fix(pyrowave): per-session raw-dmabuf zero-copy capture on the Linux NVIDIA host

A PyroWave session on an NVIDIA-auto host was forced onto CPU-RGB capture
(session_plan flipped gpu=false): Mutter blits tiled->LINEAR, we mmap +
de-pad ~30 MB, the encoder re-uploads it - three full-frame CPU touches
per frame at 5120x1440 while an HEVC session on the same box rides the
tiled EGL/CUDA zero-copy. The dmabuf passthrough + Vulkan tiled import
were already validated (8dc5d672) but only reachable via the global
PUNKTFUNK_ENCODER=pyrowave lab policy.

ZeroCopyPolicy gains pyrowave_session (from OutputFormat.pyrowave, i.e.
the negotiated codec): the capturer skips the NVENC-only EGL->CUDA
importer, takes the raw-dmabuf passthrough, and advertises the wavelet
encoder's Vulkan-importable modifiers so Mutter+NVIDIA negotiates tiled
zero-copy. The forced-CPU flip in session_plan is gone.

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

crates/pf-capture/src/lib.rs

@@ -244,8 +244,13 @@ pub struct ZeroCopyPolicy {
     /// The resolved backend produces GPU-resident frames (everything but the software encoder) —
     /// used only to phrase the CPU-fallback warning (the host `encode::resolved_backend_is_gpu`).
     pub backend_is_gpu: bool,
+    /// THIS session encodes PyroWave: the frames' consumer is the wavelet encoder's own Vulkan
+    /// device, which imports raw dmabufs on ANY vendor — so the capturer takes the raw-dmabuf
+    /// passthrough (like the VAAPI backend) instead of the EGL→CUDA import whose payloads only
+    /// NVENC can consume. Per-session (the codec is negotiated), unlike `backend_is_vaapi`.
+    pub pyrowave_session: bool,
     /// The PyroWave encoder's Vulkan-importable dmabuf modifiers for the capture's packed-RGB fourcc,
-    /// resolved when the encoder pref is `pyrowave` (the passthrough advertises them so Mutter+NVIDIA,
+    /// resolved when the session encodes PyroWave (the passthrough advertises them so Mutter+NVIDIA,
     /// which allocates tiled-only, still negotiates zero-copy). Empty otherwise.
     pub pyrowave_modifiers: Vec<u64>,
 }

