeviceMemoryProperties) {
+    VkPhysicalDeviceMemoryProperties pdev_mem_props;
+    uint32_t i;
+
+    vkGetPhysicalDeviceMemoryProperties(pdev, &pdev_mem_props);
+
+    for (i = 0; i < pdev_mem_props.memoryTypeCount; i++) {
+        const VkMemoryType *type = &pdev_mem_props.memoryTypes[i];
+
+        if ((mem_reqs->memoryTypeBits & (1 << i)) &&
+            (type->propertyFlags & prop_flags) == prop_flags) {
+            return i;
+            break;
+        }
+    }
+    return UINT32_MAX;
+}
+#endif
 static bool gsr_video_encoder_vulkan_setup_textures(gsr_video_encoder_vulkan *self, AVCodecContext *video_codec_context, AVFrame *frame) {
     const int res = av_hwframe_get_buffer(video_codec_context->hw_frames_ctx, frame, 0);
     if(res < 0) {
@@ -91,56 +100,136 @@ static bool gsr_video_encoder_vulkan_setup_textures(gsr_video_encoder_vulkan *se
         return false;
     }
 
-    //AVVkFrame *target_surface_id = (AVVkFrame*)frame->data[0];
-    self->vv = video_codec_context_get_vulkan_data(video_codec_context);
+    while(self->params.egl->glGetError()) {}
+#if 0
+    AVVkFrame *target_surface_id = (AVVkFrame*)frame->data[0];
+    AVVulkanDeviceContext* vv = video_codec_context_get_vulkan_data(video_codec_context);
+    const size_t luma_size = frame->width * frame->height;
+    if(vv) {
+        PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements = (PFN_vkGetImageMemoryRequirements)vv->get_proc_addr(vv->inst, "vkGetImageMemoryRequirements");
+        PFN_vkAllocateMemory vkAllocateMemory = (PFN_vkAllocateMemory)vv->get_proc_addr(vv->inst, "vkAllocateMemory");
+        PFN_vkGetPhysicalDeviceMemoryProperties vkGetPhysicalDeviceMemoryProperties = (PFN_vkGetPhysicalDeviceMemoryProperties)vv->get_proc_addr(vv->inst, "vkGetPhysicalDeviceMemoryProperties");
+        PFN_vkGetMemoryFdKHR vkGetMemoryFdKHR = (PFN_vkGetMemoryFdKHR)vv->get_proc_addr(vv->inst, "vkGetMemoryFdKHR");
 
-    const unsigned int internal_formats_nv12[2] = { GL_RGBA8, GL_RGBA8 };
-    const unsigned int internal_formats_p010[2] = { GL_R16, GL_RG16 };
-    const unsigned int formats[2] = { GL_RED, GL_RG };
-    const int div[2] = {1, 2}; // divide UV texture size by 2 because chroma is half size
+        VkMemoryRequirements mem_reqs = {0};
+        vkGetImageMemoryRequirements(vv->act_dev, target_surface_id->img[0], &mem_reqs);
+
+        fprintf(stderr, "size: %lu, alignment: %lu, memory bits: 0x%08x\n", mem_reqs.size, mem_reqs.alignment, mem_reqs.memoryTypeBits);
+        VkDeviceMemory mem;
+        {
+            VkExportMemoryAllocateInfo exp_mem_info;
+            VkMemoryAllocateInfo mem_alloc_info;
+            VkMemoryDedicatedAllocateInfoKHR ded_info;
+
+            memset(&exp_mem_info, 0, sizeof(exp_mem_info));
+            exp_mem_info.sType = VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO;
+            exp_mem_info.handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT;
+            
+            memset(&ded_info, 0, sizeof(ded_info));
+            ded_info.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO;
+            ded_info.image = target_surface_id->img[0];
+
+            exp_mem_info.pNext = &ded_info;
+
+            memset(&mem_alloc_info, 0, sizeof(mem_alloc_info));
+            mem_alloc_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
+            mem_alloc_info.pNext = &exp_mem_info;
+            mem_alloc_info.allocationSize = target_surface_id->size[0];
+            mem_alloc_info.memoryTypeIndex = get_memory_type_idx(vv->phys_dev, &mem_reqs, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, vkGetPhysicalDeviceMemoryProperties);
+
+            if (mem_alloc_info.memoryTypeIndex == UINT32_MAX) {
+                fprintf(stderr, "No suitable memory type index found.\n");
+                return VK_NULL_HANDLE;
+            }
+
+            if (vkAllocateMemory(vv->act_dev, &mem_alloc_info, 0, &mem) !=
+                VK_SUCCESS)
+                return VK_NULL_HANDLE;
+ 
+            fprintf(stderr, "memory: %p\n", (void*)mem);
 
-    for(int i = 0; i < 2; ++i) {
-        self->target_textures[i] = gl_create_texture(self->params.egl, video_codec_context->width / div[i], video_codec_context->height / div[i], self->params.color_depth == GSR_COLOR_DEPTH_8_BITS ? internal_formats_nv12[i] : internal_formats_p010[i], formats[i], GL_NEAREST);
-        if(self->target_textures[i] == 0) {
-            fprintf(stderr, "gsr error: gsr_video_encoder_cuda_setup_textures: failed to create opengl texture\n");
-            return false;
         }
-    }
 
-    self->params.egl->glGenBuffers(2, self->pbo_y);
+        fprintf(stderr, "target surface id: %p, %zu, %zu\n", (void*)target_surface_id->mem[0], target_surface_id->offset[0], target_surface_id->offset[1]);
+        fprintf(stderr, "vkGetMemoryFdKHR: %p\n", (void*)vkGetMemoryFdKHR);
 
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_y[0]);
-    self->params.egl->glBufferData(GL_PIXEL_PACK_BUFFER, frame->width * frame->height, 0, GL_STREAM_READ);
+        int fd = 0;
+        VkMemoryGetFdInfoKHR fd_info;
+        memset(&fd_info, 0, sizeof(fd_info));
+        fd_info.sType = VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR;
+        fd_info.memory = target_surface_id->mem[0];
+        fd_info.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT;
+        if(vkGetMemoryFdKHR(vv->act_dev, &fd_info, &fd) != VK_SUCCESS) {
+            fprintf(stderr, "failed!\n");
+        } else {
+            fprintf(stderr, "fd: %d\n", fd);
+        }
 
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_y[1]);
-    self->params.egl->glBufferData(GL_PIXEL_PACK_BUFFER, frame->width * frame->height, 0, GL_STREAM_READ);
+        fprintf(stderr, "glImportMemoryFdEXT: %p, size: %zu\n", (void*)self->params.egl->glImportMemoryFdEXT, target_surface_id->size[0]);
+        const int tiling = target_surface_id->tiling == VK_IMAGE_TILING_LINEAR ? GL_LINEAR_TILING_EXT : GL_OPTIMAL_TILING_EXT;
 
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
+        if(tiling != GL_OPTIMAL_TILING_EXT) {
+            fprintf(stderr, "tiling %d is not supported, only GL_OPTIMAL_TILING_EXT (%d) is supported\n", tiling, GL_OPTIMAL_TILING_EXT);
+        }
 
-    self->params.egl->glGenBuffers(2, self->pbo_uv);
 
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_uv[0]);
-    self->params.egl->glBufferData(GL_PIXEL_PACK_BUFFER, (frame->width/2 * frame->height/2) * 2, 0, GL_STREAM_READ);
+        unsigned int gl_memory_obj = 0;
+        self->params.egl->glCreateMemoryObjectsEXT(1, &gl_memory_obj);
 
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_uv[1]);
-    self->params.egl->glBufferData(GL_PIXEL_PACK_BUFFER, (frame->width/2 * frame->height/2) * 2, 0, GL_STREAM_READ);
+        //const int dedicated = GL_TRUE;
+        //self->params.egl->glMemoryObjectParameterivEXT(gl_memory_obj, GL_DEDICATED_MEMORY_OBJECT_EXT, &dedicated);
 
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
+        self->params.egl->glImportMemoryFdEXT(gl_memory_obj, target_surface_id->size[0], GL_HANDLE_TYPE_OPAQUE_FD_EXT, fd);
+        if(!self->params.egl->glIsMemoryObjectEXT(gl_memory_obj))
+            fprintf(stderr, "failed to create object!\n");
 
-    self->sw_frame = av_frame_alloc();
-    self->sw_frame->format = AV_PIX_FMT_NV12;
-    self->sw_frame->width = frame->width;
-    self->sw_frame->height = frame->height;
+        fprintf(stderr, "gl memory obj: %u, error: %d\n", gl_memory_obj, self->params.egl->glGetError());
 
-    // TODO: Remove
-    if(av_frame_get_buffer(self->sw_frame, 0) < 0) {
-        fprintf(stderr, "failed to allocate sw frame\n");
-    }
+        // fprintf(stderr, "0 gl error: %d\n", self->params.egl->glGetError());
+        // unsigned int vertex_buffer = 0;
+        // self->params.egl->glGenBuffers(1, &vertex_buffer);
+        // self->params.egl->glBindBuffer(GL_ARRAY_BUFFER, vertex_buffer);
+        // self->params.egl->glBufferStorageMemEXT(GL_ARRAY_BUFFER, target_surface_id->size[0], gl_memory_obj, target_surface_id->offset[0]);
+        // fprintf(stderr, "1 gl error: %d\n", self->params.egl->glGetError());
 
-    // TODO: Remove
-    if(av_frame_make_writable(self->sw_frame) < 0) {
-        fprintf(stderr, "failed to make writable\n");
-    }
+        // fprintf(stderr, "0 gl error: %d\n", self->params.egl->glGetError());
+        // unsigned int buffer = 0;
+        // self->params.egl->glCreateBuffers(1, &buffer);
+        // self->params.egl->glNamedBufferStorageMemEXT(buffer, target_surface_id->size[0], gl_memory_obj, target_surface_id->offset[0]);
+        // fprintf(stderr, "1 gl error: %d\n", self->params.egl->glGetError());
+
+        self->params.egl->glGenTextures(1, &self->target_textures[0]);
+        self->params.egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[0]);
+
+        fprintf(stderr, "1 gl error: %d\n", self->params.egl->glGetError());
+        self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_TILING_EXT, tiling);
+
+        fprintf(stderr, "tiling: %d\n", tiling);
+
+        fprintf(stderr, "2 gl error: %d\n", self->params.egl->glGetError());
+        self->params.egl->glTexStorageMem2DEXT(GL_TEXTURE_2D, 1, GL_R8, frame->width, frame->height, gl_memory_obj, target_surface_id->offset[0]);
+
+        fprintf(stderr, "3 gl error: %d\n", self->params.egl->glGetError());
+        self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
+
+        self->params.egl->glGenTextures(1, &self->target_textures[1]);
+        self->params.egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[1]);
+
+        fprintf(stderr, "1 gl error: %d\n", self->params.egl->glGetError());
+        self->params.egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_TILING_EXT, tiling);
+
+        fprintf(stderr, "tiling: %d\n", tiling);
+
+        fprintf(stderr, "2 gl error: %d\n", self->params.egl->glGetError());
+        self->params.egl->glTexStorageMem2DEXT(GL_TEXTURE_2D, 1, GL_RG8, frame->width/2, frame->height/2, gl_memory_obj, target_surface_id->offset[0] + luma_size);
+
+        fprintf(stderr, "3 gl error: %d\n", self->params.egl->glGetError());
+        self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
+
+        self->texture_sizes[0] = (vec2i){ frame->width, frame->height };
+        self->texture_sizes[1] = (vec2i){ frame->width/2, frame->height/2 };
+     }
+#endif
     return true;
 }
 
@@ -185,95 +274,12 @@ void gsr_video_encoder_vulkan_stop(gsr_video_encoder_vulkan *self, AVCodecContex
         av_buffer_unref(&self->device_ctx);
 }
 
-static void nop_free(void *opaque, uint8_t *data) {
-
-}
-
-static void gsr_video_encoder_vulkan_copy_textures_to_frame(gsr_video_encoder *encoder, AVFrame *frame, gsr_color_conversion *color_conversion) {
-    gsr_video_encoder_vulkan *self = encoder->priv;
-
-    static int counter = 0;
-    ++counter;
-
-    // AVBufferRef *av_buffer_create(uint8_t *data, size_t size,
-    //                           void (*free)(void *opaque, uint8_t *data),
-    //                           void *opaque, int flags);
-
-    while(self->params.egl->glGetError()){}
-    self->params.egl->glBindFramebuffer(GL_READ_FRAMEBUFFER, color_conversion->framebuffers[0]);
-    //fprintf(stderr, "1 gl err: %d\n", self->params.egl->glGetError());
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_y[counter % 2]);
-    self->params.egl->glBufferData(GL_PIXEL_PACK_BUFFER, frame->width * frame->height, 0, GL_STREAM_READ);
-    self->params.egl->glReadPixels(0, 0, frame->width, frame->height, GL_RED, GL_UNSIGNED_BYTE, 0);
-    //fprintf(stderr, "2 gl err: %d\n", self->params.egl->glGetError());
-
-    const int next_pbo_y = (counter + 1) % 2;
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_y[next_pbo_y]);
-    self->params.egl->glBufferData(GL_PIXEL_PACK_BUFFER, frame->width * frame->height, 0, GL_STREAM_READ);
-    //fprintf(stderr, "3 gl err: %d\n", self->params.egl->glGetError());
-    uint8_t *ptr_y = (uint8_t*)self->params.egl->glMapBuffer(GL_PIXEL_PACK_BUFFER, GL_READ_ONLY);
-    //fprintf(stderr, "4 gl err: %d\n", self->params.egl->glGetError());
-    if(!ptr_y) {
-        fprintf(stderr, "failed to map buffer y!\n");
-    }
-
-    while(self->params.egl->glGetError()){}
-    self->params.egl->glBindFramebuffer(GL_READ_FRAMEBUFFER, color_conversion->framebuffers[1]);
-    //fprintf(stderr, "5 gl err: %d\n", self->params.egl->glGetError());
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_uv[counter % 2]);
-    self->params.egl->glBufferData(GL_PIXEL_PACK_BUFFER, (frame->width/2 * frame->height/2) * 2, 0, GL_STREAM_READ);
-    //fprintf(stderr, "5.5 gl err: %d\n", self->params.egl->glGetError());
-    self->params.egl->glReadPixels(0, 0, frame->width/2, frame->height/2, GL_RG, GL_UNSIGNED_BYTE, 0);
-    //fprintf(stderr, "6 gl err: %d\n", self->params.egl->glGetError());
-
-    const int next_pbo_uv = (counter + 1) % 2;
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_uv[next_pbo_uv]);
-    self->params.egl->glBufferData(GL_PIXEL_PACK_BUFFER, (frame->width/2 * frame->height/2) * 2, 0, GL_STREAM_READ);
-    //fprintf(stderr, "7 gl err: %d\n", self->params.egl->glGetError());
-    uint8_t *ptr_uv = (uint8_t*)self->params.egl->glMapBuffer(GL_PIXEL_PACK_BUFFER, GL_READ_ONLY);
-    //fprintf(stderr, "8 gl err: %d\n", self->params.egl->glGetError());
-    if(!ptr_uv) {
-        fprintf(stderr, "failed to map buffer uv!\n");
-    }
-
-    //self->sw_frame->buf[0] = av_buffer_create(ptr_y, 3840 * 2160, nop_free, NULL, 0);
-    //self->sw_frame->buf[1] = av_buffer_create(ptr_uv, 1920 * 1080 * 2, nop_free, NULL, 0);
-    //self->sw_frame->data[0] = self->sw_frame->buf[0]->data;
-    //self->sw_frame->data[1] = self->sw_frame->buf[1]->data;
-    //self->sw_frame->extended_data[0] = self->sw_frame->data[0];
-    //self->sw_frame->extended_data[1] = self->sw_frame->data[1];
-
-    self->sw_frame->data[0] = ptr_y;
-    self->sw_frame->data[1] = ptr_uv;
-
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
-    self->params.egl->glBindFramebuffer(GL_READ_FRAMEBUFFER, 0);
-
-    //self->params.egl->glBindTexture(GL_TEXTURE_2D, self->target_textures[1]);
-    //self->params.egl->glGetTexImage(GL_TEXTURE_2D, 0, GL_RG, GL_UNSIGNED_BYTE, sw_frame->data[1]);
-
-    //self->params.egl->glBindTexture(GL_TEXTURE_2D, 0);
-
-    int ret = av_hwframe_transfer_data(frame, self->sw_frame, 0);
-    if(ret < 0) {
-        fprintf(stderr, "transfer data failed, error: %s\n", av_err2str(ret));
-    }
-
-    //av_buffer_unref(&self->sw_frame->buf[0]);
-    //av_buffer_unref(&self->sw_frame->buf[1]);
-
-    //av_frame_free(&sw_frame);
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_y[next_pbo_y]);
-    self->params.egl->glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, self->pbo_y[next_pbo_uv]);
-    self->params.egl->glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
-    self->params.egl->glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
-}
-
-static void gsr_video_encoder_vulkan_get_textures(gsr_video_encoder *encoder, unsigned int *textures, int *num_textures, gsr_destination_color *destination_color) {
+static void gsr_video_encoder_vulkan_get_textures(gsr_video_encoder *encoder, unsigned int *textures, vec2i *texture_sizes, int *num_textures, gsr_destination_color *destination_color) {
     gsr_video_encoder_vulkan *self = encoder->priv;
     textures[0] = self->target_textures[0];
     textures[1] = self->target_textures[1];
+    texture_sizes[0] = self->texture_sizes[0];
+    texture_sizes[1] = self->texture_sizes[1];
     *num_textures = 2;
     *destination_color = self->params.color_depth == GSR_COLOR_DEPTH_10_BITS ? GSR_DESTINATION_COLOR_P010 : GSR_DESTINATION_COLOR_NV12;
 }
@@ -299,7 +305,7 @@ gsr_video_encoder* gsr_video_encoder_vulkan_create(const gsr_video_encoder_vulka
 
     *encoder = (gsr_video_encoder) {
         .start = gsr_video_encoder_vulkan_start,
-        .copy_textures_to_frame = gsr_video_encoder_vulkan_copy_textures_to_frame,
+        .copy_textures_to_frame = NULL,
         .get_textures = gsr_video_encoder_vulkan_get_textures,
         .destroy = gsr_video_encoder_vulkan_destroy,
         .priv = encoder_vulkan

src/image_writer.c

@@ -13,11 +13,10 @@
 /* TODO: Support hdr/10-bit */
 bool gsr_image_writer_init_opengl(gsr_image_writer *self, gsr_egl *egl, int width, int height) {
     memset(self, 0, sizeof(*self));
-    self->source = GSR_IMAGE_WRITER_SOURCE_OPENGL;
     self->egl = egl;
     self->width = width;
     self->height = height;
-    self->texture = gl_create_texture(self->egl, self->width, self->height, GL_RGB8, GL_RGB, GL_NEAREST); /* TODO: use GL_RGB16 instead of GL_RGB8 for hdr/10-bit */
+    self->texture = gl_create_texture(self->egl, self->width, self->height, GL_RGBA8, GL_RGBA, GL_NEAREST); /* TODO: use GL_RGB16 instead of GL_RGB8 for hdr/10-bit */
     if(self->texture == 0) {
         fprintf(stderr, "gsr error: gsr_image_writer_init: failed to create texture\n");
         return false;
@@ -25,15 +24,6 @@ bool gsr_image_writer_init_opengl(gsr_image_writer *self, gsr_egl *egl, int widt
     return true;
 }
 
-bool gsr_image_writer_init_memory(gsr_image_writer *self, const void *memory, int width, int height) {
-    memset(self, 0, sizeof(*self));
-    self->source = GSR_IMAGE_WRITER_SOURCE_OPENGL;
-    self->width = width;
-    self->height = height;
-    self->memory = memory;
-    return true;
-}
-
 void gsr_image_writer_deinit(gsr_image_writer *self) {
     if(self->texture) {
         self->egl->glDeleteTextures(1, &self->texture);
@@ -50,10 +40,10 @@ static bool gsr_image_writer_write_memory_to_file(gsr_image_writer *self, const
     bool success = false;
     switch(image_format) {
         case GSR_IMAGE_FORMAT_JPEG:
-            success = stbi_write_jpg(filepath, self->width, self->height, 3, data, quality);
+            success = stbi_write_jpg(filepath, self->width, self->height, 4, data, quality);
             break;
         case GSR_IMAGE_FORMAT_PNG:
-            success = stbi_write_png(filepath, self->width, self->height, 3, data, 0);
+            success = stbi_write_png(filepath, self->width, self->height, 4, data, 0);
             break;
     }
 
@@ -64,18 +54,21 @@ static bool gsr_image_writer_write_memory_to_file(gsr_image_writer *self, const
 }
 
 static bool gsr_image_writer_write_opengl_texture_to_file(gsr_image_writer *self, const char *filepath, gsr_image_format image_format, int quality) {
-    assert(self->source == GSR_IMAGE_WRITER_SOURCE_OPENGL);
-    uint8_t *frame_data = malloc(self->width * self->height * 3);
+    uint8_t *frame_data = malloc(self->width * self->height * 4);
     if(!frame_data) {
         fprintf(stderr, "gsr error: gsr_image_writer_write_to_file: failed to allocate memory for image frame\n");
         return false;
     }
 
-    // TODO: hdr support
-    self->egl->glBindTexture(GL_TEXTURE_2D, self->texture);
-    // We could use glGetTexSubImage, but it's only available starting from opengl 4.5
-    self->egl->glGetTexImage(GL_TEXTURE_2D, 0, GL_RGB, GL_UNSIGNED_BYTE, frame_data);
-    self->egl->glBindTexture(GL_TEXTURE_2D, 0);
+    unsigned int fbo = 0;
+    self->egl->glGenFramebuffers(1, &fbo);
+    self->egl->glBindFramebuffer(GL_FRAMEBUFFER, fbo);
+    self->egl->glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, self->texture, 0);
+
+    self->egl->glReadPixels(0, 0, self->width, self->height, GL_RGBA, GL_UNSIGNED_BYTE, frame_data);
+
+    self->egl->glBindFramebuffer(GL_FRAMEBUFFER, 0);
+    self->egl->glDeleteFramebuffers(1, &fbo);
 
     self->egl->glFlush();
     self->egl->glFinish();
@@ -86,11 +79,5 @@ static bool gsr_image_writer_write_opengl_texture_to_file(gsr_image_writer *self
 }
 
 bool gsr_image_writer_write_to_file(gsr_image_writer *self, const char *filepath, gsr_image_format image_format, int quality) {
-    switch(self->source) {
-        case GSR_IMAGE_WRITER_SOURCE_OPENGL:
-            return gsr_image_writer_write_opengl_texture_to_file(self, filepath, image_format, quality);
-        case GSR_IMAGE_WRITER_SOURCE_MEMORY:
-            return gsr_image_writer_write_memory_to_file(self, filepath, image_format, quality, self->memory);
-    }
-    return false;
+    return gsr_image_writer_write_opengl_texture_to_file(self, filepath, image_format, quality);
 }

src/overclock.c

@@ -246,7 +246,7 @@ bool gsr_overclock_start(gsr_overclock *self) {
         const int transfer_rate_max_diff = query.performance_level[query.num_performance_levels - 1].mem_transfer_rate_max - query.performance_level[query.num_performance_levels - 2].mem_transfer_rate_max;
         target_transfer_rate_offset = min_int(target_transfer_rate_offset, transfer_rate_max_diff);
         if(target_transfer_rate_offset >= 0 && xnvctrl_set_attribute_offset(&self->xnvctrl, self->num_performance_levels, target_transfer_rate_offset, NVCTRL_ATTRIB_GPU_MEM_TRANSFER_RATE)) {
-            fprintf(stderr, "gsr info: gsr_overclock_start: sucessfully set memory transfer rate offset to %d\n", target_transfer_rate_offset);
+            fprintf(stderr, "gsr info: gsr_overclock_start: successfully set memory transfer rate offset to %d\n", target_transfer_rate_offset);
         } else {
             fprintf(stderr, "gsr info: gsr_overclock_start: failed to overclock memory transfer rate offset to %d\n", target_transfer_rate_offset);
         }
@@ -261,7 +261,7 @@ bool gsr_overclock_start(gsr_overclock *self) {
         const int nv_clock_max_diff = query.performance_level[query.num_performance_levels - 1].nv_clock_max - query.performance_level[query.num_performance_levels - 2].nv_clock_max;
         target_nv_clock_offset = min_int(target_nv_clock_offset, nv_clock_max_diff);
         if(target_nv_clock_offset >= 0 && xnvctrl_set_attribute_offset(&self->xnvctrl, self->num_performance_levels, target_nv_clock_offset, NVCTRL_GPU_NVCLOCK)) {
-            fprintf(stderr, "gsr info: gsr_overclock_start: sucessfully set nv clock offset to %d\n", target_nv_clock_offset);
+            fprintf(stderr, "gsr info: gsr_overclock_start: successfully set nv clock offset to %d\n", target_nv_clock_offset);
         } else {
             fprintf(stderr, "gsr info: gsr_overclock_start: failed to overclock nv clock offset to %d\n", target_nv_clock_offset);
         }

src/pipewire_audio.c

@@ -4,6 +4,11 @@
 #include <pipewire/extensions/metadata.h>
 #include <pipewire/impl-module.h>
 
