GRK/cw 4/dependencies/glm/detail/func_integer.inl

369 lines
15 KiB
Plaintext
Raw Normal View History

2022-01-12 16:07:16 +01:00
/// @ref core
/// @file glm/detail/func_integer.inl
#include "type_vec2.hpp"
#include "type_vec3.hpp"
#include "type_vec4.hpp"
#include "type_int.hpp"
#include "_vectorize.hpp"
#if(GLM_ARCH & GLM_ARCH_X86 && GLM_COMPILER & GLM_COMPILER_VC)
# include <intrin.h>
# pragma intrinsic(_BitScanReverse)
#endif//(GLM_ARCH & GLM_ARCH_X86 && GLM_COMPILER & GLM_COMPILER_VC)
#include <limits>
#if !GLM_HAS_EXTENDED_INTEGER_TYPE
# if GLM_COMPILER & GLM_COMPILER_GCC
# pragma GCC diagnostic ignored "-Wlong-long"
# endif
# if (GLM_COMPILER & GLM_COMPILER_CLANG)
# pragma clang diagnostic ignored "-Wc++11-long-long"
# endif
#endif
namespace glm{
namespace detail
{
template <typename T>
GLM_FUNC_QUALIFIER T mask(T Bits)
{
return Bits >= sizeof(T) * 8 ? ~static_cast<T>(0) : (static_cast<T>(1) << Bits) - static_cast<T>(1);
}
template <typename T, glm::precision P, template <typename, glm::precision> class vecType, bool Aligned, bool EXEC>
struct compute_bitfieldReverseStep
{
GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & v, T, T)
{
return v;
}
};
template <typename T, glm::precision P, template <typename, glm::precision> class vecType, bool Aligned>
struct compute_bitfieldReverseStep<T, P, vecType, Aligned, true>
{
GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & v, T Mask, T Shift)
{
return (v & Mask) << Shift | (v & (~Mask)) >> Shift;
}
};
template <typename T, glm::precision P, template <typename, glm::precision> class vecType, bool Aligned, bool EXEC>
struct compute_bitfieldBitCountStep
{
GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & v, T, T)
{
return v;
}
};
template <typename T, glm::precision P, template <typename, glm::precision> class vecType, bool Aligned>
struct compute_bitfieldBitCountStep<T, P, vecType, Aligned, true>
{
GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & v, T Mask, T Shift)
{
return (v & Mask) + ((v >> Shift) & Mask);
}
};
template <typename genIUType, size_t Bits>
struct compute_findLSB
{
GLM_FUNC_QUALIFIER static int call(genIUType Value)
{
if(Value == 0)
return -1;
return glm::bitCount(~Value & (Value - static_cast<genIUType>(1)));
}
};
# if GLM_HAS_BITSCAN_WINDOWS
template <typename genIUType>
struct compute_findLSB<genIUType, 32>
{
GLM_FUNC_QUALIFIER static int call(genIUType Value)
{
unsigned long Result(0);
unsigned char IsNotNull = _BitScanForward(&Result, *reinterpret_cast<unsigned long*>(&Value));
return IsNotNull ? int(Result) : -1;
}
};
# if !((GLM_COMPILER & GLM_COMPILER_VC) && (GLM_MODEL == GLM_MODEL_32))
template <typename genIUType>
struct compute_findLSB<genIUType, 64>
{
GLM_FUNC_QUALIFIER static int call(genIUType Value)
{
unsigned long Result(0);
unsigned char IsNotNull = _BitScanForward64(&Result, *reinterpret_cast<unsigned __int64*>(&Value));
return IsNotNull ? int(Result) : -1;
}
};
# endif
# endif//GLM_HAS_BITSCAN_WINDOWS
template <typename T, glm::precision P, template <class, glm::precision> class vecType, bool EXEC = true>
struct compute_findMSB_step_vec
{
GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x, T Shift)
{
return x | (x >> Shift);
}
};
template <typename T, glm::precision P, template <typename, glm::precision> class vecType>
struct compute_findMSB_step_vec<T, P, vecType, false>
{
GLM_FUNC_QUALIFIER static vecType<T, P> call(vecType<T, P> const & x, T)
{
return x;
}
};
template <typename T, glm::precision P, template <typename, glm::precision> class vecType, int>
struct compute_findMSB_vec
{
GLM_FUNC_QUALIFIER static vecType<int, P> call(vecType<T, P> const & vec)
{
vecType<T, P> x(vec);
