fix(encode/qsv): write the colour VUI unconditionally + Intel-lane colour diagnostics

The native QSV encoder only attached mfxExtVideoSignalInfo on HDR sessions —
an SDR stream carried no colour description at all (NVENC writes 709-limited
unconditionally; qsv.rs now mirrors it).

Diagnostics for the field-reported blue/magenta Intel-host colours:
- hdr-p010-selftest takes an optional WxH + GPU vendor (intel|nvidia|amd) and
  prints which adapter it tested — 64x64 on the default GPU proved nothing on
  dual-GPU boxes, and the field heights (1080/1400) are not 16-aligned.
- pf-capture: ignored live test running the converter at 1920x1080 pinned to
  the Intel adapter.
- pf-encode: qsv_live_p010_1080_colorbars_dump — known P010 bars through the
  UNALIGNED-height ingest copy (1080 src -> align16 1088 pool, the seam no
  640x480 test exercises) to Main10 HEVC, dumped for off-box decode checks.
- dxgi.rs: drop the stale 'falls back to the R10 path' claim (no such path).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

crates/pf-capture/src/windows/dxgi.rs

@@ -308,8 +308,9 @@ float2 main(float4 pos : SV_POSITION, float2 uv : TEXCOORD0) : SV_TARGET {
 /// Plane writes use per-plane render-target views of the single P010 texture: an `R16_UNORM` RTV
 /// selects plane 0 (luma, full WxH), an `R16G16_UNORM` RTV selects plane 1 (chroma, W/2 x H/2). This
 /// planar-RTV mechanism needs a D3D11.3+ runtime + driver support; [`HdrP010Converter::convert`]
-/// surfaces a clear error if `CreateRenderTargetView` rejects the plane format so the caller can fall
-/// back to the existing R10 path.
+/// surfaces a clear error if `CreateRenderTargetView` rejects the plane format. (There is no runtime
+/// fallback — the error propagates through `try_consume` and ends the session; the "R10 path" the
+/// original design referenced was never kept.)
 pub(crate) struct HdrP010Converter {
     vs: ID3D11VertexShader,
     ps_y: ID3D11PixelShader,
@@ -737,14 +738,28 @@ fn p010_reference(r: f64, g: f64, b: f64) -> (f64, f64, f64) {
 /// Y ≤ 4 codes, U/V ≤ 5 codes (rounding + chroma averaging). Prints a per-colour table + PASS/FAIL.
 #[cfg(target_os = "windows")]
 pub fn hdr_p010_selftest() -> Result<()> {
-    use windows::Win32::Graphics::Direct3D::D3D_DRIVER_TYPE_HARDWARE;
-    use windows::Win32::Graphics::Dxgi::IDXGIAdapter;
+    hdr_p010_selftest_at(64, 64, None)
+}
 