+typedef struct {
+    const gsr_pipewire_audio_port *output_port;
+    const gsr_pipewire_audio_port *input_port;
+} gsr_pipewire_audio_desired_link;
+
 static void on_core_info_cb(void *user_data, const struct pw_core_info *info) {
     gsr_pipewire_audio *self = user_data;
     //fprintf(stderr, "server name: %s\n", info->name);
@@ -29,7 +34,7 @@ static const struct pw_core_events core_events = {
 };
 
 static gsr_pipewire_audio_node* gsr_pipewire_audio_get_node_by_name_case_insensitive(gsr_pipewire_audio *self, const char *node_name, gsr_pipewire_audio_node_type node_type) {
-    for(int i = 0; i < self->num_stream_nodes; ++i) {
+    for(size_t i = 0; i < self->num_stream_nodes; ++i) {
         const gsr_pipewire_audio_node *node = &self->stream_nodes[i];
         if(node->type == node_type && strcasecmp(node->name, node_name) == 0)
             return &self->stream_nodes[i];
@@ -38,7 +43,7 @@ static gsr_pipewire_audio_node* gsr_pipewire_audio_get_node_by_name_case_insensi
 }
 
 static gsr_pipewire_audio_port* gsr_pipewire_audio_get_node_port_by_name(gsr_pipewire_audio *self, uint32_t node_id, const char *port_name) {
-    for(int i = 0; i < self->num_ports; ++i) {
+    for(size_t i = 0; i < self->num_ports; ++i) {
         if(self->ports[i].node_id == node_id && strcmp(self->ports[i].name, port_name) == 0)
             return &self->ports[i];
     }
@@ -53,6 +58,29 @@ static bool requested_link_matches_name_case_insensitive(const gsr_pipewire_audi
     return false;
 }
 
+static bool requested_link_matches_name_case_insensitive_any_type(const gsr_pipewire_audio *self, const gsr_pipewire_audio_requested_link *requested_link, const char *name) {
+    for(int i = 0; i < requested_link->num_outputs; ++i) {
+        switch(requested_link->outputs[i].type) {
+            case GSR_PIPEWIRE_AUDIO_REQUESTED_TYPE_STANDARD: {
+                if(strcasecmp(requested_link->outputs[i].name, name) == 0)
+                    return true;
+                break;
+            }
+            case GSR_PIPEWIRE_AUDIO_REQUESTED_TYPE_DEFAULT_OUTPUT: {
+                if(strcasecmp(self->default_output_device_name, name) == 0)
+                    return true;
+                break;
+            }
+            case GSR_PIPEWIRE_AUDIO_REQUESTED_TYPE_DEFAULT_INPUT: {
+                if(strcasecmp(self->default_input_device_name, name) == 0)
+                    return true;
+                break;
+            }
+        }
+    }
+    return false;
+}
+
 static bool requested_link_has_type(const gsr_pipewire_audio_requested_link *requested_link, gsr_pipewire_audio_requested_type type) {
     for(int i = 0; i < requested_link->num_outputs; ++i) {
         if(requested_link->outputs[i].type == type)
@@ -81,69 +109,68 @@ static void gsr_pipewire_get_node_input_port_by_type(gsr_pipewire_audio *self, c
     }
 }
 
-static void gsr_pipewire_get_node_output_port_by_type(gsr_pipewire_audio *self, const gsr_pipewire_audio_node *output_node, gsr_pipewire_audio_node_type output_type,
-    const gsr_pipewire_audio_port **output_fl_port, const gsr_pipewire_audio_port **output_fr_port)
-{
-    *output_fl_port = NULL;
-    *output_fr_port = NULL;
-
-    switch(output_type) {
-        case GSR_PIPEWIRE_AUDIO_NODE_TYPE_STREAM_OUTPUT:
-            *output_fl_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "output_FL");
-            *output_fr_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "output_FR");
-            break;
-        case GSR_PIPEWIRE_AUDIO_NODE_TYPE_STREAM_INPUT:
-            *output_fl_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "monitor_FL");
-            *output_fr_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "monitor_FR");
-            break;
-        case GSR_PIPEWIRE_AUDIO_NODE_TYPE_SINK_OR_SOURCE: {
-            *output_fl_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "monitor_FL");
-            *output_fr_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "monitor_FR");
-            if(!*output_fl_port || !*output_fr_port) {
-                *output_fl_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "capture_FL");
-                *output_fr_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "capture_FR");
-            }
-            if(!*output_fl_port || !*output_fr_port) {
-                const gsr_pipewire_audio_port *output_mono_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "monitor_MONO");
-                if(!output_mono_port)
-                    output_mono_port = gsr_pipewire_audio_get_node_port_by_name(self, output_node->id, "capture_MONO");
-
-                if(output_mono_port) {
-                    *output_fl_port = output_mono_port;
-                    *output_fr_port = output_mono_port;
-                }
-            }
-            break;
-        }
-    }
+static bool string_starts_with(const char *str, const char *substr) {
+    const int len = strlen(str);
+    const int substr_len = strlen(substr);
+    return len >= substr_len && memcmp(str, substr, substr_len) == 0;
 }
 
-static void gsr_pipewire_audio_establish_link(gsr_pipewire_audio *self, const gsr_pipewire_audio_port *input_fl_port, const gsr_pipewire_audio_port *input_fr_port,
-    const gsr_pipewire_audio_port *output_fl_port, const gsr_pipewire_audio_port *output_fr_port)
+static bool string_ends_with(const char *str, const char *substr) {
+    const int len = strlen(str);
+    const int substr_len = strlen(substr);
+    return len >= substr_len && memcmp(str + len - substr_len, substr, substr_len) == 0;
+}
+
+/* Returns number of desired links */
+static size_t gsr_pipewire_get_node_output_ports(gsr_pipewire_audio *self, const gsr_pipewire_audio_node *output_node,
+    gsr_pipewire_audio_desired_link *desired_links, size_t desired_links_max_size,
+    const gsr_pipewire_audio_port *input_fl_port, const gsr_pipewire_audio_port *input_fr_port)
 {
-    // TODO: Detect if link already exists before so we dont create these proxies when not needed
+    size_t num_desired_links = 0;
+    for(size_t i = 0; i < self->num_ports && num_desired_links < desired_links_max_size; ++i) {
+        if(self->ports[i].node_id != output_node->id)
+            continue;
+
+        if(string_starts_with(self->ports[i].name, "playback_"))
+            continue;
+
+        if(string_ends_with(self->ports[i].name, "_MONO") || string_ends_with(self->ports[i].name, "_FC") || string_ends_with(self->ports[i].name, "_LFE")) {
+            if(num_desired_links + 2 >= desired_links_max_size)
+                break;
+
+            desired_links[num_desired_links + 0] = (gsr_pipewire_audio_desired_link){ .output_port = &self->ports[i], .input_port = input_fl_port };
+            desired_links[num_desired_links + 1] = (gsr_pipewire_audio_desired_link){ .output_port = &self->ports[i], .input_port = input_fr_port };
+            num_desired_links += 2;
+        } else if(string_ends_with(self->ports[i].name, "_FL") || string_ends_with(self->ports[i].name, "_RL") || string_ends_with(self->ports[i].name, "_SL")) {
+            if(num_desired_links + 1 >= desired_links_max_size)
+                break;
+
+            desired_links[num_desired_links] = (gsr_pipewire_audio_desired_link){ .output_port = &self->ports[i], .input_port = input_fl_port };
+            num_desired_links += 1;
+        } else if(string_ends_with(self->ports[i].name, "_FR") || string_ends_with(self->ports[i].name, "_RR") || string_ends_with(self->ports[i].name, "_SR")) {
+            if(num_desired_links + 1 >= desired_links_max_size)
+                break;
+
+            desired_links[num_desired_links] = (gsr_pipewire_audio_desired_link){ .output_port = &self->ports[i], .input_port = input_fr_port };
+            num_desired_links += 1;
+        }
+    }
+    return num_desired_links;
+}
+
+static void gsr_pipewire_audio_establish_link(gsr_pipewire_audio *self, const gsr_pipewire_audio_port *output_port, const gsr_pipewire_audio_port *input_port) {
+    // TODO: Detect if link already exists before so we dont create these proxies when not needed.
+    // We could do that by saving which nodes have been linked with which nodes after linking them.
 
     //fprintf(stderr, "linking!\n");
     // TODO: error check and cleanup
-    {
-        struct pw_properties *props = pw_properties_new(NULL, NULL);
-        pw_properties_setf(props, PW_KEY_LINK_OUTPUT_PORT, "%u", output_fl_port->id);
-        pw_properties_setf(props, PW_KEY_LINK_INPUT_PORT, "%u", input_fl_port->id);
-        // TODO: Clean this up when removing node
-        struct pw_proxy *proxy = pw_core_create_object(self->core, "link-factory", PW_TYPE_INTERFACE_Link, PW_VERSION_LINK, &props->dict, 0);
-        //self->server_version_sync = pw_core_sync(self->core, PW_ID_CORE, self->server_version_sync);
-        pw_properties_free(props);
-    }
-
-    {
-        struct pw_properties *props = pw_properties_new(NULL, NULL);
-        pw_properties_setf(props, PW_KEY_LINK_OUTPUT_PORT, "%u", output_fr_port->id);
-        pw_properties_setf(props, PW_KEY_LINK_INPUT_PORT, "%u", input_fr_port->id);
-        // TODO: Clean this up when removing node
-        struct pw_proxy *proxy = pw_core_create_object(self->core, "link-factory", PW_TYPE_INTERFACE_Link, PW_VERSION_LINK, &props->dict, 0);
-        //self->server_version_sync = pw_core_sync(self->core, PW_ID_CORE, self->server_version_sync);
-        pw_properties_free(props);
-    }
+    struct pw_properties *props = pw_properties_new(NULL, NULL);
+    pw_properties_setf(props, PW_KEY_LINK_OUTPUT_PORT, "%u", output_port->id);
+    pw_properties_setf(props, PW_KEY_LINK_INPUT_PORT, "%u", input_port->id);
+    // TODO: Clean this up when removing node
+    struct pw_proxy *proxy = pw_core_create_object(self->core, "link-factory", PW_TYPE_INTERFACE_Link, PW_VERSION_LINK, &props->dict, 0);
+    //self->server_version_sync = pw_core_sync(self->core, PW_ID_CORE, self->server_version_sync);
+    pw_properties_free(props);
 }
 
 static void gsr_pipewire_audio_create_link(gsr_pipewire_audio *self, const gsr_pipewire_audio_requested_link *requested_link) {
@@ -158,12 +185,13 @@ static void gsr_pipewire_audio_create_link(gsr_pipewire_audio *self, const gsr_p
     if(!input_fl_port || !input_fr_port)
         return;
 
-    for(int i = 0; i < self->num_stream_nodes; ++i) {
+    gsr_pipewire_audio_desired_link desired_links[64];
+    for(size_t i = 0; i < self->num_stream_nodes; ++i) {
         const gsr_pipewire_audio_node *output_node = &self->stream_nodes[i];
         if(output_node->type != requested_link->output_type)
             continue;
 
-        const bool requested_link_matches_app = requested_link_matches_name_case_insensitive(requested_link, output_node->name);
+        const bool requested_link_matches_app = requested_link_matches_name_case_insensitive_any_type(self, requested_link, output_node->name);
         if(requested_link->inverted) {
             if(requested_link_matches_app)
                 continue;
@@ -172,18 +200,15 @@ static void gsr_pipewire_audio_create_link(gsr_pipewire_audio *self, const gsr_p
                 continue;
         }
 
-        const gsr_pipewire_audio_port *output_fl_port = NULL;
-        const gsr_pipewire_audio_port *output_fr_port = NULL;
-        gsr_pipewire_get_node_output_port_by_type(self, output_node, requested_link->output_type, &output_fl_port, &output_fr_port);
-        if(!output_fl_port || !output_fr_port)
-            continue;
-
-        gsr_pipewire_audio_establish_link(self, input_fl_port, input_fr_port, output_fl_port, output_fr_port);
+        const size_t num_desired_links = gsr_pipewire_get_node_output_ports(self, output_node, desired_links, 64, input_fl_port, input_fr_port);
+        for(size_t j = 0; j < num_desired_links; ++j) {
+            gsr_pipewire_audio_establish_link(self, desired_links[j].output_port, desired_links[j].input_port);
+        }
     }
 }
 
 static void gsr_pipewire_audio_create_links(gsr_pipewire_audio *self) {
-    for(int i = 0; i < self->num_requested_links; ++i) {
+    for(size_t i = 0; i < self->num_requested_links; ++i) {
         gsr_pipewire_audio_create_link(self, &self->requested_links[i]);
     }
 }
@@ -214,24 +239,21 @@ static void gsr_pipewire_audio_create_link_for_default_devices(gsr_pipewire_audi
     if(!stream_output_node)
         return;
 
-    const gsr_pipewire_audio_port *output_fl_port = NULL;
-    const gsr_pipewire_audio_port *output_fr_port = NULL;
-    gsr_pipewire_get_node_output_port_by_type(self, stream_output_node, requested_link->output_type, &output_fl_port, &output_fr_port);
-    if(!output_fl_port || !output_fr_port)
-        return;
-
-    gsr_pipewire_audio_establish_link(self, input_fl_port, input_fr_port, output_fl_port, output_fr_port);
-    //fprintf(stderr, "establishing a link from %u to %u\n", stream_output_node->id, stream_input_node->id);
+    gsr_pipewire_audio_desired_link desired_links[64];
+    const size_t num_desired_links = gsr_pipewire_get_node_output_ports(self, stream_output_node, desired_links, 64, input_fl_port, input_fr_port);
+    for(size_t i = 0; i < num_desired_links; ++i) {
+        gsr_pipewire_audio_establish_link(self, desired_links[i].output_port, desired_links[i].input_port);
+    }
 }
 
 static void gsr_pipewire_audio_create_links_for_default_devices(gsr_pipewire_audio *self, gsr_pipewire_audio_requested_type default_device_type) {
-    for(int i = 0; i < self->num_requested_links; ++i) {
+    for(size_t i = 0; i < self->num_requested_links; ++i) {
         gsr_pipewire_audio_create_link_for_default_devices(self, &self->requested_links[i], default_device_type);
     }
 }
 
 static void gsr_pipewire_audio_destroy_links_by_output_to_input(gsr_pipewire_audio *self, uint32_t output_node_id, uint32_t input_node_id) {
-    for(int i = 0; i < self->num_links; ++i) {
+    for(size_t i = 0; i < self->num_links; ++i) {
         if(self->links[i].output_node_id == output_node_id && self->links[i].input_node_id == input_node_id)
             pw_registry_destroy(self->registry, self->links[i].id);
     }
@@ -271,7 +293,7 @@ static void gsr_pipewire_destroy_default_device_link(gsr_pipewire_audio *self, c
 }
 
 static void gsr_pipewire_destroy_default_device_links(gsr_pipewire_audio *self, gsr_pipewire_audio_requested_type default_device_type) {
-    for(int i = 0; i < self->num_requested_links; ++i) {
+    for(size_t i = 0; i < self->num_requested_links; ++i) {
         gsr_pipewire_destroy_default_device_link(self, &self->requested_links[i], default_device_type);
     }
 }
@@ -370,30 +392,51 @@ static bool gsr_pipewire_audio_listen_on_metadata(gsr_pipewire_audio *self, uint
     return true;
 }
 
+static bool array_ensure_capacity(void **array, size_t size, size_t *capacity_items, size_t element_size) {
+    if(size + 1 >= *capacity_items) {
+        size_t new_capacity_items = *capacity_items * 2;
+        if(new_capacity_items == 0)
+            new_capacity_items = 32;
+
+        void *new_data = realloc(*array, new_capacity_items * element_size);
+        if(!new_data) {
+            fprintf(stderr, "gsr error: pipewire_audio: failed to reallocate memory\n");
+            return false;
+        }
+
+        *array = new_data;
+        *capacity_items = new_capacity_items;
+    }
+    return true;
+}
+
 static void registry_event_global(void *data, uint32_t id, uint32_t permissions,
                   const char *type, uint32_t version,
                   const struct spa_dict *props)
 {
     //fprintf(stderr, "add: id: %d, type: %s\n", (int)id, type);
-    if(!props || !type)
+    gsr_pipewire_audio *self = (gsr_pipewire_audio*)data;
+    if(!props || !type || !self->running)
         return;
 
     //pw_properties_new_dict(props);
 
-    gsr_pipewire_audio *self = (gsr_pipewire_audio*)data;
     if(strcmp(type, PW_TYPE_INTERFACE_Node) == 0) {
         const char *node_name = spa_dict_lookup(props, PW_KEY_NODE_NAME);
         const char *media_class = spa_dict_lookup(props, PW_KEY_MEDIA_CLASS);
         //fprintf(stderr, "  node id: %u, node name: %s, media class: %s\n", id, node_name, media_class);
         const bool is_stream_output = media_class && strcmp(media_class, "Stream/Output/Audio") == 0;
         const bool is_stream_input = media_class && strcmp(media_class, "Stream/Input/Audio") == 0;
-        const bool is_sink = media_class && strcmp(media_class, "Audio/Sink") == 0;
-        const bool is_source = media_class && strcmp(media_class, "Audio/Source") == 0;
-        if(self->num_stream_nodes < GSR_PIPEWIRE_AUDIO_MAX_STREAM_NODES && node_name && (is_stream_output || is_stream_input || is_sink || is_source)) {
+        const bool is_sink = media_class && string_starts_with(media_class, "Audio/Sink"); // Matches Audio/Sink/Virtual as well
+        const bool is_source = media_class && string_starts_with(media_class, "Audio/Source"); // Matches Audio/Source/Virtual as well
+        if(node_name && (is_stream_output || is_stream_input || is_sink || is_source)) {
             //const char *application_binary = spa_dict_lookup(props, PW_KEY_APP_PROCESS_BINARY);
             //const char *application_name = spa_dict_lookup(props, PW_KEY_APP_NAME);
             //fprintf(stderr, "  node name: %s, app binary: %s, app name: %s\n", node_name, application_binary, application_name);
 
+            if(!array_ensure_capacity((void**)&self->stream_nodes, self->num_stream_nodes, &self->stream_nodes_capacity_items, sizeof(gsr_pipewire_audio_node)))
+                return;
+
             char *node_name_copy = strdup(node_name);
             if(node_name_copy) {
                 self->stream_nodes[self->num_stream_nodes].id = id;
@@ -408,8 +451,6 @@ static void registry_event_global(void *data, uint32_t id, uint32_t permissions,
 
                 gsr_pipewire_audio_create_links(self);
             }
-        } else if(self->num_stream_nodes >= GSR_PIPEWIRE_AUDIO_MAX_STREAM_NODES) {
-            fprintf(stderr, "gsr error: reached the maximum amount of audio stream nodes\n");
         }
     } else if(strcmp(type, PW_TYPE_INTERFACE_Port) == 0) {
         const char *port_name = spa_dict_lookup(props, PW_KEY_PORT_NAME);
@@ -424,7 +465,10 @@ static void registry_event_global(void *data, uint32_t id, uint32_t permissions,
         const char *node_id = spa_dict_lookup(props, PW_KEY_NODE_ID);
         const int node_id_num = node_id ? atoi(node_id) : 0;
 
-        if(self->num_ports < GSR_PIPEWIRE_AUDIO_MAX_PORTS && port_name && direction >= 0 && node_id_num > 0) {
+        if(port_name && direction >= 0 && node_id_num > 0) {
+            if(!array_ensure_capacity((void**)&self->ports, self->num_ports, &self->ports_capacity_items, sizeof(gsr_pipewire_audio_port)))
+                return;
+
             //fprintf(stderr, "  port name: %s, node id: %d, direction: %s\n", port_name, node_id_num, port_direction);
             char *port_name_copy = strdup(port_name);
             if(port_name_copy) {
@@ -437,8 +481,6 @@ static void registry_event_global(void *data, uint32_t id, uint32_t permissions,
 
                 gsr_pipewire_audio_create_links(self);
             }
-        } else if(self->num_ports >= GSR_PIPEWIRE_AUDIO_MAX_PORTS) {
-            fprintf(stderr, "gsr error: reached the maximum amount of audio ports\n");
         }
     } else if(strcmp(type, PW_TYPE_INTERFACE_Link) == 0) {
         const char *output_node = spa_dict_lookup(props, PW_KEY_LINK_OUTPUT_NODE);
@@ -446,14 +488,15 @@ static void registry_event_global(void *data, uint32_t id, uint32_t permissions,
 
         const uint32_t output_node_id_num = output_node ? atoi(output_node) : 0;
         const uint32_t input_node_id_num = input_node ? atoi(input_node) : 0;
-        if(self->num_links < GSR_PIPEWIRE_AUDIO_MAX_LINKS && output_node_id_num > 0 && input_node_id_num > 0) {
+        if(output_node_id_num > 0 && input_node_id_num > 0) {
+            if(!array_ensure_capacity((void**)&self->links, self->num_links, &self->links_capacity_items, sizeof(gsr_pipewire_audio_link)))
+                return;
+
             //fprintf(stderr, "  new link (%u): %u -> %u\n", id, output_node_id_num, input_node_id_num);
             self->links[self->num_links].id = id;
             self->links[self->num_links].output_node_id = output_node_id_num;
             self->links[self->num_links].input_node_id = input_node_id_num;
             ++self->num_links;
-        } else if(self->num_ports >= GSR_PIPEWIRE_AUDIO_MAX_LINKS) {
-            fprintf(stderr, "gsr error: reached the maximum amount of audio links\n");
         }
     } else if(strcmp(type, PW_TYPE_INTERFACE_Metadata) == 0) {
         const char *name = spa_dict_lookup(props, PW_KEY_METADATA_NAME);
@@ -463,7 +506,7 @@ static void registry_event_global(void *data, uint32_t id, uint32_t permissions,
 }
 
 static bool gsr_pipewire_audio_remove_node_by_id(gsr_pipewire_audio *self, uint32_t node_id) {
-    for(int i = 0; i < self->num_stream_nodes; ++i) {
+    for(size_t i = 0; i < self->num_stream_nodes; ++i) {
         if(self->stream_nodes[i].id != node_id)
             continue;
 
