The LLVM (Low Level Virtual Machine) is virtual machine at a machine code level. It supports both stand alone code generation and JITing. The Haskell llvm package is a (relatively) high level interface to the LLVM. The high level interface makes it easy to construct LLVM code. There is also an interface to the raw low level LLVM API as exposed by the LLVM C interface.

LLVM code is organized into modules (type Module). Each module contains a number of global variables and functions (type Function). Each functions has a number of basic blocks (type BasicBlock). Each basic block has a number instructions, where each instruction produces a value (type Value).

Unlike assembly code for a real processor the assembly code for LLVM is in SSA (Static Single Assignment) form. This means that each instruction generates a new bound variable which may not be assigned again. A consequence of this is that where control flow joins from several execution paths there has to be a phi pseudo instruction if you want different variables to be joined into one.

The definition of several of the LLVM entities (Module, Function, and BasicBlock) follow the same pattern. First the entity has to be created using newX (where X is one of Module, Function, or BasicBlock), then at some later point it has to given its definition using defineX. The reason for splitting the creation and definition is that you often need to be able to refer to an entity before giving it's body, e.g., in two mutually recursive functions. The the newX and defineX function can also be done at the same time by using createX. Furthermore, an explicit name can be given to an entity by the newNamedX function; the newX function just generates a fresh name.