feat(windows): AMD (AMF) + Intel (QSV) hardware encode on the Windows host

The Windows host was NVIDIA-only (NVENC) with an openh264 software fallback. Add
AMD AMF and Intel QSV via libavcodec — the Windows analogue of the Linux VAAPI
backend — so one installer serves all three GPU vendors.

- encode/ffmpeg_win.rs: new WinVendor{Amf,Qsv} encoder. System-memory NV12/P010
  readback (default, robust) + opt-in zero-copy D3D11 (PUNKTFUNK_ZEROCOPY: shares
  the capturer's ID3D11Device; AMF takes AV_PIX_FMT_D3D11, QSV derives a QSV frames
  ctx and maps) with a system fallback for the format-group mismatch the capturer's
  video-processor fallback can produce. HDR Main10 (P010 + BT.2020/PQ VUI; an
  Rgb10a2->P010 swscale covers the shader fallback).
- encode.rs: Codec::amf_name/qsv_name; open_video + windows_resolved_backend()
  resolve PUNKTFUNK_ENCODER=auto|nvenc|amf|qsv|sw via a DXGI adapter VendorId probe.
- capture/dxgi.rs: gpu_mode mirrors the resolved backend (D3D11 NV12/P010 for AMF/QSV).
- gamestream/serverinfo.rs: GPU-aware codec advertisement (windows_codec_support;
  AV1 gated to RDNA3+/Arc, like the VAAPI path).
- Cargo.toml: amf-qsv feature (optional ffmpeg-next in the windows target block).
- CI/installer: windows-host.yml sets FFMPEG_DIR + builds --features nvenc,amf-qsv;
  the Inno installer bundles the FFmpeg DLLs; host.env default nvenc -> auto.

CI-green target; AMF/QSV not yet on-glass validated (no AMD/Intel Windows box in the
lab) — NVENC stays live-validated. An adversarial-review pass caught + fixed real
FFI bugs (AV_PIX_FMT_P010 is a macro -> P010LE; windows-rs 0.62 GetImmediateContext/
GetDesc1 return Result; AV_HWFRAME_MAP_* is a bindgen enum with no BitOr).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.gitea/workflows/windows-host.yml

@@ -20,9 +20,12 @@
 # an ephemeral self-signed cert is generated and its public .cer published next to the installer
 # (import once to LocalMachine\TrustedPublisher). See packaging/windows/pack-host-installer.ps1.
 #
-# NVENC: the host builds with --features nvenc; the only link need is nvencodeapi.lib, synthesised
+# GPU backends: the host builds with --features nvenc,amf-qsv = all three vendors in one installer.
-# from a 2-export .def with llvm-dlltool (no GPU/SDK at build time). The resulting exe is NVIDIA-only
+#  - NVENC (NVIDIA, direct SDK): the only link need is nvencodeapi.lib, synthesised from a 2-export
-# by design — CI never launches it, so no GPU is needed here.
+#    .def with llvm-dlltool (no GPU/SDK at build time).
+#  - AMF/QSV (AMD/Intel, libavcodec): link-imports the FFmpeg libs from FFMPEG_DIR (the BtbN gpl-shared
+#    tree the client uses; includes the *_amf/*_qsv encoders) and bundles its DLLs into the installer.
+# CI never launches the exe, so no GPU is needed here — this is build + Windows clippy coverage only.
 name: windows-host
 
 on:
@@ -59,6 +62,13 @@ jobs:
           # (pwsh Out-File utf8 = no BOM, unlike Windows PowerShell 5.1 — keeps the first line clean).
           "CARGO_TARGET_DIR=C:\t" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
           "CARGO_WORKSPACE_DIR=$env:GITHUB_WORKSPACE" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
+          # FFMPEG_DIR: the same BtbN gpl-shared x64 tree the Windows CLIENT links against (provisioned
+          # by scripts/ci/setup-windows-runner.ps1). The host's AMD/Intel AMF/QSV encode backend
+          # (--features amf-qsv) link-imports avcodec/avutil/swscale from it; pack-host-installer.ps1
+          # then bundles its bin\*.dll into the installer. LIBCLANG_PATH is in the runner daemon env.
+          if (-not $env:FFMPEG_DIR) {
+            "FFMPEG_DIR=C:\Users\Public\ffmpeg" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
+          }
           $v = if ($env:GITHUB_REF -like 'refs/tags/v*') {
             $env:GITHUB_REF_NAME -replace '^v', ''
           } else {
@@ -74,14 +84,15 @@ jobs:
           & packaging/windows/nvenc/gen-nvenc-importlib.ps1 -OutDir C:\t\nvenc
           "PUNKTFUNK_NVENC_LIB_DIR=C:\t\nvenc" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
 
-      - name: Build (release, nvenc)
+      - name: Build (release, nvenc + amf-qsv)
         shell: pwsh
-        run: cargo build --release -p punktfunk-host --features nvenc
+        # All-vendor host: NVENC (NVIDIA, direct SDK) + AMF/QSV (AMD/Intel, libavcodec via FFMPEG_DIR).
+        run: cargo build --release -p punktfunk-host --features nvenc,amf-qsv
 