@@ -476,7 +519,7 @@ static bool gsr_pipewire_audio_remove_node_by_id(gsr_pipewire_audio *self, uint3
 }
 
 static bool gsr_pipewire_audio_remove_port_by_id(gsr_pipewire_audio *self, uint32_t port_id) {
-    for(int i = 0; i < self->num_ports; ++i) {
+    for(size_t i = 0; i < self->num_ports; ++i) {
         if(self->ports[i].id != port_id)
             continue;
 
@@ -489,7 +532,7 @@ static bool gsr_pipewire_audio_remove_port_by_id(gsr_pipewire_audio *self, uint3
 }
 
 static bool gsr_pipewire_audio_remove_link_by_id(gsr_pipewire_audio *self, uint32_t link_id) {
-    for(int i = 0; i < self->num_links; ++i) {
+    for(size_t i = 0; i < self->num_links; ++i) {
         if(self->links[i].id != link_id)
             continue;
 
@@ -527,6 +570,7 @@ static const struct pw_registry_events registry_events = {
 
 bool gsr_pipewire_audio_init(gsr_pipewire_audio *self) {
     memset(self, 0, sizeof(*self));
+    self->running = true;
 
     pw_init(NULL, NULL);
     
@@ -574,20 +618,53 @@ bool gsr_pipewire_audio_init(gsr_pipewire_audio *self) {
     return true;
 }
 
+static gsr_pipewire_audio_link* gsr_pipewire_audio_get_first_link_to_node(gsr_pipewire_audio *self, uint32_t node_id) {
+    for(size_t i = 0; i < self->num_links; ++i) {
+        if(self->links[i].input_node_id == node_id)
+            return &self->links[i];
+    }
+    return NULL;
+}
+
+static void gsr_pipewire_audio_destroy_requested_links(gsr_pipewire_audio *self) {
+    pw_thread_loop_lock(self->thread_loop);
+
+    self->server_version_sync = pw_core_sync(self->core, PW_ID_CORE, self->server_version_sync);
+    pw_thread_loop_wait(self->thread_loop);
+
+    for(size_t requested_link_index = 0; requested_link_index < self->num_requested_links; ++requested_link_index) {
+        const gsr_pipewire_audio_node_type requested_link_node_type = self->requested_links[requested_link_index].input_type == GSR_PIPEWIRE_AUDIO_LINK_INPUT_TYPE_STREAM ? GSR_PIPEWIRE_AUDIO_NODE_TYPE_STREAM_INPUT : GSR_PIPEWIRE_AUDIO_NODE_TYPE_SINK_OR_SOURCE;
+        const gsr_pipewire_audio_node *stream_input_node = gsr_pipewire_audio_get_node_by_name_case_insensitive(self, self->requested_links[requested_link_index].input_name, requested_link_node_type);
+        if(!stream_input_node)
+            continue;
+
+        for(;;) {
+            gsr_pipewire_audio_link *link = gsr_pipewire_audio_get_first_link_to_node(self, stream_input_node->id);
+            if(!link)
+                break;
+
+            pw_registry_destroy(self->registry, link->id);
+
+            self->server_version_sync = pw_core_sync(self->core, PW_ID_CORE, self->server_version_sync);
+            pw_thread_loop_wait(self->thread_loop);
+
+            usleep(10 * 1000);
+        }
+    }
+
+    pw_thread_loop_unlock(self->thread_loop);
+}
+
 void gsr_pipewire_audio_deinit(gsr_pipewire_audio *self) {
+    self->running = false;
+
     if(self->thread_loop) {
+        /* We need to manually destroy links first, otherwise the linked audio sources will be paused when closing the program */
+        gsr_pipewire_audio_destroy_requested_links(self);
         //pw_thread_loop_wait(self->thread_loop);
         pw_thread_loop_stop(self->thread_loop);
     }
 
-    for(int i = 0; i < self->num_virtual_sink_proxies; ++i) {
-        if(self->virtual_sink_proxies[i]) {
-            pw_proxy_destroy(self->virtual_sink_proxies[i]);
-            self->virtual_sink_proxies[i] = NULL;
-        }
-    }
-    self->num_virtual_sink_proxies = 0;
-
     if(self->metadata_proxy) {
         spa_hook_remove(&self->metadata_listener);
         spa_hook_remove(&self->metadata_proxy_listener);
@@ -615,80 +692,56 @@ void gsr_pipewire_audio_deinit(gsr_pipewire_audio *self) {
         self->thread_loop = NULL;
     }
 
-    for(int i = 0; i < self->num_stream_nodes; ++i) {
-        free(self->stream_nodes[i].name);
-    }
-    self->num_stream_nodes = 0;
-
-    for(int i = 0; i < self->num_ports; ++i) {
-        free(self->ports[i].name);
-    }
-    self->num_ports = 0;
-
-    self->num_links = 0;
-
-    for(int i = 0; i < self->num_requested_links; ++i) {
-        for(int j = 0; j < self->requested_links[i].num_outputs; ++j) {
-            free(self->requested_links[i].outputs[j].name);
+    if(self->stream_nodes) {
+        for(size_t i = 0; i < self->num_stream_nodes; ++i) {
+            free(self->stream_nodes[i].name);
         }
-        free(self->requested_links[i].outputs);
-        free(self->requested_links[i].input_name);
+        self->num_stream_nodes = 0;
+        self->stream_nodes_capacity_items = 0;
+
+        free(self->stream_nodes);
+        self->stream_nodes = NULL;
+    }
+
+    if(self->ports) {
+        for(size_t i = 0; i < self->num_ports; ++i) {
+            free(self->ports[i].name);
+        }
+        self->num_ports = 0;
+        self->ports_capacity_items = 0;
+
+        free(self->ports);
+        self->ports = NULL;
+    }
+
+    if(self->links) {
+        self->num_links = 0;
+        self->links_capacity_items = 0;
+
+        free(self->links);
+        self->links = NULL;
+    }
+
+    if(self->requested_links) {
+        for(size_t i = 0; i < self->num_requested_links; ++i) {
+            for(int j = 0; j < self->requested_links[i].num_outputs; ++j) {
+                free(self->requested_links[i].outputs[j].name);
+            }
+            free(self->requested_links[i].outputs);
+            free(self->requested_links[i].input_name);
+        }
+        self->num_requested_links = 0;
+        self->requested_links_capacity_items = 0;
+
+        free(self->requested_links);
+        self->requested_links = NULL;
     }
-    self->num_requested_links = 0;
 
 #if PW_CHECK_VERSION(0, 3, 49)
     pw_deinit();
 #endif
 }
 
-static struct pw_properties* gsr_pipewire_create_null_audio_sink(const char *name) {
-    char props_str[512];
-    snprintf(props_str, sizeof(props_str), "{ factory.name=support.null-audio-sink node.name=\"%s\" media.class=Audio/Sink object.linger=false audio.position=[FL FR] monitor.channel-volumes=true monitor.passthrough=true adjust_time=0 node.description=gsr-app-sink slaves=\"\" }", name);
-    struct pw_properties *props = pw_properties_new_string(props_str);
-    if(!props) {
-        fprintf(stderr, "gsr error: gsr_pipewire_create_null_audio_sink: failed to create virtual sink properties\n");
-        return NULL;
-    }
-    return props;
-}
-
-bool gsr_pipewire_audio_create_virtual_sink(gsr_pipewire_audio *self, const char *name) {
-    if(self->num_virtual_sink_proxies == GSR_PIPEWIRE_AUDIO_MAX_VIRTUAL_SINKS) {
-        fprintf(stderr, "gsr error: gsr_pipewire_audio_create_virtual_sink: reached max number of virtual sinks\n");
-        return false;
-    }
-
-    pw_thread_loop_lock(self->thread_loop);
-
-    struct pw_properties *virtual_sink_props = gsr_pipewire_create_null_audio_sink(name);
-    if(!virtual_sink_props) {
-        pw_thread_loop_unlock(self->thread_loop);
-        return false;
-    }
-
-    struct pw_proxy *virtual_sink_proxy = pw_core_create_object(self->core, "adapter", PW_TYPE_INTERFACE_Node, PW_VERSION_NODE, &virtual_sink_props->dict, 0);
-    // TODO:
-    // If these are done then the above needs sizeof(*self) as the last argument
-    //pw_proxy_add_object_listener(virtual_sink_proxy, &pd->object_listener, &node_events, self);
-	//pw_proxy_add_listener(virtual_sink_proxy, &pd->proxy_listener, &proxy_events, self);
-    // TODO: proxy
-    pw_properties_free(virtual_sink_props);
-    if(!virtual_sink_proxy) {
-        fprintf(stderr, "gsr error: gsr_pipewire_audio_create_virtual_sink: failed to create virtual sink\n");
-        pw_thread_loop_unlock(self->thread_loop);
-        return false;
-    }
-
-    self->server_version_sync = pw_core_sync(self->core, PW_ID_CORE, self->server_version_sync);
-    pw_thread_loop_wait(self->thread_loop);
-    pw_thread_loop_unlock(self->thread_loop);
-
-    self->virtual_sink_proxies[self->num_virtual_sink_proxies] = virtual_sink_proxy;
-    ++self->num_virtual_sink_proxies;
-
-    return true;
-}
-
 static bool string_remove_suffix(char *str, const char *suffix) {
     int str_len = strlen(str);
     int suffix_len = strlen(suffix);
@@ -701,10 +754,8 @@ static bool string_remove_suffix(char *str, const char *suffix) {
 }
 
 static bool gsr_pipewire_audio_add_links_to_output(gsr_pipewire_audio *self, const char **output_names, int num_output_names, const char *input_name, gsr_pipewire_audio_node_type output_type, gsr_pipewire_audio_link_input_type input_type, bool inverted) {
-    if(self->num_requested_links >= GSR_PIPEWIRE_AUDIO_MAX_REQUESTED_LINKS) {
-        fprintf(stderr, "gsr error: reached the maximum amount of audio links\n");
+    if(!array_ensure_capacity((void**)&self->requested_links, self->num_requested_links, &self->requested_links_capacity_items, sizeof(gsr_pipewire_audio_requested_link)))
         return false;
-    }
     
     gsr_pipewire_audio_requested_output *outputs = calloc(num_output_names, sizeof(gsr_pipewire_audio_requested_output));
     if(!outputs)
@@ -744,6 +795,7 @@ static bool gsr_pipewire_audio_add_links_to_output(gsr_pipewire_audio *self, con
     self->requested_links[self->num_requested_links].inverted = inverted;
     ++self->num_requested_links;
     gsr_pipewire_audio_create_link(self, &self->requested_links[self->num_requested_links - 1]);
+    // TODO: Remove these?
     gsr_pipewire_audio_create_link_for_default_devices(self, &self->requested_links[self->num_requested_links - 1], GSR_PIPEWIRE_AUDIO_REQUESTED_TYPE_DEFAULT_OUTPUT);
     gsr_pipewire_audio_create_link_for_default_devices(self, &self->requested_links[self->num_requested_links - 1], GSR_PIPEWIRE_AUDIO_REQUESTED_TYPE_DEFAULT_INPUT);
     pw_thread_loop_unlock(self->thread_loop);
@@ -775,13 +827,17 @@ bool gsr_pipewire_audio_add_link_from_apps_to_sink_inverted(gsr_pipewire_audio *
     return gsr_pipewire_audio_add_links_to_output(self, app_names, num_app_names, sink_name_input, GSR_PIPEWIRE_AUDIO_NODE_TYPE_STREAM_OUTPUT, GSR_PIPEWIRE_AUDIO_LINK_INPUT_TYPE_SINK, true);
 }
 
+bool gsr_pipewire_audio_add_link_from_sources_to_stream(gsr_pipewire_audio *self, const char **source_names, int num_source_names, const char *stream_name_input) {
+    return gsr_pipewire_audio_add_links_to_output(self, source_names, num_source_names, stream_name_input, GSR_PIPEWIRE_AUDIO_NODE_TYPE_SINK_OR_SOURCE, GSR_PIPEWIRE_AUDIO_LINK_INPUT_TYPE_STREAM, false);
+}
+
 bool gsr_pipewire_audio_add_link_from_sources_to_sink(gsr_pipewire_audio *self, const char **source_names, int num_source_names, const char *sink_name_input) {
     return gsr_pipewire_audio_add_links_to_output(self, source_names, num_source_names, sink_name_input, GSR_PIPEWIRE_AUDIO_NODE_TYPE_SINK_OR_SOURCE, GSR_PIPEWIRE_AUDIO_LINK_INPUT_TYPE_SINK, false);
 }
 
 void gsr_pipewire_audio_for_each_app(gsr_pipewire_audio *self, gsr_pipewire_audio_app_query_callback callback, void *userdata) {
     pw_thread_loop_lock(self->thread_loop);
-    for(int i = 0; i < self->num_stream_nodes; ++i) {
+    for(int i = 0; i < (int)self->num_stream_nodes; ++i) {
         const gsr_pipewire_audio_node *node = &self->stream_nodes[i];
         if(node->type != GSR_PIPEWIRE_AUDIO_NODE_TYPE_STREAM_OUTPUT)
             continue;

src/pipewire_video.c

@@ -6,7 +6,7 @@
 #include <spa/param/video/format-utils.h>
 #include <spa/debug/types.h>
 
-#include <libdrm/drm_fourcc.h>
+#include <drm_fourcc.h>
 
 #include <fcntl.h>
 #include <unistd.h>
@@ -21,6 +21,29 @@
 #define SPA_POD_PROP_FLAG_DONT_FIXATE (1 << 4)
 #endif
 
+#if !PW_CHECK_VERSION(0, 3, 62)
+enum spa_meta_videotransform_value {
+    SPA_META_TRANSFORMATION_None = 0,    /**< no transform */
+    SPA_META_TRANSFORMATION_90,          /**< 90 degree counter-clockwise */
+    SPA_META_TRANSFORMATION_180,         /**< 180 degree counter-clockwise */
+    SPA_META_TRANSFORMATION_270,         /**< 270 degree counter-clockwise */
+    SPA_META_TRANSFORMATION_Flipped,     /**< 180 degree flipped around the vertical axis. Equivalent
+                                           * to a reflexion through the vertical line splitting the
+                                           * buffer in two equal sized parts */
+    SPA_META_TRANSFORMATION_Flipped90,   /**< flip then rotate around 90 degree counter-clockwise */
+    SPA_META_TRANSFORMATION_Flipped180,  /**< flip then rotate around 180 degree counter-clockwise */
+    SPA_META_TRANSFORMATION_Flipped270,  /**< flip then rotate around 270 degree counter-clockwise */
+};
+
+/** a transformation of the buffer */
+struct spa_meta_videotransform {
+    uint32_t transform;                  /**< orientation transformation that was applied to the buffer,
+                                           *  one of enum spa_meta_videotransform_value */
+};
+
+#define SPA_META_VideoTransform 8
+#endif
+
 #define CURSOR_META_SIZE(width, height)                                    \
     (sizeof(struct spa_meta_cursor) + sizeof(struct spa_meta_bitmap) + \
      width * height * 4)
@@ -93,8 +116,6 @@ static const struct pw_core_events core_events = {
 
 static void on_process_cb(void *user_data) {
     gsr_pipewire_video *self = user_data;
-    struct spa_meta_cursor *cursor = NULL;
-    //struct spa_meta *video_damage = NULL;
 
     /* Find the most recent buffer */
     struct pw_buffer *pw_buf = NULL;
@@ -113,13 +134,12 @@ static void on_process_cb(void *user_data) {
     }
 
     struct spa_buffer *buffer = pw_buf->buffer;
-    const bool has_buffer = buffer->datas[0].chunk->size != 0;
-    if(!has_buffer)
-        goto read_metadata;
+    const bool has_buffer = buffer->n_datas > 0 && buffer->datas[0].chunk->size != 0;
 
     pthread_mutex_lock(&self->mutex);
 
-    if(buffer->datas[0].type == SPA_DATA_DmaBuf) {
+    bool buffer_updated = false;
+    if(has_buffer && buffer->datas[0].type == SPA_DATA_DmaBuf) {
         for(size_t i = 0; i < self->dmabuf_num_planes; ++i) {
             if(self->dmabuf_data[i].fd > 0) {
                 close(self->dmabuf_data[i].fd);
@@ -137,9 +157,7 @@ static void on_process_cb(void *user_data) {
             self->dmabuf_data[i].stride = buffer->datas[i].chunk->stride;
         }
 
-        self->damaged = true;
-    } else {
-        // TODO:
+        buffer_updated = true;
     }
 
     // TODO: Move down to read_metadata
@@ -157,32 +175,51 @@ static void on_process_cb(void *user_data) {
         self->crop.valid = false;
     }
 
-    pthread_mutex_unlock(&self->mutex);
+    struct spa_meta_videotransform *video_transform = spa_buffer_find_meta_data(buffer, SPA_META_VideoTransform, sizeof(*video_transform));
+    enum spa_meta_videotransform_value transform = SPA_META_TRANSFORMATION_None;
+    if(video_transform)
+        transform = video_transform->transform;
 
-read_metadata:
+    self->rotation = GSR_MONITOR_ROT_0;
+    switch(transform) {
+        case SPA_META_TRANSFORMATION_90:
+            self->rotation = GSR_MONITOR_ROT_90;
+            break;
+        case SPA_META_TRANSFORMATION_180:
+            self->rotation = GSR_MONITOR_ROT_180;
+            break;
+        case SPA_META_TRANSFORMATION_270:
+            self->rotation = GSR_MONITOR_ROT_270;
+            break;
+        default:
+            // TODO: Support other rotations. Wayland compositors dont use them yet so it's ok to not support it now
+            break;
+    }
 
-    // video_damage = spa_buffer_find_meta(buffer, SPA_META_VideoDamage);
-    // if(video_damage) {
-    //     struct spa_meta_region *r = spa_meta_first(video_damage);
-    //     if(spa_meta_check(r, video_damage)) {
-    //         //fprintf(stderr, "damage: %d,%d %ux%u\n", r->region.position.x, r->region.position.y, r->region.size.width, r->region.size.height);
-    //         pthread_mutex_lock(&self->mutex);
-    //         self->damaged = true;
-    //         pthread_mutex_unlock(&self->mutex);
-    //     }
-    // }
+    const struct spa_meta *video_damage = spa_buffer_find_meta(buffer, SPA_META_VideoDamage);
+    if(video_damage) {
+        struct spa_meta_region *meta_region = NULL;
+        spa_meta_for_each(meta_region, video_damage) {
+            if(meta_region->region.size.width == 0 || meta_region->region.size.height == 0)
+                continue;
 
-    cursor = spa_buffer_find_meta_data(buffer, SPA_META_Cursor, sizeof(*cursor));
+            self->damaged = true;
+            break;
+        }
+    } else if(buffer_updated) {
+        self->damaged = true;
+    }
+
+    const struct spa_meta_cursor *cursor = spa_buffer_find_meta_data(buffer, SPA_META_Cursor, sizeof(*cursor));
     self->cursor.valid = cursor && spa_meta_cursor_is_valid(cursor);
     
     if (self->cursor.visible && self->cursor.valid) {
-        pthread_mutex_lock(&self->mutex);
-
         struct spa_meta_bitmap *bitmap = NULL;
         if (cursor->bitmap_offset)
             bitmap = SPA_MEMBER(cursor, cursor->bitmap_offset, struct spa_meta_bitmap);
 
-        if (bitmap && bitmap->size.width > 0 && bitmap->size.height && is_cursor_format_supported(bitmap->format)) {
+        // TODO: Maybe check if the cursor is actually visible by checking if there are visible pixels
+        if (bitmap && bitmap->size.width > 0 && bitmap->size.height > 0 && is_cursor_format_supported(bitmap->format)) {
             const uint8_t *bitmap_data = SPA_MEMBER(bitmap, bitmap->offset, uint8_t);
             fprintf(stderr, "gsr info: pipewire: cursor bitmap update, size: %dx%d, format: %s\n",
                 (int)bitmap->size.width, (int)bitmap->size.height, spa_debug_type_find_name(spa_type_video_format, bitmap->format));
@@ -199,15 +236,19 @@ read_metadata:
             self->cursor.hotspot_y = cursor->hotspot.y;
             self->cursor.width = bitmap->size.width;
             self->cursor.height = bitmap->size.height;
+            self->damaged = true;
         }
 
+        if(cursor->position.x != self->cursor.x || cursor->position.y != self->cursor.y)
+            self->damaged = true;
+
         self->cursor.x = cursor->position.x;
         self->cursor.y = cursor->position.y;
-        pthread_mutex_unlock(&self->mutex);
 
         //fprintf(stderr, "gsr info: pipewire: cursor: %d %d %d %d\n", cursor->hotspot.x, cursor->hotspot.y, cursor->position.x, cursor->position.y);
     }
 
+    pthread_mutex_unlock(&self->mutex);
     pw_stream_queue_buffer(self->stream, pw_buf);
 }
 
@@ -239,24 +280,26 @@ static void on_param_changed_cb(void *user_data, uint32_t id, const struct spa_p
          self->format.info.raw.format,
          spa_debug_type_find_name(spa_type_video_format, self->format.info.raw.format));
 
-    if(has_modifier) {
+    self->has_modifier = has_modifier;
+    if(self->has_modifier) {
         fprintf(stderr, "gsr info: pipewire:    Modifier: 0x%" PRIx64 "\n", self->format.info.raw.modifier);
     }
 
     fprintf(stderr, "gsr info: pipewire:    Size: %dx%d\n", self->format.info.raw.size.width, self->format.info.raw.size.height);
     fprintf(stderr, "gsr info: pipewire:    Framerate: %d/%d\n", self->format.info.raw.framerate.num, self->format.info.raw.framerate.denom);
 
-    uint8_t params_buffer[1024];
+    uint8_t params_buffer[2048];
     struct spa_pod_builder pod_builder = SPA_POD_BUILDER_INIT(params_buffer, sizeof(params_buffer));
-    const struct spa_pod *params[4];
+    const struct spa_pod *params[5];
+    int param_index = 0;
 
-    params[0] = spa_pod_builder_add_object(
+    params[param_index++] = spa_pod_builder_add_object(
         &pod_builder, SPA_TYPE_OBJECT_ParamMeta, SPA_PARAM_Meta,
         SPA_PARAM_META_type, SPA_POD_Id(SPA_META_VideoCrop),
         SPA_PARAM_META_size,
         SPA_POD_Int(sizeof(struct spa_meta_region)));
 
-    params[1] = spa_pod_builder_add_object(
+    params[param_index++] = spa_pod_builder_add_object(
         &pod_builder, SPA_TYPE_OBJECT_ParamMeta, SPA_PARAM_Meta,
         SPA_PARAM_META_type, SPA_POD_Id(SPA_META_VideoDamage),
         SPA_PARAM_META_size, SPA_POD_CHOICE_RANGE_Int(
@@ -264,7 +307,7 @@ static void on_param_changed_cb(void *user_data, uint32_t id, const struct spa_p
                                 sizeof(struct spa_meta_region) * 1,
                                 sizeof(struct spa_meta_region) * 16));
 
-    params[2] = spa_pod_builder_add_object(
+    params[param_index++] = spa_pod_builder_add_object(
         &pod_builder, SPA_TYPE_OBJECT_ParamMeta, SPA_PARAM_Meta,
         SPA_PARAM_META_type, SPA_POD_Id(SPA_META_Cursor),
         SPA_PARAM_META_size,
@@ -272,21 +315,41 @@ static void on_param_changed_cb(void *user_data, uint32_t id, const struct spa_p
                      CURSOR_META_SIZE(1, 1),
                      CURSOR_META_SIZE(1024, 1024)));
 
