-
Notifications
You must be signed in to change notification settings - Fork 2
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
1 parent
d8602fd
commit fec7239
Showing
11 changed files
with
539 additions
and
67 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,35 +1,179 @@ | ||
| #include "Common.ush" | ||
|
|
||
| float ComputeBoundsScreenRadiusSquared(const FVector4& BoundsOrigin, const float SphereRadius, const FVector4& ViewOrigin, const FMatrix& ProjMatrix) | ||
| #define GROUP_TILE_SIZE 64 | ||
| #define GROUP_TILE_SIZE_1 8 | ||
| #define DRAWCOMMAND_SIZE 5 | ||
|
|
||
| //[Input] | ||
| /* Layout | ||
| float3 ViewOriginPosition; | ||
| float4 ProjMatrixParameters; (ProjMatrix.M[0][0], ProjMatrix.M[1][1], ProjMatrix.M[2][3], ClusterSqureSizePerComponent) | ||
| float4 LODSettingsComponent; (LastLODScreenSizeSquared, LOD1ScreenSizeSquared, LODOnePlusDistributionScalarSquared, LastLODIndex) | ||
| */ | ||
| float4 LodCSParameters[3]; | ||
| Buffer<float4> ComponentsOriginAndRadiusSRV; | ||
|
|
||
| //[Output] | ||
| RWBuffer<uint> ClusterLodBufferUAV; | ||
|
|
||
| float ComputeBoundsScreenRadiusSquared(float4 OriginAndRadius) | ||
| { | ||
| // ignore perspective foreshortening for orthographic projections | ||
| const float DistSqr = FVector::DistSquared(BoundsOrigin, ViewOrigin) * ProjMatrix.M[2][3]; | ||
| // const float DistSqr = FVector::DistSquared(BoundsOrigin, ViewOrigin) * ProjMatrix.M[2][3]; | ||
| float3 ViewOriginPosition = LodCSParameters[0].xyz; | ||
| float3 ProjMatrixParameters = LodCSParameters[1].xyz; | ||
| const float DistSqr = dot(ViewOriginPosition - OriginAndRadius.xyz, ViewOriginPosition - OriginAndRadius.xyz) * ProjMatrixParameters.z; | ||
|
|
||
| // Get projection multiple accounting for view scaling. | ||
| const float ScreenMultiple = FMath::Max(0.5f * ProjMatrix.M[0][0], 0.5f * ProjMatrix.M[1][1]); | ||
| const float ScreenMultiple = max(0.5f * ProjMatrixParameters.x, 0.5f * ProjMatrixParameters.y); | ||
|
|
||
| // Calculate screen-space projected radius | ||
| return FMath::Square(ScreenMultiple * SphereRadius) / FMath::Max(1.0f, DistSqr); | ||
| return Square(ScreenMultiple * OriginAndRadius.w) / max(1.0f, DistSqr); | ||
| } | ||
|
|
||
| uint GetLODFromScreenSize(LODSettingsComponent LODSettings, float InScreenSizeSquared, float InViewLODScale, float& OutFractionalLOD) | ||
| uint GetLODFromScreenSize(float InScreenSizeSquared) | ||
| { | ||
| float ScreenSizeSquared = InScreenSizeSquared / InViewLODScale; | ||
|
|
||
| if (ScreenSizeSquared <= LODSettings.LastLODScreenSizeSquared) | ||
| //LODDistanceFactor Don't consider LODScale for now | ||
| //float ScreenSizeSquared = InScreenSizeSquared / InViewLODScale; | ||
| float ScreenSizeSquared = InScreenSizeSquared; | ||
| float4 LODSettings = LodCSParameters[2]; | ||
|
|
||
| uint CurLod = ScreenSizeSquared <= LODSettings.x ? (uint) LODSettings.w | ||
| : ScreenSizeSquared > LODSettings.y ? 0 | ||
| : (1 + log2(LODSettings.y / ScreenSizeSquared) / log2(LODSettings.z)); | ||
|
|
||
| return CurLod; | ||
| } | ||
|
|
||
| [numthreads(GROUP_TILE_SIZE, 1, 1)] | ||
| void ClusterComputeLODCS(uint DispatchThreadId : SV_DispatchThreadID) | ||
| { | ||
| float BoundsScreenRadiusSquared = ComputeBoundsScreenRadiusSquared(ComponentsOriginAndRadiusSRV[DispatchThreadId]); | ||