-    // 64x64, even dims. A 4x4 grid of 16x16 flat scRGB blocks (each 2x2 chroma footprint uniform →
-    // exact chroma comparison) covering pure R/G/B/white/black/gray at plausible HDR nit levels, plus
-    // a couple of bright (>1.0 scRGB) colours, then the rest is a gradient (compared on Y only).
-    const W: u32 = 64;
-    const H: u32 = 64;
+/// [`hdr_p010_selftest`] at an arbitrary even size and (optionally) on a specific GPU vendor
+/// (PCI vendor id, e.g. `0x8086` Intel / `0x10de` NVIDIA / `0x1002` AMD). The size matters on
+/// top of the 64×64 default because the field sessions run at capture resolutions whose height
+/// is NOT 16-aligned (1080 → the encoder's align16 pool seam) and a driver may treat the planar
+/// RTVs differently at real sizes; the vendor pin matters on dual-GPU boxes where the default
+/// adapter is not the one the session encodes on.
+#[cfg(target_os = "windows")]
+pub fn hdr_p010_selftest_at(w: u32, h: u32, vendor: Option<u32>) -> Result<()> {
+    use windows::Win32::Graphics::Direct3D::{D3D_DRIVER_TYPE_HARDWARE, D3D_DRIVER_TYPE_UNKNOWN};
+    use windows::Win32::Graphics::Dxgi::{CreateDXGIFactory1, IDXGIAdapter, IDXGIFactory1};
+
+    if w == 0 || h == 0 || w % 2 != 0 || h % 2 != 0 {
+        bail!("hdr-p010-selftest needs even non-zero dimensions, got {w}x{h}");
+    }
+    // A grid of 16x16 flat scRGB blocks (each 2x2 chroma footprint uniform → exact chroma
+    // comparison) covering pure R/G/B/white/black/gray at plausible HDR nit levels, plus a couple
+    // of bright (>1.0 scRGB) colours, then the rest is a gradient (compared on Y only).
+    #[allow(non_snake_case)]
+    let (W, H) = (w, h);
     const BLK: u32 = 16;
     // (name, r, g, b) scRGB linear (1.0 = 80 nits). Mix of SDR-ish and HDR (>1.0) values.
     let named: [(&str, f32, f32, f32); 8] = [
@@ -797,12 +812,36 @@ pub fn hdr_p010_selftest() -> Result<()> {
     // `fp16` outlives the synchronous `CreateTexture2D` that reads it. The mapped-pointer reads are
     // proven individually at the `read_u16` closure below.
     unsafe {
-        // Hardware D3D11 device (no adapter pin — the default GPU is fine for the self-test).
+        // Device on the requested vendor's adapter (dual-GPU boxes encode on a specific one), else
+        // the default hardware GPU. Always says which adapter ran — a PASS is only meaningful for
+        // the GPU it actually tested.
+        let adapter: Option<IDXGIAdapter> = match vendor {
+            None => None,
+            Some(want) => {
+                let factory: IDXGIFactory1 = CreateDXGIFactory1().context("dxgi factory")?;
+                let mut found = None;
+                for i in 0.. {
+                    let Ok(a) = factory.EnumAdapters(i) else {
+                        break;
+                    };
+                    let desc = a.GetDesc().context("adapter desc")?;
+                    if desc.VendorId == want {
+                        found = Some(a);
+                        break;
+                    }
+                }
+                Some(found.with_context(|| format!("no adapter with vendor id {want:#x}"))?)
+            }
+        };
         let mut device: Option<ID3D11Device> = None;
         let mut context: Option<ID3D11DeviceContext> = None;
         D3D11CreateDevice(
-            None::<&IDXGIAdapter>,
-            D3D_DRIVER_TYPE_HARDWARE,
+            adapter.as_ref(),
+            if adapter.is_some() {
+                D3D_DRIVER_TYPE_UNKNOWN
+            } else {
+                D3D_DRIVER_TYPE_HARDWARE
+            },
             HMODULE::default(),
             D3D11_CREATE_DEVICE_BGRA_SUPPORT,
             Some(&[D3D_FEATURE_LEVEL_11_0]),
@@ -814,6 +853,22 @@ pub fn hdr_p010_selftest() -> Result<()> {
         .context("D3D11CreateDevice(hardware) for hdr-p010-selftest")?;
         let device = device.context("null device")?;
         let context = context.context("null context")?;
+        {
+            let dxgi: windows::Win32::Graphics::Dxgi::IDXGIDevice =
+                device.cast().context("device -> IDXGIDevice")?;
+            let desc = dxgi.GetAdapter().context("GetAdapter")?.GetDesc()?;
+            let name = String::from_utf16_lossy(
+                &desc.Description[..desc
+                    .Description
+                    .iter()
+                    .position(|&c| c == 0)
+                    .unwrap_or(desc.Description.len())],
+            );
+            println!(
+                "adapter: {name} (vendor {:#06x}, luid {:08x}:{:08x})",
+                desc.VendorId, desc.AdapterLuid.HighPart, desc.AdapterLuid.LowPart
+            );
+        }
 
         // Source FP16 texture (initialized) + SRV.
         let src_desc = D3D11_TEXTURE2D_DESC {
@@ -1175,3 +1230,16 @@ impl VideoConverter {
         blt.context("VideoProcessorBlt")
     }
 }
+
+#[cfg(test)]
+mod hdr_selftests {
+    /// LIVE (needs the GPU): [`super::hdr_p010_selftest_at`] at the field capture size — 1080 is
+    /// NOT 16-aligned, and the planar-RTV write path is driver-specific per vendor. Pinned to the
+    /// Intel adapter (`0x8086`), so it runs on the Intel validation boxes and errors out cleanly
+    /// ("no adapter") elsewhere. `cargo test -p pf-capture -- --ignored hdr_p010 --nocapture`.
+    #[test]
+    #[ignore]
+    fn hdr_p010_selftest_intel_1080_live() {
+        super::hdr_p010_selftest_at(1920, 1080, Some(0x8086)).expect("hdr p010 selftest @1080");
+    }
+}

crates/pf-encode/src/enc/windows/qsv.rs

@@ -478,10 +478,14 @@ fn build_params(cfg: &EncodeConfig) -> ParamSet {
         b
     });
 