-    params[3] = spa_pod_builder_add_object(
+    params[param_index++] = spa_pod_builder_add_object(
         &pod_builder, SPA_TYPE_OBJECT_ParamBuffers, SPA_PARAM_Buffers,
         SPA_PARAM_BUFFERS_dataType, SPA_POD_Int(buffer_types));
 
-    pw_stream_update_params(self->stream, params, 4);
+#if PW_CHECK_VERSION(0, 3, 62)
+	if (check_pw_version(&self->server_version, 0, 3, 62)) {
+		/* Video transformation */
+		params[param_index++] = spa_pod_builder_add_object(&pod_builder, SPA_TYPE_OBJECT_ParamMeta, SPA_PARAM_Meta,
+								SPA_PARAM_META_type,
+								SPA_POD_Id(SPA_META_VideoTransform),
+								SPA_PARAM_META_size,
+								SPA_POD_Int(sizeof(struct spa_meta_videotransform)));
+	}
+#endif
+
+    pw_stream_update_params(self->stream, params, param_index);
     self->negotiated = true;
 }
 
-static void on_state_changed_cb(void *user_data, enum pw_stream_state old, enum pw_stream_state state, const char *error) {
-    (void)old;
+static void on_state_changed_cb(void *user_data, enum pw_stream_state prev_state, enum pw_stream_state new_state, const char *error) {
     gsr_pipewire_video *self = user_data;
 
-    fprintf(stderr, "gsr info: pipewire: stream %p state: \"%s\" (error: %s)\n",
-         (void*)self->stream, pw_stream_state_as_string(state),
+    fprintf(stderr, "gsr info: pipewire: stream %p previous state: \"%s\", new state: \"%s\" (error: %s)\n",
+         (void*)self->stream, pw_stream_state_as_string(prev_state), pw_stream_state_as_string(new_state),
          error ? error : "none");
+
+    pthread_mutex_lock(&self->mutex);
+    if(new_state == PW_STREAM_STATE_PAUSED) {
+        self->paused_start_secs = clock_get_monotonic_seconds();
+        self->paused = true;
+    } else {
+        self->paused = false;
+    }
+    self->streaming = (new_state == PW_STREAM_STATE_STREAMING);
+    pthread_mutex_unlock(&self->mutex);
 }
 
 static const struct pw_stream_events stream_events = {
@@ -350,15 +413,15 @@ static int64_t spa_video_format_to_drm_format(const enum spa_video_format format
         case SPA_VIDEO_FORMAT_BGRA:       return DRM_FORMAT_ARGB8888;
         case SPA_VIDEO_FORMAT_RGB:        return DRM_FORMAT_XBGR8888;
         case SPA_VIDEO_FORMAT_BGR:        return DRM_FORMAT_XRGB8888;
-        case SPA_VIDEO_FORMAT_ARGB:       return DRM_FORMAT_XRGB8888;
-        case SPA_VIDEO_FORMAT_ABGR:       return DRM_FORMAT_XRGB8888;
+        //case SPA_VIDEO_FORMAT_ARGB:       return DRM_FORMAT_BGRA8888;
+        //case SPA_VIDEO_FORMAT_ABGR:       return DRM_FORMAT_RGBA8888;
 #if PW_CHECK_VERSION(0, 3, 41)
         case SPA_VIDEO_FORMAT_xRGB_210LE: return DRM_FORMAT_XRGB2101010;
         case SPA_VIDEO_FORMAT_xBGR_210LE: return DRM_FORMAT_XBGR2101010;
         case SPA_VIDEO_FORMAT_ARGB_210LE: return DRM_FORMAT_ARGB2101010;
         case SPA_VIDEO_FORMAT_ABGR_210LE: return DRM_FORMAT_ABGR2101010;
 #endif
-        default:                    break;
+        default:                          break;
     }
     return DRM_FORMAT_INVALID;
 }
@@ -366,23 +429,23 @@ static int64_t spa_video_format_to_drm_format(const enum spa_video_format format
 #if PW_CHECK_VERSION(0, 3, 41)
 #define GSR_PIPEWIRE_VIDEO_NUM_VIDEO_FORMATS GSR_PIPEWIRE_VIDEO_MAX_VIDEO_FORMATS
 #else
-#define GSR_PIPEWIRE_VIDEO_NUM_VIDEO_FORMATS 8
+#define GSR_PIPEWIRE_VIDEO_NUM_VIDEO_FORMATS 6
 #endif
 
 static const enum spa_video_format video_formats[GSR_PIPEWIRE_VIDEO_MAX_VIDEO_FORMATS] = {
-    SPA_VIDEO_FORMAT_BGRA,
     SPA_VIDEO_FORMAT_BGRx,
     SPA_VIDEO_FORMAT_BGR,
     SPA_VIDEO_FORMAT_RGBx,
-    SPA_VIDEO_FORMAT_RGBA,
     SPA_VIDEO_FORMAT_RGB,
-    SPA_VIDEO_FORMAT_ARGB,
-    SPA_VIDEO_FORMAT_ABGR,
+    SPA_VIDEO_FORMAT_RGBA,
+    SPA_VIDEO_FORMAT_BGRA,
+    //SPA_VIDEO_FORMAT_ARGB,
+    //SPA_VIDEO_FORMAT_ABGR,
 #if PW_CHECK_VERSION(0, 3, 41)
     SPA_VIDEO_FORMAT_xRGB_210LE,
     SPA_VIDEO_FORMAT_xBGR_210LE,
     SPA_VIDEO_FORMAT_ARGB_210LE,
-    SPA_VIDEO_FORMAT_ABGR_210LE
+    SPA_VIDEO_FORMAT_ABGR_210LE,
 #endif
 };
 
@@ -438,6 +501,8 @@ static bool spa_video_format_get_modifiers(gsr_pipewire_video *self, const enum
         //modifiers[0] = DRM_FORMAT_MOD_LINEAR;
         //modifiers[1] = DRM_FORMAT_MOD_INVALID;
         //*num_modifiers = 2;
+        modifiers[0] = DRM_FORMAT_MOD_INVALID;
+        *num_modifiers = 1;
         return false;
     }
 
@@ -452,7 +517,8 @@ static bool spa_video_format_get_modifiers(gsr_pipewire_video *self, const enum
         //modifiers[0] = DRM_FORMAT_MOD_LINEAR;
         //modifiers[1] = DRM_FORMAT_MOD_INVALID;
         //*num_modifiers = 2;
-        *num_modifiers = 0;
+        modifiers[0] = DRM_FORMAT_MOD_INVALID;
+        *num_modifiers = 1;
         return false;
     }
 
@@ -467,10 +533,36 @@ static bool spa_video_format_get_modifiers(gsr_pipewire_video *self, const enum
 static void gsr_pipewire_video_init_modifiers(gsr_pipewire_video *self) {
     for(size_t i = 0; i < GSR_PIPEWIRE_VIDEO_NUM_VIDEO_FORMATS; i++) {
         self->supported_video_formats[i].format = video_formats[i];
-        int32_t num_modifiers = 0;
-        spa_video_format_get_modifiers(self, self->supported_video_formats[i].format, self->modifiers + self->num_modifiers, GSR_PIPEWIRE_VIDEO_MAX_MODIFIERS - self->num_modifiers, &num_modifiers);
+        int32_t num_modifiers_video_format = 0;
+        spa_video_format_get_modifiers(self, self->supported_video_formats[i].format, self->modifiers + self->num_modifiers, GSR_PIPEWIRE_VIDEO_MAX_MODIFIERS - self->num_modifiers, &num_modifiers_video_format);
         self->supported_video_formats[i].modifiers_index = self->num_modifiers;
-        self->supported_video_formats[i].modifiers_size = num_modifiers;
+        self->supported_video_formats[i].modifiers_size = num_modifiers_video_format;
+        self->num_modifiers += num_modifiers_video_format;
+    }
+}
+
+/* Returns the number of modifiers  */
+static size_t gsr_pipewire_video_format_remove_modifier(gsr_pipewire_video *self, gsr_video_format *video_format, uint64_t modifier) {
+    for(size_t i = 0; i < video_format->modifiers_size; ++i) {
+        if(self->modifiers[video_format->modifiers_index + i] == modifier) {
+            for(size_t j = i + 1; j < video_format->modifiers_size; ++j) {
+                self->modifiers[video_format->modifiers_index + j - 1] = self->modifiers[video_format->modifiers_index + j];
+            }
+
+            --video_format->modifiers_size;
+            break;
+        }
+    }
+    return video_format->modifiers_size;
+}
+
+static void gsr_pipewire_video_remove_modifier(gsr_pipewire_video *self, uint64_t modifier) {
+    self->num_modifiers = 0;
+    for(size_t i = 0; i < GSR_PIPEWIRE_VIDEO_NUM_VIDEO_FORMATS; i++) {
+        gsr_video_format *video_format = &self->supported_video_formats[i];
+        const size_t num_modifiers_video_format = gsr_pipewire_video_format_remove_modifier(self, video_format, modifier);
+        video_format->modifiers_index = self->num_modifiers;
+        self->num_modifiers += num_modifiers_video_format;
     }
 }
 
@@ -697,14 +789,23 @@ static EGLImage gsr_pipewire_video_create_egl_image_with_fallback(gsr_pipewire_v
     }
 
     EGLImage image = NULL;
-    if(self->no_modifiers_fallback) {
+    if(self->no_modifiers_fallback || !self->has_modifier) {
         image = gsr_pipewire_video_create_egl_image(self, fds, offsets, pitches, modifiers, false);
     } else {
         image = gsr_pipewire_video_create_egl_image(self, fds, offsets, pitches, modifiers, true);
         if(!image) {
-            fprintf(stderr, "gsr error: gsr_pipewire_video_create_egl_image_with_fallback: failed to create egl image with modifiers, trying without modifiers\n");
-            self->no_modifiers_fallback = true;
-            image = gsr_pipewire_video_create_egl_image(self, fds, offsets, pitches, modifiers, false);
+            if(self->format.info.raw.modifier == DRM_FORMAT_MOD_INVALID) {
+                fprintf(stderr, "gsr error: gsr_pipewire_video_create_egl_image_with_fallback: failed to create egl image with modifiers, trying without modifiers\n");
+                self->no_modifiers_fallback = true;
+                image = gsr_pipewire_video_create_egl_image(self, fds, offsets, pitches, modifiers, false);
+            } else {
+                fprintf(stderr, "gsr error: gsr_pipewire_video_create_egl_image_with_fallback: failed to create egl image with modifier 0x%" PRIx64 ", renegotiating with a different modifier\n", self->format.info.raw.modifier);
+                self->negotiated = false;
+                pw_thread_loop_lock(self->thread_loop);
+                gsr_pipewire_video_remove_modifier(self, self->format.info.raw.modifier);
+                pw_loop_signal_event(pw_thread_loop_get_loop(self->thread_loop), self->reneg);
+                pw_thread_loop_unlock(self->thread_loop);
+            }
         }
     }
     return image;
@@ -739,8 +840,6 @@ static void gsr_pipewire_video_update_cursor_texture(gsr_pipewire_video *self, g
     self->egl->glBindTexture(GL_TEXTURE_2D, texture_map.cursor_texture_id);
     // TODO: glTextureSubImage2D if same size
     self->egl->glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8, self->cursor.width, self->cursor.height, 0, GL_RGBA, GL_UNSIGNED_BYTE, self->cursor.data);
-    self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
     self->egl->glBindTexture(GL_TEXTURE_2D, 0);
@@ -749,58 +848,73 @@ static void gsr_pipewire_video_update_cursor_texture(gsr_pipewire_video *self, g
     self->cursor.data = NULL;
 }
 
-bool gsr_pipewire_video_map_texture(gsr_pipewire_video *self, gsr_texture_map texture_map, gsr_pipewire_video_region *region, gsr_pipewire_video_region *cursor_region, gsr_pipewire_video_dmabuf_data *dmabuf_data, int *num_dmabuf_data, uint32_t *fourcc, uint64_t *modifiers, bool *using_external_image) {
+bool gsr_pipewire_video_map_texture(gsr_pipewire_video *self, gsr_texture_map texture_map, gsr_map_texture_output *output) {
     for(int i = 0; i < GSR_PIPEWIRE_VIDEO_DMABUF_MAX_PLANES; ++i) {
-        memset(&dmabuf_data[i], 0, sizeof(gsr_pipewire_video_dmabuf_data));
+        memset(&output->dmabuf_data[i], 0, sizeof(gsr_pipewire_video_dmabuf_data));
     }
-    *num_dmabuf_data = 0;
-    *using_external_image = self->external_texture_fallback;
-    *fourcc = 0;
-    *modifiers = 0;
+    output->num_dmabuf_data = 0;
+    output->using_external_image = self->external_texture_fallback;
+    output->fourcc = 0;
+    output->modifiers = 0;
+    output->rotation = GSR_MONITOR_ROT_0;
     pthread_mutex_lock(&self->mutex);
 
-    if(!self->negotiated || self->dmabuf_data[0].fd <= 0) {
+    if(!self->negotiated || !self->streaming || self->dmabuf_data[0].fd <= 0) {
         pthread_mutex_unlock(&self->mutex);
         return false;
     }
 
     EGLImage image = gsr_pipewire_video_create_egl_image_with_fallback(self);
-    if(image) {
-        gsr_pipewire_video_bind_image_to_texture_with_fallback(self, texture_map, image);
-        *using_external_image = self->external_texture_fallback;
-        self->egl->eglDestroyImage(self->egl->egl_display, image);
+    if(!image) {
+        pthread_mutex_unlock(&self->mutex);
+        return false;
     }
 
+    gsr_pipewire_video_bind_image_to_texture_with_fallback(self, texture_map, image);
+    output->using_external_image = self->external_texture_fallback;
+    self->egl->eglDestroyImage(self->egl->egl_display, image);
+
     gsr_pipewire_video_update_cursor_texture(self, texture_map);
 
-    region->x = 0;
-    region->y = 0;
+    output->texture_width = self->format.info.raw.size.width;
+    output->texture_height = self->format.info.raw.size.height;
 
-    region->width = self->format.info.raw.size.width;
-    region->height = self->format.info.raw.size.height;
+    output->region.x = 0;
+    output->region.y = 0;
+
+    output->region.width = self->format.info.raw.size.width;
+    output->region.height = self->format.info.raw.size.height;
 
     if(self->crop.valid) {
-        region->x = self->crop.x;
-        region->y = self->crop.y;
+        output->region.x = self->crop.x;
+        output->region.y = self->crop.y;
 
-        region->width = self->crop.width;
-        region->height = self->crop.height;
+        output->region.width = self->crop.width;
+        output->region.height = self->crop.height;
+    }
+
+    // TODO: Test transform + cropping
+    if(self->rotation == GSR_MONITOR_ROT_90 || self->rotation == GSR_MONITOR_ROT_270) {
+        const int temp = output->region.width;
+        output->region.width = output->region.height;
+        output->region.height = temp;
     }
 
     /* TODO: Test if cursor hotspot is correct */
-    cursor_region->x = self->cursor.x - self->cursor.hotspot_x;
-    cursor_region->y = self->cursor.y - self->cursor.hotspot_y;
+    output->cursor_region.x = self->cursor.x - self->cursor.hotspot_x;
+    output->cursor_region.y = self->cursor.y - self->cursor.hotspot_y;
 
-    cursor_region->width = self->cursor.width;
-    cursor_region->height = self->cursor.height;
+    output->cursor_region.width = self->cursor.width;
+    output->cursor_region.height = self->cursor.height;
 
     for(size_t i = 0; i < self->dmabuf_num_planes; ++i) {
-        dmabuf_data[i] = self->dmabuf_data[i];
+        output->dmabuf_data[i] = self->dmabuf_data[i];
         self->dmabuf_data[i].fd = -1;
     }
-    *num_dmabuf_data = self->dmabuf_num_planes;
-    *fourcc = spa_video_format_to_drm_format(self->format.info.raw.format);
-    *modifiers = self->format.info.raw.modifier;
+    output->num_dmabuf_data = self->dmabuf_num_planes;
+    output->fourcc = spa_video_format_to_drm_format(self->format.info.raw.format);
+    output->modifiers = self->format.info.raw.modifier;
+    output->rotation = self->rotation;
     self->dmabuf_num_planes = 0;
 
     pthread_mutex_unlock(&self->mutex);
@@ -808,6 +922,9 @@ bool gsr_pipewire_video_map_texture(gsr_pipewire_video *self, gsr_texture_map te
 }
 
 bool gsr_pipewire_video_is_damaged(gsr_pipewire_video *self) {
+    if(!self->mutex_initialized)
+        return false;
+
     bool damaged = false;
     pthread_mutex_lock(&self->mutex);
     damaged = self->damaged;
@@ -816,7 +933,21 @@ bool gsr_pipewire_video_is_damaged(gsr_pipewire_video *self) {
 }
 
 void gsr_pipewire_video_clear_damage(gsr_pipewire_video *self) {
+    if(!self->mutex_initialized)
+        return;
+
     pthread_mutex_lock(&self->mutex);
     self->damaged = false;
     pthread_mutex_unlock(&self->mutex);
 }
+
+bool gsr_pipewire_video_should_restart(gsr_pipewire_video *self) {
+    if(!self->mutex_initialized)
+        return false;
+
+    bool should_restart = false;
+    pthread_mutex_lock(&self->mutex);
+    should_restart = self->paused && clock_get_monotonic_seconds() - self->paused_start_secs >= 3.0;
+    pthread_mutex_unlock(&self->mutex);
+    return should_restart;
+}

src/plugins.c

@@ -0,0 +1,138 @@
+#include "../include/plugins.h"
+#include "../include/utils.h"
+#include <stdio.h>
+#include <string.h>
+#include <dlfcn.h>
+#include <assert.h>
+
+static int color_depth_to_gl_internal_format(gsr_plugin_color_depth color_depth) {
+    switch(color_depth) {
+        case GSR_PLUGIN_COLOR_DEPTH_8_BITS:
+            return GL_RGBA8;
+        case GSR_PLUGIN_COLOR_DEPTH_10_BITS:
+            return GL_RGBA16;
+    }
+    assert(false);
+    return GL_RGBA8;
+}
+
+bool gsr_plugins_init(gsr_plugins *self, gsr_plugin_init_params init_params, gsr_egl *egl) {
+    memset(self, 0, sizeof(*self));
+    self->init_params = init_params;
+    self->egl = egl;
+
+    /* TODO: GL_RGB8? */
+    const unsigned int texture = gl_create_texture(egl, init_params.width, init_params.height, color_depth_to_gl_internal_format(init_params.color_depth), GL_RGBA, GL_LINEAR);
+    if(texture == 0) {
+        fprintf(stderr, "gsr error: gsr_plugins_init failed to create texture\n");
+        return false;
+    }
+    self->texture = texture;
+
+    gsr_color_conversion_params color_conversion_params = {
+        .egl = egl,
+        .destination_color = GSR_DESTINATION_COLOR_RGB8, /* TODO: Support 10-bits, use init_params.color_depth */
+        .destination_textures[0] = self->texture,
+        .destination_textures_size[0] = (vec2i){ init_params.width, init_params.height },
+        .num_destination_textures = 1,
+        .color_range = GSR_COLOR_RANGE_FULL,
+        .load_external_image_shader = false,
+        //.force_graphics_shader = false,
+    };
+    color_conversion_params.destination_textures[0] = self->texture;
+
+    if(gsr_color_conversion_init(&self->color_conversion, &color_conversion_params) != 0) {
+        fprintf(stderr, "gsr error: gsr_plugins_init failed to create color conversion\n");
+        gsr_plugins_deinit(self);
+        return false;
+    }
+    gsr_color_conversion_clear(&self->color_conversion);
+
+    return true;
+}
+
+void gsr_plugins_deinit(gsr_plugins *self) {
+    for(int i = self->num_plugins - 1; i >= 0; --i) {
+        gsr_plugin *plugin = &self->plugins[i];
+        plugin->gsr_plugin_deinit(plugin->data.userdata);
+        fprintf(stderr, "gsr info: unloaded plugin: %s\n", plugin->data.name);
+    }
+    self->num_plugins = 0;
+
+    if(self->texture > 0) {
+        self->egl->glDeleteTextures(1, &self->texture);
+        self->texture = 0;
+    }
+
+    gsr_color_conversion_deinit(&self->color_conversion);
+}
+
+bool gsr_plugins_load_plugin(gsr_plugins *self, const char *plugin_filepath) {
+    if(self->num_plugins >= GSR_MAX_PLUGINS) {
+        fprintf(stderr, "gsr error: gsr_plugins_load_plugin failed, more plugins can't load more than %d plugins. Report this as an issue\n", GSR_MAX_PLUGINS);
+        return false;
+    }
+
+    gsr_plugin plugin;
+    memset(&plugin, 0, sizeof(plugin));
+
+    plugin.lib = dlopen(plugin_filepath, RTLD_LAZY);
+    if(!plugin.lib) {
+        fprintf(stderr, "gsr error: gsr_plugins_load_plugin failed to load \"%s\", error: %s\n", plugin_filepath, dlerror());
+        return false;
+    }
+
+    plugin.gsr_plugin_init = dlsym(plugin.lib, "gsr_plugin_init");
+    if(!plugin.gsr_plugin_init) {
+        fprintf(stderr, "gsr error: gsr_plugins_load_plugin failed to find \"gsr_plugin_init\" in plugin \"%s\"\n", plugin_filepath);
+        goto fail;
+    }
+
+    plugin.gsr_plugin_deinit = dlsym(plugin.lib, "gsr_plugin_deinit");
+    if(!plugin.gsr_plugin_deinit) {
+        fprintf(stderr, "gsr error: gsr_plugins_load_plugin failed to find \"gsr_plugin_deinit\" in plugin \"%s\"\n", plugin_filepath);
+        goto fail;
+    }
+
+    if(!plugin.gsr_plugin_init(&self->init_params, &plugin.data)) {
+        fprintf(stderr, "gsr error: gsr_plugins_load_plugin failed to load plugin \"%s\", gsr_plugin_init in the plugin failed\n", plugin_filepath);
+        goto fail;
+    }
+
+    if(!plugin.data.name) {
+        fprintf(stderr, "gsr error: gsr_plugins_load_plugin failed to load plugin \"%s\", the plugin didn't set the name (gsr_plugin_init_return.name)\n", plugin_filepath);
+        goto fail;
+    }
+
+    if(plugin.data.version == 0) {
+        fprintf(stderr, "gsr error: gsr_plugins_load_plugin failed to load plugin \"%s\", the plugin didn't set the version (gsr_plugin_init_return.version)\n", plugin_filepath);
+        goto fail;
+    }
+
+    fprintf(stderr, "gsr info: loaded plugin: %s, name: %s, version: %u\n", plugin_filepath, plugin.data.name, plugin.data.version);
+    self->plugins[self->num_plugins] = plugin;
+    ++self->num_plugins;
+    return true;
+
+    fail:
+    dlclose(plugin.lib);
+    return false;
+}
+
+void gsr_plugins_draw(gsr_plugins *self) {
+    const gsr_plugin_draw_params params = {
+        .width = self->init_params.width,
+        .height = self->init_params.height,
+    };
+
+    self->egl->glBindFramebuffer(GL_FRAMEBUFFER, self->color_conversion.framebuffers[0]);
+    self->egl->glViewport(0, 0, self->init_params.width, self->init_params.height);
+
+    for(int i = 0; i < self->num_plugins; ++i) {
+        const gsr_plugin *plugin = &self->plugins[i];
+        if(plugin->data.draw)
+            plugin->data.draw(&params, plugin->data.userdata);
+    }
+
+    self->egl->glBindFramebuffer(GL_FRAMEBUFFER, 0);
+}