x = compute_findMSB_step_vec<T, P, vecType, sizeof(T) * 8 >= 8>::call(x, static_cast<T>( 1));
x = compute_findMSB_step_vec<T, P, vecType, sizeof(T) * 8 >= 8>::call(x, static_cast<T>( 2));
x = compute_findMSB_step_vec<T, P, vecType, sizeof(T) * 8 >= 8>::call(x, static_cast<T>( 4));
x = compute_findMSB_step_vec<T, P, vecType, sizeof(T) * 8 >= 16>::call(x, static_cast<T>( 8));
x = compute_findMSB_step_vec<T, P, vecType, sizeof(T) * 8 >= 32>::call(x, static_cast<T>(16));
x = compute_findMSB_step_vec<T, P, vecType, sizeof(T) * 8 >= 64>::call(x, static_cast<T>(32));
return vecType<int, P>(sizeof(T) * 8 - 1) - glm::bitCount(~x);
}
};
# if GLM_HAS_BITSCAN_WINDOWS
template <typename genIUType>
GLM_FUNC_QUALIFIER int compute_findMSB_32(genIUType Value)
{
unsigned long Result(0);
unsigned char IsNotNull = _BitScanReverse(&Result, *reinterpret_cast<unsigned long*>(&Value));
return IsNotNull ? int(Result) : -1;
}
template <typename T, glm::precision P, template<typename, glm::precision> class vecType>
struct compute_findMSB_vec<T, P, vecType, 32>
{
GLM_FUNC_QUALIFIER static vecType<int, P> call(vecType<T, P> const & x)
{
return detail::functor1<int, T, P, vecType>::call(compute_findMSB_32, x);
}
};
# if !((GLM_COMPILER & GLM_COMPILER_VC) && (GLM_MODEL == GLM_MODEL_32))
template <typename genIUType>
GLM_FUNC_QUALIFIER int compute_findMSB_64(genIUType Value)
{
unsigned long Result(0);
unsigned char IsNotNull = _BitScanReverse64(&Result, *reinterpret_cast<unsigned __int64*>(&Value));
return IsNotNull ? int(Result) : -1;
}
template <typename T, glm::precision P, template <class, glm::precision> class vecType>
struct compute_findMSB_vec<T, P, vecType, 64>
{
GLM_FUNC_QUALIFIER static vecType<int, P> call(vecType<T, P> const & x)
{
return detail::functor1<int, T, P, vecType>::call(compute_findMSB_64, x);
}
};
# endif
# endif//GLM_HAS_BITSCAN_WINDOWS
}//namespace detail
// uaddCarry
GLM_FUNC_QUALIFIER uint uaddCarry(uint const & x, uint const & y, uint & Carry)
{
uint64 const Value64(static_cast<uint64>(x) + static_cast<uint64>(y));
uint64 const Max32((static_cast<uint64>(1) << static_cast<uint64>(32)) - static_cast<uint64>(1));
Carry = Value64 > Max32 ? 1u : 0u;
return static_cast<uint32>(Value64 % (Max32 + static_cast<uint64>(1)));
}
template <precision P, template <typename, precision> class vecType>
GLM_FUNC_QUALIFIER vecType<uint, P> uaddCarry(vecType<uint, P> const & x, vecType<uint, P> const & y, vecType<uint, P> & Carry)
{
vecType<uint64, P> Value64(vecType<uint64, P>(x) + vecType<uint64, P>(y));
vecType<uint64, P> Max32((static_cast<uint64>(1) << static_cast<uint64>(32)) - static_cast<uint64>(1));
Carry = mix(vecType<uint32, P>(0), vecType<uint32, P>(1), greaterThan(Value64, Max32));
return vecType<uint32,P>(Value64 % (Max32 + static_cast<uint64>(1)));
}
// usubBorrow
GLM_FUNC_QUALIFIER uint usubBorrow(uint const & x, uint const & y, uint & Borrow)
{
GLM_STATIC_ASSERT(sizeof(uint) == sizeof(uint32), "uint and uint32 size mismatch");
Borrow = x >= y ? static_cast<uint32>(0) : static_cast<uint32>(1);
if(y >= x)
return y - x;
else
return static_cast<uint32>((static_cast<int64>(1) << static_cast<int64>(32)) + (static_cast<int64>(y) - static_cast<int64>(x)));
}
template <precision P, template <typename, precision> class vecType>
GLM_FUNC_QUALIFIER vecType<uint, P> usubBorrow(vecType<uint, P> const & x, vecType<uint, P> const & y, vecType<uint, P> & Borrow)
{
Borrow = mix(vecType<uint, P>(1), vecType<uint, P>(0), greaterThanEqual(x, y));
vecType<uint, P> const YgeX(y - x);
vecType<uint, P> const XgeY(vecType<uint32, P>((static_cast<int64>(1) << static_cast<int64>(32)) + (vecType<int64, P>(y) - vecType<int64, P>(x))));
return mix(XgeY, YgeX, greaterThanEqual(y, x));
}
// umulExtended
GLM_FUNC_QUALIFIER void umulExtended(uint const & x, uint const & y, uint & msb, uint & lsb)
{
GLM_STATIC_ASSERT(sizeof(uint) == sizeof(uint32), "uint and uint32 size mismatch");
uint64 Value64 = static_cast<uint64>(x) * static_cast<uint64>(y);
msb = static_cast<uint>(Value64 >> static_cast<uint64>(32));
lsb = static_cast<uint>(Value64);
}
template <precision P, template <typename, precision> class vecType>
GLM_FUNC_QUALIFIER void umulExtended(vecType<uint, P> const & x, vecType<uint, P> const & y, vecType<uint, P> & msb, vecType<uint, P> & lsb)
{
GLM_STATIC_ASSERT(sizeof(uint) == sizeof(uint32), "uint and uint32 size mismatch");
vecType<uint64, P> Value64(vecType<uint64, P>(x) * vecType<uint64, P>(y));
msb = vecType<uint32, P>(Value64 >> static_cast<uint64>(32));
lsb = vecType<uint32, P>(Value64);
}
// imulExtended
GLM_FUNC_QUALIFIER void imulExtended(int x, int y, int & msb, int & lsb)
{
GLM_STATIC_ASSERT(sizeof(int) == sizeof(int32), "int and int32 size mismatch");
int64 Value64 = static_cast<int64>(x) * static_cast<int64>(y);
msb = static_cast<int>(Value64 >> static_cast<int64>(32));
lsb = static_cast<int>(Value64);
}
template <precision P, template <typename, precision> class vecType>
GLM_FUNC_QUALIFIER void imulExtended(vecType<int, P> const & x, vecType<int, P> const & y, vecType<int, P> & msb, vecType<int, P> & lsb)
{
GLM_STATIC_ASSERT(sizeof(int) == sizeof(int32), "int and int32 size mismatch");
vecType<int64, P> Value64(vecType<int64, P>(x) * vecType<int64, P>(y));
lsb = vecType<int32, P>(Value64 & static_cast<int64>(0xFFFFFFFF));
msb = vecType<int32, P>((Value64 >> static_cast<int64>(32)) & static_cast<int64>(0xFFFFFFFF));
}
// bitfieldExtract
template <typename genIUType>
GLM_FUNC_QUALIFIER genIUType bitfieldExtract(genIUType Value, int Offset, int Bits)
{
return bitfieldExtract(tvec1<genIUType>(Value), Offset, Bits).x;
}
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_QUALIFIER vecType<T, P> bitfieldExtract(vecType<T, P> const & Value, int Offset, int Bits)
{
GLM_STATIC_ASSERT(std::numeric_limits<T>::is_integer, "'bitfieldExtract' only accept integer inputs");
return (Value >> static_cast<T>(Offset)) & static_cast<T>(detail::mask(Bits));
}
// bitfieldInsert
template <typename genIUType>
GLM_FUNC_QUALIFIER genIUType bitfieldInsert(genIUType const & Base, genIUType const & Insert, int Offset, int Bits)
{
return bitfieldInsert(tvec1<genIUType>(Base), tvec1<genIUType>(Insert), Offset, Bits).x;
}
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_QUALIFIER vecType<T, P> bitfieldInsert(vecType<T, P> const & Base, vecType<T, P> const & Insert, int Offset, int Bits)
{
GLM_STATIC_ASSERT(std::numeric_limits<T>::is_integer, "'bitfieldInsert' only accept integer values");
T const Mask = static_cast<T>(detail::mask(Bits) << Offset);
return (Base & ~Mask) | (Insert & Mask);
}
// bitfieldReverse
template <typename genType>
GLM_FUNC_QUALIFIER genType bitfieldReverse(genType x)
{
return bitfieldReverse(glm::tvec1<genType, glm::defaultp>(x)).x;
}
template <typename T, glm::precision P, template <typename, glm::precision> class vecType>
GLM_FUNC_QUALIFIER vecType<T, P> bitfieldReverse(vecType<T, P> const & v)
{
vecType<T, P> x(v);
x = detail::compute_bitfieldReverseStep<T, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 2>::call(x, T(0x5555555555555555ull), static_cast<T>( 1));
x = detail::compute_bitfieldReverseStep<T, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 4>::call(x, T(0x3333333333333333ull), static_cast<T>( 2));
x = detail::compute_bitfieldReverseStep<T, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 8>::call(x, T(0x0F0F0F0F0F0F0F0Full), static_cast<T>( 4));
