mirror of
https://bitbucket.org/mfeemster/fractorium.git
synced 2025-01-21 21:20:07 -05:00
3f5db01a06
This prevents a circular reference where Isaac.h requires this method, but Utils.h includes Isaac.h
1034 lines
30 KiB
C++
1034 lines
30 KiB
C++
#pragma once
|
|
|
|
#include "Isaac.h"
|
|
|
|
/// <summary>
|
|
/// Global utility classes and functions that don't really fit anywhere else, but are
|
|
/// too small to justify being in their own file.
|
|
/// </summary>
|
|
namespace EmberNs
|
|
{
|
|
/// <summary>
|
|
/// Thin wrapper around std::find_if() to relieve the caller of having to
|
|
/// pass the implicitly obvious .begin() and .end(), and then compare the results to .end().
|
|
/// </summary>
|
|
/// <param name="container">The container to call find_if() on</param>
|
|
/// <param name="pred">The lambda to call on each element</param>
|
|
/// <returns>True if pred returned true once, else false.</returns>
|
|
template<class c, class pr>
|
|
static inline bool FindIf(c& container, pr pred)
|
|
{
|
|
return std::find_if(container.begin(), container.end(), pred) != container.end();
|
|
}
|
|
|
|
/// <summary>
|
|
/// Thin wrapper around std::for_each() to relieve the caller of having to
|
|
/// pass the implicitly obvious .begin() and .end().
|
|
/// </summary>
|
|
/// <param name="container">The container to call for_each() on</param>
|
|
/// <param name="pred">The lambda to call on each element</param>
|
|
template<class c, class fn>
|
|
static inline void ForEach(c& container, fn func)
|
|
{
|
|
std::for_each(container.begin(), container.end(), func);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Thin wrapper around computing the total size of a vector.
|
|
/// </summary>
|
|
/// <param name="vec">The vector to compute the size of</param>
|
|
/// <returns>The size of one element times the length.</returns>
|
|
template<typename T>
|
|
static inline size_t SizeOf(vector<T>& vec)
|
|
{
|
|
return sizeof(vec[0]) * vec.size();
|
|
}
|
|
|
|
/// <summary>
|
|
/// After a run completes, information about what was run can be saved as strings to the comments
|
|
/// section of a jpg or png file. This class is just a container for those values.
|
|
/// </summary>
|
|
class EMBER_API EmberImageComments
|
|
{
|
|
public:
|
|
/// <summary>
|
|
/// Empty destructor.
|
|
/// Needed to eliminate warnings about inlining.
|
|
/// </summary>
|
|
~EmberImageComments()
|
|
{
|
|
}
|
|
|
|
/// <summary>
|
|
/// Set all values to the empty string.
|
|
/// </summary>
|
|
void Clear()
|
|
{
|
|
m_Genome = "";
|
|
m_Badvals = "";
|
|
m_NumIters = "";
|
|
m_Runtime = "";
|
|
}
|
|
|
|
string m_Genome;
|
|
string m_Badvals;
|
|
string m_NumIters;
|
|
string m_Runtime;
|
|
};
|
|
|
|
/// <summary>
|
|
/// Since running is an incredibly complex process with multiple points of possible failure,
|
|
/// it's important that as much information as possible is captured if something goes wrong.
|
|
/// Classes wishing to capture this failure information will derive from this class and populate
|
|
/// the vector of strings with any useful error information. Note that a small complication can occur
|
|
/// when a class derives from this class, yet also has one or more members which do too. In that case, they should
|
|
/// override the methods to aggregate the error information from themselves, as well as their members.
|
|
/// </summary>
|
|
class EMBER_API EmberReport
|
|
{
|
|
public:
|
|
/// <summary>
|
|
/// Virtual destructor needed for virtual classes.
|
|
/// </summary>
|
|
virtual ~EmberReport() { }
|
|
|
|
/// <summary>
|
|
/// Write the entire error report as a single string to the console.
|
|
/// Derived classes with members that also derive from EmberReport should override this to capture
|
|
/// their error information as well as that of their members.
|
|
/// </summary>
|
|
virtual void DumpErrorReport() { cout << ErrorReportString(); }
|
|
|
|
/// <summary>
|
|
/// Clear the error report string vector.
|
|
/// Derived classes with members that also derive from EmberReport should override this to clear
|
|
/// their error information as well as that of their members.
|
|
/// </summary>
|
|
virtual void ClearErrorReport() { m_ErrorReport.clear(); }
|
|
|
|
/// <summary>
|
|
/// Return the entire error report as a single string.
