/* * Copyright (C) 2008, 2009 Apple Inc. All rights reserved. * * 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. */ #include "config.h" #if !USE(JSVALUE32_64) #include "JIT.h" #if ENABLE(JIT) #include "CodeBlock.h" #include "GetterSetter.h" #include "JITInlineMethods.h" #include "JITStubCall.h" #include "JSArray.h" #include "JSFunction.h" #include "JSPropertyNameIterator.h" #include "Interpreter.h" #include "LinkBuffer.h" #include "RepatchBuffer.h" #include "ResultType.h" #include "SamplingTool.h" #ifndef NDEBUG #include #endif using namespace std; namespace JSC { void JIT::emit_op_get_by_val(Instruction* currentInstruction) { unsigned dst = currentInstruction[1].u.operand; unsigned base = currentInstruction[2].u.operand; unsigned property = currentInstruction[3].u.operand; emitGetVirtualRegisters(base, regT0, property, regT1); emitJumpSlowCaseIfNotImmediateInteger(regT1); #if USE(JSVALUE64) // This is technically incorrect - we're zero-extending an int32. On the hot path this doesn't matter. // We check the value as if it was a uint32 against the m_vectorLength - which will always fail if // number was signed since m_vectorLength is always less than intmax (since the total allocation // size is always less than 4Gb). As such zero extending wil have been correct (and extending the value // to 64-bits is necessary since it's used in the address calculation. We zero extend rather than sign // extending since it makes it easier to re-tag the value in the slow case. zeroExtend32ToPtr(regT1, regT1); #else emitFastArithImmToInt(regT1); #endif emitJumpSlowCaseIfNotJSCell(regT0, base); addSlowCase(branchPtr(NotEqual, Address(regT0), ImmPtr(m_globalData->jsArrayVPtr))); loadPtr(Address(regT0, OBJECT_OFFSETOF(JSArray, m_storage)), regT2); addSlowCase(branch32(AboveOrEqual, regT1, Address(regT0, OBJECT_OFFSETOF(JSArray, m_vectorLength)))); loadPtr(BaseIndex(regT2, regT1, ScalePtr, OBJECT_OFFSETOF(ArrayStorage, m_vector[0])), regT0); addSlowCase(branchTestPtr(Zero, regT0)); emitPutVirtualRegister(dst); } void JIT::compileGetDirectOffset(RegisterID base, RegisterID result, RegisterID structure, RegisterID offset, RegisterID scratch) { ASSERT(sizeof(((Structure*)0)->m_propertyStorageCapacity) == sizeof(int32_t)); ASSERT(sizeof(JSObject::inlineStorageCapacity) == sizeof(int32_t)); Jump notUsingInlineStorage = branch32(NotEqual, Address(structure, OBJECT_OFFSETOF(Structure, m_propertyStorageCapacity)), Imm32(JSObject::inlineStorageCapacity)); loadPtr(BaseIndex(base, offset, ScalePtr, OBJECT_OFFSETOF(JSObject, m_inlineStorage)), result); Jump finishedLoad = jump(); notUsingInlineStorage.link(this); loadPtr(Address(base, OBJECT_OFFSETOF(JSObject, m_externalStorage)), scratch); loadPtr(BaseIndex(scratch, offset, ScalePtr, 0), result); finishedLoad.link(this); } void JIT::emit_op_get_by_pname(Instruction* currentInstruction) { unsigned dst = currentInstruction[1].u.operand; unsigned base = currentInstruction[2].u.operand; unsigned property = currentInstruction[3].u.operand; unsigned expected = currentInstruction[4].u.operand; unsigned iter = currentInstruction[5].u.operand; unsigned i = currentInstruction[6].u.operand; emitGetVirtualRegister(property, regT0); addSlowCase(branchPtr(NotEqual, regT0, addressFor(expected))); emitGetVirtualRegisters(base, regT0, iter, regT1); emitJumpSlowCaseIfNotJSCell(regT0, base); // Test base's structure loadPtr(Address(regT0, OBJECT_OFFSETOF(JSCell, m_structure)), regT2); addSlowCase(branchPtr(NotEqual, regT2, Address(regT1, OBJECT_OFFSETOF(JSPropertyNameIterator, m_cachedStructure)))); load32(addressFor(i), regT3); sub32(Imm32(1), regT3); addSlowCase(branch32(AboveOrEqual, regT3, Address(regT1, OBJECT_OFFSETOF(JSPropertyNameIterator, m_numCacheableSlots)))); compileGetDirectOffset(regT0, regT0, regT2, regT3, regT1); emitPutVirtualRegister(dst, regT0); } void JIT::emitSlow_op_get_by_pname(Instruction* currentInstruction, Vector::iterator& iter) { unsigned dst = currentInstruction[1].u.operand; unsigned base = currentInstruction[2].u.operand; unsigned property = currentInstruction[3].u.operand; linkSlowCase(iter); linkSlowCaseIfNotJSCell(iter, base); linkSlowCase(iter); linkSlowCase(iter); JITStubCall stubCall(this, cti_op_get_by_val); stubCall.addArgument(base, regT2); stubCall.addArgument(property, regT2); stubCall.call(dst); } void JIT::emit_op_put_by_val(Instruction* currentInstruction) { unsigned base = currentInstruction[1].u.operand; unsigned property = currentInstruction[2].u.operand; unsigned value = currentInstruction[3].u.operand; emitGetVirtualRegisters(base, regT0, property, regT1); emitJumpSlowCaseIfNotImmediateInteger(regT1); #if USE(JSVALUE64) // See comment in op_get_by_val. zeroExtend32ToPtr(regT1, regT1); #else emitFastArithImmToInt(regT1); #endif emitJumpSlowCaseIfNotJSCell(regT0, base); addSlowCase(branchPtr(NotEqual, Address(regT0), ImmPtr(m_globalData->jsArrayVPtr))); addSlowCase(branch32(AboveOrEqual, regT1, Address(regT0, OBJECT_OFFSETOF(JSArray, m_vectorLength)))); loadPtr(Address(regT0, OBJECT_OFFSETOF(JSArray, m_storage)), regT2); Jump empty = branchTestPtr(Zero, BaseIndex(regT2, regT1, ScalePtr, OBJECT_OFFSETOF(ArrayStorage, m_vector[0]))); Label storeResult(this); emitGetVirtualRegister(value, regT0); storePtr(regT0, BaseIndex(regT2, regT1, ScalePtr, OBJECT_OFFSETOF(ArrayStorage, m_vector[0]))); Jump end = jump(); empty.link(this); add32(Imm32(1), Address(regT2, OBJECT_OFFSETOF(ArrayStorage, m_numValuesInVector))); branch32(Below, regT1, Address(regT2, OBJECT_OFFSETOF(ArrayStorage, m_length))).linkTo(storeResult, this); move(regT1, regT0); add32(Imm32(1), regT0); store32(regT0, Address(regT2, OBJECT_OFFSETOF(ArrayStorage, m_length))); jump().linkTo(storeResult, this); end.link(this); } void JIT::emit_op_put_by_index(Instruction* currentInstruction) { JITStubCall stubCall(this, cti_op_put_by_index); stubCall.addArgument(currentInstruction[1].u.operand, regT2); stubCall.addArgument(Imm32(currentInstruction[2].u.operand)); stubCall.addArgument(currentInstruction[3].u.operand, regT2); stubCall.call(); } void JIT::emit_op_put_getter(Instruction* currentInstruction) { JITStubCall stubCall(this, cti_op_put_getter); stubCall.addArgument(currentInstruction[1].u.operand, regT2); stubCall.addArgument(ImmPtr(&m_codeBlock->identifier(currentInstruction[2].u.operand))); stubCall.addArgument(currentInstruction[3].u.operand, regT2); stubCall.call(); } void JIT::emit_op_put_setter(Instruction* currentInstruction) { JITStubCall stubCall(this, cti_op_put_setter); stubCall.addArgument(currentInstruction[1].u.operand, regT2); stubCall.addArgument(ImmPtr(&m_codeBlock->identifier(currentInstruction[2].u.operand))); stubCall.addArgument(currentInstruction[3].u.operand, regT2); stubCall.call(); } void JIT::emit_op_del_by_id(Instruction* currentInstruction) { JITStubCall stubCall(this, cti_op_del_by_id); stubCall.addArgument(currentInstruction[2].u.operand, regT2); stubCall.addArgument(ImmPtr(&m_codeBlock->identifier(currentInstruction[3].u.operand))); stubCall.call(currentInstruction[1].u.operand); } #if !ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS) /* ------------------------------ BEGIN: !ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS) ------------------------------ */ // Treat these as nops - the call will be handed as a regular get_by_id/op_call pair. void JIT::emit_op_method_check(Instruction*) {} void JIT::emitSlow_op_method_check(Instruction*, Vector::iterator&) { ASSERT_NOT_REACHED(); } #if ENABLE(JIT_OPTIMIZE_METHOD_CALLS) #error "JIT_OPTIMIZE_METHOD_CALLS requires JIT_OPTIMIZE_PROPERTY_ACCESS" #endif void JIT::emit_op_get_by_id(Instruction* currentInstruction) { unsigned resultVReg = currentInstruction[1].u.operand; unsigned baseVReg = currentInstruction[2].u.operand; Identifier* ident = &(m_codeBlock->identifier(currentInstruction[3].u.operand)); emitGetVirtualRegister(baseVReg, regT0); JITStubCall stubCall(this, cti_op_get_by_id_generic); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(ident)); stubCall.call(resultVReg); m_propertyAccessInstructionIndex++; } void JIT::emitSlow_op_get_by_id(Instruction*, Vector::iterator&) { ASSERT_NOT_REACHED(); } void JIT::emit_op_put_by_id(Instruction* currentInstruction) { unsigned baseVReg = currentInstruction[1].u.operand; Identifier* ident = &(m_codeBlock->identifier(currentInstruction[2].u.operand)); unsigned valueVReg = currentInstruction[3].u.operand; emitGetVirtualRegisters(baseVReg, regT0, valueVReg, regT1); JITStubCall stubCall(this, cti_op_put_by_id_generic); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(ident)); stubCall.addArgument(regT1); stubCall.call(); m_propertyAccessInstructionIndex++; } void JIT::emitSlow_op_put_by_id(Instruction*, Vector::iterator&) { ASSERT_NOT_REACHED(); } #else // !ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS) /* ------------------------------ BEGIN: ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS) ------------------------------ */ #if ENABLE(JIT_OPTIMIZE_METHOD_CALLS) void JIT::emit_op_method_check(Instruction* currentInstruction) { // Assert that the following instruction is a get_by_id. ASSERT(m_interpreter->getOpcodeID((currentInstruction + OPCODE_LENGTH(op_method_check))->u.opcode) == op_get_by_id); currentInstruction += OPCODE_LENGTH(op_method_check); unsigned resultVReg = currentInstruction[1].u.operand; unsigned baseVReg = currentInstruction[2].u.operand; Identifier* ident = &(m_codeBlock->identifier(currentInstruction[3].u.operand)); emitGetVirtualRegister(baseVReg, regT0); // Do the method check - check the object & its prototype's structure inline (this is the common case). m_methodCallCompilationInfo.append(MethodCallCompilationInfo(m_propertyAccessInstructionIndex)); MethodCallCompilationInfo& info = m_methodCallCompilationInfo.last(); Jump notCell = emitJumpIfNotJSCell(regT0); BEGIN_UNINTERRUPTED_SEQUENCE(sequenceMethodCheck); Jump structureCheck = branchPtrWithPatch(NotEqual, Address(regT0, OBJECT_OFFSETOF(JSCell, m_structure)), info.structureToCompare, ImmPtr(reinterpret_cast(patchGetByIdDefaultStructure))); DataLabelPtr protoStructureToCompare, protoObj = moveWithPatch(ImmPtr(0), regT1); Jump protoStructureCheck = branchPtrWithPatch(NotEqual, Address(regT1, OBJECT_OFFSETOF(JSCell, m_structure)), protoStructureToCompare, ImmPtr(reinterpret_cast(patchGetByIdDefaultStructure))); // This will be relinked to load the function without doing a load. DataLabelPtr putFunction = moveWithPatch(ImmPtr(0), regT0); END_UNINTERRUPTED_SEQUENCE(sequenceMethodCheck); Jump match = jump(); ASSERT_JIT_OFFSET(differenceBetween(info.structureToCompare, protoObj), patchOffsetMethodCheckProtoObj); ASSERT_JIT_OFFSET(differenceBetween(info.structureToCompare, protoStructureToCompare), patchOffsetMethodCheckProtoStruct); ASSERT_JIT_OFFSET(differenceBetween(info.structureToCompare, putFunction), patchOffsetMethodCheckPutFunction); // Link