blob: 513643902b6eb576e9e2ca1f209a8d2d49cfcdf7 [file] [log] [blame]
# Copyright (C) 2011-2019 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. AND ITS CONTRIBUTORS ``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 ITS 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.
# Crash course on the language that this is written in (which I just call
# "assembly" even though it's more than that):
#
# - Mostly gas-style operand ordering. The last operand tends to be the
# destination. So "a := b" is written as "mov b, a". But unlike gas,
# comparisons are in-order, so "if (a < b)" is written as
# "bilt a, b, ...".
#
# - "b" = byte, "h" = 16-bit word, "i" = 32-bit word, "p" = pointer.
# For 32-bit, "i" and "p" are interchangeable except when an op supports one
# but not the other.
#
# - In general, valid operands for macro invocations and instructions are
# registers (eg "t0"), addresses (eg "4[t0]"), base-index addresses
# (eg "7[t0, t1, 2]"), absolute addresses (eg "0xa0000000[]"), or labels
# (eg "_foo" or ".foo"). Macro invocations can also take anonymous
# macros as operands. Instructions cannot take anonymous macros.
#
# - Labels must have names that begin with either "_" or ".". A "." label
# is local and gets renamed before code gen to minimize namespace
# pollution. A "_" label is an extern symbol (i.e. ".globl"). The "_"
# may or may not be removed during code gen depending on whether the asm
# conventions for C name mangling on the target platform mandate a "_"
# prefix.
#
# - A "macro" is a lambda expression, which may be either anonymous or
# named. But this has caveats. "macro" can take zero or more arguments,
# which may be macros or any valid operands, but it can only return
# code. But you can do Turing-complete things via continuation passing
# style: "macro foo (a, b) b(a, a) end foo(foo, foo)". Actually, don't do
# that, since you'll just crash the assembler.
#
# - An "if" is a conditional on settings. Any identifier supplied in the
# predicate of an "if" is assumed to be a #define that is available
# during code gen. So you can't use "if" for computation in a macro, but
# you can use it to select different pieces of code for different
# platforms.
#
# - Arguments to macros follow lexical scoping rather than dynamic scoping.
# Const's also follow lexical scoping and may override (hide) arguments
# or other consts. All variables (arguments and constants) can be bound
# to operands. Additionally, arguments (but not constants) can be bound
# to macros.
# The following general-purpose registers are available:
#
# - cfr and sp hold the call frame and (native) stack pointer respectively.
# They are callee-save registers, and guaranteed to be distinct from all other
# registers on all architectures.
#
# - lr is defined on non-X86 architectures (ARM64, ARM64E, ARMv7, MIPS and CLOOP)
# and holds the return PC
#
# - t0, t1, t2, t3, t4, and optionally t5, t6, and t7 are temporary registers that can get trashed on
# calls, and are pairwise distinct registers. t4 holds the JS program counter, so use
# with caution in opcodes (actually, don't use it in opcodes at all, except as PC).
#
# - r0 and r1 are the platform's customary return registers, and thus are
# two distinct registers
#
# - a0, a1, a2 and a3 are the platform's customary argument registers, and
# thus are pairwise distinct registers. Be mindful that:
# + On X86, there are no argument registers. a0 and a1 are edx and
# ecx following the fastcall convention, but you should still use the stack
# to pass your arguments. The cCall2 and cCall4 macros do this for you.
# + On X86_64_WIN, you should allocate space on the stack for the arguments,
# and the return convention is weird for > 8 bytes types. The only place we
# use > 8 bytes return values is on a cCall, and cCall2 and cCall4 handle
# this for you.
#
# - The only registers guaranteed to be caller-saved are r0, r1, a0, a1 and a2, and
# you should be mindful of that in functions that are called directly from C.
# If you need more registers, you should push and pop them like a good
# assembly citizen, because any other register will be callee-saved on X86.
#
# You can additionally assume:
#
# - a3, t2, t3, t4 and t5 are never return registers; t0, t1, a0, a1 and a2
# can be return registers.
#
# - t4 and t5 are never argument registers, t3 can only be a3, t1 can only be
# a1; but t0 and t2 can be either a0 or a2.
#
# - There are callee-save registers named csr0, csr1, ... csrN.
# The last three csr registers are used used to store the PC base and
# two special tag values (on 64-bits only). Don't use them for anything else.
#
# Additional platform-specific details (you shouldn't rely on this remaining
# true):
#
# - For consistency with the baseline JIT, t0 is always r0 (and t1 is always
# r1 on 32 bits platforms). You should use the r version when you need return
# registers, and the t version otherwise: code using t0 (or t1) should still
# work if swapped with e.g. t3, while code using r0 (or r1) should not. There
# *may* be legacy code relying on this.
#
# - On all platforms other than X86, t0 can only be a0 and t2 can only be a2.
#
# - On all platforms other than X86 and X86_64, a2 is not a return register.
# a2 is r0 on X86 (because we have so few registers) and r1 on X86_64 (because
# the ABI enforces it).
#
# The following floating-point registers are available:
#
# - ft0-ft5 are temporary floating-point registers that get trashed on calls,
# and are pairwise distinct.
#
# - fa0 and fa1 are the platform's customary floating-point argument
# registers, and are both distinct. On 64-bits platforms, fa2 and fa3 are
# additional floating-point argument registers.
#
# - fr is the platform's customary floating-point return register
#
# You can assume that ft1-ft5 or fa1-fa3 are never fr, and that ftX is never
# faY if X != Y.
# First come the common protocols that both interpreters use. Note that each
# of these must have an ASSERT() in LLIntData.cpp
# Work-around for the fact that the toolchain's awareness of armv7k / armv7s
# results in a separate slab in the fat binary, yet the offlineasm doesn't know
# to expect it.
if ARMv7k
end
if ARMv7s
end
# These declarations must match interpreter/JSStack.h.
const PtrSize = constexpr (sizeof(void*))
const MachineRegisterSize = constexpr (sizeof(CPURegister))
const SlotSize = constexpr (sizeof(Register))
if JSVALUE64
const CallFrameHeaderSlots = 5
else
const CallFrameHeaderSlots = 4
const CallFrameAlignSlots = 1
end
const JSLexicalEnvironment_variables = (sizeof JSLexicalEnvironment + SlotSize - 1) & ~(SlotSize - 1)
const DirectArguments_storage = (sizeof DirectArguments + SlotSize - 1) & ~(SlotSize - 1)
const JSInternalFieldObjectImpl_internalFields = JSInternalFieldObjectImpl::m_internalFields
const StackAlignment = constexpr (stackAlignmentBytes())
const StackAlignmentSlots = constexpr (stackAlignmentRegisters())
const StackAlignmentMask = StackAlignment - 1
const CallerFrameAndPCSize = constexpr (sizeof(CallerFrameAndPC))
const CallerFrame = 0
const ReturnPC = CallerFrame + MachineRegisterSize
const CodeBlock = ReturnPC + MachineRegisterSize
const Callee = CodeBlock + SlotSize
const ArgumentCountIncludingThis = Callee + SlotSize
const ThisArgumentOffset = ArgumentCountIncludingThis + SlotSize
const FirstArgumentOffset = ThisArgumentOffset + SlotSize
const CallFrameHeaderSize = ThisArgumentOffset
const MetadataOffsetTable16Offset = 0
const MetadataOffsetTable32Offset = constexpr UnlinkedMetadataTable::s_offset16TableSize
const NumberOfJSOpcodeIDs = constexpr numOpcodeIDs
# Some value representation constants.
if JSVALUE64
const TagOther = constexpr JSValue::OtherTag
const TagBool = constexpr JSValue::BoolTag
const TagUndefined = constexpr JSValue::UndefinedTag
const ValueEmpty = constexpr JSValue::ValueEmpty
const ValueFalse = constexpr JSValue::ValueFalse
const ValueTrue = constexpr JSValue::ValueTrue
const ValueUndefined = constexpr JSValue::ValueUndefined
const ValueNull = constexpr JSValue::ValueNull
const TagNumber = constexpr JSValue::NumberTag
const NotCellMask = constexpr JSValue::NotCellMask
else
const Int32Tag = constexpr JSValue::Int32Tag
const BooleanTag = constexpr JSValue::BooleanTag
const NullTag = constexpr JSValue::NullTag
const UndefinedTag = constexpr JSValue::UndefinedTag
const CellTag = constexpr JSValue::CellTag
const EmptyValueTag = constexpr JSValue::EmptyValueTag
const DeletedValueTag = constexpr JSValue::DeletedValueTag
const LowestTag = constexpr JSValue::LowestTag
end
if JSVALUE64
const NumberOfStructureIDEntropyBits = constexpr StructureIDTable::s_numberOfEntropyBits
const StructureEntropyBitsShift = constexpr StructureIDTable::s_entropyBitsShiftForStructurePointer
end
const maxFrameExtentForSlowPathCall = constexpr maxFrameExtentForSlowPathCall
if X86_64 or X86_64_WIN or ARM64 or ARM64E
const CalleeSaveSpaceAsVirtualRegisters = 4
elsif C_LOOP or C_LOOP_WIN
const CalleeSaveSpaceAsVirtualRegisters = 1
elsif ARMv7
const CalleeSaveSpaceAsVirtualRegisters = 1
elsif MIPS
const CalleeSaveSpaceAsVirtualRegisters = 1
else
const CalleeSaveSpaceAsVirtualRegisters = 0
end
const CalleeSaveSpaceStackAligned = (CalleeSaveSpaceAsVirtualRegisters * SlotSize + StackAlignment - 1) & ~StackAlignmentMask
# Watchpoint states
const ClearWatchpoint = constexpr ClearWatchpoint
const IsWatched = constexpr IsWatched
const IsInvalidated = constexpr IsInvalidated
# ShadowChicken data
const ShadowChickenTailMarker = constexpr ShadowChicken::Packet::tailMarkerValue
# UnaryArithProfile data
const ArithProfileInt = constexpr (UnaryArithProfile::observedIntBits())
const ArithProfileNumber = constexpr (UnaryArithProfile::observedNumberBits())
# BinaryArithProfile data
const ArithProfileIntInt = constexpr (BinaryArithProfile::observedIntIntBits())
const ArithProfileNumberInt = constexpr (BinaryArithProfile::observedNumberIntBits())
const ArithProfileIntNumber = constexpr (BinaryArithProfile::observedIntNumberBits())
const ArithProfileNumberNumber = constexpr (BinaryArithProfile::observedNumberNumberBits())
