6 files changed, 736 insertions, 0 deletions
diff --git a/Tools/NoiseSpeedTest/CMakeLists.txt b/Tools/NoiseSpeedTest/CMakeLists.txt
new file mode 100644
index 000000000..29163398b
--- /dev/null
+++ b/Tools/NoiseSpeedTest/CMakeLists.txt
@@ -0,0 +1,66 @@
+project (NoiseSpeedTest)
+
+include(../../SetFlags.cmake)
+set_flags()
+set_lib_flags()
+enable_profile()
+
+# Set include paths to the used libraries:
+include_directories(SYSTEM "../../lib")
+include_directories("../../src")
+
+set_exe_flags()
+
+# Include the shared files:
+set(SHARED_SRC
+	../../src/Logger.cpp
+	../../src/LoggerListeners.cpp
+	../../src/OSSupport/CriticalSection.cpp
+	../../src/OSSupport/File.cpp
+	../../src/OSSupport/StackTrace.cpp
+	../../src/Noise/Noise.cpp
+	../../src/StringUtils.cpp
+)
+set(SHARED_HDR
+	../../src/Noise/Noise.h
+	../../src/Noise/OctavedNoise.h
+	../../src/Noise/RidgedNoise.h
+	../../src/OSSupport/CriticalSection.h
+	../../src/OSSupport/File.h
+	../../src/OSSupport/StackTrace.h
+	../../src/StringUtils.h
+)
+
+if(WIN32)
+	list (APPEND SHARED_SRC ../../src/StackWalker.cpp)
+	list (APPEND SHARED_HDR ../../src/StackWalker.h)
+endif()
+
+source_group("Shared" FILES ${SHARED_SRC} ${SHARED_HDR})
+
+
+
+
+# Include the main source files:
+set(SOURCES
+	NoiseSpeedTest.cpp
+	Globals.cpp
+)
+set(HEADERS
+	NoiseSpeedTest.h
+	Globals.h
+)
+
+source_group("" FILES ${SOURCES} ${HEADERS})
+
+add_executable(NoiseSpeedTest
+	${SOURCES}
+	${HEADERS}
+	${SHARED_SRC}
+	${SHARED_HDR}
+)
+
+set_target_properties(
+	NoiseSpeedTest
+	PROPERTIES FOLDER Tools
+)
diff --git a/Tools/NoiseSpeedTest/Globals.cpp b/Tools/NoiseSpeedTest/Globals.cpp
new file mode 100644
index 000000000..13c6ae709
--- /dev/null
+++ b/Tools/NoiseSpeedTest/Globals.cpp
@@ -0,0 +1,10 @@
+
+// Globals.cpp
+
+// This file is used for precompiled header generation in MSVC environments
+
+#include "Globals.h"
+
+
+
+
diff --git a/Tools/NoiseSpeedTest/Globals.h b/Tools/NoiseSpeedTest/Globals.h
new file mode 100644
index 000000000..ed9ef82fe
--- /dev/null
+++ b/Tools/NoiseSpeedTest/Globals.h
@@ -0,0 +1,243 @@
+
+// Globals.h
+
+// This file gets included from every module in the project, so that global symbols may be introduced easily
+// Also used for precompiled header generation in MSVC environments
+
+
+
+
+
+// Compiler-dependent stuff:
+#if defined(_MSC_VER)
+	// MSVC produces warning C4481 on the override keyword usage, so disable the warning altogether
+	#pragma warning(disable:4481)
+	
+	// Disable some warnings that we don't care about:
+	#pragma warning(disable:4100)
+
+	#define OBSOLETE __declspec(deprecated)
+	
+	// No alignment needed in MSVC
+	#define ALIGN_8
+	#define ALIGN_16
+	
+	#define FORMATSTRING(formatIndex, va_argsIndex)
+
+	// MSVC has its own custom version of zu format
+	#define SIZE_T_FMT "%Iu"
+	#define SIZE_T_FMT_PRECISION(x) "%" #x "Iu"
+	#define SIZE_T_FMT_HEX "%Ix"
+	
+	#define NORETURN      __declspec(noreturn)
+
+#elif defined(__GNUC__)
+
+	// TODO: Can GCC explicitly mark classes as abstract (no instances can be created)?
