The compiler supports the complete Latte language including arrays. LLVM IR and x86_64 backends are available.

The code is transformed to SSA first. The following optimisations are applied:

• removal of unreachable blocks,
• removal of unused assignments,
• removal of unreachable instructions inside blocks,
• copy propagation (thanks to transformation to SSA),
• compile time calculation of constant expressions,
• removal of trivial phi instructions.

x86_64 backend stores commonly used values in processor registers. The implemented algorithm bases on SSA. Interference graph of SSA registers is built and coloured. Registers with the same colour are assigned to the same register or address in memory.

Moreover, if a function argument is passed by register and register allocation algorithm decides to keep it in register (rather than on the stack), it will preserve its original register. Similarly, if a function argument is passed on stack and the algorithm decides to keep it in memory, the argument's colour is granted the argument's place in memory.

Before function calls, all caller preserved registers are backed up on stack and restored after the called function returns.

lattelib.c defines library functions and auxiliary builtins like string concatenation.

Functions (apart from main) are appended a prefix latte_, so that they do not conflict with functions from the standard library.

The following files are of particular interest:

• CompileLatte - translates Latte AST to LLVM,
• CompilerErr - listing of possible compiler errors,
• CompilerState - monads used in the compiler,
• LLVM - definition of LLVM IR AST and pretty printing,
• TransLLVM - transformations on LLVM including optimisations and mem2reg algorithm.
• X86_64 - translates LLVM to x86_64.

Moreover, lattelib.c contains Latte library functions like printInt.