|
|
/// Derived classes with members that also derive from EmberReport should override this to capture
|
|
/// their error information as well as that of their members.
|
|
/// </summary>
|
|
/// <returns>The entire error report as a single string. Empty if no errors.</returns>
|
|
virtual string ErrorReportString() { return StaticErrorReportString(m_ErrorReport); }
|
|
|
|
/// <summary>
|
|
/// Return the entire error report as a vector of strings.
|
|
/// Derived classes with members that also derive from EmberReport should override this to capture
|
|
/// their error information as well as that of their members.
|
|
/// </summary>
|
|
/// <returns>The entire error report as a vector of strings. Empty if no errors.</returns>
|
|
virtual vector<string> ErrorReport() { return m_ErrorReport; }
|
|
|
|
/// <summary>
|
|
/// Add string to report.
|
|
/// </summary>
|
|
/// <param name="s">The string to add</param>
|
|
virtual void AddToReport(const string& s) { m_ErrorReport.push_back(s); }
|
|
|
|
/// <summary>
|
|
/// Add a vector of strings to report.
|
|
/// </summary>
|
|
/// <param name="vec">The vector of strings to add</param>
|
|
virtual void AddToReport(const vector<string>& vec) { m_ErrorReport.insert(m_ErrorReport.end(), vec.begin(), vec.end()); }
|
|
|
|
/// <summary>
|
|
/// Static function to dump a vector of strings passed in.
|
|
/// </summary>
|
|
/// <param name="errorReport">The vector of strings to dump</param>
|
|
static void StaticDumpErrorReport(const vector<string>& errorReport) { cout << StaticErrorReportString(errorReport); }
|
|
|
|
/// <summary>
|
|
/// Static function to return the entire error report passed in as a single string.
|
|
/// </summary>
|
|
/// <param name="errorReport">The vector of strings to concatenate</param>
|
|
/// <returns>A string containing all strings in the vector passed in separated by newlines</returns>
|
|
static string StaticErrorReportString(const vector<string>& errorReport)
|
|
{
|
|
stringstream ss;
|
|
|
|
ForEach(errorReport, [&](const string& s) { ss << s << endl; });
|
|
|
|
return ss.str();
|
|
}
|
|
|
|
protected:
|
|
vector<string> m_ErrorReport;
|
|
};
|
|
|
|
/// <summary>
|
|
/// Open a file in binary mode and read its entire contents into a vector of bytes. Optionally null terminate.
|
|
/// </summary>
|
|
/// <param name="filename">The full path to the file to read</param>
|
|
/// <param name="buf">The vector which will be populated with the file's contents</param>
|
|
/// <param name="nullTerminate">Whether to append a NULL character as the last element of the vector. Needed when reading text files. Default: true.</param>
|
|
/// <returns>True if successfully read and populated, else false</returns>
|
|
static bool ReadFile(const char* filename, string& buf, bool nullTerminate = true)
|
|
{
|
|
bool b = false;
|
|
FILE* f = nullptr;
|
|
|
|
try
|
|
{
|
|
fopen_s(&f, filename, "rb");//Open in binary mode.
|
|
|
|
if (f != nullptr)
|
|
{
|
|
struct _stat statBuf;
|
|
|
|
#if defined(_WIN32) || defined(__APPLE__)
|
|
int statResult = _fstat(f->_file, &statBuf);//Get data associated with file.
|
|
#else
|
|
int statResult = _fstat(f->_fileno, &statBuf);//Get data associated with file.
|
|
#endif
|
|
|
|
if (statResult == 0)//Check if statistics are valid.
|
|
{
|
|
buf.resize(statBuf.st_size + (nullTerminate ? 1 : 0));//Allocate vector to be the size of the entire file, with an optional additional character for nullptr.
|
|
|
|
if (buf.size() == static_cast<size_t>(statBuf.st_size + 1))//Ensure allocation succeeded.
|
|
{
|
|
size_t bytesRead = fread(&buf[0], 1, statBuf.st_size, f);//Read the entire file at once.
|
|
|
|
if (bytesRead == (static_cast<size_t>(statBuf.st_size)))//Ensure the number of bytes read matched what was requested.
|
|
{
|
|
if (nullTerminate)//Optionally nullptr terminate if they want to treat it as a string.
|
|
buf[buf.size() - 1] = 0;
|
|
|
|
b = true;//Success.
|
|
}
|
|
}
|
|
}
|
|
|
|
fclose(f);
|
|
}
|
|
}
|
|
catch (...)
|
|
{
|
|
if (f != nullptr)
|
|
fclose(f);
|
|
|
|
b = false;
|
|
}
|
|
|
|
return b;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Clear dest and copy all of the elements of vector source with elements of type U to the vector
|
|
/// dest with elements of type T.
