perf(tunnels): memoise the room mouth -- room ownership now cheaper than before the change

The room-wins rule cost +55% static / +36% moving worker CPU because every sample re-answered the room-mouth question. A bounded exact memo around BuildRoomMouth (4096 sets x 4 ways, 7.49 MiB, keyed on every geometry-affecting room/tunnel input) removes that: hit rate 99.91% static / 99.89% moving, and VF_FindRoomFloorCrossingWorld (19.5%/17.1% inclusive), VF_FindRoomMouthBoundary (16.0%/14.0%) and VF_EvaluateRoomLandingDensity leave the sampler's top 30 entirely.

vs the tip DLL (bf9384a runtime), same staged owner map: worker sum 29.003 s vs 30.749 s static (-5.7%), 57.232 s vs 60.923 s moving (-6.1%), request p95 -3.3% / -8.4%. Geometry unchanged by construction: canonical hash 86A99075, 64/64 tile hashes, OBJ SHA 7E0D8109..C543, identical across a repeat run and with -voxel.CrtFma3=0. Walk Connected 42,870 / 17,413; tunnel-attributed solid inside rooms 0; mouth amplitude 1.4903 cm; TilePostReachProof and DetSinCos pass. The crossing test was not rerun (the memo returns the same values; its PASS stands at 248.43 m / 26.584 s).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ

Source/VoxelForge/Private/VoxelDensityOpStack.cpp

@@ -3747,6 +3747,8 @@ namespace
             FTunnelCoreWorldEvaluation LastTunnelCoreWorldEvaluation;
             FVector LastTunnelCoreEvaluationPosition = FVector::ZeroVector;
             bool bLastTunnelCoreWorldEvaluationValid = false;
+            bool bLastRoomOwnsBottom = false;
+            bool bLastRoomOwnsBottomValid = false;
 
             /** ÉTAPE C1 — les params de la strate avec l'op de CETTE salle appliqué par-dessus.
              *  Mémo par voxel : invalidé au début de chaque `Eval`, calculé au PREMIER modificateur
@@ -3800,6 +3802,7 @@ namespace
             S.ActiveCache = Sample.RoomCache;
             S.NearestRoom = Sample.NearestRoomIndex;
             S.bLocalParamsValid = false;
+            S.bLastRoomOwnsBottomValid = false;
         }
 
         bool TryGetLastTunnelCoreWorldEvaluation(
@@ -3827,7 +3830,8 @@ namespace
                 S.LastTunnelCoreWorldEvaluation = VoxelCaveMorphology::EvaluateTunnelCoreWorld(
                     Position.X, Position.Y, Position.Z, GetCache(), nullptr,
                     VoxelGenLOD::ShouldUseSpatialIndex(false),
-                    &S.LastWarpedPosition);
+                    &S.LastWarpedPosition,
+                    S.bLastRoomOwnsBottomValid ? &S.bLastRoomOwnsBottom : nullptr);
                 S.LastTunnelCoreEvaluationPosition = Position;
                 S.bLastTunnelCoreWorldEvaluationValid = true;
             }
@@ -3922,6 +3926,7 @@ namespace
             FState& S = State();
             S.LastWorldPosition = FVector(WorldX, WorldY, WorldZ);
             S.bLastTunnelCoreWorldEvaluationValid = false;
+            S.bLastRoomOwnsBottomValid = false;
 
             // ⚠️ REMIS À -1 INCONDITIONNELLEMENT, ce que l'original ne fait pas : chez lui
             // `NearestRoomIdx` est un `thread_local` qui, quand `RoomDensity <= 0`, garde la valeur
@@ -4184,7 +4189,9 @@ namespace
             const FVector WorldTunnelPosition(WorldX, WorldY, WorldZ);
             float CaveSDF = VoxelCaveMorphology::EvaluateSDFCached(
                 WarpedX, WarpedY, WarpedZ, GetCache(), P.SDFBlendRadius,
-                &S.NearestRoom, bUseSpatialIndex, &WorldTunnelPosition);
+                &S.NearestRoom, bUseSpatialIndex, &WorldTunnelPosition,
+                &S.bLastRoomOwnsBottom);
+            S.bLastRoomOwnsBottomValid = true;
 