# Pointer Tags
const BytecodePtrTag = constexpr BytecodePtrTag
const JSEntryPtrTag = constexpr JSEntryPtrTag
const ExceptionHandlerPtrTag = constexpr ExceptionHandlerPtrTag
const NoPtrTag = constexpr NoPtrTag
const SlowPathPtrTag = constexpr SlowPathPtrTag
# Some register conventions.
# - We use a pair of registers to represent the PC: one register for the
# base of the bytecodes, and one register for the index.
# - The PC base (or PB for short) must be stored in a callee-save register.
# - C calls are still given the Instruction* rather than the PC index.
# This requires an add before the call, and a sub after.
if JSVALUE64
const PC = t4 # When changing this, make sure LLIntPC is up to date in LLIntPCRanges.h
if ARM64 or ARM64E
const metadataTable = csr6
const PB = csr7
const numberTag = csr8
const notCellMask = csr9
elsif X86_64
const metadataTable = csr1
const PB = csr2
const numberTag = csr3
const notCellMask = csr4
elsif X86_64_WIN
const metadataTable = csr3
const PB = csr4
const numberTag = csr5
const notCellMask = csr6
elsif C_LOOP or C_LOOP_WIN
const PB = csr0
const numberTag = csr1
const notCellMask = csr2
const metadataTable = csr3
end
else
const PC = t4 # When changing this, make sure LLIntPC is up to date in LLIntPCRanges.h
if C_LOOP or C_LOOP_WIN
const PB = csr0
const metadataTable = csr3
elsif ARMv7
const metadataTable = csr0
const PB = csr1
elsif MIPS
const metadataTable = csr0
const PB = csr1
else
error
end
end
if GIGACAGE_ENABLED
const GigacagePrimitiveBasePtrOffset = constexpr Gigacage::offsetOfPrimitiveGigacageBasePtr
const GigacageJSValueBasePtrOffset = constexpr Gigacage::offsetOfJSValueGigacageBasePtr
end
# Opcode offsets
const OpcodeIDNarrowSize = 1 # OpcodeID
const OpcodeIDWide16Size = 2 # Wide16 Prefix + OpcodeID
const OpcodeIDWide32Size = 2 # Wide32 Prefix + OpcodeID
macro nextInstruction()
loadb [PB, PC, 1], t0
leap _g_opcodeMap, t1
jmp [t1, t0, PtrSize], BytecodePtrTag
end
macro nextInstructionWide16()
loadb OpcodeIDNarrowSize[PB, PC, 1], t0
leap _g_opcodeMapWide16, t1
jmp [t1, t0, PtrSize], BytecodePtrTag
end
macro nextInstructionWide32()
loadb OpcodeIDNarrowSize[PB, PC, 1], t0
leap _g_opcodeMapWide32, t1
jmp [t1, t0, PtrSize], BytecodePtrTag
end
macro dispatch(advanceReg)
addp advanceReg, PC
nextInstruction()
end
macro dispatchIndirect(offsetReg)
dispatch(offsetReg)
end
macro genericDispatchOp(dispatch, size, opcodeName)
macro dispatchNarrow()
dispatch((constexpr %opcodeName%_length - 1) * 1 + OpcodeIDNarrowSize)
end
macro dispatchWide16()
dispatch((constexpr %opcodeName%_length - 1) * 2 + OpcodeIDWide16Size)
end
macro dispatchWide32()
dispatch((constexpr %opcodeName%_length - 1) * 4 + OpcodeIDWide32Size)
end
size(dispatchNarrow, dispatchWide16, dispatchWide32, macro (dispatch) dispatch() end)
end
macro dispatchOp(size, opcodeName)
genericDispatchOp(dispatch, size, opcodeName)
end
macro getu(size, opcodeStruct, fieldName, dst)
size(getuOperandNarrow, getuOperandWide16, getuOperandWide32, macro (getu)
getu(opcodeStruct, fieldName, dst)
end)
end
macro get(size, opcodeStruct, fieldName, dst)
size(getOperandNarrow, getOperandWide16, getOperandWide32, macro (get)
get(opcodeStruct, fieldName, dst)
end)
end
macro narrow(narrowFn, wide16Fn, wide32Fn, k)
k(narrowFn)
end
macro wide16(narrowFn, wide16Fn, wide32Fn, k)
k(wide16Fn)
end
macro wide32(narrowFn, wide16Fn, wide32Fn, k)
k(wide32Fn)
end
macro metadata(size, opcode, dst, scratch)
loadh (constexpr %opcode%::opcodeID * 2 + MetadataOffsetTable16Offset)[metadataTable], dst # offset = metadataTable<uint16_t*>[opcodeID]
btinz dst, .setUpOffset
loadi (constexpr %opcode%::opcodeID * 4 + MetadataOffsetTable32Offset)[metadataTable], dst # offset = metadataTable<uint32_t*>[opcodeID]
.setUpOffset:
getu(size, opcode, m_metadataID, scratch) # scratch = bytecode.m_metadataID
muli sizeof %opcode%::Metadata, scratch # scratch *= sizeof(Op::Metadata)
addi scratch, dst # offset += scratch
addp metadataTable, dst # return &metadataTable[offset]
end
macro jumpImpl(dispatchIndirect, targetOffsetReg)
btiz targetOffsetReg, .outOfLineJumpTarget
dispatchIndirect(targetOffsetReg)
.outOfLineJumpTarget:
callSlowPath(_llint_slow_path_out_of_line_jump_target)
nextInstruction()
end
macro commonOp(label, prologue, fn)
_%label%:
prologue()
fn(narrow)
if ASSERT_ENABLED
break
break
end
# FIXME: We cannot enable wide16 bytecode in Windows CLoop. With MSVC, as CLoop::execute gets larger code
# size, CLoop::execute gets higher stack height requirement. This makes CLoop::execute takes 160KB stack
# per call, causes stack overflow error easily. For now, we disable wide16 optimization for Windows CLoop.
# https://bugs.webkit.org/show_bug.cgi?id=198283
if not C_LOOP_WIN
_%label%_wide16:
prologue()
fn(wide16)
if ASSERT_ENABLED
break
break
end
end
_%label%_wide32:
prologue()
fn(wide32)
if ASSERT_ENABLED
break
break
end
end
macro op(l, fn)
commonOp(l, macro () end, macro (size)
size(fn, macro() end, macro() end, macro(gen) gen() end)
end)
end
macro llintOp(opcodeName, opcodeStruct, fn)
commonOp(llint_%opcodeName%, traceExecution, macro(size)
macro getImpl(fieldName, dst)
get(size, opcodeStruct, fieldName, dst)
end
macro dispatchImpl()
dispatchOp(size, opcodeName)
end
fn(size, getImpl, dispatchImpl)
end)
end
macro llintOpWithReturn(opcodeName, opcodeStruct, fn)
llintOp(opcodeName, opcodeStruct, macro(size, get, dispatch)
makeReturn(get, dispatch, macro (return)
fn(size, get, dispatch, return)
end)
end)
end
macro llintOpWithMetadata(opcodeName, opcodeStruct, fn)
llintOpWithReturn(opcodeName, opcodeStruct, macro (size, get, dispatch, return)
macro meta(dst, scratch)
metadata(size, opcodeStruct, dst, scratch)
end
fn(size, get, dispatch, meta, return)
end)
end
macro llintOpWithJump(opcodeName, opcodeStruct, impl)
llintOpWithMetadata(opcodeName, opcodeStruct, macro(size, get, dispatch, metadata, return)
macro jump(fieldName)
get(fieldName, t0)
jumpImpl(dispatchIndirect, t0)
end
impl(size, get, jump, dispatch)
end)
end
macro llintOpWithProfile(opcodeName, opcodeStruct, fn)
llintOpWithMetadata(opcodeName, opcodeStruct, macro(size, get, dispatch, metadata, return)
makeReturnProfiled(opcodeStruct, get, metadata, dispatch, macro (returnProfiled)
fn(size, get, dispatch, returnProfiled)
end)
end)
