update go-ffmpreg to v0.2.5 (pulls in latest tetratelabs/wazero) (#3203)

10 files changed, 472 insertions, 216 deletions

diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/abi.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/abi.go
index 4eaa13ce1..d1eaa7cd4 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/abi.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/abi.go
@@ -182,9 +182,9 @@ func (m *machine) LowerReturns(rets []ssa.Value) {
 
 // callerGenVRegToFunctionArg is the opposite of GenFunctionArgToVReg, which is used to generate the
 // caller side of the function call.
-func (m *machine) callerGenVRegToFunctionArg(a *backend.FunctionABI, argIndex int, reg regalloc.VReg, def *backend.SSAValueDefinition, slotBegin int64) {
+func (m *machine) callerGenVRegToFunctionArg(a *backend.FunctionABI, argIndex int, reg regalloc.VReg, def backend.SSAValueDefinition, slotBegin int64) {
 	arg := &a.Args[argIndex]
-	if def != nil && def.IsFromInstr() {
+	if def.IsFromInstr() {
 		// Constant instructions are inlined.
 		if inst := def.Instr; inst.Constant() {
 			val := inst.Return()
@@ -228,10 +228,9 @@ func (m *machine) callerGenFunctionReturnVReg(a *backend.FunctionABI, retIndex i
 }
 
 func (m *machine) resolveAddressModeForOffsetAndInsert(cur *instruction, offset int64, dstBits byte, rn regalloc.VReg, allowTmpRegUse bool) (*instruction, *addressMode) {
-	exct := m.executableContext
-	exct.PendingInstructions = exct.PendingInstructions[:0]
+	m.pendingInstructions = m.pendingInstructions[:0]
 	mode := m.resolveAddressModeForOffset(offset, dstBits, rn, allowTmpRegUse)
-	for _, instr := range exct.PendingInstructions {
+	for _, instr := range m.pendingInstructions {
 		cur = linkInstr(cur, instr)
 	}
 	return cur, mode
diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/abi_go_call.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/abi_go_call.go
index 99e6bb482..06f8a4a05 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/abi_go_call.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/abi_go_call.go
@@ -14,7 +14,6 @@ var calleeSavedRegistersSorted = []regalloc.VReg{
 
 // CompileGoFunctionTrampoline implements backend.Machine.
 func (m *machine) CompileGoFunctionTrampoline(exitCode wazevoapi.ExitCode, sig *ssa.Signature, needModuleContextPtr bool) []byte {
-	exct := m.executableContext
 	argBegin := 1 // Skips exec context by default.
 	if needModuleContextPtr {
 		argBegin++
@@ -26,7 +25,7 @@ func (m *machine) CompileGoFunctionTrampoline(exitCode wazevoapi.ExitCode, sig *
 
 	cur := m.allocateInstr()
 	cur.asNop0()
-	exct.RootInstr = cur
+	m.rootInstr = cur
 
 	// Execution context is always the first argument.
 	execCtrPtr := x0VReg
@@ -244,7 +243,7 @@ func (m *machine) CompileGoFunctionTrampoline(exitCode wazevoapi.ExitCode, sig *
 	ret.asRet()
 	linkInstr(cur, ret)
 
-	m.encode(m.executableContext.RootInstr)
+	m.encode(m.rootInstr)
 	return m.compiler.Buf()
 }
 
@@ -302,20 +301,18 @@ func (m *machine) restoreRegistersInExecutionContext(cur *instruction, regs []re
 }
 
 func (m *machine) lowerConstantI64AndInsert(cur *instruction, dst regalloc.VReg, v int64) *instruction {
-	exct := m.executableContext
-	exct.PendingInstructions = exct.PendingInstructions[:0]
+	m.pendingInstructions = m.pendingInstructions[:0]
 	m.lowerConstantI64(dst, v)
-	for _, instr := range exct.PendingInstructions {
+	for _, instr := range m.pendingInstructions {
 		cur = linkInstr(cur, instr)
 	}
 	return cur
 }
 
 func (m *machine) lowerConstantI32AndInsert(cur *instruction, dst regalloc.VReg, v int32) *instruction {
-	exct := m.executableContext
-	exct.PendingInstructions = exct.PendingInstructions[:0]
+	m.pendingInstructions = m.pendingInstructions[:0]
 	m.lowerConstantI32(dst, v)
-	for _, instr := range exct.PendingInstructions {
+	for _, instr := range m.pendingInstructions {
 		cur = linkInstr(cur, instr)
 	}
 	return cur
diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/instr.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/instr.go
index 7121cb538..1f563428a 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/instr.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/instr.go
@@ -36,18 +36,6 @@ type (
 	instructionKind byte
 )
 
-func asNop0(i *instruction) {
-	i.kind = nop0
-}
-
-func setNext(i, next *instruction) {
-	i.next = next
-}
-
-func setPrev(i, prev *instruction) {
-	i.prev = prev
-}
-
 // IsCall implements regalloc.Instr IsCall.
 func (i *instruction) IsCall() bool {
 	return i.kind == call
@@ -63,21 +51,6 @@ func (i *instruction) IsReturn() bool {
 	return i.kind == ret
 }
 
-// Next implements regalloc.Instr Next.
-func (i *instruction) Next() regalloc.Instr {
-	return i.next
-}
-
-// Prev implements regalloc.Instr Prev.
-func (i *instruction) Prev() regalloc.Instr {
-	return i.prev
-}
-
-// AddedBeforeRegAlloc implements regalloc.Instr AddedBeforeRegAlloc.
-func (i *instruction) AddedBeforeRegAlloc() bool {
-	return i.addedBeforeRegAlloc
-}
-
 type defKind byte
 
 const (
diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/instr_encoding.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/instr_encoding.go
index f0ede2d6a..21be9b71e 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/instr_encoding.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/instr_encoding.go
@@ -12,7 +12,7 @@ import (
 // Encode implements backend.Machine Encode.
 func (m *machine) Encode(ctx context.Context) error {
 	m.resolveRelativeAddresses(ctx)
-	m.encode(m.executableContext.RootInstr)
+	m.encode(m.rootInstr)
 	if l := len(m.compiler.Buf()); l > maxFunctionExecutableSize {
 		return fmt.Errorf("function size exceeds the limit: %d > %d", l, maxFunctionExecutableSize)
 	}
diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/lower_instr.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/lower_instr.go
index 048bf3204..f9df356c0 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/lower_instr.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/lower_instr.go
@@ -17,19 +17,18 @@ import (
 