       - name: Clippy (host, Windows)
         shell: pwsh
         # First-ever Windows lint coverage for the host (Linux CI never lints the windows-cfg code).
-        run: cargo clippy -p punktfunk-host --features nvenc -- -D warnings
+        run: cargo clippy -p punktfunk-host --features nvenc,amf-qsv -- -D warnings
 
       - name: Ensure Inno Setup
         shell: pwsh

CLAUDE.md

@@ -74,17 +74,26 @@ Low-latency desktop/game streaming stack, Linux-first, with a shared Rust protoc
   Clients auto-resolve the type from the physical controller (DS5→DualSense, DS4→DualShock 4,
   Xbox One→Xbox One). Windows-host DualShock 4 (ViGEm) is not yet wired — Windows clients asking for
   DS4 get Xbox 360 for now.
-- **Windows host: implemented and shipping (NVIDIA-only, x64-only).** `#[cfg(windows)]` backends
+- **Windows host: implemented and shipping (all-vendor, x64-only).** `#[cfg(windows)]` backends
   behind the same traits as Linux — DXGI Desktop Duplication capture (`capture/dxgi.rs`), **SudoVDA**
-  virtual display per session (`vdisplay/sudovda.rs`), NVENC encode (`--features nvenc`), SendInput +
+  virtual display per session (`vdisplay/sudovda.rs`), GPU encode (NVENC `--features nvenc`; AMD/Intel
-  **ViGEm** gamepads (`inject/gamepad_windows.rs`), WASAPI loopback + virtual mic (`audio/wasapi_*`).
+  `--features amf-qsv`), SendInput + **ViGEm** gamepads (`inject/gamepad_windows.rs`), WASAPI loopback
-  Ships as a **signed Inno Setup installer** that registers a `LocalSystem` SCM service launching into
+  + virtual mic (`audio/wasapi_*`). Ships as a **signed Inno Setup installer** that registers a
-  the interactive session for secure-desktop (UAC/lock-screen) capture (`service.rs`), bundles the
+  `LocalSystem` SCM service launching into the interactive session for secure-desktop (UAC/lock-screen)
-  SudoVDA driver, and is published by `windows-host.yml`. **HDR (10-bit)**: WGC captures the HDR
+  capture (`service.rs`), bundles the SudoVDA driver + the FFmpeg DLLs, and is published by
-  desktop as FP16/Rgb10a2 (DDA FP16 for the secure desktop), NVENC forces HEVC Main10 + BT.2020 PQ,
+  `windows-host.yml`. **Encoder is GPU-aware** (`encode.rs` `open_video` + `windows_resolved_backend`):
-  the client auto-detects PQ from the HEVC VUI — gated by `PUNKTFUNK_10BIT` + client `VIDEO_CAP_10BIT`;
+  `PUNKTFUNK_ENCODER=auto` (the host.env default) detects the DXGI adapter vendor → **NVENC** (NVIDIA,
-  **Windows host only** (the Linux host stays 8-bit, blocked upstream). Newer/less battle-tested than
+  direct SDK, `encode/nvenc.rs`), **AMF** (AMD) / **QSV** (Intel) via libavcodec
-  the Linux host; no AMD/Intel/software encode path. Packaging: `packaging/windows/`.
+  (`encode/ffmpeg_win.rs`, the Windows analogue of the Linux VAAPI backend — `WinVendor{Amf,Qsv}`,
+  system-memory NV12/P010 readback default + opt-in zero-copy D3D11 behind `PUNKTFUNK_ZEROCOPY` with a
+  system fallback), or software H.264 (`encode/sw.rs`, GPU-less). GameStream codec advertisement is
+  probed per-GPU on AMF/QSV (`windows_codec_support` → `serverinfo`, AV1 gated). **HDR (10-bit)**: WGC
+  captures the HDR desktop as FP16/Rgb10a2 (DDA FP16 for the secure desktop), the encoder forces HEVC
+  Main10 + BT.2020 PQ (NVENC ABGR10/P010; AMF/QSV P010 + a swscale Rgb10a2→P010 fallback), the client
+  auto-detects PQ from the HEVC VUI — gated by `PUNKTFUNK_10BIT` + client `VIDEO_CAP_10BIT`; **Windows
+  host only** (the Linux host stays 8-bit, blocked upstream). **AMF/QSV is CI-green but not yet
+  on-glass validated** (no AMD/Intel Windows box in the lab); NVENC is live-validated. Newer/less
+  battle-tested than the Linux host. Packaging: `packaging/windows/`.
 
