Re: Register allocation in special circumstancies

"Andrey S. Bokhanko" <ruff@mcst.ru> wrote:
> As we all know, C language has nasty (for compiler developers)
> feature: setjmp/longjmp. If there is "setjmp" point in procedure, than
> to explicitly express control flow we have to create control flow
> edges from ALMOST EVERY reachable (from "setjmp" point) operation to
> that "setjmp" point. This is TOO costly (both in memory requirements
> and compilation time), so we decided not to make such edges. Of
> course, this decision prohibit application of wide class of
> optimizations in such procedures, but this is inevitable cost (AFAIK,
> gcc compiler works exactly same way).
>
> But we can't disable one special optimization: register
> allocation. So, that's where the problem is: if r.a. based on
> classical graph coloring approach, we must construct Register
> Interference Graph (RIG), and hence we must take into account this
> nasty feature of "setjmp" points. But how to do it? I have some ideas,
> but they all are less than ideal, so any help will be gratefully
> appreciated.


In C (and I believe in C++) there is the rule: If a non-volatile variable
is modified between setjmp and longjmp then the contents of that variable
is indeterminate. I very much suspect that if you treat setjmp as a normal
function, implement it by saving and restoring registers, stackpointer and
program counter, AND keep all volatile variables out of registers then you
should be fine.

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