Incrementally implementing a simple ML compiler using LLVM

I recently tried the Low-Level Virtual Machine (LLVM) project and found that
I can use its OCaml bindings to write native-code compilers with ease.


I would like to use this technology to create a simple compiler for a
language similar to OCaml but with no baggage (e.g. I have no desire
to support 63-bit integers!). However, I am completely new to this (I
am a computational physicist/dabbler) so I'd really appreciate a
little assistance.


I already have a really bare-bones compiler for a first-order language
with a single type (int) that can compile a Fibonacci program to
x86-64 native code.


I'm not even sure what my milestones should be but I assume I should
add more types (e.g. function pointers), boxed values and a garbage
collector next.


Until now I have only been superficially aware of boxing (in the
context of optimization). What exactly is the simplest run-time
representation of a boxed value? Is it a pointer to a >=1 word-sized
block?


Is the Cheney semi-space GC a suitable starting point? Is there an
abstract interface for using a GC from generated code that I could
adhere to?


Is there anything else that I should be aware of?


--
Dr Jon D Harrop, Flying Frog Consultancy Ltd.
http://www.ffconsultancy.com/products/?u

Return to the comp.compilers page.
Search the comp.compilers archives again.