-    // HDR signalling (10-bit sessions are the HDR path on Windows — same coupling as NVENC):
-    // BT.2020/PQ colour description + the source's mastering/CLL grade at every IDR.
+    // Colour signalling, written UNCONDITIONALLY (mirrors nvenc_core.rs): the input is already
+    // CSC'd to a specific matrix — BT.709 limited for SDR (the capture-side VideoConverter),
+    // BT.2020 PQ for HDR (HdrP010Converter) — so the stream must say so. An SDR stream without a
+    // colour description leaves the choice to the decoder's "unspecified" default, and
+    // Moonlight/third-party/Android-vendor decoders default to 601 at sub-HD → mis-rendered
+    // colours. (10-bit sessions are the HDR path on Windows — same coupling as NVENC.)
     let hdr = cfg.ten_bit && cfg.codec != Codec::H264;
-    let vsi = hdr.then(|| {
+    let vsi = {
         // SAFETY: all-zero is valid; header stamped below.
         let mut b: Box<vpl::mfxExtVideoSignalInfo> = Box::new(unsafe { std::mem::zeroed() });
         b.Header.BufferId = vpl::MFX_EXTBUFF_VIDEO_SIGNAL_INFO as u32;
@@ -489,11 +493,17 @@ fn build_params(cfg: &EncodeConfig) -> ParamSet {
         b.VideoFormat = 5; // unspecified
         b.VideoFullRange = 0;
         b.ColourDescriptionPresent = 1;
-        b.ColourPrimaries = 9; // BT.2020
-        b.TransferCharacteristics = 16; // SMPTE ST 2084 (PQ)
-        b.MatrixCoefficients = 9; // BT.2020 non-constant
-        b
-    });
+        if hdr {
+            b.ColourPrimaries = 9; // BT.2020
+            b.TransferCharacteristics = 16; // SMPTE ST 2084 (PQ)
+            b.MatrixCoefficients = 9; // BT.2020 non-constant
+        } else {
+            b.ColourPrimaries = 1; // BT.709
+            b.TransferCharacteristics = 1; // BT.709
+            b.MatrixCoefficients = 1; // BT.709
+        }
+        Some(b)
+    };
     let mastering = cfg.hdr_meta.filter(|_| hdr).map(|m| {
         // SAFETY: all-zero is valid; header stamped below.
         let mut b: Box<vpl::mfxExtMasteringDisplayColourVolume> =
@@ -1994,4 +2004,171 @@ mod tests {
             "the bitrate retarget emitted a keyframe (StartNewSequence leak)"
         );
     }
+
+    /// FULL-CHAIN colour check at the field capture size: a known P010 colour-bar source at
+    /// 1920x1080 — whose height is NOT 16-aligned, so the ingest `CopySubresourceRegion` copies
+    /// into a 1920x1088 runtime pool surface whose chroma plane sits at a DIFFERENT row offset
+    /// than the source's (the seam no 640x480 test exercises) — encoded to Main10 HEVC and
+    /// dumped to `%TEMP%\pf_qsv_1080_bars.h265` for off-box decode verification against the
+    /// same codes. On-box this asserts stream shape; the pixel verdict needs a decoder.
+    #[test]
+    fn qsv_live_p010_1080_colorbars_dump() {
+        use windows::Win32::Graphics::Direct3D::D3D_DRIVER_TYPE_UNKNOWN;
+        use windows::Win32::Graphics::Direct3D11::{
+            D3D11CreateDevice, D3D11_BIND_RENDER_TARGET, D3D11_BIND_SHADER_RESOURCE,
+            D3D11_SDK_VERSION, D3D11_SUBRESOURCE_DATA, D3D11_USAGE_DEFAULT,
+        };
+        use windows::Win32::Graphics::Dxgi::Common::{DXGI_FORMAT_P010, DXGI_SAMPLE_DESC};
+        use windows::Win32::Graphics::Dxgi::{CreateDXGIFactory1, IDXGIAdapter1, IDXGIFactory4};
+
+        // (Y, Cb, Cr) 10-bit limited codes for the 8 sRGB bars white/yellow/cyan/green/magenta/
+        // red/blue/black at 80-nit SDR white under PQ/BT.2020 — the same math as pf-capture's
+        // `p010_reference` (and the bars_pq2020 client fixture). Stored MSB-aligned (`<<6`).
+        const BARS: [(u16, u16, u16); 8] = [
+            (490, 512, 512),
+            (478, 423, 518),
+            (464, 525, 473),
+            (450, 432, 476),
+            (350, 584, 585),
+            (325, 448, 598),
+            (226, 650, 535),
+            (64, 512, 512),
+        ];
+        const W: u32 = 1920;
+        const H: u32 = 1080;
+
+        init_tracing();
+        let Ok((_l, impls)) = intel_loader() else {
+            eprintln!("skipping: no VPL loader");
+            return;
+        };
+        let Some(imp) = impls.iter().find(|i| i.luid_valid) else {
+            eprintln!("skipping: no Intel VPL implementation on this box");
+            return;
+        };
+        if !probe_can_encode_10bit(Codec::H265) {
+            eprintln!("skipping: this GPU declines 10-bit HEVC");
+            return;
+        }
+
+        // P010 initial data: plane 0 = H rows of W u16 luma; plane 1 = H/2 rows of W u16
+        // (interleaved Cb,Cr pairs), same pitch. Bars are vertical: bar index = x / (W/8).
+        let bar_w = (W / 8) as usize;
