public static void Vneg_S(ArmEmitterContext context)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
if (Optimizations.UseSse2)
{
EmitScalarUnaryOpSimd32(context, (m) =>
{
if ((op.Size & 1) == 0)
{
Operand mask = X86GetScalar(context, -0f);
return(context.AddIntrinsic(Intrinsic.X86Xorps, mask, m));
}
else
{
Operand mask = X86GetScalar(context, -0d);
return(context.AddIntrinsic(Intrinsic.X86Xorpd, mask, m));
}
});
}
else
{
EmitScalarUnaryOpF32(context, (op1) => context.Negate(op1));
}
}
public static void Vcvt_FD(ArmEmitterContext context)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
int vm = op.Vm;
int vd;
if (op.Size == 3)
{
vd = FlipVdBits(op.Vd, false);
// Double to single.
Operand fp = ExtractScalar(context, OperandType.FP64, vm);
Operand res = context.ConvertToFP(OperandType.FP32, fp);
InsertScalar(context, vd, res);
}
else
{
vd = FlipVdBits(op.Vd, true);
// Single to double.
Operand fp = ExtractScalar(context, OperandType.FP32, vm);
Operand res = context.ConvertToFP(OperandType.FP64, fp);
InsertScalar(context, vd, res);
}
}
private static void EmitVcmpOrVcmpe(ArmEmitterContext context, bool signalNaNs)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
bool cmpWithZero = (op.Opc & 2) != 0;
{
int fSize = op.Size & 1;
OperandType type = fSize != 0 ? OperandType.FP64 : OperandType.FP32;
Operand ne = ExtractScalar(context, type, op.Vd);
Operand me;
if (cmpWithZero)
{
me = fSize == 0 ? ConstF(0f) : ConstF(0d);
}
else
{
me = ExtractScalar(context, type, op.Vm);
}
Delegate dlg = fSize != 0
? (Delegate) new _S32_F64_F64_Bool(SoftFloat64.FPCompare)
: (Delegate) new _S32_F32_F32_Bool(SoftFloat32.FPCompare);
Operand nzcv = context.Call(dlg, ne, me, Const(signalNaNs));
EmitSetFPSCRFlags(context, nzcv);
}
}
public static void EmitScalarUnaryOpF32(ArmEmitterContext context, Intrinsic inst32, Intrinsic inst64)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
Intrinsic inst = (op.Size & 1) != 0 ? inst64 : inst32;
EmitScalarUnaryOpSimd32(context, (m) => (inst == 0) ? m : context.AddIntrinsic(inst, m));
}
public static void EmitScalarUnaryOpF32(ArmEmitterContext context, Func1I emit)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
OperandType type = (op.Size & 1) != 0 ? OperandType.FP64 : OperandType.FP32;
Operand m = ExtractScalar(context, type, op.Vm);
InsertScalar(context, op.Vd, emit(m));
}
// VRINTX (floating-point).
public static void Vrintx_S(ArmEmitterContext context)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
bool doubleSize = (op.Size & 1) == 1;
string methodName = doubleSize ? nameof(SoftFallback.Round) : nameof(SoftFallback.RoundF);
EmitScalarUnaryOpF32(context, (op1) =>
{
MethodInfo info = typeof(SoftFallback).GetMethod(methodName);
return(context.Call(info, op1));
});
}
public static void Vabs_S(ArmEmitterContext context)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
if (Optimizations.FastFP && Optimizations.UseSse2)
{
EmitScalarUnaryOpSimd32(context, (m) =>
{
return(EmitFloatAbs(context, m, (op.Size & 1) == 0, false));
});
}
else
{
EmitScalarUnaryOpF32(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Abs), op1));
}
}
// VRINTZ (floating-point).
public static void Vrint_Z(ArmEmitterContext context)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
if (Optimizations.UseSse2)
{
EmitScalarUnaryOpSimd32(context, (m) =>
{
Intrinsic inst = (op.Size & 1) == 0 ? Intrinsic.X86Roundss : Intrinsic.X86Roundsd;
return(context.AddIntrinsic(inst, m, Const(X86GetRoundControl(FPRoundingMode.TowardsZero))));
});
}
else
{
EmitScalarUnaryOpF32(context, (op1) => EmitUnaryMathCall(context, nameof(Math.Truncate), op1));
}
}
// VRINTA/M/N/P (floating-point).
