Please review JDK-8059321

[Attila Szegedi]

Collections.synchronizedMap is okay; I usually go with a single explicit synchronized block encompassing both operations; this way you'll end up with two MONITORENTER/MONITOREXITs, but it shouldn't matter much. Maybe HotSpot does some lock coarsening itself. Again, doesn't matter much - this is fine too.

lookupInternalType does "cache = INTERNAL_TYPE_CACHE;" and then inconsistently uses both the local variable and the field subsequently. Eliminate one.

Now that you have a set instead of a map for VALID_CACHE_SET, you could use the nicely atomic

if (cache.add(key)) {
  // do stuff 
}

idiom instead of

if (cache.contains(key)) {
   return;
}
cache.add(key);
// do stuff

(and then you also don't need a local variable but can just use the field as "if (VALID_CACHE_SET.add(key)) { ... }"

If you do these, then +1 - no need for a new webrev.

Attila.

On Sep 29, 2014, at 8:45 PM, Marcus Lagergren <marcus.lagergren at oracle.com> wrote:

> OK. New webrev here http://cr.openjdk.java.net/~lagergren/8059321.2/webrev/
> 
> I experimented a bit with synchronization methods, and this seems to be the one that gives the least overhead - there is actually very little difference and 95% of the original performance increase is preserved.
> 
> (I also experimented with your ‘one extra recompile’ in CompiledFunction, and applied that diff - this brings us down another ~600 ms, which is nice indeed).
> 
> Let me know if this is semantically sound. From reading the OpenJDK code, I think it is.
> 
> /M
> 
> 
> On 29 Sep 2014, at 11:22, Aleksey Shipilev <aleksey.shipilev at oracle.com> wrote:
> 
>> Yes, that's a simple adapter:
>> http://docs.oracle.com/javase/7/docs/api/java/util/Collections.html#newSetFromMap(java.util.Map)
>> 
>> See the example there.
>> 
>> -Aleksey.
>> 
>> On 09/29/2014 10:10 PM, Marcus Lagergren wrote:
>>> Aleksey - I still need the weak semantics, because I don’t want to hold on to the strings. Would that work for the WeakHashMap and preserve semantics? #iamnotajavaprogrammer.
>>> 
>>>> 
>>>> The entire shenanigan would go away if you turn the Map into Set with
>>>> Collections.newSetFromMap(...), and then do add().
>>>> 
>>>> -Aleksey.
>>>> 
>>>> 
>>> 
>> 
>> 
>