 // LowerSingleBranch implements backend.Machine.
 func (m *machine) LowerSingleBranch(br *ssa.Instruction) {
-	ectx := m.executableContext
 	switch br.Opcode() {
 	case ssa.OpcodeJump:
-		_, _, targetBlk := br.BranchData()
+		_, _, targetBlkID := br.BranchData()
 		if br.IsFallthroughJump() {
 			return
 		}
 		b := m.allocateInstr()
-		target := ectx.GetOrAllocateSSABlockLabel(targetBlk)
-		if target == labelReturn {
+		targetBlk := m.compiler.SSABuilder().BasicBlock(targetBlkID)
+		if targetBlk.ReturnBlock() {
 			b.asRet()
 		} else {
-			b.asBr(target)
+			b.asBr(ssaBlockLabel(targetBlk))
 		}
 		m.insert(b)
 	case ssa.OpcodeBrTable:
@@ -40,7 +39,8 @@ func (m *machine) LowerSingleBranch(br *ssa.Instruction) {
 }
 
 func (m *machine) lowerBrTable(i *ssa.Instruction) {
-	index, targets := i.BrTableData()
+	index, targetBlockIDs := i.BrTableData()
+	targetBlockCount := len(targetBlockIDs.View())
 	indexOperand := m.getOperand_NR(m.compiler.ValueDefinition(index), extModeNone)
 
 	// Firstly, we have to do the bounds check of the index, and
@@ -50,7 +50,7 @@ func (m *machine) lowerBrTable(i *ssa.Instruction) {
 	// subs wzr, index, maxIndexReg
 	// csel adjustedIndex, maxIndexReg, index, hs ;; if index is higher or equal than maxIndexReg.
 	maxIndexReg := m.compiler.AllocateVReg(ssa.TypeI32)
-	m.lowerConstantI32(maxIndexReg, int32(len(targets)-1))
+	m.lowerConstantI32(maxIndexReg, int32(targetBlockCount-1))
 	subs := m.allocateInstr()
 	subs.asALU(aluOpSubS, xzrVReg, indexOperand, operandNR(maxIndexReg), false)
 	m.insert(subs)
@@ -61,24 +61,24 @@ func (m *machine) lowerBrTable(i *ssa.Instruction) {
 
 	brSequence := m.allocateInstr()
 
-	tableIndex := m.addJmpTableTarget(targets)
-	brSequence.asBrTableSequence(adjustedIndex, tableIndex, len(targets))
+	tableIndex := m.addJmpTableTarget(targetBlockIDs)
+	brSequence.asBrTableSequence(adjustedIndex, tableIndex, targetBlockCount)
 	m.insert(brSequence)
 }
 
 // LowerConditionalBranch implements backend.Machine.
 func (m *machine) LowerConditionalBranch(b *ssa.Instruction) {
-	exctx := m.executableContext
-	cval, args, targetBlk := b.BranchData()
+	cval, args, targetBlkID := b.BranchData()
 	if len(args) > 0 {
 		panic(fmt.Sprintf(
 			"conditional branch shouldn't have args; likely a bug in critical edge splitting: from %s to %s",
-			exctx.CurrentSSABlk,
-			targetBlk,
+			m.currentLabelPos.sb,
+			targetBlkID,
 		))
 	}
 
-	target := exctx.GetOrAllocateSSABlockLabel(targetBlk)
+	targetBlk := m.compiler.SSABuilder().BasicBlock(targetBlkID)
+	target := ssaBlockLabel(targetBlk)
 	cvalDef := m.compiler.ValueDefinition(cval)
 
 	switch {
@@ -791,7 +791,7 @@ func (m *machine) LowerInstr(instr *ssa.Instruction) {
 	default:
 		panic("TODO: lowering " + op.String())
 	}
-	m.executableContext.FlushPendingInstructions()
+	m.FlushPendingInstructions()
 }
 
 func (m *machine) lowerShuffle(rd regalloc.VReg, rn, rm operand, lane1, lane2 uint64) {
diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/lower_instr_operands.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/lower_instr_operands.go
index d9fbf1789..7a398c3d0 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/lower_instr_operands.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/lower_instr_operands.go
@@ -162,9 +162,9 @@ func (o operand) assignReg(v regalloc.VReg) operand {
 //
 // `mode` is used to extend the operand if the bit length is smaller than mode.bits().
 // If the operand can be expressed as operandKindImm12, `mode` is ignored.
-func (m *machine) getOperand_Imm12_ER_SR_NR(def *backend.SSAValueDefinition, mode extMode) (op operand) {
-	if def.IsFromBlockParam() {
-		return operandNR(def.BlkParamVReg)
+func (m *machine) getOperand_Imm12_ER_SR_NR(def backend.SSAValueDefinition, mode extMode) (op operand) {
+	if !def.IsFromInstr() {
+		return operandNR(m.compiler.VRegOf(def.V))
 	}
 
 	instr := def.Instr
@@ -179,9 +179,9 @@ func (m *machine) getOperand_Imm12_ER_SR_NR(def *backend.SSAValueDefinition, mod
 
