Changes

Change Node::mayUse() function to return TR_UseOnlyAliasSetInterface
This commit make Node::mayUse() function to initialize the 
AliasInterface from the TR_SymAliasSetInterface class directly instead 
of from the TR_NodeUseAliasSetInterface class. This helps remove the 
NodeAliasing class.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: d84ed6f)

Change Node::mayKill() function to return TR_UseDefAliasSetInterface
This commit make Node::mayKill() function to initialize the 
AliasInterface from the TR_SymAliasSetInterface class directly instead 
of from the TR_NodeKillAliasSetInterface class. This helps remove the 
NodeAliasing class.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: ea2fdde)

Remove Node Alias
Remove all the references and definitions of
TR_NodeKillAliasSetInterface and TR_NodeUseAliasSetInterface.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 18abdd2)

Add isRelocatable boolean inside loadAddressConstant
This boolean can be set by the caller to indicate that the instructions
generated by loadAddressConstant need to be relocated.
The reason we use this boolean is because the 
cg->comp()->compileRelocatableCode() query may not always return the 
correct answer for when relocatable code needs to be generated. For
example: a regular JIT compile done remotely. So for now, we allow the
caller to make the decision on if the instructions needs to be
relocatable when not in an AOT compile.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 27d55ea)

Correctly use loadAddressConstant API for lookup IL
lookup schmemes 3 and 4 use the loadAddressConstant API to materialize 
addresses into a register. These instructions need to be relocatable. 
This commit specifies the boolean parameter correctly in order to
indicate to the API that relocation for the instructions is necessary.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: bf75a9c)

Specify isRelocatable bool for TR_DataAddress
When generating a TR_DataAddress contant using the loadAddressConstant
API in accessStaticItem, we must provide the correct value for the
isRelocatable parameter. This will allow the routine to create
relocatable instructions for remote compilations.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 46db497)

Don't do local relocation for remote compilations
Doing a local relocation is problematic because when we do a relocation
on the client, if there is already data in the location that needs to be
patched, then we end up with garbage. This is becase we expect the patch
location to be populated with zeroes and at patch time we OR the new
data in (on Power).
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 4f716ae)

Generate relocatable code in emitFloatingPointConstant
For remote compilations, emitFloatingPointConstant emits unrelocatable
code. This commit addresses that issue.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: f26c557)

Don't emit data for remote compilations
For remote compilations, we must not emit the actual address when
generating a fixed sequence on power during a remote compilation. This
is because during relocation time on the client, the new data will be
OR'ed in. This infrastructure depends on the fact that the data in the
patch location is initially zero.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 858f77e)

Generate relocation record in ConstantDataSnippet
When emitting an addressConstant, a relocation record must be created
for ConstantDataSnippet. This commit adds the appropriate check to add a
relocation when doing a remote compile.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 35705c4)

Add relocation record for TR_DataAddress in JITServer
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 6608d75)

Remove OMR::ILOpCode::unsignedShiftLeftOpCode function
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 57977c0)

Concurrent GC Yield and Resume after CM GC Exclusive
Previously, after terminating and yielding to an exclusive access 
request, Concurrent Scavenger would only resume if it yielded to an 
external VM access request. However, CS should also resume after 
yielding to internal GC request during CM init as it completely 
terminates CS and results in missed opportunity for CS to make progress.
- upon GC termination request, `_concurrentState` is no longer 
transitioned to `concurrent_state_complete`, it's left in scan state.
- Introduced `_concurrentResumable` configuration param for
`MasterGCThread` which is set to true if the collector is resumable.
(initialized to true for Scavenger, defaults to false for balanced)
Signed-off-by: Salman Rana <salman.rana@ibm.com>
 (commit: 7a612ac)

Add OMR_THR_MCS_LOCKS flag to enable/disable the MCS lock
OMR_THR_MCS_LOCKS flag will be used to enable/disable the usage of the 
MCS lock in the OMR thread monitor.
Related: https://github.com/eclipse/omr/pull/4086.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: fa1f0e0)

Fix formatting of ifdefs
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 8d65b32)

Initialize variables, fix null checks and use OMR_ARE_*_BITS_SET macros
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 75e0da2)

Remove unnecessary NULL checks around omrthread_free_memory
omrthread_free_memory does an internal NULL check. So, wrapping 
omrthread_free_memory calls with a NULL check is irrelevant. Such NULL 
checks have been removed.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 7ad834a)

AArch64: Set destination address in ARM64ImmSymInstruction
This commit adds code for setting the destination address of "bl" 
instruction in ARM64ImmSymInstruction when possible, to improve the 
readability of trace files.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: 9c28af4)

Keep contracted LOA size not smaller than minimum LOA size
	- Keep LOA size not smaller than minimum LOA size during
resizeLOA and resetLOA, expandWithRange, redistributeFreeMemory
to avoid potential crash on some	platforms.
- Add assertion check for the case.
- Enable size heapAlignment assertion by default.
Signed-off-by: Lin Hu <linhu@ca.ibm.com>
 (commit: 3c7df4a)

Fix checkAndReplaceRotation 32/64 bit ambiguity
checkAndReplaceRotation tests for pattern validity by comparing an
expected shift amount based on the size of the operand with the constant
from the multiply node:
if ( (1 << correctLeftShiftAmount) != mulConst )
     return false;
Unfortunately, the implicit int constant '1' allows the C++ compiler to
ignore the possibility of shift amounts>32 (undefined behaviour),
resulting in 1<<n giving the same answer as 1<<(n+32). This allowed the
simplification to proceed on a 64 bit operand with shift values that
were only valid for 32 bit operands.
The shift has been updated with code crafted by Leonardo which allows
the test to correctly differentiate between 32 and 64 bit valid
patterns.
Signed-off-by: James Kingdon <jkingdon@ca.ibm.com>
 (commit: b5ea61d)

Fix the compile failure related to the lockExchange function on AIX
Details: 1540-0256 (S) A parameter of type "volatile unsigned long *"
cannot be initialized with an expression of type "volatile long *". 
1540-1205 (I) The error occurred while converting to parameter 1 of
"__fetch_and_swaplp(volatile unsigned long *, unsigned long)".
Related: https://github.com/eclipse/omr/pull/4469.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 18c567b)

remove generation of deprecated unsigned load opcodes
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 946b00e)

Remove generations of deprecated unsigned store opcodes
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 67c1af8)

