Skip to content

Commit aee6e90

Browse files
walbournwalbourn
committed
DirectXMath 3.06
1 parent fb43a89 commit aee6e90

File tree

5 files changed

+128
-166
lines changed

5 files changed

+128
-166
lines changed

‎Inc/DirectXMath.h‎

Lines changed: 8 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -17,7 +17,7 @@
1717
#error DirectX Math requires C++
1818
#endif
1919

20-
#define DIRECTX_MATH_VERSION 305
20+
#define DIRECTX_MATH_VERSION 306
2121

2222
#if !defined(_XM_BIGENDIAN_) && !defined(_XM_LITTLEENDIAN_)
2323
#if defined(_M_X64) || defined(_M_IX86) || defined(_M_ARM)
@@ -29,6 +29,7 @@
2929
#endif
3030
#endif // !_XM_BIGENDIAN_ && !_XM_LITTLEENDIAN_
3131

32+
3233
#if defined(_MSC_VER) && !defined(_M_ARM) && (!_MANAGED) && (!_M_CEE) && (!defined(_M_IX86_FP) || (_M_IX86_FP > 1)) && !defined(_XM_NO_INTRINSICS_) && !defined(_XM_VECTORCALL_)
3334
#if ((_MSC_FULL_VER >= 170065501) && (_MSC_VER < 1800)) || (_MSC_FULL_VER >= 180020418)
3435
#define _XM_VECTORCALL_ 1
@@ -279,8 +280,8 @@ typedef const XMVECTOR FXMVECTOR;
279280
typedef const XMVECTOR& FXMVECTOR;
280281
#endif
281282

282-
// Fix-up for (4th) XMVECTOR parameter to pass in-register for ARM, Xbox 360, and vector call; by reference otherwise
283-
#if ( defined(_M_ARM) || defined(_XM_VMX128_INTRINSICS_) || _XM_VECTORCALL_ ) && !defined(_XM_NO_INTRINSICS_)
283+
// Fix-up for (4th) XMVECTOR parameter to pass in-register for ARM, Xbox 360, and x64 vector call; by reference otherwise
284+
#if ( defined(_M_ARM) || defined(_XM_VMX128_INTRINSICS_) || (_XM_VECTORCALL_ && !defined(_M_IX86) ) ) && !defined(_XM_NO_INTRINSICS_)
284285
typedef const XMVECTOR GXMVECTOR;
285286
#else
286287
typedef const XMVECTOR& GXMVECTOR;
@@ -1392,6 +1393,10 @@ XMVECTOR XM_CALLCONV XMColorXYZToRGB( FXMVECTOR xyz );
13921393
XMVECTOR XM_CALLCONV XMColorXYZToSRGB( FXMVECTOR xyz );
13931394
XMVECTOR XM_CALLCONV XMColorSRGBToXYZ( FXMVECTOR srgb );
13941395

1396+
XMVECTOR XM_CALLCONV XMColorRGBToSRGB( FXMVECTOR rgb );
1397+
XMVECTOR XM_CALLCONV XMColorSRGBToRGB( FXMVECTOR srgb );
1398+
1399+
13951400
/****************************************************************************
13961401
*
13971402
* Miscellaneous operations

‎Inc/DirectXMathMisc.inl‎

Lines changed: 36 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1985,6 +1985,42 @@ inline XMVECTOR XM_CALLCONV XMColorSRGBToXYZ( FXMVECTOR srgb )
19851985
return XMVectorSelect( srgb, clr, g_XMSelect1110 );
19861986
}
19871987

