https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102147
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Well, ira_conflict_vector_profitable_p is dependent on sizeof (ira_object_t)
but it's a decision about the representation of conflicts. The question is now
whether at any point processing of the two representations can yield different
answers. That might happen if we somewhere iterate over conflicts?
I suppose the simplest fix for the actual bootstrap issue is to make
ira_conflict_vector_profitable_p independent of sizeof (ira_object_t)
and sizeof (IRA_INT_TYPE) and simplify
return (2 * sizeof (ira_object_t) * (num + 1)
< 3 * nw * sizeof (IRA_INT_TYPE));
for example to
return (2 * (num + 1) < 3 * nw);
(note 'nw' is also rounded up to IRA_INT_BITS).
