GoaLitiuM/FlaxEngine
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Refactor Global Surface Atlas objects buffer into separate tiles buffer to reduce memory usage and increase cache hit ratio

Browse Source
This commit is contained in:
Wojciech Figat
2022-04-22 14:24:06 +02:00
parent 9d205cbb7d
commit 77dcc9b7a3
5 changed files with 79 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

BIN
Content/Shaders/GlobalSurfaceAtlas.flax (Stored with Git LFS)
View File

Binary file not shown.

62
Source/Engine/Renderer/GlobalSurfaceAtlasPass.cpp
View File

@@ -18,7 +18,8 @@
#include "Engine/Utilities/RectPack.h" #include "Engine/Utilities/RectPack.h"
// This must match HLSL // This must match HLSL
#define GLOBAL_SURFACE_ATLAS_OBJECT_SIZE (5 + 6 * 5) // Amount of float4s per-object #define GLOBAL_SURFACE_ATLAS_OBJECT_BUFFER_STRIDE 6 // Amount of float4s per-object
#define GLOBAL_SURFACE_ATLAS_TILE_BUFFER_STRIDE 5 // Amount of float4s per-tile
#define GLOBAL_SURFACE_ATLAS_TILE_PADDING 1 // 1px padding to prevent color bleeding between tiles #define GLOBAL_SURFACE_ATLAS_TILE_PADDING 1 // 1px padding to prevent color bleeding between tiles
#define GLOBAL_SURFACE_ATLAS_TILE_PROJ_PLANE_OFFSET 0.1f // Small offset to prevent clipping with the closest triangles (shifts near and far planes) #define GLOBAL_SURFACE_ATLAS_TILE_PROJ_PLANE_OFFSET 0.1f // Small offset to prevent clipping with the closest triangles (shifts near and far planes)
#define GLOBAL_SURFACE_ATLAS_DEBUG_FORCE_REDRAW_TILES 0 // Forces to redraw all object tiles every frame #define GLOBAL_SURFACE_ATLAS_DEBUG_FORCE_REDRAW_TILES 0 // Forces to redraw all object tiles every frame
@@ -55,6 +56,7 @@ struct GlobalSurfaceAtlasTile : RectPack<GlobalSurfaceAtlasTile, uint16>
Vector3 ViewPosition; Vector3 ViewPosition;
Vector3 ViewBoundsSize; Vector3 ViewBoundsSize;
Matrix ViewMatrix; Matrix ViewMatrix;
uint16 TileIndex;
GlobalSurfaceAtlasTile(uint16 x, uint16 y, uint16 width, uint16 height) GlobalSurfaceAtlasTile(uint16 x, uint16 y, uint16 width, uint16 height)
: RectPack<GlobalSurfaceAtlasTile, uint16>(x, y, width, height) : RectPack<GlobalSurfaceAtlasTile, uint16>(x, y, width, height)
@@ -118,13 +120,16 @@ public:
GPUTexture* AtlasGBuffer2 = nullptr; GPUTexture* AtlasGBuffer2 = nullptr;
GPUTexture* AtlasDirectLight = nullptr; GPUTexture* AtlasDirectLight = nullptr;
DynamicTypedBuffer ObjectsBuffer; DynamicTypedBuffer ObjectsBuffer;
DynamicTypedBuffer TilesBuffer;
uint32 ObjectIndexCounter; uint32 ObjectIndexCounter;
uint16 TileIndexCounter;
GlobalSurfaceAtlasPass::BindingData Result; GlobalSurfaceAtlasPass::BindingData Result;
GlobalSurfaceAtlasTile* AtlasTiles = nullptr; // TODO: optimize with a single allocation for atlas tiles GlobalSurfaceAtlasTile* AtlasTiles = nullptr; // TODO: optimize with a single allocation for atlas tiles
Dictionary<Actor*, GlobalSurfaceAtlasObject> Objects; Dictionary<Actor*, GlobalSurfaceAtlasObject> Objects;
GlobalSurfaceAtlasCustomBuffer() GlobalSurfaceAtlasCustomBuffer()
: ObjectsBuffer(256 * GLOBAL_SURFACE_ATLAS_OBJECT_SIZE, PixelFormat::R32G32B32A32_Float, false, TEXT("GlobalSurfaceAtlas.ObjectsBuffer")) : ObjectsBuffer(256 * GLOBAL_SURFACE_ATLAS_OBJECT_BUFFER_STRIDE, PixelFormat::R32G32B32A32_Float, false, TEXT("GlobalSurfaceAtlas.ObjectsBuffer"))