+	#define abstract
+	
+	#define OBSOLETE __attribute__((deprecated))
+
+	#define ALIGN_8 __attribute__((aligned(8)))
+	#define ALIGN_16 __attribute__((aligned(16)))
+
+	// Some portability macros :)
+	#define stricmp strcasecmp
+	
+	#define FORMATSTRING(formatIndex,va_argsIndex)
+
+	#define SIZE_T_FMT "%zu"
+	#define SIZE_T_FMT_PRECISION(x) "%" #x "zu"
+	#define SIZE_T_FMT_HEX "%zx"
+	
+	#define NORETURN      __attribute((__noreturn__))
+#else
+
+	#error "You are using an unsupported compiler, you might need to #define some stuff here for your compiler"
+	
+	/*
+	// Copy and uncomment this into another #elif section based on your compiler identification
+	
+	// Explicitly mark classes as abstract (no instances can be created)
+	#define abstract
+	
+	// Mark functions as obsolete, so that their usage results in a compile-time warning
+	#define OBSOLETE
+
+	// Mark types / variables for alignment. Do the platforms need it?
+	#define ALIGN_8
+	#define ALIGN_16
+	*/
+	
+	#define FORMATSTRING(formatIndex,va_argsIndex) __attribute__((format (printf, formatIndex, va_argsIndex)))
+
+#endif
+
+
+
+
+
+// Integral types with predefined sizes:
+typedef long long Int64;
+typedef int       Int32;
+typedef short     Int16;
+
+typedef unsigned long long UInt64;
+typedef unsigned int       UInt32;
+typedef unsigned short     UInt16;
+
+typedef unsigned char Byte;
+
+
+
+
+
+// A macro to disallow the copy constructor and operator= functions
+// This should be used in the private: declarations for any class that shouldn't allow copying itself
+#define DISALLOW_COPY_AND_ASSIGN(TypeName) \
+	TypeName(const TypeName &); \
+	void operator=(const TypeName &)
+
+// A macro that is used to mark unused function parameters, to avoid pedantic warnings in gcc
+#define UNUSED(X) (void)(X)
+
+
+
+
+// OS-dependent stuff:
+#ifdef _WIN32
+	#define WIN32_LEAN_AND_MEAN
+	#include <Windows.h>
+	#include <winsock2.h>
+	#include <ws2tcpip.h>
+	
+	// Windows SDK defines min and max macros, messing up with our std::min and std::max usage
+	#undef min
+	#undef max
+	
+	// Windows SDK defines GetFreeSpace as a constant, probably a Win16 API remnant
+	#ifdef GetFreeSpace
+		#undef GetFreeSpace
+	#endif  // GetFreeSpace
+	
+	#define SocketError WSAGetLastError()
+#else
+	#include <sys/types.h>
+	#include <sys/stat.h>   // for mkdir
+	#include <sys/time.h>
+	#include <sys/socket.h>
+	#include <netinet/in.h>
+	#include <arpa/inet.h>
+	#include <netdb.h>
+	#include <time.h>
+	#include <dirent.h>
+	#include <errno.h>
+	#include <iostream>
+	#include <unistd.h>
+
+	#include <cstdio>
+	#include <cstring>
+	#include <pthread.h>
+	#include <semaphore.h>
+	#include <errno.h>
+	#include <fcntl.h>
+	
+	typedef int SOCKET;
+	enum
+	{
+		INVALID_SOCKET = -1,
+	};
+	#define closesocket close
+	#define SocketError errno
+#if !defined(ANDROID_NDK)
+	#include <tr1/memory>
+#endif
+#endif
+
+#if !defined(ANDROID_NDK)
+	#define USE_SQUIRREL
+#endif
+
+#if defined(ANDROID_NDK)
+	#define FILE_IO_PREFIX "/sdcard/mcserver/"
+#else
+	#define FILE_IO_PREFIX ""
+#endif
+
+
+
+
+
+// CRT stuff:
+#include <assert.h>
+#include <stdio.h>
+#include <math.h>
+#include <stdarg.h>
+#include <time.h>
+
+
+
+
+
+// STL stuff:
+#include <vector>
+#include <list>
+#include <deque>
+#include <string>
+#include <map>
+#include <algorithm>
+#include <memory>
+
+
+
+
+
+// Common headers (without macros):
+#include "StringUtils.h"
+#include "OSSupport/CriticalSection.h"
+#include "OSSupport/Event.h"
+#include "OSSupport/IsThread.h"
+#include "OSSupport/File.h"
+
+
+
+
+
+// Common definitions:
+
+/// Evaluates to the number of elements in an array (compile-time!)
