perf(particlesys): Implement batched rendering for particles (#)

Author: Mauller

Core/GameEngine/Include/GameClient/ParticleSys.h

@@ -602,7 +602,7 @@ class ParticleSystem : public MemoryPoolObject,
 
 	void setInitialDelay( UnsignedInt delay ) { m_delayLeft = delay; }
 
-	AsciiString getParticleTypeName() { return m_particleTypeName; }	///< return the name of the particles
+	const AsciiString& getParticleTypeName() const { return m_particleTypeName; }	///< return the name of the particles
 	Bool isUsingDrawables() { return (m_particleType == DRAWABLE) ? true : false; }
 	Bool isUsingStreak() { return (m_particleType == STREAK) ? true : false; }
 	Bool isUsingSmudge() { return (m_particleType == SMUDGE) ? true : false; }

Core/GameEngineDevice/Source/W3DDevice/GameClient/W3DParticleSys.cpp

@@ -46,6 +46,10 @@
 
 W3DParticleSystemManager::W3DParticleSystemManager()
 {
+	m_batchBillboard = true;
+	m_batchShaderType = ParticleSystemInfo::INVALID_SHADER;
+	m_batchTexture = nullptr;
+
 	m_pointGroup = nullptr;
 	m_streakLine = nullptr;
 	m_posBuffer = nullptr;
@@ -77,6 +81,11 @@ W3DParticleSystemManager::~W3DParticleSystemManager()
 		REF_PTR_RELEASE(m_streakLine);
 	}
 
+	if (m_batchTexture)
+	{
+		REF_PTR_RELEASE(m_batchTexture);
+	}
+
 	REF_PTR_RELEASE(m_posBuffer);
 	REF_PTR_RELEASE(m_RGBABuffer);
 	REF_PTR_RELEASE(m_sizeBuffer);
@@ -159,24 +168,41 @@ void W3DParticleSystemManager::doParticles(RenderInfoClass &rinfo)
 		if (sys->isUsingDrawables())
 			continue;
 
+		// TheSuperHackers @performance Mauller 16/08/2026 Test if the particle system has any visible particles that can be drawn
+		// Earlier visibility testing prevents the particle texture lookup which can cause a batch flush
+		int particleCount = 0;
+		for (Particle* vp = sys->getFirstParticle(); vp; vp = vp->m_systemNext)
+		{
+			const Coord3D* pos = vp->getPosition();
+			Real psize = vp->getSize();
+			vp->setIsCulled(true);
+
+			//Test if particle is at the screen or terrain edges.
+			if (WWMath::Fabs(pos->x - bcX) > (beX + psize))
+				continue;
+
+			if (WWMath::Fabs(pos->y - bcY) > (beY + psize))
+				continue;
+
+			if (WWMath::Fabs(pos->z - bcZ) > (beZ + psize))
+				continue;
+
+			vp->setIsCulled(false);
+			particleCount++;
+		}
+
+		// Particle system has no particles on screen
+		if (particleCount == 0)
+			continue;
+
 		//temporary hack that checks if texture name starts with "SMUD" - if so, we can assume it's a smudge type
 		if (/*sys->isUsingSmudge()*/ *((DWORD *)sys->getParticleTypeName().str()) == 0x44554D53)
 		{
 			if (drawSmudge)
 			{
 				for (Particle *p = sys->getFirstParticle(); p; p = p->m_systemNext)
 				{
-					const Coord3D *pos = p->getPosition();
-					Real psize = p->getSize();
-
-					//Cull particle to edges of screen and terrain.
-					if (WWMath::Fabs( pos->x - bcX ) > ( beX + psize ) )
-						continue;
-
-					if (WWMath::Fabs( pos->y - bcY ) > ( beY + psize ) )
-						continue;
-
-					if (WWMath::Fabs( pos->z - bcZ ) > ( beZ + psize ) )
+					if (p->isCulled())
 						continue;
 
 					if (Smudge *smudge = TheSmudgeManager->findSmudge(p))
@@ -189,10 +215,31 @@ void W3DParticleSystemManager::doParticles(RenderInfoClass &rinfo)
 			continue;
 		}
 