| uint Lod = GetLODFromScreenSize(BoundsScreenRadiusSquared); | ||
| uint ClusterSqureSizePerComponent = (uint)LodCSParameters[1].w; | ||
| uint StartClusterIndex = DispatchThreadId * ClusterSqureSizePerComponent; | ||
|
|
||
| LOOP | ||
| for (uint ClusterIndex = 0; ClusterIndex < ClusterSqureSizePerComponent; ++ClusterIndex) | ||
| { | ||
| ClusterLodBufferUAV[StartClusterIndex + ClusterIndex] = Lod; | ||
| } | ||
| } | ||
|
|
||
| //[Input] | ||
| struct ClusterInputData | ||
| { | ||
| float3 BoundCenter; | ||
|
|
||
| float3 BoundExtent; | ||
| }; | ||
|
|
||
| float4 ViewFrustumPermutedPlanes[8]; | ||
| uint TotalCluster; | ||
| uint ClusterSizePerComponent; | ||
| uint2 LandscapeComponentSize; | ||
| StructuredBuffer<ClusterInputData> ClusterInputData_SRV; | ||
| Buffer<uint> ClusterLodBufferSRV; | ||
|
|
||
| //[Output] | ||
| RWBuffer<uint> ClusterOutBufferUAV; | ||
| RWBuffer<uint> DrawCommandBufferUAV; | ||
|
|
||
| bool IntersectBox8Plane(in float3 Center, in float3 Extent, out bool InsideNearPlane) | ||
| { | ||
| float4 DistX_0 = Center.xxxx * ViewFrustumPermutedPlanes[0]; | ||
| float4 DistY_0 = Center.yyyy * ViewFrustumPermutedPlanes[1] + DistX_0; | ||
| float4 DistZ_0 = Center.zzzz * ViewFrustumPermutedPlanes[2] + DistY_0; | ||
| float4 Distance_0 = DistZ_0 - ViewFrustumPermutedPlanes[3]; | ||
|
|
||
| float4 PushX_0 = Extent.xxxx * abs(ViewFrustumPermutedPlanes[0]); | ||
| float4 PushY_0 = Extent.yyyy * abs(ViewFrustumPermutedPlanes[1]) + PushX_0; | ||
| float4 PushOut_0 = Extent.zzzz * abs(ViewFrustumPermutedPlanes[2]) + PushY_0; | ||
|
|
||
| if (any(Distance_0 > PushOut_0)) | ||
| { | ||
| OutFractionalLOD = LODSettings.LastLODIndex; | ||
| return LODSettings.LastLODIndex; | ||
| return false; | ||
| } | ||
| else if (ScreenSizeSquared > LODSettings.LOD1ScreenSizeSquared) | ||
|
|
||
| InsideNearPlane = Distance_0.x < -PushOut_0.x; | ||
|
|
||
| float4 DistX_1 = Center.xxxx * ViewFrustumPermutedPlanes[4]; | ||
| float4 DistY_1 = Center.yyyy * ViewFrustumPermutedPlanes[5] + DistX_1; | ||
| float4 DistZ_1 = Center.zzzz * ViewFrustumPermutedPlanes[6] + DistY_1; | ||
| float4 Distance_1 = DistZ_1 - ViewFrustumPermutedPlanes[7]; | ||
|
|
||
| float4 PushX_1 = Extent.xxxx * abs(ViewFrustumPermutedPlanes[4]); | ||
| float4 PushY_1 = Extent.yyyy * abs(ViewFrustumPermutedPlanes[5]) + PushX_1; | ||
| float4 PushOut_1 = Extent.zzzz * abs(ViewFrustumPermutedPlanes[6]) + PushY_1; | ||
|
|
||
| if (any(Distance_1 > PushOut_1)) | ||
| { | ||
| OutFractionalLOD = (LODSettings.LOD0ScreenSizeSquared - FMath::Min(ScreenSizeSquared, LODSettings.LOD0ScreenSizeSquared)) / (LODSettings.LOD0ScreenSizeSquared - LODSettings.LOD1ScreenSizeSquared); | ||
| return 0; | ||
| return false; | ||
| } | ||
| else | ||
|
|
||
| return true; | ||
| } | ||
|
|
||
| uint GetLinearIndexByClusterIndex(in int2 ClusterIndex) | ||
| { | ||
| uint2 ClampSize = clamp(ClusterIndex, int2(0, 0), int2(LandscapeComponentSize * ClusterSizePerComponent) - int2(1, 1)); | ||
| uint ClusterSqureSizePerComponent = ClusterSizePerComponent * ClusterSizePerComponent; | ||
| uint2 ClusterOffset = ClampSize & (ClusterSizePerComponent - 1); | ||
| uint2 ComponentOffset = ClampSize / ClusterSizePerComponent; | ||
|
|
||
| uint offset_1 = ComponentOffset.y * ClusterSqureSizePerComponent * LandscapeComponentSize.x + ComponentOffset.x * ClusterSqureSizePerComponent; | ||
