https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106692
anlauf at gcc dot gnu.org changed:
What |Removed |Added
----------------------------------------------------------------------------
Known to work| |6.4.1
CC| |anlauf at gcc dot gnu.org
Known to fail|7.5.0 |7.4.1
--- Comment #15 from anlauf at gcc dot gnu.org ---
(In reply to Tobias Burnus from comment #13)
> > What do others think?
>
> I am a bit unsure. Cray pointers are weird and it is not quite clear how
> they are used in real world code.
>
> Your modification causes a missed optimization for code like:
>
> var = NULL
> if (var == NULL)
>
> as the volatile not only prevents seeing this but also reloads the variable
> instead of using the value in the register.
>
> But as Cray pointers are weird, pointer checks are rare and volatile is
> localized, it might actually be okay?
I have tried replacing TREE_VOLATILE by TREE_STATIC, i.e.
+ // Hack: prevent optimization of comparison of Cray pointers (PR106692)
+ if (expr->value.op.op1->expr_type == EXPR_VARIABLE
+ && expr->value.op.op1->ts.type == BT_INTEGER
+ && expr->value.op.op1->symtree
+ && expr->value.op.op1->symtree->n.sym->attr.cray_pointer)
+ TREE_STATIC (lse.expr) = 1;
[...]
and this works too. The optimized dump shows that the above if () is
still optimized away at -Og and higher, unless there is a PRINT statement
in between. (Another missed optimization around PRINT statements?).
Any other suggestion? Otherwise I'll wrap up and submit.