Add relocation records correctly inside ConstantDataSnippets
Use FE query to decide if data represented by RamMethodConst and
JNITargetAddress need relocation records.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: b289414)

Create TR_MethodCallAddress relocation for resolved methods
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: d743c95)

Disable ExpandInstructionsPhase on codegens that don't use it
Previously, ExpandInstructionsPhase would iterate over every instruction 
and call a virtual method on each, even if the codegen in question 
wasn't making use of it. Since only the Power codegen currently has 
plans to make immediate use of this phase, other codegens do not need to 
be doing this. To avoid performing unnecessary work, expandInstructions 
is now stubbed out on all codegens other than Power.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: bcc882c)

Handle __fetch_and_swap(lp) across XLC versions used by OMR consumers
For XLC versions < 13.1.3, the following __fetch_and_swap(lp) 
definitions are supported: i) __fetch_and_swaplp(volatile long *, long) 
ii) __fetch_and_swap(volatile int *, int)
For XLC versions >= 13.1.3, the following __fetch_and_swap(lp) 
definitions are supported: i) __fetch_and_swaplp(volatile unsigned long
*, unsigned long) ii) __fetch_and_swap(volatile unsigned int *, unsigned
int)
Co-Authored-By: Robert Young <rwy0717@gmail.com>
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 5166fff)

AArch64: Implement PrefetchEvaluator
This commit implements a PrefetchEvaluator for AArch64. It does not
generate "PRFM" instructions yet.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: b725d82)

Use linkage entry point API for recursive method calls
When encoding a recursive compiled method call instruction, ask the 
linkage for the compiled method entry point.
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: 9754e65)

Remove deprecated X86 outlined allocation helpers
Not used in OMR nor any known downstream project.
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: 5a94db8)

Remove Unused Options
TR_EnableLargePages and TR_EnableLargeCodePages are not being used
anywhere. They might be remnants of removed code.
Signed-off-by: AlenBadel <Alen.Badel@ibm.com>
 (commit: d2daaaf)

Refactor callsite constructor macro
The old macro doesn't support callsite subclasses for defining more 
complicated constructor other than initailizing the fields that exists 
in the super class. Adding this ability by breaking the single macro 
into three parts 1. TR_ALLOC 2. constructor signature and super classes 
field initialization 3. constructor method body
Signed-off-by: Yi Zhang <yizhang@ca.ibm.com>
 (commit: 38a8038)

Refactor the optimizer to allocate UseDefInfo through a central location
This change implements the recommended refactoring suggested by
@andrewcraik in the review comments for 
https://github.com/eclipse/omr/pull/2823.
The background is that we need to allow setting of the _hasCallAsUses 
flag depending on the language front-end. However this is currently hard 
coded to false. Following is a copy of Andrew's recommendation.
Setting _hasCallAsUses should be handled as part of the constructor if 
it needs to be configured by different languages. This is the first time 
we are going to have a language specific override of a flag like this 
for UseDefInfo. Looking at how it is currently used instances of 
TR_UseDef_info are allocated in the OMROptimizer.cpp, 
ValueNumberInfo.cpp and LoadExtensions.cpp. There is some variation 
between them in how they configure UseDefInfo for globals. I think the 
best way to handle this is to create a new API on the Optimizer object
(eg OMROptimizer.hpp and OMROptimizer.cpp) which all current allocation 
sites switch to calling - the method will accept as parameters the bools 
which vary between the different request sites within the compiler. The 
value for _hasCallAsUses could be either stored as a flag in the 
optimizer or by overriding the build method to change how UseDefInfo is 
built. I would prefer the later since it means languages have a common 
place to configure UseDefInfo according to their needs. This small 
refactor would ideally be done as a separate commit and pull request 
ahead of changes introducing the new flag to the constructor. I hope 
that makes sense - if it doesn't do keep asking questions and I'll try 
to clarify more on how this would work.
An added bonus of this is that we could improve the UseDefInfo memory 
management in a follow-on item which would be a another good reason to 
Refactor the optimizer to allocate UseDefInfo through a central 
location.
End quote.
Note that this change does not address the construction of 
TR_OSRDefInfo, which extends the TR_UseDefInfo class.
To allow language front-ends to control this flag when use defs are 
analysed we need a callsAsUses flag to be held by the Optimizer and we
need methods to get and set the flag. This flag will be used when
constructing TR_UseDefInfo objects.
The callsAsUses parameter is controlled by a virtual method in the 
Optimizer that front-end language implementations may override to change
the default behaviour.
Also the TR_UseDefInfo constructor has been made protected to avoid 
users making a mistake trying to construct it directly.
Signed-off-by: Dibyendu Majumdar <mobile@majumdar.org.uk>
 (commit: 775d7cf)

Forward format args to trace function in nonfatal assertion
TR::Assertion is passing the format string, but not the variadic 
arguments, to fatal assert. This is causing crashes in the printing 
helper.
This commit introduces a new function, va_fatal_assertion, which accepts 
a va_list argument. Both assertion and fatal_assertion call this helper.
Using this helper, the arguments should be forwarded along correctly to 
the printing helper.
Fixes: #4423 Signed-off-by: Robert Young <rwy0717@gmail.com>
 (commit: 3a66828)

Remove avl_internal.h
Header does not provide anything
Signed-off-by: Devin Nakamura <devinn@ca.ibm.com>
 (commit: 446aa57)

AArch64: Set first instruction to block in BBStartEvaluator
Set first instruction to the block of the node.
Signed-off-by: Akira Saitoh <saiaki@jp.ibm.com>
 (commit: 62eb789)

AArch64: Add floatRemainder and doubleRemainder helpers
Add floatRemainder and doubleRemainder helper entries.
Signed-off-by: Akira Saitoh <saiaki@jp.ibm.com>
 (commit: 28444ec)

Include correct concrete class header files
There exist many instances where we do not include the concrete 
implementation of the headerfile. Fix all instances.
Fixes #3973
Signed-off-by: Aidan Ha <qbha@edu.uwaterloo.ca>
 (commit: 51665bb)

Add AArch64 helpers to getRuntimeHelperName()
This commit adds AArch64 helpers to getRuntimeHelperName() in Debug.cpp.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: c62b512)