-		/// @todo lorenzen sez: declare these outside the sys loop, and put some in registers
-		// initialize them here still, of course
+		// TheSuperHackers @performance Ronin/Mauller 09/08/2026 Implement batched rendering for similar particles
+		// Particles with the same blending will now be batched onto a single texture surface before being drawn
+		// If a different particle type appears before the batch is filled, the previous batch will be drawn first
+		TextureClass *texture = W3DDisplay::m_assetManager->Get_Texture( sys->getParticleTypeName().str() );
+		const Bool canBatch = !( m_streakLine && sys->isUsingStreak() ) && ( sys->getVolumeParticleDepth() <= 1 );
+		if (!canBatch ||
+			texture != m_batchTexture ||
+			sys->getShaderType() != m_batchShaderType ||
+			sys->shouldBillboard() != m_batchBillboard)
+		{
+			flushParticleBatch(rinfo, pointCount);
+		}
+
+		// setup a new particle batch texture if prior batch was flushed
+		if (canBatch && m_batchTexture == nullptr)
+		{
+			m_batchTexture = texture;
+			m_batchTexture->Add_Ref();
+			m_batchShaderType = sys->getShaderType();
+			m_batchBillboard = sys->shouldBillboard();
+		}
+
+		Int startCount = pointCount;
+
 		// build W3D particle buffer
-		pointCount = 0;
 		Vector3 *posArray = m_posBuffer->Get_Array();
 		Real *sizeArray = m_sizeBuffer->Get_Array();
 		Vector4 *RGBAArray = m_RGBABuffer->Get_Array();
@@ -206,18 +253,11 @@ void W3DParticleSystemManager::doParticles(RenderInfoClass &rinfo)
 		//set-up all the per-particle
 		for (Particle *p = sys->getFirstParticle(); p; p = p->m_systemNext)
 		{
-			pos = p->getPosition();
-			psize = p->getSize();
-
-			//Cull particle to edges of screen and terrain.
-			if (WWMath::Fabs(pos->x - bcX) > (beX + psize))
-				continue;
-
-			if (WWMath::Fabs(pos->y - bcY) > (beY + psize))
+			if (p->isCulled())
 				continue;
 
-			if (WWMath::Fabs(pos->z - bcZ) > (beZ + psize))
-				continue;
+			pos = p->getPosition();
+			psize = p->getSize();
 
 			m_fieldParticleCount += ( sys->getPriority() == AREA_EFFECT && sys->m_isGroundAligned != FALSE );
 
@@ -239,13 +279,29 @@ void W3DParticleSystemManager::doParticles(RenderInfoClass &rinfo)
 			angleArray[pointCount] = (uint8)(p->getAngle() * 255.0f / (2.0f * PI));
 
 			if (++pointCount == MAX_POINTS_PER_GROUP)
-				break;
+			{
+				if (!canBatch)
+				{
+					break;
+				}
+
+				// TheSuperHackers @info The Buffer is full mid-system so draw what we have and carry on with the SAME system
+				// This prevents particles being dropped. Bank the stats first as the flush resets count to 0.
+				m_onScreenParticleCount += (pointCount - startCount);
+				flushParticleBatch(rinfo, pointCount);
+				m_batchTexture = texture;
+				m_batchTexture->Add_Ref();
+				m_batchShaderType = sys->getShaderType();
+				m_batchBillboard = sys->shouldBillboard();
+				startCount = 0;
+			}
 		}
 
-		if ( pointCount == 0 )
+		if (pointCount == startCount)
+		{
+			texture->Release_Ref();
 			continue;	//this system has no particles to render
-
-		TextureClass *texture = W3DDisplay::m_assetManager->Get_Texture( sys->getParticleTypeName().str() );
+		}
 
 		if ( m_streakLine && sys->isUsingStreak() && (pointCount >= 2) )
 		{
@@ -298,51 +354,67 @@ void W3DParticleSystemManager::doParticles(RenderInfoClass &rinfo)
 
 			if ( m_pointGroup ) // this catches the particle and volumeparticle cases
 			{
-				// render all the systems' particles
-				m_pointGroup->Set_Texture( texture );
-				texture->Release_Ref();//release reference since it's held by pointGroup
-				m_pointGroup->Set_Flag( PointGroupClass::TRANSFORM, true );	// transform to screen space
-
-				switch( sys->getShaderType() )
+				if ( sys->getVolumeParticleDepth() > 1 )
 				{
-					case ParticleSystemInfo::ADDITIVE:
-						m_pointGroup->Set_Shader( ShaderClass::_PresetAdditiveSpriteShader );
-						break;
-					case ParticleSystemInfo::ALPHA:
-						m_pointGroup->Set_Shader( ShaderClass::_PresetAlphaSpriteShader );
-						break;
-					case ParticleSystemInfo::ALPHA_TEST:
-						m_pointGroup->Set_Shader( ShaderClass::_PresetATestSpriteShader );
-						break;
-					case ParticleSystemInfo::MULTIPLY:
-						m_pointGroup->Set_Shader( ShaderClass::_PresetMultiplicativeSpriteShader );
-						break;
-				}
+					m_pointGroup->Set_Texture( texture );
+					texture->Release_Ref();//release reference since it's held by pointGroup
+					m_pointGroup->Set_Flag( PointGroupClass::TRANSFORM, true );	// transform to screen space
 