1988+
//------------------------------------------------------------------------------
1989+
1990+
inline XMVECTOR XM_CALLCONV XMColorRGBToSRGB( FXMVECTOR rgb )
1991+
{
1992+
static const XMVECTORF32 Cutoff = { 0.0031308f, 0.0031308f, 0.0031308f, 1.f };
1993+
static const XMVECTORF32 Linear = { 12.92f, 12.92f, 12.92f, 1.f };
1994+
static const XMVECTORF32 Scale = { 1.055f, 1.055f, 1.055f, 1.f };
1995+
static const XMVECTORF32 Bias = { 0.055f, 0.055f, 0.055f, 0.f };
1996+
static const XMVECTORF32 InvGamma = { 1.0f/2.4f, 1.0f/2.4f, 1.0f/2.4f, 1.f };
1997+
1998+
XMVECTOR V = XMVectorSaturate(rgb);
1999+
XMVECTOR V0 = XMVectorMultiply( V, Linear );
2000+
XMVECTOR V1 = Scale * XMVectorPow( V, InvGamma ) - Bias;
2001+
XMVECTOR select = XMVectorLess( V, Cutoff );
2002+
V = XMVectorSelect( V1, V0, select );
2003+
return XMVectorSelect( rgb, V, g_XMSelect1110 );
2004+
}
2005+
2006+
//------------------------------------------------------------------------------
2007+
2008+
inline XMVECTOR XM_CALLCONV XMColorSRGBToRGB( FXMVECTOR srgb )
2009+
{
2010+
static const XMVECTORF32 Cutoff = { 0.04045f, 0.04045f, 0.04045f, 1.f };
2011+
static const XMVECTORF32 ILinear = { 1.f/12.92f, 1.f/12.92f, 1.f/12.92f, 1.f };
2012+
static const XMVECTORF32 Scale = { 1.f/1.055f, 1.f/1.055f, 1.f/1.055f, 1.f };
2013+
static const XMVECTORF32 Bias = { 0.055f, 0.055f, 0.055f, 0.f };
2014+
static const XMVECTORF32 Gamma = { 2.4f, 2.4f, 2.4f, 1.f };
2015+
2016+
XMVECTOR V = XMVectorSaturate(srgb);
2017+
XMVECTOR V0 = XMVectorMultiply( V, ILinear );
2018+
XMVECTOR V1 = XMVectorPow( (V + Bias) * Scale, Gamma );
2019+
XMVECTOR select = XMVectorGreater( V, Cutoff );
2020+
V = XMVectorSelect( V0, V1, select );
2021+
return XMVectorSelect( srgb, V, g_XMSelect1110 );
2022+
}
2023+
19882024
/****************************************************************************
19892025
*
19902026
* Miscellaneous

‎Inc/DirectXMathVector.inl‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -2323,7 +2323,6 @@ inline XMVECTOR XM_CALLCONV XMVectorMax
23232323

23242324
//------------------------------------------------------------------------------
23252325

2326-
#if defined(_XM_NO_INTRINSICS_)
23272326
namespace Internal
23282327
{
23292328
inline float round_to_nearest( float x )
@@ -2345,7 +2344,8 @@ namespace Internal
23452344
return i + 1.f;
23462345
}
23472346
};
2348-
#else
2347+
2348+
#if !defined(_XM_NO_INTRINSICS_)
23492349
#pragma float_control(push)
23502350
#pragma float_control(precise, on)
23512351
#endif

‎Inc/DirectXPackedVector.h‎

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -921,6 +921,7 @@ XMVECTOR XM_CALLCONV XMLoadXDec4(_In_ const XMXDEC4* pSource);
921921
XMVECTOR XM_CALLCONV XMLoadDecN4(_In_ const XMDECN4* pSource);
922922
XMVECTOR XM_CALLCONV XMLoadDec4(_In_ const XMDEC4* pSource);
923923
XMVECTOR XM_CALLCONV XMLoadUDecN4(_In_ const XMUDECN4* pSource);
924+
XMVECTOR XM_CALLCONV XMLoadUDecN4_XR(_In_ const XMUDECN4* pSource);
924925
XMVECTOR XM_CALLCONV XMLoadUDec4(_In_ const XMUDEC4* pSource);
925926
XMVECTOR XM_CALLCONV XMLoadByteN4(_In_ const XMBYTEN4* pSource);
926927
XMVECTOR XM_CALLCONV XMLoadByte4(_In_ const XMBYTE4* pSource);
@@ -962,6 +963,7 @@ void XM_CALLCONV XMStoreXDec4(_Out_ XMXDEC4* pDestination, _In_ FXMVECTOR
962963
void XM_CALLCONV XMStoreDecN4(_Out_ XMDECN4* pDestination, _In_ FXMVECTOR V);
963964
void XM_CALLCONV XMStoreDec4(_Out_ XMDEC4* pDestination, _In_ FXMVECTOR V);
964965
void XM_CALLCONV XMStoreUDecN4(_Out_ XMUDECN4* pDestination, _In_ FXMVECTOR V);
966+
void XM_CALLCONV XMStoreUDecN4_XR(_Out_ XMUDECN4* pDestination, _In_ FXMVECTOR V);
965967
void XM_CALLCONV XMStoreUDec4(_Out_ XMUDEC4* pDestination, _In_ FXMVECTOR V);
966968
void XM_CALLCONV XMStoreByteN4(_Out_ XMBYTEN4* pDestination, _In_ FXMVECTOR V);
967969
void XM_CALLCONV XMStoreByte4(_Out_ XMBYTE4* pDestination, _In_ FXMVECTOR V);