Remove all the definitions of deprecated ILOpCode TR:: iumul
Remove all of its definitions, including its declaration, property and 
all other lists/enums that defines the behavior of thos ILOpcode. Also
removed functions: ARM64::iumulEvaluator(), constrainIumul() TR::iumul
is completely removed after this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 6ff92ce)

Rewrote parser in tril
Signed-off-by: Yuehan-Lin <Yuehan.Lin@ibm.com>
 (commit: 393044a)

Created Function Prototypes for OMRSocketAPI
Created Unimplemented Function Prototypes in omrsock.c
-Modified omr files to be able to see and compile omrsock code
-Added new omrsock.c file for common for all operating systems, this 
includes function prototypes proposed by the Socket API proprosal
-Added omrsock variables needed by the function prototypes. They are 
created in omrportsock.h
Issue: #4102
Co-authored-by: Babneet Singh <sbabneet@ca.ibm.com>
Signed-off-by: Haley Cao <haleycao88@hotmail.com>
 (commit: cf8987d)

Remove isAOT check in order to enable relo logging for JITServer
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 19eb6bf)

Fix missing defines in Power SPP files when building with CMake
For legacy reasons, the assembly files for Power are run through the C 
preprocessor separately from when they are assembled. However, some 
defines (namely platform ones such as TR_HOST_64BIT) were not being 
correctly passed to the C preprocessor when running these assembly files 
through it when building with CMake. This has now been corrected.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: b199666)

Fix OMRMonitor.cpp failing to compile on AIX
Previously, OMRMonitor.hpp contained a workaround for an XLC issue with 
private delete operators in it. However, this workaround was only used 
when compiling on z/OS, meaning that attempting to build the OMR 
compiler on AIX with XLC would fail with an error in OMRMonitor.cpp. 
This workaround has now been enabled on AIX.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: e355f32)

Remove all the definictions of deprected ILOpCode TR::cmul
There is no definitions and declarations to this opcode. Removed all 
commentted codes that reference this TR::cmul. Also removed 
cmulEvaluator() functions. TR::cmul is completely removed from OMR after 
this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: f6a932c)

Remove all the definitions of deprecated ILOpCode TR::buternary
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of this ILOpCode. TR::buternery
will be completely removed after this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: eed978c)

Remove all the definitions of deprecated ILOpCode TR::iuternary
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of this ILOpCode. TR::iuternery
will be completely removed after this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 7b0e52e)

Remove all the definitions of deprecated ILOpCode TR::luternary
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of this ILOpCode. TR::luternery
will be completely removed after this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 6695ff8)

Remove all the definitions of deprecated ILOpCode TR::suternary
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of this ILOpCode. TR::suternery
will be completely removed after this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 01d3018)

Enable Tril and test compiler hacks on AIX
Previously, there were a number of hacks to work around xlc issues in 
Tril. However, these hacks were not being enabled on AIX since they were 
explicitly guarded by checking J9ZOS390, as z/OS was previously the only 
platform supported by Tril that used xlc. In order to allow Tril to 
compile on AIX, these hacks have now been enabled on AIX.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 107365c)

Fix mmap with MAP_ANONYMOUS not working on AIX
It turns out that AIX's implementation of the mmap function requires 
that the file descriptor be set to -1 when using MAP_ANONYMOUS. 
Previously, there were a couple places that were trying to use 
MAP_ANONYMOUS without setting the file descriptor to -1, resulting in 
them erroring out. This has been corrected and they should now work 
correctly.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 675f519)

Fix incorrect use of PRId64/PRIu64 in ShiftAndRotateTest
Previously, a number of places in ShiftAndRotateTest were using PRId64 
and PRIu64 directly in their format strings. However, this does not work 
correctly on AIX. These have been corrected to use OMR_PRId64 and 
OMR_PRIu64 instead.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: ca854e6)

Make Tril::SimpleCompiler return a function descriptor on AIX
On most platforms, function pointers are simply pointers to the first 
instruction in the function. However, on AIX and z/OS this is not true: 
they are instead pointers to data structures that themselves contain 
such a pointer. Support for calling Tril-compiled functions has now been 
added on AIX by adding the ability for Tril::SimpleCompiler to create 
such a function descriptor on AIX.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 9146ec1)

Skip Tril tests affected by #3525 on Power BE
A number of Tril tests are currently affected by #3525, in which the 
code generators spill sub-integer arguments to stack slots incorrectly. 
While this works correctly on little endian machines, it does not work 
on big endian machines. Previously, such tests were already being 
skipped on Z, but this issue also affects the Power codegen when run on 
a big-endian machine. In order to allow Tril tests to be run on AIX, 
such tests are now skipped when run on Power BE.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 5240cd3)

Fix tril_compiler and tril_dumper failing to link on AIX
Due to failing to link against libpthread, neither the tril_compiler nor 
tril_dumper executables would link correctly on AIX. This has been 
corrected by forcing the tril library itself to link against libpthread, 
which in turn forces tril_compiler and tril_dumper to link against it as 
well.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 7e7a800)

Add hacks for running JitBuilder on AIX
Previously, JitBuilder did not work properly on AIX due to similar 
problems to what Tril was previously facing. Since enabling Tril tests 
also enables JitBuilder tests, it is necessary to ensure that JitBuilder 
works at least minimally on AIX before Tril tests can be run in pull 
request builds for AIX.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 24bad33)

Disable Tril's IlGenTest on AIX
The IlGenTest from Tril's internal tests is currently broken on AIX, 
since it is not valid on AIX to use the return value of 
compileMethodFromDetails as a function pointer. Until such a time as 
this test is fixed, it has been temporarily disabled on AIX.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: f816690)

Disable test compiler tests on AIX
The test compiler tests make the assumption that function pointers can 
simply be called using the regular call opcodes in the IL. This is not 
true on AIX, since function pointers do not actually point to the 
address of the first instruction of the function on AIX. As a result, 
the test compiler tests will always segfault when run on AIX. However, 
since the test compiler tests are effectively deprecated and were never 
previously run on AIX in the first place, this is not considered to be 
something worth fixing. Thus, the test compiler tests will no longer be 
run on AIX when using ctest.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: c29da09)

Enable the JIT for AIX CI jobs
Previously, the AIX pull request and build CI jobs explicitly disabled 
the JIT and all related components. However, recent work has now made it 
possible to compile the OMR JIT and run JitBuilder and Tril tests on 
AIX. Since this now works correctly, the JIT is no longer disabled in 
the AIX CI jobs.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 3a9a956)