the failure cases here. notCell.link(this); structureCheck.link(this); protoStructureCheck.link(this); // Do a regular(ish) get_by_id (the slow case will be link to // cti_op_get_by_id_method_check instead of cti_op_get_by_id. compileGetByIdHotPath(resultVReg, baseVReg, ident, m_propertyAccessInstructionIndex++); match.link(this); emitPutVirtualRegister(resultVReg); // We've already generated the following get_by_id, so make sure it's skipped over. m_bytecodeIndex += OPCODE_LENGTH(op_get_by_id); } void JIT::emitSlow_op_method_check(Instruction* currentInstruction, Vector::iterator& iter) { currentInstruction += OPCODE_LENGTH(op_method_check); unsigned resultVReg = currentInstruction[1].u.operand; unsigned baseVReg = currentInstruction[2].u.operand; Identifier* ident = &(m_codeBlock->identifier(currentInstruction[3].u.operand)); compileGetByIdSlowCase(resultVReg, baseVReg, ident, iter, true); // We've already generated the following get_by_id, so make sure it's skipped over. m_bytecodeIndex += OPCODE_LENGTH(op_get_by_id); } #else //!ENABLE(JIT_OPTIMIZE_METHOD_CALLS) // Treat these as nops - the call will be handed as a regular get_by_id/op_call pair. void JIT::emit_op_method_check(Instruction*) {} void JIT::emitSlow_op_method_check(Instruction*, Vector::iterator&) { ASSERT_NOT_REACHED(); } #endif void JIT::emit_op_get_by_id(Instruction* currentInstruction) { unsigned resultVReg = currentInstruction[1].u.operand; unsigned baseVReg = currentInstruction[2].u.operand; Identifier* ident = &(m_codeBlock->identifier(currentInstruction[3].u.operand)); emitGetVirtualRegister(baseVReg, regT0); compileGetByIdHotPath(resultVReg, baseVReg, ident, m_propertyAccessInstructionIndex++); emitPutVirtualRegister(resultVReg); } void JIT::compileGetByIdHotPath(int, int baseVReg, Identifier*, unsigned propertyAccessInstructionIndex) { // As for put_by_id, get_by_id requires the offset of the Structure and the offset of the access to be patched. // Additionally, for get_by_id we need patch the offset of the branch to the slow case (we patch this to jump // to array-length / prototype access tranpolines, and finally we also the the property-map access offset as a label // to jump back to if one of these trampolies finds a match. emitJumpSlowCaseIfNotJSCell(regT0, baseVReg); BEGIN_UNINTERRUPTED_SEQUENCE(sequenceGetByIdHotPath); Label hotPathBegin(this); m_propertyAccessCompilationInfo[propertyAccessInstructionIndex].hotPathBegin = hotPathBegin; DataLabelPtr structureToCompare; Jump structureCheck = branchPtrWithPatch(NotEqual, Address(regT0, OBJECT_OFFSETOF(JSCell, m_structure)), structureToCompare, ImmPtr(reinterpret_cast(patchGetByIdDefaultStructure))); addSlowCase(structureCheck); ASSERT_JIT_OFFSET(differenceBetween(hotPathBegin, structureToCompare), patchOffsetGetByIdStructure); ASSERT_JIT_OFFSET(differenceBetween(hotPathBegin, structureCheck), patchOffsetGetByIdBranchToSlowCase) Label externalLoad = loadPtrWithPatchToLEA(Address(regT0, OBJECT_OFFSETOF(JSObject, m_externalStorage)), regT0); Label externalLoadComplete(this); ASSERT_JIT_OFFSET(differenceBetween(hotPathBegin, externalLoad), patchOffsetGetByIdExternalLoad); ASSERT_JIT_OFFSET(differenceBetween(externalLoad, externalLoadComplete), patchLengthGetByIdExternalLoad); DataLabel32 displacementLabel = loadPtrWithAddressOffsetPatch(Address(regT0, patchGetByIdDefaultOffset), regT0); ASSERT_JIT_OFFSET(differenceBetween(hotPathBegin, displacementLabel), patchOffsetGetByIdPropertyMapOffset); Label putResult(this); END_UNINTERRUPTED_SEQUENCE(sequenceGetByIdHotPath); ASSERT_JIT_OFFSET(differenceBetween(hotPathBegin, putResult), patchOffsetGetByIdPutResult); } void JIT::emitSlow_op_get_by_id(Instruction* currentInstruction, Vector::iterator& iter) { unsigned resultVReg = currentInstruction[1].u.operand; unsigned baseVReg = currentInstruction[2].u.operand; Identifier* ident = &(m_codeBlock->identifier(currentInstruction[3].u.operand)); compileGetByIdSlowCase(resultVReg, baseVReg, ident, iter, false); } void JIT::compileGetByIdSlowCase(int resultVReg, int baseVReg, Identifier* ident, Vector::iterator& iter, bool isMethodCheck) { // As for the hot path of get_by_id, above, we ensure that we can use an architecture specific offset // so that we only need track one pointer into the slow case code - we track a pointer to the location // of the call (which we can use to look up the patch information), but should a array-length or // prototype access trampoline fail we want to bail out back to here. To do so we can subtract back // the distance from the call to the head of the slow case. linkSlowCaseIfNotJSCell(iter, baseVReg); linkSlowCase(iter); BEGIN_UNINTERRUPTED_SEQUENCE(sequenceGetByIdSlowCase); #ifndef NDEBUG Label coldPathBegin(this); #endif JITStubCall stubCall(this, isMethodCheck ? cti_op_get_by_id_method_check : cti_op_get_by_id); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(ident)); Call call = stubCall.call(resultVReg); END_UNINTERRUPTED_SEQUENCE(sequenceGetByIdSlowCase); ASSERT_JIT_OFFSET(differenceBetween(coldPathBegin, call), patchOffsetGetByIdSlowCaseCall); // Track the location of the call; this will be used to recover patch information. m_propertyAccessCompilationInfo[m_propertyAccessInstructionIndex].callReturnLocation = call; m_propertyAccessInstructionIndex++; } void JIT::emit_op_put_by_id(Instruction* currentInstruction) { unsigned baseVReg = currentInstruction[1].u.operand; unsigned valueVReg = currentInstruction[3].u.operand; unsigned propertyAccessInstructionIndex = m_propertyAccessInstructionIndex++; // In order to be able to patch both the Structure, and the object offset, we store one pointer, // to just after the arguments have been loaded into registers 'hotPathBegin', and we generate code // such that the Structure & offset are always at the same distance from this. emitGetVirtualRegisters(baseVReg, regT0, valueVReg, regT1); // Jump to a slow case if either the base object is an immediate, or if the Structure does not match. emitJumpSlowCaseIfNotJSCell(regT0, baseVReg); BEGIN_UNINTERRUPTED_SEQUENCE(sequencePutById); Label hotPathBegin(this); m_propertyAccessCompilationInfo[propertyAccessInstructionIndex].hotPathBegin = hotPathBegin; // It is important that the following instruction plants a 32bit immediate, in order that it can be patched over. DataLabelPtr structureToCompare; addSlowCase(branchPtrWithPatch(NotEqual, Address(regT0, OBJECT_OFFSETOF(JSCell, m_structure)), structureToCompare, ImmPtr(reinterpret_cast(patchGetByIdDefaultStructure)))); ASSERT_JIT_OFFSET(differenceBetween(hotPathBegin, structureToCompare), patchOffsetPutByIdStructure); // Plant a load from a bogus ofset in the object's property map; we will patch this later, if it is to be used. Label externalLoad = loadPtrWithPatchToLEA(Address(regT0, OBJECT_OFFSETOF(JSObject, m_externalStorage)), regT0); Label externalLoadComplete(this); ASSERT_JIT_OFFSET(differenceBetween(hotPathBegin, externalLoad), patchOffsetPutByIdExternalLoad); ASSERT_JIT_OFFSET(differenceBetween(externalLoad, externalLoadComplete), patchLengthPutByIdExternalLoad); DataLabel32 displacementLabel = storePtrWithAddressOffsetPatch(regT1, Address(regT0, patchGetByIdDefaultOffset)); END_UNINTERRUPTED_SEQUENCE(sequencePutById); ASSERT_JIT_OFFSET(differenceBetween(hotPathBegin, displacementLabel), patchOffsetPutByIdPropertyMapOffset); } void JIT::emitSlow_op_put_by_id(Instruction* currentInstruction, Vector::iterator& iter) { unsigned baseVReg = currentInstruction[1].u.operand; Identifier* ident = &(m_codeBlock->identifier(currentInstruction[2].u.operand)); unsigned propertyAccessInstructionIndex = m_propertyAccessInstructionIndex++; linkSlowCaseIfNotJSCell(iter, baseVReg); linkSlowCase(iter); JITStubCall stubCall(this, cti_op_put_by_id); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(ident)); stubCall.addArgument(regT1); Call call = stubCall.call(); // Track the location of the call; this will be used to recover patch information. m_propertyAccessCompilationInfo[propertyAccessInstructionIndex].callReturnLocation = call; } // Compile a store into an object's property storage. May overwrite the // value in objectReg. void JIT::compilePutDirectOffset(RegisterID base, RegisterID value, Structure* structure, size_t cachedOffset) { int offset = cachedOffset * sizeof(JSValue); if (structure->isUsingInlineStorage()) offset += OBJECT_OFFSETOF(JSObject, m_inlineStorage); else loadPtr(Address(base, OBJECT_OFFSETOF(JSObject, m_externalStorage)), base); storePtr(value, Address(base, offset)); } // Compile a load from an object's property storage. May overwrite base. void JIT::compileGetDirectOffset(RegisterID base, RegisterID result, Structure* structure, size_t cachedOffset) { int offset = cachedOffset * sizeof(JSValue); if (structure->isUsingInlineStorage()) offset += OBJECT_OFFSETOF(JSObject, m_inlineStorage); else loadPtr(Address(base, OBJECT_OFFSETOF(JSObject, m_externalStorage)), base); loadPtr(Address(base, offset), result); } void JIT::compileGetDirectOffset(JSObject* base, RegisterID temp, RegisterID result, size_t cachedOffset) { if (base->isUsingInlineStorage()) loadPtr(static_cast(&base->m_inlineStorage[cachedOffset]), result); else { PropertyStorage* protoPropertyStorage = &base->m_externalStorage; loadPtr(static_cast(protoPropertyStorage), temp); loadPtr(Address(temp, cachedOffset * sizeof(JSValue)), result); } } void JIT::testPrototype(Structure* structure, JumpList& failureCases) { if (structure->m_prototype.isNull()) return; move(ImmPtr(&asCell(structure->m_prototype)->m_structure), regT2); move(ImmPtr(asCell(structure->m_prototype)->m_structure), regT3); failureCases.append(branchPtr(NotEqual, Address(regT2), regT3)); } void JIT::privateCompilePutByIdTransition(StructureStubInfo* stubInfo, Structure* oldStructure, Structure* newStructure, size_t cachedOffset, StructureChain* chain, ReturnAddressPtr returnAddress) { JumpList failureCases; // Check eax is an object of the right Structure. failureCases.append(emitJumpIfNotJSCell(regT0)); failureCases.append(branchPtr(NotEqual, Address(regT0, OBJECT_OFFSETOF(JSCell, m_structure)), ImmPtr(oldStructure))); testPrototype(oldStructure, failureCases); // ecx = baseObject->m_structure for (RefPtr* it = chain->head(); *it; ++it) testPrototype(it->get(), failureCases); Call callTarget; // emit a call only if storage realloc is needed bool willNeedStorageRealloc = oldStructure->propertyStorageCapacity() != newStructure->propertyStorageCapacity(); if (willNeedStorageRealloc) { // This trampoline was called to like a JIT stub; before we can can call again we need to // remove the return address from the stack, to prevent the stack from becoming misaligned. preserveReturnAddressAfterCall(regT3); JITStubCall stubCall(this, cti_op_put_by_id_transition_realloc); stubCall.skipArgument(); // base