Counter Strike : Global Offensive Source Code
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

2567 lines
73 KiB

/*++
Copyright (c) Microsoft Corporation. All rights reserved.
Module Name:
xnamathmisc.inl
Abstract:
XNA math library for Windows and Xbox 360: Quaternion, plane, and color functions.
--*/
#if defined(_MSC_VER) && (_MSC_VER > 1000)
#pragma once
#endif
#ifndef __XNAMATHMISC_INL__
#define __XNAMATHMISC_INL__
/****************************************************************************
*
* Quaternion
*
****************************************************************************/
//------------------------------------------------------------------------------
// Comparison operations
//------------------------------------------------------------------------------
//------------------------------------------------------------------------------
XMFINLINE BOOL XMQuaternionEqual
(
FXMVECTOR Q1,
FXMVECTOR Q2
)
{
return XMVector4Equal(Q1, Q2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMQuaternionNotEqual
(
FXMVECTOR Q1,
FXMVECTOR Q2
)
{
return XMVector4NotEqual(Q1, Q2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMQuaternionIsNaN
(
FXMVECTOR Q
)
{
return XMVector4IsNaN(Q);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMQuaternionIsInfinite
(
FXMVECTOR Q
)
{
return XMVector4IsInfinite(Q);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMQuaternionIsIdentity
(
FXMVECTOR Q
)
{
#if defined(_XM_NO_INTRINSICS_)
return XMVector4Equal(Q, g_XMIdentityR3.v);
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR vTemp = _mm_cmpeq_ps(Q,g_XMIdentityR3);
return (_mm_movemask_ps(vTemp)==0x0f) ? true : false;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
// Computation operations
//------------------------------------------------------------------------------
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionDot
(
FXMVECTOR Q1,
FXMVECTOR Q2
)
{
return XMVector4Dot(Q1, Q2);
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionMultiply
(
FXMVECTOR Q1,
FXMVECTOR Q2
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR NegativeQ1;
XMVECTOR Q2X;
XMVECTOR Q2Y;
XMVECTOR Q2Z;
XMVECTOR Q2W;
XMVECTOR Q1WZYX;
XMVECTOR Q1ZWXY;
XMVECTOR Q1YXWZ;
XMVECTOR Result;
CONST XMVECTORU32 ControlWZYX = {XM_PERMUTE_0W, XM_PERMUTE_1Z, XM_PERMUTE_0Y, XM_PERMUTE_1X};
CONST XMVECTORU32 ControlZWXY = {XM_PERMUTE_0Z, XM_PERMUTE_0W, XM_PERMUTE_1X, XM_PERMUTE_1Y};
CONST XMVECTORU32 ControlYXWZ = {XM_PERMUTE_1Y, XM_PERMUTE_0X, XM_PERMUTE_0W, XM_PERMUTE_1Z};
NegativeQ1 = XMVectorNegate(Q1);
Q2W = XMVectorSplatW(Q2);
Q2X = XMVectorSplatX(Q2);
Q2Y = XMVectorSplatY(Q2);
Q2Z = XMVectorSplatZ(Q2);
Q1WZYX = XMVectorPermute(Q1, NegativeQ1, ControlWZYX.v);
Q1ZWXY = XMVectorPermute(Q1, NegativeQ1, ControlZWXY.v);
Q1YXWZ = XMVectorPermute(Q1, NegativeQ1, ControlYXWZ.v);
Result = XMVectorMultiply(Q1, Q2W);
Result = XMVectorMultiplyAdd(Q1WZYX, Q2X, Result);
Result = XMVectorMultiplyAdd(Q1ZWXY, Q2Y, Result);
Result = XMVectorMultiplyAdd(Q1YXWZ, Q2Z, Result);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
static CONST XMVECTORF32 g_ControlWZYX = { 1.0f,-1.0f, 1.0f,-1.0f};
static CONST XMVECTORF32 g_ControlZWXY = { 1.0f, 1.0f,-1.0f,-1.0f};
static CONST XMVECTORF32 g_ControlYXWZ = {-1.0f, 1.0f, 1.0f,-1.0f};
// Copy to SSE registers and use as few as possible for x86
XMVECTOR Q2X = Q2;
XMVECTOR Q2Y = Q2;
XMVECTOR Q2Z = Q2;
XMVECTOR vResult = Q2;
// Splat with one instruction
vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(3,3,3,3));
Q2X = _mm_shuffle_ps(Q2X,Q2X,_MM_SHUFFLE(0,0,0,0));
Q2Y = _mm_shuffle_ps(Q2Y,Q2Y,_MM_SHUFFLE(1,1,1,1));
Q2Z = _mm_shuffle_ps(Q2Z,Q2Z,_MM_SHUFFLE(2,2,2,2));
// Retire Q1 and perform Q1*Q2W
vResult = _mm_mul_ps(vResult,Q1);
XMVECTOR Q1Shuffle = Q1;
// Shuffle the copies of Q1
Q1Shuffle = _mm_shuffle_ps(Q1Shuffle,Q1Shuffle,_MM_SHUFFLE(0,1,2,3));
// Mul by Q1WZYX
Q2X = _mm_mul_ps(Q2X,Q1Shuffle);
Q1Shuffle = _mm_shuffle_ps(Q1Shuffle,Q1Shuffle,_MM_SHUFFLE(2,3,0,1));
// Flip the signs on y and z
Q2X = _mm_mul_ps(Q2X,g_ControlWZYX);
// Mul by Q1ZWXY
Q2Y = _mm_mul_ps(Q2Y,Q1Shuffle);
Q1Shuffle = _mm_shuffle_ps(Q1Shuffle,Q1Shuffle,_MM_SHUFFLE(0,1,2,3));
// Flip the signs on z and w
Q2Y = _mm_mul_ps(Q2Y,g_ControlZWXY);
// Mul by Q1YXWZ
Q2Z = _mm_mul_ps(Q2Z,Q1Shuffle);
vResult = _mm_add_ps(vResult,Q2X);
// Flip the signs on x and w
Q2Z = _mm_mul_ps(Q2Z,g_ControlYXWZ);
Q2Y = _mm_add_ps(Q2Y,Q2Z);
vResult = _mm_add_ps(vResult,Q2Y);
return vResult;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionLengthSq
(
FXMVECTOR Q
)
{
return XMVector4LengthSq(Q);
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionReciprocalLength
(
FXMVECTOR Q
)
{
return XMVector4ReciprocalLength(Q);
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionLength
(
FXMVECTOR Q
)
{
return XMVector4Length(Q);
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionNormalizeEst
(
FXMVECTOR Q
)
{
return XMVector4NormalizeEst(Q);
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionNormalize
(
FXMVECTOR Q
)
{
return XMVector4Normalize(Q);
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionConjugate
(
FXMVECTOR Q
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR Result = {
-Q.x,
-Q.y,
-Q.z,
Q.w
};
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
static const XMVECTORF32 g_XMNegativeOne3 = {-1.0f,-1.0f,-1.0f,1.0f};
XMVECTOR Result = _mm_mul_ps(Q,g_XMNegativeOne3);
return Result;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionInverse
(
FXMVECTOR Q
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR Conjugate;
XMVECTOR L;
XMVECTOR Control;
XMVECTOR Result;
CONST XMVECTOR Zero = XMVectorZero();
L = XMVector4LengthSq(Q);
Conjugate = XMQuaternionConjugate(Q);
Control = XMVectorLessOrEqual(L, g_XMEpsilon.v);
L = XMVectorReciprocal(L);
Result = XMVectorMultiply(Conjugate, L);
Result = XMVectorSelect(Result, Zero, Control);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR Conjugate;
XMVECTOR L;
XMVECTOR Control;
XMVECTOR Result;
XMVECTOR Zero = XMVectorZero();
L = XMVector4LengthSq(Q);
Conjugate = XMQuaternionConjugate(Q);
Control = XMVectorLessOrEqual(L, g_XMEpsilon);
Result = _mm_div_ps(Conjugate,L);
Result = XMVectorSelect(Result, Zero, Control);
return Result;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionLn
(
FXMVECTOR Q
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR Q0;
XMVECTOR QW;
XMVECTOR Theta;
XMVECTOR SinTheta;
XMVECTOR S;
XMVECTOR ControlW;
XMVECTOR Result;
static CONST XMVECTOR OneMinusEpsilon = {1.0f - 0.00001f, 1.0f - 0.00001f, 1.0f - 0.00001f, 1.0f - 0.00001f};
QW = XMVectorSplatW(Q);
Q0 = XMVectorSelect(g_XMSelect1110.v, Q, g_XMSelect1110.v);
ControlW = XMVectorInBounds(QW, OneMinusEpsilon);
Theta = XMVectorACos(QW);
SinTheta = XMVectorSin(Theta);
S = XMVectorReciprocal(SinTheta);
S = XMVectorMultiply(Theta, S);
Result = XMVectorMultiply(Q0, S);
Result = XMVectorSelect(Q0, Result, ControlW);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
static CONST XMVECTORF32 OneMinusEpsilon = {1.0f - 0.00001f, 1.0f - 0.00001f, 1.0f - 0.00001f, 1.0f - 0.00001f};
static CONST XMVECTORF32 NegOneMinusEpsilon = {-(1.0f - 0.00001f), -(1.0f - 0.00001f),-(1.0f - 0.00001f),-(1.0f - 0.00001f)};
// Get W only
XMVECTOR QW = _mm_shuffle_ps(Q,Q,_MM_SHUFFLE(3,3,3,3));
// W = 0
XMVECTOR Q0 = _mm_and_ps(Q,g_XMMask3);
// Use W if within bounds
XMVECTOR ControlW = _mm_cmple_ps(QW,OneMinusEpsilon);
XMVECTOR vTemp2 = _mm_cmpge_ps(QW,NegOneMinusEpsilon);
ControlW = _mm_and_ps(ControlW,vTemp2);
// Get theta
XMVECTOR vTheta = XMVectorACos(QW);
// Get Sine of theta
vTemp2 = XMVectorSin(vTheta);
// theta/sine of theta
vTheta = _mm_div_ps(vTheta,vTemp2);
// Here's the answer
vTheta = _mm_mul_ps(vTheta,Q0);
// Was W in bounds? If not, return input as is
vTheta = XMVectorSelect(Q0,vTheta,ControlW);
return vTheta;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionExp
(
FXMVECTOR Q
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR Theta;
XMVECTOR SinTheta;
XMVECTOR CosTheta;
XMVECTOR S;
XMVECTOR Control;
XMVECTOR Zero;
XMVECTOR Result;
Theta = XMVector3Length(Q);
XMVectorSinCos(&SinTheta, &CosTheta, Theta);
S = XMVectorReciprocal(Theta);
S = XMVectorMultiply(SinTheta, S);
Result = XMVectorMultiply(Q, S);
Zero = XMVectorZero();