Remove Flex/Bison from AppVeyor and Travis
With the merge of #4345 we no longer depend on Flex and Bison being 
installed for our testing. AppVeyor seems to be broken at the moment in 
that checksums are failing when installing this package. We fix the 
issue by removing the dependency as it is no longer needed.
Signed-off-by: Filip Jeremic <fjeremic@ca.ibm.com>
 (commit: 578eb17)

Remove the last few mentions of flex/lex/bison/yacc
Signed-off-by: Robert Young <rwy0717@gmail.com>
 (commit: b17af5c)

Created Test Stubs for OMR Socket API
Created test stubs in omrsockTest.cpp, in test set PortSockTest in Port
Tests
-Created 2 helper function stubs, one for setting up  a server socket,
one
for setting up a client socket. These functions will be called multiple
times,
in multiple tests.
-Created 4 test stubs in PortSockTest.
-Will need to add other tests during implementation.
Issue: #4102
Signed-off-by: Haley Cao <haleycao88@hotmail.com>
 (commit: 99cc5fa)

Add code to allocate and free OMRThreadMCSNodes
Two static functions, allocate_mcs_nodes and free_mcs_nodes, are added 
to allocate and free an OMRThreadMCSNodes structure respectively.
The code is wrapped using OMR_THR_MCS_LOCKS ifdefs.
Related: https://github.com/eclipse/omr/pull/4086.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: e7ad4e9)

Initialize J9Thread->mcsNodes
The entry points where J9Thread->mcsNodes needs to be allocated and 
initialized: 1) omrthread_attach_ex 2) threadCreate
The exit point where J9Thread->mcsNodes needs to be freed: 1)
threadDestroy
OMRTHREAD_ERR_CANT_ALLOC_MCS_NODES added to tag errors related to 
allocating J9Thread->mcsNodes.
The code is wrapped using OMR_THR_MCS_LOCKS ifdefs.
Related: https://github.com/eclipse/omr/pull/4086.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 9752b52)

Initialize and reset J9ThreadMonitor->queueTail
The entry point where J9ThreadMonitor->queueTail needs to be initialized 
is: 1) monitor_init
The two exit points where J9ThreadMonitor->queueTail needs to be reset 
are: 1) monitor_free_nolock 2) monitor_free
For initialization and reset, queueTail is set to NULL.
The code is wrapped using OMR_THR_MCS_LOCKS ifdefs.
Related: https://github.com/eclipse/omr/pull/4086.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 2a693b3)

Remove PPCAlignedLabelInstruction
Previously, the Power codegen had a class of instruction called 
PPCAlignedLabelInstruction. This class was meant to be used for labels 
that should be aligned on a given boundary, but it was unused. On top of 
that, there's no reason to conflate labels and alignment at all: we 
could theoretically want to align instructions without adding a label.
For these reasons, this old class of instruction is being removed.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: add4358)

AArch64: Add VirtualGuardNOPInstruction
This commit adds VirtualGuardNOPInstruction to aarch64 codegen.
Signed-off-by: Akira Saitoh <saiaki@jp.ibm.com>
 (commit: 5c3348c)

AArch64: Implement virtualGuardHelper function
Implement virtualGuardHelper function and use it from ificmpHelper.
Signed-off-by: Akira Saitoh <saiaki@jp.ibm.com>
 (commit: fe4fcf3)

Relocated FrontEnd.hpp and modified relevent directives
Signed-off-by: Yuehan-Lin <Yuehan.Lin@ibm.com>
 (commit: d993cf6)

Add an alignment nop instruction class on Power
In some cases, it is beneficial to ensure that an instruction is aligned 
to a specific number of bytes by padding the instruction stream with 
nops. The Power codegen was previously doing this ad-hoc in the binary 
encoder, but a better solution would be to allow tree evaluators to 
signal where this should happen. This is now possible using the new 
PPCAlignmentNopInstruction class of instruction. When binary encoded, 
this instruction will generate the necessary number of nops to ensure 
that the next instruction is aligned on the requested boundary.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 76b04ee)

Refactor alignment of loop start labels on Power
Previously, the Power codegen would perform alignment of loop start 
labels in the binary encoding loop by checking the node for which a 
label instruction was emitted. This is surprising behaviour, so it has 
been replaced by using the new PPCAlignmentNopInstruction.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 358b1ff)

Make group-ending nops apparent in the tracefile
Previously, the genop instruction had the unhelpful name of "nop", 
making it indistinguishable in tracefiles from the normal nop 
instruction. This caused some confusion when debugging the new loop 
label alignment, as it made it seem like the group-ending nop was not 
being emitted. In order to make the difference apparent, group-ending 
nops will now show up as genop in the tracefile.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: d42c647)

remove all definitions of deprecated ILOpcode TR::cshl
Since most of its denitions are aready removed. I only removed the 
comments and a few evaluators associate with this ILOpcode. TR::cshl
will be completely removed aftet this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 0d4c1ae)

emove all the definitions of deprecated ILOpcode TR::iushl
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of this ILOpCode. TR::iushl will
be completely removed after this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: df4c1ca)

Remove all the definitions of deprecated ILOpCode TR::lushl
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of this ILOpCode. Also removes a
reference to TR::lushl in function Z::genericLongShiftSingle() TR::lushl
will be completely removed after this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: a6e9687)

Remove all the definitions of deprecated ILOpcode TR::ishfl/TR::lshfl
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of these 2 ILOpCode. TR::ishfl and
TR::lshfl  will be completely removed after this commit.
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 0060a4f)

Remove all the definitions of deprecated ILOpcode TR::iushfl/TR::lushfl
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of these 2 ILOpCode. TR::iushfl
and TR::lushfl will be completely removed after this commit
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 5e70828)

Remove all the definitions of deprecated ILOpcode TR::bshfl/TR::sshfl
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of these 2 ILOpCode. TR::bshfl and
TR::sshfl will be completely removed after this commit
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 2f58068)

Remove all the definitions of deprecated ILOpcode TR::bushfl/TR::sushfl
Remove all of its definitions, declarations, properties and all other 
lists/enums that defines the behavior of these 2 ILOpCode. TR::bushfl
and TR::sushfl will be completely removed after this commit
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 1669582)