stubCall.skipArgument(); // ident stubCall.skipArgument(); // value stubCall.addArgument(Imm32(oldStructure->propertyStorageCapacity())); stubCall.addArgument(Imm32(newStructure->propertyStorageCapacity())); stubCall.call(regT0); emitGetJITStubArg(2, regT1); restoreReturnAddressBeforeReturn(regT3); } // Assumes m_refCount can be decremented easily, refcount decrement is safe as // codeblock should ensure oldStructure->m_refCount > 0 sub32(Imm32(1), AbsoluteAddress(oldStructure->addressOfCount())); add32(Imm32(1), AbsoluteAddress(newStructure->addressOfCount())); storePtr(ImmPtr(newStructure), Address(regT0, OBJECT_OFFSETOF(JSCell, m_structure))); // write the value compilePutDirectOffset(regT0, regT1, newStructure, cachedOffset); ret(); ASSERT(!failureCases.empty()); failureCases.link(this); restoreArgumentReferenceForTrampoline(); Call failureCall = tailRecursiveCall(); LinkBuffer patchBuffer(this, m_codeBlock->executablePool()); patchBuffer.link(failureCall, FunctionPtr(cti_op_put_by_id_fail)); if (willNeedStorageRealloc) { ASSERT(m_calls.size() == 1); patchBuffer.link(m_calls[0].from, FunctionPtr(cti_op_put_by_id_transition_realloc)); } CodeLocationLabel entryLabel = patchBuffer.finalizeCodeAddendum(); stubInfo->stubRoutine = entryLabel; RepatchBuffer repatchBuffer(m_codeBlock); repatchBuffer.relinkCallerToTrampoline(returnAddress, entryLabel); } void JIT::patchGetByIdSelf(CodeBlock* codeBlock, StructureStubInfo* stubInfo, Structure* structure, size_t cachedOffset, ReturnAddressPtr returnAddress) { RepatchBuffer repatchBuffer(codeBlock); // We don't want to patch more than once - in future go to cti_op_get_by_id_generic. // Should probably go to cti_op_get_by_id_fail, but that doesn't do anything interesting right now. repatchBuffer.relinkCallerToFunction(returnAddress, FunctionPtr(cti_op_get_by_id_self_fail)); int offset = sizeof(JSValue) * cachedOffset; // If we're patching to use inline storage, convert the initial load to a lea; this avoids the extra load // and makes the subsequent load's offset automatically correct if (structure->isUsingInlineStorage()) repatchBuffer.repatchLoadPtrToLEA(stubInfo->hotPathBegin.instructionAtOffset(patchOffsetGetByIdExternalLoad)); // Patch the offset into the propoerty map to load from, then patch the Structure to look for. repatchBuffer.repatch(stubInfo->hotPathBegin.dataLabelPtrAtOffset(patchOffsetGetByIdStructure), structure); repatchBuffer.repatch(stubInfo->hotPathBegin.dataLabel32AtOffset(patchOffsetGetByIdPropertyMapOffset), offset); } void JIT::patchMethodCallProto(CodeBlock* codeBlock, MethodCallLinkInfo& methodCallLinkInfo, JSFunction* callee, Structure* structure, JSObject* proto, ReturnAddressPtr returnAddress) { RepatchBuffer repatchBuffer(codeBlock); ASSERT(!methodCallLinkInfo.cachedStructure); methodCallLinkInfo.cachedStructure = structure; structure->ref(); Structure* prototypeStructure = proto->structure(); methodCallLinkInfo.cachedPrototypeStructure = prototypeStructure; prototypeStructure->ref(); repatchBuffer.repatch(methodCallLinkInfo.structureLabel, structure); repatchBuffer.repatch(methodCallLinkInfo.structureLabel.dataLabelPtrAtOffset(patchOffsetMethodCheckProtoObj), proto); repatchBuffer.repatch(methodCallLinkInfo.structureLabel.dataLabelPtrAtOffset(patchOffsetMethodCheckProtoStruct), prototypeStructure); repatchBuffer.repatch(methodCallLinkInfo.structureLabel.dataLabelPtrAtOffset(patchOffsetMethodCheckPutFunction), callee); repatchBuffer.relinkCallerToFunction(returnAddress, FunctionPtr(cti_op_get_by_id)); } void JIT::patchPutByIdReplace(CodeBlock* codeBlock, StructureStubInfo* stubInfo, Structure* structure, size_t cachedOffset, ReturnAddressPtr returnAddress) { RepatchBuffer repatchBuffer(codeBlock); // We don't want to patch more than once - in future go to cti_op_put_by_id_generic. // Should probably go to cti_op_put_by_id_fail, but that doesn't do anything interesting right now. repatchBuffer.relinkCallerToFunction(returnAddress, FunctionPtr(cti_op_put_by_id_generic)); int offset = sizeof(JSValue) * cachedOffset; // If we're patching to use inline storage, convert the initial load to a lea; this avoids the extra load // and makes the subsequent load's offset automatically correct if (structure->isUsingInlineStorage()) repatchBuffer.repatchLoadPtrToLEA(stubInfo->hotPathBegin.instructionAtOffset(patchOffsetPutByIdExternalLoad)); // Patch the offset into the propoerty map to load from, then patch the Structure to look for. repatchBuffer.repatch(stubInfo->hotPathBegin.dataLabelPtrAtOffset(patchOffsetPutByIdStructure), structure); repatchBuffer.repatch(stubInfo->hotPathBegin.dataLabel32AtOffset(patchOffsetPutByIdPropertyMapOffset), offset); } void JIT::privateCompilePatchGetArrayLength(ReturnAddressPtr returnAddress) { StructureStubInfo* stubInfo = &m_codeBlock->getStubInfo(returnAddress); // Check eax is an array Jump failureCases1 = branchPtr(NotEqual, Address(regT0), ImmPtr(m_globalData->jsArrayVPtr)); // Checks out okay! - get the length from the storage loadPtr(Address(regT0, OBJECT_OFFSETOF(JSArray, m_storage)), regT2); load32(Address(regT2, OBJECT_OFFSETOF(ArrayStorage, m_length)), regT2); Jump failureCases2 = branch32(Above, regT2, Imm32(JSImmediate::maxImmediateInt)); emitFastArithIntToImmNoCheck(regT2, regT0); Jump success = jump(); LinkBuffer patchBuffer(this, m_codeBlock->executablePool()); // Use the patch information to link the