Improved g++ compatibility. Still not tested well enough.

This commit is contained in:
Leonetienne
2022-01-20 23:09:53 +01:00
parent 131a2239b3
commit 58c369e025
8 changed files with 115 additions and 5 deletions

View File

@@ -32,6 +32,7 @@
<ItemGroup>
<ClInclude Include="Collider.h" />
<ClInclude Include="Constants.h" />
<ClInclude Include="gcccompat.h" />
<ClInclude Include="Math.h" />
<ClInclude Include="Matrix4x4.h" />
<ClInclude Include="Quaternion.h" />

View File

@@ -80,5 +80,8 @@
<ClInclude Include="Random.h">
<Filter>Headerdateien</Filter>
</ClInclude>
<ClInclude Include="gcccompat.h">
<Filter>Headerdateien</Filter>
</ClInclude>
</ItemGroup>
</Project>

View File

@@ -1,9 +1,12 @@
#include "Quaternion.h"
#include "Constants.h"
#include <algorithm>
#include <functional>
//#define _EULE_NO_INTRINSICS_
#ifndef _EULE_NO_INTRINSICS_
#include <immintrin.h>
#include "gcccompat.h"
#endif
using namespace Eule;

View File

@@ -1,5 +1,5 @@
#include "Random.h"
#include <Array>
#include <array>
using namespace Eule;

View File

@@ -19,6 +19,7 @@ using namespace Eule;
*/
// Good, optimized chad version for doubles
template<>
double Vector2<double>::DotProduct(const Vector2<double>& other) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -44,6 +45,7 @@ double Vector2<double>::DotProduct(const Vector2<double>& other) const
}
// Slow, lame version for intcels
template<>
double Vector2<int>::DotProduct(const Vector2<int>& other) const
{
int iDot = (x * other.x) +
@@ -55,6 +57,7 @@ double Vector2<int>::DotProduct(const Vector2<int>& other) const
// Good, optimized chad version for doubles
template<>
double Vector2<double>::CrossProduct(const Vector2<double>& other) const
{
return (x * other.y) -
@@ -62,6 +65,7 @@ double Vector2<double>::CrossProduct(const Vector2<double>& other) const
}
// Slow, lame version for intcels
template<>
double Vector2<int>::CrossProduct(const Vector2<int>& other) const
{
int iCross = (x * other.y) -
@@ -73,6 +77,7 @@ double Vector2<int>::CrossProduct(const Vector2<int>& other) const
// Good, optimized chad version for doubles
template<>
double Vector2<double>::SqrMagnitude() const
{
// x.DotProduct(x) == x.SqrMagnitude()
@@ -80,6 +85,7 @@ double Vector2<double>::SqrMagnitude() const
}
// Slow, lame version for intcels
template<>
double Vector2<int>::SqrMagnitude() const
{
int iSqrMag = x*x + y*y;
@@ -93,7 +99,7 @@ double Vector2<T>::Magnitude() const
}
template<>
Vector2<double> Vector2<double>::VectorScale(const Vector2<double>& scalar) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -124,6 +130,7 @@ Vector2<double> Vector2<double>::VectorScale(const Vector2<double>& scalar) cons
#endif
}
template<>
Vector2<int> Vector2<int>::VectorScale(const Vector2<int>& scalar) const
{
return Vector2<int>(
@@ -143,6 +150,7 @@ Vector2<double> Vector2<T>::Normalize() const
}
// Method to normalize a Vector2d