 // getOperand_MaybeNegatedImm12_ER_SR_NR is almost the same as getOperand_Imm12_ER_SR_NR, but this might negate the immediate value.
 // If the immediate value is negated, the second return value is true, otherwise always false.
-func (m *machine) getOperand_MaybeNegatedImm12_ER_SR_NR(def *backend.SSAValueDefinition, mode extMode) (op operand, negatedImm12 bool) {
-	if def.IsFromBlockParam() {
-		return operandNR(def.BlkParamVReg), false
+func (m *machine) getOperand_MaybeNegatedImm12_ER_SR_NR(def backend.SSAValueDefinition, mode extMode) (op operand, negatedImm12 bool) {
+	if !def.IsFromInstr() {
+		return operandNR(m.compiler.VRegOf(def.V)), false
 	}
 
 	instr := def.Instr
@@ -193,7 +193,7 @@ func (m *machine) getOperand_MaybeNegatedImm12_ER_SR_NR(def *backend.SSAValueDef
 		}
 
 		signExtended := int64(c)
-		if def.SSAValue().Type().Bits() == 32 {
+		if def.V.Type().Bits() == 32 {
 			signExtended = (signExtended << 32) >> 32
 		}
 		negatedWithoutSign := -signExtended
@@ -208,9 +208,9 @@ func (m *machine) getOperand_MaybeNegatedImm12_ER_SR_NR(def *backend.SSAValueDef
 // ensureValueNR returns an operand of either operandKindER, operandKindSR, or operandKindNR from the given value (defined by `def).
 //
 // `mode` is used to extend the operand if the bit length is smaller than mode.bits().
-func (m *machine) getOperand_ER_SR_NR(def *backend.SSAValueDefinition, mode extMode) (op operand) {
-	if def.IsFromBlockParam() {
-		return operandNR(def.BlkParamVReg)
+func (m *machine) getOperand_ER_SR_NR(def backend.SSAValueDefinition, mode extMode) (op operand) {
+	if !def.IsFromInstr() {
+		return operandNR(m.compiler.VRegOf(def.V))
 	}
 
 	if m.compiler.MatchInstr(def, ssa.OpcodeSExtend) || m.compiler.MatchInstr(def, ssa.OpcodeUExtend) {
@@ -251,9 +251,9 @@ func (m *machine) getOperand_ER_SR_NR(def *backend.SSAValueDefinition, mode extM
 // ensureValueNR returns an operand of either operandKindSR or operandKindNR from the given value (defined by `def).
 //
 // `mode` is used to extend the operand if the bit length is smaller than mode.bits().
-func (m *machine) getOperand_SR_NR(def *backend.SSAValueDefinition, mode extMode) (op operand) {
-	if def.IsFromBlockParam() {
-		return operandNR(def.BlkParamVReg)
+func (m *machine) getOperand_SR_NR(def backend.SSAValueDefinition, mode extMode) (op operand) {
+	if !def.IsFromInstr() {
+		return operandNR(m.compiler.VRegOf(def.V))
 	}
 
 	if m.compiler.MatchInstr(def, ssa.OpcodeIshl) {
@@ -273,9 +273,9 @@ func (m *machine) getOperand_SR_NR(def *backend.SSAValueDefinition, mode extMode
 }
 
 // getOperand_ShiftImm_NR returns an operand of either operandKindShiftImm or operandKindNR from the given value (defined by `def).
-func (m *machine) getOperand_ShiftImm_NR(def *backend.SSAValueDefinition, mode extMode, shiftBitWidth byte) (op operand) {
-	if def.IsFromBlockParam() {
-		return operandNR(def.BlkParamVReg)
+func (m *machine) getOperand_ShiftImm_NR(def backend.SSAValueDefinition, mode extMode, shiftBitWidth byte) (op operand) {
+	if !def.IsFromInstr() {
+		return operandNR(m.compiler.VRegOf(def.V))
 	}
 
 	instr := def.Instr
@@ -289,28 +289,18 @@ func (m *machine) getOperand_ShiftImm_NR(def *backend.SSAValueDefinition, mode e
 // ensureValueNR returns an operand of operandKindNR from the given value (defined by `def).
 //
 // `mode` is used to extend the operand if the bit length is smaller than mode.bits().
-func (m *machine) getOperand_NR(def *backend.SSAValueDefinition, mode extMode) (op operand) {
+func (m *machine) getOperand_NR(def backend.SSAValueDefinition, mode extMode) (op operand) {
 	var v regalloc.VReg
-	if def.IsFromBlockParam() {
-		v = def.BlkParamVReg
+	if def.IsFromInstr() && def.Instr.Constant() {
+		// We inline all the constant instructions so that we could reduce the register usage.
+		v = m.lowerConstant(def.Instr)
+		def.Instr.MarkLowered()
 	} else {
-		instr := def.Instr
-		if instr.Constant() {
-			// We inline all the constant instructions so that we could reduce the register usage.
-			v = m.lowerConstant(instr)
-			instr.MarkLowered()
-		} else {
-			if n := def.N; n == 0 {
-				v = m.compiler.VRegOf(instr.Return())
-			} else {
-				_, rs := instr.Returns()
-				v = m.compiler.VRegOf(rs[n-1])
-			}
-		}
+		v = m.compiler.VRegOf(def.V)
 	}
 
 	r := v
-	switch inBits := def.SSAValue().Type().Bits(); {
+	switch inBits := def.V.Type().Bits(); {
 	case mode == extModeNone:
 	case inBits == 32 && (mode == extModeZeroExtend32 || mode == extModeSignExtend32):
 	case inBits == 32 && mode == extModeZeroExtend64:
diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine.go
index 5f584f928..00e6b238f 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine.go
@@ -3,6 +3,7 @@ package arm64
 import (
 	"context"
 	"fmt"
+	"math"
 	"strings"
 
