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

Optimize DDGI probes border pixels to be copied within probe update, rather than via separate dispatch

Browse Source
This commit is contained in:
Wojtek Figat
2024-07-18 19:51:01 +02:00
parent aeff147b6d
commit 03898a064a
3 changed files with 85 additions and 114 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
Source/Engine/Renderer/GI/DynamicDiffuseGlobalIllumination.cpp
View File

@@ -219,10 +219,6 @@ bool DynamicDiffuseGlobalIlluminationPass::setupResources()
_csTraceRays[3] = shader->GetCS("CS_TraceRays", 3);
_csUpdateProbesIrradiance = shader->GetCS("CS_UpdateProbes", 0);
_csUpdateProbesDistance = shader->GetCS("CS_UpdateProbes", 1);
_csUpdateBordersIrradianceRow = shader->GetCS("CS_UpdateBorders", 0);
_csUpdateBordersIrradianceCollumn = shader->GetCS("CS_UpdateBorders", 1);
_csUpdateBordersDistanceRow = shader->GetCS("CS_UpdateBorders", 2);
_csUpdateBordersDistanceCollumn = shader->GetCS("CS_UpdateBorders", 3);
auto device = GPUDevice::Instance;
auto psDesc = GPUPipelineState::Description::DefaultFullscreenTriangle;
if (!_psIndirectLighting)
@@ -250,10 +246,6 @@ void DynamicDiffuseGlobalIlluminationPass::OnShaderReloading(Asset* obj)
_csTraceRays[3] = nullptr;
_csUpdateProbesIrradiance = nullptr;
_csUpdateProbesDistance = nullptr;
_csUpdateBordersIrradianceRow = nullptr;
_csUpdateBordersIrradianceCollumn = nullptr;
_csUpdateBordersDistanceRow = nullptr;
_csUpdateBordersDistanceCollumn = nullptr;
SAFE_DELETE_GPU_RESOURCE(_psIndirectLighting);
invalidateResources();
}
@@ -542,7 +534,6 @@ bool DynamicDiffuseGlobalIlluminationPass::RenderInner(RenderContext& renderCont
// Update probes
{
PROFILE_GPU_CPU_NAMED("Probes Update");
bool anyDirty = false;
uint32 threadGroupsX, threadGroupsY;
#if DDGI_DEBUG_STATS
uint32 zero[4] = {};
@@ -552,7 +543,6 @@ bool DynamicDiffuseGlobalIlluminationPass::RenderInner(RenderContext& renderCont
{
if (cascadeSkipUpdate[cascadeIndex])
continue;
anyDirty = true;
// Classify probes (activation/deactivation and relocation)
{
@@ -667,33 +657,6 @@ bool DynamicDiffuseGlobalIlluminationPass::RenderInner(RenderContext& renderCont
}
}
#endif
// Update probes border pixels
if (anyDirty)
{
PROFILE_GPU_CPU_NAMED("Update Borders");
// Irradiance
context->BindUA(0, ddgiData.Result.ProbesIrradiance);
threadGroupsX = Math::DivideAndRoundUp(probesCountTotalX * (DDGI_PROBE_RESOLUTION_IRRADIANCE + 2), DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE);
threadGroupsY = Math::DivideAndRoundUp(probesCountTotalY, DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE);
context->Dispatch(_csUpdateBordersIrradianceRow, threadGroupsX, threadGroupsY, 1);
threadGroupsX = Math::DivideAndRoundUp(probesCountTotalX, DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE);
threadGroupsY = Math::DivideAndRoundUp(probesCountTotalY * (DDGI_PROBE_RESOLUTION_IRRADIANCE + 2), DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE);
context->Dispatch(_csUpdateBordersIrradianceCollumn, threadGroupsX, threadGroupsY, 1);
// Distance
context->BindUA(0, ddgiData.Result.ProbesDistance);
threadGroupsX = Math::DivideAndRoundUp(probesCountTotalX * (DDGI_PROBE_RESOLUTION_DISTANCE + 2), DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE);
threadGroupsY = Math::DivideAndRoundUp(probesCountTotalY, DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE);
context->Dispatch(_csUpdateBordersDistanceRow, threadGroupsX, threadGroupsY, 1);
threadGroupsX = Math::DivideAndRoundUp(probesCountTotalX, DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE);
threadGroupsY = Math::DivideAndRoundUp(probesCountTotalY * (DDGI_PROBE_RESOLUTION_DISTANCE + 2), DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE);
context->Dispatch(_csUpdateBordersDistanceCollumn, threadGroupsX, threadGroupsY, 1);
context->ResetUA();
context->ResetSR();
}
}
return false;

4
Source/Engine/Renderer/GI/DynamicDiffuseGlobalIllumination.h
View File

@@ -47,10 +47,6 @@ private:
GPUShaderProgramCS* _csTraceRays[4];
GPUShaderProgramCS* _csUpdateProbesIrradiance;
GPUShaderProgramCS* _csUpdateProbesDistance;
GPUShaderProgramCS* _csUpdateBordersIrradianceRow;
GPUShaderProgramCS* _csUpdateBordersIrradianceCollumn;
GPUShaderProgramCS* _csUpdateBordersDistanceRow;
GPUShaderProgramCS* _csUpdateBordersDistanceCollumn;
GPUPipelineState* _psIndirectLighting;
#if USE_EDITOR
AssetReference<Model> _debugModel;

