Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
/*
|
Support compiling catch in the DFG
https://bugs.webkit.org/show_bug.cgi?id=174590
<rdar://problem/34047845>
Reviewed by Filip Pizlo.
JSTests:
* microbenchmarks/delta-blue-try-catch.js: Added.
(exception):
(value):
(OrderedCollection):
(OrderedCollection.prototype.add):
(OrderedCollection.prototype.at):
(OrderedCollection.prototype.size):
(OrderedCollection.prototype.removeFirst):
(OrderedCollection.prototype.remove):
(Strength):
(Strength.stronger):
(Strength.weaker):
(Strength.weakestOf):
(Strength.strongest):
(Strength.prototype.nextWeaker):
(Constraint):
(Constraint.prototype.addConstraint):
(Constraint.prototype.satisfy):
(Constraint.prototype.destroyConstraint):
(Constraint.prototype.isInput):
(UnaryConstraint):
(UnaryConstraint.prototype.addToGraph):
(UnaryConstraint.prototype.chooseMethod):
(UnaryConstraint.prototype.isSatisfied):
(UnaryConstraint.prototype.markInputs):
(UnaryConstraint.prototype.output):
(UnaryConstraint.prototype.recalculate):
(UnaryConstraint.prototype.markUnsatisfied):
(UnaryConstraint.prototype.inputsKnown):
(UnaryConstraint.prototype.removeFromGraph):
(StayConstraint):
(StayConstraint.prototype.execute):
(EditConstraint.prototype.isInput):
(EditConstraint.prototype.execute):
(BinaryConstraint):
(BinaryConstraint.prototype.chooseMethod):
(BinaryConstraint.prototype.addToGraph):
(BinaryConstraint.prototype.isSatisfied):
(BinaryConstraint.prototype.markInputs):
(BinaryConstraint.prototype.input):
(BinaryConstraint.prototype.output):
(BinaryConstraint.prototype.recalculate):
(BinaryConstraint.prototype.markUnsatisfied):
(BinaryConstraint.prototype.inputsKnown):
(BinaryConstraint.prototype.removeFromGraph):
(ScaleConstraint):
(ScaleConstraint.prototype.addToGraph):
(ScaleConstraint.prototype.removeFromGraph):
(ScaleConstraint.prototype.markInputs):
(ScaleConstraint.prototype.execute):
(ScaleConstraint.prototype.recalculate):
(EqualityConstraint):
(EqualityConstraint.prototype.execute):
(Variable):
(Variable.prototype.addConstraint):
(Variable.prototype.removeConstraint):
(Planner):
(Planner.prototype.incrementalAdd):
(Planner.prototype.incrementalRemove):
(Planner.prototype.newMark):
(Planner.prototype.makePlan):
(Planner.prototype.extractPlanFromConstraints):
(Planner.prototype.addPropagate):
(Planner.prototype.removePropagateFrom):
(Planner.prototype.addConstraintsConsumingTo):
(Plan):
(Plan.prototype.addConstraint):
(Plan.prototype.size):
(Plan.prototype.constraintAt):
(Plan.prototype.execute):
(chainTest):
(projectionTest):
(change):
(deltaBlue):
* microbenchmarks/fake-iterators-that-throw-when-finished.js: Added.
(assert):
(Numbers):
(Numbers.prototype.next):
(return.Transpose):
(return.Transpose.prototype.next):
(transpose):
(verifyEven):
(verifyString):
(foo):
(runIterators):
* microbenchmarks/try-catch-word-count.js: Added.
(let.assert):
(EOF):
(let.texts):
(let.o.apply):
(foo):
(bar):
(f):
(run):
(test1):
(test2):
(test3):
(fn):
(A):
(B):
(A.prototype.getValue):
(B.prototype.getParentValue):
(strlen):
(sum.0):
(test):
(result.test.o):
(set add.set add):
(set forEach):
(stringHash):
(set if):
(testFunction):
(set delete.set has.set add):
* stress/catch-set-argument-speculation-failure.js: Added.
(o):
(e):
(e2):
(escape):
(baz):
(noInline.run):
(noInline):
* stress/osr-enter-to-catch-with-set-local-type-check-failure.js: Added.
(foo):
(e):
(baz):
(bar):
Source/JavaScriptCore:
This patch implements OSR entry into op_catch in the DFG. We will support OSR entry
into the FTL in a followup: https://bugs.webkit.org/show_bug.cgi?id=175396
To implement catch in the DFG, this patch introduces the concept of multiple
entrypoints into CPS/LoadStore DFG IR. A lot of this patch is stringing this concept
through the DFG. Many phases used to assume that Graph::block(0) is the only root, and this
patch contains many straight forward changes generalizing the code to handle more than
one entrypoint.
A main building block of this is moving to two CFG types: SSACFG and CPSCFG. SSACFG
is the same CFG we used to have. CPSCFG is a new type that introduces a fake root
that has an outgoing edge to all the entrypoints. This allows our existing graph algorithms
to Just Work over CPSCFG. For example, there is now the concept of SSADominators vs CPSDominators,
and SSANaturalLoops vs CPSNaturalLoops.
The way we compile the catch entrypoint is by bootstrapping the state
of the program by loading all live bytecode locals from a buffer. The OSR
entry code will store all live values into that buffer before jumping to
the entrypoint. The OSR entry code is also responsible for performing type
proofs of the arguments before doing an OSR entry. If there is a type
mismatch, it's not legal to OSR enter into the DFG compilation. Currently,
each catch entrypoint knows the argument type proofs it must perform to enter
into the DFG. Currently, all entrypoints' arguments flush format are unified
via ArgumentPosition, but this is just an implementation detail. The code is
written more generally to assume that each entrypoint may perform its own distinct
proof.
op_catch now performs value profiling for all live bytecode locals in the
LLInt and baseline JIT. This information is then fed into the DFG via the
ExtractCatchLocal node in the prediction propagation phase.
This patch also changes how we generate op_catch in bytecode. All op_catches
are now split out at the end of the program in bytecode. This ensures that
no op_catch is inside a try block. This is needed to ensure correctness in
the DFGLiveCatchVariablePreservationPhase. That phase only inserts flushes
before SetLocals inside a try block. If an op_catch were in a try block, this
would cause the phase to insert a Flush before one of the state bootstrapping
SetLocals, which would generate invalid IR. Moving op_catch to be generated on
its own at the end of a bytecode stream seemed like the most elegant solution since
it better represents that we treat op_catch as an entrypoint. This is true
both in the DFG and in the baseline and LLInt: we don't reach an op_catch
via normal control flow. Because op_catch cannot throw, this will not break
any previous semantics of op_catch. Logically, it'd be valid to split try
blocks around any non-throwing bytecode operation.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* bytecode/BytecodeDumper.cpp:
(JSC::BytecodeDumper<Block>::dumpBytecode):
* bytecode/BytecodeList.json:
* bytecode/BytecodeUseDef.h:
(JSC::computeUsesForBytecodeOffset):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::finishCreation):
(JSC::CodeBlock::ensureCatchLivenessIsComputedForBytecodeOffset):
(JSC::CodeBlock::updateAllPredictionsAndCountLiveness):
(JSC::CodeBlock::validate):
* bytecode/CodeBlock.h:
* bytecode/ValueProfile.h:
(JSC::ValueProfile::ValueProfile):
(JSC::ValueProfileAndOperandBuffer::ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::~ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::forEach):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::generate):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::emitCatch):
(JSC::BytecodeGenerator::emitEnumeration):
* bytecompiler/BytecodeGenerator.h:
* bytecompiler/NodesCodegen.cpp:
(JSC::TryNode::emitBytecode):
* dfg/DFGAbstractInterpreterInlines.h:
(JSC::DFG::AbstractInterpreter<AbstractStateType>::executeEffects):
* dfg/DFGBackwardsCFG.h:
(JSC::DFG::BackwardsCFG::BackwardsCFG):
* dfg/DFGBasicBlock.cpp:
(JSC::DFG::BasicBlock::BasicBlock):
* dfg/DFGBasicBlock.h:
(JSC::DFG::BasicBlock::findTerminal const):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::setDirect):
(JSC::DFG::ByteCodeParser::flush):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::DelayedSetLocal):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::execute):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::parseCodeBlock):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCFG.h:
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::roots):
(JSC::DFG::CPSCFG::CPSCFG):
(JSC::DFG::selectCFG):
* dfg/DFGCPSRethreadingPhase.cpp:
(JSC::DFG::CPSRethreadingPhase::specialCaseArguments):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGClobberize.h:
(JSC::DFG::clobberize):
* dfg/DFGControlEquivalenceAnalysis.h:
(JSC::DFG::ControlEquivalenceAnalysis::ControlEquivalenceAnalysis):
* dfg/DFGDCEPhase.cpp:
(JSC::DFG::DCEPhase::run):
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDoesGC.cpp:
(JSC::DFG::doesGC):
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::ensureDominatorsForCFG):
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::EdgeDominates):
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
(JSC::DFG::FixupPhase::fixupChecksInBlock):
* dfg/DFGFlushFormat.h:
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::unboxLoopNode):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::dump):
(JSC::DFG::Graph::determineReachability):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::blocksInPreOrder):
(JSC::DFG::Graph::blocksInPostOrder):
(JSC::DFG::Graph::ensureCPSDominators):
(JSC::DFG::Graph::ensureSSADominators):
(JSC::DFG::Graph::ensureCPSNaturalLoops):
(JSC::DFG::Graph::ensureSSANaturalLoops):
(JSC::DFG::Graph::ensureBackwardsCFG):
(JSC::DFG::Graph::ensureBackwardsDominators):
(JSC::DFG::Graph::ensureControlEquivalenceAnalysis):
(JSC::DFG::Graph::methodOfGettingAValueProfileFor):
(JSC::DFG::Graph::clearCPSCFGData):
(JSC::DFG::Graph::ensureDominators): Deleted.
(JSC::DFG::Graph::ensurePrePostNumbering): Deleted.
(JSC::DFG::Graph::ensureNaturalLoops): Deleted.
* dfg/DFGGraph.h:
(JSC::DFG::Graph::willCatchExceptionInMachineFrame):
(JSC::DFG::Graph::isEntrypoint const):
* dfg/DFGInPlaceAbstractState.cpp:
(JSC::DFG::InPlaceAbstractState::initialize):
(JSC::DFG::InPlaceAbstractState::mergeToSuccessors):
* dfg/DFGJITCode.cpp:
(JSC::DFG::JITCode::shrinkToFit):
* dfg/DFGJITCode.h:
(JSC::DFG::JITCode::catchOSREntryDataForBytecodeIndex):
(JSC::DFG::JITCode::finalizeCatchOSREntrypoints):
(JSC::DFG::JITCode::appendCatchEntrypoint):
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::compile):
(JSC::DFG::JITCompiler::compileFunction):
(JSC::DFG::JITCompiler::noticeCatchEntrypoint):
(JSC::DFG::JITCompiler::noticeOSREntry):
(JSC::DFG::JITCompiler::makeCatchOSREntryBuffer):
* dfg/DFGJITCompiler.h:
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::run):
(JSC::DFG::LiveCatchVariablePreservationPhase::isValidFlushLocation):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlockForTryCatch):
(JSC::DFG::LiveCatchVariablePreservationPhase::newVariableAccessData):
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException): Deleted.
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock): Deleted.
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGMaximalFlushInsertionPhase.cpp:
(JSC::DFG::MaximalFlushInsertionPhase::run):
(JSC::DFG::MaximalFlushInsertionPhase::treatRegularBlock):
(JSC::DFG::MaximalFlushInsertionPhase::treatRootBlock):
* dfg/DFGMayExit.cpp:
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::isSwitch const):
(JSC::DFG::Node::successor):
(JSC::DFG::Node::catchOSREntryIndex const):
(JSC::DFG::Node::catchLocalPrediction):
(JSC::DFG::Node::isSwitch): Deleted.
* dfg/DFGNodeType.h:
* dfg/DFGOSREntry.cpp:
(JSC::DFG::prepareCatchOSREntry):
* dfg/DFGOSREntry.h:
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGOSRExitCompilerCommon.cpp:
(JSC::DFG::handleExitCounts):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering): Deleted.
(JSC::DFG::PrePostNumbering::~PrePostNumbering): Deleted.
(WTF::printInternal): Deleted.
* dfg/DFGPrePostNumbering.h:
(): Deleted.
(JSC::DFG::PrePostNumbering::preNumber const): Deleted.
(JSC::DFG::PrePostNumbering::postNumber const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::isDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::edgeKind const): Deleted.
* dfg/DFGPredictionInjectionPhase.cpp:
(JSC::DFG::PredictionInjectionPhase::run):
* dfg/DFGPredictionPropagationPhase.cpp:
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
(JSC::DFG::performSSAConversion):
* dfg/DFGSafeToExecute.h:
(JSC::DFG::safeToExecute):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compileCurrentBlock):
(JSC::DFG::SpeculativeJIT::checkArgumentTypes):
(JSC::DFG::SpeculativeJIT::createOSREntries):
(JSC::DFG::SpeculativeJIT::linkOSREntries):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGStaticExecutionCountEstimationPhase.cpp:
(JSC::DFG::StaticExecutionCountEstimationPhase::run):
* dfg/DFGStrengthReductionPhase.cpp:
(JSC::DFG::StrengthReductionPhase::handleNode):
* dfg/DFGTierUpCheckInjectionPhase.cpp:
(JSC::DFG::TierUpCheckInjectionPhase::run):
(JSC::DFG::TierUpCheckInjectionPhase::buildNaturalLoopToLoopHintMap):
* dfg/DFGTypeCheckHoistingPhase.cpp:
(JSC::DFG::TypeCheckHoistingPhase::run):
* dfg/DFGValidate.cpp:
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToB3.cpp:
(JSC::FTL::DFG::LowerDFGToB3::lower):
(JSC::FTL::DFG::LowerDFGToB3::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToB3::isValid):
* jit/JIT.h:
* jit/JITInlines.h:
(JSC::JIT::callOperation):
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOperations.cpp:
* jit/JITOperations.h:
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* llint/LLIntSlowPaths.h:
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
Source/WTF:
This patch generalizes the BackwardsGraph fake root into a more generalizable
class called SingleRootGraph. SingleRootGraph exposes the general graph interface
used in Dominators and NaturalLoops. SingleRootGraph takes as input a graph with
the normal graph interface, but also allows the input graph to contain more than
one root. SingleRootGraph then exposes a single root, which it creates, that has
an outgoing edge to all the roots in the original graph.
* WTF.xcodeproj/project.pbxproj:
* wtf/BackwardsGraph.h:
(WTF::BackwardsGraph::dump const):
(WTF::BackwardsGraph::rootName): Deleted.
(WTF::BackwardsGraph::Node::Node): Deleted.
(WTF::BackwardsGraph::Node::root): Deleted.
(WTF::BackwardsGraph::Node::operator== const): Deleted.
(WTF::BackwardsGraph::Node::operator!= const): Deleted.
(WTF::BackwardsGraph::Node::operator bool const): Deleted.
(WTF::BackwardsGraph::Node::isRoot const): Deleted.
(WTF::BackwardsGraph::Node::node const): Deleted.
(): Deleted.
(WTF::BackwardsGraph::Set::Set): Deleted.
(WTF::BackwardsGraph::Set::add): Deleted.
(WTF::BackwardsGraph::Set::remove): Deleted.
(WTF::BackwardsGraph::Set::contains): Deleted.
(WTF::BackwardsGraph::Set::dump const): Deleted.
(WTF::BackwardsGraph::Map::Map): Deleted.
(WTF::BackwardsGraph::Map::clear): Deleted.
(WTF::BackwardsGraph::Map::size const): Deleted.
(WTF::BackwardsGraph::Map::operator[]): Deleted.
(WTF::BackwardsGraph::Map::operator[] const): Deleted.
* wtf/Dominators.h:
(WTF::Dominators::Dominators):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf const):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl const):
* wtf/SingleRootGraph.h: Added.
(WTF::SingleRootGraphNode::rootName):
(WTF::SingleRootGraphNode::SingleRootGraphNode):
(WTF::SingleRootGraphNode::root):
(WTF::SingleRootGraphNode::operator== const):
(WTF::SingleRootGraphNode::operator!= const):
(WTF::SingleRootGraphNode::operator bool const):
(WTF::SingleRootGraphNode::isRoot const):
(WTF::SingleRootGraphNode::node const):
(WTF::SingleRootGraphSet::add):
(WTF::SingleRootGraphSet::remove):
(WTF::SingleRootGraphSet::contains):
(WTF::SingleRootGraphSet::dump const):
(WTF::SingleRootMap::SingleRootMap):
(WTF::SingleRootMap::clear):
(WTF::SingleRootMap::size const):
(WTF::SingleRootMap::operator[]):
(WTF::SingleRootMap::operator[] const):
(WTF::SingleRootGraph::SingleRootGraph):
(WTF::SingleRootGraph::root const):
(WTF::SingleRootGraph::newMap):
(WTF::SingleRootGraph::successors const):
(WTF::SingleRootGraph::predecessors const):
(WTF::SingleRootGraph::index const):
(WTF::SingleRootGraph::node const):
(WTF::SingleRootGraph::numNodes const):
(WTF::SingleRootGraph::dump const):
(WTF::SingleRootGraph::assertIsConsistent const):
Canonical link: https://commits.webkit.org/192644@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@221196 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2017-08-25 18:26:15 +00:00
|
|
|
* Copyright (C) 2011-2017 Apple Inc. All rights reserved.
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
*
|
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
|
* are met:
|
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
|
*
|
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY
|
|
|
|
|
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
|
|
|
|
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR
|
|
|
|
|
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
|
|
|
|
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
|
|
|
|
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
|
|
|
|
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
|
|
|
|
|
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
|
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
|
|
|
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
|
*/
|
|
|
|
|
|
2018-10-15 14:24:49 +00:00
|
|
|
#pragma once
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
|
2017-07-26 01:58:36 +00:00
|
|
|
#include <wtf/CommaPrinter.h>
|
FTL B3 should be able to flag the tag constants as being super important so that B3 can hoist them and Air can force them into registers
https://bugs.webkit.org/show_bug.cgi?id=151955
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Taught B3 about the concept of "fast constants". A client of B3 can now tell B3 which
constants are super important. B3 will not spill the constant in that case and will ensure
that the constant is materialized only once: statically once, and dynamically once per
procedure execution. The hoistFastConstants() algorithm in B3MoveConstants.cpp achieves this
by first picking the lowest common dominator of all uses of each fast constant, and then
picking the materialization point by finding the lowest dominator of that dominator that is
tied for lowest block frequency. In practice, the second step ensures that this is the lowest
point in the program that is not in a loop (i.e. executes no more than once dynamically per
procedure invocation).
Taught Air about the concept of "fast tmps". B3 tells Air that a tmp is fast if it is used to
hold the materialization of a fast constant. IRC will use the lowest possible spill score for
fast tmps. In practice, this ensures that fast constants are never spilled.
Added a small snippet of code to FTL::LowerDFGToLLVM that makes both of the tag constants
into fast constants.
My hope is that this very brute-force heuristic is good enough that we don't have to think
about constants for a while. Based on my experience with how LLVM's constant hoisting works
out, the heuristic in this patch is going to be tough to beat. LLVM's constant hoisting does
good things when it hoists the tags, and usually causes nothing but problems when it hoists
anything else. This is because there is no way a low-level compiler to really understand how
a constant materialization impacts some operation's contribution to the overall execution
time of a procedure. But, in the FTL we know that constant materializations for type checks
are a bummer because we are super comfortable placing type checks on the hottest of paths. So
those are the last paths where extra instructions should be added by the compiler. On the
other hand, all other large constant uses are on relatively cold paths, or paths that are
already expensive for other reasons. For example, global variable accesses have to
materialize a pointer to the global. But that's not really a big deal, since a load from a
global involves first the load itself and then type checks on the result - so probably the
constant materialization is just not interesting. A store to a global often involves a store
barrier, so the constant materialization is really not interesting. This patch codifies this
heuristic in a pact between Air, B3, and the FTL: FTL demands that B3 pin the two tags in
registers, and B3 relays the demand to Air.
* JavaScriptCore.xcodeproj/project.pbxproj:
* b3/B3CFG.h: Added.
(JSC::B3::CFG::CFG):
(JSC::B3::CFG::root):
(JSC::B3::CFG::newMap):
(JSC::B3::CFG::successors):
(JSC::B3::CFG::predecessors):
(JSC::B3::CFG::index):
(JSC::B3::CFG::node):
(JSC::B3::CFG::numNodes):
(JSC::B3::CFG::dump):
* b3/B3Dominators.h: Added.
(JSC::B3::Dominators::Dominators):
* b3/B3IndexMap.h:
(JSC::B3::IndexMap::resize):
(JSC::B3::IndexMap::size):
(JSC::B3::IndexMap::operator[]):
* b3/B3LowerMacros.cpp:
* b3/B3LowerToAir.cpp:
(JSC::B3::Air::LowerToAir::tmp):
* b3/B3MoveConstants.cpp:
* b3/B3Opcode.h:
(JSC::B3::constPtrOpcode):
(JSC::B3::isConstant):
* b3/B3Procedure.cpp:
(JSC::B3::Procedure::Procedure):
(JSC::B3::Procedure::resetReachability):
(JSC::B3::Procedure::invalidateCFG):
(JSC::B3::Procedure::dump):
(JSC::B3::Procedure::deleteValue):
(JSC::B3::Procedure::dominators):
(JSC::B3::Procedure::addFastConstant):
(JSC::B3::Procedure::isFastConstant):
(JSC::B3::Procedure::addDataSection):
* b3/B3Procedure.h:
(JSC::B3::Procedure::size):
(JSC::B3::Procedure::cfg):
(JSC::B3::Procedure::setLastPhaseName):
* b3/B3ReduceStrength.cpp:
* b3/B3ValueInlines.h:
(JSC::B3::Value::isConstant):
(JSC::B3::Value::isInteger):
* b3/B3ValueKey.h:
(JSC::B3::ValueKey::canMaterialize):
(JSC::B3::ValueKey::isConstant):
* b3/air/AirCode.cpp:
(JSC::B3::Air::Code::findNextBlock):
(JSC::B3::Air::Code::addFastTmp):
* b3/air/AirCode.h:
(JSC::B3::Air::Code::specials):
(JSC::B3::Air::Code::isFastTmp):
(JSC::B3::Air::Code::setLastPhaseName):
* b3/air/AirIteratedRegisterCoalescing.cpp:
* dfg/DFGDominators.h:
* dfg/DFGSSACalculator.cpp:
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
Source/WTF:
Remove some remaining DFG-specific snippets from Dominators. This used to be a non-template
DFG class, and some time ago I hoisted it into WTF and made it generic. But since the only
user of the class was the DFG, this class still had a handful of DFG-specific snippets that
didn't compile when I started using it from B3.
Also renamed immediateDominatorOf() to idom(). This is the sort of abbreviation that we
wouldn't ordinarily want to have in WebKit. But WebKit does allow for abbreviations that are
"more canonical". The term "idom" is definitely more canonical than "immediateDominatorOf".
* wtf/Dominators.h:
(WTF::Dominators::dominates):
(WTF::Dominators::idom):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::ValidationContext::handleErrors):
Canonical link: https://commits.webkit.org/170119@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@193682 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-12-08 02:46:22 +00:00
|
|
|
#include <wtf/FastBitVector.h>
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
#include <wtf/GraphNodeWorklist.h>
|
2020-05-08 15:39:39 +00:00
|
|
|
#include <wtf/Vector.h>
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
|
|
|
|
|
namespace WTF {
|
|
|
|
|
|
|
|
|
|
// This is a utility for finding the dominators of a graph. Dominators are almost universally used
|
|
|
|
|
// for control flow graph analysis, so this code will refer to the graph's "nodes" as "blocks". In
|
|
|
|
|
// that regard this code is kind of specialized for the various JSC compilers, but you could use it
|
|
|
|
|
// for non-compiler things if you are OK with referring to your "nodes" as "blocks".
|
|
|
|
|
|
|
|
|
|
template<typename Graph>
|
|
|
|
|
class Dominators {
|
2019-08-12 20:57:15 +00:00
|
|
|
WTF_MAKE_FAST_ALLOCATED;
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
public:
|
Support compiling catch in the DFG
https://bugs.webkit.org/show_bug.cgi?id=174590
<rdar://problem/34047845>
Reviewed by Filip Pizlo.
JSTests:
* microbenchmarks/delta-blue-try-catch.js: Added.
(exception):
(value):
(OrderedCollection):
(OrderedCollection.prototype.add):
(OrderedCollection.prototype.at):
(OrderedCollection.prototype.size):
(OrderedCollection.prototype.removeFirst):
(OrderedCollection.prototype.remove):
(Strength):
(Strength.stronger):
(Strength.weaker):
(Strength.weakestOf):
(Strength.strongest):
(Strength.prototype.nextWeaker):
(Constraint):
(Constraint.prototype.addConstraint):
(Constraint.prototype.satisfy):
(Constraint.prototype.destroyConstraint):
(Constraint.prototype.isInput):
(UnaryConstraint):
(UnaryConstraint.prototype.addToGraph):
(UnaryConstraint.prototype.chooseMethod):
(UnaryConstraint.prototype.isSatisfied):
(UnaryConstraint.prototype.markInputs):
(UnaryConstraint.prototype.output):
(UnaryConstraint.prototype.recalculate):
(UnaryConstraint.prototype.markUnsatisfied):
(UnaryConstraint.prototype.inputsKnown):
(UnaryConstraint.prototype.removeFromGraph):
(StayConstraint):
(StayConstraint.prototype.execute):
(EditConstraint.prototype.isInput):
(EditConstraint.prototype.execute):
(BinaryConstraint):
(BinaryConstraint.prototype.chooseMethod):
(BinaryConstraint.prototype.addToGraph):
(BinaryConstraint.prototype.isSatisfied):
(BinaryConstraint.prototype.markInputs):
(BinaryConstraint.prototype.input):
(BinaryConstraint.prototype.output):
(BinaryConstraint.prototype.recalculate):
(BinaryConstraint.prototype.markUnsatisfied):
(BinaryConstraint.prototype.inputsKnown):
(BinaryConstraint.prototype.removeFromGraph):
(ScaleConstraint):
(ScaleConstraint.prototype.addToGraph):
(ScaleConstraint.prototype.removeFromGraph):
(ScaleConstraint.prototype.markInputs):
(ScaleConstraint.prototype.execute):
(ScaleConstraint.prototype.recalculate):
(EqualityConstraint):
(EqualityConstraint.prototype.execute):
(Variable):
(Variable.prototype.addConstraint):
(Variable.prototype.removeConstraint):
(Planner):
(Planner.prototype.incrementalAdd):
(Planner.prototype.incrementalRemove):
(Planner.prototype.newMark):
(Planner.prototype.makePlan):
(Planner.prototype.extractPlanFromConstraints):
(Planner.prototype.addPropagate):
(Planner.prototype.removePropagateFrom):
(Planner.prototype.addConstraintsConsumingTo):
(Plan):
(Plan.prototype.addConstraint):
(Plan.prototype.size):
(Plan.prototype.constraintAt):
(Plan.prototype.execute):
(chainTest):
(projectionTest):
(change):
(deltaBlue):
* microbenchmarks/fake-iterators-that-throw-when-finished.js: Added.
(assert):
(Numbers):
(Numbers.prototype.next):
(return.Transpose):
(return.Transpose.prototype.next):
(transpose):
(verifyEven):
(verifyString):
(foo):
(runIterators):
* microbenchmarks/try-catch-word-count.js: Added.
(let.assert):
(EOF):
(let.texts):
(let.o.apply):
(foo):
(bar):
(f):
(run):
(test1):
(test2):
(test3):
(fn):
(A):
(B):
(A.prototype.getValue):
(B.prototype.getParentValue):
(strlen):
(sum.0):
(test):
(result.test.o):
(set add.set add):
(set forEach):
(stringHash):
(set if):
(testFunction):
(set delete.set has.set add):
* stress/catch-set-argument-speculation-failure.js: Added.
(o):
(e):
(e2):
(escape):
(baz):
(noInline.run):
(noInline):
* stress/osr-enter-to-catch-with-set-local-type-check-failure.js: Added.
(foo):
(e):
(baz):
(bar):
Source/JavaScriptCore:
This patch implements OSR entry into op_catch in the DFG. We will support OSR entry
into the FTL in a followup: https://bugs.webkit.org/show_bug.cgi?id=175396
To implement catch in the DFG, this patch introduces the concept of multiple
entrypoints into CPS/LoadStore DFG IR. A lot of this patch is stringing this concept
through the DFG. Many phases used to assume that Graph::block(0) is the only root, and this
patch contains many straight forward changes generalizing the code to handle more than
one entrypoint.
A main building block of this is moving to two CFG types: SSACFG and CPSCFG. SSACFG
is the same CFG we used to have. CPSCFG is a new type that introduces a fake root
that has an outgoing edge to all the entrypoints. This allows our existing graph algorithms
to Just Work over CPSCFG. For example, there is now the concept of SSADominators vs CPSDominators,
and SSANaturalLoops vs CPSNaturalLoops.
The way we compile the catch entrypoint is by bootstrapping the state
of the program by loading all live bytecode locals from a buffer. The OSR
entry code will store all live values into that buffer before jumping to
the entrypoint. The OSR entry code is also responsible for performing type
proofs of the arguments before doing an OSR entry. If there is a type
mismatch, it's not legal to OSR enter into the DFG compilation. Currently,
each catch entrypoint knows the argument type proofs it must perform to enter
into the DFG. Currently, all entrypoints' arguments flush format are unified
via ArgumentPosition, but this is just an implementation detail. The code is
written more generally to assume that each entrypoint may perform its own distinct
proof.
op_catch now performs value profiling for all live bytecode locals in the
LLInt and baseline JIT. This information is then fed into the DFG via the
ExtractCatchLocal node in the prediction propagation phase.