 	"github.com/tetratelabs/wazero/internal/engine/wazevo/backend"
@@ -14,12 +15,33 @@ import (
 type (
 	// machine implements backend.Machine.
 	machine struct {
-		compiler          backend.Compiler
-		executableContext *backend.ExecutableContextT[instruction]
-		currentABI        *backend.FunctionABI
+		compiler   backend.Compiler
+		currentABI *backend.FunctionABI
+		instrPool  wazevoapi.Pool[instruction]
+		// labelPositionPool is the pool of labelPosition. The id is the label where
+		// if the label is less than the maxSSABlockID, it's the ssa.BasicBlockID.
+		labelPositionPool wazevoapi.IDedPool[labelPosition]
 
-		regAlloc   regalloc.Allocator
-		regAllocFn *backend.RegAllocFunction[*instruction, *machine]
+		// nextLabel is the next label to be allocated. The first free label comes after maxSSABlockID
+		// so that we can have an identical label for the SSA block ID, which is useful for debugging.
+		nextLabel label
+		// rootInstr is the first instruction of the function.
+		rootInstr *instruction
+		// currentLabelPos is the currently-compiled ssa.BasicBlock's labelPosition.
+		currentLabelPos *labelPosition
+		// orderedSSABlockLabelPos is the ordered list of labelPosition in the generated code for each ssa.BasicBlock.
+		orderedSSABlockLabelPos []*labelPosition
+		// returnLabelPos is the labelPosition for the return block.
+		returnLabelPos labelPosition
+		// perBlockHead and perBlockEnd are the head and tail of the instruction list per currently-compiled ssa.BasicBlock.
+		perBlockHead, perBlockEnd *instruction
+		// pendingInstructions are the instructions which are not yet emitted into the instruction list.
+		pendingInstructions []*instruction
+		// maxSSABlockID is the maximum ssa.BasicBlockID in the current function.
+		maxSSABlockID label
+
+		regAlloc   regalloc.Allocator[*instruction, *labelPosition, *regAllocFn]
+		regAllocFn regAllocFn
 
 		amodePool wazevoapi.Pool[addressMode]
 
@@ -35,6 +57,8 @@ type (
 
 		// jmpTableTargets holds the labels of the jump table targets.
 		jmpTableTargets [][]uint32
+		// jmpTableTargetNext is the index to the jmpTableTargets slice to be used for the next jump table.
+		jmpTableTargetsNext int
 
 		// spillSlotSize is the size of the stack slot in bytes used for spilling registers.
 		// During the execution of the function, the stack looks like:
@@ -91,45 +115,132 @@ type (
 		nextLabel label
 		offset    int64
 	}
+)
+
+type (
+	// label represents a position in the generated code which is either
+	// a real instruction or the constant InstructionPool (e.g. jump tables).
+	//
+	// This is exactly the same as the traditional "label" in assembly code.
+	label uint32
 
-	labelPosition = backend.LabelPosition[instruction]
-	label         = backend.Label
+	// labelPosition represents the regions of the generated code which the label represents.
+	// This implements regalloc.Block.
+	labelPosition struct {
+		// sb is not nil if this corresponds to a ssa.BasicBlock.
+		sb ssa.BasicBlock
+		// cur is used to walk through the instructions in the block during the register allocation.
+		cur,
+		// begin and end are the first and last instructions of the block.
+		begin, end *instruction
+		// binaryOffset is the offset in the binary where the label is located.
+		binaryOffset int64
+	}
 )
 
 const (
-	labelReturn  = backend.LabelReturn
-	labelInvalid = backend.LabelInvalid
+	labelReturn  label = math.MaxUint32
+	labelInvalid       = labelReturn - 1
 )
 
+// String implements backend.Machine.
+func (l label) String() string {
+	return fmt.Sprintf("L%d", l)
+}
+
+func resetLabelPosition(l *labelPosition) {
+	*l = labelPosition{}
+}
+
 // NewBackend returns a new backend for arm64.
 func NewBackend() backend.Machine {
 	m := &machine{
 		spillSlots:        make(map[regalloc.VRegID]int64),
-		executableContext: newExecutableContext(),
-		regAlloc:          regalloc.NewAllocator(regInfo),
+		regAlloc:          regalloc.NewAllocator[*instruction, *labelPosition, *regAllocFn](regInfo),
 		amodePool:         wazevoapi.NewPool[addressMode](resetAddressMode),
+		instrPool:         wazevoapi.NewPool[instruction](resetInstruction),
+		labelPositionPool: wazevoapi.NewIDedPool[labelPosition](resetLabelPosition),
 	}
+	m.regAllocFn.m = m
 	return m
 }
 
-func newExecutableContext() *backend.ExecutableContextT[instruction] {
-	return backend.NewExecutableContextT[instruction](resetInstruction, setNext, setPrev, asNop0)
+func ssaBlockLabel(sb ssa.BasicBlock) label {
+	if sb.ReturnBlock() {
+		return labelReturn
+	}
+	return label(sb.ID())
 }
 
-// ExecutableContext implements backend.Machine.
-func (m *machine) ExecutableContext() backend.ExecutableContext {
-	return m.executableContext
+// getOrAllocateSSABlockLabelPosition returns the labelPosition for the given basic block.
+func (m *machine) getOrAllocateSSABlockLabelPosition(sb ssa.BasicBlock) *labelPosition {
+	if sb.ReturnBlock() {
+		m.returnLabelPos.sb = sb
+		return &m.returnLabelPos
+	}
+
+	l := ssaBlockLabel(sb)
+	pos := m.labelPositionPool.GetOrAllocate(int(l))
+	pos.sb = sb
+	return pos
 }
 
