my test PR #1
Open
my test PR #1
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llxia
pushed a commit
that referenced
this pull request
May 31, 2023
Disable CLassLookahead because it is possible for static final field value to be changes via reflection or the unsafe API. 1. Disable ClassLookahead. Similar to GVP, ClassLookahead is using static analysis of static final field initialization code to provide field information that is later used to make assumptions about the field contents which might not be true if reflection or the unsafe API is used to change static final field values. An environment variable is introduced to allow ClassLookahead to be re-enable. 2. Apply a fix to ClassLookahead so that it does not not generate field information for static final fields that can be initialized with different values on different code paths. This fix is being applied in case the ClassLookahead code is resurrected in the future since `#1` above has disabled it globally. 3.Remove the code in IlGen what would replace array length checks with hard coded obj array length values. This code would look for static final fields that contain ClassLookahead fieldInfo and use the existence of that information to permit the inspection of a heap object and replace array length checks with a constant based on the current heap object's array length. This fix is also applied as a safety measure because the disabling of ClassLookahead (`eclipse-openj9#2` above) would effectively disable this code path. In general the only safe way to inspect at compile time heap objects pointed to by static final fields is to us the "Static Final Field Folding" optimization which uses OSR to invalidate any code which made an assumptions based on the contents of a static final field when that fields contents may have been modified by reflection. Signed-off-by: Kevin Langman <[email protected]>
…e_RB_copy_stats Read Barrier bytes stats
Signed-off-by: Jason Feng <[email protected]>
Use add() (not addU64) with uintptr_t RB counters. Signed-off-by: Aleksandar Micic <[email protected]>
…e_getReferenceField Delete unused JITServer message `KnownObjectTable_getReferenceField`
…e_RB_copy_stats Match operation with the type
JDK25 adds JVM_CreateThreadSnapshot, removes isCarrierThreadLocalPresent
This change fixes a type mismatch when compiled on 32-bit. Signed-off-by: Adrian Popescu <[email protected]>
Tracepoints don't evaluate their arguments if UT_TRACE_OVERHEAD <= 0. Signed-off-by: Keith W. Campbell <[email protected]>
Add NULL check for moduleObject in GC
Fix type mismatch on 32-bit
Add "if defined(J9VM_GC_SPARSE_HEAP_ALLOCATION)" guard to the recognized call transformer for vectorizedMismatch's offheap-exclusive code. Signed-off-by: Luke Li <[email protected]>
…orizedMismatchFix Add Guards to the OffHeap-exclusive Code for vectorizedMismatch Transformer
If a monitor is already detached, do not detach it again. The detached monitor is already in enteredMonitors list.Do not modify its J9ObjectMonitor->next which will overwrite the extsing entry. issue eclipse-openj9#21864 Signed-off-by: Hang Shao <[email protected]>
…ngClassFromFieldOrStatic Use the actual declaring class for static final field folding
- Use J9_LOCK_IS_FLATLOCKED macro instead of count check - Move reserved check before setting newLockWord Signed-off-by: Jack Lu <[email protected]>
Separate side-effects from tracepoint use
Casting to arrays was considered unlikely and therefore not fast-pathed in JITed code. However, checkcasts of non-primitive references to j/l/O arrays should always succeed and can be easily checked for without calling a helper. Modify inline checkcast to do so. Signed-off-by: Daryl Maier <[email protected]>
…etDeclaringClassFromFieldOrStatic Suppress getDeclaringClassFromFieldOrStatic() in JITServer non-SVM AOT
Virtual memory amount for ZOS31 is limited to 2G, Any attempt to allocate larger memory ends in abend S878 system failure. Too large size for -Xmx can not be calculated by default, it only can be provided in command line option -Xmx. So, we need to check that virtual memory request is not larger than 2047m (leave 1m for extra protection) anf fail JVM startup if it is requested too large. Please note that real amount of the virtual memory for allocation is much smaller than 2047m (about 1700m). An attempt to allocate more memory leads to normal JVM startup failure. The newly introduced error message is exactly the same as that failure message. Signed-off-by: Dmitri Pivkine <[email protected]>