 ## What's left
 
@@ -243,6 +252,7 @@ crates/punktfunk-host/
   vdisplay/{kwin,gamescope,mutter,wlroots}.rs   per-compositor client-sized virtual outputs
   zerocopy/{egl,cuda,vulkan}.rs         dmabuf → CUDA → NVENC (tiled via EGL/GL, LINEAR via Vulkan)
   inject/{libei,wlr,gamepad,dualsense}.rs   input backends (uinput xpad + UHID DualSense)
+  encode/{nvenc,linux,vaapi,ffmpeg_win,sw}.rs   per-GPU encoders (NVENC · Linux NVENC/CUDA · VAAPI · AMF/QSV · openh264)
   capture.rs · encode.rs · audio.rs · spike.rs · punktfunk1.rs · mgmt.rs · native_pairing.rs
 clients/probe/    punktfunk/1 reference/probe client (headless test/measurement tool)
 clients/linux/    native Linux client (GTK4/libadwaita · FFmpeg · PipeWire · SDL3)
@@ -250,7 +260,7 @@ clients/windows/  native Windows client (WinUI 3 via windows-reactor · D3D11 ·
 clients/apple/    native macOS/iOS/tvOS client (Swift · VideoToolbox · GameController)
 clients/android/  native Android client (Kotlin app + native/ Rust JNI core over punktfunk-core)
 clients/decky/    Steam Deck Decky plugin
-crates/punktfunk-host/src/{capture/dxgi,vdisplay/sudovda,inject/gamepad_windows,audio/wasapi_*,service}.rs   Windows host backends
+crates/punktfunk-host/src/{capture/dxgi,vdisplay/sudovda,encode/ffmpeg_win,inject/gamepad_windows,audio/wasapi_*,service}.rs   Windows host backends
 web/                          TanStack web console over the mgmt API (status · devices · pairing)
 packaging/                    apt(deb) · RPM/COPR · Arch/sysext · Flatpak · Bazzite bootc · Windows host installer (per-dir READMEs)
 tools/{loss-harness,latency-probe}/     measurement (plan §10)

crates/punktfunk-host/Cargo.toml

@@ -184,6 +184,12 @@ vigem-client = { version = "0.1", features = ["unstable_xtarget_notification"] }
 # the crate builds on docs.rs/CI. We enable it so the GPU-less VM/CI compiles; the DirectX NVENC path
 # never calls CUDA at runtime, so the pinned CUDA bindings version is irrelevant.
 nvidia-video-codec-sdk = { version = "0.4", features = ["ci-check"], optional = true }
+# AMD (AMF) + Intel (QSV) hardware encode on Windows via libavcodec — the analogue of the Linux
+# VAAPI backend (`src/encode/ffmpeg_win.rs`). Optional + behind the `amf-qsv` feature because it
+# link-imports the FFmpeg libs at build time (needs a `FFMPEG_DIR` with the AMF/QSV encoders — the
+# same BtbN gpl-shared tree the Windows client uses) and pulls the shared `avcodec/avutil/...` DLLs
+# at runtime. `ffmpeg-sys-next` auto-detects the FFmpeg version (7.x/avcodec-61 or 8.x/62).
+ffmpeg-next = { version = "8", optional = true }
 
 [features]
 # NVENC hardware encode (Windows). OFF by default: it pulls the NVENC SDK, and the host then needs
@@ -191,3 +197,7 @@ nvidia-video-codec-sdk = { version = "0.4", features = ["ci-check"], optional =
 # time — i.e. `nvencodeapi.lib` from the NVIDIA Video Codec SDK (or an import lib generated from
 # nvEncodeAPI64.dll) on the linker path. Build the GPU host with `--features nvenc`.
 nvenc = ["dep:nvidia-video-codec-sdk"]
+# AMD/Intel hardware encode on Windows (AMF/QSV via ffmpeg-next). OFF by default: it needs a
+# `FFMPEG_DIR` (BtbN gpl-shared, includes `*_amf`/`*_qsv`) at build time and bundles the FFmpeg
+# DLLs at runtime. Build the all-vendor GPU host with `--features nvenc,amf-qsv`.
+amf-qsv = ["dep:ffmpeg-next"]

crates/punktfunk-host/src/capture/dxgi.rs

@@ -2157,9 +2157,14 @@ impl DuplCapturer {
                 .ok()
                 .and_then(|s| s.parse().ok())
                 .unwrap_or((2000 / refresh_hz.max(1)).max(100));
-            let gpu_mode = std::env::var("PUNKTFUNK_ENCODER")
+            // Produce GPU-resident D3D11 frames (zero-copy NVENC, or the NV12/P010 the AMF/QSV
-                .map(|v| matches!(v.to_ascii_lowercase().as_str(), "nvenc" | "hw" | "nvidia"))
+            // backends read back / import) whenever the resolved encode backend is a GPU one — so the
-                .unwrap_or(false);
+            // capturer's output format matches the encoder's input. Only the software (GPU-less) path
+            // takes CPU staging. Mirrors `encode::open_video`'s dispatch exactly.
+            let gpu_mode = !matches!(
+                crate::encode::windows_resolved_backend(),
+                crate::encode::WindowsBackend::Software
+            );
             // Read the source display's HDR mastering metadata while we still hold `output` (it is
             // moved into the struct below). Only meaningful for an HDR (FP16) duplication.
             let is_hdr_init = dd.ModeDesc.Format == DXGI_FORMAT_R16G16B16A16_FLOAT;

crates/punktfunk-host/src/encode.rs

@@ -49,6 +49,26 @@ impl Codec {
             Codec::Av1 => "av1_vaapi",
         }
     }
+
+    /// The FFmpeg AMD **AMF** encoder name (the Windows AMD backend). Selected by name (the codec id
+    /// would pick the software encoder). AV1 (`av1_amf`) is RDNA3+/RX 7000+ — probe, never assume.
+    pub fn amf_name(self) -> &'static str {
+        match self {
+            Codec::H264 => "h264_amf",
+            Codec::H265 => "hevc_amf",
+            Codec::Av1 => "av1_amf",
+        }
+    }
+
+    /// The FFmpeg Intel **QSV** encoder name (the Windows Intel backend). Selected by name. AV1
+    /// (`av1_qsv`) is Arc/Xe2+; HEVC Main10 is Gen9.5+ — probe, never assume.
+    pub fn qsv_name(self) -> &'static str {
+        match self {
+            Codec::H264 => "h264_qsv",
+            Codec::H265 => "hevc_qsv",
+            Codec::Av1 => "av1_qsv",
+        }
+    }
 }
 