Control = XMVectorNearEqual(Theta, Zero, g_XMEpsilon.v);
Result = XMVectorSelect(Result, Q, Control);
Result = XMVectorSelect(CosTheta, Result, g_XMSelect1110.v);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR Theta;
XMVECTOR SinTheta;
XMVECTOR CosTheta;
XMVECTOR S;
XMVECTOR Control;
XMVECTOR Zero;
XMVECTOR Result;
Theta = XMVector3Length(Q);
XMVectorSinCos(&SinTheta, &CosTheta, Theta);
S = _mm_div_ps(SinTheta,Theta);
Result = _mm_mul_ps(Q, S);
Zero = XMVectorZero();
Control = XMVectorNearEqual(Theta, Zero, g_XMEpsilon);
Result = XMVectorSelect(Result,Q,Control);
Result = _mm_and_ps(Result,g_XMMask3);
CosTheta = _mm_and_ps(CosTheta,g_XMMaskW);
Result = _mm_or_ps(Result,CosTheta);
return Result;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMINLINE XMVECTOR XMQuaternionSlerp
(
FXMVECTOR Q0,
FXMVECTOR Q1,
FLOAT t
)
{
XMVECTOR T = XMVectorReplicate(t);
return XMQuaternionSlerpV(Q0, Q1, T);
}
//------------------------------------------------------------------------------
XMINLINE XMVECTOR XMQuaternionSlerpV
(
FXMVECTOR Q0,
FXMVECTOR Q1,
FXMVECTOR T
)
{
#if defined(_XM_NO_INTRINSICS_)
// Result = Q0 * sin((1.0 - t) * Omega) / sin(Omega) + Q1 * sin(t * Omega) / sin(Omega)
XMVECTOR Omega;
XMVECTOR CosOmega;
XMVECTOR SinOmega;
XMVECTOR InvSinOmega;
XMVECTOR V01;
XMVECTOR C1000;
XMVECTOR SignMask;
XMVECTOR S0;
XMVECTOR S1;
XMVECTOR Sign;
XMVECTOR Control;
XMVECTOR Result;
XMVECTOR Zero;
CONST XMVECTOR OneMinusEpsilon = {1.0f - 0.00001f, 1.0f - 0.00001f, 1.0f - 0.00001f, 1.0f - 0.00001f};
XMASSERT((T.v[1] == T.v[0]) && (T.v[2] == T.v[0]) && (T.v[3] == T.v[0]));
CosOmega = XMQuaternionDot(Q0, Q1);
Zero = XMVectorZero();
Control = XMVectorLess(CosOmega, Zero);
Sign = XMVectorSelect(g_XMOne.v, g_XMNegativeOne.v, Control);
CosOmega = XMVectorMultiply(CosOmega, Sign);
Control = XMVectorLess(CosOmega, OneMinusEpsilon);
SinOmega = XMVectorNegativeMultiplySubtract(CosOmega, CosOmega, g_XMOne.v);
SinOmega = XMVectorSqrt(SinOmega);
Omega = XMVectorATan2(SinOmega, CosOmega);
SignMask = XMVectorSplatSignMask();
C1000 = XMVectorSetBinaryConstant(1, 0, 0, 0);
V01 = XMVectorShiftLeft(T, Zero, 2);
SignMask = XMVectorShiftLeft(SignMask, Zero, 3);
V01 = XMVectorXorInt(V01, SignMask);
V01 = XMVectorAdd(C1000, V01);
InvSinOmega = XMVectorReciprocal(SinOmega);
S0 = XMVectorMultiply(V01, Omega);
S0 = XMVectorSin(S0);
S0 = XMVectorMultiply(S0, InvSinOmega);
S0 = XMVectorSelect(V01, S0, Control);
S1 = XMVectorSplatY(S0);
S0 = XMVectorSplatX(S0);
S1 = XMVectorMultiply(S1, Sign);
Result = XMVectorMultiply(Q0, S0);
Result = XMVectorMultiplyAdd(Q1, S1, Result);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
// Result = Q0 * sin((1.0 - t) * Omega) / sin(Omega) + Q1 * sin(t * Omega) / sin(Omega)
XMVECTOR Omega;
XMVECTOR CosOmega;
XMVECTOR SinOmega;
XMVECTOR V01;
XMVECTOR S0;
XMVECTOR S1;
XMVECTOR Sign;
XMVECTOR Control;
XMVECTOR Result;
XMVECTOR Zero;
static const XMVECTORF32 OneMinusEpsilon = {1.0f - 0.00001f, 1.0f - 0.00001f, 1.0f - 0.00001f, 1.0f - 0.00001f};
static const XMVECTORI32 g_XMSignMask2 = {0x80000000,0x00000000,0x00000000,0x00000000};
static const XMVECTORI32 g_XMMaskXY = {0xFFFFFFFF,0xFFFFFFFF,0x00000000,0x00000000};
XMASSERT((XMVectorGetY(T) == XMVectorGetX(T)) && (XMVectorGetZ(T) == XMVectorGetX(T)) && (XMVectorGetW(T) == XMVectorGetX(T)));
CosOmega = XMQuaternionDot(Q0, Q1);
Zero = XMVectorZero();
Control = XMVectorLess(CosOmega, Zero);
Sign = XMVectorSelect(g_XMOne, g_XMNegativeOne, Control);
CosOmega = _mm_mul_ps(CosOmega, Sign);
Control = XMVectorLess(CosOmega, OneMinusEpsilon);
SinOmega = _mm_mul_ps(CosOmega,CosOmega);
SinOmega = _mm_sub_ps(g_XMOne,SinOmega);
SinOmega = _mm_sqrt_ps(SinOmega);
Omega = XMVectorATan2(SinOmega, CosOmega);
V01 = _mm_shuffle_ps(T,T,_MM_SHUFFLE(2,3,0,1));
V01 = _mm_and_ps(V01,g_XMMaskXY);
V01 = _mm_xor_ps(V01,g_XMSignMask2);
V01 = _mm_add_ps(g_XMIdentityR0, V01);
S0 = _mm_mul_ps(V01, Omega);
S0 = XMVectorSin(S0);
S0 = _mm_div_ps(S0, SinOmega);
S0 = XMVectorSelect(V01, S0, Control);
S1 = XMVectorSplatY(S0);
S0 = XMVectorSplatX(S0);
S1 = _mm_mul_ps(S1, Sign);
Result = _mm_mul_ps(Q0, S0);
S1 = _mm_mul_ps(S1, Q1);
Result = _mm_add_ps(Result,S1);
return Result;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionSquad
(
FXMVECTOR Q0,
FXMVECTOR Q1,
FXMVECTOR Q2,
CXMVECTOR Q3,
FLOAT t
)
{
XMVECTOR T = XMVectorReplicate(t);
return XMQuaternionSquadV(Q0, Q1, Q2, Q3, T);
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionSquadV
(
FXMVECTOR Q0,
FXMVECTOR Q1,
FXMVECTOR Q2,
CXMVECTOR Q3,
CXMVECTOR T
)
{
XMVECTOR Q03;
XMVECTOR Q12;
XMVECTOR TP;
XMVECTOR Two;
XMVECTOR Result;
XMASSERT( (XMVectorGetY(T) == XMVectorGetX(T)) && (XMVectorGetZ(T) == XMVectorGetX(T)) && (XMVectorGetW(T) == XMVectorGetX(T)) );
TP = T;
Two = XMVectorSplatConstant(2, 0);
Q03 = XMQuaternionSlerpV(Q0, Q3, T);
Q12 = XMQuaternionSlerpV(Q1, Q2, T);
TP = XMVectorNegativeMultiplySubtract(TP, TP, TP);
TP = XMVectorMultiply(TP, Two);
Result = XMQuaternionSlerpV(Q03, Q12, TP);
return Result;
}
//------------------------------------------------------------------------------
XMINLINE VOID XMQuaternionSquadSetup
(
XMVECTOR* pA,
XMVECTOR* pB,
XMVECTOR* pC,
FXMVECTOR Q0,
FXMVECTOR Q1,
FXMVECTOR Q2,
CXMVECTOR Q3
)
{
XMVECTOR SQ0, SQ2, SQ3;
XMVECTOR InvQ1, InvQ2;
XMVECTOR LnQ0, LnQ1, LnQ2, LnQ3;
XMVECTOR ExpQ02, ExpQ13;
XMVECTOR LS01, LS12, LS23;
XMVECTOR LD01, LD12, LD23;
XMVECTOR Control0, Control1, Control2;
XMVECTOR NegativeOneQuarter;
XMASSERT(pA);
XMASSERT(pB);
XMASSERT(pC);
LS12 = XMQuaternionLengthSq(XMVectorAdd(Q1, Q2));
LD12 = XMQuaternionLengthSq(XMVectorSubtract(Q1, Q2));
SQ2 = XMVectorNegate(Q2);
Control1 = XMVectorLess(LS12, LD12);
SQ2 = XMVectorSelect(Q2, SQ2, Control1);
LS01 = XMQuaternionLengthSq(XMVectorAdd(Q0, Q1));
LD01 = XMQuaternionLengthSq(XMVectorSubtract(Q0, Q1));
SQ0 = XMVectorNegate(Q0);
LS23 = XMQuaternionLengthSq(XMVectorAdd(SQ2, Q3));
LD23 = XMQuaternionLengthSq(XMVectorSubtract(SQ2, Q3));
SQ3 = XMVectorNegate(Q3);
Control0 = XMVectorLess(LS01, LD01);
Control2 = XMVectorLess(LS23, LD23);
SQ0 = XMVectorSelect(Q0, SQ0, Control0);
SQ3 = XMVectorSelect(Q3, SQ3, Control2);
InvQ1 = XMQuaternionInverse(Q1);
InvQ2 = XMQuaternionInverse(SQ2);
LnQ0 = XMQuaternionLn(XMQuaternionMultiply(InvQ1, SQ0));
LnQ2 = XMQuaternionLn(XMQuaternionMultiply(InvQ1, SQ2));
LnQ1 = XMQuaternionLn(XMQuaternionMultiply(InvQ2, Q1));
LnQ3 = XMQuaternionLn(XMQuaternionMultiply(InvQ2, SQ3));
NegativeOneQuarter = XMVectorSplatConstant(-1, 2);
ExpQ02 = XMVectorMultiply(XMVectorAdd(LnQ0, LnQ2), NegativeOneQuarter);
ExpQ13 = XMVectorMultiply(XMVectorAdd(LnQ1, LnQ3), NegativeOneQuarter);
ExpQ02 = XMQuaternionExp(ExpQ02);
ExpQ13 = XMQuaternionExp(ExpQ13);
*pA = XMQuaternionMultiply(Q1, ExpQ02);
*pB = XMQuaternionMultiply(SQ2, ExpQ13);
*pC = SQ2;
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionBaryCentric
(
FXMVECTOR Q0,
FXMVECTOR Q1,
FXMVECTOR Q2,
FLOAT f,
FLOAT g
)
{
XMVECTOR Q01;
XMVECTOR Q02;
FLOAT s;
XMVECTOR Result;
s = f + g;
if (s < 0.00001f && s > -0.00001f)
{
Result = Q0;
}
else
{
Q01 = XMQuaternionSlerp(Q0, Q1, s);
Q02 = XMQuaternionSlerp(Q0, Q2, s);
Result = XMQuaternionSlerp(Q01, Q02, g / s);
}
return Result;
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionBaryCentricV
(
FXMVECTOR Q0,
FXMVECTOR Q1,
FXMVECTOR Q2,
CXMVECTOR F,
CXMVECTOR G
)
{
XMVECTOR Q01;
XMVECTOR Q02;
XMVECTOR S, GS;
XMVECTOR Epsilon;
XMVECTOR Result;
XMASSERT( (XMVectorGetY(F) == XMVectorGetX(F)) && (XMVectorGetZ(F) == XMVectorGetX(F)) && (XMVectorGetW(F) == XMVectorGetX(F)) );
XMASSERT( (XMVectorGetY(G) == XMVectorGetX(G)) && (XMVectorGetZ(G) == XMVectorGetX(G)) && (XMVectorGetW(G) == XMVectorGetX(G)) );
Epsilon = XMVectorSplatConstant(1, 16);
S = XMVectorAdd(F, G);
if (XMVector4InBounds(S, Epsilon))
{
Result = Q0;
}
else
{
Q01 = XMQuaternionSlerpV(Q0, Q1, S);
Q02 = XMQuaternionSlerpV(Q0, Q2, S);
GS = XMVectorReciprocal(S);
GS = XMVectorMultiply(G, GS);
Result = XMQuaternionSlerpV(Q01, Q02, GS);
}
return Result;
}
//------------------------------------------------------------------------------
// Transformation operations
//------------------------------------------------------------------------------
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionIdentity()
{
#if defined(_XM_NO_INTRINSICS_)
