public static void Dup_Gp(ArmEmitterContext context)
{
OpCodeSimdIns op = (OpCodeSimdIns)context.CurrOp;
Operand n = GetIntOrZR(context, op.Rn);
if (Optimizations.UseSse2)
{
switch (op.Size)
{
case 0: n = context.ZeroExtend8(n.Type, n); n = context.Multiply(n, Const(n.Type, 0x01010101)); break;
case 1: n = context.ZeroExtend16(n.Type, n); n = context.Multiply(n, Const(n.Type, 0x00010001)); break;
case 2: n = context.ZeroExtend32(n.Type, n); break;
}
Operand res = context.VectorInsert(context.VectorZero(), n, 0);
if (op.Size < 3)
{
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(0xf0));
}
else
{
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(0));
}
}
else
{
res = context.AddIntrinsic(Intrinsic.X86Movlhps, res, res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
Operand res = context.VectorZero();
int elems = op.GetBytesCount() >> op.Size;
for (int index = 0; index < elems; index++)
{
res = EmitVectorInsert(context, res, n, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
}
public static void Dup_V(ArmEmitterContext context)
{
OpCodeSimdIns op = (OpCodeSimdIns)context.CurrOp;
if (Optimizations.UseSse2)
{
Operand res = GetVec(op.Rn);
if (op.Size == 0)
{
if (op.DstIndex != 0)
{
res = context.AddIntrinsic(Intrinsic.X86Psrldq, res, Const(op.DstIndex));
}
res = context.AddIntrinsic(Intrinsic.X86Punpcklbw, res, res);
res = context.AddIntrinsic(Intrinsic.X86Punpcklwd, res, res);
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(0));
}
else if (op.Size == 1)
{
if (op.DstIndex != 0)
{
res = context.AddIntrinsic(Intrinsic.X86Psrldq, res, Const(op.DstIndex * 2));
}
res = context.AddIntrinsic(Intrinsic.X86Punpcklwd, res, res);
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(0));
}
else if (op.Size == 2)
{
int mask = op.DstIndex * 0b01010101;
res = context.AddIntrinsic(Intrinsic.X86Shufps, res, res, Const(mask));
}
else if (op.DstIndex == 0 && op.RegisterSize != RegisterSize.Simd64)
{
res = context.AddIntrinsic(Intrinsic.X86Movlhps, res, res);
}
else if (op.DstIndex == 1)
{
res = context.AddIntrinsic(Intrinsic.X86Movhlps, res, res);
}
if (op.RegisterSize == RegisterSize.Simd64)
{
res = context.VectorZeroUpper64(res);
}
context.Copy(GetVec(op.Rd), res);
}
else
{
Operand ne = EmitVectorExtractZx(context, op.Rn, op.DstIndex, op.Size);
Operand res = context.VectorZero();
int elems = op.GetBytesCount() >> op.Size;
for (int index = 0; index < elems; index++)
{
res = EmitVectorInsert(context, res, ne, index, op.Size);
}
context.Copy(GetVec(op.Rd), res);
}
}