 /// A hardware encoder. One per session; runs on the encode thread.
@@ -198,16 +218,15 @@ pub fn open_video(
     #[cfg(target_os = "windows")]
     {
         let _ = cuda; // always false on Windows (no Cuda payload)
-        let _ = bit_depth; // used by the NVENC path below; the software H.264 path is 8-bit only
+                      // NVIDIA → NVENC (direct SDK), AMD → AMF, Intel → QSV (both libavcodec), else → software
-        let pref = std::env::var("PUNKTFUNK_ENCODER")
+                      // H.264. `auto` (the default) resolves from the DXGI adapter vendor.
-            .unwrap_or_default()
+        match windows_resolved_backend() {
-            .to_ascii_lowercase();
+            WindowsBackend::Nvenc => {
-        if matches!(pref.as_str(), "nvenc" | "hw" | "nvidia") {
                 // Hardware path: NVENC over D3D11. The DXGI capturer switches to its zero-copy
                 // FramePayload::D3d11 output under the same env var so capture + encode share textures.
                 #[cfg(feature = "nvenc")]
                 {
-                let enc = nvenc::NvencD3d11Encoder::open(
+                    nvenc::NvencD3d11Encoder::open(
                         codec,
                         format,
                         width,
@@ -215,32 +234,67 @@ pub fn open_video(
                         fps,
                         bitrate_bps,
                         bit_depth,
-                )?;
+                    )
-                return Ok(Box::new(enc) as Box<dyn Encoder>);
+                    .map(|e| Box::new(e) as Box<dyn Encoder>)
                 }
                 #[cfg(not(feature = "nvenc"))]
                 {
                     anyhow::bail!(
-                    "NVENC requested but this host was built without it — rebuild with \
+                        "NVENC requested/detected but this host was built without it — rebuild \
-                     `--features nvenc` (needs the NVENC SDK's nvencodeapi.lib at link time)"
+                         with `--features nvenc` (needs the NVENC SDK's nvencodeapi.lib at link time)"
-                );
+                    )
                 }
             }
+            backend @ (WindowsBackend::Amf | WindowsBackend::Qsv) => {
+                // AMD AMF / Intel QSV via libavcodec (the Windows analogue of the Linux VAAPI path).
+                #[cfg(feature = "amf-qsv")]
+                {
+                    let vendor = if matches!(backend, WindowsBackend::Amf) {
+                        ffmpeg_win::WinVendor::Amf
+                    } else {
+                        ffmpeg_win::WinVendor::Qsv
+                    };
+                    ffmpeg_win::FfmpegWinEncoder::open(
+                        vendor,
+                        codec,
+                        format,
+                        width,
+                        height,
+                        fps,
+                        bitrate_bps,
+                        bit_depth,
+                    )
+                    .map(|e| Box::new(e) as Box<dyn Encoder>)
+                }
+                #[cfg(not(feature = "amf-qsv"))]
+                {
+                    let _ = backend;
+                    anyhow::bail!(
+                        "AMD/Intel (AMF/QSV) encode requested/detected but this host was built \
+                         without it — rebuild with `--features amf-qsv` (needs ffmpeg-next + a \
+                         FFMPEG_DIR with the AMF/QSV encoders at build time)"
+                    )
+                }
+            }
+            WindowsBackend::Software => {
                 anyhow::ensure!(
                     codec == Codec::H264,
                     "the Windows software encoder supports H.264 only; client negotiated {codec:?} \
-             (set PUNKTFUNK_ENCODER=nvenc for a GPU host, or request H264)"
+                     (build a GPU backend: --features nvenc or amf-qsv, or request H264)"
                 );
+                let _ = bit_depth; // the software H.264 path is 8-bit only
                                    // Software H.264 realistically caps far below the negotiated hardware rates.
                 const SW_BITRATE_CEIL: u64 = 100_000_000;
-        let enc = sw::OpenH264Encoder::open(
+                sw::OpenH264Encoder::open(
                     format,
                     width,
                     height,
                     fps,
                     bitrate_bps.min(SW_BITRATE_CEIL),
-        )?;
+                )
-        Ok(Box::new(enc) as Box<dyn Encoder>)
+                .map(|e| Box::new(e) as Box<dyn Encoder>)
+            }
+        }
     }
     #[cfg(not(any(target_os = "linux", target_os = "windows")))]
     {
@@ -339,7 +393,7 @@ pub fn linux_zero_copy_is_vaapi() -> bool {
 /// Which codecs the active GPU can actually ENCODE. Used to build the GameStream codec
 /// advertisement so a client never negotiates a codec the GPU can't do (AV1 encode is narrow —
 /// Intel Arc/Xe2+, AMD RDNA3+/RDNA4 — so it must be probed, not assumed).
-#[cfg(target_os = "linux")]
+#[cfg(any(target_os = "linux", target_os = "windows"))]
 #[derive(Clone, Copy, Debug)]
 pub struct CodecSupport {
     pub h264: bool,
@@ -370,6 +424,123 @@ pub fn vaapi_codec_support() -> CodecSupport {
     })
 }
 
