Object inlining, aka object fusing
raffaello.giulietti at supsi.ch
Wed Jun 29 13:36:53 UTC 2016
Well, it seems to me that ObjectLayout has a stricter semantics than
object inlining as advocated in Christian Wimmer's work. The latter is
far more flexible and transparent.
Of course, that also means that ObjectLayout probably requires less
implementational efforts than object inlining.
It's a pity that object layout could not make it into the official
On 2016-06-29 00:17, Tom Rodriguez wrote:
> The ObjectLayout stuff from Azul presupposes control over the grouping of objects in a way that’s seems compatible with object inlining. Maybe if that’s ever supported in hotspot it could be leveraged for this. It would probably depend on how flexible the implementation in the GC is.
>> On Jun 28, 2016, at 4:32 PM, Christian Wimmer <christian.wimmer at oracle.com> wrote:
>> The object inlining was a research prototype (like many other PhD theses). It was never intended to be production-ready, so it is not in the product version. It was also way before Graal got started (object inlining is implemented in the client compiler only). Graal would certainly have saved me some time back in the days...
>> On 06/28/2016 01:46 PM, Raffaello Giulietti wrote:
>>> Hi all,
>>> this is perhaps not the right place for questions related to object
>>> memory layout in HotSpot, but I dare to ask anyway given the tremendous
>>> expertise about this JVM internals by the Graal developers reading this
>>> Some years ago, Christian Wimmer and Hanspeter Mössenböck conducted
>>> research on object inlining
>>> (http://www.ssw.uni-linz.ac.at/Research/Projects/JVM/AOI.html) and
>>> experimented the idea in HotSpot.
>>> Anybody knows whether their code managed to be officially integrated in
>>> the stock HotSpot VM, and thus also in the JVMCI extended version used
>>> by Truffle/Graal?
More information about the graal-dev