This patch also changes how we generate op_catch in bytecode. All op_catches
are now split out at the end of the program in bytecode. This ensures that
no op_catch is inside a try block. This is needed to ensure correctness in
the DFGLiveCatchVariablePreservationPhase. That phase only inserts flushes
before SetLocals inside a try block. If an op_catch were in a try block, this
would cause the phase to insert a Flush before one of the state bootstrapping
SetLocals, which would generate invalid IR. Moving op_catch to be generated on
its own at the end of a bytecode stream seemed like the most elegant solution since
it better represents that we treat op_catch as an entrypoint. This is true
both in the DFG and in the baseline and LLInt: we don't reach an op_catch
via normal control flow. Because op_catch cannot throw, this will not break
any previous semantics of op_catch. Logically, it'd be valid to split try
blocks around any non-throwing bytecode operation.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* bytecode/BytecodeDumper.cpp:
(JSC::BytecodeDumper<Block>::dumpBytecode):
* bytecode/BytecodeList.json:
* bytecode/BytecodeUseDef.h:
(JSC::computeUsesForBytecodeOffset):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::finishCreation):
(JSC::CodeBlock::ensureCatchLivenessIsComputedForBytecodeOffset):
(JSC::CodeBlock::updateAllPredictionsAndCountLiveness):
(JSC::CodeBlock::validate):
* bytecode/CodeBlock.h:
* bytecode/ValueProfile.h:
(JSC::ValueProfile::ValueProfile):
(JSC::ValueProfileAndOperandBuffer::ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::~ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::forEach):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::generate):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::emitCatch):
(JSC::BytecodeGenerator::emitEnumeration):
* bytecompiler/BytecodeGenerator.h:
* bytecompiler/NodesCodegen.cpp:
(JSC::TryNode::emitBytecode):
* dfg/DFGAbstractInterpreterInlines.h:
(JSC::DFG::AbstractInterpreter<AbstractStateType>::executeEffects):
* dfg/DFGBackwardsCFG.h:
(JSC::DFG::BackwardsCFG::BackwardsCFG):
* dfg/DFGBasicBlock.cpp:
(JSC::DFG::BasicBlock::BasicBlock):
* dfg/DFGBasicBlock.h:
(JSC::DFG::BasicBlock::findTerminal const):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::setDirect):
(JSC::DFG::ByteCodeParser::flush):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::DelayedSetLocal):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::execute):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::parseCodeBlock):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCFG.h:
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::roots):
(JSC::DFG::CPSCFG::CPSCFG):
(JSC::DFG::selectCFG):
* dfg/DFGCPSRethreadingPhase.cpp:
(JSC::DFG::CPSRethreadingPhase::specialCaseArguments):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGClobberize.h:
(JSC::DFG::clobberize):
* dfg/DFGControlEquivalenceAnalysis.h:
(JSC::DFG::ControlEquivalenceAnalysis::ControlEquivalenceAnalysis):
* dfg/DFGDCEPhase.cpp:
(JSC::DFG::DCEPhase::run):
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDoesGC.cpp:
(JSC::DFG::doesGC):
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::ensureDominatorsForCFG):
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::EdgeDominates):
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
(JSC::DFG::FixupPhase::fixupChecksInBlock):
* dfg/DFGFlushFormat.h:
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::unboxLoopNode):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::dump):
(JSC::DFG::Graph::determineReachability):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::blocksInPreOrder):
(JSC::DFG::Graph::blocksInPostOrder):
(JSC::DFG::Graph::ensureCPSDominators):
(JSC::DFG::Graph::ensureSSADominators):
(JSC::DFG::Graph::ensureCPSNaturalLoops):
(JSC::DFG::Graph::ensureSSANaturalLoops):
(JSC::DFG::Graph::ensureBackwardsCFG):
(JSC::DFG::Graph::ensureBackwardsDominators):
(JSC::DFG::Graph::ensureControlEquivalenceAnalysis):
(JSC::DFG::Graph::methodOfGettingAValueProfileFor):
(JSC::DFG::Graph::clearCPSCFGData):
(JSC::DFG::Graph::ensureDominators): Deleted.
(JSC::DFG::Graph::ensurePrePostNumbering): Deleted.
(JSC::DFG::Graph::ensureNaturalLoops): Deleted.
* dfg/DFGGraph.h:
(JSC::DFG::Graph::willCatchExceptionInMachineFrame):
(JSC::DFG::Graph::isEntrypoint const):
* dfg/DFGInPlaceAbstractState.cpp:
(JSC::DFG::InPlaceAbstractState::initialize):
(JSC::DFG::InPlaceAbstractState::mergeToSuccessors):
* dfg/DFGJITCode.cpp:
(JSC::DFG::JITCode::shrinkToFit):
* dfg/DFGJITCode.h:
(JSC::DFG::JITCode::catchOSREntryDataForBytecodeIndex):
(JSC::DFG::JITCode::finalizeCatchOSREntrypoints):
(JSC::DFG::JITCode::appendCatchEntrypoint):
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::compile):
(JSC::DFG::JITCompiler::compileFunction):
(JSC::DFG::JITCompiler::noticeCatchEntrypoint):
(JSC::DFG::JITCompiler::noticeOSREntry):
(JSC::DFG::JITCompiler::makeCatchOSREntryBuffer):
* dfg/DFGJITCompiler.h:
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::run):
(JSC::DFG::LiveCatchVariablePreservationPhase::isValidFlushLocation):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlockForTryCatch):
(JSC::DFG::LiveCatchVariablePreservationPhase::newVariableAccessData):
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException): Deleted.
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock): Deleted.
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGMaximalFlushInsertionPhase.cpp:
(JSC::DFG::MaximalFlushInsertionPhase::run):
(JSC::DFG::MaximalFlushInsertionPhase::treatRegularBlock):
(JSC::DFG::MaximalFlushInsertionPhase::treatRootBlock):
* dfg/DFGMayExit.cpp:
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::isSwitch const):
(JSC::DFG::Node::successor):
(JSC::DFG::Node::catchOSREntryIndex const):
(JSC::DFG::Node::catchLocalPrediction):
(JSC::DFG::Node::isSwitch): Deleted.
* dfg/DFGNodeType.h:
* dfg/DFGOSREntry.cpp:
(JSC::DFG::prepareCatchOSREntry):
* dfg/DFGOSREntry.h:
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGOSRExitCompilerCommon.cpp:
(JSC::DFG::handleExitCounts):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering): Deleted.
(JSC::DFG::PrePostNumbering::~PrePostNumbering): Deleted.
(WTF::printInternal): Deleted.
* dfg/DFGPrePostNumbering.h:
(): Deleted.
(JSC::DFG::PrePostNumbering::preNumber const): Deleted.
(JSC::DFG::PrePostNumbering::postNumber const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::isDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::edgeKind const): Deleted.
* dfg/DFGPredictionInjectionPhase.cpp:
(JSC::DFG::PredictionInjectionPhase::run):
* dfg/DFGPredictionPropagationPhase.cpp:
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
(JSC::DFG::performSSAConversion):
* dfg/DFGSafeToExecute.h:
(JSC::DFG::safeToExecute):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compileCurrentBlock):
(JSC::DFG::SpeculativeJIT::checkArgumentTypes):
(JSC::DFG::SpeculativeJIT::createOSREntries):
(JSC::DFG::SpeculativeJIT::linkOSREntries):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGStaticExecutionCountEstimationPhase.cpp:
(JSC::DFG::StaticExecutionCountEstimationPhase::run):
* dfg/DFGStrengthReductionPhase.cpp:
(JSC::DFG::StrengthReductionPhase::handleNode):
* dfg/DFGTierUpCheckInjectionPhase.cpp:
(JSC::DFG::TierUpCheckInjectionPhase::run):
(JSC::DFG::TierUpCheckInjectionPhase::buildNaturalLoopToLoopHintMap):
* dfg/DFGTypeCheckHoistingPhase.cpp:
(JSC::DFG::TypeCheckHoistingPhase::run):
* dfg/DFGValidate.cpp:
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToB3.cpp:
(JSC::FTL::DFG::LowerDFGToB3::lower):
(JSC::FTL::DFG::LowerDFGToB3::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToB3::isValid):
* jit/JIT.h:
* jit/JITInlines.h:
(JSC::JIT::callOperation):
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOperations.cpp:
* jit/JITOperations.h:
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* llint/LLIntSlowPaths.h:
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
Source/WTF:
This patch generalizes the BackwardsGraph fake root into a more generalizable
class called SingleRootGraph. SingleRootGraph exposes the general graph interface
used in Dominators and NaturalLoops. SingleRootGraph takes as input a graph with
the normal graph interface, but also allows the input graph to contain more than
one root. SingleRootGraph then exposes a single root, which it creates, that has
an outgoing edge to all the roots in the original graph.
* WTF.xcodeproj/project.pbxproj:
* wtf/BackwardsGraph.h:
(WTF::BackwardsGraph::dump const):
(WTF::BackwardsGraph::rootName): Deleted.
(WTF::BackwardsGraph::Node::Node): Deleted.
(WTF::BackwardsGraph::Node::root): Deleted.
(WTF::BackwardsGraph::Node::operator== const): Deleted.
(WTF::BackwardsGraph::Node::operator!= const): Deleted.
(WTF::BackwardsGraph::Node::operator bool const): Deleted.
(WTF::BackwardsGraph::Node::isRoot const): Deleted.
(WTF::BackwardsGraph::Node::node const): Deleted.
(): Deleted.
(WTF::BackwardsGraph::Set::Set): Deleted.
(WTF::BackwardsGraph::Set::add): Deleted.
(WTF::BackwardsGraph::Set::remove): Deleted.
(WTF::BackwardsGraph::Set::contains): Deleted.
(WTF::BackwardsGraph::Set::dump const): Deleted.
(WTF::BackwardsGraph::Map::Map): Deleted.
(WTF::BackwardsGraph::Map::clear): Deleted.
(WTF::BackwardsGraph::Map::size const): Deleted.
(WTF::BackwardsGraph::Map::operator[]): Deleted.
(WTF::BackwardsGraph::Map::operator[] const): Deleted.
* wtf/Dominators.h:
(WTF::Dominators::Dominators):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf const):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl const):
* wtf/SingleRootGraph.h: Added.
(WTF::SingleRootGraphNode::rootName):
(WTF::SingleRootGraphNode::SingleRootGraphNode):
(WTF::SingleRootGraphNode::root):
(WTF::SingleRootGraphNode::operator== const):
(WTF::SingleRootGraphNode::operator!= const):
(WTF::SingleRootGraphNode::operator bool const):
(WTF::SingleRootGraphNode::isRoot const):
(WTF::SingleRootGraphNode::node const):
(WTF::SingleRootGraphSet::add):
(WTF::SingleRootGraphSet::remove):
(WTF::SingleRootGraphSet::contains):
(WTF::SingleRootGraphSet::dump const):
(WTF::SingleRootMap::SingleRootMap):
(WTF::SingleRootMap::clear):
(WTF::SingleRootMap::size const):
(WTF::SingleRootMap::operator[]):
(WTF::SingleRootMap::operator[] const):
(WTF::SingleRootGraph::SingleRootGraph):
(WTF::SingleRootGraph::root const):
(WTF::SingleRootGraph::newMap):
(WTF::SingleRootGraph::successors const):
(WTF::SingleRootGraph::predecessors const):
(WTF::SingleRootGraph::index const):
(WTF::SingleRootGraph::node const):
(WTF::SingleRootGraph::numNodes const):
(WTF::SingleRootGraph::dump const):
(WTF::SingleRootGraph::assertIsConsistent const):
Canonical link: https://commits.webkit.org/192644@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@221196 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2017-08-25 18:26:15 +00:00
|
|
|
using List = typename Graph::List;
|
|
|
|
|
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
Dominators(Graph& graph, bool selfCheck = false)
|
|
|
|
|
: m_graph(graph)
|
2016-05-19 21:25:29 +00:00
|
|
|
, m_data(graph.template newMap<BlockData>())
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
{
|
|
|
|
|
LengauerTarjan lengauerTarjan(m_graph);
|
|
|
|
|
lengauerTarjan.compute();
|
|
|
|
|
|
|
|
|
|
// From here we want to build a spanning tree with both upward and downward links and we want
|
|
|
|
|
// to do a search over this tree to compute pre and post numbers that can be used for dominance
|
|
|
|
|
// tests.
|
|
|
|
|
|
|
|
|
|
for (unsigned blockIndex = m_graph.numNodes(); blockIndex--;) {
|
|
|
|
|
typename Graph::Node block = m_graph.node(blockIndex);
|
|
|
|
|
if (!block)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
typename Graph::Node idomBlock = lengauerTarjan.immediateDominator(block);
|
|
|
|
|
m_data[block].idomParent = idomBlock;
|
|
|
|
|
if (idomBlock)
|
|
|
|
|
m_data[idomBlock].idomKids.append(block);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
unsigned nextPreNumber = 0;
|
|
|
|
|
unsigned nextPostNumber = 0;
|
|
|
|
|
|
|
|
|
|
// Plain stack-based worklist because we are guaranteed to see each block exactly once anyway.
|
|
|
|
|
Vector<GraphNodeWithOrder<typename Graph::Node>> worklist;
|
2016-05-19 21:25:29 +00:00
|
|
|
worklist.append(GraphNodeWithOrder<typename Graph::Node>(m_graph.root(), GraphVisitOrder::Pre));
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
while (!worklist.isEmpty()) {
|
|
|
|
|
GraphNodeWithOrder<typename Graph::Node> item = worklist.takeLast();
|
|
|
|
|
switch (item.order) {
|
|
|
|
|
case GraphVisitOrder::Pre:
|
|
|
|
|
m_data[item.node].preNumber = nextPreNumber++;
|
|
|
|
|
worklist.append(GraphNodeWithOrder<typename Graph::Node>(item.node, GraphVisitOrder::Post));
|
|
|
|
|
for (typename Graph::Node kid : m_data[item.node].idomKids)
|
|
|
|
|
worklist.append(GraphNodeWithOrder<typename Graph::Node>(kid, GraphVisitOrder::Pre));
|
|
|
|
|
break;
|
|
|
|
|
case GraphVisitOrder::Post:
|
|
|
|
|
m_data[item.node].postNumber = nextPostNumber++;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (selfCheck) {
|
|
|
|
|
// Check our dominator calculation:
|
|
|
|
|
// 1) Check that our range-based ancestry test is the same as a naive ancestry test.
|
|
|
|
|
// 2) Check that our notion of who dominates whom is identical to a naive (not
|
|
|
|
|
// Lengauer-Tarjan) dominator calculation.
|
|
|
|
|
|
|
|
|
|
ValidationContext context(m_graph, *this);
|
|
|
|
|
|
|
|
|
|
for (unsigned fromBlockIndex = m_graph.numNodes(); fromBlockIndex--;) {
|
|
|
|
|
typename Graph::Node fromBlock = m_graph.node(fromBlockIndex);
|
|
|
|
|
if (!fromBlock || m_data[fromBlock].preNumber == UINT_MAX)
|
|
|
|
|
continue;
|
|
|
|
|
for (unsigned toBlockIndex = m_graph.numNodes(); toBlockIndex--;) {
|
|
|
|
|
typename Graph::Node toBlock = m_graph.node(toBlockIndex);
|
|
|
|
|
if (!toBlock || m_data[toBlock].preNumber == UINT_MAX)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
if (dominates(fromBlock, toBlock) != naiveDominates(fromBlock, toBlock))
|
|
|
|
|
context.reportError(fromBlock, toBlock, "Range-based domination check is broken");
|
|
|
|
|
if (dominates(fromBlock, toBlock) != context.naiveDominators.dominates(fromBlock, toBlock))
|
|
|
|
|
context.reportError(fromBlock, toBlock, "Lengauer-Tarjan domination is broken");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
context.handleErrors();
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
bool strictlyDominates(typename Graph::Node from, typename Graph::Node to) const
|
|
|
|
|
{
|
|
|
|
|
return m_data[to].preNumber > m_data[from].preNumber
|
|
|
|
|
&& m_data[to].postNumber < m_data[from].postNumber;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
bool dominates(typename Graph::Node from, typename Graph::Node to) const
|
|
|
|
|
{
|
|
|
|
|
return from == to || strictlyDominates(from, to);
|
|
|
|
|
}
|
FTL B3 should be able to flag the tag constants as being super important so that B3 can hoist them and Air can force them into registers
https://bugs.webkit.org/show_bug.cgi?id=151955
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Taught B3 about the concept of "fast constants". A client of B3 can now tell B3 which
constants are super important. B3 will not spill the constant in that case and will ensure
that the constant is materialized only once: statically once, and dynamically once per
procedure execution. The hoistFastConstants() algorithm in B3MoveConstants.cpp achieves this
by first picking the lowest common dominator of all uses of each fast constant, and then
picking the materialization point by finding the lowest dominator of that dominator that is
tied for lowest block frequency. In practice, the second step ensures that this is the lowest
point in the program that is not in a loop (i.e. executes no more than once dynamically per
procedure invocation).
Taught Air about the concept of "fast tmps". B3 tells Air that a tmp is fast if it is used to
hold the materialization of a fast constant. IRC will use the lowest possible spill score for
fast tmps. In practice, this ensures that fast constants are never spilled.
Added a small snippet of code to FTL::LowerDFGToLLVM that makes both of the tag constants
into fast constants.
My hope is that this very brute-force heuristic is good enough that we don't have to think
about constants for a while. Based on my experience with how LLVM's constant hoisting works
out, the heuristic in this patch is going to be tough to beat. LLVM's constant hoisting does
good things when it hoists the tags, and usually causes nothing but problems when it hoists
anything else. This is because there is no way a low-level compiler to really understand how
a constant materialization impacts some operation's contribution to the overall execution
time of a procedure. But, in the FTL we know that constant materializations for type checks
are a bummer because we are super comfortable placing type checks on the hottest of paths. So
those are the last paths where extra instructions should be added by the compiler. On the
other hand, all other large constant uses are on relatively cold paths, or paths that are
already expensive for other reasons. For example, global variable accesses have to
materialize a pointer to the global. But that's not really a big deal, since a load from a
global involves first the load itself and then type checks on the result - so probably the
constant materialization is just not interesting. A store to a global often involves a store
barrier, so the constant materialization is really not interesting. This patch codifies this
heuristic in a pact between Air, B3, and the FTL: FTL demands that B3 pin the two tags in
registers, and B3 relays the demand to Air.
* JavaScriptCore.xcodeproj/project.pbxproj:
* b3/B3CFG.h: Added.
(JSC::B3::CFG::CFG):
(JSC::B3::CFG::root):
(JSC::B3::CFG::newMap):
(JSC::B3::CFG::successors):
(JSC::B3::CFG::predecessors):
(JSC::B3::CFG::index):
(JSC::B3::CFG::node):
(JSC::B3::CFG::numNodes):
(JSC::B3::CFG::dump):
* b3/B3Dominators.h: Added.
(JSC::B3::Dominators::Dominators):
* b3/B3IndexMap.h:
(JSC::B3::IndexMap::resize):
(JSC::B3::IndexMap::size):
(JSC::B3::IndexMap::operator[]):
* b3/B3LowerMacros.cpp:
* b3/B3LowerToAir.cpp:
(JSC::B3::Air::LowerToAir::tmp):
* b3/B3MoveConstants.cpp:
* b3/B3Opcode.h:
(JSC::B3::constPtrOpcode):
(JSC::B3::isConstant):
* b3/B3Procedure.cpp:
(JSC::B3::Procedure::Procedure):
(JSC::B3::Procedure::resetReachability):
(JSC::B3::Procedure::invalidateCFG):
(JSC::B3::Procedure::dump):
(JSC::B3::Procedure::deleteValue):
(JSC::B3::Procedure::dominators):
(JSC::B3::Procedure::addFastConstant):
(JSC::B3::Procedure::isFastConstant):
(JSC::B3::Procedure::addDataSection):
* b3/B3Procedure.h:
(JSC::B3::Procedure::size):
(JSC::B3::Procedure::cfg):
(JSC::B3::Procedure::setLastPhaseName):
* b3/B3ReduceStrength.cpp:
* b3/B3ValueInlines.h:
(JSC::B3::Value::isConstant):
(JSC::B3::Value::isInteger):
* b3/B3ValueKey.h:
(JSC::B3::ValueKey::canMaterialize):
(JSC::B3::ValueKey::isConstant):
* b3/air/AirCode.cpp:
(JSC::B3::Air::Code::findNextBlock):
(JSC::B3::Air::Code::addFastTmp):
* b3/air/AirCode.h:
(JSC::B3::Air::Code::specials):
(JSC::B3::Air::Code::isFastTmp):
(JSC::B3::Air::Code::setLastPhaseName):
* b3/air/AirIteratedRegisterCoalescing.cpp:
* dfg/DFGDominators.h:
* dfg/DFGSSACalculator.cpp:
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
Source/WTF:
Remove some remaining DFG-specific snippets from Dominators. This used to be a non-template
DFG class, and some time ago I hoisted it into WTF and made it generic. But since the only
user of the class was the DFG, this class still had a handful of DFG-specific snippets that
didn't compile when I started using it from B3.
Also renamed immediateDominatorOf() to idom(). This is the sort of abbreviation that we
wouldn't ordinarily want to have in WebKit. But WebKit does allow for abbreviations that are
"more canonical". The term "idom" is definitely more canonical than "immediateDominatorOf".
* wtf/Dominators.h:
(WTF::Dominators::dominates):
(WTF::Dominators::idom):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::ValidationContext::handleErrors):
Canonical link: https://commits.webkit.org/170119@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@193682 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-12-08 02:46:22 +00:00
|
|
|
|
|
|
|
|
// Returns the immediate dominator of this block. Returns null for the root block.
|
|
|
|
|
typename Graph::Node idom(typename Graph::Node block) const
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
{
|
|
|
|
|
return m_data[block].idomParent;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template<typename Functor>
|
|
|
|
|
void forAllStrictDominatorsOf(typename Graph::Node to, const Functor& functor) const
|
|
|
|
|
{
|
|
|
|
|
for (typename Graph::Node block = m_data[to].idomParent; block; block = m_data[block].idomParent)
|
|
|
|
|
functor(block);
|
|
|
|
|
}
|
|
|
|
|
|
B3::moveConstants should be able to edit code to minimize the number of constants
https://bugs.webkit.org/show_bug.cgi?id=162764
Reviewed by Saam Barati.
Source/JavaScriptCore:
There are some interesting cases where we can reduce the number of constant materializations if
we teach moveConstants() how to edit code. The two examples that this patch supports are:
- Loads and stores from a constant pointer. Since loads and stores get an offset for free
and the instruction selector is really good at handling it, and since we can query Air to
see what kinds of offsets are legal, we can sometimes avoid using a constant pointer that
is specific to the absolute address of that load and instead pick some other constant
that is within offset distance of ours.
- Add and Sub by a constant (x + c, x - c). Since x + c = x - -c and x - c = x + -c, we can
flip Add to Sub or vice versa if the negated constant is available.
This change makes moveConstants() pick the most dominant constant that works for an value. In
the case of memory accesses, it uses Air::Arg::isValidAddrForm() to work out what other
constants would work. In the case of Add/Sub, it simply looks for the negated constant. This
should result in something like a minimal number of constants since these rules always pick the
most dominant constant that works - so if an Add's constant is already most dominant then
nothing changes, but if the negated one is more dominant then it becomes a Sub.
This is a 0.5% speed-up on LongSpider and neutral elsewhere. It's a speed-up because the
absolute address thing reduces the number of address materializations that we have to do, while
the add/sub thing prevents us from having to materialize 0x1000000000000 to box doubles.
However, this may introduce a pathology, which I've filed a bug for: bug 162796.
* b3/B3MoveConstants.cpp:
* b3/B3MoveConstants.h:
* b3/B3UseCounts.h:
* b3/air/AirFixObviousSpills.cpp:
* b3/testb3.cpp:
(JSC::B3::testMoveConstants):
(JSC::B3::run):
Source/WTF:
I thought it would be a good idea to document the fact that dominator traversal happens in a
particular order for a reason.
* wtf/Dominators.h:
Canonical link: https://commits.webkit.org/180753@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@206682 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2016-09-30 22:29:24 +00:00
|
|
|
// Note: This will visit the dominators starting with the 'to' node and moving up the idom tree
|
|
|
|
|
// until it gets to the root. Some clients of this function, like B3::moveConstants(), rely on this
|
|
|
|
|
// order.
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
template<typename Functor>
|
|
|
|
|
void forAllDominatorsOf(typename Graph::Node to, const Functor& functor) const
|
|
|
|
|
{
|
|
|
|
|
for (typename Graph::Node block = to; block; block = m_data[block].idomParent)
|
|
|
|
|
functor(block);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template<typename Functor>
|
|
|
|
|
void forAllBlocksStrictlyDominatedBy(typename Graph::Node from, const Functor& functor) const
|
|
|
|
|
{
|
|
|
|
|
Vector<typename Graph::Node, 16> worklist;
|
|
|
|
|
worklist.appendVector(m_data[from].idomKids);
|
|
|
|
|
while (!worklist.isEmpty()) {
|
|
|
|
|
typename Graph::Node block = worklist.takeLast();
|
|
|
|
|
functor(block);
|
|
|
|
|
worklist.appendVector(m_data[block].idomKids);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template<typename Functor>
|
|
|
|
|
void forAllBlocksDominatedBy(typename Graph::Node from, const Functor& functor) const
|
|
|
|
|
{
|
|
|
|
|
Vector<typename Graph::Node, 16> worklist;
|
|
|
|
|
worklist.append(from);
|
|
|
|
|
while (!worklist.isEmpty()) {
|
|
|
|
|
typename Graph::Node block = worklist.takeLast();
|
|
|
|
|
functor(block);
|
|
|
|
|
worklist.appendVector(m_data[block].idomKids);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typename Graph::Set strictDominatorsOf(typename Graph::Node to) const
|
|
|
|
|
{
|
|
|
|
|
typename Graph::Set result;
|
|
|
|
|
forAllStrictDominatorsOf(
|
|
|
|
|
to,
|
|
|
|
|
[&] (typename Graph::Node node) {
|
|
|
|
|
result.add(node);
|
|
|
|
|
});
|
|
|
|
|
return result;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typename Graph::Set dominatorsOf(typename Graph::Node to) const
|
|
|
|
|
{
|
|
|
|
|
typename Graph::Set result;
|
|
|
|
|
forAllDominatorsOf(
|
|
|
|
|
to,
|
|
|
|
|
[&] (typename Graph::Node node) {
|
|
|
|
|
result.add(node);
|
|
|
|
|
});
|
|
|
|
|
return result;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typename Graph::Set blocksStrictlyDominatedBy(typename Graph::Node from) const
|
|
|
|
|
{
|
|
|
|
|
typename Graph::Set result;
|
|
|
|
|
forAllBlocksStrictlyDominatedBy(
|
|
|
|
|
from,
|
|
|
|
|
[&] (typename Graph::Node node) {
|
|
|
|
|
result.add(node);
|
|
|
|
|
});
|
|
|
|
|
return result;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typename Graph::Set blocksDominatedBy(typename Graph::Node from) const
|
|
|
|
|
{
|
|
|
|
|
typename Graph::Set result;
|
|
|
|
|
forAllBlocksDominatedBy(
|
|
|
|
|
from,
|
|
|
|
|
[&] (typename Graph::Node node) {
|
|
|
|
|
result.add(node);
|
|
|
|
|
});
|
|
|
|
|
return result;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template<typename Functor>
|
|
|
|
|
void forAllBlocksInDominanceFrontierOf(
|
|
|
|
|
typename Graph::Node from, const Functor& functor) const
|
|
|
|
|
{
|
|
|
|
|
typename Graph::Set set;
|
|
|
|
|
forAllBlocksInDominanceFrontierOfImpl(
|
|
|
|
|
from,
|
|
|
|
|
[&] (typename Graph::Node block) {
|
|
|
|
|
if (set.add(block))
|
|
|
|
|
functor(block);
|
|
|
|
|
});
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typename Graph::Set dominanceFrontierOf(typename Graph::Node from) const
|
|
|
|
|
{
|
|
|
|
|
typename Graph::Set result;
|
|
|
|
|
forAllBlocksInDominanceFrontierOf(
|
|
|
|
|
from,
|
|
|
|
|
[&] (typename Graph::Node node) {
|
|
|
|
|
result.add(node);
|
|
|
|
|
});
|
|
|
|
|
return result;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template<typename Functor>
|
Support compiling catch in the DFG
https://bugs.webkit.org/show_bug.cgi?id=174590
<rdar://problem/34047845>
Reviewed by Filip Pizlo.
JSTests:
* microbenchmarks/delta-blue-try-catch.js: Added.
(exception):
(value):
(OrderedCollection):
(OrderedCollection.prototype.add):
(OrderedCollection.prototype.at):
(OrderedCollection.prototype.size):
(OrderedCollection.prototype.removeFirst):
(OrderedCollection.prototype.remove):
(Strength):
(Strength.stronger):
(Strength.weaker):
(Strength.weakestOf):
(Strength.strongest):
(Strength.prototype.nextWeaker):
(Constraint):
(Constraint.prototype.addConstraint):
(Constraint.prototype.satisfy):
(Constraint.prototype.destroyConstraint):
(Constraint.prototype.isInput):
(UnaryConstraint):
(UnaryConstraint.prototype.addToGraph):
(UnaryConstraint.prototype.chooseMethod):
(UnaryConstraint.prototype.isSatisfied):
(UnaryConstraint.prototype.markInputs):
(UnaryConstraint.prototype.output):
(UnaryConstraint.prototype.recalculate):
(UnaryConstraint.prototype.markUnsatisfied):
(UnaryConstraint.prototype.inputsKnown):
(UnaryConstraint.prototype.removeFromGraph):
(StayConstraint):
(StayConstraint.prototype.execute):
(EditConstraint.prototype.isInput):
(EditConstraint.prototype.execute):
(BinaryConstraint):
(BinaryConstraint.prototype.chooseMethod):
(BinaryConstraint.prototype.addToGraph):
(BinaryConstraint.prototype.isSatisfied):
(BinaryConstraint.prototype.markInputs):
(BinaryConstraint.prototype.input):
(BinaryConstraint.prototype.output):
(BinaryConstraint.prototype.recalculate):
(BinaryConstraint.prototype.markUnsatisfied):
(BinaryConstraint.prototype.inputsKnown):
(BinaryConstraint.prototype.removeFromGraph):
(ScaleConstraint):
(ScaleConstraint.prototype.addToGraph):
(ScaleConstraint.prototype.removeFromGraph):
(ScaleConstraint.prototype.markInputs):
(ScaleConstraint.prototype.execute):
(ScaleConstraint.prototype.recalculate):
(EqualityConstraint):
(EqualityConstraint.prototype.execute):
(Variable):
(Variable.prototype.addConstraint):
(Variable.prototype.removeConstraint):
(Planner):
(Planner.prototype.incrementalAdd):
(Planner.prototype.incrementalRemove):
(Planner.prototype.newMark):
(Planner.prototype.makePlan):
(Planner.prototype.extractPlanFromConstraints):
(Planner.prototype.addPropagate):
(Planner.prototype.removePropagateFrom):
(Planner.prototype.addConstraintsConsumingTo):
(Plan):
(Plan.prototype.addConstraint):
(Plan.prototype.size):
(Plan.prototype.constraintAt):
(Plan.prototype.execute):
(chainTest):
(projectionTest):
(change):
(deltaBlue):
* microbenchmarks/fake-iterators-that-throw-when-finished.js: Added.
(assert):
(Numbers):
(Numbers.prototype.next):
(return.Transpose):
(return.Transpose.prototype.next):
(transpose):
(verifyEven):
(verifyString):
(foo):
(runIterators):
* microbenchmarks/try-catch-word-count.js: Added.
(let.assert):
(EOF):
(let.texts):
(let.o.apply):
(foo):
(bar):
(f):
(run):
(test1):
(test2):
(test3):
(fn):
(A):
(B):
(A.prototype.getValue):
(B.prototype.getParentValue):
(strlen):
(sum.0):
(test):
(result.test.o):
(set add.set add):
(set forEach):
(stringHash):
(set if):
(testFunction):
(set delete.set has.set add):
* stress/catch-set-argument-speculation-failure.js: Added.
(o):
(e):
(e2):
(escape):
(baz):
(noInline.run):
(noInline):
* stress/osr-enter-to-catch-with-set-local-type-check-failure.js: Added.
(foo):
(e):
(baz):
(bar):
Source/JavaScriptCore:
This patch implements OSR entry into op_catch in the DFG. We will support OSR entry
into the FTL in a followup: https://bugs.webkit.org/show_bug.cgi?id=175396
To implement catch in the DFG, this patch introduces the concept of multiple
entrypoints into CPS/LoadStore DFG IR. A lot of this patch is stringing this concept
through the DFG. Many phases used to assume that Graph::block(0) is the only root, and this
patch contains many straight forward changes generalizing the code to handle more than
one entrypoint.
A main building block of this is moving to two CFG types: SSACFG and CPSCFG. SSACFG
is the same CFG we used to have. CPSCFG is a new type that introduces a fake root
that has an outgoing edge to all the entrypoints. This allows our existing graph algorithms
to Just Work over CPSCFG. For example, there is now the concept of SSADominators vs CPSDominators,
and SSANaturalLoops vs CPSNaturalLoops.