failure cases back to the original slow case routine. CodeLocationLabel slowCaseBegin = stubInfo->callReturnLocation.labelAtOffset(-patchOffsetGetByIdSlowCaseCall); patchBuffer.link(failureCases1, slowCaseBegin); patchBuffer.link(failureCases2, slowCaseBegin); // On success return back to the hot patch code, at a point it will perform the store to dest for us. patchBuffer.link(success, stubInfo->hotPathBegin.labelAtOffset(patchOffsetGetByIdPutResult)); // Track the stub we have created so that it will be deleted later. CodeLocationLabel entryLabel = patchBuffer.finalizeCodeAddendum(); stubInfo->stubRoutine = entryLabel; // Finally patch the jump to slow case back in the hot path to jump here instead. CodeLocationJump jumpLocation = stubInfo->hotPathBegin.jumpAtOffset(patchOffsetGetByIdBranchToSlowCase); RepatchBuffer repatchBuffer(m_codeBlock); repatchBuffer.relink(jumpLocation, entryLabel); // We don't want to patch more than once - in future go to cti_op_put_by_id_generic. repatchBuffer.relinkCallerToFunction(returnAddress, FunctionPtr(cti_op_get_by_id_array_fail)); } void JIT::privateCompileGetByIdProto(StructureStubInfo* stubInfo, Structure* structure, Structure* prototypeStructure, const Identifier& ident, const PropertySlot& slot, size_t cachedOffset, ReturnAddressPtr returnAddress, CallFrame* callFrame) { // The prototype object definitely exists (if this stub exists the CodeBlock is referencing a Structure that is // referencing the prototype object - let's speculatively load it's table nice and early!) JSObject* protoObject = asObject(structure->prototypeForLookup(callFrame)); // Check eax is an object of the right Structure. Jump failureCases1 = checkStructure(regT0, structure); // Check the prototype object's Structure had not changed. Structure** prototypeStructureAddress = &(protoObject->m_structure); #if CPU(X86_64) move(ImmPtr(prototypeStructure), regT3); Jump failureCases2 = branchPtr(NotEqual, AbsoluteAddress(prototypeStructureAddress), regT3); #else Jump failureCases2 = branchPtr(NotEqual, AbsoluteAddress(prototypeStructureAddress), ImmPtr(prototypeStructure)); #endif bool needsStubLink = false; // Checks out okay! if (slot.cachedPropertyType() == PropertySlot::Getter) { needsStubLink = true; compileGetDirectOffset(protoObject, regT1, regT1, cachedOffset); JITStubCall stubCall(this, cti_op_get_by_id_getter_stub); stubCall.addArgument(regT1); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else if (slot.cachedPropertyType() == PropertySlot::Custom) { needsStubLink = true; JITStubCall stubCall(this, cti_op_get_by_id_custom_stub); stubCall.addArgument(ImmPtr(protoObject)); stubCall.addArgument(ImmPtr(FunctionPtr(slot.customGetter()).executableAddress())); stubCall.addArgument(ImmPtr(const_cast(&ident))); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else compileGetDirectOffset(protoObject, regT1, regT0, cachedOffset); Jump success = jump(); LinkBuffer patchBuffer(this, m_codeBlock->executablePool()); // Use the patch information to link the failure cases back to the original slow case routine. CodeLocationLabel slowCaseBegin = stubInfo->callReturnLocation.labelAtOffset(-patchOffsetGetByIdSlowCaseCall); patchBuffer.link(failureCases1, slowCaseBegin); patchBuffer.link(failureCases2, slowCaseBegin); // On success return back to the hot patch code, at a point it will perform the store to dest for us. patchBuffer.link(success, stubInfo->hotPathBegin.labelAtOffset(patchOffsetGetByIdPutResult)); if (needsStubLink) { for (Vector::iterator iter = m_calls.begin(); iter != m_calls.end(); ++iter) { if (iter->to) patchBuffer.link(iter->from, FunctionPtr(iter->to)); } } // Track the stub we have created so that it will be deleted later. CodeLocationLabel entryLabel = patchBuffer.finalizeCodeAddendum(); stubInfo->stubRoutine = entryLabel; // Finally patch the jump to slow case back in the hot path to jump here instead. CodeLocationJump jumpLocation = stubInfo->hotPathBegin.jumpAtOffset(patchOffsetGetByIdBranchToSlowCase); RepatchBuffer repatchBuffer(m_codeBlock); repatchBuffer.relink(jumpLocation, entryLabel); // We don't want to patch more than once - in future go to cti_op_put_by_id_generic. repatchBuffer.relinkCallerToFunction(returnAddress, FunctionPtr(cti_op_get_by_id_proto_list)); } void JIT::privateCompileGetByIdSelfList(StructureStubInfo* stubInfo, PolymorphicAccessStructureList* polymorphicStructures, int currentIndex, Structure* structure, const Identifier& ident, const PropertySlot& slot, size_t cachedOffset) { Jump failureCase = checkStructure(regT0, structure); bool needsStubLink = false; if (slot.cachedPropertyType() == PropertySlot::Getter) { needsStubLink = true; if (!structure->isUsingInlineStorage()) { move(regT0, regT1); compileGetDirectOffset(regT1, regT1, structure, cachedOffset); } else compileGetDirectOffset(regT0, regT1, structure, cachedOffset); JITStubCall stubCall(this, cti_op_get_by_id_getter_stub); stubCall.addArgument(regT1); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else if (slot.cachedPropertyType() == PropertySlot::Custom) { needsStubLink = true; JITStubCall stubCall(this, cti_op_get_by_id_custom_stub); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(FunctionPtr(slot.customGetter()).executableAddress())); stubCall.addArgument(ImmPtr(const_cast(&ident))); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else compileGetDirectOffset(regT0, regT0, structure, cachedOffset); Jump