307 lines
17 KiB
C#
307 lines
17 KiB
C#
using System.Diagnostics;
|
|
|
|
namespace Unity.Burst.Intrinsics
|
|
{
|
|
public unsafe static partial class X86
|
|
{
|
|
/// <summary>
|
|
/// F16C intrinsics
|
|
/// </summary>
|
|
public static class F16C
|
|
{
|
|
/// <summary>
|
|
/// Evaluates to true at compile time if F16C intrinsics are supported.
|
|
///
|
|
/// Burst ties F16C support to AVX2 support to simplify feature sets to support.
|
|
/// </summary>
|
|
public static bool IsF16CSupported { get { return Avx2.IsAvx2Supported; } }
|
|
|
|
/// <summary>
|
|
/// Converts a half (hiding in a ushort) to a float (hiding in a uint).
|
|
/// </summary>
|
|
/// <param name="h">The half to convert</param>
|
|
/// <returns>The float result</returns>
|
|
[DebuggerStepThrough]
|
|
private static uint HalfToFloat(ushort h)
|
|
{
|
|
var signed = (h & 0x8000u) != 0;
|
|
var exponent = (h >> 10) & 0x1fu;
|
|
var mantissa = h & 0x3ffu;
|
|
|
|
var result = signed ? 0x80000000u : 0u;
|
|
|
|
if (!(exponent == 0 && mantissa == 0))
|
|
{
|
|
// Denormal (converts to normalized)
|
|
if (exponent == 0)
|
|
{
|
|
// Adjust mantissa so it's normalized (and keep track of exponent adjustment)
|
|
exponent = -1;
|
|
do
|
|
{
|
|
exponent++;
|
|
mantissa <<= 1;
|
|
} while ((mantissa & 0x400) == 0);
|
|
|
|
result |= (uint)((127 - 15 - exponent) << 23);
|
|
|
|
// Have to re-mask the mantissa here because we've been shifting bits up.
|
|
result |= (mantissa & 0x3ff) << 13;
|
|
}
|
|
else
|
|
{
|
|
var isInfOrNan = exponent == 0x1f;
|
|
result |= (uint)(isInfOrNan ? 255 : (127 - 15 + exponent) << 23);
|
|
result |= mantissa << 13;
|
|
}
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Convert packed half-precision (16-bit) floating-point elements in a to packed single-precision (32-bit) floating-point elements, and store the results in dst.
|
|
/// </summary>
|
|
/// <remarks>
|
|
/// **** vcvtph2ps xmm, xmm
|
|
/// </remarks>
|
|
/// <param name="a">Vector a</param>
|
|
/// <returns>Vector</returns>
|
|
[DebuggerStepThrough]
|
|
public static v128 cvtph_ps(v128 a)
|
|
{
|
|
return new v128(HalfToFloat(a.UShort0), HalfToFloat(a.UShort1), HalfToFloat(a.UShort2), HalfToFloat(a.UShort3));
|
|
}
|
|
|
|
/// <summary>
|
|
/// Convert packed half-precision (16-bit) floating-point elements in a to packed single-precision (32-bit) floating-point elements, and store the results in dst.
|
|
/// </summary>
|
|
/// <remarks>
|
|
/// **** vcvtph2ps ymm, xmm
|
|
/// </remarks>
|
|
/// <param name="a">Vector a</param>
|
|
/// <returns>Vector</returns>
|
|
[DebuggerStepThrough]
|
|
public static v256 mm256_cvtph_ps(v128 a)
|
|
{
|
|
return new v256(HalfToFloat(a.UShort0), HalfToFloat(a.UShort1), HalfToFloat(a.UShort2), HalfToFloat(a.UShort3), HalfToFloat(a.UShort4), HalfToFloat(a.UShort5), HalfToFloat(a.UShort6), HalfToFloat(a.UShort7));
|
|
}
|
|
|
|
// Using ftp://ftp.fox-toolkit.org/pub/fasthalffloatconversion.pdf
|
|
private static readonly ushort[] BaseTable =
|
|
{
|
|
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
|
|
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
|
|
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
|
|
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
|
|
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
|
|
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
|
|
0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, 0x0100,
|
|
0x0200, 0x0400, 0x0800, 0x0C00, 0x1000, 0x1400, 0x1800, 0x1C00, 0x2000, 0x2400, 0x2800, 0x2C00, 0x3000, 0x3400, 0x3800, 0x3C00,
|
|
0x4000, 0x4400, 0x4800, 0x4C00, 0x5000, 0x5400, 0x5800, 0x5C00, 0x6000, 0x6400, 0x6800, 0x6C00, 0x7000, 0x7400, 0x7800, 0x7C00,
|
|
0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00,
|
|
0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00,
|
|
0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00,
|
|
0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00,
|
|
0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00,
|
|
0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00,
|
|
0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00, 0x7C00,
|
|
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
|
|
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
|
|
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
|
|
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
|
|
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
|
|
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000,
|
|
0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8000, 0x8001, 0x8002, 0x8004, 0x8008, 0x8010, 0x8020, 0x8040, 0x8080, 0x8100,
|
|
0x8200, 0x8400, 0x8800, 0x8C00, 0x9000, 0x9400, 0x9800, 0x9C00, 0xA000, 0xA400, 0xA800, 0xAC00, 0xB000, 0xB400, 0xB800, 0xBC00,
|
|
0xC000, 0xC400, 0xC800, 0xCC00, 0xD000, 0xD400, 0xD800, 0xDC00, 0xE000, 0xE400, 0xE800, 0xEC00, 0xF000, 0xF400, 0xF800, 0xFC00,
|
|
0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00,
|
|
0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00,
|
|
0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00,
|
|
0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00,
|
|
0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00,
|
|
0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00,
|
|
0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00, 0xFC00,
|
|
};
|
|
|
|
private static readonly sbyte[] ShiftTable =
|
|
{
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
|
|
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 13,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
|
|
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
|
|
24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 24, 13,
|
|
};
|
|
|
|
/// <summary>
|
|
/// Converts a float (hiding in a uint) to a half (hiding in a ushort).
|
|
/// </summary>
|
|
/// <param name="f">The float to convert</param>
|
|
/// <param name="rounding">Rounding mode</param>
|
|
/// <returns>The half result</returns>
|
|
[DebuggerStepThrough]
|
|
private static ushort FloatToHalf(uint f, int rounding)
|
|
{
|
|
var exponentAndSign = f >> 23;
|
|
var shift = ShiftTable[exponentAndSign];
|
|
|
|
var result = (uint)(BaseTable[exponentAndSign] + (ushort)((f & 0x7FFFFFu) >> shift));
|
|
|
|
// Check if the result is not Inf or NaN.
|
|
var isFinite = (result & 0x7C00) != 0x7C00;
|
|
var isNegative = (result & 0x8000) != 0;
|
|
|
|
if (rounding == (int)RoundingMode.FROUND_NINT_NOEXC)
|
|
{
|
|
var fWithRoundingBitPreserved = (f & 0x7FFFFFu) >> (shift - 1);
|
|
|
|
if ((exponentAndSign & 0xFF) == 102)
|
|
{
|
|
result++;
|
|
}
|
|
if (isFinite && ((fWithRoundingBitPreserved & 0x1u) != 0))
|
|
{
|
|
result++;
|
|
}
|
|
}
|
|
else if (rounding == (int)RoundingMode.FROUND_TRUNC_NOEXC)
|
|
{
|
|
if (!isFinite)
|
|
{
|
|
result -= (uint)(~shift & 0x1);
|
|
}
|
|
}
|
|
else if (rounding == (int)RoundingMode.FROUND_CEIL_NOEXC)
|
|
{
|
|
if (isFinite && !isNegative)
|
|
{
|
|
if ((exponentAndSign <= 102) && (exponentAndSign != 0))
|
|
{
|
|
result++;
|
|
}
|
|
else if ((f & 0x7FFFFFu & ((1u << shift) - 1u)) != 0)
|
|
{
|
|
result++;
|
|
}
|
|
}
|
|
|
|
var resultIsNegativeInf = (result == 0xFC00);
|
|
var inputIsNotNegativeInfOrNan = (exponentAndSign != 0x1FF);
|
|
|
|
if (resultIsNegativeInf && inputIsNotNegativeInfOrNan)
|
|
{
|
|
result--;
|
|
}
|
|
}
|
|
else if (rounding == (int)RoundingMode.FROUND_FLOOR_NOEXC)
|
|
{
|
|
if (isFinite && isNegative)
|
|
{
|
|
if ((exponentAndSign <= 358) && (exponentAndSign != 256))
|
|
{
|
|
result++;
|
|
}
|
|
else if ((f & 0x7FFFFFu & ((1u << shift) - 1u)) != 0)
|
|
{
|
|
result++;
|
|
}
|
|
}
|
|
|
|
var resultIsPositiveInf = (result == 0x7C00);
|
|
var inputIsNotPositiveInfOrNan = (exponentAndSign != 0xFF);
|
|
|
|
if (resultIsPositiveInf && inputIsNotPositiveInfOrNan)
|
|
{
|
|
result--;
|
|
}
|
|
}
|
|
|
|
return (ushort)result;
|
|
}
|
|
|
|
/// <summary>
|
|
/// Convert packed single-precision (32-bit) floating-point elements in a to packed half-precision (16-bit) floating-point elements, and store the results in dst.
