Enumerator 

Legal  The operation is expected to be selectable directly by the target, and no transformation is necessary.

NarrowScalar  The operation should be synthesized from multiple instructions acting on a narrower scalar basetype.
For example a 64bit add might be implemented in terms of 32bit addwithcarry.

WidenScalar  The operation should be implemented in terms of a wider scalar basetype.
For example a <2 x s8> add could be implemented as a <2 x s32> add (ignoring the high bits).

FewerElements  The (vector) operation should be implemented by splitting it into subvectors where the operation is legal.
For example a <8 x s64> add might be implemented as 4 separate <2 x s64> adds.

MoreElements  The (vector) operation should be implemented by widening the input vector and ignoring the lanes added by doing so.
For example <2 x i8> is rarely legal, but you might perform an <8 x i8> and then only look at the first two results.

Lower  The operation itself must be expressed in terms of simpler actions on this target.
E.g. a SREM replaced by an SDIV and subtraction.

Libcall  The operation should be implemented as a call to some kind of runtime support library.
For example this usually happens on machines that don't support floatingpoint operations natively.

Custom  The target wants to do something special with this combination of operand and type.
A callback will be issued when it is needed.

Unsupported  This operation is completely unsupported on the target.
A programming error has occurred.

NotFound  Sentinel value for when no action was found in the specified table.

UseLegacyRules  Fall back onto the old rules.
TODO: Remove this once we've migrated