template<>
void Vector2<double>::NormalizeSelf()
{
double length = Magnitude();
@@ -184,6 +192,7 @@ void Vector2<double>::NormalizeSelf()
// You can't normalize an int vector, ffs!
// But we need an implementation for T=int
template<>
void Vector2<int>::NormalizeSelf()
{
std::cerr << "Stop normalizing int-vectors!!" << std::endl;
@@ -195,6 +204,7 @@ void Vector2<int>::NormalizeSelf()
// Good, optimized chad version for doubles
template<>
void Vector2<double>::LerpSelf(const Vector2<double>& other, double t)
{
const double it = 1.0 - t; // Inverse t
@@ -235,6 +245,7 @@ void Vector2<double>::LerpSelf(const Vector2<double>& other, double t)
// Slow, lame version for intcels
template<>
void Vector2<int>::LerpSelf(const Vector2<int>& other, double t)
{
const double it = 1.0 - t; // Inverse t
@@ -245,6 +256,7 @@ void Vector2<int>::LerpSelf(const Vector2<int>& other, double t)
return;
}
template<>
Vector2<double> Vector2<double>::Lerp(const Vector2<double>& other, double t) const
{
Vector2d copy(*this);
@@ -253,6 +265,7 @@ Vector2<double> Vector2<double>::Lerp(const Vector2<double>& other, double t) co
return copy;
}
template<>
Vector2<double> Vector2<int>::Lerp(const Vector2<int>& other, double t) const
{
Vector2d copy(this->ToDouble());
@@ -312,7 +325,7 @@ Vector2<double> Vector2<T>::ToDouble() const
return Vector2<double>((double)x, (double)y);
}
template<>
Vector2<double> Vector2<double>::operator+(const Vector2<double>& other) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -353,6 +366,7 @@ Vector2<T> Vector2<T>::operator+(const Vector2<T>& other) const
template<>
void Vector2<double>::operator+=(const Vector2<double>& other)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -391,6 +405,7 @@ void Vector2<T>::operator+=(const Vector2<T>& other)
template<>
Vector2<double> Vector2<double>::operator-(const Vector2<double>& other) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -431,6 +446,7 @@ Vector2<T> Vector2<T>::operator-(const Vector2<T>& other) const
template<>
void Vector2<double>::operator-=(const Vector2<double>& other)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -469,6 +485,7 @@ void Vector2<T>::operator-=(const Vector2<T>& other)
template<>
Vector2<double> Vector2<double>::operator*(const double scale) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -510,6 +527,7 @@ Vector2<T> Vector2<T>::operator*(const T scale) const
template<>
void Vector2<double>::operator*=(const double scale)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -548,6 +566,7 @@ void Vector2<T>::operator*=(const T scale)
template<>
Vector2<double> Vector2<double>::operator/(const double scale) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -589,6 +608,7 @@ Vector2<T> Vector2<T>::operator/(const T scale) const
template<>
void Vector2<double>::operator/=(const double scale)
{
#ifndef _EULE_NO_INTRINSICS_