return g_XMIdentityR3.v;
#elif defined(_XM_SSE_INTRINSICS_)
return g_XMIdentityR3;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionRotationRollPitchYaw
(
FLOAT Pitch,
FLOAT Yaw,
FLOAT Roll
)
{
XMVECTOR Angles;
XMVECTOR Q;
Angles = XMVectorSet(Pitch, Yaw, Roll, 0.0f);
Q = XMQuaternionRotationRollPitchYawFromVector(Angles);
return Q;
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionRotationRollPitchYawFromVector
(
FXMVECTOR Angles // <Pitch, Yaw, Roll, 0>
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR Q, Q0, Q1;
XMVECTOR P0, P1, Y0, Y1, R0, R1;
XMVECTOR HalfAngles;
XMVECTOR SinAngles, CosAngles;
static CONST XMVECTORU32 ControlPitch = {XM_PERMUTE_0X, XM_PERMUTE_1X, XM_PERMUTE_1X, XM_PERMUTE_1X};
static CONST XMVECTORU32 ControlYaw = {XM_PERMUTE_1Y, XM_PERMUTE_0Y, XM_PERMUTE_1Y, XM_PERMUTE_1Y};
static CONST XMVECTORU32 ControlRoll = {XM_PERMUTE_1Z, XM_PERMUTE_1Z, XM_PERMUTE_0Z, XM_PERMUTE_1Z};
static CONST XMVECTOR Sign = {1.0f, -1.0f, -1.0f, 1.0f};
HalfAngles = XMVectorMultiply(Angles, g_XMOneHalf.v);
XMVectorSinCos(&SinAngles, &CosAngles, HalfAngles);
P0 = XMVectorPermute(SinAngles, CosAngles, ControlPitch.v);
Y0 = XMVectorPermute(SinAngles, CosAngles, ControlYaw.v);
R0 = XMVectorPermute(SinAngles, CosAngles, ControlRoll.v);
P1 = XMVectorPermute(CosAngles, SinAngles, ControlPitch.v);
Y1 = XMVectorPermute(CosAngles, SinAngles, ControlYaw.v);
R1 = XMVectorPermute(CosAngles, SinAngles, ControlRoll.v);
Q1 = XMVectorMultiply(P1, Sign);
Q0 = XMVectorMultiply(P0, Y0);
Q1 = XMVectorMultiply(Q1, Y1);
Q0 = XMVectorMultiply(Q0, R0);
Q = XMVectorMultiplyAdd(Q1, R1, Q0);
return Q;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR Q, Q0, Q1;
XMVECTOR P0, P1, Y0, Y1, R0, R1;
XMVECTOR HalfAngles;
XMVECTOR SinAngles, CosAngles;
static CONST XMVECTORI32 ControlPitch = {XM_PERMUTE_0X, XM_PERMUTE_1X, XM_PERMUTE_1X, XM_PERMUTE_1X};
static CONST XMVECTORI32 ControlYaw = {XM_PERMUTE_1Y, XM_PERMUTE_0Y, XM_PERMUTE_1Y, XM_PERMUTE_1Y};
static CONST XMVECTORI32 ControlRoll = {XM_PERMUTE_1Z, XM_PERMUTE_1Z, XM_PERMUTE_0Z, XM_PERMUTE_1Z};
static CONST XMVECTORF32 g_XMSign = {1.0f, -1.0f, -1.0f, 1.0f};
HalfAngles = _mm_mul_ps(Angles, g_XMOneHalf);
XMVectorSinCos(&SinAngles, &CosAngles, HalfAngles);
P0 = XMVectorPermute(SinAngles, CosAngles, ControlPitch);
Y0 = XMVectorPermute(SinAngles, CosAngles, ControlYaw);
R0 = XMVectorPermute(SinAngles, CosAngles, ControlRoll);
P1 = XMVectorPermute(CosAngles, SinAngles, ControlPitch);
Y1 = XMVectorPermute(CosAngles, SinAngles, ControlYaw);
R1 = XMVectorPermute(CosAngles, SinAngles, ControlRoll);
Q1 = _mm_mul_ps(P1, g_XMSign);
Q0 = _mm_mul_ps(P0, Y0);
Q1 = _mm_mul_ps(Q1, Y1);
Q0 = _mm_mul_ps(Q0, R0);
Q = _mm_mul_ps(Q1, R1);
Q = _mm_add_ps(Q,Q0);
return Q;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionRotationNormal
(
FXMVECTOR NormalAxis,
FLOAT Angle
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR Q;
XMVECTOR N;
XMVECTOR Scale;
N = XMVectorSelect(g_XMOne.v, NormalAxis, g_XMSelect1110.v);
XMScalarSinCos(&Scale.v[2], &Scale.v[3], 0.5f * Angle);
Scale.v[0] = Scale.v[1] = Scale.v[2];
Q = XMVectorMultiply(N, Scale);
return Q;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR N = _mm_and_ps(NormalAxis,g_XMMask3);
N = _mm_or_ps(N,g_XMIdentityR3);
XMVECTOR Scale = _mm_set_ps1(0.5f * Angle);
XMVECTOR vSine;
XMVECTOR vCosine;
XMVectorSinCos(&vSine,&vCosine,Scale);
Scale = _mm_and_ps(vSine,g_XMMask3);
vCosine = _mm_and_ps(vCosine,g_XMMaskW);
Scale = _mm_or_ps(Scale,vCosine);
N = _mm_mul_ps(N,Scale);
return N;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMQuaternionRotationAxis
(
FXMVECTOR Axis,
FLOAT Angle
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR Normal;
XMVECTOR Q;
XMASSERT(!XMVector3Equal(Axis, XMVectorZero()));
XMASSERT(!XMVector3IsInfinite(Axis));
Normal = XMVector3Normalize(Axis);
Q = XMQuaternionRotationNormal(Normal, Angle);
return Q;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR Normal;
XMVECTOR Q;
XMASSERT(!XMVector3Equal(Axis, XMVectorZero()));
XMASSERT(!XMVector3IsInfinite(Axis));
Normal = XMVector3Normalize(Axis);
Q = XMQuaternionRotationNormal(Normal, Angle);
return Q;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMINLINE XMVECTOR XMQuaternionRotationMatrix
(
CXMMATRIX M
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR Q0, Q1, Q2;
XMVECTOR M00, M11, M22;
XMVECTOR CQ0, CQ1, C;
XMVECTOR CX, CY, CZ, CW;
XMVECTOR SQ1, Scale;
XMVECTOR Rsq, Sqrt, VEqualsInfinity, VEqualsZero, Select;
XMVECTOR A, B, P;
XMVECTOR PermuteSplat, PermuteSplatT;
XMVECTOR SignB, SignBT;
XMVECTOR PermuteControl, PermuteControlT;
XMVECTOR Zero;
XMVECTOR Result;
static CONST XMVECTOR OneQuarter = {0.25f, 0.25f, 0.25f, 0.25f};
static CONST XMVECTOR SignPNNP = {1.0f, -1.0f, -1.0f, 1.0f};
static CONST XMVECTOR SignNPNP = {-1.0f, 1.0f, -1.0f, 1.0f};
static CONST XMVECTOR SignNNPP = {-1.0f, -1.0f, 1.0f, 1.0f};
static CONST XMVECTOR SignPNPP = {1.0f, -1.0f, 1.0f, 1.0f};
static CONST XMVECTOR SignPPNP = {1.0f, 1.0f, -1.0f, 1.0f};
static CONST XMVECTOR SignNPPP = {-1.0f, 1.0f, 1.0f, 1.0f};
static CONST XMVECTOR SignNNNX = {-1.0f, -1.0f, -1.0f, 2.0e-126f};
static CONST XMVECTORU32 Permute0X0X0Y0W = {XM_PERMUTE_0X, XM_PERMUTE_0X, XM_PERMUTE_0Y, XM_PERMUTE_0W};
static CONST XMVECTORU32 Permute0Y0Z0Z1W = {XM_PERMUTE_0Y, XM_PERMUTE_0Z, XM_PERMUTE_0Z, XM_PERMUTE_1W};
static CONST XMVECTORU32 SplatX = {XM_PERMUTE_0X, XM_PERMUTE_0X, XM_PERMUTE_0X, XM_PERMUTE_0X};
static CONST XMVECTORU32 SplatY = {XM_PERMUTE_0Y, XM_PERMUTE_0Y, XM_PERMUTE_0Y, XM_PERMUTE_0Y};
static CONST XMVECTORU32 SplatZ = {XM_PERMUTE_0Z, XM_PERMUTE_0Z, XM_PERMUTE_0Z, XM_PERMUTE_0Z};
static CONST XMVECTORU32 SplatW = {XM_PERMUTE_0W, XM_PERMUTE_0W, XM_PERMUTE_0W, XM_PERMUTE_0W};
static CONST XMVECTORU32 PermuteC = {XM_PERMUTE_0X, XM_PERMUTE_0Z, XM_PERMUTE_1X, XM_PERMUTE_1Y};
static CONST XMVECTORU32 PermuteA = {XM_PERMUTE_0Y, XM_PERMUTE_1Y, XM_PERMUTE_1Z, XM_PERMUTE_0W};
static CONST XMVECTORU32 PermuteB = {XM_PERMUTE_1X, XM_PERMUTE_1W, XM_PERMUTE_0Z, XM_PERMUTE_0W};
static CONST XMVECTORU32 Permute0 = {XM_PERMUTE_0X, XM_PERMUTE_1X, XM_PERMUTE_1Z, XM_PERMUTE_1Y};
static CONST XMVECTORU32 Permute1 = {XM_PERMUTE_1X, XM_PERMUTE_0Y, XM_PERMUTE_1Y, XM_PERMUTE_1Z};
static CONST XMVECTORU32 Permute2 = {XM_PERMUTE_1Z, XM_PERMUTE_1Y, XM_PERMUTE_0Z, XM_PERMUTE_1X};
static CONST XMVECTORU32 Permute3 = {XM_PERMUTE_1Y, XM_PERMUTE_1Z, XM_PERMUTE_1X, XM_PERMUTE_0W};
M00 = XMVectorSplatX(M.r[0]);
M11 = XMVectorSplatY(M.r[1]);
M22 = XMVectorSplatZ(M.r[2]);
Q0 = XMVectorMultiply(SignPNNP, M00);
Q0 = XMVectorMultiplyAdd(SignNPNP, M11, Q0);
Q0 = XMVectorMultiplyAdd(SignNNPP, M22, Q0);
Q1 = XMVectorAdd(Q0, g_XMOne.v);
Rsq = XMVectorReciprocalSqrt(Q1);
Zero = XMVectorZero();
VEqualsInfinity = XMVectorEqualInt(Q1, g_XMInfinity.v);
VEqualsZero = XMVectorEqual(Q1, Zero);
Sqrt = XMVectorMultiply(Q1, Rsq);
Select = XMVectorEqualInt(VEqualsInfinity, VEqualsZero);
Q1 = XMVectorSelect(Q1, Sqrt, Select);
Q1 = XMVectorMultiply(Q1, g_XMOneHalf.v);
SQ1 = XMVectorMultiply(Rsq, g_XMOneHalf.v);
CQ0 = XMVectorPermute(Q0, Q0, Permute0X0X0Y0W.v);
CQ1 = XMVectorPermute(Q0, SignNNNX, Permute0Y0Z0Z1W.v);
C = XMVectorGreaterOrEqual(CQ0, CQ1);
CX = XMVectorSplatX(C);
CY = XMVectorSplatY(C);
CZ = XMVectorSplatZ(C);
CW = XMVectorSplatW(C);
PermuteSplat = XMVectorSelect(SplatZ.v, SplatY.v, CZ);
SignB = XMVectorSelect(SignNPPP, SignPPNP, CZ);
PermuteControl = XMVectorSelect(Permute2.v, Permute1.v, CZ);
PermuteSplat = XMVectorSelect(PermuteSplat, SplatZ.v, CX);
SignB = XMVectorSelect(SignB, SignNPPP, CX);
PermuteControl = XMVectorSelect(PermuteControl, Permute2.v, CX);
PermuteSplatT = XMVectorSelect(PermuteSplat,SplatX.v, CY);
SignBT = XMVectorSelect(SignB, SignPNPP, CY);
PermuteControlT = XMVectorSelect(PermuteControl,Permute0.v, CY);
PermuteSplat = XMVectorSelect(PermuteSplat, PermuteSplatT, CX);
SignB = XMVectorSelect(SignB, SignBT, CX);
PermuteControl = XMVectorSelect(PermuteControl, PermuteControlT, CX);
PermuteSplat = XMVectorSelect(PermuteSplat,SplatW.v, CW);
SignB = XMVectorSelect(SignB, SignNNNX, CW);
PermuteControl = XMVectorSelect(PermuteControl,Permute3.v, CW);
Scale = XMVectorPermute(SQ1, SQ1, PermuteSplat);
P = XMVectorPermute(M.r[1], M.r[2],PermuteC.v); // {M10, M12, M20, M21}
