mirror of
https://github.com/JGRennison/OpenTTD-patches.git
synced 2024-11-17 21:25:40 +00:00
297 lines
16 KiB
C++
297 lines
16 KiB
C++
/* $Id$ */
|
|
|
|
/** @file helpers.hpp */
|
|
|
|
#ifndef HELPERS_HPP
|
|
#define HELPERS_HPP
|
|
|
|
#include "macros.h"
|
|
|
|
/** When allocating using malloc/calloc in C++ it is usually needed to cast the return value
|
|
* from void* to the proper pointer type. Another alternative would be MallocT<> as follows */
|
|
template <typename T> FORCEINLINE T* MallocT(size_t num_elements)
|
|
{
|
|
T *t_ptr = (T*)malloc(num_elements * sizeof(T));
|
|
return t_ptr;
|
|
}
|
|
/** When allocating using malloc/calloc in C++ it is usually needed to cast the return value
|
|
* from void* to the proper pointer type. Another alternative would be MallocT<> as follows */
|
|
template <typename T> FORCEINLINE T* CallocT(size_t num_elements)
|
|
{
|
|
T *t_ptr = (T*)calloc(num_elements, sizeof(T));
|
|
return t_ptr;
|
|
}
|
|
/** When allocating using malloc/calloc in C++ it is usually needed to cast the return value
|
|
* from void* to the proper pointer type. Another alternative would be MallocT<> as follows */
|
|
template <typename T> FORCEINLINE T* ReallocT(T* t_ptr, size_t num_elements)
|
|
{
|
|
t_ptr = (T*)realloc(t_ptr, num_elements * sizeof(T));
|
|
return t_ptr;
|
|
}
|
|
|
|
|
|
/** type safe swap operation */
|
|
template<typename T> void Swap(T& a, T& b)
|
|
{
|
|
T t = a;
|
|
a = b;
|
|
b = t;
|
|
}
|
|
|
|
|
|
/** returns the absolute value of (scalar) variable. @note assumes variable to be signed */
|
|
template <typename T> static inline T myabs(T a) { return a < (T)0 ? -a : a; }
|
|
/** returns the (absolute) difference between two (scalar) variables */
|
|
template <typename T> static inline T delta(T a, T b) { return a < b ? b - a : a - b; }
|
|
|
|
/** Some enums need to have allowed incrementing (i.e. StationClassID) */
|
|
#define DECLARE_POSTFIX_INCREMENT(type) \
|
|
FORCEINLINE type operator ++(type& e, int) \
|
|
{ \
|
|
type e_org = e; \
|
|
e = (type)((int)e + 1); \
|
|
return e_org; \
|
|
} \
|
|
FORCEINLINE type operator --(type& e, int) \
|
|
{ \
|
|
type e_org = e; \
|
|
e = (type)((int)e - 1); \
|
|
return e_org; \
|
|
}
|
|
|
|
|
|
|
|
/** Operators to allow to work with enum as with type safe bit set in C++ */
|
|
# define DECLARE_ENUM_AS_BIT_SET(mask_t) \
|
|
FORCEINLINE mask_t operator | (mask_t m1, mask_t m2) {return (mask_t)((int)m1 | m2);} \
|
|
FORCEINLINE mask_t operator & (mask_t m1, mask_t m2) {return (mask_t)((int)m1 & m2);} \
|
|
FORCEINLINE mask_t operator ^ (mask_t m1, mask_t m2) {return (mask_t)((int)m1 ^ m2);} \
|
|
FORCEINLINE mask_t& operator |= (mask_t& m1, mask_t m2) {m1 = m1 | m2; return m1;} \
|
|
FORCEINLINE mask_t& operator &= (mask_t& m1, mask_t m2) {m1 = m1 & m2; return m1;} \
|
|
FORCEINLINE mask_t& operator ^= (mask_t& m1, mask_t m2) {m1 = m1 ^ m2; return m1;} \
|
|
FORCEINLINE mask_t operator ~(mask_t m) {return (mask_t)(~(int)m);}
|
|
|
|
|
|
/** Informative template class exposing basic enumeration properties used by several
|
|
* other templates below. Here we have only forward declaration. For each enum type
|
|
* we will create specialization derived from MakeEnumPropsT<>.
|
|
* i.e.:
|
|
* template <> struct EnumPropsT<Track> : MakeEnumPropsT<Track, byte, TRACK_BEGIN, TRACK_END, INVALID_TRACK> {};
|
|
* followed by:
|
|
* typedef TinyEnumT<Track> TrackByte;
|
|
*/
|
|
template <typename Tenum_t> struct EnumPropsT;
|
|
|
|
/** Helper template class that makes basic properties of given enumeration type visible
|
|
* from outsize. It is used as base class of several EnumPropsT specializations each
|
|
* dedicated to one of commonly used enumeration types.