-// RegAlloc implements backend.Machine Function.
-func (m *machine) RegAlloc() {
-	rf := m.regAllocFn
-	for _, pos := range m.executableContext.OrderedBlockLabels {
-		rf.AddBlock(pos.SB, pos.L, pos.Begin, pos.End)
+// LinkAdjacentBlocks implements backend.Machine.
+func (m *machine) LinkAdjacentBlocks(prev, next ssa.BasicBlock) {
+	prevPos, nextPos := m.getOrAllocateSSABlockLabelPosition(prev), m.getOrAllocateSSABlockLabelPosition(next)
+	prevPos.end.next = nextPos.begin
+}
+
+// StartBlock implements backend.Machine.
+func (m *machine) StartBlock(blk ssa.BasicBlock) {
+	m.currentLabelPos = m.getOrAllocateSSABlockLabelPosition(blk)
+	labelPos := m.currentLabelPos
+	end := m.allocateNop()
+	m.perBlockHead, m.perBlockEnd = end, end
+	labelPos.begin, labelPos.end = end, end
+	m.orderedSSABlockLabelPos = append(m.orderedSSABlockLabelPos, labelPos)
+}
+
+// EndBlock implements ExecutableContext.
+func (m *machine) EndBlock() {
+	// Insert nop0 as the head of the block for convenience to simplify the logic of inserting instructions.
+	m.insertAtPerBlockHead(m.allocateNop())
+
+	m.currentLabelPos.begin = m.perBlockHead
+
+	if m.currentLabelPos.sb.EntryBlock() {
+		m.rootInstr = m.perBlockHead
+	}
+}
+
+func (m *machine) insertAtPerBlockHead(i *instruction) {
+	if m.perBlockHead == nil {
+		m.perBlockHead = i
+		m.perBlockEnd = i
+		return
 	}
 
+	i.next = m.perBlockHead
+	m.perBlockHead.prev = i
+	m.perBlockHead = i
+}
+
+// FlushPendingInstructions implements backend.Machine.
+func (m *machine) FlushPendingInstructions() {
+	l := len(m.pendingInstructions)
+	if l == 0 {
+		return
+	}
+	for i := l - 1; i >= 0; i-- { // reverse because we lower instructions in reverse order.
+		m.insertAtPerBlockHead(m.pendingInstructions[i])
+	}
+	m.pendingInstructions = m.pendingInstructions[:0]
+}
+
+// RegAlloc implements backend.Machine Function.
+func (m *machine) RegAlloc() {
 	m.regAllocStarted = true
-	m.regAlloc.DoAllocation(rf)
+	m.regAlloc.DoAllocation(&m.regAllocFn)
 	// Now that we know the final spill slot size, we must align spillSlotSize to 16 bytes.
 	m.spillSlotSize = (m.spillSlotSize + 15) &^ 15
 }
@@ -146,13 +257,22 @@ func (m *machine) Reset() {
 	m.clobberedRegs = m.clobberedRegs[:0]
 	m.regAllocStarted = false
 	m.regAlloc.Reset()
-	m.regAllocFn.Reset()
 	m.spillSlotSize = 0
 	m.unresolvedAddressModes = m.unresolvedAddressModes[:0]
 	m.maxRequiredStackSizeForCalls = 0
-	m.executableContext.Reset()
-	m.jmpTableTargets = m.jmpTableTargets[:0]
+	m.jmpTableTargetsNext = 0
 	m.amodePool.Reset()
+	m.instrPool.Reset()
+	m.labelPositionPool.Reset()
+	m.pendingInstructions = m.pendingInstructions[:0]
+	m.perBlockHead, m.perBlockEnd, m.rootInstr = nil, nil, nil
+	m.orderedSSABlockLabelPos = m.orderedSSABlockLabelPos[:0]
+}
+
+// StartLoweringFunction implements backend.Machine StartLoweringFunction.
+func (m *machine) StartLoweringFunction(maxBlockID ssa.BasicBlockID) {
+	m.maxSSABlockID = label(maxBlockID)
+	m.nextLabel = label(maxBlockID) + 1
 }
 
 // SetCurrentABI implements backend.Machine SetCurrentABI.
@@ -168,12 +288,11 @@ func (m *machine) DisableStackCheck() {
 // SetCompiler implements backend.Machine.
 func (m *machine) SetCompiler(ctx backend.Compiler) {
 	m.compiler = ctx
-	m.regAllocFn = backend.NewRegAllocFunction[*instruction, *machine](m, ctx.SSABuilder(), ctx)
+	m.regAllocFn.ssaB = ctx.SSABuilder()
 }
 
 func (m *machine) insert(i *instruction) {
-	ectx := m.executableContext
-	ectx.PendingInstructions = append(ectx.PendingInstructions, i)
+	m.pendingInstructions = append(m.pendingInstructions, i)
 }
 
 func (m *machine) insertBrTargetLabel() label {
@@ -183,18 +302,18 @@ func (m *machine) insertBrTargetLabel() label {
 }
 
 func (m *machine) allocateBrTarget() (nop *instruction, l label) {
-	ectx := m.executableContext
-	l = ectx.AllocateLabel()
+	l = m.nextLabel
+	m.nextLabel++
 	nop = m.allocateInstr()
 	nop.asNop0WithLabel(l)
-	pos := ectx.GetOrAllocateLabelPosition(l)
-	pos.Begin, pos.End = nop, nop
+	pos := m.labelPositionPool.GetOrAllocate(int(l))
+	pos.begin, pos.end = nop, nop
 	return
 }
 
 // allocateInstr allocates an instruction.
 func (m *machine) allocateInstr() *instruction {
-	instr := m.executableContext.InstructionPool.Allocate()
+	instr := m.instrPool.Allocate()
 	if !m.regAllocStarted {
 		instr.addedBeforeRegAlloc = true
 	}
@@ -251,7 +370,6 @@ func (m *machine) resolveAddressingMode(arg0offset, ret0offset int64, i *instruc
 
