[Bug tree-optimization/65818] [6 Regression] libiberty/vprintf-support.c:41:1: ICE: in expand_i fn_va_arg_1, at tree-stdarg.c:1095

["vries at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65818

--- Comment #7 from vries at gcc dot gnu.org ---
Hmm, this patch has failures for x86_64:

- va_arg (ap, void):
  gcc.c-torture/compile/pr48767.c 
- vla:
  gcc.c-torture/execute/20020412-1.c
  gcc.dg/compat/struct-by-value-22
  gcc.dg/lto/20090706-1
  gcc.target/i386/memcpy-strategy-4.c

this patch is more conservative:
...
diff --git a/gcc/gimplify.c b/gcc/gimplify.c
index 7786e16..e23f510 100644
--- a/gcc/gimplify.c
+++ b/gcc/gimplify.c
@@ -9336,7 +9336,12 @@ gimplify_va_arg_internal (tree valist, tree type,
location_t loc,
   else
     gimplify_expr (&valist, pre_p, post_p, is_gimple_min_lval, fb_lvalue);

-  return targetm.gimplify_va_arg_expr (valist, type, pre_p, post_p);
+  tree expr = targetm.gimplify_va_arg_expr (valist, type, pre_p, post_p);
+  gcc_assert (TREE_CODE (expr) == MEM_REF);
+  if (!is_gimple_mem_ref_addr (TREE_OPERAND (expr, 0)))
+    gimplify_expr (&TREE_OPERAND (expr, 0), pre_p, post_p,
+                  is_gimple_mem_ref_addr, fb_rvalue);
+  return expr;
 }

 /* Gimplify __builtin_va_arg, aka VA_ARG_EXPR, which is not really a
...

It allows at least pr48767.c, 20020412-1.c and memcpy-strategy-4.c to pass.