wip(standalone): GPU step 27 -- slot-overflow fallback, -SkipCpuDetailBand (slower), -AffinityMask; full machine cached 0.93 s

Luna: normal completion path reads the raw vertex count before residency; an
over-cap count or failed readback falls back to CPU meshing. Max seen 18,153 of
19,456 vertices at (2,-9,-4,lod1) -- 1,303 headroom for this seed only.
Consumer map: 320 GPU near, 640 CPU near (skirted), 1,114 GPU far, 152 CPU far.
Exceptions are exported separately from roughness. Opt-in -SkipCpuDetailBand skips
CPU roughness for GPU-owned near visuals (lazy exact reconstruction on query).
Runner gains -AffinityMask (default 0xFF00) and -NormalPriority.

My runs outside the sandbox (Step27.exe, marker removed first):
  half machine (0xFF00, BelowNormal), interleaved:
    CPU 1.374 / 1.285 / 1.296 s
    GPU miss 1.330 s; cached 1.151 / 1.143 / 1.129 s (1,434 GPU tiles)
    GPU cached + -SkipCpuDetailBand 1.316 / 1.303 / 1.307 s -- SLOWER: detail apply
      331 -> 297 ms, but intervals+signs 112 -> 286 ms (the plan), shape 557 -> 692 ms
  full machine (0xFFFF, Normal), interleaved:
    CPU 1.004 / 1.015 / 0.974 s
    GPU cached 0.931 / 0.930 / 0.925 s -- first sub-second launches
Audits (Luna): hashes w1/w8, walk 5,642/5,642, near (skip) / far 0 sign/topology
diffs vs CPU and WARP, auto + cold survival pass, ten captures pass.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ

Standalone/Core/include/VoxelForgeCore/VoxelForgeCore.h

@@ -124,6 +124,13 @@ namespace VoxelForgeCore
     enum class DetailVariant : std::uint8_t { coherent, smooth, perVoxel };
     enum class TileClass : std::uint8_t { allSolid, allAir, mixed };
 
+    struct DetailBandTileRegion
+    {
+        Vec3i originVoxel{};
+        std::int32_t cells = 0;
+        std::int32_t step = 1;
+    };
+
     struct ShapeBuildRequest
     {
         Vec3i minVoxel{};
@@ -139,6 +146,12 @@ namespace VoxelForgeCore
         // World-wide walkability is a validation pass, not an input to the compiled field.
         // Keep it enabled for benchmarks/audits; the interactive viewer can defer it.
         bool validatePassageRoutes = true;
+        // Near GPU visual tiles evaluate owner roughness in the compute shader. Their exact
+        // sample lattice may omit CPU proxy-band work; CPU density queries still use the exact
+        // analytic field for those omitted samples. CPU visual tiles remain fully packed.
+        std::vector<DetailBandTileRegion> gpuVisualDetailTiles;
+        std::vector<DetailBandTileRegion> cpuVisualDetailTiles;
+        bool skipCpuDetailBandForGpuVisuals = false;
     };
 
     struct ShapeBuildStats
@@ -281,6 +294,8 @@ namespace VoxelForgeCore
         std::uint64_t protectedSealSampleCount = 0;
         std::uint64_t protectedThinSampleCount = 0;
         std::uint64_t protectedSolidOverlaySampleCount = 0;
+        std::uint64_t nearPackedCandidateSampleCount = 0;
+        std::uint64_t nearPackedSkippedSampleCount = 0;
     };
 