, TilesBuffer(256 * GLOBAL_SURFACE_ATLAS_TILE_BUFFER_STRIDE * 3 / 4, PixelFormat::R32G32B32A32_Float, false, TEXT("GlobalSurfaceAtlas.TilesBuffer"))
{ {
} }
@@ -133,6 +138,7 @@ public:
LastFrameAtlasDefragmentation = Engine::FrameCount; LastFrameAtlasDefragmentation = Engine::FrameCount;
SAFE_DELETE(AtlasTiles); SAFE_DELETE(AtlasTiles);
ObjectsBuffer.Clear(); ObjectsBuffer.Clear();
TilesBuffer.Clear();
Objects.Clear(); Objects.Clear();
} }
@@ -344,20 +350,28 @@ bool GlobalSurfaceAtlasPass::Render(RenderContext& renderContext, GPUContext* co
Half2 min(minPos * posToClipMul + posToClipAdd), max(maxPos * posToClipMul + posToClipAdd); \ Half2 min(minPos * posToClipMul + posToClipAdd), max(maxPos * posToClipMul + posToClipAdd); \
Vector2 minUV(0, 0), maxUV(1, 1); \ Vector2 minUV(0, 0), maxUV(1, 1); \
auto* quad = _vertexBuffer->WriteReserve<AtlasTileVertex>(6); \ auto* quad = _vertexBuffer->WriteReserve<AtlasTileVertex>(6); \
quad[0] = { { max }, { maxUV }, (uint16)object.Index, (uint16)tileIndex }; \ quad[0] = { { max }, { maxUV }, (uint16)object.Index, tile->TileIndex }; \
quad[1] = { { min.X, max.Y }, { minUV.X, maxUV.Y }, (uint16)object.Index, (uint16)tileIndex }; \ quad[1] = { { min.X, max.Y }, { minUV.X, maxUV.Y }, (uint16)object.Index, tile->TileIndex }; \
quad[2] = { { min }, { minUV }, (uint16)object.Index, (uint16)tileIndex }; \ quad[2] = { { min }, { minUV }, (uint16)object.Index, tile->TileIndex }; \
quad[3] = quad[2]; \ quad[3] = quad[2]; \
quad[4] = { { max.X, min.Y }, { maxUV.X, minUV.Y }, (uint16)object.Index, (uint16)tileIndex }; \ quad[4] = { { max.X, min.Y }, { maxUV.X, minUV.Y }, (uint16)object.Index, tile->TileIndex }; \
quad[5] = quad[0] quad[5] = quad[0]
#define VB_DRAW() \ #define VB_DRAW() \
_vertexBuffer->Flush(context); \ _vertexBuffer->Flush(context); \
auto vb = _vertexBuffer->GetBuffer(); \ auto vb = _vertexBuffer->GetBuffer(); \
context->BindVB(ToSpan(&vb, 1)); \ context->BindVB(ToSpan(&vb, 1)); \
context->DrawInstanced(_vertexBuffer->Data.Count() / sizeof(AtlasTileVertex), 1); context->DrawInstanced(_vertexBuffer->Data.Count() / sizeof(AtlasTileVertex), 1);
// Add objects into the atlas // Add objects into the atlas
surfaceAtlasData.ObjectsBuffer.Clear(); surfaceAtlasData.ObjectsBuffer.Clear();
surfaceAtlasData.TilesBuffer.Clear();
surfaceAtlasData.ObjectIndexCounter = 0; surfaceAtlasData.ObjectIndexCounter = 0;
{
// Tile at index 0 is invalid
surfaceAtlasData.TileIndexCounter = 1;
auto* tileData = surfaceAtlasData.TilesBuffer.WriteReserve<Vector4>(GLOBAL_SURFACE_ATLAS_TILE_BUFFER_STRIDE);
Platform::MemoryClear(tileData, sizeof(Vector4) * GLOBAL_SURFACE_ATLAS_TILE_BUFFER_STRIDE);
}
_dirtyObjectsBuffer.Clear(); _dirtyObjectsBuffer.Clear();
{ {
PROFILE_CPU_NAMED("Draw"); PROFILE_CPU_NAMED("Draw");
@@ -468,23 +482,22 @@ bool GlobalSurfaceAtlasPass::Render(RenderContext& renderContext, GPUContext* co
Matrix::Invert(object->Bounds.Transformation, worldToLocalBounds); Matrix::Invert(object->Bounds.Transformation, worldToLocalBounds);
// TODO: cache data for static objects to optimize CPU perf (move ObjectsBuffer into surfaceAtlasData) // TODO: cache data for static objects to optimize CPU perf (move ObjectsBuffer into surfaceAtlasData)
object->Index = surfaceAtlasData.ObjectIndexCounter++; object->Index = surfaceAtlasData.ObjectIndexCounter++;