 // resolveRelativeAddresses resolves the relative addresses before encoding.
 func (m *machine) resolveRelativeAddresses(ctx context.Context) {
-	ectx := m.executableContext
 	for {
 		if len(m.unresolvedAddressModes) > 0 {
 			arg0offset, ret0offset := m.arg0OffsetFromSP(), m.ret0OffsetFromSP()
@@ -265,35 +383,36 @@ func (m *machine) resolveRelativeAddresses(ctx context.Context) {
 
 		var fn string
 		var fnIndex int
-		var labelToSSABlockID map[label]ssa.BasicBlockID
+		var labelPosToLabel map[*labelPosition]label
 		if wazevoapi.PerfMapEnabled {
-			fn = wazevoapi.GetCurrentFunctionName(ctx)
-			labelToSSABlockID = make(map[label]ssa.BasicBlockID)
-			for i, l := range ectx.SsaBlockIDToLabels {
-				labelToSSABlockID[l] = ssa.BasicBlockID(i)
+			labelPosToLabel = make(map[*labelPosition]label)
+			for i := 0; i <= m.labelPositionPool.MaxIDEncountered(); i++ {
+				labelPosToLabel[m.labelPositionPool.Get(i)] = label(i)
 			}
+
+			fn = wazevoapi.GetCurrentFunctionName(ctx)
 			fnIndex = wazevoapi.GetCurrentFunctionIndex(ctx)
 		}
 
 		// Next, in order to determine the offsets of relative jumps, we have to calculate the size of each label.
 		var offset int64
-		for i, pos := range ectx.OrderedBlockLabels {
-			pos.BinaryOffset = offset
+		for i, pos := range m.orderedSSABlockLabelPos {
+			pos.binaryOffset = offset
 			var size int64
-			for cur := pos.Begin; ; cur = cur.next {
+			for cur := pos.begin; ; cur = cur.next {
 				switch cur.kind {
 				case nop0:
 					l := cur.nop0Label()
-					if pos := ectx.LabelPositions[l]; pos != nil {
-						pos.BinaryOffset = offset + size
+					if pos := m.labelPositionPool.Get(int(l)); pos != nil {
+						pos.binaryOffset = offset + size
 					}
 				case condBr:
 					if !cur.condBrOffsetResolved() {
 						var nextLabel label
-						if i < len(ectx.OrderedBlockLabels)-1 {
+						if i < len(m.orderedSSABlockLabelPos)-1 {
 							// Note: this is only used when the block ends with fallthrough,
 							// therefore can be safely assumed that the next block exists when it's needed.
-							nextLabel = ectx.OrderedBlockLabels[i+1].L
+							nextLabel = ssaBlockLabel(m.orderedSSABlockLabelPos[i+1].sb)
 						}
 						m.condBrRelocs = append(m.condBrRelocs, condBrReloc{
 							cbr: cur, currentLabelPos: pos, offset: offset + size,
@@ -302,21 +421,14 @@ func (m *machine) resolveRelativeAddresses(ctx context.Context) {
 					}
 				}
 				size += cur.size()
-				if cur == pos.End {
+				if cur == pos.end {
 					break
 				}
 			}
 
 			if wazevoapi.PerfMapEnabled {
 				if size > 0 {
-					l := pos.L
-					var labelStr string
-					if blkID, ok := labelToSSABlockID[l]; ok {
-						labelStr = fmt.Sprintf("%s::SSA_Block[%s]", l, blkID)
-					} else {
-						labelStr = l.String()
-					}
-					wazevoapi.PerfMap.AddModuleEntry(fnIndex, offset, uint64(size), fmt.Sprintf("%s:::::%s", fn, labelStr))
+					wazevoapi.PerfMap.AddModuleEntry(fnIndex, offset, uint64(size), fmt.Sprintf("%s:::::%s", fn, labelPosToLabel[pos]))
 				}
 			}
 			offset += size
@@ -330,7 +442,7 @@ func (m *machine) resolveRelativeAddresses(ctx context.Context) {
 			offset := reloc.offset
 
 			target := cbr.condBrLabel()
-			offsetOfTarget := ectx.LabelPositions[target].BinaryOffset
+			offsetOfTarget := m.labelPositionPool.Get(int(target)).binaryOffset
 			diff := offsetOfTarget - offset
 			if divided := diff >> 2; divided < minSignedInt19 || divided > maxSignedInt19 {
 				// This case the conditional branch is too huge. We place the trampoline instructions at the end of the current block,
@@ -351,11 +463,11 @@ func (m *machine) resolveRelativeAddresses(ctx context.Context) {
 	}
 
 	var currentOffset int64
-	for cur := ectx.RootInstr; cur != nil; cur = cur.next {
+	for cur := m.rootInstr; cur != nil; cur = cur.next {
 		switch cur.kind {
 		case br:
 			target := cur.brLabel()
-			offsetOfTarget := ectx.LabelPositions[target].BinaryOffset
+			offsetOfTarget := m.labelPositionPool.Get(int(target)).binaryOffset
 			diff := offsetOfTarget - currentOffset
 			divided := diff >> 2
 			if divided < minSignedInt26 || divided > maxSignedInt26 {
@@ -366,7 +478,7 @@ func (m *machine) resolveRelativeAddresses(ctx context.Context) {
 		case condBr:
 			if !cur.condBrOffsetResolved() {
 				target := cur.condBrLabel()
-				offsetOfTarget := ectx.LabelPositions[target].BinaryOffset
+				offsetOfTarget := m.labelPositionPool.Get(int(target)).binaryOffset
 				diff := offsetOfTarget - currentOffset
 				if divided := diff >> 2; divided < minSignedInt19 || divided > maxSignedInt19 {
 					panic("BUG: branch relocation for large conditional branch larger than 19-bit range must be handled properly")
@@ -378,7 +490,7 @@ func (m *machine) resolveRelativeAddresses(ctx context.Context) {
 			targets := m.jmpTableTargets[tableIndex]
 			for i := range targets {
 				l := label(targets[i])
-				offsetOfTarget := ectx.LabelPositions[l].BinaryOffset
+				offsetOfTarget := m.labelPositionPool.Get(int(l)).binaryOffset
 				diff := offsetOfTarget - (currentOffset + brTableSequenceOffsetTableBegin)
 				targets[i] = uint32(diff)
 			}
@@ -399,7 +511,7 @@ const (
 )
 
 func (m *machine) insertConditionalJumpTrampoline(cbr *instruction, currentBlk *labelPosition, nextLabel label) {
-	cur := currentBlk.End
+	cur := currentBlk.end
 	originalTarget := cbr.condBrLabel()
 	endNext := cur.next
 