A = XMVectorPermute(M.r[0], P, PermuteA.v); // {M01, M12, M20, M03}
B = XMVectorPermute(M.r[0], P, PermuteB.v); // {M10, M21, M02, M03}
Q2 = XMVectorMultiplyAdd(SignB, B, A);
Q2 = XMVectorMultiply(Q2, Scale);
Result = XMVectorPermute(Q1, Q2, PermuteControl);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR Q0, Q1, Q2;
XMVECTOR M00, M11, M22;
XMVECTOR CQ0, CQ1, C;
XMVECTOR CX, CY, CZ, CW;
XMVECTOR SQ1, Scale;
XMVECTOR Rsq, Sqrt, VEqualsInfinity, VEqualsZero, Select;
XMVECTOR A, B, P;
XMVECTOR PermuteSplat, PermuteSplatT;
XMVECTOR SignB, SignBT;
XMVECTOR PermuteControl, PermuteControlT;
XMVECTOR Zero;
XMVECTOR Result;
static CONST XMVECTORF32 OneQuarter = {0.25f, 0.25f, 0.25f, 0.25f};
static CONST XMVECTORF32 SignPNNP = {1.0f, -1.0f, -1.0f, 1.0f};
static CONST XMVECTORF32 SignNPNP = {-1.0f, 1.0f, -1.0f, 1.0f};
static CONST XMVECTORF32 SignNNPP = {-1.0f, -1.0f, 1.0f, 1.0f};
static CONST XMVECTORF32 SignPNPP = {1.0f, -1.0f, 1.0f, 1.0f};
static CONST XMVECTORF32 SignPPNP = {1.0f, 1.0f, -1.0f, 1.0f};
static CONST XMVECTORF32 SignNPPP = {-1.0f, 1.0f, 1.0f, 1.0f};
static CONST XMVECTORF32 SignNNNX = {-1.0f, -1.0f, -1.0f, 2.0e-126f};
static CONST XMVECTORI32 Permute0X0X0Y0W = {XM_PERMUTE_0X, XM_PERMUTE_0X, XM_PERMUTE_0Y, XM_PERMUTE_0W};
static CONST XMVECTORI32 Permute0Y0Z0Z1W = {XM_PERMUTE_0Y, XM_PERMUTE_0Z, XM_PERMUTE_0Z, XM_PERMUTE_1W};
static CONST XMVECTORI32 SplatX = {XM_PERMUTE_0X, XM_PERMUTE_0X, XM_PERMUTE_0X, XM_PERMUTE_0X};
static CONST XMVECTORI32 SplatY = {XM_PERMUTE_0Y, XM_PERMUTE_0Y, XM_PERMUTE_0Y, XM_PERMUTE_0Y};
static CONST XMVECTORI32 SplatZ = {XM_PERMUTE_0Z, XM_PERMUTE_0Z, XM_PERMUTE_0Z, XM_PERMUTE_0Z};
static CONST XMVECTORI32 SplatW = {XM_PERMUTE_0W, XM_PERMUTE_0W, XM_PERMUTE_0W, XM_PERMUTE_0W};
static CONST XMVECTORI32 PermuteC = {XM_PERMUTE_0X, XM_PERMUTE_0Z, XM_PERMUTE_1X, XM_PERMUTE_1Y};
static CONST XMVECTORI32 PermuteA = {XM_PERMUTE_0Y, XM_PERMUTE_1Y, XM_PERMUTE_1Z, XM_PERMUTE_0W};
static CONST XMVECTORI32 PermuteB = {XM_PERMUTE_1X, XM_PERMUTE_1W, XM_PERMUTE_0Z, XM_PERMUTE_0W};
static CONST XMVECTORI32 Permute0 = {XM_PERMUTE_0X, XM_PERMUTE_1X, XM_PERMUTE_1Z, XM_PERMUTE_1Y};
static CONST XMVECTORI32 Permute1 = {XM_PERMUTE_1X, XM_PERMUTE_0Y, XM_PERMUTE_1Y, XM_PERMUTE_1Z};
static CONST XMVECTORI32 Permute2 = {XM_PERMUTE_1Z, XM_PERMUTE_1Y, XM_PERMUTE_0Z, XM_PERMUTE_1X};
static CONST XMVECTORI32 Permute3 = {XM_PERMUTE_1Y, XM_PERMUTE_1Z, XM_PERMUTE_1X, XM_PERMUTE_0W};
M00 = XMVectorSplatX(M.r[0]);
M11 = XMVectorSplatY(M.r[1]);
M22 = XMVectorSplatZ(M.r[2]);
Q0 = XMVectorMultiply(SignPNNP, M00);
Q0 = XMVectorMultiplyAdd(SignNPNP, M11, Q0);
Q0 = XMVectorMultiplyAdd(SignNNPP, M22, Q0);
Q1 = XMVectorAdd(Q0, g_XMOne);
Rsq = XMVectorReciprocalSqrt(Q1);
Zero = XMVectorZero();
VEqualsInfinity = XMVectorEqualInt(Q1, g_XMInfinity);
VEqualsZero = XMVectorEqual(Q1, Zero);
Sqrt = XMVectorMultiply(Q1, Rsq);
Select = XMVectorEqualInt(VEqualsInfinity, VEqualsZero);
Q1 = XMVectorSelect(Q1, Sqrt, Select);
Q1 = XMVectorMultiply(Q1, g_XMOneHalf);
SQ1 = XMVectorMultiply(Rsq, g_XMOneHalf);
CQ0 = XMVectorPermute(Q0, Q0, Permute0X0X0Y0W);
CQ1 = XMVectorPermute(Q0, SignNNNX, Permute0Y0Z0Z1W);
C = XMVectorGreaterOrEqual(CQ0, CQ1);
CX = XMVectorSplatX(C);
CY = XMVectorSplatY(C);
CZ = XMVectorSplatZ(C);
CW = XMVectorSplatW(C);
PermuteSplat = XMVectorSelect(SplatZ, SplatY, CZ);
SignB = XMVectorSelect(SignNPPP, SignPPNP, CZ);
PermuteControl = XMVectorSelect(Permute2, Permute1, CZ);
PermuteSplat = XMVectorSelect(PermuteSplat, SplatZ, CX);
SignB = XMVectorSelect(SignB, SignNPPP, CX);
PermuteControl = XMVectorSelect(PermuteControl, Permute2, CX);
PermuteSplatT = XMVectorSelect(PermuteSplat,SplatX, CY);
SignBT = XMVectorSelect(SignB, SignPNPP, CY);
PermuteControlT = XMVectorSelect(PermuteControl,Permute0, CY);
PermuteSplat = XMVectorSelect(PermuteSplat, PermuteSplatT, CX);
SignB = XMVectorSelect(SignB, SignBT, CX);
PermuteControl = XMVectorSelect(PermuteControl, PermuteControlT, CX);
PermuteSplat = XMVectorSelect(PermuteSplat,SplatW, CW);
SignB = XMVectorSelect(SignB, SignNNNX, CW);
PermuteControl = XMVectorSelect(PermuteControl,Permute3, CW);
Scale = XMVectorPermute(SQ1, SQ1, PermuteSplat);
P = XMVectorPermute(M.r[1], M.r[2],PermuteC); // {M10, M12, M20, M21}
A = XMVectorPermute(M.r[0], P, PermuteA); // {M01, M12, M20, M03}
B = XMVectorPermute(M.r[0], P, PermuteB); // {M10, M21, M02, M03}
Q2 = XMVectorMultiplyAdd(SignB, B, A);
Q2 = XMVectorMultiply(Q2, Scale);
Result = XMVectorPermute(Q1, Q2, PermuteControl);
return Result;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
// Conversion operations
//------------------------------------------------------------------------------
//------------------------------------------------------------------------------
XMFINLINE VOID XMQuaternionToAxisAngle
(
XMVECTOR* pAxis,
FLOAT* pAngle,
FXMVECTOR Q
)
{
XMASSERT(pAxis);
XMASSERT(pAngle);
*pAxis = Q;
#if defined(_XM_SSE_INTRINSICS_) && !defined(_XM_NO_INTRINSICS_)
*pAngle = 2.0f * acosf(XMVectorGetW(Q));
#else
*pAngle = 2.0f * XMScalarACos(XMVectorGetW(Q));
#endif
}
/****************************************************************************
*
* Plane
*
****************************************************************************/
//------------------------------------------------------------------------------
// Comparison operations
//------------------------------------------------------------------------------
//------------------------------------------------------------------------------
XMFINLINE BOOL XMPlaneEqual
(
FXMVECTOR P1,
FXMVECTOR P2
)
{
return XMVector4Equal(P1, P2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMPlaneNearEqual
(
FXMVECTOR P1,
FXMVECTOR P2,
FXMVECTOR Epsilon
)
{
XMVECTOR NP1 = XMPlaneNormalize(P1);
XMVECTOR NP2 = XMPlaneNormalize(P2);
return XMVector4NearEqual(NP1, NP2, Epsilon);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMPlaneNotEqual
(
FXMVECTOR P1,
FXMVECTOR P2
)
{
return XMVector4NotEqual(P1, P2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMPlaneIsNaN
(
FXMVECTOR P
)
{
return XMVector4IsNaN(P);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMPlaneIsInfinite
(
FXMVECTOR P
)
{
return XMVector4IsInfinite(P);
}
//------------------------------------------------------------------------------
// Computation operations
//------------------------------------------------------------------------------
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMPlaneDot
(
FXMVECTOR P,
FXMVECTOR V
)
{
#if defined(_XM_NO_INTRINSICS_)
return XMVector4Dot(P, V);
#elif defined(_XM_SSE_INTRINSICS_)
__m128 vTemp2 = V;
__m128 vTemp = _mm_mul_ps(P,vTemp2);
vTemp2 = _mm_shuffle_ps(vTemp2,vTemp,_MM_SHUFFLE(1,0,0,0)); // Copy X to the Z position and Y to the W position
vTemp2 = _mm_add_ps(vTemp2,vTemp); // Add Z = X+Z; W = Y+W;
vTemp = _mm_shuffle_ps(vTemp,vTemp2,_MM_SHUFFLE(0,3,0,0)); // Copy W to the Z position
vTemp = _mm_add_ps(vTemp,vTemp2); // Add Z and W together
return _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(2,2,2,2)); // Splat Z and return
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMPlaneDotCoord
(
FXMVECTOR P,
FXMVECTOR V
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR V3;
XMVECTOR Result;
// Result = P[0] * V[0] + P[1] * V[1] + P[2] * V[2] + P[3]
V3 = XMVectorSelect(g_XMOne.v, V, g_XMSelect1110.v);
Result = XMVector4Dot(P, V3);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR vTemp2 = _mm_and_ps(V,g_XMMask3);
vTemp2 = _mm_or_ps(vTemp2,g_XMIdentityR3);
XMVECTOR vTemp = _mm_mul_ps(P,vTemp2);
vTemp2 = _mm_shuffle_ps(vTemp2,vTemp,_MM_SHUFFLE(1,0,0,0)); // Copy X to the Z position and Y to the W position
vTemp2 = _mm_add_ps(vTemp2,vTemp); // Add Z = X+Z; W = Y+W;
vTemp = _mm_shuffle_ps(vTemp,vTemp2,_MM_SHUFFLE(0,3,0,0)); // Copy W to the Z position
vTemp = _mm_add_ps(vTemp,vTemp2); // Add Z and W together
return _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(2,2,2,2)); // Splat Z and return
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMPlaneDotNormal
(
FXMVECTOR P,
FXMVECTOR V
)
{
return XMVector3Dot(P, V);
}
//------------------------------------------------------------------------------
// XMPlaneNormalizeEst uses a reciprocal estimate and
// returns QNaN on zero and infinite vectors.