+// ---------------------------------------------------------------------------------------------
+// Windows backend selection (the analogue of the Linux nvidia_present / linux_zero_copy_is_vaapi
+// logic). NVIDIA → NVENC, AMD → AMF, Intel → QSV; `auto` (default) reads the DXGI adapter vendor.
+// ---------------------------------------------------------------------------------------------
+
+#[cfg(target_os = "windows")]
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub(crate) enum WindowsBackend {
+    Nvenc,
+    Amf,
+    Qsv,
+    Software,
+}
+
+#[cfg(target_os = "windows")]
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+enum GpuVendor {
+    Nvidia,
+    Amd,
+    Intel,
+}
+
+/// Resolve the active Windows encode backend from `PUNKTFUNK_ENCODER` (`auto` → the DXGI adapter
+/// vendor). Shared by [`open_video`] and the GameStream codec advertisement so both agree.
+#[cfg(target_os = "windows")]
+pub(crate) fn windows_resolved_backend() -> WindowsBackend {
+    let pref = std::env::var("PUNKTFUNK_ENCODER")
+        .unwrap_or_default()
+        .to_ascii_lowercase();
+    match pref.as_str() {
+        "nvenc" | "hw" | "nvidia" | "cuda" => WindowsBackend::Nvenc,
+        "amf" | "amd" => WindowsBackend::Amf,
+        "qsv" | "intel" => WindowsBackend::Qsv,
+        "sw" | "software" | "openh264" => WindowsBackend::Software,
+        _ => match windows_gpu_vendor() {
+            Some(GpuVendor::Nvidia) => WindowsBackend::Nvenc,
+            Some(GpuVendor::Amd) => WindowsBackend::Amf,
+            Some(GpuVendor::Intel) => WindowsBackend::Qsv,
+            None => WindowsBackend::Software,
+        },
+    }
+}
+
+/// True if the active Windows backend is the libavcodec AMF/QSV path (so the codec advertisement
+/// consults a real GPU probe rather than the NVENC static superset). Always false when the
+/// `amf-qsv` feature is off — there's then no ffmpeg backend to probe.
+#[cfg(target_os = "windows")]
+pub fn windows_backend_is_ffmpeg() -> bool {
+    cfg!(feature = "amf-qsv")
+        && matches!(
+            windows_resolved_backend(),
+            WindowsBackend::Amf | WindowsBackend::Qsv
+        )
+}
+
+/// Detect the host GPU vendor from the first hardware DXGI adapter (Windows has no `/dev/nvidia*`
+/// probe). Cached. NVIDIA=0x10DE, AMD=0x1002, Intel=0x8086; the software/WARP adapter is skipped.
+#[cfg(target_os = "windows")]
+fn windows_gpu_vendor() -> Option<GpuVendor> {
+    use std::sync::OnceLock;
+    use windows::Win32::Graphics::Dxgi::{
+        CreateDXGIFactory1, IDXGIFactory1, DXGI_ADAPTER_FLAG_SOFTWARE,
+    };
+    static CACHE: OnceLock<Option<GpuVendor>> = OnceLock::new();
+    *CACHE.get_or_init(|| unsafe {
+        let factory: IDXGIFactory1 = CreateDXGIFactory1().ok()?;
+        let mut i = 0u32;
+        while let Ok(adapter) = factory.EnumAdapters1(i) {
+            i += 1;
+            // windows-rs 0.62: GetDesc1 returns the desc by value (no out-param).
+            let Ok(desc) = adapter.GetDesc1() else {
+                continue;
+            };
+            if (desc.Flags & DXGI_ADAPTER_FLAG_SOFTWARE.0 as u32) != 0 {
+                continue; // skip the Microsoft Basic Render / WARP adapter
+            }
+            match desc.VendorId {
+                0x10DE => return Some(GpuVendor::Nvidia),
+                0x1002 => return Some(GpuVendor::Amd),
+                0x8086 => return Some(GpuVendor::Intel),
+                _ => continue,
+            }
+        }
+        None
+    })
+}
+
+/// Probe the active Windows AMF/QSV backend for its encodable codecs (cached; opens a tiny encoder
+/// per codec, once). Mirrors [`vaapi_codec_support`]; called only when [`windows_backend_is_ffmpeg`]
+/// is true. AV1 is narrow (AMD RDNA3+, Intel Arc/Xe2+), so it must be probed, not assumed.
+#[cfg(all(target_os = "windows", feature = "amf-qsv"))]
+pub fn windows_codec_support() -> CodecSupport {
+    use std::sync::OnceLock;
+    static CACHE: OnceLock<CodecSupport> = OnceLock::new();
+    *CACHE.get_or_init(|| {
+        let vendor = match windows_resolved_backend() {
+            WindowsBackend::Qsv => ffmpeg_win::WinVendor::Qsv,
+            _ => ffmpeg_win::WinVendor::Amf,
+        };
+        let caps = CodecSupport {
+            h264: ffmpeg_win::probe_can_encode(vendor, Codec::H264),
+            h265: ffmpeg_win::probe_can_encode(vendor, Codec::H265),
+            av1: ffmpeg_win::probe_can_encode(vendor, Codec::Av1),
+        };
+        tracing::info!(
+            backend = ?vendor,
+            h264 = caps.h264,
+            h265 = caps.h265,
+            av1 = caps.av1,
+            "Windows AMF/QSV encode capabilities probed"
+        );
+        caps
+    })
+}
+
+#[cfg(all(target_os = "windows", feature = "amf-qsv"))]
+mod ffmpeg_win;
 #[cfg(target_os = "linux")]
 mod linux;
 #[cfg(all(target_os = "windows", feature = "nvenc"))]

crates/punktfunk-host/src/gamestream/serverinfo.rs

@@ -50,8 +50,28 @@ pub fn serverinfo_xml(host: &Host, https: bool, paired: bool) -> String {
 fn codec_mode_support() -> u32 {
     #[cfg(target_os = "linux")]
     if crate::encode::linux_zero_copy_is_vaapi() {
+        if let Some(m) = probed_mask(crate::encode::vaapi_codec_support()) {
+            return m;
+        }
+    }
+    // Windows AMD/Intel (AMF/QSV): advertise only what the GPU actually encodes (AV1 is narrow, an
+    // old iGPU might lack HEVC). NVENC and the GPU-less software path keep the static superset.
+    #[cfg(all(target_os = "windows", feature = "amf-qsv"))]
+    if crate::encode::windows_backend_is_ffmpeg() {
+        if let Some(m) = probed_mask(crate::encode::windows_codec_support()) {
+            return m;
+        }
+    }
+    SERVER_CODEC_MODE_SUPPORT
+}
+
+/// Turn a probed [`CodecSupport`](crate::encode::CodecSupport) into a `ServerCodecModeSupport` mask,
+/// or `None` if the probe found nothing — meaning the GPU wasn't usable at probe time (GPU-less CI,
+/// a misconfigured/wrong-vendor host), NOT that it encodes zero codecs; the caller then advertises
+/// the static superset (pre-probe behaviour) rather than claiming nothing.
+#[cfg(any(target_os = "linux", all(target_os = "windows", feature = "amf-qsv")))]
+fn probed_mask(caps: crate::encode::CodecSupport) -> Option<u32> {
     use super::{SCM_AV1_MAIN8, SCM_H264, SCM_HEVC};
-        let caps = crate::encode::vaapi_codec_support();
     let mut m = 0;
     if caps.h264 {
         m |= SCM_H264;
@@ -62,15 +82,7 @@ fn codec_mode_support() -> u32 {
     if caps.av1 {
         m |= SCM_AV1_MAIN8;
     }
-        // Only trust a probe that actually found an encoder. An empty result means VAAPI wasn't
+    (m != 0).then_some(m)
-        // usable at probe time (no VA display — a GPU-less CI box, or a misconfigured host), NOT
-        // that the GPU encodes nothing; advertise the static superset (pre-probe behaviour) rather
-        // than claiming zero codecs.
-        if m != 0 {
-            return m;
-        }
-    }
-    SERVER_CODEC_MODE_SUPPORT
 }
 
 #[cfg(test)]

crates/punktfunk-host/src/service.rs

@@ -603,7 +603,8 @@ fn uninstall() -> Result<()> {
     Ok(())
 }
 
-/// Write a default `host.env` if none exists, so a fresh install streams with NVENC out of the box.
+/// Write a default `host.env` if none exists, so a fresh install streams out of the box. The encoder
+/// defaults to `auto` — the host picks NVENC (NVIDIA) / AMF (AMD) / QSV (Intel) from the GPU vendor.
 fn ensure_default_host_env() -> Result<()> {
     let path = host_env_path();
     if path.exists() {
@@ -616,7 +617,9 @@ fn ensure_default_host_env() -> Result<()> {
         # KEY=VALUE per line; '#' comments. Restart the service after editing:\n\
         #   punktfunk-host service stop && punktfunk-host service start\n\
         \n\
-        PUNKTFUNK_ENCODER=nvenc\n\
+        # Encode backend: auto (default) detects the GPU vendor — NVIDIA->nvenc, AMD->amf, Intel->qsv.\n\
+        # Force one with nvenc | amf | qsv | sw (software H.264). amf/qsv need an FFmpeg-built host.\n\
+        PUNKTFUNK_ENCODER=auto\n\
         PUNKTFUNK_VIDEO_SOURCE=virtual\n\
         PUNKTFUNK_SECURE_DDA=1\n\
         RUST_LOG=info\n\
@@ -625,7 +628,7 @@ fn ensure_default_host_env() -> Result<()> {
         # compat). Use `serve` for a SECURE native-only host (no GameStream #5/#9 surface).\n\
         # PUNKTFUNK_HOST_CMD=serve --gamestream\n\
         \n\
-        # Force a specific NVENC render GPU by name substring (multi-GPU boxes only):\n\
+        # Force a specific render GPU by name substring (multi-GPU boxes only):\n\
         # PUNKTFUNK_RENDER_ADAPTER=4090\n";
     std::fs::write(&path, default).with_context(|| format!("write {}", path.display()))?;
     println!("Wrote default config: {}", path.display());

