VPKnownObject Constraints for JITServer #7565
Conversation
compiler/optimizer/VPConstraint.cpp
Outdated
| vp->addConstraint(constraint, hash); | ||
| } | ||
| } | ||
| TR_J9VMBase::ObjectClassInfo ci = vp->comp()->fej9()->getObjectClassInfoFromKnownObjectIndex |
There was a problem hiding this comment.
Where is TR_J9VMBase::ObjectClassInfo defined?
There was a problem hiding this comment.
Note that getObjectClassInfoFromKnownObjectIndex() already exists, but does not do exactly what the code it replaces does.
There was a problem hiding this comment.
The ObjectClassInfo is defined under class TR_J9VMBase in VMJ9.h, I see a few other structs defined like that as well.
luke-li-2003
force-pushed
the
ServerVPKnownObject
branch
from
9286f68
to
c9499b1
Compare
compiler/optimizer/VPConstraint.cpp
Outdated
| } | ||
| } | ||
| TR_OpaqueClassBlock *clazz = vp->comp()->fej9()->getObjectClassFromKnownObjectIndex | ||
| (vp->comp(), index, isJavaLangClass); |
There was a problem hiding this comment.
This implementation eliminates the assert that made sure that, whenever the parameter isJavaLangClass is true, clazz is the class the java/lang/class
|
This PR depends on eclipse-openj9/openj9#20695 |
luke-li-2003
force-pushed
the
ServerVPKnownObject
branch
2 times, most recently
from
1a9e704
to
a855f6d
Compare
compiler/optimizer/VPConstraint.cpp
Outdated
| bool matchJavaLangClass; | ||
| TR_OpaqueClassBlock *clazz = vp->comp()->fej9()->getObjectClassFromKnownObjectIndex | ||
| (vp->comp(), index, &matchJavaLangClass); | ||
| TR_ASSERT(matchJavaLangClass == isJavaLangClass, |
There was a problem hiding this comment.
I would prefer this assert to be made FATAL.
Also, I propose the following message:
"Use createForJavaLangClass if and only if the object is an instance of java/lang/Class;
compiler/optimizer/VPConstraint.cpp
Outdated
| @@ -1385,7 +1375,7 @@ TR::VPConstString *TR::VPConstString::create(OMR::ValuePropagation *vp, TR::Symb | |||
| TR::VMAccessCriticalSection vpConstStringCriticalSection(vp->comp(), | |||
| TR::VMAccessCriticalSection::tryToAcquireVMAccess); | |||
|
|
|||
| if (vpConstStringCriticalSection.hasVMAccess()) | |||
| if (vpConstStringCriticalSection.hasVMAccess() && 0) | |||
There was a problem hiding this comment.
I guess these "&& 0" are hacks that will be removed
| if (vp->comp()->isOutOfProcessCompilation()) | ||
| return false; | ||
|
|
||
| TR::KnownObjectTable *knot = vp->comp()->getKnownObjectTable(); |
There was a problem hiding this comment.
We should keep this test that makes sure knot exists
luke-li-2003
force-pushed
the
ServerVPKnownObject
branch
4 times, most recently
from
e764ea2
to
55816c2
Compare
Ensure VPKnownObject works in server mode with appropriate queries. Signed-off-by: Luke Li <[email protected]>
luke-li-2003
force-pushed
the
ServerVPKnownObject
branch
from
55816c2
to
e9ba29c
Compare
|
I think we can make this PR ready for review by the OMR committers. |
|
jenkins build all |
|
Jenkins build all |
|
@mpirvu may I ask you to formally approve when you are good with these changes ? Since you were having a conversation with Luke. |
Ensure VPKnownObject works in server mode with appropriate queries.