On 03/06/19 19:18 +0200, François Dumont wrote:
Sorry for the mis-understanding but the core of this patch has already
been committed after your approval here:
https://gcc.gnu.org/ml/libstdc++/2019-05/msg00023.html
OK, thanks. I didn't realise it superseded this one. I thought there
might still be bits of this that would help. But looking at it again,
I see either the code touched by this patch has changed anyway, or
your patch a month ago made equivalent changes.
It was considered as PR 90277 fix but eventually I also change tests
to make those implementation-details agnostics.
However the performance test didn't make it, I'll re-submit it in
another patch more dedicated to small size optimization.
OK, that would be useful. Thanks.
It might still be useful to know more about why -O3 helps so much, but
that might require one of the middle end experts to analyze it.
Thanks!