|
|
* @param Tenum_t enumeration type that you want to describe
|
|
* @param Tstorage_t what storage type would be sufficient (i.e. byte)
|
|
* @param Tbegin first valid value from the contiguous range (i.e. TRACK_BEGIN)
|
|
* @param Tend one past the last valid value from the contiguous range (i.e. TRACK_END)
|
|
* @param Tinvalid value used as invalid value marker (i.e. INVALID_TRACK)
|
|
*/
|
|
template <typename Tenum_t, typename Tstorage_t, Tenum_t Tbegin, Tenum_t Tend, Tenum_t Tinvalid>
|
|
struct MakeEnumPropsT {
|
|
typedef Tenum_t type; ///< enum type (i.e. Trackdir)
|
|
typedef Tstorage_t storage; ///< storage type (i.e. byte)
|
|
static const Tenum_t begin = Tbegin; ///< lowest valid value (i.e. TRACKDIR_BEGIN)
|
|
static const Tenum_t end = Tend; ///< one after the last valid value (i.e. TRACKDIR_END)
|
|
static const Tenum_t invalid = Tinvalid; ///< what value is used as invalid value (i.e. INVALID_TRACKDIR)
|
|
};
|
|
|
|
|
|
|
|
/** In some cases we use byte or uint16 to store values that are defined as enum. It is
|
|
* necessary in order to control the sizeof() such values. Some compilers make enum
|
|
* the same size as int (4 or 8 bytes instead of 1 or 2). As a consequence the strict
|
|
* compiler type-checking causes errors like:
|
|
* 'HasPowerOnRail' : cannot convert parameter 1 from 'byte' to 'RailType' when
|
|
* u->u.rail.railtype is passed as argument or type RailType. In such cases it is better
|
|
* to teach the compiler that u->u.rail.railtype is to be treated as RailType. */
|
|
template <typename Tenum_t> struct TinyEnumT;
|
|
|
|
/** The general declaration of TinyEnumT<> (above) */
|
|
template <typename Tenum_t> struct TinyEnumT
|
|
{
|
|
typedef Tenum_t enum_type; ///< expose our enumeration type (i.e. Trackdir) to outside
|
|
typedef EnumPropsT<Tenum_t> Props; ///< make easier access to our enumeration propeties
|
|
typedef typename Props::storage storage_type; ///< small storage type
|
|
static const enum_type begin = Props::begin; ///< enum beginning (i.e. TRACKDIR_BEGIN)
|
|
static const enum_type end = Props::end; ///< enum end (i.e. TRACKDIR_END)
|
|
static const enum_type invalid = Props::invalid;///< invalid value (i.e. INVALID_TRACKDIR)
|
|
|
|
storage_type m_val; ///< here we hold the actual value in small (i.e. byte) form
|
|
|
|
/** Cast operator - invoked then the value is assigned to the Tenum_t type */
|
|
FORCEINLINE operator enum_type () const
|
|
{
|
|
return (enum_type)m_val;
|
|
}
|
|
|
|
/** Assignment operator (from Tenum_t type) */
|
|
FORCEINLINE TinyEnumT& operator = (enum_type e)
|
|
{
|
|
m_val = (storage_type)e; return *this;
|
|
}
|
|
|
|
/** postfix ++ operator on tiny type */
|
|
FORCEINLINE TinyEnumT& operator ++ (int)
|
|
{
|
|
if (++m_val >= end) m_val -= (storage_type)(end - begin);
|
|
return *this;
|
|
}
|
|
};
|
|
|
|
template <typename T> void ClrBitT(T &t, int bit_index)
|
|
{
|
|
t = (T)(t & ~((T)1 << bit_index));
|
|
}
|
|
|
|
template <typename T> void SetBitT(T &t, int bit_index)
|
|
{
|
|
t = (T)(t | ((T)1 << bit_index));
|
|
}
|
|
|
|
template <typename T> void ToggleBitT(T &t, int bit_index)
|
|
{
|
|
t = (T)(t ^ ((T)1 << bit_index));
|
|
}
|
|
|
|
/**
|
|
* Overflow safe template for integers, i.e. integers that will never overflow
|
|
* you multiply the maximum value with 2, or add 2, or substract somethng from
|
|
* the minimum value, etc.
|
|
* @param T the type these integers are stored with.
|
|
* @param T_MAX the maximum value for the integers.
