On Thu, 30 Sept 2021 at 15:49, Adhemerval Zanella
<adhemerval.zanella@linaro.org> wrote:
>
>
>
> On 30/09/2021 07:43, Florian Weimer via Libc-alpha wrote:
> > * Jonathan Wakely via Libc-alpha:
> >
> >> diff --git a/bits/stdlib-bsearch.h b/bits/stdlib-bsearch.h
> >> index 4132dc6af0..d688ed2e15 100644
> >> --- a/bits/stdlib-bsearch.h
> >> +++ b/bits/stdlib-bsearch.h
> >> @@ -29,14 +29,21 @@ bsearch (const void *__key, const void *__base, size_t __nmemb, size_t __size,
> >>    while (__l < __u)
> >>      {
> >>        __idx = (__l + __u) / 2;
> >> -      __p = (void *) (((const char *) __base) + (__idx * __size));
> >> +      __p = (const void *) (((const char *) __base) + (__idx * __size));
> >>        __comparison = (*__compar) (__key, __p);
> >>        if (__comparison < 0)
> >>      __u = __idx;
> >>        else if (__comparison > 0)
> >>      __l = __idx + 1;
> >>        else
> >> +#if __GNUC_PREREQ(4, 6)
> >> +# pragma GCC diagnostic push
> >> +# pragma GCC diagnostic ignored "-Wcast-qual"
> >> +#endif
> >>      return (void *) __p;
> >> +#if __GNUC_PREREQ(4, 6)
> >> +# pragma GCC diagnostic pop
> >> +#endif
> >>      }
> >>
> >>    return NULL;
> >
> > Patch looks okay, thanks.
> >
> > Reviewed-by: Florian Weimer <fweimer@redhat.com>
> >
> > Florian
> >
>
> I am seeing a lot of failure on x86_64 with gcc 11.1 after this
> patch landed:
>
> x86_64-linux-gnu$ grep ^FAIL tests.sum
> FAIL: catgets/de/libc.cat
> FAIL: catgets/test-gencat
> FAIL: catgets/test1.cat
> FAIL: catgets/tst-catgets
> FAIL: debug/tst-chk1
> FAIL: debug/tst-chk2
> FAIL: debug/tst-chk3
> FAIL: debug/tst-chk4
> FAIL: debug/tst-chk5
> FAIL: debug/tst-chk6
> FAIL: debug/tst-lfschk1
> FAIL: debug/tst-lfschk2
> FAIL: debug/tst-lfschk3
> FAIL: debug/tst-lfschk4
> FAIL: debug/tst-lfschk5
> FAIL: debug/tst-lfschk6
> [...]
>
> For instance some math tests shows ulps failures that does not
> make much sense:
>
> $ ./testrun.sh math/test-double-cacos
> testing double (without inline functions)
> Failure: Test: Imaginary part of: cacos_downward (-0x1p-52 - 0x1.0000000000001p+0 i)
> Result:
>  is:          8.8137358701954271e-01   0x1.c34366179d424p-1
>  should be:   8.8137358701954315e-01   0x1.c34366179d428p-1
>  difference:  4.4408920985006261e-16   0x1.0000000000000p-51
>  ulp       :  4.0000
>  max.ulp   :  3.0000
> Failure: Test: Imaginary part of: cacos_downward (-0x1p-52 - 0x1.000002p+0 i)
> Result:
>  is:          8.8137367131323707e-01   0x1.c34368ebb10d9p-1
>  should be:   8.8137367131323751e-01   0x1.c34368ebb10ddp-1
>  difference:  4.4408920985006261e-16   0x1.0000000000000p-51
>  ulp       :  4.0000
>  max.ulp   :  3.0000
> [...]
>
> Reverting fixes it.

Huh, I wonder why I didn't see these. Sorry. I'll see if Joseph's
suggestion helps (as attached) or I'll revert it.