devirtualize

Devirtualize is a compiler and runtime optimization that replaces a virtual call with a direct call to a specific implementation when the target type can be determined, enabling inlining and reducing the overhead of dynamic dispatch. The goal is to improve performance at polymorphic call sites without changing program semantics.

How it works: Static analysis and profile-guided information identify call sites where only a single concrete subtype is possible (monomorphic call sites) or where the receiver’s type is known at runtime. If a method is final or the type is known, the compiler may emit a direct call to the specific implementation instead of performing a vtable lookup. JIT runtimes may use inline caching (IC) or polymorphic inline caches to remember observed types and devirtualize subsequent calls or inline the target method.

Usage and languages: In statically typed languages such as C++, devirtualization is a common optimization performed by optimizing compilers. In managed runtimes like the Java Virtual Machine or .NET CLR, JIT compilers attempt devirtualization when receiver types are predictable or methods are not polymorphic. Devirtualization is also used in dynamic language runtimes with type feedback and inline caches.

Limitations and trade-offs: Devirtualization is only applicable when the incoming type information is reliable; if new subclasses may be introduced at runtime, devirtualized paths may be invalidated. Semantics remain the same; correctness is preserved because the optimization preserves dynamic dispatch where required. Marking methods as final or classes as final often facilitates devirtualization.