On Wed, 8 Dec 2021 at 19:27, Jonathan Wakely wrote:
>
> On Wed, 8 Dec 2021 at 19:21, Jonathan Wakely wrote:
> >
> > On Wed, 8 Dec 2021 at 19:17, Rainer Orth wrote:
> > >
> > > Hi Jonathan,
> > >
> > > > I've pushed this change to trunk now (it was posted and reviewed in
> > > > stage 1, I just didn't get around to pushing it until now).
> > > >
> > > > The final version of the patch is attached to this mail.
> > >
> > > unfortunately, it breaks Solaris/SPARC bootstrap:
> > >
> > > In file included from 
> > > /var/gcc/regression/master/11.4-gcc-gas/build/sparc-sun-solaris2.11/sparcv9/libstdc++-v3/include/bits/shared_ptr.h:53,
> > >                  from 
> > > /var/gcc/regression/master/11.4-gcc-gas/build/sparc-sun-solaris2.11/sparcv9/libstdc++-v3/include/memory:77,
> > >                  from 
> > > /vol/gcc/src/hg/master/local/libstdc++-v3/include/precompiled/stdc++.h:82:
> > > /var/gcc/regression/master/11.4-gcc-gas/build/sparc-sun-solaris2.11/sparcv9/libstdc++-v3/include/bits/shared_ptr_base.h:
> > >  In member function 'void std::_Sp_counted_base<_Lp>::_M_release() [with 
> > > __gnu_cxx::_Lock_policy _Lp = __gnu_cxx::_S_atomic]':
> > > /var/gcc/regression/master/11.4-gcc-gas/build/sparc-sun-solaris2.11/sparcv9/libstdc++-v3/include/bits/shared_ptr_base.h:329:26:
> > >  error: right operand of shift expression '(1 << 64)' is greater than or 
> > > equal to the precision 64 of the left operand [-fpermissive]
> > >   329 |             = 1LL + (1LL << (__CHAR_BIT__ * 
> > > sizeof(_Atomic_word)));
> > >       |                     ~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > > make[9]: *** [Makefile:1875: 
> > > sparc-sun-solaris2.11/bits/stdc++.h.gch/O2ggnu++0x.gch] Error 1
> > >
> > > For 64-bit SPARC, _Atomic_word is long.
> >
> > Ah yes, so we need to disable this optimization. Patch coming up ...
>
> Gah, I remembered to check that:
>
>       constexpr bool __double_word
>         = sizeof(long long) == 2 * sizeof(_Atomic_word);
>       // The ref-count members follow the vptr, so are aligned to
>       // alignof(void*).
>       constexpr bool __aligned = __alignof(long long) <= alignof(void*);
>       if _GLIBCXX17_CONSTEXPR (__lock_free && __double_word && __aligned)
>
>
> But for C++11 and C++14 that is a normal runtime condition not
> if-constexpr, so the undefined shift still gets compiled, even though
> it can never be reached at runtime.

Fixed like so. Tested sparc-sun-solaris2.11, pushed to trunk.
commit c15aa46cca0649b68613d3292cf71c7cc57ef78f
Author: Jonathan Wakely <jwak...@redhat.com>
Date:   Wed Dec 8 19:36:24 2021

    libstdc++: Fix undefined shift when _Atomic_word is 64-bit
    
    The check for _Atomic_word being 32-bit is just a normal runtime
    condition for C++11 and C++14, because it doesn't use if-constexpr. That
    means the 1LL << (CHAR_BIT * sizeof(_Atomic_word)) expression expands to
    1LL << 64 on Solaris, which is ill-formed.
    
    This adds another indirection so that the shift width is zero if the
    code is unreachable.
    
    libstdc++-v3/ChangeLog:
    
            * include/bits/shared_ptr_base.h (_Sp_counted_base::_M_release()):
            Make shift width conditional on __double_word condition.

diff --git a/libstdc++-v3/include/bits/shared_ptr_base.h 
b/libstdc++-v3/include/bits/shared_ptr_base.h
index 90ad30947b0..f315d8f354f 100644
--- a/libstdc++-v3/include/bits/shared_ptr_base.h
+++ b/libstdc++-v3/include/bits/shared_ptr_base.h
@@ -325,8 +325,9 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
       constexpr bool __aligned = __alignof(long long) <= alignof(void*);
       if _GLIBCXX17_CONSTEXPR (__lock_free && __double_word && __aligned)
        {
-         constexpr long long __unique_ref
-           = 1LL + (1LL << (__CHAR_BIT__ * sizeof(_Atomic_word)));
+         constexpr int __wordbits = __CHAR_BIT__ * sizeof(_Atomic_word);
+         constexpr int __shiftbits = __double_word ? __wordbits : 0;
+         constexpr long long __unique_ref = 1LL + (1LL << __shiftbits);
          auto __both_counts = reinterpret_cast<long long*>(&_M_use_count);
 
          _GLIBCXX_SYNCHRONIZATION_HAPPENS_BEFORE(&_M_weak_count);

Reply via email to