             //---------------------------------------------------------------
             // PITS & CHEMINÉES — coordonnées RÉELLES, SmoothMin dans le même canal SDF
@@ -4275,7 +4282,8 @@ namespace
             {
                 S.LastTunnelCoreWorldEvaluation = VoxelCaveMorphology::EvaluateTunnelCoreWorld(
                     WorldX, WorldY, WorldZ, GetCache(), nullptr,
-                    bUseSpatialIndex, &S.LastWarpedPosition);
+                    bUseSpatialIndex, &S.LastWarpedPosition,
+                    &S.bLastRoomOwnsBottom);
                 S.LastTunnelCoreEvaluationPosition = S.LastWorldPosition;
                 S.bLastTunnelCoreWorldEvaluationValid = true;
 

Source/VoxelForge/Private/VoxelGenerator.cpp

@@ -3294,6 +3294,10 @@ float UVoxelGenerator::GetDensityAt(float WorldX, float WorldY, float WorldZ) co
         float BlockTunnelCoreSweptFloorRadius = 0.0f;
         bool bUsedFusedEvaluator = false;
         FTunnelCoreWorldEvaluation FusedTunnelCore;
+        bool bRoomOwnsBottom = false;
+        bool bRoomOwnsBottomValid = false;
+        FVector NativeWarpedPosition = FVector::ZeroVector;
+        bool bNativeWarpedPositionValid = false;
 
         // Le seul point d'entrée de la pile dans le chemin de production. Elle rend la convention
         // MC (négatif = solide) comme les fonctions d'archétype, donc les disturbances et la couche
@@ -3322,7 +3326,9 @@ float UVoxelGenerator::GetDensityAt(float WorldX, float WorldY, float WorldZ) co
                     WorldX, WorldY, WorldZ, CP_Tunnel, CP_TunnelFP, LayoutVersion,
                     /*bApplyLegacyStructuralPosts=*/false,
                     &FusedTunnelCore,
-                    /*bCollectFusedDiagnostics=*/true);
+                    /*bCollectFusedDiagnostics=*/true,
+                    &bRoomOwnsBottom);
+                bRoomOwnsBottomValid = true;
             }
 
             const bool bIntegerLatticePoint =
@@ -3434,7 +3440,12 @@ float UVoxelGenerator::GetDensityAt(float WorldX, float WorldY, float WorldZ) co
             // Underwater shares tunnel rock (water table is a render-side overlay).
         Result = GetDensityWithParams(WorldX, WorldY, WorldZ, CP_Tunnel,
                                           CP_TunnelFP, LayoutVersion,
-                                          /*bApplyLegacyStructuralPosts=*/false);       break;
+                                          /*bApplyLegacyStructuralPosts=*/false,
+                                          nullptr, false, &bRoomOwnsBottom,
+                                          &NativeWarpedPosition);
+            bRoomOwnsBottomValid = true;
+            bNativeWarpedPositionValid = true;
+            break;
         }
         DensityCoreTimer.End();
 
@@ -3475,18 +3486,27 @@ float UVoxelGenerator::GetDensityAt(float WorldX, float WorldY, float WorldZ) co
         {
             VoxelDensityProfile::FScopedTimer ProfileTimer(
                 VoxelDensityProfile::EBucket::TunnelCorePosts);
-            FVector RoomQueryPosition(WorldX, WorldY, WorldZ);
+            FVector RoomQueryPosition;
+            if (bNativeWarpedPositionValid)
+            {
+                RoomQueryPosition = NativeWarpedPosition;
+            }
+            else
+            {
+                RoomQueryPosition = FVector(WorldX, WorldY, WorldZ);
                 if (CP_Tunnel.VerticalScale != 1.0f && CP_Tunnel.VerticalScale > 0.0f)
                 {
                     RoomQueryPosition.Z = WorldZ / CP_Tunnel.VerticalScale;
                 }
                 RoomQueryPosition = VoxelCaveMorphology::ApplyCaveWarp(
                     RoomQueryPosition, CP_Tunnel, static_cast<uint32>(Seed));
+            }
             PreDisturbanceTunnelCore = VoxelCaveMorphology::EvaluateTunnelCoreWorld(
                 WorldX, WorldY, WorldZ, ActiveTunnelCoreCache->Cache, nullptr,
                 VoxelGenLOD::ShouldUseSpatialIndex(
                     CP_UseOpStack ? bUsedFusedEvaluator : true),
-                &RoomQueryPosition);
+                &RoomQueryPosition,
+                bRoomOwnsBottomValid ? &bRoomOwnsBottom : nullptr);
             bHavePreDisturbanceTunnelCore = true;
         }
 