x = detail::compute_bitfieldReverseStep<T, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 16>::call(x, T(0x00FF00FF00FF00FFull), static_cast<T>( 8));
x = detail::compute_bitfieldReverseStep<T, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 32>::call(x, T(0x0000FFFF0000FFFFull), static_cast<T>(16));
x = detail::compute_bitfieldReverseStep<T, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 64>::call(x, T(0x00000000FFFFFFFFull), static_cast<T>(32));
return x;
}
// bitCount
template <typename genType>
GLM_FUNC_QUALIFIER int bitCount(genType x)
{
return bitCount(glm::tvec1<genType, glm::defaultp>(x)).x;
}
template <typename T, glm::precision P, template <typename, glm::precision> class vecType>
GLM_FUNC_QUALIFIER vecType<int, P> bitCount(vecType<T, P> const & v)
{
vecType<typename detail::make_unsigned<T>::type, P> x(*reinterpret_cast<vecType<typename detail::make_unsigned<T>::type, P> const *>(&v));
x = detail::compute_bitfieldBitCountStep<typename detail::make_unsigned<T>::type, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 2>::call(x, typename detail::make_unsigned<T>::type(0x5555555555555555ull), typename detail::make_unsigned<T>::type( 1));
x = detail::compute_bitfieldBitCountStep<typename detail::make_unsigned<T>::type, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 4>::call(x, typename detail::make_unsigned<T>::type(0x3333333333333333ull), typename detail::make_unsigned<T>::type( 2));
x = detail::compute_bitfieldBitCountStep<typename detail::make_unsigned<T>::type, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 8>::call(x, typename detail::make_unsigned<T>::type(0x0F0F0F0F0F0F0F0Full), typename detail::make_unsigned<T>::type( 4));
x = detail::compute_bitfieldBitCountStep<typename detail::make_unsigned<T>::type, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 16>::call(x, typename detail::make_unsigned<T>::type(0x00FF00FF00FF00FFull), typename detail::make_unsigned<T>::type( 8));
x = detail::compute_bitfieldBitCountStep<typename detail::make_unsigned<T>::type, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 32>::call(x, typename detail::make_unsigned<T>::type(0x0000FFFF0000FFFFull), typename detail::make_unsigned<T>::type(16));
x = detail::compute_bitfieldBitCountStep<typename detail::make_unsigned<T>::type, P, vecType, detail::is_aligned<P>::value, sizeof(T) * 8>= 64>::call(x, typename detail::make_unsigned<T>::type(0x00000000FFFFFFFFull), typename detail::make_unsigned<T>::type(32));
return vecType<int, P>(x);
}
// findLSB
template <typename genIUType>
GLM_FUNC_QUALIFIER int findLSB(genIUType Value)
{
GLM_STATIC_ASSERT(std::numeric_limits<genIUType>::is_integer, "'findLSB' only accept integer values");
return detail::compute_findLSB<genIUType, sizeof(genIUType) * 8>::call(Value);
}
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_QUALIFIER vecType<int, P> findLSB(vecType<T, P> const & x)
{
GLM_STATIC_ASSERT(std::numeric_limits<T>::is_integer, "'findLSB' only accept integer values");
return detail::functor1<int, T, P, vecType>::call(findLSB, x);
}
// findMSB
template <typename genIUType>
GLM_FUNC_QUALIFIER int findMSB(genIUType x)
{
GLM_STATIC_ASSERT(std::numeric_limits<genIUType>::is_integer, "'findMSB' only accept integer values");
return findMSB(tvec1<genIUType>(x)).x;
}
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_QUALIFIER vecType<int, P> findMSB(vecType<T, P> const & x)
{
GLM_STATIC_ASSERT(std::numeric_limits<T>::is_integer, "'findMSB' only accept integer values");
return detail::compute_findMSB_vec<T, P, vecType, sizeof(T) * 8>::call(x);
}
}//namespace glm
#if GLM_ARCH != GLM_ARCH_PURE && GLM_HAS_UNRESTRICTED_UNIONS
# include "func_integer_simd.inl"
#endif