On Tue, 2022-12-06 at 09:49 -0600, Richard Henderson wrote:
> On 12/6/22 09:28, Ilya Leoshkevich wrote:
> > > + switch (TCG_TARGET_CALL_ARG_I64) {
> > > + case TCG_CALL_ARG_EVEN:
> >
> > On a s390x host with gcc-11.0.1-0.3.1.ibm.fc34.s390x I get:
> >
> > FAILED: libqemu-aarch64-softmmu.fa.p/tcg_tcg.c.o
> > ../tcg/tcg.c: In function ‘init_call_layout’:
> > ../tcg/tcg.c:739:13: error: case value ‘1’ not in enumerated type
> > [-Werror=switch]
> > 739 | case TCG_CALL_ARG_EVEN:
> > | ^~~~
> >
> > The following helps:
>
> Yes, I found and fixed this since.
>
> > --- a/tcg/tcg.c
> > +++ b/tcg/tcg.c
> > @@ -735,7 +735,7 @@ static void init_call_layout(TCGHelperInfo
> > *info)
> > break;
> >
> > case TCG_TYPE_I64:
> > - switch (TCG_TARGET_CALL_ARG_I64) {
> > + switch ((TCGCallArgumentKind)TCG_TARGET_CALL_ARG_I64)
> > {
> > case TCG_CALL_ARG_EVEN:
> > layout_arg_even(&cum);
> > /* fall through */
> >
> > This looks like a gcc bug to me.
>
> The gcc "bug" is only in not being sufficiently verbose. It should
> say something about
> *differing* enumerated types, and perhaps name them.
>
> Back in patch 20, tcg/s390x/tcg-target.h,
>
> -#define TCG_TARGET_CALL_ARG_I64 TCG_CALL_RET_NORMAL
> +#define TCG_TARGET_CALL_ARG_I64 TCG_CALL_ARG_NORMAL
>
>
> r~
I looked at this line and completely missed the RET vs ARG difference.
Your diff fixes the issue for me too, of course.
Thanks!