|
|
/// </summary>
|
|
/// <param name="dest">The vector of type T to copy to</param>
|
|
/// <param name="source">The vector of type U to copy from</param>
|
|
template <typename T, typename U>
|
|
static void CopyVec(vector<T>& dest, const vector<U>& source)
|
|
{
|
|
dest.clear();
|
|
dest.resize(source.size());
|
|
|
|
for (size_t i = 0; i < source.size(); i++)
|
|
dest[i] = static_cast<T>(source[i]);//Valid assignment operator between T and U types must be defined somewhere.
|
|
}
|
|
|
|
/// <summary>
|
|
/// Clear dest and copy all of the elements of vector source with elements of type U to the vector
|
|
/// dest with elements of type T.
|
|
/// Call a function on each element after it's been copied.
|
|
/// </summary>
|
|
/// <param name="dest">The vector of type T to copy to</param>
|
|
/// <param name="source">The vector of type U to copy from</param>
|
|
/// <param name="perElementOperation">A function to call on each element after it's copied</param>
|
|
template <typename T, typename U>
|
|
static void CopyVec(vector<T>& dest, const vector<U>& source, std::function<void(T& t)> perElementOperation)
|
|
{
|
|
dest.clear();
|
|
dest.resize(source.size());
|
|
|
|
for (size_t i = 0; i < source.size(); i++)
|
|
{
|
|
dest[i] = static_cast<T>(source[i]);//Valid assignment operator between T and U types must be defined somewhere.
|
|
perElementOperation(dest[i]);
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Clear a vector of pointers to any type by checking each element for nullptr and calling delete on it, then clearing the entire vector.
|
|
/// Optionally call array delete if the elements themselves are pointers to dynamically allocated arrays.
|
|
/// </summary>
|
|
/// <param name="vec">The vector to be cleared</param>
|
|
/// <param name="arrayDelete">Whether to call delete or delete []. Default: false.</param>
|
|
template <typename T>
|
|
static void ClearVec(vector<T*>& vec, bool arrayDelete = false)
|
|
{
|
|
for (uint i = 0; i < vec.size(); i++)
|
|
{
|
|
if (vec[i] != nullptr)
|
|
{
|
|
if (arrayDelete)
|
|
delete [] vec[i];
|
|
else
|
|
delete vec[i];
|
|
}
|
|
|
|
vec[i] = nullptr;
|
|
}
|
|
|
|
vec.clear();
|
|
}
|
|
|
|
/// <summary>
|
|
/// Thin wrapper around passing a vector to memset() to relieve
|
|
/// the caller of having to pass the size.
|
|
/// </summary>
|
|
/// <param name="vec">The vector to memset</param>
|
|
/// <param name="val">The value to set each element to, default 0.</param>
|
|
template<typename T>
|
|
static inline void Memset(vector<T>& vec, int val = 0)
|
|
{
|
|
memset(static_cast<void*>(vec.data()), val, SizeOf(vec));
|
|
}
|
|
|
|
/// <summary>
|
|
/// System floor() extremely slow because it accounts for various error conditions.
|
|
/// This is a much faster version that works on data that is not NaN.
|
|
/// </summary>
|
|
/// <param name="x">The value to return the floor of</param>
|
|
/// <returns>The floored value</returns>
|
|
template <typename T>
|
|
static inline int Floor(T val)
|
|
{
|
|
if (val >= 0)
|
|
{
|
|
return static_cast<int>(val);
|
|
}
|
|
else
|
|
{
|
|
int i = static_cast<int>(val);//Truncate.
|
|
return i - (i > val);//Convert trunc to floor.
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Clamp and return a value to be greater than or equal to a specified minimum and less than
|
|
/// or equal to a specified maximum.
|
|
/// </summary>
|
|
/// <param name="val">The value to be clamped</param>
|
|
/// <param name="min">A value which the clamped value must be greater than or equal to</param>
|
|
/// <param name="max">A value which the clamped value must be less than or equal to</param>
|
|
/// <returns>The clamped value</returns>
|
|
template <typename T>
|
|
static inline T Clamp(T val, T min, T max)
|
|
{
|
|
if (val < min)
|
|
return min;
|
|
else if (val > max)
|
|
return max;
|
|
else
|
|
return val;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Clamp and return a value to be greater than or equal to a specified minimum and less than
|
|
/// or equal to a specified maximum. If lesser, the value is fmod(val - min, max - min). If greater,
|
|
/// the value is max - fmod(max - val, max - min).