‎Inc/DirectXPackedVector.inl‎

Lines changed: 80 additions & 161 deletions
Original file line numberDiff line numberDiff line change
@@ -605,104 +605,16 @@ inline XMVECTOR XM_CALLCONV PackedVector::XMLoadFloat3SE
605605
{
606606
assert(pSource);
607607

608-
__declspec(align(16)) uint32_t Result[4];
609-
uint32_t Mantissa;
610-
uint32_t Exponent, ExpBits;
611-
612-
if ( pSource->e == 0x1f ) // INF or NAN
613-
{
614-
Result[0] = 0x7f800000 | (pSource->xm << 14);
615-
Result[1] = 0x7f800000 | (pSource->ym << 14);
616-
Result[2] = 0x7f800000 | (pSource->zm << 14);
617-
}
618-
else if ( pSource->e != 0 ) // The values are all normalized
619-
{
620-
Exponent = pSource->e;
621-
622-
ExpBits = (Exponent + 112) << 23;
623-
624-
Mantissa = pSource->xm;
625-
Result[0] = ExpBits | (Mantissa << 14);
626-
627-
Mantissa = pSource->ym;
628-
Result[1] = ExpBits | (Mantissa << 14);
629-
630-
Mantissa = pSource->zm;
631-
Result[2] = ExpBits | (Mantissa << 14);
632-
}
633-
else
634-
{
635-
// X Channel
636-
Mantissa = pSource->xm;
637-
638-
if (Mantissa != 0) // The value is denormalized
639-
{
640-
// Normalize the value in the resulting float
641-
Exponent = 1;
642-
643-
do
644-
{
645-
Exponent--;
646-
Mantissa <<= 1;
647-
} while ((Mantissa & 0x200) == 0);
648-
649-
Mantissa &= 0x1FF;
650-
}
651-
else // The value is zero
652-
{
653-
Exponent = (uint32_t)-112;
654-
}
655-
656-
Result[0] = ((Exponent + 112) << 23) | (Mantissa << 14);
657-
658-
// Y Channel
659-
Mantissa = pSource->ym;
660-
661-
if (Mantissa != 0) // The value is denormalized
662-
{
663-
// Normalize the value in the resulting float
664-
Exponent = 1;
608+
union { float f; int32_t i; } fi;
609+
fi.i = 0x33800000 + (pSource->e << 23);
610+
float Scale = fi.f;
665611

666-
do
667-
{
668-
Exponent--;
669-
Mantissa <<= 1;
670-
} while ((Mantissa & 0x200) == 0);
671-
672-
Mantissa &= 0x1FF;
673-
}
674-
else // The value is zero
675-
{
676-
Exponent = (uint32_t)-112;
677-
}
678-
679-
Result[1] = ((Exponent + 112) << 23) | (Mantissa << 14);
680-
681-
// Z Channel
682-
Mantissa = pSource->zm;
683-
684-
if (Mantissa != 0) // The value is denormalized
685-
{
686-
// Normalize the value in the resulting float
687-
Exponent = 1;
688-
689-
do
690-
{
691-
Exponent--;
692-
Mantissa <<= 1;
693-
} while ((Mantissa & 0x200) == 0);
694-
695-
Mantissa &= 0x1FF;
696-
}
697-
else // The value is zero
698-
{
699-
Exponent = (uint32_t)-112;
700-
}
701-
702-
Result[2] = ((Exponent + 112) << 23) | (Mantissa << 14);
703-
}
704-
705-
return XMLoadFloat3A( reinterpret_cast<const XMFLOAT3A*>(&Result) );
612+
XMVECTORF32 v = {
613+
Scale * float( pSource->xm ),
614+
Scale * float( pSource->ym ),
615+
Scale * float( pSource->zm ),
616+
1.0f };
617+
return v;
706618
}
707619