public static void Vrint_RM(ArmEmitterContext context)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
OperandType floatSize = op.RegisterSize == RegisterSize.Int64 ? OperandType.FP64 : OperandType.FP32;
int rm = op.Opc2 & 3;
if (Optimizations.UseSse2 && rm != 0b00)
{
EmitScalarUnaryOpSimd32(context, (m) =>
{
Intrinsic inst = (op.Size & 1) == 0 ? Intrinsic.X86Roundss : Intrinsic.X86Roundsd;
FPRoundingMode roundMode = RMToRoundMode(rm);
return(context.AddIntrinsic(inst, m, Const(X86GetRoundControl(roundMode))));
});
}
else
{
Operand toConvert = ExtractScalar(context, floatSize, op.Vm);
switch (rm)
{
case 0b00: // Away
toConvert = EmitRoundMathCall(context, MidpointRounding.AwayFromZero, toConvert);
break;
case 0b01: // Nearest
toConvert = EmitRoundMathCall(context, MidpointRounding.ToEven, toConvert);
break;
case 0b10: // Towards positive infinity
toConvert = EmitUnaryMathCall(context, nameof(Math.Ceiling), toConvert);
break;
case 0b11: // Towards negative infinity
toConvert = EmitUnaryMathCall(context, nameof(Math.Floor), toConvert);
break;
}
InsertScalar(context, op.Vd, toConvert);
}
}
public static void EmitScalarUnaryOpSimd32(ArmEmitterContext context, Func1I scalarFunc)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
bool doubleSize = (op.Size & 1) != 0;
int shift = doubleSize ? 1 : 2;
Operand m = GetVecA32(op.Vm >> shift);
Operand d = GetVecA32(op.Vd >> shift);
m = EmitSwapScalar(context, m, op.Vm, doubleSize);
Operand res = scalarFunc(m);
// Insert scalar into vector.
res = EmitScalarInsert(context, d, res, op.Vd, doubleSize);
context.Copy(d, res);
}
private static void EmitSse41ConvertInt32(ArmEmitterContext context, FPRoundingMode roundMode, bool signed)
{
// A port of the similar round function in InstEmitSimdCvt.
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
bool doubleSize = (op.Size & 1) != 0;
int shift = doubleSize ? 1 : 2;
Operand n = GetVecA32(op.Vm >> shift);
n = EmitSwapScalar(context, n, op.Vm, doubleSize);
if (!doubleSize)
{
Operand nRes = context.AddIntrinsic(Intrinsic.X86Cmpss, n, n, Const((int)CmpCondition.OrderedQ));
nRes = context.AddIntrinsic(Intrinsic.X86Pand, nRes, n);
nRes = context.AddIntrinsic(Intrinsic.X86Roundss, nRes, Const(X86GetRoundControl(roundMode)));
Operand zero = context.VectorZero();
Operand nCmp;
Operand nIntOrLong2 = null;
if (!signed)
{
nCmp = context.AddIntrinsic(Intrinsic.X86Cmpss, nRes, zero, Const((int)CmpCondition.NotLessThanOrEqual));
nRes = context.AddIntrinsic(Intrinsic.X86Pand, nRes, nCmp);
}
int fpMaxVal = 0x4F000000; // 2.14748365E9f (2147483648)
Operand fpMaxValMask = X86GetScalar(context, fpMaxVal);
Operand nIntOrLong = context.AddIntrinsicInt(Intrinsic.X86Cvtss2si, nRes);
if (!signed)
{
nRes = context.AddIntrinsic(Intrinsic.X86Subss, nRes, fpMaxValMask);
nCmp = context.AddIntrinsic(Intrinsic.X86Cmpss, nRes, zero, Const((int)CmpCondition.NotLessThanOrEqual));
nRes = context.AddIntrinsic(Intrinsic.X86Pand, nRes, nCmp);
nIntOrLong2 = context.AddIntrinsicInt(Intrinsic.X86Cvtss2si, nRes);
}
nRes = context.AddIntrinsic(Intrinsic.X86Cmpss, nRes, fpMaxValMask, Const((int)CmpCondition.NotLessThan));
Operand nInt = context.AddIntrinsicInt(Intrinsic.X86Cvtsi2si, nRes);
Operand dRes;
if (signed)
{
dRes = context.BitwiseExclusiveOr(nIntOrLong, nInt);
}
else
{
dRes = context.BitwiseExclusiveOr(nIntOrLong2, nInt);
dRes = context.Add(dRes, nIntOrLong);
}
InsertScalar(context, op.Vd, dRes);
}
else
{
Operand nRes = context.AddIntrinsic(Intrinsic.X86Cmpsd, n, n, Const((int)CmpCondition.OrderedQ));
nRes = context.AddIntrinsic(Intrinsic.X86Pand, nRes, n);
nRes = context.AddIntrinsic(Intrinsic.X86Roundsd, nRes, Const(X86GetRoundControl(roundMode)));
Operand zero = context.VectorZero();
Operand nCmp;
Operand nIntOrLong2 = null;
if (!signed)
{
