3 files changed, 219 insertions, 32 deletions

diff --git a/src/Generating/Noise3DGenerator.cpp b/src/Generating/Noise3DGenerator.cpp
index d051948c4..78b739d32 100644
--- a/src/Generating/Noise3DGenerator.cpp
+++ b/src/Generating/Noise3DGenerator.cpp
@@ -69,6 +69,15 @@ static class cInterpolNoiseSpeedTest
 public:
 	cInterpolNoiseSpeedTest(void)
 	{
+		TestSpeed2D();
+		TestSpeed3D();
+		printf("InterpolNoise speed comparison finished.\n");
+	}
+
+
+	/** Compare the speed of the 3D InterpolNoise vs 3D CubicNoise. */
+	void TestSpeed3D(void)
+	{
 		printf("Evaluating 3D noise performance...\n");
 		static const int SIZE_X = 128;
 		static const int SIZE_Y = 128;
@@ -99,6 +108,38 @@ public:
 		printf("3D noise performance comparison finished.\n");
 	}
 
+
+	/** Compare the speed of the 2D InterpolNoise vs 2D CubicNoise. */
+	void TestSpeed2D(void)
+	{
+		printf("Evaluating 2D noise performance...\n");
+		static const int SIZE_X = 128;
+		static const int SIZE_Y = 128;
+		static const NOISE_DATATYPE MUL = 80;
+		std::unique_ptr<NOISE_DATATYPE[]> arr(new NOISE_DATATYPE[SIZE_X * SIZE_Y]);
+		cTimer timer;
+
+		// Test the cInterpolNoise:
+		cInterpolNoise<Interp5Deg> interpNoise(1);
+		long long start = timer.GetNowTime();
+		for (int i = 0; i < 500; i++)
+		{
+			interpNoise.Generate2D(arr.get(), SIZE_X, SIZE_Y, MUL * i, MUL * i + MUL, 0, MUL);
+		}
+		long long end = timer.GetNowTime();
+		printf("InterpolNoise took %.02f sec\n", static_cast<float>(end - start) / 1000);
+
+		// Test the cCubicNoise:
+		cCubicNoise cubicNoise(1);
+		start = timer.GetNowTime();
+		for (int i = 0; i < 500; i++)
+		{
+			cubicNoise.Generate2D(arr.get(), SIZE_X, SIZE_Y, MUL * i, MUL * i + MUL, 0, MUL);
+		}
+		end = timer.GetNowTime();
+		printf("CubicNoise took %.02f sec\n", static_cast<float>(end - start) / 1000);
+		printf("2D noise performance comparison finished.\n");
+	}
 } g_InterpolNoiseSpeedTest;
 #endif
 
@@ -120,6 +161,12 @@ cNoise3DGenerator::cNoise3DGenerator(cChunkGenerator & a_ChunkGenerator) :
 	m_Perlin.AddOctave(8,  0.125);
 	m_Perlin.AddOctave(16, 0.0625);
 
+	m_Cubic.AddOctave(1,  1);
+	m_Cubic.AddOctave(2,  0.5);
+	m_Cubic.AddOctave(4,  0.25);
+	m_Cubic.AddOctave(8,  0.125);
+	m_Cubic.AddOctave(16, 0.0625);
+
 	#if 0
 	// DEBUG: Test the noise generation:
 	// NOTE: In order to be able to run MCS with this code, you need to increase the default thread stack size
@@ -201,8 +248,8 @@ void cNoise3DGenerator::Initialize(cIniFile & a_IniFile)
 {
 	// Params:
 	m_SeaLevel            =                 a_IniFile.GetValueSetI("Generator", "Noise3DSeaLevel", 62);
-	m_HeightAmplification = (NOISE_DATATYPE)a_IniFile.GetValueSetF("Generator", "Noise3DHeightAmplification", 0);
-	m_MidPoint            = (NOISE_DATATYPE)a_IniFile.GetValueSetF("Generator", "Noise3DMidPoint", 75);
+	m_HeightAmplification = (NOISE_DATATYPE)a_IniFile.GetValueSetF("Generator", "Noise3DHeightAmplification", 0.1);
+	m_MidPoint            = (NOISE_DATATYPE)a_IniFile.GetValueSetF("Generator", "Noise3DMidPoint", 68);
 	m_FrequencyX          = (NOISE_DATATYPE)a_IniFile.GetValueSetF("Generator", "Noise3DFrequencyX", 8);
 	m_FrequencyY          = (NOISE_DATATYPE)a_IniFile.GetValueSetF("Generator", "Noise3DFrequencyY", 8);
 	m_FrequencyZ          = (NOISE_DATATYPE)a_IniFile.GetValueSetF("Generator", "Noise3DFrequencyZ", 8);
@@ -280,23 +327,23 @@ void cNoise3DGenerator::GenerateNoiseArray(int a_ChunkX, int a_ChunkZ, NOISE_DAT
 