     struct EditBrush

Standalone/Core/src/ShapeWorldInternal.h

@@ -64,6 +64,9 @@ public:
     {
         uint64 Lines=0,Candidates=0,DetailEvaluations=0,Flips=0;
         uint64 LocalBandSkips=0;
+        uint64 GpuVisualCandidates=0,CpuVisualCandidates=0,GpuOnlyCandidates=0;
+        uint64 GpuVisualDetailEvaluations=0,GpuOnlyDetailEvaluations=0;
+        uint64 GpuOnlySkippedCandidateMasks=0;
         uint64 DistanceSamples=0,DistanceSampleNanoseconds=0;
         uint64 DetailSamples=0,DetailSampleNanoseconds=0;
         uint64 ReliefSampleNanoseconds=0,LayerLookupSampleNanoseconds=0;
@@ -277,7 +280,10 @@ public:
                std::string& Error,
                const VoxelForgeCore::JobScheduler* InScheduler=nullptr,
                bool bInRoundTunnels=false,bool bInProfileDetailBand=false,
-               bool bValidatePassageRoutes=true)
+               bool bValidatePassageRoutes=true,
+               const std::vector<VoxelForgeCore::DetailBandTileRegion>& InGpuVisualDetailTiles={},
+               const std::vector<VoxelForgeCore::DetailBandTileRegion>& InCpuVisualDetailTiles={},
+               bool bInSkipCpuDetailBandForGpuVisuals=false)
     {
         using Clock=std::chrono::steady_clock;
         const auto TotalStart=Clock::now();
@@ -285,6 +291,9 @@ public:
         Scheduler=InScheduler;RoundTunnelsEnabled=bInRoundTunnels;
         bAnalyticOnly=false;
         bProfileDetailBand=bInProfileDetailBand;
+        bSkipCpuDetailBandForGpuVisuals=bInSkipCpuDetailBandForGpuVisuals;
+        if(bSkipCpuDetailBandForGpuVisuals&&InGpuVisualDetailTiles.empty())
+        {Error="GPU visual detail-band skipping requires at least one GPU near tile region";return false;}
         for(int32 A=0;A<3;++A)
         {
             Min[A]=(&MinVoxels.X)[A]-2;
@@ -322,6 +331,11 @@ public:
         BuildPackedRockIntervals(Stats.RockIntervalLines);
         BuildCandidateBands(Stats.CandidateAxisLines,Stats.CandidateMergeLines,
             Stats.CandidateMergeSeconds);
+        if(!BuildGpuVisualDetailNodeSets(InGpuVisualDetailTiles,InCpuVisualDetailTiles,Error))
+            return false;
+        GpuVisualProxyValidWords.clear();
+        if(bSkipCpuDetailBandForGpuVisuals)
+            GpuVisualProxyValidWords.assign(CandidateWords.size(),0u);
         Stats.IntervalSeconds=std::chrono::duration<double>(Clock::now()-Stage).count();
 
         Stage=Clock::now();
@@ -763,8 +777,17 @@ public:
                 const uint64 Mask=uint64{1}<<(Zi&63u);
                 if((CandidateWords[Word]&Mask)==0)
                     Out.nearPackedProxy[SampleIndex]=0u;
+                else if(IsSkippedGpuVisualDetailNode(WX,WY,WZ)
+                    &&(GpuVisualProxyValidWords[Word]&Mask)==0)
+                {
+                    // UINT_MAX asks ApplyNearFieldDetailCS to use its own exact
+                    // GPU roughness result instead of a CPU packed proxy.
+                    Out.nearPackedProxy[SampleIndex]=std::numeric_limits<uint32>::max();
+                    ++Out.nearPackedSkippedSampleCount;
+                }
                 else
                 {
+                    ++Out.nearPackedCandidateSampleCount;
                     const size_t Node=Zi*static_cast<size_t>(Size[1])*Size[0]+Column;
                     const int32 Q=std::abs(static_cast<int32>(SurfaceDistanceQ[Node]));
                     const uint32 Magnitude=static_cast<uint32>(std::clamp(Q,1,32767));
@@ -910,6 +933,8 @@ public:
     {
         int32 IX=0,IY=0,IZ=0;
         const bool bInteger=NearInteger(X,IX)&&NearInteger(Y,IY)&&NearInteger(Z,IZ);
+        if(bInteger&&IsSkippedGpuVisualDetailNode(IX,IY,IZ))
+            return GpuSkippedDetailDensityAt(IX,IY,IZ,X,Y,Z);
         const bool bPacked=!bAnalyticOnly&&bInteger&&Contains(IX,IY,IZ);
         if(bPacked)
             return PackedDensityAt(IX,IY,IZ,IsRock(IX,IY,IZ),X,Y,Z);
@@ -918,6 +943,9 @@ public:
 
