VAAPI: opt-in UU_VAAPI_BUFFER_MS for the VBV buffer (#47)

The 100-500 kbit default keeps NVENC keyframes inside the buffer, but VAAPI quantizes a
whole desktop keyframe when the buffer is that small: on a 2560x1440 H.264 VAAPI stream
capped at 30 Mbit/s the first IDR measured 59,910 bytes and the picture blurred once per
GOP before sharpening back.

UU_VAAPI_BUFFER_MS (10-1000 ms, VAAPI only) sizes the buffer as bitrate x ms/1000. With
250 ms on the same input the first IDR measured 142,397 bytes and the worst-frame Y-SSIM
over 180 static frames went from 0.9492 to 0.9933, with no rise in average bitrate.

Opt-in: unset, out of range or unparsable keeps the old buffer.

Assisted-by: DeepSeek V4.1 Flash-1017
This commit is contained in:
NuoFang authored and GitHub committed 2026-10-18 15:31:00 +08:00
1 parent 696cf10e33
commit cb8253febf
3 files changed
+44 -1

No files matched your search

+8
View File
@@ -280,6 +280,7 @@ Common environment variables:
| `UU_INPUT_BACKEND` | `auto` (default: uinput on Wayland when accessible) / `uinput` / `native` |
| `UU_TERMINAL_SHORTCUTS=0` | Disable the Ctrl+V paste rewrite in terminal windows |
| `UU_VIDEO_ENCODER` | `cpu`: libx264; `vaapi`: opt-in Intel/AMD; `nvenc`: force NVENC; unset: NVENC only when an NVIDIA driver is loaded, else CPU |
| `UU_VAAPI_BUFFER_MS` | VAAPI VBV buffer in ms (10-1000, VAAPI only); raise it when keyframes are squeezed into a blur at high resolution |
| `UU_FFMPEG_ROOT` | Use FFmpeg from this prefix |
| `UU_AUDIO=0` | Disable sound; `UU_AUDIO_SOURCE` picks the PulseAudio source |
| `UU_REMOTE_ASSIST=0` | Disable remote assistance (no guest assistance before sign-in either) |
@@ -366,6 +367,13 @@ with a warning. Accepted on Intel UHD 770 (iHD) with the default FFmpeg process
viewer (presets, fps, scaling, reconnect; no fallback, ~40% less CPU than libx264). AMD and the `inproc-av`
VAAPI path are not hardware-verified.
The default VAAPI VBV buffer (100-500 kbit) squeezes keyframes hard at high resolution: on a 2560x1440
H.264 VAAPI stream capped at 30 Mbit/s the first IDR measured 59,910 bytes and the picture blurred at
every GOP before sharpening back. `UU_VAAPI_BUFFER_MS=250` (buffer = bitrate cap x ms/1000) restored that
first IDR to 142,397 bytes and raised the worst-frame Y-SSIM from 0.9492 to 0.9933 without raising the
average bitrate. Off by default; unset or out of the 10-1000 ms range keeps the old buffer. Larger
keyframes can queue briefly on a weak link, so start at 250 ms.
### Virtual audio / 虚拟声卡
The viewer's 静音运行 switch creates the null sink `uu-remote-sink`,
+7
View File
@@ -250,6 +250,7 @@ systemctl --user restart uuremote-native.service
| `UU_INPUT_BACKEND` | `auto`(默认,Wayland 下可用时用 uinput)/ `uinput` / `native` |
| `UU_TERMINAL_SHORTCUTS=0` | 关闭终端窗口里的 Ctrl+V 粘贴转换 |
| `UU_VIDEO_ENCODER` | `cpu` 强制 CPU;`vaapi` 显式启用 Intel/AMD 硬编;`nvenc` 强制 NVENC;未设置时检测到 NVIDIA 驱动才用 NVENC,否则用 CPU |
| `UU_VAAPI_BUFFER_MS` | VAAPI 的 VBV 缓冲(10–1000 ms,仅 `UU_VIDEO_ENCODER=vaapi` 时生效);高分辨率下关键帧被压糊时用它换稳定画质 |