docs/windows-host.md

@@ -143,12 +143,15 @@ rustc 1.96 clippy is stricter than the Linux CI image on shared code, e.g. `need
    driver DLL — `lib /def:nvenc.def /machine:x64 /out:nvencodeapi.lib` where `nvenc.def` lists
    `NvEncodeAPICreateInstance` and `NvEncodeAPIGetMaxSupportedVersion` — and set
    `PUNKTFUNK_NVENC_LIB_DIR` to its directory.
-3. `cargo build -p punktfunk-host --features nvenc` (needs NASM + CMake for aws-lc-rs; libclang for
+3. `cargo build -p punktfunk-host --features nvenc,amf-qsv` for the all-vendor GPU host (NVENC for
-   any ffmpeg-using client). Default build (no feature) uses the openh264 software encoder.
+   NVIDIA; AMD AMF + Intel QSV via libavcodec — `amf-qsv` needs `FFMPEG_DIR` with the `*_amf`/`*_qsv`
+   encoders at build, e.g. the BtbN gpl-shared tree, and the FFmpeg DLLs on PATH at run; NASM + CMake
+   for aws-lc-rs; libclang for ffmpeg-sys-next). Default build (no feature) = openh264 software encoder.
 4. Run in the **interactive session** (not a Session-0 service / not over SSH — SendInput + DXGI
-   Desktop Duplication need a desktop): `serve` or `punktfunk1-host --source virtual`. Set
+   Desktop Duplication need a desktop): `serve` or `punktfunk1-host --source virtual`.
-   `PUNKTFUNK_ENCODER=nvenc` to select NVENC (the DXGI capturer switches to zero-copy D3D11 output to
+   `PUNKTFUNK_ENCODER=auto` (default) picks the backend from the GPU vendor — `nvenc`/`amf`/`qsv`/`sw`
-   match). The SudoVDA monitor activates once a real GPU drives WDDM, so capture + NVENC then work.
+   force one. The DXGI capturer emits zero-copy D3D11 NV12/P010 to match any GPU backend; the SudoVDA
+   monitor activates once a real GPU drives WDDM, so capture + encode then work.
 
 ### Dev loop (this repo → the Windows VM)
 
@@ -224,7 +227,7 @@ This is the highest-value first move and is **fully doable GPU-less**.
 | **VirtualDisplay** | KWin/gamescope/Mutter/Sway | **SudoVDA** IOCTLs (below) + `SetDisplayConfig` mode-set | ✅ likely (WARP) — *spike* |
 | **Capture** | PipeWire/dmabuf | **DXGI Desktop Duplication** primary, **WGC** fallback → `ID3D11Texture2D`; add `FramePayload::D3d11` | ⚠️ DDA-on-WARP unreliable; WGC-on-WARP unverified — *spike* |
 | **Zero-copy** | dmabuf→EGL/Vulkan→CUDA | register `ID3D11Texture2D` with NVENC (`NV_ENC_DEVICE_TYPE_DIRECTX`) — no CUDA bridge | ❌ needs real GPU |
-| **Encode** | ffmpeg `*_nvenc` | `openh264` SW (default on VM) + `nvidia-video-codec-sdk` HW (real GPU); behind `PUNKTFUNK_ENCODER` | SW ✅ / HW ❌ |
+| **Encode** | ffmpeg `*_nvenc`/`*_vaapi` | `nvidia-video-codec-sdk` (NVIDIA) + libavcodec `*_amf`/`*_qsv` (AMD/Intel, `encode/ffmpeg_win.rs`, `--features amf-qsv`) + `openh264` SW fallback; vendor-auto via `PUNKTFUNK_ENCODER` | NVENC ✅ live / AMF/QSV CI-only |
 | **Input kbd/mouse** | libei / wlr | **SendInput** with `MOUSEEVENTF_VIRTUALDESK` absolute mapping onto the virtual desktop rect (skip the VK→evdev table — client sends Win VKs; use `KEYEVENTF_SCANCODE`+`EXTENDEDKEY`) | ✅ |
 | **Gamepad** | uinput xpad + FF | **ViGEmBus** via `vigem-client` (`Xbox360Wired`); rumble via `request_notification()`→`XNotification{large,small}` | ✅ (install driver) |
 | **Audio capture** | PipeWire sink monitor | **WASAPI loopback** via the `wasapi` crate (48 kHz stereo f32 → existing Opus) | ⚠️ needs an audio endpoint |

packaging/windows/pack-host-installer.ps1

@@ -25,6 +25,7 @@ param(
     [string]$Publisher = 'CN=unom',
     [string]$PfxBase64 = $env:MSIX_CERT_PFX_B64,                    # reuse the client's signing secret
     [string]$PfxPassword = $env:MSIX_CERT_PASSWORD,
+    [string]$FfmpegDir = $env:FFMPEG_DIR,                           # bundle its bin\*.dll (amf-qsv build)
     [switch]$NoDriver,                                              # build without the bundled SudoVDA driver
     [switch]$NoSign                                                 # skip signing (local debug)
 )
@@ -146,6 +147,24 @@ if (-not $NoDriver) {
 }
 else { Write-Host "-NoDriver: building installer WITHOUT the bundled SudoVDA driver" }
 