|
|
/// </summary>
|
|
/// <param name="val">The value to be clamped</param>
|
|
/// <param name="min">A value which the clamped value must be greater than or equal to</param>
|
|
/// <param name="max">A value which the clamped value must be less than or equal to</param>
|
|
/// <returns>The clamped and modded value</returns>
|
|
template <typename T>
|
|
static inline T ClampMod(T val, T min, T max)
|
|
{
|
|
if (val < min)
|
|
return min + fmod(val - min, max - min);
|
|
else if (val > max)
|
|
return max - fmod(max - val, max - min);
|
|
else
|
|
return val;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Similar to Clamp(), but clamps a reference value in place rather than returning.
|
|
/// </summary>
|
|
/// <param name="val">The reference value to be clamped in place</param>
|
|
/// <param name="min">A value which the clamped value must be greater than or equal to</param>
|
|
/// <param name="max">A value which the clamped value must be less than or equal to</param>
|
|
template <typename T>
|
|
static inline void ClampRef(T& val, T min, T max)
|
|
{
|
|
if (val < min)
|
|
val = min;
|
|
else if (val > max)
|
|
val = max;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Similar to Clamp(), but clamps a reference value in place rather than returning.
|
|
/// </summary>
|
|
/// <param name="val">The reference value to be clamped in place</param>
|
|
/// <param name="gte">A value which the clamped value must be less than or equal to</param>
|
|
template <typename T>
|
|
static inline void ClampLteRef(T& val, T lte)
|
|
{
|
|
if (val > lte)
|
|
val = lte;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Clamp and return a value to be greater than or equal to a specified value.
|
|
/// Useful for ensuring something is not less than zero.
|
|
/// </summary>
|
|
/// <param name="val">The value to be clamped</param>
|
|
/// <param name="gte">A value which the clamped value must be greater than or equal to</param>
|
|
/// <returns>The clamped value</returns>
|
|
template <typename T>
|
|
static inline T ClampGte(T val, T gte)
|
|
{
|
|
return (val < gte) ? gte : val;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Similar to Clamp(), but clamps a reference value in place rather than returning.
|
|
/// </summary>
|
|
/// <param name="val">The reference value to be clamped in place</param>
|
|
/// <param name="gte">A value which the clamped value must be greater than or equal to</param>
|
|
template <typename T>
|
|
static inline void ClampGteRef(T& val, T gte)
|
|
{
|
|
if (val < gte)
|
|
val = gte;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Thin wrapper around a call to ClampGte() with a gte value of zero.
|
|
/// </summary>
|
|
/// <param name="val">The value to be clamped</param>
|
|
/// <returns>The clamped value</returns>
|
|
template <typename T>
|
|
static inline T ClampGte0(T val)
|
|
{
|
|
return ClampGte<T>(val, 0);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Thin wrapper around a call to ClampGteRef() with a gte value of zero.
|
|
/// </summary>
|
|
/// <param name="val">The reference value to be clamped in place</param>
|
|
template <typename T>
|
|
static inline void ClampGte0Ref(T& val)
|
|
{
|
|
ClampGteRef<T>(val, 0);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return a value rounded up or down. Works for positive and negative numbers.
|
|
/// </summary>
|
|
/// <param name="r">The value to round</param>
|
|
/// <returns>The rounded value</returns>
|
|
template <typename T>
|
|
static inline T Round(T r)
|
|
{
|
|
return (r > 0) ? static_cast<T>(Floor<T>(r + T(0.5))) : ceil(r - T(0.5));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Special rounding for certain variations, gotten from Apophysis.
|
|
/// </summary>
|
|
/// <param name="x">The value to round</param>
|
|
/// <returns>The rounded value</returns>
|
|
static inline float LRint(float x)
|
|
{
|
|
int temp = (x >= 0 ? static_cast<int>(x + 0.5f) : static_cast<int>(x - 0.5f));
|
|
return static_cast<float>(temp);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Special rounding for certain variations, gotten from Apophysis.
|
|
/// </summary>
|
|
/// <param name="x">The value to round</param>
|
|
/// <returns>The rounded value</returns>
|
|
static inline double LRint(double x)
|
|
{
|
|
glm::int64_t temp = (x >= 0 ? static_cast<int64_t>(x + 0.5) : static_cast<int64_t>(x - 0.5));
|
|
return static_cast<double>(temp);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Never really understood what this did.