The way we compile the catch entrypoint is by bootstrapping the state
of the program by loading all live bytecode locals from a buffer. The OSR
entry code will store all live values into that buffer before jumping to
the entrypoint. The OSR entry code is also responsible for performing type
proofs of the arguments before doing an OSR entry. If there is a type
mismatch, it's not legal to OSR enter into the DFG compilation. Currently,
each catch entrypoint knows the argument type proofs it must perform to enter
into the DFG. Currently, all entrypoints' arguments flush format are unified
via ArgumentPosition, but this is just an implementation detail. The code is
written more generally to assume that each entrypoint may perform its own distinct
proof.
op_catch now performs value profiling for all live bytecode locals in the
LLInt and baseline JIT. This information is then fed into the DFG via the
ExtractCatchLocal node in the prediction propagation phase.
This patch also changes how we generate op_catch in bytecode. All op_catches
are now split out at the end of the program in bytecode. This ensures that
no op_catch is inside a try block. This is needed to ensure correctness in
the DFGLiveCatchVariablePreservationPhase. That phase only inserts flushes
before SetLocals inside a try block. If an op_catch were in a try block, this
would cause the phase to insert a Flush before one of the state bootstrapping
SetLocals, which would generate invalid IR. Moving op_catch to be generated on
its own at the end of a bytecode stream seemed like the most elegant solution since
it better represents that we treat op_catch as an entrypoint. This is true
both in the DFG and in the baseline and LLInt: we don't reach an op_catch
via normal control flow. Because op_catch cannot throw, this will not break
any previous semantics of op_catch. Logically, it'd be valid to split try
blocks around any non-throwing bytecode operation.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* bytecode/BytecodeDumper.cpp:
(JSC::BytecodeDumper<Block>::dumpBytecode):
* bytecode/BytecodeList.json:
* bytecode/BytecodeUseDef.h:
(JSC::computeUsesForBytecodeOffset):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::finishCreation):
(JSC::CodeBlock::ensureCatchLivenessIsComputedForBytecodeOffset):
(JSC::CodeBlock::updateAllPredictionsAndCountLiveness):
(JSC::CodeBlock::validate):
* bytecode/CodeBlock.h:
* bytecode/ValueProfile.h:
(JSC::ValueProfile::ValueProfile):
(JSC::ValueProfileAndOperandBuffer::ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::~ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::forEach):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::generate):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::emitCatch):
(JSC::BytecodeGenerator::emitEnumeration):
* bytecompiler/BytecodeGenerator.h:
* bytecompiler/NodesCodegen.cpp:
(JSC::TryNode::emitBytecode):
* dfg/DFGAbstractInterpreterInlines.h:
(JSC::DFG::AbstractInterpreter<AbstractStateType>::executeEffects):
* dfg/DFGBackwardsCFG.h:
(JSC::DFG::BackwardsCFG::BackwardsCFG):
* dfg/DFGBasicBlock.cpp:
(JSC::DFG::BasicBlock::BasicBlock):
* dfg/DFGBasicBlock.h:
(JSC::DFG::BasicBlock::findTerminal const):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::setDirect):
(JSC::DFG::ByteCodeParser::flush):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::DelayedSetLocal):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::execute):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::parseCodeBlock):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCFG.h:
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::roots):
(JSC::DFG::CPSCFG::CPSCFG):
(JSC::DFG::selectCFG):
* dfg/DFGCPSRethreadingPhase.cpp:
(JSC::DFG::CPSRethreadingPhase::specialCaseArguments):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGClobberize.h:
(JSC::DFG::clobberize):
* dfg/DFGControlEquivalenceAnalysis.h:
(JSC::DFG::ControlEquivalenceAnalysis::ControlEquivalenceAnalysis):
* dfg/DFGDCEPhase.cpp:
(JSC::DFG::DCEPhase::run):
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDoesGC.cpp:
(JSC::DFG::doesGC):
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::ensureDominatorsForCFG):
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::EdgeDominates):
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
(JSC::DFG::FixupPhase::fixupChecksInBlock):
* dfg/DFGFlushFormat.h:
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::unboxLoopNode):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::dump):
(JSC::DFG::Graph::determineReachability):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::blocksInPreOrder):
(JSC::DFG::Graph::blocksInPostOrder):
(JSC::DFG::Graph::ensureCPSDominators):
(JSC::DFG::Graph::ensureSSADominators):
(JSC::DFG::Graph::ensureCPSNaturalLoops):
(JSC::DFG::Graph::ensureSSANaturalLoops):
(JSC::DFG::Graph::ensureBackwardsCFG):
(JSC::DFG::Graph::ensureBackwardsDominators):
(JSC::DFG::Graph::ensureControlEquivalenceAnalysis):
(JSC::DFG::Graph::methodOfGettingAValueProfileFor):
(JSC::DFG::Graph::clearCPSCFGData):
(JSC::DFG::Graph::ensureDominators): Deleted.
(JSC::DFG::Graph::ensurePrePostNumbering): Deleted.
(JSC::DFG::Graph::ensureNaturalLoops): Deleted.
* dfg/DFGGraph.h:
(JSC::DFG::Graph::willCatchExceptionInMachineFrame):
(JSC::DFG::Graph::isEntrypoint const):
* dfg/DFGInPlaceAbstractState.cpp:
(JSC::DFG::InPlaceAbstractState::initialize):
(JSC::DFG::InPlaceAbstractState::mergeToSuccessors):
* dfg/DFGJITCode.cpp:
(JSC::DFG::JITCode::shrinkToFit):
* dfg/DFGJITCode.h:
(JSC::DFG::JITCode::catchOSREntryDataForBytecodeIndex):
(JSC::DFG::JITCode::finalizeCatchOSREntrypoints):
(JSC::DFG::JITCode::appendCatchEntrypoint):
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::compile):
(JSC::DFG::JITCompiler::compileFunction):
(JSC::DFG::JITCompiler::noticeCatchEntrypoint):
(JSC::DFG::JITCompiler::noticeOSREntry):
(JSC::DFG::JITCompiler::makeCatchOSREntryBuffer):
* dfg/DFGJITCompiler.h:
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::run):
(JSC::DFG::LiveCatchVariablePreservationPhase::isValidFlushLocation):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlockForTryCatch):
(JSC::DFG::LiveCatchVariablePreservationPhase::newVariableAccessData):
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException): Deleted.
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock): Deleted.
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGMaximalFlushInsertionPhase.cpp:
(JSC::DFG::MaximalFlushInsertionPhase::run):
(JSC::DFG::MaximalFlushInsertionPhase::treatRegularBlock):
(JSC::DFG::MaximalFlushInsertionPhase::treatRootBlock):
* dfg/DFGMayExit.cpp:
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::isSwitch const):
(JSC::DFG::Node::successor):
(JSC::DFG::Node::catchOSREntryIndex const):
(JSC::DFG::Node::catchLocalPrediction):
(JSC::DFG::Node::isSwitch): Deleted.
* dfg/DFGNodeType.h:
* dfg/DFGOSREntry.cpp:
(JSC::DFG::prepareCatchOSREntry):
* dfg/DFGOSREntry.h:
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGOSRExitCompilerCommon.cpp:
(JSC::DFG::handleExitCounts):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering): Deleted.
(JSC::DFG::PrePostNumbering::~PrePostNumbering): Deleted.
(WTF::printInternal): Deleted.
* dfg/DFGPrePostNumbering.h:
(): Deleted.
(JSC::DFG::PrePostNumbering::preNumber const): Deleted.
(JSC::DFG::PrePostNumbering::postNumber const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::isDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::edgeKind const): Deleted.
* dfg/DFGPredictionInjectionPhase.cpp:
(JSC::DFG::PredictionInjectionPhase::run):
* dfg/DFGPredictionPropagationPhase.cpp:
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
(JSC::DFG::performSSAConversion):
* dfg/DFGSafeToExecute.h:
(JSC::DFG::safeToExecute):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compileCurrentBlock):
(JSC::DFG::SpeculativeJIT::checkArgumentTypes):
(JSC::DFG::SpeculativeJIT::createOSREntries):
(JSC::DFG::SpeculativeJIT::linkOSREntries):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGStaticExecutionCountEstimationPhase.cpp:
(JSC::DFG::StaticExecutionCountEstimationPhase::run):
* dfg/DFGStrengthReductionPhase.cpp:
(JSC::DFG::StrengthReductionPhase::handleNode):
* dfg/DFGTierUpCheckInjectionPhase.cpp:
(JSC::DFG::TierUpCheckInjectionPhase::run):
(JSC::DFG::TierUpCheckInjectionPhase::buildNaturalLoopToLoopHintMap):
* dfg/DFGTypeCheckHoistingPhase.cpp:
(JSC::DFG::TypeCheckHoistingPhase::run):
* dfg/DFGValidate.cpp:
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToB3.cpp:
(JSC::FTL::DFG::LowerDFGToB3::lower):
(JSC::FTL::DFG::LowerDFGToB3::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToB3::isValid):
* jit/JIT.h:
* jit/JITInlines.h:
(JSC::JIT::callOperation):
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOperations.cpp:
* jit/JITOperations.h:
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* llint/LLIntSlowPaths.h:
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
Source/WTF:
This patch generalizes the BackwardsGraph fake root into a more generalizable
class called SingleRootGraph. SingleRootGraph exposes the general graph interface
used in Dominators and NaturalLoops. SingleRootGraph takes as input a graph with
the normal graph interface, but also allows the input graph to contain more than
one root. SingleRootGraph then exposes a single root, which it creates, that has
an outgoing edge to all the roots in the original graph.
* WTF.xcodeproj/project.pbxproj:
* wtf/BackwardsGraph.h:
(WTF::BackwardsGraph::dump const):
(WTF::BackwardsGraph::rootName): Deleted.
(WTF::BackwardsGraph::Node::Node): Deleted.
(WTF::BackwardsGraph::Node::root): Deleted.
(WTF::BackwardsGraph::Node::operator== const): Deleted.
(WTF::BackwardsGraph::Node::operator!= const): Deleted.
(WTF::BackwardsGraph::Node::operator bool const): Deleted.
(WTF::BackwardsGraph::Node::isRoot const): Deleted.
(WTF::BackwardsGraph::Node::node const): Deleted.
(): Deleted.
(WTF::BackwardsGraph::Set::Set): Deleted.
(WTF::BackwardsGraph::Set::add): Deleted.
(WTF::BackwardsGraph::Set::remove): Deleted.
(WTF::BackwardsGraph::Set::contains): Deleted.
(WTF::BackwardsGraph::Set::dump const): Deleted.
(WTF::BackwardsGraph::Map::Map): Deleted.
(WTF::BackwardsGraph::Map::clear): Deleted.
(WTF::BackwardsGraph::Map::size const): Deleted.
(WTF::BackwardsGraph::Map::operator[]): Deleted.
(WTF::BackwardsGraph::Map::operator[] const): Deleted.
* wtf/Dominators.h:
(WTF::Dominators::Dominators):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf const):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl const):
* wtf/SingleRootGraph.h: Added.
(WTF::SingleRootGraphNode::rootName):
(WTF::SingleRootGraphNode::SingleRootGraphNode):
(WTF::SingleRootGraphNode::root):
(WTF::SingleRootGraphNode::operator== const):
(WTF::SingleRootGraphNode::operator!= const):
(WTF::SingleRootGraphNode::operator bool const):
(WTF::SingleRootGraphNode::isRoot const):
(WTF::SingleRootGraphNode::node const):
(WTF::SingleRootGraphSet::add):
(WTF::SingleRootGraphSet::remove):
(WTF::SingleRootGraphSet::contains):
(WTF::SingleRootGraphSet::dump const):
(WTF::SingleRootMap::SingleRootMap):
(WTF::SingleRootMap::clear):
(WTF::SingleRootMap::size const):
(WTF::SingleRootMap::operator[]):
(WTF::SingleRootMap::operator[] const):
(WTF::SingleRootGraph::SingleRootGraph):
(WTF::SingleRootGraph::root const):
(WTF::SingleRootGraph::newMap):
(WTF::SingleRootGraph::successors const):
(WTF::SingleRootGraph::predecessors const):
(WTF::SingleRootGraph::index const):
(WTF::SingleRootGraph::node const):
(WTF::SingleRootGraph::numNodes const):
(WTF::SingleRootGraph::dump const):
(WTF::SingleRootGraph::assertIsConsistent const):
Canonical link: https://commits.webkit.org/192644@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@221196 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2017-08-25 18:26:15 +00:00
|
|
|
void forAllBlocksInIteratedDominanceFrontierOf(const List& from, const Functor& functor)
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
{
|
|
|
|
|
forAllBlocksInPrunedIteratedDominanceFrontierOf(
|
|
|
|
|
from,
|
|
|
|
|
[&] (typename Graph::Node block) -> bool {
|
|
|
|
|
functor(block);
|
|
|
|
|
return true;
|
|
|
|
|
});
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// This is a close relative of forAllBlocksInIteratedDominanceFrontierOf(), which allows the
|
|
|
|
|
// given functor to return false to indicate that we don't wish to consider the given block.
|
|
|
|
|
// Useful for computing pruned SSA form.
|
|
|
|
|
template<typename Functor>
|
|
|
|
|
void forAllBlocksInPrunedIteratedDominanceFrontierOf(
|
Support compiling catch in the DFG
https://bugs.webkit.org/show_bug.cgi?id=174590
<rdar://problem/34047845>
Reviewed by Filip Pizlo.
JSTests:
* microbenchmarks/delta-blue-try-catch.js: Added.
(exception):
(value):
(OrderedCollection):
(OrderedCollection.prototype.add):
(OrderedCollection.prototype.at):
(OrderedCollection.prototype.size):
(OrderedCollection.prototype.removeFirst):
(OrderedCollection.prototype.remove):
(Strength):
(Strength.stronger):
(Strength.weaker):
(Strength.weakestOf):
(Strength.strongest):
(Strength.prototype.nextWeaker):
(Constraint):
(Constraint.prototype.addConstraint):
(Constraint.prototype.satisfy):
(Constraint.prototype.destroyConstraint):
(Constraint.prototype.isInput):
(UnaryConstraint):
(UnaryConstraint.prototype.addToGraph):
(UnaryConstraint.prototype.chooseMethod):
(UnaryConstraint.prototype.isSatisfied):
(UnaryConstraint.prototype.markInputs):
(UnaryConstraint.prototype.output):
(UnaryConstraint.prototype.recalculate):
(UnaryConstraint.prototype.markUnsatisfied):
(UnaryConstraint.prototype.inputsKnown):
(UnaryConstraint.prototype.removeFromGraph):
(StayConstraint):
(StayConstraint.prototype.execute):
(EditConstraint.prototype.isInput):
(EditConstraint.prototype.execute):
(BinaryConstraint):
(BinaryConstraint.prototype.chooseMethod):
(BinaryConstraint.prototype.addToGraph):
(BinaryConstraint.prototype.isSatisfied):
(BinaryConstraint.prototype.markInputs):
(BinaryConstraint.prototype.input):
(BinaryConstraint.prototype.output):
(BinaryConstraint.prototype.recalculate):
(BinaryConstraint.prototype.markUnsatisfied):
(BinaryConstraint.prototype.inputsKnown):
(BinaryConstraint.prototype.removeFromGraph):
(ScaleConstraint):
(ScaleConstraint.prototype.addToGraph):
(ScaleConstraint.prototype.removeFromGraph):
(ScaleConstraint.prototype.markInputs):
(ScaleConstraint.prototype.execute):
(ScaleConstraint.prototype.recalculate):
(EqualityConstraint):
(EqualityConstraint.prototype.execute):
(Variable):
(Variable.prototype.addConstraint):
(Variable.prototype.removeConstraint):
(Planner):
(Planner.prototype.incrementalAdd):
(Planner.prototype.incrementalRemove):
(Planner.prototype.newMark):
(Planner.prototype.makePlan):
(Planner.prototype.extractPlanFromConstraints):
(Planner.prototype.addPropagate):
(Planner.prototype.removePropagateFrom):
(Planner.prototype.addConstraintsConsumingTo):
(Plan):
(Plan.prototype.addConstraint):
(Plan.prototype.size):
(Plan.prototype.constraintAt):
(Plan.prototype.execute):
(chainTest):
(projectionTest):
(change):
(deltaBlue):
* microbenchmarks/fake-iterators-that-throw-when-finished.js: Added.
(assert):
(Numbers):
(Numbers.prototype.next):
(return.Transpose):
(return.Transpose.prototype.next):
(transpose):
(verifyEven):
(verifyString):
(foo):
(runIterators):
* microbenchmarks/try-catch-word-count.js: Added.
(let.assert):
(EOF):
(let.texts):
(let.o.apply):
(foo):
(bar):
(f):
(run):
(test1):
(test2):
(test3):
(fn):
(A):
(B):
(A.prototype.getValue):
(B.prototype.getParentValue):
(strlen):
(sum.0):
(test):
(result.test.o):
(set add.set add):
(set forEach):
(stringHash):
(set if):
(testFunction):
(set delete.set has.set add):
* stress/catch-set-argument-speculation-failure.js: Added.
(o):
(e):
(e2):
(escape):
(baz):
(noInline.run):
(noInline):
* stress/osr-enter-to-catch-with-set-local-type-check-failure.js: Added.
(foo):
(e):
(baz):
(bar):
Source/JavaScriptCore:
This patch implements OSR entry into op_catch in the DFG. We will support OSR entry
into the FTL in a followup: https://bugs.webkit.org/show_bug.cgi?id=175396
To implement catch in the DFG, this patch introduces the concept of multiple
entrypoints into CPS/LoadStore DFG IR. A lot of this patch is stringing this concept
through the DFG. Many phases used to assume that Graph::block(0) is the only root, and this
patch contains many straight forward changes generalizing the code to handle more than
one entrypoint.
A main building block of this is moving to two CFG types: SSACFG and CPSCFG. SSACFG
is the same CFG we used to have. CPSCFG is a new type that introduces a fake root
that has an outgoing edge to all the entrypoints. This allows our existing graph algorithms
to Just Work over CPSCFG. For example, there is now the concept of SSADominators vs CPSDominators,
and SSANaturalLoops vs CPSNaturalLoops.
The way we compile the catch entrypoint is by bootstrapping the state
of the program by loading all live bytecode locals from a buffer. The OSR
entry code will store all live values into that buffer before jumping to
the entrypoint. The OSR entry code is also responsible for performing type
proofs of the arguments before doing an OSR entry. If there is a type
mismatch, it's not legal to OSR enter into the DFG compilation. Currently,
each catch entrypoint knows the argument type proofs it must perform to enter
into the DFG. Currently, all entrypoints' arguments flush format are unified
via ArgumentPosition, but this is just an implementation detail. The code is
written more generally to assume that each entrypoint may perform its own distinct
proof.
op_catch now performs value profiling for all live bytecode locals in the
LLInt and baseline JIT. This information is then fed into the DFG via the
ExtractCatchLocal node in the prediction propagation phase.
This patch also changes how we generate op_catch in bytecode. All op_catches
are now split out at the end of the program in bytecode. This ensures that
no op_catch is inside a try block. This is needed to ensure correctness in
the DFGLiveCatchVariablePreservationPhase. That phase only inserts flushes
before SetLocals inside a try block. If an op_catch were in a try block, this
would cause the phase to insert a Flush before one of the state bootstrapping
SetLocals, which would generate invalid IR. Moving op_catch to be generated on
its own at the end of a bytecode stream seemed like the most elegant solution since
it better represents that we treat op_catch as an entrypoint. This is true
both in the DFG and in the baseline and LLInt: we don't reach an op_catch
via normal control flow. Because op_catch cannot throw, this will not break
any previous semantics of op_catch. Logically, it'd be valid to split try
blocks around any non-throwing bytecode operation.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* bytecode/BytecodeDumper.cpp:
(JSC::BytecodeDumper<Block>::dumpBytecode):
* bytecode/BytecodeList.json:
* bytecode/BytecodeUseDef.h:
(JSC::computeUsesForBytecodeOffset):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::finishCreation):
(JSC::CodeBlock::ensureCatchLivenessIsComputedForBytecodeOffset):
(JSC::CodeBlock::updateAllPredictionsAndCountLiveness):
(JSC::CodeBlock::validate):
* bytecode/CodeBlock.h:
* bytecode/ValueProfile.h:
(JSC::ValueProfile::ValueProfile):
(JSC::ValueProfileAndOperandBuffer::ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::~ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::forEach):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::generate):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::emitCatch):
(JSC::BytecodeGenerator::emitEnumeration):
* bytecompiler/BytecodeGenerator.h:
* bytecompiler/NodesCodegen.cpp:
(JSC::TryNode::emitBytecode):
* dfg/DFGAbstractInterpreterInlines.h:
(JSC::DFG::AbstractInterpreter<AbstractStateType>::executeEffects):
* dfg/DFGBackwardsCFG.h:
(JSC::DFG::BackwardsCFG::BackwardsCFG):
* dfg/DFGBasicBlock.cpp:
(JSC::DFG::BasicBlock::BasicBlock):
* dfg/DFGBasicBlock.h:
(JSC::DFG::BasicBlock::findTerminal const):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::setDirect):
(JSC::DFG::ByteCodeParser::flush):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::DelayedSetLocal):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::execute):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::parseCodeBlock):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCFG.h:
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::roots):
(JSC::DFG::CPSCFG::CPSCFG):
(JSC::DFG::selectCFG):
* dfg/DFGCPSRethreadingPhase.cpp:
(JSC::DFG::CPSRethreadingPhase::specialCaseArguments):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGClobberize.h:
(JSC::DFG::clobberize):
* dfg/DFGControlEquivalenceAnalysis.h:
(JSC::DFG::ControlEquivalenceAnalysis::ControlEquivalenceAnalysis):
* dfg/DFGDCEPhase.cpp:
(JSC::DFG::DCEPhase::run):
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDoesGC.cpp:
(JSC::DFG::doesGC):
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::ensureDominatorsForCFG):
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::EdgeDominates):
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
(JSC::DFG::FixupPhase::fixupChecksInBlock):
* dfg/DFGFlushFormat.h:
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::unboxLoopNode):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::dump):
(JSC::DFG::Graph::determineReachability):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::blocksInPreOrder):
(JSC::DFG::Graph::blocksInPostOrder):
(JSC::DFG::Graph::ensureCPSDominators):
(JSC::DFG::Graph::ensureSSADominators):
(JSC::DFG::Graph::ensureCPSNaturalLoops):
(JSC::DFG::Graph::ensureSSANaturalLoops):
(JSC::DFG::Graph::ensureBackwardsCFG):
(JSC::DFG::Graph::ensureBackwardsDominators):
(JSC::DFG::Graph::ensureControlEquivalenceAnalysis):
(JSC::DFG::Graph::methodOfGettingAValueProfileFor):
(JSC::DFG::Graph::clearCPSCFGData):
(JSC::DFG::Graph::ensureDominators): Deleted.
(JSC::DFG::Graph::ensurePrePostNumbering): Deleted.
(JSC::DFG::Graph::ensureNaturalLoops): Deleted.
* dfg/DFGGraph.h:
(JSC::DFG::Graph::willCatchExceptionInMachineFrame):
(JSC::DFG::Graph::isEntrypoint const):
* dfg/DFGInPlaceAbstractState.cpp:
(JSC::DFG::InPlaceAbstractState::initialize):
(JSC::DFG::InPlaceAbstractState::mergeToSuccessors):
* dfg/DFGJITCode.cpp:
(JSC::DFG::JITCode::shrinkToFit):
* dfg/DFGJITCode.h:
(JSC::DFG::JITCode::catchOSREntryDataForBytecodeIndex):
(JSC::DFG::JITCode::finalizeCatchOSREntrypoints):
(JSC::DFG::JITCode::appendCatchEntrypoint):
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::compile):
(JSC::DFG::JITCompiler::compileFunction):
(JSC::DFG::JITCompiler::noticeCatchEntrypoint):
(JSC::DFG::JITCompiler::noticeOSREntry):
(JSC::DFG::JITCompiler::makeCatchOSREntryBuffer):
* dfg/DFGJITCompiler.h:
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::run):
(JSC::DFG::LiveCatchVariablePreservationPhase::isValidFlushLocation):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlockForTryCatch):
(JSC::DFG::LiveCatchVariablePreservationPhase::newVariableAccessData):
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException): Deleted.
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock): Deleted.
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGMaximalFlushInsertionPhase.cpp:
(JSC::DFG::MaximalFlushInsertionPhase::run):
(JSC::DFG::MaximalFlushInsertionPhase::treatRegularBlock):
(JSC::DFG::MaximalFlushInsertionPhase::treatRootBlock):
* dfg/DFGMayExit.cpp:
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::isSwitch const):
(JSC::DFG::Node::successor):
(JSC::DFG::Node::catchOSREntryIndex const):
(JSC::DFG::Node::catchLocalPrediction):
(JSC::DFG::Node::isSwitch): Deleted.
* dfg/DFGNodeType.h:
* dfg/DFGOSREntry.cpp:
(JSC::DFG::prepareCatchOSREntry):
* dfg/DFGOSREntry.h:
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGOSRExitCompilerCommon.cpp:
(JSC::DFG::handleExitCounts):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering): Deleted.
(JSC::DFG::PrePostNumbering::~PrePostNumbering): Deleted.
(WTF::printInternal): Deleted.
* dfg/DFGPrePostNumbering.h:
(): Deleted.
(JSC::DFG::PrePostNumbering::preNumber const): Deleted.
(JSC::DFG::PrePostNumbering::postNumber const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::isDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::edgeKind const): Deleted.
* dfg/DFGPredictionInjectionPhase.cpp:
(JSC::DFG::PredictionInjectionPhase::run):
* dfg/DFGPredictionPropagationPhase.cpp:
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
(JSC::DFG::performSSAConversion):
* dfg/DFGSafeToExecute.h:
(JSC::DFG::safeToExecute):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compileCurrentBlock):
(JSC::DFG::SpeculativeJIT::checkArgumentTypes):
(JSC::DFG::SpeculativeJIT::createOSREntries):
(JSC::DFG::SpeculativeJIT::linkOSREntries):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGStaticExecutionCountEstimationPhase.cpp:
(JSC::DFG::StaticExecutionCountEstimationPhase::run):
* dfg/DFGStrengthReductionPhase.cpp:
(JSC::DFG::StrengthReductionPhase::handleNode):
* dfg/DFGTierUpCheckInjectionPhase.cpp:
(JSC::DFG::TierUpCheckInjectionPhase::run):
(JSC::DFG::TierUpCheckInjectionPhase::buildNaturalLoopToLoopHintMap):
* dfg/DFGTypeCheckHoistingPhase.cpp:
(JSC::DFG::TypeCheckHoistingPhase::run):
* dfg/DFGValidate.cpp:
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToB3.cpp:
(JSC::FTL::DFG::LowerDFGToB3::lower):
(JSC::FTL::DFG::LowerDFGToB3::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToB3::isValid):
* jit/JIT.h:
* jit/JITInlines.h:
(JSC::JIT::callOperation):
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOperations.cpp:
* jit/JITOperations.h:
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* llint/LLIntSlowPaths.h:
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
Source/WTF:
This patch generalizes the BackwardsGraph fake root into a more generalizable
class called SingleRootGraph. SingleRootGraph exposes the general graph interface
used in Dominators and NaturalLoops. SingleRootGraph takes as input a graph with
the normal graph interface, but also allows the input graph to contain more than
one root. SingleRootGraph then exposes a single root, which it creates, that has
an outgoing edge to all the roots in the original graph.
* WTF.xcodeproj/project.pbxproj:
* wtf/BackwardsGraph.h:
(WTF::BackwardsGraph::dump const):
(WTF::BackwardsGraph::rootName): Deleted.
(WTF::BackwardsGraph::Node::Node): Deleted.
(WTF::BackwardsGraph::Node::root): Deleted.
(WTF::BackwardsGraph::Node::operator== const): Deleted.
(WTF::BackwardsGraph::Node::operator!= const): Deleted.
(WTF::BackwardsGraph::Node::operator bool const): Deleted.
(WTF::BackwardsGraph::Node::isRoot const): Deleted.
(WTF::BackwardsGraph::Node::node const): Deleted.
(): Deleted.
(WTF::BackwardsGraph::Set::Set): Deleted.
(WTF::BackwardsGraph::Set::add): Deleted.
(WTF::BackwardsGraph::Set::remove): Deleted.
(WTF::BackwardsGraph::Set::contains): Deleted.
(WTF::BackwardsGraph::Set::dump const): Deleted.
(WTF::BackwardsGraph::Map::Map): Deleted.
(WTF::BackwardsGraph::Map::clear): Deleted.
(WTF::BackwardsGraph::Map::size const): Deleted.
(WTF::BackwardsGraph::Map::operator[]): Deleted.
(WTF::BackwardsGraph::Map::operator[] const): Deleted.
* wtf/Dominators.h:
(WTF::Dominators::Dominators):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf const):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl const):
* wtf/SingleRootGraph.h: Added.
(WTF::SingleRootGraphNode::rootName):
(WTF::SingleRootGraphNode::SingleRootGraphNode):
(WTF::SingleRootGraphNode::root):
(WTF::SingleRootGraphNode::operator== const):
(WTF::SingleRootGraphNode::operator!= const):
(WTF::SingleRootGraphNode::operator bool const):
(WTF::SingleRootGraphNode::isRoot const):
(WTF::SingleRootGraphNode::node const):
(WTF::SingleRootGraphSet::add):
(WTF::SingleRootGraphSet::remove):
(WTF::SingleRootGraphSet::contains):
(WTF::SingleRootGraphSet::dump const):
(WTF::SingleRootMap::SingleRootMap):
(WTF::SingleRootMap::clear):
(WTF::SingleRootMap::size const):
(WTF::SingleRootMap::operator[]):
(WTF::SingleRootMap::operator[] const):
(WTF::SingleRootGraph::SingleRootGraph):
(WTF::SingleRootGraph::root const):
(WTF::SingleRootGraph::newMap):
(WTF::SingleRootGraph::successors const):
(WTF::SingleRootGraph::predecessors const):
(WTF::SingleRootGraph::index const):
(WTF::SingleRootGraph::node const):
(WTF::SingleRootGraph::numNodes const):
(WTF::SingleRootGraph::dump const):
(WTF::SingleRootGraph::assertIsConsistent const):
Canonical link: https://commits.webkit.org/192644@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@221196 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2017-08-25 18:26:15 +00:00
|
|
|
const List& from, const Functor& functor)
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
{
|
|
|
|
|
typename Graph::Set set;
|
|
|
|
|
forAllBlocksInIteratedDominanceFrontierOfImpl(
|
|
|
|
|
from,
|
|
|
|
|
[&] (typename Graph::Node block) -> bool {
|
|
|
|
|
if (!set.add(block))
|
|
|
|
|
return false;
|
|
|
|
|
return functor(block);
|
|
|
|
|
});
|
|
|
|
|
}
|
|
|
|
|
|
Support compiling catch in the DFG
https://bugs.webkit.org/show_bug.cgi?id=174590
<rdar://problem/34047845>
Reviewed by Filip Pizlo.
JSTests:
* microbenchmarks/delta-blue-try-catch.js: Added.
