//
// Helps determine whether a function should be speculatively jitted.
// This function is only used once and is used in a time-critical area, so
// be careful with it (moving it around actually caused around a 5% perf
// regression on a test).
//
bool CodeGenWorkItem::ShouldSpeculativelyJit(uint byteCodeSizeGenerated) const
{
if(!functionBody->DoFullJit())
{
return false;
}
byteCodeSizeGenerated += this->GetByteCodeCount();
if(CONFIG_FLAG(ProfileBasedSpeculativeJit))
{
Assert(!CONFIG_ISENABLED(Js::NoDynamicProfileInMemoryCacheFlag));
// JIT this now if we are under the speculation cap.
return
byteCodeSizeGenerated < (uint)CONFIG_FLAG(SpeculationCap) ||
(
byteCodeSizeGenerated < (uint)CONFIG_FLAG(ProfileBasedSpeculationCap) &&
this->ShouldSpeculativelyJitBasedOnProfile()
);
}
else
{
return byteCodeSizeGenerated < (uint)CONFIG_FLAG(SpeculationCap);
}
}
BYTE* WebAssemblySharedArrayBuffer::AllocBuffer(uint32 length, uint32 maxLength)
{
#if ENABLE_FAST_ARRAYBUFFER
if (CONFIG_FLAG(WasmFastArray))
{
return (BYTE*)WasmVirtualAllocator(length);
}
#endif
#ifdef _WIN32
if (CONFIG_FLAG(WasmSharedArrayVirtualBuffer))
{
return (BYTE*)AllocWrapper(length, maxLength);
}
#endif
AssertOrFailFast(maxLength >= length);
uint32 additionalSize = maxLength - length;
if (additionalSize > 0)
{
// SharedArrayBuffer::Init already requested External Memory for `length`, we need to request the balance
if (!this->GetRecycler()->RequestExternalMemoryAllocation(additionalSize))
{
// Failed to request for more memory
return nullptr;
}
}
// Allocate the full size of the buffer if we can't do VirtualAlloc
return HeapNewNoThrowArray(BYTE, maxLength);
}
void
SwitchIRBuilder::BuildCaseBrInstr(uint32 targetOffset)
{
Assert(m_isAsmJs || m_profiledSwitchInstr);
int start = 0;
int end = m_caseNodes->Count() - 1;
if (m_caseNodes->Count() <= CONFIG_FLAG(MaxLinearIntCaseCount))
{
BuildLinearTraverseInstr(start, end, targetOffset);
ResetCaseNodes();
return;
}
RefineCaseNodes();
BuildOptimizedIntegerCaseInstrs(targetOffset);
ResetCaseNodes(); // clear the list for the next new set of integers - or for a new switch case statement
//optimization is definitely performed when the number of cases is greater than the threshold
if (end - start > CONFIG_FLAG(MaxLinearIntCaseCount) - 1) // -1 for handling zero index as the base
{
BuildBailOnNotInteger();
}
}
void PropertyRecordUsageCache::RegisterCacheMiss()
{
this->hitRate -= (int)CONFIG_FLAG(PropertyCacheMissPenalty);
if (this->hitRate < (int)CONFIG_FLAG(PropertyCacheMissReset))
{
this->hitRate = 0;
}
}
void PropertyString::LogCacheMiss()
{
this->hitRate -= (int)CONFIG_FLAG(StringCacheMissPenalty);
if (this->hitRate < (int)CONFIG_FLAG(StringCacheMissReset))
{
this->hitRate = 0;
}
}
bool
AutoSystemInfo::VirtualSseAvailable(const int sseLevel) const
{
#ifdef ENABLE_DEBUG_CONFIG_OPTIONS
return CONFIG_FLAG(Sse) < 0 || CONFIG_FLAG(Sse) >= sseLevel;
#else
return true;
#endif
}
void InliningThreshold::SetAggressiveHeuristics()
{
int limit = CONFIG_FLAG(AggressiveInlineThreshold);
inlineThreshold = limit;
constructorInlineThreshold = limit;
outsideLoopInlineThreshold = limit;
leafInlineThreshold = limit;
loopInlineThreshold = limit;
polymorphicInlineThreshold = limit;
maxNumberOfInlineesWithLoop = CONFIG_FLAG(MaxNumberOfInlineesWithLoop);
inlineCountMax = CONFIG_FLAG(AggressiveInlineCountMax);
}
void
SwitchIRBuilder::BuildBinaryTraverseInstr(int start, int end, uint32 defaultLeafBranch)
{
int mid;
if (start > end)
{
return;
}
if (end - start <= CONFIG_FLAG(MaxLinearIntCaseCount) - 1) // -1 for handling zero index as the base
{
//if only 3 elements, then do linear search on the elements
BuildLinearTraverseInstr(start, end, defaultLeafBranch);
return;
}
mid = start + ((end - start + 1) / 2);
CaseNode* midNode = m_caseNodes->Item(mid);
CaseNode* startNode = m_caseNodes->Item(start);
// if the value that we are switching on is greater than the start case value
// then we branch right to the right half of the binary search
IR::BranchInstr* caseInstr = startNode->GetCaseInstr();
IR::BranchInstr* branchInstr = IR::BranchInstr::New(m_geOp, nullptr, caseInstr->GetSrc1(), midNode->GetLowerBound(), m_func);
branchInstr->m_isSwitchBr = true;
m_adapter->AddBranchInstr(branchInstr, startNode->GetOffset(), midNode->GetOffset(), true);
BuildBinaryTraverseInstr(start, mid - 1, defaultLeafBranch);
BuildBinaryTraverseInstr(mid, end, defaultLeafBranch);
}
void InterpreterThunkEmitter::NewOOPJITThunkBlock()
{
InterpreterThunkInfoIDL thunkInfo;
HRESULT hr = JITManager::GetJITManager()->NewInterpreterThunkBlock(
this->scriptContext->GetRemoteScriptAddr(),
this->isAsmInterpreterThunk,
&thunkInfo
);
JITManager::HandleServerCallResult(hr);
BYTE* buffer = (BYTE*)thunkInfo.thunkBlockAddr;
if (!CONFIG_FLAG(OOPCFGRegistration))
{
this->scriptContext->GetThreadContext()->SetValidCallTargetForCFG(buffer);
}
// Update object state only at the end when everything has succeeded - and no exceptions can be thrown.
auto block = this->thunkBlocks.PrependNode(allocator, buffer);
#if PDATA_ENABLED
void* pdataTable;
PDataManager::RegisterPdata((PRUNTIME_FUNCTION)thunkInfo.pdataTableStart, (ULONG_PTR)buffer, (ULONG_PTR)thunkInfo.epilogEndAddr, &pdataTable);
block->SetPdata(pdataTable);
#else
Unused(block);
#endif
this->thunkCount = thunkInfo.thunkCount;
this->thunkBuffer = (BYTE*)thunkInfo.thunkBlockAddr;
}
bool WebAssemblySharedArrayBuffer::IsValidVirtualBufferLength(uint length) const
{
#if ENABLE_FAST_ARRAYBUFFER
if (CONFIG_FLAG(WasmFastArray))
{
return true;
}
#endif
#ifdef _WIN32
if (CONFIG_FLAG(WasmSharedArrayVirtualBuffer))
{
return true;
}
#endif
return false;
}
size_t
Output::VTrace(const wchar_t* shortPrefixFormat, const wchar_t* prefix, const wchar_t *form, va_list argptr)
{
size_t retValue = 0;
if (CONFIG_FLAG(RichTraceFormat))
{
InterlockedIncrement(&s_traceEntryId);
retValue += Output::Print(L"[%d ~%d %s] ", s_traceEntryId, ::GetCurrentThreadId(), prefix);
}
else
{
retValue += Output::Print(shortPrefixFormat, prefix);
}
retValue += Output::VPrint(form, argptr);
// Print stack trace.
if (s_stackTraceHelper)
{
const ULONG c_framesToSkip = 2; // Skip 2 frames -- Output::VTrace and Output::Trace.
const ULONG c_frameCount = 10; // TODO: make it configurable.
const wchar_t callStackPrefix[] = L"call stack:";
if (s_inMemoryLogger)
{
// Trace just addresses of functions, avoid symbol info as it takes too much memory.
// One line for whole stack trace for easier parsing on the jd side.
const size_t c_msgCharCount = _countof(callStackPrefix) + (1 + sizeof(void*) * 2) * c_frameCount; // 2 hexadecimal digits per byte + 1 for space.
wchar_t callStackMsg[c_msgCharCount];
void* frames[c_frameCount];
size_t start = 0;
size_t temp;
temp = _snwprintf_s(callStackMsg, _countof(callStackMsg), _TRUNCATE, L"%s", callStackPrefix);
Assert(temp != -1);
start += temp;
ULONG framesObtained = s_stackTraceHelper->GetStackTrace(c_framesToSkip, c_frameCount, frames);
Assert(framesObtained <= c_frameCount);
for (ULONG i = 0; i < framesObtained && i < c_frameCount; ++i)
{
Assert(_countof(callStackMsg) >= start);
temp = _snwprintf_s(callStackMsg + start, _countof(callStackMsg) - start, _TRUNCATE, L" %p", frames[i]);
Assert(temp != -1);
start += temp;
}
retValue += Output::Print(L"%s\n", callStackMsg);
}
else
{
// Trace with full symbol info.
retValue += Output::Print(L"%s\n", callStackPrefix);
retValue += s_stackTraceHelper->PrintStackTrace(c_framesToSkip, c_frameCount);
}
}
return retValue;
}
void DynamicType::PrepareForTypeSnapshotEnumeration()
{
if(!GetIsLocked() && CONFIG_FLAG(TypeSnapshotEnumeration))
{
// Lock the type and handler, enabling us to enumerate properties of the type snapshotted
// at the beginning of enumeration, despite property changes made by script during enumeration.
LockType(); // Note: this only works for type handlers that support locking.
}
}
DynamicType * DynamicObjectPropertyEnumerator::GetTypeToEnumerate() const
{
return
GetSnapShotSemantics() &&
initialType->GetIsLocked() &&
CONFIG_FLAG(TypeSnapshotEnumeration)
? PointerValue(initialType)
: object->GetDynamicType();
}
double
DateImplementation::DoubleToTvUtc(double tv)
{
if (JavascriptNumber::IsNan(tv) || tv < ktvMin || tv > ktvMax)
{
return JavascriptNumber::NaN;
}
return CONFIG_FLAG(HighPrecisionDate)? tv : ConvertToInteger(tv);
}
void
JITOutput::FinalizeNativeCode()
{
CustomHeap::Allocation * allocation = GetAllocation();
#if ENABLE_OOP_NATIVE_CODEGEN
if (JITManager::GetJITManager()->IsJITServer())
{
m_func->GetOOPCodeGenAllocators()->emitBufferManager.CompletePreviousAllocation(m_oopAlloc);
#if defined(_CONTROL_FLOW_GUARD) && !defined(_M_ARM)
if (!m_func->IsLoopBody() && CONFIG_FLAG(UseJITTrampoline))
{
allocation->thunkAddress = m_func->GetOOPThreadContext()->GetJITThunkEmitter()->CreateThunk(m_outputData->codeAddress);
}
#endif
}
else
#endif
{
m_func->GetInProcCodeGenAllocators()->emitBufferManager.CompletePreviousAllocation(m_inProcAlloc);
m_func->GetInProcJITEntryPointInfo()->SetInProcJITNativeCodeData(m_func->GetNativeCodeDataAllocator()->Finalize());
m_func->GetInProcJITEntryPointInfo()->GetJitTransferData()->SetRawData(m_func->GetTransferDataAllocator()->Finalize());
#if !FLOATVAR
CodeGenNumberChunk * numberChunks = m_func->GetNumberAllocator()->Finalize();
m_func->GetInProcJITEntryPointInfo()->SetNumberChunks(numberChunks);
#endif
#if defined(_CONTROL_FLOW_GUARD) && !defined(_M_ARM)
if (!m_func->IsLoopBody() && CONFIG_FLAG(UseJITTrampoline))
{
allocation->thunkAddress = m_func->GetInProcThreadContext()->GetJITThunkEmitter()->CreateThunk(m_outputData->codeAddress);
}
#endif
}
m_outputData->thunkAddress = allocation->thunkAddress;
if (!allocation->thunkAddress && CONFIG_FLAG(OOPCFGRegistration))
{
PVOID callTarget = (PVOID)m_outputData->codeAddress;
#ifdef _M_ARM
callTarget = (PVOID)((uintptr_t)callTarget | 0x1);
#endif
m_func->GetThreadContextInfo()->SetValidCallTargetForCFG(callTarget);
}
}
void BackgroundJobProcessor::InitializeThreadCount()
{
if (CONFIG_FLAG(ForceMaxJitThreadCount))
{
this->maxThreadCount = CONFIG_FLAG(MaxJitThreadCount);
}
else if (AutoSystemInfo::Data.IsLowMemoryProcess())
{
// In a low-memory scenario, don't spin up multiple threads, regardless of how many cores we have.
this->maxThreadCount = 1;
}
else
{
int processorCount = AutoSystemInfo::Data.GetNumberOfPhysicalProcessors();
//There is 2 threads already in play, one UI (main) thread and a GC thread. So subtract 2 from processorCount to account for the same.
this->maxThreadCount = max(1, min(processorCount - 2, CONFIG_FLAG(MaxJitThreadCount)));
}
}
template <typename TAlloc> inline
void
JITThunkEmitter<TAlloc>::FreeThunk(uintptr_t thunkAddress)
{
AutoCriticalSection autoCs(&this->cs);
BVIndex thunkIndex = GetThunkIndexFromAddress(thunkAddress);
if (thunkIndex >= this->freeThunks.Length() || this->freeThunks.TestAndSet(thunkIndex))
{
Assert(UNREACHED);
this->firstBitToCheck = 0;
return;
}
if (thunkIndex < firstBitToCheck)
{
this->firstBitToCheck = thunkIndex;
}
if (CONFIG_FLAG(OOPCFGRegistration))
{
#if ENABLE_OOP_NATIVE_CODEGEN
if (JITManager::GetJITManager()->IsJITServer())
{
HANDLE fileHandle = nullptr;
PVOID baseAddress = nullptr;
bool found = this->codeAllocator->GetFileInfo((PVOID)thunkAddress, &fileHandle, &baseAddress);
AssertOrFailFast(found);
this->threadContext->SetValidCallTargetFile((PVOID)thunkAddress, fileHandle, baseAddress, false);
}
else
#endif
{
this->threadContext->SetValidCallTargetForCFG((PVOID)thunkAddress, false);
}
}
uintptr_t pageStartAddress = GetThunkPageStart(thunkAddress);
if (IsThunkPageEmpty(pageStartAddress))
{
this->codeAllocator->Free((PVOID)pageStartAddress, AutoSystemInfo::PageSize, MEM_DECOMMIT);
}
else
{
char * localAddress = (char *)this->codeAllocator->AllocLocal((PVOID)thunkAddress, ThunkSize);
if (localAddress == nullptr)
{
return;
}
UnprotectPage(localAddress);
memset(localAddress, 0xCC, ThunkSize);
ProtectPage(localAddress);
this->codeAllocator->FreeLocal(localAddress);
}
FlushInstructionCache(this->processHandle, (PVOID)thunkAddress, ThunkSize);
}
void WebAssemblySharedArrayBuffer::ValidateBuffer()
{
#if DBG && _WIN32
if (CONFIG_FLAG(WasmSharedArrayVirtualBuffer))
{
MEMORY_BASIC_INFORMATION info = { 0 };
size_t size = 0;
size_t allocationSize = 0;
// Make sure the beggining of the buffer is committed memory to the expected size
if (sharedContents->bufferLength > 0)
{
size = VirtualQuery((LPCVOID)sharedContents->buffer, &info, sizeof(info));
Assert(size > 0);
allocationSize = info.RegionSize + ((uintptr_t)info.BaseAddress - (uintptr_t)info.AllocationBase);
Assert(allocationSize == sharedContents->bufferLength && info.State == MEM_COMMIT && info.Type == MEM_PRIVATE);
}
// Make sure the end of the buffer is reserved memory to the expected size
size_t expectedAllocationSize = sharedContents->maxBufferLength;
#if ENABLE_FAST_ARRAYBUFFER
if (CONFIG_FLAG(WasmFastArray))
{
expectedAllocationSize = MAX_WASM__ARRAYBUFFER_LENGTH;
}
#endif
// If the whole buffer has been committed, no need to verify this
if (expectedAllocationSize > sharedContents->bufferLength)
{
size = VirtualQuery((LPCVOID)(sharedContents->buffer + sharedContents->bufferLength), &info, sizeof(info));
Assert(size > 0);
allocationSize = info.RegionSize + ((uintptr_t)info.BaseAddress - (uintptr_t)info.AllocationBase);
Assert(allocationSize == expectedAllocationSize && info.State == MEM_RESERVE && info.Type == MEM_PRIVATE);
}
}
#endif
}
void
EmitBufferManager<TAlloc, TPreReservedAlloc, SyncObject>::FreeAllocations(bool release)
{
#if PDATA_ENABLED && defined(_WIN32)
DelayDeletingFunctionTable::Clear();
#endif
AutoRealOrFakeCriticalSection<SyncObject> autoCs(&this->criticalSection);
#if DBG_DUMP
if (!release && PHASE_STATS1(Js::EmitterPhase))
{
this->DumpAndResetStats(Js::Configuration::Global.flags.Filename);
}
#endif
TEmitBufferAllocation * allocation = this->allocations;
while (allocation != nullptr)
{
#ifdef ENABLE_DEBUG_CONFIG_OPTIONS
if(CONFIG_FLAG(CheckEmitBufferPermissions))
{
CheckBufferPermissions(allocation);
}
#endif
if (release)
{
this->allocationHeap.Free(allocation->allocation);
}
else if ((scriptContext != nullptr) && allocation->recorded)
{
// In case of ThunkEmitter the script context would be null and we don't want to track that as code size.
this->scriptContext->GetThreadContext()->SubCodeSize(allocation->bytesCommitted);
allocation->recorded = false;
}
allocation = allocation->nextAllocation;
}
if (release)
{
this->allocations = nullptr;
}
else
{
this->allocationHeap.DecommitAll();
}
}
void WritePerfHint(PerfHints hint, Js::FunctionBody * functionBody, uint byteCodeOffset /*= Js::Constants::NoByteCodeOffset*/)
{
Assert(functionBody);
Assert(((uint)hint) < _countof(s_perfHintContainer));
PerfHintItem item = s_perfHintContainer[(uint)hint];
int level = CONFIG_FLAG(PerfHintLevel);
Assert(level <= (int)PerfHintLevels::VERBOSE);
if ((int)item.level <= level)
{
ULONG lineNumber = functionBody->GetLineNumber();
LONG columnNumber = functionBody->GetColumnNumber();
if (byteCodeOffset != Js::Constants::NoByteCodeOffset)
{
functionBody->GetLineCharOffset(byteCodeOffset, &lineNumber, &columnNumber, false /*canAllocateLineCache*/);
// returned values are 0-based. Adjusting.
lineNumber++;
columnNumber++;
}
// We will print the short name.
TCHAR shortName[255];
Js::FunctionBody::GetShortNameFromUrl(functionBody->GetSourceName(), shortName, 255);
OUTPUT_TRACE(Js::PerfHintPhase, _u("%s : %s {\n Function : %s [%s @ %u, %u]\n Consequences : %s\n Suggestion : %s\n}\n"),
item.isNotOptimized ? _u("Not optimized") : _u("Optimized"),
item.description,
functionBody->GetExternalDisplayName(),
shortName,
lineNumber,
columnNumber,
item.consequences,
item.suggestion);
Output::Flush();
}
}
void Utf8SourceInfo::EnsureInitialized(int initialFunctionCount)
{
ThreadContext* threadContext = ThreadContext::GetContextForCurrentThread();
Recycler* recycler = threadContext->GetRecycler();
if (this->functionBodyDictionary == nullptr)
{
// This collection is allocated with leaf allocation policy. The references to the function body
// here does not keep the function alive. However, the functions remove themselves at finalize
// so if a function actually is in this map, it means that it is alive.
this->functionBodyDictionary = RecyclerNew(recycler, FunctionBodyDictionary, recycler,
initialFunctionCount, threadContext->GetEtwRundownCriticalSection());
}
if (CONFIG_FLAG(DeferTopLevelTillFirstCall) && !m_deferredFunctionsInitialized)
{
Assert(this->m_deferredFunctionsDictionary == nullptr);
this->m_deferredFunctionsDictionary = RecyclerNew(recycler, DeferredFunctionsDictionary, recycler,
initialFunctionCount, threadContext->GetEtwRundownCriticalSection());
m_deferredFunctionsInitialized = true;
}
}
void JsBuiltInEngineInterfaceExtensionObject::EnsureJsBuiltInByteCode(ScriptContext * scriptContext)
{
if (jsBuiltInByteCode == nullptr)
{
SourceContextInfo* sourceContextInfo = RecyclerNewStructZ(scriptContext->GetRecycler(), SourceContextInfo);
sourceContextInfo->dwHostSourceContext = Js::Constants::JsBuiltInSourceContext;
sourceContextInfo->isHostDynamicDocument = true;
sourceContextInfo->sourceContextId = Js::Constants::JsBuiltInSourceContextId;
Assert(sourceContextInfo != nullptr);
SRCINFO si;
memset(&si, 0, sizeof(si));
si.sourceContextInfo = sourceContextInfo;
SRCINFO *hsi = scriptContext->AddHostSrcInfo(&si);
uint32 flags = fscrJsBuiltIn | (CONFIG_FLAG(CreateFunctionProxy) && !scriptContext->IsProfiling() ? fscrAllowFunctionProxy : 0);
HRESULT hr = Js::ByteCodeSerializer::DeserializeFromBuffer(scriptContext, flags, (LPCUTF8)nullptr, hsi, (byte*)Library_Bytecode_JsBuiltIn, nullptr, &jsBuiltInByteCode);
IfFailAssertMsgAndThrowHr(hr, "Failed to deserialize JsBuiltIn.js bytecode - very probably the bytecode needs to be rebuilt.");
this->SetHasBytecode();
}
}
bool Configuration::EnableJitInDebugMode()
{
return CONFIG_FLAG(EnableJitInDiagMode);
}
bool PropertyString::ShouldUseCache() const
{
return this->hitRate > (int)CONFIG_FLAG(StringCacheMissThreshold);
}
void
SwitchIRBuilder::BuildMultiBrCaseInstrForStrings(uint32 targetOffset)
{
Assert(m_caseNodes && m_caseNodes->Count() && m_profiledSwitchInstr && !m_isAsmJs);
if (m_caseNodes->Count() < CONFIG_FLAG(MaxLinearStringCaseCount))
{
int start = 0;
int end = m_caseNodes->Count() - 1;
BuildLinearTraverseInstr(start, end, targetOffset);
ResetCaseNodes();
return;
}
IR::Opnd * srcOpnd = m_caseNodes->Item(0)->GetCaseInstr()->GetSrc1(); // Src1 is same in all the caseNodes
IR::MultiBranchInstr * multiBranchInstr = IR::MultiBranchInstr::New(Js::OpCode::MultiBr, srcOpnd, m_func);
uint32 lastCaseOffset = m_caseNodes->Item(m_caseNodes->Count() - 1)->GetOffset();
uint caseCount = m_caseNodes->Count();
bool generateDictionary = true;
char16 minChar = USHORT_MAX;
char16 maxChar = 0;
// Either the jump table is within the limit (<= 128) or it is dense (<= 2 * case Count)
uint const maxJumpTableSize = max<uint>(CONFIG_FLAG(MaxSingleCharStrJumpTableSize), CONFIG_FLAG(MaxSingleCharStrJumpTableRatio) * caseCount);
if (this->m_seenOnlySingleCharStrCaseNodes)
{
generateDictionary = false;
for (uint i = 0; i < caseCount; i++)
{
Js::JavascriptString * str = m_caseNodes->Item(i)->GetSrc2StringConstLocal();
Assert(str->GetLength() == 1);
char16 currChar = str->GetString()[0];
minChar = min(minChar, currChar);
maxChar = max(maxChar, currChar);
if ((uint)(maxChar - minChar) > maxJumpTableSize)
{
generateDictionary = true;
break;
}
}
}
if (generateDictionary)
{
multiBranchInstr->CreateBranchTargetsAndSetDefaultTarget(caseCount, IR::MultiBranchInstr::StrDictionary, targetOffset);
//Adding normal cases to the instruction (except the default case, which we do it later)
for (uint i = 0; i < caseCount; i++)
{
Js::JavascriptString * str = m_caseNodes->Item(i)->GetSrc2StringConstLocal();
uint32 caseTargetOffset = m_caseNodes->Item(i)->GetTargetOffset();
multiBranchInstr->AddtoDictionary(caseTargetOffset, str, m_caseNodes->Item(i)->GetSrc2StringConst());
}
}
else
{
// If we are only going to save 16 entries, just start from 0 so we don't have to subtract
if (minChar < 16)
{
minChar = 0;
}
multiBranchInstr->m_baseCaseValue = minChar;
multiBranchInstr->m_lastCaseValue = maxChar;
uint jumpTableSize = maxChar - minChar + 1;
multiBranchInstr->CreateBranchTargetsAndSetDefaultTarget(jumpTableSize, IR::MultiBranchInstr::SingleCharStrJumpTable, targetOffset);
for (uint i = 0; i < jumpTableSize; i++)
{
// Initialize all the entries to the default target first.
multiBranchInstr->AddtoJumpTable(targetOffset, i);
}
//Adding normal cases to the instruction (except the default case, which we do it later)
for (uint i = 0; i < caseCount; i++)
{
Js::JavascriptString * str = m_caseNodes->Item(i)->GetSrc2StringConstLocal();
Assert(str->GetLength() == 1);
uint32 caseTargetOffset = m_caseNodes->Item(i)->GetTargetOffset();
multiBranchInstr->AddtoJumpTable(caseTargetOffset, str->GetString()[0] - minChar);
}
}
multiBranchInstr->m_isSwitchBr = true;
m_adapter->CreateRelocRecord(multiBranchInstr, lastCaseOffset, targetOffset);
m_adapter->AddInstr(multiBranchInstr, lastCaseOffset);
BuildBailOnNotString();
ResetCaseNodes();
}
// This only enables collection of the inlinee data, we are much more aggressive here.
// Actual decision of whether something is inlined or not is taken in CommitInlineIntoInliner
bool InliningDecider::DeciderInlineIntoInliner(Js::FunctionBody * inlinee, Js::FunctionBody * inliner, bool isConstructorCall, bool isPolymorphicCall, uint16 constantArgInfo, uint recursiveInlineDepth, bool allowRecursiveInlining)
{
if (!CanRecursivelyInline(inlinee, inliner, allowRecursiveInlining, recursiveInlineDepth))
{
return false;
}
if (inlinee->GetIsAsmjsMode() || inliner->GetIsAsmjsMode())
{
return false;
}
if (PHASE_FORCE(Js::InlinePhase, this->topFunc) ||
PHASE_FORCE(Js::InlinePhase, inliner) ||
PHASE_FORCE(Js::InlinePhase, inlinee))
{
return true;
}
if (PHASE_OFF(Js::InlinePhase, this->topFunc) ||
PHASE_OFF(Js::InlinePhase, inliner) ||
PHASE_OFF(Js::InlinePhase, inlinee))
{
return false;
}
if (PHASE_FORCE(Js::InlineTreePhase, this->topFunc) ||
PHASE_FORCE(Js::InlineTreePhase, inliner))
{
return true;
}
if (PHASE_FORCE(Js::InlineAtEveryCallerPhase, inlinee))
{
return true;
}
uint inlineeByteCodeCount = inlinee->GetByteCodeWithoutLDACount();
// Heuristics are hit in the following order (Note *order* is important)
// 1. Leaf function: If the inlinee is a leaf (but not a constructor or a polymorphic call) inline threshold is LeafInlineThreshold (60). Also it can have max 1 loop
// 2. Constant Function Argument: If the inlinee candidate has a constant argument and that argument is used for branching, then the inline threshold is ConstantArgumentInlineThreshold (157)
// 3. InlineThreshold: If an inlinee candidate exceeds InlineThreshold just don't inline no matter what.
// Following are additional constraint for an inlinee which meets InlineThreshold (Rule no 3)
// 4. Rule for inlinee with loops:
// 4a. Only single loop in inlinee is permitted.
// 4b. Should not have polymorphic field access.
// 4c. Should not be a constructor.
// 4d. Should meet LoopInlineThreshold (25)
// 5. Rule for polymorphic inlinee:
// 4a. Should meet PolymorphicInlineThreshold (32)
// 6. Rule for constructors:
// 5a. Always inline if inlinee has polymorphic field access (as we have cloned runtime data).
// 5b. If inlinee is monomorphic, inline only small constructors. They are governed by ConstructorInlineThreshold (21)
// 7. Rule for inlinee which is not interpreted enough (as we might not have all the profile data):
// 7a. As of now it is still governed by the InlineThreshold. Plan to play with this in future.
// 8. Rest should be inlined.
uint16 mask = constantArgInfo & inlinee->m_argUsedForBranch;
if (mask && inlineeByteCodeCount < (uint)CONFIG_FLAG(ConstantArgumentInlineThreshold))
{
return true;
}
int inlineThreshold = threshold.inlineThreshold;
if (!isPolymorphicCall && !isConstructorCall && IsInlineeLeaf(inlinee) && (inlinee->GetLoopCount() <= 2))
{
// Inlinee is a leaf function
if (inlinee->GetLoopCount() == 0 || GetNumberOfInlineesWithLoop() <= (uint)threshold.maxNumberOfInlineesWithLoop) // Don't inlinee too many inlinees with loops.
{
// Negative LeafInlineThreshold disable the threshold
if (threshold.leafInlineThreshold >= 0)
{
inlineThreshold += threshold.leafInlineThreshold - threshold.inlineThreshold;
}
}
}
#if ENABLE_DEBUG_CONFIG_OPTIONS
char16 debugStringBuffer[MAX_FUNCTION_BODY_DEBUG_STRING_SIZE];
char16 debugStringBuffer2[MAX_FUNCTION_BODY_DEBUG_STRING_SIZE];
char16 debugStringBuffer3[MAX_FUNCTION_BODY_DEBUG_STRING_SIZE];
#endif
if (inlinee->GetHasLoops())
{
if (threshold.loopInlineThreshold < 0 || // Negative LoopInlineThreshold disable inlining with loop
GetNumberOfInlineesWithLoop() >(uint)threshold.maxNumberOfInlineesWithLoop || // See if we are inlining too many inlinees with loops.
(inlinee->GetLoopCount() > 2) || // Allow at most 2 loops.
inlinee->GetHasNestedLoop() || // Nested loops are not a good inlinee candidate
isConstructorCall || // If the function is constructor with loops, don't inline.
PHASE_OFF(Js::InlineFunctionsWithLoopsPhase, this->topFunc))
{
INLINE_TESTTRACE(_u("INLINING: Skip Inline: Has loops \tBytecode size: %d \tgetNumberOfInlineesWithLoop: %d\tloopCount: %d\thasNestedLoop: %B\tisConstructorCall:%B\tInlinee: %s (%s)\tCaller: %s (%s) \tRoot: %s (%s)\n"),
inlinee->GetByteCodeCount(),
GetNumberOfInlineesWithLoop(),
inlinee->GetLoopCount(),
inlinee->GetHasNestedLoop(),
isConstructorCall,
inlinee->GetDisplayName(), inlinee->GetDebugNumberSet(debugStringBuffer),
inliner->GetDisplayName(), inliner->GetDebugNumberSet(debugStringBuffer2),
topFunc->GetDisplayName(), topFunc->GetDebugNumberSet(debugStringBuffer3));
// Don't inline function with loops
return false;
}
else
{
inlineThreshold -= (threshold.inlineThreshold > threshold.loopInlineThreshold) ? threshold.inlineThreshold - threshold.loopInlineThreshold : 0;
}
}
if (isPolymorphicCall)
{
if (threshold.polymorphicInlineThreshold < 0 || // Negative PolymorphicInlineThreshold disable inlining
isConstructorCall)
{
INLINE_TESTTRACE(_u("INLINING: Skip Inline: Polymorphic call under PolymorphicInlineThreshold: %d \tBytecode size: %d\tInlinee: %s (%s)\tCaller: %s (%s) \tRoot: %s (%s)\n"),
threshold.polymorphicInlineThreshold,
inlinee->GetByteCodeCount(),
inlinee->GetDisplayName(), inlinee->GetDebugNumberSet(debugStringBuffer),
inliner->GetDisplayName(), inliner->GetDebugNumberSet(debugStringBuffer2),
topFunc->GetDisplayName(), topFunc->GetDebugNumberSet(debugStringBuffer3));
return false;
}
else
{
inlineThreshold -= (threshold.inlineThreshold > threshold.polymorphicInlineThreshold) ? threshold.inlineThreshold - threshold.polymorphicInlineThreshold : 0;
}
}
if (isConstructorCall)
{
#pragma prefast(suppress: 6285, "logical-or of constants is by design")
if (PHASE_OFF(Js::InlineConstructorsPhase, this->topFunc) ||
PHASE_OFF(Js::InlineConstructorsPhase, inliner) ||
PHASE_OFF(Js::InlineConstructorsPhase, inlinee) ||
!CONFIG_FLAG(CloneInlinedPolymorphicCaches))
{
return false;
}
if (PHASE_FORCE(Js::InlineConstructorsPhase, this->topFunc) ||
PHASE_FORCE(Js::InlineConstructorsPhase, inliner) ||
PHASE_FORCE(Js::InlineConstructorsPhase, inlinee))
{
return true;
}
if (inlinee->HasDynamicProfileInfo() && inlinee->GetAnyDynamicProfileInfo()->HasPolymorphicFldAccess())
{
// As of now this is not dependent on bytecodeInlinedThreshold.
return true;
}
// Negative ConstructorInlineThreshold always disable constructor inlining
if (threshold.constructorInlineThreshold < 0)
{
INLINE_TESTTRACE(_u("INLINING: Skip Inline: Constructor with no polymorphic field access \tBytecode size: %d\tInlinee: %s (%s)\tCaller: %s (%s) \tRoot: %s (%s)\n"),
inlinee->GetByteCodeCount(),
inlinee->GetDisplayName(), inlinee->GetDebugNumberSet(debugStringBuffer),
inliner->GetDisplayName(), inliner->GetDebugNumberSet(debugStringBuffer2),
topFunc->GetDisplayName(), topFunc->GetDebugNumberSet(debugStringBuffer3));
// Don't inline constructor that does not have a polymorphic field access, or if cloning polymorphic inline
// caches is disabled
return false;
}
else
{
inlineThreshold -= (threshold.inlineThreshold > threshold.constructorInlineThreshold) ? threshold.inlineThreshold - threshold.constructorInlineThreshold : 0;
}
}
if (threshold.forLoopBody)
{
inlineThreshold /= CONFIG_FLAG(InlineInLoopBodyScaleDownFactor);
}
if (inlineThreshold > 0 && inlineeByteCodeCount <= (uint)inlineThreshold)
{
if (inlinee->GetLoopCount())
{
IncrementNumberOfInlineesWithLoop();
}
return true;
}
else
{
return false;
}
}
bool AutoSystemInfo::IsLowMemoryProcess()
{
ULONG64 commit = ULONG64(-1);
this->GetAvailableCommit(&commit);
return commit <= CONFIG_FLAG(LowMemoryCap);
}
void
SwitchIRBuilder::BuildOptimizedIntegerCaseInstrs(uint32 targetOffset)
{
int startjmpTableIndex = 0;
int endjmpTableIndex = 0;
int startBinaryTravIndex = 0;
int endBinaryTravIndex = 0;
IR::MultiBranchInstr * multiBranchInstr = nullptr;
/*
* Algorithm to find chunks of consecutive integers in a given set of case arms(sorted)
* -Start and end indices for jump table and binary tree are maintained.
* -The corresponding start and end indices indicate that they are suitable candidate for their respective category(binaryTree/jumpTable)
* -All holes are filled with an offset corresponding to the default fallthrough instruction and each block is filled with an offset corresponding to the start of the next block
* A Block here refers either to a jump table or to a binary tree.
* -Blocks of BinaryTrav/Jump table are traversed in a linear fashion.
**/
for (int currentIndex = 0; currentIndex < m_caseNodes->Count() - 1; currentIndex++)
{
int nextIndex = currentIndex + 1;
//Check if there is no missing value between subsequent case arms
if (m_caseNodes->Item(currentIndex)->GetSrc2IntConst() + 1 != m_caseNodes->Item(nextIndex)->GetSrc2IntConst())
{
//value of the case nodes are guaranteed to be 32 bits or less than 32bits at this point(if it is more, the Switch Opt will not kick in)
Assert(nextIndex == endjmpTableIndex + 1);
int64 speculatedEndJmpCaseValue = m_caseNodes->Item(nextIndex)->GetSrc2IntConst();
int64 endJmpCaseValue = m_caseNodes->Item(endjmpTableIndex)->GetSrc2IntConst();
int64 startJmpCaseValue = m_caseNodes->Item(startjmpTableIndex)->GetSrc2IntConst();
int64 speculatedJmpTableSize = speculatedEndJmpCaseValue - startJmpCaseValue + 1;
int64 jmpTableSize = endJmpCaseValue - startJmpCaseValue + 1;
int numFilledEntries = nextIndex - startjmpTableIndex + 1;
//Checks if the % of filled entries(unique targets from the case arms) in the jump table is within the threshold
if (speculatedJmpTableSize != 0 && ((numFilledEntries)* 100 / speculatedJmpTableSize) < (100 - CONFIG_FLAG(SwitchOptHolesThreshold)))
{
if (jmpTableSize >= CONFIG_FLAG(MinSwitchJumpTableSize))
{
uint32 fallThrOffset = m_caseNodes->Item(endjmpTableIndex)->GetOffset();
TryBuildBinaryTreeOrMultiBrForSwitchInts(multiBranchInstr, fallThrOffset, startjmpTableIndex, endjmpTableIndex, startBinaryTravIndex, targetOffset);
//Reset start/end indices of BinaryTrav to the next index.
startBinaryTravIndex = nextIndex;
endBinaryTravIndex = nextIndex;
}
//Reset start/end indices of the jump table to the next index.
startjmpTableIndex = nextIndex;
endjmpTableIndex = nextIndex;
}
else
{
endjmpTableIndex++;
}
}
else
{
endjmpTableIndex++;
}
}
int64 endJmpCaseValue = m_caseNodes->Item(endjmpTableIndex)->GetSrc2IntConst();
int64 startJmpCaseValue = m_caseNodes->Item(startjmpTableIndex)->GetSrc2IntConst();
int64 jmpTableSize = endJmpCaseValue - startJmpCaseValue + 1;
if (jmpTableSize < CONFIG_FLAG(MinSwitchJumpTableSize))
{
endBinaryTravIndex = endjmpTableIndex;
BuildBinaryTraverseInstr(startBinaryTravIndex, endBinaryTravIndex, targetOffset);
if (multiBranchInstr)
{
FixUpMultiBrJumpTable(multiBranchInstr, multiBranchInstr->GetNextRealInstr()->GetByteCodeOffset());
multiBranchInstr = nullptr;
}
}
else
{
uint32 fallthrOffset = m_caseNodes->Item(endjmpTableIndex)->GetOffset();
TryBuildBinaryTreeOrMultiBrForSwitchInts(multiBranchInstr, fallthrOffset, startjmpTableIndex, endjmpTableIndex, startBinaryTravIndex, targetOffset);
FixUpMultiBrJumpTable(multiBranchInstr, targetOffset);
}
}
void ConfigParser::ProcessConfiguration(HANDLE hmod)
{
#if defined(ENABLE_DEBUG_CONFIG_OPTIONS)
bool hasOutput = false;
char16 modulename[_MAX_PATH];
GetModuleFileName((HMODULE)hmod, modulename, _MAX_PATH);
// Win32 specific console creation code
// xplat-todo: Consider having this mechanism available on other
// platforms
// Not a pressing need since ChakraCore runs only in consoles by
// default so we don't need to allocate a second console for this
#if CONFIG_CONSOLE_AVAILABLE
if (Js::Configuration::Global.flags.Console)
{
int fd;
FILE *fp;
// fail usually means there is an existing console. We don't really care.
AllocConsole();
fd = _open_osfhandle((intptr_t)GetStdHandle(STD_OUTPUT_HANDLE), O_TEXT);
fp = _wfdopen(fd, _u("w"));
if (fp != nullptr)
{
*stdout = *fp;
setvbuf(stdout, nullptr, _IONBF, 0);
fd = _open_osfhandle((intptr_t)GetStdHandle(STD_ERROR_HANDLE), O_TEXT);
fp = _wfdopen(fd, _u("w"));
if (fp != nullptr)
{
*stderr = *fp;
setvbuf(stderr, nullptr, _IONBF, 0);
char16 buffer[_MAX_PATH + 70];
if (ConfigParserAPI::FillConsoleTitle(buffer, _MAX_PATH + 20, modulename))
{
SetConsoleTitle(buffer);
}
hasOutput = true;
}
}
}
#endif
if (Js::Configuration::Global.flags.IsEnabled(Js::OutputFileFlag)
&& Js::Configuration::Global.flags.OutputFile != nullptr)
{
SetOutputFile(Js::Configuration::Global.flags.OutputFile, Js::Configuration::Global.flags.OutputFileOpenMode);
hasOutput = true;
}
if (Js::Configuration::Global.flags.DebugWindow)
{
Output::UseDebuggerWindow();
hasOutput = true;
}
#ifdef ENABLE_TRACE
if (CONFIG_FLAG(InMemoryTrace))
{
Output::SetInMemoryLogger(
Js::MemoryLogger::Create(::GetOutputAllocator1(),
CONFIG_FLAG(InMemoryTraceBufferSize) * 3)); // With stack each trace is 3 entries (header, msg, stack).
hasOutput = true;
}
#ifdef STACK_BACK_TRACE
if (CONFIG_FLAG(TraceWithStack))
{
Output::SetStackTraceHelper(Js::StackTraceHelper::Create(::GetOutputAllocator2()));
}
#endif // STACK_BACK_TRACE
#endif // ENABLE_TRACE
if (hasOutput)
{
ConfigParserAPI::DisplayInitialOutput(modulename);
Output::Print(_u("\n"));
Js::Configuration::Global.flags.VerboseDump();
Output::Flush();
}
if (Js::Configuration::Global.flags.ForceSerialized)
{
// Can't generate or execute byte code under forced serialize
Js::Configuration::Global.flags.GenerateByteCodeBufferReturnsCantGenerate = true;
Js::Configuration::Global.flags.ExecuteByteCodeBufferReturnsInvalidByteCode = true;
}
ForcedMemoryConstraint::Apply();
#endif
#ifdef MEMSPECT_TRACKING
bool all = false;
if (Js::Configuration::Global.flags.Memspect.IsEnabled(Js::AllPhase))
{
all = true;
}
if (all || Js::Configuration::Global.flags.Memspect.IsEnabled(Js::RecyclerPhase))
{
RecyclerMemoryTracking::Activate();
}
if (all || Js::Configuration::Global.flags.Memspect.IsEnabled(Js::PageAllocatorPhase))
{
PageTracking::Activate();
}
if (all || Js::Configuration::Global.flags.Memspect.IsEnabled(Js::ArenaPhase))
{
ArenaMemoryTracking::Activate();
}
#endif
}
void InliningThreshold::SetHeuristics()
{
inlineThreshold = CONFIG_FLAG(InlineThreshold);
// Inline less aggressively in large functions since the register pressure is likely high.
// Small functions shouldn't be a problem.
if (topFunc->GetByteCodeWithoutLDACount() > 800)
{
inlineThreshold -= CONFIG_FLAG(InlineThresholdAdjustCountInLargeFunction);
}
else if (topFunc->GetByteCodeWithoutLDACount() > 200)
{
inlineThreshold -= CONFIG_FLAG(InlineThresholdAdjustCountInMediumSizedFunction);
}
else if (topFunc->GetByteCodeWithoutLDACount() < 50)
{
inlineThreshold += CONFIG_FLAG(InlineThresholdAdjustCountInSmallFunction);
}
constructorInlineThreshold = CONFIG_FLAG(ConstructorInlineThreshold);
outsideLoopInlineThreshold = CONFIG_FLAG(OutsideLoopInlineThreshold);
leafInlineThreshold = CONFIG_FLAG(LeafInlineThreshold);
loopInlineThreshold = CONFIG_FLAG(LoopInlineThreshold);
polymorphicInlineThreshold = CONFIG_FLAG(PolymorphicInlineThreshold);
maxNumberOfInlineesWithLoop = CONFIG_FLAG(MaxNumberOfInlineesWithLoop);
constantArgumentInlineThreshold = CONFIG_FLAG(ConstantArgumentInlineThreshold);
inlineCountMax = CONFIG_FLAG(InlineCountMax);
}