708620
//------------------------------------------------------------------------------
@@ -1012,6 +924,31 @@ inline XMVECTOR XM_CALLCONV PackedVector::XMLoadUDecN4
1012924
#endif // _XM_VMX128_INTRINSICS_
1013925
}
1014926

927+
928+
//------------------------------------------------------------------------------
929+
_Use_decl_annotations_
930+
inline XMVECTOR XM_CALLCONV PackedVector::XMLoadUDecN4_XR
931+
(
932+
const XMUDECN4* pSource
933+
)
934+
{
935+
assert(pSource);
936+
937+
int32_t ElementX = pSource->v & 0x3FF;
938+
int32_t ElementY = (pSource->v >> 10) & 0x3FF;
939+
int32_t ElementZ = (pSource->v >> 20) & 0x3FF;
940+
941+
XMVECTORF32 vResult = {
942+
(float)(ElementX - 0x180) / 510.0f,
943+
(float)(ElementY - 0x180) / 510.0f,
944+
(float)(ElementZ - 0x180) / 510.0f,
945+
(float)(pSource->v >> 30) / 3.0f
946+
};
947+
948+
return vResult.v;
949+
}
950+
951+
1015952
//------------------------------------------------------------------------------
1016953
_Use_decl_annotations_
1017954
inline XMVECTOR XM_CALLCONV PackedVector::XMLoadUDec4
@@ -1814,77 +1751,33 @@ inline void XM_CALLCONV PackedVector::XMStoreFloat3SE
18141751
{
18151752
assert(pDestination);
18161753

1817-
__declspec(align(16)) uint32_t IValue[4];
1818-
XMStoreFloat3A( reinterpret_cast<XMFLOAT3A*>(&IValue), V );
1754+
XMFLOAT3A tmp;
1755+
XMStoreFloat3A( &tmp, V );
18191756

1820-
uint32_t Exp[3];
1821-
uint32_t Frac[3];
1757+
static const float maxf9 = float(0x1FF << 7);
1758+
static const float minf9 = float(1.f / (1 << 16));
18221759

1823-
// X, Y, Z Channels (5-bit exponent, 9-bit mantissa)
1824-
for(uint32_t j=0; j < 3; ++j)
1825-
{
1826-
uint32_t Sign = IValue[j] & 0x80000000;
1827-
uint32_t I = IValue[j] & 0x7FFFFFFF;
1760+
float x = (tmp.x >= 0.f) ? ( (tmp.x > maxf9) ? maxf9 : tmp.x ) : 0.f;
1761+
float y = (tmp.y >= 0.f) ? ( (tmp.y > maxf9) ? maxf9 : tmp.y ) : 0.f;
1762+
float z = (tmp.z >= 0.f) ? ( (tmp.z > maxf9) ? maxf9 : tmp.z ) : 0.f;
18281763

1829-
if ((I & 0x7F800000) == 0x7F800000)
1830-
{
1831-
// INF or NAN
1832-
Exp[j] = 0x1f;
1833-
if (( I & 0x7FFFFF ) != 0)
1834-
{
1835-
Frac[j] = ((I>>14)|(I>>5)|(I))&0x1ff;
1836-
}
1837-
else if ( Sign )
1838-
{
1839-
// -INF is clamped to 0 since 3SE is positive only
1840-
Exp[j] = Frac[j] = 0;
1841-
}
1842-
}
1843-
else if ( Sign )
1844-
{
1845-
// 3SE is positive only, so clamp to zero
1846-
Exp[j] = Frac[j] = 0;
1847-
}
1848-
else if (I > 0x477FC000U)
1849-
{
1850-
// The number is too large, set to max
1851-
Exp[j] = 0x1e;
1852-
Frac[j] = 0x1ff;
1853-
}
1854-
else
1855-
{
1856-
if (I < 0x38800000U)
1857-
{
1858-
// The number is too small to be represented as a normalized float9
1859-
// Convert it to a denormalized value.
1860-
uint32_t Shift = 113U - (I >> 23U);
1861-
I = (0x800000U | (I & 0x7FFFFFU)) >> Shift;
1862-
}
1863-
else
1864-
{
1865-
// Rebias the exponent to represent the value as a normalized float9
1866-
I += 0xC8000000U;
1867-
}
1868-
1869-
uint32_t T = ((I + 0x1FFFU + ((I >> 14U) & 1U)) >> 14U)&0x3fffU;
1764+
const float max_xy = (x > y) ? x : y;
1765+
const float max_xyz = (max_xy > z) ? max_xy : z;
18701766

