[Bug libgcc/71744] Concurrently throwing exceptions is not scalable

["gleb at scylladb dot com" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71744

--- Comment #22 from Gleb Natapov <gleb at scylladb dot com> ---
(In reply to torvald from comment #21)
> (In reply to Jakub Jelinek from comment #17)
> > (In reply to torvald from comment #15)
> > > > Similarly, the 64 recursive locks in libc, again, significant amount of
> > > > memory
> > > > per process that doesn't care about exceptions,
> > > 
> > > That might be reducable if we build custom locks and don't use pthread ones,
> > > but we'll at least have 64b per lock.
> > 
> > Right now the lock is I think 40 bytes per lock, which is 64 * 40 bytes per
> > process.  That is just too much (but of course the 64x locking a recursive
> > lock is even worse).
> 
> Remembering more of the discussion we had about this in the past, then even
> with the improved rwlock, Gleb reports that there is a slowdown in Gleb's
> tests because of cache contention -- which would mean that we may have to
> use one cacheline per lock.

Right, my patch lacks it, but cache alignment will definitely be an
improvement.