eclipse-openj9
X: implement general case multianewarrayEvaluator for 2D arrays
Currently multianewarray is only implemented on x86 for 2D arrays with zero first or second dimension, otherwise calling out to a VM helper. This PR adds logic to do allocation and initialisation inline for nonzero 2D arrays.
@ehsankianifar @luke-li-2003 @zl-wang could I get your review since you've looked at multianewarray for Z & P?
Internal sanity testing: https://hyc-runtimes-jenkins.swg-devops.com/job/Pipeline-Build-Test-Personal/29776/
I haven't added support for off-heap in this PR. I will need to add logic to special case the zero length arrays for that, so I am planning on doing that in a future change.
Benchmarking with this change is showing a regression in the dacapo graphchi benchmark, reverting to draft while I investigate that
I haven't run 32-bit testing, but I'm assuming multianewarrayEvaluator is only supported on 64-bit based on https://github.com/eclipse-openj9/openj9/blob/bbd11673d4f8f34489254feb9aa687e32fce6159/runtime/compiler/x/codegen/J9TreeEvaluator.cpp#L1762
No great rush, but can the reviews for this keep moving forward ? It would be nice to have it in a nightly build and do a fresh batch of DaCapo runs (with several other changes already merged now).
Just a reminder to keep the next code complete date (3 weeks or so away I think) in mind, and aim to get this in well before that date, to avoid missing the next release.
@Spencer-Comin @hzongaro : who owns the next step on this PR? Our goal is to have this in for the 0.57 code split on Nov. 14. Thank you.
I am addressing testing failures, unless I get those resolved tomorrow unfortunately it is not likely this will make the 0.57 split
I am working on fixing register dependency issues between the OOL helper call and the mainline. I've gotten access to Bob, so hopefully that provides some new insights, but Bob has told me "This is extremely frustrating".
Can you share what problems you're seeing? I suspect "Daryl" might do a better job than "Bob" for this kind of problem.
@0xdaryl my last attempt at resolving the problem hasn't failed local testing yet. I'll update this PR to reflect my development branch in a few minutes once that completes, and point out where I'm seeing problems.