nCmp = context.AddIntrinsic(Intrinsic.X86Cmpsd, nRes, zero, Const((int)CmpCondition.NotLessThanOrEqual));
nRes = context.AddIntrinsic(Intrinsic.X86Pand, nRes, nCmp);
}
long fpMaxVal = 0x41E0000000000000L; // 2147483648.0000000d (2147483648)
Operand fpMaxValMask = X86GetScalar(context, fpMaxVal);
Operand nIntOrLong = context.AddIntrinsicInt(Intrinsic.X86Cvtsd2si, nRes);
if (!signed)
{
nRes = context.AddIntrinsic(Intrinsic.X86Subsd, nRes, fpMaxValMask);
nCmp = context.AddIntrinsic(Intrinsic.X86Cmpsd, nRes, zero, Const((int)CmpCondition.NotLessThanOrEqual));
nRes = context.AddIntrinsic(Intrinsic.X86Pand, nRes, nCmp);
nIntOrLong2 = context.AddIntrinsicInt(Intrinsic.X86Cvtsd2si, nRes);
}
nRes = context.AddIntrinsic(Intrinsic.X86Cmpsd, nRes, fpMaxValMask, Const((int)CmpCondition.NotLessThan));
Operand nLong = context.AddIntrinsicLong(Intrinsic.X86Cvtsi2si, nRes);
nLong = context.ConvertI64ToI32(nLong);
Operand dRes;
if (signed)
{
dRes = context.BitwiseExclusiveOr(nIntOrLong, nLong);
}
else
{
dRes = context.BitwiseExclusiveOr(nIntOrLong2, nLong);
dRes = context.Add(dRes, nIntOrLong);
}
InsertScalar(context, op.Vd, dRes);
}
}
private static void EmitVcmpOrVcmpe(ArmEmitterContext context, bool signalNaNs)
{
OpCode32SimdS op = (OpCode32SimdS)context.CurrOp;
bool cmpWithZero = (op.Opc & 2) != 0;
int sizeF = op.Size & 1;
if (Optimizations.FastFP && (signalNaNs ? Optimizations.UseAvx : Optimizations.UseSse2))
{
CmpCondition cmpOrdered = signalNaNs ? CmpCondition.OrderedS : CmpCondition.OrderedQ;
bool doubleSize = sizeF != 0;
int shift = doubleSize ? 1 : 2;
Operand m = GetVecA32(op.Vm >> shift);
Operand n = GetVecA32(op.Vd >> shift);
n = EmitSwapScalar(context, n, op.Vd, doubleSize);
m = cmpWithZero ? context.VectorZero() : EmitSwapScalar(context, m, op.Vm, doubleSize);
Operand lblNaN = Label();
Operand lblEnd = Label();
if (!doubleSize)
{
Operand ordMask = context.AddIntrinsic(Intrinsic.X86Cmpss, n, m, Const((int)cmpOrdered));
Operand isOrdered = context.AddIntrinsicInt(Intrinsic.X86Cvtsi2si, ordMask);
context.BranchIfFalse(lblNaN, isOrdered);
Operand cf = context.AddIntrinsicInt(Intrinsic.X86Comissge, n, m);
Operand zf = context.AddIntrinsicInt(Intrinsic.X86Comisseq, n, m);
Operand nf = context.AddIntrinsicInt(Intrinsic.X86Comisslt, n, m);
SetFpFlag(context, FPState.VFlag, Const(0));
SetFpFlag(context, FPState.CFlag, cf);
SetFpFlag(context, FPState.ZFlag, zf);
SetFpFlag(context, FPState.NFlag, nf);
}
else
{
Operand ordMask = context.AddIntrinsic(Intrinsic.X86Cmpsd, n, m, Const((int)cmpOrdered));
Operand isOrdered = context.AddIntrinsicLong(Intrinsic.X86Cvtsi2si, ordMask);
context.BranchIfFalse(lblNaN, isOrdered);
Operand cf = context.AddIntrinsicInt(Intrinsic.X86Comisdge, n, m);
Operand zf = context.AddIntrinsicInt(Intrinsic.X86Comisdeq, n, m);
Operand nf = context.AddIntrinsicInt(Intrinsic.X86Comisdlt, n, m);
SetFpFlag(context, FPState.VFlag, Const(0));
SetFpFlag(context, FPState.CFlag, cf);
SetFpFlag(context, FPState.ZFlag, zf);
SetFpFlag(context, FPState.NFlag, nf);
}
context.Branch(lblEnd);
context.MarkLabel(lblNaN);
SetFpFlag(context, FPState.VFlag, Const(1));
SetFpFlag(context, FPState.CFlag, Const(1));
SetFpFlag(context, FPState.ZFlag, Const(0));
SetFpFlag(context, FPState.NFlag, Const(0));
context.MarkLabel(lblEnd);
}
else
{
OperandType type = sizeF != 0 ? OperandType.FP64 : OperandType.FP32;
Operand ne = ExtractScalar(context, type, op.Vd);
Operand me;
if (cmpWithZero)
{
me = sizeF == 0 ? ConstF(0f) : ConstF(0d);
}
else
{
me = ExtractScalar(context, type, op.Vm);
}
MethodInfo info = sizeF != 0
? typeof(SoftFloat64).GetMethod(nameof(SoftFloat64.FPCompare))
: typeof(SoftFloat32).GetMethod(nameof(SoftFloat32.FPCompare));
Operand nzcv = context.Call(info, ne, me, Const(signalNaNs));
EmitSetFpscrNzcv(context, nzcv);
}
}
