rustdesk/src/server/video_service.rs
2021-05-02 00:37:32 +08:00

389 lines
12 KiB
Rust

// 24FPS (actually 23.976FPS) is what video professionals ages ago determined to be the
// slowest playback rate that still looks smooth enough to feel real.
// Our eyes can see a slight difference and even though 30FPS actually shows
// more information and is more realistic.
// 60FPS is commonly used in game, teamviewer 12 support this for video editing user.
// how to capture with mouse cursor:
// https://docs.microsoft.com/zh-cn/windows/win32/direct3ddxgi/desktop-dup-api?redirectedfrom=MSDN
// 实现了硬件编解码和音频抓取,还绘制了鼠标
// https://github.com/PHZ76/DesktopSharing
// dxgi memory leak issue
// https://stackoverflow.com/questions/47801238/memory-leak-in-creating-direct2d-device
// but per my test, it is more related to AcquireNextFrame,
// https://forums.developer.nvidia.com/t/dxgi-outputduplication-memory-leak-when-using-nv-but-not-amd-drivers/108582
// to-do:
// https://slhck.info/video/2017/03/01/rate-control.html
use super::*;
use scrap::{Capturer, Config, Display, EncodeFrame, Encoder, VideoCodecId, STRIDE_ALIGN};
use std::{
io::ErrorKind::WouldBlock,
time::{self, Instant},
};
const WAIT_BASE: i32 = 17;
pub const NAME: &'static str = "video";
lazy_static::lazy_static! {
static ref CURRENT_DISPLAY: Arc<Mutex<usize>> = Arc::new(Mutex::new(usize::MAX));
static ref LAST_ACTIVE: Arc<Mutex<Instant>> = Arc::new(Mutex::new(Instant::now()));
static ref SWITCH: Arc<Mutex<bool>> = Default::default();
static ref INTERNAL_LATENCIES: Arc<Mutex<HashMap<i32, i64>>> = Default::default();
static ref TEST_LATENCIES: Arc<Mutex<HashMap<i32, i64>>> = Default::default();
static ref IMAGE_QUALITIES: Arc<Mutex<HashMap<i32, i32>>> = Default::default();
}
pub fn new() -> GenericService {
let sp = GenericService::new(NAME, true);
sp.run(run);
sp
}
fn run(sp: GenericService) -> ResultType<()> {
let fps = 30;
let spf = time::Duration::from_secs_f32(1. / (fps as f32));
let (n, current, display) = get_current_display()?;
let (origin, width, height) = (display.origin(), display.width(), display.height());
log::debug!(
"#displays={}, current={}, origin: {:?}, width={}, height={}",
n,
current,
&origin,
width,
height
);
// Capturer object is expensive, avoiding to create it frequently.
let mut c = Capturer::new(display, true).with_context(|| "Failed to create capturer")?;
let q = get_image_quality();
let (bitrate, rc_min_quantizer, rc_max_quantizer, speed) = get_quality(width, height, q);
log::info!("bitrate={}, rc_min_quantizer={}", bitrate, rc_min_quantizer);
let mut wait = WAIT_BASE;
let cfg = Config {
width: width as _,
height: height as _,
timebase: [1, 1000], // Output timestamp precision
bitrate,
codec: VideoCodecId::VP9,
rc_min_quantizer,
rc_max_quantizer,
speed,
};
let mut vpx;
match Encoder::new(&cfg, 1) {
Ok(x) => vpx = x,
Err(err) => bail!("Failed to create encoder: {}", err),
}
if *SWITCH.lock().unwrap() {
