Implement kms vaapi capture

28 files changed, 1306 insertions, 197 deletions

diff --git a/.gitignore b/.gitignore
index 7efd9fe..8bedc1d 100644
--- a/.gitignore
+++ b/.gitignore
@@ -10,6 +10,7 @@ tests/compile_commands.json
 
 *.o
 gpu-screen-recorder
+gsr-kms-server
 
 *.mp4
 *.flv
diff --git a/README.md b/README.md
index 1cf23cc..4bd2db3 100644
--- a/README.md
+++ b/README.md
@@ -6,7 +6,7 @@ This screen recorder can be used for recording your desktop offline, for live st
 where only the last few seconds are saved.
 
 ## Note
-This software works only on x11 and with an nvidia gpu.\
+This software works only on x11.\
 If you are using a variable refresh rate monitor then choose to record "screen-direct". This will allow variable refresh rate to work when recording fullscreen applications. Note that some applications such as mpv will not work in fullscreen mode. A fix is being developed for this.\
 For screen capture to work with PRIME (laptops with a nvidia gpu), you must set the primary GPU to use your dedicated nvidia graphics card. You can do this by selecting "NVIDIA (Performance Mode) in nvidia settings:\
 ![](https://dec05eba.com/images/nvidia-settings-prime.png)\
@@ -35,11 +35,11 @@ You can also install gpu screen recorder ([the gtk gui version](https://git.dec0
 
 # Dependencies
 ## AMD
-`libglvnd (which provides libgl and libegl), mesa, ffmpeg (libavcodec, libavformat, libavutil, libswresample, libavfilter), libx11, libxcomposite, libxrandr, libpulse, libva, libva-mesa-driver`.
+`libglvnd (which provides libgl and libegl), mesa, ffmpeg (libavcodec, libavformat, libavutil, libswresample, libavfilter), libx11, libxcomposite, libxrandr, libpulse, libva, libva-mesa-driver, libdrm, libcap`.
 ## Intel
-`libglvnd (which provides libgl and libegl), mesa, ffmpeg (libavcodec, libavformat, libavutil, libswresample, libavfilter), libx11, libxcomposite, libxrandr, libpulse, libva, libva-intel-driver`.
+`libglvnd (which provides libgl and libegl), mesa, ffmpeg (libavcodec, libavformat, libavutil, libswresample, libavfilter), libx11, libxcomposite, libxrandr, libpulse, libva, libva-intel-driver, libdrm, libcap`.
 ## NVIDIA
-`libglvnd (which provides libgl and libegl), ffmpeg (libavcodec, libavformat, libavutil, libswresample, libavfilter), libx11, libxcomposite, libxrandr, libpulse, cuda (libnvidia-compute), nvenc (libnvidia-encode), libva`. Additionally, you need to have `nvfbc (libnvidia-fbc1)` installed when using nvfbc and `xnvctrl (libxnvctrl0)` when using the `-oc` option.
+`libglvnd (which provides libgl and libegl), ffmpeg (libavcodec, libavformat, libavutil, libswresample, libavfilter), libx11, libxcomposite, libxrandr, libpulse, cuda (libnvidia-compute), nvenc (libnvidia-encode), libva, libdrm, libcap`. Additionally, you need to have `nvfbc (libnvidia-fbc1)` installed when using nvfbc and `xnvctrl (libxnvctrl0)` when using the `-oc` option.
 
 # How to use
 Run `scripts/interactive.sh` or run gpu-screen-recorder directly, for example: `gpu-screen-recorder -w $(xdotool selectwindow) -c mp4 -f 60 -a "$(pactl get-default-sink).monitor" -o test_video.mp4` then stop the screen recorder with Ctrl+C, which will also save the recording. You can change -w to -w screen if you want to record all monitors or if you want to record a specific monitor then you can use -w monitor-name, for example -w HDMI-0 (use xrandr command to find the name of your monitor. The name can also be found in your desktop environments display settings).\
diff --git a/build.sh b/build.sh
index 3f44f37..591728b 100755
--- a/build.sh
+++ b/build.sh
@@ -1,21 +1,37 @@
 #!/bin/sh -e
 
-#libdrm
-dependencies="libavcodec libavformat libavutil x11 xcomposite xrandr libpulse libswresample libavfilter libva"
-includes="$(pkg-config --cflags $dependencies)"
-libs="$(pkg-config --libs $dependencies) -ldl -pthread -lm"
-opts="-O2 -g0 -DNDEBUG"
-gcc -c src/capture/capture.c $opts $includes
-gcc -c src/capture/nvfbc.c $opts $includes
-gcc -c src/capture/xcomposite_cuda.c $opts $includes
-gcc -c src/capture/xcomposite_vaapi.c $opts $includes
-gcc -c src/egl.c $opts $includes
-gcc -c src/cuda.c $opts $includes
-gcc -c src/xnvctrl.c $opts $includes
-gcc -c src/overclock.c $opts $includes
-gcc -c src/window_texture.c $opts $includes
-gcc -c src/time.c $opts $includes
-g++ -c src/sound.cpp $opts $includes
-g++ -c src/main.cpp $opts $includes
-g++ -o gpu-screen-recorder -O2 capture.o nvfbc.o egl.o cuda.o xnvctrl.o overclock.o window_texture.o time.o xcomposite_cuda.o xcomposite_vaapi.o sound.o main.o -s $libs
-echo "Successfully built gpu-screen-recorder"
+build_gsr_kms_server() {
+    dependencies="libdrm"
+    includes="$(pkg-config --cflags $dependencies)"
+    libs="$(pkg-config --libs $dependencies) -ldl"
+    opts="-O2 -g0 -DNDEBUG"
+    gcc -c src/kms/kms_server.c $opts $includes
+    gcc -o gsr-kms-server -O2 kms_server.o -s $libs
+}
+
+build_gsr() {
+    dependencies="libavcodec libavformat libavutil x11 xcomposite xrandr libpulse libswresample libavfilter libva libcap"
+    includes="$(pkg-config --cflags $dependencies)"
+    libs="$(pkg-config --libs $dependencies) -ldl -pthread -lm"
+    opts="-O2 -g0 -DNDEBUG"
+    gcc -c src/capture/capture.c $opts $includes
+    gcc -c src/capture/nvfbc.c $opts $includes
+    gcc -c src/capture/xcomposite_cuda.c $opts $includes
+    gcc -c src/capture/xcomposite_vaapi.c $opts $includes
+    gcc -c src/capture/kms_vaapi.c $opts $includes
+    gcc -c src/kms/kms_client.c $opts $includes
+    gcc -c src/egl.c $opts $includes
+    gcc -c src/cuda.c $opts $includes
+    gcc -c src/xnvctrl.c $opts $includes
+    gcc -c src/overclock.c $opts $includes
+    gcc -c src/window_texture.c $opts $includes
+    gcc -c src/utils.c $opts $includes
+    gcc -c src/library_loader.c $opts $includes
+    g++ -c src/sound.cpp $opts $includes
+    g++ -c src/main.cpp $opts $includes
+    g++ -o gpu-screen-recorder -O2 capture.o nvfbc.o kms_client.o egl.o cuda.o xnvctrl.o overclock.o window_texture.o utils.o library_loader.o xcomposite_cuda.o xcomposite_vaapi.o kms_vaapi.o sound.o main.o -s $libs
+}
+
+build_gsr_kms_server
+build_gsr
+echo "Successfully built gpu-screen-recorder"
\ No newline at end of file
diff --git a/include/capture/kms_vaapi.h b/include/capture/kms_vaapi.h
new file mode 100644
index 0000000..5cf8606
--- /dev/null
+++ b/include/capture/kms_vaapi.h
@@ -0,0 +1,18 @@
+#ifndef GSR_CAPTURE_KMS_VAAPI_H
+#define GSR_CAPTURE_KMS_VAAPI_H
+
+#include "capture.h"
+#include "../vec2.h"
+#include <X11/X.h>
+
+typedef struct _XDisplay Display;
+
+typedef struct {
+    Display *dpy;
+    const char *display_to_capture; /* if this is "screen", then the entire x11 screen is captured (all displays). A copy is made of this */
+    const char *program_dir; /* ref */
+} gsr_capture_kms_vaapi_params;
+
+gsr_capture* gsr_capture_kms_vaapi_create(const gsr_capture_kms_vaapi_params *params);
+
+#endif /* GSR_CAPTURE_KMS_VAAPI_H */
diff --git a/include/kms/kms_client.h b/include/kms/kms_client.h
new file mode 100644
index 0000000..bb4325b
--- /dev/null
+++ b/include/kms/kms_client.h
@@ -0,0 +1,21 @@
+#ifndef GSR_KMS_CLIENT_H
+#define GSR_KMS_CLIENT_H
+
+#include "kms_shared.h"
+#include <sys/types.h>
+
+typedef struct {
+    pid_t kms_server_pid;
+    int socket_fd;
+    int client_fd;
+    char socket_path[27];
+    char *card_path;
+} gsr_kms_client;
+
+/* |card_path| should be a path to card, for example /dev/dri/card0 */
+int gsr_kms_client_init(gsr_kms_client *self, const char *card_path, const char *program_dir);
+void gsr_kms_client_deinit(gsr_kms_client *self);
+
+int gsr_kms_client_get_kms(gsr_kms_client *self, gsr_kms_response *response);
+
+#endif /* #define GSR_KMS_CLIENT_H */
diff --git a/include/kms/kms_shared.h b/include/kms/kms_shared.h
new file mode 100644
index 0000000..f63c2f3
--- /dev/null
+++ b/include/kms/kms_shared.h
@@ -0,0 +1,45 @@
+#ifndef GSR_KMS_SHARED_H
+#define GSR_KMS_SHARED_H
+
+#include <stdint.h>
+
+typedef enum {
+    KMS_REQUEST_TYPE_GET_KMS
+} gsr_kms_request_type;
+
+typedef enum {
+    KMS_RESULT_OK,
+    KMS_RESULT_INVALID_REQUEST,
+    KMS_RESULT_FAILED_TO_OPEN_CARD,
+    KMS_RESULT_INSUFFICIENT_PERMISSIONS,
+    KMS_RESULT_FAILED_TO_GET_KMS,
+    KMS_RESULT_NO_KMS_AVAILABLE,
+    KMS_RESULT_FAILED_TO_SEND
+} gsr_kms_result;
+
+typedef struct {
+    int type; /* gsr_kms_request_type */
+    union {
+        char card_path[255];
+    } data;
+} gsr_kms_request;
+
+typedef struct {
+    int fd;
+    uint32_t width;
+    uint32_t height;
+    uint32_t pitch;
+    uint32_t offset;
+    uint32_t pixel_format;
+    uint64_t modifier;
+} gsr_kms_response_fd;
+
+typedef struct {
+    int result; /* gsr_kms_result */
+    union {
+        char err_msg[128];
+        gsr_kms_response_fd fd;
+    } data;
+} gsr_kms_response;
+
+#endif /* #define GSR_KMS_SHARED_H */
diff --git a/include/library_loader.h b/include/library_loader.h
index 1622521..47bc9f0 100644
--- a/include/library_loader.h
+++ b/include/library_loader.h
@@ -1,42 +1,17 @@
 #ifndef GSR_LIBRARY_LOADER_H
 #define GSR_LIBRARY_LOADER_H
 