(exception):
(value):
(OrderedCollection):
(OrderedCollection.prototype.add):
(OrderedCollection.prototype.at):
(OrderedCollection.prototype.size):
(OrderedCollection.prototype.removeFirst):
(OrderedCollection.prototype.remove):
(Strength):
(Strength.stronger):
(Strength.weaker):
(Strength.weakestOf):
(Strength.strongest):
(Strength.prototype.nextWeaker):
(Constraint):
(Constraint.prototype.addConstraint):
(Constraint.prototype.satisfy):
(Constraint.prototype.destroyConstraint):
(Constraint.prototype.isInput):
(UnaryConstraint):
(UnaryConstraint.prototype.addToGraph):
(UnaryConstraint.prototype.chooseMethod):
(UnaryConstraint.prototype.isSatisfied):
(UnaryConstraint.prototype.markInputs):
(UnaryConstraint.prototype.output):
(UnaryConstraint.prototype.recalculate):
(UnaryConstraint.prototype.markUnsatisfied):
(UnaryConstraint.prototype.inputsKnown):
(UnaryConstraint.prototype.removeFromGraph):
(StayConstraint):
(StayConstraint.prototype.execute):
(EditConstraint.prototype.isInput):
(EditConstraint.prototype.execute):
(BinaryConstraint):
(BinaryConstraint.prototype.chooseMethod):
(BinaryConstraint.prototype.addToGraph):
(BinaryConstraint.prototype.isSatisfied):
(BinaryConstraint.prototype.markInputs):
(BinaryConstraint.prototype.input):
(BinaryConstraint.prototype.output):
(BinaryConstraint.prototype.recalculate):
(BinaryConstraint.prototype.markUnsatisfied):
(BinaryConstraint.prototype.inputsKnown):
(BinaryConstraint.prototype.removeFromGraph):
(ScaleConstraint):
(ScaleConstraint.prototype.addToGraph):
(ScaleConstraint.prototype.removeFromGraph):
(ScaleConstraint.prototype.markInputs):
(ScaleConstraint.prototype.execute):
(ScaleConstraint.prototype.recalculate):
(EqualityConstraint):
(EqualityConstraint.prototype.execute):
(Variable):
(Variable.prototype.addConstraint):
(Variable.prototype.removeConstraint):
(Planner):
(Planner.prototype.incrementalAdd):
(Planner.prototype.incrementalRemove):
(Planner.prototype.newMark):
(Planner.prototype.makePlan):
(Planner.prototype.extractPlanFromConstraints):
(Planner.prototype.addPropagate):
(Planner.prototype.removePropagateFrom):
(Planner.prototype.addConstraintsConsumingTo):
(Plan):
(Plan.prototype.addConstraint):
(Plan.prototype.size):
(Plan.prototype.constraintAt):
(Plan.prototype.execute):
(chainTest):
(projectionTest):
(change):
(deltaBlue):
* microbenchmarks/fake-iterators-that-throw-when-finished.js: Added.
(assert):
(Numbers):
(Numbers.prototype.next):
(return.Transpose):
(return.Transpose.prototype.next):
(transpose):
(verifyEven):
(verifyString):
(foo):
(runIterators):
* microbenchmarks/try-catch-word-count.js: Added.
(let.assert):
(EOF):
(let.texts):
(let.o.apply):
(foo):
(bar):
(f):
(run):
(test1):
(test2):
(test3):
(fn):
(A):
(B):
(A.prototype.getValue):
(B.prototype.getParentValue):
(strlen):
(sum.0):
(test):
(result.test.o):
(set add.set add):
(set forEach):
(stringHash):
(set if):
(testFunction):
(set delete.set has.set add):
* stress/catch-set-argument-speculation-failure.js: Added.
(o):
(e):
(e2):
(escape):
(baz):
(noInline.run):
(noInline):
* stress/osr-enter-to-catch-with-set-local-type-check-failure.js: Added.
(foo):
(e):
(baz):
(bar):
Source/JavaScriptCore:
This patch implements OSR entry into op_catch in the DFG. We will support OSR entry
into the FTL in a followup: https://bugs.webkit.org/show_bug.cgi?id=175396
To implement catch in the DFG, this patch introduces the concept of multiple
entrypoints into CPS/LoadStore DFG IR. A lot of this patch is stringing this concept
through the DFG. Many phases used to assume that Graph::block(0) is the only root, and this
patch contains many straight forward changes generalizing the code to handle more than
one entrypoint.
A main building block of this is moving to two CFG types: SSACFG and CPSCFG. SSACFG
is the same CFG we used to have. CPSCFG is a new type that introduces a fake root
that has an outgoing edge to all the entrypoints. This allows our existing graph algorithms
to Just Work over CPSCFG. For example, there is now the concept of SSADominators vs CPSDominators,
and SSANaturalLoops vs CPSNaturalLoops.
The way we compile the catch entrypoint is by bootstrapping the state
of the program by loading all live bytecode locals from a buffer. The OSR
entry code will store all live values into that buffer before jumping to
the entrypoint. The OSR entry code is also responsible for performing type
proofs of the arguments before doing an OSR entry. If there is a type
mismatch, it's not legal to OSR enter into the DFG compilation. Currently,
each catch entrypoint knows the argument type proofs it must perform to enter
into the DFG. Currently, all entrypoints' arguments flush format are unified
via ArgumentPosition, but this is just an implementation detail. The code is
written more generally to assume that each entrypoint may perform its own distinct
proof.
op_catch now performs value profiling for all live bytecode locals in the
LLInt and baseline JIT. This information is then fed into the DFG via the
ExtractCatchLocal node in the prediction propagation phase.
This patch also changes how we generate op_catch in bytecode. All op_catches
are now split out at the end of the program in bytecode. This ensures that
no op_catch is inside a try block. This is needed to ensure correctness in
the DFGLiveCatchVariablePreservationPhase. That phase only inserts flushes
before SetLocals inside a try block. If an op_catch were in a try block, this
would cause the phase to insert a Flush before one of the state bootstrapping
SetLocals, which would generate invalid IR. Moving op_catch to be generated on
its own at the end of a bytecode stream seemed like the most elegant solution since
it better represents that we treat op_catch as an entrypoint. This is true
both in the DFG and in the baseline and LLInt: we don't reach an op_catch
via normal control flow. Because op_catch cannot throw, this will not break
any previous semantics of op_catch. Logically, it'd be valid to split try
blocks around any non-throwing bytecode operation.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* bytecode/BytecodeDumper.cpp:
(JSC::BytecodeDumper<Block>::dumpBytecode):
* bytecode/BytecodeList.json:
* bytecode/BytecodeUseDef.h:
(JSC::computeUsesForBytecodeOffset):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::finishCreation):
(JSC::CodeBlock::ensureCatchLivenessIsComputedForBytecodeOffset):
(JSC::CodeBlock::updateAllPredictionsAndCountLiveness):
(JSC::CodeBlock::validate):
* bytecode/CodeBlock.h:
* bytecode/ValueProfile.h:
(JSC::ValueProfile::ValueProfile):
(JSC::ValueProfileAndOperandBuffer::ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::~ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::forEach):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::generate):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::emitCatch):
(JSC::BytecodeGenerator::emitEnumeration):
* bytecompiler/BytecodeGenerator.h:
* bytecompiler/NodesCodegen.cpp:
(JSC::TryNode::emitBytecode):
* dfg/DFGAbstractInterpreterInlines.h:
(JSC::DFG::AbstractInterpreter<AbstractStateType>::executeEffects):
* dfg/DFGBackwardsCFG.h:
(JSC::DFG::BackwardsCFG::BackwardsCFG):
* dfg/DFGBasicBlock.cpp:
(JSC::DFG::BasicBlock::BasicBlock):
* dfg/DFGBasicBlock.h:
(JSC::DFG::BasicBlock::findTerminal const):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::setDirect):
(JSC::DFG::ByteCodeParser::flush):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::DelayedSetLocal):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::execute):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::parseCodeBlock):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCFG.h:
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::roots):
(JSC::DFG::CPSCFG::CPSCFG):
(JSC::DFG::selectCFG):
* dfg/DFGCPSRethreadingPhase.cpp:
(JSC::DFG::CPSRethreadingPhase::specialCaseArguments):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGClobberize.h:
(JSC::DFG::clobberize):
* dfg/DFGControlEquivalenceAnalysis.h:
(JSC::DFG::ControlEquivalenceAnalysis::ControlEquivalenceAnalysis):
* dfg/DFGDCEPhase.cpp:
(JSC::DFG::DCEPhase::run):
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDoesGC.cpp:
(JSC::DFG::doesGC):
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::ensureDominatorsForCFG):
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::EdgeDominates):
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
(JSC::DFG::FixupPhase::fixupChecksInBlock):
* dfg/DFGFlushFormat.h:
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::unboxLoopNode):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::dump):
(JSC::DFG::Graph::determineReachability):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::blocksInPreOrder):
(JSC::DFG::Graph::blocksInPostOrder):
(JSC::DFG::Graph::ensureCPSDominators):
(JSC::DFG::Graph::ensureSSADominators):
(JSC::DFG::Graph::ensureCPSNaturalLoops):
(JSC::DFG::Graph::ensureSSANaturalLoops):
(JSC::DFG::Graph::ensureBackwardsCFG):
(JSC::DFG::Graph::ensureBackwardsDominators):
(JSC::DFG::Graph::ensureControlEquivalenceAnalysis):
(JSC::DFG::Graph::methodOfGettingAValueProfileFor):
(JSC::DFG::Graph::clearCPSCFGData):
(JSC::DFG::Graph::ensureDominators): Deleted.
(JSC::DFG::Graph::ensurePrePostNumbering): Deleted.
(JSC::DFG::Graph::ensureNaturalLoops): Deleted.
* dfg/DFGGraph.h:
(JSC::DFG::Graph::willCatchExceptionInMachineFrame):
(JSC::DFG::Graph::isEntrypoint const):
* dfg/DFGInPlaceAbstractState.cpp:
(JSC::DFG::InPlaceAbstractState::initialize):
(JSC::DFG::InPlaceAbstractState::mergeToSuccessors):
* dfg/DFGJITCode.cpp:
(JSC::DFG::JITCode::shrinkToFit):
* dfg/DFGJITCode.h:
(JSC::DFG::JITCode::catchOSREntryDataForBytecodeIndex):
(JSC::DFG::JITCode::finalizeCatchOSREntrypoints):
(JSC::DFG::JITCode::appendCatchEntrypoint):
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::compile):
(JSC::DFG::JITCompiler::compileFunction):
(JSC::DFG::JITCompiler::noticeCatchEntrypoint):
(JSC::DFG::JITCompiler::noticeOSREntry):
(JSC::DFG::JITCompiler::makeCatchOSREntryBuffer):
* dfg/DFGJITCompiler.h:
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::run):
(JSC::DFG::LiveCatchVariablePreservationPhase::isValidFlushLocation):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlockForTryCatch):
(JSC::DFG::LiveCatchVariablePreservationPhase::newVariableAccessData):
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException): Deleted.
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock): Deleted.
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGMaximalFlushInsertionPhase.cpp:
(JSC::DFG::MaximalFlushInsertionPhase::run):
(JSC::DFG::MaximalFlushInsertionPhase::treatRegularBlock):
(JSC::DFG::MaximalFlushInsertionPhase::treatRootBlock):
* dfg/DFGMayExit.cpp:
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::isSwitch const):
(JSC::DFG::Node::successor):
(JSC::DFG::Node::catchOSREntryIndex const):
(JSC::DFG::Node::catchLocalPrediction):
(JSC::DFG::Node::isSwitch): Deleted.
* dfg/DFGNodeType.h:
* dfg/DFGOSREntry.cpp:
(JSC::DFG::prepareCatchOSREntry):
* dfg/DFGOSREntry.h:
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGOSRExitCompilerCommon.cpp:
(JSC::DFG::handleExitCounts):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering): Deleted.
(JSC::DFG::PrePostNumbering::~PrePostNumbering): Deleted.
(WTF::printInternal): Deleted.
* dfg/DFGPrePostNumbering.h:
(): Deleted.
(JSC::DFG::PrePostNumbering::preNumber const): Deleted.
(JSC::DFG::PrePostNumbering::postNumber const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::isDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::edgeKind const): Deleted.
* dfg/DFGPredictionInjectionPhase.cpp:
(JSC::DFG::PredictionInjectionPhase::run):
* dfg/DFGPredictionPropagationPhase.cpp:
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
(JSC::DFG::performSSAConversion):
* dfg/DFGSafeToExecute.h:
(JSC::DFG::safeToExecute):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compileCurrentBlock):
(JSC::DFG::SpeculativeJIT::checkArgumentTypes):
(JSC::DFG::SpeculativeJIT::createOSREntries):
(JSC::DFG::SpeculativeJIT::linkOSREntries):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGStaticExecutionCountEstimationPhase.cpp:
(JSC::DFG::StaticExecutionCountEstimationPhase::run):
* dfg/DFGStrengthReductionPhase.cpp:
(JSC::DFG::StrengthReductionPhase::handleNode):
* dfg/DFGTierUpCheckInjectionPhase.cpp:
(JSC::DFG::TierUpCheckInjectionPhase::run):
(JSC::DFG::TierUpCheckInjectionPhase::buildNaturalLoopToLoopHintMap):
* dfg/DFGTypeCheckHoistingPhase.cpp:
(JSC::DFG::TypeCheckHoistingPhase::run):
* dfg/DFGValidate.cpp:
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToB3.cpp:
(JSC::FTL::DFG::LowerDFGToB3::lower):
(JSC::FTL::DFG::LowerDFGToB3::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToB3::isValid):
* jit/JIT.h:
* jit/JITInlines.h:
(JSC::JIT::callOperation):
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOperations.cpp:
* jit/JITOperations.h:
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* llint/LLIntSlowPaths.h:
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
Source/WTF:
This patch generalizes the BackwardsGraph fake root into a more generalizable
class called SingleRootGraph. SingleRootGraph exposes the general graph interface
used in Dominators and NaturalLoops. SingleRootGraph takes as input a graph with
the normal graph interface, but also allows the input graph to contain more than
one root. SingleRootGraph then exposes a single root, which it creates, that has
an outgoing edge to all the roots in the original graph.
* WTF.xcodeproj/project.pbxproj:
* wtf/BackwardsGraph.h:
(WTF::BackwardsGraph::dump const):
(WTF::BackwardsGraph::rootName): Deleted.
(WTF::BackwardsGraph::Node::Node): Deleted.
(WTF::BackwardsGraph::Node::root): Deleted.
(WTF::BackwardsGraph::Node::operator== const): Deleted.
(WTF::BackwardsGraph::Node::operator!= const): Deleted.
(WTF::BackwardsGraph::Node::operator bool const): Deleted.
(WTF::BackwardsGraph::Node::isRoot const): Deleted.
(WTF::BackwardsGraph::Node::node const): Deleted.
(): Deleted.
(WTF::BackwardsGraph::Set::Set): Deleted.
(WTF::BackwardsGraph::Set::add): Deleted.
(WTF::BackwardsGraph::Set::remove): Deleted.
(WTF::BackwardsGraph::Set::contains): Deleted.
(WTF::BackwardsGraph::Set::dump const): Deleted.
(WTF::BackwardsGraph::Map::Map): Deleted.
(WTF::BackwardsGraph::Map::clear): Deleted.
(WTF::BackwardsGraph::Map::size const): Deleted.
(WTF::BackwardsGraph::Map::operator[]): Deleted.
(WTF::BackwardsGraph::Map::operator[] const): Deleted.
* wtf/Dominators.h:
(WTF::Dominators::Dominators):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf const):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl const):
* wtf/SingleRootGraph.h: Added.
(WTF::SingleRootGraphNode::rootName):
(WTF::SingleRootGraphNode::SingleRootGraphNode):
(WTF::SingleRootGraphNode::root):
(WTF::SingleRootGraphNode::operator== const):
(WTF::SingleRootGraphNode::operator!= const):
(WTF::SingleRootGraphNode::operator bool const):
(WTF::SingleRootGraphNode::isRoot const):
(WTF::SingleRootGraphNode::node const):
(WTF::SingleRootGraphSet::add):
(WTF::SingleRootGraphSet::remove):
(WTF::SingleRootGraphSet::contains):
(WTF::SingleRootGraphSet::dump const):
(WTF::SingleRootMap::SingleRootMap):
(WTF::SingleRootMap::clear):
(WTF::SingleRootMap::size const):
(WTF::SingleRootMap::operator[]):
(WTF::SingleRootMap::operator[] const):
(WTF::SingleRootGraph::SingleRootGraph):
(WTF::SingleRootGraph::root const):
(WTF::SingleRootGraph::newMap):
(WTF::SingleRootGraph::successors const):
(WTF::SingleRootGraph::predecessors const):
(WTF::SingleRootGraph::index const):
(WTF::SingleRootGraph::node const):
(WTF::SingleRootGraph::numNodes const):
(WTF::SingleRootGraph::dump const):
(WTF::SingleRootGraph::assertIsConsistent const):
Canonical link: https://commits.webkit.org/192644@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@221196 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2017-08-25 18:26:15 +00:00
|
|
|
typename Graph::Set iteratedDominanceFrontierOf(const List& from) const
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
{
|
|
|
|
|
typename Graph::Set result;
|
|
|
|
|
forAllBlocksInIteratedDominanceFrontierOfImpl(
|
|
|
|
|
from,
|
|
|
|
|
[&] (typename Graph::Node node) -> bool {
|
|
|
|
|
return result.add(node);
|
|
|
|
|
});
|
|
|
|
|
return result;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void dump(PrintStream& out) const
|
|
|
|
|
{
|
|
|
|
|
for (unsigned blockIndex = 0; blockIndex < m_data.size(); ++blockIndex) {
|
|
|
|
|
if (m_data[blockIndex].preNumber == UINT_MAX)
|
|
|
|
|
continue;
|
|
|
|
|
|
2016-05-19 21:25:29 +00:00
|
|
|
out.print(" Block #", blockIndex, ": idom = ", m_graph.dump(m_data[blockIndex].idomParent), ", idomKids = [");
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
CommaPrinter comma;
|
|
|
|
|
for (unsigned i = 0; i < m_data[blockIndex].idomKids.size(); ++i)
|
2016-05-19 21:25:29 +00:00
|
|
|
out.print(comma, m_graph.dump(m_data[blockIndex].idomKids[i]));
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
out.print("], pre/post = ", m_data[blockIndex].preNumber, "/", m_data[blockIndex].postNumber, "\n");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
// This implements Lengauer and Tarjan's "A Fast Algorithm for Finding Dominators in a Flowgraph"
|
|
|
|
|
// (TOPLAS 1979). It uses the "simple" implementation of LINK and EVAL, which yields an O(n log n)
|
|
|
|
|
// solution. The full paper is linked below; this code attempts to closely follow the algorithm as
|
|
|
|
|
// it is presented in the paper; in particular sections 3 and 4 as well as appendix B.
|
|
|
|
|
// https://www.cs.princeton.edu/courses/archive/fall03/cs528/handouts/a%20fast%20algorithm%20for%20finding.pdf
|
|
|
|
|
//
|
|
|
|
|
// This code is very subtle. The Lengauer-Tarjan algorithm is incredibly deep to begin with. The
|
|
|
|
|
// goal of this code is to follow the code in the paper, however our implementation must deviate
|
|
|
|
|
// from the paper when it comes to recursion. The authors had used recursion to implement DFS, and
|
|
|
|
|
// also to implement the "simple" EVAL. We convert both of those into worklist-based solutions.
|
|
|
|
|
// Finally, once the algorithm gives us immediate dominators, we implement dominance tests by
|
|
|
|
|
// walking the dominator tree and computing pre and post numbers. We then use the range inclusion
|
|
|
|
|
// check trick that was first discovered by Paul F. Dietz in 1982 in "Maintaining order in a linked
|
|
|
|
|
// list" (see http://dl.acm.org/citation.cfm?id=802184).
|
|
|
|
|
|
|
|
|
|
class LengauerTarjan {
|
2019-08-12 20:57:15 +00:00
|
|
|
WTF_MAKE_FAST_ALLOCATED;
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
public:
|
|
|
|
|
LengauerTarjan(Graph& graph)
|
|
|
|
|
: m_graph(graph)
|
|
|
|
|
, m_data(graph.template newMap<BlockData>())
|
|
|
|
|
{
|
|
|
|
|
for (unsigned blockIndex = m_graph.numNodes(); blockIndex--;) {
|
|
|
|
|
typename Graph::Node block = m_graph.node(blockIndex);
|
|
|
|
|
if (!block)
|
|
|
|
|
continue;
|
|
|
|
|
m_data[block].label = block;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void compute()
|
|
|
|
|
{
|
|
|
|
|
computeDepthFirstPreNumbering(); // Step 1.
|
|
|
|
|
computeSemiDominatorsAndImplicitImmediateDominators(); // Steps 2 and 3.
|
|
|
|
|
computeExplicitImmediateDominators(); // Step 4.
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typename Graph::Node immediateDominator(typename Graph::Node block)
|
|
|
|
|
{
|
|
|
|
|
return m_data[block].dom;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
void computeDepthFirstPreNumbering()
|
|
|
|
|
{
|
|
|
|
|
// Use a block worklist that also tracks the index inside the successor list. This is
|
|
|
|
|
// necessary for ensuring that we don't attempt to visit a successor until the previous
|
|
|
|
|
// successors that we had visited are fully processed. This ends up being revealed in the
|
|
|
|
|
// output of this method because the first time we see an edge to a block, we set the
|
|
|
|
|
// block's parent. So, if we have:
|
|
|
|
|
//
|
|
|
|
|
// A -> B
|
|
|
|
|
// A -> C
|
|
|
|
|
// B -> C
|
|
|
|
|
//
|
|
|
|
|
// And we're processing A, then we want to ensure that if we see A->B first (and hence set
|
|
|
|
|
// B's prenumber before we set C's) then we also end up setting C's parent to B by virtue
|
|
|
|
|
// of not noticing A->C until we're done processing B.
|
|
|
|
|
|
|
|
|
|
ExtendedGraphNodeWorklist<typename Graph::Node, unsigned, typename Graph::Set> worklist;
|
2016-05-19 21:25:29 +00:00
|
|
|
worklist.push(m_graph.root(), 0);
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
|
|
|
|
|
while (GraphNodeWith<typename Graph::Node, unsigned> item = worklist.pop()) {
|
|
|
|
|
typename Graph::Node block = item.node;
|
|
|
|
|
unsigned successorIndex = item.data;
|
|
|
|
|
|
|
|
|
|
// We initially push with successorIndex = 0 regardless of whether or not we have any
|
|
|
|
|
// successors. This is so that we can assign our prenumber. Subsequently we get pushed
|
|
|
|
|
// with higher successorIndex values, but only if they are in range.
|
|
|
|
|
ASSERT(!successorIndex || successorIndex < m_graph.successors(block).size());
|
|
|
|
|
|
|
|
|
|
if (!successorIndex) {
|
|
|
|
|
m_data[block].semiNumber = m_blockByPreNumber.size();
|
|
|
|
|
m_blockByPreNumber.append(block);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (successorIndex < m_graph.successors(block).size()) {
|
|
|
|
|
unsigned nextSuccessorIndex = successorIndex + 1;
|
|
|
|
|
if (nextSuccessorIndex < m_graph.successors(block).size())
|
|
|
|
|
worklist.forcePush(block, nextSuccessorIndex);
|
|
|
|
|
|
|
|
|
|
typename Graph::Node successorBlock = m_graph.successors(block)[successorIndex];
|
|
|
|
|
if (worklist.push(successorBlock, 0))
|
|
|
|
|
m_data[successorBlock].parent = block;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void computeSemiDominatorsAndImplicitImmediateDominators()
|
|
|
|
|
{
|
|
|
|
|
for (unsigned currentPreNumber = m_blockByPreNumber.size(); currentPreNumber-- > 1;) {
|
|
|
|
|
typename Graph::Node block = m_blockByPreNumber[currentPreNumber];
|
|
|
|
|
BlockData& blockData = m_data[block];
|
|
|
|
|
|
|
|
|
|
// Step 2:
|
|
|
|
|
for (typename Graph::Node predecessorBlock : m_graph.predecessors(block)) {
|
|
|
|
|
typename Graph::Node intermediateBlock = eval(predecessorBlock);
|
|
|
|
|
blockData.semiNumber = std::min(
|
|
|
|
|
m_data[intermediateBlock].semiNumber, blockData.semiNumber);
|
|
|
|
|
}
|
|
|
|
|
unsigned bucketPreNumber = blockData.semiNumber;
|
|
|
|
|
ASSERT(bucketPreNumber <= currentPreNumber);
|
|
|
|
|
m_data[m_blockByPreNumber[bucketPreNumber]].bucket.append(block);
|
|
|
|
|
link(blockData.parent, block);
|
|
|
|
|
|
|
|
|
|
// Step 3:
|
|
|
|
|
for (typename Graph::Node semiDominee : m_data[blockData.parent].bucket) {
|
|
|
|
|
typename Graph::Node possibleDominator = eval(semiDominee);
|
|
|
|
|
BlockData& semiDomineeData = m_data[semiDominee];
|
|
|
|
|
ASSERT(m_blockByPreNumber[semiDomineeData.semiNumber] == blockData.parent);
|
|
|
|
|
BlockData& possibleDominatorData = m_data[possibleDominator];
|
|
|
|
|
if (possibleDominatorData.semiNumber < semiDomineeData.semiNumber)
|
|
|
|
|
semiDomineeData.dom = possibleDominator;
|
|
|
|
|
else
|
|
|
|
|
semiDomineeData.dom = blockData.parent;
|
|
|
|
|
}
|
|
|
|
|
m_data[blockData.parent].bucket.clear();
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void computeExplicitImmediateDominators()
|
|
|
|
|
{
|
|
|
|
|
for (unsigned currentPreNumber = 1; currentPreNumber < m_blockByPreNumber.size(); ++currentPreNumber) {
|
|
|
|
|
typename Graph::Node block = m_blockByPreNumber[currentPreNumber];
|
|
|
|
|
BlockData& blockData = m_data[block];
|
|
|
|
|
|
|
|
|
|
if (blockData.dom != m_blockByPreNumber[blockData.semiNumber])
|
|
|
|
|
blockData.dom = m_data[blockData.dom].dom;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void link(typename Graph::Node from, typename Graph::Node to)
|
|
|
|
|
{
|
|
|
|
|
m_data[to].ancestor = from;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typename Graph::Node eval(typename Graph::Node block)
|
|
|
|
|
{
|
|
|
|
|
if (!m_data[block].ancestor)
|
|
|
|
|
return block;
|
|
|
|
|
|
|
|
|
|
compress(block);
|
|
|
|
|
return m_data[block].label;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void compress(typename Graph::Node initialBlock)
|
|
|
|
|
{
|
|
|
|
|
// This was meant to be a recursive function, but we don't like recursion because we don't
|
|
|
|
|
// want to blow the stack. The original function will call compress() recursively on the
|
|
|
|
|
// ancestor of anything that has an ancestor. So, we populate our worklist with the
|
|
|
|
|
// recursive ancestors of initialBlock. Then we process the list starting from the block
|
|
|
|
|
// that is furthest up the ancestor chain.
|
|
|
|
|
|
|
|
|
|
typename Graph::Node ancestor = m_data[initialBlock].ancestor;
|
|
|
|
|
ASSERT(ancestor);
|
|
|
|
|
if (!m_data[ancestor].ancestor)
|
|
|
|
|
return;
|
|
|
|
|
|
|
|
|
|
Vector<typename Graph::Node, 16> stack;
|
|
|
|
|
for (typename Graph::Node block = initialBlock; block; block = m_data[block].ancestor)
|
|
|
|
|
stack.append(block);
|
|
|
|
|
|
|
|
|
|
// We only care about blocks that have an ancestor that has an ancestor. The last two
|
|
|
|
|
// elements in the stack won't satisfy this property.
|
|
|
|
|
ASSERT(stack.size() >= 2);
|
|
|
|
|
ASSERT(!m_data[stack[stack.size() - 1]].ancestor);
|
|
|
|
|
ASSERT(!m_data[m_data[stack[stack.size() - 2]].ancestor].ancestor);
|
|
|
|
|
|
|
|
|
|
for (unsigned i = stack.size() - 2; i--;) {
|
|
|
|
|
typename Graph::Node block = stack[i];
|
|
|
|
|
typename Graph::Node& labelOfBlock = m_data[block].label;
|
|
|
|
|
typename Graph::Node& ancestorOfBlock = m_data[block].ancestor;
|
|
|
|
|
ASSERT(ancestorOfBlock);
|
|
|
|
|
ASSERT(m_data[ancestorOfBlock].ancestor);
|
|
|
|
|
|
|
|
|
|
typename Graph::Node labelOfAncestorOfBlock = m_data[ancestorOfBlock].label;
|
|
|
|
|
|
|
|
|
|
if (m_data[labelOfAncestorOfBlock].semiNumber < m_data[labelOfBlock].semiNumber)
|
|
|
|
|
labelOfBlock = labelOfAncestorOfBlock;
|
|
|
|
|
ancestorOfBlock = m_data[ancestorOfBlock].ancestor;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
struct BlockData {
|
2019-08-12 20:57:15 +00:00
|
|
|
WTF_MAKE_STRUCT_FAST_ALLOCATED;
|
|
|
|
|
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
BlockData()
|
|
|
|
|
: parent(nullptr)
|
|
|
|
|
, preNumber(UINT_MAX)
|
|
|
|
|
, semiNumber(UINT_MAX)
|
|
|
|
|
, ancestor(nullptr)
|
|
|
|
|
, label(nullptr)
|
|
|
|
|
, dom(nullptr)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typename Graph::Node parent;
|
|
|
|
|
unsigned preNumber;
|
|
|
|
|
unsigned semiNumber;
|
|
|
|
|
typename Graph::Node ancestor;
|
|
|
|
|
typename Graph::Node label;
|
|
|
|
|
Vector<typename Graph::Node> bucket;
|
|
|
|
|
typename Graph::Node dom;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
Graph& m_graph;
|
|
|
|
|
typename Graph::template Map<BlockData> m_data;
|
|
|
|
|
Vector<typename Graph::Node> m_blockByPreNumber;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
class NaiveDominators {
|
2019-08-12 20:57:15 +00:00
|
|
|
WTF_MAKE_FAST_ALLOCATED;
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
public:
|
|
|
|
|
NaiveDominators(Graph& graph)
|
|
|
|
|
: m_graph(graph)
|
|
|
|
|
{
|
|
|
|
|
// This implements a naive dominator solver.
|
|
|
|
|
|
|
|
|
|
ASSERT(!graph.predecessors(graph.root()).size());
|
|
|
|
|
|
|
|
|
|
unsigned numBlocks = graph.numNodes();
|
|
|
|
|
|
|
|
|
|
// Allocate storage for the dense dominance matrix.
|
|
|
|
|
m_results.grow(numBlocks);
|
|
|
|
|
for (unsigned i = numBlocks; i--;)
|
|
|
|
|
m_results[i].resize(numBlocks);
|
|
|
|
|
m_scratch.resize(numBlocks);
|
|
|
|
|
|
|
|
|
|
// We know that the entry block is only dominated by itself.
|
|
|
|
|
m_results[0].clearAll();
|
FastBitVector should have efficient and easy-to-use vector-vector operations
https://bugs.webkit.org/show_bug.cgi?id=161847
Reviewed by Saam Barati.
Source/JavaScriptCore:
Adapt existing users of FastBitVector to the new API.
* bytecode/BytecodeLivenessAnalysis.cpp:
(JSC::BytecodeLivenessAnalysis::computeKills):
(JSC::BytecodeLivenessAnalysis::dumpResults):
* bytecode/BytecodeLivenessAnalysisInlines.h:
(JSC::operandThatIsNotAlwaysLiveIsLive):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::stepOverInstruction):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::runLivenessFixpoint):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::validate):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::flushForTerminal):
* dfg/DFGForAllKills.h:
(JSC::DFG::forAllKilledOperands):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::forAllLocalsLiveInBytecode):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::cleanMustHandleValuesIfNecessary):
Source/WTF:
FastBitVector is a bitvector representation that supports manual dynamic resizing and is
optimized for speed, not space. (BitVector supports automatic dynamic resizing and is
optimized for space, while Bitmap is sized statically and is optimized for both speed and
space.) This change greatly increases the power of FastBitVector. We will use these new
powers for changing the JSC GC to use FastBitVectors to track sets of MarkedBlocks (bug
161581) instead of using a combination of Vectors and doubly-linked lists.