View File

@@ -19,6 +19,7 @@ using namespace Eule;
*/
// Good, optimized chad version for doubles
template<>
double Vector3<double>::DotProduct(const Vector3<double>& other) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -45,6 +46,7 @@ double Vector3<double>::DotProduct(const Vector3<double>& other) const
}
// Slow, lame version for intcels
template<>
double Vector3<int>::DotProduct(const Vector3<int>& other) const
{
int iDot = (x * other.x) + (y * other.y) + (z * other.z);
@@ -54,6 +56,7 @@ double Vector3<int>::DotProduct(const Vector3<int>& other) const
// Good, optimized chad version for doubles
template<>
Vector3<double> Vector3<double>::CrossProduct(const Vector3<double>& other) const
{
Vector3<double> cp;
@@ -65,6 +68,7 @@ Vector3<double> Vector3<double>::CrossProduct(const Vector3<double>& other) cons
}
// Slow, lame version for intcels
template<>
Vector3<double> Vector3<int>::CrossProduct(const Vector3<int>& other) const
{
Vector3<double> cp;
@@ -78,6 +82,7 @@ Vector3<double> Vector3<int>::CrossProduct(const Vector3<int>& other) const
// Good, optimized chad version for doubles
template<>
double Vector3<double>::SqrMagnitude() const
{
// x.DotProduct(x) == x.SqrMagnitude()
@@ -85,6 +90,7 @@ double Vector3<double>::SqrMagnitude() const
}
// Slow, lame version for intcels
template<>
double Vector3<int>::SqrMagnitude() const
{
int iSqrMag = x*x + y*y + z*z;
@@ -99,6 +105,7 @@ double Vector3<T>::Magnitude() const
template<>
Vector3<double> Vector3<double>::VectorScale(const Vector3<double>& scalar) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -132,6 +139,7 @@ Vector3<double> Vector3<double>::VectorScale(const Vector3<double>& scalar) cons
#endif
}
template<>
Vector3<int> Vector3<int>::VectorScale(const Vector3<int>& scalar) const
{
return Vector3<int>(
@@ -153,6 +161,7 @@ Vector3<double> Vector3<T>::Normalize() const
}
// Method to normalize a Vector3d
template<>
void Vector3<double>::NormalizeSelf()
{
const double length = Magnitude();
@@ -197,6 +206,7 @@ void Vector3<double>::NormalizeSelf()
// You can't normalize an int vector, ffs!
// But we need an implementation for T=int
template<>
void Vector3<int>::NormalizeSelf()
{
std::cerr << "Stop normalizing int-vectors!!" << std::endl;
@@ -266,6 +276,7 @@ const T& Vector3<T>::operator[](std::size_t idx) const
// Good, optimized chad version for doubles
template<>
void Vector3<double>::LerpSelf(const Vector3<double>& other, double t)
{
const double it = 1.0 - t; // Inverse t
@@ -308,6 +319,7 @@ void Vector3<double>::LerpSelf(const Vector3<double>& other, double t)
// Slow, lame version for intcels
template<>
void Vector3<int>::LerpSelf(const Vector3<int>& other, double t)
{
const double it = 1.0 - t; // Inverse t
@@ -319,6 +331,7 @@ void Vector3<int>::LerpSelf(const Vector3<int>& other, double t)
return;
}
template<>
Vector3<double> Vector3<double>::Lerp(const Vector3<double>& other, double t) const
{
Vector3d copy(*this);
@@ -327,6 +340,7 @@ Vector3<double> Vector3<double>::Lerp(const Vector3<double>& other, double t) co
return copy;
}
template<>
Vector3<double> Vector3<int>::Lerp(const Vector3<int>& other, double t) const
{
Vector3d copy(this->ToDouble());
@@ -337,6 +351,7 @@ Vector3<double> Vector3<int>::Lerp(const Vector3<int>& other, double t) const
template<>
Vector3<double> Vector3<double>::operator+(const Vector3<double>& other) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -380,6 +395,7 @@ Vector3<T> Vector3<T>::operator+(const Vector3<T>& other) const
template<>
void Vector3<double>::operator+=(const Vector3<double>& other)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -421,6 +437,7 @@ void Vector3<T>::operator+=(const Vector3<T>& other)
template<>
Vector3<double> Vector3<double>::operator-(const Vector3<double>& other) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -464,6 +481,7 @@ Vector3<T> Vector3<T>::operator-(const Vector3<T>& other) const
template<>
void Vector3<double>::operator-=(const Vector3<double>& other)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -505,6 +523,7 @@ void Vector3<T>::operator-=(const Vector3<T>& other)
template<>
Vector3<double> Vector3<double>::operator*(const double scale) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -549,6 +568,7 @@ Vector3<T> Vector3<T>::operator*(const T scale) const
template<>
void Vector3<double>::operator*=(const double scale)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -590,6 +610,7 @@ void Vector3<T>::operator*=(const T scale)
template<>
Vector3<double> Vector3<double>::operator/(const double scale) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -634,6 +655,7 @@ Vector3<T> Vector3<T>::operator/(const T scale) const
template<>
void Vector3<double>::operator/=(const double scale)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -674,6 +696,7 @@ void Vector3<T>::operator/=(const T scale)
// Good, optimized chad version for doubles
template<>
Vector3<double> Vector3<double>::operator*(const Matrix4x4& mat) const
{
Vector3<double> newVec;
@@ -727,6 +750,7 @@ Vector3<double> Vector3<double>::operator*(const Matrix4x4& mat) const
}
// Slow, lame version for intcels
template<>
Vector3<int> Vector3<int>::operator*(const Matrix4x4& mat) const
{
Vector3<double> newVec;
@@ -751,6 +775,7 @@ Vector3<int> Vector3<int>::operator*(const Matrix4x4& mat) const
// Good, optimized chad version for doubles
template<>
void Vector3<double>::operator*=(const Matrix4x4& mat)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -832,6 +857,7 @@ void Vector3<T>::operator=(Vector3<T>&& other) noexcept
}
// Slow, lame version for intcels
template<>
void Vector3<int>::operator*=(const Matrix4x4& mat)
{
Vector3<double> buffer(x, y, z);