-#include <dlfcn.h>
 #include <stdbool.h>
-#include <stdio.h>
 
 typedef struct {
     void **func;
     const char *name;
 } dlsym_assign;
 
-static void* dlsym_print_fail(void *handle, const char *name, bool required) {
-    dlerror();
-    void *sym = dlsym(handle, name);
-    char *err_str = dlerror();
-
-    if(!sym)
-        fprintf(stderr, "%s: dlsym(handle, \"%s\") failed, error: %s\n", required ? "error" : "warning", name, err_str ? err_str : "(null)");
-
-    return sym;
-}
-
+void* dlsym_print_fail(void *handle, const char *name, bool required);
 /* |dlsyms| should be null terminated */
-static bool dlsym_load_list(void *handle, const dlsym_assign *dlsyms) {
-    bool success = true;
-    for(int i = 0; dlsyms[i].func; ++i) {
-        *dlsyms[i].func = dlsym_print_fail(handle, dlsyms[i].name, true);
-        if(!*dlsyms[i].func)
-            success = false;
-    }
-    return success;
-}
-
+bool dlsym_load_list(void *handle, const dlsym_assign *dlsyms);
 /* |dlsyms| should be null terminated */
-static void dlsym_load_list_optional(void *handle, const dlsym_assign *dlsyms) {
-    for(int i = 0; dlsyms[i].func; ++i) {
-        *dlsyms[i].func = dlsym_print_fail(handle, dlsyms[i].name, false);
-    }
-}
+void dlsym_load_list_optional(void *handle, const dlsym_assign *dlsyms);
 
 #endif /* GSR_LIBRARY_LOADER_H */
diff --git a/include/time.h b/include/time.h
deleted file mode 100644
index 150c655..0000000
--- a/include/time.h
+++ /dev/null
@@ -1,6 +0,0 @@
-#ifndef GSR_TIME_H
-#define GSR_TIME_H
-
-double clock_get_monotonic_seconds();
-
-#endif /* GSR_TIME_H */
diff --git a/include/utils.h b/include/utils.h
new file mode 100644
index 0000000..f4925e5
--- /dev/null
+++ b/include/utils.h
@@ -0,0 +1,26 @@
+#ifndef GSR_UTILS_H
+#define GSR_UTILS_H
+
+#include "vec2.h"
+#include <stdbool.h>
+#include <X11/extensions/Xrandr.h>
+
+typedef struct {
+    vec2i pos;
+    vec2i size;
+} gsr_monitor;
+
+typedef struct {
+    const char *name;
+    int name_len;
+    gsr_monitor *monitor;
+    bool found_monitor;
+} get_monitor_by_name_userdata;
+
+double clock_get_monotonic_seconds(void);
+
+typedef void (*active_monitor_callback)(const XRROutputInfo *output_info, const XRRCrtcInfo *crt_info, const XRRModeInfo *mode_info, void *userdata);
+void for_each_active_monitor_output(Display *display, active_monitor_callback callback, void *userdata);
+bool get_monitor_by_name(Display *display, const char *name, gsr_monitor *monitor);
+
+#endif /* GSR_UTILS_H */
diff --git a/install.sh b/install.sh
index aa4c802..f898534 100755
--- a/install.sh
+++ b/install.sh
@@ -6,7 +6,8 @@ cd "$script_dir"
 [ $(id -u) -ne 0 ] && echo "You need root privileges to run the install script" && exit 1
 
 ./build.sh
-install -Dm755 "gpu-screen-recorder" "/usr/local/bin/gpu-screen-recorder"
+rm -f "/usr/local/bin/gpu-screen-recorder"
+install -Dm755 "gsr-kms-server" "/usr/bin/gsr-kms-server"
 install -Dm755 "gpu-screen-recorder" "/usr/bin/gpu-screen-recorder"
 
 [ -f "/proc/driver/nvidia/version" ] && ./install_coolbits.sh
diff --git a/install_ubuntu_amd.sh b/install_ubuntu_amd.sh
index 5af70dc..8404b49 100755
--- a/install_ubuntu_amd.sh
+++ b/install_ubuntu_amd.sh
@@ -9,6 +9,6 @@ set -e
 apt-get -y install build-essential\
 	libswresample-dev libavformat-dev libavcodec-dev libavutil-dev libavfilter-dev\
 	libglvnd-dev libx11-dev libxcomposite-dev libxrandr-dev\
-	libpulse-dev libva2 libegl-mesa0 mesa-va-drivers
+	libpulse-dev libva2 libegl-mesa0 mesa-va-drivers libdrm-dev libcap-dev
 
 ./install.sh
diff --git a/install_ubuntu_intel.sh b/install_ubuntu_intel.sh
index 782376d..f01e354 100755
--- a/install_ubuntu_intel.sh
+++ b/install_ubuntu_intel.sh
@@ -9,6 +9,6 @@ set -e
 apt-get -y install build-essential\
 	libswresample-dev libavformat-dev libavcodec-dev libavutil-dev libavfilter-dev\
 	libglvnd-dev libx11-dev libxcomposite-dev libxrandr-dev\
-	libpulse-dev libva2 libegl-mesa0 intel-media-va-driver
+	libpulse-dev libva2 libegl-mesa0 intel-media-va-driver libdrm-dev libcap-dev
 
 ./install.sh
diff --git a/install_ubuntu_nvidia.sh b/install_ubuntu_nvidia.sh
index d776c93..1eeb7f5 100755
--- a/install_ubuntu_nvidia.sh
+++ b/install_ubuntu_nvidia.sh
@@ -9,6 +9,7 @@ set -e
 apt-get -y install build-essential\
 	libswresample-dev libavformat-dev libavcodec-dev libavutil-dev libavfilter-dev\
 	libglvnd-dev libx11-dev libxcomposite-dev libxrandr-dev\
-	libpulse-dev libva2 libxnvctrl0 libnvidia-compute libnvidia-encode libnvidia-fbc1
+	libpulse-dev libva2 libxnvctrl0 libnvidia-compute libnvidia-encode libnvidia-fbc1\
+	libdrm-dev libcap-dev
 
 ./install.sh
diff --git a/project.conf b/project.conf
index 5b9358f..2f3637d 100644
--- a/project.conf
+++ b/project.conf
@@ -4,6 +4,9 @@ type = "executable"
 version = "1.3.0"
 platforms = ["posix"]
 