Runtime compressed refs work
Update SlotObject for runtime compressed refs.
Signed-off-by: Graham Chapman <graham_chapman@ca.ibm.com>
 (commit: ac1e00f)

Fix mis-spellings of 'count'
[ci skip]
Signed-off-by: Keith W. Campbell <keithc@ca.ibm.com>
 (commit: 546e9c3)

Add FE query to add relocations for TR_HelperAddress
This query can be overriden by consumer projects such as OpenJ9 
depending on project requirements. A use of an older version of this
query is also replaced in the X codegen.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 464c81d)

Add omrstr_convert target code page CP_ACP/CP_THREAD_ACP
Added target encoding J9STR_CODE_WINDEFAULTACP/J9STR_CODE_WINTHREADACP 
for Windows code page CP_ACP/CP_THREAD_ACP.
Signed-off-by: Jason Feng <fengj@ca.ibm.com>
 (commit: 1fc5b76)

Apply suggestions from code review
Accept suggested and requested changes
- Deletion of the test entry and exit reports
- Grammar fixes
- Added more descriptive test names
Issue: #4102
Co-Authored-By: Babneet Singh <sbabneet@ca.ibm.com>
Signed-off-by: Haley Cao <haleycao88@hotmail.com>
 (commit: adfbbf4)

Having X86 return true value from 'getSupportsHardwareSQRT()'
Having X86 return true value from 'getSupportsHardwareSQRT()', in order
to make use of the reduce from `StrictMath.sqrt()` and `Math.sqrt()` to
`dsqrt` in `J9RecognizedCallTransformer.cpp`.
Signed-off-by: simonameng <simonameng97@gmail.com>
 (commit: 64b26da)

Adding 'dsqrtEvaluator' in x86
Implementing `dsqrtEvaluator` in x86. `StrictMath.sqrt()` and
`Math.sqrt()` will be reduced to `dsqrt` instead of calling
`inlineMathSQRT` function.
Signed-off-by: simonameng <simonameng97@gmail.com>
 (commit: 48d13f2)

Remove references to OpenJ9 private linkage
J9 private linkage has been largely isolated to OpenJ9 and moved into
the J9 namespace.  Remove unused references in OMR and for code guarded
in project-specific macros reference the appropriate private linkage
class in the J9 namespace.
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: c6cd450)

Fix IA32PrivateLinkage static_cast typo
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: d1e0170)

Change Dwarf_CU_Context files list from vector to map
The current dwarf parser for OSX does not actually prune duplicate file
names, which leads to wasted time parsing the file name list later on.
Switching the container for the file names from list to map would allow
us to remove the duplicate file names from the list and thus improve the
performance of ddrgen. On AIX, we keep a separate set for this purpose
and this change would also do away with that. We need a map instead of a
set so that dwarf_srcfiles returns the file list in the correct order
indicated by the DW_AT_decl_file attribute.
Signed-off-by: Mike Zhang <mike.h.zhang@ibm.com>
 (commit: 4b9df41)

gtest: Do not reference temporaries past their lifetimes
This is an error in gtest, not in OMR.
This same error is repeated in a few places, causing different types of 
bad behaviour on AIX, including breaking almost all logging and option 
processing.
Rather than bind a const mytype& to a temporary result, copy the result 
into a local, which ensures it is safe to access the value later in the 
program.
When a reference is bound to a temporary value, that reference is 
invalid when the value dies, at the end of the "full expression" that 
created it. Using this reference to a temporary in a later expression 
results in a kind of "use after destruction" error.
Fixes: #4537 Signed-off-by: Robert Young <rwy0717@gmail.com>
 (commit: 1e71054)

AArch64: Handle resolved class pointer in loadRelocatableConstant()
This commit adds code to loadRelocatableConstant() in OMRMemoryReference 
for AArch64, so that it can handle resolved class pointer.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: 4ef5339)

AArch64: Add "nop" and "vgdnop" to ARM64Debug.cpp
This commit adds "nop" and "vgdnop" instructions to ARM64Debug.cpp.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: 7f2e813)

AArch64: Implement ARM64Trg1MemSrc1Instruction
This commit implements ARM64Trg1MemSrc1Instruction, for generating 
AArch64 "store exclusive" instructions.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: 57218a3)

Peephole optimizations for BNDCHK evaluator on Z
- `CLR` and `BRC` instructions are reduced to `CLRJ`
- Reduction happens only if CC is not consumed by any instruction
 in the immediate window of 5 instructions, in both branch taken and
 fall through paths
- We do not need to check if CC is consumed in branch taken path if
 branch target is a Snippet
Signed-off-by: Shubham Verma <shubhamv.sv@gmail.com>
 (commit: d1c6a04)

Fix Bugs in splitPostGRA
In post GRA block splitter there are some issue which we found in tests. 
This commits fixes following two issues. 1. While creating approprite
symbol reference for to be uncommoned node to
  store on temp slot or global register, it checks if that node is
representing
  internal pointer. In such case it needs get corresponding pinning
array symref
  if already exists and need to mark created symref with it. In case if
corresponding
  pinning array symref does not exist then it needs to find it and
create new pinning
  array pointer symref. Issue was, it was only checking variable load
nodes for already
  existing pinning array pointer symref while with splitting block after
GRA, we can have
  regLoad node which already contains it. 2. Second issue we found was
altering the Basic Block semantics and adding treetops
  in between branch treetop and BBEnd treetop in the Basic block.
Signed-off-by: Rahil Shah <rahil@ca.ibm.com>
 (commit: 7eb746e)

Fix spelling of 'label'
* collapse multiple blank lines into a single blank line
Signed-off-by: Keith W. Campbell <keithc@ca.ibm.com>
 (commit: 1c4c564)

Removed API within DAA for which Java implementations do not exist
- Removed TR::com_ibm_dataaccess_ByteArrayUtils_trailingZeros
- Removed
TR::com_ibm_dataaccess_ByteArrayUtils_trailingZerosQuadWordAtATime_
Signed-off-by: Shubham Verma <shubhamv.sv@gmail.com>
 (commit: 20af3ad)

Removed helper method for `inlineTrailingZerosQuadWordAtATime_`
Signed-off-by: Shubham Verma <shubhamv.sv@gmail.com>
 (commit: cb6d3ec)

Fix compile error with Concurrent Scavenger enabled on macOS
Removed unused private field `_bytesToScan` from
`ConcurrentScavengeTask` to resolve:
```
/ConcurrentScavengeTask.hpp:38:18: error: private field '_bytesToScan' 
is not used [-Werror,-Wunused-private-field] uintptr_t const
_bytesToScan;   /**< The number of bytes that this must scan before it
will stop trying to do more work */
```
Signed-off-by: Salman Rana <salman.rana@ibm.com>
 (commit: 601ca0e)

Make PPCAlignmentNopInstruction::setAlignment private
Previously, the PPCAlignmentNopInstruction::setAlignment function had 
public visibility. However, a point was brought up during code review 
that there's really not any good reason for this function to be called 
after an instruction is created and calling it after certain analyses 
are run could cause problems.
Since there's no known valid reason to ever call this function outside 
the constructors, it is being made private to prevent its accidental 
misuse.
Signed-off-by: Ben Thomas <ben@benthomas.ca>
 (commit: 62dadbb)

Used fgets to read file in parseFile
Issue:4524
Signed-off-by: Yuehan-Lin <Yuehan.Lin@ibm.com>
 (commit: dc817b9)

Runtime compressed refs work
Update HeapLinkedFreeHeader for runtime compressed refs.
Signed-off-by: Graham Chapman <graham_chapman@ca.ibm.com>
 (commit: 6944795)

Add findOr{Create,Fabricate}ShadowSymbol functions
Add findOrCreateShadowSymbol function and findOrFabricateShadowSymbol to
accept a TR_OpaqueClassBlock. These will be needed for supporting value
types. The behavior of these are project specific, so their 
implementation in OMR is empty. Downstream projects should override 
these as needed.
Co-authored-by: Yiling Han <Yiling.Han@ibm.com> Signed-off-by: Leonardo
Banderali <leonardo2718@protonmail.com>
 (commit: fb355ed)

Remove unsigned support in Z::imulEvaluator
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 1e69d6f)

Fix issue with GC during idle cycle
Fix for https://github.com/eclipse/openj9/issues/7679
- Forced system gc resulting from idleness now ignores any uncompleted 
concurrent global cycle by aborting it and restarting in STW.
- Introduced new abort reason for verbose reporting purposes:
`ABORT_COLLECTION_IDLE_GC`
Signed-off-by: Salman Rana <salman.rana@ibm.com>
 (commit: 290fef8)

Fix missing conversion in splitPostGRA
There was an issue with splitPostGRA where it was creating temp slot for
the node which was either of Int8 or Int16 type. To maintain alignment,
it needed conversion to Int32 type before storing it into memory.
splitPostGRA was missing this conversion and which was causing issues on
Z where it was storing a byte into memory without conversion and then
loads a byte from that slot causing us to load most significant byte
which contains 0 instead of the original value. This commit adds the
missing conversion piece of code for such nodes.
Signed-off-by: Rahil Shah <rahil@ca.ibm.com>
 (commit: 76484b9)

AArch64: Implement ARM64Trg1ImmSymInstruction
This commit implements ARM64Trg1ImmSymInstruction, for generating PC
relative load register instructions.
Signed-off-by: Akira Saitoh <saiaki@jp.ibm.com>
 (commit: 2cd0f0b)

Don't default initialize isRelocatable boolean
This boolean will now be used to represent if a relocation is needed for
remote as well as AOT compiles. Doing so consolidates the two questions
that the routine needs to answer. Since OMR is not aware of remote
compilations, the decision on whether a relocation is required will not
have to be made explicitly by the caller.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 313b2fa)

Rename doc/compiler/README
Give it an `md` extension so that it is automatically rendered properly 
by GitHub.
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: 509443c)

Compact trigger adjustment
Dark matter and paging compact triggers currently only compare potential 
free memory gains vs current free memory (with aim to reduce Global GC 
frequency). While trigger thresholds are set rather high, they can still 
easily trigger with Tenure being fairly full.
To account for the cost of work by doing the compact, the gains are not
just compared with total free memory but with total heap size, too.
That alone would make the trigger way less probable to occur, so as a 
compromise: the weight of total heap that account into the formula is 
set to 50%, and thresholds are cut by a factor of 2 or so.
Signed-off-by: Aleksandar Micic <amicic@ca.ibm.com>
 (commit: 538fd9b)

Update uses of loadAddressConstant API
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: c71a701)

Fixes bug found when assigning a new real register
When we assign a new real register to a virtual register (for GPRs), we
were not updating the virtual register to hold its new real register.
This led to a discrepancy between the real->virtual and virtual->real
mapping between registers. This commit ensures that the virtual register
now knows of its new real register.
Signed-off-by: Pushkar Bettadpur <pushkar.bettadpur@gmail.com>
 (commit: a12a2d0)

CMake: Add ability to handle split debug info
Add new flag OMR_SEPARATE_DEBUG_INFO. If set, debug info for a target is
copied into <target_file>.dbg, and debug info from <target_file> is 
stripped.
Signed-off-by: Devin Nakamura <devinn@ca.ibm.com>
 (commit: ecc077e)

Move codegen/FrontEnd.cpp to env/FrontEnd.cpp
Signed-off-by: Yuehan-Lin <Yuehan.Lin@ibm.com>
 (commit: 0995e26)

Move NullChkEvaluators to OpenJ9 for ARM, Power, Z and Aarch64 codegen
This commit moves the following functions from OMR to OpenJ9 for ARM,
Power, Z and Aarch64 codegen implementations:
- NULLCHKEvaluator
- resolveAndNULLCHKEvaluator
- evaluateNULLCHKWithPossibleResolve
All these functions are moved to OpenJ9 and are no longer needed in OMR.
Fixes: #3530 Signed-off-by: Md. Ariful Haque <mhaque5@unb.ca>
 (commit: 8588a6e)

Add new needRelocationsForLookupEvaluationData query
This query will maintain the current behaviour in OMR, while also
allowing downstream projects to override its behaviour.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: c158ab0)

Update loadAddressConstant API consumers
This commit updates the consumers of this API to use it correctly after
the new API changes.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: fae6238)

Share implementation to re-reserve trampolines on code cache switch
Each code generator duplicates logic to re-reserve a trampoline for a 
call target if the code cache is switched between the time the
trampoline is first reserved (likely instruction selection) and binary
encoding.
The logic is nearly the same in all cases.  Provide a common version 
that is shared by all code generators.
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: 4b5d965)

