Add intrinsics support (#121)
* Initial intrinsics support * Update tests to work with the new Vector128 type and intrinsics * Drop SSE4.1 requirement * Fix copy-paste mistake
This commit is contained in:
gdkchan
GitHub
parent
8e306b3ac1
commit
f9f111bc85
36 changed files with 1658 additions and 1111 deletions
|
|
@ -1,20 +1,11 @@
|
|||
using ChocolArm64.State;
|
||||
using ChocolArm64.Translation;
|
||||
using System;
|
||||
using System.Numerics;
|
||||
using System.Runtime.CompilerServices;
|
||||
|
||||
namespace ChocolArm64.Instruction
|
||||
{
|
||||
static class ASoftFallback
|
||||
{
|
||||
public static void EmitCall(AILEmitterCtx Context, string Name64, string Name128)
|
||||
{
|
||||
bool IsSimd64 = Context.CurrOp.RegisterSize == ARegisterSize.SIMD64;
|
||||
|
||||
Context.EmitCall(typeof(ASoftFallback), IsSimd64 ? Name64 : Name128);
|
||||
}
|
||||
|
||||
public static void EmitCall(AILEmitterCtx Context, string MthdName)
|
||||
{
|
||||
Context.EmitCall(typeof(ASoftFallback), MthdName);
|
||||
|
|
@ -160,78 +151,6 @@ namespace ChocolArm64.Instruction
|
|||
throw new ArgumentException(nameof(Size));
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public static int SatF32ToS32(float Value)
|
||||
{
|
||||
if (float.IsNaN(Value)) return 0;
|
||||
|
||||
return Value > int.MaxValue ? int.MaxValue :
|
||||
Value < int.MinValue ? int.MinValue : (int)Value;
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public static long SatF32ToS64(float Value)
|
||||
{
|
||||
if (float.IsNaN(Value)) return 0;
|
||||
|
||||
return Value > long.MaxValue ? long.MaxValue :
|
||||
Value < long.MinValue ? long.MinValue : (long)Value;
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public static uint SatF32ToU32(float Value)
|
||||
{
|
||||
if (float.IsNaN(Value)) return 0;
|
||||
|
||||
return Value > uint.MaxValue ? uint.MaxValue :
|
||||
Value < uint.MinValue ? uint.MinValue : (uint)Value;
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public static ulong SatF32ToU64(float Value)
|
||||
{
|
||||
if (float.IsNaN(Value)) return 0;
|
||||
|
||||
return Value > ulong.MaxValue ? ulong.MaxValue :
|
||||
Value < ulong.MinValue ? ulong.MinValue : (ulong)Value;
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public static int SatF64ToS32(double Value)
|
||||
{
|
||||
if (double.IsNaN(Value)) return 0;
|
||||
|
||||
return Value > int.MaxValue ? int.MaxValue :
|
||||
Value < int.MinValue ? int.MinValue : (int)Value;
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public static long SatF64ToS64(double Value)
|
||||
{
|
||||
if (double.IsNaN(Value)) return 0;
|
||||
|
||||
return Value > long.MaxValue ? long.MaxValue :
|
||||
Value < long.MinValue ? long.MinValue : (long)Value;
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public static uint SatF64ToU32(double Value)
|
||||
{
|
||||
if (double.IsNaN(Value)) return 0;
|
||||
|
||||
return Value > uint.MaxValue ? uint.MaxValue :
|
||||
Value < uint.MinValue ? uint.MinValue : (uint)Value;
|
||||
}
|
||||
|
||||
[MethodImpl(MethodImplOptions.AggressiveInlining)]
|
||||
public static ulong SatF64ToU64(double Value)
|
||||
{
|
||||
if (double.IsNaN(Value)) return 0;
|
||||
|
||||
return Value > ulong.MaxValue ? ulong.MaxValue :
|
||||
Value < ulong.MinValue ? ulong.MinValue : (ulong)Value;
|
||||
}
|
||||
|
||||
public static long SMulHi128(long LHS, long RHS)
|
||||
{
|
||||
return (long)(BigInteger.Multiply(LHS, RHS) >> 64);