+#define ARRAYCOUNT(X) (sizeof(X) / sizeof(*(X)))
+
+/// Allows arithmetic expressions like "32 KiB" (but consider using parenthesis around it, "(32 KiB)" )
+#define KiB * 1024
+#define MiB * 1024 * 1024
+
+/// Faster than (int)floorf((float)x / (float)div)
+#define FAST_FLOOR_DIV( x, div ) ( (x) < 0 ? (((int)x / div) - 1) : ((int)x / div) )
+
+// Own version of assert() that writes failed assertions to the log for review
+#ifdef  NDEBUG
+	#define ASSERT(x) ((void)0)
+#else
+	#define ASSERT assert
+#endif
+
+// Pretty much the same as ASSERT() but stays in Release builds
+#define VERIFY( x ) ( !!(x) || ( LOGERROR("Verification failed: %s, file %s, line %i", #x, __FILE__, __LINE__ ), exit(1), 0 ) )
+
+
+
+
+
+/// A generic interface used mainly in ForEach() functions
+template <typename Type> class cItemCallback
+{
+public:
+	/// Called for each item in the internal list; return true to stop the loop, or false to continue enumerating
+	virtual bool Item(Type * a_Type) = 0;
+	virtual ~cItemCallback() {}
+} ;
+
+
+
+
diff --git a/Tools/NoiseSpeedTest/NoiseSpeedTest.cpp b/Tools/NoiseSpeedTest/NoiseSpeedTest.cpp
new file mode 100644
index 000000000..6857b1a27
--- /dev/null
+++ b/Tools/NoiseSpeedTest/NoiseSpeedTest.cpp
@@ -0,0 +1,149 @@
+// NoiseSpeedTest.cpp
+
+// Implements the main app entrypoint
+
+/*
+This program compares the performance of the highly-optimized noise implementation in Cuberite, and the Simplex noise.
+Since the Simplex noise is not yet implemented in Cuberite, an own implementation is provided.
+Also, the performance difference between using a float and double as datatype is measured, by using a templatized Simplex noise.
+
+The testing is done on a usage of the generator that is typical for the Cuberite's terrain generator: generate a 3D array of numbers with
+not much variance in the coords. The exact sizes and coord ranges were adapted from the cNoise3DComposable generator.