     float DensityAtLattice(int32 X,int32 Y,int32 Z) const
     {
+        if(IsSkippedGpuVisualDetailNode(X,Y,Z))
+            return GpuSkippedDetailDensityAt(X,Y,Z,static_cast<float>(X),
+                static_cast<float>(Y),static_cast<float>(Z));
         const bool bPacked=!bAnalyticOnly&&Contains(X,Y,Z);
         if(bPacked)
             return PackedDensityAt(X,Y,Z,IsRock(X,Y,Z),static_cast<float>(X),
@@ -931,6 +959,39 @@ public:
         return AnalyticDensityAt(X,Y,Z,false);
     }
 
+    float GpuSkippedDetailDensityAt(int32 X,int32 Y,int32 Z,
+                                    float FX,float FY,float FZ) const
+    {
+        const size_t Column=static_cast<size_t>(Y-Min[1])*Size[0]
+            +static_cast<size_t>(X-Min[0]);
+        const size_t Zi=static_cast<size_t>(Z-Min[2]);
+        const size_t Word=Column*WordsPerZ+(Zi>>6);
+        const uint64 Mask=uint64{1}<<(Zi&63u);
+        if(!GpuVisualProxyValidWords.empty()&&(GpuVisualProxyValidWords[Word]&Mask))
+            return PackedDensityAt(X,Y,Z,IsRock(X,Y,Z),FX,FY,FZ);
+
+        if(CandidateWords[Word]&Mask)
+        {
+            // The GPU-only compile omitted this packed cache entry. Reconstruct the
+            // CPU reference's un-clamped detail density and the same signed Q8.8
+            // proxy when an exceptional CPU mesh or field query needs the sample.
+            const float ExactDensity=AnalyticDensityAt(FX,FY,FZ,false,
+                nullptr,nullptr,nullptr,false);
+            int32 Q=static_cast<int32>(std::lround(ExactDensity*256.0f));
+            Q=std::clamp(Q,-32767,32767);
+            if(Q==0)Q=ExactDensity<0.0f?-1:1;
+            float Density=static_cast<float>(Q)*(1.0f/256.0f);
+            const bool bRock=ExactDensity<=0.0f;
+            if((Density<0.0f)!=bRock)
+                Density=(bRock?-1.0f:1.0f)*std::max(std::abs(Density),1.0f/256.0f);
+            return ApplyEditBrushes(Density,FX,FY,FZ);
+        }
+
+        const bool bRock=((RockWords[Word]&Mask)!=0);
+        constexpr float ProxyBandEdge=DetailSupremum+0.125f;
+        return ApplyEditBrushes(bRock?-ProxyBandEdge:ProxyBandEdge,FX,FY,FZ);
+    }
+
     VoxelForgeCore::DetailFieldSample DetailSampleAt(float X,float Y,float Z) const
     {
         VoxelForgeCore::DetailFieldSample Sample;
@@ -957,6 +1018,18 @@ public:
                 OutDensity[Lane]=DensityAtLattice(XStart+Lane*XStep,Y,Z);
             return;
         }
+        if(bSkipCpuDetailBandForGpuVisuals)
+        {
+            bool bContainsSkippedNode=false;
+            for(int32 Lane=0;Lane<Count;++Lane)
+                bContainsSkippedNode|=IsSkippedGpuVisualDetailNode(XStart+Lane*XStep,Y,Z);
+            if(bContainsSkippedNode)
+            {
+                for(int32 Lane=0;Lane<Count;++Lane)
+                    OutDensity[Lane]=DensityAtLattice(XStart+Lane*XStep,Y,Z);
+                return;
+            }
+        }
         bool bFast=EditBrushes.empty()&&Y>=Min[1]&&Y<=Max[1]
             &&Z>=Min[2]&&Z<=Max[2]
             &&XStart>=Min[0]
@@ -1028,7 +1101,8 @@ private:
     float AnalyticDensityAt(float X,float Y,float Z,bool bClampNearZero=true,
                             float* OutDetailDisplacement=nullptr,
                             float* OutCaveInfluence=nullptr,
-                            float* OutFloorInfluence=nullptr) const
+                            float* OutFloorInfluence=nullptr,
+                            bool bApplyEdits=true) const
     {
         if(OutDetailDisplacement)*OutDetailDisplacement=0.0f;
         if(OutCaveInfluence)*OutCaveInfluence=0.0f;
@@ -1045,7 +1119,7 @@ private:
             if((Density<0.0f)!=bRock||(bClampNearZero&&std::abs(Density)<0.05f))
                 Density=(bRock?-1.0f:1.0f)*std::max(std::abs(Density),
                     bClampNearZero?0.05f:1.0e-7f);
-            return ApplyEditBrushes(Density,X,Y,Z);
+            return bApplyEdits?ApplyEditBrushes(Density,X,Y,Z):Density;
         }
         ESurfaceFamily SurfaceFamily=ESurfaceFamily::Strata;float SurfaceSDF=100000.0f;
         bool bFloorSurface=false,bCriticalAir=false;float CaveInfluence=0.0f,FloorInfluence=0.0f;
@@ -1074,7 +1148,7 @@ private:
         if((Density<0.0f)!=bRock||(bClampNearZero&&std::abs(Density)<0.05f))
             Density=(bRock?-1.0f:1.0f)*std::max(std::abs(Density),
                 bClampNearZero?0.05f:1.0e-7f);
-        return ApplyEditBrushes(Density,X,Y,Z);
+        return bApplyEdits?ApplyEditBrushes(Density,X,Y,Z):Density;
     }
 