src/replay_buffer/replay_buffer.c

@@ -0,0 +1,57 @@
+#include "../../include/replay_buffer/replay_buffer.h"
+#include "../../include/replay_buffer/replay_buffer_ram.h"
+#include "../../include/replay_buffer/replay_buffer_disk.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+
+gsr_replay_buffer* gsr_replay_buffer_create(gsr_replay_storage replay_storage, const char *replay_directory, double replay_buffer_time, size_t replay_buffer_num_packets) {
+    gsr_replay_buffer *replay_buffer = NULL;
+    switch(replay_storage) {
+        case GSR_REPLAY_STORAGE_RAM:
+            replay_buffer = gsr_replay_buffer_ram_create(replay_buffer_num_packets);
+            break;
+        case GSR_REPLAY_STORAGE_DISK:
+            replay_buffer = gsr_replay_buffer_disk_create(replay_directory, replay_buffer_time);
+            break;
+    }
+    return replay_buffer;
+}
+
+void gsr_replay_buffer_destroy(gsr_replay_buffer *self) {
+    self->destroy(self);
+    free(self);
+}
+
+bool gsr_replay_buffer_append(gsr_replay_buffer *self, const AVPacket *av_packet, double timestamp) {
+    return self->append(self, av_packet, timestamp);
+}
+
+void gsr_replay_buffer_clear(gsr_replay_buffer *self) {
+    self->clear(self);
+}
+
+AVPacket* gsr_replay_buffer_iterator_get_packet(gsr_replay_buffer *self, gsr_replay_buffer_iterator iterator) {
+    return self->iterator_get_packet(self, iterator);
+}
+
+uint8_t* gsr_replay_buffer_iterator_get_packet_data(gsr_replay_buffer *self, gsr_replay_buffer_iterator iterator) {
+    return self->iterator_get_packet_data(self, iterator);
+}
+
+gsr_replay_buffer* gsr_replay_buffer_clone(gsr_replay_buffer *self) {
+    return self->clone(self);
+}
+
+gsr_replay_buffer_iterator gsr_replay_buffer_find_packet_index_by_time_passed(gsr_replay_buffer *self, int seconds) {
+    return self->find_packet_index_by_time_passed(self, seconds);
+}
+
+gsr_replay_buffer_iterator gsr_replay_buffer_find_keyframe(gsr_replay_buffer *self, gsr_replay_buffer_iterator start_iterator, int stream_index, bool invert_stream_index) {
+    return self->find_keyframe(self, start_iterator, stream_index, invert_stream_index);
+}
+
+bool gsr_replay_buffer_iterator_next(gsr_replay_buffer *self, gsr_replay_buffer_iterator *iterator) {
+    return self->iterator_next(self, iterator);
+}

src/replay_buffer/replay_buffer_disk.c

@@ -0,0 +1,423 @@
+#include "../../include/replay_buffer/replay_buffer_disk.h"
+#include "../../include/utils.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <time.h>
+#include <errno.h>
+#include <assert.h>
+
+#define REPLAY_BUFFER_FILE_SIZE_BYTES 1024 * 1024 * 256 /* 256MB */
+#define FILE_PREFIX "Replay"
+
+static void gsr_replay_buffer_disk_set_impl_funcs(gsr_replay_buffer_disk *self);
+
+static void gsr_av_packet_disk_init(gsr_av_packet_disk *self, const AVPacket *av_packet, size_t data_index, double timestamp) {
+    self->packet = *av_packet;
+    self->packet.data = NULL;
+    self->data_index = data_index;
+    self->timestamp = timestamp;
+}
+
+static gsr_replay_buffer_file* gsr_replay_buffer_file_create(char *replay_directory, size_t replay_storage_counter, double timestamp, int *replay_storage_fd) {
+    gsr_replay_buffer_file *self = calloc(1, sizeof(gsr_replay_buffer_file));
+    if(!self) {
+        fprintf(stderr, "gsr error: gsr_av_packet_file_init: failed to create buffer file\n");
+        return NULL;
+    }
+
+    if(create_directory_recursive(replay_directory) != 0) {
+        fprintf(stderr, "gsr error: gsr_av_packet_file_init: failed to create replay directory: %s\n", replay_directory);
+        free(self);
+        return NULL;
+    }
+
+    char filename[PATH_MAX];
+    snprintf(filename, sizeof(filename), "%s/%s_%d.gsr", replay_directory, FILE_PREFIX, (int)replay_storage_counter);
+    *replay_storage_fd = creat(filename, 0700);
+    if(*replay_storage_fd <= 0) {
+        fprintf(stderr, "gsr error: gsr_av_packet_file_init: failed to create replay file: %s\n", filename);
+        free(self);
+        return NULL;
+    }
+
+    self->id = replay_storage_counter;
+    self->start_timestamp = timestamp;
+    self->end_timestamp = timestamp;
+    self->ref_counter = 1;
+    self->fd = -1;
+
+    self->packets = NULL;
+    self->capacity_num_packets = 0;
+    self->num_packets = 0;
+    return self;
+}
+
+static gsr_replay_buffer_file* gsr_replay_buffer_file_ref(gsr_replay_buffer_file *self) {
+    if(self->ref_counter >= 1)
+        ++self->ref_counter;
+    return self;
+}
+
+static void gsr_replay_buffer_file_free(gsr_replay_buffer_file *self, const char *replay_directory) {
+    self->ref_counter = 0;
+
+    if(self->fd > 0) {
+        close(self->fd);
+        self->fd = -1;
+    }
+
+    char filename[PATH_MAX];
+    snprintf(filename, sizeof(filename), "%s/%s_%d.gsr", replay_directory, FILE_PREFIX, (int)self->id);
+    remove(filename);
+
+    if(self->packets) {
+        free(self->packets);
+        self->packets = NULL;
+    }
+    self->num_packets = 0;
+    self->capacity_num_packets = 0;
+
+    free(self);
+}
+
+static void gsr_replay_buffer_file_unref(gsr_replay_buffer_file *self, const char *replay_directory) {
+    if(self->ref_counter > 0)
+        --self->ref_counter;
+
+    if(self->ref_counter <= 0)
+        gsr_replay_buffer_file_free(self, replay_directory);
+}
+
+static void gsr_replay_buffer_disk_clear(gsr_replay_buffer *replay_buffer) {
+    gsr_replay_buffer_disk *self = (gsr_replay_buffer_disk*)replay_buffer;
+
+    for(size_t i = 0; i < self->num_files; ++i) {
+        gsr_replay_buffer_file_unref(self->files[i], self->replay_directory);
+    }
+    self->num_files = 0;
+
+    if(self->storage_fd > 0) {
+        close(self->storage_fd);
+        self->storage_fd = 0;
+    }
+
+    self->storage_num_bytes_written = 0;
+}
+
+static void gsr_replay_buffer_disk_destroy(gsr_replay_buffer *replay_buffer) {
+    gsr_replay_buffer_disk *self = (gsr_replay_buffer_disk*)replay_buffer;
+    gsr_replay_buffer_disk_clear(replay_buffer);
+
+    if(self->owns_directory) {
+        remove(self->replay_directory);
+        self->owns_directory = false;
+    }
+}
+
+static bool file_write_all(int fd, const uint8_t *data, size_t size, size_t *bytes_written_total) {
+    *bytes_written_total = 0;
+    while(*bytes_written_total < size) {
+        const ssize_t bytes_written = write(fd, data + *bytes_written_total, size - *bytes_written_total);
+        if(bytes_written == -1) {
+            if(errno == EAGAIN)
+                continue;
+            else
+                return false;
+        }
+        *bytes_written_total += bytes_written;
+    }
+    return true;
+}
+
+static bool gsr_replay_buffer_disk_create_next_file(gsr_replay_buffer_disk *self, double timestamp) {
+    if(self->num_files + 1 >= GSR_REPLAY_BUFFER_CAPACITY_NUM_FILES) {
+        fprintf(stderr, "gsr error: gsr_replay_buffer_disk_create_next_file: too many replay buffer files created! (> %d), either reduce the replay buffer time or report this as a bug\n", (int)GSR_REPLAY_BUFFER_CAPACITY_NUM_FILES);
+        return false;
+    }
+
+    gsr_replay_buffer_file *replay_buffer_file = gsr_replay_buffer_file_create(self->replay_directory, self->storage_counter, timestamp, &self->storage_fd);
+    if(!replay_buffer_file)
+        return false;
+
+    self->files[self->num_files] = replay_buffer_file;
+    ++self->num_files;
+    ++self->storage_counter;
+    return true;
+}
+
+static bool gsr_replay_buffer_disk_append_to_current_file(gsr_replay_buffer_disk *self, const AVPacket *av_packet, double timestamp) {
+    gsr_replay_buffer_file *replay_buffer_file = self->files[self->num_files - 1];
+    replay_buffer_file->end_timestamp = timestamp;
+
+    if(replay_buffer_file->num_packets + 1 >= replay_buffer_file->capacity_num_packets) {
+        size_t new_capacity_num_packets = replay_buffer_file->capacity_num_packets * 2;
+        if(new_capacity_num_packets == 0)
+            new_capacity_num_packets = 256;
+
+        void *new_packets = realloc(replay_buffer_file->packets, new_capacity_num_packets * sizeof(gsr_av_packet_disk));
+        if(!new_packets) {
+            fprintf(stderr, "gsr error: gsr_replay_buffer_disk_append_to_current_file: failed to reallocate replay buffer file packets\n");
+            return false;
+        }
+
+        replay_buffer_file->capacity_num_packets = new_capacity_num_packets;
+        replay_buffer_file->packets = new_packets;
+    }
+
+    gsr_av_packet_disk *packet = &replay_buffer_file->packets[replay_buffer_file->num_packets];
+    gsr_av_packet_disk_init(packet, av_packet, self->storage_num_bytes_written, timestamp);
+    ++replay_buffer_file->num_packets;
+
+    size_t bytes_written = 0;
+    const bool file_written = file_write_all(self->storage_fd, av_packet->data, av_packet->size, &bytes_written);
+    self->storage_num_bytes_written += bytes_written;
+    if(self->storage_num_bytes_written >= REPLAY_BUFFER_FILE_SIZE_BYTES) {
+        self->storage_num_bytes_written = 0;
+        close(self->storage_fd);
+        self->storage_fd = 0;
+    }
+
+    return file_written;
+}
+
+static void gsr_replay_buffer_disk_remove_first_file(gsr_replay_buffer_disk *self) {
+    gsr_replay_buffer_file_unref(self->files[0], self->replay_directory);
+    for(size_t i = 1; i < self->num_files; ++i) {
+        self->files[i - 1] = self->files[i];
+    }
+    --self->num_files;
+}
+
+static bool gsr_replay_buffer_disk_append(gsr_replay_buffer *replay_buffer, const AVPacket *av_packet, double timestamp) {
+    gsr_replay_buffer_disk *self = (gsr_replay_buffer_disk*)replay_buffer;
+    bool success = false;
+
+    if(self->storage_fd <= 0) {
+        if(!gsr_replay_buffer_disk_create_next_file(self, timestamp))
+            goto done;
+    }
+
+    const bool data_written = gsr_replay_buffer_disk_append_to_current_file(self, av_packet, timestamp);
+
+    if(self->num_files > 1) {
+        const double buffer_time_accumulated = timestamp - self->files[1]->start_timestamp;
+        if(buffer_time_accumulated >= self->replay_buffer_time)
+            gsr_replay_buffer_disk_remove_first_file(self);
+    }
+
+    success = data_written;
+
+    done:
+    return success;
+}
+
+static AVPacket* gsr_replay_buffer_disk_iterator_get_packet(gsr_replay_buffer *replay_buffer, gsr_replay_buffer_iterator iterator) {
+    gsr_replay_buffer_disk *self = (gsr_replay_buffer_disk*)replay_buffer;
+    assert(iterator.file_index < self->num_files);
+    assert(iterator.packet_index < self->files[iterator.file_index]->num_packets);
+    return &self->files[iterator.file_index]->packets[iterator.packet_index].packet;
+}
+
+static uint8_t* gsr_replay_buffer_disk_iterator_get_packet_data(gsr_replay_buffer *replay_buffer, gsr_replay_buffer_iterator iterator) {
+    gsr_replay_buffer_disk *self = (gsr_replay_buffer_disk*)replay_buffer;
+    assert(iterator.file_index < self->num_files);
+    gsr_replay_buffer_file *file = self->files[iterator.file_index];
+    assert(iterator.packet_index < file->num_packets);
+
+    if(file->fd <= 0) {
+        char filename[PATH_MAX];
+        snprintf(filename, sizeof(filename), "%s/%s_%d.gsr", self->replay_directory, FILE_PREFIX, (int)file->id);
+        file->fd = open(filename, O_RDONLY);
+        if(file->fd <= 0) {
+            fprintf(stderr, "gsr error: gsr_replay_buffer_disk_iterator_get_packet_data: failed to open file\n");
+            return NULL;
+        }
+    }
+
+    const gsr_av_packet_disk *packet = &self->files[iterator.file_index]->packets[iterator.packet_index];
+    if(lseek(file->fd, packet->data_index, SEEK_SET) == -1) {
+        fprintf(stderr, "gsr error: gsr_replay_buffer_disk_iterator_get_packet_data: failed to seek\n");
+        return NULL;
+    }
+
+    uint8_t *packet_data = malloc(packet->packet.size);
+    if(read(file->fd, packet_data, packet->packet.size) != packet->packet.size) {
+        fprintf(stderr, "gsr error: gsr_replay_buffer_disk_iterator_get_packet_data: failed to read data from file\n");
+        free(packet_data);
+        return NULL;
+    }
+
+    return packet_data;
+}
+
+static gsr_replay_buffer* gsr_replay_buffer_disk_clone(gsr_replay_buffer *replay_buffer) {
+    gsr_replay_buffer_disk *self = (gsr_replay_buffer_disk*)replay_buffer;
+    gsr_replay_buffer_disk *destination = calloc(1, sizeof(gsr_replay_buffer_disk));
+    if(!destination)
+        return NULL;
+
+    gsr_replay_buffer_disk_set_impl_funcs(destination);
+
+    destination->replay_buffer_time = self->replay_buffer_time;
+    destination->storage_counter = self->storage_counter;
+    destination->storage_num_bytes_written = self->storage_num_bytes_written;
+    destination->storage_fd = 0; // We only want to read from the clone. If there is a need to write to it in the future then TODO change this
+
+    for(size_t i = 0; i < self->num_files; ++i) {
+        destination->files[i] = gsr_replay_buffer_file_ref(self->files[i]);
+    }
+    destination->num_files = self->num_files;
+
+    snprintf(destination->replay_directory, sizeof(destination->replay_directory), "%s", self->replay_directory);
+    destination->owns_directory = false;
+
+    return (gsr_replay_buffer*)destination;
+}
+
+/* Binary search */
+static size_t gsr_replay_buffer_file_find_packet_index_by_time_passed(const gsr_replay_buffer_file *self, int seconds) {
+    const double now = clock_get_monotonic_seconds();
+    if(self->num_packets == 0) {
+        return 0;
+    }
+
+    size_t lower_bound = 0;
+    size_t upper_bound = self->num_packets;
+    size_t index = 0;
+
+    for(;;) {
+        index = lower_bound + (upper_bound - lower_bound) / 2;
+        const gsr_av_packet_disk *packet = &self->packets[index];
+        const double time_passed_since_packet = now - packet->timestamp;
+        if(time_passed_since_packet >= seconds) {
+            if(lower_bound == index)
+                break;
+            lower_bound = index;
+        } else {
+            if(upper_bound == index)
+                break;
+            upper_bound = index;
+        }
+    }
+
+    return index;
+}
+
+/* Binary search */
+static gsr_replay_buffer_iterator gsr_replay_buffer_disk_find_file_index_by_time_passed(gsr_replay_buffer *replay_buffer, int seconds) {
+    gsr_replay_buffer_disk *self = (gsr_replay_buffer_disk*)replay_buffer;
+
+    const double now = clock_get_monotonic_seconds();
+    if(self->num_files == 0) {
+        return (gsr_replay_buffer_iterator){0, 0};
+    }
+
+    size_t lower_bound = 0;
+    size_t upper_bound = self->num_files;
+    size_t file_index = 0;
+
+    for(;;) {
+        file_index = lower_bound + (upper_bound - lower_bound) / 2;
+        const gsr_replay_buffer_file *file = self->files[file_index];
+        const double time_passed_since_file_start = now - file->start_timestamp;
+        const double time_passed_since_file_end = now - file->end_timestamp;
+        if(time_passed_since_file_start >= seconds && time_passed_since_file_end <= seconds) {
+            break;
+        } else if(time_passed_since_file_start >= seconds) {
+            if(lower_bound == file_index)
+                break;
+            lower_bound = file_index;
+        } else {
+            if(upper_bound == file_index)
+                break;
+            upper_bound = file_index;
+        }
+    }
+
+    const gsr_replay_buffer_file *file = self->files[file_index];
+    const size_t packet_index = gsr_replay_buffer_file_find_packet_index_by_time_passed(file, seconds);
+
+    return (gsr_replay_buffer_iterator){packet_index, file_index};
+}
+
+static gsr_replay_buffer_iterator gsr_replay_buffer_disk_find_keyframe(gsr_replay_buffer *replay_buffer, gsr_replay_buffer_iterator start_iterator, int stream_index, bool invert_stream_index) {
+    gsr_replay_buffer_disk *self = (gsr_replay_buffer_disk*)replay_buffer;
+    gsr_replay_buffer_iterator keyframe_iterator = {(size_t)-1, 0};
+    size_t packet_index = start_iterator.packet_index;
+    for(size_t file_index = start_iterator.file_index; file_index < self->num_files; ++file_index) {
+        const gsr_replay_buffer_file *file = self->files[file_index];
+        for(; packet_index < file->num_packets; ++packet_index) {
+            const gsr_av_packet_disk *packet = &file->packets[packet_index];
+            if((packet->packet.flags & AV_PKT_FLAG_KEY) && (invert_stream_index ? packet->packet.stream_index != stream_index : packet->packet.stream_index == stream_index)) {
+                keyframe_iterator.packet_index = packet_index;
+                keyframe_iterator.file_index = file_index;
+                goto done;
+            }
+        }
+        packet_index = 0;
+    }
+    done:
+    return keyframe_iterator;
+}
+
+static bool gsr_replay_buffer_disk_iterator_next(gsr_replay_buffer *replay_buffer, gsr_replay_buffer_iterator *iterator) {
+    gsr_replay_buffer_disk *self = (gsr_replay_buffer_disk*)replay_buffer;
+    if(iterator->file_index >= self->num_files)
+        return false;
+
+    if(iterator->packet_index + 1 >= self->files[iterator->file_index]->num_packets) {
+        if(iterator->file_index + 1 >= self->num_files)
+            return false;
+
+        if(self->files[iterator->file_index + 1]->num_packets == 0)
+            return false;
+
+        ++iterator->file_index;
+        iterator->packet_index = 0;
+        return true;
+    } else {
+        ++iterator->packet_index;
+        return true;
+    }
+}
+
+static void get_current_time(char *time_str, size_t time_str_size) {
+    time_t now = time(NULL);
+    struct tm *t = localtime(&now);
+    strftime(time_str, time_str_size - 1, "%Y-%m-%d_%H-%M-%S", t);
+}
+
+static void gsr_replay_buffer_disk_set_impl_funcs(gsr_replay_buffer_disk *self) {
+    self->replay_buffer.destroy = gsr_replay_buffer_disk_destroy;
+    self->replay_buffer.append = gsr_replay_buffer_disk_append;
+    self->replay_buffer.clear = gsr_replay_buffer_disk_clear;
+    self->replay_buffer.iterator_get_packet = gsr_replay_buffer_disk_iterator_get_packet;
+    self->replay_buffer.iterator_get_packet_data = gsr_replay_buffer_disk_iterator_get_packet_data;
+    self->replay_buffer.clone = gsr_replay_buffer_disk_clone;
+    self->replay_buffer.find_packet_index_by_time_passed = gsr_replay_buffer_disk_find_file_index_by_time_passed;
+    self->replay_buffer.find_keyframe = gsr_replay_buffer_disk_find_keyframe;
+    self->replay_buffer.iterator_next = gsr_replay_buffer_disk_iterator_next;
+}
+
+gsr_replay_buffer* gsr_replay_buffer_disk_create(const char *replay_directory, double replay_buffer_time) {
+    assert(replay_buffer_time > 0);
+    gsr_replay_buffer_disk *replay_buffer = calloc(1, sizeof(gsr_replay_buffer_disk));
+    if(!replay_buffer)
+        return NULL;
+
+    char time_str[128];
+    get_current_time(time_str, sizeof(time_str));
+
+    replay_buffer->num_files = 0;
+    replay_buffer->storage_counter = 0;
+    replay_buffer->replay_buffer_time = replay_buffer_time;
+    snprintf(replay_buffer->replay_directory, sizeof(replay_buffer->replay_directory), "%s/gsr-replay-%s.gsr", replay_directory, time_str);
+    replay_buffer->owns_directory = true;
+
+    gsr_replay_buffer_disk_set_impl_funcs(replay_buffer);
+    return (gsr_replay_buffer*)replay_buffer;
+}

