skip entity culling and wind compute if they are zero anyway

turanszkij · Jan 11, 2024 · 0aa9eea · 0aa9eea
1 parent bf5d86e
commit 0aa9eea
Show file tree

Hide file tree

Showing 4 changed files with 64 additions and 33 deletions.
diff --git a/WickedEngine/wiRenderPath3D.cpp b/WickedEngine/wiRenderPath3D.cpp
@@ -925,6 +925,7 @@ namespace wi
 
 			wi::renderer::ComputeTiledLightCulling(
 				tiledLightResources,
+				visibility_main,
 				debugUAV,
 				cmd
 			);
@@ -1132,6 +1133,7 @@ namespace wi
 
 				wi::renderer::ComputeTiledLightCulling(
 					tiledLightResources_planarReflection,
+					visibility_reflection,
 					Texture(),
 					cmd
 				);

diff --git a/WickedEngine/wiRenderer.cpp b/WickedEngine/wiRenderer.cpp
@@ -4265,15 +4265,25 @@ void UpdateRenderData(
 	{
 		auto range = wi::profiler::BeginRangeGPU("Wind", cmd);
 		device->EventBegin("Wind", cmd);
-
-		device->BindComputeShader(&shaders[CSTYPE_WIND], cmd);
-		device->BindUAV(&textures[TEXTYPE_3D_WIND], 0, cmd);
-		device->BindUAV(&textures[TEXTYPE_3D_WIND_PREV], 1, cmd);
-		const TextureDesc& desc = textures[TEXTYPE_3D_WIND].GetDesc();
-		device->Dispatch(desc.width / 8, desc.height / 8, desc.depth / 8, cmd);
+		if (
+			vis.scene->weather.windDirection.x == 0 &&
+			vis.scene->weather.windDirection.y == 0 &&
+			vis.scene->weather.windDirection.z == 0
+			)
+		{
+			device->ClearUAV(&textures[TEXTYPE_3D_WIND], 0, cmd);
+			device->ClearUAV(&textures[TEXTYPE_3D_WIND_PREV], 0, cmd);
+		}
+		else
+		{
+			device->BindComputeShader(&shaders[CSTYPE_WIND], cmd);
+			device->BindUAV(&textures[TEXTYPE_3D_WIND], 0, cmd);
+			device->BindUAV(&textures[TEXTYPE_3D_WIND_PREV], 1, cmd);
+			const TextureDesc& desc = textures[TEXTYPE_3D_WIND].GetDesc();
+			device->Dispatch(desc.width / 8, desc.height / 8, desc.depth / 8, cmd);
+		}
 		barrier_stack.push_back(GPUBarrier::Image(&textures[TEXTYPE_3D_WIND], ResourceState::UNORDERED_ACCESS, textures[TEXTYPE_3D_WIND].desc.layout));
 		barrier_stack.push_back(GPUBarrier::Image(&textures[TEXTYPE_3D_WIND_PREV], ResourceState::UNORDERED_ACCESS, textures[TEXTYPE_3D_WIND_PREV].desc.layout));
-
 		device->EventEnd(cmd);
 		wi::profiler::EndRange(range);
 	}
@@ -8837,13 +8847,48 @@ void CreateTiledLightResources(TiledLightResources& res, XMUINT2 resolution)
 }
 void ComputeTiledLightCulling(
 	const TiledLightResources& res,
+	const Visibility& vis,
 	const Texture& debugUAV,
 	CommandList cmd
 )
 {
-	BindCommonResources(cmd);
 	auto range = wi::profiler::BeginRangeGPU("Entity Culling", cmd);
 