|
||||
|
|
@ -241,239 +160,5 @@ namespace ChocolArm64.Instruction
|
|||
{
|
||||
return (ulong)(BigInteger.Multiply(LHS, RHS) >> 64);
|
||||
}
|
||||
|
||||
public static int CountSetBits8(byte Value)
|
||||
{
|
||||
return ((Value >> 0) & 1) + ((Value >> 1) & 1) +
|
||||
((Value >> 2) & 1) + ((Value >> 3) & 1) +
|
||||
((Value >> 4) & 1) + ((Value >> 5) & 1) +
|
||||
((Value >> 6) & 1) + (Value >> 7);
|
||||
}
|
||||
|
||||
public static float MaxF(float val1, float val2)
|
||||
{
|
||||
if (val1 == 0.0 && val2 == 0.0)
|
||||
{
|
||||
if (BitConverter.SingleToInt32Bits(val1) < 0 && BitConverter.SingleToInt32Bits(val2) < 0)
|
||||
return -0.0f;
|
||||
|
||||
return 0.0f;
|
||||
}
|
||||
|
||||
if (val1 > val2)
|
||||
return val1;
|
||||
|
||||
if (float.IsNaN(val1))
|
||||
return val1;
|
||||
|
||||
return val2;
|
||||
}
|
||||
|
||||
public static double Max(double val1, double val2)
|
||||
{
|
||||
if (val1 == 0.0 && val2 == 0.0)
|
||||
{
|
||||
if (BitConverter.DoubleToInt64Bits(val1) < 0 && BitConverter.DoubleToInt64Bits(val2) < 0)
|
||||
return -0.0;
|
||||
|
||||
return 0.0;
|
||||
}
|
||||
|
||||
if (val1 > val2)
|
||||
return val1;
|
||||
|
||||
if (double.IsNaN(val1))
|
||||
return val1;
|
||||
|
||||
return val2;
|
||||
}
|
||||
|
||||
public static float MinF(float val1, float val2)
|
||||
{
|
||||
if (val1 == 0.0 && val2 == 0.0)
|
||||
{
|
||||
if (BitConverter.SingleToInt32Bits(val1) < 0 || BitConverter.SingleToInt32Bits(val2) < 0)
|
||||
return -0.0f;
|
||||
|
||||
return 0.0f;
|
||||
}
|
||||
|
||||
if (val1 < val2)
|
||||
return val1;
|
||||
|
||||
if (float.IsNaN(val1))
|
||||
return val1;
|
||||
|
||||
return val2;
|
||||
}
|
||||
|
||||
public static double Min(double val1, double val2)
|
||||
{
|
||||
if (val1 == 0.0 && val2 == 0.0)
|
||||
{
|
||||
if (BitConverter.DoubleToInt64Bits(val1) < 0 || BitConverter.DoubleToInt64Bits(val2) < 0)
|
||||
return -0.0;
|
||||
|
||||
return 0.0;
|
||||
}
|
||||
|
||||
if (val1 < val2)
|
||||
return val1;
|
||||
|
||||
if (double.IsNaN(val1))
|
||||
return val1;
|
||||
|
||||
return val2;
|
||||
}
|
||||
|
||||
public static float RoundF(float Value, int Fpcr)
|
||||
{
|
||||
switch ((ARoundMode)((Fpcr >> 22) & 3))
|
||||
{
|
||||
case ARoundMode.ToNearest: return MathF.Round (Value);
|
||||
case ARoundMode.TowardsPlusInfinity: return MathF.Ceiling (Value);
|
||||
case ARoundMode.TowardsMinusInfinity: return MathF.Floor (Value);
|
||||
case ARoundMode.TowardsZero: return MathF.Truncate(Value);
|
||||
}
|
||||
|
||||
throw new InvalidOperationException();
|
||||
}
|
||||
|
||||
public static double Round(double Value, int Fpcr)
|
||||
{
|
||||
switch ((ARoundMode)((Fpcr >> 22) & 3))
|
||||
{
|
||||
case ARoundMode.ToNearest: return Math.Round (Value);
|
||||
case ARoundMode.TowardsPlusInfinity: return Math.Ceiling (Value);
|
||||
case ARoundMode.TowardsMinusInfinity: return Math.Floor (Value);
|
||||
case ARoundMode.TowardsZero: return Math.Truncate(Value);
|
||||
}
|
||||
|
||||
throw new InvalidOperationException();
|
||||
}
|
||||
|
||||
public static AVec Tbl1_V64(AVec Vector, AVec Tb0)
|
||||
{
|
||||
return Tbl(Vector, 8, Tb0);
|
||||
}
|
||||
|
||||
public static AVec Tbl1_V128(AVec Vector, AVec Tb0)
|
||||
{
|
||||
return Tbl(Vector, 16, Tb0);
|
||||
}
|
||||
|
||||
public static AVec Tbl2_V64(AVec Vector, AVec Tb0, AVec Tb1)
|
||||
{
|
||||
return Tbl(Vector, 8, Tb0, Tb1);
|
||||
}
|
||||
|