crates/pf-capture/src/linux/mod.rs

@@ -255,9 +255,11 @@ fn spawn_pipewire(
         want_hdr
     };
     // Mirror of the thread's `vaapi_passthrough` decision (deterministic from here: on a VAAPI
-    // backend the EGL→CUDA importer is never built) — kept on the capturer so `next_frame`'s
-    // negotiation-timeout branch knows a failed negotiation was the LINEAR-dmabuf offer.
-    let vaapi_dmabuf = zerocopy && !force_shm && policy.backend_is_vaapi;
+    // backend or a PyroWave session the EGL→CUDA importer is never built) — kept on the capturer
+    // so `next_frame`'s negotiation-timeout branch knows a failed negotiation was the raw-dmabuf
+    // passthrough offer.
+    let vaapi_dmabuf =
+        zerocopy && !force_shm && (policy.backend_is_vaapi || policy.pyrowave_session);
     let join = thread::Builder::new()
         .name("punktfunk-pipewire".into())
         .spawn(move || {
@@ -1945,16 +1947,18 @@ mod pipewire {
         // Build the GPU importer up front — normally the ISOLATED worker process
         // (design/zerocopy-worker-isolation.md), so a driver fault on a dying compositor's
         // dmabuf kills the worker, not this host. If it fails, log and fall back to the CPU path
-        // (we simply won't request dmabuf below). Skipped entirely when the encode backend is
-        // VAAPI: those frames go to the raw-dmabuf passthrough, and building the importer there
-        // would waste a CUDA probe — or worse, on an NVIDIA box forced to PUNKTFUNK_ENCODER=vaapi,
-        // succeed and produce CUDA payloads the VAAPI encoder must reject. Also skipped once
-        // repeated worker deaths latched the import off (a wedged GPU stack must not crash-loop).
+        // (we simply won't request dmabuf below). Skipped entirely when the frames go to the
+        // raw-dmabuf passthrough — the encode backend is VAAPI, or the SESSION encodes PyroWave
+        // (its Vulkan device imports raw dmabufs on any vendor): building the importer there
+        // would waste a CUDA probe — or worse, succeed and produce CUDA payloads only NVENC can
+        // consume. Also skipped once repeated worker deaths latched the import off (a wedged GPU
+        // stack must not crash-loop).
         let backend_is_vaapi = policy.backend_is_vaapi;
+        let raw_passthrough = backend_is_vaapi || policy.pyrowave_session;
         // HDR never builds the EGL→CUDA importer: its de-tile blit renders into 8-bit RGBA8,
         // which would silently crush the 10-bit depth. The HDR consumers are the CPU mmap path
         // (LINEAR de-pad → X2Rgb10 CPU frames) and the VAAPI raw-dmabuf passthrough.
-        let mut importer = if zerocopy && !backend_is_vaapi && !want_hdr {
+        let mut importer = if zerocopy && !raw_passthrough && !want_hdr {
             if pf_zerocopy::gpu_import_disabled() {
                 tracing::warn!(
                     "zero-copy GPU import disabled after repeated import-worker deaths — using CPU path"
@@ -1980,9 +1984,11 @@ mod pipewire {
         // host. KWin/gamescope don't need it (they blit into the buffer, so no read-before-render
         // race).
         let force_shm = std::env::var("PUNKTFUNK_FORCE_SHM").as_deref() == Ok("1");
-        // VAAPI zero-copy passthrough: zero-copy on, no EGL→CUDA importer (any non-NVIDIA host), and
-        // the encoder backend is VAAPI → hand the raw dmabuf to the encoder (it imports + GPU-CSCs).
-        let vaapi_passthrough = zerocopy && !force_shm && importer.is_none() && backend_is_vaapi;
+        // Raw-dmabuf zero-copy passthrough: zero-copy on, no EGL→CUDA importer, and the frames'
+        // consumer imports raw dmabufs itself — the VAAPI backend (libva import + GPU CSC) or a
+        // PyroWave session (the wavelet encoder's own Vulkan device, any vendor) → hand the raw
+        // dmabuf straight to the encoder.
+        let vaapi_passthrough = zerocopy && !force_shm && importer.is_none() && raw_passthrough;
         // Modifiers our import stack handles for BGRx: the EGL-importable (tiled) set, plus LINEAR
         // (0) — NVIDIA's EGL won't list it, but LINEAR dmabufs (gamescope's only offer) import via
         // CUDA external memory instead. For the VAAPI passthrough path we advertise LINEAR only:
@@ -1998,8 +2004,9 @@ mod pipewire {
         // advertisement with every modifier its device samples from, so compositors that
         // never allocate LINEAR (Mutter+NVIDIA) still negotiate zero-copy dmabufs. The modifiers
         // were resolved by the facade (`ZeroCopyPolicy::pyrowave_modifiers`) — non-empty only when
-        // the host's `pyrowave` feature is on AND the encoder pref is `pyrowave` — so capture never
-        // calls back into `encode` and needs no feature gate of its own (the emptiness check gates it).
+        // the host's `pyrowave` feature is on AND the session (or the global encoder pref) is
+        // PyroWave — so capture never calls back into `encode` and needs no feature gate of its
+        // own (the emptiness check gates it).
         if vaapi_passthrough && !policy.pyrowave_modifiers.is_empty() {
             for &m in &policy.pyrowave_modifiers {
                 if !modifiers.contains(&m) {
@@ -2019,11 +2026,11 @@ mod pipewire {
             );
         } else if zerocopy && !want_dmabuf {
             tracing::warn!("zero-copy: no importable dmabuf modifiers — using CPU path");
-        } else if vaapi_passthrough {
+        } else if vaapi_passthrough && policy.pyrowave_modifiers.is_empty() {
             tracing::info!(
                 "zero-copy: advertising LINEAR dmabuf for direct VAAPI import (GPU CSC)"
             );
-        } else if want_dmabuf {
+        } else if want_dmabuf && !vaapi_passthrough {
             tracing::info!(
                 count = modifiers.len(),
                 sample = ?&modifiers[..modifiers.len().min(6)],

