------- Comment #23 from pcarlini at suse dot de 2006-05-22 08:54 ------- (In reply to comment #22)
> This would mean a non-enable-threads compiled libstdc++ would skip the calls > to __exchange_and_add? Of course the idea would not be that of "skipping", instead providing a version of __exchange_and_add not using slow atomics and inlined. > And to clarify, there would be no other way to skip these calls except > making a separate non-threaded libstdc++? Yes. At least for this PR. If you can provide compelling experimental evidence that normal users can pay the overhead of checking each __gthread_active_p() each time, then we could reconsider that, in a *separate* PR. Also, since we are moving anyway to not-reference-counted strings (already provided in ext/), there is not much point for more invasive and complex changes. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24704