|
|
/// </summary>
|
|
/// <param name="r">The value to round</param>
|
|
/// <returns>The rounded value</returns>
|
|
template <typename T>
|
|
static inline T Round6(T r)
|
|
{
|
|
r *= 1e6;
|
|
|
|
if (r < 0)
|
|
r -= 1;
|
|
|
|
return static_cast<T>(1e-6 * static_cast<int>(r + T(0.5)));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return -1 if the value is less than 0, 1 if it's greater and
|
|
/// 0 if it's equal to 0.
|
|
/// </summary>
|
|
/// <param name="v">The value to inspect</param>
|
|
/// <returns>-1, 0 or 1</returns>
|
|
template <typename T>
|
|
static inline T Sign(T v)
|
|
{
|
|
return (v < 0) ? static_cast<T>(-1) : (v > 0) ? static_cast<T>(1) : static_cast<T>(0);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return -1 if the value is less than 0, 1 if it's greater.
|
|
/// This differs from Sign() in that it doesn't return 0.
|
|
/// </summary>
|
|
/// <param name="v">The value to inspect</param>
|
|
/// <returns>-1 or 1</returns>
|
|
template <typename T>
|
|
static inline T SignNz(T v)
|
|
{
|
|
return (v < 0) ? static_cast<T>(-1) : static_cast<T>(1);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return the square of the passed in value.
|
|
/// This is useful when the value is a result of a computation
|
|
/// rather than a fixed number. Otherwise, use the SQR macro.
|
|
/// </summary>
|
|
/// <param name="v">The value to square</param>
|
|
/// <returns>The squared value</returns>
|
|
template <typename T>
|
|
static inline T Sqr(T t)
|
|
{
|
|
return t * t;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Taking the square root of numbers close to zero is dangerous. If x is negative
|
|
/// due to floating point errors, it can return NaN results.
|
|
/// </summary>
|
|
template <typename T>
|
|
static inline T SafeSqrt(T x)
|
|
{
|
|
if (x <= 0)
|
|
return 0;
|
|
|
|
return sqrt(x);
|
|
}
|
|
|
|
template <typename T>
|
|
static inline T SafeTan(T x)
|
|
{
|
|
return x;
|
|
}
|
|
|
|
template <>
|
|
#ifdef _WIN32
|
|
static
|
|
#endif
|
|
float SafeTan<float>(float x)
|
|
{
|
|
return tan(Clamp<float>(x, FLOAT_MIN_TAN, FLOAT_MAX_TAN));
|
|
}
|
|
|
|
template <>
|
|
#ifdef _WIN32
|
|
static
|
|
#endif
|
|
double SafeTan<double>(double x)
|
|
{
|
|
return tan(x);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return the cube of the passed in value.
|
|
/// This is useful when the value is a result of a computation
|
|
/// rather than a fixed number. Otherwise, use the CUBE macro.
|
|
/// </summary>
|
|
/// <param name="v">The value to cube</param>
|
|
/// <returns>The cubed value</returns>
|
|
template <typename T>
|
|
static inline T Cube(T t)
|
|
{
|
|
return t * t * t;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return the hypotenuse of the passed in values.
|
|
/// </summary>
|
|
/// <param name="x">The x distance</param>
|
|
/// <param name="y">The y distance</param>
|
|
/// <returns>The hypotenuse</returns>
|
|
template <typename T>
|
|
static inline T Hypot(T x, T y)
|
|
{
|
|
return sqrt(SQR(x) + SQR(y));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Spread the values.
|
|
/// </summary>
|
|
/// <param name="x">The x distance</param>
|
|
/// <param name="y">The y distance</param>
|
|
/// <returns>The spread</returns>
|
|
template <typename T>
|
|
static inline T Spread(T x, T y)
|
|
{
|
|
return Hypot<T>(x, y) * ((x) > 0 ? 1 : -1);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Unsure.
|
|
/// </summary>
|
|
/// <param name="x">The x distance</param>
|
|
/// <param name="y">The y distance</param>
|
|
/// <returns>The powq4</returns>
|
|
template <typename T>
|
|
static inline T Powq4(T x, T y)
|
|
{
|
|
return pow(fabs(x), y) * SignNz(x);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Unsure.
|
|
/// </summary>
|
|
/// <param name="x">The x distance</param>
|
|
/// <param name="y">The y distance</param>
|
|
/// <returns>The powq4c</returns>
|
|
template <typename T>
|
|
static inline T Powq4c(T x, T y)
|
|
{
|
|
return y == 1 ? x : Powq4(x, y);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return EPS if the passed in value was zero, else return the value.
|
|
/// </summary>
|
|
/// <param name="x">The value</param>
|
|
/// <param name="y">The y distance</param>
|
|
/// <returns>EPS or the value if it was non-zero</returns>
|
|
template <typename T>
|
|
static inline T Zeps(T x)
|
|
{
|
|
return x == 0 ? EPS : x;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Interpolate a given percentage between two values.