|
|
* @param T_MIN the minimum value for the integers.
|
|
*/
|
|
template <class T, T T_MAX, T T_MIN>
|
|
class OverflowSafeInt
|
|
{
|
|
private:
|
|
/** The non-overflow safe backend to store the value in. */
|
|
T m_value;
|
|
public:
|
|
OverflowSafeInt() : m_value(0) { }
|
|
|
|
OverflowSafeInt(const OverflowSafeInt& other) { this->m_value = other.m_value; }
|
|
OverflowSafeInt(const int64 int_) { this->m_value = int_; }
|
|
|
|
FORCEINLINE OverflowSafeInt& operator = (const OverflowSafeInt& other) { this->m_value = other.m_value; return *this; }
|
|
|
|
FORCEINLINE OverflowSafeInt operator - () const { return OverflowSafeInt(-this->m_value); }
|
|
|
|
/**
|
|
* Safe implementation of addition.
|
|
* @param other the amount to add
|
|
* @note when the addition would yield more than T_MAX (or less than T_MIN),
|
|
* it will be T_MAX (respectively T_MIN).
|
|
*/
|
|
FORCEINLINE OverflowSafeInt& operator += (const OverflowSafeInt& other)
|
|
{
|
|
if ((T_MAX - myabs(other.m_value)) < myabs(this->m_value) &&
|
|
(this->m_value < 0) == (other.m_value < 0)) {
|
|
this->m_value = (this->m_value < 0) ? T_MIN : T_MAX ;
|
|
} else {
|
|
this->m_value += other.m_value;
|
|
}
|
|
return *this;
|
|
}
|
|
|
|
/* Operators for addition and substraction */
|
|
FORCEINLINE OverflowSafeInt operator + (const OverflowSafeInt& other) const { OverflowSafeInt result = *this; result += other; return result; }
|
|
FORCEINLINE OverflowSafeInt operator + (const int other) const { OverflowSafeInt result = *this; result += (int64)other; return result; }
|
|
FORCEINLINE OverflowSafeInt operator + (const uint other) const { OverflowSafeInt result = *this; result += (int64)other; return result; }
|
|
FORCEINLINE OverflowSafeInt& operator -= (const OverflowSafeInt& other) { return *this += (-other); }
|
|
FORCEINLINE OverflowSafeInt operator - (const OverflowSafeInt& other) const { OverflowSafeInt result = *this; result -= other; return result; }
|
|
FORCEINLINE OverflowSafeInt operator - (const int other) const { OverflowSafeInt result = *this; result -= (int64)other; return result; }
|
|
FORCEINLINE OverflowSafeInt operator - (const uint other) const { OverflowSafeInt result = *this; result -= (int64)other; return result; }
|
|
|
|
FORCEINLINE OverflowSafeInt& operator ++ (int) { return *this += 1; }
|
|
FORCEINLINE OverflowSafeInt& operator -- (int) { return *this += -1; }
|
|
|
|
/**
|
|
* Safe implementation of multiplication.
|
|
* @param factor the factor to multiply this with.
|
|
* @note when the multiplication would yield more than T_MAX (or less than T_MIN),
|
|
* it will be T_MAX (respectively T_MIN).
|
|
*/
|
|
FORCEINLINE OverflowSafeInt& operator *= (const int factor)
|
|
{
|
|
if (factor != 0 && (T_MAX / myabs(factor)) < myabs(this->m_value)) {
|
|
this->m_value = ((this->m_value < 0) == (factor < 0)) ? T_MAX : T_MIN ;
|
|
} else {
|
|
this->m_value *= factor ;
|
|
}
|
|
return *this;
|
|
}
|
|
|
|
/* Operators for multiplication */
|
|
FORCEINLINE OverflowSafeInt operator * (const int64 factor) const { OverflowSafeInt result = *this; result *= factor; return result; }
|
|
FORCEINLINE OverflowSafeInt operator * (const int factor) const { OverflowSafeInt result = *this; result *= (int64)factor; return result; }
|
|
FORCEINLINE OverflowSafeInt operator * (const uint factor) const { OverflowSafeInt result = *this; result *= (int64)factor; return result; }
|
|
FORCEINLINE OverflowSafeInt operator * (const uint16 factor) const { OverflowSafeInt result = *this; result *= (int64)factor; return result; }
|
|
FORCEINLINE OverflowSafeInt operator * (const byte factor) const { OverflowSafeInt result = *this; result *= (int64)factor; return result; }
|
|
|
|
/* Operators for division */
|
|
FORCEINLINE OverflowSafeInt& operator /= (const int divisor) { this->m_value /= divisor; return *this; }
|
|
FORCEINLINE OverflowSafeInt operator / (const OverflowSafeInt& divisor) const { OverflowSafeInt result = *this; result /= divisor.m_value; return result; }
|
|
FORCEINLINE OverflowSafeInt operator / (const int divisor) const { OverflowSafeInt result = *this; result /= divisor; return result; }
|
|
FORCEINLINE OverflowSafeInt operator / (const uint divisor) const { OverflowSafeInt result = *this; result /= (int)divisor; return result; }
|
|
|
|
/* Operators for modulo */
|
|
FORCEINLINE OverflowSafeInt& operator %= (const int divisor) { this->m_value %= divisor; return *this; }
|
|
FORCEINLINE OverflowSafeInt operator % (const int divisor) const { OverflowSafeInt result = *this; result %= divisor; return result; }
|
|
|
|
/* Operators for shifting */
|
|