@@ -422,32 +534,27 @@ func (m *machine) insertConditionalJumpTrampoline(cbr *instruction, currentBlk *
 	cur = linkInstr(cur, br)
 
 	// Update the end of the current block.
-	currentBlk.End = cur
+	currentBlk.end = cur
 
 	linkInstr(cur, endNext)
 }
 
 // Format implements backend.Machine.
 func (m *machine) Format() string {
-	ectx := m.executableContext
 	begins := map[*instruction]label{}
-	for _, pos := range ectx.LabelPositions {
+	for l := label(0); l < m.nextLabel; l++ {
+		pos := m.labelPositionPool.Get(int(l))
 		if pos != nil {
-			begins[pos.Begin] = pos.L
+			begins[pos.begin] = l
 		}
 	}
 
-	irBlocks := map[label]ssa.BasicBlockID{}
-	for i, l := range ectx.SsaBlockIDToLabels {
-		irBlocks[l] = ssa.BasicBlockID(i)
-	}
-
 	var lines []string
-	for cur := ectx.RootInstr; cur != nil; cur = cur.next {
+	for cur := m.rootInstr; cur != nil; cur = cur.next {
 		if l, ok := begins[cur]; ok {
 			var labelStr string
-			if blkID, ok := irBlocks[l]; ok {
-				labelStr = fmt.Sprintf("%s (SSA Block: %s):", l, blkID)
+			if l <= m.maxSSABlockID {
+				labelStr = fmt.Sprintf("%s (SSA Block: blk%d):", l, int(l))
 			} else {
 				labelStr = fmt.Sprintf("%s:", l)
 			}
@@ -508,13 +615,17 @@ func (m *machine) frameSize() int64 {
 	return s
 }
 
-func (m *machine) addJmpTableTarget(targets []ssa.BasicBlock) (index int) {
-	// TODO: reuse the slice!
-	labels := make([]uint32, len(targets))
-	for j, target := range targets {
-		labels[j] = uint32(m.executableContext.GetOrAllocateSSABlockLabel(target))
+func (m *machine) addJmpTableTarget(targets ssa.Values) (index int) {
+	if m.jmpTableTargetsNext == len(m.jmpTableTargets) {
+		m.jmpTableTargets = append(m.jmpTableTargets, make([]uint32, 0, len(targets.View())))
+	}
+
+	index = m.jmpTableTargetsNext
+	m.jmpTableTargetsNext++
+	m.jmpTableTargets[index] = m.jmpTableTargets[index][:0]
+	for _, targetBlockID := range targets.View() {
+		target := m.compiler.SSABuilder().BasicBlock(ssa.BasicBlockID(targetBlockID))
+		m.jmpTableTargets[index] = append(m.jmpTableTargets[index], uint32(target.ID()))
 	}
-	index = len(m.jmpTableTargets)
-	m.jmpTableTargets = append(m.jmpTableTargets, labels)
 	return
 }
diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine_pro_epi_logue.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine_pro_epi_logue.go
index d9032f921..c646a8fab 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine_pro_epi_logue.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine_pro_epi_logue.go
@@ -15,9 +15,7 @@ func (m *machine) PostRegAlloc() {
 
 // setupPrologue initializes the prologue of the function.
 func (m *machine) setupPrologue() {
-	ectx := m.executableContext
-
-	cur := ectx.RootInstr
+	cur := m.rootInstr
 	prevInitInst := cur.next
 
 	//
@@ -196,21 +194,20 @@ func (m *machine) createFrameSizeSlot(cur *instruction, s int64) *instruction {
 // 1. Removes the redundant copy instruction.
 // 2. Inserts the epilogue.
 func (m *machine) postRegAlloc() {
-	ectx := m.executableContext
-	for cur := ectx.RootInstr; cur != nil; cur = cur.next {
+	for cur := m.rootInstr; cur != nil; cur = cur.next {
 		switch cur.kind {
 		case ret:
 			m.setupEpilogueAfter(cur.prev)
 		case loadConstBlockArg:
 			lc := cur
 			next := lc.next
-			m.executableContext.PendingInstructions = m.executableContext.PendingInstructions[:0]
+			m.pendingInstructions = m.pendingInstructions[:0]
 			m.lowerLoadConstantBlockArgAfterRegAlloc(lc)
-			for _, instr := range m.executableContext.PendingInstructions {
+			for _, instr := range m.pendingInstructions {
 				cur = linkInstr(cur, instr)
 			}
 			linkInstr(cur, next)
-			m.executableContext.PendingInstructions = m.executableContext.PendingInstructions[:0]
+			m.pendingInstructions = m.pendingInstructions[:0]
 		default:
 			// Removes the redundant copy instruction.
 			if cur.IsCopy() && cur.rn.realReg() == cur.rd.RealReg() {
@@ -432,11 +429,9 @@ func (m *machine) insertStackBoundsCheck(requiredStackSize int64, cur *instructi
 
 // CompileStackGrowCallSequence implements backend.Machine.
 func (m *machine) CompileStackGrowCallSequence() []byte {
-	ectx := m.executableContext
-
 	cur := m.allocateInstr()
 	cur.asNop0()
-	ectx.RootInstr = cur
+	m.rootInstr = cur
 
 	// Save the callee saved and argument registers.
 	cur = m.saveRegistersInExecutionContext(cur, saveRequiredRegs)
@@ -458,16 +453,14 @@ func (m *machine) CompileStackGrowCallSequence() []byte {
 	ret.asRet()
 	linkInstr(cur, ret)
 