|
|
/// </summary>
|
|
/// <param name="a">The first value to interpolate between.</param>
|
|
/// <param name="b">The secod value to interpolate between.</param>
|
|
/// <param name="p">The percentage between the two values to calculate.</param>
|
|
/// <returns>The interpolated value.</returns>
|
|
template <typename T>
|
|
static inline T Lerp(T a, T b, T p)
|
|
{
|
|
return a + (b - a) * p;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Thin wrapper around a call to modf that discards the integer portion
|
|
/// and returns the signed fractional portion.
|
|
/// </summary>
|
|
/// <param name="v">The value to retrieve the signed fractional portion of.</param>
|
|
/// <returns>The signed fractional portion of v.</returns>
|
|
template <typename T>
|
|
static inline T Fabsmod(T v)
|
|
{
|
|
T dummy;
|
|
|
|
return modf(v, &dummy);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Unsure.
|
|
/// </summary>
|
|
/// <param name="p">Unsure.</param>
|
|
/// <param name="amp">Unsure.</param>
|
|
/// <param name="ph">Unsure.</param>
|
|
/// <returns>Unsure.</returns>
|
|
template <typename T>
|
|
static inline T Fosc(T p, T amp, T ph)
|
|
{
|
|
return T(0.5) - cos(p * amp + ph) * T(0.5);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Unsure.
|
|
/// </summary>
|
|
/// <param name="p">Unsure.</param>
|
|
/// <param name="ph">Unsure.</param>
|
|
/// <returns>Unsure.</returns>
|
|
template <typename T>
|
|
static inline T Foscn(T p, T ph)
|
|
{
|
|
return T(0.5) - cos(p + ph) * T(0.5);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Log scale from Apophysis.
|
|
/// </summary>
|
|
/// <param name="x">The value to log scale</param>
|
|
/// <returns>The log scaled value</returns>
|
|
template <typename T>
|
|
static inline T LogScale(T x)
|
|
{
|
|
return x == 0 ? 0 : log((fabs(x) + 1) * T(M_E)) * SignNz(x) / T(M_E);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Log map from Apophysis.
|
|
/// </summary>
|
|
/// <param name="x">The value to log map</param>
|
|
/// <returns>The log mapped value</returns>
|
|
template <typename T>
|
|
static inline T LogMap(T x)
|
|
{
|
|
return x == 0 ? 0 : (T(M_E) + log(x * T(M_E))) * T(0.25) * SignNz(x);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Thin wrapper around calling xmlStrcmp() on an Xml tag to tell
|
|
/// if its name is a given value.
|
|
/// </summary>
|
|
/// <param name="name">The name of the tag of the to inspect</param>
|
|
/// <param name="val">The value compare against</param>
|
|
/// <returns>True if the comparison matched, else false</returns>
|
|
static inline bool Compare(const xmlChar* name, const char* val)
|
|
{
|
|
return xmlStrcmp(name, XC(val)) != 0;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Determine whether the specified value is very close to zero.
|
|
/// This is useful for determining equality of float/double types.
|
|
/// </summary>
|
|
/// <param name="val">The value to compare against</param>
|
|
/// <param name="tolerance">The tolerance. Default: 1e-6.</param>
|
|
/// <returns>True if the value was very close to zero, else false</returns>
|
|
template <typename T>
|
|
static inline bool IsNearZero(T val, T tolerance = 1e-6)
|
|
{
|
|
return (val > -tolerance && val < tolerance);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Determine whether a specified value is very close to another value.
|
|
/// This is useful for determining equality of float/double types.
|
|
/// </summary>
|
|
/// <param name="val1">The first value.</param>
|
|
/// <param name="val2">The second value.</param>
|
|
/// <param name="tolerance">The tolerance. Default: 1e-6.</param>
|
|
/// <returns>True if the values were very close to each other, else false</returns>
|
|
template <typename T>
|
|
static inline bool IsClose(T val1, T val2, T tolerance = 1e-6)
|
|
{
|
|
return IsNearZero(val1 - val2, tolerance);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Put an angular measurement in degrees into the range of -180 - 180.
|
|
/// </summary>
|
|
/// <param name="angle">The angle to normalize</param>
|
|
/// <returns>The normalized angle in a range of -180 - 180</returns>
|
|
template <typename T>
|
|
static inline T NormalizeDeg180(T angle)
|
|
{
|
|
angle = fmod(angle, 360);
|
|
|
|
if (angle > 180)
|
|
{
|
|
angle -= 360;
|
|
}
|
|
else if (angle < -180)
|
|
{
|
|
angle += 360;
|
|
}
|
|
|
|
return angle;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Put an angular measurement in degrees into the range of 0 - 360.