AArch64: Fix code generation for adr and adrp instructions
This commit fixes the wrong encoding of the immediate value for "adr" 
and "adrp" instructions. Those two instructions encode the lower 2 bits
of the immediate value in bits 29-30 of the instruction word, and the
upper 19 bits of the immediate value in bits 5-23 of the instruction
word. There was a mistake with masking and shifting the upper 19 bits.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: d54781f)

Runtime compressed refs work
Remove obsolete API from HeapLinkedFreeHeader.
Signed-off-by: Graham Chapman <graham_chapman@ca.ibm.com>
 (commit: c6ec740)

Break unneeded class dependencies on OMRLinkage
Do not include unnecessary header files to prevent downstream breakages.
Introduce include directives and forward declarations as appropriate to 
compensate for removing header files from the extensible OMRLinkage.hpp
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: 6de2de3)

Only unblock an asynchronous signal if it is used
Currently, all supported asynchronous signals are unblocked during 
startup. A signal is unblocked even when a signal handler is not 
installed for it. This conflicts with an OMR consumer, who wants to 
block a set of signals. The OMR consumer blocks a set of signals but the 
OMR signal library unblocks those signals at startup. OMR signal library 
should only unblock a signal if a signal handler is registered for it. 
In other words, a signal should not be unblocked if it is not used i.e. 
a signal handler is not registered for it. 
Fixes: https://github.com/eclipse/openj9/issues/7749
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 717082e)

Add new overload for loadAddressConstant
This new overload allows the consumer of the API to determine if a
relocation record is needed. The overload is added because it will allow
all consumers of this API to use the new definition of
loadAddressConstant, while not breaking anything in downstream projects.
Once the new definition has been accepted by downstream projects, they
can begin to use the new API as well since if they choose to do so.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: e01ecad)

AArch64: Add extraRegister to OMRMemoryReference
This commit adds extraRegister and related code to OMRMemoryReference
for AArch64. It will be used by UnresolvedSnippet.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: 3377516)

Implement omrthread_mcs_lock, which is part of the MCS Lock API
Related: https://github.com/eclipse/omr/pull/4086.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: c196dd3)

Implement omrthread_mcs_trylock, which is part of the MCS Lock API
Related: https://github.com/eclipse/omr/pull/4086.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 2c3bdd4)

Implement omrthread_mcs_unlock, which is part of the MCS Lock API
Related: https://github.com/eclipse/omr/pull/4086.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: cf627e1)

Implement omrthread_mcs_node_allocate, which is part of the MCS Lock API
Related: https://github.com/eclipse/omr/pull/4086.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: 46dd324)

Implement omrthread_mcs_node_free, which is part of the MCS Lock API
Related: https://github.com/eclipse/omr/pull/4086.
Signed-off-by: Babneet Singh <sbabneet@ca.ibm.com>
 (commit: a81723b)

Relocate getMonClass and addMonClass to OpenJ9
The OMR::CodeGenerator functions- getMonClass and addMonClass are 
relocated to OpenJ9 CodeGenerator and removed from OMR.
The changes also include the array- _monitorMapping moving from OMR to
OpenJ9.
Closes: GitHub issue #1869
Signed-off-by: Md. Alvee Noor <mnoor@unb.ca>
 (commit: 2339b26)

CMake: Update FindLibElf
Use data from pkg-config if available
Signed-off-by: Devin Nakamura <devinn@ca.ibm.com>
 (commit: db0d2fd)

Runtime compressed refs work
Add new constructor to ForwardedHeader in preparation for the real work.
Signed-off-by: Graham Chapman <graham_chapman@ca.ibm.com>
 (commit: 91483b0)

Fixes to match prototypes to old socket API
Decided to match with old API until new socket API is done implementing.
-Made sure the function prototypes have the same arguments
Issue: #4102
Signed-off-by: Haley Cao <haleycao88@hotmail.com>
 (commit: a2f1d3e)

Defined private copy constructor for TypeDictionary
Signed-off-by: Yuehan-Lin <Yuehan.Lin@ibm.com>
 (commit: 2c3f67d)

Avoid buffer overflow in expandString()
The test in the old code (formatCursor >= MAX_IMAGE_PATH_LENGTH - 1) 
only ensured there was room for one more character (plus the 
NUL-terminator) in any iteration of the loop and thus the expansion of
'%p', '%d' or '%t' might not fit.
This was highlighted by a warning issued by gcc 9.
Signed-off-by: Keith W. Campbell <keithc@ca.ibm.com>
 (commit: 9d712f3)

Avoid buffer overflow in count_threads()
strncpy() does not add a NUL-terminator if not part of the specified 
segment of the source, so if the file name was more than 11 characters 
long, buf might not be NUL-terminated and the following strcat() may
write beyond the end of the buffer.
This was highlighted by a warning issued by gcc 9.
Signed-off-by: Keith W. Campbell <keithc@ca.ibm.com>
 (commit: 448e32e)

Ensure strings are NUL-terminated in omrosdump_helpers
strncpy() does not add a NUL-terminator if not part of the specified 
segment of the source, so the result may not be NUL-terminated; use 
OMR's str_printf() instead. Affected functions are deriveCoreFileName() 
and renameDump().
This was highlighted by a warning issued by gcc 9.
Also fixed typo in corePatternFormat parameter name.
Signed-off-by: Keith W. Campbell <keithc@ca.ibm.com>
 (commit: 7a932ad)

Ensure string is NUL-terminated in omrsysinfo_get_OS_type()
strncpy() does not add a NUL-terminator if not part of the specified 
segment of the source, so the result may not be NUL-terminated. Even
though the buffer is explicitly NUL-terminated, the compiler still 
issues a warning: use memcpy() instead.
This was highlighted by a warning issued by gcc 9.
Signed-off-by: Keith W. Campbell <keithc@ca.ibm.com>
 (commit: b399ceb)

Avoid compiler warning in omrvmem_numa_get_node_details()
Even though strncat() always adds a NUL-terminator, gcc 9 issues a
warning about its use: use OMR's str_printf() function instead.
Signed-off-by: Keith W. Campbell <keithc@ca.ibm.com>
 (commit: 5a11f35)

Introduce new query to replace JITServer checks
This commit introduces a new query, 
canEmitDataForExternallyRelocatableInstructions, that can be used to
determind if data can be emitted safely in an instruction that is the
target of a relocation record.
Signed-off-by: Dhruv Chopra <Dhruv.C.Chopra@ibm.com>
 (commit: 7e87e0c)