@@ -3679,19 +3699,28 @@ float UVoxelGenerator::GetDensityAt(float WorldX, float WorldY, float WorldZ) co
             {
                 VoxelDensityProfile::FScopedTimer ProfileTimer(
                     VoxelDensityProfile::EBucket::TunnelCorePosts);
-                FVector RoomQueryPosition(WorldX, WorldY, WorldZ);
+                FVector RoomQueryPosition;
+                if (bNativeWarpedPositionValid)
+                {
+                    RoomQueryPosition = NativeWarpedPosition;
+                }
+                else
+                {
+                    RoomQueryPosition = FVector(WorldX, WorldY, WorldZ);
                     if (CP_Tunnel.VerticalScale != 1.0f && CP_Tunnel.VerticalScale > 0.0f)
                     {
                         RoomQueryPosition.Z = WorldZ / CP_Tunnel.VerticalScale;
                     }
                     RoomQueryPosition = VoxelCaveMorphology::ApplyCaveWarp(
                         RoomQueryPosition, CP_Tunnel, static_cast<uint32>(Seed));
+                }
                 TunnelCore = VoxelCaveMorphology::EvaluateTunnelCoreWorld(
                     WorldX, WorldY, WorldZ, ActiveTunnelCoreCache->Cache,
                     nullptr,
                     VoxelGenLOD::ShouldUseSpatialIndex(
                         CP_UseOpStack ? bUsedFusedEvaluator : true),
-                    &RoomQueryPosition);
+                    &RoomQueryPosition,
+                    bRoomOwnsBottomValid ? &bRoomOwnsBottom : nullptr);
                 bHaveTunnelCore = true;
             }
             if (bHaveTunnelCore)
@@ -3828,7 +3857,9 @@ float UVoxelGenerator::GetDensityWithParams(float WorldX, float WorldY, float Wo
                                              uint32 ParamsFingerprint, uint32 LayoutVersion,
                                              bool bApplyLegacyStructuralPosts,
                                              FTunnelCoreWorldEvaluation* OutTunnelCore,
-                                             bool bCollectFusedDiagnostics) const
+                                             bool bCollectFusedDiagnostics,
+                                             bool* OutRoomOwnsBottom,
+                                             FVector* OutWarpedPosition) const
 {
     //=========================================================================
     // STRATE DENSITY FUNCTION (Morphology Pipeline)
@@ -3859,6 +3890,14 @@ float UVoxelGenerator::GetDensityWithParams(float WorldX, float WorldY, float Wo
     {
         *OutTunnelCore = FTunnelCoreWorldEvaluation();
     }
+    if (OutRoomOwnsBottom != nullptr)
+    {
+        *OutRoomOwnsBottom = false;
+    }
+    if (OutWarpedPosition != nullptr)
+    {
+        *OutWarpedPosition = FVector::ZeroVector;
+    }
     if (bCollectFusedDiagnostics && VoxelDensityProfile::AreCountersEnabled())
     {
         VoxelDensityProfile::AddCounter(
@@ -3930,6 +3969,10 @@ float UVoxelGenerator::GetDensityWithParams(float WorldX, float WorldY, float Wo
         WarpedY += WarpY * VOXEL_NOISE_SCALE * WS;
         WarpedZ += WarpZ * VOXEL_NOISE_SCALE * WS;
     }
+    if (OutWarpedPosition != nullptr)
+    {
+        *OutWarpedPosition = FVector(WarpedX, WarpedY, WarpedZ);
+    }
 
