Nathan Grasso Bronson
Nathan Grasso Bronson
That's also what I would expect, but I see it with both ska::bytell_hash_set and ska::flat_hash_set when trying to run the benchmark with them.
I observe that in dbg builds the `assert(res.front().size() == res.back().size());` in GenerateSets fails. In opt builds the benchmark crashes.
I think it will be reasonable to just use the minimum size during Transpose, which isn't much work (I've already done it).
Another issue is that the early-rehash behavior is dependent on key insertion order, so changes to GenerateSet will also be required for Density::kMax. Some ways to fix it would be...
6989aafc963dbf16e343afcf32d7ad84a6719240 includes a basic TMap implementation (pred-rc-enum).
Perhaps a concurrent bounding box (or convex hull) algorithm?
Thanks, the heap dump might be very helpful. I won't be able to work on this during the next week, but I will dig into it as soon as possible
I am working on it, but I have newborn twins so it is hard to achieve the required mental clarity. I have a repro and a theory. Thanks for reporting...
Given a compiler plugin, is targeting `TxnFieldUpdater` actually the best strategy?
@ot, is anybody at Meta working on a functioning implementation of `rcu_barrier()`? If not, would you be willing to take it as an external contribution? Rockset has a barrier implementation...