Memory Segment efficient array handling

[Maurizio Cimadamore]

On 01/04/2021 14:33, Ty Young wrote:
> Is the expectation that people have to buy and read a metaphorical 
> "Java <X> for dummies"book every new Java release just so they know 
> what the current set of optimizations are and in what situations they 
> do and don't come into play? 

On this point - the optimization we're talking about here is escape 
analysis. One of the first docs I find on Google on the topic is

https://wiki.openjdk.java.net/display/HotSpot/EscapeAnalysis

Which is a 2009 document. So, it's not like we're talking about bleeding 
edge here. It's a technique that has been around for years and which 
works with some degree of success (see below).

In the case of the simple bulk copy routines discussed here, it is easy 
to see how all the intermediate heap segment and the various slices 
never leave the method where they are used - they are just temporary 
artifacts. In such cases, Hotspot will try hard (as the benchmark shows) 
to remove allocation. But this is nothing new.

A different thing is e.g. to write a method that computes a slice *and* 
returns it - in that case the slice "escapes" the method - so in such a 
case you can't robustly rely on Hotspot to be able to figure it out (in 
the case where the slicing method is only called in one place and 
inlined there, well, maybe Hotspot could detect it doesn't escape that 
new context).

AFAIK, the biggest impediment to escape analysis is dependency on 
control flow. So, if you have a method which branches, and create a 
slice in one case and a different slice in another, in general Hotspot 
will refuse to escape-analyze that. GraalVM has a better escape-analysis 
logic (known as "partial escape analysis") which helps out in that case; 
Valhalla completely side-steps the issue by marking, at declaration 
site, which classes can reliably avoid allocation.

Since, usual disclaimer, I'm not a VM engineer, a muuuch better 
explanation on the topic, with some benchmarks, written by Alex 
Shipilev, can be found here:

https://shipilev.net/jvm/anatomy-quarks/18-scalar-replacement/

Maurizio