embedded3 PERF-3 P3-1: the device predictor's interior fast path (bit 11, default ON; RTX_E3_PREDICT_FAST=0 = uniform path), the face update templated on the component + unrolled (392-byte local frame -> 64), exact-zero divisions/sqrt skipped on the fast path; all byte-identical (DFG 2D ny 122, slab flag ny 62, DFG 3D ny 62; device tests); predictor 22.4 -> 12.5 ms/step at DFG 3D ny 62

Co-Authored-By: Claude Fable 5.1 <[email protected]>
2026-09-19 20:29:57 -05:00
parent 77ecc54b24
commit b126915759
2 changed files with 63 additions and 24 deletions
@@ -306,7 +306,11 @@ impl DeviceStep {
                // bit 9: the advancing-wall friction closure (A3-i).
                + 512 * i32::from(self.solver.params.wall_advancing)
                // bit 10: no convective exchange with prescribed faces (A2).
-                + 1024 * i32::from(self.solver.params.exchange_convection_off),
+                + 1024 * i32::from(self.solver.params.exchange_convection_off)
+                // bit 11: the predictor's interior fast path (PERF-3 P3-1; bit-identical
+                // by construction and gated so; default ON, `RTX_E3_PREDICT_FAST=0` restores
+                // the uniform path).
+                + 2048 * i32::from(!std::env::var("RTX_E3_PREDICT_FAST").is_ok_and(|v| v == "0")),
            dx: g.dx,
            dy: g.dy,
            dz: g.dz,