+        let mut init = vec![0u16; (W as usize) * (H as usize + H as usize / 2)];
+        for y in 0..H as usize {
+            for x in 0..W as usize {
+                init[y * W as usize + x] = BARS[(x / bar_w).min(7)].0 << 6;
+            }
+        }
+        let chroma_base = (W as usize) * (H as usize);
+        for cy in 0..(H as usize / 2) {
+            for cx in 0..(W as usize / 2) {
+                let (_, cb, cr) = BARS[((cx * 2) / bar_w).min(7)];
+                init[chroma_base + cy * W as usize + cx * 2] = cb << 6;
+                init[chroma_base + cy * W as usize + cx * 2 + 1] = cr << 6;
+            }
+        }
+
+        // SAFETY: self-contained harness on one thread/device (same contract as `drive_live`);
+        // the initial-data pointer outlives the synchronous CreateTexture2D that reads it.
+        let (device, tex) = unsafe {
+            let luid = windows::Win32::Foundation::LUID {
+                LowPart: u32::from_le_bytes(imp.luid[..4].try_into().unwrap()),
+                HighPart: i32::from_le_bytes(imp.luid[4..].try_into().unwrap()),
+            };
+            let factory: IDXGIFactory4 = CreateDXGIFactory1().expect("dxgi factory");
+            let adapter: IDXGIAdapter1 = factory.EnumAdapterByLuid(luid).expect("intel adapter");
+            let mut device = None;
+            D3D11CreateDevice(
+                &adapter,
+                D3D_DRIVER_TYPE_UNKNOWN,
+                windows::Win32::Foundation::HMODULE::default(),
+                Default::default(),
+                None,
+                D3D11_SDK_VERSION,
+                Some(&mut device),
+                None,
+                None,
+            )
+            .expect("d3d11 device on intel adapter");
+            let device: ID3D11Device = device.expect("device");
+            let desc = D3D11_TEXTURE2D_DESC {
+                Width: W,
+                Height: H,
+                MipLevels: 1,
+                ArraySize: 1,
+                Format: DXGI_FORMAT_P010,
+                SampleDesc: DXGI_SAMPLE_DESC {
+                    Count: 1,
+                    Quality: 0,
+                },
+                Usage: D3D11_USAGE_DEFAULT,
+                BindFlags: (D3D11_BIND_RENDER_TARGET.0 | D3D11_BIND_SHADER_RESOURCE.0) as u32,
+                CPUAccessFlags: 0,
+                MiscFlags: 0,
+            };
+            let data = D3D11_SUBRESOURCE_DATA {
+                pSysMem: init.as_ptr() as *const std::ffi::c_void,
+                SysMemPitch: W * 2,
+                SysMemSlicePitch: 0,
+            };
+            let mut t: Option<ID3D11Texture2D> = None;
+            device
+                .CreateTexture2D(&desc, Some(&data), Some(&mut t))
+                .expect("bar texture");
+            (device.clone(), t.expect("texture"))
+        };
+
+        let mut enc = QsvEncoder::open(
+            Codec::H265,
+            PixelFormat::P010,
+            W,
+            H,
+            30,
+            10_000_000,
+            10,
+            ChromaFormat::Yuv420,
+        )
+        .expect("open");
+        enc.set_hdr_meta(Some(test_hdr_meta()));
+        let mut stream = Vec::new();
+        let mut aus = 0usize;
+        let mut keyframes = 0usize;
+        for i in 0..12u32 {
+            let frame = CapturedFrame {
+                width: W,
+                height: H,
+                pts_ns: i as u64 * 33_333_333,
+                format: PixelFormat::P010,
+                payload: FramePayload::D3d11(pf_frame::dxgi::D3d11Frame {
+                    texture: tex.clone(),
+                    device: device.clone(),
+                    pyro: None,
+                }),
+                cursor: None,
+            };
+            enc.submit_indexed(&frame, i).expect("submit");
+            if let Some(au) = enc.poll().expect("poll") {
+                aus += 1;
+                keyframes += au.keyframe as usize;
+                stream.extend_from_slice(&au.data);
+            }
+        }
+        enc.flush().expect("flush");
+        while let Some(au) = enc.poll().expect("drain") {
+            aus += 1;
+            keyframes += au.keyframe as usize;
+            stream.extend_from_slice(&au.data);
+        }
+        assert!(aus >= 10, "expected ≥10 AUs, got {aus}");
+        assert!(keyframes >= 1, "expected an IDR in the dump");
+        let path = std::env::temp_dir().join("pf_qsv_1080_bars.h265");
+        std::fs::write(&path, &stream).expect("write dump");
+        println!(
+            "wrote {} AUs ({} bytes, {keyframes} keyframes) to {}",
+            aus,
+            stream.len(),
+            path.display()
+        );
+    }
 }