auto* objectData = surfaceAtlasData.ObjectsBuffer.WriteReserve<Vector4>(GLOBAL_SURFACE_ATLAS_OBJECT_SIZE); auto* objectData = surfaceAtlasData.ObjectsBuffer.WriteReserve<Vector4>(GLOBAL_SURFACE_ATLAS_OBJECT_BUFFER_STRIDE);
objectData[0] = *(Vector4*)&e.Bounds; objectData[0] = *(Vector4*)&e.Bounds;
objectData[1] = Vector4(worldToLocalBounds.M11, worldToLocalBounds.M12, worldToLocalBounds.M13, worldToLocalBounds.M41); objectData[1] = Vector4(worldToLocalBounds.M11, worldToLocalBounds.M12, worldToLocalBounds.M13, worldToLocalBounds.M41);
objectData[2] = Vector4(worldToLocalBounds.M21, worldToLocalBounds.M22, worldToLocalBounds.M23, worldToLocalBounds.M42); objectData[2] = Vector4(worldToLocalBounds.M21, worldToLocalBounds.M22, worldToLocalBounds.M23, worldToLocalBounds.M42);
objectData[3] = Vector4(worldToLocalBounds.M31, worldToLocalBounds.M32, worldToLocalBounds.M33, worldToLocalBounds.M43); objectData[3] = Vector4(worldToLocalBounds.M31, worldToLocalBounds.M32, worldToLocalBounds.M33, worldToLocalBounds.M43);
objectData[4] = Vector4(object->Bounds.Extents, 0.0f); objectData[4] = Vector4(object->Bounds.Extents, 0.0f); // w unused
objectData[5] = Vector4::Zero; // w unused
auto tileIndices = reinterpret_cast<uint16*>(&objectData[5]); // xyz used for tile indices packed into uint16
// TODO: try to optimize memory footprint (eg. merge scale into extents and use rotation+offset but reconstruct rotation from two axes with sign) // TODO: try to optimize memory footprint (eg. merge scale into extents and use rotation+offset but reconstruct rotation from two axes with sign)
for (int32 tileIndex = 0; tileIndex < 6; tileIndex++) for (int32 tileIndex = 0; tileIndex < 6; tileIndex++)
{ {
auto* tile = object->Tiles[tileIndex]; auto* tile = object->Tiles[tileIndex];
const int32 tileStart = 5 + tileIndex * 5;
if (!tile) if (!tile)
{
// Disable tile
objectData[tileStart + 4] = Vector4::Zero;
continue; continue;
} tile->TileIndex = surfaceAtlasData.TileIndexCounter++;
tileIndices[tileIndex] = tile->TileIndex;
// Setup view to render object from the side // Setup view to render object from the side
Vector3 xAxis, yAxis, zAxis = Vector3::Zero; Vector3 xAxis, yAxis, zAxis = Vector3::Zero;
@@ -517,11 +530,12 @@ bool GlobalSurfaceAtlasPass::Render(RenderContext& renderContext, GPUContext* co
// Per-tile data // Per-tile data
const float tileWidth = (float)tile->Width - GLOBAL_SURFACE_ATLAS_TILE_PADDING; const float tileWidth = (float)tile->Width - GLOBAL_SURFACE_ATLAS_TILE_PADDING;
const float tileHeight = (float)tile->Height - GLOBAL_SURFACE_ATLAS_TILE_PADDING; const float tileHeight = (float)tile->Height - GLOBAL_SURFACE_ATLAS_TILE_PADDING;
objectData[tileStart + 0] = Vector4(tile->X, tile->Y, tileWidth, tileHeight) * resolutionInv; auto* tileData = surfaceAtlasData.TilesBuffer.WriteReserve<Vector4>(GLOBAL_SURFACE_ATLAS_TILE_BUFFER_STRIDE);
objectData[tileStart + 1] = Vector4(tile->ViewMatrix.M11, tile->ViewMatrix.M12, tile->ViewMatrix.M13, tile->ViewMatrix.M41); tileData[0] = Vector4(tile->X, tile->Y, tileWidth, tileHeight) * resolutionInv;
objectData[tileStart + 2] = Vector4(tile->ViewMatrix.M21, tile->ViewMatrix.M22, tile->ViewMatrix.M23, tile->ViewMatrix.M42); tileData[1] = Vector4(tile->ViewMatrix.M11, tile->ViewMatrix.M12, tile->ViewMatrix.M13, tile->ViewMatrix.M41);
objectData[tileStart + 3] = Vector4(tile->ViewMatrix.M31, tile->ViewMatrix.M32, tile->ViewMatrix.M33, tile->ViewMatrix.M43); tileData[2] = Vector4(tile->ViewMatrix.M21, tile->ViewMatrix.M22, tile->ViewMatrix.M23, tile->ViewMatrix.M42);
objectData[tileStart + 4] = Vector4(tile->ViewBoundsSize, 1.0f); tileData[3] = Vector4(tile->ViewMatrix.M31, tile->ViewMatrix.M32, tile->ViewMatrix.M33, tile->ViewMatrix.M43);
tileData[4] = Vector4(tile->ViewBoundsSize, 0.0f); // w unused
} }
} }
} }
@@ -548,6 +562,7 @@ bool GlobalSurfaceAtlasPass::Render(RenderContext& renderContext, GPUContext* co
{ {
PROFILE_GPU_CPU("Update Objects"); PROFILE_GPU_CPU("Update Objects");
surfaceAtlasData.ObjectsBuffer.Flush(context); surfaceAtlasData.ObjectsBuffer.Flush(context);
surfaceAtlasData.TilesBuffer.Flush(context);
} }
// Rasterize world geometry material properties into Global Surface Atlas // Rasterize world geometry material properties into Global Surface Atlas
@@ -670,6 +685,7 @@ bool GlobalSurfaceAtlasPass::Render(RenderContext& renderContext, GPUContext* co
result.Atlas[3] = surfaceAtlasData.AtlasGBuffer2; result.Atlas[3] = surfaceAtlasData.AtlasGBuffer2;
result.Atlas[4] = surfaceAtlasData.AtlasDirectLight; result.Atlas[4] = surfaceAtlasData.AtlasDirectLight;
result.Objects = surfaceAtlasData.ObjectsBuffer.GetBuffer(); result.Objects = surfaceAtlasData.ObjectsBuffer.GetBuffer();
result.Tiles = surfaceAtlasData.TilesBuffer.GetBuffer();
result.GlobalSurfaceAtlas.Resolution = (float)resolution; result.GlobalSurfaceAtlas.Resolution = (float)resolution;
result.GlobalSurfaceAtlas.ObjectsCount = surfaceAtlasData.Objects.Count(); result.GlobalSurfaceAtlas.ObjectsCount = surfaceAtlasData.Objects.Count();
surfaceAtlasData.Result = result; surfaceAtlasData.Result = result;
@@ -693,10 +709,11 @@ bool GlobalSurfaceAtlasPass::Render(RenderContext& renderContext, GPUContext* co
context->BindSR(2, surfaceAtlasData.AtlasGBuffer2->View()); context->BindSR(2, surfaceAtlasData.AtlasGBuffer2->View());
context->BindSR(3, surfaceAtlasData.AtlasDepth->View()); context->BindSR(3, surfaceAtlasData.AtlasDepth->View());
context->BindSR(4, surfaceAtlasData.ObjectsBuffer.GetBuffer()->View()); context->BindSR(4, surfaceAtlasData.ObjectsBuffer.GetBuffer()->View());
context->BindSR(5, surfaceAtlasData.TilesBuffer.GetBuffer()->View());
for (int32 i = 0; i < 4; i++) for (int32 i = 0; i < 4; i++)
{ {
context->BindSR(i + 5, bindingDataSDF.Cascades[i]->ViewVolume()); context->BindSR(i + 6, bindingDataSDF.Cascades[i]->ViewVolume());
context->BindSR(i + 9, bindingDataSDF.CascadeMips[i]->ViewVolume()); context->BindSR(i + 10, bindingDataSDF.CascadeMips[i]->ViewVolume());
} }
context->BindCB(0, _cb0); context->BindCB(0, _cb0);
Data0 data; Data0 data;
@@ -827,13 +844,14 @@ void GlobalSurfaceAtlasPass::RenderDebug(RenderContext& renderContext, GPUContex
context->BindSR(i + 4, bindingDataSDF.CascadeMips[i]->ViewVolume()); context->BindSR(i + 4, bindingDataSDF.CascadeMips[i]->ViewVolume());
} }
context->BindSR(8, bindingData.Objects ? bindingData.Objects->View() : nullptr); context->BindSR(8, bindingData.Objects ? bindingData.Objects->View() : nullptr);
context->BindSR(9, bindingData.Atlas[0]->View()); context->BindSR(9, bindingData.Tiles ? bindingData.Tiles->View() : nullptr);
context->BindSR(10, bindingData.Atlas[0]->View());
{ {
//GPUTexture* tex = bindingData.Atlas[1]; // Preview diffuse //GPUTexture* tex = bindingData.Atlas[1]; // Preview diffuse
//GPUTexture* tex = bindingData.Atlas[2]; // Preview normals //GPUTexture* tex = bindingData.Atlas[2]; // Preview normals