| uint offset_2 = ClusterOffset.x + ClusterOffset.y * ClusterSizePerComponent; | ||
|
|
||
| return offset_1 + offset_2; | ||
| } | ||
|
|
||
| uint2 GetLinearIndexByClusterIndexBatch(in uint4 ClusterIndex) | ||
| { | ||
| uint4 ClampSize = clamp((int4) ClusterIndex, int4(0, 0, 0, 0), int4(LandscapeComponentSize.xyxy * ClusterSizePerComponent) - int4(1, 1, 1, 1)); | ||
| uint4 ClusterOffset = ClampSize & (ClusterSizePerComponent - 1); | ||
| uint4 ComponentOffset = ClampSize / ClusterSizePerComponent; | ||
| uint ClusterSqureSizePerComponent = ClusterSizePerComponent * ClusterSizePerComponent; | ||
| uint2 offset_1 = ComponentOffset.yw * ClusterSqureSizePerComponent * LandscapeComponentSize.x + ComponentOffset.xz * ClusterSqureSizePerComponent; | ||
| uint2 offset_2 = ClusterOffset.xz + ClusterOffset.yw * ClusterSizePerComponent; | ||
|
|
||
| return offset_1 + offset_2; | ||
| } | ||
|
|
||
| [numthreads(GROUP_TILE_SIZE_1, GROUP_TILE_SIZE_1, 1)] | ||
| void LandscapeGpuCullingCS(uint2 DispatchThreadId : SV_DispatchThreadID) | ||
| { | ||
| //保证一个Wrap访问的内存连续, Cache friend | ||
| uint CenterLinearIndex = GetLinearIndexByClusterIndex(DispatchThreadId); | ||
| ClusterInputData RenderData = ClusterInputData_SRV[CenterLinearIndex]; | ||
| uint ClusterLod = ClusterLodBufferSRV[CenterLinearIndex]; | ||
| bool InsideNearPlane; | ||
| bool bIsFrustumVisible = IntersectBox8Plane(RenderData.BoundCenter, RenderData.BoundExtent, InsideNearPlane); | ||
|
|
||
| BRANCH | ||
| if (!bIsFrustumVisible) | ||
| { | ||
| // No longer linear fraction, but worth the cache misses | ||
| OutFractionalLOD = 1 + FMath::LogX(LODSettings.LODOnePlusDistributionScalarSquared, LODSettings.LOD1ScreenSizeSquared / ScreenSizeSquared); | ||
| return (int8) OutFractionalLOD; | ||
| return; | ||
| } | ||
|
|
||
| //打包对应数据到输出数据中 | ||
| uint PackOutputData = 0; | ||
| uint DownLod = ClusterInputData_SRV[GetLinearIndexByClusterIndex(int2(0, 1) + (int2)DispatchThreadId)]; | ||
| uint LeftLod = ClusterInputData_SRV[GetLinearIndexByClusterIndex(int2(-1, 0) + (int2) DispatchThreadId)]; | ||
| uint TopLod = ClusterInputData_SRV[GetLinearIndexByClusterIndex(int2(0, -1) + (int2) DispatchThreadId)]; | ||
| uint RightLod = ClusterInputData_SRV[GetLinearIndexByClusterIndex(int2(1, 0) + (int2) DispatchThreadId)]; | ||
|
|
||
| PackOutputData = PackOutputData | (DispatchThreadId.x & 0xFF); | ||
| PackOutputData = PackOutputData | ((DispatchThreadId.y << 8) & 0xFF00); // There may be an error in the ARM register? | ||
| PackOutputData = PackOutputData | ((DownLod << 16) & 0x70000); | ||
| PackOutputData = PackOutputData | ((LeftLod << 19) & 0x380000); | ||
| PackOutputData = PackOutputData | ((TopLod << 22) & 0x1C0000); | ||
| PackOutputData = PackOutputData | ((RightLod << 25) & 0xE000000); | ||
| PackOutputData = PackOutputData | ((ClusterLod << 28) & 0x70000000); | ||
|
|
||
| //计算对应LOD级别的offset | ||
| uint IndirectDrawInstanceCountIndex = ClusterLod * DRAWCOMMAND_SIZE * 1; //InstanceCount | ||
| uint OutPutOffset = ClusterLod * TotalCluster; | ||
| uint CurrentLodCount = 0; | ||
| InterlockedAdd(DrawCommandBufferUAV[IndirectDrawInstanceCountIndex], 1, CurrentLodCount); | ||
| ClusterOutBufferUAV[CurrentLodCount + OutPutOffset] = PackOutputData; | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.