PERF-2 P0: intra-step profiler — StepTimers on the overset solver (RTX_PROFILE; overlap build / predictors / patch BiCGSTAB / background Poisson with its setup-vs-iterate split / round exchange / apply / end exchange), PoissonSolution carries setup_ns and iterate_ns, the harness times advance / restore / snapshot / load sampling / force / FEA predictor and prints the wall split of the coupled phase; no clock is read when profiling is off; the reclassified/step progress denominator fixed (was ×4)

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

crates/specialized/rtx-fsi/tests/fsi2_harness/overset_march.rs

@@ -377,6 +377,8 @@ pub fn run_march_overset(case: BenchmarkCase, config: &OversetMarchConfig) -> Ov
         f
     });
     let (t_fluid, t_structure) = (std::cell::Cell::new(0.0_f64), std::cell::Cell::new(0.0_f64));
+    // PERF-2 P0: the FEA predictor step (outside both buckets before).
+    let t_predictor = std::cell::Cell::new(0.0_f64);
     let prev_area = std::cell::Cell::new(fluid.borrow().shared.read().unwrap().area());
     let save_from: f64 = std::env::var("RTX_FSI2O_SAVE_FROM")
         .ok()
@@ -395,9 +397,11 @@ pub fn run_march_overset(case: BenchmarkCase, config: &OversetMarchConfig) -> Ov
             let v = extract_velocity(&flag_state);
             d_n.iter().zip(&v).map(|(d, v)| d + dt * v).collect()
         } else {
+            let ps = std::time::Instant::now();
             flag.borrow_mut()
                 .set_nodal_forces(&with_fict(&committed_nodal, &extract_accel(&flag_state)));
             let (predicted, _) = flag.borrow_mut().step(&flag_state).unwrap();
+            t_predictor.set(t_predictor.get() + ps.elapsed().as_secs_f64());
             extract(&predicted)
         };
         if robin_alpha > 0.0 {
@@ -582,7 +586,7 @@ pub fn run_march_overset(case: BenchmarkCase, config: &OversetMarchConfig) -> Ov
                 total_subiterations as f64 / (step + 1) as f64,
                 fl.rounds_total.get() as f64 / fl.correctors_total.get().max(1) as f64,
                 fl.reclassified_total.get() as f64
-                    / (rigid_steps + (step + 1) * cfg.subcycle * 4).max(1) as f64,
+                    / (rigid_steps + (step + 1) * cfg.subcycle).max(1) as f64,
                 fl.regen_seconds.get(),
                 t_fluid.get(),
                 phase_start.elapsed().as_secs_f64()
@@ -590,6 +594,37 @@ pub fn run_march_overset(case: BenchmarkCase, config: &OversetMarchConfig) -> Ov
         }
     }
     let fl = fluid.borrow();
+    // PERF-2 P0: the wall split of the coupled phase (every bucket the
+    // harness can see; `other` is what none of them caught).
+    {
+        let wall = phase_start.elapsed().as_secs_f64().max(1e-300);
+        let pct = |x: f64| 100.0 * x / wall;
+        let (adv, res, snap, samp, force, regen) = (
+            fl.t_advance.get(),
+            fl.t_restore.get(),
+            fl.t_snapshot.get(),
+            fl.t_sample.get(),
+            fl.t_force.get(),
+            fl.regen_seconds.get(),
+        );
+        let (st, pr) = (t_structure.get(), t_predictor.get());
+        let other = wall - (adv + res + snap + samp + force + regen + st + pr);
+        println!(
+            "  wall split over the coupled phase ({wall:.0} s): fluid advance {adv:.0} s ({:.1}%), patch regeneration {regen:.0} s ({:.1}%), restore {res:.0} s ({:.1}%), snapshot {snap:.0} s ({:.1}%), load sampling {samp:.0} s ({:.1}%), force {force:.0} s ({:.1}%), structure {st:.0} s ({:.1}%), FEA predictor {pr:.0} s ({:.1}%), other {other:.0} s ({:.1}%)",
+            pct(adv),
+            pct(regen),
+            pct(res),
+            pct(snap),
+            pct(samp),
+            pct(force),
+            pct(st),
+            pct(pr),
+            pct(other)
+        );
+        if let Some(line) = fl.profile_line() {
+            println!("{line}");
+        }
+    }
     let final_state_finite = flag_state.displacement.iter().all(|v| v.is_finite());
     let steps_done = times.len();
     OversetMarchResult {