//GPUTexture* tex = bindingData.Atlas[3]; // Preview roughness/metalness/ao //GPUTexture* tex = bindingData.Atlas[3]; // Preview roughness/metalness/ao
GPUTexture* tex = bindingData.Atlas[4]; // Preview direct light GPUTexture* tex = bindingData.Atlas[4]; // Preview direct light
context->BindSR(10, tex->View()); context->BindSR(11, tex->View());
} }
context->SetState(_psDebug); context->SetState(_psDebug);
context->SetRenderTarget(output->View()); context->SetRenderTarget(output->View());

1
Source/Engine/Renderer/GlobalSurfaceAtlasPass.h
View File

@@ -23,6 +23,7 @@ public:
{ {
GPUTexture* Atlas[5]; GPUTexture* Atlas[5];
GPUBuffer* Objects; GPUBuffer* Objects;
GPUBuffer* Tiles;
GlobalSurfaceAtlasData GlobalSurfaceAtlas; GlobalSurfaceAtlasData GlobalSurfaceAtlas;
}; };

48
Source/Shaders/GlobalSurfaceAtlas.hlsl
View File

@@ -4,7 +4,8 @@
#include "./Flax/Collisions.hlsl" #include "./Flax/Collisions.hlsl"
// This must match C++ // This must match C++
#define GLOBAL_SURFACE_ATLAS_OBJECT_SIZE (5 + 6 * 5) // Amount of float4s per-object #define GLOBAL_SURFACE_ATLAS_OBJECT_BUFFER_STRIDE 6 // Amount of float4s per-object
#define GLOBAL_SURFACE_ATLAS_TILE_BUFFER_STRIDE 5 // Amount of float4s per-tile
#define GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD 0.1f // Cut-off value for tiles transitions blending during sampling #define GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD 0.1f // Cut-off value for tiles transitions blending during sampling
#define GLOBAL_SURFACE_ATLAS_TILE_PROJ_PLANE_OFFSET 0.1f // Small offset to prevent clipping with the closest triangles (shifts near and far planes) #define GLOBAL_SURFACE_ATLAS_TILE_PROJ_PLANE_OFFSET 0.1f // Small offset to prevent clipping with the closest triangles (shifts near and far planes)
@@ -13,7 +14,6 @@ struct GlobalSurfaceTile
float4 AtlasRectUV; float4 AtlasRectUV;
float4x4 WorldToLocal; float4x4 WorldToLocal;
float3 ViewBoundsSize; float3 ViewBoundsSize;
bool Enabled;
}; };
struct GlobalSurfaceObject struct GlobalSurfaceObject
@@ -22,24 +22,26 @@ struct GlobalSurfaceObject
float BoundsRadius; float BoundsRadius;
float4x4 WorldToLocal; float4x4 WorldToLocal;
float3 Extent; float3 Extent;
uint TileIndices[6];
}; };
float4 LoadGlobalSurfaceAtlasObjectBounds(Buffer<float4> objects, uint objectIndex) float4 LoadGlobalSurfaceAtlasObjectBounds(Buffer<float4> objects, uint objectIndex)
{ {
// This must match C++ // This must match C++
const uint objectStart = objectIndex * GLOBAL_SURFACE_ATLAS_OBJECT_SIZE; const uint objectStart = objectIndex * GLOBAL_SURFACE_ATLAS_OBJECT_BUFFER_STRIDE;
return objects.Load(objectStart); return objects.Load(objectStart);
} }
GlobalSurfaceObject LoadGlobalSurfaceAtlasObject(Buffer<float4> objects, uint objectIndex) GlobalSurfaceObject LoadGlobalSurfaceAtlasObject(Buffer<float4> objects, uint objectIndex)
{ {
// This must match C++ // This must match C++
const uint objectStart = objectIndex * GLOBAL_SURFACE_ATLAS_OBJECT_SIZE; const uint objectStart = objectIndex * GLOBAL_SURFACE_ATLAS_OBJECT_BUFFER_STRIDE;
float4 vector0 = objects.Load(objectStart + 0); float4 vector0 = objects.Load(objectStart + 0);
float4 vector1 = objects.Load(objectStart + 1); float4 vector1 = objects.Load(objectStart + 1);
float4 vector2 = objects.Load(objectStart + 2); float4 vector2 = objects.Load(objectStart + 2);
float4 vector3 = objects.Load(objectStart + 3); float4 vector3 = objects.Load(objectStart + 3);
float4 vector4 = objects.Load(objectStart + 4); // w unused float4 vector4 = objects.Load(objectStart + 4); // w unused
float4 vector5 = objects.Load(objectStart + 5); // w unused
GlobalSurfaceObject object = (GlobalSurfaceObject)0; GlobalSurfaceObject object = (GlobalSurfaceObject)0;
object.BoundsPosition = vector0.xyz; object.BoundsPosition = vector0.xyz;
object.BoundsRadius = vector0.w; object.BoundsRadius = vector0.w;
@@ -48,19 +50,27 @@ GlobalSurfaceObject LoadGlobalSurfaceAtlasObject(Buffer<float4> objects, uint ob
object.WorldToLocal[2] = float4(vector3.xyz, 0.0f); object.WorldToLocal[2] = float4(vector3.xyz, 0.0f);
object.WorldToLocal[3] = float4(vector1.w, vector2.w, vector3.w, 1.0f); object.WorldToLocal[3] = float4(vector1.w, vector2.w, vector3.w, 1.0f);
object.Extent = vector4.xyz; object.Extent = vector4.xyz;
uint vector5x = asuint(vector5.x);
uint vector5y = asuint(vector5.y);
uint vector5z = asuint(vector5.z);
object.TileIndices[0] = vector5x & 0xffff; // Limitation on max 65k active tiles
object.TileIndices[1] = vector5x >> 16;
object.TileIndices[2] = vector5y & 0xffff;
object.TileIndices[3] = vector5y >> 16;
object.TileIndices[4] = vector5z & 0xffff;
object.TileIndices[5] = vector5z >> 16;
return object; return object;
} }
GlobalSurfaceTile LoadGlobalSurfaceAtlasTile(Buffer<float4> objects, uint objectIndex, uint tileIndex) GlobalSurfaceTile LoadGlobalSurfaceAtlasTile(Buffer<float4> objects, uint tileIndex)
{ {
// This must match C++ // This must match C++
const uint objectStart = objectIndex * GLOBAL_SURFACE_ATLAS_OBJECT_SIZE; const uint tileStart = tileIndex * GLOBAL_SURFACE_ATLAS_TILE_BUFFER_STRIDE;
const uint tileStart = objectStart + 5 + tileIndex * 5;
float4 vector0 = objects.Load(tileStart + 0); float4 vector0 = objects.Load(tileStart + 0);
float4 vector1 = objects.Load(tileStart + 1); float4 vector1 = objects.Load(tileStart + 1);
float4 vector2 = objects.Load(tileStart + 2); float4 vector2 = objects.Load(tileStart + 2);
float4 vector3 = objects.Load(tileStart + 3); float4 vector3 = objects.Load(tileStart + 3);
float4 vector4 = objects.Load(tileStart + 4); float4 vector4 = objects.Load(tileStart + 4); // w unused
GlobalSurfaceTile tile = (GlobalSurfaceTile)0; GlobalSurfaceTile tile = (GlobalSurfaceTile)0;
tile.AtlasRectUV = vector0.xyzw; tile.AtlasRectUV = vector0.xyzw;
tile.WorldToLocal[0] = float4(vector1.xyz, 0.0f); tile.WorldToLocal[0] = float4(vector1.xyz, 0.0f);
@@ -68,7 +78,6 @@ GlobalSurfaceTile LoadGlobalSurfaceAtlasTile(Buffer<float4> objects, uint object
tile.WorldToLocal[2] = float4(vector3.xyz, 0.0f); tile.WorldToLocal[2] = float4(vector3.xyz, 0.0f);
tile.WorldToLocal[3] = float4(vector1.w, vector2.w, vector3.w, 1.0f); tile.WorldToLocal[3] = float4(vector1.w, vector2.w, vector3.w, 1.0f);
tile.ViewBoundsSize = vector4.xyz; tile.ViewBoundsSize = vector4.xyz;
tile.Enabled = vector4.w > 0;
return tile; return tile;
} }
@@ -138,7 +147,7 @@ float4 SampleGlobalSurfaceAtlasTile(const GlobalSurfaceAtlasData data, GlobalSur
} }
// Samples the Global Surface Atlas and returns the lighting (with opacity) at the given world location (and direction). // Samples the Global Surface Atlas and returns the lighting (with opacity) at the given world location (and direction).
