Backend

The backend of a compiler takes the representation of the language it’s compiling and emits its target language. Over time, libraries were created to modularize this process and make it easier for a compiler designer to target common architectures.

There are two main ways to translate a language to machine code: Ahead-of-time Compilation and Just-in-time Compilation. AOT Compilers emit machine code immediately, while JIT Compilers won’t translate code until it needs to be executed. This separates backend frameworks into Code Generators and Virtual Machines. There isn’t a huge difference between these terms, but usually CGs refer to backends that use AOT compilation, and VMs refer to backends that use JIT compilation. This line can be blurred with frameworks like LLVM that provide both.

Code Generators

Code Generators offer libraries and tools for generating native code for a variety of architectures. There are many platforms one can choose, and as such this list is not comprehensive. These usually consist of an Intermediate Representation that a compiler will emit. The library will take that IR and produce machine code for one of the instruction sets it supports.

Popular Code Generators:

• LLVM
• CraneLift
• QBE
• GRIN

Virtual Machines

Virtual Machines act similarly, except one can translate code to a VM’s instruction set and execute it at a later date. This usually requires bundling the VM with your program or expecting the VM to be on the user’s machine. The VM will handle translating the program to machine code at runtime through JIT compilation.

Popular Virtual Machines:

• JVM
• CLR