This change splits FastBitVector into two parts:
- A thing that manages the storage of a bitvector: a uint32_t array and a size_t numBits.
We call this the word view.
- A thing that takes some kind of abstract array of uint32_t's and does bitvector
operations to it. We call this the FastBitVectorImpl.
FastBitVectorImpl and word views are immutable. The FastBitVector class is a subclass of
FastBitVectorImpl specialized on a word view that owns its words and has additional
support for mutable operations.
Doing this allows us to efficiently support things like this without any unnecessary
memory allocation or copying:
FastBitVector a, b, c; // Assume that there is code to initialize these.
a &= b | ~c;
Previously, this kind of operation would not be efficient, because "~c" would have to
create a whole new FastBitVector. But now, it just returns a FastBitVectorImpl whose
underlying word view bitnots (~) its words on the fly. Using template magic, this can get
pretty complex. For example "b | ~c" returns a FastBitVectorImpl that wraps a word view
whose implementation of WordView::word(size_t index) is something like:
uint32_t word(size_t index) { return b.m_words.word(index) | ~c.m_words.word(index); }
FastBitVectorImpl supports all of the fast bulk bitvector operations, like
forEachSetBit(), bitCount(), etc. So, when you say "a &= b | ~c", the actual
implementation is going to run these bit operations on word granularity directly over the
storage inside a, b, c.
The use of operator overloading is worth explaining a bit. Previously, FastBitVector
avoided operator overloading. For example, the &= operation was called filter(). I think
that this was a pretty good approach at the time. I tried using non-operator methods in
this FastBitVector rewrite, but I found it very odd to say things like:
a.filter(b.bitOr(c.bitNot()));
I think that it's harder to see what is going on here, then using operators, because infix
notation is always better.
* WTF.xcodeproj/project.pbxproj:
* wtf/BitVector.h:
(WTF::BitVector::findBitInWord): Deleted.
* wtf/CMakeLists.txt:
* wtf/Dominators.h:
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::pruneDominators):
* wtf/FastBitVector.cpp: Removed.
* wtf/FastBitVector.h:
(WTF::fastBitVectorArrayLength):
(WTF::FastBitVectorWordView::FastBitVectorWordView):
(WTF::FastBitVectorWordView::numBits):
(WTF::FastBitVectorWordView::word):
(WTF::FastBitVectorWordOwner::FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::~FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::view):
(WTF::FastBitVectorWordOwner::operator=):
(WTF::FastBitVectorWordOwner::setAll):
(WTF::FastBitVectorWordOwner::clearAll):
(WTF::FastBitVectorWordOwner::set):
(WTF::FastBitVectorWordOwner::numBits):
(WTF::FastBitVectorWordOwner::arrayLength):
(WTF::FastBitVectorWordOwner::resize):
(WTF::FastBitVectorWordOwner::word):
(WTF::FastBitVectorWordOwner::words):
(WTF::FastBitVectorAndWords::FastBitVectorAndWords):
(WTF::FastBitVectorAndWords::view):
(WTF::FastBitVectorAndWords::numBits):
(WTF::FastBitVectorAndWords::word):
(WTF::FastBitVectorOrWords::FastBitVectorOrWords):
(WTF::FastBitVectorOrWords::view):
(WTF::FastBitVectorOrWords::numBits):
(WTF::FastBitVectorOrWords::word):
(WTF::FastBitVectorNotWords::FastBitVectorNotWords):
(WTF::FastBitVectorNotWords::view):
(WTF::FastBitVectorNotWords::numBits):
(WTF::FastBitVectorNotWords::word):
(WTF::FastBitVectorImpl::FastBitVectorImpl):
(WTF::FastBitVectorImpl::numBits):
(WTF::FastBitVectorImpl::size):
(WTF::FastBitVectorImpl::arrayLength):
(WTF::FastBitVectorImpl::operator==):
(WTF::FastBitVectorImpl::operator!=):
(WTF::FastBitVectorImpl::at):
(WTF::FastBitVectorImpl::operator[]):
(WTF::FastBitVectorImpl::bitCount):
(WTF::FastBitVectorImpl::operator&):
(WTF::FastBitVectorImpl::operator|):
(WTF::FastBitVectorImpl::operator~):
(WTF::FastBitVectorImpl::forEachSetBit):
(WTF::FastBitVectorImpl::forEachClearBit):
(WTF::FastBitVectorImpl::forEachBit):
(WTF::FastBitVectorImpl::findBit):
(WTF::FastBitVectorImpl::findSetBit):
(WTF::FastBitVectorImpl::findClearBit):
(WTF::FastBitVectorImpl::dump):
(WTF::FastBitVectorImpl::atImpl):
(WTF::FastBitVector::FastBitVector):
(WTF::FastBitVector::operator=):
(WTF::FastBitVector::resize):
(WTF::FastBitVector::setAll):
(WTF::FastBitVector::clearAll):
(WTF::FastBitVector::setAndCheck):
(WTF::FastBitVector::operator|=):
(WTF::FastBitVector::operator&=):
(WTF::FastBitVector::at):
(WTF::FastBitVector::operator[]):
(WTF::FastBitVector::BitReference::BitReference):
(WTF::FastBitVector::BitReference::operator bool):
(WTF::FastBitVector::BitReference::operator=):
(WTF::FastBitVector::~FastBitVector): Deleted.
(WTF::FastBitVector::numBits): Deleted.
(WTF::FastBitVector::set): Deleted.
(WTF::FastBitVector::equals): Deleted.
(WTF::FastBitVector::merge): Deleted.
(WTF::FastBitVector::filter): Deleted.
(WTF::FastBitVector::exclude): Deleted.
(WTF::FastBitVector::clear): Deleted.
(WTF::FastBitVector::get): Deleted.
(WTF::FastBitVector::bitCount): Deleted.
(WTF::FastBitVector::forEachSetBit): Deleted.
(WTF::FastBitVector::arrayLength): Deleted.
* wtf/StdLibExtras.h:
(WTF::findBitInWord):
Canonical link: https://commits.webkit.org/180007@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@205794 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2016-09-12 04:03:36 +00:00
|
|
|
m_results[0][0] = true;
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
|
|
|
|
|
// Find all of the valid blocks.
|
|
|
|
|
m_scratch.clearAll();
|
|
|
|
|
for (unsigned i = numBlocks; i--;) {
|
|
|
|
|
if (!graph.node(i))
|
|
|
|
|
continue;
|
FastBitVector should have efficient and easy-to-use vector-vector operations
https://bugs.webkit.org/show_bug.cgi?id=161847
Reviewed by Saam Barati.
Source/JavaScriptCore:
Adapt existing users of FastBitVector to the new API.
* bytecode/BytecodeLivenessAnalysis.cpp:
(JSC::BytecodeLivenessAnalysis::computeKills):
(JSC::BytecodeLivenessAnalysis::dumpResults):
* bytecode/BytecodeLivenessAnalysisInlines.h:
(JSC::operandThatIsNotAlwaysLiveIsLive):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::stepOverInstruction):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::runLivenessFixpoint):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::validate):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::flushForTerminal):
* dfg/DFGForAllKills.h:
(JSC::DFG::forAllKilledOperands):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::forAllLocalsLiveInBytecode):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::cleanMustHandleValuesIfNecessary):
Source/WTF:
FastBitVector is a bitvector representation that supports manual dynamic resizing and is
optimized for speed, not space. (BitVector supports automatic dynamic resizing and is
optimized for space, while Bitmap is sized statically and is optimized for both speed and
space.) This change greatly increases the power of FastBitVector. We will use these new
powers for changing the JSC GC to use FastBitVectors to track sets of MarkedBlocks (bug
161581) instead of using a combination of Vectors and doubly-linked lists.
This change splits FastBitVector into two parts:
- A thing that manages the storage of a bitvector: a uint32_t array and a size_t numBits.
We call this the word view.
- A thing that takes some kind of abstract array of uint32_t's and does bitvector
operations to it. We call this the FastBitVectorImpl.
FastBitVectorImpl and word views are immutable. The FastBitVector class is a subclass of
FastBitVectorImpl specialized on a word view that owns its words and has additional
support for mutable operations.
Doing this allows us to efficiently support things like this without any unnecessary
memory allocation or copying:
FastBitVector a, b, c; // Assume that there is code to initialize these.
a &= b | ~c;
Previously, this kind of operation would not be efficient, because "~c" would have to
create a whole new FastBitVector. But now, it just returns a FastBitVectorImpl whose
underlying word view bitnots (~) its words on the fly. Using template magic, this can get
pretty complex. For example "b | ~c" returns a FastBitVectorImpl that wraps a word view
whose implementation of WordView::word(size_t index) is something like:
uint32_t word(size_t index) { return b.m_words.word(index) | ~c.m_words.word(index); }
FastBitVectorImpl supports all of the fast bulk bitvector operations, like
forEachSetBit(), bitCount(), etc. So, when you say "a &= b | ~c", the actual
implementation is going to run these bit operations on word granularity directly over the
storage inside a, b, c.
The use of operator overloading is worth explaining a bit. Previously, FastBitVector
avoided operator overloading. For example, the &= operation was called filter(). I think
that this was a pretty good approach at the time. I tried using non-operator methods in
this FastBitVector rewrite, but I found it very odd to say things like:
a.filter(b.bitOr(c.bitNot()));
I think that it's harder to see what is going on here, then using operators, because infix
notation is always better.
* WTF.xcodeproj/project.pbxproj:
* wtf/BitVector.h:
(WTF::BitVector::findBitInWord): Deleted.
* wtf/CMakeLists.txt:
* wtf/Dominators.h:
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::pruneDominators):
* wtf/FastBitVector.cpp: Removed.
* wtf/FastBitVector.h:
(WTF::fastBitVectorArrayLength):
(WTF::FastBitVectorWordView::FastBitVectorWordView):
(WTF::FastBitVectorWordView::numBits):
(WTF::FastBitVectorWordView::word):
(WTF::FastBitVectorWordOwner::FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::~FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::view):
(WTF::FastBitVectorWordOwner::operator=):
(WTF::FastBitVectorWordOwner::setAll):
(WTF::FastBitVectorWordOwner::clearAll):
(WTF::FastBitVectorWordOwner::set):
(WTF::FastBitVectorWordOwner::numBits):
(WTF::FastBitVectorWordOwner::arrayLength):
(WTF::FastBitVectorWordOwner::resize):
(WTF::FastBitVectorWordOwner::word):
(WTF::FastBitVectorWordOwner::words):
(WTF::FastBitVectorAndWords::FastBitVectorAndWords):
(WTF::FastBitVectorAndWords::view):
(WTF::FastBitVectorAndWords::numBits):
(WTF::FastBitVectorAndWords::word):
(WTF::FastBitVectorOrWords::FastBitVectorOrWords):
(WTF::FastBitVectorOrWords::view):
(WTF::FastBitVectorOrWords::numBits):
(WTF::FastBitVectorOrWords::word):
(WTF::FastBitVectorNotWords::FastBitVectorNotWords):
(WTF::FastBitVectorNotWords::view):
(WTF::FastBitVectorNotWords::numBits):
(WTF::FastBitVectorNotWords::word):
(WTF::FastBitVectorImpl::FastBitVectorImpl):
(WTF::FastBitVectorImpl::numBits):
(WTF::FastBitVectorImpl::size):
(WTF::FastBitVectorImpl::arrayLength):
(WTF::FastBitVectorImpl::operator==):
(WTF::FastBitVectorImpl::operator!=):
(WTF::FastBitVectorImpl::at):
(WTF::FastBitVectorImpl::operator[]):
(WTF::FastBitVectorImpl::bitCount):
(WTF::FastBitVectorImpl::operator&):
(WTF::FastBitVectorImpl::operator|):
(WTF::FastBitVectorImpl::operator~):
(WTF::FastBitVectorImpl::forEachSetBit):
(WTF::FastBitVectorImpl::forEachClearBit):
(WTF::FastBitVectorImpl::forEachBit):
(WTF::FastBitVectorImpl::findBit):
(WTF::FastBitVectorImpl::findSetBit):
(WTF::FastBitVectorImpl::findClearBit):
(WTF::FastBitVectorImpl::dump):
(WTF::FastBitVectorImpl::atImpl):
(WTF::FastBitVector::FastBitVector):
(WTF::FastBitVector::operator=):
(WTF::FastBitVector::resize):
(WTF::FastBitVector::setAll):
(WTF::FastBitVector::clearAll):
(WTF::FastBitVector::setAndCheck):
(WTF::FastBitVector::operator|=):
(WTF::FastBitVector::operator&=):
(WTF::FastBitVector::at):
(WTF::FastBitVector::operator[]):
(WTF::FastBitVector::BitReference::BitReference):
(WTF::FastBitVector::BitReference::operator bool):
(WTF::FastBitVector::BitReference::operator=):
(WTF::FastBitVector::~FastBitVector): Deleted.
(WTF::FastBitVector::numBits): Deleted.
(WTF::FastBitVector::set): Deleted.
(WTF::FastBitVector::equals): Deleted.
(WTF::FastBitVector::merge): Deleted.
(WTF::FastBitVector::filter): Deleted.
(WTF::FastBitVector::exclude): Deleted.
(WTF::FastBitVector::clear): Deleted.
(WTF::FastBitVector::get): Deleted.
(WTF::FastBitVector::bitCount): Deleted.
(WTF::FastBitVector::forEachSetBit): Deleted.
(WTF::FastBitVector::arrayLength): Deleted.
* wtf/StdLibExtras.h:
(WTF::findBitInWord):
Canonical link: https://commits.webkit.org/180007@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@205794 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2016-09-12 04:03:36 +00:00
|
|
|
m_scratch[i] = true;
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Mark all nodes as dominated by everything.
|
|
|
|
|
for (unsigned i = numBlocks; i-- > 1;) {
|
|
|
|
|
if (!graph.node(i) || !graph.predecessors(graph.node(i)).size())
|
|
|
|
|
m_results[i].clearAll();
|
|
|
|
|
else
|
FastBitVector should have efficient and easy-to-use vector-vector operations
https://bugs.webkit.org/show_bug.cgi?id=161847
Reviewed by Saam Barati.
Source/JavaScriptCore:
Adapt existing users of FastBitVector to the new API.
* bytecode/BytecodeLivenessAnalysis.cpp:
(JSC::BytecodeLivenessAnalysis::computeKills):
(JSC::BytecodeLivenessAnalysis::dumpResults):
* bytecode/BytecodeLivenessAnalysisInlines.h:
(JSC::operandThatIsNotAlwaysLiveIsLive):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::stepOverInstruction):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::runLivenessFixpoint):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::validate):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::flushForTerminal):
* dfg/DFGForAllKills.h:
(JSC::DFG::forAllKilledOperands):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::forAllLocalsLiveInBytecode):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::cleanMustHandleValuesIfNecessary):
Source/WTF:
FastBitVector is a bitvector representation that supports manual dynamic resizing and is
optimized for speed, not space. (BitVector supports automatic dynamic resizing and is
optimized for space, while Bitmap is sized statically and is optimized for both speed and
space.) This change greatly increases the power of FastBitVector. We will use these new
powers for changing the JSC GC to use FastBitVectors to track sets of MarkedBlocks (bug
161581) instead of using a combination of Vectors and doubly-linked lists.
This change splits FastBitVector into two parts:
- A thing that manages the storage of a bitvector: a uint32_t array and a size_t numBits.
We call this the word view.
- A thing that takes some kind of abstract array of uint32_t's and does bitvector
operations to it. We call this the FastBitVectorImpl.
FastBitVectorImpl and word views are immutable. The FastBitVector class is a subclass of
FastBitVectorImpl specialized on a word view that owns its words and has additional
support for mutable operations.
Doing this allows us to efficiently support things like this without any unnecessary
memory allocation or copying:
FastBitVector a, b, c; // Assume that there is code to initialize these.
a &= b | ~c;
Previously, this kind of operation would not be efficient, because "~c" would have to
create a whole new FastBitVector. But now, it just returns a FastBitVectorImpl whose
underlying word view bitnots (~) its words on the fly. Using template magic, this can get
pretty complex. For example "b | ~c" returns a FastBitVectorImpl that wraps a word view
whose implementation of WordView::word(size_t index) is something like:
uint32_t word(size_t index) { return b.m_words.word(index) | ~c.m_words.word(index); }
FastBitVectorImpl supports all of the fast bulk bitvector operations, like
forEachSetBit(), bitCount(), etc. So, when you say "a &= b | ~c", the actual
implementation is going to run these bit operations on word granularity directly over the
storage inside a, b, c.
The use of operator overloading is worth explaining a bit. Previously, FastBitVector
avoided operator overloading. For example, the &= operation was called filter(). I think
that this was a pretty good approach at the time. I tried using non-operator methods in
this FastBitVector rewrite, but I found it very odd to say things like:
a.filter(b.bitOr(c.bitNot()));
I think that it's harder to see what is going on here, then using operators, because infix
notation is always better.
* WTF.xcodeproj/project.pbxproj:
* wtf/BitVector.h:
(WTF::BitVector::findBitInWord): Deleted.
* wtf/CMakeLists.txt:
* wtf/Dominators.h:
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::pruneDominators):
* wtf/FastBitVector.cpp: Removed.
* wtf/FastBitVector.h:
(WTF::fastBitVectorArrayLength):
(WTF::FastBitVectorWordView::FastBitVectorWordView):
(WTF::FastBitVectorWordView::numBits):
(WTF::FastBitVectorWordView::word):
(WTF::FastBitVectorWordOwner::FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::~FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::view):
(WTF::FastBitVectorWordOwner::operator=):
(WTF::FastBitVectorWordOwner::setAll):
(WTF::FastBitVectorWordOwner::clearAll):
(WTF::FastBitVectorWordOwner::set):
(WTF::FastBitVectorWordOwner::numBits):
(WTF::FastBitVectorWordOwner::arrayLength):
(WTF::FastBitVectorWordOwner::resize):
(WTF::FastBitVectorWordOwner::word):
(WTF::FastBitVectorWordOwner::words):
(WTF::FastBitVectorAndWords::FastBitVectorAndWords):
(WTF::FastBitVectorAndWords::view):
(WTF::FastBitVectorAndWords::numBits):
(WTF::FastBitVectorAndWords::word):
(WTF::FastBitVectorOrWords::FastBitVectorOrWords):
(WTF::FastBitVectorOrWords::view):
(WTF::FastBitVectorOrWords::numBits):
(WTF::FastBitVectorOrWords::word):
(WTF::FastBitVectorNotWords::FastBitVectorNotWords):
(WTF::FastBitVectorNotWords::view):
(WTF::FastBitVectorNotWords::numBits):
(WTF::FastBitVectorNotWords::word):
(WTF::FastBitVectorImpl::FastBitVectorImpl):
(WTF::FastBitVectorImpl::numBits):
(WTF::FastBitVectorImpl::size):
(WTF::FastBitVectorImpl::arrayLength):
(WTF::FastBitVectorImpl::operator==):
(WTF::FastBitVectorImpl::operator!=):
(WTF::FastBitVectorImpl::at):
(WTF::FastBitVectorImpl::operator[]):
(WTF::FastBitVectorImpl::bitCount):
(WTF::FastBitVectorImpl::operator&):
(WTF::FastBitVectorImpl::operator|):
(WTF::FastBitVectorImpl::operator~):
(WTF::FastBitVectorImpl::forEachSetBit):
(WTF::FastBitVectorImpl::forEachClearBit):
(WTF::FastBitVectorImpl::forEachBit):
(WTF::FastBitVectorImpl::findBit):
(WTF::FastBitVectorImpl::findSetBit):
(WTF::FastBitVectorImpl::findClearBit):
(WTF::FastBitVectorImpl::dump):
(WTF::FastBitVectorImpl::atImpl):
(WTF::FastBitVector::FastBitVector):
(WTF::FastBitVector::operator=):
(WTF::FastBitVector::resize):
(WTF::FastBitVector::setAll):
(WTF::FastBitVector::clearAll):
(WTF::FastBitVector::setAndCheck):
(WTF::FastBitVector::operator|=):
(WTF::FastBitVector::operator&=):
(WTF::FastBitVector::at):
(WTF::FastBitVector::operator[]):
(WTF::FastBitVector::BitReference::BitReference):
(WTF::FastBitVector::BitReference::operator bool):
(WTF::FastBitVector::BitReference::operator=):
(WTF::FastBitVector::~FastBitVector): Deleted.
(WTF::FastBitVector::numBits): Deleted.
(WTF::FastBitVector::set): Deleted.
(WTF::FastBitVector::equals): Deleted.
(WTF::FastBitVector::merge): Deleted.
(WTF::FastBitVector::filter): Deleted.
(WTF::FastBitVector::exclude): Deleted.
(WTF::FastBitVector::clear): Deleted.
(WTF::FastBitVector::get): Deleted.
(WTF::FastBitVector::bitCount): Deleted.
(WTF::FastBitVector::forEachSetBit): Deleted.
(WTF::FastBitVector::arrayLength): Deleted.
* wtf/StdLibExtras.h:
(WTF::findBitInWord):
Canonical link: https://commits.webkit.org/180007@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@205794 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2016-09-12 04:03:36 +00:00
|
|
|
m_results[i] = m_scratch;
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Iteratively eliminate nodes that are not dominator.
|
|
|
|
|
bool changed;
|
|
|
|
|
do {
|
|
|
|
|
changed = false;
|
|
|
|
|
// Prune dominators in all non entry blocks: forward scan.
|
|
|
|
|
for (unsigned i = 1; i < numBlocks; ++i)
|
|
|
|
|
changed |= pruneDominators(i);
|
|
|
|
|
|
|
|
|
|
if (!changed)
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
// Prune dominators in all non entry blocks: backward scan.
|
|
|
|
|
changed = false;
|
|
|
|
|
for (unsigned i = numBlocks; i-- > 1;)
|
|
|
|
|
changed |= pruneDominators(i);
|
|
|
|
|
} while (changed);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
bool dominates(unsigned from, unsigned to) const
|
|
|
|
|
{
|
FastBitVector should have efficient and easy-to-use vector-vector operations
https://bugs.webkit.org/show_bug.cgi?id=161847
Reviewed by Saam Barati.
Source/JavaScriptCore:
Adapt existing users of FastBitVector to the new API.
* bytecode/BytecodeLivenessAnalysis.cpp:
(JSC::BytecodeLivenessAnalysis::computeKills):
(JSC::BytecodeLivenessAnalysis::dumpResults):
* bytecode/BytecodeLivenessAnalysisInlines.h:
(JSC::operandThatIsNotAlwaysLiveIsLive):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::stepOverInstruction):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::runLivenessFixpoint):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::validate):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::flushForTerminal):
* dfg/DFGForAllKills.h:
(JSC::DFG::forAllKilledOperands):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::forAllLocalsLiveInBytecode):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::cleanMustHandleValuesIfNecessary):
Source/WTF:
FastBitVector is a bitvector representation that supports manual dynamic resizing and is
optimized for speed, not space. (BitVector supports automatic dynamic resizing and is
optimized for space, while Bitmap is sized statically and is optimized for both speed and
space.) This change greatly increases the power of FastBitVector. We will use these new
powers for changing the JSC GC to use FastBitVectors to track sets of MarkedBlocks (bug
161581) instead of using a combination of Vectors and doubly-linked lists.
This change splits FastBitVector into two parts:
- A thing that manages the storage of a bitvector: a uint32_t array and a size_t numBits.
We call this the word view.
- A thing that takes some kind of abstract array of uint32_t's and does bitvector
operations to it. We call this the FastBitVectorImpl.
FastBitVectorImpl and word views are immutable. The FastBitVector class is a subclass of
FastBitVectorImpl specialized on a word view that owns its words and has additional
support for mutable operations.
Doing this allows us to efficiently support things like this without any unnecessary
memory allocation or copying:
FastBitVector a, b, c; // Assume that there is code to initialize these.
a &= b | ~c;
Previously, this kind of operation would not be efficient, because "~c" would have to
create a whole new FastBitVector. But now, it just returns a FastBitVectorImpl whose
underlying word view bitnots (~) its words on the fly. Using template magic, this can get
pretty complex. For example "b | ~c" returns a FastBitVectorImpl that wraps a word view
whose implementation of WordView::word(size_t index) is something like:
uint32_t word(size_t index) { return b.m_words.word(index) | ~c.m_words.word(index); }
FastBitVectorImpl supports all of the fast bulk bitvector operations, like
forEachSetBit(), bitCount(), etc. So, when you say "a &= b | ~c", the actual
implementation is going to run these bit operations on word granularity directly over the
storage inside a, b, c.
The use of operator overloading is worth explaining a bit. Previously, FastBitVector
avoided operator overloading. For example, the &= operation was called filter(). I think
that this was a pretty good approach at the time. I tried using non-operator methods in
this FastBitVector rewrite, but I found it very odd to say things like:
a.filter(b.bitOr(c.bitNot()));
I think that it's harder to see what is going on here, then using operators, because infix
notation is always better.
* WTF.xcodeproj/project.pbxproj:
* wtf/BitVector.h:
(WTF::BitVector::findBitInWord): Deleted.
* wtf/CMakeLists.txt:
* wtf/Dominators.h:
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::pruneDominators):
* wtf/FastBitVector.cpp: Removed.
* wtf/FastBitVector.h:
(WTF::fastBitVectorArrayLength):
(WTF::FastBitVectorWordView::FastBitVectorWordView):
(WTF::FastBitVectorWordView::numBits):
(WTF::FastBitVectorWordView::word):
(WTF::FastBitVectorWordOwner::FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::~FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::view):
(WTF::FastBitVectorWordOwner::operator=):
(WTF::FastBitVectorWordOwner::setAll):
(WTF::FastBitVectorWordOwner::clearAll):
(WTF::FastBitVectorWordOwner::set):
(WTF::FastBitVectorWordOwner::numBits):
(WTF::FastBitVectorWordOwner::arrayLength):
(WTF::FastBitVectorWordOwner::resize):
(WTF::FastBitVectorWordOwner::word):
(WTF::FastBitVectorWordOwner::words):
(WTF::FastBitVectorAndWords::FastBitVectorAndWords):
(WTF::FastBitVectorAndWords::view):
(WTF::FastBitVectorAndWords::numBits):
(WTF::FastBitVectorAndWords::word):
(WTF::FastBitVectorOrWords::FastBitVectorOrWords):
(WTF::FastBitVectorOrWords::view):
(WTF::FastBitVectorOrWords::numBits):
(WTF::FastBitVectorOrWords::word):
(WTF::FastBitVectorNotWords::FastBitVectorNotWords):
(WTF::FastBitVectorNotWords::view):
(WTF::FastBitVectorNotWords::numBits):
(WTF::FastBitVectorNotWords::word):
(WTF::FastBitVectorImpl::FastBitVectorImpl):
(WTF::FastBitVectorImpl::numBits):
(WTF::FastBitVectorImpl::size):
(WTF::FastBitVectorImpl::arrayLength):
(WTF::FastBitVectorImpl::operator==):
(WTF::FastBitVectorImpl::operator!=):
(WTF::FastBitVectorImpl::at):
(WTF::FastBitVectorImpl::operator[]):
(WTF::FastBitVectorImpl::bitCount):
(WTF::FastBitVectorImpl::operator&):
(WTF::FastBitVectorImpl::operator|):
(WTF::FastBitVectorImpl::operator~):
(WTF::FastBitVectorImpl::forEachSetBit):
(WTF::FastBitVectorImpl::forEachClearBit):
(WTF::FastBitVectorImpl::forEachBit):
(WTF::FastBitVectorImpl::findBit):
(WTF::FastBitVectorImpl::findSetBit):
(WTF::FastBitVectorImpl::findClearBit):
(WTF::FastBitVectorImpl::dump):
(WTF::FastBitVectorImpl::atImpl):
(WTF::FastBitVector::FastBitVector):
(WTF::FastBitVector::operator=):
(WTF::FastBitVector::resize):
(WTF::FastBitVector::setAll):
(WTF::FastBitVector::clearAll):
(WTF::FastBitVector::setAndCheck):
(WTF::FastBitVector::operator|=):
(WTF::FastBitVector::operator&=):
(WTF::FastBitVector::at):
(WTF::FastBitVector::operator[]):
(WTF::FastBitVector::BitReference::BitReference):
(WTF::FastBitVector::BitReference::operator bool):
(WTF::FastBitVector::BitReference::operator=):
(WTF::FastBitVector::~FastBitVector): Deleted.
(WTF::FastBitVector::numBits): Deleted.
(WTF::FastBitVector::set): Deleted.
(WTF::FastBitVector::equals): Deleted.
(WTF::FastBitVector::merge): Deleted.
(WTF::FastBitVector::filter): Deleted.
(WTF::FastBitVector::exclude): Deleted.
(WTF::FastBitVector::clear): Deleted.
(WTF::FastBitVector::get): Deleted.
(WTF::FastBitVector::bitCount): Deleted.
(WTF::FastBitVector::forEachSetBit): Deleted.
(WTF::FastBitVector::arrayLength): Deleted.
* wtf/StdLibExtras.h:
(WTF::findBitInWord):
Canonical link: https://commits.webkit.org/180007@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@205794 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2016-09-12 04:03:36 +00:00
|
|
|
return m_results[to][from];
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
bool dominates(typename Graph::Node from, typename Graph::Node to) const
|
|
|
|
|
{
|
FTL B3 should be able to flag the tag constants as being super important so that B3 can hoist them and Air can force them into registers
https://bugs.webkit.org/show_bug.cgi?id=151955
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Taught B3 about the concept of "fast constants". A client of B3 can now tell B3 which
constants are super important. B3 will not spill the constant in that case and will ensure
that the constant is materialized only once: statically once, and dynamically once per
procedure execution. The hoistFastConstants() algorithm in B3MoveConstants.cpp achieves this
by first picking the lowest common dominator of all uses of each fast constant, and then
picking the materialization point by finding the lowest dominator of that dominator that is
tied for lowest block frequency. In practice, the second step ensures that this is the lowest
point in the program that is not in a loop (i.e. executes no more than once dynamically per
procedure invocation).
Taught Air about the concept of "fast tmps". B3 tells Air that a tmp is fast if it is used to
hold the materialization of a fast constant. IRC will use the lowest possible spill score for
fast tmps. In practice, this ensures that fast constants are never spilled.
Added a small snippet of code to FTL::LowerDFGToLLVM that makes both of the tag constants
into fast constants.
My hope is that this very brute-force heuristic is good enough that we don't have to think
about constants for a while. Based on my experience with how LLVM's constant hoisting works
out, the heuristic in this patch is going to be tough to beat. LLVM's constant hoisting does
good things when it hoists the tags, and usually causes nothing but problems when it hoists
anything else. This is because there is no way a low-level compiler to really understand how
a constant materialization impacts some operation's contribution to the overall execution
time of a procedure. But, in the FTL we know that constant materializations for type checks
are a bummer because we are super comfortable placing type checks on the hottest of paths. So
those are the last paths where extra instructions should be added by the compiler. On the
other hand, all other large constant uses are on relatively cold paths, or paths that are
already expensive for other reasons. For example, global variable accesses have to
materialize a pointer to the global. But that's not really a big deal, since a load from a
global involves first the load itself and then type checks on the result - so probably the
constant materialization is just not interesting. A store to a global often involves a store
barrier, so the constant materialization is really not interesting. This patch codifies this
heuristic in a pact between Air, B3, and the FTL: FTL demands that B3 pin the two tags in
registers, and B3 relays the demand to Air.
* JavaScriptCore.xcodeproj/project.pbxproj:
* b3/B3CFG.h: Added.