end
if X86_64_WIN
const extraTempReg = t0
else
const extraTempReg = t5
end
# Constants for reasoning about value representation.
const TagOffset = constexpr TagOffset
const PayloadOffset = constexpr PayloadOffset
# Constant for reasoning about butterflies.
const IsArray = constexpr IsArray
const IndexingShapeMask = constexpr IndexingShapeMask
const NoIndexingShape = constexpr NoIndexingShape
const Int32Shape = constexpr Int32Shape
const DoubleShape = constexpr DoubleShape
const ContiguousShape = constexpr ContiguousShape
const ArrayStorageShape = constexpr ArrayStorageShape
const SlowPutArrayStorageShape = constexpr SlowPutArrayStorageShape
const CopyOnWrite = constexpr CopyOnWrite
# Type constants.
const StringType = constexpr StringType
const SymbolType = constexpr SymbolType
const ObjectType = constexpr ObjectType
const FinalObjectType = constexpr FinalObjectType
const JSFunctionType = constexpr JSFunctionType
const ArrayType = constexpr ArrayType
const DerivedArrayType = constexpr DerivedArrayType
const ProxyObjectType = constexpr ProxyObjectType
# The typed array types need to be numbered in a particular order because of the manually written
# switch statement in get_by_val and put_by_val.
const Int8ArrayType = constexpr Int8ArrayType
const Uint8ArrayType = constexpr Uint8ArrayType
const Uint8ClampedArrayType = constexpr Uint8ClampedArrayType
const Int16ArrayType = constexpr Int16ArrayType
const Uint16ArrayType = constexpr Uint16ArrayType
const Int32ArrayType = constexpr Int32ArrayType
const Uint32ArrayType = constexpr Uint32ArrayType
const Float32ArrayType = constexpr Float32ArrayType
const Float64ArrayType = constexpr Float64ArrayType
const FirstTypedArrayType = constexpr FirstTypedArrayType
const NumberOfTypedArrayTypesExcludingDataView = constexpr NumberOfTypedArrayTypesExcludingDataView
# Type flags constants.
const MasqueradesAsUndefined = constexpr MasqueradesAsUndefined
const ImplementsDefaultHasInstance = constexpr ImplementsDefaultHasInstance
# Bytecode operand constants.
const FirstConstantRegisterIndexNarrow = constexpr FirstConstantRegisterIndex8
const FirstConstantRegisterIndexWide16 = constexpr FirstConstantRegisterIndex16
const FirstConstantRegisterIndexWide32 = constexpr FirstConstantRegisterIndex
# Code type constants.
const GlobalCode = constexpr GlobalCode
const EvalCode = constexpr EvalCode
const FunctionCode = constexpr FunctionCode
const ModuleCode = constexpr ModuleCode
# The interpreter steals the tag word of the argument count.
const LLIntReturnPC = ArgumentCountIncludingThis + TagOffset
# String flags.
const isRopeInPointer = constexpr JSString::isRopeInPointer
const HashFlags8BitBuffer = constexpr StringImpl::s_hashFlag8BitBuffer
# Copied from PropertyOffset.h
const firstOutOfLineOffset = constexpr firstOutOfLineOffset
# ResolveType
const GlobalProperty = constexpr GlobalProperty
const GlobalVar = constexpr GlobalVar
const GlobalLexicalVar = constexpr GlobalLexicalVar
const ClosureVar = constexpr ClosureVar
const LocalClosureVar = constexpr LocalClosureVar
const ModuleVar = constexpr ModuleVar
const GlobalPropertyWithVarInjectionChecks = constexpr GlobalPropertyWithVarInjectionChecks
const GlobalVarWithVarInjectionChecks = constexpr GlobalVarWithVarInjectionChecks
const GlobalLexicalVarWithVarInjectionChecks = constexpr GlobalLexicalVarWithVarInjectionChecks
const ClosureVarWithVarInjectionChecks = constexpr ClosureVarWithVarInjectionChecks
const ResolveTypeMask = constexpr GetPutInfo::typeBits
const InitializationModeMask = constexpr GetPutInfo::initializationBits
const InitializationModeShift = constexpr GetPutInfo::initializationShift
const NotInitialization = constexpr InitializationMode::NotInitialization
const MarkedBlockSize = constexpr MarkedBlock::blockSize
const MarkedBlockMask = ~(MarkedBlockSize - 1)
const MarkedBlockFooterOffset = constexpr MarkedBlock::offsetOfFooter
const PreciseAllocationHeaderSize = constexpr (PreciseAllocation::headerSize())
const PreciseAllocationVMOffset = (PreciseAllocation::m_weakSet + WeakSet::m_vm - PreciseAllocationHeaderSize)
const BlackThreshold = constexpr blackThreshold
const VectorBufferOffset = Vector::m_buffer
const VectorSizeOffset = Vector::m_size
# Some common utilities.
macro crash()
if C_LOOP or C_LOOP_WIN
cloopCrash
else
call _llint_crash
end
end
macro assert(assertion)
if ASSERT_ENABLED
assertion(.ok)
crash()
.ok:
end
end
macro assert_with(assertion, crash)
if ASSERT_ENABLED
assertion(.ok)
crash()
.ok:
end
end
# The probe macro can be used to insert some debugging code without perturbing scalar
# registers. Presently, the probe macro only preserves scalar registers. Hence, the
# C probe callback function should not trash floating point registers.
#
# The macro you pass to probe() can pass whatever registers you like to your probe
# callback function. However, you need to be mindful of which of the registers are
# also used as argument registers, and ensure that you don't trash the register value
# before storing it in the probe callback argument register that you desire.
#
# Here's an example of how it's used:
#
# probe(
# macro()
# move cfr, a0 # pass the CallFrame* as arg0.
# move t0, a1 # pass the value of register t0 as arg1.
# call _cProbeCallbackFunction # to do whatever you want.
# end
# )
#
if X86_64 or ARM64 or ARM64E or ARMv7
macro probe(action)
# save all the registers that the LLInt may use.
if ARM64 or ARM64E or ARMv7
push cfr, lr
end
push a0, a1
push a2, a3
push t0, t1
push t2, t3
push t4, t5
if ARM64 or ARM64E
push csr0, csr1
push csr2, csr3
push csr4, csr5
push csr6, csr7
push csr8, csr9
elsif ARMv7
push csr0, csr1
end
action()
# restore all the registers we saved previously.
if ARM64 or ARM64E
pop csr9, csr8
pop csr7, csr6
pop csr5, csr4
pop csr3, csr2
pop csr1, csr0
elsif ARMv7
pop csr1, csr0
end
pop t5, t4
pop t3, t2
pop t1, t0
pop a3, a2
pop a1, a0
if ARM64 or ARM64E or ARMv7
pop lr, cfr
end
end
else
macro probe(action)
end
end
macro checkStackPointerAlignment(tempReg, location)
if ASSERT_ENABLED
if ARM64 or ARM64E or C_LOOP or C_LOOP_WIN
# ARM64 and ARM64E will check for us!
# C_LOOP or C_LOOP_WIN does not need the alignment, and can use a little perf
# improvement from avoiding useless work.
else
if ARMv7
# ARM can't do logical ops with the sp as a source
move sp, tempReg
andp StackAlignmentMask, tempReg
else
andp sp, StackAlignmentMask, tempReg
end
btpz tempReg, .stackPointerOkay
move location, tempReg
break
.stackPointerOkay:
end
end
end
if C_LOOP or C_LOOP_WIN or ARM64 or ARM64E or X86_64 or X86_64_WIN
const CalleeSaveRegisterCount = 0
elsif ARMv7
const CalleeSaveRegisterCount = 7
elsif MIPS
const CalleeSaveRegisterCount = 3
elsif X86 or X86_WIN
const CalleeSaveRegisterCount = 3
end
const CalleeRegisterSaveSize = CalleeSaveRegisterCount * MachineRegisterSize
# VMEntryTotalFrameSize includes the space for struct VMEntryRecord and the
# callee save registers rounded up to keep the stack aligned
const VMEntryTotalFrameSize = (CalleeRegisterSaveSize + sizeof VMEntryRecord + StackAlignment - 1) & ~StackAlignmentMask
macro pushCalleeSaves()
if C_LOOP or C_LOOP_WIN or ARM64 or ARM64E or X86_64 or X86_64_WIN
elsif ARMv7
emit "push {r4-r6, r8-r11}"
elsif MIPS
emit "addiu $sp, $sp, -12"
emit "sw $s0, 0($sp)" # csr0/metaData
emit "sw $s1, 4($sp)" # csr1/PB
emit "sw $s4, 8($sp)"
# save $gp to $s4 so that we can restore it after a function call
emit "move $s4, $gp"
elsif X86
emit "push %esi"
emit "push %edi"
emit "push %ebx"
elsif X86_WIN
emit "push esi"
emit "push edi"
emit "push ebx"
end
end
macro popCalleeSaves()
if C_LOOP or C_LOOP_WIN or ARM64 or ARM64E or X86_64 or X86_64_WIN
elsif ARMv7
emit "pop {r4-r6, r8-r11}"
elsif MIPS
emit "lw $s0, 0($sp)"
emit "lw $s1, 4($sp)"
emit "lw $s4, 8($sp)"
emit "addiu $sp, $sp, 12"
elsif X86
emit "pop %ebx"
emit "pop %edi"
emit "pop %esi"
elsif X86_WIN
emit "pop ebx"
emit "pop edi"
emit "pop esi"
end
end
macro preserveCallerPCAndCFR()
if C_LOOP or C_LOOP_WIN or ARMv7 or MIPS
push lr
push cfr
elsif X86 or X86_WIN or X86_64 or X86_64_WIN
push cfr
elsif ARM64 or ARM64E
push cfr, lr
else
error
end
move sp, cfr
end
macro restoreCallerPCAndCFR()
move cfr, sp
if C_LOOP or C_LOOP_WIN or ARMv7 or MIPS
pop cfr
pop lr
elsif X86 or X86_WIN or X86_64 or X86_64_WIN
pop cfr
elsif ARM64 or ARM64E
pop lr, cfr
end
end
macro preserveCalleeSavesUsedByLLInt()
subp CalleeSaveSpaceStackAligned, sp
if C_LOOP or C_LOOP_WIN
storep metadataTable, -PtrSize[cfr]
