22 Apr
2026
22 Apr
'26
9:58 p.m.
El 21/04/2026 a las 18:25, Joaquin M López Muñoz via Boost escribió:
At this microlevel, it's hard to beat a local mask-based loop as boost::container::hub uses.
Very hard. I was crazy trying to know why nest was behind in the for_each bench without luck. It turns that doing: pbb->mask & (full << 1 << std::size_t(n)); vs (note no std::size_t conversion on n, which is an int) pbb->mask & (full << 1 << n); was a 5% improvement because converting int to uint64 before shifting was a noticeable impact, I was quite shocked. I think the porting performance is on par with the proposal in my machine, in case you want to test it: https://github.com/boostorg/container/blob/develop/experimental/bench_hub.cp... Ion