4 ms·

I couldn't agree more with this, it's very frustrating. I recently worked through most of the "make a lisp" guide [0], and it was excellent introduction to a si

I couldn't agree more with this, it's very frustrating. I recently worked through most of the "make a lisp" guide [0], and it was excellent introduction to a simple lisp interpreter. But now I want to try my hand at a compiler (either producing machine code or maybe custom VM bytecode) and the lack of resources for that is bewildering. Just as you say, I've found that most online texts on compilers seem to spend most of their time on parsing, which is something I already feel relatively comfortable with. Does anyone know of a good resource on how to transform an AST to machine code? I've skimmed Crafting Interpreters and the Bytecode VM section seems to spend an inexplicable amount of time on not only parsing, but minutia having nothing to do with compilers (e.g. writing a custom dynamic array implementation in C). [0] https://github.com/kanaka/mal/blob/master/process/guide.md https://github.com/kanaka/mal/blob/master/process/guide.md

fao_ 7y ago
Check out "Modern Compiler Implementation" by Appel and also the Red (Purple?) Dragon book.
- kerkeslager 7y ago
  sigh Have you actually done this? Do you really think we haven't? The Appel book in particular is very good, but it only covers assembly generation at a high level. None of the examples target a real-world architecture.
  - fao_ 7y ago
    I don't know what you have done. The red dragon book goes over targeting an intermediate form that's basically assembly language, it also covers register allocation in depth. As far as I can tell, most of the texts after that is in research papers and real world examples (https://c9x.me/compile/ https://c9x.me/compile/)
mras0 7y ago
I don't have a good resource, but having just finished a small C-compiler myself[0] after previously getting stuck on the codegen part, I have a few suggestions: Keep it simple and focus on making it composable. By that I mean if you're targeting x86 keep the current value in [r|e]ax (forget about non-integer stuff for now :), if it's a stack based VM: focus on the stack top. A numeric literal loads [r|e]ax (or pushes it on the stack). Binary operators first pushes computes the lhs; pushes the result; computes rhs, combines and leaves the result in [r|e]ax (or the stack top). Output to some kind of text format and lean on existing tools. It's probably not a bad idea to output C, LLVM bitcode, JVM/.NET bytecode or x86 assembly at first rather than going all the way to machine code starting out. Postpone thinking about optimizations and reduce your level of ambition. All of my previous attempts at finishing compiler projects failed because the complexity got out of hand. For instance in a C compiler don't try to optimize expressions of smaller-than-int types. Maybe all expressions in your language can be calculated using doubles? Finally as GP says: You need to be comfortable with the output language (assembly or otherwise). Study the output of similar compilers and translate code by hand into assembly to get a feel of what your compiler should be doing. [0]: https://github.com/mras0/scc https://github.com/mras0/scc
jwhitlark 7y ago
Lisp in Small Pieces https://www.amazon.com/dp/0521545668/ref=cm_sw_r_other_apa_i_q2KfEbQRY3CFP https://www.amazon.com/dp/0521545668/ref=cm_sw_r_other_apa_i... is excellent, and I think would help.

fao_ 7y ago

Check out "Modern Compiler Implementation" by Appel and also the Red (Purple?) Dragon book.

kerkeslager 7y ago
sigh Have you actually done this? Do you really think we haven't? The Appel book in particular is very good, but it only covers assembly generation at a high level. None of the examples target a real-world architecture.
- fao_ 7y ago
  I don't know what you have done. The red dragon book goes over targeting an intermediate form that's basically assembly language, it also covers register allocation in depth. As far as I can tell, most of the texts after that is in research papers and real world examples (https://c9x.me/compile/ https://c9x.me/compile/)

mras0 7y ago

I don't have a good resource, but having just finished a small C-compiler myself[0] after previously getting stuck on the codegen part, I have a few suggestions: Keep it simple and focus on making it composable. By that I mean if you're targeting x86 keep the current value in [r|e]ax (forget about non-integer stuff for now :), if it's a stack based VM: focus on the stack top. A numeric literal loads [r|e]ax (or pushes it on the stack). Binary operators first pushes computes the lhs; pushes the result; computes rhs, combines and leaves the result in [r|e]ax (or the stack top). Output to some kind of text format and lean on existing tools. It's probably not a bad idea to output C, LLVM bitcode, JVM/.NET bytecode or x86 assembly at first rather than going all the way to machine code starting out. Postpone thinking about optimizations and reduce your level of ambition. All of my previous attempts at finishing compiler projects failed because the complexity got out of hand. For instance in a C compiler don't try to optimize expressions of smaller-than-int types. Maybe all expressions in your language can be calculated using doubles? Finally as GP says: You need to be comfortable with the output language (assembly or otherwise). Study the output of similar compilers and translate code by hand into assembly to get a feel of what your compiler should be doing. [0]: https://github.com/mras0/scc https://github.com/mras0/scc

jwhitlark 7y ago

Lisp in Small Pieces https://www.amazon.com/dp/0521545668/ref=cm_sw_r_other_apa_i_q2KfEbQRY3CFP https://www.amazon.com/dp/0521545668/ref=cm_sw_r_other_apa_i... is excellent, and I think would help.