(JSC::B3::CFG::CFG):
(JSC::B3::CFG::root):
(JSC::B3::CFG::newMap):
(JSC::B3::CFG::successors):
(JSC::B3::CFG::predecessors):
(JSC::B3::CFG::index):
(JSC::B3::CFG::node):
(JSC::B3::CFG::numNodes):
(JSC::B3::CFG::dump):
* b3/B3Dominators.h: Added.
(JSC::B3::Dominators::Dominators):
* b3/B3IndexMap.h:
(JSC::B3::IndexMap::resize):
(JSC::B3::IndexMap::size):
(JSC::B3::IndexMap::operator[]):
* b3/B3LowerMacros.cpp:
* b3/B3LowerToAir.cpp:
(JSC::B3::Air::LowerToAir::tmp):
* b3/B3MoveConstants.cpp:
* b3/B3Opcode.h:
(JSC::B3::constPtrOpcode):
(JSC::B3::isConstant):
* b3/B3Procedure.cpp:
(JSC::B3::Procedure::Procedure):
(JSC::B3::Procedure::resetReachability):
(JSC::B3::Procedure::invalidateCFG):
(JSC::B3::Procedure::dump):
(JSC::B3::Procedure::deleteValue):
(JSC::B3::Procedure::dominators):
(JSC::B3::Procedure::addFastConstant):
(JSC::B3::Procedure::isFastConstant):
(JSC::B3::Procedure::addDataSection):
* b3/B3Procedure.h:
(JSC::B3::Procedure::size):
(JSC::B3::Procedure::cfg):
(JSC::B3::Procedure::setLastPhaseName):
* b3/B3ReduceStrength.cpp:
* b3/B3ValueInlines.h:
(JSC::B3::Value::isConstant):
(JSC::B3::Value::isInteger):
* b3/B3ValueKey.h:
(JSC::B3::ValueKey::canMaterialize):
(JSC::B3::ValueKey::isConstant):
* b3/air/AirCode.cpp:
(JSC::B3::Air::Code::findNextBlock):
(JSC::B3::Air::Code::addFastTmp):
* b3/air/AirCode.h:
(JSC::B3::Air::Code::specials):
(JSC::B3::Air::Code::isFastTmp):
(JSC::B3::Air::Code::setLastPhaseName):
* b3/air/AirIteratedRegisterCoalescing.cpp:
* dfg/DFGDominators.h:
* dfg/DFGSSACalculator.cpp:
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
Source/WTF:
Remove some remaining DFG-specific snippets from Dominators. This used to be a non-template
DFG class, and some time ago I hoisted it into WTF and made it generic. But since the only
user of the class was the DFG, this class still had a handful of DFG-specific snippets that
didn't compile when I started using it from B3.
Also renamed immediateDominatorOf() to idom(). This is the sort of abbreviation that we
wouldn't ordinarily want to have in WebKit. But WebKit does allow for abbreviations that are
"more canonical". The term "idom" is definitely more canonical than "immediateDominatorOf".
* wtf/Dominators.h:
(WTF::Dominators::dominates):
(WTF::Dominators::idom):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::ValidationContext::handleErrors):
Canonical link: https://commits.webkit.org/170119@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@193682 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-12-08 02:46:22 +00:00
|
|
|
return dominates(m_graph.index(from), m_graph.index(to));
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void dump(PrintStream& out) const
|
|
|
|
|
{
|
|
|
|
|
for (unsigned blockIndex = 0; blockIndex < m_graph.numNodes(); ++blockIndex) {
|
|
|
|
|
typename Graph::Node block = m_graph.node(blockIndex);
|
|
|
|
|
if (!block)
|
|
|
|
|
continue;
|
|
|
|
|
out.print(" Block ", m_graph.dump(block), ":");
|
|
|
|
|
for (unsigned otherIndex = 0; otherIndex < m_graph.numNodes(); ++otherIndex) {
|
|
|
|
|
if (!dominates(m_graph.index(block), otherIndex))
|
|
|
|
|
continue;
|
|
|
|
|
out.print(" ", m_graph.dump(m_graph.node(otherIndex)));
|
|
|
|
|
}
|
|
|
|
|
out.print("\n");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
bool pruneDominators(unsigned idx)
|
|
|
|
|
{
|
|
|
|
|
typename Graph::Node block = m_graph.node(idx);
|
|
|
|
|
|
|
|
|
|
if (!block || !m_graph.predecessors(block).size())
|
|
|
|
|
return false;
|
|
|
|
|
|
|
|
|
|
// Find the intersection of dom(preds).
|
FastBitVector should have efficient and easy-to-use vector-vector operations
https://bugs.webkit.org/show_bug.cgi?id=161847
Reviewed by Saam Barati.
Source/JavaScriptCore:
Adapt existing users of FastBitVector to the new API.
* bytecode/BytecodeLivenessAnalysis.cpp:
(JSC::BytecodeLivenessAnalysis::computeKills):
(JSC::BytecodeLivenessAnalysis::dumpResults):
* bytecode/BytecodeLivenessAnalysisInlines.h:
(JSC::operandThatIsNotAlwaysLiveIsLive):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::stepOverInstruction):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::runLivenessFixpoint):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::validate):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::flushForTerminal):
* dfg/DFGForAllKills.h:
(JSC::DFG::forAllKilledOperands):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::forAllLocalsLiveInBytecode):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::cleanMustHandleValuesIfNecessary):
Source/WTF:
FastBitVector is a bitvector representation that supports manual dynamic resizing and is
optimized for speed, not space. (BitVector supports automatic dynamic resizing and is
optimized for space, while Bitmap is sized statically and is optimized for both speed and
space.) This change greatly increases the power of FastBitVector. We will use these new
powers for changing the JSC GC to use FastBitVectors to track sets of MarkedBlocks (bug
161581) instead of using a combination of Vectors and doubly-linked lists.
This change splits FastBitVector into two parts:
- A thing that manages the storage of a bitvector: a uint32_t array and a size_t numBits.
We call this the word view.
- A thing that takes some kind of abstract array of uint32_t's and does bitvector
operations to it. We call this the FastBitVectorImpl.
FastBitVectorImpl and word views are immutable. The FastBitVector class is a subclass of
FastBitVectorImpl specialized on a word view that owns its words and has additional
support for mutable operations.
Doing this allows us to efficiently support things like this without any unnecessary
memory allocation or copying:
FastBitVector a, b, c; // Assume that there is code to initialize these.
a &= b | ~c;
Previously, this kind of operation would not be efficient, because "~c" would have to
create a whole new FastBitVector. But now, it just returns a FastBitVectorImpl whose
underlying word view bitnots (~) its words on the fly. Using template magic, this can get
pretty complex. For example "b | ~c" returns a FastBitVectorImpl that wraps a word view
whose implementation of WordView::word(size_t index) is something like:
uint32_t word(size_t index) { return b.m_words.word(index) | ~c.m_words.word(index); }
FastBitVectorImpl supports all of the fast bulk bitvector operations, like
forEachSetBit(), bitCount(), etc. So, when you say "a &= b | ~c", the actual
implementation is going to run these bit operations on word granularity directly over the
storage inside a, b, c.
The use of operator overloading is worth explaining a bit. Previously, FastBitVector
avoided operator overloading. For example, the &= operation was called filter(). I think
that this was a pretty good approach at the time. I tried using non-operator methods in
this FastBitVector rewrite, but I found it very odd to say things like:
a.filter(b.bitOr(c.bitNot()));
I think that it's harder to see what is going on here, then using operators, because infix
notation is always better.
* WTF.xcodeproj/project.pbxproj:
* wtf/BitVector.h:
(WTF::BitVector::findBitInWord): Deleted.
* wtf/CMakeLists.txt:
* wtf/Dominators.h:
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::pruneDominators):
* wtf/FastBitVector.cpp: Removed.
* wtf/FastBitVector.h:
(WTF::fastBitVectorArrayLength):
(WTF::FastBitVectorWordView::FastBitVectorWordView):
(WTF::FastBitVectorWordView::numBits):
(WTF::FastBitVectorWordView::word):
(WTF::FastBitVectorWordOwner::FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::~FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::view):
(WTF::FastBitVectorWordOwner::operator=):
(WTF::FastBitVectorWordOwner::setAll):
(WTF::FastBitVectorWordOwner::clearAll):
(WTF::FastBitVectorWordOwner::set):
(WTF::FastBitVectorWordOwner::numBits):
(WTF::FastBitVectorWordOwner::arrayLength):
(WTF::FastBitVectorWordOwner::resize):
(WTF::FastBitVectorWordOwner::word):
(WTF::FastBitVectorWordOwner::words):
(WTF::FastBitVectorAndWords::FastBitVectorAndWords):
(WTF::FastBitVectorAndWords::view):
(WTF::FastBitVectorAndWords::numBits):
(WTF::FastBitVectorAndWords::word):
(WTF::FastBitVectorOrWords::FastBitVectorOrWords):
(WTF::FastBitVectorOrWords::view):
(WTF::FastBitVectorOrWords::numBits):
(WTF::FastBitVectorOrWords::word):
(WTF::FastBitVectorNotWords::FastBitVectorNotWords):
(WTF::FastBitVectorNotWords::view):
(WTF::FastBitVectorNotWords::numBits):
(WTF::FastBitVectorNotWords::word):
(WTF::FastBitVectorImpl::FastBitVectorImpl):
(WTF::FastBitVectorImpl::numBits):
(WTF::FastBitVectorImpl::size):
(WTF::FastBitVectorImpl::arrayLength):
(WTF::FastBitVectorImpl::operator==):
(WTF::FastBitVectorImpl::operator!=):
(WTF::FastBitVectorImpl::at):
(WTF::FastBitVectorImpl::operator[]):
(WTF::FastBitVectorImpl::bitCount):
(WTF::FastBitVectorImpl::operator&):
(WTF::FastBitVectorImpl::operator|):
(WTF::FastBitVectorImpl::operator~):
(WTF::FastBitVectorImpl::forEachSetBit):
(WTF::FastBitVectorImpl::forEachClearBit):
(WTF::FastBitVectorImpl::forEachBit):
(WTF::FastBitVectorImpl::findBit):
(WTF::FastBitVectorImpl::findSetBit):
(WTF::FastBitVectorImpl::findClearBit):
(WTF::FastBitVectorImpl::dump):
(WTF::FastBitVectorImpl::atImpl):
(WTF::FastBitVector::FastBitVector):
(WTF::FastBitVector::operator=):
(WTF::FastBitVector::resize):
(WTF::FastBitVector::setAll):
(WTF::FastBitVector::clearAll):
(WTF::FastBitVector::setAndCheck):
(WTF::FastBitVector::operator|=):
(WTF::FastBitVector::operator&=):
(WTF::FastBitVector::at):
(WTF::FastBitVector::operator[]):
(WTF::FastBitVector::BitReference::BitReference):
(WTF::FastBitVector::BitReference::operator bool):
(WTF::FastBitVector::BitReference::operator=):
(WTF::FastBitVector::~FastBitVector): Deleted.
(WTF::FastBitVector::numBits): Deleted.
(WTF::FastBitVector::set): Deleted.
(WTF::FastBitVector::equals): Deleted.
(WTF::FastBitVector::merge): Deleted.
(WTF::FastBitVector::filter): Deleted.
(WTF::FastBitVector::exclude): Deleted.
(WTF::FastBitVector::clear): Deleted.
(WTF::FastBitVector::get): Deleted.
(WTF::FastBitVector::bitCount): Deleted.
(WTF::FastBitVector::forEachSetBit): Deleted.
(WTF::FastBitVector::arrayLength): Deleted.
* wtf/StdLibExtras.h:
(WTF::findBitInWord):
Canonical link: https://commits.webkit.org/180007@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@205794 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2016-09-12 04:03:36 +00:00
|
|
|
m_scratch = m_results[m_graph.index(m_graph.predecessors(block)[0])];
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
for (unsigned j = m_graph.predecessors(block).size(); j-- > 1;)
|
FastBitVector should have efficient and easy-to-use vector-vector operations
https://bugs.webkit.org/show_bug.cgi?id=161847
Reviewed by Saam Barati.
Source/JavaScriptCore:
Adapt existing users of FastBitVector to the new API.
* bytecode/BytecodeLivenessAnalysis.cpp:
(JSC::BytecodeLivenessAnalysis::computeKills):
(JSC::BytecodeLivenessAnalysis::dumpResults):
* bytecode/BytecodeLivenessAnalysisInlines.h:
(JSC::operandThatIsNotAlwaysLiveIsLive):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::stepOverInstruction):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::runLivenessFixpoint):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::validate):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::flushForTerminal):
* dfg/DFGForAllKills.h:
(JSC::DFG::forAllKilledOperands):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::forAllLocalsLiveInBytecode):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::cleanMustHandleValuesIfNecessary):
Source/WTF:
FastBitVector is a bitvector representation that supports manual dynamic resizing and is
optimized for speed, not space. (BitVector supports automatic dynamic resizing and is
optimized for space, while Bitmap is sized statically and is optimized for both speed and
space.) This change greatly increases the power of FastBitVector. We will use these new
powers for changing the JSC GC to use FastBitVectors to track sets of MarkedBlocks (bug
161581) instead of using a combination of Vectors and doubly-linked lists.
This change splits FastBitVector into two parts:
- A thing that manages the storage of a bitvector: a uint32_t array and a size_t numBits.
We call this the word view.
- A thing that takes some kind of abstract array of uint32_t's and does bitvector
operations to it. We call this the FastBitVectorImpl.
FastBitVectorImpl and word views are immutable. The FastBitVector class is a subclass of
FastBitVectorImpl specialized on a word view that owns its words and has additional
support for mutable operations.
Doing this allows us to efficiently support things like this without any unnecessary
memory allocation or copying:
FastBitVector a, b, c; // Assume that there is code to initialize these.
a &= b | ~c;
Previously, this kind of operation would not be efficient, because "~c" would have to
create a whole new FastBitVector. But now, it just returns a FastBitVectorImpl whose
underlying word view bitnots (~) its words on the fly. Using template magic, this can get
pretty complex. For example "b | ~c" returns a FastBitVectorImpl that wraps a word view
whose implementation of WordView::word(size_t index) is something like:
uint32_t word(size_t index) { return b.m_words.word(index) | ~c.m_words.word(index); }
FastBitVectorImpl supports all of the fast bulk bitvector operations, like
forEachSetBit(), bitCount(), etc. So, when you say "a &= b | ~c", the actual
implementation is going to run these bit operations on word granularity directly over the
storage inside a, b, c.
The use of operator overloading is worth explaining a bit. Previously, FastBitVector
avoided operator overloading. For example, the &= operation was called filter(). I think
that this was a pretty good approach at the time. I tried using non-operator methods in
this FastBitVector rewrite, but I found it very odd to say things like:
a.filter(b.bitOr(c.bitNot()));
I think that it's harder to see what is going on here, then using operators, because infix
notation is always better.
* WTF.xcodeproj/project.pbxproj:
* wtf/BitVector.h:
(WTF::BitVector::findBitInWord): Deleted.
* wtf/CMakeLists.txt:
* wtf/Dominators.h:
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::pruneDominators):
* wtf/FastBitVector.cpp: Removed.
* wtf/FastBitVector.h:
(WTF::fastBitVectorArrayLength):
(WTF::FastBitVectorWordView::FastBitVectorWordView):
(WTF::FastBitVectorWordView::numBits):
(WTF::FastBitVectorWordView::word):
(WTF::FastBitVectorWordOwner::FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::~FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::view):
(WTF::FastBitVectorWordOwner::operator=):
(WTF::FastBitVectorWordOwner::setAll):
(WTF::FastBitVectorWordOwner::clearAll):
(WTF::FastBitVectorWordOwner::set):
(WTF::FastBitVectorWordOwner::numBits):
(WTF::FastBitVectorWordOwner::arrayLength):
(WTF::FastBitVectorWordOwner::resize):
(WTF::FastBitVectorWordOwner::word):
(WTF::FastBitVectorWordOwner::words):
(WTF::FastBitVectorAndWords::FastBitVectorAndWords):
(WTF::FastBitVectorAndWords::view):
(WTF::FastBitVectorAndWords::numBits):
(WTF::FastBitVectorAndWords::word):
(WTF::FastBitVectorOrWords::FastBitVectorOrWords):
(WTF::FastBitVectorOrWords::view):
(WTF::FastBitVectorOrWords::numBits):
(WTF::FastBitVectorOrWords::word):
(WTF::FastBitVectorNotWords::FastBitVectorNotWords):
(WTF::FastBitVectorNotWords::view):
(WTF::FastBitVectorNotWords::numBits):
(WTF::FastBitVectorNotWords::word):
(WTF::FastBitVectorImpl::FastBitVectorImpl):
(WTF::FastBitVectorImpl::numBits):
(WTF::FastBitVectorImpl::size):
(WTF::FastBitVectorImpl::arrayLength):
(WTF::FastBitVectorImpl::operator==):
(WTF::FastBitVectorImpl::operator!=):
(WTF::FastBitVectorImpl::at):
(WTF::FastBitVectorImpl::operator[]):
(WTF::FastBitVectorImpl::bitCount):
(WTF::FastBitVectorImpl::operator&):
(WTF::FastBitVectorImpl::operator|):
(WTF::FastBitVectorImpl::operator~):
(WTF::FastBitVectorImpl::forEachSetBit):
(WTF::FastBitVectorImpl::forEachClearBit):
(WTF::FastBitVectorImpl::forEachBit):
(WTF::FastBitVectorImpl::findBit):
(WTF::FastBitVectorImpl::findSetBit):
(WTF::FastBitVectorImpl::findClearBit):
(WTF::FastBitVectorImpl::dump):
(WTF::FastBitVectorImpl::atImpl):
(WTF::FastBitVector::FastBitVector):
(WTF::FastBitVector::operator=):
(WTF::FastBitVector::resize):
(WTF::FastBitVector::setAll):
(WTF::FastBitVector::clearAll):
(WTF::FastBitVector::setAndCheck):
(WTF::FastBitVector::operator|=):
(WTF::FastBitVector::operator&=):
(WTF::FastBitVector::at):
(WTF::FastBitVector::operator[]):
(WTF::FastBitVector::BitReference::BitReference):
(WTF::FastBitVector::BitReference::operator bool):
(WTF::FastBitVector::BitReference::operator=):
(WTF::FastBitVector::~FastBitVector): Deleted.
(WTF::FastBitVector::numBits): Deleted.
(WTF::FastBitVector::set): Deleted.
(WTF::FastBitVector::equals): Deleted.
(WTF::FastBitVector::merge): Deleted.
(WTF::FastBitVector::filter): Deleted.
(WTF::FastBitVector::exclude): Deleted.
(WTF::FastBitVector::clear): Deleted.
(WTF::FastBitVector::get): Deleted.
(WTF::FastBitVector::bitCount): Deleted.
(WTF::FastBitVector::forEachSetBit): Deleted.
(WTF::FastBitVector::arrayLength): Deleted.
* wtf/StdLibExtras.h:
(WTF::findBitInWord):
Canonical link: https://commits.webkit.org/180007@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@205794 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2016-09-12 04:03:36 +00:00
|
|
|
m_scratch &= m_results[m_graph.index(m_graph.predecessors(block)[j])];
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
|
|
|
|
|
// The block is also dominated by itself.
|
FastBitVector should have efficient and easy-to-use vector-vector operations
https://bugs.webkit.org/show_bug.cgi?id=161847
Reviewed by Saam Barati.
Source/JavaScriptCore:
Adapt existing users of FastBitVector to the new API.
* bytecode/BytecodeLivenessAnalysis.cpp:
(JSC::BytecodeLivenessAnalysis::computeKills):
(JSC::BytecodeLivenessAnalysis::dumpResults):
* bytecode/BytecodeLivenessAnalysisInlines.h:
(JSC::operandThatIsNotAlwaysLiveIsLive):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::stepOverInstruction):
(JSC::BytecodeLivenessPropagation<DerivedAnalysis>::runLivenessFixpoint):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::validate):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::flushForTerminal):
* dfg/DFGForAllKills.h:
(JSC::DFG::forAllKilledOperands):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::forAllLocalsLiveInBytecode):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::cleanMustHandleValuesIfNecessary):
Source/WTF:
FastBitVector is a bitvector representation that supports manual dynamic resizing and is
optimized for speed, not space. (BitVector supports automatic dynamic resizing and is
optimized for space, while Bitmap is sized statically and is optimized for both speed and
space.) This change greatly increases the power of FastBitVector. We will use these new
powers for changing the JSC GC to use FastBitVectors to track sets of MarkedBlocks (bug
161581) instead of using a combination of Vectors and doubly-linked lists.
This change splits FastBitVector into two parts:
- A thing that manages the storage of a bitvector: a uint32_t array and a size_t numBits.
We call this the word view.
- A thing that takes some kind of abstract array of uint32_t's and does bitvector
operations to it. We call this the FastBitVectorImpl.
FastBitVectorImpl and word views are immutable. The FastBitVector class is a subclass of
FastBitVectorImpl specialized on a word view that owns its words and has additional
support for mutable operations.
Doing this allows us to efficiently support things like this without any unnecessary
memory allocation or copying:
FastBitVector a, b, c; // Assume that there is code to initialize these.
a &= b | ~c;
Previously, this kind of operation would not be efficient, because "~c" would have to
create a whole new FastBitVector. But now, it just returns a FastBitVectorImpl whose
underlying word view bitnots (~) its words on the fly. Using template magic, this can get
pretty complex. For example "b | ~c" returns a FastBitVectorImpl that wraps a word view
whose implementation of WordView::word(size_t index) is something like:
uint32_t word(size_t index) { return b.m_words.word(index) | ~c.m_words.word(index); }
FastBitVectorImpl supports all of the fast bulk bitvector operations, like
forEachSetBit(), bitCount(), etc. So, when you say "a &= b | ~c", the actual
implementation is going to run these bit operations on word granularity directly over the
storage inside a, b, c.
The use of operator overloading is worth explaining a bit. Previously, FastBitVector
avoided operator overloading. For example, the &= operation was called filter(). I think
that this was a pretty good approach at the time. I tried using non-operator methods in
this FastBitVector rewrite, but I found it very odd to say things like:
a.filter(b.bitOr(c.bitNot()));
I think that it's harder to see what is going on here, then using operators, because infix
notation is always better.
* WTF.xcodeproj/project.pbxproj:
* wtf/BitVector.h:
(WTF::BitVector::findBitInWord): Deleted.
* wtf/CMakeLists.txt:
* wtf/Dominators.h:
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::pruneDominators):
* wtf/FastBitVector.cpp: Removed.
* wtf/FastBitVector.h:
(WTF::fastBitVectorArrayLength):
(WTF::FastBitVectorWordView::FastBitVectorWordView):
(WTF::FastBitVectorWordView::numBits):
(WTF::FastBitVectorWordView::word):
(WTF::FastBitVectorWordOwner::FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::~FastBitVectorWordOwner):
(WTF::FastBitVectorWordOwner::view):
(WTF::FastBitVectorWordOwner::operator=):
(WTF::FastBitVectorWordOwner::setAll):
(WTF::FastBitVectorWordOwner::clearAll):
(WTF::FastBitVectorWordOwner::set):
(WTF::FastBitVectorWordOwner::numBits):
(WTF::FastBitVectorWordOwner::arrayLength):
(WTF::FastBitVectorWordOwner::resize):
(WTF::FastBitVectorWordOwner::word):
(WTF::FastBitVectorWordOwner::words):
(WTF::FastBitVectorAndWords::FastBitVectorAndWords):
(WTF::FastBitVectorAndWords::view):
(WTF::FastBitVectorAndWords::numBits):
(WTF::FastBitVectorAndWords::word):
(WTF::FastBitVectorOrWords::FastBitVectorOrWords):
(WTF::FastBitVectorOrWords::view):
(WTF::FastBitVectorOrWords::numBits):
(WTF::FastBitVectorOrWords::word):
(WTF::FastBitVectorNotWords::FastBitVectorNotWords):
(WTF::FastBitVectorNotWords::view):
(WTF::FastBitVectorNotWords::numBits):
(WTF::FastBitVectorNotWords::word):
(WTF::FastBitVectorImpl::FastBitVectorImpl):
(WTF::FastBitVectorImpl::numBits):
(WTF::FastBitVectorImpl::size):
(WTF::FastBitVectorImpl::arrayLength):
(WTF::FastBitVectorImpl::operator==):
(WTF::FastBitVectorImpl::operator!=):
(WTF::FastBitVectorImpl::at):
(WTF::FastBitVectorImpl::operator[]):
(WTF::FastBitVectorImpl::bitCount):
(WTF::FastBitVectorImpl::operator&):
(WTF::FastBitVectorImpl::operator|):
(WTF::FastBitVectorImpl::operator~):
(WTF::FastBitVectorImpl::forEachSetBit):
(WTF::FastBitVectorImpl::forEachClearBit):
(WTF::FastBitVectorImpl::forEachBit):
(WTF::FastBitVectorImpl::findBit):
(WTF::FastBitVectorImpl::findSetBit):
(WTF::FastBitVectorImpl::findClearBit):
(WTF::FastBitVectorImpl::dump):
(WTF::FastBitVectorImpl::atImpl):
(WTF::FastBitVector::FastBitVector):
(WTF::FastBitVector::operator=):
(WTF::FastBitVector::resize):
(WTF::FastBitVector::setAll):
(WTF::FastBitVector::clearAll):
(WTF::FastBitVector::setAndCheck):
(WTF::FastBitVector::operator|=):
(WTF::FastBitVector::operator&=):
(WTF::FastBitVector::at):
(WTF::FastBitVector::operator[]):
(WTF::FastBitVector::BitReference::BitReference):
(WTF::FastBitVector::BitReference::operator bool):
(WTF::FastBitVector::BitReference::operator=):
(WTF::FastBitVector::~FastBitVector): Deleted.
(WTF::FastBitVector::numBits): Deleted.
(WTF::FastBitVector::set): Deleted.
(WTF::FastBitVector::equals): Deleted.
(WTF::FastBitVector::merge): Deleted.
(WTF::FastBitVector::filter): Deleted.
(WTF::FastBitVector::exclude): Deleted.
(WTF::FastBitVector::clear): Deleted.
(WTF::FastBitVector::get): Deleted.
(WTF::FastBitVector::bitCount): Deleted.
(WTF::FastBitVector::forEachSetBit): Deleted.
(WTF::FastBitVector::arrayLength): Deleted.
* wtf/StdLibExtras.h:
(WTF::findBitInWord):
Canonical link: https://commits.webkit.org/180007@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@205794 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2016-09-12 04:03:36 +00:00
|
|
|
m_scratch[idx] = true;
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
|
|
|
|
|
return m_results[idx].setAndCheck(m_scratch);
|
|
|
|
|
}
|
|
|
|
|
|
FTL B3 should be able to flag the tag constants as being super important so that B3 can hoist them and Air can force them into registers
https://bugs.webkit.org/show_bug.cgi?id=151955
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Taught B3 about the concept of "fast constants". A client of B3 can now tell B3 which
constants are super important. B3 will not spill the constant in that case and will ensure
that the constant is materialized only once: statically once, and dynamically once per
procedure execution. The hoistFastConstants() algorithm in B3MoveConstants.cpp achieves this
by first picking the lowest common dominator of all uses of each fast constant, and then
picking the materialization point by finding the lowest dominator of that dominator that is
tied for lowest block frequency. In practice, the second step ensures that this is the lowest
point in the program that is not in a loop (i.e. executes no more than once dynamically per
procedure invocation).
Taught Air about the concept of "fast tmps". B3 tells Air that a tmp is fast if it is used to
hold the materialization of a fast constant. IRC will use the lowest possible spill score for
fast tmps. In practice, this ensures that fast constants are never spilled.
Added a small snippet of code to FTL::LowerDFGToLLVM that makes both of the tag constants
into fast constants.
My hope is that this very brute-force heuristic is good enough that we don't have to think
about constants for a while. Based on my experience with how LLVM's constant hoisting works
out, the heuristic in this patch is going to be tough to beat. LLVM's constant hoisting does
good things when it hoists the tags, and usually causes nothing but problems when it hoists
anything else. This is because there is no way a low-level compiler to really understand how
a constant materialization impacts some operation's contribution to the overall execution
time of a procedure. But, in the FTL we know that constant materializations for type checks
are a bummer because we are super comfortable placing type checks on the hottest of paths. So
those are the last paths where extra instructions should be added by the compiler. On the
other hand, all other large constant uses are on relatively cold paths, or paths that are
already expensive for other reasons. For example, global variable accesses have to
materialize a pointer to the global. But that's not really a big deal, since a load from a
global involves first the load itself and then type checks on the result - so probably the
constant materialization is just not interesting. A store to a global often involves a store
barrier, so the constant materialization is really not interesting. This patch codifies this
heuristic in a pact between Air, B3, and the FTL: FTL demands that B3 pin the two tags in
registers, and B3 relays the demand to Air.
* JavaScriptCore.xcodeproj/project.pbxproj:
* b3/B3CFG.h: Added.
(JSC::B3::CFG::CFG):
(JSC::B3::CFG::root):
(JSC::B3::CFG::newMap):
(JSC::B3::CFG::successors):
(JSC::B3::CFG::predecessors):
(JSC::B3::CFG::index):
(JSC::B3::CFG::node):
(JSC::B3::CFG::numNodes):
(JSC::B3::CFG::dump):
* b3/B3Dominators.h: Added.
(JSC::B3::Dominators::Dominators):
* b3/B3IndexMap.h:
(JSC::B3::IndexMap::resize):
(JSC::B3::IndexMap::size):
(JSC::B3::IndexMap::operator[]):
* b3/B3LowerMacros.cpp:
* b3/B3LowerToAir.cpp:
(JSC::B3::Air::LowerToAir::tmp):
* b3/B3MoveConstants.cpp:
* b3/B3Opcode.h:
(JSC::B3::constPtrOpcode):
(JSC::B3::isConstant):
* b3/B3Procedure.cpp:
(JSC::B3::Procedure::Procedure):
(JSC::B3::Procedure::resetReachability):
(JSC::B3::Procedure::invalidateCFG):
(JSC::B3::Procedure::dump):
(JSC::B3::Procedure::deleteValue):
(JSC::B3::Procedure::dominators):
(JSC::B3::Procedure::addFastConstant):
(JSC::B3::Procedure::isFastConstant):
(JSC::B3::Procedure::addDataSection):
* b3/B3Procedure.h:
(JSC::B3::Procedure::size):
(JSC::B3::Procedure::cfg):
(JSC::B3::Procedure::setLastPhaseName):
* b3/B3ReduceStrength.cpp:
* b3/B3ValueInlines.h:
(JSC::B3::Value::isConstant):
(JSC::B3::Value::isInteger):
* b3/B3ValueKey.h:
(JSC::B3::ValueKey::canMaterialize):
(JSC::B3::ValueKey::isConstant):
* b3/air/AirCode.cpp:
(JSC::B3::Air::Code::findNextBlock):
(JSC::B3::Air::Code::addFastTmp):
* b3/air/AirCode.h:
(JSC::B3::Air::Code::specials):
(JSC::B3::Air::Code::isFastTmp):
(JSC::B3::Air::Code::setLastPhaseName):
* b3/air/AirIteratedRegisterCoalescing.cpp:
* dfg/DFGDominators.h:
* dfg/DFGSSACalculator.cpp:
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
Source/WTF:
Remove some remaining DFG-specific snippets from Dominators. This used to be a non-template
DFG class, and some time ago I hoisted it into WTF and made it generic. But since the only
user of the class was the DFG, this class still had a handful of DFG-specific snippets that
didn't compile when I started using it from B3.
Also renamed immediateDominatorOf() to idom(). This is the sort of abbreviation that we
wouldn't ordinarily want to have in WebKit. But WebKit does allow for abbreviations that are
"more canonical". The term "idom" is definitely more canonical than "immediateDominatorOf".
* wtf/Dominators.h:
(WTF::Dominators::dominates):
(WTF::Dominators::idom):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::ValidationContext::handleErrors):
Canonical link: https://commits.webkit.org/170119@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@193682 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-12-08 02:46:22 +00:00
|
|
|
Graph& m_graph;
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
Vector<FastBitVector> m_results; // For each block, the bitvector of blocks that dominate it.
|
|
|
|
|
FastBitVector m_scratch; // A temporary bitvector with bit for each block. We recycle this to save new/deletes.
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
struct ValidationContext {
|
2019-08-12 20:57:15 +00:00
|
|
|
WTF_MAKE_STRUCT_FAST_ALLOCATED;
|
|
|
|
|
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
ValidationContext(Graph& graph, Dominators& dominators)
|
|
|
|
|
: graph(graph)
|
|
|
|
|
, dominators(dominators)
|
|
|
|
|
, naiveDominators(graph)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void reportError(typename Graph::Node from, typename Graph::Node to, const char* message)
|
|
|
|
|
{
|
|
|
|
|
Error error;
|
|
|
|
|
error.from = from;
|
|
|
|
|
error.to = to;
|
|
|
|
|
error.message = message;
|
|
|
|
|
errors.append(error);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void handleErrors()
|
|
|
|
|
{
|
|
|
|
|
if (errors.isEmpty())
|
|
|
|
|
return;
|
|
|
|
|
|
|
|
|
|
dataLog("DFG DOMINATOR VALIDATION FAILED:\n");
|
|
|
|
|
dataLog("\n");
|
|
|
|
|
dataLog("For block domination relationships:\n");
|
|
|
|
|
for (unsigned i = 0; i < errors.size(); ++i) {
|
|
|
|
|
dataLog(
|
|
|
|
|
" ", graph.dump(errors[i].from), " -> ", graph.dump(errors[i].to),
|
|
|
|
|
" (", errors[i].message, ")\n");
|
|
|
|
|
}
|
|
|
|
|
dataLog("\n");
|
|
|
|
|
dataLog("Control flow graph:\n");
|
|
|
|
|
for (unsigned blockIndex = 0; blockIndex < graph.numNodes(); ++blockIndex) {
|
|
|
|
|
typename Graph::Node block = graph.node(blockIndex);
|
|
|
|
|
if (!block)
|
|
|
|
|
continue;
|
|
|
|
|
dataLog(" Block ", graph.dump(graph.node(blockIndex)), ": successors = [");
|
|
|
|
|
CommaPrinter comma;
|
FTL B3 should be able to flag the tag constants as being super important so that B3 can hoist them and Air can force them into registers
https://bugs.webkit.org/show_bug.cgi?id=151955
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Taught B3 about the concept of "fast constants". A client of B3 can now tell B3 which
constants are super important. B3 will not spill the constant in that case and will ensure
that the constant is materialized only once: statically once, and dynamically once per
procedure execution. The hoistFastConstants() algorithm in B3MoveConstants.cpp achieves this
by first picking the lowest common dominator of all uses of each fast constant, and then
picking the materialization point by finding the lowest dominator of that dominator that is
tied for lowest block frequency. In practice, the second step ensures that this is the lowest
point in the program that is not in a loop (i.e. executes no more than once dynamically per
procedure invocation).
Taught Air about the concept of "fast tmps". B3 tells Air that a tmp is fast if it is used to
hold the materialization of a fast constant. IRC will use the lowest possible spill score for
fast tmps. In practice, this ensures that fast constants are never spilled.
Added a small snippet of code to FTL::LowerDFGToLLVM that makes both of the tag constants
into fast constants.
My hope is that this very brute-force heuristic is good enough that we don't have to think
about constants for a while. Based on my experience with how LLVM's constant hoisting works
out, the heuristic in this patch is going to be tough to beat. LLVM's constant hoisting does
good things when it hoists the tags, and usually causes nothing but problems when it hoists
anything else. This is because there is no way a low-level compiler to really understand how
a constant materialization impacts some operation's contribution to the overall execution
time of a procedure. But, in the FTL we know that constant materializations for type checks
are a bummer because we are super comfortable placing type checks on the hottest of paths. So
those are the last paths where extra instructions should be added by the compiler. On the
other hand, all other large constant uses are on relatively cold paths, or paths that are
already expensive for other reasons. For example, global variable accesses have to
materialize a pointer to the global. But that's not really a big deal, since a load from a
global involves first the load itself and then type checks on the result - so probably the
constant materialization is just not interesting. A store to a global often involves a store
barrier, so the constant materialization is really not interesting. This patch codifies this
heuristic in a pact between Air, B3, and the FTL: FTL demands that B3 pin the two tags in
registers, and B3 relays the demand to Air.
* JavaScriptCore.xcodeproj/project.pbxproj:
* b3/B3CFG.h: Added.
(JSC::B3::CFG::CFG):
(JSC::B3::CFG::root):
(JSC::B3::CFG::newMap):
(JSC::B3::CFG::successors):
(JSC::B3::CFG::predecessors):
(JSC::B3::CFG::index):
(JSC::B3::CFG::node):
(JSC::B3::CFG::numNodes):
(JSC::B3::CFG::dump):
* b3/B3Dominators.h: Added.
(JSC::B3::Dominators::Dominators):
* b3/B3IndexMap.h:
(JSC::B3::IndexMap::resize):
(JSC::B3::IndexMap::size):
(JSC::B3::IndexMap::operator[]):
* b3/B3LowerMacros.cpp:
* b3/B3LowerToAir.cpp:
(JSC::B3::Air::LowerToAir::tmp):
* b3/B3MoveConstants.cpp:
* b3/B3Opcode.h:
(JSC::B3::constPtrOpcode):
(JSC::B3::isConstant):
* b3/B3Procedure.cpp:
(JSC::B3::Procedure::Procedure):
(JSC::B3::Procedure::resetReachability):
(JSC::B3::Procedure::invalidateCFG):
(JSC::B3::Procedure::dump):
(JSC::B3::Procedure::deleteValue):
(JSC::B3::Procedure::dominators):
(JSC::B3::Procedure::addFastConstant):
(JSC::B3::Procedure::isFastConstant):
(JSC::B3::Procedure::addDataSection):
* b3/B3Procedure.h:
(JSC::B3::Procedure::size):
(JSC::B3::Procedure::cfg):
(JSC::B3::Procedure::setLastPhaseName):
* b3/B3ReduceStrength.cpp:
* b3/B3ValueInlines.h:
(JSC::B3::Value::isConstant):
(JSC::B3::Value::isInteger):
* b3/B3ValueKey.h:
(JSC::B3::ValueKey::canMaterialize):
(JSC::B3::ValueKey::isConstant):
* b3/air/AirCode.cpp:
(JSC::B3::Air::Code::findNextBlock):
(JSC::B3::Air::Code::addFastTmp):
* b3/air/AirCode.h:
(JSC::B3::Air::Code::specials):
(JSC::B3::Air::Code::isFastTmp):
(JSC::B3::Air::Code::setLastPhaseName):
* b3/air/AirIteratedRegisterCoalescing.cpp:
* dfg/DFGDominators.h:
* dfg/DFGSSACalculator.cpp:
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
Source/WTF:
Remove some remaining DFG-specific snippets from Dominators. This used to be a non-template
DFG class, and some time ago I hoisted it into WTF and made it generic. But since the only
user of the class was the DFG, this class still had a handful of DFG-specific snippets that
didn't compile when I started using it from B3.
Also renamed immediateDominatorOf() to idom(). This is the sort of abbreviation that we
wouldn't ordinarily want to have in WebKit. But WebKit does allow for abbreviations that are
"more canonical". The term "idom" is definitely more canonical than "immediateDominatorOf".
* wtf/Dominators.h:
(WTF::Dominators::dominates):
(WTF::Dominators::idom):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::ValidationContext::handleErrors):
Canonical link: https://commits.webkit.org/170119@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@193682 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-12-08 02:46:22 +00:00
|
|
|
for (auto successor : graph.successors(block))
|
|
|
|
|
dataLog(comma, graph.dump(successor));
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
dataLog("], predecessors = [");
|
|
|
|
|
comma = CommaPrinter();
|
FTL B3 should be able to flag the tag constants as being super important so that B3 can hoist them and Air can force them into registers
https://bugs.webkit.org/show_bug.cgi?id=151955
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Taught B3 about the concept of "fast constants". A client of B3 can now tell B3 which
constants are super important. B3 will not spill the constant in that case and will ensure
that the constant is materialized only once: statically once, and dynamically once per
procedure execution. The hoistFastConstants() algorithm in B3MoveConstants.cpp achieves this
by first picking the lowest common dominator of all uses of each fast constant, and then
picking the materialization point by finding the lowest dominator of that dominator that is
tied for lowest block frequency. In practice, the second step ensures that this is the lowest
point in the program that is not in a loop (i.e. executes no more than once dynamically per
procedure invocation).
Taught Air about the concept of "fast tmps". B3 tells Air that a tmp is fast if it is used to
hold the materialization of a fast constant. IRC will use the lowest possible spill score for
fast tmps. In practice, this ensures that fast constants are never spilled.
Added a small snippet of code to FTL::LowerDFGToLLVM that makes both of the tag constants
into fast constants.
My hope is that this very brute-force heuristic is good enough that we don't have to think
about constants for a while. Based on my experience with how LLVM's constant hoisting works
out, the heuristic in this patch is going to be tough to beat. LLVM's constant hoisting does
good things when it hoists the tags, and usually causes nothing but problems when it hoists
anything else. This is because there is no way a low-level compiler to really understand how
a constant materialization impacts some operation's contribution to the overall execution
time of a procedure. But, in the FTL we know that constant materializations for type checks
are a bummer because we are super comfortable placing type checks on the hottest of paths. So
those are the last paths where extra instructions should be added by the compiler. On the
other hand, all other large constant uses are on relatively cold paths, or paths that are
already expensive for other reasons. For example, global variable accesses have to
materialize a pointer to the global. But that's not really a big deal, since a load from a
global involves first the load itself and then type checks on the result - so probably the
constant materialization is just not interesting. A store to a global often involves a store
barrier, so the constant materialization is really not interesting. This patch codifies this
heuristic in a pact between Air, B3, and the FTL: FTL demands that B3 pin the two tags in
registers, and B3 relays the demand to Air.
* JavaScriptCore.xcodeproj/project.pbxproj:
* b3/B3CFG.h: Added.
(JSC::B3::CFG::CFG):
(JSC::B3::CFG::root):
(JSC::B3::CFG::newMap):
(JSC::B3::CFG::successors):
(JSC::B3::CFG::predecessors):
(JSC::B3::CFG::index):
(JSC::B3::CFG::node):
(JSC::B3::CFG::numNodes):
(JSC::B3::CFG::dump):
* b3/B3Dominators.h: Added.
(JSC::B3::Dominators::Dominators):
* b3/B3IndexMap.h:
(JSC::B3::IndexMap::resize):
(JSC::B3::IndexMap::size):
(JSC::B3::IndexMap::operator[]):
* b3/B3LowerMacros.cpp:
* b3/B3LowerToAir.cpp:
(JSC::B3::Air::LowerToAir::tmp):
* b3/B3MoveConstants.cpp:
* b3/B3Opcode.h:
(JSC::B3::constPtrOpcode):
(JSC::B3::isConstant):
* b3/B3Procedure.cpp:
(JSC::B3::Procedure::Procedure):
(JSC::B3::Procedure::resetReachability):
(JSC::B3::Procedure::invalidateCFG):
(JSC::B3::Procedure::dump):
(JSC::B3::Procedure::deleteValue):
(JSC::B3::Procedure::dominators):
(JSC::B3::Procedure::addFastConstant):
(JSC::B3::Procedure::isFastConstant):
(JSC::B3::Procedure::addDataSection):
* b3/B3Procedure.h:
(JSC::B3::Procedure::size):
(JSC::B3::Procedure::cfg):
(JSC::B3::Procedure::setLastPhaseName):
* b3/B3ReduceStrength.cpp:
* b3/B3ValueInlines.h:
(JSC::B3::Value::isConstant):
(JSC::B3::Value::isInteger):
* b3/B3ValueKey.h:
(JSC::B3::ValueKey::canMaterialize):
(JSC::B3::ValueKey::isConstant):
* b3/air/AirCode.cpp:
(JSC::B3::Air::Code::findNextBlock):
(JSC::B3::Air::Code::addFastTmp):
* b3/air/AirCode.h:
(JSC::B3::Air::Code::specials):
(JSC::B3::Air::Code::isFastTmp):
(JSC::B3::Air::Code::setLastPhaseName):
* b3/air/AirIteratedRegisterCoalescing.cpp:
* dfg/DFGDominators.h:
* dfg/DFGSSACalculator.cpp:
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
Source/WTF:
Remove some remaining DFG-specific snippets from Dominators. This used to be a non-template
DFG class, and some time ago I hoisted it into WTF and made it generic. But since the only
user of the class was the DFG, this class still had a handful of DFG-specific snippets that
didn't compile when I started using it from B3.
Also renamed immediateDominatorOf() to idom(). This is the sort of abbreviation that we
wouldn't ordinarily want to have in WebKit. But WebKit does allow for abbreviations that are
"more canonical". The term "idom" is definitely more canonical than "immediateDominatorOf".
* wtf/Dominators.h:
(WTF::Dominators::dominates):
(WTF::Dominators::idom):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::ValidationContext::handleErrors):
Canonical link: https://commits.webkit.org/170119@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@193682 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-12-08 02:46:22 +00:00
|
|
|
for (auto predecessor : graph.predecessors(block))
|
|
|
|
|
dataLog(comma, graph.dump(predecessor));
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
dataLog("]\n");
|
|
|
|
|
}
|
|
|
|
|
dataLog("\n");
|
|
|
|
|
dataLog("Lengauer-Tarjan Dominators:\n");
|
|
|
|
|
dataLog(dominators);
|
|
|
|
|
dataLog("\n");
|
|
|
|
|
dataLog("Naive Dominators:\n");
|
|
|
|
|
naiveDominators.dump(WTF::dataFile());
|
|
|
|
|
dataLog("\n");
|
|
|
|
|
dataLog("Graph at time of failure:\n");
|
|
|
|
|
dataLog(graph);
|
|
|
|
|
dataLog("\n");
|
|
|
|
|
dataLog("DFG DOMINATOR VALIDATION FAILIED!\n");
|
|
|
|
|
CRASH();
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
Graph& graph;
|
|
|
|
|
Dominators& dominators;
|
|
|
|
|
NaiveDominators naiveDominators;
|
|
|
|
|
|
|
|
|
|
struct Error {
|
2019-08-12 20:57:15 +00:00
|
|
|
WTF_MAKE_STRUCT_FAST_ALLOCATED;
|
|
|
|
|
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
typename Graph::Node from;
|
|
|
|
|
typename Graph::Node to;
|
|
|
|
|
const char* message;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
Vector<Error> errors;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
bool naiveDominates(typename Graph::Node from, typename Graph::Node to) const
|
|
|
|
|
{
|
|
|
|
|
for (typename Graph::Node block = to; block; block = m_data[block].idomParent) {
|
|
|
|
|
if (block == from)
|
|
|
|
|
return true;
|
|
|
|
|
}
|
|
|
|
|
return false;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template<typename Functor>
|
|
|
|
|
void forAllBlocksInDominanceFrontierOfImpl(
|
|
|
|
|
typename Graph::Node from, const Functor& functor) const
|
|
|
|
|
{
|
|
|
|
|
// Paraphrasing from http://en.wikipedia.org/wiki/Dominator_(graph_theory):
|
|
|
|
|
// "The dominance frontier of a block 'from' is the set of all blocks 'to' such that
|
|
|
|
|
// 'from' dominates an immediate predecessor of 'to', but 'from' does not strictly
|
|
|
|
|
// dominate 'to'."
|
|
|
|
|
//
|
|
|
|
|
// A useful corner case to remember: a block may be in its own dominance frontier if it has
|
|
|
|
|
// a loop edge to itself, since it dominates itself and so it dominates its own immediate
|
|
|
|
|
// predecessor, and a block never strictly dominates itself.
|
|
|
|
|
|
|
|
|
|
forAllBlocksDominatedBy(
|
|
|
|
|
from,
|
|
|
|
|
[&] (typename Graph::Node block) {
|
|
|
|
|
for (typename Graph::Node to : m_graph.successors(block)) {
|
|
|
|
|
if (!strictlyDominates(from, to))
|
|
|
|
|
functor(to);
|
|
|
|
|
}
|
|
|
|
|
});
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
template<typename Functor>
|
|
|
|
|
void forAllBlocksInIteratedDominanceFrontierOfImpl(
|
Support compiling catch in the DFG
https://bugs.webkit.org/show_bug.cgi?id=174590
<rdar://problem/34047845>
Reviewed by Filip Pizlo.
JSTests:
* microbenchmarks/delta-blue-try-catch.js: Added.
(exception):
(value):
(OrderedCollection):
(OrderedCollection.prototype.add):
(OrderedCollection.prototype.at):
(OrderedCollection.prototype.size):
(OrderedCollection.prototype.removeFirst):
(OrderedCollection.prototype.remove):
(Strength):
(Strength.stronger):
(Strength.weaker):
(Strength.weakestOf):
(Strength.strongest):
(Strength.prototype.nextWeaker):
(Constraint):
(Constraint.prototype.addConstraint):
(Constraint.prototype.satisfy):
(Constraint.prototype.destroyConstraint):
(Constraint.prototype.isInput):
(UnaryConstraint):
(UnaryConstraint.prototype.addToGraph):
(UnaryConstraint.prototype.chooseMethod):
(UnaryConstraint.prototype.isSatisfied):
(UnaryConstraint.prototype.markInputs):
(UnaryConstraint.prototype.output):
(UnaryConstraint.prototype.recalculate):
(UnaryConstraint.prototype.markUnsatisfied):
(UnaryConstraint.prototype.inputsKnown):
(UnaryConstraint.prototype.removeFromGraph):
(StayConstraint):
(StayConstraint.prototype.execute):
(EditConstraint.prototype.isInput):
(EditConstraint.prototype.execute):
(BinaryConstraint):
(BinaryConstraint.prototype.chooseMethod):
(BinaryConstraint.prototype.addToGraph):
(BinaryConstraint.prototype.isSatisfied):
(BinaryConstraint.prototype.markInputs):
(BinaryConstraint.prototype.input):
(BinaryConstraint.prototype.output):
(BinaryConstraint.prototype.recalculate):
(BinaryConstraint.prototype.markUnsatisfied):
(BinaryConstraint.prototype.inputsKnown):
(BinaryConstraint.prototype.removeFromGraph):
(ScaleConstraint):
(ScaleConstraint.prototype.addToGraph):
(ScaleConstraint.prototype.removeFromGraph):
(ScaleConstraint.prototype.markInputs):
(ScaleConstraint.prototype.execute):
(ScaleConstraint.prototype.recalculate):
(EqualityConstraint):
(EqualityConstraint.prototype.execute):
(Variable):
(Variable.prototype.addConstraint):
(Variable.prototype.removeConstraint):
(Planner):
(Planner.prototype.incrementalAdd):
(Planner.prototype.incrementalRemove):
(Planner.prototype.newMark):
(Planner.prototype.makePlan):
(Planner.prototype.extractPlanFromConstraints):
(Planner.prototype.addPropagate):
(Planner.prototype.removePropagateFrom):
(Planner.prototype.addConstraintsConsumingTo):
(Plan):
(Plan.prototype.addConstraint):
(Plan.prototype.size):
(Plan.prototype.constraintAt):
(Plan.prototype.execute):
(chainTest):
(projectionTest):
(change):
(deltaBlue):
* microbenchmarks/fake-iterators-that-throw-when-finished.js: Added.
(assert):
(Numbers):
(Numbers.prototype.next):
(return.Transpose):
(return.Transpose.prototype.next):
(transpose):
(verifyEven):
(verifyString):
(foo):
(runIterators):
* microbenchmarks/try-catch-word-count.js: Added.
(let.assert):
(EOF):
(let.texts):
(let.o.apply):
(foo):
(bar):
(f):
(run):
(test1):
(test2):
(test3):
(fn):
(A):
(B):
(A.prototype.getValue):
(B.prototype.getParentValue):
(strlen):
(sum.0):
(test):
(result.test.o):
(set add.set add):
(set forEach):
(stringHash):
(set if):
(testFunction):
(set delete.set has.set add):
* stress/catch-set-argument-speculation-failure.js: Added.
(o):
(e):
(e2):
(escape):
(baz):
(noInline.run):
(noInline):
* stress/osr-enter-to-catch-with-set-local-type-check-failure.js: Added.
(foo):
(e):
(baz):
(bar):
Source/JavaScriptCore:
This patch implements OSR entry into op_catch in the DFG. We will support OSR entry
into the FTL in a followup: https://bugs.webkit.org/show_bug.cgi?id=175396
To implement catch in the DFG, this patch introduces the concept of multiple
entrypoints into CPS/LoadStore DFG IR. A lot of this patch is stringing this concept
through the DFG. Many phases used to assume that Graph::block(0) is the only root, and this
patch contains many straight forward changes generalizing the code to handle more than
one entrypoint.
A main building block of this is moving to two CFG types: SSACFG and CPSCFG. SSACFG
is the same CFG we used to have. CPSCFG is a new type that introduces a fake root
that has an outgoing edge to all the entrypoints. This allows our existing graph algorithms
to Just Work over CPSCFG. For example, there is now the concept of SSADominators vs CPSDominators,
and SSANaturalLoops vs CPSNaturalLoops.
The way we compile the catch entrypoint is by bootstrapping the state
of the program by loading all live bytecode locals from a buffer. The OSR
entry code will store all live values into that buffer before jumping to
the entrypoint. The OSR entry code is also responsible for performing type
proofs of the arguments before doing an OSR entry. If there is a type
mismatch, it's not legal to OSR enter into the DFG compilation. Currently,
each catch entrypoint knows the argument type proofs it must perform to enter
into the DFG. Currently, all entrypoints' arguments flush format are unified
via ArgumentPosition, but this is just an implementation detail. The code is
written more generally to assume that each entrypoint may perform its own distinct
proof.
op_catch now performs value profiling for all live bytecode locals in the
LLInt and baseline JIT. This information is then fed into the DFG via the
ExtractCatchLocal node in the prediction propagation phase.
This patch also changes how we generate op_catch in bytecode. All op_catches
are now split out at the end of the program in bytecode. This ensures that
no op_catch is inside a try block. This is needed to ensure correctness in
the DFGLiveCatchVariablePreservationPhase. That phase only inserts flushes
before SetLocals inside a try block. If an op_catch were in a try block, this
would cause the phase to insert a Flush before one of the state bootstrapping
SetLocals, which would generate invalid IR. Moving op_catch to be generated on
its own at the end of a bytecode stream seemed like the most elegant solution since
it better represents that we treat op_catch as an entrypoint. This is true
both in the DFG and in the baseline and LLInt: we don't reach an op_catch
via normal control flow. Because op_catch cannot throw, this will not break
any previous semantics of op_catch. Logically, it'd be valid to split try
blocks around any non-throwing bytecode operation.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* bytecode/BytecodeDumper.cpp:
(JSC::BytecodeDumper<Block>::dumpBytecode):
* bytecode/BytecodeList.json:
* bytecode/BytecodeUseDef.h:
(JSC::computeUsesForBytecodeOffset):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::finishCreation):
(JSC::CodeBlock::ensureCatchLivenessIsComputedForBytecodeOffset):
(JSC::CodeBlock::updateAllPredictionsAndCountLiveness):
(JSC::CodeBlock::validate):
* bytecode/CodeBlock.h:
* bytecode/ValueProfile.h:
(JSC::ValueProfile::ValueProfile):
(JSC::ValueProfileAndOperandBuffer::ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::~ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::forEach):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::generate):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::emitCatch):
(JSC::BytecodeGenerator::emitEnumeration):
* bytecompiler/BytecodeGenerator.h:
* bytecompiler/NodesCodegen.cpp:
(JSC::TryNode::emitBytecode):
* dfg/DFGAbstractInterpreterInlines.h:
(JSC::DFG::AbstractInterpreter<AbstractStateType>::executeEffects):
* dfg/DFGBackwardsCFG.h:
(JSC::DFG::BackwardsCFG::BackwardsCFG):
* dfg/DFGBasicBlock.cpp:
(JSC::DFG::BasicBlock::BasicBlock):
* dfg/DFGBasicBlock.h:
(JSC::DFG::BasicBlock::findTerminal const):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::setDirect):
(JSC::DFG::ByteCodeParser::flush):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::DelayedSetLocal):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::execute):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::parseCodeBlock):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCFG.h:
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::roots):
(JSC::DFG::CPSCFG::CPSCFG):
(JSC::DFG::selectCFG):
* dfg/DFGCPSRethreadingPhase.cpp:
(JSC::DFG::CPSRethreadingPhase::specialCaseArguments):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGClobberize.h:
(JSC::DFG::clobberize):
* dfg/DFGControlEquivalenceAnalysis.h:
(JSC::DFG::ControlEquivalenceAnalysis::ControlEquivalenceAnalysis):
* dfg/DFGDCEPhase.cpp:
(JSC::DFG::DCEPhase::run):
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDoesGC.cpp:
(JSC::DFG::doesGC):
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::ensureDominatorsForCFG):
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::EdgeDominates):
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
(JSC::DFG::FixupPhase::fixupChecksInBlock):
* dfg/DFGFlushFormat.h:
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::unboxLoopNode):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::dump):
(JSC::DFG::Graph::determineReachability):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::blocksInPreOrder):
(JSC::DFG::Graph::blocksInPostOrder):
(JSC::DFG::Graph::ensureCPSDominators):
(JSC::DFG::Graph::ensureSSADominators):
(JSC::DFG::Graph::ensureCPSNaturalLoops):
(JSC::DFG::Graph::ensureSSANaturalLoops):
(JSC::DFG::Graph::ensureBackwardsCFG):
(JSC::DFG::Graph::ensureBackwardsDominators):
(JSC::DFG::Graph::ensureControlEquivalenceAnalysis):
(JSC::DFG::Graph::methodOfGettingAValueProfileFor):
(JSC::DFG::Graph::clearCPSCFGData):
(JSC::DFG::Graph::ensureDominators): Deleted.
(JSC::DFG::Graph::ensurePrePostNumbering): Deleted.
(JSC::DFG::Graph::ensureNaturalLoops): Deleted.
* dfg/DFGGraph.h:
(JSC::DFG::Graph::willCatchExceptionInMachineFrame):
(JSC::DFG::Graph::isEntrypoint const):
* dfg/DFGInPlaceAbstractState.cpp:
(JSC::DFG::InPlaceAbstractState::initialize):
(JSC::DFG::InPlaceAbstractState::mergeToSuccessors):
* dfg/DFGJITCode.cpp:
(JSC::DFG::JITCode::shrinkToFit):
* dfg/DFGJITCode.h:
(JSC::DFG::JITCode::catchOSREntryDataForBytecodeIndex):
(JSC::DFG::JITCode::finalizeCatchOSREntrypoints):
(JSC::DFG::JITCode::appendCatchEntrypoint):
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::compile):
(JSC::DFG::JITCompiler::compileFunction):
(JSC::DFG::JITCompiler::noticeCatchEntrypoint):
(JSC::DFG::JITCompiler::noticeOSREntry):
(JSC::DFG::JITCompiler::makeCatchOSREntryBuffer):
* dfg/DFGJITCompiler.h:
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::run):
(JSC::DFG::LiveCatchVariablePreservationPhase::isValidFlushLocation):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlockForTryCatch):
(JSC::DFG::LiveCatchVariablePreservationPhase::newVariableAccessData):
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException): Deleted.
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock): Deleted.
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGMaximalFlushInsertionPhase.cpp:
(JSC::DFG::MaximalFlushInsertionPhase::run):
(JSC::DFG::MaximalFlushInsertionPhase::treatRegularBlock):
(JSC::DFG::MaximalFlushInsertionPhase::treatRootBlock):
* dfg/DFGMayExit.cpp:
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::isSwitch const):
(JSC::DFG::Node::successor):
(JSC::DFG::Node::catchOSREntryIndex const):
(JSC::DFG::Node::catchLocalPrediction):
(JSC::DFG::Node::isSwitch): Deleted.
* dfg/DFGNodeType.h:
* dfg/DFGOSREntry.cpp:
(JSC::DFG::prepareCatchOSREntry):
* dfg/DFGOSREntry.h:
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGOSRExitCompilerCommon.cpp:
(JSC::DFG::handleExitCounts):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering): Deleted.
(JSC::DFG::PrePostNumbering::~PrePostNumbering): Deleted.
(WTF::printInternal): Deleted.
* dfg/DFGPrePostNumbering.h:
(): Deleted.
(JSC::DFG::PrePostNumbering::preNumber const): Deleted.
(JSC::DFG::PrePostNumbering::postNumber const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::isDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::edgeKind const): Deleted.
* dfg/DFGPredictionInjectionPhase.cpp:
(JSC::DFG::PredictionInjectionPhase::run):
* dfg/DFGPredictionPropagationPhase.cpp:
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
(JSC::DFG::performSSAConversion):
* dfg/DFGSafeToExecute.h:
(JSC::DFG::safeToExecute):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compileCurrentBlock):
(JSC::DFG::SpeculativeJIT::checkArgumentTypes):
(JSC::DFG::SpeculativeJIT::createOSREntries):
(JSC::DFG::SpeculativeJIT::linkOSREntries):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGStaticExecutionCountEstimationPhase.cpp:
(JSC::DFG::StaticExecutionCountEstimationPhase::run):
* dfg/DFGStrengthReductionPhase.cpp:
(JSC::DFG::StrengthReductionPhase::handleNode):
* dfg/DFGTierUpCheckInjectionPhase.cpp:
(JSC::DFG::TierUpCheckInjectionPhase::run):
(JSC::DFG::TierUpCheckInjectionPhase::buildNaturalLoopToLoopHintMap):
* dfg/DFGTypeCheckHoistingPhase.cpp:
(JSC::DFG::TypeCheckHoistingPhase::run):
* dfg/DFGValidate.cpp:
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToB3.cpp:
(JSC::FTL::DFG::LowerDFGToB3::lower):
(JSC::FTL::DFG::LowerDFGToB3::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToB3::isValid):
* jit/JIT.h:
* jit/JITInlines.h:
(JSC::JIT::callOperation):
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOperations.cpp:
* jit/JITOperations.h:
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* llint/LLIntSlowPaths.h:
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
Source/WTF:
This patch generalizes the BackwardsGraph fake root into a more generalizable
class called SingleRootGraph. SingleRootGraph exposes the general graph interface
used in Dominators and NaturalLoops. SingleRootGraph takes as input a graph with
the normal graph interface, but also allows the input graph to contain more than
one root. SingleRootGraph then exposes a single root, which it creates, that has
an outgoing edge to all the roots in the original graph.
* WTF.xcodeproj/project.pbxproj:
* wtf/BackwardsGraph.h:
(WTF::BackwardsGraph::dump const):
(WTF::BackwardsGraph::rootName): Deleted.
(WTF::BackwardsGraph::Node::Node): Deleted.
(WTF::BackwardsGraph::Node::root): Deleted.
(WTF::BackwardsGraph::Node::operator== const): Deleted.
(WTF::BackwardsGraph::Node::operator!= const): Deleted.
(WTF::BackwardsGraph::Node::operator bool const): Deleted.
(WTF::BackwardsGraph::Node::isRoot const): Deleted.
(WTF::BackwardsGraph::Node::node const): Deleted.
(): Deleted.
