PERF-2 P0-b: CG iterations per Poisson solve in the profile (ms per V-cycle) and sub-timers inside the patch regeneration (outline, hull + offset, ring projection, Winslow sweeps, respace, warm bookkeeping, mesh finalisation)

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.rs

@@ -1164,43 +1164,58 @@ impl OversetFluid {
     /// setup / iterate split.
     pub fn profile_line(&self) -> Option<String> {
         let t = self.solver.timers()?;
-        let (ps, pi, pc) = self.solver.background().poisson_profile();
+        let (ps, pi, pc, pk) = self.solver.background().poisson_profile();
+        let rg = rtx_cfd::mesh::patch_gen::regen_profile();
         let s = |ns: u64| ns as f64 * 1e-9;
         let adv = s(t.advance_ns).max(1e-300);
         let pct = |ns: u64| 100.0 * s(ns) / adv;
-        Some(format!(
-            "  advance split over {} steps ({} rounds), {:.0} s: overlap build {:.0} s ({:.1}%), predictor bg {:.0} s ({:.1}%), predictor patch {:.0} s ({:.1}%), bg Poisson {:.0} s ({:.1}%) [setup {:.0} s, iterate {:.0} s, {} solves], patch BiCGSTAB {:.0} s ({:.1}%), round exchange {:.0} s ({:.1}%), apply {:.0} s ({:.1}%), end exchange {:.0} s ({:.1}%), other {:.0} s",
-            t.steps,
-            t.rounds,
-            adv,
-            s(t.overlap_build_ns),
-            pct(t.overlap_build_ns),
-            s(t.predictor_bg_ns),
-            pct(t.predictor_bg_ns),
-            s(t.predictor_patch_ns),
-            pct(t.predictor_patch_ns),
-            s(t.bg_solve_ns),
-            pct(t.bg_solve_ns),
-            s(ps),
-            s(pi),
-            pc,
-            s(t.patch_solve_ns),
-            pct(t.patch_solve_ns),
-            s(t.round_exchange_ns),
-            pct(t.round_exchange_ns),
-            s(t.apply_ns),
-            pct(t.apply_ns),
-            s(t.end_exchange_ns),
-            pct(t.end_exchange_ns),
-            adv - s(t.overlap_build_ns
-                + t.predictor_bg_ns
-                + t.predictor_patch_ns
-                + t.bg_solve_ns
-                + t.patch_solve_ns
-                + t.round_exchange_ns
-                + t.apply_ns
-                + t.end_exchange_ns),
-        ))
+        Some(
+            format!(
+                "  advance split over {} steps ({} rounds), {:.0} s: overlap build {:.0} s ({:.1}%), predictor bg {:.0} s ({:.1}%), predictor patch {:.0} s ({:.1}%), bg Poisson {:.0} s ({:.1}%) [setup {:.0} s, iterate {:.0} s, {} solves, {} CG iterations, {:.2} ms per V-cycle], patch BiCGSTAB {:.0} s ({:.1}%), round exchange {:.0} s ({:.1}%), apply {:.0} s ({:.1}%), end exchange {:.0} s ({:.1}%), other {:.0} s",
+                t.steps,
+                t.rounds,
+                adv,
+                s(t.overlap_build_ns),
+                pct(t.overlap_build_ns),
+                s(t.predictor_bg_ns),
+                pct(t.predictor_bg_ns),
+                s(t.predictor_patch_ns),
+                pct(t.predictor_patch_ns),
+                s(t.bg_solve_ns),
+                pct(t.bg_solve_ns),
+                s(ps),
+                s(pi),
+                pc,
+                pk,
+                1e3 * s(pi) / (pk + pc).max(1) as f64,
+                s(t.patch_solve_ns),
+                pct(t.patch_solve_ns),
+                s(t.round_exchange_ns),
+                pct(t.round_exchange_ns),
+                s(t.apply_ns),
+                pct(t.apply_ns),
+                s(t.end_exchange_ns),
+                pct(t.end_exchange_ns),
+                adv - s(t.overlap_build_ns
+                    + t.predictor_bg_ns
+                    + t.predictor_patch_ns
+                    + t.bg_solve_ns
+                    + t.patch_solve_ns
+                    + t.round_exchange_ns
+                    + t.apply_ns
+                    + t.end_exchange_ns),
+            ) + &format!(
+                "\n  regeneration split ({} builds): outline {:.0} s, hull + offset {:.0} s, ring projection {:.0} s, Winslow sweeps {:.0} s, respace {:.0} s, warm bookkeeping {:.0} s, mesh finalisation {:.0} s",
+                rg[7],
+                s(rg[0]),
+                s(rg[1]),
+                s(rg[2]),
+                s(rg[3]),
+                s(rg[4]),
+                s(rg[5]),
+                s(rg[6])
+            ),
+        )
     }
 
     pub fn time(&self) -> f64 {