success = jump(); LinkBuffer patchBuffer(this, m_codeBlock->executablePool()); if (needsStubLink) { for (Vector::iterator iter = m_calls.begin(); iter != m_calls.end(); ++iter) { if (iter->to) patchBuffer.link(iter->from, FunctionPtr(iter->to)); } } // Use the patch information to link the failure cases back to the original slow case routine. CodeLocationLabel lastProtoBegin = polymorphicStructures->list[currentIndex - 1].stubRoutine; if (!lastProtoBegin) lastProtoBegin = stubInfo->callReturnLocation.labelAtOffset(-patchOffsetGetByIdSlowCaseCall); patchBuffer.link(failureCase, lastProtoBegin); // On success return back to the hot patch code, at a point it will perform the store to dest for us. patchBuffer.link(success, stubInfo->hotPathBegin.labelAtOffset(patchOffsetGetByIdPutResult)); CodeLocationLabel entryLabel = patchBuffer.finalizeCodeAddendum(); structure->ref(); polymorphicStructures->list[currentIndex].set(entryLabel, structure); // Finally patch the jump to slow case back in the hot path to jump here instead. CodeLocationJump jumpLocation = stubInfo->hotPathBegin.jumpAtOffset(patchOffsetGetByIdBranchToSlowCase); RepatchBuffer repatchBuffer(m_codeBlock); repatchBuffer.relink(jumpLocation, entryLabel); } void JIT::privateCompileGetByIdProtoList(StructureStubInfo* stubInfo, PolymorphicAccessStructureList* prototypeStructures, int currentIndex, Structure* structure, Structure* prototypeStructure, const Identifier& ident, const PropertySlot& slot, size_t cachedOffset, CallFrame* callFrame) { // The prototype object definitely exists (if this stub exists the CodeBlock is referencing a Structure that is // referencing the prototype object - let's speculatively load it's table nice and early!) JSObject* protoObject = asObject(structure->prototypeForLookup(callFrame)); // Check eax is an object of the right Structure. Jump failureCases1 = checkStructure(regT0, structure); // Check the prototype object's Structure had not changed. Structure** prototypeStructureAddress = &(protoObject->m_structure); #if CPU(X86_64) move(ImmPtr(prototypeStructure), regT3); Jump failureCases2 = branchPtr(NotEqual, AbsoluteAddress(prototypeStructureAddress), regT3); #else Jump failureCases2 = branchPtr(NotEqual, AbsoluteAddress(prototypeStructureAddress), ImmPtr(prototypeStructure)); #endif // Checks out okay! bool needsStubLink = false; if (slot.cachedPropertyType() == PropertySlot::Getter) { needsStubLink = true; compileGetDirectOffset(protoObject, regT1, regT1, cachedOffset); JITStubCall stubCall(this, cti_op_get_by_id_getter_stub); stubCall.addArgument(regT1); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else if (slot.cachedPropertyType() == PropertySlot::Custom) { needsStubLink = true; JITStubCall stubCall(this, cti_op_get_by_id_custom_stub); stubCall.addArgument(ImmPtr(protoObject)); stubCall.addArgument(ImmPtr(FunctionPtr(slot.customGetter()).executableAddress())); stubCall.addArgument(ImmPtr(const_cast(&ident))); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else compileGetDirectOffset(protoObject, regT1, regT0, cachedOffset); Jump success = jump(); LinkBuffer patchBuffer(this, m_codeBlock->executablePool()); if (needsStubLink) { for (Vector::iterator iter = m_calls.begin(); iter != m_calls.end(); ++iter) { if (iter->to) patchBuffer.link(iter->from, FunctionPtr(iter->to)); } } // Use the patch information to link the failure cases back to the original slow case routine. CodeLocationLabel lastProtoBegin = prototypeStructures->list[currentIndex - 1].stubRoutine; patchBuffer.link(failureCases1, lastProtoBegin); patchBuffer.link(failureCases2, lastProtoBegin); // On success return back to the hot patch code, at a point it will perform the store to dest for us. patchBuffer.link(success, stubInfo->hotPathBegin.labelAtOffset(patchOffsetGetByIdPutResult)); CodeLocationLabel entryLabel = patchBuffer.finalizeCodeAddendum(); structure->ref(); prototypeStructure->ref(); prototypeStructures->list[currentIndex].set(entryLabel, structure, prototypeStructure); // Finally patch the jump to slow case back in the hot path to jump here instead. CodeLocationJump jumpLocation = stubInfo->hotPathBegin.jumpAtOffset(patchOffsetGetByIdBranchToSlowCase); RepatchBuffer repatchBuffer(m_codeBlock); repatchBuffer.relink(jumpLocation, entryLabel); } void JIT::privateCompileGetByIdChainList(StructureStubInfo* stubInfo, PolymorphicAccessStructureList* prototypeStructures, int currentIndex, Structure* structure, StructureChain* chain, size_t count, const Identifier& ident, const PropertySlot& slot, size_t cachedOffset, CallFrame* callFrame) { ASSERT(count); JumpList bucketsOfFail; // Check eax is an object of the right Structure. Jump baseObjectCheck = checkStructure(regT0, structure); bucketsOfFail.append(baseObjectCheck); Structure* currStructure = structure; RefPtr* chainEntries = chain->head(); JSObject* protoObject = 0; for (unsigned i = 0; i < count; ++i) { protoObject = asObject(currStructure->prototypeForLookup(callFrame)); currStructure = chainEntries[i].get(); // Check the prototype object's Structure had not changed. Structure** prototypeStructureAddress = &(protoObject->m_structure); #if CPU(X86_64) move(ImmPtr(currStructure), regT3); bucketsOfFail.append(branchPtr(NotEqual, AbsoluteAddress(prototypeStructureAddress), regT3)); #else bucketsOfFail.append(branchPtr(NotEqual, AbsoluteAddress(prototypeStructureAddress), ImmPtr(currStructure))); #endif } ASSERT(protoObject); bool needsStubLink = false; if (slot.cachedPropertyType() == PropertySlot::Getter) { needsStubLink = true; compileGetDirectOffset(protoObject, regT1, regT1, cachedOffset); JITStubCall stubCall(this, cti_op_get_by_id_getter_stub); stubCall.addArgument(regT1); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else if (slot.cachedPropertyType() == PropertySlot::Custom) { needsStubLink = true; JITStubCall stubCall(this, cti_op_get_by_id_custom_stub); stubCall.addArgument(ImmPtr(protoObject)); stubCall.addArgument(ImmPtr(FunctionPtr(slot.customGetter()).executableAddress())); stubCall.addArgument(ImmPtr(const_cast(&ident))); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else compileGetDirectOffset(protoObject, regT1, regT0, cachedOffset); Jump success = jump(); LinkBuffer patchBuffer(this, m_codeBlock->executablePool()); if (needsStubLink) { for (Vector::iterator iter = m_calls.begin(); iter != m_calls.end(); ++iter) { if (iter->to) patchBuffer.link(iter->from, FunctionPtr(iter->to)); } } // Use the patch information to link the failure cases back to the original slow case routine. CodeLocationLabel lastProtoBegin = prototypeStructures->list[currentIndex - 1].stubRoutine; patchBuffer.link(bucketsOfFail, lastProtoBegin); // On success return back to the hot patch code, at a point it will perform the store to dest for us. patchBuffer.link(success, stubInfo->hotPathBegin.labelAtOffset(patchOffsetGetByIdPutResult)); CodeLocationLabel entryLabel = patchBuffer.finalizeCodeAddendum(); // Track the stub we have created so that it will be deleted later. structure->ref(); chain->ref(); prototypeStructures->list[currentIndex].set(entryLabel, structure, chain); // Finally patch the jump to slow case back in the hot path to jump here instead. CodeLocationJump jumpLocation = stubInfo->hotPathBegin.jumpAtOffset(patchOffsetGetByIdBranchToSlowCase); RepatchBuffer repatchBuffer(m_codeBlock); repatchBuffer.relink(jumpLocation, entryLabel); } void JIT::privateCompileGetByIdChain(StructureStubInfo* stubInfo, Structure* structure, StructureChain* chain, size_t count, const Identifier& ident, const PropertySlot& slot, size_t cachedOffset, ReturnAddressPtr returnAddress, CallFrame* callFrame) { ASSERT(count); JumpList bucketsOfFail; // Check eax is an object of the right Structure. bucketsOfFail.append(checkStructure(regT0, structure)); Structure* currStructure = structure; RefPtr* chainEntries = chain->head(); JSObject* protoObject = 0; for (unsigned i = 0; i < count; ++i) { protoObject = asObject(currStructure->prototypeForLookup(callFrame)); currStructure = chainEntries[i].get(); // Check the prototype object's Structure had not changed. Structure** prototypeStructureAddress = &(protoObject->m_structure); #if CPU(X86_64) move(ImmPtr(currStructure), regT3); bucketsOfFail.append(branchPtr(NotEqual, AbsoluteAddress(prototypeStructureAddress), regT3)); #else bucketsOfFail.append(branchPtr(NotEqual, AbsoluteAddress(prototypeStructureAddress), ImmPtr(currStructure))); #endif } ASSERT(protoObject); bool needsStubLink = false; if (slot.cachedPropertyType() == PropertySlot::Getter) { needsStubLink = true; compileGetDirectOffset(protoObject, regT1, regT1, cachedOffset); JITStubCall stubCall(this, cti_op_get_by_id_getter_stub); stubCall.addArgument(regT1); stubCall.addArgument(regT0); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else if (slot.cachedPropertyType() == PropertySlot::Custom) { needsStubLink = true; JITStubCall stubCall(this, cti_op_get_by_id_custom_stub); stubCall.addArgument(ImmPtr(protoObject)); stubCall.addArgument(ImmPtr(FunctionPtr(slot.customGetter()).executableAddress())); stubCall.addArgument(ImmPtr(const_cast(&ident))); stubCall.addArgument(ImmPtr(stubInfo->callReturnLocation.executableAddress())); stubCall.call(); } else compileGetDirectOffset(protoObject, regT1, regT0, cachedOffset); Jump success = jump(); LinkBuffer patchBuffer(this, m_codeBlock->executablePool()); if (needsStubLink) { for (Vector::iterator iter = m_calls.begin(); iter != m_calls.end(); ++iter) { if (iter->to) patchBuffer.link(iter->from, FunctionPtr(iter->to)); } } // Use the patch information to link the failure cases back to the original slow case routine. patchBuffer.link(bucketsOfFail, stubInfo->callReturnLocation.labelAtOffset(-patchOffsetGetByIdSlowCaseCall)); // On success return back to the hot patch code, at a point it will perform the store to dest for us. patchBuffer.link(success, stubInfo->hotPathBegin.labelAtOffset(patchOffsetGetByIdPutResult)); // Track the stub we have created so that it will be deleted later. CodeLocationLabel entryLabel = patchBuffer.finalizeCodeAddendum(); stubInfo->stubRoutine = entryLabel; // Finally patch the jump to slow case back in the hot path to jump here instead. CodeLocationJump jumpLocation = stubInfo->hotPathBegin.jumpAtOffset(patchOffsetGetByIdBranchToSlowCase); RepatchBuffer repatchBuffer(m_codeBlock); repatchBuffer.relink(jumpLocation, entryLabel); // We don't want to patch more than once - in future go to cti_op_put_by_id_generic. repatchBuffer.relinkCallerToFunction(returnAddress, FunctionPtr(cti_op_get_by_id_proto_list)); } /* ------------------------------ END: !ENABLE / ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS) ------------------------------ */ #endif // !ENABLE(JIT_OPTIMIZE_PROPERTY_ACCESS) } // namespace JSC #endif // ENABLE(JIT) #endif // !USE(JSVALUE32_64)