crates/punktfunk-host/src/main.rs

@@ -322,7 +322,33 @@ fn real_main() -> Result<()> {
         // `PUNKTFUNK_HDR_SHADER_P010` colour math without green-screening a live HDR stream. Prints
         // PASS/FAIL + max Y/Cb/Cr error.
         #[cfg(target_os = "windows")]
-        Some("hdr-p010-selftest") => crate::capture::dxgi::hdr_p010_selftest(),
+        Some("hdr-p010-selftest") => {
+            // Optional args: a `WxH` size (default 64x64 — pass the real capture size: heights
+            // like 1080 are NOT 16-aligned and exercise a different driver path) and a GPU
+            // vendor (`intel`|`nvidia`|`amd` — dual-GPU boxes otherwise test the default
+            // adapter, which may not be the one that encodes).
+            let mut size = (64u32, 64u32);
+            let mut vendor = None;
+            for a in args.iter().skip(2) {
+                match a.as_str() {
+                    "intel" => vendor = Some(0x8086),
+                    "nvidia" => vendor = Some(0x10de),
+                    "amd" => vendor = Some(0x1002),
+                    s => {
+                        let parsed = s
+                            .split_once('x')
+                            .and_then(|(w, h)| Some((w.parse().ok()?, h.parse().ok()?)));
+                        match parsed {
+                            Some(wh) => size = wh,
+                            None => anyhow::bail!(
+                                "hdr-p010-selftest: unrecognized arg {s:?} (want WxH or intel|nvidia|amd)"
+                            ),
+                        }
+                    }
+                }
+            }
+            crate::capture::dxgi::hdr_p010_selftest_at(size.0, size.1, vendor)
+        }
         // Linux HDR readiness probe (GNOME 50+ portal path): prints whether a monitor is currently
         // in BT.2100 (HDR) colour mode, whether the NVENC/VAAPI backend probes Main10 for
         // HEVC/AV1, and the GameStream HDR capability the two combine into — the "why isn't my