||||
public static AVec Tbl2_V128(AVec Vector, AVec Tb0, AVec Tb1)
|
||||
{
|
||||
return Tbl(Vector, 16, Tb0, Tb1);
|
||||
}
|
||||
|
||||
public static AVec Tbl3_V64(AVec Vector, AVec Tb0, AVec Tb1, AVec Tb2)
|
||||
{
|
||||
return Tbl(Vector, 8, Tb0, Tb1, Tb2);
|
||||
}
|
||||
|
||||
public static AVec Tbl3_V128(AVec Vector, AVec Tb0, AVec Tb1, AVec Tb2)
|
||||
{
|
||||
return Tbl(Vector, 16, Tb0, Tb1, Tb2);
|
||||
}
|
||||
|
||||
public static AVec Tbl4_V64(AVec Vector, AVec Tb0, AVec Tb1, AVec Tb2, AVec Tb3)
|
||||
{
|
||||
return Tbl(Vector, 8, Tb0, Tb1, Tb2, Tb3);
|
||||
}
|
||||
|
||||
public static AVec Tbl4_V128(AVec Vector, AVec Tb0, AVec Tb1, AVec Tb2, AVec Tb3)
|
||||
{
|
||||
return Tbl(Vector, 16, Tb0, Tb1, Tb2, Tb3);
|
||||
}
|
||||
|
||||
private static AVec Tbl(AVec Vector, int Bytes, params AVec[] Tb)
|
||||
{
|
||||
AVec Res = new AVec();
|
||||
|
||||
byte[] Table = new byte[Tb.Length * 16];
|
||||
|
||||
for (int Index = 0; Index < Tb.Length; Index++)
|
||||
for (int Index2 = 0; Index2 < 16; Index2++)
|
||||
{
|
||||
Table[Index * 16 + Index2] = (byte)VectorExtractIntZx(Tb[Index], Index2, 0);
|
||||
}
|
||||
|
||||
for (int Index = 0; Index < Bytes; Index++)
|
||||
{
|
||||
byte TblIdx = (byte)VectorExtractIntZx(Vector, Index, 0);
|
||||
|
||||
if (TblIdx < Table.Length)
|
||||
{
|
||||
Res = VectorInsertInt(Table[TblIdx], Res, Index, 0);
|
||||
}
|
||||
}
|
||||
|
||||
return Res;
|
||||
}
|
||||
|
||||
public static ulong VectorExtractIntZx(AVec Vector, int Index, int Size)
|
||||
{
|
||||
switch (Size)
|
||||
{
|
||||
case 0: return Vector.ExtractByte (Index);
|
||||
case 1: return Vector.ExtractUInt16(Index);
|
||||
case 2: return Vector.ExtractUInt32(Index);
|
||||
case 3: return Vector.ExtractUInt64(Index);
|
||||
}
|
||||
|
||||
throw new ArgumentOutOfRangeException(nameof(Size));
|
||||
}
|
||||
|
||||
public static long VectorExtractIntSx(AVec Vector, int Index, int Size)
|
||||
{
|
||||
switch (Size)
|
||||
{
|
||||
case 0: return (sbyte)Vector.ExtractByte (Index);
|
||||
case 1: return (short)Vector.ExtractUInt16(Index);
|
||||
case 2: return (int)Vector.ExtractUInt32(Index);
|
||||
case 3: return (long)Vector.ExtractUInt64(Index);
|
||||
}
|
||||
|
||||
throw new ArgumentOutOfRangeException(nameof(Size));
|
||||
}
|
||||
|
||||
public static float VectorExtractSingle(AVec Vector, int Index)
|
||||
{
|
||||
return Vector.ExtractSingle(Index);
|
||||
}
|
||||
|
||||
public static double VectorExtractDouble(AVec Vector, int Index)
|
||||
{
|
||||
return Vector.ExtractDouble(Index);
|
||||
}
|
||||
|
||||
public static AVec VectorInsertSingle(float Value, AVec Vector, int Index)
|
||||
{
|
||||
return AVec.InsertSingle(Vector, Index, Value);
|
||||
}
|
||||
|
||||
public static AVec VectorInsertDouble(double Value, AVec Vector, int Index)
|
||||
{
|
||||
return AVec.InsertDouble(Vector, Index, Value);
|
||||
}
|
||||
|
||||
public static AVec VectorInsertInt(ulong Value, AVec Vector, int Index, int Size)
|
||||
{
|
||||
switch (Size)
|
||||
{
|
||||
case 0: return AVec.InsertByte (Vector, Index, (byte)Value);
|
||||
case 1: return AVec.InsertUInt16(Vector, Index, (ushort)Value);
|
||||
case 2: return AVec.InsertUInt32(Vector, Index, (uint)Value);
|
||||
case 3: return AVec.InsertUInt64(Vector, Index, (ulong)Value);
|
||||
}
|
||||
|
||||
throw new ArgumentOutOfRangeException(nameof(Size));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue