Software Concurrent Scavenge Support on Power

Sets OMR_GC_CONCURRENT_SCAVENGER in the configure files to enable
concurrent scavenge.

Modifies mminit.cpp to be able to set softwareRangeCheckReadBarrier to
true to enable software based concurrent scavenge.

Implemented irdbarEvaluator and ardbarEvaluator which are used to handle
loads that might need to call out to the readbarrier under Concurrent
Scavenge GC policy.

Updated VMinlineCompareAndSwapObject to be aware of the possiblity of
calling out to the read barrier under Concurrent Scavenge GC policy.

Implemented a new version of arraycopyEvaluator for reference array copies
that calls out to the referenceArrayCopy if current scavenge is currently
happening.

Added jitReadBarrier and jitReferenceArrayCopy as helpers to call out to
J9ReadBarrier and referenceArrayCopy.

Updated thisThreadGetConcurrentScavengeActiveByteAddressOffset and
thisThreadGetEvacuateBaseAddressOffset to recognize Power as well as X86.

Updated insertPrefetchIfNecessary to recognize the irdbari opcode.

Updated copyright dates on modified files.

TM is current not compatible with read barriers. This change will disable
TM if read barriers are needed.

Issue: eclipse-openj9#3421
Signed-off-by: jimmyk <jimmyk@ca.ibm.com>

runtime/codert_vm/pnathelp.m4

@@ -1,4 +1,4 @@
-dnl Copyright (c) 2017, 2018 IBM Corp. and others
+dnl Copyright (c) 2017, 2019 IBM Corp. and others
 dnl
 dnl This program and the accompanying materials are made available under
 dnl the terms of the Eclipse Public License 2.0 which accompanies this
@@ -852,6 +852,63 @@ START_PROC(jitFillOSRBufferReturn)
 	CINTERP(J9TR_bcloop_exit_interpreter, 0)
 END_PROC(jitFillOSRBufferReturn)
 
+dnl Expects r3 to already contain the vmThread.
+dnl Expects r4 to already contain the address being loaded from.
+START_PROC(jitReadBarrier)
+ifdef({OMR_GC_CONCURRENT_SCAVENGER},{
+	staddr r0,JIT_GPR_SAVE_SLOT(0)
+	SAVE_LR
+	SAVE_JIT_GOT_REGISTERS
+	INIT_JIT_TOC
+	SWITCH_TO_C_STACK
+	SAVE_C_VOLATILE_REGS
+
+	laddr r5,J9TR_VMThread_javaVM(J9VMTHREAD)
+	laddr r5,J9TR_JavaVM_memoryManagerFunctions(r5)
+	laddr FUNC_PTR,J9TR_J9MemoryManagerFunctions_J9ReadBarrier(r5)
+	CALL_INDIRECT
+
+	RESTORE_C_VOLATILE_REGS
+	RESTORE_JIT_GOT_REGISTERS
+	RESTORE_LR
+	SWITCH_TO_JAVA_STACK
+	laddr r0,JIT_GPR_SAVE_SLOT(0)
+	blr
+},{
+dnl jitReadBarrier is not supported if OMR_GC_CONCURRENT_SCAVENGER is not set
+	trap
+})
+END_PROC(jitReadBarrier)
+
+dnl Expects r3 to already contain vmThread.
+dnl Expects r4 to already contain srcObj.
+dnl Expects r5 to already contain dstObj.
+dnl Expects r6 to already contain srcAddr.
+dnl Expects r7 to already contain dstAddr.
+dnl Expects r8 to already contain length.
+START_PROC(jitReferenceArrayCopy)
+	staddr r0,JIT_GPR_SAVE_SLOT(0)
+	SAVE_LR
+	SAVE_JIT_GOT_REGISTERS
+	INIT_JIT_TOC
+	SWITCH_TO_C_STACK
+	SAVE_ALL_REGS
+
+	laddr r9,J9TR_VMThread_javaVM(J9VMTHREAD)
+	laddr r9,J9TR_JavaVM_memoryManagerFunctions(r9)
+	laddr FUNC_PTR,J9TR_J9MemoryManagerFunctions_referenceArrayCopy(r9)
+	CALL_INDIRECT
+	staddr r3,J9TR_VMThread_returnValue(J9VMTHREAD)
+
+	RESTORE_ALL_REGS
+	RESTORE_JIT_GOT_REGISTERS
+	RESTORE_LR
+	SWITCH_TO_JAVA_STACK
+	laddr r0,JIT_GPR_SAVE_SLOT(0)
+	laddr r3,J9TR_VMThread_returnValue(J9VMTHREAD)
+	blr
+END_PROC(jitReferenceArrayCopy)
+
 ifdef({ASM_J9VM_ENV_DATA64},{
 