|
|
/// </summary>
|
|
/// <param name="angle">The angle to normalize</param>
|
|
/// <returns>The normalized angle in a range of 0 - 360</returns>
|
|
template <typename T>
|
|
static inline T NormalizeDeg360(T angle)
|
|
{
|
|
if (angle > 360 || angle < -360)
|
|
angle = fmod(angle, 360);
|
|
|
|
if (angle < 0)
|
|
angle += 360;
|
|
|
|
return angle;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return a lower case copy of a string.
|
|
/// </summary>
|
|
/// <param name="str">The string to copy and make lower case</param>
|
|
/// <returns>The lower case string</returns>
|
|
static string ToLower(const string& str)
|
|
{
|
|
string lower;
|
|
|
|
lower.resize(str.size());//Allocate the destination space.
|
|
std::transform(str.begin(), str.end(), lower.begin(), ::tolower);//Convert the source string to lower case storing the result in the destination string.
|
|
return lower;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return an upper case copy of a string.
|
|
/// </summary>
|
|
/// <param name="str">The string to copy and make upper case</param>
|
|
/// <returns>The upper case string</returns>
|
|
static string ToUpper(const string& str)
|
|
{
|
|
string upper;
|
|
|
|
upper.resize(str.size());//Allocate the destination space.
|
|
std::transform(str.begin(), str.end(), upper.begin(), ::toupper);//Convert the source string to lower case storing the result in the destination string.
|
|
return upper;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return a copy of a string with leading and trailing occurrences of a specified character removed.
|
|
/// The default character is a space.
|
|
/// </summary>
|
|
/// <param name="str">The string to trim</param>
|
|
/// <param name="ch">The character to trim. Default: space.</param>
|
|
/// <returns>The trimmed string</returns>
|
|
static string Trim(const string& str, char ch = ' ')
|
|
{
|
|
string ret;
|
|
|
|
if (str != "")
|
|
{
|
|
size_t firstChar = str.find_first_not_of(ch);
|
|
size_t lastChar = str.find_last_not_of(ch);
|
|
|
|
if (firstChar == string::npos)
|
|
firstChar = 0;
|
|
|
|
if (lastChar == string::npos)
|
|
lastChar = str.size();
|
|
|
|
ret = str.substr(firstChar, lastChar - firstChar + 1);
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return a copy of a file path string with the path portion removed.
|
|
/// </summary>
|
|
/// <param name="filename">The string to retrieve the path from</param>
|
|
/// <returns>The path portion of the string</returns>
|
|
static string GetPath(const string& filename)
|
|
{
|
|
string s;
|
|
const size_t lastSlash = filename.find_last_of("\\/");
|
|
|
|
if (std::string::npos != lastSlash)
|
|
s = filename.substr(0, lastSlash + 1);
|
|
else
|
|
s = "";
|
|
|
|
return s;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Placeholder for a templated function to query the value of a specified system environment variable
|
|
/// of a specific type. This function does nothing as the functions for specific types implement the behavior
|
|
/// via template specialization.
|
|
/// </summary>
|
|
/// <param name="name">The name of the environment variable to query</param>
|
|
/// <param name="def">The default value to return if the environment variable was not present</param>
|
|
/// <returns>The value of the specified environment variable if found, else default</returns>
|
|
template <typename T>
|
|
static inline T Arg(char* name, T def)
|
|
{
|
|
T t;
|
|
return t;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Template specialization for Arg<>() with a type of int.
|
|
/// </summary>
|
|
/// <param name="name">The name of the environment variable to query</param>
|
|
/// <param name="def">The default value to return if the environment variable was not present</param>
|
|
/// <returns>The value of the specified environment variable if found, else default</returns>
|
|
template <>
|
|
#ifdef _WIN32
|
|
static
|
|
#endif
|
|
int Arg<int>(char* name, int def)
|
|
{
|
|
char* ch;
|
|
int returnVal;
|
|
#ifdef WIN32
|
|
size_t len;
|
|
errno_t err = _dupenv_s(&ch, &len, name);
|
|
#else
|
|
int err = 1;
|
|
ch = getenv(name);
|
|
#endif
|
|
|
|
if (err || !ch)
|
|
returnVal = def;
|
|
else
|
|
returnVal = atoi(ch);
|
|
|
|
#ifdef WIN32
|
|
free(ch);
|
|
#endif
|
|
return returnVal;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Template specialization for Arg<>() with a type of uint.