     //=========================================================================
     // STEP 4: SDF MORPHOLOGY (rooms + tunnels) — CACHED PER CHUNK
@@ -4187,7 +4230,8 @@ float UVoxelGenerator::GetDensityWithParams(float WorldX, float WorldY, float Wo
             SDFCache, Params.SDFBlendRadius,
             &NearestRoomIdx,
             bUseSpatialIndex,
-            &WorldTunnelPosition
+            &WorldTunnelPosition,
+            OutRoomOwnsBottom
         );
 
         // ── PIT & CHIMNEY SDF INTEGRATION ──
@@ -4307,7 +4351,7 @@ float UVoxelGenerator::GetDensityWithParams(float WorldX, float WorldY, float Wo
         const FVector RoomQueryPosition(WarpedX, WarpedY, WarpedZ);
         *OutTunnelCore = VoxelCaveMorphology::EvaluateTunnelCoreWorld(
             WorldX, WorldY, WorldZ, SDFCache, nullptr, bUseSpatialIndex,
-            &RoomQueryPosition);
+            &RoomQueryPosition, OutRoomOwnsBottom);
     }
 
     //=========================================================================

Source/VoxelForge/Public/VoxelCaveMorphology.h

@@ -1049,6 +1049,10 @@ namespace VoxelCaveMorphology
     //                                  supplied, rooms remain at the query position but the tunnel
     //                                  uses its authored world chain. Null preserves the legacy
     //                                  all-query behavior.
+    // @param OutRoomOwnsBottom      — optional out: true when the query position is inside any
+    //                                  cached room's raw authored shape. When the caller evaluates
+    //                                  the tunnel core for this same sample, pass this result back
+    //                                  to EvaluateTunnelCoreWorld to avoid a second room query.
     // (Room shape variety is baked into FCachedRoom by BuildChunkCache — no per-voxel roll.)
     // @return negative = inside cave, positive = solid rock
     VOXELFORGE_API float EvaluateSDFCached(
@@ -1057,7 +1061,8 @@ namespace VoxelCaveMorphology
         float SDFBlendRadius,
         int32* OutNearestRoomIdx = nullptr,
         bool bUseSpatialIndex = true,
-        const FVector* WorldTunnelPosition = nullptr
+        const FVector* WorldTunnelPosition = nullptr,
+        bool* OutRoomOwnsBottom = nullptr
     );
 
     // Evaluate the raw union of cached graph-tunnel capsules. This is intentionally separate from
@@ -1087,7 +1092,10 @@ namespace VoxelCaveMorphology
         // The room graph is evaluated in warped coordinates while the tunnel core is evaluated in
         // the authored world frame. Pass the former when available so room ownership is tested in
         // the same domain as the generic room field.
-        const FVector* RoomQueryPosition = nullptr
+        const FVector* RoomQueryPosition = nullptr,
+        // Optional result from EvaluateSDFCached for this exact sample. When supplied, it avoids
+        // repeating the room-shape containment query; null preserves standalone-call behavior.
+        const bool* CachedRoomOwnsBottom = nullptr
     );
 
     // Per-tile broad reach for the world-space core/floor post.  The bound includes the swept

Source/VoxelForge/Public/VoxelGenerator.h

@@ -240,7 +240,14 @@ public:
                                FTunnelCoreWorldEvaluation* OutTunnelCore = nullptr,
                                // Diagnostics are opt-in so ordinary/reference calls do not pay
                                // for fused-path counters.
-                               bool bCollectFusedDiagnostics = false) const;
+                               bool bCollectFusedDiagnostics = false,
+                               // Optional same-sample hand-off from EvaluateSDFCached. This is
+                               // only a room-shape ownership bit; it does not alter the scalar
+                               // result and avoids a duplicate room-index query in native paths.
+                               bool* OutRoomOwnsBottom = nullptr,
+                               // Optional same-sample warped position. Native GetDensityAt uses
+                               // this to feed the core post without applying the cave warp twice.
+                               FVector* OutWarpedPosition = nullptr) const;
 
     /**
      * Densité pour une strate Slab (FlatPlain / CrystalChamber).