Fix a bug around generic functions using DXR RayQuery (#1309)

The DXR `RayQuery` type is our first generic type defined in the stdlib that is marked as a target intrinsic but does *not* map to a custom `IRType` case. Instead, a reference to `RayQuery<T>` is encoded in the IR as an ordinary `specialize` instruction.

Unfortunately, this doesn't play nice with the current specialization logic, which considered a `specialize` instruction to not represent a "fully specialized" instruction, which then inhibits specialization of generics/functions that use such an instruction.

The fix here was to add more nuanced logic to the check for "fully specialized"-ness, so that it considers a `specialize` to already be fully specialized when the generic it applies to represents a target intrinsic.

I also added a note that the whole notion of "fully specialized"-ness that we use isn't really the right thing for the specialization pass, and how we really ought to use a notion of what is or is not a "value."

This change doesn't include a test because the only way to trigger the issue is using the DXR 1.1 `RayQuery` type, and that type is not supported in current release versions of DXC.

0 files changed, 0 insertions, 0 deletions