 public:
@@ -1111,6 +1185,9 @@ public:
 
     bool IsRock(int32 X,int32 Y,int32 Z) const
     {
+        if(IsSkippedGpuVisualDetailNode(X,Y,Z))
+            return AnalyticDensityAt(static_cast<float>(X),static_cast<float>(Y),
+                static_cast<float>(Z))<=0.0f;
         if(!Contains(X,Y,Z))return EvaluateFast(X,Y,Z);
         const size_t Line=static_cast<size_t>(Y-Min[1])*Size[0]
             +static_cast<size_t>(X-Min[0]);
@@ -3257,6 +3334,167 @@ private:
         MergeSeconds=std::max(0.0,MergeElapsed-MergeTiming.WallSeconds);
     }
 
+    bool BuildGpuVisualDetailNodeSets(
+        const std::vector<VoxelForgeCore::DetailBandTileRegion>& GpuTiles,
+        const std::vector<VoxelForgeCore::DetailBandTileRegion>& CpuTiles,
+        std::string& Error)
+    {
+        if(GpuTiles.empty()&&CpuTiles.empty())
+        {
+            GpuVisualDetailWords.clear();CpuVisualDetailWords.clear();
+            CpuVisualDetailGuardWords.clear();
+            if(bProfileDetailBand)
+            {
+                GpuVisualDetailWords.assign(CandidateWords.size(),0u);
+                CpuVisualDetailWords.assign(CandidateWords.size(),0u);
+                CpuVisualDetailGuardWords.assign(CandidateWords.size(),0u);
+            }
+            GpuVisualDetailTileCount=CpuVisualDetailTileCount=0;
+            GpuVisualDetailGridSamples=CpuVisualDetailGridSamples=0;
+            GpuVisualDetailNodeCount=CpuVisualDetailNodeCount=0;
+            CpuVisualDetailGuardNodeCount=0;
+            return true;
+        }
+        const auto BuildSet=[&](const std::vector<VoxelForgeCore::DetailBandTileRegion>& Tiles,
+            std::vector<uint64>& Nodes,uint64& GridSamples)->bool
+        {
+            Nodes.clear();GridSamples=0;
+            uint64 ReserveSamples=0;
+            for(const auto& Tile:Tiles)
+            {
+                if(Tile.cells<=0||Tile.cells>256||Tile.step<=0)
+                {Error="GPU visual detail-band tile region has invalid cells or step";return false;}
+                const uint64 Dim=static_cast<uint64>(Tile.cells)+3u;
+                const uint64 TileSamples=Dim*Dim*Dim;
+                if(TileSamples>std::numeric_limits<uint64>::max()-ReserveSamples)
+                {Error="GPU visual detail-band tile sample count overflowed 64 bits";return false;}
+                ReserveSamples+=TileSamples;
+            }
+            GridSamples=ReserveSamples;
+            const auto IsCandidate=[&](int32 X,int32 Y,int32 Z)
+            {
+                if(X<Min[0]||X>Max[0]||Y<Min[1]||Y>Max[1]||Z<Min[2]||Z>Max[2])
+                    return false;
+                const size_t Column=static_cast<size_t>(Y-Min[1])*Size[0]
+                    +static_cast<size_t>(X-Min[0]);
+                const size_t Zi=static_cast<size_t>(Z-Min[2]);
+                return (CandidateWords[Column*WordsPerZ+(Zi>>6)]
+                    &(uint64{1}<<(Zi&63u)))!=0;
+            };
+            Nodes.assign(CandidateWords.size(),0u);
+            for(const auto& Tile:Tiles)
+            for(int32 Z=-1;Z<=Tile.cells+1;++Z)
+            for(int32 Y=-1;Y<=Tile.cells+1;++Y)
+            for(int32 X=-1;X<=Tile.cells+1;++X)
+            {
+                const int32 WX=Tile.originVoxel.x+X*Tile.step;
+                const int32 WY=Tile.originVoxel.y+Y*Tile.step;
+                const int32 WZ=Tile.originVoxel.z+Z*Tile.step;
+                if(!IsCandidate(WX,WY,WZ))continue;
+                const size_t Column=static_cast<size_t>(WY-Min[1])*Size[0]
+                    +static_cast<size_t>(WX-Min[0]);
+                const size_t Zi=static_cast<size_t>(WZ-Min[2]);
+                Nodes[Column*WordsPerZ+(Zi>>6)]|=uint64{1}<<(Zi&63u);
+            }
+            return true;
+        };
+        if(!BuildSet(GpuTiles,GpuVisualDetailWords,GpuVisualDetailGridSamples)
+            ||!BuildSet(CpuTiles,CpuVisualDetailWords,CpuVisualDetailGridSamples))return false;
+        GpuVisualDetailNodeCount=0;CpuVisualDetailNodeCount=0;
+        for(const uint64 Bits:GpuVisualDetailWords)GpuVisualDetailNodeCount+=std::popcount(Bits);
+        for(const uint64 Bits:CpuVisualDetailWords)CpuVisualDetailNodeCount+=std::popcount(Bits);
+        CpuVisualDetailGuardWords.assign(CandidateWords.size(),0u);
+        if(bSkipCpuDetailBandForGpuVisuals&&!CpuTiles.empty())
+        {
+            constexpr int32 AxisOffsets[6][3]={{-1,0,0},{1,0,0},{0,-1,0},
+                {0,1,0},{0,0,-1},{0,0,1}};
+            for(size_t Word=0;Word<CpuVisualDetailWords.size();++Word)
+            {
+                uint64 Bits=CpuVisualDetailWords[Word];
+                const size_t Column=Word/WordsPerZ;
+                const int32 X=Min[0]+static_cast<int32>(Column%Size[0]);
+                const int32 Y=Min[1]+static_cast<int32>(Column/Size[0]);
+                const size_t ZiBase=(Word%WordsPerZ)*64u;
+                while(Bits)
+                {
+                    const uint32 Bit=std::countr_zero(Bits);Bits&=Bits-1;
+                    const size_t Zi=ZiBase+Bit;
+                    if(Zi>=static_cast<size_t>(Size[2]))continue;
+                    const int32 Z=Min[2]+static_cast<int32>(Zi);
+                    for(const auto& Offset:AxisOffsets)
+                    {
+                        const int32 GX=X+Offset[0],GY=Y+Offset[1],GZ=Z+Offset[2];
+                        if(GX<Min[0]||GX>Max[0]||GY<Min[1]||GY>Max[1]
+                            ||GZ<Min[2]||GZ>Max[2])continue;
+                        const size_t GuardColumn=static_cast<size_t>(GY-Min[1])*Size[0]
+                            +static_cast<size_t>(GX-Min[0]);
+                        const size_t GuardZi=static_cast<size_t>(GZ-Min[2]);
+                        const size_t GuardWord=GuardColumn*WordsPerZ+(GuardZi>>6);
+                        const uint64 GuardMask=uint64{1}<<(GuardZi&63u);
+                        if(CandidateWords[GuardWord]&GuardMask)
+                            CpuVisualDetailGuardWords[GuardWord]|=GuardMask;
+                    }
+                }
+            }
+        }
+        CpuVisualDetailGuardNodeCount=0;
+        for(const uint64 Bits:CpuVisualDetailGuardWords)
+            CpuVisualDetailGuardNodeCount+=std::popcount(Bits);
+        GpuVisualDetailTileCount=GpuTiles.size();CpuVisualDetailTileCount=CpuTiles.size();
+        if(!GpuTiles.empty()||!CpuTiles.empty())
+        {
+            std::printf("[GpuVisualDetailBandPlan] gpu_tiles=%llu gpu_sample_nodes=%llu gpu_unique_candidate_nodes=%llu cpu_near_tiles=%llu cpu_sample_nodes=%llu cpu_unique_candidate_nodes=%llu cpu_boundary_guard_candidate_nodes=%llu skip_enabled=%d\n",
+                static_cast<unsigned long long>(GpuVisualDetailTileCount),
+                static_cast<unsigned long long>(GpuVisualDetailGridSamples),
+                static_cast<unsigned long long>(GpuVisualDetailNodeCount),
+                static_cast<unsigned long long>(CpuVisualDetailTileCount),
+                static_cast<unsigned long long>(CpuVisualDetailGridSamples),
+                static_cast<unsigned long long>(CpuVisualDetailNodeCount),
+                static_cast<unsigned long long>(CpuVisualDetailGuardNodeCount),
+                bSkipCpuDetailBandForGpuVisuals?1:0);