1871-
Exp[j] = (T & 0x3E00) >> 9;
1872-
Frac[j] = T & 0x1ff;
1873-
}
1874-
}
1767+
const float maxColor = (max_xyz > minf9) ? max_xyz : minf9;
1768+
1769+
union { float f; int32_t i; } fi;
1770+
fi.f = maxColor;
1771+
fi.i &= 0xFF800000; // cut off fraction
18751772

1876-
// Adjust to a shared exponent
1877-
uint32_t T = XMMax( Exp[0], XMMax( Exp[1], Exp[2] ) );
1773+
pDestination->e = (fi.i - 0x37800000) >> 23;
18781774

1879-
Frac[0] = Frac[0] >> (T - Exp[0]);
1880-
Frac[1] = Frac[1] >> (T - Exp[1]);
1881-
Frac[2] = Frac[2] >> (T - Exp[2]);
1775+
fi.i = 0x83000000 - fi.i;
1776+
float ScaleR = fi.f;
18821777

1883-
// Store packed into memory
1884-
pDestination->xm = Frac[0];
1885-
pDestination->ym = Frac[1];
1886-
pDestination->zm = Frac[2];
1887-
pDestination->e = T;
1778+
pDestination->xm = static_cast<uint32_t>( Internal::round_to_nearest(x * ScaleR) );
1779+
pDestination->ym = static_cast<uint32_t>( Internal::round_to_nearest(y * ScaleR) );
1780+
pDestination->zm = static_cast<uint32_t>( Internal::round_to_nearest(z * ScaleR) );
18881781
}
18891782

18901783
//------------------------------------------------------------------------------
@@ -2260,6 +2153,32 @@ inline void XM_CALLCONV PackedVector::XMStoreUDecN4
22602153
#endif // _XM_VMX128_INTRINSICS_
22612154
}
22622155

2156+
//------------------------------------------------------------------------------
2157+
_Use_decl_annotations_
2158+
inline void XM_CALLCONV PackedVector::XMStoreUDecN4_XR
2159+
(
2160+
XMUDECN4* pDestination,
2161+
FXMVECTOR V
2162+
)
2163+
{
2164+
assert(pDestination);
2165+
2166+
static const XMVECTORF32 Scale = { 510.0f, 510.0f, 510.0f, 3.0f };
2167+
static const XMVECTORF32 Bias = { 384.0f, 384.0f, 384.0f, 0.0f };
2168+
static const XMVECTORF32 C = { 1023.f, 1023.f, 1023.f, 3.f };
2169+
2170+
XMVECTOR N = XMVectorMultiplyAdd( V, Scale, Bias );
2171+
N = XMVectorClamp( N, g_XMZero, C );
2172+
2173+
XMFLOAT4A tmp;
2174+
XMStoreFloat4A(&tmp, N );
2175+
2176+
pDestination->v = ((uint32_t)tmp.w << 30)
2177+
| (((uint32_t)tmp.z & 0x3FF) << 20)
2178+
| (((uint32_t)tmp.y & 0x3FF) << 10)
2179+
| (((uint32_t)tmp.x & 0x3FF));
2180+
}
2181+
22632182
//------------------------------------------------------------------------------
22642183
_Use_decl_annotations_
22652184
inline void XM_CALLCONV PackedVector::XMStoreUDec4

0 commit comments

Comments
 (0)