| `UU_FFMPEG_ROOT` | 使用指定目录下的 FFmpeg |
| `UU_AUDIO=0` | 关闭声音转发;`UU_AUDIO_SOURCE` 指定 PulseAudio 采集源 |
| `UU_LOGIN_WINDOW=0` | 凭证缺失时不自动弹出登录窗口 |
@@ -323,6 +324,12 @@ flowchart LR
VBR 使用控制端码率预设及 CQ 对应的最小 QP,无 B 帧、async depth=1,并保留现有 GOP/VFR 规则。
Intel 核显(UHD 770,iHD 驱动)已实机验收:默认的外部 FFmpeg 管线用官方 Mac 控制端测试了各档画质、帧率、缩放和重连,全程硬件编码、无回退,CPU 占用比 libx264 低约 40%。**AMD 显卡和 `--features inproc-av` 管线的 VAAPI 尚未实机验证。**
高分辨率下 VAAPI 的默认 VBV 缓冲(100–500 kbit)会把关键帧压得过狠:2560×1440 H.264 VAAPI、30M
上限、静止画面上实测首个 IDR 只有 59,910 字节,画面每隔一个 GOP 糊一下再慢慢恢复。设
`UU_VAAPI_BUFFER_MS=250`(按 `码率上限 × 毫秒 / 1000` 放大 VBV)后首个 IDR 为 142,397 字节,
最低帧 Y-SSIM 从 0.9492 升到 0.9933,平均码率没有上升。默认不启用,未设置或超出 10–1000 ms
仍用原缓冲;代价是关键帧变大,弱网下可能出现瞬时排队,建议从 250 ms 试起。
### 静音运行与虚拟声卡
控制端开启「静音运行」后,程序创建虚拟输出 `uu-remote-sink`,保存原默认输出,
+29 -1
View File
@@ -480,7 +480,21 @@ async fn read_inprocess(
/// whole IDR frame, otherwise large desktops get near-grey keyframes that sharpen over the GOP (#8).
/// Half a second of bitrate still starved repeated 0.5 s-GOP keyframes at 3296x2036/2M; one second did not.
fn buffer_kbits(max_kbps: u32, encoder: Encoder) -> u32 {
if encoder == Encoder::Cpu { max_kbps.clamp(500, 8000) } else { (max_kbps / 8).clamp(100, 500) }
if encoder == Encoder::Cpu { max_kbps.clamp(500, 8000) }
else if encoder == Encoder::Vaapi {
let requested = std::env::var("UU_VAAPI_BUFFER_MS").ok();
vaapi_buffer_kbits(max_kbps, requested.as_deref()).unwrap_or((max_kbps / 8).clamp(100, 500))
}
else { (max_kbps / 8).clamp(100, 500) }
}
/// Opt-in VAAPI VBV buffer: `码率上限 × 毫秒 / 1000` for `UU_VAAPI_BUFFER_MS` in 10-1000 ms.
/// The 100-500 kbit default is sized for NVENC; VAAPI quantizes a whole desktop keyframe when
/// the buffer is that small, which reads as a blur that sharpens over each GOP (#42).
/// Unset, out of range or unparsable values keep the default buffer.
fn vaapi_buffer_kbits(max_kbps: u32, value: Option<&str>) -> Option<u32> {
let ms = value?.trim().parse::<u32>().ok().filter(|ms| (10..=1000).contains(ms))?;
Some((max_kbps.saturating_mul(ms) / 1000).max(1))
}
fn buffer_for(settings: &VideoSettings, encoder: Encoder) -> u32 {
@@ -1066,6 +1080,20 @@ mod tests {
assert_eq!(buffer_kbits(30_000, Encoder::Cpu), 8000);
}
#[test]
fn vaapi_buffer_env_is_opt_in_and_bounded() {
assert_eq!(vaapi_buffer_kbits(30_000, None), None);
assert_eq!(vaapi_buffer_kbits(30_000, Some("250")), Some(7500));
assert_eq!(vaapi_buffer_kbits(30_000, Some(" 10 ")), Some(300));
assert_eq!(vaapi_buffer_kbits(30_000, Some("1000")), Some(30_000));
// Out of the documented range, empty or unparsable keeps the default buffer.
for value in ["", "9", "1001", "0", "abc", "-250", "250ms"] {
assert_eq!(vaapi_buffer_kbits(30_000, Some(value)), None, "{value}");
}
// A tiny cap must not produce a zero-sized buffer.
assert_eq!(vaapi_buffer_kbits(1, Some("10")), Some(1));
}
#[test]
fn vaapi_is_explicit_and_upload_follows_scale() {
assert_eq!(Encoder::from_setting(None, true), Encoder::Nvenc);