XMFINLINE XMVECTOR XMPlaneNormalizeEst
(
FXMVECTOR P
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR Result;
Result = XMVector3ReciprocalLength(P);
Result = XMVectorMultiply(P, Result);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
// Perform the dot product
XMVECTOR vDot = _mm_mul_ps(P,P);
// x=Dot.y, y=Dot.z
XMVECTOR vTemp = _mm_shuffle_ps(vDot,vDot,_MM_SHUFFLE(2,1,2,1));
// Result.x = x+y
vDot = _mm_add_ss(vDot,vTemp);
// x=Dot.z
vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(1,1,1,1));
// Result.x = (x+y)+z
vDot = _mm_add_ss(vDot,vTemp);
// Splat x
vDot = _mm_shuffle_ps(vDot,vDot,_MM_SHUFFLE(0,0,0,0));
// Get the reciprocal
vDot = _mm_rsqrt_ps(vDot);
// Get the reciprocal
vDot = _mm_mul_ps(vDot,P);
return vDot;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMPlaneNormalize
(
FXMVECTOR P
)
{
#if defined(_XM_NO_INTRINSICS_)
FLOAT fLengthSq = sqrtf((P.x*P.x)+(P.y*P.y)+(P.z*P.z));
// Prevent divide by zero
if (fLengthSq) {
fLengthSq = 1.0f/fLengthSq;
}
{
XMVECTOR vResult = {
P.x*fLengthSq,
P.y*fLengthSq,
P.z*fLengthSq,
P.w*fLengthSq
};
return vResult;
}
#elif defined(_XM_SSE_INTRINSICS_)
// Perform the dot product on x,y and z only
XMVECTOR vLengthSq = _mm_mul_ps(P,P);
XMVECTOR vTemp = _mm_shuffle_ps(vLengthSq,vLengthSq,_MM_SHUFFLE(2,1,2,1));
vLengthSq = _mm_add_ss(vLengthSq,vTemp);
vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(1,1,1,1));
vLengthSq = _mm_add_ss(vLengthSq,vTemp);
vLengthSq = _mm_shuffle_ps(vLengthSq,vLengthSq,_MM_SHUFFLE(0,0,0,0));
// Prepare for the division
XMVECTOR vResult = _mm_sqrt_ps(vLengthSq);
// Failsafe on zero (Or epsilon) length planes
// If the length is infinity, set the elements to zero
vLengthSq = _mm_cmpneq_ps(vLengthSq,g_XMInfinity);
// Reciprocal mul to perform the normalization
vResult = _mm_div_ps(P,vResult);
// Any that are infinity, set to zero
vResult = _mm_and_ps(vResult,vLengthSq);
return vResult;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMPlaneIntersectLine
(
FXMVECTOR P,
FXMVECTOR LinePoint1,
FXMVECTOR LinePoint2
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR V1;
XMVECTOR V2;
XMVECTOR D;
XMVECTOR ReciprocalD;
XMVECTOR VT;
XMVECTOR Point;
XMVECTOR Zero;
XMVECTOR Control;
XMVECTOR Result;
V1 = XMVector3Dot(P, LinePoint1);
V2 = XMVector3Dot(P, LinePoint2);
D = XMVectorSubtract(V1, V2);
ReciprocalD = XMVectorReciprocal(D);
VT = XMPlaneDotCoord(P, LinePoint1);
VT = XMVectorMultiply(VT, ReciprocalD);
Point = XMVectorSubtract(LinePoint2, LinePoint1);
Point = XMVectorMultiplyAdd(Point, VT, LinePoint1);
Zero = XMVectorZero();
Control = XMVectorNearEqual(D, Zero, g_XMEpsilon.v);
Result = XMVectorSelect(Point, g_XMQNaN.v, Control);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR V1;
XMVECTOR V2;
XMVECTOR D;
XMVECTOR VT;
XMVECTOR Point;
XMVECTOR Zero;
XMVECTOR Control;
XMVECTOR Result;
V1 = XMVector3Dot(P, LinePoint1);
V2 = XMVector3Dot(P, LinePoint2);
D = _mm_sub_ps(V1, V2);
VT = XMPlaneDotCoord(P, LinePoint1);
VT = _mm_div_ps(VT, D);
Point = _mm_sub_ps(LinePoint2, LinePoint1);
Point = _mm_mul_ps(Point,VT);
Point = _mm_add_ps(Point,LinePoint1);
Zero = XMVectorZero();
Control = XMVectorNearEqual(D, Zero, g_XMEpsilon);
Result = XMVectorSelect(Point, g_XMQNaN, Control);
return Result;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMINLINE VOID XMPlaneIntersectPlane
(
XMVECTOR* pLinePoint1,
XMVECTOR* pLinePoint2,
FXMVECTOR P1,
FXMVECTOR P2
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR V1;
XMVECTOR V2;
XMVECTOR V3;
XMVECTOR LengthSq;
XMVECTOR RcpLengthSq;
XMVECTOR Point;
XMVECTOR P1W;
XMVECTOR P2W;
XMVECTOR Control;
XMVECTOR LinePoint1;
XMVECTOR LinePoint2;
XMASSERT(pLinePoint1);
XMASSERT(pLinePoint2);
V1 = XMVector3Cross(P2, P1);
LengthSq = XMVector3LengthSq(V1);
V2 = XMVector3Cross(P2, V1);
P1W = XMVectorSplatW(P1);
Point = XMVectorMultiply(V2, P1W);
V3 = XMVector3Cross(V1, P1);
P2W = XMVectorSplatW(P2);
Point = XMVectorMultiplyAdd(V3, P2W, Point);
RcpLengthSq = XMVectorReciprocal(LengthSq);
LinePoint1 = XMVectorMultiply(Point, RcpLengthSq);
LinePoint2 = XMVectorAdd(LinePoint1, V1);
Control = XMVectorLessOrEqual(LengthSq, g_XMEpsilon.v);
*pLinePoint1 = XMVectorSelect(LinePoint1,g_XMQNaN.v, Control);
*pLinePoint2 = XMVectorSelect(LinePoint2,g_XMQNaN.v, Control);
#elif defined(_XM_SSE_INTRINSICS_)
XMASSERT(pLinePoint1);
XMASSERT(pLinePoint2);
XMVECTOR V1;
XMVECTOR V2;
XMVECTOR V3;
XMVECTOR LengthSq;
XMVECTOR Point;
XMVECTOR P1W;
XMVECTOR P2W;
XMVECTOR Control;
XMVECTOR LinePoint1;
XMVECTOR LinePoint2;
V1 = XMVector3Cross(P2, P1);
LengthSq = XMVector3LengthSq(V1);
V2 = XMVector3Cross(P2, V1);
P1W = _mm_shuffle_ps(P1,P1,_MM_SHUFFLE(3,3,3,3));
Point = _mm_mul_ps(V2, P1W);
V3 = XMVector3Cross(V1, P1);
P2W = _mm_shuffle_ps(P2,P2,_MM_SHUFFLE(3,3,3,3));
V3 = _mm_mul_ps(V3,P2W);
Point = _mm_add_ps(Point,V3);
LinePoint1 = _mm_div_ps(Point,LengthSq);
LinePoint2 = _mm_add_ps(LinePoint1, V1);
Control = XMVectorLessOrEqual(LengthSq, g_XMEpsilon);
*pLinePoint1 = XMVectorSelect(LinePoint1,g_XMQNaN, Control);
*pLinePoint2 = XMVectorSelect(LinePoint2,g_XMQNaN, Control);
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMPlaneTransform
(
FXMVECTOR P,
CXMMATRIX M
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR X;
XMVECTOR Y;
XMVECTOR Z;
XMVECTOR W;
XMVECTOR Result;
W = XMVectorSplatW(P);
Z = XMVectorSplatZ(P);
Y = XMVectorSplatY(P);
X = XMVectorSplatX(P);
Result = XMVectorMultiply(W, M.r[3]);
Result = XMVectorMultiplyAdd(Z, M.r[2], Result);
Result = XMVectorMultiplyAdd(Y, M.r[1], Result);
Result = XMVectorMultiplyAdd(X, M.r[0], Result);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR X = _mm_shuffle_ps(P,P,_MM_SHUFFLE(0,0,0,0));
XMVECTOR Y = _mm_shuffle_ps(P,P,_MM_SHUFFLE(1,1,1,1));
XMVECTOR Z = _mm_shuffle_ps(P,P,_MM_SHUFFLE(2,2,2,2));
XMVECTOR W = _mm_shuffle_ps(P,P,_MM_SHUFFLE(3,3,3,3));
X = _mm_mul_ps(X, M.r[0]);
Y = _mm_mul_ps(Y, M.r[1]);
Z = _mm_mul_ps(Z, M.r[2]);
W = _mm_mul_ps(W, M.r[3]);
X = _mm_add_ps(X,Z);
Y = _mm_add_ps(Y,W);
X = _mm_add_ps(X,Y);
return X;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMFLOAT4* XMPlaneTransformStream
(
XMFLOAT4* pOutputStream,
UINT OutputStride,
CONST XMFLOAT4* pInputStream,
UINT InputStride,
UINT PlaneCount,
CXMMATRIX M
)
{
return XMVector4TransformStream(pOutputStream,
OutputStride,
pInputStream,
InputStride,
PlaneCount,
M);
}
//------------------------------------------------------------------------------
// Conversion operations
//------------------------------------------------------------------------------
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMPlaneFromPointNormal
(
FXMVECTOR Point,
FXMVECTOR Normal
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR W;
XMVECTOR Result;
W = XMVector3Dot(Point, Normal);
W = XMVectorNegate(W);
Result = XMVectorSelect(W, Normal, g_XMSelect1110.v);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR W;
XMVECTOR Result;
W = XMVector3Dot(Point,Normal);
W = _mm_mul_ps(W,g_XMNegativeOne);
Result = _mm_and_ps(Normal,g_XMMask3);
W = _mm_and_ps(W,g_XMMaskW);
Result = _mm_or_ps(Result,W);
return Result;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMPlaneFromPoints
(
FXMVECTOR Point1,
FXMVECTOR Point2,
FXMVECTOR Point3
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR N;
XMVECTOR D;
XMVECTOR V21;
XMVECTOR V31;
XMVECTOR Result;
V21 = XMVectorSubtract(Point1, Point2);
V31 = XMVectorSubtract(Point1, Point3);
N = XMVector3Cross(V21, V31);
N = XMVector3Normalize(N);
D = XMPlaneDotNormal(N, Point1);
D = XMVectorNegate(D);
Result = XMVectorSelect(D, N, g_XMSelect1110.v);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR N;
