public void Visit(TernaryOp ins)
{
Vector4f a = ctx.ReadValue (ins.Source1);
Vector4f b = ctx.ReadValue (ins.Source2);
Vector4f c = ctx.ReadValue (ins.Source3);
Vector4f res = new Vector4f ();
switch (ins.Operation) {
case TernaryOpKind.Cmp: {// a >= 0 ? b : c
//m = a < [0,0,0,0]
Vector4f mask = a.CompareLessThan (new Vector4f ()); //we change to a <b
//res = (m & C) | (~m & B) -- this could be replaced by a blendps
res = (mask & c) | mask.AndNot (b);
break;
}
case TernaryOpKind.Mad: {// a * b + c
res = (a * b) + c;
break;
}
case TernaryOpKind.SinCos: { // sin(a) , cos(a)
//XXX maybe we should use the macro expansion provided in the driver docs for HLSL
res.X = (float)Math.Cos (a.X);
res.Y = (float)Math.Sin (a.X);
break;
}
case TernaryOpKind.Lrp: //c + a * (b - c)
//res = c + (a * (b - c));
res = a * b + (Vector4f.One - a) * c;
break;
case TernaryOpKind.Dp2Add: //res = a.r * b.r + a.g * b.g + c.swizzle
res = a * b;
//XX we could use Hadd here
res = res + res.Shuffle (ShuffleSel.XFromY) + c;
res = res.Shuffle (ShuffleSel.ExpandX);
break;
default:
throw new Exception ("Cant handle " + ins.Operation);
}
if (Tracing.Enabled) Console.WriteLine ("{0} {1} {2} {3} => {4}/{5}/{6} == {7}", ins.Source1, ins.Operation, ins.Source2, ins.Source3, a, b, c, res);
ctx.StoreValue (ins.Dest, res);
}
public void Visit(TernaryOp ins)
{
VisitSrcReg (ins.Source1);
VisitSrcReg (ins.Source2);
VisitSrcReg (ins.Source3);
VisitDestReg (ins.Dest);
}
public void Visit(TernaryOp ins)
{
ctx.EmitTernary (ins);
}
public void Visit(TernaryOp ins)
{
}
internal void EmitTernary(TernaryOp ins)
{
//FIXME it might be an issue if arguments are not of type Vector4f
switch (ins.Operation) {
case TernaryOpKind.Cmp: {
var mask = DeclareTmp (typeof (Vector4f));
//mask = 0
ilgen.Emit (OpCodes.Ldloca, mask);
ilgen.Emit (OpCodes.Initobj, typeof (Vector4f));
//a < tmp
LoadValue (ins.Source1);
ilgen.Emit (OpCodes.Ldloc, mask);
ilgen.Emit (OpCodes.Call, typeof (VectorOperations).GetMethod ("CompareLessThan", new Type[] { typeof (Vector4f), typeof (Vector4f)}));
ilgen.Emit (OpCodes.Stloc, mask);
//mask & c
ilgen.Emit (OpCodes.Ldloc, mask);
LoadValue (ins.Source3);
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_BitwiseAnd"));
//mask.AndNot (b);
ilgen.Emit (OpCodes.Ldloc, mask);
LoadValue (ins.Source2);
ilgen.Emit (OpCodes.Call, typeof (VectorOperations).GetMethod ("AndNot", new Type[] { typeof (Vector4f), typeof (Vector4f)}));
//res = (mask & c) | mask.AndNot (b);
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_BitwiseOr"));
StoreValue (ins.Dest);
break;
}
case TernaryOpKind.Mad: //a * b + c
LoadValue (ins.Source1);
LoadValue (ins.Source2);
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_Multiply"));
LoadValue (ins.Source3);
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_Addition"));
StoreValue (ins.Dest);
break;
case TernaryOpKind.SinCos: //.x = cos (a.X) .y = cos (a.X)
//XXX maybe we should use the macro expansion provided in the driver docs for HLSL
switch (ins.Dest.WriteMask) {
case 1: //R cos
ApplyMathScalarUnaryToElement (ins.Dest, ins.Source1, "Cos", "X");
break;
case 2: //G Sin
ApplyMathScalarUnaryToElement (ins.Dest, ins.Source1, "Sin", "Y");
break;
case 3: //RG
ApplyMathScalarUnaryToElement (ins.Dest, ins.Source1, "Cos", "X");
ApplyMathScalarUnaryToElement (ins.Dest, ins.Source1, "Sin", "Y");
break;
default:
throw new Exception ("Invalid writeMask for sincos " + ins.Dest.WriteMask.ToString ("X"));
}
break;
case TernaryOpKind.Lrp: //a * b + (1 - a) * c;
//a * b
LoadValue (ins.Source1);
LoadValue (ins.Source2);
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_Multiply"));
//1 - a
EmitTempVector4f (1);
LoadValue (ins.Source1);
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_Subtraction"));
// (1 - a) * c
LoadValue (ins.Source3);
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_Multiply"));
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_Addition"));
StoreValue (ins.Dest);
break;
case TernaryOpKind.Dp2Add: {//res = a.r * b.r + a.g * b.g + c.swizzle
LoadValue (ins.Source1);
LoadValue (ins.Source2);
LoadValue (ins.Source3);
ilgen.Emit (OpCodes.Call, typeof (SimdExtras).GetMethod ("Dp2Add"));
StoreValue (ins.Dest);
break;
//a * b
LoadValue (ins.Source1);
LoadValue (ins.Source2);
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_Multiply"));
ilgen.Emit (OpCodes.Dup);
//FIXME we could use HorizontalAdd here
EmitShuffle (ShuffleSel.XFromY); //[x,y,_,_] [y,y,_,_]
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_Addition")); //[x + y,_,_,_]
//+c.swizzle
LoadValue (ins.Source3);
ilgen.Emit (OpCodes.Call, typeof (Vector4f).GetMethod ("op_Addition"));
EmitShuffle (ShuffleSel.ExpandX);
StoreValue (ins.Dest);
break;
}
default:
throw new Exception ("can't handle ternop " + ins.Operation);
}
}