+[config]
+ignore_dirs = ["src/kms"]
+
 [dependencies]
 libavcodec = ">=58"
 libavformat = ">=58"
diff --git a/src/capture/kms_vaapi.c b/src/capture/kms_vaapi.c
new file mode 100644
index 0000000..7e42011
--- /dev/null
+++ b/src/capture/kms_vaapi.c
@@ -0,0 +1,470 @@
+#include "../../include/capture/kms_vaapi.h"
+#include "../../include/kms/kms_client.h"
+#include "../../include/egl.h"
+#include "../../include/utils.h"
+#include <stdlib.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <assert.h>
+#include <X11/Xlib.h>
+#include <libavutil/hwcontext.h>
+#include <libavutil/hwcontext_vaapi.h>
+#include <libavutil/frame.h>
+#include <libavcodec/avcodec.h>
+#include <va/va.h>
+#include <va/va_drmcommon.h>
+
+typedef struct {
+    gsr_capture_kms_vaapi_params params;
+    bool should_stop;
+    bool stop_is_error;
+    bool created_hw_frame;
+
+    gsr_egl egl;
+    
+    gsr_kms_client kms_client;
+
+    uint32_t fourcc;
+    uint64_t modifiers;
+    int dmabuf_fd;
+    uint32_t pitch;
+    uint32_t offset;
+    vec2i kms_size;
+
+    vec2i capture_pos;
+    vec2i capture_size;
+    bool screen_capture;
+
+    VADisplay va_dpy;
+    VAConfigID config_id;
+    VAContextID context_id;
+    VASurfaceID input_surface;
+    VABufferID buffer_id;
+    VARectangle input_region;
+} gsr_capture_kms_vaapi;
+
+static int max_int(int a, int b) {
+    return a > b ? a : b;
+}
+
+static void gsr_capture_kms_vaapi_stop(gsr_capture *cap, AVCodecContext *video_codec_context);
+
+static bool drm_create_codec_context(gsr_capture_kms_vaapi *cap_kms, AVCodecContext *video_codec_context) {
+    AVBufferRef *device_ctx;
+    if(av_hwdevice_ctx_create(&device_ctx, AV_HWDEVICE_TYPE_VAAPI, "/dev/dri/renderD128", NULL, 0) < 0) {
+        fprintf(stderr, "Error: Failed to create hardware device context\n");
+        return false;
+    }
+
+    AVBufferRef *frame_context = av_hwframe_ctx_alloc(device_ctx);
+    if(!frame_context) {
+        fprintf(stderr, "Error: Failed to create hwframe context\n");
+        av_buffer_unref(&device_ctx);
+        return false;
+    }
+
+    AVHWFramesContext *hw_frame_context =
+        (AVHWFramesContext *)frame_context->data;
+    hw_frame_context->width = video_codec_context->width;
+    hw_frame_context->height = video_codec_context->height;
+    hw_frame_context->sw_format = AV_PIX_FMT_NV12;//AV_PIX_FMT_0RGB32;//AV_PIX_FMT_YUV420P;//AV_PIX_FMT_0RGB32;//AV_PIX_FMT_NV12;
+    hw_frame_context->format = video_codec_context->pix_fmt;
+    hw_frame_context->device_ref = device_ctx;
+    hw_frame_context->device_ctx = (AVHWDeviceContext*)device_ctx->data;
+
+    hw_frame_context->initial_pool_size = 1; // TODO: (and in other places)
+
+    AVVAAPIDeviceContext *vactx =((AVHWDeviceContext*)device_ctx->data)->hwctx;
+    cap_kms->va_dpy = vactx->display;
+
+    if (av_hwframe_ctx_init(frame_context) < 0) {
+        fprintf(stderr, "Error: Failed to initialize hardware frame context "
+                        "(note: ffmpeg version needs to be > 4.0)\n");
+        av_buffer_unref(&device_ctx);
+        //av_buffer_unref(&frame_context);
+        return false;
+    }
+
+    video_codec_context->hw_device_ctx = av_buffer_ref(device_ctx);
+    video_codec_context->hw_frames_ctx = av_buffer_ref(frame_context);
+    return true;
+}
+
+#define DRM_FORMAT_MOD_INVALID 72057594037927935
+
+// TODO: On monitor reconfiguration, find monitor x, y, width and height again. Do the same for nvfbc.
+
+static int gsr_capture_kms_vaapi_start(gsr_capture *cap, AVCodecContext *video_codec_context) {
+    gsr_capture_kms_vaapi *cap_kms = cap->priv;
+
+    // TODO: Allow specifying another card, and in other places (TODO: Use /dev/dri/renderD128?)
+    if(gsr_kms_client_init(&cap_kms->kms_client, "/dev/dri/card0", cap_kms->params.program_dir) != 0) {
+        return -1;
+    }
+
+    gsr_kms_response kms_response;
+    if(gsr_kms_client_get_kms(&cap_kms->kms_client, &kms_response) != 0) {
+        fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_start: failed to get kms, error: %d (%s)\n", kms_response.result, kms_response.data.err_msg);
+        gsr_capture_kms_vaapi_stop(cap, video_codec_context);
+        return -1;
+    }
+
+    cap_kms->dmabuf_fd = kms_response.data.fd.fd;
+    cap_kms->pitch = kms_response.data.fd.pitch;
+    cap_kms->offset = kms_response.data.fd.offset;
+    cap_kms->fourcc = kms_response.data.fd.pixel_format;
+    cap_kms->modifiers = kms_response.data.fd.modifier;
+
+    // TODO: Update on monitor reconfiguration, make sure to update on window focus change (maybe for kms update each second?), needs to work on focus change to the witcher 3 on full window, fullscreen firefox, etc
+    gsr_monitor monitor;
+    if(strcmp(cap_kms->params.display_to_capture, "screen") == 0) {
+        monitor.pos.x = 0;
+        monitor.pos.y = 0;
+        monitor.size.x = XWidthOfScreen(DefaultScreenOfDisplay(cap_kms->params.dpy));
+        monitor.size.y = XHeightOfScreen(DefaultScreenOfDisplay(cap_kms->params.dpy));
+        cap_kms->screen_capture = true;
+    } else if(!get_monitor_by_name(cap_kms->params.dpy, cap_kms->params.display_to_capture, &monitor)) {
+        fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_start: failed to find monitor by name \"%s\"\n", cap_kms->params.display_to_capture);
+        gsr_capture_kms_vaapi_stop(cap, video_codec_context);
+        return -1;
+    }
+
+    cap_kms->capture_pos = monitor.pos;
+    cap_kms->capture_size = monitor.size;
+
+    cap_kms->kms_size.x = kms_response.data.fd.width;
+    cap_kms->kms_size.y = kms_response.data.fd.height;
+
+    if(!gsr_egl_load(&cap_kms->egl, cap_kms->params.dpy)) {
+        fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_start: failed to load opengl\n");
+        return -1;
+    }
+
+    /* Disable vsync */
+    cap_kms->egl.eglSwapInterval(cap_kms->egl.egl_display, 0);
+
+    video_codec_context->width = max_int(2, cap_kms->capture_size.x & ~1);
+    video_codec_context->height = max_int(2, cap_kms->capture_size.y & ~1);
+
+    if(!drm_create_codec_context(cap_kms, video_codec_context)) {
+        gsr_capture_kms_vaapi_stop(cap, video_codec_context);
+        return -1;
+    }
+
+    //cap_kms->window_resize_timer = clock_get_monotonic_seconds(); // TODO:
+    return 0;
+}
+
+static void gsr_capture_kms_vaapi_tick(gsr_capture *cap, AVCodecContext *video_codec_context, AVFrame **frame) {
+    gsr_capture_kms_vaapi *cap_kms = cap->priv;
+
+    // TODO:
+    //cap_kms->egl.glClear(GL_COLOR_BUFFER_BIT);
+
+    //const double window_resize_timeout = 1.0; // 1 second
+    if(!cap_kms->created_hw_frame) {
+        if(cap_kms->buffer_id) {
+            vaDestroyBuffer(cap_kms->va_dpy, cap_kms->buffer_id);
+            cap_kms->buffer_id = 0;
+        }
+
+        if(cap_kms->context_id) {
+            vaDestroyContext(cap_kms->va_dpy, cap_kms->context_id);
+            cap_kms->context_id = 0;
+        }
+
+        if(cap_kms->config_id) {
+            vaDestroyConfig(cap_kms->va_dpy, cap_kms->config_id);
+            cap_kms->config_id = 0;
+        }
+
+        if(cap_kms->input_surface) {
+            vaDestroySurfaces(cap_kms->va_dpy, &cap_kms->input_surface, 1);
+            cap_kms->input_surface = 0;
+        }
+
+        if(!cap_kms->created_hw_frame) {
+            cap_kms->created_hw_frame = true;
+            av_frame_free(frame);
+            *frame = av_frame_alloc();
+            if(!frame) {
+                fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_tick: failed to allocate frame\n");
+                cap_kms->should_stop = true;
+                cap_kms->stop_is_error = true;
+                return;
+            }
+            (*frame)->format = video_codec_context->pix_fmt;
+            (*frame)->width = video_codec_context->width;
+            (*frame)->height = video_codec_context->height;
+            (*frame)->color_range = video_codec_context->color_range;
+            (*frame)->color_primaries = video_codec_context->color_primaries;
+            (*frame)->color_trc = video_codec_context->color_trc;
+            (*frame)->colorspace = video_codec_context->colorspace;
+            (*frame)->chroma_location = video_codec_context->chroma_sample_location;
+
+            int res = av_hwframe_get_buffer(video_codec_context->hw_frames_ctx, *frame, 0);
+            if(res < 0) {
+                fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_tick: av_hwframe_get_buffer failed: %d\n", res);
+                cap_kms->should_stop = true;
+                cap_kms->stop_is_error = true;
+                return;
+            }
+        }
+
+        uintptr_t dmabuf = cap_kms->dmabuf_fd;
+
+        VASurfaceAttribExternalBuffers buf = {0};
+        buf.pixel_format = VA_FOURCC_BGRX; // VA_FOURCC_XRGB
+        buf.width = cap_kms->kms_size.x;
+        buf.height = cap_kms->kms_size.y;
+        buf.data_size = cap_kms->kms_size.y * cap_kms->pitch;
+        buf.num_planes = 1;
+        buf.pitches[0] = cap_kms->pitch;
+        buf.offsets[0] = cap_kms->offset;
+        buf.buffers = &dmabuf;
+        buf.num_buffers = 1;
+        buf.flags = 0;
+        buf.private_data = 0;
+
+        VADRMFormatModifierList modifier_list = {0};
+        modifier_list.modifiers = &cap_kms->modifiers;
+        modifier_list.num_modifiers = 1;
+
+        #define VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME        0x20000000
+
+        VASurfaceAttrib attribs[3] = {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; // TODO: prime1 instead?
+        attribs[1].type = VASurfaceAttribExternalBufferDescriptor;
+        attribs[1].flags = VA_SURFACE_ATTRIB_SETTABLE;
+        attribs[1].value.type = VAGenericValueTypePointer;
+        attribs[1].value.value.p = &buf;
+
+        int num_attribs = 2;
+        if(cap_kms->modifiers != DRM_FORMAT_MOD_INVALID) {
+            attribs[2].type = VASurfaceAttribDRMFormatModifiers;
+            attribs[2].flags = VA_SURFACE_ATTRIB_SETTABLE;
+            attribs[2].value.type = VAGenericValueTypePointer;
+            attribs[2].value.value.p = &modifier_list;