# Next ARMv7 and MIPS differ in how we store metadataTable and PB,
# because this codes needs to be in sync with how registers are
# restored in Baseline JIT (specifically in emitRestoreCalleeSavesFor).
# emitRestoreCalleeSavesFor restores registers in order instead of by name.
# However, ARMv7 and MIPS differ in the order in which registers are assigned
# to metadataTable and PB, therefore they can also not have the same saving
# order.
elsif ARMv7
storep metadataTable, -4[cfr]
storep PB, -8[cfr]
elsif MIPS
storep PB, -4[cfr]
storep metadataTable, -8[cfr]
elsif ARM64 or ARM64E
emit "stp x27, x28, [x29, #-16]"
emit "stp x25, x26, [x29, #-32]"
elsif X86
elsif X86_WIN
elsif X86_64
storep csr4, -8[cfr]
storep csr3, -16[cfr]
storep csr2, -24[cfr]
storep csr1, -32[cfr]
elsif X86_64_WIN
storep csr6, -8[cfr]
storep csr5, -16[cfr]
storep csr4, -24[cfr]
storep csr3, -32[cfr]
end
end
macro restoreCalleeSavesUsedByLLInt()
if C_LOOP or C_LOOP_WIN
loadp -PtrSize[cfr], metadataTable
# To understand why ARMv7 and MIPS differ in restore order,
# see comment in preserveCalleeSavesUsedByLLInt
elsif ARMv7
loadp -4[cfr], metadataTable
loadp -8[cfr], PB
elsif MIPS
loadp -4[cfr], PB
loadp -8[cfr], metadataTable
elsif ARM64 or ARM64E
emit "ldp x25, x26, [x29, #-32]"
emit "ldp x27, x28, [x29, #-16]"
elsif X86
elsif X86_WIN
elsif X86_64
loadp -32[cfr], csr1
loadp -24[cfr], csr2
loadp -16[cfr], csr3
loadp -8[cfr], csr4
elsif X86_64_WIN
loadp -32[cfr], csr3
loadp -24[cfr], csr4
loadp -16[cfr], csr5
loadp -8[cfr], csr6
end
end
macro copyCalleeSavesToEntryFrameCalleeSavesBuffer(entryFrame)
if ARM64 or ARM64E or X86_64 or X86_64_WIN or ARMv7 or MIPS
vmEntryRecord(entryFrame, entryFrame)
leap VMEntryRecord::calleeSaveRegistersBuffer[entryFrame], entryFrame
if ARM64 or ARM64E
storeq csr0, [entryFrame]
storeq csr1, 8[entryFrame]
storeq csr2, 16[entryFrame]
storeq csr3, 24[entryFrame]
storeq csr4, 32[entryFrame]
storeq csr5, 40[entryFrame]
storeq csr6, 48[entryFrame]
storeq csr7, 56[entryFrame]
storeq csr8, 64[entryFrame]
storeq csr9, 72[entryFrame]
stored csfr0, 80[entryFrame]
stored csfr1, 88[entryFrame]
stored csfr2, 96[entryFrame]
stored csfr3, 104[entryFrame]
stored csfr4, 112[entryFrame]
stored csfr5, 120[entryFrame]
stored csfr6, 128[entryFrame]
stored csfr7, 136[entryFrame]
elsif X86_64
storeq csr0, [entryFrame]
storeq csr1, 8[entryFrame]
storeq csr2, 16[entryFrame]
storeq csr3, 24[entryFrame]
storeq csr4, 32[entryFrame]
elsif X86_64_WIN
storeq csr0, [entryFrame]
storeq csr1, 8[entryFrame]
storeq csr2, 16[entryFrame]
storeq csr3, 24[entryFrame]
storeq csr4, 32[entryFrame]
storeq csr5, 40[entryFrame]
storeq csr6, 48[entryFrame]
elsif ARMv7 or MIPS
storep csr0, [entryFrame]
storep csr1, 4[entryFrame]
end
end
end
macro copyCalleeSavesToVMEntryFrameCalleeSavesBuffer(vm, temp)
if ARM64 or ARM64E or X86_64 or X86_64_WIN or ARMv7 or MIPS
loadp VM::topEntryFrame[vm], temp
copyCalleeSavesToEntryFrameCalleeSavesBuffer(temp)
end
end
macro restoreCalleeSavesFromVMEntryFrameCalleeSavesBuffer(vm, temp)
if ARM64 or ARM64E or X86_64 or X86_64_WIN or ARMv7 or MIPS
loadp VM::topEntryFrame[vm], temp
vmEntryRecord(temp, temp)
leap VMEntryRecord::calleeSaveRegistersBuffer[temp], temp
if ARM64 or ARM64E
loadq [temp], csr0
loadq 8[temp], csr1
loadq 16[temp], csr2
loadq 24[temp], csr3
loadq 32[temp], csr4
loadq 40[temp], csr5
loadq 48[temp], csr6
loadq 56[temp], csr7
loadq 64[temp], csr8
loadq 72[temp], csr9
loadd 80[temp], csfr0
loadd 88[temp], csfr1
loadd 96[temp], csfr2
loadd 104[temp], csfr3
loadd 112[temp], csfr4
loadd 120[temp], csfr5
loadd 128[temp], csfr6
loadd 136[temp], csfr7
elsif X86_64
loadq [temp], csr0
loadq 8[temp], csr1
loadq 16[temp], csr2
loadq 24[temp], csr3
loadq 32[temp], csr4
elsif X86_64_WIN
loadq [temp], csr0
loadq 8[temp], csr1
loadq 16[temp], csr2
loadq 24[temp], csr3
loadq 32[temp], csr4
loadq 40[temp], csr5
loadq 48[temp], csr6
elsif ARMv7 or MIPS
loadp [temp], csr0
loadp 4[temp], csr1
end
end
end
macro preserveReturnAddressAfterCall(destinationRegister)
if C_LOOP or C_LOOP_WIN or ARMv7 or ARM64 or ARM64E or MIPS
# In C_LOOP or C_LOOP_WIN case, we're only preserving the bytecode vPC.
move lr, destinationRegister
elsif X86 or X86_WIN or X86_64 or X86_64_WIN
pop destinationRegister
else
error
end
end
macro functionPrologue()
tagReturnAddress sp
if X86 or X86_WIN or X86_64 or X86_64_WIN
push cfr
elsif ARM64 or ARM64E
push cfr, lr
elsif C_LOOP or C_LOOP_WIN or ARMv7 or MIPS
push lr
push cfr
end
move sp, cfr
end
macro functionEpilogue()
if X86 or X86_WIN or X86_64 or X86_64_WIN
pop cfr
elsif ARM64 or ARM64E
pop lr, cfr
elsif C_LOOP or C_LOOP_WIN or ARMv7 or MIPS
pop cfr
pop lr
end
end
macro vmEntryRecord(entryFramePointer, resultReg)
subp entryFramePointer, VMEntryTotalFrameSize, resultReg
end
macro getFrameRegisterSizeForCodeBlock(codeBlock, size)
loadi CodeBlock::m_numCalleeLocals[codeBlock], size
lshiftp 3, size
addp maxFrameExtentForSlowPathCall, size
end
macro restoreStackPointerAfterCall()
loadp CodeBlock[cfr], t2
getFrameRegisterSizeForCodeBlock(t2, t2)
if ARMv7
subp cfr, t2, t2
move t2, sp
else
subp cfr, t2, sp
end
end
macro traceExecution()
if TRACING
callSlowPath(_llint_trace)
end
end
macro defineOSRExitReturnLabel(opcodeName, size)
macro defineNarrow()
if not C_LOOP_WIN
_%opcodeName%_return_location:
end
end
macro defineWide16()
if not C_LOOP_WIN
_%opcodeName%_return_location_wide16:
end
end
macro defineWide32()
if not C_LOOP_WIN
_%opcodeName%_return_location_wide32:
end
end
size(defineNarrow, defineWide16, defineWide32, macro (f) f() end)
end
macro callTargetFunction(opcodeName, size, opcodeStruct, dispatch, callee, callPtrTag)