 	// Precalculate a "height" array:
 	NOISE_DATATYPE Height[DIM_X * DIM_Z];  // Output for the cubic noise heightmap ("source")
-	m_Cubic.Generate2D(Height, DIM_X, DIM_Z, StartX / 25, EndX / 25, StartZ / 25, EndZ / 25);
+	m_Cubic.Generate2D(Height, DIM_X, DIM_Z, StartX / 5, EndX / 5, StartZ / 5, EndZ / 5);
 	for (size_t i = 0; i < ARRAYCOUNT(Height); i++)
 	{
-		Height[i] = std::abs(Height[i]) * m_HeightAmplification + 1;
+		Height[i] = Height[i] * m_HeightAmplification;
 	}
 
 	// Modify the noise by height data:
 	for (int y = 0; y < DIM_Y; y++)
 	{
-		NOISE_DATATYPE AddHeight = (y * UPSCALE_Y - m_MidPoint) / 20;
-		AddHeight *= AddHeight * AddHeight;
+		NOISE_DATATYPE AddHeight = (y * UPSCALE_Y - m_MidPoint) / 30;
+		// AddHeight *= AddHeight * AddHeight;
 		for (int z = 0; z < DIM_Z; z++)
 		{
 			NOISE_DATATYPE * CurRow = &(NoiseO[y * DIM_X + z * DIM_X * DIM_Y]);
 			for (int x = 0; x < DIM_X; x++)
 			{
-				CurRow[x] += AddHeight / Height[x + DIM_X * z];
+				CurRow[x] += AddHeight + Height[x + DIM_X * z];
 			}
 		}
 	}
diff --git a/src/Generating/Noise3DGenerator.h b/src/Generating/Noise3DGenerator.h
index d198c5498..07767ba84 100644
--- a/src/Generating/Noise3DGenerator.h
+++ b/src/Generating/Noise3DGenerator.h
@@ -47,7 +47,8 @@ protected:
 	/** The base 3D noise source for the actual composition */
 	cOctavedNoise<cInterp5DegNoise> m_Perlin;
 
-	cCubicNoise  m_Cubic;    // The noise used for heightmap directing
+	/** The noise used for heightmap directing. */
+	cOctavedNoise<cInterp5DegNoise> m_Cubic;
 	
 	int            m_SeaLevel;
 	NOISE_DATATYPE m_HeightAmplification;
diff --git a/src/Noise/InterpolNoise.h b/src/Noise/InterpolNoise.h
index 69e7194c9..683b54563 100644
--- a/src/Noise/InterpolNoise.h
+++ b/src/Noise/InterpolNoise.h
@@ -18,6 +18,134 @@
 