|
|
///
|
|
/// Rounding is done according to the rounding parameter, which can be one of:
|
|
/// </summary>
|
|
/// <remarks>
|
|
/// **** cvtps2ph xmm, xmm, imm
|
|
/// </remarks>
|
|
/// <param name="a">Vector a</param>
|
|
/// <param name="rounding">Rounding mode</param>
|
|
/// <returns>Vector</returns>
|
|
[DebuggerStepThrough]
|
|
public static v128 cvtps_ph(v128 a, int rounding)
|
|
{
|
|
if (rounding == (int)RoundingMode.FROUND_RINT_NOEXC)
|
|
{
|
|
switch (MXCSR & MXCSRBits.RoundingControlMask)
|
|
{
|
|
case MXCSRBits.RoundToNearest:
|
|
rounding = (int)RoundingMode.FROUND_NINT_NOEXC;
|
|
break;
|
|
case MXCSRBits.RoundDown:
|
|
rounding = (int)RoundingMode.FROUND_FLOOR_NOEXC;
|
|
break;
|
|
case MXCSRBits.RoundUp:
|
|
rounding = (int)RoundingMode.FROUND_CEIL_NOEXC;
|
|
break;
|
|
case MXCSRBits.RoundTowardZero:
|
|
rounding = (int)RoundingMode.FROUND_TRUNC_NOEXC;
|
|
break;
|
|
}
|
|
}
|
|
|
|
return new v128(FloatToHalf(a.UInt0, rounding), FloatToHalf(a.UInt1, rounding), FloatToHalf(a.UInt2, rounding), FloatToHalf(a.UInt3, rounding), 0, 0, 0, 0);
|
|
}
|
|
|
|
/// <summary>
|
|
/// Convert packed single-precision (32-bit) floating-point elements in a to packed half-precision (16-bit) floating-point elements, and store the results in dst.
|
|
///
|
|
/// Rounding is done according to the rounding parameter, which can be one of:
|
|
/// </summary>
|
|
/// <remarks>
|
|
/// **** cvtps2ph xmm, ymm, imm
|
|
/// </remarks>
|
|
/// <param name="a">Vector a</param>
|
|
/// <param name="rounding">Rounding mode</param>
|
|
/// <returns>Vector</returns>
|
|
[DebuggerStepThrough]
|
|
public static v128 mm256_cvtps_ph(v256 a, int rounding)
|
|
{
|
|
if (rounding == (int)RoundingMode.FROUND_RINT_NOEXC)
|
|
{
|
|
switch (MXCSR & MXCSRBits.RoundingControlMask)
|
|
{
|
|
case MXCSRBits.RoundToNearest:
|
|
rounding = (int)RoundingMode.FROUND_NINT_NOEXC;
|
|
break;
|
|
case MXCSRBits.RoundDown:
|
|
rounding = (int)RoundingMode.FROUND_FLOOR_NOEXC;
|
|
break;
|
|
case MXCSRBits.RoundUp:
|
|
rounding = (int)RoundingMode.FROUND_CEIL_NOEXC;
|
|
break;
|
|
case MXCSRBits.RoundTowardZero:
|
|
rounding = (int)RoundingMode.FROUND_TRUNC_NOEXC;
|
|
break;
|
|
}
|
|
}
|
|
|
|
return new v128(FloatToHalf(a.UInt0, rounding), FloatToHalf(a.UInt1, rounding), FloatToHalf(a.UInt2, rounding), FloatToHalf(a.UInt3, rounding), FloatToHalf(a.UInt4, rounding), FloatToHalf(a.UInt5, rounding), FloatToHalf(a.UInt6, rounding), FloatToHalf(a.UInt7, rounding));
|
|
}
|
|
}
|
|
}
|
|
}
|