> Hi,
> the alias-2.c testcase fails on targets with anchors. The reason is that
> the variable itself is anchored while the alias is not and they point to the
> same location. I folllowed the docs of SYMBOL_REF claiming that
> SYMBOL_REF_DECL if the symbol is label and tought it is safe to disambiguate
> them.
What kind of distinction do you mean between "label" and non-label here?
I don't think !SYMBOL_REF_DECL tells us anything useful about the
symbol. I think it's more a case of: if SYMBOF_REF_DECL is present,
we can assume that what it says is accurate. If it isn't present we
can't assume anything.
So was it...
Jan Hubicka <hubi...@ucw.cz> writes:
> @@ -2323,26 +2367,14 @@
>
> if (GET_CODE (x) == SYMBOL_REF && GET_CODE (y) == SYMBOL_REF)
> {
> - tree x_decl = SYMBOL_REF_DECL (x);
> - tree y_decl = SYMBOL_REF_DECL (y);
> - int cmp;
> + int cmp = compare_base_symbol_refs (x,y);
>
> - if (!x_decl || !y_decl)
> - {
> - /* Label and normal symbol are never the same. */
> - if (x_decl != y_decl)
> - return 0;
> - return offset_overlap_p (c, xsize, ysize);
...this part of the code that was causing the problem? That doesn't
seem valid even without the anchor stuff. I think the starting position
should be that we can only use offset_overlap_p if XSTR (x) == XSTR (y)
and need to assume a conflict otherwise. E.g. I think in principle it's
valid for a target to create symbol_refs for different parts of a single
artificial object.
I agree there are other refinements you can do on top of that,
e.g. that two block symbols in different blocks can never conflict.
But the patch seems to be treating anchors as an exception to the rule,
whereas I think they're just one instance of the rule.
Thanks,
Richard