+            ++num_attribs;
+        }
+        
+        VAStatus va_status = vaCreateSurfaces(cap_kms->va_dpy, VA_RT_FORMAT_RGB32, cap_kms->kms_size.x, cap_kms->kms_size.y, &cap_kms->input_surface, 1, attribs, num_attribs);
+        if(va_status != VA_STATUS_SUCCESS) {
+            fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_tick: vaCreateSurfaces failed: %d\n", va_status);
+            cap_kms->should_stop = true;
+            cap_kms->stop_is_error = true;
+            return;
+        }
+
+        //vaBeginPicture(cap_kms->va_dpy, )
+
+        va_status = vaCreateConfig(cap_kms->va_dpy, VAProfileNone, VAEntrypointVideoProc, NULL, 0, &cap_kms->config_id);
+        if(va_status != VA_STATUS_SUCCESS) {
+            fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_tick: vaCreateConfig failed: %d\n", va_status);
+            cap_kms->should_stop = true;
+            cap_kms->stop_is_error = true;
+            return;
+        }
+
+        VASurfaceID target_surface_id = (uintptr_t)(*frame)->data[3];
+        va_status = vaCreateContext(cap_kms->va_dpy, cap_kms->config_id, cap_kms->kms_size.x, cap_kms->kms_size.y, VA_PROGRESSIVE, &target_surface_id, 1, &cap_kms->context_id);
+        if(va_status != VA_STATUS_SUCCESS) {
+            fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_tick: vaCreateContext failed: %d\n", va_status);
+            cap_kms->should_stop = true;
+            cap_kms->stop_is_error = true;
+            return;
+        }
+
+        cap_kms->input_region = (VARectangle) {
+            .x = cap_kms->capture_pos.x,
+            .y = cap_kms->capture_pos.y,
+            .width = cap_kms->capture_size.x,
+            .height = cap_kms->capture_size.y
+        };
+
+        // Copying a surface to another surface will automatically perform the color conversion. Thanks vaapi!
+        VAProcPipelineParameterBuffer params = {0};
+        params.surface = cap_kms->input_surface;
+        if(cap_kms->screen_capture)
+            params.surface_region = NULL;
+        else
+            params.surface_region = &cap_kms->input_region;
+        params.output_region = NULL;
+        params.output_background_color = 0;
+        params.filter_flags = VA_FRAME_PICTURE;
+        // TODO: Colors
+        params.input_color_properties.color_range = (*frame)->color_range == AVCOL_RANGE_JPEG ? VA_SOURCE_RANGE_FULL : VA_SOURCE_RANGE_REDUCED;
+        params.output_color_properties.color_range = (*frame)->color_range == AVCOL_RANGE_JPEG ? VA_SOURCE_RANGE_FULL : VA_SOURCE_RANGE_REDUCED;
+
+        va_status = vaCreateBuffer(cap_kms->va_dpy, cap_kms->context_id, VAProcPipelineParameterBufferType, sizeof(params), 1, &params, &cap_kms->buffer_id);
+        if(va_status != VA_STATUS_SUCCESS) {
+            fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_tick: vaCreateBuffer failed: %d\n", va_status);
+            cap_kms->should_stop = true;
+            cap_kms->stop_is_error = true;
+            return;
+        }
+
+        // Clear texture with black background because the source texture (window_texture_get_opengl_texture_id(&cap_kms->window_texture))
+        // might be smaller than cap_kms->target_texture_id
+        // TODO:
+        //cap_kms->egl.glClearTexImage(cap_kms->target_texture_id, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
+    }
+}
+
+static bool gsr_capture_kms_vaapi_should_stop(gsr_capture *cap, bool *err) {
+    gsr_capture_kms_vaapi *cap_kms = cap->priv;
+    if(cap_kms->should_stop) {
+        if(err)
+            *err = cap_kms->stop_is_error;
+        return true;
+    }
+
+    if(err)
+        *err = false;
+    return false;
+}
+
+static int gsr_capture_kms_vaapi_capture(gsr_capture *cap, AVFrame *frame) {
+    gsr_capture_kms_vaapi *cap_kms = cap->priv;
+
+    VASurfaceID target_surface_id = (uintptr_t)frame->data[3];
+
+    VAStatus va_status = vaBeginPicture(cap_kms->va_dpy, cap_kms->context_id, target_surface_id);
+    if(va_status != VA_STATUS_SUCCESS) {
+        static bool error_printed = false;
+        if(!error_printed) {
+            error_printed = true;
+            fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_tick: vaBeginPicture failed: %d\n", va_status);
+        }
+        return -1;
+    }
+
+    va_status = vaRenderPicture(cap_kms->va_dpy, cap_kms->context_id, &cap_kms->buffer_id, 1);
+    if(va_status != VA_STATUS_SUCCESS) {
+        vaEndPicture(cap_kms->va_dpy, cap_kms->context_id);
+        static bool error_printed = false;
+        if(!error_printed) {
+            error_printed = true;
+            fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_tick: vaRenderPicture failed: %d\n", va_status);
+        }
+        return -1;
+    }
+
+    va_status = vaEndPicture(cap_kms->va_dpy, cap_kms->context_id);
+    if(va_status != VA_STATUS_SUCCESS) {
+        static bool error_printed = false;
+        if(!error_printed) {
+            error_printed = true;
+            fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_tick: vaEndPicture failed: %d\n", va_status);
+        }
+        return -1;
+    }
+
+    // TODO: Needed?
+    //vaSyncSurface(cap_kms->va_dpy, cap_kms->input_surface);
+    //vaSyncSurface(cap_kms->va_dpy, target_surface_id);
+
+    // TODO: Remove
+    //cap_kms->egl.eglSwapBuffers(cap_kms->egl.egl_display, cap_kms->egl.egl_surface);
+
+    return 0;
+}
+
+static void gsr_capture_kms_vaapi_stop(gsr_capture *cap, AVCodecContext *video_codec_context) {
+    gsr_capture_kms_vaapi *cap_kms = cap->priv;
+
+    if(cap_kms->buffer_id) {
+        vaDestroyBuffer(cap_kms->va_dpy, cap_kms->buffer_id);
+        cap_kms->buffer_id = 0;
+    }
+
+    if(cap_kms->context_id) {
+        vaDestroyContext(cap_kms->va_dpy, cap_kms->context_id);
+        cap_kms->context_id = 0;
+    }
+
+    if(cap_kms->config_id) {
+        vaDestroyConfig(cap_kms->va_dpy, cap_kms->config_id);
+        cap_kms->config_id = 0;
+    }
+
+    if(cap_kms->input_surface) {
+        vaDestroySurfaces(cap_kms->va_dpy, &cap_kms->input_surface, 1);
+        cap_kms->input_surface = 0;
+    }
+
+    if(cap_kms->dmabuf_fd > 0) {
+        close(cap_kms->dmabuf_fd);
+        cap_kms->dmabuf_fd = 0;
+    }
+
+    if(video_codec_context->hw_device_ctx)
+        av_buffer_unref(&video_codec_context->hw_device_ctx);
+    if(video_codec_context->hw_frames_ctx)
+        av_buffer_unref(&video_codec_context->hw_frames_ctx);
+
+    gsr_egl_unload(&cap_kms->egl);
+    gsr_kms_client_deinit(&cap_kms->kms_client);
+}
+
+static void gsr_capture_kms_vaapi_destroy(gsr_capture *cap, AVCodecContext *video_codec_context) {
+    (void)video_codec_context;
+    gsr_capture_kms_vaapi *cap_kms = cap->priv;
+    if(cap->priv) {
+        gsr_capture_kms_vaapi_stop(cap, video_codec_context);
+        free((void*)cap_kms->params.display_to_capture);
+        cap_kms->params.display_to_capture = NULL;
+        free(cap->priv);
+        cap->priv = NULL;
+    }
+    free(cap);
+}
+
+gsr_capture* gsr_capture_kms_vaapi_create(const gsr_capture_kms_vaapi_params *params) {
+    if(!params) {
+        fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_create params is NULL\n");
+        return NULL;
+    }
+
+    gsr_capture *cap = calloc(1, sizeof(gsr_capture));
+    if(!cap)
+        return NULL;
+
+    gsr_capture_kms_vaapi *cap_kms = calloc(1, sizeof(gsr_capture_kms_vaapi));
+    if(!cap_kms) {
+        free(cap);
+        return NULL;
+    }
+
+    Display *display = XOpenDisplay(NULL);
+    if(!display) {
+        fprintf(stderr, "gsr error: gsr_capture_kms_vaapi_create failed: XOpenDisplay failed\n");
+        free(cap);
+        free(cap_kms);
+        return NULL;
+    }
+
+    const char *display_to_capture = strdup(params->display_to_capture);
+    if(!display_to_capture) {
+        free(cap);
+        free(cap_kms);
+        return NULL;
+    }
+
+    cap_kms->params = *params;
+    cap_kms->params.display_to_capture = display_to_capture;
+    
+    *cap = (gsr_capture) {
+        .start = gsr_capture_kms_vaapi_start,
+        .tick = gsr_capture_kms_vaapi_tick,
+        .should_stop = gsr_capture_kms_vaapi_should_stop,
+        .capture = gsr_capture_kms_vaapi_capture,
+        .destroy = gsr_capture_kms_vaapi_destroy,
+        .priv = cap_kms
+    };
+
+    return cap;
+}
diff --git a/src/capture/nvfbc.c b/src/capture/nvfbc.c
index a7bb16a..c45712d 100644
--- a/src/capture/nvfbc.c
+++ b/src/capture/nvfbc.c
@@ -444,6 +444,7 @@ static void gsr_capture_nvfbc_destroy(gsr_capture *cap, AVCodecContext *video_co
         gsr_cuda_unload(&cap_nvfbc->cuda);
         dlclose(cap_nvfbc->library);
         free((void*)cap_nvfbc->params.display_to_capture);
+        cap_nvfbc->params.display_to_capture = NULL;
         free(cap->priv);
         cap->priv = NULL;
     }
diff --git a/src/capture/xcomposite_cuda.c b/src/capture/xcomposite_cuda.c
index 2b0d1a1..32b0d83 100644
--- a/src/capture/xcomposite_cuda.c
+++ b/src/capture/xcomposite_cuda.c
@@ -2,7 +2,7 @@
 #include "../../include/egl.h"
 #include "../../include/cuda.h"
 #include "../../include/window_texture.h"
-#include "../../include/time.h"
+#include "../../include/utils.h"
 #include <libavutil/hwcontext.h>
 #include <libavutil/hwcontext_cuda.h>
 #include <libavutil/frame.h>
diff --git a/src/capture/xcomposite_vaapi.c b/src/capture/xcomposite_vaapi.c
index cfe99dc..f12a978 100644
--- a/src/capture/xcomposite_vaapi.c
+++ b/src/capture/xcomposite_vaapi.c
@@ -1,13 +1,12 @@
 #include "../../include/capture/xcomposite_vaapi.h"
 #include "../../include/egl.h"
 #include "../../include/window_texture.h"
-#include "../../include/time.h"
+#include "../../include/utils.h"
 #include <stdlib.h>
 #include <stdio.h>
 #include <unistd.h>
 #include <assert.h>
 #include <X11/Xlib.h>
-#include <X11/extensions/Xcomposite.h>
 #include <libavutil/hwcontext.h>
 #include <libavutil/hwcontext_vaapi.h>
 #include <libavutil/frame.h>
@@ -399,6 +398,8 @@ static void gsr_capture_xcomposite_vaapi_tick(gsr_capture *cap, AVCodecContext *
         buf.flags = 0;
         buf.private_data = 0;
 