+# --- stage the FFmpeg shared DLLs (AMD/Intel AMF/QSV build) ------------------------------------
+# A host built with --features amf-qsv link-imports avcodec/avutil/swscale/... so the shared DLLs
+# MUST sit next to the exe (it won't start otherwise). Bundle them from $FfmpegDir\bin — the same
+# BtbN gpl-shared tree the build linked against. A nvenc/software-only build doesn't import them, so
+# this is a harmless extra there; skipped entirely when $FfmpegDir is unset.
+$ffmpegBinSrc = if ($FfmpegDir) { Join-Path $FfmpegDir 'bin' } else { $null }
+if ($ffmpegBinSrc -and (Test-Path $ffmpegBinSrc)) {
+    $dlls = Get-ChildItem -Path $ffmpegBinSrc -Filter '*.dll' -ErrorAction SilentlyContinue
+    if ($dlls) {
+        $ffmpegStage = Join-Path $OutDir 'ffmpeg'
+        New-Item -ItemType Directory -Force -Path $ffmpegStage | Out-Null
+        $dlls | ForEach-Object { Copy-Item $_.FullName -Destination $ffmpegStage -Force }
+        $defines += "/DFfmpegBin=$ffmpegStage"
+        Write-Host "bundling $($dlls.Count) FFmpeg DLL(s) from $ffmpegBinSrc"
+    }
+}
+else { Write-Host "no FFMPEG_DIR\bin -> installer built WITHOUT FFmpeg DLLs (nvenc/software-only host)" }
+
 # --- build the installer (from the non-redirected copy under C:\t) -----------------------------
 Write-Host "==> ISCC $($defines -join ' ') $issLocal"
 & $iscc @defines $issLocal

packaging/windows/punktfunk-host.iss

@@ -32,6 +32,11 @@
 #ifdef StageDir
   #define WithDriver
 #endif
+; FfmpegBin (a dir of FFmpeg shared DLLs) is optional — present when the host is built with
+; --features amf-qsv (the AMD/Intel AMF/QSV encode backend link-imports the FFmpeg libs).
+#ifdef FfmpegBin
+  #define WithFfmpeg
+#endif
 
 [Setup]
 AppId={{7C9E6A52-1F4B-4E8D-A3C7-2B5D8F1E0A93}
@@ -68,6 +73,12 @@ Name: "startservice"; Description: "Start the punktfunk host service now (also s
 Source: "{#BinDir}\punktfunk-host.exe"; DestDir: "{app}"; Flags: ignoreversion
 Source: "{#HostEnv}"; DestDir: "{app}"; Flags: ignoreversion
 Source: "{#Readme}"; DestDir: "{app}"; DestName: "README.txt"; Flags: ignoreversion
+#ifdef WithFfmpeg
+; FFmpeg shared DLLs (avcodec/avutil/swscale/...) laid down next to the exe — the AMD/Intel
+; (AMF/QSV) encode backend link-imports them, so the exe won't start without them. NVENC/software-
+; only builds simply omit this block.
+Source: "{#FfmpegBin}\*.dll"; DestDir: "{app}"; Flags: ignoreversion
+#endif
 #ifdef WithDriver
 ; The driver payload + nefconc.exe + install-sudovda.ps1, extracted to {tmp} and removed after install.
 Source: "{#StageDir}\*"; DestDir: "{tmp}\sudovda"; Flags: deleteafterinstall recursesubdirs createallsubdirs; Tasks: installdriver

scripts/windows/host.env.example

@@ -9,9 +9,11 @@
 # Format: KEY=VALUE per line; '#' starts a comment. The service loads these into its environment
 # and passes PUNKTFUNK_* and RUST_LOG through to the host it launches into the active session.
 
-# Hardware encode via NVENC (NVIDIA). The host must be the `--features nvenc` build. Falls back to
+# Hardware encode backend. `auto` (default) detects the GPU vendor: NVIDIA->nvenc (direct SDK),
-# the software encoder automatically if NVENC is unavailable.
+# AMD->amf, Intel->qsv (both libavcodec). Force one with: nvenc | amf | qsv | sw (software H.264).
-PUNKTFUNK_ENCODER=nvenc
+# nvenc needs the `--features nvenc` build; amf/qsv need the `--features amf-qsv` build (FFmpeg DLLs
+# ship in the installer). The published installer is built with all three.
+PUNKTFUNK_ENCODER=auto
 
 # Video source: `virtual` creates a per-client virtual display (SudoVDA) at the client's exact
 # resolution + refresh — the flagship mode. Requires the SudoVDA indirect display driver installed.