src/replay_buffer/replay_buffer_ram.c

@@ -0,0 +1,237 @@
+#include "../../include/replay_buffer/replay_buffer_ram.h"
+#include "../../include/utils.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+
+#include <libavutil/mem.h>
+
+static void gsr_replay_buffer_ram_set_impl_funcs(gsr_replay_buffer_ram *self);
+
+static gsr_av_packet_ram* gsr_av_packet_ram_create(const AVPacket *av_packet, double timestamp) {
+    gsr_av_packet_ram *self = malloc(sizeof(gsr_av_packet_ram));
+    if(!self)
+        return NULL;
+
+    self->ref_counter = 1;
+    self->packet = *av_packet;
+    self->timestamp = timestamp;
+    // Why are we doing this you ask? there is a ffmpeg bug that causes cpu usage to increase over time when you have
+    // packets that are not being free'd until later. So we copy the packet data, free the packet and then reconstruct
+    // the packet later on when we need it, to keep packets alive only for a short period.
+    self->packet.data = av_memdup(av_packet->data, av_packet->size);
+    if(!self->packet.data) {
+        free(self);
+        return NULL;
+    }
+
+    return self;
+}
+
+static gsr_av_packet_ram* gsr_av_packet_ram_ref(gsr_av_packet_ram *self) {
+    if(self->ref_counter >= 1)
+        ++self->ref_counter;
+    return self;
+}
+
+static void gsr_av_packet_ram_free(gsr_av_packet_ram *self) {
+    self->ref_counter = 0;
+    if(self->packet.data) {
+        av_free(self->packet.data);
+        self->packet.data = NULL;
+    }
+    free(self);
+}
+
+static void gsr_av_packet_ram_unref(gsr_av_packet_ram *self) {
+    if(self->ref_counter >= 1)
+        --self->ref_counter;
+
+    if(self->ref_counter <= 0)
+        gsr_av_packet_ram_free(self);
+}
+
+static void gsr_replay_buffer_ram_destroy(gsr_replay_buffer *replay_buffer) {
+    gsr_replay_buffer_ram *self = (gsr_replay_buffer_ram*)replay_buffer;
+    for(size_t i = 0; i < self->num_packets; ++i) {
+        if(self->packets[i]) {
+            gsr_av_packet_ram_unref(self->packets[i]);
+            self->packets[i] = NULL;
+        }
+    }
+    self->num_packets = 0;
+
+    if(self->packets) {
+        free(self->packets);
+        self->packets = NULL;
+    }
+
+    self->capacity_num_packets = 0;
+    self->index = 0;
+}
+
+static bool gsr_replay_buffer_ram_append(gsr_replay_buffer *replay_buffer, const AVPacket *av_packet, double timestamp) {
+    gsr_replay_buffer_ram *self = (gsr_replay_buffer_ram*)replay_buffer;
+    gsr_av_packet_ram *packet = gsr_av_packet_ram_create(av_packet, timestamp);
+    if(!packet)
+        return false;
+
+    if(self->packets[self->index]) {
+        gsr_av_packet_ram_unref(self->packets[self->index]);
+        self->packets[self->index] = NULL;
+    }
+    self->packets[self->index] = packet;
+
+    self->index = (self->index + 1) % self->capacity_num_packets;
+    ++self->num_packets;
+    if(self->num_packets > self->capacity_num_packets)
+        self->num_packets = self->capacity_num_packets;
+
+    return true;
+}
+
+static void gsr_replay_buffer_ram_clear(gsr_replay_buffer *replay_buffer) {
+    gsr_replay_buffer_ram *self = (gsr_replay_buffer_ram*)replay_buffer;
+    for(size_t i = 0; i < self->num_packets; ++i) {
+        if(self->packets[i]) {
+            gsr_av_packet_ram_unref(self->packets[i]);
+            self->packets[i] = NULL;
+        }
+    }
+    self->num_packets = 0;
+    self->index = 0;
+}
+
+static gsr_av_packet_ram* gsr_replay_buffer_ram_get_packet_at_index(gsr_replay_buffer *replay_buffer, size_t index) {
+    gsr_replay_buffer_ram *self = (gsr_replay_buffer_ram*)replay_buffer;
+    assert(index < self->num_packets);
+    size_t start_index = 0;
+    if(self->num_packets < self->capacity_num_packets)
+        start_index = self->num_packets - self->index;
+    else
+        start_index = self->index;
+
+    const size_t offset = (start_index + index) % self->capacity_num_packets;
+    return self->packets[offset];
+}
+
+static AVPacket* gsr_replay_buffer_ram_iterator_get_packet(gsr_replay_buffer *replay_buffer, gsr_replay_buffer_iterator iterator) {
+    return &gsr_replay_buffer_ram_get_packet_at_index(replay_buffer, iterator.packet_index)->packet;
+}
+
+static uint8_t* gsr_replay_buffer_ram_iterator_get_packet_data(gsr_replay_buffer *replay_buffer, gsr_replay_buffer_iterator iterator) {
+    (void)replay_buffer;
+    (void)iterator;
+    return NULL;
+}
+
+static gsr_replay_buffer* gsr_replay_buffer_ram_clone(gsr_replay_buffer *replay_buffer) {
+    gsr_replay_buffer_ram *self = (gsr_replay_buffer_ram*)replay_buffer;
+    gsr_replay_buffer_ram *destination = calloc(1, sizeof(gsr_replay_buffer_ram));
+    if(!destination)
+        return NULL;
+
+    gsr_replay_buffer_ram_set_impl_funcs(destination);
+
+    destination->capacity_num_packets = self->capacity_num_packets;
+    destination->index = self->index;
+    destination->packets = calloc(destination->capacity_num_packets, sizeof(gsr_av_packet_ram*));
+    if(!destination->packets) {
+        free(destination);
+        return NULL;
+    }
+
+    destination->num_packets = self->num_packets;
+    for(size_t i = 0; i < destination->num_packets; ++i) {
+        destination->packets[i] = gsr_av_packet_ram_ref(self->packets[i]);
+    }
+
+    return (gsr_replay_buffer*)destination;
+}
+
+/* Binary search */
+static gsr_replay_buffer_iterator gsr_replay_buffer_ram_find_packet_index_by_time_passed(gsr_replay_buffer *replay_buffer, int seconds) {
+    gsr_replay_buffer_ram *self = (gsr_replay_buffer_ram*)replay_buffer;
+
+    const double now = clock_get_monotonic_seconds();
+    if(self->num_packets == 0) {
+        return (gsr_replay_buffer_iterator){0, 0};
+    }
+
+    size_t lower_bound = 0;
+    size_t upper_bound = self->num_packets;
+    size_t index = 0;
+
+    for(;;) {
+        index = lower_bound + (upper_bound - lower_bound) / 2;
+        const gsr_av_packet_ram *packet = gsr_replay_buffer_ram_get_packet_at_index(replay_buffer, index);
+        const double time_passed_since_packet = now - packet->timestamp;
+        if(time_passed_since_packet >= seconds) {
+            if(lower_bound == index)
+                break;
+            lower_bound = index;
+        } else {
+            if(upper_bound == index)
+                break;
+            upper_bound = index;
+        }
+    }
+
+    return (gsr_replay_buffer_iterator){index, 0};
+}
+
+static gsr_replay_buffer_iterator gsr_replay_buffer_ram_find_keyframe(gsr_replay_buffer *replay_buffer, gsr_replay_buffer_iterator start_iterator, int stream_index, bool invert_stream_index) {
+    gsr_replay_buffer_ram *self = (gsr_replay_buffer_ram*)replay_buffer;
+    size_t keyframe_index = (size_t)-1;
+    for(size_t i = start_iterator.packet_index; i < self->num_packets; ++i) {
+        const gsr_av_packet_ram *packet = gsr_replay_buffer_ram_get_packet_at_index(replay_buffer, i);
+        if((packet->packet.flags & AV_PKT_FLAG_KEY) && (invert_stream_index ? packet->packet.stream_index != stream_index : packet->packet.stream_index == stream_index)) {
+            keyframe_index = i;
+            break;
+        }
+    }
+    return (gsr_replay_buffer_iterator){keyframe_index, 0};
+}
+
+static bool gsr_replay_buffer_ram_iterator_next(gsr_replay_buffer *replay_buffer, gsr_replay_buffer_iterator *iterator) {
+    gsr_replay_buffer_ram *self = (gsr_replay_buffer_ram*)replay_buffer;
+    if(iterator->packet_index + 1 < self->num_packets) {
+        ++iterator->packet_index;
+        return true;
+    } else {
+        return false;
+    }
+}
+
+static void gsr_replay_buffer_ram_set_impl_funcs(gsr_replay_buffer_ram *self) {
+    self->replay_buffer.destroy = gsr_replay_buffer_ram_destroy;
+    self->replay_buffer.append = gsr_replay_buffer_ram_append;
+    self->replay_buffer.clear = gsr_replay_buffer_ram_clear;
+    self->replay_buffer.iterator_get_packet = gsr_replay_buffer_ram_iterator_get_packet;
+    self->replay_buffer.iterator_get_packet_data = gsr_replay_buffer_ram_iterator_get_packet_data;
+    self->replay_buffer.clone = gsr_replay_buffer_ram_clone;
+    self->replay_buffer.find_packet_index_by_time_passed = gsr_replay_buffer_ram_find_packet_index_by_time_passed;
+    self->replay_buffer.find_keyframe = gsr_replay_buffer_ram_find_keyframe;
+    self->replay_buffer.iterator_next = gsr_replay_buffer_ram_iterator_next;
+}
+
+gsr_replay_buffer* gsr_replay_buffer_ram_create(size_t replay_buffer_num_packets) {
+    assert(replay_buffer_num_packets > 0);
+    gsr_replay_buffer_ram *replay_buffer = calloc(1, sizeof(gsr_replay_buffer_ram));
+    if(!replay_buffer)
+        return NULL;
+
+    replay_buffer->capacity_num_packets = replay_buffer_num_packets;
+    replay_buffer->num_packets = 0;
+    replay_buffer->index = 0;
+    replay_buffer->packets = calloc(replay_buffer->capacity_num_packets, sizeof(gsr_av_packet_ram*));
+    if(!replay_buffer->packets) {
+        gsr_replay_buffer_ram_destroy(&replay_buffer->replay_buffer);
+        free(replay_buffer);
+        return NULL;
+    }
+
+    gsr_replay_buffer_ram_set_impl_funcs(replay_buffer);
+    return (gsr_replay_buffer*)replay_buffer;
+}

src/shader.c

@@ -3,14 +3,16 @@
 #include <stdio.h>
 #include <assert.h>
 
+static bool print_compile_errors = false;
+
 static int min_int(int a, int b) {
     return a < b ? a : b;
 }
 
-static unsigned int loader_shader(gsr_egl *egl, unsigned int type, const char *source) {
+static unsigned int load_shader(gsr_egl *egl, unsigned int type, const char *source) {
     unsigned int shader_id = egl->glCreateShader(type);
     if(shader_id == 0) {
-        fprintf(stderr, "gsr error: loader_shader: failed to create shader, error: %d\n", egl->glGetError());
+        fprintf(stderr, "gsr error: load_shader: failed to create shader, error: %d\n", egl->glGetError());
         return 0;
     }
 
@@ -23,10 +25,10 @@ static unsigned int loader_shader(gsr_egl *egl, unsigned int type, const char *s
         int info_length = 0;
         egl->glGetShaderiv(shader_id, GL_INFO_LOG_LENGTH, &info_length);
         
-        if(info_length > 1) {
+        if(info_length > 1 && print_compile_errors) {
             char info_log[4096];
             egl->glGetShaderInfoLog(shader_id, min_int(4096, info_length), NULL, info_log);
-            fprintf(stderr, "gsr error: loader shader: failed to compile shader, error:\n%s\nshader source:\n%s\n", info_log, source);
+            fprintf(stderr, "gsr error: load_shader: failed to compile shader, error:\n%s\nshader source:\n%s\n", info_log, source);
         }
 
         egl->glDeleteShader(shader_id);
@@ -41,23 +43,24 @@ static unsigned int load_program(gsr_egl *egl, const char *vertex_shader, const
     unsigned int fragment_shader_id = 0;
     unsigned int program_id = 0;
     int linked = 0;
+    bool success = false;
 
     if(vertex_shader) {
-        vertex_shader_id = loader_shader(egl, GL_VERTEX_SHADER, vertex_shader);
+        vertex_shader_id = load_shader(egl, GL_VERTEX_SHADER, vertex_shader);
         if(vertex_shader_id == 0)
-            goto err;
+            goto done;
     }
 
     if(fragment_shader) {
-        fragment_shader_id = loader_shader(egl, GL_FRAGMENT_SHADER, fragment_shader);
+        fragment_shader_id = load_shader(egl, GL_FRAGMENT_SHADER, fragment_shader);
         if(fragment_shader_id == 0)
-            goto err;
+            goto done;
     }
 
     program_id = egl->glCreateProgram();
     if(program_id == 0) {
         fprintf(stderr, "gsr error: load_program: failed to create shader program, error: %d\n", egl->glGetError());
-        goto err;
+        goto done;
     }
 
     if(vertex_shader_id)
@@ -79,24 +82,21 @@ static unsigned int load_program(gsr_egl *egl, const char *vertex_shader, const
             fprintf(stderr, "gsr error: load program: linking shader program failed, error:\n%s\n", info_log);            
         }
 
-        goto err;
+        goto done;
     }
 
+    success = true;
+    done:
+
+    if(!success) {
+        if(program_id)
+            egl->glDeleteProgram(program_id);
+    }
     if(fragment_shader_id)
         egl->glDeleteShader(fragment_shader_id);
     if(vertex_shader_id)
         egl->glDeleteShader(vertex_shader_id);
-
     return program_id;
-
-    err:
-    if(program_id)
-        egl->glDeleteProgram(program_id);
-    if(fragment_shader_id)
-        egl->glDeleteShader(fragment_shader_id);
-    if(vertex_shader_id)
-        egl->glDeleteShader(vertex_shader_id);
-    return 0;
 }
 
 int gsr_shader_init(gsr_shader *self, gsr_egl *egl, const char *vertex_shader, const char *fragment_shader) {
@@ -105,7 +105,7 @@ int gsr_shader_init(gsr_shader *self, gsr_egl *egl, const char *vertex_shader, c
     self->program_id = 0;
 
     if(!vertex_shader && !fragment_shader) {
-        fprintf(stderr, "gsr error: gsr_shader_init: vertex shader and fragment shader can't be NULL at the same time\n");
+        fprintf(stderr, "gsr error: gsr_shader_init: vertex and fragment shader can't be NULL at the same time\n");
         return -1;
     }
 
@@ -141,3 +141,7 @@ void gsr_shader_use(gsr_shader *self) {
 void gsr_shader_use_none(gsr_shader *self) {
     self->egl->glUseProgram(0);
 }
+
+void gsr_shader_enable_debug_output(bool enable) {
+    print_compile_errors = enable;
+}

src/sound.cpp

@@ -59,17 +59,26 @@ struct pa_handle {
     std::mutex reconnect_mutex;
     DeviceType device_type;
     char stream_name[256];
+    char node_name[256];
     bool reconnect;
     double reconnect_last_tried_seconds;
 
     char device_name[DEVICE_NAME_MAX_SIZE];
     char default_output_device_name[DEVICE_NAME_MAX_SIZE];
     char default_input_device_name[DEVICE_NAME_MAX_SIZE];
+
+    pa_proplist *proplist;
+    bool connected;
 };
 
 static void pa_sound_device_free(pa_handle *p) {
     assert(p);
 
+    if(p->proplist) {
+        pa_proplist_free(p->proplist);
+        p->proplist = NULL;
+    }
+
     if (p->stream) {
         pa_stream_unref(p->stream);
         p->stream = NULL;
@@ -154,7 +163,7 @@ static bool startup_get_default_devices(pa_handle *p, const char *device_name) {
     }
 
     if(p->default_output_device_name[0] == '\0') {
-        fprintf(stderr, "Error: failed to find default audio output device\n");
+        fprintf(stderr, "gsr error: failed to find default audio output device\n");
         return false;
     }
 
@@ -186,10 +195,11 @@ static pa_handle* pa_sound_device_new(const char *server,
     p = pa_xnew0(pa_handle, 1);
     p->attr = *attr;
     p->ss = *ss;
+    snprintf(p->node_name, sizeof(p->node_name), "%s", name);
     snprintf(p->stream_name, sizeof(p->stream_name), "%s", stream_name);
 
     p->reconnect = true;
-    p->reconnect_last_tried_seconds = clock_get_monotonic_seconds() - 1000.0;
+    p->reconnect_last_tried_seconds = clock_get_monotonic_seconds() - (RECONNECT_TRY_TIMEOUT_SECONDS * 1000.0 * 2.0);
     p->default_output_device_name[0] = '\0';
     p->default_input_device_name[0] = '\0';
     p->device_type = DeviceType::STANDARD;
@@ -197,7 +207,7 @@ static pa_handle* pa_sound_device_new(const char *server,
     const int buffer_size = attr->fragsize;
     void *buffer = malloc(buffer_size);
     if(!buffer) {
-        fprintf(stderr, "Error: failed to allocate buffer for audio\n");
+        fprintf(stderr, "gsr error: failed to allocate buffer for audio\n");
         *rerror = -1;
         return NULL;
     }
@@ -206,10 +216,17 @@ static pa_handle* pa_sound_device_new(const char *server,
     p->output_length = buffer_size;
     p->output_index = 0;
 
+    p->proplist = pa_proplist_new();
+    pa_proplist_sets(p->proplist, PA_PROP_MEDIA_ROLE, "production");
+    if(strcmp(device_name, "") == 0) {
+        pa_proplist_sets(p->proplist, "node.autoconnect", "false");
+        pa_proplist_sets(p->proplist, "node.dont-reconnect", "true");
+    }
+
     if (!(p->mainloop = pa_mainloop_new()))
         goto fail;
 
-    if (!(p->context = pa_context_new(pa_mainloop_get_api(p->mainloop), name)))
+    if (!(p->context = pa_context_new_with_proplist(pa_mainloop_get_api(p->mainloop), p->node_name, p->proplist)))
         goto fail;
 
     if (pa_context_connect(p->context, server, PA_CONTEXT_NOFLAGS, NULL) < 0) {
@@ -239,6 +256,7 @@ static pa_handle* pa_sound_device_new(const char *server,
     if(pa)
         pa_operation_unref(pa);
 
+    p->connected = true;
     return p;
 
 fail:
@@ -248,19 +266,67 @@ fail:
     return NULL;
 }
 
+static void pa_sound_device_update_context_status(pa_handle *p) {
+    if(p->connected || !p->context || pa_context_get_state(p->context) != PA_CONTEXT_READY)
+        return;
+
+    p->connected = true;
+    pa_context_set_subscribe_callback(p->context, subscribe_cb, p);
+    pa_operation *pa = pa_context_subscribe(p->context, PA_SUBSCRIPTION_MASK_SERVER, NULL, NULL);
+    if(pa)
+        pa_operation_unref(pa);
+}
+
+static bool pa_sound_device_handle_context_recreate(pa_handle *p) {
+    if(p->context) {
+        pa_context_disconnect(p->context);
+        pa_context_unref(p->context);
+        p->context = NULL;
+        p->connected = false;
+    }
+
+    if (!(p->context = pa_context_new_with_proplist(pa_mainloop_get_api(p->mainloop), p->node_name, p->proplist))) {
+        fprintf(stderr, "gsr error: pa_context_new_with_proplist failed\n");
+        goto fail;
+    }
+
+    if(pa_context_connect(p->context, nullptr, PA_CONTEXT_NOFLAGS, NULL) < 0) {
+        fprintf(stderr, "gsr error: pa_context_connect failed\n");
+        goto fail;
+    }
+
+    pa_mainloop_iterate(p->mainloop, 0, NULL);
+    pa_sound_device_update_context_status(p);
+    return true;
+
+    fail:
+    if(p->context) {
+        pa_context_disconnect(p->context);
+        pa_context_unref(p->context);
+        p->context = NULL;
+    }
+    return false;
+}
+
 static bool pa_sound_device_should_reconnect(pa_handle *p, double now, char *device_name, size_t device_name_size) {
     std::lock_guard<std::mutex> lock(p->reconnect_mutex);
+
+    if(!p->reconnect && (!p->stream || !PA_STREAM_IS_GOOD(pa_stream_get_state(p->stream)))) {
+        p->reconnect = true;
+        p->reconnect_last_tried_seconds = now;
+    }
+
     if(p->reconnect && now - p->reconnect_last_tried_seconds >= RECONNECT_TRY_TIMEOUT_SECONDS) {
         p->reconnect_last_tried_seconds = now;
         // TODO: Size check
         snprintf(device_name, device_name_size, "%s", p->device_name);
         return true;
     }
+
     return false;
 }
 
 static bool pa_sound_device_handle_reconnect(pa_handle *p, char *device_name, size_t device_name_size, double now) {
-    int r;
     if(!pa_sound_device_should_reconnect(p, now, device_name, device_name_size))
         return true;
 
@@ -268,34 +334,33 @@ static bool pa_sound_device_handle_reconnect(pa_handle *p, char *device_name, si
         pa_stream_disconnect(p->stream);
         pa_stream_unref(p->stream);
         p->stream = NULL;
+
+        pa_sound_device_handle_context_recreate(p);
+        if(!p->connected)
+            return false;
     }
 
-    if(!(p->stream = pa_stream_new(p->context, p->stream_name, &p->ss, NULL))) {
+    pa_proplist *proplist = pa_proplist_new();
+    // This prevents microphone recording indicator from being shown on KDE
+    pa_proplist_sets(proplist, "node.virtual", "true");
+
+    if(!(p->stream = pa_stream_new_with_proplist(p->context, p->stream_name, &p->ss, NULL, proplist))) {
         //pa_context_errno(p->context);
+        pa_proplist_free(proplist);
         return false;
     }
 
-    r = pa_stream_connect_record(p->stream, device_name, &p->attr,
-        (pa_stream_flags_t)(PA_STREAM_INTERPOLATE_TIMING|PA_STREAM_ADJUST_LATENCY|PA_STREAM_AUTO_TIMING_UPDATE));
+    pa_proplist_free(proplist);
+
+    const int r = pa_stream_connect_record(p->stream, device_name, &p->attr,
+        (pa_stream_flags_t)(PA_STREAM_INTERPOLATE_TIMING|PA_STREAM_ADJUST_LATENCY|PA_STREAM_AUTO_TIMING_UPDATE|PA_STREAM_DONT_MOVE));
 
     if(r < 0) {
         //pa_context_errno(p->context);
         return false;
     }
 
-    for(;;) {
-        pa_stream_state_t state = pa_stream_get_state(p->stream);
-
-        if(state == PA_STREAM_READY)
-            break;
-
-        if(!PA_STREAM_IS_GOOD(state)) {
-            //pa_context_errno(p->context);
-            return false;
-        }
-
-        pa_mainloop_iterate(p->mainloop, 1, NULL);
-    }
+    pa_mainloop_iterate(p->mainloop, 0, NULL);
 
     std::lock_guard<std::mutex> lock(p->reconnect_mutex);
     p->reconnect = false;
@@ -314,7 +379,21 @@ static int pa_sound_device_read(pa_handle *p, double timeout_seconds) {
     pa_usec_t latency = 0;
     int negative = 0;
 
-    if(!pa_sound_device_handle_reconnect(p, device_name, sizeof(device_name), start_time))
+    pa_mainloop_iterate(p->mainloop, 0, NULL);
+
+    if(!p->context) {
+        if(!pa_sound_device_handle_context_recreate(p))
+            goto fail;
+    }
+
+    pa_sound_device_update_context_status(p);
+    if(!p->connected)
+        goto fail;
+
+    if(!pa_sound_device_handle_reconnect(p, device_name, sizeof(device_name), start_time) || !p->stream)
+        goto fail;
+
+    if(pa_stream_get_state(p->stream) != PA_STREAM_READY)
         goto fail;
 
     CHECK_DEAD_GOTO(p, rerror, fail);
@@ -410,7 +489,7 @@ static int audio_format_to_get_bytes_per_sample(AudioFormat audio_format) {
     return 2;
 }
 
-int sound_device_get_by_name(SoundDevice *device, const char *device_name, const char *description, unsigned int num_channels, unsigned int period_frame_size, AudioFormat audio_format) {
+int sound_device_get_by_name(SoundDevice *device, const char *node_name, const char *device_name, const char *description, unsigned int num_channels, unsigned int period_frame_size, AudioFormat audio_format) {
     pa_sample_spec ss;
     ss.format = audio_format_to_pulse_audio_format(audio_format);
     ss.rate = 48000;
@@ -424,9 +503,9 @@ int sound_device_get_by_name(SoundDevice *device, const char *device_name, const
     buffer_attr.maxlength = buffer_attr.fragsize;
 
     int error = 0;
-    pa_handle *handle = pa_sound_device_new(nullptr, description, device_name, description, &ss, &buffer_attr, &error);
+    pa_handle *handle = pa_sound_device_new(nullptr, node_name, device_name, description, &ss, &buffer_attr, &error);
     if(!handle) {
-        fprintf(stderr, "Error: pa_sound_device_new() failed: %s. Audio input device %s might not be valid\n", pa_strerror(error), device_name);
+        fprintf(stderr, "gsr error: pa_sound_device_new() failed: %s. Audio input device %s might not be valid\n", pa_strerror(error), device_name);
         return -1;
     }
 

src/utils.c

@@ -1,5 +1,6 @@
 #include "../include/utils.h"
 #include "../include/window/window.h"
+#include "../include/capture/capture.h"
 
