embedded3 S2-2a/S2-3: moving bodies on the device (phased tables, host rebuild per step, impose kernel; gate test), the per-span cut wall route, the flag-wake driver and its geometry pre-flight

Co-Authored-By: Claude Fable 5.1 <[email protected]>

crates/specialized/rtx-cfd/src/kernels/cuda/e3_cut.cu

@@ -307,3 +307,23 @@ extern "C" __global__ void e3_cut_add_p(E3Params g, E3Ptrs f, E3Cut m)
     if (!m.active[t]) return;
     f.p[t] += f.pp[m.owner[t]];
 }
+
+/* The prescribed interior faces of component c take the surface velocity
+ * (the host `impose` for a cut mask; `open` = the instantaneous kinds). */
+extern "C" __global__ void e3_cut_impose(E3Params g, E3Ptrs f, E3Cut m, int c)
+{
+    int t = blockIdx.x * blockDim.x + threadIdx.x;
+    int ni = c == 0 ? g.nx + 1 : g.nx;
+    int nj = c == 1 ? g.ny + 1 : g.ny;
+    int nk = c == 2 ? g.nz + 1 : g.nz;
+    if (t >= ni * nj * nk) return;
+    int i = t % ni; int j = (t / ni) % nj; int k = t / (ni * nj);
+    if (c == 0 && (i == 0 || i == g.nx)) return;
+    if (c == 1 && (j == 0 || j == g.ny)) return;
+    if (c == 2) { if (g.periodic_z) { if (k == g.nz) return; } else if (k == 0 || k == g.nz) return; }
+    const int* open = c == 0 ? m.open_u : (c == 1 ? m.open_v : m.open_w);
+    if (open[t]) return;
+    const double* ubt = c == 0 ? m.ub_u : (c == 1 ? m.ub_v : m.ub_w);
+    double* out = c == 0 ? f.u : (c == 1 ? f.v : f.w);
+    out[t] = ubt[t];
+}