if C_LOOP or C_LOOP_WIN
cloopCallJSFunction callee
else
call callee, callPtrTag
end
if ARMv7 or MIPS
# It is required in ARMv7 and MIPs because global label definitions
# for those architectures generates a set of instructions
# that can clobber LLInt execution, resulting in unexpected
# crashes.
restoreStackPointerAfterCall()
dispatchAfterCall(size, opcodeStruct, dispatch)
end
defineOSRExitReturnLabel(opcodeName, size)
restoreStackPointerAfterCall()
dispatchAfterCall(size, opcodeStruct, dispatch)
end
macro prepareForRegularCall(callee, temp1, temp2, temp3, callPtrTag)
addp CallerFrameAndPCSize, sp
end
# sp points to the new frame
macro prepareForTailCall(callee, temp1, temp2, temp3, callPtrTag)
restoreCalleeSavesUsedByLLInt()
loadi PayloadOffset + ArgumentCountIncludingThis[cfr], temp2
loadp CodeBlock[cfr], temp1
loadi CodeBlock::m_numParameters[temp1], temp1
bilteq temp1, temp2, .noArityFixup
move temp1, temp2
.noArityFixup:
# We assume < 2^28 arguments
muli SlotSize, temp2
addi StackAlignment - 1 + CallFrameHeaderSize, temp2
andi ~StackAlignmentMask, temp2
move cfr, temp1
addp temp2, temp1
loadi PayloadOffset + ArgumentCountIncludingThis[sp], temp2
# We assume < 2^28 arguments
muli SlotSize, temp2
addi StackAlignment - 1 + CallFrameHeaderSize, temp2
andi ~StackAlignmentMask, temp2
if ARMv7 or ARM64 or ARM64E or C_LOOP or C_LOOP_WIN or MIPS
addp CallerFrameAndPCSize, sp
subi CallerFrameAndPCSize, temp2
loadp CallerFrameAndPC::returnPC[cfr], lr
else
addp PtrSize, sp
subi PtrSize, temp2
loadp PtrSize[cfr], temp3
storep temp3, [sp]
end
if ARM64E
addp 16, cfr, temp3
untagReturnAddress temp3
end
subp temp2, temp1
loadp [cfr], cfr
.copyLoop:
if ARM64 and not ADDRESS64
subi MachineRegisterSize, temp2
loadq [sp, temp2, 1], temp3
storeq temp3, [temp1, temp2, 1]
btinz temp2, .copyLoop
else
subi PtrSize, temp2
loadp [sp, temp2, 1], temp3
storep temp3, [temp1, temp2, 1]
btinz temp2, .copyLoop
end
move temp1, sp
jmp callee, callPtrTag
end
macro slowPathForCall(opcodeName, size, opcodeStruct, dispatch, slowPath, prepareCall)
callCallSlowPath(
slowPath,
# Those are r0 and r1
macro (callee, calleeFramePtr)
btpz calleeFramePtr, .dontUpdateSP
move calleeFramePtr, sp
prepareCall(callee, t2, t3, t4, SlowPathPtrTag)
.dontUpdateSP:
callTargetFunction(%opcodeName%_slow, size, opcodeStruct, dispatch, callee, SlowPathPtrTag)
end)
end
macro getterSetterOSRExitReturnPoint(opName, size)
crash() # We don't reach this in straight line code. We only reach it via returning to the code below when reconstructing stack frames during OSR exit.
defineOSRExitReturnLabel(opName, size)
restoreStackPointerAfterCall()
loadi LLIntReturnPC[cfr], PC
end
macro arrayProfile(offset, cellAndIndexingType, metadata, scratch)
const cell = cellAndIndexingType
const indexingType = cellAndIndexingType
loadi JSCell::m_structureID[cell], scratch
storei scratch, offset + ArrayProfile::m_lastSeenStructureID[metadata]
loadb JSCell::m_indexingTypeAndMisc[cell], indexingType
end
macro skipIfIsRememberedOrInEden(cell, slowPath)
memfence
bba JSCell::m_cellState[cell], BlackThreshold, .done
slowPath()
.done:
end
macro notifyWrite(set, slow)
bbneq WatchpointSet::m_state[set], IsInvalidated, slow
end
macro checkSwitchToJIT(increment, action)
loadp CodeBlock[cfr], t0
baddis increment, CodeBlock::m_llintExecuteCounter + BaselineExecutionCounter::m_counter[t0], .continue
action()
.continue:
end
macro checkSwitchToJITForEpilogue()
checkSwitchToJIT(
10,
macro ()
callSlowPath(_llint_replace)
end)
end
macro assertNotConstant(size, index)
size(FirstConstantRegisterIndexNarrow, FirstConstantRegisterIndexWide16, FirstConstantRegisterIndexWide32, macro (FirstConstantRegisterIndex)
assert(macro (ok) bilt index, FirstConstantRegisterIndex, ok end)
end)
end
macro functionForCallCodeBlockGetter(targetRegister)
if JSVALUE64
loadp Callee[cfr], targetRegister
else
loadp Callee + PayloadOffset[cfr], targetRegister
end
loadp JSFunction::m_executableOrRareData[targetRegister], targetRegister
btpz targetRegister, (constexpr JSFunction::rareDataTag), .isExecutable
loadp (FunctionRareData::m_executable - (constexpr JSFunction::rareDataTag))[targetRegister], targetRegister
.isExecutable:
loadp FunctionExecutable::m_codeBlockForCall[targetRegister], targetRegister
loadp ExecutableToCodeBlockEdge::m_codeBlock[targetRegister], targetRegister
end
macro functionForConstructCodeBlockGetter(targetRegister)
if JSVALUE64
loadp Callee[cfr], targetRegister
else
loadp Callee + PayloadOffset[cfr], targetRegister
end
loadp JSFunction::m_executableOrRareData[targetRegister], targetRegister
btpz targetRegister, (constexpr JSFunction::rareDataTag), .isExecutable
loadp (FunctionRareData::m_executable - (constexpr JSFunction::rareDataTag))[targetRegister], targetRegister
.isExecutable:
loadp FunctionExecutable::m_codeBlockForConstruct[targetRegister], targetRegister
loadp ExecutableToCodeBlockEdge::m_codeBlock[targetRegister], targetRegister
end
macro notFunctionCodeBlockGetter(targetRegister)
loadp CodeBlock[cfr], targetRegister
end
macro functionCodeBlockSetter(sourceRegister)
storep sourceRegister, CodeBlock[cfr]
end
macro notFunctionCodeBlockSetter(sourceRegister)
# Nothing to do!
end
macro convertCalleeToVM(callee)
btpnz callee, (constexpr PreciseAllocation::halfAlignment), .preciseAllocation
andp MarkedBlockMask, callee
loadp MarkedBlockFooterOffset + MarkedBlock::Footer::m_vm[callee], callee
jmp .done
.preciseAllocation:
loadp PreciseAllocationVMOffset[callee], callee
.done:
end
# Do the bare minimum required to execute code. Sets up the PC, leave the CodeBlock*
# in t1. May also trigger prologue entry OSR.
macro prologue(codeBlockGetter, codeBlockSetter, osrSlowPath, traceSlowPath)
# Set up the call frame and check if we should OSR.
tagReturnAddress sp
preserveCallerPCAndCFR()
if TRACING
subp maxFrameExtentForSlowPathCall, sp
callSlowPath(traceSlowPath)
addp maxFrameExtentForSlowPathCall, sp
end
codeBlockGetter(t1)
codeBlockSetter(t1)
if not (C_LOOP or C_LOOP_WIN)
baddis 5, CodeBlock::m_llintExecuteCounter + BaselineExecutionCounter::m_counter[t1], .continue
if JSVALUE64
move cfr, a0
move PC, a1
cCall2(osrSlowPath)
else
# We are after the function prologue, but before we have set up sp from the CodeBlock.
# Temporarily align stack pointer for this call.
subp 8, sp
move cfr, a0
move PC, a1
cCall2(osrSlowPath)
addp 8, sp
end
btpz r0, .recover
move cfr, sp # restore the previous sp
# pop the callerFrame since we will jump to a function that wants to save it
if ARM64 or ARM64E
pop lr, cfr
untagReturnAddress sp
elsif ARMv7 or MIPS
pop cfr
pop lr
else
pop cfr
end
jmp r0, JSEntryPtrTag
.recover:
notFunctionCodeBlockGetter(t1)
.continue:
end
preserveCalleeSavesUsedByLLInt()
# Set up the PC.
loadp CodeBlock::m_instructionsRawPointer[t1], PB
move 0, PC
# Get new sp in t0 and check stack height.
getFrameRegisterSizeForCodeBlock(t1, t0)
subp cfr, t0, t0
bpa t0, cfr, .needStackCheck
loadp CodeBlock::m_vm[t1], t2
if C_LOOP or C_LOOP_WIN
bpbeq VM::m_cloopStackLimit[t2], t0, .stackHeightOK
else
bpbeq VM::m_softStackLimit[t2], t0, .stackHeightOK
end
.needStackCheck:
# Stack height check failed - need to call a slow_path.
# Set up temporary stack pointer for call including callee saves
subp maxFrameExtentForSlowPathCall, sp
callSlowPath(_llint_stack_check)
bpeq r1, 0, .stackHeightOKGetCodeBlock
# We're throwing before the frame is fully set up. This frame will be
# ignored by the unwinder. So, let's restore the callee saves before we
# start unwinding. We need to do this before we change the cfr.
restoreCalleeSavesUsedByLLInt()
move r1, cfr
jmp _llint_throw_from_slow_path_trampoline
.stackHeightOKGetCodeBlock:
# Stack check slow path returned that the stack was ok.
# Since they were clobbered, need to get CodeBlock and new sp
notFunctionCodeBlockGetter(t1)
getFrameRegisterSizeForCodeBlock(t1, t0)
subp cfr, t0, t0
.stackHeightOK:
if X86_64 or ARM64
# We need to start zeroing from sp as it has been adjusted after saving callee saves.
move sp, t2
move t0, sp
.zeroStackLoop:
bpeq sp, t2, .zeroStackDone
subp PtrSize, t2
storep 0, [t2]
jmp .zeroStackLoop
.zeroStackDone:
else
move t0, sp
end
loadp CodeBlock::m_metadata[t1], metadataTable
if JSVALUE64
move TagNumber, numberTag
addq TagOther, numberTag, notCellMask
end
end
# Expects that CodeBlock is in t1, which is what prologue() leaves behind.
# Must call dispatch(0) after calling this.
macro functionInitialization(profileArgSkip)
# Profile the arguments. Unfortunately, we have no choice but to do this. This
# code is pretty horrendous because of the difference in ordering between
# arguments and value profiles, the desire to have a simple loop-down-to-zero
# loop, and the desire to use only three registers so as to preserve the PC and
# the code block. It is likely that this code should be rewritten in a more
# optimal way for architectures that have more than five registers available
# for arbitrary use in the interpreter.
loadi CodeBlock::m_numParameters[t1], t0
addp -profileArgSkip, t0 # Use addi because that's what has the peephole
assert(macro (ok) bpgteq t0, 0, ok end)
btpz t0, .argumentProfileDone
loadp CodeBlock::m_argumentValueProfiles + RefCountedArray::m_data[t1], t3
btpz t3, .argumentProfileDone # When we can't JIT, we don't allocate any argument value profiles.
mulp sizeof ValueProfile, t0, t2 # Aaaaahhhh! Need strength reduction!
lshiftp 3, t0 # offset of last JSValue arguments on the stack.
addp t2, t3 # pointer to end of ValueProfile array in CodeBlock::m_argumentValueProfiles.
.argumentProfileLoop:
if JSVALUE64
loadq ThisArgumentOffset - 8 + profileArgSkip * 8[cfr, t0], t2
subp sizeof ValueProfile, t3
storeq t2, profileArgSkip * sizeof ValueProfile + ValueProfile::m_buckets[t3]
else
loadi ThisArgumentOffset + TagOffset - 8 + profileArgSkip * 8[cfr, t0], t2
subp sizeof ValueProfile, t3
storei t2, profileArgSkip * sizeof ValueProfile + ValueProfile::m_buckets + TagOffset[t3]
loadi ThisArgumentOffset + PayloadOffset - 8 + profileArgSkip * 8[cfr, t0], t2
storei t2, profileArgSkip * sizeof ValueProfile + ValueProfile::m_buckets + PayloadOffset[t3]
end
baddpnz -8, t0, .argumentProfileLoop
.argumentProfileDone:
end
macro doReturn()
restoreCalleeSavesUsedByLLInt()
restoreCallerPCAndCFR()
ret
end
# This break instruction is needed so that the synthesized llintPCRangeStart label
# doesn't point to the exact same location as vmEntryToJavaScript which comes after it.
# Otherwise, libunwind will report vmEntryToJavaScript as llintPCRangeStart in
# stack traces.
break
# stub to call into JavaScript or Native functions
# EncodedJSValue vmEntryToJavaScript(void* code, VM* vm, ProtoCallFrame* protoFrame)
# EncodedJSValue vmEntryToNativeFunction(void* code, VM* vm, ProtoCallFrame* protoFrame)
if C_LOOP or C_LOOP_WIN
_llint_vm_entry_to_javascript:
else
global _vmEntryToJavaScript
_vmEntryToJavaScript:
end
doVMEntry(makeJavaScriptCall)
if C_LOOP or C_LOOP_WIN
_llint_vm_entry_to_native:
else
global _vmEntryToNative
_vmEntryToNative:
end
doVMEntry(makeHostFunctionCall)
if not (C_LOOP or C_LOOP_WIN)
# void sanitizeStackForVMImpl(VM* vm)
global _sanitizeStackForVMImpl
_sanitizeStackForVMImpl:
tagReturnAddress sp
# We need three non-aliased caller-save registers. We are guaranteed
# this for a0, a1 and a2 on all architectures.
if X86 or X86_WIN
loadp 4[sp], a0
end
const vmOrStartSP = a0
const address = a1
const zeroValue = a2
loadp VM::m_lastStackTop[vmOrStartSP], address
move sp, zeroValue
storep zeroValue, VM::m_lastStackTop[vmOrStartSP]
move sp, vmOrStartSP
bpbeq sp, address, .zeroFillDone
move address, sp
move 0, zeroValue
.zeroFillLoop:
storep zeroValue, [address]
addp PtrSize, address
bpa vmOrStartSP, address, .zeroFillLoop
.zeroFillDone:
move vmOrStartSP, sp
ret
# VMEntryRecord* vmEntryRecord(const EntryFrame* entryFrame)
global _vmEntryRecord
_vmEntryRecord:
tagReturnAddress sp
if X86 or X86_WIN
loadp 4[sp], a0
end
vmEntryRecord(a0, r0)
ret
end
if C_LOOP or C_LOOP_WIN
# Dummy entry point the C Loop uses to initialize.
_llint_entry:
crash()
else
macro initPCRelative(kind, pcBase)
if X86_64 or X86_64_WIN or X86 or X86_WIN
call _%kind%_relativePCBase
_%kind%_relativePCBase:
pop pcBase
elsif ARM64 or ARM64E
elsif ARMv7
_%kind%_relativePCBase:
move pc, pcBase
subp 3, pcBase # Need to back up the PC and set the Thumb2 bit
elsif MIPS
la _%kind%_relativePCBase, pcBase
setcallreg pcBase # needed to set $t9 to the right value for the .cpload created by the label.
_%kind%_relativePCBase:
end
end
# The PC base is in t3, as this is what _llint_entry leaves behind through
# initPCRelative(t3)
macro setEntryAddressCommon(kind, index, label, map)
if X86_64
leap (label - _%kind%_relativePCBase)[t3], t4
move index, t5
storep t4, [map, t5, 8]
elsif X86_64_WIN
leap (label - _%kind%_relativePCBase)[t3], t4
move index, t0
storep t4, [map, t0, 8]
elsif X86 or X86_WIN
leap (label - _%kind%_relativePCBase)[t3], t4
move index, t5
storep t4, [map, t5, 4]
elsif ARM64 or ARM64E
pcrtoaddr label, t3
move index, t4
storep t3, [map, t4, PtrSize]
elsif ARMv7
mvlbl (label - _%kind%_relativePCBase), t4
addp t4, t3, t4
move index, t5
storep t4, [map, t5, 4]
elsif MIPS
la label, t4
la _%kind%_relativePCBase, t3
subp t3, t4
addp t4, t3, t4
move index, t5
storep t4, [map, t5, 4]
end
end
macro includeEntriesAtOffset(kind, fn)
macro setEntryAddress(index, label)
setEntryAddressCommon(kind, index, label, a0)
end
macro setEntryAddressWide16(index, label)
setEntryAddressCommon(kind, index, label, a1)
end
macro setEntryAddressWide32(index, label)
setEntryAddressCommon(kind, index, label, a2)
end
fn()
end
macro entry(kind, initialize)
global _%kind%_entry
_%kind%_entry:
functionPrologue()
pushCalleeSaves()
if X86 or X86_WIN
loadp 20[sp], a0
loadp 24[sp], a1
loadp 28[sp], a2
end
initPCRelative(kind, t3)
# Include generated bytecode initialization file.
includeEntriesAtOffset(kind, initialize)
popCalleeSaves()
functionEpilogue()