log::debug!("Broadcasting display switch");
let mut misc = Misc::new();
misc.set_switch_display(SwitchDisplay {
display: current as _,
x: origin.0 as _,
y: origin.1 as _,
width: width as _,
height: height as _,
..Default::default()
});
let mut msg_out = Message::new();
msg_out.set_misc(misc);
*SWITCH.lock().unwrap() = false;
sp.send(msg_out);
}
#[cfg(windows)]
if !c.is_gdi() {
// dxgi duplicateoutput has no output if display no change, so we use gdi this as workaround
if c.set_gdi() {
// dgx capture release has memory leak somehow, so use gdi always before fixing it, just sacrificing some cpu
/*
if let Ok(frame) = c.frame(wait as _) {
handle_one_frame(&sp, &frame, 0, &mut vpx)?;
}
c.cancel_gdi();
*/
}
}
let start = time::Instant::now();
let mut crc = (0, 0);
let mut last_sent = time::Instant::now();
while sp.ok() {
if *SWITCH.lock().unwrap() {
bail!("SWITCH");
}
if current != *CURRENT_DISPLAY.lock().unwrap() {
*SWITCH.lock().unwrap() = true;
bail!("SWITCH");
}
if get_image_quality() != q {
bail!("SWITCH");
}
#[cfg(windows)]
{
if crate::platform::windows::desktop_changed() {
bail!("Desktop changed");
}
}
let now = time::Instant::now();
*LAST_ACTIVE.lock().unwrap() = now;
if get_latency() < 1000 || last_sent.elapsed().as_millis() > 1000 {
match c.frame(wait as _) {
Ok(frame) => {
let time = now - start;
let ms = (time.as_secs() * 1000 + time.subsec_millis() as u64) as i64;
handle_one_frame(&sp, &frame, ms, &mut crc, &mut vpx)?;
last_sent = now;
}
Err(ref e) if e.kind() == WouldBlock => {
// https://github.com/NVIDIA/video-sdk-samples/tree/master/nvEncDXGIOutputDuplicationSample
wait = WAIT_BASE - now.elapsed().as_millis() as i32;
if wait < 0 {
wait = 0
}
continue;
}
Err(err) => {
return Err(err.into());
}
}
}
let elapsed = now.elapsed();
// may need to enable frame(timeout)
log::trace!("{:?} {:?}", time::Instant::now(), elapsed);
if elapsed < spf {
std::thread::sleep(spf - elapsed);
}
}
Ok(())
}
#[inline]
fn create_msg(vp9s: Vec<VP9>) -> Message {
let mut msg_out = Message::new();
let mut vf = VideoFrame::new();
vf.set_vp9s(VP9s {
frames: vp9s.into(),
..Default::default()
});
msg_out.set_video_frame(vf);
msg_out
}
#[inline]
fn create_frame(frame: &EncodeFrame) -> VP9 {
VP9 {
data: frame.data.to_vec(),
key: frame.key,
pts: frame.pts,
..Default::default()
}
}
#[inline]
fn handle_one_frame(
sp: &GenericService,
frame: &[u8],
ms: i64,
crc: &mut (u32, u32),
vpx: &mut Encoder,
) -> ResultType<()> {
sp.snapshot(|sps| {
// so that new sub and old sub share the same encoder after switch
if sps.has_subscribes() {
bail!("SWITCH");
}
Ok(())
})?;
let mut hasher = crc32fast::Hasher::new();
hasher.update(frame);
let checksum = hasher.finalize();
if checksum != crc.0 {
crc.0 = checksum;
crc.1 = 0;
} else {
crc.1 += 1;
}
if crc.1 <= 180 && crc.1 % 5 == 0 {
let mut frames = Vec::new();
for ref frame in vpx
.encode(ms, frame, STRIDE_ALIGN)
.with_context(|| "Failed to encode")?
