PERF-2 P3-ii: the device V-cycle as the CG's preconditioner — poisson/device.rs (one CUDA runtime per process, persistent per-operator buffers, the mg_vcycle.cu kernels at K = 1; upload r, run the V-cycle, download z; the f64 CG unchanged), MultigridParameters::device, Prepared holds the device hierarchy, the CG driver destructures the prepared operator instead of cloning it; EmbeddedPisoSolver::set_poisson_device, overset pass-through, harness knob RTX_FSI2O_MG_DEVICE=1; export_levels factored out; the quarantine's dangling cfg attribute fixed

Co-Authored-By: Claude Fable 5.1 <[email protected]>
Claude-Session: https://claude.ai/code/session_01YJPeT6WA2e7YvAnS875AHL

crates/specialized/rtx-cfd/src/solvers/incompressible/embedded/mod.rs

@@ -180,6 +180,9 @@ pub struct EmbeddedPisoSolver {
     pcg_cache: std::cell::RefCell<super::poisson::PcgCache>,
     /// PERF-2 P2: threads for the multigrid's red-black maps (default 1).
     poisson_threads: usize,
+    /// PERF-2 P3-ii: the V-cycle on the CUDA device (needs the `cuda`
+    /// feature and the red-black smoother; default off).
+    poisson_device: bool,
     moving: bool,
     /// Mask hysteresis band in multiples of the min cell size (0 = off).
     mask_hysteresis: f64,
@@ -209,6 +212,7 @@ impl EmbeddedPisoSolver {
             poisson_profile: std::cell::Cell::new((0, 0, 0, 0)),
             pcg_cache: std::cell::RefCell::new(super::poisson::PcgCache::default()),
             poisson_threads: 1,
+            poisson_device: false,
             moving: false,
             mask_hysteresis: 0.0,
             time: 0.0,
@@ -234,6 +238,11 @@ impl EmbeddedPisoSolver {
         self.poisson_threads = threads.max(1);
     }
 
+    /// The V-cycle on the CUDA device (PERF-2 P3-ii; a regime, band-gated).
+    pub fn set_poisson_device(&mut self, on: bool) {
+        self.poisson_device = on;
+    }
+
     /// Mask hysteresis for the moving-body rebuild, as a fraction of the
     /// min cell size (default 0, exactly the plain rebuild). With a band,
     /// a cell within `band * h_min` of the surface keeps the

crates/specialized/rtx-cfd/src/solvers/incompressible/embedded/projection.rs

@@ -301,6 +301,7 @@ impl EmbeddedPisoSolver {
                     precision: self.parameters.poisson_precision,
                     smoother: self.parameters.poisson_smoother,
                     threads: self.poisson_threads,
+                    device: self.poisson_device,
                     ..MultigridParameters::default()
                 },
                 inner_stop,