+        // TODO: Use VASurfaceAttribDRMFormatModifiers to set modifier if modifier is not INVALID
+
         #define VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME        0x20000000
 
         VASurfaceAttrib attribs[2] = {0};
@@ -448,10 +449,13 @@ static void gsr_capture_xcomposite_vaapi_tick(gsr_capture *cap, AVCodecContext *
         // Copying a surface to another surface will automatically perform the color conversion. Thanks vaapi!
         VAProcPipelineParameterBuffer params = {0};
         params.surface = cap_xcomp->input_surface;
-        params.surface_region = NULL; // TODO: Use when using kmsgrab to restrict region to captured monitor
-        params.output_region = &cap_xcomp->output_region;
+        params.surface_region = NULL;
+        if(cap_xcomp->params.follow_focused)
+            params.output_region = &cap_xcomp->output_region;
+        else
+            params.output_region = NULL;
         params.output_background_color = 0;
-        //params.filter_flags = VA_FRAME_PICTURE;
+        params.filter_flags = VA_FRAME_PICTURE;
         // TODO: Colors
         params.input_color_properties.color_range = (*frame)->color_range == AVCOL_RANGE_JPEG ? VA_SOURCE_RANGE_FULL : VA_SOURCE_RANGE_REDUCED;
         params.output_color_properties.color_range = (*frame)->color_range == AVCOL_RANGE_JPEG ? VA_SOURCE_RANGE_FULL : VA_SOURCE_RANGE_REDUCED;
diff --git a/src/cuda.c b/src/cuda.c
index ea14e8b..32b447e 100644
--- a/src/cuda.c
+++ b/src/cuda.c
@@ -1,6 +1,8 @@
 #include "../include/cuda.h"
 #include "../include/library_loader.h"
 #include <string.h>
+#include <stdio.h>
+#include <dlfcn.h>
 
 bool gsr_cuda_load(gsr_cuda *self, Display *display, bool do_overclock) {
     memset(self, 0, sizeof(gsr_cuda));
diff --git a/src/egl.c b/src/egl.c
index b991ff4..eb5bf9a 100644
--- a/src/egl.c
+++ b/src/egl.c
@@ -1,6 +1,8 @@
 #include "../include/egl.h"
 #include "../include/library_loader.h"
 #include <string.h>
+#include <stdio.h>
+#include <dlfcn.h>
 
