[SM6.9] Lower vector any/all calls to vector op

[V-FEXrt]

Fixes #7687

Implement vector reduction lowing for calls to any and all

[github-actions]

✅ With the latest revision this PR passed the Python code formatter.

[damyanp]

LGTM, but should aim for approval from someone more experienced than me.

[hekota]

Few comments, plus the scalarization using just and needs to be updated.

[hekota]

Suggested change

	bool NeedsConvertToInt = EltTy->isFloatingPointTy();
	for (unsigned I = 0; I < Ty->getVectorNumElements(); I++) {
	Value *Elt = Builder.CreateExtractElement(Arg, I);
	if (NeedsConvertToInt)
	Elt = GenerateCmpNEZero(Elt, Builder);
	bool NeedsConvertToBool = EltTy->isFloatingPointTy();
	for (unsigned I = 0; I < Ty->getVectorNumElements(); I++) {
	Value *Elt = Builder.CreateExtractElement(Arg, I);
	if (NeedsConvertToBool)
	Elt = GenerateCmpNEZero(Elt, Builder);

Please rename, this is not really converting float value to integer ;)

[V-FEXrt]

ack for the future! This actually goes away entirely with the latest commit!

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[hekota]

The operations are declared as "Bitwise AND/OR reduction of the vector returning a scalar", but in reality, we only ever call it with a bool vector, right? I am not sure if that's a problem or not.

[V-FEXrt]

The operations are declared as "Bitwise AND/OR reduction of the vector returning a scalar", but in reality, we only ever call it with a bool vector, right? I am not sure if that's a problem or not.

With the currently lowering, yes. It's only every called with an i1 vec. But there isn't a reason to add that restriction. For example, it might be nice to add an intrinsic to hlsl for "unary vector and" and "unary vector or" in the future but restricting the opcode would prevent that

[hekota]

LGTM!