XMVECTOR D;
XMVECTOR V21;
XMVECTOR V31;
XMVECTOR Result;
V21 = _mm_sub_ps(Point1, Point2);
V31 = _mm_sub_ps(Point1, Point3);
N = XMVector3Cross(V21, V31);
N = XMVector3Normalize(N);
D = XMPlaneDotNormal(N, Point1);
D = _mm_mul_ps(D,g_XMNegativeOne);
N = _mm_and_ps(N,g_XMMask3);
D = _mm_and_ps(D,g_XMMaskW);
Result = _mm_or_ps(D,N);
return Result;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
/****************************************************************************
*
* Color
*
****************************************************************************/
//------------------------------------------------------------------------------
// Comparison operations
//------------------------------------------------------------------------------
//------------------------------------------------------------------------------
XMFINLINE BOOL XMColorEqual
(
FXMVECTOR C1,
FXMVECTOR C2
)
{
return XMVector4Equal(C1, C2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMColorNotEqual
(
FXMVECTOR C1,
FXMVECTOR C2
)
{
return XMVector4NotEqual(C1, C2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMColorGreater
(
FXMVECTOR C1,
FXMVECTOR C2
)
{
return XMVector4Greater(C1, C2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMColorGreaterOrEqual
(
FXMVECTOR C1,
FXMVECTOR C2
)
{
return XMVector4GreaterOrEqual(C1, C2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMColorLess
(
FXMVECTOR C1,
FXMVECTOR C2
)
{
return XMVector4Less(C1, C2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMColorLessOrEqual
(
FXMVECTOR C1,
FXMVECTOR C2
)
{
return XMVector4LessOrEqual(C1, C2);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMColorIsNaN
(
FXMVECTOR C
)
{
return XMVector4IsNaN(C);
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMColorIsInfinite
(
FXMVECTOR C
)
{
return XMVector4IsInfinite(C);
}
//------------------------------------------------------------------------------
// Computation operations
//------------------------------------------------------------------------------
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMColorNegative
(
FXMVECTOR vColor
)
{
#if defined(_XM_NO_INTRINSICS_)
// XMASSERT(XMVector4GreaterOrEqual(C, XMVectorReplicate(0.0f)));
// XMASSERT(XMVector4LessOrEqual(C, XMVectorReplicate(1.0f)));
XMVECTOR vResult = {
1.0f - vColor.x,
1.0f - vColor.y,
1.0f - vColor.z,
vColor.w
};
return vResult;
#elif defined(_XM_SSE_INTRINSICS_)
// Negate only x,y and z.
XMVECTOR vTemp = _mm_xor_ps(vColor,g_XMNegate3);
// Add 1,1,1,0 to -x,-y,-z,w
return _mm_add_ps(vTemp,g_XMOne3);
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMColorModulate
(
FXMVECTOR C1,
FXMVECTOR C2
)
{
return XMVectorMultiply(C1, C2);
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMColorAdjustSaturation
(
FXMVECTOR vColor,
FLOAT fSaturation
)
{
#if defined(_XM_NO_INTRINSICS_)
CONST XMVECTOR gvLuminance = {0.2125f, 0.7154f, 0.0721f, 0.0f};
// Luminance = 0.2125f * C[0] + 0.7154f * C[1] + 0.0721f * C[2];
// Result = (C - Luminance) * Saturation + Luminance;
FLOAT fLuminance = (vColor.x*gvLuminance.x)+(vColor.y*gvLuminance.y)+(vColor.z*gvLuminance.z);
XMVECTOR vResult = {
((vColor.x - fLuminance)*fSaturation)+fLuminance,
((vColor.y - fLuminance)*fSaturation)+fLuminance,
((vColor.z - fLuminance)*fSaturation)+fLuminance,
vColor.w};
return vResult;
#elif defined(_XM_SSE_INTRINSICS_)
static const XMVECTORF32 gvLuminance = {0.2125f, 0.7154f, 0.0721f, 0.0f};
// Mul RGB by intensity constants
XMVECTOR vLuminance = _mm_mul_ps(vColor,gvLuminance);
// vResult.x = vLuminance.y, vResult.y = vLuminance.y,
// vResult.z = vLuminance.z, vResult.w = vLuminance.z
XMVECTOR vResult = vLuminance;
vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(2,2,1,1));
// vLuminance.x += vLuminance.y
vLuminance = _mm_add_ss(vLuminance,vResult);
// Splat vLuminance.z
vResult = _mm_shuffle_ps(vResult,vResult,_MM_SHUFFLE(2,2,2,2));
// vLuminance.x += vLuminance.z (Dot product)
vLuminance = _mm_add_ss(vLuminance,vResult);
// Splat vLuminance
vLuminance = _mm_shuffle_ps(vLuminance,vLuminance,_MM_SHUFFLE(0,0,0,0));
// Splat fSaturation
XMVECTOR vSaturation = _mm_set_ps1(fSaturation);
// vResult = ((vColor-vLuminance)*vSaturation)+vLuminance;
vResult = _mm_sub_ps(vColor,vLuminance);
vResult = _mm_mul_ps(vResult,vSaturation);
vResult = _mm_add_ps(vResult,vLuminance);
// Retain w from the source color
vLuminance = _mm_shuffle_ps(vResult,vColor,_MM_SHUFFLE(3,2,2,2)); // x = vResult.z,y = vResult.z,z = vColor.z,w=vColor.w
vResult = _mm_shuffle_ps(vResult,vLuminance,_MM_SHUFFLE(3,0,1,0)); // x = vResult.x,y = vResult.y,z = vResult.z,w=vColor.w
return vResult;
#elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS)
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMColorAdjustContrast
(
FXMVECTOR vColor,
FLOAT fContrast
)
{
#if defined(_XM_NO_INTRINSICS_)
// Result = (vColor - 0.5f) * fContrast + 0.5f;
XMVECTOR vResult = {
((vColor.x-0.5f) * fContrast) + 0.5f,
((vColor.y-0.5f) * fContrast) + 0.5f,
((vColor.z-0.5f) * fContrast) + 0.5f,
vColor.w // Leave W untouched
};
return vResult;
#elif defined(_XM_SSE_INTRINSICS_)
XMVECTOR vScale = _mm_set_ps1(fContrast); // Splat the scale
XMVECTOR vResult = _mm_sub_ps(vColor,g_XMOneHalf); // Subtract 0.5f from the source (Saving source)
vResult = _mm_mul_ps(vResult,vScale); // Mul by scale
vResult = _mm_add_ps(vResult,g_XMOneHalf); // Add 0.5f
// Retain w from the source color
vScale = _mm_shuffle_ps(vResult,vColor,_MM_SHUFFLE(3,2,2,2)); // x = vResult.z,y = vResult.z,z = vColor.z,w=vColor.w
vResult = _mm_shuffle_ps(vResult,vScale,_MM_SHUFFLE(3,0,1,0)); // x = vResult.x,y = vResult.y,z = vResult.z,w=vColor.w
return vResult;
#elif defined(XM_NO_MISALIGNED_VECTOR_ACCESS)
#endif // _XM_VMX128_INTRINSICS_
}
/****************************************************************************
*
* Miscellaneous
*
****************************************************************************/
//------------------------------------------------------------------------------
XMINLINE BOOL XMVerifyCPUSupport()
{
#if defined(_XM_NO_INTRINSICS_) || !defined(_XM_SSE_INTRINSICS_)
return TRUE;
#else // _XM_SSE_INTRINSICS_
// Note that on Windows 2000 or older, SSE2 detection is not supported so this will always fail
// Detecting SSE2 on older versions of Windows would require using cpuid directly
return ( IsProcessorFeaturePresent( PF_XMMI_INSTRUCTIONS_AVAILABLE ) && IsProcessorFeaturePresent( PF_XMMI64_INSTRUCTIONS_AVAILABLE ) );
#endif
}
//------------------------------------------------------------------------------
#define XMASSERT_LINE_STRING_SIZE 16
XMINLINE VOID XMAssert
(
CONST CHAR* pExpression,
CONST CHAR* pFileName,
UINT LineNumber
)
{
CHAR aLineString[XMASSERT_LINE_STRING_SIZE];
CHAR* pLineString;
UINT Line;
aLineString[XMASSERT_LINE_STRING_SIZE - 2] = '0';
aLineString[XMASSERT_LINE_STRING_SIZE - 1] = '\0';
for (Line = LineNumber, pLineString = aLineString + XMASSERT_LINE_STRING_SIZE - 2;
Line != 0 && pLineString >= aLineString;
Line /= 10, pLineString--)
{
*pLineString = (CHAR)('0' + (Line % 10));
}
#ifndef NO_OUTPUT_DEBUG_STRING
OutputDebugStringA("Assertion failed: ");
OutputDebugStringA(pExpression);
OutputDebugStringA(", file ");
OutputDebugStringA(pFileName);
OutputDebugStringA(", line ");
OutputDebugStringA(pLineString + 1);
OutputDebugStringA("\r\n");
#else
DbgPrint("Assertion failed: %s, file %s, line %d\r\n", pExpression, pFileName, LineNumber);
#endif
__debugbreak();
}
//------------------------------------------------------------------------------
XMFINLINE XMVECTOR XMFresnelTerm
(
FXMVECTOR CosIncidentAngle,
FXMVECTOR RefractionIndex
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR G;
XMVECTOR D, S;
XMVECTOR V0, V1, V2, V3;
XMVECTOR Result;
// Result = 0.5f * (g - c)^2 / (g + c)^2 * ((c * (g + c) - 1)^2 / (c * (g - c) + 1)^2 + 1) where
// c = CosIncidentAngle
// g = sqrt(c^2 + RefractionIndex^2 - 1)
XMASSERT(!XMVector4IsInfinite(CosIncidentAngle));
G = XMVectorMultiplyAdd(RefractionIndex, RefractionIndex, g_XMNegativeOne.v);
G = XMVectorMultiplyAdd(CosIncidentAngle, CosIncidentAngle, G);
G = XMVectorAbs(G);
G = XMVectorSqrt(G);
S = XMVectorAdd(G, CosIncidentAngle);
D = XMVectorSubtract(G, CosIncidentAngle);
V0 = XMVectorMultiply(D, D);
V1 = XMVectorMultiply(S, S);
V1 = XMVectorReciprocal(V1);
V0 = XMVectorMultiply(g_XMOneHalf.v, V0);
V0 = XMVectorMultiply(V0, V1);
V2 = XMVectorMultiplyAdd(CosIncidentAngle, S, g_XMNegativeOne.v);
V3 = XMVectorMultiplyAdd(CosIncidentAngle, D, g_XMOne.v);
V2 = XMVectorMultiply(V2, V2);
V3 = XMVectorMultiply(V3, V3);
V3 = XMVectorReciprocal(V3);
V2 = XMVectorMultiplyAdd(V2, V3, g_XMOne.v);
Result = XMVectorMultiply(V0, V2);
Result = XMVectorSaturate(Result);
return Result;
#elif defined(_XM_SSE_INTRINSICS_)
// Result = 0.5f * (g - c)^2 / (g + c)^2 * ((c * (g + c) - 1)^2 / (c * (g - c) + 1)^2 + 1) where
// c = CosIncidentAngle
// g = sqrt(c^2 + RefractionIndex^2 - 1)
XMASSERT(!XMVector4IsInfinite(CosIncidentAngle));
// G = sqrt(abs((RefractionIndex^2-1) + CosIncidentAngle^2))
XMVECTOR G = _mm_mul_ps(RefractionIndex,RefractionIndex);
XMVECTOR vTemp = _mm_mul_ps(CosIncidentAngle,CosIncidentAngle);
G = _mm_sub_ps(G,g_XMOne);
vTemp = _mm_add_ps(vTemp,G);
// max((0-vTemp),vTemp) == abs(vTemp)
// The abs is needed to deal with refraction and cosine being zero
G = _mm_setzero_ps();
G = _mm_sub_ps(G,vTemp);
G = _mm_max_ps(G,vTemp);
// Last operation, the sqrt()
G = _mm_sqrt_ps(G);
// Calc G-C and G+C
XMVECTOR GAddC = _mm_add_ps(G,CosIncidentAngle);
XMVECTOR GSubC = _mm_sub_ps(G,CosIncidentAngle);
// Perform the term (0.5f *(g - c)^2) / (g + c)^2
XMVECTOR vResult = _mm_mul_ps(GSubC,GSubC);
vTemp = _mm_mul_ps(GAddC,GAddC);
vResult = _mm_mul_ps(vResult,g_XMOneHalf);
vResult = _mm_div_ps(vResult,vTemp);
// Perform the term ((c * (g + c) - 1)^2 / (c * (g - c) + 1)^2 + 1)
GAddC = _mm_mul_ps(GAddC,CosIncidentAngle);
GSubC = _mm_mul_ps(GSubC,CosIncidentAngle);
GAddC = _mm_sub_ps(GAddC,g_XMOne);
GSubC = _mm_add_ps(GSubC,g_XMOne);
GAddC = _mm_mul_ps(GAddC,GAddC);
GSubC = _mm_mul_ps(GSubC,GSubC);
GAddC = _mm_div_ps(GAddC,GSubC);
GAddC = _mm_add_ps(GAddC,g_XMOne);
// Multiply the two term parts
vResult = _mm_mul_ps(vResult,GAddC);
// Clamp to 0.0 - 1.0f
vResult = _mm_max_ps(vResult,g_XMZero);
vResult = _mm_min_ps(vResult,g_XMOne);
return vResult;
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE BOOL XMScalarNearEqual
(
FLOAT S1,
FLOAT S2,
FLOAT Epsilon
)
{
FLOAT Delta = S1 - S2;
#if defined(_XM_NO_INTRINSICS_)
UINT AbsDelta = *(UINT*)&Delta & 0x7FFFFFFF;
return (*(FLOAT*)&AbsDelta <= Epsilon);
#elif defined(_XM_SSE_INTRINSICS_)
return (fabsf(Delta) <= Epsilon);
#else
return (__fabs(Delta) <= Epsilon);
#endif
}
//------------------------------------------------------------------------------
// Modulo the range of the given angle such that -XM_PI <= Angle < XM_PI
XMFINLINE FLOAT XMScalarModAngle
(
FLOAT Angle
)
{
// Note: The modulo is performed with unsigned math only to work
// around a precision error on numbers that are close to PI
float fTemp;
#if defined(_XM_NO_INTRINSICS_) || !defined(_XM_VMX128_INTRINSICS_)
// Normalize the range from 0.0f to XM_2PI
Angle = Angle + XM_PI;
// Perform the modulo, unsigned
fTemp = fabsf(Angle);
fTemp = fTemp - (XM_2PI * (FLOAT)((INT)(fTemp/XM_2PI)));
// Restore the number to the range of -XM_PI to XM_PI-epsilon
fTemp = fTemp - XM_PI;
// If the modulo'd value was negative, restore negation
if (Angle<0.0f) {
fTemp = -fTemp;
}
return fTemp;
#else
#endif
}
//------------------------------------------------------------------------------
XMINLINE FLOAT XMScalarSin
(
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
FLOAT ValueMod;
FLOAT ValueSq;
XMVECTOR V0123, V0246, V1357, V9111315, V17192123;
XMVECTOR V1, V7, V8;
XMVECTOR R0, R1, R2;
ValueMod = XMScalarModAngle(Value);
// sin(V) ~= V - V^3 / 3! + V^5 / 5! - V^7 / 7! + V^9 / 9! - V^11 / 11! + V^13 / 13! - V^15 / 15! +
// V^17 / 17! - V^19 / 19! + V^21 / 21! - V^23 / 23! (for -PI <= V < PI)
ValueSq = ValueMod * ValueMod;
V0123 = XMVectorSet(1.0f, ValueMod, ValueSq, ValueSq * ValueMod);
V1 = XMVectorSplatY(V0123);
V0246 = XMVectorMultiply(V0123, V0123);
V1357 = XMVectorMultiply(V0246, V1);
V7 = XMVectorSplatW(V1357);
V8 = XMVectorMultiply(V7, V1);
V9111315 = XMVectorMultiply(V1357, V8);
V17192123 = XMVectorMultiply(V9111315, V8);
R0 = XMVector4Dot(V1357, g_XMSinCoefficients0.v);
R1 = XMVector4Dot(V9111315, g_XMSinCoefficients1.v);
R2 = XMVector4Dot(V17192123, g_XMSinCoefficients2.v);
return R0.v[0] + R1.v[0] + R2.v[0];
#elif defined(_XM_SSE_INTRINSICS_)
return sinf( Value );
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMINLINE FLOAT XMScalarCos
(
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
FLOAT ValueMod;
FLOAT ValueSq;
XMVECTOR V0123, V0246, V8101214, V16182022;
XMVECTOR V2, V6, V8;
XMVECTOR R0, R1, R2;
ValueMod = XMScalarModAngle(Value);
// cos(V) ~= 1 - V^2 / 2! + V^4 / 4! - V^6 / 6! + V^8 / 8! - V^10 / 10! +
// V^12 / 12! - V^14 / 14! + V^16 / 16! - V^18 / 18! + V^20 / 20! - V^22 / 22! (for -PI <= V < PI)
ValueSq = ValueMod * ValueMod;
V0123 = XMVectorSet(1.0f, ValueMod, ValueSq, ValueSq * ValueMod);
V0246 = XMVectorMultiply(V0123, V0123);
V2 = XMVectorSplatZ(V0123);
V6 = XMVectorSplatW(V0246);
V8 = XMVectorMultiply(V6, V2);
V8101214 = XMVectorMultiply(V0246, V8);
V16182022 = XMVectorMultiply(V8101214, V8);
R0 = XMVector4Dot(V0246, g_XMCosCoefficients0.v);
R1 = XMVector4Dot(V8101214, g_XMCosCoefficients1.v);
R2 = XMVector4Dot(V16182022, g_XMCosCoefficients2.v);
return R0.v[0] + R1.v[0] + R2.v[0];
#elif defined(_XM_SSE_INTRINSICS_)
return cosf(Value);
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMINLINE VOID XMScalarSinCos
(
FLOAT* pSin,
FLOAT* pCos,
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
FLOAT ValueMod;
FLOAT ValueSq;
XMVECTOR V0123, V0246, V1357, V8101214, V9111315, V16182022, V17192123;
XMVECTOR V1, V2, V6, V8;
XMVECTOR S0, S1, S2, C0, C1, C2;
XMASSERT(pSin);
XMASSERT(pCos);
ValueMod = XMScalarModAngle(Value);
// sin(V) ~= V - V^3 / 3! + V^5 / 5! - V^7 / 7! + V^9 / 9! - V^11 / 11! + V^13 / 13! - V^15 / 15! +
// V^17 / 17! - V^19 / 19! + V^21 / 21! - V^23 / 23! (for -PI <= V < PI)
// cos(V) ~= 1 - V^2 / 2! + V^4 / 4! - V^6 / 6! + V^8 / 8! - V^10 / 10! +
// V^12 / 12! - V^14 / 14! + V^16 / 16! - V^18 / 18! + V^20 / 20! - V^22 / 22! (for -PI <= V < PI)
ValueSq = ValueMod * ValueMod;
V0123 = XMVectorSet(1.0f, ValueMod, ValueSq, ValueSq * ValueMod);
V1 = XMVectorSplatY(V0123);
V2 = XMVectorSplatZ(V0123);
V0246 = XMVectorMultiply(V0123, V0123);
V1357 = XMVectorMultiply(V0246, V1);
V6 = XMVectorSplatW(V0246);
V8 = XMVectorMultiply(V6, V2);
V8101214 = XMVectorMultiply(V0246, V8);
V9111315 = XMVectorMultiply(V1357, V8);
V16182022 = XMVectorMultiply(V8101214, V8);
V17192123 = XMVectorMultiply(V9111315, V8);
C0 = XMVector4Dot(V0246, g_XMCosCoefficients0.v);
S0 = XMVector4Dot(V1357, g_XMSinCoefficients0.v);
C1 = XMVector4Dot(V8101214, g_XMCosCoefficients1.v);
S1 = XMVector4Dot(V9111315, g_XMSinCoefficients1.v);
C2 = XMVector4Dot(V16182022, g_XMCosCoefficients2.v);