(WTF::BackwardsGraph::Set::Set): Deleted.
(WTF::BackwardsGraph::Set::add): Deleted.
(WTF::BackwardsGraph::Set::remove): Deleted.
(WTF::BackwardsGraph::Set::contains): Deleted.
(WTF::BackwardsGraph::Set::dump const): Deleted.
(WTF::BackwardsGraph::Map::Map): Deleted.
(WTF::BackwardsGraph::Map::clear): Deleted.
(WTF::BackwardsGraph::Map::size const): Deleted.
(WTF::BackwardsGraph::Map::operator[]): Deleted.
(WTF::BackwardsGraph::Map::operator[] const): Deleted.
* wtf/Dominators.h:
(WTF::Dominators::Dominators):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf const):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl const):
* wtf/SingleRootGraph.h: Added.
(WTF::SingleRootGraphNode::rootName):
(WTF::SingleRootGraphNode::SingleRootGraphNode):
(WTF::SingleRootGraphNode::root):
(WTF::SingleRootGraphNode::operator== const):
(WTF::SingleRootGraphNode::operator!= const):
(WTF::SingleRootGraphNode::operator bool const):
(WTF::SingleRootGraphNode::isRoot const):
(WTF::SingleRootGraphNode::node const):
(WTF::SingleRootGraphSet::add):
(WTF::SingleRootGraphSet::remove):
(WTF::SingleRootGraphSet::contains):
(WTF::SingleRootGraphSet::dump const):
(WTF::SingleRootMap::SingleRootMap):
(WTF::SingleRootMap::clear):
(WTF::SingleRootMap::size const):
(WTF::SingleRootMap::operator[]):
(WTF::SingleRootMap::operator[] const):
(WTF::SingleRootGraph::SingleRootGraph):
(WTF::SingleRootGraph::root const):
(WTF::SingleRootGraph::newMap):
(WTF::SingleRootGraph::successors const):
(WTF::SingleRootGraph::predecessors const):
(WTF::SingleRootGraph::index const):
(WTF::SingleRootGraph::node const):
(WTF::SingleRootGraph::numNodes const):
(WTF::SingleRootGraph::dump const):
(WTF::SingleRootGraph::assertIsConsistent const):
Canonical link: https://commits.webkit.org/192644@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@221196 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2017-08-25 18:26:15 +00:00
|
|
|
const List& from, const Functor& functor) const
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
{
|
Support compiling catch in the DFG
https://bugs.webkit.org/show_bug.cgi?id=174590
<rdar://problem/34047845>
Reviewed by Filip Pizlo.
JSTests:
* microbenchmarks/delta-blue-try-catch.js: Added.
(exception):
(value):
(OrderedCollection):
(OrderedCollection.prototype.add):
(OrderedCollection.prototype.at):
(OrderedCollection.prototype.size):
(OrderedCollection.prototype.removeFirst):
(OrderedCollection.prototype.remove):
(Strength):
(Strength.stronger):
(Strength.weaker):
(Strength.weakestOf):
(Strength.strongest):
(Strength.prototype.nextWeaker):
(Constraint):
(Constraint.prototype.addConstraint):
(Constraint.prototype.satisfy):
(Constraint.prototype.destroyConstraint):
(Constraint.prototype.isInput):
(UnaryConstraint):
(UnaryConstraint.prototype.addToGraph):
(UnaryConstraint.prototype.chooseMethod):
(UnaryConstraint.prototype.isSatisfied):
(UnaryConstraint.prototype.markInputs):
(UnaryConstraint.prototype.output):
(UnaryConstraint.prototype.recalculate):
(UnaryConstraint.prototype.markUnsatisfied):
(UnaryConstraint.prototype.inputsKnown):
(UnaryConstraint.prototype.removeFromGraph):
(StayConstraint):
(StayConstraint.prototype.execute):
(EditConstraint.prototype.isInput):
(EditConstraint.prototype.execute):
(BinaryConstraint):
(BinaryConstraint.prototype.chooseMethod):
(BinaryConstraint.prototype.addToGraph):
(BinaryConstraint.prototype.isSatisfied):
(BinaryConstraint.prototype.markInputs):
(BinaryConstraint.prototype.input):
(BinaryConstraint.prototype.output):
(BinaryConstraint.prototype.recalculate):
(BinaryConstraint.prototype.markUnsatisfied):
(BinaryConstraint.prototype.inputsKnown):
(BinaryConstraint.prototype.removeFromGraph):
(ScaleConstraint):
(ScaleConstraint.prototype.addToGraph):
(ScaleConstraint.prototype.removeFromGraph):
(ScaleConstraint.prototype.markInputs):
(ScaleConstraint.prototype.execute):
(ScaleConstraint.prototype.recalculate):
(EqualityConstraint):
(EqualityConstraint.prototype.execute):
(Variable):
(Variable.prototype.addConstraint):
(Variable.prototype.removeConstraint):
(Planner):
(Planner.prototype.incrementalAdd):
(Planner.prototype.incrementalRemove):
(Planner.prototype.newMark):
(Planner.prototype.makePlan):
(Planner.prototype.extractPlanFromConstraints):
(Planner.prototype.addPropagate):
(Planner.prototype.removePropagateFrom):
(Planner.prototype.addConstraintsConsumingTo):
(Plan):
(Plan.prototype.addConstraint):
(Plan.prototype.size):
(Plan.prototype.constraintAt):
(Plan.prototype.execute):
(chainTest):
(projectionTest):
(change):
(deltaBlue):
* microbenchmarks/fake-iterators-that-throw-when-finished.js: Added.
(assert):
(Numbers):
(Numbers.prototype.next):
(return.Transpose):
(return.Transpose.prototype.next):
(transpose):
(verifyEven):
(verifyString):
(foo):
(runIterators):
* microbenchmarks/try-catch-word-count.js: Added.
(let.assert):
(EOF):
(let.texts):
(let.o.apply):
(foo):
(bar):
(f):
(run):
(test1):
(test2):
(test3):
(fn):
(A):
(B):
(A.prototype.getValue):
(B.prototype.getParentValue):
(strlen):
(sum.0):
(test):
(result.test.o):
(set add.set add):
(set forEach):
(stringHash):
(set if):
(testFunction):
(set delete.set has.set add):
* stress/catch-set-argument-speculation-failure.js: Added.
(o):
(e):
(e2):
(escape):
(baz):
(noInline.run):
(noInline):
* stress/osr-enter-to-catch-with-set-local-type-check-failure.js: Added.
(foo):
(e):
(baz):
(bar):
Source/JavaScriptCore:
This patch implements OSR entry into op_catch in the DFG. We will support OSR entry
into the FTL in a followup: https://bugs.webkit.org/show_bug.cgi?id=175396
To implement catch in the DFG, this patch introduces the concept of multiple
entrypoints into CPS/LoadStore DFG IR. A lot of this patch is stringing this concept
through the DFG. Many phases used to assume that Graph::block(0) is the only root, and this
patch contains many straight forward changes generalizing the code to handle more than
one entrypoint.
A main building block of this is moving to two CFG types: SSACFG and CPSCFG. SSACFG
is the same CFG we used to have. CPSCFG is a new type that introduces a fake root
that has an outgoing edge to all the entrypoints. This allows our existing graph algorithms
to Just Work over CPSCFG. For example, there is now the concept of SSADominators vs CPSDominators,
and SSANaturalLoops vs CPSNaturalLoops.
The way we compile the catch entrypoint is by bootstrapping the state
of the program by loading all live bytecode locals from a buffer. The OSR
entry code will store all live values into that buffer before jumping to
the entrypoint. The OSR entry code is also responsible for performing type
proofs of the arguments before doing an OSR entry. If there is a type
mismatch, it's not legal to OSR enter into the DFG compilation. Currently,
each catch entrypoint knows the argument type proofs it must perform to enter
into the DFG. Currently, all entrypoints' arguments flush format are unified
via ArgumentPosition, but this is just an implementation detail. The code is
written more generally to assume that each entrypoint may perform its own distinct
proof.
op_catch now performs value profiling for all live bytecode locals in the
LLInt and baseline JIT. This information is then fed into the DFG via the
ExtractCatchLocal node in the prediction propagation phase.
This patch also changes how we generate op_catch in bytecode. All op_catches
are now split out at the end of the program in bytecode. This ensures that
no op_catch is inside a try block. This is needed to ensure correctness in
the DFGLiveCatchVariablePreservationPhase. That phase only inserts flushes
before SetLocals inside a try block. If an op_catch were in a try block, this
would cause the phase to insert a Flush before one of the state bootstrapping
SetLocals, which would generate invalid IR. Moving op_catch to be generated on
its own at the end of a bytecode stream seemed like the most elegant solution since
it better represents that we treat op_catch as an entrypoint. This is true
both in the DFG and in the baseline and LLInt: we don't reach an op_catch
via normal control flow. Because op_catch cannot throw, this will not break
any previous semantics of op_catch. Logically, it'd be valid to split try
blocks around any non-throwing bytecode operation.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* bytecode/BytecodeDumper.cpp:
(JSC::BytecodeDumper<Block>::dumpBytecode):
* bytecode/BytecodeList.json:
* bytecode/BytecodeUseDef.h:
(JSC::computeUsesForBytecodeOffset):
* bytecode/CodeBlock.cpp:
(JSC::CodeBlock::finishCreation):
(JSC::CodeBlock::ensureCatchLivenessIsComputedForBytecodeOffset):
(JSC::CodeBlock::updateAllPredictionsAndCountLiveness):
(JSC::CodeBlock::validate):
* bytecode/CodeBlock.h:
* bytecode/ValueProfile.h:
(JSC::ValueProfile::ValueProfile):
(JSC::ValueProfileAndOperandBuffer::ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::~ValueProfileAndOperandBuffer):
(JSC::ValueProfileAndOperandBuffer::forEach):
* bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::generate):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::emitCatch):
(JSC::BytecodeGenerator::emitEnumeration):
* bytecompiler/BytecodeGenerator.h:
* bytecompiler/NodesCodegen.cpp:
(JSC::TryNode::emitBytecode):
* dfg/DFGAbstractInterpreterInlines.h:
(JSC::DFG::AbstractInterpreter<AbstractStateType>::executeEffects):
* dfg/DFGBackwardsCFG.h:
(JSC::DFG::BackwardsCFG::BackwardsCFG):
* dfg/DFGBasicBlock.cpp:
(JSC::DFG::BasicBlock::BasicBlock):
* dfg/DFGBasicBlock.h:
(JSC::DFG::BasicBlock::findTerminal const):
* dfg/DFGByteCodeParser.cpp:
(JSC::DFG::ByteCodeParser::setDirect):
(JSC::DFG::ByteCodeParser::flush):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::DelayedSetLocal):
(JSC::DFG::ByteCodeParser::DelayedSetLocal::execute):
(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::parseCodeBlock):
(JSC::DFG::ByteCodeParser::parse):
* dfg/DFGCFG.h:
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::roots):
(JSC::DFG::CPSCFG::CPSCFG):
(JSC::DFG::selectCFG):
* dfg/DFGCPSRethreadingPhase.cpp:
(JSC::DFG::CPSRethreadingPhase::specialCaseArguments):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGClobberize.h:
(JSC::DFG::clobberize):
* dfg/DFGControlEquivalenceAnalysis.h:
(JSC::DFG::ControlEquivalenceAnalysis::ControlEquivalenceAnalysis):
* dfg/DFGDCEPhase.cpp:
(JSC::DFG::DCEPhase::run):
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDoesGC.cpp:
(JSC::DFG::doesGC):
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::ensureDominatorsForCFG):
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::EdgeDominates):
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGFixupPhase.cpp:
(JSC::DFG::FixupPhase::fixupNode):
(JSC::DFG::FixupPhase::fixupChecksInBlock):
* dfg/DFGFlushFormat.h:
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::unboxLoopNode):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::dump):
(JSC::DFG::Graph::determineReachability):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::blocksInPreOrder):
(JSC::DFG::Graph::blocksInPostOrder):
(JSC::DFG::Graph::ensureCPSDominators):
(JSC::DFG::Graph::ensureSSADominators):
(JSC::DFG::Graph::ensureCPSNaturalLoops):
(JSC::DFG::Graph::ensureSSANaturalLoops):
(JSC::DFG::Graph::ensureBackwardsCFG):
(JSC::DFG::Graph::ensureBackwardsDominators):
(JSC::DFG::Graph::ensureControlEquivalenceAnalysis):
(JSC::DFG::Graph::methodOfGettingAValueProfileFor):
(JSC::DFG::Graph::clearCPSCFGData):
(JSC::DFG::Graph::ensureDominators): Deleted.
(JSC::DFG::Graph::ensurePrePostNumbering): Deleted.
(JSC::DFG::Graph::ensureNaturalLoops): Deleted.
* dfg/DFGGraph.h:
(JSC::DFG::Graph::willCatchExceptionInMachineFrame):
(JSC::DFG::Graph::isEntrypoint const):
* dfg/DFGInPlaceAbstractState.cpp:
(JSC::DFG::InPlaceAbstractState::initialize):
(JSC::DFG::InPlaceAbstractState::mergeToSuccessors):
* dfg/DFGJITCode.cpp:
(JSC::DFG::JITCode::shrinkToFit):
* dfg/DFGJITCode.h:
(JSC::DFG::JITCode::catchOSREntryDataForBytecodeIndex):
(JSC::DFG::JITCode::finalizeCatchOSREntrypoints):
(JSC::DFG::JITCode::appendCatchEntrypoint):
* dfg/DFGJITCompiler.cpp:
(JSC::DFG::JITCompiler::compile):
(JSC::DFG::JITCompiler::compileFunction):
(JSC::DFG::JITCompiler::noticeCatchEntrypoint):
(JSC::DFG::JITCompiler::noticeOSREntry):
(JSC::DFG::JITCompiler::makeCatchOSREntryBuffer):
* dfg/DFGJITCompiler.h:
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLiveCatchVariablePreservationPhase.cpp:
(JSC::DFG::LiveCatchVariablePreservationPhase::run):
(JSC::DFG::LiveCatchVariablePreservationPhase::isValidFlushLocation):
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlockForTryCatch):
(JSC::DFG::LiveCatchVariablePreservationPhase::newVariableAccessData):
(JSC::DFG::LiveCatchVariablePreservationPhase::willCatchException): Deleted.
(JSC::DFG::LiveCatchVariablePreservationPhase::handleBlock): Deleted.
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGMaximalFlushInsertionPhase.cpp:
(JSC::DFG::MaximalFlushInsertionPhase::run):
(JSC::DFG::MaximalFlushInsertionPhase::treatRegularBlock):
(JSC::DFG::MaximalFlushInsertionPhase::treatRootBlock):
* dfg/DFGMayExit.cpp:
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::NaturalLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::isSwitch const):
(JSC::DFG::Node::successor):
(JSC::DFG::Node::catchOSREntryIndex const):
(JSC::DFG::Node::catchLocalPrediction):
(JSC::DFG::Node::isSwitch): Deleted.
* dfg/DFGNodeType.h:
* dfg/DFGOSREntry.cpp:
(JSC::DFG::prepareCatchOSREntry):
* dfg/DFGOSREntry.h:
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGOSRExitCompilerCommon.cpp:
(JSC::DFG::handleExitCounts):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering): Deleted.
(JSC::DFG::PrePostNumbering::~PrePostNumbering): Deleted.
(WTF::printInternal): Deleted.
* dfg/DFGPrePostNumbering.h:
(): Deleted.
(JSC::DFG::PrePostNumbering::preNumber const): Deleted.
(JSC::DFG::PrePostNumbering::postNumber const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isAncestorOf const): Deleted.
(JSC::DFG::PrePostNumbering::isStrictDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::isDescendantOf const): Deleted.
(JSC::DFG::PrePostNumbering::edgeKind const): Deleted.
* dfg/DFGPredictionInjectionPhase.cpp:
(JSC::DFG::PredictionInjectionPhase::run):
* dfg/DFGPredictionPropagationPhase.cpp:
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
(JSC::DFG::performSSAConversion):
* dfg/DFGSafeToExecute.h:
(JSC::DFG::safeToExecute):
* dfg/DFGSpeculativeJIT.cpp:
(JSC::DFG::SpeculativeJIT::compileCurrentBlock):
(JSC::DFG::SpeculativeJIT::checkArgumentTypes):
(JSC::DFG::SpeculativeJIT::createOSREntries):
(JSC::DFG::SpeculativeJIT::linkOSREntries):
* dfg/DFGSpeculativeJIT32_64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGSpeculativeJIT64.cpp:
(JSC::DFG::SpeculativeJIT::compile):
* dfg/DFGStaticExecutionCountEstimationPhase.cpp:
(JSC::DFG::StaticExecutionCountEstimationPhase::run):
* dfg/DFGStrengthReductionPhase.cpp:
(JSC::DFG::StrengthReductionPhase::handleNode):
* dfg/DFGTierUpCheckInjectionPhase.cpp:
(JSC::DFG::TierUpCheckInjectionPhase::run):
(JSC::DFG::TierUpCheckInjectionPhase::buildNaturalLoopToLoopHintMap):
* dfg/DFGTypeCheckHoistingPhase.cpp:
(JSC::DFG::TypeCheckHoistingPhase::run):
* dfg/DFGValidate.cpp:
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToB3.cpp:
(JSC::FTL::DFG::LowerDFGToB3::lower):
(JSC::FTL::DFG::LowerDFGToB3::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToB3::isValid):
* jit/JIT.h:
* jit/JITInlines.h:
(JSC::JIT::callOperation):
* jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOpcodes32_64.cpp:
(JSC::JIT::emit_op_catch):
* jit/JITOperations.cpp:
* jit/JITOperations.h:
* llint/LLIntSlowPaths.cpp:
(JSC::LLInt::LLINT_SLOW_PATH_DECL):
* llint/LLIntSlowPaths.h:
* llint/LowLevelInterpreter32_64.asm:
* llint/LowLevelInterpreter64.asm:
Source/WTF:
This patch generalizes the BackwardsGraph fake root into a more generalizable
class called SingleRootGraph. SingleRootGraph exposes the general graph interface
used in Dominators and NaturalLoops. SingleRootGraph takes as input a graph with
the normal graph interface, but also allows the input graph to contain more than
one root. SingleRootGraph then exposes a single root, which it creates, that has
an outgoing edge to all the roots in the original graph.
* WTF.xcodeproj/project.pbxproj:
* wtf/BackwardsGraph.h:
(WTF::BackwardsGraph::dump const):
(WTF::BackwardsGraph::rootName): Deleted.
(WTF::BackwardsGraph::Node::Node): Deleted.
(WTF::BackwardsGraph::Node::root): Deleted.
(WTF::BackwardsGraph::Node::operator== const): Deleted.
(WTF::BackwardsGraph::Node::operator!= const): Deleted.
(WTF::BackwardsGraph::Node::operator bool const): Deleted.
(WTF::BackwardsGraph::Node::isRoot const): Deleted.
(WTF::BackwardsGraph::Node::node const): Deleted.
(): Deleted.
(WTF::BackwardsGraph::Set::Set): Deleted.
(WTF::BackwardsGraph::Set::add): Deleted.
(WTF::BackwardsGraph::Set::remove): Deleted.
(WTF::BackwardsGraph::Set::contains): Deleted.
(WTF::BackwardsGraph::Set::dump const): Deleted.
(WTF::BackwardsGraph::Map::Map): Deleted.
(WTF::BackwardsGraph::Map::clear): Deleted.
(WTF::BackwardsGraph::Map::size const): Deleted.
(WTF::BackwardsGraph::Map::operator[]): Deleted.
(WTF::BackwardsGraph::Map::operator[] const): Deleted.
* wtf/Dominators.h:
(WTF::Dominators::Dominators):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf const):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl const):
* wtf/SingleRootGraph.h: Added.
(WTF::SingleRootGraphNode::rootName):
(WTF::SingleRootGraphNode::SingleRootGraphNode):
(WTF::SingleRootGraphNode::root):
(WTF::SingleRootGraphNode::operator== const):
(WTF::SingleRootGraphNode::operator!= const):
(WTF::SingleRootGraphNode::operator bool const):
(WTF::SingleRootGraphNode::isRoot const):
(WTF::SingleRootGraphNode::node const):
(WTF::SingleRootGraphSet::add):
(WTF::SingleRootGraphSet::remove):
(WTF::SingleRootGraphSet::contains):
(WTF::SingleRootGraphSet::dump const):
(WTF::SingleRootMap::SingleRootMap):
(WTF::SingleRootMap::clear):
(WTF::SingleRootMap::size const):
(WTF::SingleRootMap::operator[]):
(WTF::SingleRootMap::operator[] const):
(WTF::SingleRootGraph::SingleRootGraph):
(WTF::SingleRootGraph::root const):
(WTF::SingleRootGraph::newMap):
(WTF::SingleRootGraph::successors const):
(WTF::SingleRootGraph::predecessors const):
(WTF::SingleRootGraph::index const):
(WTF::SingleRootGraph::node const):
(WTF::SingleRootGraph::numNodes const):
(WTF::SingleRootGraph::dump const):
(WTF::SingleRootGraph::assertIsConsistent const):
Canonical link: https://commits.webkit.org/192644@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@221196 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2017-08-25 18:26:15 +00:00
|
|
|
List worklist = from;
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
while (!worklist.isEmpty()) {
|
|
|
|
|
typename Graph::Node block = worklist.takeLast();
|
|
|
|
|
forAllBlocksInDominanceFrontierOfImpl(
|
|
|
|
|
block,
|
|
|
|
|
[&] (typename Graph::Node otherBlock) {
|
|
|
|
|
if (functor(otherBlock))
|
|
|
|
|
worklist.append(otherBlock);
|
|
|
|
|
});
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
struct BlockData {
|
2019-08-12 20:57:15 +00:00
|
|
|
WTF_MAKE_STRUCT_FAST_ALLOCATED;
|
|
|
|
|
|
Dominators should be factored out of the DFG
https://bugs.webkit.org/show_bug.cgi?id=150764
Reviewed by Geoffrey Garen.
Source/JavaScriptCore:
Factored DFGDominators.h/DFGDominators.cpp into WTF. To do this, I made two changes to the
DFG:
1) DFG now has a CFG abstraction called DFG::CFG. The cool thing about this is that in the
future if we wanted to support inverted dominators, we could do it by just creating a
DFG::BackwardCFG.
2) Got rid of DFG::Analysis. From now on, an Analysis being invalidated is expressed by the
DFG::Graph having a null pointer for that analysis. When we "run" the analysis, we
just instantiate it. This makes it much more natural to integrate WTF::Dominators into
the DFG.
* CMakeLists.txt:
* JavaScriptCore.xcodeproj/project.pbxproj:
* dfg/DFGAnalysis.h: Removed.
* dfg/DFGCFG.h: Added.
(JSC::DFG::CFG::CFG):
(JSC::DFG::CFG::root):
(JSC::DFG::CFG::newMap<T>):
(JSC::DFG::CFG::successors):
(JSC::DFG::CFG::predecessors):
(JSC::DFG::CFG::index):
(JSC::DFG::CFG::node):
(JSC::DFG::CFG::numNodes):
(JSC::DFG::CFG::dump):
* dfg/DFGCSEPhase.cpp:
* dfg/DFGDisassembler.cpp:
(JSC::DFG::Disassembler::createDumpList):
* dfg/DFGDominators.cpp: Removed.
* dfg/DFGDominators.h:
(JSC::DFG::Dominators::Dominators):
(JSC::DFG::Dominators::strictlyDominates): Deleted.
(JSC::DFG::Dominators::dominates): Deleted.
(JSC::DFG::Dominators::immediateDominatorOf): Deleted.
(JSC::DFG::Dominators::forAllStrictDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllDominatorsOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksStrictlyDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksDominatedBy): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf): Deleted.
(JSC::DFG::Dominators::forAllBlocksInDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl): Deleted.
(JSC::DFG::Dominators::BlockData::BlockData): Deleted.
* dfg/DFGEdgeDominates.h:
(JSC::DFG::EdgeDominates::operator()):
* dfg/DFGGraph.cpp:
(JSC::DFG::Graph::Graph):
(JSC::DFG::Graph::dumpBlockHeader):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::substituteGetLocal):
(JSC::DFG::Graph::handleAssertionFailure):
(JSC::DFG::Graph::ensureDominators):
(JSC::DFG::Graph::ensurePrePostNumbering):
(JSC::DFG::Graph::ensureNaturalLoops):
(JSC::DFG::Graph::valueProfileFor):
* dfg/DFGGraph.h:
(JSC::DFG::Graph::hasDebuggerEnabled):
* dfg/DFGLICMPhase.cpp:
(JSC::DFG::LICMPhase::run):
(JSC::DFG::LICMPhase::attemptHoist):
* dfg/DFGLoopPreHeaderCreationPhase.cpp:
(JSC::DFG::createPreHeader):
(JSC::DFG::LoopPreHeaderCreationPhase::run):
* dfg/DFGNaturalLoops.cpp:
(JSC::DFG::NaturalLoop::dump):
(JSC::DFG::NaturalLoops::NaturalLoops):
(JSC::DFG::NaturalLoops::~NaturalLoops):
(JSC::DFG::NaturalLoops::loopsOf):
(JSC::DFG::NaturalLoops::computeDependencies): Deleted.
(JSC::DFG::NaturalLoops::compute): Deleted.
* dfg/DFGNaturalLoops.h:
(JSC::DFG::NaturalLoops::numLoops):
* dfg/DFGNode.h:
(JSC::DFG::Node::SuccessorsIterable::end):
(JSC::DFG::Node::SuccessorsIterable::size):
(JSC::DFG::Node::SuccessorsIterable::at):
(JSC::DFG::Node::SuccessorsIterable::operator[]):
* dfg/DFGOSREntrypointCreationPhase.cpp:
(JSC::DFG::OSREntrypointCreationPhase::run):
* dfg/DFGObjectAllocationSinkingPhase.cpp:
* dfg/DFGPlan.cpp:
(JSC::DFG::Plan::compileInThreadImpl):
* dfg/DFGPrePostNumbering.cpp:
(JSC::DFG::PrePostNumbering::PrePostNumbering):
(JSC::DFG::PrePostNumbering::~PrePostNumbering):
(JSC::DFG::PrePostNumbering::compute): Deleted.
* dfg/DFGPrePostNumbering.h:
(JSC::DFG::PrePostNumbering::preNumber):
(JSC::DFG::PrePostNumbering::postNumber):
* dfg/DFGPutStackSinkingPhase.cpp:
* dfg/DFGSSACalculator.cpp:
(JSC::DFG::SSACalculator::nonLocalReachingDef):
(JSC::DFG::SSACalculator::reachingDefAtTail):
* dfg/DFGSSACalculator.h:
(JSC::DFG::SSACalculator::computePhis):
* dfg/DFGSSAConversionPhase.cpp:
(JSC::DFG::SSAConversionPhase::run):
* ftl/FTLLink.cpp:
(JSC::FTL::link):
* ftl/FTLLowerDFGToLLVM.cpp:
(JSC::FTL::DFG::LowerDFGToLLVM::lower):
(JSC::FTL::DFG::LowerDFGToLLVM::safelyInvalidateAfterTermination):
(JSC::FTL::DFG::LowerDFGToLLVM::isValid):
Source/WTF:
This takes what used to be DFGDominators.h/DFGDominators.cpp and turns it into a generic
algorithm that can take any abstract graph. The idea is that you create Dominators<CFG> and
pass it a CFG object, which defines the types of graph nodes and methods for getting
successors, predecessors, etc. The DFG now defines a class called CFG, which is a wrapper for
DFG::Graph that conforms to the thing that wtf/Dominators.h expects.
When doing things to graphs, it's common to refer to the things in the graph as "nodes".
Because I intend to reuse the CFG abstraction with many graph algorithms, that abstraction uses
the term "node" to refer to a DFG basic block. But in Dominators, the method and variable names
still use "block". This is because although Dominators are applicable to any kind of directed
graph, it's super unlikely that we will ever use them for anything but compilers. Indeed, the
only reason why I'm factoring them out of the DFG is so that I can use them with B3 and Air.
This has the nice side effect that a user of WTF::Dominators<JSC::DFG::CFG> will see familiar
terminology like "blocksStrictlyDominatedBy(...)" instead of "nodesStrictlyDominatedBy(...)",
which would be super confusing.
Overall, wtf/Dominators.h is a combination of what used to be in DFGDominators.h,
DFGDominators.cpp, DFGNaiveDominators.h, and DFGNaiveDominators.cpp. I only changed code when I
had to in order to make it generic.
* WTF.xcodeproj/project.pbxproj:
* wtf/CMakeLists.txt:
* wtf/Dominators.h: Added.
(WTF::Dominators::Dominators):
(WTF::Dominators::compute):
(WTF::Dominators::strictlyDominates):
(WTF::Dominators::dominates):
(WTF::Dominators::immediateDominatorOf):
(WTF::Dominators::forAllStrictDominatorsOf):
(WTF::Dominators::forAllDominatorsOf):
(WTF::Dominators::forAllBlocksStrictlyDominatedBy):
(WTF::Dominators::forAllBlocksDominatedBy):
(WTF::Dominators::strictDominatorsOf):
(WTF::Dominators::dominatorsOf):
(WTF::Dominators::blocksStrictlyDominatedBy):
(WTF::Dominators::blocksDominatedBy):
(WTF::Dominators::forAllBlocksInDominanceFrontierOf):
(WTF::Dominators::dominanceFrontierOf):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOf):
(WTF::Dominators::forAllBlocksInPrunedIteratedDominanceFrontierOf):
(WTF::Dominators::iteratedDominanceFrontierOf):
(WTF::Dominators::dump):
(WTF::Dominators::LengauerTarjan::LengauerTarjan):
(WTF::Dominators::LengauerTarjan::compute):
(WTF::Dominators::LengauerTarjan::immediateDominator):
(WTF::Dominators::LengauerTarjan::computeDepthFirstPreNumbering):
(WTF::Dominators::LengauerTarjan::computeSemiDominatorsAndImplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::computeExplicitImmediateDominators):
(WTF::Dominators::LengauerTarjan::link):
(WTF::Dominators::LengauerTarjan::eval):
(WTF::Dominators::LengauerTarjan::compress):
(WTF::Dominators::LengauerTarjan::BlockData::BlockData):
(WTF::Dominators::NaiveDominators::NaiveDominators):
(WTF::Dominators::NaiveDominators::dominates):
(WTF::Dominators::NaiveDominators::dump):
(WTF::Dominators::NaiveDominators::pruneDominators):
(WTF::Dominators::ValidationContext::ValidationContext):
(WTF::Dominators::ValidationContext::reportError):
(WTF::Dominators::ValidationContext::handleErrors):
(WTF::Dominators::naiveDominates):
(WTF::Dominators::forAllBlocksInDominanceFrontierOfImpl):
(WTF::Dominators::forAllBlocksInIteratedDominanceFrontierOfImpl):
(WTF::Dominators::BlockData::BlockData):
Canonical link: https://commits.webkit.org/168962@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@191870 268f45cc-cd09-0410-ab3c-d52691b4dbfc
2015-11-02 00:48:03 +00:00
|
|
|
BlockData()
|
|
|
|
|
: idomParent(nullptr)
|
|
|
|
|
, preNumber(UINT_MAX)
|
|
|
|
|
, postNumber(UINT_MAX)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
Vector<typename Graph::Node> idomKids;
|
|
|
|
|
typename Graph::Node idomParent;
|
|
|
|
|
|
|
|
|
|
unsigned preNumber;
|
|
|
|
|
unsigned postNumber;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
Graph& m_graph;
|
|
|
|
|
typename Graph::template Map<BlockData> m_data;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
} // namespace WTF
|
|
|
|
|
|
|
|
|
|
using WTF::Dominators;
|