Changeset 221602 in webkit for trunk/Source/JavaScriptCore/dfg/DFGJITCode.cpp

Timestamp:

Sep 4, 2017, 8:21:33 PM (8 years ago)

Author:

[email protected]

Message:

Support compiling catch in the FTL
​https://bugs.webkit.org/show_bug.cgi?id=175396

Reviewed by Filip Pizlo.

This patch implements op_catch in the FTL. It extends the DFG implementation
by supporting multiple entrypoints in DFG-SSA. This patch implements this
by introducing an EntrySwitch node. When converting to SSA, we introduce a new
root block with an EntrySwitch that has the previous DFG entrypoints as its
successors. By convention, we pick the zeroth entry point index to be the
op_enter entrypoint. Like in B3, in DFG-SSA, EntrySwitch just acts like a
switch over the entrypoint index argument. DFG::EntrySwitch in the FTL
simply lowers to B3::EntrySwitch. The EntrySwitch in the root block that
SSAConversion creates can not exit because we would both not know where to exit
to in the program: we would not have valid OSR exit state. This design also
mandates that anything we hoist above EntrySwitch in the new root block
can not exit since they also do not have valid OSR exit state.

This patch also adds a new metadata node named InitializeEntrypointArguments.
InitializeEntrypointArguments is a metadata node that initializes the flush format for
the arguments at a given entrypoint. For a given entrypoint index, this node
tells AI and OSRAvailabilityAnalysis what the flush format for each argument
is. This allows each individual entrypoint to have an independent set of
argument types. Currently, this won't happen in practice because ArgumentPosition
unifies flush formats, but this is an implementation detail we probably want
to modify in the future. SSAConversion will add InitializeEntrypointArguments
to the beginning of each of the original DFG entrypoint blocks.

This patch also adds the ability to specify custom prologue code generators in Air.
This allows the FTL to specify a custom prologue for catch entrypoints that
matches the op_catch OSR entry calling convention that the DFG uses. This way,
the baseline JIT code OSR enters into op_catch the same way both in the DFG
and the FTL. In the future, we can use this same mechanism to perform stack
overflow checks instead of using a patchpoint.

• b3/air/AirCode.cpp:

(JSC::B3::Air::Code::isEntrypoint):
(JSC::B3::Air::Code::entrypointIndex):

• b3/air/AirCode.h:

(JSC::B3::Air::Code::setPrologueForEntrypoint):
(JSC::B3::Air::Code::prologueGeneratorForEntrypoint):

• b3/air/AirGenerate.cpp:

(JSC::B3::Air::generate):

• dfg/DFGAbstractInterpreterInlines.h:

(JSC::DFG::AbstractInterpreter<AbstractStateType>::executeEffects):

• dfg/DFGBasicBlock.h:
• dfg/DFGByteCodeParser.cpp:

(JSC::DFG::ByteCodeParser::parseBlock):
(JSC::DFG::ByteCodeParser::parse):

• dfg/DFGCFG.h:

(JSC::DFG::selectCFG):

• dfg/DFGClobberize.h:

(JSC::DFG::clobberize):

• dfg/DFGClobbersExitState.cpp:

(JSC::DFG::clobbersExitState):

• dfg/DFGCommonData.cpp:

(JSC::DFG::CommonData::shrinkToFit):
(JSC::DFG::CommonData::finalizeCatchEntrypoints):

• dfg/DFGCommonData.h:

(JSC::DFG::CommonData::catchOSREntryDataForBytecodeIndex):
(JSC::DFG::CommonData::appendCatchEntrypoint):

• dfg/DFGDoesGC.cpp:

(JSC::DFG::doesGC):

• dfg/DFGFixupPhase.cpp:

(JSC::DFG::FixupPhase::fixupNode):

• dfg/DFGGraph.cpp:

(JSC::DFG::Graph::dump):
(JSC::DFG::Graph::invalidateCFG):
(JSC::DFG::Graph::ensureCPSCFG):
(JSC::DFG::Graph::methodOfGettingAValueProfileFor):

• dfg/DFGGraph.h:

(JSC::DFG::Graph::isEntrypoint):

• dfg/DFGInPlaceAbstractState.cpp:

(JSC::DFG::InPlaceAbstractState::initialize):
(JSC::DFG::InPlaceAbstractState::mergeToSuccessors):

• dfg/DFGJITCode.cpp:

(JSC::DFG::JITCode::shrinkToFit):
(JSC::DFG::JITCode::finalizeOSREntrypoints):

• dfg/DFGJITCode.h:

(JSC::DFG::JITCode::catchOSREntryDataForBytecodeIndex): Deleted.
(JSC::DFG::JITCode::appendCatchEntrypoint): Deleted.

• dfg/DFGJITCompiler.cpp:

(JSC::DFG::JITCompiler::noticeCatchEntrypoint):
(JSC::DFG::JITCompiler::makeCatchOSREntryBuffer):

• dfg/DFGMayExit.cpp:
• dfg/DFGNode.h:

(JSC::DFG::Node::isEntrySwitch):
(JSC::DFG::Node::isTerminal):
(JSC::DFG::Node::entrySwitchData):
(JSC::DFG::Node::numSuccessors):
(JSC::DFG::Node::successor):
(JSC::DFG::Node::entrypointIndex):

• dfg/DFGNodeType.h:
• dfg/DFGOSRAvailabilityAnalysisPhase.cpp:

(JSC::DFG::OSRAvailabilityAnalysisPhase::run):
(JSC::DFG::LocalOSRAvailabilityCalculator::executeNode):

• dfg/DFGOSREntry.cpp:

(JSC::DFG::prepareCatchOSREntry):

• dfg/DFGOSREntry.h:
• dfg/DFGOSREntrypointCreationPhase.cpp:

(JSC::DFG::OSREntrypointCreationPhase::run):

• dfg/DFGPredictionPropagationPhase.cpp:
• dfg/DFGSSAConversionPhase.cpp:

(JSC::DFG::SSAConversionPhase::SSAConversionPhase):
(JSC::DFG::SSAConversionPhase::run):

• dfg/DFGSafeToExecute.h:

(JSC::DFG::safeToExecute):

• dfg/DFGSpeculativeJIT.cpp:

(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/DFGValidate.cpp:
• ftl/FTLCapabilities.cpp:

(JSC::FTL::canCompile):

• ftl/FTLCompile.cpp:

(JSC::FTL::compile):

• ftl/FTLLowerDFGToB3.cpp:

(JSC::FTL::DFG::LowerDFGToB3::lower):
(JSC::FTL::DFG::LowerDFGToB3::compileNode):
(JSC::FTL::DFG::LowerDFGToB3::compileExtractCatchLocal):
(JSC::FTL::DFG::LowerDFGToB3::compileGetStack):
(JSC::FTL::DFG::LowerDFGToB3::compileEntrySwitch):
(JSC::FTL::DFG::LowerDFGToB3::speculate):
(JSC::FTL::DFG::LowerDFGToB3::appendOSRExitDescriptor):
(JSC::FTL::DFG::LowerDFGToB3::appendOSRExit):
(JSC::FTL::DFG::LowerDFGToB3::blessSpeculation):

• ftl/FTLOutput.cpp:

(JSC::FTL::Output::entrySwitch):

• ftl/FTLOutput.h:
• jit/JITOperations.cpp:

File:

• trunk/Source/JavaScriptCore/dfg/DFGJITCode.cpp (modified) (3 diffs)

trunk/Source/JavaScriptCore/dfg/DFGJITCode.cpp

@@ -64 +64 @@
     osrEntry.shrinkToFit();
     osrExit.shrinkToFit();
-    catchEntrypoints.shrinkToFit();
     speculationRecovery.shrinkToFit();
     minifiedDFG.prepareAndShrink();
@@ -244 +243 @@
         return a.m_bytecodeIndex < b.m_bytecodeIndex;
     };
-    std::sort(catchEntrypoints.begin(), catchEntrypoints.end(), comparator);
     std::sort(osrEntry.begin(), osrEntry.end(), comparator);
 
@@ -252 +250 @@
             ASSERT(osrVector[i].m_bytecodeIndex <= osrVector[i + 1].m_bytecodeIndex);
     };
-    verifyIsSorted(catchEntrypoints);
     verifyIsSorted(osrEntry);
 #endif