+	// Initial barriers to put all resources into UAV:
+	{
+		GPUBarrier barriers[] = {
+			GPUBarrier::Buffer(&res.tileFrustums, ResourceState::SHADER_RESOURCE, ResourceState::UNORDERED_ACCESS),
+			GPUBarrier::Buffer(&res.entityTiles_Transparent, ResourceState::SHADER_RESOURCE, ResourceState::UNORDERED_ACCESS),
+			GPUBarrier::Buffer(&res.entityTiles_Opaque, ResourceState::SHADER_RESOURCE, ResourceState::UNORDERED_ACCESS),
+		};
+		device->Barrier(barriers, arraysize(barriers), cmd);
+	}
+
+	if (
+		vis.visibleLights.empty() &&
+		vis.visibleDecals.empty() &&
+		vis.visibleEnvProbes.empty()
+		)
+	{
+		device->EventBegin("Tiled Entity Clear Only", cmd);
+		device->ClearUAV(&res.tileFrustums, 0, cmd);
+		device->ClearUAV(&res.entityTiles_Transparent, 0, cmd);
+		device->ClearUAV(&res.entityTiles_Opaque, 0, cmd);
+		{
+			GPUBarrier barriers[] = {
+				GPUBarrier::Buffer(&res.tileFrustums, ResourceState::UNORDERED_ACCESS, ResourceState::SHADER_RESOURCE),
+				GPUBarrier::Buffer(&res.entityTiles_Transparent, ResourceState::UNORDERED_ACCESS, ResourceState::SHADER_RESOURCE),
+				GPUBarrier::Buffer(&res.entityTiles_Opaque, ResourceState::UNORDERED_ACCESS, ResourceState::SHADER_RESOURCE),
+			};
+			device->Barrier(barriers, arraysize(barriers), cmd);
+		}
+		device->EventEnd(cmd);
+		wi::profiler::EndRange(range);
+		return;
+	}
+
+	BindCommonResources(cmd);
+
 	// Frustum computation
 	{
 		device->EventBegin("Tile Frustums", cmd);
@@ -8854,20 +8899,20 @@ void ComputeTiledLightCulling(
 		};
 		device->BindUAVs(uavs, 0, arraysize(uavs), cmd);
 
-		{
-			GPUBarrier barriers[] = {
-				GPUBarrier::Buffer(&res.tileFrustums, ResourceState::SHADER_RESOURCE, ResourceState::UNORDERED_ACCESS)
-			};
-			device->Barrier(barriers, arraysize(barriers), cmd);
-		}
-
 		device->Dispatch(
 			(res.tileCount.x + TILED_CULLING_BLOCKSIZE - 1) / TILED_CULLING_BLOCKSIZE,
 			(res.tileCount.y + TILED_CULLING_BLOCKSIZE - 1) / TILED_CULLING_BLOCKSIZE,
 			1,
 			cmd
 		);
 
+		{
+			GPUBarrier barriers[] = {
+				GPUBarrier::Buffer(&res.tileFrustums, ResourceState::UNORDERED_ACCESS, ResourceState::SHADER_RESOURCE),
+			};
+			device->Barrier(barriers, arraysize(barriers), cmd);
+		}
+
 		device->EventEnd(cmd);
 	}
 
@@ -8893,25 +8938,8 @@ void ComputeTiledLightCulling(
 		};
 		device->BindUAVs(uavs, 0, arraysize(uavs), cmd);
 
-		{
-			GPUBarrier barriers[] = {
-				GPUBarrier::Buffer(&res.tileFrustums, ResourceState::UNORDERED_ACCESS, ResourceState::SHADER_RESOURCE),
-				GPUBarrier::Buffer(&res.entityTiles_Transparent, ResourceState::SHADER_RESOURCE, ResourceState::UNORDERED_ACCESS),
-				GPUBarrier::Buffer(&res.entityTiles_Opaque, ResourceState::SHADER_RESOURCE, ResourceState::UNORDERED_ACCESS),
-			};
-			device->Barrier(barriers, arraysize(barriers), cmd);
-		}
-
 		device->Dispatch(res.tileCount.x, res.tileCount.y, 1, cmd);
 
-		{
-			GPUBarrier barriers[] = {
-				GPUBarrier::Buffer(&res.entityTiles_Opaque, ResourceState::UNORDERED_ACCESS, ResourceState::SHADER_RESOURCE),
-				GPUBarrier::Buffer(&res.entityTiles_Transparent, ResourceState::UNORDERED_ACCESS, ResourceState::SHADER_RESOURCE),
-			};
-			device->Barrier(barriers, arraysize(barriers), cmd);
-		}
-
 		device->EventEnd(cmd);
 	}
 

diff --git a/WickedEngine/wiRenderer.h b/WickedEngine/wiRenderer.h
@@ -305,6 +305,7 @@ namespace wi::renderer
 	// Compute light grid tiles
 	void ComputeTiledLightCulling(
 		const TiledLightResources& res,
+		const Visibility& vis,
 		const wi::graphics::Texture& debugUAV,
 		wi::graphics::CommandList cmd
 	);

diff --git a/WickedEngine/wiVersion.cpp b/WickedEngine/wiVersion.cpp
@@ -9,7 +9,7 @@ namespace wi::version
 	// minor features, major updates, breaking compatibility changes
 	const int minor = 71;
 	// minor bug fixes, alterations, refactors, updates
-	const int revision = 357;
+	const int revision = 358;
 
 	const std::string version_string = std::to_string(major) + "." + std::to_string(minor) + "." + std::to_string(revision);