https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118952
--- Comment #3 from Richard Sandiford <rsandifo at gcc dot gnu.org> ---
(In reply to ktkachov from comment #2)
> Thanks, yeah I don't see PR34678 getting generally resolved any time soon.
> Is there something we could do with these particular builtins to make them a
> stronger optimisation/reordering barrier in the meantime though?
Not sure, but it seems unlikely. Scheduling/optimisation barriers are kind of
anathema to SSA representations. Given the SSA expressions:
a = b + c
...
d = a - c
it should be possible to fold d to b without having to look through "...".
I think we'll just to have to recommend to users that any code written for a
particular FPCR setting should be in its own noipa function. That would be:
memcpy(&fsrc, &src, 4);
double d = (double) fsrc;
memcpy(&dst, &d, 8);
in the example.