 
 ////////////////////////////////////////////////////////////////////////////////
+// cInterpolCell2D:
+
+template <typename T>
+class cInterpolCell2D
+{
+public:
+	cInterpolCell2D(
+		const cNoise & a_Noise,    ///< Noise to use for generating the random values
+		NOISE_DATATYPE * a_Array,  ///< Array to generate into [x + a_SizeX * y]
+		int a_SizeX, int a_SizeY,  ///< Count of the array, in each direction
+		const NOISE_DATATYPE * a_FracX,  ///< Pointer to the array that stores the X fractional values
+		const NOISE_DATATYPE * a_FracY   ///< Pointer to the attay that stores the Y fractional values
+	):
+		m_Noise(a_Noise),
+		m_WorkRnds(&m_Workspace1),
+		m_CurFloorX(0),
+		m_CurFloorY(0),
+		m_Array(a_Array),
+		m_SizeX(a_SizeX),
+		m_SizeY(a_SizeY),
+		m_FracX(a_FracX),
+		m_FracY(a_FracY)
+	{
+	}
+	
+
+	/** Generates part of the output noise array using the current m_WorkRnds[] values */
+	void Generate(
+		int a_FromX, int a_ToX,
+		int a_FromY, int a_ToY
+	)
+	{
+		for (int y = a_FromY; y < a_ToY; y++)
+		{
+			NOISE_DATATYPE Interp[2];
+			NOISE_DATATYPE FracY = T::coeff(m_FracY[y]);
+			Interp[0] = Lerp((*m_WorkRnds)[0][0], (*m_WorkRnds)[0][1], FracY);
+			Interp[1] = Lerp((*m_WorkRnds)[1][0], (*m_WorkRnds)[1][1], FracY);
+			int idx = y * m_SizeX + a_FromX;
+			for (int x = a_FromX; x < a_ToX; x++)
+			{
+				m_Array[idx++] = Lerp(Interp[0], Interp[1], T::coeff(m_FracX[x]));
+			}  // for x
+		}  // for y
+	}
+	
+
+	/** Initializes m_WorkRnds[] with the specified values of the noise at the specified integral coords. */
+	void InitWorkRnds(int a_FloorX, int a_FloorY)
+	{
+		m_CurFloorX = a_FloorX;
+		m_CurFloorY = a_FloorY;
+		(*m_WorkRnds)[0][0] = m_Noise.IntNoise2D(m_CurFloorX,     m_CurFloorY);
+		(*m_WorkRnds)[0][1] = m_Noise.IntNoise2D(m_CurFloorX,     m_CurFloorY + 1);
+		(*m_WorkRnds)[1][0] = m_Noise.IntNoise2D(m_CurFloorX + 1, m_CurFloorY);
+		(*m_WorkRnds)[1][1] = m_Noise.IntNoise2D(m_CurFloorX + 1, m_CurFloorY + 1);
+	}
+	
+
+	/** Updates m_WorkRnds[] for the new integral coords */
+	void Move(int a_NewFloorX, int a_NewFloorY)
+	{
+		// Swap the doublebuffer:
+		int OldFloorX = m_CurFloorX;
+		int OldFloorY = m_CurFloorY;
+		Workspace * OldWorkRnds = m_WorkRnds;
+		m_WorkRnds = (m_WorkRnds == &m_Workspace1) ? &m_Workspace2 : &m_Workspace1;
+	
+		// Reuse as much of the old workspace as possible:
+		// TODO: Try out if simply calculating all 4 elements each time is faster than this monster loop
+		int DiffX = OldFloorX - a_NewFloorX;
+		int DiffY = OldFloorY - a_NewFloorY;
+		for (int x = 0; x < 2; x++)
+		{
+			int cx = a_NewFloorX + x;
+			int OldX = x - DiffX;  // Where would this X be in the old grid?
+			for (int y = 0; y < 2; y++)
+			{
+				int cy = a_NewFloorY + y;
+				int OldY = y - DiffY;  // Where would this Y be in the old grid?
+				if ((OldX >= 0) && (OldX < 2) && (OldY >= 0) && (OldY < 2))
+				{
+					(*m_WorkRnds)[x][y] = (*OldWorkRnds)[OldX][OldY];
+				}
+				else
+				{
+					(*m_WorkRnds)[x][y] = (NOISE_DATATYPE)m_Noise.IntNoise2D(cx, cy);
+				}
+			}
+		}
+		m_CurFloorX = a_NewFloorX;
+		m_CurFloorY = a_NewFloorY;
+	}
+
+protected:
+	typedef NOISE_DATATYPE Workspace[2][2];
+	
+	/** The noise used for generating the values at integral coords. */
+	const cNoise & m_Noise;
+	
+	/** The current random values; points to either m_Workspace1 or m_Workspace2 (doublebuffering) */
+	Workspace * m_WorkRnds;
+
+	/** Buffer 1 for workspace doublebuffering, used in Move() */
+	Workspace m_Workspace1;
+
+	/** Buffer 2 for workspace doublebuffering, used in Move() */
+	Workspace m_Workspace2;
+
+	/** Coords of the currently calculated m_WorkRnds[]. */
+	int m_CurFloorX, m_CurFloorY;
+	
+	/** The output array to generate into. */
+	NOISE_DATATYPE * m_Array;
+
+	/** Dimensions of the output array. */
+	int m_SizeX, m_SizeY;
+
+	/** Arrays holding the fractional values of the coords in each direction. */
+	const NOISE_DATATYPE * m_FracX;
+	const NOISE_DATATYPE * m_FracY;
+} ;
+
+
+
+
+
+////////////////////////////////////////////////////////////////////////////////
 // cInterpolCell3D:
 