- maintain a cache of all classes that are needed for boxing and unboxing - classes are passed to intrinsics, record location of class parameters in the table - make sure result type of an intrinsic is stored in the table and used for boxing/unboxing - record result type of 'compare' in the methodTable instead of the source type
…pes_pr Box and unbox all types in VectorAPIExpansion
Do not re-detach a monitor that is already detached
setCallCount() and getCallCount() is a pair of interpreter profiler routines used by the optimizer indirectly to derive block frequencies. However, block frequencies can/should be inferred solely based on branch information provided by IProfiler. Moreover, setcallCount() generates "artificial" IProfiler entries based on normalized basic block frequencies which may not be directly comparable to raw frequencies coming from the IProfiler. Signed-off-by: Marius <[email protected]>
This file is removed in OMR. Signed-off-by: Spencer Comin <[email protected]>
Signed-off-by: Peter Shipton <[email protected]>
If the operation that's being profiled is instanceof or checkcast, the code in the quickTest block and the helper call block should test the result of the load of the vft-symbol for profiling. The previous change only added the indirect load of the vft-symbol in the IL generated for the helper call. This change adds the missing load of the vft-symbol to the quickTest block as well. Signed-off-by: Henry Zongaro <[email protected]>
…xOfStr AArch64: Inline StringLatin1.indexOf([BI[BII)I
…ffHeap AArch64: Add support for off-heap arrays in some intrinsics
Fix JNI return types
The OpenXL replacement for `-fno-eliminate-unused-debug-types` is `-qdbxextra`, but only for version 17.1 or newer. Signed-off-by: Keith W. Campbell <[email protected]>
Update DDR compile options for OpenXL
Enable bigCallee size adjustment during ECS for JDK8 and 11
Modify scavenger GC count
This commit reverts the inlineIntrinsicIndexOf() part of PR eclipse-openj9#22192, due to the failure with JDK 24. Fixes: eclipse-openj9#22205
AArch64: Revert eclipse-openj9#22192 partially
…ects Fix Vector API boxing defects
The initialization of classObject of J9Class is done by loadWarmClassFromSnapshot() and make initializeSnapshotClassObject() redundant. Other modification includes if the call is failed, go to exit or error processing. Fixes: eclipse-openj9#22201
Enable JFR SystemProcess test on z/OS
- Improve assert about the type of intrinsic operand - Correct return type of Convert intrinsic since it can it be Vector or Mask
Apply fixes in VectorAPIExpansion
RAM Class Persistence: Remove redundant call to initializeSnapshotClassObject
Only set VirtualThread.blockPermit for a virtual thread in the BLOCKING state when the lock is available. This prevents virtual threads from being unblocked prematurely. Signed-off-by: Theresa Mammarella <[email protected]>
…test Set blockPermit for BLOCKING thread when lock is free
Check the lastITable and iTable before using interface lookup helper. First check the lastITable if it fails, check iTable entries. If any entry matches, calculate the entry point and jumps to the entry. If no entry matches, calls the helper method. Signed-off-by: Ehsan Kiani Far <[email protected]>
Previously when performing Class.isAssignableFrom, only the trivial case of class equality was done inline and any other case relied on a callout to a C Helper. With this change, an inlined superclass test is added along with a castClassCache test, avoiding the C Helper in the most common cases. Additionally, this change checks for class depth at compile time when possible to fast-fail when fromClassDepth <= toClassDepth. Signed-off-by: Matthew Hall <[email protected]>
To faciltate a non-compilation thread invoking the deserializer, which can happen outside the context of a compilation, the APIs in the Deserializer need to be updated to no longer require a TR::Compilation object directly. Signed-off-by: Irwin D'Souza <[email protected]>
…rom_genSuperclassTest Generate inline superclass test for Class.isAssignableFrom on Z
Add AOTCache datastructure diagram
Build 118 has been rotated out. Update to 123. Signed-off-by: Adam Brousseau <[email protected]>
…k15_update Bump to newer jdk24 aix build for bootjdk
Refactor the Deserializer to use a helper struct
resolves: eclipse-openj9#22232 Signed-off-by: Lan Xia <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.