On October 20, 2018 1:00:52 AM GMT+02:00, Jakub Jelinek <ja...@redhat.com>
wrote:
>Hi!
>
>COMPOUND_LITERAL_EXPRs are removed from static initializers in
>record_references_in_initializer, unfortunately decode_addr_const can
>be
>called from const_hash_1 from output_constant_def before that happens
>and as record_references_in_initializer needs a varpool node, we can't
>call
>it during the hashing used to check if we have such a constant already.
>
>The following patch handles COMPOUND_LITERAL_EXPRs like they will be
>handled
>later on for the purpose of hashing them.
>
>Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
OK.
Richard.
>2018-10-19 Jakub Jelinek <ja...@redhat.com>
>
> PR middle-end/87647
> * varasm.c (decode_addr_const): Handle COMPOUND_LITERAL_EXPR.
>
> * gcc.c-torture/compile/pr87647.c: New test.
>
>--- gcc/varasm.c.jj 2018-10-11 09:05:48.124511816 +0200
>+++ gcc/varasm.c 2018-10-19 13:27:41.921160889 +0200
>@@ -2953,6 +2953,11 @@ decode_addr_const (tree exp, struct addr
> gen_rtx_SYMBOL_REF (Pmode, "origin of addresses"));
> break;
>
>+ case COMPOUND_LITERAL_EXPR:
>+ gcc_assert (COMPOUND_LITERAL_EXPR_DECL (target));
>+ x = DECL_RTL (COMPOUND_LITERAL_EXPR_DECL (target));
>+ break;
>+
> default:
> gcc_unreachable ();
> }
>--- gcc/testsuite/gcc.c-torture/compile/pr87647.c.jj 2018-10-19
>13:30:28.797388068 +0200
>+++ gcc/testsuite/gcc.c-torture/compile/pr87647.c 2018-10-19
>13:30:00.778853626 +0200
>@@ -0,0 +1,15 @@
>+/* PR middle-end/87647 */
>+
>+struct A {};
>+struct A *const b = &(struct A) {};
>+struct B { char *s; struct A *t; };
>+void bar (struct B *);
>+
>+void
>+foo (void)
>+{
>+ struct B a[] = { "", b, "", b, "", b, "", b, "", b, "", b, "", b,
>"", b,
>+ "", b, "", b, "", b, "", b, "", b, "", b, "", b, "", b,
>+ "", b };
>+ bar (a);
>+}
>
> Jakub