 #include <time.h>
 #include <string.h>
@@ -14,13 +15,13 @@
 
 #include <xf86drmMode.h>
 #include <xf86drm.h>
-#include <libdrm/drm_fourcc.h>
 #include <X11/Xatom.h>
 #include <X11/extensions/Xrandr.h>
-#include <va/va_drmcommon.h>
 #include <libavcodec/avcodec.h>
 #include <libavutil/hwcontext_vaapi.h>
 
+#define DRM_NUM_BUF_ATTRS 4
+
 double clock_get_monotonic_seconds(void) {
     struct timespec ts;
     ts.tv_sec = 0;
@@ -110,7 +111,7 @@ void for_each_active_monitor_output_x11_not_cached(Display *display, active_moni
                 // but gpu screen recorder captures the drm framebuffer instead of x11 api. This drm framebuffer which doesn't increase in size when using xrandr scaling.
                 // Maybe a better option would be to get the drm crtc size instead.
                 const XRRModeInfo *mode_info = get_mode_info(screen_res, crt_info->mode);
-                if(mode_info && out_info->nameLen < (int)sizeof(display_name)) {
+                if(mode_info) {
                     snprintf(display_name, sizeof(display_name), "%.*s", (int)out_info->nameLen, out_info->name);
                     const gsr_monitor_rotation rotation = x11_rotation_to_gsr_rotation(crt_info->rotation);
                     const vec2i monitor_size = get_monitor_size_rotated(mode_info->width, mode_info->height, rotation);
@@ -120,6 +121,8 @@ void for_each_active_monitor_output_x11_not_cached(Display *display, active_moni
                         .name_len = out_info->nameLen,
                         .pos = { .x = crt_info->x, .y = crt_info->y },
                         .size = monitor_size,
+                        .logical_pos = { .x = crt_info->x, .y = crt_info->y },
+                        .logical_size = monitor_size,
                         .connector_id = x11_output_get_connector_id(display, screen_res->outputs[i], randr_connector_id_atom),
                         .rotation = rotation,
                         .monitor_identifier = out_info->crtc
@@ -148,25 +151,28 @@ int get_connector_type_by_name(const char *name) {
         return 3;
     else if(len >= 4 && strncmp(name, "eDP-", 4) == 0)
         return 4;
+    else if(len >= 4 && strncmp(name, "DVI-", 4) == 0)
+        return 5;
     else
         return -1;
 }
 
-drm_connector_type_count* drm_connector_types_get_index(drm_connector_type_count *type_counts, int *num_type_counts, int connector_type) {
-    for(int i = 0; i < *num_type_counts; ++i) {
-        if(type_counts[i].type == connector_type)
-            return &type_counts[i];
+int get_connector_type_id_by_name(const char *name) {
+    int len = strlen(name);
+    int num_start = 0;
+    for(int i = len - 1; i >= 0; --i) {
+        const bool is_num = name[i] >= '0' && name[i] <= '9';
+        if(!is_num) {
+            num_start = i + 1;
+            break;
+        }
     }
 
-    if(*num_type_counts == CONNECTOR_TYPE_COUNTS)
-        return NULL;
+    const int num_len = len - num_start;
+    if(num_len <= 0)
+        return -1;
 
-    const int index = *num_type_counts;
-    type_counts[index].type = connector_type;
-    type_counts[index].count = 0;
-    type_counts[index].count_active = 0;
-    ++*num_type_counts;
-    return &type_counts[index];
+    return atoi(name + num_start);
 }
 
 uint32_t monitor_identifier_from_type_and_count(int monitor_type_index, int monitor_type_count) {
@@ -197,9 +203,6 @@ static void for_each_active_monitor_output_drm(const char *card_path, active_mon
 
     drmSetClientCap(fd, DRM_CLIENT_CAP_ATOMIC, 1);
 
-    drm_connector_type_count type_counts[CONNECTOR_TYPE_COUNTS];
-    int num_type_counts = 0;
-
     char display_name[256];
     drmModeResPtr resources = drmModeGetResources(fd);
     if(resources) {
@@ -208,35 +211,31 @@ static void for_each_active_monitor_output_drm(const char *card_path, active_mon
             if(!connector)
                 continue;
 
-            drm_connector_type_count *connector_type = drm_connector_types_get_index(type_counts, &num_type_counts, connector->connector_type);
-            const char *connection_name = drmModeGetConnectorTypeName(connector->connector_type);
-            const int connection_name_len = strlen(connection_name);
-            if(connector_type)
-                ++connector_type->count;
-
             if(connector->connection != DRM_MODE_CONNECTED) {
                 drmModeFreeConnector(connector);
                 continue;
             }
 
-            if(connector_type)
-                ++connector_type->count_active;
-
             uint64_t crtc_id = 0;
             connector_get_property_by_name(fd, connector, "CRTC_ID", &crtc_id);
 
             drmModeCrtcPtr crtc = drmModeGetCrtc(fd, crtc_id);
-            if(connector_type && crtc_id > 0 && crtc && connection_name_len + 5 < (int)sizeof(display_name)) {
-                const int display_name_len = snprintf(display_name, sizeof(display_name), "%s-%d", connection_name, connector_type->count);
+            const char *connection_name = drmModeGetConnectorTypeName(connector->connector_type);
+
+            if(connection_name && crtc_id > 0 && crtc) {
                 const int connector_type_index_name = get_connector_type_by_name(display_name);
-                gsr_monitor monitor = {
+                const int display_name_len = snprintf(display_name, sizeof(display_name), "%s-%u", connection_name, connector->connector_type_id);
+
+                const gsr_monitor monitor = {
                     .name = display_name,
                     .name_len = display_name_len,
                     .pos = { .x = crtc->x, .y = crtc->y },
                     .size = { .x = (int)crtc->width, .y = (int)crtc->height },
+                    .logical_pos = { .x = crtc->x, .y = crtc->y },
+                    .logical_size = { .x = (int)crtc->width, .y = (int)crtc->height },
                     .connector_id = connector->connector_id,
                     .rotation = GSR_MONITOR_ROT_0,
-                    .monitor_identifier = connector_type_index_name != -1 ? monitor_identifier_from_type_and_count(connector_type_index_name, connector_type->count_active) : 0
+                    .monitor_identifier = connector_type_index_name != -1 ? monitor_identifier_from_type_and_count(connector_type_index_name, connector->connector_type_id) : 0
                 };
                 callback(&monitor, userdata);
             }
@@ -269,6 +268,8 @@ static void get_monitor_by_name_callback(const gsr_monitor *monitor, void *userd
     if(!data->found_monitor && strcmp(data->name, monitor->name) == 0) {
         data->monitor->pos = monitor->pos;
         data->monitor->size = monitor->size;
+        data->monitor->logical_pos = monitor->logical_pos;
+        data->monitor->logical_size = monitor->logical_size;
         data->monitor->connector_id = monitor->connector_id;
         data->monitor->rotation = monitor->rotation;
         data->monitor->monitor_identifier = monitor->monitor_identifier;
@@ -293,13 +294,9 @@ typedef struct {
     bool match_found;
 } get_monitor_by_connector_id_userdata;
 
-static bool vec2i_eql(vec2i a, vec2i b) {
-    return a.x == b.x && a.y == b.y;
-}
-
-static void get_monitor_by_name_and_size_callback(const gsr_monitor *monitor, void *userdata) {
+static void get_monitor_by_name_wayland_callback(const gsr_monitor *monitor, void *userdata) {
     get_monitor_by_connector_id_userdata *data = (get_monitor_by_connector_id_userdata*)userdata;
-    if(monitor->name && data->monitor->name && strcmp(monitor->name, data->monitor->name) == 0 && vec2i_eql(monitor->size, data->monitor->size)) {
+    if(monitor->name && data->monitor->name && strcmp(monitor->name, data->monitor->name) == 0) {
         data->rotation = monitor->rotation;
         data->position = monitor->pos;
         data->match_found = true;
@@ -328,7 +325,7 @@ bool drm_monitor_get_display_server_data(const gsr_window *window, const gsr_mon
             userdata.rotation = GSR_MONITOR_ROT_0;
             userdata.position = (vec2i){0, 0};
             userdata.match_found = false;
-            gsr_window_for_each_active_monitor_output_cached(window, get_monitor_by_name_and_size_callback, &userdata);
+            gsr_window_for_each_active_monitor_output_cached(window, get_monitor_by_name_wayland_callback, &userdata);
             if(userdata.match_found) {
                 *monitor_rotation = userdata.rotation;
                 *monitor_position = userdata.position;
@@ -364,13 +361,9 @@ bool gl_get_gpu_info(gsr_egl *egl, gsr_gpu_info *info) {
     bool supported = true;
     const unsigned char *gl_vendor = egl->glGetString(GL_VENDOR);
     const unsigned char *gl_renderer = egl->glGetString(GL_RENDERER);
-    const unsigned char *gl_version = egl->glGetString(GL_VERSION);
 
     info->gpu_version = 0;
     info->is_steam_deck = false;
-    info->driver_major = 0;
-    info->driver_minor = 0;
-    info->driver_patch = 0;
 
     if(!gl_vendor) {
         fprintf(stderr, "gsr error: failed to get gpu vendor\n");
@@ -396,6 +389,8 @@ bool gl_get_gpu_info(gsr_egl *egl, gsr_gpu_info *info) {
         info->vendor = GSR_GPU_VENDOR_INTEL;
     else if(strstr((const char*)gl_vendor, "NVIDIA"))
         info->vendor = GSR_GPU_VENDOR_NVIDIA;
+    else if(strstr((const char*)gl_vendor, "Broadcom"))
+        info->vendor = GSR_GPU_VENDOR_BROADCOM;
     else {
         fprintf(stderr, "gsr error: unknown gpu vendor: %s\n", gl_vendor);
         supported = false;
@@ -408,33 +403,10 @@ bool gl_get_gpu_info(gsr_egl *egl, gsr_gpu_info *info) {
         info->is_steam_deck = strstr((const char*)gl_renderer, "vangogh") != NULL;
     }
 
-    if(gl_version) {
-        const char *mesa_p = strstr((const char*)gl_version, "Mesa ");
-        if(mesa_p) {
-            mesa_p += 5;
-            int major = 0;
-            int minor = 0;
-            int patch = 0;
-            if(sscanf(mesa_p, "%d.%d.%d", &major, &minor, &patch) == 3) {
-                info->driver_major = major;
-                info->driver_minor = minor;
-                info->driver_patch = patch;
-            }
-        }
-    }
-
     end:
     return supported;
 }
 
-bool version_greater_than(int major, int minor, int patch, int other_major, int other_minor, int other_patch) {
-    return (major > other_major) || (major == other_major && minor > other_minor) || (major == other_major && minor == other_minor && patch > other_patch);
-}
-
-bool gl_driver_version_greater_than(const gsr_gpu_info *gpu_info, int major, int minor, int patch) {
-    return version_greater_than(gpu_info->driver_major, gpu_info->driver_minor, gpu_info->driver_patch, major, minor, patch);
-}
-
 bool try_card_has_valid_plane(const char *card_path) {
     drmVersion *ver = NULL;
     drmModePlaneResPtr planes = NULL;
@@ -543,6 +515,41 @@ int create_directory_recursive(char *path) {
 }
 
 void setup_dma_buf_attrs(intptr_t *img_attr, uint32_t format, uint32_t width, uint32_t height, const int *fds, const uint32_t *offsets, const uint32_t *pitches, const uint64_t *modifiers, int num_planes, bool use_modifier) {
+    const uint32_t plane_fd_attrs[DRM_NUM_BUF_ATTRS] = {
+        EGL_DMA_BUF_PLANE0_FD_EXT,
+        EGL_DMA_BUF_PLANE1_FD_EXT,
+        EGL_DMA_BUF_PLANE2_FD_EXT,
+        EGL_DMA_BUF_PLANE3_FD_EXT
+    };
+
+    const uint32_t plane_offset_attrs[DRM_NUM_BUF_ATTRS] = {
+        EGL_DMA_BUF_PLANE0_OFFSET_EXT,
+        EGL_DMA_BUF_PLANE1_OFFSET_EXT,
+        EGL_DMA_BUF_PLANE2_OFFSET_EXT,
+        EGL_DMA_BUF_PLANE3_OFFSET_EXT
+    };
+
+    const uint32_t plane_pitch_attrs[DRM_NUM_BUF_ATTRS] = {
+        EGL_DMA_BUF_PLANE0_PITCH_EXT,
+        EGL_DMA_BUF_PLANE1_PITCH_EXT,
+        EGL_DMA_BUF_PLANE2_PITCH_EXT,
+        EGL_DMA_BUF_PLANE3_PITCH_EXT
+    };
+
+    const uint32_t plane_modifier_lo_attrs[DRM_NUM_BUF_ATTRS] = {
+        EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT,
+        EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT,
+        EGL_DMA_BUF_PLANE2_MODIFIER_LO_EXT,
+        EGL_DMA_BUF_PLANE3_MODIFIER_LO_EXT
+    };
+
+    const uint32_t plane_modifier_hi_attrs[DRM_NUM_BUF_ATTRS] = {
+        EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT,
+        EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT,
+        EGL_DMA_BUF_PLANE2_MODIFIER_HI_EXT,
+        EGL_DMA_BUF_PLANE3_MODIFIER_HI_EXT
+    };
+
     size_t img_attr_index = 0;
 
     img_attr[img_attr_index++] = EGL_LINUX_DRM_FOURCC_EXT;
@@ -554,79 +561,23 @@ void setup_dma_buf_attrs(intptr_t *img_attr, uint32_t format, uint32_t width, ui
     img_attr[img_attr_index++] = EGL_HEIGHT;
     img_attr[img_attr_index++] = height;
 
-    if(num_planes >= 1) {
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE0_FD_EXT;
-        img_attr[img_attr_index++] = fds[0];
+    assert(num_planes <= DRM_NUM_BUF_ATTRS);
+    for(int i = 0; i < num_planes; ++i) {
+        img_attr[img_attr_index++] = plane_fd_attrs[i];
+        img_attr[img_attr_index++] = fds[i];
 
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE0_OFFSET_EXT;
-        img_attr[img_attr_index++] = offsets[0];
+        img_attr[img_attr_index++] = plane_offset_attrs[i];
+        img_attr[img_attr_index++] = offsets[i];
 
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE0_PITCH_EXT;
-        img_attr[img_attr_index++] = pitches[0];
+        img_attr[img_attr_index++] = plane_pitch_attrs[i];
+        img_attr[img_attr_index++] = pitches[i];
 
         if(use_modifier) {
-            img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE0_MODIFIER_LO_EXT;
-            img_attr[img_attr_index++] = modifiers[0] & 0xFFFFFFFFULL;
+            img_attr[img_attr_index++] = plane_modifier_lo_attrs[i];
+            img_attr[img_attr_index++] = modifiers[i] & 0xFFFFFFFFULL;
 
-            img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE0_MODIFIER_HI_EXT;
-            img_attr[img_attr_index++] = modifiers[0] >> 32ULL;
-        }
-    }
-
-    if(num_planes >= 2) {
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE1_FD_EXT;
-        img_attr[img_attr_index++] = fds[1];
-
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE1_OFFSET_EXT;
-        img_attr[img_attr_index++] = offsets[1];
-
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE1_PITCH_EXT;
-        img_attr[img_attr_index++] = pitches[1];
-
-        if(use_modifier) {
-            img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE1_MODIFIER_LO_EXT;
-            img_attr[img_attr_index++] = modifiers[1] & 0xFFFFFFFFULL;
-
-            img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE1_MODIFIER_HI_EXT;
-            img_attr[img_attr_index++] = modifiers[1] >> 32ULL;
-        }
-    }
-
-    if(num_planes >= 3) {
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE2_FD_EXT;
-        img_attr[img_attr_index++] = fds[2];
-
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE2_OFFSET_EXT;
-        img_attr[img_attr_index++] = offsets[2];
-
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE2_PITCH_EXT;
-        img_attr[img_attr_index++] = pitches[2];
-
-        if(use_modifier) {
-            img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE2_MODIFIER_LO_EXT;
-            img_attr[img_attr_index++] = modifiers[2] & 0xFFFFFFFFULL;
-
-            img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE2_MODIFIER_HI_EXT;
-            img_attr[img_attr_index++] = modifiers[2] >> 32ULL;
-        }
-    }
-
-    if(num_planes >= 4) {
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE3_FD_EXT;
-        img_attr[img_attr_index++] = fds[3];
-
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE3_OFFSET_EXT;
-        img_attr[img_attr_index++] = offsets[3];
-
-        img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE3_PITCH_EXT;
-        img_attr[img_attr_index++] = pitches[3];
-
-        if(use_modifier) {
-            img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE3_MODIFIER_LO_EXT;
-            img_attr[img_attr_index++] = modifiers[3] & 0xFFFFFFFFULL;
-
-            img_attr[img_attr_index++] = EGL_DMA_BUF_PLANE3_MODIFIER_HI_EXT;
-            img_attr[img_attr_index++] = modifiers[3] >> 32ULL;
+            img_attr[img_attr_index++] = plane_modifier_hi_attrs[i];
+            img_attr[img_attr_index++] = modifiers[i] >> 32ULL;
         }
     }
 
@@ -634,268 +585,6 @@ void setup_dma_buf_attrs(intptr_t *img_attr, uint32_t format, uint32_t width, ui
     assert(img_attr_index <= 44);
 }
 
