Hi!
On Thu, Nov 10, 2016 at 06:00:29PM +0100, Jakub Jelinek wrote:
> On arm/aarch64 we ICE because some decls that make it here has non-NULL
> DECL_SIZE, which is a VAR_DECL rather than CONST_INT (or DECL_SIZE that
> doesn't fit into shwi would ICE similarly). While it is arguably a FE bug
> that it creates for VLA initialization from STRING_CST such a decl,
> I believe we have some PRs about it already open.
> I think it won't hurt to check for the large sizes properly even in this
> function though, and punt on unexpected cases, or even extremely large ones.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, Yvan said he is going
> to test on arm and/or aarch64. Ok for trunk if the testing there passes
> too?
>
> 2016-11-10 Jakub Jelinek <ja...@redhat.com>
>
> PR middle-end/78201
> * varasm.c (default_use_anchors_for_symbol_p): Fix a comment typo.
> Don't test decl != NULL. Don't look at DECL_SIZE, but DECL_SIZE_UNIT
> instead, return false if it is NULL, or doesn't fit into uhwi, or
> is larger or equal to targetm.max_anchor_offset.
>
> * g++.dg/opt/pr78201.C: New test.
I'd like to ping this patch, Yvan has successfully tested it on arm and
aarch64.
Jakub