-				/// @todo Use both QUADS and TRIS for particles
-				m_pointGroup->Set_Point_Mode( PointGroupClass::QUADS );
-				m_pointGroup->Set_Arrays( m_posBuffer, m_RGBABuffer, nullptr, m_sizeBuffer, m_angleBuffer, nullptr, pointCount);
-				m_pointGroup->Set_Billboard(sys->shouldBillboard());
+					switch( sys->getShaderType() )
+					{
+						case ParticleSystemInfo::ADDITIVE:
+							m_pointGroup->Set_Shader( ShaderClass::_PresetAdditiveSpriteShader );
+							break;
+						case ParticleSystemInfo::ALPHA:
+							m_pointGroup->Set_Shader( ShaderClass::_PresetAlphaSpriteShader );
+							break;
+						case ParticleSystemInfo::ALPHA_TEST:
+							m_pointGroup->Set_Shader( ShaderClass::_PresetATestSpriteShader );
+							break;
+						case ParticleSystemInfo::MULTIPLY:
+							m_pointGroup->Set_Shader( ShaderClass::_PresetMultiplicativeSpriteShader );
+							break;
+					}
 
-				/// @todo Support animated texture particles
-				/// @todo lorenzen sez: unimplemented code wastes cpu cycles
-				m_pointGroup->Set_Point_Frame( 0 );
+					/// @todo Use both QUADS and TRIS for particles
+					m_pointGroup->Set_Point_Mode( PointGroupClass::QUADS );
+					m_pointGroup->Set_Arrays( m_posBuffer, m_RGBABuffer, nullptr, m_sizeBuffer, m_angleBuffer, nullptr, pointCount );
+					m_pointGroup->Set_Billboard(sys->shouldBillboard());
+					m_pointGroup->Set_Point_Frame( 0 );
 
-				//RENDER IT!
-				if( sys->getVolumeParticleDepth() > 1 )
-				{
 					m_pointGroup->RenderVolumeParticle( rinfo, sys->getVolumeParticleDepth() );
-				}
+					m_onScreenParticleCount += (pointCount - startCount);
+					pointCount = startCount;
+				}				
 				else
-					m_pointGroup->Render( rinfo );
+				{
+					if ( m_batchTexture == nullptr )
+					{
+						m_batchTexture    = texture;
+						m_batchShaderType = sys->getShaderType();
+						m_batchBillboard  = sys->shouldBillboard();
+					}
+					else
+					{
+						texture->Release_Ref();	// same key as the pending batch so drop the duplicate ref
+					}
 
+					if ( pointCount >= MAX_POINTS_PER_GROUP )
+					{
+						flushParticleBatch(rinfo, pointCount);
+					}
+				}
+			}
+			else
+			{
+				texture->Release_Ref();
 			}
 		}
 
 
 		/// @todo lorenzen sez: this should be debug only:
 		//add particle count to total
-		m_onScreenParticleCount += pointCount;
+		m_onScreenParticleCount += (pointCount - startCount);
 
 	/*
 		// draw the wind vector for this particle system on the screen
@@ -364,6 +436,9 @@ void W3DParticleSystemManager::doParticles(RenderInfoClass &rinfo)
 
 	}
 
+	// TheSuperHackers @info Flush the last batch if one is pending
+	flushParticleBatch(rinfo, pointCount);
+
 		/// @todo lorenzen sez: this should be debug only:
 	TheParticleSystemManager->setOnScreenParticleCount(m_onScreenParticleCount);
 