 BEGIN_RETURN_POINT(jitExitInterpreterJ)

runtime/compiler/env/VMJ9.cpp

@@ -1920,11 +1920,11 @@ UDATA TR_J9VMBase::thisThreadGetConcurrentScavengeActiveByteAddressOffset()
 UDATA TR_J9VMBase::thisThreadGetEvacuateBaseAddressOffset()
    {
 #if defined(OMR_GC_CONCURRENT_SCAVENGER)
-#if defined(J9VM_GC_COMPRESSED_POINTERS) && defined(TR_TARGET_X86)
+#if defined(J9VM_GC_COMPRESSED_POINTERS) && !defined(TR_TARGET_S390)
    return offsetof(J9VMThread, readBarrierRangeCheckBaseCompressed);
 #else
    return offsetof(J9VMThread, readBarrierRangeCheckBase);
-#endif /* defined(J9VM_GC_COMPRESSED_POINTERS) && defined(TR_TARGET_X86) */
+#endif /* defined(J9VM_GC_COMPRESSED_POINTERS) && !defined(TR_TARGET_S390) */
 #else /* defined(OMR_GC_CONCURRENT_SCAVENGER) */
    TR_ASSERT(0,"Field readBarrierRangeCheckBase does not exists in J9VMThread.");
    return 0;
@@ -1937,11 +1937,11 @@ UDATA TR_J9VMBase::thisThreadGetEvacuateBaseAddressOffset()
 UDATA TR_J9VMBase::thisThreadGetEvacuateTopAddressOffset()
    {
 #if defined(OMR_GC_CONCURRENT_SCAVENGER)
-#if defined(J9VM_GC_COMPRESSED_POINTERS) && defined(TR_TARGET_X86)
+#if defined(J9VM_GC_COMPRESSED_POINTERS) && !defined(TR_TARGET_S390)
    return offsetof(J9VMThread, readBarrierRangeCheckTopCompressed);
 #else
    return offsetof(J9VMThread, readBarrierRangeCheckTop);
-#endif /* defined(J9VM_GC_COMPRESSED_POINTERS) && defined(TR_TARGET_X86) */
+#endif /* defined(J9VM_GC_COMPRESSED_POINTERS) && !defined(TR_TARGET_S390) */
 #else /* defined(OMR_GC_CONCURRENT_SCAVENGER) */
    TR_ASSERT(0,"Field readBarrierRangeCheckTop does not exists in J9VMThread.");
    return 0;

runtime/compiler/p/codegen/J9CodeGenerator.cpp

@@ -619,7 +619,10 @@ J9::Power::CodeGenerator::insertPrefetchIfNecessary(TR::Node *node, TR::Register
       }
 
    if (node->getOpCodeValue() == TR::aloadi ||
-      (TR::Compiler->target.is64Bit() && comp()->useCompressedPointers() && node->getOpCodeValue() == TR::iloadi && comp()->getMethodHotness() >= hot))
+         (TR::Compiler->target.is64Bit() &&
+          comp()->useCompressedPointers() &&
+          (node->getOpCodeValue() == TR::iloadi || node->getOpCodeValue() == TR::irdbari) &&
+          comp()->getMethodHotness() >= hot))
       {
       TR::Node *firstChild = node->getFirstChild();
       optDisabled = disableIteratorPrefetch;