Value types is the set of non-array object types. These types are
assignable, can be used as function parameters, and as return types
(along with void), and in other scenarios where the type must have usual
value-like behaviour with no weirdness (such as array decay).
`mlir::Type` is now used as the default C++ type for representing all
MLIR types, even when `ValueType` would be a valid constraint, unless
doing so provides a clear benefit in that specific context.
* PrimitiveType is now an interface.
* VoidKind is now VoidType.
* FloatKind is now FloatType.
* The integer kinds are now IntegerType with IntegerKind.
* DecayOp - array or function to pointer decay. Not shown in C.
* BitCastOp - bit-preserving conversion. Emitted as bit_cast if
necessary.
* ExtendOp - zero- or sign-extending conversion from any integer type to
any wider integer type.
* TruncateOp - truncating conversion from any integer type to any
narrower integer type.
* PtrResizeOp - resizes a pointer without changing its pointee type.
Currently there is only a single default target used for all C emission.
However, the emitted code is actually compiled using multiple
implementations (32-bit and 64-bit targets), some of which use a 32-bit
wide long. This causes integer literals to be emitted with wrong sizes
on those targets. This change forces all 64-bit literals to be emitted
using long long, which is 64 bits wide in all relevant implementations.
Rework which information is transmitted in the pipeline metadata,
avoiding redundancy and moving some information there instead of
returning it every time a request is made.
Introduce the support to decouple the `BaseType` from `BasePointer`
pointee type.
This enable us to wrap pointee typees into an implicit array. This
enables pointer arithmetic rewriting of `*(p + i)` as `p[i]`.
Lauri Vasama