 /** Holds a cache of the last calculated integral noise values and interpolates between them en masse.
@@ -212,33 +340,44 @@ public:
 		NOISE_DATATYPE a_StartY, NOISE_DATATYPE a_EndY   ///< Noise-space coords of the array in the Y direction
 	) const
 	{
-		// Check params:
-		ASSERT(a_SizeX > 1);
-		ASSERT(a_SizeY > 1);
-
-		// Generate the noise:
-		size_t idx = 0;
-		for (int y = 0; y < a_SizeY; y++)
+		ASSERT(a_SizeX > 0);
+		ASSERT(a_SizeY > 0);
+		ASSERT(a_SizeX < MAX_SIZE);
+		ASSERT(a_SizeY < MAX_SIZE);
+		ASSERT(a_StartX < a_EndX);
+		ASSERT(a_StartY < a_EndY);
+	
+		// Calculate the integral and fractional parts of each coord:
+		int FloorX[MAX_SIZE];
+		int FloorY[MAX_SIZE];
+		NOISE_DATATYPE FracX[MAX_SIZE];
+		NOISE_DATATYPE FracY[MAX_SIZE];
+		int SameX[MAX_SIZE];
+		int SameY[MAX_SIZE];
+		int NumSameX, NumSameY;
+		CalcFloorFrac(a_SizeX, a_StartX, a_EndX, FloorX, FracX, SameX, NumSameX);
+		CalcFloorFrac(a_SizeY, a_StartY, a_EndY, FloorY, FracY, SameY, NumSameY);
+	
+		cInterpolCell2D<T> Cell(m_Noise, a_Array, a_SizeX, a_SizeY, FracX, FracY);
+	
+		Cell.InitWorkRnds(FloorX[0], FloorY[0]);
+	
+		// Calculate query values using Cell:
+		int FromY = 0;
+		for (int y = 0; y < NumSameY; y++)
 		{
-			NOISE_DATATYPE ratioY = static_cast<NOISE_DATATYPE>(y) / (a_SizeY - 1);
-			NOISE_DATATYPE noiseY = Lerp(a_StartY, a_EndY, ratioY);
-			int noiseYInt = FAST_FLOOR(noiseY);
-			NOISE_DATATYPE ratioY = typename T(noiseY - static_cast<NOISE_DATATYPE>(noiseYInt));
-			for (int x = 0; x < a_SizeX; x++)
+			int ToY = FromY + SameY[y];
+			int FromX = 0;
+			int CurFloorY = FloorY[FromY];
+			for (int x = 0; x < NumSameX; x++)
 			{
-				NOISE_DATATYPE ratioX = static_cast<NOISE_DATATYPE>(x) / (a_SizeX - 1);
-				NOISE_DATATYPE noiseX = Lerp(a_StartX, a_EndX, ratioX);
-				int noiseXInt = FAST_FLOOR(noiseX);
-				NOISE_DATATYPE ratioX = typename T(noiseX - static_cast<NOISE_DATATYPE>(noiseXInt));
-
-				NOISE_DATATYPE valx0y0 = m_Noise.IntNoise2D(noiseXInt, noiseYInt);
-				NOISE_DATATYPE valx1y0 = m_Noise.IntNoise2D(noiseXInt + 1, noiseYInt);
-				NOISE_DATATYPE valx0y1 = m_Noise.IntNoise2D(noiseXInt, noiseYInt + 1);
-				NOISE_DATATYPE valx1y1 = m_Noise.IntNoise2D(noiseXInt + 1, noiseYInt + 1);
-				NOISE_DATATYPE valx0 = Lerp(valx0y0, valx0y1, ratioY);
-				NOISE_DATATYPE valx1 = Lerp(valx1y1, valx1y1, ratioY);
-				a_Array[idx++] = Lerp(valx0, valx1, ratioX);
+				int ToX = FromX + SameX[x];
+				Cell.Generate(FromX, ToX, FromY, ToY);
+				Cell.Move(FloorX[ToX], CurFloorY);
+				FromX = ToX;
 			}  // for x
+			Cell.Move(FloorX[0], FloorY[ToY]);
+			FromY = ToY;
 		}  // for y
 	}