@@ -377,3 +452,42 @@ void W3DParticleSystemManager::doParticles(RenderInfoClass &rinfo)
 		((W3DSmudgeManager *)TheSmudgeManager)->render(rinfo);
 	}
 }
+
+void W3DParticleSystemManager::flushParticleBatch(RenderInfoClass& rinfo, UnsignedInt& pointCount)
+{
+	if (pointCount > 0 && m_batchTexture != nullptr && m_pointGroup != nullptr)
+	{
+		m_pointGroup->Set_Texture(m_batchTexture);
+
+		switch (m_batchShaderType)
+		{
+		case ParticleSystemInfo::ADDITIVE:
+			m_pointGroup->Set_Shader(ShaderClass::_PresetAdditiveSpriteShader);
+			break;
+		case ParticleSystemInfo::ALPHA:
+			m_pointGroup->Set_Shader(ShaderClass::_PresetAlphaSpriteShader);
+			break;
+		case ParticleSystemInfo::ALPHA_TEST:
+			m_pointGroup->Set_Shader(ShaderClass::_PresetATestSpriteShader);
+			break;
+		case ParticleSystemInfo::MULTIPLY:
+			m_pointGroup->Set_Shader(ShaderClass::_PresetMultiplicativeSpriteShader);
+			break;
+		}
+
+		m_pointGroup->Set_Flag(PointGroupClass::TRANSFORM, true);
+		m_pointGroup->Set_Point_Mode(PointGroupClass::QUADS);
+		m_pointGroup->Set_Arrays(m_posBuffer, m_RGBABuffer, nullptr, m_sizeBuffer, m_angleBuffer, nullptr, pointCount);
+		m_pointGroup->Set_Billboard(m_batchBillboard);
+		m_pointGroup->Set_Point_Frame(0);
+		m_pointGroup->Render(rinfo);
+	}
+
+	if (m_batchTexture != nullptr)
+	{
+		m_batchTexture->Release_Ref();
+		m_batchTexture = nullptr;
+	}
+
+	pointCount = 0;
+}

Generals/Code/GameEngineDevice/Include/W3DDevice/GameClient/W3DParticleSys.h

@@ -50,13 +50,19 @@ class W3DParticleSystemManager : public ParticleSystemManager
 	virtual Int getOnScreenParticleCount() override { return m_onScreenParticleCount; }
 
 private:
+	void flushParticleBatch(RenderInfoClass& rinfo, UnsignedInt& pointCount);
+
 	enum { MAX_POINTS_PER_GROUP = 512 };
 
+	TextureClass *m_batchTexture;								///< the texture used as the drawing surface for batched particle draws
 	PointGroupClass *m_pointGroup;							///< the point group that contains all of the particles
 	StreakLineClass *m_streakLine;							///< the streak class that contains all of the streaks
 	ShareBufferClass<Vector3> *m_posBuffer;			///< array of particle positions
 	ShareBufferClass<Vector4> *m_RGBABuffer;		///< array of particle color and alpha
 	ShareBufferClass<float> *m_sizeBuffer;			///< array of particle sizes
 	ShareBufferClass<uint8> *m_angleBuffer;			///< array of particle orientations
+
+	ParticleSystemInfo::ParticleShaderType m_batchShaderType;
 	Bool m_readyToRender;											///< if true, it is OK to render
+	Bool m_batchBillboard;
 };

GeneralsMD/Code/GameEngineDevice/Include/W3DDevice/GameClient/W3DParticleSys.h

@@ -50,13 +50,19 @@ class W3DParticleSystemManager : public ParticleSystemManager
 	virtual Int getOnScreenParticleCount() override { return m_onScreenParticleCount; }
 
 private:
+	void flushParticleBatch(RenderInfoClass& rinfo, UnsignedInt& pointCount);
+
 	enum { MAX_POINTS_PER_GROUP = 512 };
 
+	TextureClass *m_batchTexture;								///< the texture used as the drawing surface for batched particle draws
 	PointGroupClass *m_pointGroup;							///< the point group that contains all of the particles
 	StreakLineClass *m_streakLine;							///< the streak class that contains all of the streaks
 	ShareBufferClass<Vector3> *m_posBuffer;			///< array of particle positions
 	ShareBufferClass<Vector4> *m_RGBABuffer;		///< array of particle color and alpha
 	ShareBufferClass<float> *m_sizeBuffer;			///< array of particle sizes
 	ShareBufferClass<uint8> *m_angleBuffer;			///< array of particle orientations
+
+	ParticleSystemInfo::ParticleShaderType m_batchShaderType;
 	Bool m_readyToRender;											///< if true, it is OK to render
+	Bool m_batchBillboard;
 };