float4 SampleGlobalSurfaceAtlas(const GlobalSurfaceAtlasData data, Buffer<float4> objects, Texture2D depth, Texture2D atlas, float3 worldPosition, float3 worldNormal) float4 SampleGlobalSurfaceAtlas(const GlobalSurfaceAtlasData data, Buffer<float4> objects, Buffer<float4> tiles, Texture2D depth, Texture2D atlas, float3 worldPosition, float3 worldNormal)
{ {
float4 result = float4(0, 0, 0, 0); float4 result = float4(0, 0, 0, 0);
float surfaceThreshold = 20.0f; // Additional threshold between object or tile size compared with input data (error due to SDF or LOD incorrect appearance) float surfaceThreshold = 20.0f; // Additional threshold between object or tile size compared with input data (error due to SDF or LOD incorrect appearance)
@@ -157,25 +166,24 @@ float4 SampleGlobalSurfaceAtlas(const GlobalSurfaceAtlasData data, Buffer<float4
continue; continue;
// Sample tiles based on the directionality // Sample tiles based on the directionality
// TODO: place enabled tiles mask in object data to skip reading disabled tiles
float3 localNormal = normalize(mul(worldNormal, (float3x3)object.WorldToLocal)); float3 localNormal = normalize(mul(worldNormal, (float3x3)object.WorldToLocal));
float3 localNormalSq = localNormal * localNormal; float3 localNormalSq = localNormal * localNormal;
if (localNormalSq.x > GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD * GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD) uint tileIndex = object.TileIndices[localNormal.x > 0.0f ? 0 : 1];
if (localNormalSq.x > GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD * GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD && tileIndex != 0)
{ {
uint tileIndex = localNormal.x > 0.0f ? 0 : 1; GlobalSurfaceTile tile = LoadGlobalSurfaceAtlasTile(tiles, tileIndex);
GlobalSurfaceTile tile = LoadGlobalSurfaceAtlasTile(objects, objectIndex, tileIndex);
result += SampleGlobalSurfaceAtlasTile(data, tile, depth, atlas, worldPosition, worldNormal, surfaceThreshold); result += SampleGlobalSurfaceAtlasTile(data, tile, depth, atlas, worldPosition, worldNormal, surfaceThreshold);
} }
if (localNormalSq.y > GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD * GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD) tileIndex = object.TileIndices[localNormal.y > 0.0f ? 2 : 3];
if (localNormalSq.y > GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD * GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD && tileIndex != 0)
{ {
uint tileIndex = localNormal.y > 0.0f ? 2 : 3; GlobalSurfaceTile tile = LoadGlobalSurfaceAtlasTile(tiles, tileIndex);
GlobalSurfaceTile tile = LoadGlobalSurfaceAtlasTile(objects, objectIndex, tileIndex);
result += SampleGlobalSurfaceAtlasTile(data, tile, depth, atlas, worldPosition, worldNormal, surfaceThreshold); result += SampleGlobalSurfaceAtlasTile(data, tile, depth, atlas, worldPosition, worldNormal, surfaceThreshold);
} }
if (localNormalSq.z > GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD * GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD) tileIndex = object.TileIndices[localNormal.z > 0.0f ? 4 : 5];
if (localNormalSq.z > GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD * GLOBAL_SURFACE_ATLAS_TILE_NORMAL_THRESHOLD && tileIndex != 0)
{ {
uint tileIndex = localNormal.z > 0.0f ? 4 : 5; GlobalSurfaceTile tile = LoadGlobalSurfaceAtlasTile(tiles, tileIndex);
GlobalSurfaceTile tile = LoadGlobalSurfaceAtlasTile(objects, objectIndex, tileIndex);