158
Source/Shaders/GI/DDGI.shader
View File

@@ -415,16 +415,22 @@ void CS_TraceRays(uint3 DispatchThreadId : SV_DispatchThreadID)
#endif
#if defined(_CS_UpdateProbes) || defined(_CS_UpdateBorders)
#if defined(_CS_UpdateProbes)
#if DDGI_PROBE_UPDATE_MODE == 0
// Update irradiance
#define DDGI_PROBE_RESOLUTION DDGI_PROBE_RESOLUTION_IRRADIANCE
groupshared float4 CachedProbesTraceRadiance[DDGI_TRACE_RAYS_LIMIT];
groupshared float OutputInstability[DDGI_PROBE_RESOLUTION * DDGI_PROBE_RESOLUTION];
#else
// Update distance
#define DDGI_PROBE_RESOLUTION DDGI_PROBE_RESOLUTION_DISTANCE
groupshared float CachedProbesTraceDistance[DDGI_TRACE_RAYS_LIMIT];
#endif
// Source: https://github.com/turanszkij/WickedEngine
#define BorderOffsetsSize (4 * DDGI_PROBE_RESOLUTION + 4)
#if DDGI_PROBE_RESOLUTION == 6
static const uint4 BorderOffsets[BorderOffsetsSize] = {
uint4(6, 1, 1, 0),
uint4(5, 1, 2, 0),
@@ -457,12 +463,77 @@ static const uint4 BorderOffsets[BorderOffsetsSize] = {
uint4(1, 1, 7, 7),
uint4(6, 1, 0, 7),
uint4(1, 6, 7, 0),
uint4(6, 6, 0, 0),
uint4(6, 6, 0, 0)
};
#elif DDGI_PROBE_RESOLUTION == 14
static const uint4 BorderOffsets[BorderOffsetsSize] = {
uint4(14, 1, 1, 0),
uint4(13, 1, 2, 0),
uint4(12, 1, 3, 0),
uint4(11, 1, 4, 0),
uint4(10, 1, 5, 0),
uint4(9, 1, 6, 0),
uint4(8, 1, 7, 0),
uint4(7, 1, 8, 0),
uint4(6, 1, 9, 0),
uint4(5, 1, 10, 0),
uint4(4, 1, 11, 0),
uint4(3, 1, 12, 0),
uint4(2, 1, 13, 0),
uint4(1, 1, 14, 0),
uint4(14, 14, 1, 15),
uint4(13, 14, 2, 15),
uint4(12, 14, 3, 15),
uint4(11, 14, 4, 15),
uint4(10, 14, 5, 15),
uint4(9, 14, 6, 15),
uint4(8, 14, 7, 15),
uint4(7, 14, 8, 15),
uint4(6, 14, 9, 15),
uint4(5, 14, 10, 15),
uint4(4, 14, 11, 15),
uint4(3, 14, 12, 15),
uint4(2, 14, 13, 15),
uint4(1, 14, 14, 15),
uint4(1, 14, 0, 1),
uint4(1, 13, 0, 2),
uint4(1, 12, 0, 3),
uint4(1, 11, 0, 4),
uint4(1, 10, 0, 5),
uint4(1, 9, 0, 6),
uint4(1, 8, 0, 7),
uint4(1, 7, 0, 8),
uint4(1, 6, 0, 9),
uint4(1, 5, 0, 10),
uint4(1, 4, 0, 11),
uint4(1, 3, 0, 12),
uint4(1, 2, 0, 13),
uint4(1, 1, 0, 14),
uint4(14, 14, 15, 1),
uint4(14, 13, 15, 2),
uint4(14, 12, 15, 3),
uint4(14, 11, 15, 4),
uint4(14, 10, 15, 5),
uint4(14, 9, 15, 6),
uint4(14, 8, 15, 7),
uint4(14, 7, 15, 8),
uint4(14, 6, 15, 9),
uint4(14, 5, 15, 10),
uint4(14, 4, 15, 11),
uint4(14, 3, 15, 12),
uint4(14, 2, 15, 13),
uint4(14, 1, 15, 14),
uint4(14, 14, 0, 0),
uint4(1, 14, 15, 0),
uint4(14, 1, 0, 15),
uint4(1, 1, 15, 15)
};
#else
// Update distance
#define DDGI_PROBE_RESOLUTION DDGI_PROBE_RESOLUTION_DISTANCE
groupshared float CachedProbesTraceDistance[DDGI_TRACE_RAYS_LIMIT];
#error "Unsupported probe size for border values copy."
#endif
groupshared float3 CachedProbesTraceDirection[DDGI_TRACE_RAYS_LIMIT];
@@ -635,9 +706,11 @@ void CS_UpdateProbes(uint3 GroupThreadId : SV_GroupThreadID, uint3 GroupId : SV_
RWOutput[outputCoords] = result;
GroupMemoryBarrierWithGroupSync();
uint2 baseCoords = GetDDGIProbeTexelCoords(DDGI, CascadeIndex, probeIndex) * (DDGI_PROBE_RESOLUTION + 2);
#if DDGI_PROBE_UPDATE_MODE == 0
// The first thread updates the probe attention based on the instability of all texels
GroupMemoryBarrierWithGroupSync();
BRANCH
if (GroupIndex == 0 && probeState != DDGI_PROBE_STATE_INACTIVE)
{