View File

@@ -19,6 +19,7 @@ using namespace Eule;
*/
// Good, optimized chad version for doubles
template<>
double Vector4<double>::SqrMagnitude() const
{
return (x * x) +
@@ -28,6 +29,7 @@ double Vector4<double>::SqrMagnitude() const
}
// Slow, lame version for intcels
template<>
double Vector4<int>::SqrMagnitude() const
{
int iSqrMag = x*x + y*y + z*z + w*w;
@@ -41,6 +43,7 @@ double Vector4<T>::Magnitude() const
}
template<>
Vector4<double> Vector4<double>::VectorScale(const Vector4<double>& scalar) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -76,6 +79,7 @@ Vector4<double> Vector4<double>::VectorScale(const Vector4<double>& scalar) cons
}
template<>
Vector4<int> Vector4<int>::VectorScale(const Vector4<int>& scalar) const
{
return Vector4<int>(
@@ -97,7 +101,8 @@ Vector4<double> Vector4<T>::Normalize() const
return norm;
}
// Method to normalize a Vector43d
// Method to normalize a Vector4d
template<>
void Vector4<double>::NormalizeSelf()
{
double length = Magnitude();
@@ -145,6 +150,7 @@ void Vector4<double>::NormalizeSelf()
// You can't normalize an int vector, ffs!
// But we need an implementation for T=int
template<>
void Vector4<int>::NormalizeSelf()
{
std::cerr << "Stop normalizing int-vectors!!" << std::endl;
@@ -220,6 +226,7 @@ const T& Vector4<T>::operator[](std::size_t idx) const
// Good, optimized chad version for doubles
template<>
void Vector4<double>::LerpSelf(const Vector4<double>& other, double t)
{
const double it = 1.0 - t; // Inverse t
@@ -264,6 +271,7 @@ void Vector4<double>::LerpSelf(const Vector4<double>& other, double t)
// Slow, lame version for intcels
template<>
void Vector4<int>::LerpSelf(const Vector4<int>& other, double t)
{
const double it = 1.0 - t;
@@ -276,6 +284,7 @@ void Vector4<int>::LerpSelf(const Vector4<int>& other, double t)
return;
}
template<>
Vector4<double> Vector4<double>::Lerp(const Vector4<double>& other, double t) const
{
Vector4d copy(*this);
@@ -284,6 +293,7 @@ Vector4<double> Vector4<double>::Lerp(const Vector4<double>& other, double t) co
return copy;
}
template<>
Vector4<double> Vector4<int>::Lerp(const Vector4<int>& other, double t) const
{
Vector4d copy(this->ToDouble());
@@ -294,6 +304,7 @@ Vector4<double> Vector4<int>::Lerp(const Vector4<int>& other, double t) const
template<>
Vector4<double> Vector4<double>::operator+(const Vector4<double>& other) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -340,6 +351,7 @@ Vector4<T> Vector4<T>::operator+(const Vector4<T>& other) const
template<>
void Vector4<double>::operator+=(const Vector4<double>& other)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -384,6 +396,7 @@ void Vector4<T>::operator+=(const Vector4<T>& other)
template<>
Vector4<double> Vector4<double>::operator-(const Vector4<double>& other) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -430,6 +443,7 @@ Vector4<T> Vector4<T>::operator-(const Vector4<T>& other) const
template<>
void Vector4<double>::operator-=(const Vector4<double>& other)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -474,6 +488,7 @@ void Vector4<T>::operator-=(const Vector4<T>& other)
template<>
Vector4<double> Vector4<double>::operator*(const double scale) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -521,6 +536,7 @@ Vector4<T> Vector4<T>::operator*(const T scale) const
template<>
void Vector4<double>::operator*=(const double scale)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -565,6 +581,7 @@ void Vector4<T>::operator*=(const T scale)
template<>
Vector4<double> Vector4<double>::operator/(const double scale) const
{
#ifndef _EULE_NO_INTRINSICS_
@@ -612,6 +629,7 @@ Vector4<T> Vector4<T>::operator/(const T scale) const
template<>
void Vector4<double>::operator/=(const double scale)
{
#ifndef _EULE_NO_INTRINSICS_
@@ -668,6 +686,7 @@ bool Vector4<T>::operator==(const Vector4<T>& other) const
// Good, optimized chad version for doubles
template<>
Vector4<double> Vector4<double>::operator*(const Matrix4x4& mat) const
{
Vector4<double> newVec;
@@ -681,6 +700,7 @@ Vector4<double> Vector4<double>::operator*(const Matrix4x4& mat) const
}
// Slow, lame version for intcels
template<>
Vector4<int> Vector4<int>::operator*(const Matrix4x4& mat) const
{
Vector4<double> newVec;
@@ -701,6 +721,7 @@ Vector4<int> Vector4<int>::operator*(const Matrix4x4& mat) const
// Good, optimized chad version for doubles
template<>
void Vector4<double>::operator*=(const Matrix4x4& mat)
{
Vector4<double> buffer = *this;
@@ -749,6 +770,7 @@ void Vector4<T>::operator=(Vector4<T>&& other) noexcept
}
// Slow, lame version for intcels
template<>
void Vector4<int>::operator*=(const Matrix4x4& mat)
{
Vector4<double> buffer(x, y, z, w);

35
Eule/gcccompat.h Normal file
View File

@@ -0,0 +1,35 @@
#pragma once
/*
* Some intrinsic functions such as _mm_sincos_pd are not available on g++ by default (requires some specific library).
* So let's just "re"define them manually if we're on g++.
* This way the code still works, even with the other intrinsics enabled.
*/
#if (__GNUC__ && __cplusplus)
#include <immintrin.h>
#include <math.h>
inline __m256d _mm256_sincos_pd(__m256d* __cos, __m256d __vec)
{
double vec[4];
_mm256_storeu_pd(vec, __vec);
// Manually calculate cosines
*__cos = _mm256_set_pd(
cos(vec[3]),
cos(vec[2]),
cos(vec[1]),
cos(vec[0])
);
// Manually calculate sines
return _mm256_set_pd(
sin(vec[3]),
sin(vec[2]),
sin(vec[1]),
sin(vec[0])
);
}
#endif