|
|
/// </summary>
|
|
/// <param name="name">The name of the environment variable to query</param>
|
|
/// <param name="def">The default value to return if the environment variable was not present</param>
|
|
/// <returns>The value of the specified environment variable if found, else default</returns>
|
|
template <>
|
|
#ifdef _WIN32
|
|
static
|
|
#endif
|
|
uint Arg<uint>(char* name, uint def)
|
|
{
|
|
return Arg<int>(name, static_cast<int>(def));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Template specialization for Arg<>() with a type of bool.
|
|
/// </summary>
|
|
/// <param name="name">The name of the environment variable to query</param>
|
|
/// <param name="def">The default value to return if the environment variable was not present</param>
|
|
/// <returns>The value of the specified environment variable if found, else default</returns>
|
|
template <>
|
|
#ifdef _WIN32
|
|
static
|
|
#endif
|
|
bool Arg<bool>(char* name, bool def)
|
|
{
|
|
return (Arg<int>(name, -999) != -999) ? true : def;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Template specialization for Arg<>() with a type of double.
|
|
/// </summary>
|
|
/// <param name="name">The name of the environment variable to query</param>
|
|
/// <param name="def">The default value to return if the environment variable was not present</param>
|
|
/// <returns>The value of the specified environment variable if found, else default</returns>
|
|
template <>
|
|
#ifdef _WIN32
|
|
static
|
|
#endif
|
|
double Arg<double>(char* name, double def)
|
|
{
|
|
char* ch;
|
|
double returnVal;
|
|
#ifdef WIN32
|
|
size_t len;
|
|
errno_t err = _dupenv_s(&ch, &len, name);
|
|
#else
|
|
int err = 1;
|
|
ch = getenv(name);
|
|
#endif
|
|
|
|
if (err || !ch)
|
|
returnVal = def;
|
|
else
|
|
returnVal = atof(ch);
|
|
|
|
#ifdef WIN32
|
|
free(ch);
|
|
#endif
|
|
return returnVal;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Template specialization for Arg<>() with a type of string.
|
|
/// </summary>
|
|
/// <param name="name">The name of the environment variable to query</param>
|
|
/// <param name="def">The default value to return if the environment variable was not present</param>
|
|
/// <returns>The value of the specified environment variable if found, else default</returns>
|
|
template <>
|
|
#ifdef _WIN32
|
|
static
|
|
#endif
|
|
string Arg<string>(char* name, string def)
|
|
{
|
|
char* ch;
|
|
string returnVal;
|
|
#ifdef WIN32
|
|
size_t len;
|
|
errno_t err = _dupenv_s(&ch, &len, name);
|
|
#else
|
|
int err = 1;
|
|
ch = getenv(name);
|
|
#endif
|
|
|
|
if (err || !ch)
|
|
{
|
|
if (def != "")
|
|
returnVal = def;
|
|
}
|
|
else
|
|
returnVal = string(ch);
|
|
|
|
#ifdef WIN32
|
|
free(ch);
|
|
#endif
|
|
return returnVal;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Replaces all instances of a value within a collection, with the specified value.
|
|
/// Taken from a StackOverflow.com post.
|
|
/// Modified to account for the scenario where the find and replace strings each start with
|
|
/// the same character.
|
|
/// Template argument should be any STL container.
|
|
/// </summary>
|
|
/// <param name="source">Collection to replace values in</param>
|
|
/// <param name="find">The value to replace</param>
|
|
/// <param name="replace">The value to replace with</param>
|
|
/// <returns>The number of instances replaced</returns>
|
|
template<typename T>
|
|
static uint FindAndReplace(T& source, const T& find, const T& replace)
|
|
{
|
|
uint replaceCount = 0;
|
|
typename T::size_type fLen = find.size();
|
|
typename T::size_type rLen = replace.size();
|
|
|
|
for (typename T::size_type pos = 0; (pos = source.find(find, pos)) != T::npos; pos += rLen)
|
|
{
|
|
typename T::size_type pos2 = source.find(replace, pos);
|
|
|
|
if (pos != pos2)
|
|
{
|
|
replaceCount++;
|
|
source.replace(pos, fLen, replace);
|
|
}
|
|
}
|
|
|
|
return replaceCount;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Return a character pointer to a version string composed of the EMBER_OS and EMBER_VERSION values.
|
|
/// </summary>
|
|
static inline const char* EmberVersion()
|
|
{
|
|
return EMBER_OS "-" EMBER_VERSION;
|
|
}
|
|
}
|