ret
end
# Entry point for the llint to initialize.
entry(llint, macro()
include InitBytecodes
end)
end // not (C_LOOP or C_LOOP_WIN)
_llint_op_wide16:
nextInstructionWide16()
_llint_op_wide32:
nextInstructionWide32()
macro noWide(label)
_%label%_wide16:
crash()
_%label%_wide32:
crash()
end
noWide(llint_op_wide16)
noWide(llint_op_wide32)
noWide(llint_op_enter)
op(llint_program_prologue, macro ()
prologue(notFunctionCodeBlockGetter, notFunctionCodeBlockSetter, _llint_entry_osr, _llint_trace_prologue)
dispatch(0)
end)
op(llint_module_program_prologue, macro ()
prologue(notFunctionCodeBlockGetter, notFunctionCodeBlockSetter, _llint_entry_osr, _llint_trace_prologue)
dispatch(0)
end)
op(llint_eval_prologue, macro ()
prologue(notFunctionCodeBlockGetter, notFunctionCodeBlockSetter, _llint_entry_osr, _llint_trace_prologue)
dispatch(0)
end)
op(llint_function_for_call_prologue, macro ()
prologue(functionForCallCodeBlockGetter, functionCodeBlockSetter, _llint_entry_osr_function_for_call, _llint_trace_prologue_function_for_call)
functionInitialization(0)
dispatch(0)
end)
op(llint_function_for_construct_prologue, macro ()
prologue(functionForConstructCodeBlockGetter, functionCodeBlockSetter, _llint_entry_osr_function_for_construct, _llint_trace_prologue_function_for_construct)
functionInitialization(1)
dispatch(0)
end)
op(llint_function_for_call_arity_check, macro ()
prologue(functionForCallCodeBlockGetter, functionCodeBlockSetter, _llint_entry_osr_function_for_call_arityCheck, _llint_trace_arityCheck_for_call)
functionArityCheck(.functionForCallBegin, _slow_path_call_arityCheck)
.functionForCallBegin:
functionInitialization(0)
dispatch(0)
end)
op(llint_function_for_construct_arity_check, macro ()
prologue(functionForConstructCodeBlockGetter, functionCodeBlockSetter, _llint_entry_osr_function_for_construct_arityCheck, _llint_trace_arityCheck_for_construct)
functionArityCheck(.functionForConstructBegin, _slow_path_construct_arityCheck)
.functionForConstructBegin:
functionInitialization(1)
dispatch(0)
end)
# Value-representation-specific code.
if JSVALUE64
include LowLevelInterpreter64
else
include LowLevelInterpreter32_64
end
# Value-representation-agnostic code.
macro slowPathOp(opcodeName)
llintOp(op_%opcodeName%, unused, macro (unused, unused, dispatch)
callSlowPath(_slow_path_%opcodeName%)
dispatch()
end)
end
slowPathOp(create_cloned_arguments)
slowPathOp(create_arguments_butterfly)
slowPathOp(create_direct_arguments)
slowPathOp(create_lexical_environment)
slowPathOp(create_rest)
slowPathOp(create_scoped_arguments)
slowPathOp(create_this)
slowPathOp(create_promise)
slowPathOp(create_generator)
slowPathOp(create_async_generator)
slowPathOp(define_accessor_property)
slowPathOp(define_data_property)
slowPathOp(enumerator_generic_pname)
slowPathOp(enumerator_structure_pname)
slowPathOp(get_by_id_with_this)
slowPathOp(get_by_val_with_this)
slowPathOp(get_direct_pname)
slowPathOp(get_enumerable_length)
slowPathOp(get_property_enumerator)
slowPathOp(greater)
slowPathOp(greatereq)
slowPathOp(has_generic_property)
slowPathOp(has_indexed_property)
slowPathOp(has_structure_property)
slowPathOp(in_by_id)
slowPathOp(in_by_val)
slowPathOp(is_function)
slowPathOp(is_object_or_null)
slowPathOp(less)
slowPathOp(lesseq)
slowPathOp(mod)
slowPathOp(new_array_buffer)
slowPathOp(new_array_with_spread)
slowPathOp(pow)
slowPathOp(push_with_scope)
slowPathOp(put_by_id_with_this)
slowPathOp(put_by_val_with_this)
slowPathOp(resolve_scope_for_hoisting_func_decl_in_eval)
slowPathOp(spread)
slowPathOp(strcat)
slowPathOp(throw_static_error)
slowPathOp(to_index_string)
slowPathOp(typeof)
slowPathOp(unreachable)
slowPathOp(new_promise)
slowPathOp(new_generator)
macro llintSlowPathOp(opcodeName)
llintOp(op_%opcodeName%, unused, macro (unused, unused, dispatch)
callSlowPath(_llint_slow_path_%opcodeName%)
dispatch()
end)
end
llintSlowPathOp(del_by_id)
llintSlowPathOp(del_by_val)
llintSlowPathOp(instanceof)
llintSlowPathOp(instanceof_custom)
llintSlowPathOp(new_array)
llintSlowPathOp(new_array_with_size)
llintSlowPathOp(new_async_func)
llintSlowPathOp(new_async_func_exp)
llintSlowPathOp(new_async_generator_func)
llintSlowPathOp(new_async_generator_func_exp)
llintSlowPathOp(new_func)
llintSlowPathOp(new_func_exp)
llintSlowPathOp(new_generator_func)
llintSlowPathOp(new_generator_func_exp)
llintSlowPathOp(new_object)
llintSlowPathOp(new_regexp)
llintSlowPathOp(put_getter_by_id)
llintSlowPathOp(put_getter_by_val)
llintSlowPathOp(put_getter_setter_by_id)
llintSlowPathOp(put_setter_by_id)
llintSlowPathOp(put_setter_by_val)
llintSlowPathOp(set_function_name)
llintSlowPathOp(super_sampler_begin)
llintSlowPathOp(super_sampler_end)
llintSlowPathOp(throw)
llintSlowPathOp(try_get_by_id)
llintOp(op_switch_string, unused, macro (unused, unused, unused)
callSlowPath(_llint_slow_path_switch_string)
nextInstruction()
end)
equalityComparisonOp(eq, OpEq,
macro (left, right, result) cieq left, right, result end)
equalityComparisonOp(neq, OpNeq,
macro (left, right, result) cineq left, right, result end)
compareUnsignedOp(below, OpBelow,
macro (left, right, result) cib left, right, result end)
compareUnsignedOp(beloweq, OpBeloweq,
macro (left, right, result) cibeq left, right, result end)
llintOpWithJump(op_jmp, OpJmp, macro (size, get, jump, dispatch)
jump(m_targetLabel)
end)
llintJumpTrueOrFalseOp(jtrue, OpJtrue,
# Misc primitive
macro (value, target) btinz value, 1, target end,
# Truthy Cell
macro (dispatch) end)
llintJumpTrueOrFalseOp(jfalse, OpJfalse,
# Misc primitive
macro (value, target) btiz value, 1, target end,
# Truthy Cell
macro (dispatch) dispatch() end)
compareJumpOp(
jless, OpJless,
macro (left, right, target) bilt left, right, target end,
macro (left, right, target) bdlt left, right, target end)
compareJumpOp(
jnless, OpJnless,
macro (left, right, target) bigteq left, right, target end,
macro (left, right, target) bdgtequn left, right, target end)
compareJumpOp(
jgreater, OpJgreater,
macro (left, right, target) bigt left, right, target end,
macro (left, right, target) bdgt left, right, target end)
compareJumpOp(
jngreater, OpJngreater,
macro (left, right, target) bilteq left, right, target end,
macro (left, right, target) bdltequn left, right, target end)
compareJumpOp(
jlesseq, OpJlesseq,
macro (left, right, target) bilteq left, right, target end,
macro (left, right, target) bdlteq left, right, target end)
compareJumpOp(
jnlesseq, OpJnlesseq,
macro (left, right, target) bigt left, right, target end,
macro (left, right, target) bdgtun left, right, target end)
compareJumpOp(
jgreatereq, OpJgreatereq,
macro (left, right, target) bigteq left, right, target end,
macro (left, right, target) bdgteq left, right, target end)
compareJumpOp(
jngreatereq, OpJngreatereq,
macro (left, right, target) bilt left, right, target end,
macro (left, right, target) bdltun left, right, target end)
equalityJumpOp(
jeq, OpJeq,
macro (left, right, target) bieq left, right, target end)
equalityJumpOp(
jneq, OpJneq,
macro (left, right, target) bineq left, right, target end)
compareUnsignedJumpOp(
jbelow, OpJbelow,
macro (left, right, target) bib left, right, target end)
compareUnsignedJumpOp(
jbeloweq, OpJbeloweq,
macro (left, right, target) bibeq left, right, target end)
preOp(inc, OpInc,
macro (value, slow) baddio 1, value, slow end)
preOp(dec, OpDec,
macro (value, slow) bsubio 1, value, slow end)
llintOp(op_loop_hint, OpLoopHint, macro (unused, unused, dispatch)
checkSwitchToJITForLoop()
dispatch()
end)
llintOp(op_check_traps, OpCheckTraps, macro (unused, unused, dispatch)
loadp CodeBlock[cfr], t1
loadp CodeBlock::m_vm[t1], t1
loadb VM::m_traps+VMTraps::m_needTrapHandling[t1], t0
btpnz t0, .handleTraps
.afterHandlingTraps:
dispatch()
.handleTraps:
callTrapHandler(.throwHandler)
jmp .afterHandlingTraps
.throwHandler:
jmp _llint_throw_from_slow_path_trampoline
end)