crates/punktfunk-host/src/capture.rs

@@ -26,24 +26,35 @@ pub use pf_capture::{dxgi, synthetic_nv12};
 /// capture→encode cycle). Resolved here (the host facade) and threaded in, so the edge stays one-way
 /// (plan §2.4 / §W6).
 #[cfg(target_os = "linux")]
-fn zero_copy_policy() -> pf_capture::ZeroCopyPolicy {
+fn zero_copy_policy(pyrowave_session: bool) -> pf_capture::ZeroCopyPolicy {
     let backend_is_vaapi = crate::encode::linux_zero_copy_is_vaapi();
+    // The raw-dmabuf passthrough serves a PyroWave session on ANY vendor (the wavelet encoder's
+    // own Vulkan device imports the dmabuf) — per-session from the negotiated codec, plus the
+    // global `PUNKTFUNK_ENCODER=pyrowave` lab lever (which also flips `backend_is_vaapi`).
     #[cfg(feature = "pyrowave")]
-    let pyrowave_modifiers =
-        if backend_is_vaapi && pf_host_config::config().encoder_pref.as_str() == "pyrowave" {
-            // BGRx is the capture path's canonical packed-RGB format (the modifier advertisement keys
-            // on it). `drm_fourcc(Bgrx)` is always `Some`.
-            pf_frame::drm_fourcc(PixelFormat::Bgrx)
-                .map(crate::encode::pyrowave_capture_modifiers)
-                .unwrap_or_default()
-        } else {
-            Vec::new()
-        };
+    let pyrowave_session =
+        pyrowave_session || pf_host_config::config().encoder_pref.as_str() == "pyrowave";
+    #[cfg(not(feature = "pyrowave"))]
+    let pyrowave_session = {
+        let _ = pyrowave_session;
+        false
+    };
+    #[cfg(feature = "pyrowave")]
+    let pyrowave_modifiers = if pyrowave_session {
+        // BGRx is the capture path's canonical packed-RGB format (the modifier advertisement keys
+        // on it). `drm_fourcc(Bgrx)` is always `Some`.
+        pf_frame::drm_fourcc(PixelFormat::Bgrx)
+            .map(crate::encode::pyrowave_capture_modifiers)
+            .unwrap_or_default()
+    } else {
+        Vec::new()
+    };
     #[cfg(not(feature = "pyrowave"))]
     let pyrowave_modifiers = Vec::new();
     pf_capture::ZeroCopyPolicy {
         backend_is_vaapi,
         backend_is_gpu: crate::encode::resolved_backend_is_gpu(),
+        pyrowave_session,
         pyrowave_modifiers,
     }
 }
@@ -57,7 +68,9 @@ pub fn open_portal_monitor(want_hdr: bool) -> Result<Box<dyn Capturer>> {
     // session so it inherits that grant headlessly; wlroots/Sway has no RemoteDesktop portal,
     // so use a plain ScreenCast session there.
     let anchored = crate::inject::default_backend() == crate::inject::Backend::Libei;
-    pf_capture::open_portal_monitor(anchored, want_hdr, zero_copy_policy())
+    // Monitor mirrors never carry the native PyroWave plane (GameStream protocol) — per-session
+    // passthrough is virtual-output-only; the global encoder-pref lever still applies inside.
+    pf_capture::open_portal_monitor(anchored, want_hdr, zero_copy_policy(false))
 }
 
 #[cfg(not(target_os = "linux"))]
@@ -88,7 +101,7 @@ pub fn capture_virtual_output(
         vout.keepalive,
         want.gpu,
         want.chroma_444,
-        zero_copy_policy(),
+        zero_copy_policy(want.pyrowave),
     )
 }
 

crates/punktfunk-host/src/session_plan.rs

@@ -155,36 +155,22 @@ impl SessionPlan {
             }
             gpu && !force_cpu_for_nvenc_444
         };
-        // PyroWave on an NVIDIA-auto host: the `gpu` capture path resolves to the EGL→CUDA
-        // import that only NVENC can consume — the wavelet backend ingests raw dmabufs
-        // (the AMD/Intel path) or CPU RGB. Flip THIS session to CPU RGB capture; the
-        // Phase-2 exit sessions ran exactly this shape at 60 fps (the encode itself stays
-        // sub-ms GPU compute). Per-session raw-dmabuf passthrough on NVIDIA (true
-        // zero-copy without the PUNKTFUNK_ENCODER=pyrowave capture policy) is the
-        // follow-up; the AMD/Intel dmabuf path is untouched.
-        #[cfg(target_os = "linux")]
-        let gpu = {
-            let pyro_needs_cpu = self.codec == crate::encode::Codec::PyroWave
-                && !crate::encode::linux_zero_copy_is_vaapi();
-            if gpu && pyro_needs_cpu {
-                tracing::info!(
-                    "PyroWave session on the NVIDIA capture path: GPU (CUDA) capture disabled \
-                     for this session — frames arrive as CPU RGB and upload to the wavelet \
-                     encoder (raw-dmabuf zero-copy on NVIDIA is a follow-up)"
-                );
-            }
-            gpu && !pyro_needs_cpu
-        };
+        // PyroWave on Linux keeps `gpu = true`: the capture facade sees `pyrowave` below and
+        // routes the session onto the raw-dmabuf passthrough (the wavelet encoder's own Vulkan
+        // device imports the compositor's dmabuf on ANY vendor — `ZeroCopyPolicy::pyrowave_session`
+        // advertises its importable modifiers, so Mutter+NVIDIA negotiates tiled zero-copy instead
+        // of the old forced CPU-RGB readback). The EGL→CUDA importer is skipped there — its
+        // payloads only NVENC consumes.
         crate::capture::OutputFormat {
             gpu,
             hdr: self.hdr,
             // 4:4:4 needs a full-chroma source: on Windows this keeps the capturer on RGB (not the
             // default NV12/P010 video-engine output) so NVENC can CSC to 4:4:4.
             chroma_444: self.chroma.is_444(),
-            // PyroWave (Windows): the IDD-push capturer makes its NV12 out-ring shareable + signals a
-            // shared fence so the wavelet encoder can zero-copy-import the texture into its own Vulkan
-            // device. Inert on Linux (the wavelet backend ingests dmabufs / CPU RGB there — handled
-            // by the `gpu` flips above, not this flag).
+            // PyroWave: on Windows the IDD-push capturer makes its NV12 out-ring shareable + signals
+            // a shared fence so the wavelet encoder can zero-copy-import the texture into its own
+            // Vulkan device; on Linux the capture facade flips the zero-copy policy to the
+            // raw-dmabuf passthrough (see above).
             pyrowave: self.codec == crate::encode::Codec::PyroWave,
         }
     }