Diogo Netto

We should understand why this is not reproducible on master and what was the "fix". Would help to get some data on what was the "fix" commit. I can try...

> I'm not sure what you mean by versioninfo? ``` help?> versioninfo search: versioninfo varinfo versioninfo(io::IO=stdout; verbose::Bool=false) Print information about the version of Julia in use. The output is controlled...

Yes, was mostly interested on what architecture you were running.

We have aligned allocation functions. If that's indeed the culprit we could just use them.

> And the second side, is actual run-time alignment. Unless you intentionally go out of your way, the actual run-time alignment guarantees are rather weak, since there is a trade-off:...

CC: @gbaraldi and @vtjnash.

Marking as draft until the x64 failure is investigated.

> is the linked list an efficient enough data structure for this Probably? Note that the sweeping code will need to access the GC bits in `bigval_t` and the `next/prev`...

Re-ran the benchmarks and looks close to performance neutral on the serial & multithreaded GCBenchmarks. Serial ones were run with 1 thread, multithreaded ones with 8 mutators and 8 GC...

> Could this expose parallelism for sweeping the big objs? Yes. We can make the list of big objects lock-free (similar to what we do in the `jl_gc_page_stack_t`) in order...