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

Optimize GPU particles simulation by scheduling larger sims first to overlap more work

Browse Source
This commit is contained in:
Wojtek Figat
2025-08-11 13:24:11 +02:00
parent a18314c669
commit 9f14bb7279
1 changed files with 28 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
Source/Engine/Particles/Particles.cpp
View File

@@ -1333,7 +1333,19 @@ void UpdateGPU(RenderTask* task, GPUContext* context)
ParticleEffect* Effect; ParticleEffect* Effect;
ParticleEmitter* Emitter; ParticleEmitter* Emitter;
int32 EmitterIndex; int32 EmitterIndex;
ParticleEmitterInstance& Data; ParticleEmitterInstance* Data;
bool operator<(const GPUSim& other) const
{
// Sort by particle count (larger effects start first)
if (Data->Buffer->GPU.ParticlesCountMax != other.Data->Buffer->GPU.ParticlesCountMax)
return Data->Buffer->GPU.ParticlesCountMax > other.Data->Buffer->GPU.ParticlesCountMax;
if (Emitter->Capacity != other.Emitter->Capacity)
return Emitter->Capacity > other.Emitter->Capacity;
// Merge emitters together (compute pipeline switches)
return (uintptr)Emitter < (uintptr)other.Emitter;
}
}; };
Array<GPUSim, RendererAllocation> sims; Array<GPUSim, RendererAllocation> sims;
sims.EnsureCapacity(Math::AlignUp(GpuUpdateList.Count(), 64)); // Preallocate with some slack sims.EnsureCapacity(Math::AlignUp(GpuUpdateList.Count(), 64)); // Preallocate with some slack
@@ -1364,11 +1376,14 @@ void UpdateGPU(RenderTask* task, GPUContext* context)
emitter->GPU.PreSim(context, emitter, effect, emitterIndex, data); emitter->GPU.PreSim(context, emitter, effect, emitterIndex, data);
continue; continue;
} }
sims.Add({ effect, emitter, emitterIndex, data }); sims.Add({ effect, emitter, emitterIndex, &data });
} }
} }
GpuUpdateList.Clear(); GpuUpdateList.Clear();
// Sort particles by emitter type to reduce compute pipeline switches
Sorting::QuickSort(sims);
// Pre-pass with buffers setup // Pre-pass with buffers setup
{ {
PROFILE_CPU_NAMED("PreSim"); PROFILE_CPU_NAMED("PreSim");
@@ -1376,14 +1391,14 @@ void UpdateGPU(RenderTask* task, GPUContext* context)
GPUMemoryPass pass(context); GPUMemoryPass pass(context);
for (GPUSim& sim : sims) for (GPUSim& sim : sims)
{ {
if (sim.Data.Buffer->GPU.PendingClear) if (sim.Data->Buffer->GPU.PendingClear)
pass.Transition(sim.Data.Buffer->GPU.Buffer, GPUResourceAccess::CopyWrite); pass.Transition(sim.Data->Buffer->GPU.Buffer, GPUResourceAccess::CopyWrite);
pass.Transition(sim.Data.Buffer->GPU.BufferSecondary, GPUResourceAccess::CopyWrite); pass.Transition(sim.Data->Buffer->GPU.BufferSecondary, GPUResourceAccess::CopyWrite);
} }
for (GPUSim& sim : sims) for (GPUSim& sim : sims)
{ {
sim.Emitter->GPU.PreSim(context, sim.Emitter, sim.Effect, sim.EmitterIndex, sim.Data); sim.Emitter->GPU.PreSim(context, sim.Emitter, sim.Effect, sim.EmitterIndex, *sim.Data);
} }
} }
@@ -1394,13 +1409,13 @@ void UpdateGPU(RenderTask* task, GPUContext* context)
GPUComputePass pass(context); GPUComputePass pass(context);
for (GPUSim& sim : sims) for (GPUSim& sim : sims)
{ {
pass.Transition(sim.Data.Buffer->GPU.Buffer, GPUResourceAccess::ShaderReadCompute); pass.Transition(sim.Data->Buffer->GPU.Buffer, GPUResourceAccess::ShaderReadCompute);
pass.Transition(sim.Data.Buffer->GPU.BufferSecondary, GPUResourceAccess::UnorderedAccess); pass.Transition(sim.Data->Buffer->GPU.BufferSecondary, GPUResourceAccess::UnorderedAccess);
} }
for (GPUSim& sim : sims) for (GPUSim& sim : sims)
{ {
sim.Emitter->GPU.Sim(context, sim.Emitter, sim.Effect, sim.EmitterIndex, sim.Data); sim.Emitter->GPU.Sim(context, sim.Emitter, sim.Effect, sim.EmitterIndex, *sim.Data);
} }
} }
@@ -1411,16 +1426,16 @@ void UpdateGPU(RenderTask* task, GPUContext* context)
GPUMemoryPass pass(context); GPUMemoryPass pass(context);
for (GPUSim& sim : sims) for (GPUSim& sim : sims)
{ {
if (sim.Data.CustomData.HasItems()) if (sim.Data->CustomData.HasItems())
{ {
pass.Transition(sim.Data.Buffer->GPU.BufferSecondary, GPUResourceAccess::CopyRead); pass.Transition(sim.Data->Buffer->GPU.BufferSecondary, GPUResourceAccess::CopyRead);
pass.Transition(sim.Data.Buffer->GPU.Buffer, GPUResourceAccess::CopyWrite); pass.Transition(sim.Data->Buffer->GPU.Buffer, GPUResourceAccess::CopyWrite);
} }
} }
for (GPUSim& sim : sims) for (GPUSim& sim : sims)
{ {
sim.Emitter->GPU.PostSim(context, sim.Emitter, sim.Effect, sim.EmitterIndex, sim.Data); sim.Emitter->GPU.PostSim(context, sim.Emitter, sim.Effect, sim.EmitterIndex, *sim.Data);
} }
} }