# Returns the packet pointer in t0.
macro acquireShadowChickenPacket(slow)
loadp CodeBlock[cfr], t1
loadp CodeBlock::m_vm[t1], t1
loadp VM::m_shadowChicken[t1], t2
loadp ShadowChicken::m_logCursor[t2], t0
bpaeq t0, ShadowChicken::m_logEnd[t2], slow
addp sizeof ShadowChicken::Packet, t0, t1
storep t1, ShadowChicken::m_logCursor[t2]
end
llintOp(op_nop, OpNop, macro (unused, unused, dispatch)
dispatch()
end)
# we can't use callOp because we can't pass `call` as the opcode name, since it's an instruction name
commonCallOp(op_call, _llint_slow_path_call, OpCall, prepareForRegularCall, macro (getu, metadata)
arrayProfileForCall(OpCall, getu)
end)
macro callOp(opcodeName, opcodeStruct, prepareCall, fn)
commonCallOp(op_%opcodeName%, _llint_slow_path_%opcodeName%, opcodeStruct, prepareCall, fn)
end
callOp(tail_call, OpTailCall, prepareForTailCall, macro (getu, metadata)
arrayProfileForCall(OpTailCall, getu)
checkSwitchToJITForEpilogue()
# reload metadata since checkSwitchToJITForEpilogue() might have trashed t5
metadata(t5, t0)
end)
callOp(construct, OpConstruct, prepareForRegularCall, macro (getu, metadata) end)
macro branchIfException(exceptionTarget)
loadp CodeBlock[cfr], t3
loadp CodeBlock::m_vm[t3], t3
btpz VM::m_exception[t3], .noException
jmp exceptionTarget
.noException:
end
macro doCallVarargs(opcodeName, size, opcodeStruct, dispatch, frameSlowPath, slowPath, prepareCall)
callSlowPath(frameSlowPath)
branchIfException(_llint_throw_from_slow_path_trampoline)
# calleeFrame in r1
if JSVALUE64
move r1, sp
else
# The calleeFrame is not stack aligned, move down by CallerFrameAndPCSize to align
if ARMv7
subp r1, CallerFrameAndPCSize, t2
move t2, sp
else
subp r1, CallerFrameAndPCSize, sp
end
end
slowPathForCall(opcodeName, size, opcodeStruct, dispatch, slowPath, prepareCall)
end
llintOp(op_call_varargs, OpCallVarargs, macro (size, get, dispatch)
doCallVarargs(op_call_varargs, size, OpCallVarargs, dispatch, _llint_slow_path_size_frame_for_varargs, _llint_slow_path_call_varargs, prepareForRegularCall)
end)
llintOp(op_tail_call_varargs, OpTailCallVarargs, macro (size, get, dispatch)
checkSwitchToJITForEpilogue()
# We lie and perform the tail call instead of preparing it since we can't
# prepare the frame for a call opcode
doCallVarargs(op_tail_call_varargs, size, OpTailCallVarargs, dispatch, _llint_slow_path_size_frame_for_varargs, _llint_slow_path_tail_call_varargs, prepareForTailCall)
end)
llintOp(op_tail_call_forward_arguments, OpTailCallForwardArguments, macro (size, get, dispatch)
checkSwitchToJITForEpilogue()
# We lie and perform the tail call instead of preparing it since we can't
# prepare the frame for a call opcode
doCallVarargs(op_tail_call_forward_arguments, size, OpTailCallForwardArguments, dispatch, _llint_slow_path_size_frame_for_forward_arguments, _llint_slow_path_tail_call_forward_arguments, prepareForTailCall)
end)
llintOp(op_construct_varargs, OpConstructVarargs, macro (size, get, dispatch)
doCallVarargs(op_construct_varargs, size, OpConstructVarargs, dispatch, _llint_slow_path_size_frame_for_varargs, _llint_slow_path_construct_varargs, prepareForRegularCall)
end)
# Eval is executed in one of two modes:
#
# 1) We find that we're really invoking eval() in which case the
# execution is perfomed entirely inside the slow_path, and it
# returns the PC of a function that just returns the return value
# that the eval returned.
#
# 2) We find that we're invoking something called eval() that is not
# the real eval. Then the slow_path returns the PC of the thing to
# call, and we call it.
#
# This allows us to handle two cases, which would require a total of
# up to four pieces of state that cannot be easily packed into two
# registers (C functions can return up to two registers, easily):
#
# - The call frame register. This may or may not have been modified
# by the slow_path, but the convention is that it returns it. It's not
# totally clear if that's necessary, since the cfr is callee save.
# But that's our style in this here interpreter so we stick with it.
#
# - A bit to say if the slow_path successfully executed the eval and has
# the return value, or did not execute the eval but has a PC for us
# to call.
#
# - Either:
# - The JS return value (two registers), or
#
# - The PC to call.
#
# It turns out to be easier to just always have this return the cfr
# and a PC to call, and that PC may be a dummy thunk that just
# returns the JS value that the eval returned.
_llint_op_call_eval:
slowPathForCall(
op_call_eval_narrow,
narrow,
OpCallEval,
macro () dispatchOp(narrow, op_call_eval) end,
_llint_slow_path_call_eval,
prepareForRegularCall)
_llint_op_call_eval_wide16:
slowPathForCall(
op_call_eval_wide16,
wide16,
OpCallEval,
macro () dispatchOp(wide16, op_call_eval) end,
_llint_slow_path_call_eval_wide16,
prepareForRegularCall)
_llint_op_call_eval_wide32:
slowPathForCall(
op_call_eval_wide32,
wide32,
OpCallEval,
macro () dispatchOp(wide32, op_call_eval) end,
_llint_slow_path_call_eval_wide32,
prepareForRegularCall)
commonOp(llint_generic_return_point, macro () end, macro (size)
dispatchAfterCall(size, OpCallEval, macro ()
dispatchOp(size, op_call_eval)
end)
end)
llintOp(op_identity_with_profile, OpIdentityWithProfile, macro (unused, unused, dispatch)
dispatch()
end)
llintOp(op_yield, OpYield, macro (unused, unused, unused)
notSupported()
end)
llintOp(op_create_generator_frame_environment, OpYield, macro (unused, unused, unused)
notSupported()
end)
llintOp(op_debug, OpDebug, macro (unused, unused, dispatch)
loadp CodeBlock[cfr], t0
loadi CodeBlock::m_debuggerRequests[t0], t0
btiz t0, .opDebugDone
callSlowPath(_llint_slow_path_debug)
.opDebugDone:
dispatch()
end)
op(llint_native_call_trampoline, macro ()
nativeCallTrampoline(NativeExecutable::m_function)
end)
op(llint_native_construct_trampoline, macro ()
nativeCallTrampoline(NativeExecutable::m_constructor)
end)
op(llint_internal_function_call_trampoline, macro ()
internalFunctionCallTrampoline(InternalFunction::m_functionForCall)
end)
op(llint_internal_function_construct_trampoline, macro ()
internalFunctionCallTrampoline(InternalFunction::m_functionForConstruct)
end)
op(checkpoint_osr_exit_from_inlined_call_trampoline, macro ()
if (JSVALUE64 and not (C_LOOP or C_LOOP_WIN)) or ARMv7 or MIPS
restoreStackPointerAfterCall()
# Make sure we move r0 to a1 first since r0 might be the same as a0, for instance, on arm.
if ARMv7 or MIPS
# Given _slow_path_checkpoint_osr_exit_from_inlined_call has
# parameters as CallFrame* and EncodedJSValue,
# we need to store call result on a2, a3 and call frame on a0,
# leaving a1 as dummy value.
move r1, a3
move r0, a2
move cfr, a0
# We don't call saveStateForCCall() because we are going to use the bytecodeIndex from our side state.
cCall4(_slow_path_checkpoint_osr_exit_from_inlined_call)
else
move r0, a1
move cfr, a0
# We don't call saveStateForCCall() because we are going to use the bytecodeIndex from our side state.
cCall2(_slow_path_checkpoint_osr_exit_from_inlined_call)
end
restoreStateAfterCCall()
branchIfException(_llint_throw_from_slow_path_trampoline)
jmp r1, JSEntryPtrTag
else
notSupported()
end
end)
op(checkpoint_osr_exit_trampoline, macro ()
# FIXME: We can probably dispatch to the checkpoint handler directly but this was easier
# and probably doesn't matter for performance.
if (JSVALUE64 and not (C_LOOP or C_LOOP_WIN)) or ARMv7 or MIPS
restoreStackPointerAfterCall()
move cfr, a0
# We don't call saveStateForCCall() because we are going to use the bytecodeIndex from our side state.
cCall2(_slow_path_checkpoint_osr_exit)
restoreStateAfterCCall()
branchIfException(_llint_throw_from_slow_path_trampoline)
jmp r1, JSEntryPtrTag
else
notSupported()
end
end)
# Lastly, make sure that we can link even though we don't support all opcodes.
# These opcodes should never arise when using LLInt or either JIT. We assert
# as much.
macro notSupported()
if ASSERT_ENABLED
crash()
else
# We should use whatever the smallest possible instruction is, just to
# ensure that there is a gap between instruction labels. If multiple
# smallest instructions exist, we should pick the one that is most
# likely result in execution being halted. Currently that is the break
# instruction on all architectures we're interested in. (Break is int3
# on Intel, which is 1 byte, and bkpt on ARMv7, which is 2 bytes.)
break
end
end
// FIXME: We should not need the X86_64_WIN condition here, since WEBASSEMBLY should already be false on Windows
// https://bugs.webkit.org/show_bug.cgi?id=203716
if WEBASSEMBLY and not X86_64_WIN
entry(wasm, macro()
include InitWasm
end)
macro wasmScope()
# Wrap the script in a macro since it overwrites some of the LLInt macros,
# but we don't want to interfere with the LLInt opcodes
include WebAssembly
end
wasmScope()
else
# These need to be defined even when WebAssembly is disabled
op(wasm_function_prologue, macro ()
crash()
end)
op(wasm_function_prologue_no_tls, macro ()
crash()
end)
end