[llvm] [DirectX][DXIL] Distinguish return type for overload type resolution. (PR #85646)

[S. Bharadwaj Yadavalli via llvm-commits]

================
@@ -255,6 +255,9 @@ class DXILOpMapping<int opCode, DXILOpClass opClass,
 }
 
 // Concrete definition of DXIL Operation mapping to corresponding LLVM intrinsic
+def IsInf : DXILOpMapping<9, isSpecialFloat, int_dx_isinf,
+                         "Determines if the specified value is infinite.",
+                         [llvm_i1_ty, llvm_halforfloat_ty]>;
----------------
bharadwajy wrote:

> Is it ok that these are scalar types? isInf returns a bool vector if the input is a float or half vector. Will we eventually have to expand type here to indicate we may be a scalar or vector?

Per my reading of [`hctdb.py`](https://github.com/microsoft/DirectXShaderCompiler/blob/6f4f758e9b3be1991956f99600982755596e29dc/utils/hct/hctdb.py#L1341), the DXIL Op `Isinf` is a scalar operation with `half` and `float` as scalar overloads. Accordingly, this record fully specifies the op.

That said, I am working with a principle and design wherein the properties of HLSL intrinsic to which this DXIL Op maps (upon elementwise expansion) will be represented as TableGen `Attr` records.

https://github.com/llvm/llvm-project/pull/85646