+*/
+
+#include "Globals.h"
+#include "Noise/Noise.h"
+#include "Noise/InterpolNoise.h"
+#include "SimplexNoise.h"
+
+
+
+
+
+/// The sizes of the interpolated noise that are calculated:
+static const int SIZE_X = 33;
+static const int SIZE_Y = 5;
+static const int SIZE_Z = 5;
+
+
+
+
+
+static void measureClassicNoise(int a_NumIterations)
+{
+	cInterp5DegNoise noise(1);
+	NOISE_DATATYPE total = 0;
+	auto timeStart = std::chrono::high_resolution_clock::now();
+	for (int i = 0; i < a_NumIterations; ++i)
+	{
+		NOISE_DATATYPE out[SIZE_X * SIZE_Y * SIZE_Z];
+		int blockX = i * 16;
+		int blockZ = i * 16;
+		NOISE_DATATYPE startX = 0;
+		NOISE_DATATYPE endX = 257 / 80.0f;
+		NOISE_DATATYPE startY = blockX / 40.0f;
+		NOISE_DATATYPE endY = (blockX + 16) / 40.0f;
+		NOISE_DATATYPE startZ = blockZ / 40.0f;
+		NOISE_DATATYPE endZ = (blockZ + 16) / 40.0f;
+		noise.Generate3D(out, SIZE_X, SIZE_Y, SIZE_Z, startX, endX, startY, endY, startZ, endZ);
+		total += out[0];  // Do not let the optimizer optimize the whole calculation away
+	}
+	auto timeEnd = std::chrono::high_resolution_clock::now();
+	auto msec = std::chrono::duration_cast<std::chrono::milliseconds>(timeEnd - timeStart);
+	printf("Classic noise took %d milliseconds, returned total %f\n", static_cast<int>(msec.count()), total);
+}
+
+
+
+
+
+static void measureSimplexNoiseFloat(int a_NumIterations)
+{
+	typedef float DATATYPE;
+	cSimplexNoise<DATATYPE> noise(1);
+	DATATYPE total = 0;
+	auto timeStart = std::chrono::high_resolution_clock::now();
+	for (int i = 0; i < a_NumIterations; ++i)
+	{
+		DATATYPE out[SIZE_X * SIZE_Y * SIZE_Z];
+		int blockX = i * 16;
+		int blockZ = i * 16;
+		DATATYPE startX = 0;
+		DATATYPE endX = 257 / 80.0f;
+		DATATYPE startY = blockX / 40.0f;
+		DATATYPE endY = (blockX + 16) / 40.0f;
+		DATATYPE startZ = blockZ / 40.0f;
+		DATATYPE endZ = (blockZ + 16) / 40.0f;
+		noise.Generate3D(out, SIZE_X, SIZE_Y, SIZE_Z, startX, endX, startY, endY, startZ, endZ);
+		total += out[0];  // Do not let the optimizer optimize the whole calculation away
+	}
+	auto timeEnd = std::chrono::high_resolution_clock::now();
+	auto msec = std::chrono::duration_cast<std::chrono::milliseconds>(timeEnd - timeStart);
+	printf("SimplexNoise<float> took %d milliseconds, returned total %f\n", static_cast<int>(msec.count()), total);
+}
+
+
+
+
+
+static void measureSimplexNoiseDouble(int a_NumIterations)
+{
+	typedef double DATATYPE;
+	cSimplexNoise<DATATYPE> noise(1);
+	DATATYPE total = 0;
+	auto timeStart = std::chrono::high_resolution_clock::now();
+	for (int i = 0; i < a_NumIterations; ++i)
+	{
+		DATATYPE out[SIZE_X * SIZE_Y * SIZE_Z];
+		int blockX = i * 16;
+		int blockZ = i * 16;
+		DATATYPE startX = 0;
+		DATATYPE endX = 257 / 80.0f;
+		DATATYPE startY = blockX / 40.0f;
+		DATATYPE endY = (blockX + 16) / 40.0f;
+		DATATYPE startZ = blockZ / 40.0f;
+		DATATYPE endZ = (blockZ + 16) / 40.0f;
+		noise.Generate3D(out, SIZE_X, SIZE_Y, SIZE_Z, startX, endX, startY, endY, startZ, endZ);
+		total += out[0];  // Do not let the optimizer optimize the whole calculation away
+	}
+	auto timeEnd = std::chrono::high_resolution_clock::now();
+	auto msec = std::chrono::duration_cast<std::chrono::milliseconds>(timeEnd - timeStart);
+	printf("SimplexNoise<double> took %d milliseconds, returned total %f\n", static_cast<int>(msec.count()), total);
+}
+
+
+
+
+
+int main(int argc, char ** argv)
+{
+	int numIterations = 10000;
+	if (argc > 1)
+	{
+		numIterations = std::atoi(argv[1]);
+		if (numIterations < 10)
+		{
+			printf("Invalid number of iterations, using 1000 instead\n");
+			numIterations = 1000;
+		}
+	}
+
+	// Perform each test twice, to account for cache-warmup:
+	measureClassicNoise(numIterations);
+	measureClassicNoise(numIterations);
+	measureSimplexNoiseFloat(numIterations);
+	measureSimplexNoiseFloat(numIterations);
+	measureSimplexNoiseDouble(numIterations);
+	measureSimplexNoiseDouble(numIterations);
+
+	// If build on Windows using MSVC, wait for a keypress before ending:
+	#ifdef _MSC_VER
+		getchar();
+	#endif
+
+	return 0;
+}
+
+
+
+
diff --git a/Tools/NoiseSpeedTest/NoiseSpeedTest.h b/Tools/NoiseSpeedTest/NoiseSpeedTest.h
new file mode 100644
index 000000000..74a0108e6
--- /dev/null
+++ b/Tools/NoiseSpeedTest/NoiseSpeedTest.h
@@ -0,0 +1,5 @@
+// NoiseSpeedTest.h
+
+
+
+
diff --git a/Tools/NoiseSpeedTest/SimplexNoise.h b/Tools/NoiseSpeedTest/SimplexNoise.h
new file mode 100644
index 000000000..33af8f007
--- /dev/null
+++ b/Tools/NoiseSpeedTest/SimplexNoise.h
@@ -0,0 +1,263 @@
+// SimplexNoise.h
+
+// Declares and implements the simplex noise, using a template parameter for the underlying datatype
+
+/*
+Note:
+This code has been adapted from the public domain code by Stefan Gustavson, available at
+http://staffwww.itn.liu.se/~stegu/simplexnoise/simplexnoise.pdf
+*/
+
+#include <random>
+
+
+
+
+
+template<typename Datatype>
+class cSimplexNoise
+{
+public:
+	cSimplexNoise(int a_Seed)
+	{
+		// Based on the seed, initialize the permutation table, using a simple LCG and swapping
+
+		// Initialize with sorted sequence:
+		for (size_t i = 0; i < ARRAYCOUNT(m_Perm) / 2; i++)
+		{
+			m_Perm[i] = static_cast<int>(i);
+		}
+
+		// Use swaps to randomize:
+		std::linear_congruential_engine<unsigned, 48271, 0, 2147483647> lcg(a_Seed);
+		for (size_t i = 0; i < 2000; i++)
+		{
+			std::swap(m_Perm[lcg() % (ARRAYCOUNT(m_Perm) / 2)], m_Perm[lcg() % (ARRAYCOUNT(m_Perm) / 2)]);
+		}
+
+		// Copy to the upper half of the buffer (to avoid the need for modulo when accessing neighbors):
+		for (size_t i = ARRAYCOUNT(m_Perm) / 2; i < ARRAYCOUNT(m_Perm); i++)
+		{
+			m_Perm[i] = m_Perm[i - ARRAYCOUNT(m_Perm) / 2];
+		}
+
+		// Copy to the "modulo 12" table to optimize away four modulo ops per value calculation:
+		for (size_t i = 0; i < ARRAYCOUNT(m_Perm); i++)
+		{
+			m_PermMod12[i] = m_Perm[i] % 12;
+		}
+	}
+
+
+
+	/** Returns a dot product of an int vector with a Datatype vector. */
+	inline Datatype dot(const int * g, const Datatype x, const Datatype y, const Datatype z)
+	{
+		return g[0] * x + g[1] * y + g[2] * z;
+	}
+
+
+
+	/** Returns a dot product of two Datatype vectors. */
+	inline Datatype dot(const Datatype * g, const Datatype x, const Datatype y, const Datatype z)
+	{
+		return g[0] * x + g[1] * y + g[2] * z;
+	}
+
+
+
+	/** Returns the floor of the specified value, already type-cast to proper int type. */
+	inline int datafloor(const Datatype a_Val)
+	{
+		return (a_Val > 0) ? static_cast<int>(a_Val) : static_cast<int>(a_Val - 1);  // This is faster than std::floor()
+	}
+
+
+	/** Returns a single noise value based on the 3D coords. */
+	Datatype GetValueAt3D(const Datatype a_X, const Datatype a_Y, const Datatype a_Z)
+	{
+		// The gradients are the midpoints of the vertices of a cube.
+		static const Datatype grad3[12][3] = {
+			{1, 1, 0}, {-1,  1, 0}, {1, -1,  0}, {-1, -1,  0},
+			{1, 0, 1}, {-1,  0, 1}, {1,  0, -1}, {-1,  0, -1},
+			{0, 1, 1}, { 0, -1, 1}, {0,  1, -1}, { 0, -1, -1}
+		}; 
+
+		// Skew factors:
+		static const Datatype F3 = static_cast<Datatype>(1.0 / 3.0);
+		static const Datatype G3 = static_cast<Datatype>(1.0 / 6.0);
+
+		// Noise contributions from the four corners:
+		Datatype n0, n1, n2, n3;
+
+		// Skew the input space to determine which simplex cell we're in
+		Datatype s = (a_X + a_Y + a_Z) * F3;
+		int i = datafloor(a_X + s);
+		int j = datafloor(a_Y + s);
+		int k = datafloor(a_Z + s);
+
+		// Unskew back into the XYZ space to calculate the distances from cell origin:
+		Datatype t = (i + j + k) * G3;
+		Datatype X0 = i - t;
+		Datatype Y0 = j - t;
+		Datatype Z0 = k - t;
+		Datatype x0 = a_X - X0;
+		Datatype y0 = a_Y - Y0;
+		Datatype z0 = a_Z - Z0;
+
+		// For the 3D case, the simplex shape is a slightly irregular tetrahedron.
+		// Determine which simplex we are in.
+		int i1, j1, k1;  // Offsets for second corner of simplex in IJK coords
+		int i2, j2, k2;  // Offsets for third  corner of simplex in IJK coords
+		if (x0 >= y0)
+		{
+			if (y0 >= z0)
+			{
+				// X Y Z order
+				i1 = 1; j1 = 0; k1 = 0; i2 = 1; j2 = 1; k2 = 0;
+			}
+			else if (x0 >= z0)
+			{
+				// X Z Y order
+				i1 = 1; j1 = 0; k1 = 0; i2 = 1; j2 = 0; k2 = 1;
+			}
+			else
+			{
+				// Z X Y order
+				i1 = 0; j1 = 0; k1 = 1; i2 = 1; j2 = 0; k2 = 1;
+			}
+		}
+		else
+		{
+			if (y0 < z0)
+			{
+				// Z Y X order
+				i1 = 0; j1 = 0; k1 = 1; i2 = 0; j2 = 1; k2 = 1;
+			}
+			else if (x0 < z0)
+			{
+				// Y Z X order
+				i1 = 0; j1 = 1; k1 = 0; i2 = 0; j2 = 1; k2 = 1;
+			}
+			else
+			{
+				// Y X Z order
+				i1 = 0; j1 = 1; k1 = 0; i2 = 1; j2 = 1; k2 = 0;
+			}
+		}
+
+		// A step of (1, 0, 0) in IJK means a step of (1 - c,    -c,    -c) in XYZ,
+		// a step of (0, 1, 0) in IJK means a step of (-c,    1 - c,    -c) in XYZ, and
+		// a step of (0, 0, 1) in IJK means a step of (-c,       -c, 1 - c) in XYZ, where c = G3 = 1 / 6.
+		Datatype x1 = x0 - i1 + G3; // Offsets for second corner in XYZ coords
+		Datatype y1 = y0 - j1 + G3;
+		Datatype z1 = z0 - k1 + G3;
+		Datatype x2 = x0 - i2 + static_cast<Datatype>(2) * G3; // Offsets for third corner in XYZ coords
+		Datatype y2 = y0 - j2 + static_cast<Datatype>(2) * G3;
+		Datatype z2 = z0 - k2 + static_cast<Datatype>(2) * G3;
+		Datatype x3 = x0 - static_cast<Datatype>(1) + static_cast<Datatype>(3) * G3; // Offsets for last corner in XYZ coords
+		Datatype y3 = y0 - static_cast<Datatype>(1) + static_cast<Datatype>(3) * G3;
+		Datatype z3 = z0 - static_cast<Datatype>(1) + static_cast<Datatype>(3) * G3;
+
+		// Work out the hashed gradient indices of the four simplex corners
+		int ii = i & 255;
+		int jj = j & 255;
+		int kk = k & 255;
+		int gi0 = m_PermMod12[ii + m_Perm[jj + m_Perm[kk]]];
+		int gi1 = m_PermMod12[ii + i1 + m_Perm[jj + j1 + m_Perm[kk + k1]]];
+		int gi2 = m_PermMod12[ii + i2 + m_Perm[jj + j2 + m_Perm[kk + k2]]];
+		int gi3 = m_PermMod12[ii + 1 + m_Perm[jj + 1 + m_Perm[kk + 1]]];
+
+		// Calculate the contribution from the four corners
+		Datatype t0 = static_cast<Datatype>(0.6) - x0 * x0 - y0 * y0 - z0 * z0;
+		if (t0 < 0)
+		{
+			n0 = 0.0;
+		}
+		else
+		{
+			t0 *= t0;
+			n0 = t0 * t0 * dot(grad3[gi0], x0, y0, z0);
+		}
+
+		Datatype t1 = static_cast<Datatype>(0.6) - x1 * x1 - y1 * y1 - z1 * z1;
+		if (t1 < 0)
+		{
+			n1 = 0.0;
+		}
+		else
+		{
+			t1 *= t1;
+			n1 = t1 * t1 * dot(grad3[gi1], x1, y1, z1);
+		}
+
+		Datatype t2 = static_cast<Datatype>(0.6) - x2 * x2 - y2 * y2 - z2 * z2;
+		if (t2 < 0)
+		{
+			n2 = 0.0;
+		}
+		else
+		{
+			t2 *= t2;
+			n2 = t2 * t2 * dot(grad3[gi2], x2, y2, z2);
+		}
+
+		Datatype t3 = static_cast<Datatype>(0.6) - x3 * x3 - y3 * y3 - z3 * z3;
+		if (t3 < 0)
+		{
+			n3 = 0.0;
+		}
+		else
+		{
+			t3 *= t3;
+			n3 = t3 * t3 * dot(grad3[gi3], x3, y3, z3);
+		}
+
+		// Add contributions from each corner to get the final noise value.
+		// The result is scaled to stay just inside [-1, 1]
+		return static_cast<Datatype>(32) * (n0 + n1 + n2 + n3);
+	}
+
+
+
+
+	/** Generates the 3D version of the SImplex noise.
+	a_Out is the 3D array into which the noise is output. Organized as [x + a_SizeX * y + a_SizeX * a_SizeY * z].
+	a_SizeX, a_SizeY, a_SizeZ are the dimensions of the a_Out array.
+	a_Start and a_End are the coords of the 3D array in the noise-space. */
+	void Generate3D(
+		Datatype * a_Out,
+		int a_SizeX, int a_SizeY, int a_SizeZ,
+		Datatype a_StartX, Datatype a_EndX,
+		Datatype a_StartY, Datatype a_EndY,
+		Datatype a_StartZ, Datatype a_EndZ
+	)
+	{
+		Datatype * out = a_Out;
+		for (int z = 0; z < a_SizeZ; ++z)
+		{
+			Datatype nz = a_StartZ + z * (a_EndZ - a_StartZ) / a_SizeZ;
+			for (int y = 0; y < a_SizeY; ++y)
+			{
+				Datatype ny = a_StartY + y * (a_EndY - a_StartY) / a_SizeY;
+				for (int x = 0; x < a_SizeX; ++x)
+				{
+					Datatype nx = a_StartX + x * (a_EndX - a_StartX) / a_SizeX;
+					*out = GetValueAt3D(nx, ny, nz);
+					++out;
+				}  // for x
+			}  // for y
+		}  // for z
+	}
+
+protected:
+	/** The permutation table, initialized by the seed. */
+	int m_Perm[512];
+
+	/** A copy of the permutation table, with each item modulo 12, to avoid 4 modulo operations per value calculation. */
+	int m_PermMod12[512];
+};
+
+
+
+