x64 JIT support for virtual threads (Loom) #15552
Conversation
|
Jenkins test sanity all jdk17,jdk19 |
|
Jenkins test sanity xlinux,xmac,win jdk17 |
|
Jenkins test sanity xlinux jdk17,jdk19 |
* thisThreadGetOwnedMonitorCountOffset() * thisThreadGetCallOutCountOffset() Signed-off-by: Daryl Maier <maier@ca.ibm.com>
* callOutCount * ownedMonitorCount Signed-off-by: Daryl Maier <maier@ca.ibm.com>
Required for virtual thread support. Signed-off-by: Daryl Maier <maier@ca.ibm.com>
Only required for hard realtime support, which has been long deprecated. Signed-off-by: Daryl Maier <maier@ca.ibm.com>
This label is not needed any longer and its presence is interfering with adding monitor counters for virtual threads. Signed-off-by: Daryl Maier <maier@ca.ibm.com>
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
Required for virtual thread support. Signed-off-by: Daryl Maier <maier@ca.ibm.com>
|
Jenkins test sanity win,xlinux,osx jdk17,jdk19 |
|
@BradleyWood @r30shah : could I ask you to review this please? Additional internal testing on JDK 17 and 19 on all platforms has succeeded. |
| @@ -1350,6 +1350,18 @@ TR::Register *J9::X86::AMD64::JNILinkage::buildDirectJNIDispatch(TR::Node *callN | |||
|
|
|||
| if (isGPUHelper) | |||
| callNode->setSymbolReference(callSymRef); //change back to callSymRef afterwards | |||
|
|
|||
| #if JAVA_SPEC_VERSION >= 19 | |||
There was a problem hiding this comment.
On monitor entry and exit code, we only update the counter if it is >= Java19 and 64-Bit platform, should we be guarding the code with 64-Bit platform check as well?
There was a problem hiding this comment.
No. This file is specific to 64-bit only (AMD64) and an additional 64-bit check would be redundant.
There was a problem hiding this comment.
I should have look at that more carefully, thanks a lot for pointing this out.
Though, I actually want to see if @gacholio's suggestion on Z PR for Loom #15752 (comment) also applies to all other codegens?
There was a problem hiding this comment.
This is really a policy decision, not a technical one. It's quite unlikely that 32-bit will make a return, though Z is perhaps the one place where it might be requested.
There was a problem hiding this comment.
Thanks a lot @gacholio for the explanation. It makes sense.
|
@joransiu since you spent time reviewing the corresponding Z changes I'm wondering if you'd be able to merge this PR when you have a chance? Thanks! |
|
@joransiu : thanks for the review. Would you mind merging please? Rahil has approved it as well. |
ownedMonitorCountJ9VMThread field on monitor enter/exitcallOutCountJ9VMThread field for JNI dispatchIssue: #15175