 static bool gsr_egl_create_window(gsr_egl *self) {
     EGLConfig  ecfg;
diff --git a/src/kms/kms_client.c b/src/kms/kms_client.c
new file mode 100644
index 0000000..ededbe9
--- /dev/null
+++ b/src/kms/kms_client.c
@@ -0,0 +1,234 @@
+#include "../../include/kms/kms_client.h"
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <unistd.h>
+#include <signal.h>
+#include <limits.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+#include <sys/wait.h>
+#include <sys/capability.h>
+
+static int send_msg_to_server(int server_fd, gsr_kms_request *request) {
+    struct iovec iov;
+    iov.iov_base = request;
+    iov.iov_len = sizeof(*request);
+
+    struct msghdr request_message = {0};
+    request_message.msg_iov = &iov;
+    request_message.msg_iovlen = 1;
+
+    return sendmsg(server_fd, &request_message, 0);
+}
+
+static int recv_msg_from_server(int server_fd, gsr_kms_response *response) {
+    struct iovec iov;
+    iov.iov_base = response;
+    iov.iov_len = sizeof(*response);
+
+    struct msghdr response_message = {0};
+    response_message.msg_iov = &iov;
+    response_message.msg_iovlen = 1;
+
+    char cmsgbuf[CMSG_SPACE(sizeof(int))];
+    memset(cmsgbuf, 0, sizeof(cmsgbuf));
+    response_message.msg_control = cmsgbuf;
+    response_message.msg_controllen = sizeof(cmsgbuf);
+
+    int res = recvmsg(server_fd, &response_message, MSG_WAITALL);
+    if(res <= 0)
+        return res;
+
+    if(response->result == KMS_RESULT_OK)
+        response->data.fd.fd = 0;
+
+    struct cmsghdr *cmsg = CMSG_FIRSTHDR(&response_message);
+    if(cmsg) {
+        fprintf(stderr, "got cmsg, %d\n", cmsg->cmsg_type);
+        if(cmsg->cmsg_type == SCM_RIGHTS) {
+            int kms_fd = 0;
+            memcpy(&kms_fd, CMSG_DATA(cmsg), sizeof(int));
+            fprintf(stderr, "kms fd: %d\n", kms_fd);
+            response->data.fd.fd = kms_fd;
+        }
+    }
+
+    return res;
+}
+
+int gsr_kms_client_init(gsr_kms_client *self, const char *card_path, const char *program_dir) {
+    self->kms_server_pid = -1;
+    self->card_path = NULL;
+    self->socket_fd = -1;
+    self->client_fd = -1;
+    self->socket_path[0] = '\0';
+    struct sockaddr_un local_addr = {0};
+    struct sockaddr_un remote_addr = {0};
+
+    // TODO: Check if gsr-kms-server is installed
+    // TODO: Check if pkexec is installed
+
+    char server_filepath[PATH_MAX];
+    snprintf(server_filepath, sizeof(server_filepath), "%s/%s", program_dir, "gsr-kms-server");
+
+    int has_perm = 0;
+    cap_t kms_server_cap = cap_get_file(server_filepath);
+    if(kms_server_cap) {
+        cap_flag_value_t res = 0;
+        cap_get_flag(kms_server_cap, CAP_SYS_ADMIN, CAP_PERMITTED, &res);
+        if(res == CAP_SET) {
+            //fprintf(stderr, "has permission!\n");
+            has_perm = 1;
+        } else {
+            //fprintf(stderr, "No permission:(\n");
+        }
+        cap_free(kms_server_cap);
+    } else {
+        if(errno == ENODATA)
+            fprintf(stderr, "gsr info: gsr_kms_client_init: gsr-kms-server is missing sys_admin cap and will require root authentication. To bypass this automatically, run: sudo setcap cap_sys_admin+ep '%s'\n", server_filepath);
+        else
+            fprintf(stderr, "failed to get cap\n");
+    }
+
+    self->card_path = strdup(card_path);
+    if(!self->card_path) {
+        fprintf(stderr, "gsr error: gsr_kms_client_init: failed to duplicate card_path\n");
+        goto err;
+    }
+
+    strcpy(self->socket_path, "/tmp/gsr-kms-socket-XXXXXX");
+    if(!tmpnam(self->socket_path)) {
+        fprintf(stderr, "gsr error: gsr_kms_client_init: mkstemp failed, error: %s\n", strerror(errno));
+        goto err;
+    }
+
+    self->socket_fd = socket(AF_UNIX, SOCK_STREAM, 0);
+    if(self->socket_fd == -1) {
+        fprintf(stderr, "gsr error: gsr_kms_client_init: socket failed, error: %s\n", strerror(errno));
+        goto err;
+    }
+
+    local_addr.sun_family = AF_UNIX;
+    strncpy(local_addr.sun_path, self->socket_path, sizeof(local_addr.sun_path));
+    if(bind(self->socket_fd, (struct sockaddr*)&local_addr, sizeof(local_addr.sun_family) + strlen(local_addr.sun_path)) == -1) {
+        fprintf(stderr, "gsr error: gsr_kms_client_init: failed to bind socket, error: %s\n", strerror(errno));
+        goto err;
+    }
+
+    if(listen(self->socket_fd, 1) == -1) {
+        fprintf(stderr, "gsr error: gsr_kms_client_init: failed to listen on socket, error: %s\n", strerror(errno));
+        goto err;
+    }
+
+    pid_t pid = fork();
+    if(pid == -1) {
+        fprintf(stderr, "gsr error: gsr_kms_client_init: fork failed, error: %s\n", strerror(errno));
+        goto err;
+    } else if(pid == 0) { /* child */
+        if(has_perm) {
+            const char *args[] = { server_filepath, self->socket_path, NULL };
+            execvp(args[0], (char *const*)args);
+        } else {
+            const char *args[] = { "pkexec", server_filepath, self->socket_path, NULL };
+            execvp(args[0], (char *const*)args);
+        }
+        perror("execvp");
+        _exit(127);
+    } else { /* parent */
+        self->kms_server_pid = pid;
+    }
+
+    fprintf(stderr, "gsr info: gsr_kms_client_init: waiting for client to connect\n");
+    for(;;) {
+        struct timeval tv;
+        fd_set rfds;
+        FD_ZERO(&rfds);
+        FD_SET(self->socket_fd, &rfds);
+
+        tv.tv_sec = 0;
+        tv.tv_usec = 100 * 1000; // 100 ms
+
+        int select_res = select(1 + self->socket_fd, &rfds, NULL, NULL, &tv);
+        if(select_res > 0) {
+            socklen_t sock_len = 0;
+            self->client_fd = accept(self->socket_fd, (struct sockaddr*)&remote_addr, &sock_len);
+            if(self->client_fd == -1) {
+                fprintf(stderr, "gsr error: gsr_kms_client_init: accept failed on socket, error: %s\n", strerror(errno));
+                goto err;
+            }
+            break;
+        } else {
+            int status;
+            int wait_result = waitpid(self->kms_server_pid, &status, WNOHANG);
+            if(wait_result > 0) {
+                fprintf(stderr, "gsr error: gsr_kms_client_init: waitpid failed on kms server, error: %s\n", strerror(errno));
+                goto err;
+            } else if(wait_result > 0) {
+                fprintf(stderr, "gsr error: gsr_kms_client_init: kms server died\n");
+                goto err;
+            }
+        }
+    }
+    fprintf(stderr, "gsr info: gsr_kms_client_init: client connected\n");
+
+    return 0;
+
+    err:
+    gsr_kms_client_deinit(self);
+    return -1;
+}
+
+void gsr_kms_client_deinit(gsr_kms_client *self) {
+    if(self->card_path) {
+        free(self->card_path);
+        self->card_path = NULL;
+    }
+
+    if(self->client_fd != -1) {
+        close(self->client_fd);
+        self->client_fd = -1;
+    }
+
+    if(self->socket_fd != -1) {
+        close(self->socket_fd);
+        self->socket_fd = -1;
+    }
+
+    if(self->kms_server_pid != -1) {
+        kill(self->kms_server_pid, SIGINT);
+        int status;
+        waitpid(self->kms_server_pid, &status, 0);
+        self->kms_server_pid = -1;
+    }
+
+    if(self->socket_path[0] != '\0') {
+        remove(self->socket_path);
+        self->socket_path[0] = '\0';
+    }
+}
+
+int gsr_kms_client_get_kms(gsr_kms_client *self, gsr_kms_response *response) {
+    response->result = KMS_RESULT_FAILED_TO_SEND;
+    strcpy(response->data.err_msg, "failed to send");
+
+    gsr_kms_request request;
+    request.type = KMS_REQUEST_TYPE_GET_KMS;
+    strcpy(request.data.card_path, self->card_path);
+    if(send_msg_to_server(self->client_fd, &request) == -1) {
+        fprintf(stderr, "gsr error: gsr_kms_client_get_kms: failed to send request message to server\n");
+        return -1;
+    }
+
+    const int recv_res = recv_msg_from_server(self->client_fd, response);
+    if(recv_res == 0) {
+        fprintf(stderr, "gsr warning: gsr_kms_client_get_kms: kms server shut down\n");
+        return -1;
+    } else if(recv_res == -1) {
+        fprintf(stderr, "gsr error: gsr_kms_client_get_kms: failed to receive response\n");
+        return -1;
+    }
+
+    return 0;
+}
diff --git a/src/kms/kms_server.c b/src/kms/kms_server.c
new file mode 100644
index 0000000..f811050
--- /dev/null
+++ b/src/kms/kms_server.c
@@ -0,0 +1,242 @@
+#include "../../include/kms/kms_shared.h"
+
+#include <asm-generic/socket.h>
+#include <stdio.h>
+#include <string.h>
+#include <errno.h>
+#include <stdlib.h>
+
+#include <unistd.h>
+#include <fcntl.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+
+#include <xf86drm.h>
+#include <xf86drmMode.h>
+
+#define DRM_CLIENT_CAP_UNIVERSAL_PLANES 2
+
+static int max_int(int a, int b) {
+    return a > b ? a : b;
+}
+
+static int send_msg_to_client(int client_fd, gsr_kms_response *response, int *fds, int num_fds) {
+    struct iovec iov;
+    iov.iov_base = response;
+    iov.iov_len = sizeof(*response);
+
+    struct msghdr response_message = {0};
+    response_message.msg_iov = &iov;
+    response_message.msg_iovlen = 1;
+
+    char cmsgbuf[CMSG_SPACE(sizeof(int)) * max_int(1, num_fds)];
+    memset(cmsgbuf, 0, sizeof(cmsgbuf));
+
+    if(num_fds > 0) {
+        response_message.msg_control = cmsgbuf;
+        response_message.msg_controllen = sizeof(cmsgbuf);
+
+        int total_msg_len = 0;
+        struct cmsghdr *cmsg = NULL;
+        for(int i = 0; i < num_fds; ++i) {
+            if(i == 0)
+                cmsg = CMSG_FIRSTHDR(&response_message);
+            else
+                cmsg = CMSG_NXTHDR(&response_message, cmsg);
+
+            cmsg->cmsg_level = SOL_SOCKET;
+            cmsg->cmsg_type = SCM_RIGHTS;
+            cmsg->cmsg_len = CMSG_LEN(sizeof(int));
+            memcpy(CMSG_DATA(cmsg), &fds[i], sizeof(int));
+            total_msg_len += cmsg->cmsg_len;
+        }
+
+        response_message.msg_controllen = total_msg_len;
+    }
+
+    return sendmsg(client_fd, &response_message, 0);
+}
+
+static int get_kms(const char *card_path, gsr_kms_response *response) {
+    response->result = KMS_RESULT_OK;
+    response->data.fd.fd = 0;
+    response->data.fd.width = 0;
+    response->data.fd.height = 0;
+
+    const int drmfd = open(card_path, O_RDONLY);
+    if (drmfd < 0) {
+        response->result = KMS_RESULT_FAILED_TO_OPEN_CARD;
+        snprintf(response->data.err_msg, sizeof(response->data.err_msg), "failed to open %s, error: %s", card_path, strerror(errno));
+        return -1;
+    }
+
+    if (0 != drmSetClientCap(drmfd, DRM_CLIENT_CAP_UNIVERSAL_PLANES, 1)) {
+        response->result = KMS_RESULT_INSUFFICIENT_PERMISSIONS;
+        snprintf(response->data.err_msg, sizeof(response->data.err_msg), "drmSetClientCap failed, error: %s", strerror(errno));
+        return -1;
+    }
+
+    drmModePlaneResPtr planes = drmModeGetPlaneResources(drmfd);
+    if (!planes) {
+        response->result = KMS_RESULT_FAILED_TO_GET_KMS;
+        snprintf(response->data.err_msg, sizeof(response->data.err_msg), "failed to access planes, error: %s", strerror(errno));
+        return -1;
+    }
+
+    fprintf(stderr, "DRM planes %d:\n", planes->count_planes);
+    for (uint32_t i = 0; i < planes->count_planes; ++i) {
+        drmModePlanePtr plane = drmModeGetPlane(drmfd, planes->planes[i]);
+        if (!plane) {
+            fprintf(stderr, "Cannot get drmModePlanePtr for plane %#x: %s (%d)\n", planes->planes[i], strerror(errno), errno);
+            continue;
+        }
+
+        fprintf(stderr, "\t%d: fb_id=%#x\n", i, plane->fb_id);
+
+        if (!plane->fb_id)
+            goto plane_continue;
+
+        drmModeFB2Ptr drmfb = drmModeGetFB2(drmfd, plane->fb_id);
+        if (!drmfb) {
+            fprintf(stderr, "Cannot get drmModeFBPtr for fb %#x: %s (%d)\n", plane->fb_id, strerror(errno), errno);
+        } else {
+            if (!drmfb->handles[0]) {
+                fprintf(stderr, "\t\tFB handle for fb %#x is NULL\n", plane->fb_id);
+                fprintf(stderr, "\t\tPossible reason: not permitted to get FB handles. Do `sudo setcap cap_sys_admin+ep`\n");
+            } else {
+                int fb_fd = -1;
+                const int ret = drmPrimeHandleToFD(drmfd, drmfb->handles[0], 0, &fb_fd);
+                if (ret != 0 || fb_fd == -1) {
+                    fprintf(stderr, "Cannot get fd for fb %#x handle %#x: %s (%d)\n", plane->fb_id, drmfb->handles[0], strerror(errno), errno);
+                } else if(drmfb->width * drmfb->height > response->data.fd.width * response->data.fd.height) {
+                    if(response->data.fd.fd != 0) {
+                        close(response->data.fd.fd);
+                        response->data.fd.fd = 0;
+                    }
+
+                    response->data.fd.fd = fb_fd;
+                    response->data.fd.width = drmfb->width;
+                    response->data.fd.height = drmfb->height;
+                    response->data.fd.pitch = drmfb->pitches[0];
+                    response->data.fd.offset = drmfb->offsets[0];
+                    response->data.fd.pixel_format = drmfb->pixel_format;
+                    response->data.fd.modifier = drmfb->modifier;
+                    fprintf(stderr, "kms width: %u, height: %u, pixel format: %u, modifier: %lu\n", response->data.fd.width, response->data.fd.height, response->data.fd.pixel_format, response->data.fd.modifier);
+                } else {
+                    close(fb_fd);
+                }
+            }
+            drmModeFreeFB2(drmfb);
+        }
+
+    plane_continue:
+        drmModeFreePlane(plane);
+    }
+
+    drmModeFreePlaneResources(planes);
+    close(drmfd); // TODO?
+
+    if(response->data.fd.fd == 0) {
+        response->result = KMS_RESULT_NO_KMS_AVAILABLE;
+        snprintf(response->data.err_msg, sizeof(response->data.err_msg), "no kms found");
+        return -1;
+    }
+
+    return 0;
+}
+
+int main(int argc, char **argv) {
+    if(argc != 2) {
+        fprintf(stderr, "usage: kms_server <domain_socket_path>\n");
+        return 1;
+    }
+
+    const char *domain_socket_path = argv[1];
+    int socket_fd = socket(AF_UNIX, SOCK_STREAM, 0);
+    if(socket_fd == -1) {
+        fprintf(stderr, "kms server error: failed to create socket, error: %s\n", strerror(errno));
+        return 2;
+    }
+
+    fprintf(stderr, "kms server info: connecting to the server\n");
+    for(;;) {
+        struct sockaddr_un remote_addr = {0};
+        remote_addr.sun_family = AF_UNIX;
+        strncpy(remote_addr.sun_path, domain_socket_path, sizeof(remote_addr.sun_path));
+        // TODO: Check if parent disconnected
+        if(connect(socket_fd, (struct sockaddr*)&remote_addr, sizeof(remote_addr.sun_family) + strlen(remote_addr.sun_path)) == -1) {
+            if(errno == ECONNREFUSED || errno == ENOENT)
+                continue; // Host not ready yet? TODO: sleep
+            if(errno == EISCONN) // TODO?
+                break;
+            fprintf(stderr, "kms server error: connect failed, error: %s (%d)\n", strerror(errno), errno);
+            return 2;
+        }
+    }
+    fprintf(stderr, "kms server info: connected to the server\n");
+
+    int res = 0;
+    for(;;) {
+        gsr_kms_request request;
+        struct iovec iov;
+        iov.iov_base = &request;
+        iov.iov_len = sizeof(request);
+
+        struct msghdr request_message = {0};
+        request_message.msg_iov = &iov;
+        request_message.msg_iovlen = 1;
+        const int recv_res = recvmsg(socket_fd, &request_message, MSG_WAITALL);
+        if(recv_res == 0) {
+            fprintf(stderr, "kms server info: kms client shutdown, shutting down the server\n");
+            res = 3;
+            goto done;
+        } else if(recv_res == -1) {
+            const int err = errno;
+            fprintf(stderr, "kms server error: failed to read all data in client request (error: %s), ignoring\n", strerror(err));
+            if(err == EBADF) {
+                fprintf(stderr, "kms server error: invalid client fd, shutting down the server\n");
+                res = 3;
+                goto done;
+            }
+            continue;
+        }
+        request.data.card_path[254] = '\0';
+
+        switch(request.type) {
+            case KMS_REQUEST_TYPE_GET_KMS: {
+                gsr_kms_response response;
+                int kms_fd = 0;
+                if (get_kms(request.data.card_path, &response) == 0) {
+                    kms_fd = response.data.fd.fd;
+                }
+
+                if(send_msg_to_client(socket_fd, &response, &kms_fd, kms_fd == 0 ? 0 : 1) == -1) {
+                    fprintf(stderr, "kms server error: failed to respond to client KMS_REQUEST_TYPE_GET_KMS request\n");
+                    if(kms_fd != 0)
+                        close(kms_fd);
+                    break;
+                }
+
+                if(kms_fd != 0)
+                    close(kms_fd);
+
+                break;
+            }
+            default: {
+                gsr_kms_response response;
+                response.result = KMS_RESULT_INVALID_REQUEST;
+                snprintf(response.data.err_msg, sizeof(response.data.err_msg), "invalid request type %d, expected %d (%s)", request.type, KMS_REQUEST_TYPE_GET_KMS, "KMS_REQUEST_TYPE_GET_KMS");
+                fprintf(stderr, "%s\n", response.data.err_msg);
+                if(send_msg_to_client(socket_fd, &response, NULL, 0) == -1) {
+                    fprintf(stderr, "kms server error: failed to respond to client request\n");
+                    break;
+                }
+                break;
+            }
+        }
+    }
+
+    done:
+    close(socket_fd);
+    return res;
+}
diff --git a/src/library_loader.c b/src/library_loader.c
new file mode 100644
index 0000000..0aeee9b
--- /dev/null
+++ b/src/library_loader.c
@@ -0,0 +1,34 @@
+#include "../include/library_loader.h"
+
+#include <dlfcn.h>
+#include <stdbool.h>
+#include <stdio.h>
+
+void* dlsym_print_fail(void *handle, const char *name, bool required) {
+    dlerror();
+    void *sym = dlsym(handle, name);
+    char *err_str = dlerror();
+
+    if(!sym)
+        fprintf(stderr, "%s: dlsym(handle, \"%s\") failed, error: %s\n", required ? "error" : "warning", name, err_str ? err_str : "(null)");
+
+    return sym;
+}
+
+/* |dlsyms| should be null terminated */
+bool dlsym_load_list(void *handle, const dlsym_assign *dlsyms) {
+    bool success = true;
+    for(int i = 0; dlsyms[i].func; ++i) {
+        *dlsyms[i].func = dlsym_print_fail(handle, dlsyms[i].name, true);
+        if(!*dlsyms[i].func)
+            success = false;
+    }
+    return success;
+}
+
+/* |dlsyms| should be null terminated */
+void dlsym_load_list_optional(void *handle, const dlsym_assign *dlsyms) {
+    for(int i = 0; dlsyms[i].func; ++i) {
+        *dlsyms[i].func = dlsym_print_fail(handle, dlsyms[i].name, false);
+    }
+}
diff --git a/src/main.cpp b/src/main.cpp
index 6b794ad..5beabd8 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -2,8 +2,9 @@ extern "C" {
 #include "../include/capture/nvfbc.h"
 #include "../include/capture/xcomposite_cuda.h"
 #include "../include/capture/xcomposite_vaapi.h"
+#include "../include/capture/kms_vaapi.h"
 #include "../include/egl.h"
-#include "../include/time.h"
+#include "../include/utils.h"
 }
 
