[llvm-dev] Concurrent Hashmap?

[antlists via llvm-dev]

On 08/04/2021 20:44, Reid Kleckner via llvm-dev wrote:
> I recommend splitting up symbol resolution into a fully data parallel 
> phase and a "shuffle" phase that does the concurrent hash table 
> insertion. This makes rehashing very simple: if the table gets too full, 
> simply abort the concurrent insertions, sync up the worker threads, and 
> retry with a bigger table.

Could you do a dynamic hash instead? In other words grow the hash 
instead of starting over - what's going to be the cost of retrying as 
opposed to growing?

Cheers,
Wol