Global thread-local free-list allocator

[Herbie Melbourne]

On Thursday, 16 March 2023 at 20:53:28 UTC, Witold wrote:
> I ported a classic DeltaBlue benchmark to D, and for 
> allocations I use a mix of malloc/free (for structs with some 
> trailing inline arrays), and new for some other structs and 
> arrays.
>
> After running it, it felt kind of slow (I did not compare it to 
> any other language yet, so no idea if it is actually slower, 
> just feels slow), so I did some profiling (perf / hotspot 
> mostly), and noticed that substantial amount of time is spent 
> in gc (both allocation and collection). This is on Linux, and 
> benchmark indeed creates a lot of garbage, as I rerun and rerun 
> it in a loop.

care to share some numbers?
did you try the forking GC on Linux, too ?