S2 = XMVector4Dot(V17192123, g_XMSinCoefficients2.v);
*pCos = C0.v[0] + C1.v[0] + C2.v[0];
*pSin = S0.v[0] + S1.v[0] + S2.v[0];
#elif defined(_XM_SSE_INTRINSICS_)
XMASSERT(pSin);
XMASSERT(pCos);
*pSin = sinf(Value);
*pCos = cosf(Value);
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMINLINE FLOAT XMScalarASin
(
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
FLOAT AbsValue, Value2, Value3, D;
XMVECTOR AbsV, R0, R1, Result;
XMVECTOR V3;
*(UINT*)&AbsValue = *(UINT*)&Value & 0x7FFFFFFF;
Value2 = Value * AbsValue;
Value3 = Value * Value2;
D = (Value - Value2) / sqrtf(1.00000011921f - AbsValue);
AbsV = XMVectorReplicate(AbsValue);
V3.v[0] = Value3;
V3.v[1] = 1.0f;
V3.v[2] = Value3;
V3.v[3] = 1.0f;
R1 = XMVectorSet(D, D, Value, Value);
R1 = XMVectorMultiply(R1, V3);
R0 = XMVectorMultiplyAdd(AbsV, g_XMASinCoefficients0.v, g_XMASinCoefficients1.v);
R0 = XMVectorMultiplyAdd(AbsV, R0, g_XMASinCoefficients2.v);
Result = XMVector4Dot(R0, R1);
return Result.v[0];
#elif defined(_XM_SSE_INTRINSICS_)
return asinf(Value);
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMINLINE FLOAT XMScalarACos
(
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
return XM_PIDIV2 - XMScalarASin(Value);
#elif defined(_XM_SSE_INTRINSICS_)
return acosf(Value);
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE FLOAT XMScalarSinEst
(
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
FLOAT ValueSq;
XMVECTOR V;
XMVECTOR Y;
XMVECTOR Result;
XMASSERT(Value >= -XM_PI);
XMASSERT(Value < XM_PI);
// sin(V) ~= V - V^3 / 3! + V^5 / 5! - V^7 / 7! (for -PI <= V < PI)
ValueSq = Value * Value;
V = XMVectorSet(1.0f, Value, ValueSq, ValueSq * Value);
Y = XMVectorSplatY(V);
V = XMVectorMultiply(V, V);
V = XMVectorMultiply(V, Y);
Result = XMVector4Dot(V, g_XMSinEstCoefficients.v);
return Result.v[0];
#elif defined(_XM_SSE_INTRINSICS_)
XMASSERT(Value >= -XM_PI);
XMASSERT(Value < XM_PI);
float ValueSq = Value*Value;
XMVECTOR vValue = _mm_set_ps1(Value);
XMVECTOR vTemp = _mm_set_ps(ValueSq * Value,ValueSq,Value,1.0f);
vTemp = _mm_mul_ps(vTemp,vTemp);
vTemp = _mm_mul_ps(vTemp,vValue);
// vTemp = Value,Value^3,Value^5,Value^7
vTemp = _mm_mul_ps(vTemp,g_XMSinEstCoefficients);
vValue = _mm_shuffle_ps(vValue,vTemp,_MM_SHUFFLE(1,0,0,0)); // Copy X to the Z position and Y to the W position
vValue = _mm_add_ps(vValue,vTemp); // Add Z = X+Z; W = Y+W;
vTemp = _mm_shuffle_ps(vTemp,vValue,_MM_SHUFFLE(0,3,0,0)); // Copy W to the Z position
vTemp = _mm_add_ps(vTemp,vValue); // Add Z and W together
vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(2,2,2,2)); // Splat Z and return
#if defined(_MSC_VER) && (_MSC_VER>=1500)
return _mm_cvtss_f32(vTemp);
#else
return vTemp.m128_f32[0];
#endif
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE FLOAT XMScalarCosEst
(
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
FLOAT ValueSq;
XMVECTOR V;
XMVECTOR Result;
XMASSERT(Value >= -XM_PI);
XMASSERT(Value < XM_PI);
// cos(V) ~= 1 - V^2 / 2! + V^4 / 4! - V^6 / 6! (for -PI <= V < PI)
ValueSq = Value * Value;
V = XMVectorSet(1.0f, Value, ValueSq, ValueSq * Value);
V = XMVectorMultiply(V, V);
Result = XMVector4Dot(V, g_XMCosEstCoefficients.v);
return Result.v[0];
#elif defined(_XM_SSE_INTRINSICS_)
XMASSERT(Value >= -XM_PI);
XMASSERT(Value < XM_PI);
float ValueSq = Value*Value;
XMVECTOR vValue = _mm_setzero_ps();
XMVECTOR vTemp = _mm_set_ps(ValueSq * Value,ValueSq,Value,1.0f);
vTemp = _mm_mul_ps(vTemp,vTemp);
// vTemp = 1.0f,Value^2,Value^4,Value^6
vTemp = _mm_mul_ps(vTemp,g_XMCosEstCoefficients);
vValue = _mm_shuffle_ps(vValue,vTemp,_MM_SHUFFLE(1,0,0,0)); // Copy X to the Z position and Y to the W position
vValue = _mm_add_ps(vValue,vTemp); // Add Z = X+Z; W = Y+W;
vTemp = _mm_shuffle_ps(vTemp,vValue,_MM_SHUFFLE(0,3,0,0)); // Copy W to the Z position
vTemp = _mm_add_ps(vTemp,vValue); // Add Z and W together
vTemp = _mm_shuffle_ps(vTemp,vTemp,_MM_SHUFFLE(2,2,2,2)); // Splat Z and return
#if defined(_MSC_VER) && (_MSC_VER>=1500)
return _mm_cvtss_f32(vTemp);
#else
return vTemp.m128_f32[0];
#endif
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE VOID XMScalarSinCosEst
(
FLOAT* pSin,
FLOAT* pCos,
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
FLOAT ValueSq;
XMVECTOR V, Sin, Cos;
XMVECTOR Y;
XMASSERT(pSin);
XMASSERT(pCos);
XMASSERT(Value >= -XM_PI);
XMASSERT(Value < XM_PI);
// sin(V) ~= V - V^3 / 3! + V^5 / 5! - V^7 / 7! (for -PI <= V < PI)
// cos(V) ~= 1 - V^2 / 2! + V^4 / 4! - V^6 / 6! (for -PI <= V < PI)
ValueSq = Value * Value;
V = XMVectorSet(1.0f, Value, ValueSq, Value * ValueSq);
Y = XMVectorSplatY(V);
Cos = XMVectorMultiply(V, V);
Sin = XMVectorMultiply(Cos, Y);
Cos = XMVector4Dot(Cos, g_XMCosEstCoefficients.v);
Sin = XMVector4Dot(Sin, g_XMSinEstCoefficients.v);
*pCos = Cos.v[0];
*pSin = Sin.v[0];
#elif defined(_XM_SSE_INTRINSICS_)
XMASSERT(pSin);
XMASSERT(pCos);
XMASSERT(Value >= -XM_PI);
XMASSERT(Value < XM_PI);
float ValueSq = Value * Value;
XMVECTOR Cos = _mm_set_ps(Value * ValueSq,ValueSq,Value,1.0f);
XMVECTOR Sin = _mm_set_ps1(Value);
Cos = _mm_mul_ps(Cos,Cos);
Sin = _mm_mul_ps(Sin,Cos);
// Cos = 1.0f,Value^2,Value^4,Value^6
Cos = XMVector4Dot(Cos,g_XMCosEstCoefficients);
_mm_store_ss(pCos,Cos);
// Sin = Value,Value^3,Value^5,Value^7
Sin = XMVector4Dot(Sin, g_XMSinEstCoefficients);
_mm_store_ss(pSin,Sin);
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE FLOAT XMScalarASinEst
(
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR VR, CR, CS;
XMVECTOR Result;
FLOAT AbsV, V2, D;
CONST FLOAT OnePlusEps = 1.00000011921f;
*(UINT*)&AbsV = *(UINT*)&Value & 0x7FFFFFFF;
V2 = Value * AbsV;
D = OnePlusEps - AbsV;
CS = XMVectorSet(Value, 1.0f, 1.0f, V2);
VR = XMVectorSet(sqrtf(D), Value, V2, D * AbsV);
CR = XMVectorMultiply(CS, g_XMASinEstCoefficients.v);
Result = XMVector4Dot(VR, CR);
return Result.v[0];
#elif defined(_XM_SSE_INTRINSICS_)
CONST FLOAT OnePlusEps = 1.00000011921f;
FLOAT AbsV = fabsf(Value);
FLOAT V2 = Value * AbsV; // Square with sign retained
FLOAT D = OnePlusEps - AbsV;
XMVECTOR Result = _mm_set_ps(V2,1.0f,1.0f,Value);
XMVECTOR VR = _mm_set_ps(D * AbsV,V2,Value,sqrtf(D));
Result = _mm_mul_ps(Result, g_XMASinEstCoefficients);
Result = XMVector4Dot(VR,Result);
#if defined(_MSC_VER) && (_MSC_VER>=1500)
return _mm_cvtss_f32(Result);
#else
return Result.m128_f32[0];
#endif
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
//------------------------------------------------------------------------------
XMFINLINE FLOAT XMScalarACosEst
(
FLOAT Value
)
{
#if defined(_XM_NO_INTRINSICS_)
XMVECTOR VR, CR, CS;
XMVECTOR Result;
FLOAT AbsV, V2, D;
CONST FLOAT OnePlusEps = 1.00000011921f;
// return XM_PIDIV2 - XMScalarASin(Value);
*(UINT*)&AbsV = *(UINT*)&Value & 0x7FFFFFFF;
V2 = Value * AbsV;
D = OnePlusEps - AbsV;
CS = XMVectorSet(Value, 1.0f, 1.0f, V2);
VR = XMVectorSet(sqrtf(D), Value, V2, D * AbsV);
CR = XMVectorMultiply(CS, g_XMASinEstCoefficients.v);
Result = XMVector4Dot(VR, CR);
return XM_PIDIV2 - Result.v[0];
#elif defined(_XM_SSE_INTRINSICS_)
CONST FLOAT OnePlusEps = 1.00000011921f;
FLOAT AbsV = fabsf(Value);
FLOAT V2 = Value * AbsV; // Value^2 retaining sign
FLOAT D = OnePlusEps - AbsV;
XMVECTOR Result = _mm_set_ps(V2,1.0f,1.0f,Value);
XMVECTOR VR = _mm_set_ps(D * AbsV,V2,Value,sqrtf(D));
Result = _mm_mul_ps(Result,g_XMASinEstCoefficients);
Result = XMVector4Dot(VR,Result);
#if defined(_MSC_VER) && (_MSC_VER>=1500)
return XM_PIDIV2 - _mm_cvtss_f32(Result);
#else
return XM_PIDIV2 - Result.m128_f32[0];
#endif
#else // _XM_VMX128_INTRINSICS_
#endif // _XM_VMX128_INTRINSICS_
}
#endif // __XNAMATHMISC_INL__