 #include <assert.h>
@@ -18,13 +19,10 @@ extern "C" {
 #include <signal.h>
 #include <sys/stat.h>
 
-#include <unistd.h>
-#include <fcntl.h>
+#include <libgen.h>
 
 #include "../include/sound.hpp"
 
-#include <X11/extensions/Xrandr.h>
-
 extern "C" {
 #include <libavutil/pixfmt.h>
 #include <libavcodec/avcodec.h>
@@ -55,71 +53,6 @@ static const int VIDEO_STREAM_INDEX = 0;
 
 static thread_local char av_error_buffer[AV_ERROR_MAX_STRING_SIZE];
 
-static const XRRModeInfo* get_mode_info(const XRRScreenResources *sr, RRMode id) {
-    for(int i = 0; i < sr->nmode; ++i) {
-        if(sr->modes[i].id == id)
-            return &sr->modes[i];
-    }    
-    return nullptr;
-}
-
-typedef void (*active_monitor_callback)(const XRROutputInfo *output_info, const XRRCrtcInfo *crt_info, const XRRModeInfo *mode_info, void *userdata);
-
-static void for_each_active_monitor_output(Display *display, active_monitor_callback callback, void *userdata) {
-    XRRScreenResources *screen_res = XRRGetScreenResources(display, DefaultRootWindow(display));
-    if(!screen_res)
-        return;
-
-    for(int i = 0; i < screen_res->noutput; ++i) {
-        XRROutputInfo *out_info = XRRGetOutputInfo(display, screen_res, screen_res->outputs[i]);
-        if(out_info && out_info->crtc && out_info->connection == RR_Connected) {
-            XRRCrtcInfo *crt_info = XRRGetCrtcInfo(display, screen_res, out_info->crtc);
-            if(crt_info && crt_info->mode) {
-                const XRRModeInfo *mode_info = get_mode_info(screen_res, crt_info->mode);
-                if(mode_info)
-                    callback(out_info, crt_info, mode_info, userdata);
-            }
-            if(crt_info)
-                XRRFreeCrtcInfo(crt_info);
-        }
-        if(out_info)
-            XRRFreeOutputInfo(out_info);
-    }    
-
-    XRRFreeScreenResources(screen_res);
-}
-
-typedef struct {
-    vec2i pos;
-    vec2i size;
-} gsr_monitor;
-
-typedef struct {
-    const char *name;
-    int name_len;
-    gsr_monitor *monitor;
-    bool found_monitor;
-} get_monitor_by_name_userdata;
-
-static void get_monitor_by_name_callback(const XRROutputInfo *output_info, const XRRCrtcInfo *crt_info, const XRRModeInfo *mode_info, void *userdata) {
-    get_monitor_by_name_userdata *data = (get_monitor_by_name_userdata*)userdata;
-    if(!data->found_monitor && data->name_len == output_info->nameLen && memcmp(data->name, output_info->name, data->name_len) == 0) {
-        data->monitor->pos = { crt_info->x, crt_info->y };
-        data->monitor->size = { (int)crt_info->width, (int)crt_info->height };
-        data->found_monitor = true;
-    }
-}
-
-static bool get_monitor_by_name(Display *display, const char *name, gsr_monitor *monitor) {
-    get_monitor_by_name_userdata userdata;
-    userdata.name = name;
-    userdata.name_len = strlen(name);
-    userdata.monitor = monitor;
-    userdata.found_monitor = false;
-    for_each_active_monitor_output(display, get_monitor_by_name_callback, &userdata);
-    return userdata.found_monitor;
-}
-
 static void monitor_output_callback_print(const XRROutputInfo *output_info, const XRRCrtcInfo *crt_info, const XRRModeInfo *mode_info, void *userdata) {
     fprintf(stderr, "    \"%.*s\"    (%dx%d+%d+%d)\n", output_info->nameLen, output_info->name, (int)crt_info->width, (int)crt_info->height, crt_info->x, crt_info->y);
 }
@@ -375,7 +308,7 @@ static AVCodecContext *create_video_codec_context(AVPixelFormat pix_fmt,
     }
     codec_context->max_b_frames = 0;
     codec_context->pix_fmt = pix_fmt;
-    //codec_context->color_range = AVCOL_RANGE_JPEG;
+    codec_context->color_range = AVCOL_RANGE_JPEG; // TODO: Amd/nvidia?
     //codec_context->color_primaries = AVCOL_PRI_BT709;
     //codec_context->color_trc = AVCOL_TRC_BT709;
     //codec_context->colorspace = AVCOL_SPC_BT709;
@@ -434,7 +367,8 @@ static AVCodecContext *create_video_codec_context(AVPixelFormat pix_fmt,
         // TODO: More options, better options
         //codec_context->bit_rate = codec_context->width * codec_context->height;
         av_opt_set(codec_context->priv_data, "rc_mode", "CQP", 0);
-        codec_context->global_quality = 4;
+        //codec_context->global_quality = 4;
+        codec_context->compression_level = 2;
     }
 
     //codec_context->rc_max_rate = codec_context->bit_rate;
@@ -687,7 +621,7 @@ static void open_video(AVCodecContext *codec_context, VideoQuality video_quality
 
         if(codec_context->codec_id == AV_CODEC_ID_H264) {
             av_dict_set(&options, "profile", "high", 0);
-            av_dict_set_int(&options, "quality", 14, 0);
+            av_dict_set_int(&options, "quality", 7, 0);
         } else {
             av_dict_set(&options, "profile", "main", 0);
         }
@@ -736,7 +670,7 @@ static void usage() {
     fprintf(stderr, "        and the video will only be saved when the gpu-screen-recorder is closed. This feature is similar to Nvidia's instant replay feature.\n");
     fprintf(stderr, "        This option has be between 5 and 1200. Note that the replay buffer size will not always be precise, because of keyframes. Optional, disabled by default.\n");
     fprintf(stderr, "\n");
-    fprintf(stderr, "  -k    Video codec to use. Should be either 'auto', 'h264' or 'h265'. Defaults to 'auto' which defaults to 'h265' unless recording at fps higher than 60.\n");
+    fprintf(stderr, "  -k    Video codec to use. Should be either 'auto', 'h264' or 'h265'. Defaults to 'auto' which defaults to 'h265' unless recording at fps higher than 60. Defaults to 'h264' on intel.\n");
     fprintf(stderr, "        Forcefully set to 'h264' if -c is 'flv'.\n");
     fprintf(stderr, "\n");
     fprintf(stderr, "  -ac   Audio codec to use. Should be either 'aac', 'opus' or 'flac'. Defaults to 'opus' for .mp4/.mkv files, otherwise defaults to 'aac'.\n");
@@ -1174,6 +1108,14 @@ int main(int argc, char **argv) {
     signal(SIGINT, int_handler);
     signal(SIGUSR1, save_replay_handler);
 
+    if(argc == 0)
+        usage();
+
+    char *program_dir = dirname(argv[0]);
+    char program_dir_full[PATH_MAX];
+    program_dir_full[0] = '\0';
+    realpath(program_dir, program_dir_full);
+
     //av_log_set_level(AV_LOG_TRACE);
 
     std::map<std::string, Arg> args = {
@@ -1408,11 +1350,6 @@ int main(int argc, char **argv) {
 
         follow_focused = true;
     } else if(contains_non_hex_number(window_str)) {
-        if(gpu_inf.vendor != GPU_VENDOR_NVIDIA) {
-            fprintf(stderr, "Error: recording a monitor is only supported on NVIDIA right now. Record \"focused\" instead for convenient fullscreen window recording\n");
-            return 2;
-        }
-
         if(strcmp(window_str, "screen") != 0 && strcmp(window_str, "screen-direct") != 0 && strcmp(window_str, "screen-direct-force") != 0) {
             gsr_monitor gmon;
             if(!get_monitor_by_name(dpy, window_str, &gmon)) {
@@ -1425,31 +1362,46 @@ int main(int argc, char **argv) {
             }
         }
 