FORCEINLINE OverflowSafeInt& operator <<= (const int shift) { this->m_value <<= shift; return *this; }
|
|
FORCEINLINE OverflowSafeInt operator << (const int shift) const { OverflowSafeInt result = *this; result <<= shift; return result; }
|
|
FORCEINLINE OverflowSafeInt& operator >>= (const int shift) { this->m_value >>= shift; return *this; }
|
|
FORCEINLINE OverflowSafeInt operator >> (const int shift) const { OverflowSafeInt result = *this; result >>= shift; return result; }
|
|
|
|
/* Operators for (in)equality when comparing overflow safe ints */
|
|
FORCEINLINE bool operator == (const OverflowSafeInt& other) const { return this->m_value == other.m_value; }
|
|
FORCEINLINE bool operator != (const OverflowSafeInt& other) const { return !(*this == other); }
|
|
FORCEINLINE bool operator > (const OverflowSafeInt& other) const { return this->m_value > other.m_value; }
|
|
FORCEINLINE bool operator >= (const OverflowSafeInt& other) const { return this->m_value >= other.m_value; }
|
|
FORCEINLINE bool operator < (const OverflowSafeInt& other) const { return !(*this >= other); }
|
|
FORCEINLINE bool operator <= (const OverflowSafeInt& other) const { return !(*this > other); }
|
|
|
|
/* Operators for (in)equality when comparing non-overflow safe ints */
|
|
FORCEINLINE bool operator == (const int other) const { return this->m_value == other; }
|
|
FORCEINLINE bool operator != (const int other) const { return !(*this == other); }
|
|
FORCEINLINE bool operator > (const int other) const { return this->m_value > other; }
|
|
FORCEINLINE bool operator >= (const int other) const { return this->m_value >= other; }
|
|
FORCEINLINE bool operator < (const int other) const { return !(*this >= other); }
|
|
FORCEINLINE bool operator <= (const int other) const { return !(*this > other); }
|
|
|
|
FORCEINLINE operator int64 () const { return this->m_value; }
|
|
};
|
|
|
|
/* Sometimes we got int64 operator OverflowSafeInt instead of vice versa. Handle that properly */
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator + (int64 a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return b + a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator - (int64 a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return -b + a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator * (int64 a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return b * a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator / (int64 a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return (OverflowSafeInt<T, T_MAX, T_MIN>)a / (int)b; }
|
|
|
|
/* Sometimes we got int operator OverflowSafeInt instead of vice versa. Handle that properly */
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator + (int a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return b + a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator - (int a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return -b + a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator * (int a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return b * a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator / (int a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return (OverflowSafeInt<T, T_MAX, T_MIN>)a / (int)b; }
|
|
|
|
/* Sometimes we got uint operator OverflowSafeInt instead of vice versa. Handle that properly */
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator + (uint a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return b + a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator - (uint a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return -b + a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator * (uint a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return b * a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator / (uint a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return (OverflowSafeInt<T, T_MAX, T_MIN>)a / (int)b; }
|
|
|
|
/* Sometimes we got byte operator OverflowSafeInt instead of vice versa. Handle that properly */
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator + (byte a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return b + a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator - (byte a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return -b + a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator * (byte a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return b * a; }
|
|
template <class T, int64 T_MAX, int64 T_MIN> FORCEINLINE OverflowSafeInt<T, T_MAX, T_MIN> operator / (byte a, OverflowSafeInt<T, T_MAX, T_MIN> b) { return (OverflowSafeInt<T, T_MAX, T_MIN>)a / (int)b; }
|
|
|
|
#endif /* HELPERS_HPP */
|