+        }
+        return true;
+    }
+
+    bool IsGpuVisualDetailNode(int32 X,int32 Y,int32 Z) const
+    {
+        if(!Contains(X,Y,Z)||GpuVisualDetailWords.empty())return false;
+        const size_t Column=static_cast<size_t>(Y-Min[1])*Size[0]
+            +static_cast<size_t>(X-Min[0]);
+        const size_t Zi=static_cast<size_t>(Z-Min[2]);
+        return (GpuVisualDetailWords[Column*WordsPerZ+(Zi>>6)]
+            &(uint64{1}<<(Zi&63u)))!=0;
+    }
+
+    bool IsCpuVisualDetailNode(int32 X,int32 Y,int32 Z) const
+    {
+        if(!Contains(X,Y,Z)||CpuVisualDetailWords.empty())return false;
+        const size_t Column=static_cast<size_t>(Y-Min[1])*Size[0]
+            +static_cast<size_t>(X-Min[0]);
+        const size_t Zi=static_cast<size_t>(Z-Min[2]);
+        return (CpuVisualDetailWords[Column*WordsPerZ+(Zi>>6)]
+            &(uint64{1}<<(Zi&63u)))!=0;
+    }
+
+    bool IsSkippedGpuVisualDetailWord(size_t Word,uint64 Mask) const
+    {
+        return bSkipCpuDetailBandForGpuVisuals
+            &&(GpuVisualDetailWords[Word]&Mask)!=0
+            &&(CpuVisualDetailWords[Word]&Mask)==0
+            &&(CpuVisualDetailGuardWords[Word]&Mask)==0;
+    }
+
+    bool IsSkippedGpuVisualDetailNode(int32 X,int32 Y,int32 Z) const
+    {
+        if(!bSkipCpuDetailBandForGpuVisuals||!Contains(X,Y,Z)
+            ||GpuVisualDetailWords.empty())return false;
+        const size_t Column=static_cast<size_t>(Y-Min[1])*Size[0]
+            +static_cast<size_t>(X-Min[0]);
+        const size_t Zi=static_cast<size_t>(Z-Min[2]);
+        return IsSkippedGpuVisualDetailWord(Column*WordsPerZ+(Zi>>6),
+            uint64{1}<<(Zi&63u));
+    }
+
     void ApplyDetailBand(FParallelTiming& Timing)
     {
         using Clock=std::chrono::steady_clock;
@@ -3326,6 +3564,27 @@ private:
                         const FEffectiveRockSample& Sample=Samples[Lane];
                         const float SDF=Sample.Distance;
                         if(std::abs(SDF)>DetailSupremum+0.125f)continue;
+                        const size_t VisualWordIndex=WordBase+(Zi>>6);
+                        const uint64 VisualMask=uint64{1}<<(Zi&63u);
+                        const bool bGpuVisualNode=(bProfile||bSkipCpuDetailBandForGpuVisuals)
+                            &&(GpuVisualDetailWords[VisualWordIndex]&VisualMask)!=0;
+                        const bool bCpuVisualNode=(bProfile||bSkipCpuDetailBandForGpuVisuals)
+                            &&(CpuVisualDetailWords[VisualWordIndex]&VisualMask)!=0;
+                        const bool bSkipDetail=bSkipCpuDetailBandForGpuVisuals&&bGpuVisualNode
+                            &&!bCpuVisualNode
+                            &&(CpuVisualDetailGuardWords[VisualWordIndex]&VisualMask)==0;
+                        if(bProfile)
+                        {
+                            if(bGpuVisualNode)++Profile.GpuVisualCandidates;
+                            if(bCpuVisualNode)++Profile.CpuVisualCandidates;
+                            if(bGpuVisualNode&&!bCpuVisualNode)++Profile.GpuOnlyCandidates;
+                        }
+                        if(bSkipDetail)
+                        {
+                            ValidBits|=uint64{1}<<(Zi&63u);
+                            if(bProfile)++Profile.GpuOnlySkippedCandidateMasks;
+                            continue;
+                        }
                         const float SurfaceSDF=Sample.SurfaceSDF;
                         const float CaveInfluence=Sample.CaveInfluence;
                         const float FloorInfluence=Sample.FloorInfluence;
@@ -3354,6 +3613,9 @@ private:
                         if(DetailFade>0.0f)
                         {
                             if(bProfile)++Profile.DetailEvaluations;
+                            if(bProfile&&bGpuVisualNode)++Profile.GpuVisualDetailEvaluations;
+                            if(bProfile&&bGpuVisualNode&&!bCpuVisualNode)
+                                ++Profile.GpuOnlyDetailEvaluations;
                             Displaced+=DetailNoiseOnLine(X,Y,Z,Sample.SurfaceSDF,
                                 CaveInfluence,FloorInfluence,PassageInfluence,NoiseLine,bTimeDetail?&Profile:nullptr,
                                 DetailField,bDetailFieldKnown,bMacroKnown,MacroValue)*DetailFade;
@@ -3418,6 +3680,8 @@ private:
             uint64 DetailSamples=0,DetailSampleNs=0,ReliefSampleNs=0,LocalBandSkips=0;
             uint64 LayerLookupSampleNs=0,OwnerSampleNs=0,OwnerFieldCalls=0,OwnerActiveCalls=0;
             uint64 OwnerCacheHits=0,OwnerCacheMisses=0;
+            uint64 GpuVisualCandidates=0,CpuVisualCandidates=0,GpuOnlyCandidates=0;
+            uint64 GpuVisualDetailCalls=0,GpuOnlyDetailCalls=0,GpuOnlySkippedMasks=0;
             for(int32 Worker=0;Worker<WorkerCount;++Worker)
             {
                 const FDetailWorkerProfile& P=WorkerProfile[static_cast<size_t>(Worker)];
@@ -3432,6 +3696,12 @@ private:
                 OwnerActiveCalls+=P.OwnerRoughnessActiveCalls;
                 OwnerCacheHits+=P.OwnerNoiseCacheHits;
                 OwnerCacheMisses+=P.OwnerNoiseCacheMisses;
+                GpuVisualCandidates+=P.GpuVisualCandidates;
+                CpuVisualCandidates+=P.CpuVisualCandidates;
+                GpuOnlyCandidates+=P.GpuOnlyCandidates;
+                GpuVisualDetailCalls+=P.GpuVisualDetailEvaluations;
+                GpuOnlyDetailCalls+=P.GpuOnlyDetailEvaluations;
+                GpuOnlySkippedMasks+=P.GpuOnlySkippedCandidateMasks;
             }
             const double DistanceAverageNs=DistanceSamples?static_cast<double>(DistanceSampleNs)/DistanceSamples:0.0;
             const double DetailAverageNs=DetailSamples?static_cast<double>(DetailSampleNs)/DetailSamples:0.0;
@@ -3456,6 +3726,14 @@ private:
                 static_cast<unsigned long long>(OwnerCacheHits),
                 static_cast<unsigned long long>(OwnerCacheMisses),
                 static_cast<unsigned long long>(LocalBandSkips));
+            std::printf("[GpuDetailConsumerProfile] gpu_near_proxy_nodes=%llu cpu_near_mesh_nodes=%llu gpu_only_nodes=%llu gpu_near_detail_noise_evaluations=%llu gpu_only_detail_noise_evaluations=%llu gpu_only_skipped_candidate_masks=%llu skip_enabled=%d\n",
+                static_cast<unsigned long long>(GpuVisualCandidates),
+                static_cast<unsigned long long>(CpuVisualCandidates),
+                static_cast<unsigned long long>(GpuOnlyCandidates),
+                static_cast<unsigned long long>(GpuVisualDetailCalls),
+                static_cast<unsigned long long>(GpuOnlyDetailCalls),
+                static_cast<unsigned long long>(GpuOnlySkippedMasks),
+                bSkipCpuDetailBandForGpuVisuals?1:0);
             for(int32 Worker=0;Worker<WorkerCount;++Worker)
             {
                 const FDetailWorkerProfile& P=WorkerProfile[static_cast<size_t>(Worker)];
@@ -3491,6 +3769,8 @@ private:
                     +static_cast<size_t>(X-Min[0]);
                 StoreSurfaceDistance(Node,Density);
                 CandidateWords[Word]|=Mask;
+                if(IsSkippedGpuVisualDetailNode(X,Y,Z))
+                    GpuVisualProxyValidWords[Word]|=Mask;
                 ++AddedThisPass;
 