-        const char *capture_target = window_str;
-        bool direct_capture = strcmp(window_str, "screen-direct") == 0;
-        if(direct_capture) {
-            capture_target = "screen";
-            // TODO: Temporary disable direct capture because push model causes stuttering when it's direct capturing. This might be a nvfbc bug. This does not happen when using a compositor.
-            direct_capture = false;
-            fprintf(stderr, "Warning: screen-direct has temporary been disabled as it causes stuttering. This is likely a NvFBC bug. Falling back to \"screen\".\n");
-        }
+        if(gpu_inf.vendor == GPU_VENDOR_NVIDIA) {
+            const char *capture_target = window_str;
+            bool direct_capture = strcmp(window_str, "screen-direct") == 0;
+            if(direct_capture) {
+                capture_target = "screen";
+                // TODO: Temporary disable direct capture because push model causes stuttering when it's direct capturing. This might be a nvfbc bug. This does not happen when using a compositor.
+                direct_capture = false;
+                fprintf(stderr, "Warning: screen-direct has temporary been disabled as it causes stuttering. This is likely a NvFBC bug. Falling back to \"screen\".\n");
+            }
 
-        if(strcmp(window_str, "screen-direct-force") == 0) {
-            direct_capture = true;
-            capture_target = "screen";
-        }
+            if(strcmp(window_str, "screen-direct-force") == 0) {
+                direct_capture = true;
+                capture_target = "screen";
+            }
 
-        gsr_capture_nvfbc_params nvfbc_params;
-        nvfbc_params.dpy = dpy;
-        nvfbc_params.display_to_capture = capture_target;
-        nvfbc_params.fps = fps;
-        nvfbc_params.pos = { 0, 0 };
-        nvfbc_params.size = { 0, 0 };
-        nvfbc_params.direct_capture = direct_capture;
-        nvfbc_params.overclock = overclock;
-        capture = gsr_capture_nvfbc_create(&nvfbc_params);
-        if(!capture)
-            return 1;
+            gsr_capture_nvfbc_params nvfbc_params;
+            nvfbc_params.dpy = dpy;
+            nvfbc_params.display_to_capture = capture_target;
+            nvfbc_params.fps = fps;
+            nvfbc_params.pos = { 0, 0 };
+            nvfbc_params.size = { 0, 0 };
+            nvfbc_params.direct_capture = direct_capture;
+            nvfbc_params.overclock = overclock;
+            capture = gsr_capture_nvfbc_create(&nvfbc_params);
+            if(!capture)
+                return 1;
+        } else {
+            const char *capture_target = window_str;
+            if(strcmp(window_str, "screen-direct") == 0 || strcmp(window_str, "screen-direct-force") == 0) {
+                capture_target = "screen";
+            }
+
+            gsr_capture_kms_vaapi_params kms_params;
+            kms_params.dpy = dpy;
+            kms_params.display_to_capture = capture_target;
+            kms_params.program_dir = program_dir_full;
+            capture = gsr_capture_kms_vaapi_create(&kms_params);
+            if(!capture)
+                return 1;
+        }
     } else {
         errno = 0;
         src_window_id = strtol(window_str, nullptr, 0);
@@ -1565,23 +1517,36 @@ int main(int argc, char **argv) {
     const double target_fps = 1.0 / (double)fps;
 
     if(strcmp(video_codec_to_use, "auto") == 0) {
-        const AVCodec *h265_codec = find_h265_encoder(gpu_inf.vendor);
-
-        // h265 generally allows recording at a higher resolution than h264 on nvidia cards. On a gtx 1080 4k is the max resolution for h264 but for h265 it's 8k.
-        // Another important info is that when recording at a higher fps than.. 60? h265 has very bad performance. For example when recording at 144 fps the fps drops to 1
-        // while with h264 the fps doesn't drop.
-        if(!h265_codec) {
-            fprintf(stderr, "Info: using h264 encoder because a codec was not specified and your gpu does not support h265\n");
-            video_codec_to_use = "h264";
-            video_codec = VideoCodec::H264;
-        } else if(fps > 60) {
-            fprintf(stderr, "Info: using h264 encoder because a codec was not specified and fps is more than 60\n");
-            video_codec_to_use = "h264";
-            video_codec = VideoCodec::H264;
+        if(gpu_inf.vendor == GPU_VENDOR_INTEL) {
+            const AVCodec *h264_codec = find_h264_encoder(gpu_inf.vendor);
+            if(!h264_codec) {
+                fprintf(stderr, "Info: using h265 encoder because a codec was not specified and your gpu does not support h264\n");
+                video_codec_to_use = "h265";
+                video_codec = VideoCodec::H265;
+            } else {
+                fprintf(stderr, "Info: using h264 encoder because a codec was not specified\n");
+                video_codec_to_use = "h264";
+                video_codec = VideoCodec::H264;
+            }
         } else {
-            fprintf(stderr, "Info: using h265 encoder because a codec was not specified\n");
-            video_codec_to_use = "h265";
-            video_codec = VideoCodec::H265;
+            const AVCodec *h265_codec = find_h265_encoder(gpu_inf.vendor);
+
+            // h265 generally allows recording at a higher resolution than h264 on nvidia cards. On a gtx 1080 4k is the max resolution for h264 but for h265 it's 8k.
+            // Another important info is that when recording at a higher fps than.. 60? h265 has very bad performance. For example when recording at 144 fps the fps drops to 1
+            // while with h264 the fps doesn't drop.
+            if(!h265_codec) {
+                fprintf(stderr, "Info: using h264 encoder because a codec was not specified and your gpu does not support h265\n");
+                video_codec_to_use = "h264";
+                video_codec = VideoCodec::H264;
+            } else if(fps > 60) {
+                fprintf(stderr, "Info: using h264 encoder because a codec was not specified and fps is more than 60\n");
+                video_codec_to_use = "h264";
+                video_codec = VideoCodec::H264;
+            } else {
+                fprintf(stderr, "Info: using h265 encoder because a codec was not specified\n");
+                video_codec_to_use = "h265";
+                video_codec = VideoCodec::H265;
+            }
         }
     }
 
@@ -1730,7 +1695,7 @@ int main(int argc, char **argv) {
 
     const double start_time_pts = clock_get_monotonic_seconds();
 
-    double start_time = clock_get_monotonic_seconds();
+    double start_time = clock_get_monotonic_seconds(); // todo - target_fps to make first frame start immediately?
     double frame_timer_start = start_time;
     int fps_counter = 0;
 
diff --git a/src/sound.cpp b/src/sound.cpp
index 762d962..f29e474 100644
--- a/src/sound.cpp
+++ b/src/sound.cpp
@@ -1,6 +1,6 @@
 #include "../include/sound.hpp"
 extern "C" {
-#include "../include/time.h"
+#include "../include/utils.h"
 }
 
 #include <stdlib.h>
diff --git a/src/time.c b/src/time.c
deleted file mode 100644
index b2dcdea..0000000
--- a/src/time.c
+++ /dev/null
@@ -1,10 +0,0 @@
-#include "../include/time.h"
-#include <time.h>
-
-double clock_get_monotonic_seconds() {
-    struct timespec ts;
-    ts.tv_sec = 0;
-    ts.tv_nsec = 0;
-    clock_gettime(CLOCK_MONOTONIC, &ts);
-    return (double)ts.tv_sec + (double)ts.tv_nsec * 0.000000001;
-}
diff --git a/src/utils.c b/src/utils.c
new file mode 100644
index 0000000..bf0a0c1
--- /dev/null
+++ b/src/utils.c
@@ -0,0 +1,62 @@
+#include "../include/utils.h"
+#include <time.h>
+#include <string.h>
+
+double clock_get_monotonic_seconds(void) {
+    struct timespec ts;
+    ts.tv_sec = 0;
+    ts.tv_nsec = 0;
+    clock_gettime(CLOCK_MONOTONIC, &ts);
+    return (double)ts.tv_sec + (double)ts.tv_nsec * 0.000000001;
+}
+
+static const XRRModeInfo* get_mode_info(const XRRScreenResources *sr, RRMode id) {
+    for(int i = 0; i < sr->nmode; ++i) {
+        if(sr->modes[i].id == id)
+            return &sr->modes[i];
+    }    
+    return NULL;
+}
+
+void for_each_active_monitor_output(Display *display, active_monitor_callback callback, void *userdata) {
+    XRRScreenResources *screen_res = XRRGetScreenResources(display, DefaultRootWindow(display));
+    if(!screen_res)
+        return;
+
+    for(int i = 0; i < screen_res->noutput; ++i) {
+        XRROutputInfo *out_info = XRRGetOutputInfo(display, screen_res, screen_res->outputs[i]);
+        if(out_info && out_info->crtc && out_info->connection == RR_Connected) {
+            XRRCrtcInfo *crt_info = XRRGetCrtcInfo(display, screen_res, out_info->crtc);
+            if(crt_info && crt_info->mode) {
+                const XRRModeInfo *mode_info = get_mode_info(screen_res, crt_info->mode);
+                if(mode_info)
+                    callback(out_info, crt_info, mode_info, userdata);
+            }
+            if(crt_info)
+                XRRFreeCrtcInfo(crt_info);
+        }
+        if(out_info)
+            XRRFreeOutputInfo(out_info);
+    }    
+
+    XRRFreeScreenResources(screen_res);
+}
+
+static void get_monitor_by_name_callback(const XRROutputInfo *output_info, const XRRCrtcInfo *crt_info, const XRRModeInfo *mode_info, void *userdata) {
+    get_monitor_by_name_userdata *data = (get_monitor_by_name_userdata*)userdata;
+    if(!data->found_monitor && data->name_len == output_info->nameLen && memcmp(data->name, output_info->name, data->name_len) == 0) {
+        data->monitor->pos = (vec2i){ .x = crt_info->x, .y = crt_info->y };
+        data->monitor->size = (vec2i){ .x = (int)crt_info->width, .y = (int)crt_info->height };
+        data->found_monitor = true;
+    }
+}
+
+bool get_monitor_by_name(Display *display, const char *name, gsr_monitor *monitor) {
+    get_monitor_by_name_userdata userdata;
+    userdata.name = name;
+    userdata.name_len = strlen(name);
+    userdata.monitor = monitor;
+    userdata.found_monitor = false;
+    for_each_active_monitor_output(display, get_monitor_by_name_callback, &userdata);
+    return userdata.found_monitor;
+}
diff --git a/src/xnvctrl.c b/src/xnvctrl.c
index 94d48d2..b738455 100644
--- a/src/xnvctrl.c
+++ b/src/xnvctrl.c
@@ -1,6 +1,8 @@
 #include "../include/xnvctrl.h"
 #include "../include/library_loader.h"
 #include <string.h>
+#include <stdio.h>
+#include <dlfcn.h>
 
 bool gsr_xnvctrl_load(gsr_xnvctrl *self, Display *display) {
     memset(self, 0, sizeof(gsr_xnvctrl));