Cull geometry based on the culling distances in the nodes. Should give some sort of speedup on lower end pcs where the CPU is predominantly the bottle neck.

Src/Graphics/Legacy3D/Models.cpp

@@ -473,8 +473,8 @@ void CLegacy3D::InsertVertex(ModelCache *Cache, const Vertex *V, const Poly *P,
   GLfloat   texWidth  = (GLfloat) (32<<((P->header[3]>>3)&7));
   GLfloat   texHeight = (GLfloat) (32<<((P->header[3]>>0)&7));
   TexSheet  *texSheet = fmtToTexSheet[texFormat]; // get X, Y offset of texture sheet within texture map
-  GLfloat   texBaseX = texSheet->xOffset + GetTextureBaseX(P);
+  GLfloat   texBaseX = (GLfloat)(texSheet->xOffset + GetTextureBaseX(P));
-  GLfloat   texBaseY = texSheet->yOffset + GetTextureBaseY(P);
+  GLfloat   texBaseY = (GLfloat)(texSheet->yOffset + GetTextureBaseY(P));
 
   /*
    * Lighting and Color Modulation:

Src/Graphics/New3D/Model.cpp

@@ -4,9 +4,10 @@ namespace New3D {
 
 NodeAttributes::NodeAttributes()
 {
-	currentTexOffsetX = 0;
+	currentTexOffsetX	= 0;
-	currentTexOffsetY = 0;
+	currentTexOffsetY	= 0;
-	currentPage = 0;
+	currentPage			= 0;
+	currentClipStatus	= Clip::INTERCEPT;
 }
 
 bool NodeAttributes::Push()
@@ -20,9 +21,10 @@ bool NodeAttributes::Push()
 		return false;
 	}
 
-	na.page = currentPage;
+	na.page			= currentPage;
-	na.texOffsetX = currentTexOffsetX;
+	na.texOffsetX	= currentTexOffsetX;
-	na.texOffsetY = currentTexOffsetY;
+	na.texOffsetY	= currentTexOffsetY;
+	na.clip			= currentClipStatus;
 
 	m_vecAttribs.push_back(na);
 
@@ -37,9 +39,10 @@ bool NodeAttributes::Pop()
 
 	auto last = &m_vecAttribs.back();
 
-	currentPage = last->page;
+	currentPage			= last->page;
-	currentTexOffsetX = last->texOffsetX;
+	currentTexOffsetX	= last->texOffsetX;
-	currentTexOffsetY = last->texOffsetY;
+	currentTexOffsetY	= last->texOffsetY;
+	currentClipStatus	= last->clip;
 
 	m_vecAttribs.pop_back();
 
@@ -53,9 +56,11 @@ bool NodeAttributes::StackLimit()
 
 void NodeAttributes::Reset()
 {
-	currentPage = 0;
+	currentPage			= 0;
-	currentTexOffsetX = 0;
+	currentTexOffsetX	= 0;
-	currentTexOffsetY = 0;
+	currentTexOffsetY	= 0;
+	currentClipStatus	= Clip::INTERCEPT;
+
 	m_vecAttribs.clear();
 }
 

Src/Graphics/New3D/Model.h

@@ -100,6 +100,8 @@ struct Viewport
 	int		priority;
 };
 
+enum class Clip { INSIDE, OUTSIDE, INTERCEPT };
+
 class NodeAttributes
 {
 public:
@@ -114,6 +116,7 @@ public:
 	int currentTexOffsetX;
 	int currentTexOffsetY;
 	int currentPage;
+	Clip currentClipStatus;
 
 private:
 
@@ -122,6 +125,7 @@ private:
 		int texOffsetX;
 		int texOffsetY;
 		int page;
+		Clip clip;
 	};
 	std::vector<NodeAttribs> m_vecAttribs;
 };

Src/Graphics/New3D/New3D.cpp

@@ -342,6 +342,7 @@ void CNew3D::DescendCullingNode(UINT32 addr)
 	UINT32			matrixOffset, child1Ptr, sibling2Ptr;
 	float			x, y, z;
 	int				tx, ty;
+	BBox			bbox;
 
 	if (m_nodeAttribs.StackLimit()) {
 		return;
@@ -358,8 +359,6 @@ void CNew3D::DescendCullingNode(UINT32 addr)
 	sibling2Ptr		= node[0x08 - m_offset] & 0x1FFFFFF;	// mask colour table bits
 	matrixOffset	= node[0x03 - m_offset] & 0xFFF;
 
-	float test = ToFloat(Convert16BitProFloat(node[9 - m_offset] >> 16));
-
 	if ((node[0x00] & 0x07) != 0x06) {						// colour table seems to indicate no siblings
 		if (!(sibling2Ptr & 0x1000000) && sibling2Ptr) {
 			DescendCullingNode(sibling2Ptr);				// no need to mask bit, would already be zero
@@ -400,24 +399,38 @@ void CNew3D::DescendCullingNode(UINT32 addr)
 	// multiply matrix, if specified
 	else if (matrixOffset) {
 		MultMatrix(matrixOffset,m_modelMat);
-	}		
+	}
 
-	// Descend down first link
+	if (m_nodeAttribs.currentClipStatus != Clip::INSIDE) {
-	if ((node[0x00] & 0x08))	// 4-element LOD table
-	{
-		lodTable = TranslateCullingAddress(child1Ptr);
 
-		if (NULL != lodTable) {
+		float distance = ToFloat(Convert16BitProFloat(node[9 - m_offset] & 0xFFFF));
-			if ((node[0x03 - m_offset] & 0x20000000)) {
+
-				DescendCullingNode(lodTable[0] & 0xFFFFFF);
+		CalcBox(distance, bbox);
-			}
+		TransformBox(m_modelMat, bbox);
-			else {
+
-				DrawModel(lodTable[0] & 0xFFFFFF);	//TODO
+		m_nodeAttribs.currentClipStatus = ClipBox(bbox, m_planes);
+	}
+
+	if (m_nodeAttribs.currentClipStatus != Clip::OUTSIDE) {
+
+		// Descend down first link
+		if ((node[0x00] & 0x08))	// 4-element LOD table
+		{
+			lodTable = TranslateCullingAddress(child1Ptr);
+
+			if (NULL != lodTable) {
+				if ((node[0x03 - m_offset] & 0x20000000)) {
+					DescendCullingNode(lodTable[0] & 0xFFFFFF);
+				}
+				else {
+					DrawModel(lodTable[0] & 0xFFFFFF);	//TODO
+				}
 			}
 		}
-	}
+		else {
-	else {
+			DescendNodePtr(child1Ptr);
-		DescendNodePtr(child1Ptr);
+		}
+
 	}
 
 	m_modelMat.PopMatrix();
@@ -671,6 +684,9 @@ void CNew3D::RenderViewport(UINT32 addr)
 			vp->projectionMatrix.Frustum(l, r, b, t, near, far);
 		}
 
+		// calculate frustum planes
+		CalcFrustumPlanes(m_planes, vp->projectionMatrix);
+
 		// Lighting (note that sun vector points toward sun -- away from vertex)
 		vp->lightingParams[0] = *(float *)&vpnode[0x05];								// sun X
 		vp->lightingParams[1] = *(float *)&vpnode[0x06];								// sun Y
@@ -1209,5 +1225,165 @@ float CNew3D::ToFloat(UINT32 a1)
 	return *(float*)(&a1);
 }
 
+void CNew3D::CalcFrustumPlanes(Plane p[6], const float* matrix)
+{
+	// Left Plane
+	p[0].a = matrix[3] + matrix[0];
+	p[0].b = matrix[7] + matrix[4];
+	p[0].c = matrix[11] + matrix[8];
+	p[0].d = matrix[15] + matrix[12];
+	p[0].Normalise();
+
+	// Right Plane
+	p[1].a = matrix[3] - matrix[0];
+	p[1].b = matrix[7] - matrix[4];
+	p[1].c = matrix[11] - matrix[8];
+	p[1].d = matrix[15] - matrix[12];
+	p[1].Normalise();
+
+	// Bottom Plane
+	p[2].a = matrix[3] + matrix[1];
+	p[2].b = matrix[7] + matrix[5];
+	p[2].c = matrix[11] + matrix[9];
+	p[2].d = matrix[15] + matrix[13];
+	p[2].Normalise();
+
+	// Top Plane
+	p[3].a = matrix[3] - matrix[1];
+	p[3].b = matrix[7] - matrix[5];
+	p[3].c = matrix[11] - matrix[9];
+	p[3].d = matrix[15] - matrix[13];
+	p[3].Normalise();
+
+	// Near Plane
+	p[4].a = matrix[3] + matrix[2];
+	p[4].b = matrix[7] + matrix[6];
+	p[4].c = matrix[11] + matrix[10];
+	p[4].d = matrix[15] + matrix[14];
+	p[4].Normalise();
+
+	// Far Plane
+	p[5].a = matrix[3] - matrix[2];
+	p[5].b = matrix[7] - matrix[6];
+	p[5].c = matrix[11] - matrix[10];
+	p[5].d = matrix[15] - matrix[14];
+	p[5].Normalise();
+}
+
+void CNew3D::CalcBox(float distance, BBox& box)
+{
+	//bottom left front
+	box.points[0][0] = -distance;
+	box.points[0][1] = -distance;
+	box.points[0][2] = distance;
+	box.points[0][3] = 1;
+
+	//bottom left back
+	box.points[1][0] = -distance;
+	box.points[1][1] = -distance;
+	box.points[1][2] = -distance;
+	box.points[1][3] = 1;
+
+	//bottom right back
+	box.points[2][0] = distance;
+	box.points[2][1] = -distance;
+	box.points[2][2] = -distance;
+	box.points[2][3] = 1;
+
+	//bottom right front
+	box.points[3][0] = distance;
+	box.points[3][1] = -distance;
+	box.points[3][2] = distance;
+	box.points[3][3] = 1;
+
+	//top left front
+	box.points[4][0] = -distance;
+	box.points[4][1] = distance;
+	box.points[4][2] = distance;
+	box.points[4][3] = 1;
+
+	//top left back
+	box.points[5][0] = -distance;
+	box.points[5][1] = distance;
+	box.points[5][2] = -distance;
+	box.points[5][3] = 1;
+
+	//top right back
+	box.points[6][0] = distance;
+	box.points[6][1] = distance;
+	box.points[6][2] = -distance;
+	box.points[6][3] = 1;
+
+	//top right front
+	box.points[7][0] = distance;
+	box.points[7][1] = distance;
+	box.points[7][2] = distance;
+	box.points[7][3] = 1;
+}
+
+void CNew3D::MultVec(const float matrix[16], const float in[4], float out[4]) 
+{
+	for (int i = 0; i < 4; i++) {
+		out[i] =
+			in[0] * matrix[0 * 4 + i] +
+			in[1] * matrix[1 * 4 + i] +
+			in[2] * matrix[2 * 4 + i] +
+			in[3] * matrix[3 * 4 + i];
+	}
+}
+
+void CNew3D::TransformBox(const float *m, BBox& box)
+{
+	for (int i = 0; i < 8; i++) {
+		float v[4];
+		MultVec(m, box.points[i], v);
+		box.points[i][0] = v[0];
+		box.points[i][1] = v[1];
+		box.points[i][2] = v[2];
+	}
+}
+
+Clip CNew3D::ClipBox(BBox& box, Plane planes[6])
+{
+	int count = 0;
+
+	for (int i = 0; i < 8; i++) {
+
+		int temp = 0;
+
+		for (int j = 0; j < 6; j++) {
+			if (planes[j].DistanceToPoint(box.points[i]) >= 0) {
+				temp++;
+			}
+		}
+
+		if (temp == 6) count++;		// point is inside all 6 frustum planes
+	}
+
+	if (count == 8)	return Clip::INSIDE;
+	if (count > 0)	return Clip::INTERCEPT;
+	
+	//if we got here all points are outside of the view frustum
+	//check for all points being side same of any plane, means box outside of view
+
+	for (int i = 0; i < 6; i++) {
+
+		int temp = 0;
+
+		for (int j = 0; j < 8; j++) {
+			if (planes[i].DistanceToPoint(box.points[j]) >= 0) {
+				float distance = planes[i].DistanceToPoint(box.points[j]);
+				temp++;
+			}
+		}
+
+		if (temp == 0) return Clip::OUTSIDE;
+	}
+
+	//if we got here, box is traversing view frustum
+
+	return Clip::INTERCEPT;
+}
+
 } // New3D
 

Src/Graphics/New3D/New3D.h

@@ -37,6 +37,8 @@
 #include "R3DShader.h"
 #include "VBO.h"
 #include "R3DData.h"
+#include "Plane.h"
+#include "Vec.h"
 
 namespace New3D {
 
@@ -218,6 +220,19 @@ private:
 
 	VBO m_vbo;								// large VBO to hold our poly data, start of VBO is ROM data, ram polys follow
 	R3DShader m_r3dShader;
+
+	Plane m_planes[6];
+
+	struct BBox
+	{
+		V4::Vec4 points[8];
+	};
+
+	void CalcFrustumPlanes	(Plane p[6], const float* matrix);
+	void CalcBox			(float distance, BBox& box);
+	void TransformBox		(const float *m, BBox& box);
+	void MultVec			(const float matrix[16], const float in[4], float out[4]);
+	Clip ClipBox			(BBox& box, Plane planes[6]);
 };
 
 } // New3D

Src/Graphics/New3D/Plane.h

@@ -0,0 +1,23 @@
+#ifndef _PLANE_H_
+#define _PLANE_H_
+
+#include <cmath>
+
+struct Plane
+{
+	float a, b, c, d;
+
+	void Normalise() {
+		float temp = std::sqrt((a * a) + (b * b) + (c * c));
+		a /= temp;
+		b /= temp;
+		c /= temp;
+		d /= temp;
+	}
+
+	float DistanceToPoint(const float v[3]) {
+		return a*v[0] + b*v[1] + c*v[2] + d;
+	}
+};
+
+#endif

Src/Graphics/New3D/Vec.h

@@ -32,6 +32,11 @@ namespace V3
 	bool	cmp				(const Vec3 a, const Vec3 b);
 	void	clamp			(Vec3 a, float _min, float _max);
 }
+
+namespace V4
+{
+	typedef float Vec4[4];
+}
 } // New3D
 
 #endif

VS2008/Supermodel.vcxproj

@@ -535,6 +535,7 @@ xcopy /D /Y "$(ProjectDir)\SDL\$(Platform)\$(Configuration)\SDL.dll" "$(TargetDi
     <ClInclude Include="..\Src\Graphics\New3D\Mat4.h" />
     <ClInclude Include="..\Src\Graphics\New3D\Model.h" />
     <ClInclude Include="..\Src\Graphics\New3D\New3D.h" />
+    <ClInclude Include="..\Src\Graphics\New3D\Plane.h" />
     <ClInclude Include="..\Src\Graphics\New3D\PolyHeader.h" />
     <ClInclude Include="..\Src\Graphics\New3D\R3DData.h" />
     <ClInclude Include="..\Src\Graphics\New3D\R3DShader.h" />

VS2008/Supermodel.vcxproj.filters

@@ -784,6 +784,9 @@
     <ClInclude Include="..\Src\Util\BMPFile.h">
       <Filter>Header Files\Util</Filter>
     </ClInclude>
+    <ClInclude Include="..\Src\Graphics\New3D\Plane.h">
+      <Filter>Source Files\Graphics\New</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <CustomBuild Include="..\Src\Debugger\ReadMe.txt">