                 const bool bExactRock=Density<0.0f;
@@ -4541,6 +4821,14 @@ private:
     std::unique_ptr<int16[]> SurfaceDistanceQ;
     std::array<std::vector<uint64>,3> CandidateAxisWords;
     std::vector<uint64> CandidateWords;
+    bool bSkipCpuDetailBandForGpuVisuals=false;
+    std::vector<uint64> GpuVisualDetailWords,CpuVisualDetailWords;
+    std::vector<uint64> CpuVisualDetailGuardWords;
+    uint64 GpuVisualDetailTileCount=0,CpuVisualDetailTileCount=0;
+    uint64 GpuVisualDetailGridSamples=0,CpuVisualDetailGridSamples=0;
+    uint64 GpuVisualDetailNodeCount=0,CpuVisualDetailNodeCount=0;
+    uint64 CpuVisualDetailGuardNodeCount=0;
+    std::vector<uint64> GpuVisualProxyValidWords;
     FBuildStats Stats;
     bool bAnalyticOnly=false;
 };

Standalone/Core/src/VoxelForgeCore.cpp

@@ -306,7 +306,8 @@ namespace VoxelForgeCore
             ToNative(request.clipMinVoxel),ToNative(request.clipMaxVoxel),
             static_cast<int32>(std::clamp<std::uint32_t>(request.workerLimit,1,8)),ToNative(request.detail),
             request.includeFarWorld,error,&scheduler,request.roundTunnels,request.profileDetailBand,
-            request.validatePassageRoutes);
+            request.validatePassageRoutes,request.gpuVisualDetailTiles,
+            request.cpuVisualDetailTiles,request.skipCpuDetailBandForGpuVisuals);
         if(ok)
         {
             impl_->shape=std::move(shape);