Kwok Cheung Yeung <k...@codesourcery.com> writes:
> Hello
>
> On 06/08/2020 1:23 pm, Tom de Vries wrote:
> > +static char AC[4];
> > +static char init_qi[4] = { -30,-30,-50,-50 };
> > +static char test_qi[4] = { -115,-115,25,25 };
> > +
> > +static void
> > +do_qi (void)
> > +{
> > + if (__sync_val_compare_and_swap(AC+0, -30, -115) != -30)
> > + abort ();
>
> If 'char' is unsigned by default, then init_qi will contain { 226, 226, 206,
> 206} and test_qi { 141, 141, 25, 25 }, which will result in the comparison
> against -30 failing when the previous value of AC[0] is implicitly promoted
> to
> signed int. This can be fixed by making the array element types explicitly
> signed.
>
> This issue is tracked as issue 96519 on the tracker. I have checked that the
> test now passes on PowerPC and Aarch64. Is the fix okay for trunk?
>
> Thanks
>
> Kwok
>
> commit fc6ac3af45a238da0bd65e020ae6f0f165b57b87
> Author: Kwok Cheung Yeung <k...@codesourcery.com>
> Date: Tue Aug 11 09:41:10 2020 -0700
>
> Fix gcc.dg/ia64-sync-5.c for architectures with unsigned char as default
> (PR 96519)
>
> If char is unsigned, then comparisons of the char array elements against
> negative integers in the test will fail as values in the array will always
> be positive, and will remain so when promoted to signed int.
>
> 2020-08-11 Kwok Cheung Yeung <k...@codesourcery.com>
>
> PR testsuite/96519
>
> gcc/testsuite/
> * gcc.dg/ia64-sync-5.c (AC, init_qi, test_qi): Change element type to
> signed char.
OK, thanks.
Richard
