Some (basic) stuff about compilers
"Compilers: Backend to Frontend and Back to Front Again" tutorial
Building a compiler, extending a grammar step by step
- Input: toy scheme
- Output: x86 executable
- The tutorial uses scheme as the implementation language, I used Haskell
Introduces tagged pointer representation
Puts the integer representation in %eax register
Introduce stack to save intermediate values
Introduces an environment to map variables to stack positions
Introduces a calling convention putting parameters in the stack
Another approach: multipass compilers
"Essentials of compilation" book: https://github.com/IUCompilerCourse/Essentials-of-Compilation/releases/tag/racket-MIT-press
https://brinckerhoff.org/clements/2194-csc431/essentials-of-compilation.pdf
General idea
- Input is scheme like. Parsing is not a priority.
- Multiple intermediate languages
- One pass mapping one to the other
- Final pass outputs x86_64 assembler
A way of validating:
- Uniquify: make var names unique
- remove-complex: makes arguments either vars or consts, but not complex ops
- explicate-control: makes order of execution explicit
- uncover-local: ❌ not in the edition I'm reading 😅. Just collecting variables names and putting them in the info
- select-instr: transforms C into a pseudo x86 with vars
❌
- assign-homes: places vars into the stack
- patch-instr: make sure that each instruction adheres to
the restrictions regarding which arguments can be memory references - before print-x86: put a prelude and main labels
❌
Example: explicate-control
Other passes we can do: partial-evaluation
Next topics in the book which I haven't got to
- Register allocation: not everything needs to be in the stack
- Apparently there's some type checking and garbage collection in later chapters!