This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Problem with std_expand_builtin_va_start on targets where sizetype and ptr_type_node don't have the same mode


On 05 Jul 2007 18:22:50 -0700, Ian Lance Taylor <iant@google.com> wrote:
"Zack Weinberg" <zackw@panix.com> writes:

[std_expand_builtin_va_start]

>   rtx va_r = expand_normal (valist);
>   emit_move_insn (va_r, nextarg);
>
> but this is a part of the compiler I am not at all familiar with...

I think you would want something more like
    rtx va_r = expand_expr (valist, NULL_RTX, VOIDmode, EXPAND_WRITE);
    convert_move (va_r, nextarg, 0);

No regressions on arm-elf (simulator, default and -mthumb); in fact, execute/20040709-1.c stops ICEing at -O3 -fomit-frame-pointer, but I think this is spurious, as that test has no variadic functions in it. OK to commit?

zw

2007-07-06  Ian Lance Taylor <iant@google.com>
            Zack Weinberg <zackw@panix.com>

       PR 32441
       * builtins.c (std_expand_builtin_va_start): Don't use make_tree.

==================================================================
--- builtins.c  (revision 126344)
+++ builtins.c  (local)
@@ -4649,14 +4649,8 @@ std_build_builtin_va_list (void)
void
std_expand_builtin_va_start (tree valist, rtx nextarg)
{
-  tree t;
-  t = make_tree (sizetype, nextarg);
-  t = fold_convert (ptr_type_node, t);
-
-  t = build2 (MODIFY_EXPR, TREE_TYPE (valist), valist, t);
-  TREE_SIDE_EFFECTS (t) = 1;
-
-  expand_expr (t, const0_rtx, VOIDmode, EXPAND_NORMAL);
+  rtx va_r = expand_expr (valist, NULL_RTX, VOIDmode, EXPAND_WRITE);
+  convert_move (va_r, nextarg, 0);
}

/* Expand EXP, a call to __builtin_va_start. */


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]