@@ -665,7 +738,6 @@ void CS_UpdateProbes(uint3 GroupThreadId : SV_GroupThreadID, uint3 GroupId : SV_
#if DDGI_DEBUG_INSTABILITY
// Copy border pixels
uint2 baseCoords = GetDDGIProbeTexelCoords(DDGI, CascadeIndex, probeIndex) * (DDGI_PROBE_RESOLUTION + 2);
for (uint borderIndex = GroupIndex; borderIndex < BorderOffsetsSize; borderIndex += DDGI_PROBE_RESOLUTION * DDGI_PROBE_RESOLUTION)
{
uint4 borderOffsets = BorderOffsets[borderIndex];
@@ -673,73 +745,13 @@ void CS_UpdateProbes(uint3 GroupThreadId : SV_GroupThreadID, uint3 GroupId : SV_
}
#endif
#endif
}
// Compute shader for updating probes irradiance or distance texture borders (fills gaps between probes to support bilinear filtering)
META_CS(true, FEATURE_LEVEL_SM5)
META_PERMUTATION_2(DDGI_PROBE_UPDATE_MODE=0, BORDER_ROW=1)
META_PERMUTATION_2(DDGI_PROBE_UPDATE_MODE=0, BORDER_ROW=0)
META_PERMUTATION_2(DDGI_PROBE_UPDATE_MODE=1, BORDER_ROW=1)
META_PERMUTATION_2(DDGI_PROBE_UPDATE_MODE=1, BORDER_ROW=0)
[numthreads(DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE, DDGI_PROBE_UPDATE_BORDERS_GROUP_SIZE, 1)]
void CS_UpdateBorders(uint3 DispatchThreadId : SV_DispatchThreadID)
{
#define COPY_PIXEL RWOutput[threadCoordinates] = RWOutput[copyCoordinates]
#define COPY_PIXEL_DEBUG RWOutput[threadCoordinates] = float4(5, 0, 0, 1)
uint probeSideLength = DDGI_PROBE_RESOLUTION + 2;
uint probeSideLengthMinusOne = probeSideLength - 1;
uint2 copyCoordinates = uint2(0, 0);
uint2 threadCoordinates = DispatchThreadId.xy;
#if BORDER_ROW
threadCoordinates.y *= probeSideLength;
uint corner = DispatchThreadId.x % probeSideLength;
#else
threadCoordinates.x *= probeSideLength;
uint corner = threadCoordinates.y % probeSideLength;
#endif
if (corner == 0 || corner == probeSideLengthMinusOne)
{
#if !BORDER_ROW
// Left corner
copyCoordinates.x = threadCoordinates.x + DDGI_PROBE_RESOLUTION;
copyCoordinates.y = threadCoordinates.y - sign((int)corner - 1) * DDGI_PROBE_RESOLUTION;
COPY_PIXEL;
// Right corner
threadCoordinates.x += probeSideLengthMinusOne;
copyCoordinates.x = threadCoordinates.x - DDGI_PROBE_RESOLUTION;
COPY_PIXEL;
#endif
return;
}
#if BORDER_ROW
// Top row
uint probeStart = uint(threadCoordinates.x / probeSideLength) * probeSideLength;
uint offset = probeSideLengthMinusOne - (threadCoordinates.x % probeSideLength);
copyCoordinates = uint2(probeStart + offset, threadCoordinates.y + 1);
#else
// Left column
uint probeStart = uint(threadCoordinates.y / probeSideLength) * probeSideLength;
uint offset = probeSideLengthMinusOne - (threadCoordinates.y % probeSideLength);
copyCoordinates = uint2(threadCoordinates.x + 1, probeStart + offset);
#endif
COPY_PIXEL;
#if BORDER_ROW
// Bottom row
threadCoordinates.y += probeSideLengthMinusOne;
copyCoordinates = uint2(probeStart + offset, threadCoordinates.y - 1);
#else
// Right column
threadCoordinates.x += probeSideLengthMinusOne;
copyCoordinates = uint2(threadCoordinates.x - 1, probeStart + offset);
#endif
COPY_PIXEL;
#undef COPY_PIXEL
#undef COPY_PIXEL_DEBUG
// Copy border pixels
for (uint borderIndex = GroupIndex; borderIndex < BorderOffsetsSize; borderIndex += DDGI_PROBE_RESOLUTION * DDGI_PROBE_RESOLUTION)
{
uint4 borderOffsets = BorderOffsets[borderIndex];
RWOutput[baseCoords + borderOffsets.zw] = RWOutput[baseCoords + borderOffsets.xy];
}
}
#endif