-static VADisplay video_codec_context_get_vaapi_display(AVCodecContext *video_codec_context) {
-    AVBufferRef *hw_frames_ctx = video_codec_context->hw_frames_ctx;
-    if(!hw_frames_ctx)
-        return NULL;
-
-    AVHWFramesContext *hw_frame_context = (AVHWFramesContext*)hw_frames_ctx->data;
-    AVHWDeviceContext *device_context = (AVHWDeviceContext*)hw_frame_context->device_ctx;
-    if(device_context->type != AV_HWDEVICE_TYPE_VAAPI)
-        return NULL;
-
-    AVVAAPIDeviceContext *vactx = device_context->hwctx;
-    return vactx->display;
-}
-
-bool video_codec_context_is_vaapi(AVCodecContext *video_codec_context) {
-    if(!video_codec_context)
-        return false;
-
-    AVBufferRef *hw_frames_ctx = video_codec_context->hw_frames_ctx;
-    if(!hw_frames_ctx)
-        return false;
-
-    AVHWFramesContext *hw_frame_context = (AVHWFramesContext*)hw_frames_ctx->data;
-    AVHWDeviceContext *device_context = (AVHWDeviceContext*)hw_frame_context->device_ctx;
-    return device_context->type == AV_HWDEVICE_TYPE_VAAPI;
-}
-
-static uint32_t drm_fourcc_to_va_fourcc(uint32_t drm_fourcc) {
-    switch(drm_fourcc) {
-        case DRM_FORMAT_XRGB8888: return VA_FOURCC_BGRX;
-        case DRM_FORMAT_XBGR8888: return VA_FOURCC_RGBX;
-        case DRM_FORMAT_RGBX8888: return VA_FOURCC_XBGR;
-        case DRM_FORMAT_BGRX8888: return VA_FOURCC_XRGB;
-        case DRM_FORMAT_ARGB8888: return VA_FOURCC_BGRA;
-        case DRM_FORMAT_ABGR8888: return VA_FOURCC_RGBA;
-        case DRM_FORMAT_RGBA8888: return VA_FOURCC_ABGR;
-        case DRM_FORMAT_BGRA8888: return VA_FOURCC_ARGB;
-        default:                  return drm_fourcc;
-    }
-}
-
-bool vaapi_copy_drm_planes_to_video_surface(AVCodecContext *video_codec_context, AVFrame *video_frame, vec2i source_pos, vec2i source_size, vec2i dest_pos, vec2i dest_size, uint32_t format, vec2i size, const int *fds, const uint32_t *offsets, const uint32_t *pitches, const uint64_t *modifiers, int num_planes) {
-    VAConfigID config_id = 0;
-    VAContextID context_id = 0;
-    VASurfaceID input_surface_id = 0;
-    VABufferID buffer_id = 0;
-    bool success = true;
-
-    VADisplay va_dpy = video_codec_context_get_vaapi_display(video_codec_context);
-    if(!va_dpy) {
-        success = false;
-        goto done;
-    }
-
-    VAStatus va_status = vaCreateConfig(va_dpy, VAProfileNone, VAEntrypointVideoProc, NULL, 0, &config_id);
-    if(va_status != VA_STATUS_SUCCESS) {
-        fprintf(stderr, "gsr error: vaapi_copy_drm_planes_to_video_surface: vaCreateConfig failed, error: %s\n", vaErrorStr(va_status));
-        success = false;
-        goto done;
-    }
-
-    VASurfaceID output_surface_id = (uintptr_t)video_frame->data[3];
-    va_status = vaCreateContext(va_dpy, config_id, size.x, size.y, VA_PROGRESSIVE, &output_surface_id, 1, &context_id);
-    if(va_status != VA_STATUS_SUCCESS) {
-        fprintf(stderr, "gsr error: vaapi_copy_drm_planes_to_video_surface: vaCreateContext failed, error: %s\n", vaErrorStr(va_status));
-        success = false;
-        goto done;
-    }
-
-    VADRMPRIMESurfaceDescriptor buf = {0};
-    buf.fourcc = drm_fourcc_to_va_fourcc(format);//VA_FOURCC_BGRX; // TODO: VA_FOURCC_BGRA, VA_FOURCC_X2R10G10B10
-    buf.width = size.x;
-    buf.height = size.y;
-    buf.num_objects = num_planes;
-    buf.num_layers = 1;
-    buf.layers[0].drm_format = format;
-    buf.layers[0].num_planes = buf.num_objects;
-    for(int i = 0; i < num_planes; ++i) {
-        buf.objects[i].fd = fds[i];
-        buf.objects[i].size = size.y * pitches[i]; // TODO:
-        buf.objects[i].drm_format_modifier = modifiers[i];
-
-        buf.layers[0].object_index[i] = i;
-        buf.layers[0].offset[i] = offsets[i];
-        buf.layers[0].pitch[i] = pitches[i];
-    }
-
-    VASurfaceAttrib attribs[2] = {0};
-    attribs[0].type = VASurfaceAttribMemoryType;
-    attribs[0].flags = VA_SURFACE_ATTRIB_SETTABLE;
-    attribs[0].value.type = VAGenericValueTypeInteger;
-    attribs[0].value.value.i = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME_2;
-    attribs[1].type = VASurfaceAttribExternalBufferDescriptor;
-    attribs[1].flags = VA_SURFACE_ATTRIB_SETTABLE;
-    attribs[1].value.type = VAGenericValueTypePointer;
-    attribs[1].value.value.p = &buf;
-    
-    // TODO: RT_FORMAT with 10 bit/hdr, VA_RT_FORMAT_RGB32_10
-    // TODO: Max size same as source_size
-    va_status = vaCreateSurfaces(va_dpy, VA_RT_FORMAT_RGB32, size.x, size.y, &input_surface_id, 1, attribs, 2);
-    if(va_status != VA_STATUS_SUCCESS) {
-        fprintf(stderr, "gsr error: vaapi_copy_drm_planes_to_video_surface: vaCreateSurfaces failed, error: %s\n", vaErrorStr(va_status));
-        success = false;
-        goto done;
-    }
-
-    const VARectangle source_region = {
-        .x = source_pos.x,
-        .y = source_pos.y,
-        .width = source_size.x,
-        .height = source_size.y
-    };
-
-    const VARectangle output_region = {
-        .x = dest_pos.x,
-        .y = dest_pos.y,
-        .width = dest_size.x,
-        .height = dest_size.y
-    };
-
-    const bool scaled = dest_size.x != source_size.x || dest_size.y != source_size.y;
-
-    // Copying a surface to another surface will automatically perform the color conversion. Thanks vaapi!
-    VAProcPipelineParameterBuffer params = {0};
-    params.surface = input_surface_id;
-    params.surface_region = NULL;
-    params.surface_region = &source_region;
-    params.output_region = &output_region;
-    params.output_background_color = 0;
-    params.filter_flags = scaled ? (VA_FILTER_SCALING_HQ | VA_FILTER_INTERPOLATION_BILINEAR) : 0;
-    params.pipeline_flags = VA_PROC_PIPELINE_FAST;
-
-    params.input_color_properties.colour_primaries = 1;
-    params.input_color_properties.transfer_characteristics = 1;
-    params.input_color_properties.matrix_coefficients = 1;
-    params.surface_color_standard = VAProcColorStandardBT709; // TODO:
-    params.input_color_properties.color_range = video_frame->color_range == AVCOL_RANGE_JPEG ? VA_SOURCE_RANGE_FULL : VA_SOURCE_RANGE_REDUCED;
-
-    params.output_color_properties.colour_primaries = 1;
-    params.output_color_properties.transfer_characteristics = 1;
-    params.output_color_properties.matrix_coefficients = 1;
-    params.output_color_standard = VAProcColorStandardBT709; // TODO:
-    params.output_color_properties.color_range = video_frame->color_range == AVCOL_RANGE_JPEG ? VA_SOURCE_RANGE_FULL : VA_SOURCE_RANGE_REDUCED;
-
-    params.processing_mode = VAProcPerformanceMode;
-
-    // VAProcPipelineCaps pipeline_caps = {0};
-    // va_status = vaQueryVideoProcPipelineCaps(self->va_dpy,
-    //                                    self->context_id,
-    //                                    NULL, 0,
-    //                                    &pipeline_caps);
-    // if(va_status == VA_STATUS_SUCCESS) {
-    //     fprintf(stderr, "pipeline_caps: %u, %u\n", (unsigned int)pipeline_caps.rotation_flags, pipeline_caps.blend_flags);
-    // }
-
-    // TODO: params.output_hdr_metadata
-
-    // TODO:
-    // if (first surface to render)
-    //     pipeline_param->output_background_color = 0xff000000; // black
-
-    va_status = vaCreateBuffer(va_dpy, context_id, VAProcPipelineParameterBufferType, sizeof(params), 1, &params, &buffer_id);
-    if(va_status != VA_STATUS_SUCCESS) {
-        fprintf(stderr, "gsr error: vaapi_copy_drm_planes_to_video_surface: vaCreateBuffer failed, error: %d\n", va_status);
-        success = false;
-        goto done;
-    }
-
-    va_status = vaBeginPicture(va_dpy, context_id, output_surface_id);
-    if(va_status != VA_STATUS_SUCCESS) {
-        fprintf(stderr, "gsr error: vaapi_copy_drm_planes_to_video_surface: vaBeginPicture failed, error: %d\n", va_status);
-        success = false;
-        goto done;
-    }
-
-    va_status = vaRenderPicture(va_dpy, context_id, &buffer_id, 1);
-    if(va_status != VA_STATUS_SUCCESS) {
-        vaEndPicture(va_dpy, context_id);
-        fprintf(stderr, "gsr error: vaapi_copy_drm_planes_to_video_surface: vaRenderPicture failed, error: %d\n", va_status);
-        success = false;
-        goto done;
-    }
-
-    va_status = vaEndPicture(va_dpy, context_id);
-    if(va_status != VA_STATUS_SUCCESS) {
-        fprintf(stderr, "gsr error: vaapi_copy_drm_planes_to_video_surface: vaEndPicture failed, error: %d\n", va_status);
-        success = false;
-        goto done;
-    }
-
-    // vaSyncBuffer(va_dpy, buffer_id, 1000 * 1000 * 1000);
-    // vaSyncSurface(va_dpy, input_surface_id);
-    // vaSyncSurface(va_dpy, output_surface_id);
-
-    done:
-    if(buffer_id)
-        vaDestroyBuffer(va_dpy, buffer_id);
-
-    if(input_surface_id)
-        vaDestroySurfaces(va_dpy, &input_surface_id, 1);
-
-    if(context_id)
-        vaDestroyContext(va_dpy, context_id);
-
-    if(config_id)
-        vaDestroyConfig(va_dpy, config_id);
-
-    return success;
-}
-
-bool vaapi_copy_egl_image_to_video_surface(gsr_egl *egl, EGLImage image, vec2i source_pos, vec2i source_size, vec2i dest_pos, vec2i dest_size, AVCodecContext *video_codec_context, AVFrame *video_frame) {
-    if(!image)
-        return false;
-
-    int texture_fourcc = 0;
-    int texture_num_planes = 0;
-    uint64_t texture_modifiers = 0;
-    if(!egl->eglExportDMABUFImageQueryMESA(egl->egl_display, image, &texture_fourcc, &texture_num_planes, &texture_modifiers)) {
-        fprintf(stderr, "gsr error: gsr_capture_xcomposite_vaapi_tick: eglExportDMABUFImageQueryMESA failed\n");
-        return false;
-    }
-
-    if(texture_num_planes <= 0 || texture_num_planes > 8) {
-        fprintf(stderr, "gsr error: gsr_capture_xcomposite_vaapi_tick: expected planes size to be 0<planes<=8 for drm buf, got %d planes\n", texture_num_planes);
-        return false;
-    }
-
-    int texture_fds[8];
-    int32_t texture_strides[8];
-    int32_t texture_offsets[8];
-
-    while(egl->eglGetError() != EGL_SUCCESS){}
-    if(!egl->eglExportDMABUFImageMESA(egl->egl_display, image, texture_fds, texture_strides, texture_offsets)) {
-        fprintf(stderr, "gsr error: gsr_capture_xcomposite_vaapi_tick: eglExportDMABUFImageMESA failed, error: %d\n", egl->eglGetError());
-        return false;
-    }
-
-    int fds[8];
-    uint32_t offsets[8];
-    uint32_t pitches[8];
-    uint64_t modifiers[8];
-    for(int i = 0; i < texture_num_planes; ++i) {
-        fds[i] = texture_fds[i];
-        offsets[i] = texture_offsets[i];
-        pitches[i] = texture_strides[i];
-        modifiers[i] = texture_modifiers;
-
-        if(fds[i] == -1)
-            texture_num_planes = i;
-    }
-    const bool success = texture_num_planes > 0 && vaapi_copy_drm_planes_to_video_surface(video_codec_context, video_frame, source_pos, source_size, dest_pos, dest_size, texture_fourcc, source_size, fds, offsets, pitches, modifiers, texture_num_planes);
-
-    for(int i = 0; i < texture_num_planes; ++i) {
-        if(texture_fds[i] > 0) {
-            close(texture_fds[i]);
-            texture_fds[i] = -1;
-        }
-    }
-
-    return success;
-}
-
 vec2i scale_keep_aspect_ratio(vec2i from, vec2i to) {
     if(from.x == 0 || from.y == 0)
         return (vec2i){0, 0};
@@ -913,17 +602,82 @@ vec2i scale_keep_aspect_ratio(vec2i from, vec2i to) {
     return from;
 }
 
+vec2i gsr_capture_get_target_position(vec2i output_size, gsr_capture_metadata *capture_metadata) {
+    vec2i target_pos = {0, 0};
+
+    switch(capture_metadata->halign) {
+        case GSR_CAPTURE_ALIGN_START:
+            break;
+        case GSR_CAPTURE_ALIGN_CENTER:
+            target_pos.x = capture_metadata->video_size.x/2 - output_size.x/2;
+            break;
+        case GSR_CAPTURE_ALIGN_END:
+            target_pos.x = capture_metadata->video_size.x - output_size.x;
+            break;
+    }
+
+    switch(capture_metadata->valign) {
+        case GSR_CAPTURE_ALIGN_START:
+            break;
+        case GSR_CAPTURE_ALIGN_CENTER:
+            target_pos.y = capture_metadata->video_size.y/2 - output_size.y/2;
+            break;
+        case GSR_CAPTURE_ALIGN_END:
+            target_pos.y = capture_metadata->video_size.y - output_size.y;
+            break;
+    }
+
+    target_pos.x += capture_metadata->position.x;
+    target_pos.y += capture_metadata->position.y;
+    return target_pos;
+}
+
+
 unsigned int gl_create_texture(gsr_egl *egl, int width, int height, int internal_format, unsigned int format, int filter) {
     unsigned int texture_id = 0;
     egl->glGenTextures(1, &texture_id);
     egl->glBindTexture(GL_TEXTURE_2D, texture_id);
-    egl->glTexImage2D(GL_TEXTURE_2D, 0, internal_format, width, height, 0, format, GL_UNSIGNED_BYTE, NULL);
+    //egl->glTexImage2D(GL_TEXTURE_2D, 0, internal_format, width, height, 0, format, GL_UNSIGNED_BYTE, NULL);
+    // Needed for hevc_10bit for nvenc (cuGraphicsGLRegisterImage)
+    egl->glTexStorage2D(GL_TEXTURE_2D, 1, internal_format, width, height);
 
-    egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, filter);
     egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
 
     egl->glBindTexture(GL_TEXTURE_2D, 0);
     return texture_id;
 }
+
+/* TODO: Test with optimus and open kernel modules */
+bool get_nvidia_driver_version(int *major, int *minor) {
+    *major = 0;
+    *minor = 0;
+
+    FILE *f = fopen("/proc/driver/nvidia/version", "rb");
+    if(!f) {
+        fprintf(stderr, "gsr warning: failed to get nvidia driver version (failed to read /proc/driver/nvidia/version)\n");
+        return false;
+    }
+
+    char buffer[2048];
+    size_t bytes_read = fread(buffer, 1, sizeof(buffer) - 1, f);
+    buffer[bytes_read] = '\0';
+
+    bool success = false;
+    const char *p = strstr(buffer, "Kernel Module");
+    if(p) {
+        p += 13;
+        int driver_major_version = 0, driver_minor_version = 0;
+        if(sscanf(p, "%d.%d", &driver_major_version, &driver_minor_version) == 2) {
+            *major = driver_major_version;
+            *minor = driver_minor_version;
+            success = true;
+        }
+    }
+
+    if(!success)
+        fprintf(stderr, "gsr warning: failed to get nvidia driver version\n");
+
+    fclose(f);
+    return success;
+}

src/window/wayland.c

@@ -9,27 +9,33 @@
 #include <stdint.h>
 #include <wayland-client.h>
 #include <wayland-egl.h>
+#include "xdg-output-unstable-v1-client-protocol.h"
 
 #define GSR_MAX_OUTPUTS 32
 
+typedef struct gsr_window_wayland gsr_window_wayland;
+
 typedef struct {
     uint32_t wl_name;
-    void *output;
+    struct wl_output *output;
+    struct zxdg_output_v1 *xdg_output;
     vec2i pos;
     vec2i size;
+    vec2i logical_size;
     int32_t transform;
     char *name;
 } gsr_wayland_output;
 
-typedef struct {
-    void *display;
-    void *window;
-    void *registry;
-    void *surface;
-    void *compositor;
+struct gsr_window_wayland {
+    struct wl_display *display;
+    struct wl_egl_window *window;
+    struct wl_registry *registry;
+    struct wl_surface *surface;
+    struct wl_compositor *compositor;
     gsr_wayland_output outputs[GSR_MAX_OUTPUTS];
     int num_outputs;
-} gsr_window_wayland;
+    struct zxdg_output_manager_v1 *xdg_output_manager;
+};
 
 static void output_handle_geometry(void *data, struct wl_output *wl_output,
         int32_t x, int32_t y, int32_t phys_width, int32_t phys_height,
@@ -95,15 +101,14 @@ static const struct wl_output_listener output_listener = {
 static void registry_add_object(void *data, struct wl_registry *registry, uint32_t name, const char *interface, uint32_t version) {
     (void)version;
     gsr_window_wayland *window_wayland = data;
-    if (strcmp(interface, "wl_compositor") == 0) {
-        if(window_wayland->compositor) {
-            wl_compositor_destroy(window_wayland->compositor);
-            window_wayland->compositor = NULL;
-        }
+    if(strcmp(interface, "wl_compositor") == 0) {
+        if(window_wayland->compositor)
+            return;
+
         window_wayland->compositor = wl_registry_bind(registry, name, &wl_compositor_interface, 1);
     } else if(strcmp(interface, wl_output_interface.name) == 0) {
         if(version < 4) {
-            fprintf(stderr, "gsr warning: wl output interface version is < 4, expected >= 4 to capture a monitor. Using KMS capture instead\n");
+            fprintf(stderr, "gsr warning: wl output interface version is < 4, expected >= 4 to capture a monitor\n");
             return;
         }
 
@@ -119,10 +124,21 @@ static void registry_add_object(void *data, struct wl_registry *registry, uint32
             .output = wl_registry_bind(registry, name, &wl_output_interface, 4),
             .pos = { .x = 0, .y = 0 },
             .size = { .x = 0, .y = 0 },
+            .logical_size = { .x = 0, .y = 0 },
             .transform = 0,
             .name = NULL,
         };
         wl_output_add_listener(gsr_output->output, &output_listener, gsr_output);
+    } else if(strcmp(interface, zxdg_output_manager_v1_interface.name) == 0) {
+        if(version < 1) {
+            fprintf(stderr, "gsr warning: xdg output interface version is < 1, expected >= 1 to capture a monitor\n");
+            return;
+        }
+
+        if(window_wayland->xdg_output_manager)
+            return;
+
+        window_wayland->xdg_output_manager = wl_registry_bind(registry, name, &zxdg_output_manager_v1_interface, 1);
     }
 }
 
@@ -130,6 +146,7 @@ static void registry_remove_object(void *data, struct wl_registry *registry, uin
     (void)data;
     (void)registry;
     (void)name;
+    // TODO: Remove output
 }
 
 static struct wl_registry_listener registry_listener = {
@@ -137,6 +154,96 @@ static struct wl_registry_listener registry_listener = {
     .global_remove = registry_remove_object,
 };
 
+static void xdg_output_logical_position(void *data, struct zxdg_output_v1 *zxdg_output_v1, int32_t x, int32_t y) {
+    (void)zxdg_output_v1;
+    gsr_wayland_output *gsr_xdg_output = data;
+    gsr_xdg_output->pos.x = x;
+    gsr_xdg_output->pos.y = y;
+}
+
+static void xdg_output_handle_logical_size(void *data, struct zxdg_output_v1 *xdg_output, int32_t width, int32_t height) {
+    (void)xdg_output;
+    gsr_wayland_output *gsr_xdg_output = data;
+    gsr_xdg_output->logical_size.x = width;
+    gsr_xdg_output->logical_size.y = height;
+}
+
+static void xdg_output_handle_done(void *data, struct zxdg_output_v1 *xdg_output) {
+    (void)data;
+    (void)xdg_output;
+}
+
+static void xdg_output_handle_name(void *data, struct zxdg_output_v1 *xdg_output, const char *name) {
+    (void)data;
+    (void)xdg_output;
+    (void)name;
+}
+
+static void xdg_output_handle_description(void *data, struct zxdg_output_v1 *xdg_output, const char *description) {
+    (void)data;
+    (void)xdg_output;
+    (void)description;
+}
+
+static const struct zxdg_output_v1_listener xdg_output_listener = {
+    .logical_position = xdg_output_logical_position,
+    .logical_size = xdg_output_handle_logical_size,
+    .done = xdg_output_handle_done,
+    .name = xdg_output_handle_name,
+    .description = xdg_output_handle_description,
+};
+
+static void gsr_window_wayland_set_monitor_outputs_from_xdg_output(gsr_window_wayland *self) {
+    if(!self->xdg_output_manager) {
+        fprintf(stderr, "gsr warning: zxdg_output_manager not found. registered monitor positions might be incorrect\n");
+        return;
+    }
+
+    for(int i = 0; i < self->num_outputs; ++i) {
+        self->outputs[i].xdg_output = zxdg_output_manager_v1_get_xdg_output(self->xdg_output_manager, self->outputs[i].output);
+        zxdg_output_v1_add_listener(self->outputs[i].xdg_output, &xdg_output_listener, &self->outputs[i]);
+    }
+
+    // Fetch xdg_output
+    wl_display_roundtrip(self->display);
+}
+
+// static int monitor_sort_x_pos(const void* a, const void* b) {
+//     const gsr_wayland_output *arg1 = *(const gsr_wayland_output**)a;
+//     const gsr_wayland_output *arg2 = *(const gsr_wayland_output**)b;
+//     return arg1->logical_pos.x - arg2->logical_pos.x;
+// }
+
+// static int monitor_sort_y_pos(const void* a, const void* b) {
+//     const gsr_wayland_output *arg1 = *(const gsr_wayland_output**)a;
+//     const gsr_wayland_output *arg2 = *(const gsr_wayland_output**)b;
+//     return arg1->logical_pos.y - arg2->logical_pos.y;
+// }
+
+static void gsr_window_wayland_set_monitor_real_positions(gsr_window_wayland *self) {
+    gsr_wayland_output *sorted_outputs[GSR_MAX_OUTPUTS];
+    for(int i = 0; i < self->num_outputs; ++i) {
+        sorted_outputs[i] = &self->outputs[i];
+    }
+
+    // TODO: set correct physical positions
+
+    // qsort(sorted_outputs, self->num_outputs, sizeof(gsr_wayland_output*), monitor_sort_x_pos);
+    // int x_pos = 0;
+    // for(int i = 0; i < self->num_outputs; ++i) {
+    //     fprintf(stderr, "monitor: %s\n", sorted_outputs[i]->name);
+    //     sorted_outputs[i]->pos.x = x_pos;
+    //     x_pos += sorted_outputs[i]->logical_size.x;
+    // }
+
+    // qsort(sorted_outputs, self->num_outputs, sizeof(gsr_wayland_output*), monitor_sort_y_pos);
+    // int y_pos = 0;
+    // for(int i = 0; i < self->num_outputs; ++i) {
+    //     sorted_outputs[i]->pos.y = y_pos;
+    //     y_pos += sorted_outputs[i]->logical_size.y;
+    // }
+}
+
 static void gsr_window_wayland_deinit(gsr_window_wayland *self) {
     if(self->window) {
         wl_egl_window_destroy(self->window);
@@ -158,9 +265,19 @@ static void gsr_window_wayland_deinit(gsr_window_wayland *self) {
             free(self->outputs[i].name);
             self->outputs[i].name = NULL;
         }
+
+        if(self->outputs[i].xdg_output) {
+            zxdg_output_v1_destroy(self->outputs[i].xdg_output);
+            self->outputs[i].output = NULL;
+        }
     }
     self->num_outputs = 0;
 
+    if(self->xdg_output_manager) {
+        zxdg_output_manager_v1_destroy(self->xdg_output_manager);
+        self->xdg_output_manager = NULL;
+    }
+
     if(self->compositor) {
         wl_compositor_destroy(self->compositor);
         self->compositor = NULL;
@@ -193,6 +310,9 @@ static bool gsr_window_wayland_init(gsr_window_wayland *self) {
     // Fetch wl_output
     wl_display_roundtrip(self->display);
 
+    gsr_window_wayland_set_monitor_outputs_from_xdg_output(self);
+    gsr_window_wayland_set_monitor_real_positions(self);
+
     if(!self->compositor) {
         fprintf(stderr, "gsr error: gsr_window_wayland_init failed: failed to find compositor\n");
         goto fail;
@@ -256,34 +376,44 @@ static gsr_monitor_rotation wayland_transform_to_gsr_rotation(int32_t rot) {
     return GSR_MONITOR_ROT_0;
 }
 
+static vec2i get_monitor_size_rotated(int width, int height, gsr_monitor_rotation rotation) {
+    vec2i size = { .x = width, .y = height };
+    if(rotation == GSR_MONITOR_ROT_90 || rotation == GSR_MONITOR_ROT_270) {
+        int tmp_x = size.x;
+        size.x = size.y;
+        size.y = tmp_x;
+    }
+    return size;
+}
+
 static void gsr_window_wayland_for_each_active_monitor_output_cached(const gsr_window *window, active_monitor_callback callback, void *userdata) {
     const gsr_window_wayland *self = window->priv;
-    drm_connector_type_count type_counts[CONNECTOR_TYPE_COUNTS];
-    int num_type_counts = 0;
-
     for(int i = 0; i < self->num_outputs; ++i) {
         const gsr_wayland_output *output = &self->outputs[i];
         if(!output->name)
             continue;
 
-        const int connector_type_index = get_connector_type_by_name(output->name);
-        drm_connector_type_count *connector_type = NULL;
-        if(connector_type_index != -1)
-            connector_type = drm_connector_types_get_index(type_counts, &num_type_counts, connector_type_index);
-        
-        if(connector_type) {
-            ++connector_type->count;
-            ++connector_type->count_active;
-        }
+        const gsr_monitor_rotation rotation = wayland_transform_to_gsr_rotation(output->transform);
 
+        vec2i size = { .x = output->size.x, .y = output->size.y };
+        size = get_monitor_size_rotated(size.x, size.y, rotation);
+
+        vec2i logical_size = { .x = output->logical_size.x, .y = output->logical_size.y };
+        if(logical_size.x == 0 || logical_size.y == 0)
+            logical_size = size;
+
+        const int connector_type_index = get_connector_type_by_name(output->name);
+        const int connector_type_id = get_connector_type_id_by_name(output->name);
         const gsr_monitor monitor = {
             .name = output->name,
             .name_len = strlen(output->name),
             .pos = { .x = output->pos.x, .y = output->pos.y },
-            .size = { .x = output->size.x, .y = output->size.y },
+            .size = size,
+            .logical_pos = { .x = output->pos.x, .y = output->pos.y },
+            .logical_size = logical_size,
             .connector_id = 0,
-            .rotation = wayland_transform_to_gsr_rotation(output->transform),
-            .monitor_identifier = connector_type ? monitor_identifier_from_type_and_count(connector_type_index, connector_type->count_active) : 0
+            .rotation = rotation,
+            .monitor_identifier = (connector_type_index != -1 && connector_type_id != -1) ? monitor_identifier_from_type_and_count(connector_type_index, connector_type_id) : 0
         };
         callback(&monitor, userdata);
     }

src/window/x11.c

@@ -121,6 +121,8 @@ static void gsr_window_x11_for_each_active_monitor_output_cached(const gsr_windo
             .name_len = strlen(output->name),
             .pos = output->pos,
             .size = output->size,
+            .logical_pos = output->pos,
+            .logical_size = output->size,
             .connector_id = output->connector_id,
             .rotation = output->rotation,
             .monitor_identifier = output->monitor_identifier

src/window_texture.c

@@ -20,6 +20,8 @@ int window_texture_init(WindowTexture *window_texture, Display *display, Window
     window_texture->texture_id = 0;
     window_texture->redirected = 0;
     window_texture->egl = egl;
+    window_texture->window_width = 0;
+    window_texture->window_height = 0;
     
     if(!x11_supports_composite_named_window_pixmap(display))
         return 1;
@@ -67,6 +69,11 @@ int window_texture_on_resize(WindowTexture *self) {
         EGL_NONE,
     };
 
+    Window root_window;
+    int window_x, window_y;
+    unsigned int window_border, window_depth;
+    XGetGeometry(self->display, self->window, &root_window, &window_x, &window_y, &self->window_width, &self->window_height, &window_border, &window_depth);
+
     pixmap = XCompositeNameWindowPixmap(self->display, self->window);
     if(!pixmap) {
         result = 2;
@@ -85,8 +92,6 @@ int window_texture_on_resize(WindowTexture *self) {
         texture_id = self->texture_id;
     }
 
-    self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
-    self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
     self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     self->egl->glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);