your mail

[Matthew Fluet]

> I was thinking of doing it another way.  Since I want %esp reserved
> for the C stack at every point, it seems like less work to hide the
> existence of %esp from the x86-codegen altogether.  Is there an easy
> way to do that?

Not that immediately comes to mind.  It wouldn't be so bad to completely
hide %esp from the codegen, except that we still make a lot of FFI calls,
which use push instructions; hence, want to have %esp available.  The way
things are set up in the allocator, I force a push instruction to have
it's base in the %esp register, but it's using the same allocation code as
everything else, so it wants to see %esp in the set of potential
registers.  (Likewise, the register allocator needs to know that %esp
exists and has the c_stackP address.)