AArch64: Correct the condition code for fcmp/dcmp
This commit corrects the AArch64 condiiton codes used for fcmp and dcmp 
operations.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: c3d0cf1)

AArch64: Implement igotoEvaluator
This commit adds implementation of igotoEvaluator.
Signed-off-by: Akira Saitoh <saiaki@jp.ibm.com>
 (commit: 9eaddae)

Remove stale comment
The comment was incorrect as the VM is *not* called to do lowering. 
Furthermore, it provides no additional insight into what the code is 
doing beyond what the function name already provides. Removing the 
comment therefore seems appropriate.
Signed-off-by: Leonardo Banderali <leonardo2718@protonmail.com>
 (commit: f937c1b)

Clean up lowerTreesPreChildrenVisit
Signed-off-by: Leonardo Banderali <leonardo2718@protonmail.com>
 (commit: c0b6b9e)

Add option disabling inline JSR292 without peeking
Signed-off-by: Yi Zhang <yizhang@ca.ibm.com>
 (commit: 67022e8)

Change isInteger to is32BitOp in lowerTreesPreChildrenVisit
Signed-off-by: Leonardo Banderali <leonardo2718@protonmail.com>
 (commit: a010217)

Add eaEscapeHelperSymbol
The OpenJ9 Escape Analysis optimization currently overloads the meaning
of the OSR helper callees as part of its analysis. This change
introduces a new non-helper symbol for this optimization pass to use to
avoid overloaded meanings and the resultant confustion in the codebase.
Signed-off-by: Andrew Craik <ajcraik@ca.ibm.com>
 (commit: 424b0a5)

Remove creation of nodes with deprecated unsigned Constant opcodes
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 2e4a572)

Remove Creation of nodes with deprecated unsigned Equality opcodes
Specifically, remove creation of opcodes in Equality compare and
equality compare and branch
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: d3ec5fa)

Fatal assert any creation of node with deprecated unsigned opcode of
equality
Add TR_ASSERT_FATAL to ensure there is no more creatation or recreation 
of node with deprecated opcode in the equality compare and equality
compare branch category
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 2c0db86)

Fatal assert any creation of nodes with deprecated constant opcodes
Signed-off-by: Bohao(Aaron) Wang <aaronwang0407@gmail.com>
 (commit: 14de6f2)

Trust maps file parser if no errors occur
If OMRPORT_VMEM_ALLOC_QUICK is used for virtual memory allocation we are 
trying to find available memory interval by scanning maps file in 
findAvailableMemoryBlockNoMalloc(). In case if it returns NULL (memory 
interval with requested parameters is not available) we have choice to 
trust this result and fail memory allocation request or continue with 
slow allocation path to try any possible address range one by one.
In the past we noticed that sometimes (with unknown conditions 
unfortunately) parser returns NULL, however slow path was able to 
allocate memory eventually. So current code always execute slow path if 
parsing fails. It creates possible delays in JVM startup time. 
Currently parser in findAvailableMemoryBlockNoMalloc() can not 
differentiate between cases parsing was complete with no error but no 
memory range found or an error occur at the time of parsing. I believe 
in first case we can trust parser results and fail allocation request 
immediately. In second case however call of slow allocation path is 
necessary.
So I am suggesting to add new output parameter that would report about 
error occur at the parsing time. 
Currently parser can not receive EOF condition because omrfile_read() 
would mask it by returning -1. I added a recognition of EOF to parser. 
Also I added a number of trace points to simplify analysis in case of 
the failure in future. Please note that these trace points at the moment 
because this call occur early in startup and Trace Engine has not been 
initialized yet. However we are planning to fix this problem in the 
future so these trace points become useful.
Also (by request from Service and have in mind we do not know why it 
failed exactly) I added possibility to revert new behaviour to old one 
by introducing a new key for Port Library. It will be used in case of 
emergency to return back to old mode if this feature still be not 
reliable enough.
Signed-off-by: Dmitri Pivkine <Dmitri_Pivkine@ca.ibm.com>
 (commit: abf241c)

Move the calls to AC_PROG_CC/CXX higher
Autoconf has a rudimentary macro-prerequisite feature, which allows 
macros and tests to automatically invoke dependencies.
Because of this system, AC_PROG_CC is being automatically invoked in the 
middle of a check for the size of void*, which is much earlier than 
where we manually invoke the macros. For clarity, I'm moving the calls 
to AC_PROG_CC/CXX higher, to the beginning of the "build environment" 
section of the script, and a few lines before the sizeof_void_p test.
Signed-off-by: Robert Young <rwy0717@gmail.com>
 (commit: bd69387)

AArch64: Add call to TR_Debug::print(TR::ARM64RecompilationSnippet)
This commit adds a call to TR_Debug::print() for 
ARM64RecompilationSnippet.
Signed-off-by: KONNO Kazuhiro <konno@jp.ibm.com>
 (commit: c8c7b8e)

Move arithmeticNeedsLiteralFromPool function from OMR to OpenJ9
This commit moves arithmeticNeedsLiteralFromPool function from OMR to 
OpenJ9. arithmeticNeedsLiteralFromPool is not used in OMR, but it is 
used in OpenJ9 for Z codegen implementation.
Fixes: #1870 Signed-off-by: Md. Ariful Haque <mhaque5@unb.ca>
 (commit: 37bcd6a)

Remove unused features from S390LabelInstruction
The following are not used in OMR nor any known downstream project.
* `targetReg` parameter from constructors
* `_handle` field
* PreservedForListing flag
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: cb5ee35)

Add new 'MinimalTest' suite to compilertriltest.
This suite contains a really minimal tests such as "return 42" or
"return arg + 10". These tests are useful when porting compiler to a new
architecture, such as RISC-V.
Signed-off-by: Jan Vrany <jan.vrany@fit.cvut.cz> Signed-off-by: Boris
Shingarov <shingarov@labware.com>
 (commit: f492938)

Remove unused isHint from Z register dependencies
Seems to have been required for access registers.
Also remove functions predicated on isHint being true.
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: f19de60)

Remove unused conflictsResolved field from Z
RegisterDependencyConditions
Fold code assuming it is always false.
Signed-off-by: Daryl Maier <maier@ca.ibm.com>
 (commit: 48e9d87)