{
frames.push(create_frame(frame));
}
for ref frame in vpx.flush().with_context(|| "Failed to flush")? {
frames.push(create_frame(frame));
}
// to-do: flush periodically, e.g. 1 second
if frames.len() > 0 {
sp.send(create_msg(frames));
}
}
Ok(())
}
pub fn get_displays() -> ResultType<(usize, Vec<DisplayInfo>)> {
// switch to primary display if long time (30 seconds) no users
if LAST_ACTIVE.lock().unwrap().elapsed().as_secs() >= 30 {
*CURRENT_DISPLAY.lock().unwrap() = usize::MAX;
}
let mut displays = Vec::new();
let mut primary = 0;
for (i, d) in Display::all()?.iter().enumerate() {
if d.is_primary() {
primary = i;
}
displays.push(DisplayInfo {
x: d.origin().0 as _,
y: d.origin().1 as _,
width: d.width() as _,
height: d.height() as _,
name: d.name(),
online: d.is_online(),
..Default::default()
});
}
let mut lock = CURRENT_DISPLAY.lock().unwrap();
if *lock >= displays.len() {
*lock = primary
}
Ok((*lock, displays))
}
pub fn switch_display(i: i32) {
let i = i as usize;
if let Ok((_, displays)) = get_displays() {
if i < displays.len() {
*CURRENT_DISPLAY.lock().unwrap() = i;
}
}
}
pub fn refresh() {
*SWITCH.lock().unwrap() = true;
}
fn get_primary() -> usize {
if let Ok(all) = Display::all() {
for (i, d) in all.iter().enumerate() {
if d.is_primary() {
return i;
}
}
}
0
}
pub fn switch_to_primary() {
switch_display(get_primary() as _);
}
fn get_current_display() -> ResultType<(usize, usize, Display)> {
let mut current = *CURRENT_DISPLAY.lock().unwrap() as usize;
let mut displays = Display::all()?;
if displays.len() == 0 {
bail!("No displays");
}
let n = displays.len();
if current >= n {
current = 0;
for (i, d) in displays.iter().enumerate() {
if d.is_primary() {
current = i;
break;
}
}
*CURRENT_DISPLAY.lock().unwrap() = current;
}
return Ok((n, current, displays.remove(current)));
}
#[inline]
fn update_latency(id: i32, latency: i64, latencies: &mut HashMap<i32, i64>) {
if latency <= 0 {
latencies.remove(&id);
} else {
latencies.insert(id, latency);
}
}
pub fn update_test_latency(id: i32, latency: i64) {
update_latency(id, latency, &mut *TEST_LATENCIES.lock().unwrap());
}
pub fn update_internal_latency(id: i32, latency: i64) {
update_latency(id, latency, &mut *INTERNAL_LATENCIES.lock().unwrap());
}
pub fn get_latency() -> i64 {
INTERNAL_LATENCIES
.lock()
.unwrap()
.values()
.max()
.unwrap_or(&0)
.clone()
}
fn convert_quality(q: i32) -> i32 {
let q = {
if q == ImageQuality::Balanced.value() {
(100 * 2 / 3, 12)
} else if q == ImageQuality::Low.value() {
(100 / 2, 18)
} else if q == ImageQuality::Best.value() {
(100, 12)
} else {
let bitrate = q >> 8 & 0xFF;
let quantizer = q & 0xFF;
(bitrate * 2, (100 - quantizer) * 36 / 100)
}
};
if q.0 <= 0 {
0
} else {
q.0 << 8 | q.1
}
}
pub fn update_image_quality(id: i32, q: Option<i32>) {
match q {
Some(q) => {
let q = convert_quality(q);
if q > 0 {
IMAGE_QUALITIES.lock().unwrap().insert(id, q);
} else {
IMAGE_QUALITIES.lock().unwrap().remove(&id);
}
}
None => {
IMAGE_QUALITIES.lock().unwrap().remove(&id);
}
}
}
fn get_image_quality() -> i32 {
IMAGE_QUALITIES
.lock()
.unwrap()
.values()
.min()
.unwrap_or(&convert_quality(ImageQuality::Balanced.value()))
.clone()
}
#[inline]
fn get_quality(w: usize, h: usize, q: i32) -> (u32, u32, u32, i32) {
// https://www.nvidia.com/en-us/geforce/guides/broadcasting-guide/
let bitrate = q >> 8 & 0xFF;
let quantizer = q & 0xFF;
let b = ((w * h) / 1000) as u32;
(bitrate as u32 * b / 100, quantizer as _, 56, 7)
}