result += SampleGlobalSurfaceAtlasTile(data, tile, depth, atlas, worldPosition, worldNormal, surfaceThreshold); result += SampleGlobalSurfaceAtlasTile(data, tile, depth, atlas, worldPosition, worldNormal, surfaceThreshold);
} }
} }

14
Source/Shaders/GlobalSurfaceAtlas.shader
View File

@@ -67,8 +67,9 @@ void PS_Clear(out float4 Light : SV_Target0, out float4 RT0 : SV_Target1, out fl
// GBuffer+Depth at 0-3 slots // GBuffer+Depth at 0-3 slots
Buffer<float4> GlobalSurfaceAtlasObjects : register(t4); Buffer<float4> GlobalSurfaceAtlasObjects : register(t4);
Texture3D<float> GlobalSDFTex[4] : register(t5); Buffer<float4> GlobalSurfaceAtlasTiles : register(t5);
Texture3D<float> GlobalSDFMip[4] : register(t9); Texture3D<float> GlobalSDFTex[4] : register(t6);
Texture3D<float> GlobalSDFMip[4] : register(t10);
// Pixel shader for Global Surface Atlas shading with direct light contribution // Pixel shader for Global Surface Atlas shading with direct light contribution
META_PS(true, FEATURE_LEVEL_SM5) META_PS(true, FEATURE_LEVEL_SM5)
@@ -78,7 +79,7 @@ float4 PS_DirectLighting(AtlasVertexOutput input) : SV_Target
{ {
// Load current tile info // Load current tile info
//GlobalSurfaceObject object = LoadGlobalSurfaceAtlasObject(GlobalSurfaceAtlasObjects, input.Index.x); //GlobalSurfaceObject object = LoadGlobalSurfaceAtlasObject(GlobalSurfaceAtlasObjects, input.Index.x);
GlobalSurfaceTile tile = LoadGlobalSurfaceAtlasTile(GlobalSurfaceAtlasObjects, input.Index.x, input.Index.y); GlobalSurfaceTile tile = LoadGlobalSurfaceAtlasTile(GlobalSurfaceAtlasTiles, input.Index.y);
float2 atlasUV = input.TileUV * tile.AtlasRectUV.zw + tile.AtlasRectUV.xy; float2 atlasUV = input.TileUV * tile.AtlasRectUV.zw + tile.AtlasRectUV.xy;
// Load GBuffer sample from atlas // Load GBuffer sample from atlas
@@ -161,8 +162,9 @@ float4 PS_DirectLighting(AtlasVertexOutput input) : SV_Target
Texture3D<float> GlobalSDFTex[4] : register(t0); Texture3D<float> GlobalSDFTex[4] : register(t0);
Texture3D<float> GlobalSDFMip[4] : register(t4); Texture3D<float> GlobalSDFMip[4] : register(t4);
Buffer<float4> GlobalSurfaceAtlasObjects : register(t8); Buffer<float4> GlobalSurfaceAtlasObjects : register(t8);
Texture2D GlobalSurfaceAtlasDepth : register(t9); Buffer<float4> GlobalSurfaceAtlasTiles : register(t9);
Texture2D GlobalSurfaceAtlasTex : register(t10); Texture2D GlobalSurfaceAtlasDepth : register(t10);
Texture2D GlobalSurfaceAtlasTex : register(t11);
// Pixel shader for Global Surface Atlas debug drawing // Pixel shader for Global Surface Atlas debug drawing
META_PS(true, FEATURE_LEVEL_SM5) META_PS(true, FEATURE_LEVEL_SM5)
@@ -185,7 +187,7 @@ float4 PS_Debug(Quad_VS2PS input) : SV_Target
//return float4(hit.HitNormal * 0.5f + 0.5f, 1); //return float4(hit.HitNormal * 0.5f + 0.5f, 1);
// Sample Global Surface Atlas at the hit location // Sample Global Surface Atlas at the hit location
float4 surfaceColor = SampleGlobalSurfaceAtlas(GlobalSurfaceAtlas, GlobalSurfaceAtlasObjects, GlobalSurfaceAtlasDepth, GlobalSurfaceAtlasTex, hit.GetHitPosition(trace), -viewRay); float4 surfaceColor = SampleGlobalSurfaceAtlas(GlobalSurfaceAtlas, GlobalSurfaceAtlasObjects, GlobalSurfaceAtlasTiles, GlobalSurfaceAtlasDepth, GlobalSurfaceAtlasTex, hit.GetHitPosition(trace), -viewRay);
return float4(surfaceColor.rgb, 1); return float4(surfaceColor.rgb, 1);
} }