-	m.encode(ectx.RootInstr)
+	m.encode(m.rootInstr)
 	return m.compiler.Buf()
 }
 
 func (m *machine) addsAddOrSubStackPointer(cur *instruction, rd regalloc.VReg, diff int64, add bool) *instruction {
-	ectx := m.executableContext
-
-	ectx.PendingInstructions = ectx.PendingInstructions[:0]
+	m.pendingInstructions = m.pendingInstructions[:0]
 	m.insertAddOrSubStackPointer(rd, diff, add)
-	for _, inserted := range ectx.PendingInstructions {
+	for _, inserted := range m.pendingInstructions {
 		cur = linkInstr(cur, inserted)
 	}
 	return cur
diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine_regalloc.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine_regalloc.go
index c7eb92cc2..f2ed53ae5 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine_regalloc.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/machine_regalloc.go
@@ -3,18 +3,226 @@ package arm64
 // This file implements the interfaces required for register allocations. See backend.RegAllocFunctionMachine.
 
 import (
-	"github.com/tetratelabs/wazero/internal/engine/wazevo/backend"
 	"github.com/tetratelabs/wazero/internal/engine/wazevo/backend/regalloc"
 	"github.com/tetratelabs/wazero/internal/engine/wazevo/ssa"
 )
 
-// ClobberedRegisters implements backend.RegAllocFunctionMachine.
-func (m *machine) ClobberedRegisters(regs []regalloc.VReg) {
-	m.clobberedRegs = append(m.clobberedRegs[:0], regs...)
+// regAllocFn implements regalloc.Function.
+type regAllocFn struct {
+	ssaB                   ssa.Builder
+	m                      *machine
+	loopNestingForestRoots []ssa.BasicBlock
+	blockIter              int
 }
 
-// Swap implements backend.RegAllocFunctionMachine.
-func (m *machine) Swap(cur *instruction, x1, x2, tmp regalloc.VReg) {
+// PostOrderBlockIteratorBegin implements regalloc.Function.
+func (f *regAllocFn) PostOrderBlockIteratorBegin() *labelPosition {
+	f.blockIter = len(f.m.orderedSSABlockLabelPos) - 1
+	return f.PostOrderBlockIteratorNext()
+}
+
+// PostOrderBlockIteratorNext implements regalloc.Function.
+func (f *regAllocFn) PostOrderBlockIteratorNext() *labelPosition {
+	if f.blockIter < 0 {
+		return nil
+	}
+	b := f.m.orderedSSABlockLabelPos[f.blockIter]
+	f.blockIter--
+	return b
+}
+
+// ReversePostOrderBlockIteratorBegin implements regalloc.Function.
+func (f *regAllocFn) ReversePostOrderBlockIteratorBegin() *labelPosition {
+	f.blockIter = 0
+	return f.ReversePostOrderBlockIteratorNext()
+}
+
+// ReversePostOrderBlockIteratorNext implements regalloc.Function.
+func (f *regAllocFn) ReversePostOrderBlockIteratorNext() *labelPosition {
+	if f.blockIter >= len(f.m.orderedSSABlockLabelPos) {
+		return nil
+	}
+	b := f.m.orderedSSABlockLabelPos[f.blockIter]
+	f.blockIter++
+	return b
+}
+
+// ClobberedRegisters implements regalloc.Function.
+func (f *regAllocFn) ClobberedRegisters(regs []regalloc.VReg) {
+	f.m.clobberedRegs = append(f.m.clobberedRegs[:0], regs...)
+}
+
+// LoopNestingForestRoots implements regalloc.Function.
+func (f *regAllocFn) LoopNestingForestRoots() int {
+	f.loopNestingForestRoots = f.ssaB.LoopNestingForestRoots()
+	return len(f.loopNestingForestRoots)
+}
+
+// LoopNestingForestRoot implements regalloc.Function.
+func (f *regAllocFn) LoopNestingForestRoot(i int) *labelPosition {
+	root := f.loopNestingForestRoots[i]
+	pos := f.m.getOrAllocateSSABlockLabelPosition(root)
+	return pos
+}
+
+// LowestCommonAncestor implements regalloc.Function.
+func (f *regAllocFn) LowestCommonAncestor(blk1, blk2 *labelPosition) *labelPosition {
+	sb := f.ssaB.LowestCommonAncestor(blk1.sb, blk2.sb)
+	pos := f.m.getOrAllocateSSABlockLabelPosition(sb)
+	return pos
+}
+
+// Idom implements regalloc.Function.
+func (f *regAllocFn) Idom(blk *labelPosition) *labelPosition {
+	sb := f.ssaB.Idom(blk.sb)
+	pos := f.m.getOrAllocateSSABlockLabelPosition(sb)
+	return pos
+}
+
+// SwapBefore implements regalloc.Function.
+func (f *regAllocFn) SwapBefore(x1, x2, tmp regalloc.VReg, instr *instruction) {
+	f.m.swap(instr.prev, x1, x2, tmp)
+}
+
+// StoreRegisterBefore implements regalloc.Function.
+func (f *regAllocFn) StoreRegisterBefore(v regalloc.VReg, instr *instruction) {
+	m := f.m
+	m.insertStoreRegisterAt(v, instr, false)
+}
+
+// StoreRegisterAfter implements regalloc.Function.
+func (f *regAllocFn) StoreRegisterAfter(v regalloc.VReg, instr *instruction) {
+	m := f.m
+	m.insertStoreRegisterAt(v, instr, true)
+}
+
+// ReloadRegisterBefore implements regalloc.Function.
+func (f *regAllocFn) ReloadRegisterBefore(v regalloc.VReg, instr *instruction) {
+	m := f.m
+	m.insertReloadRegisterAt(v, instr, false)
+}
+
+// ReloadRegisterAfter implements regalloc.Function.
+func (f *regAllocFn) ReloadRegisterAfter(v regalloc.VReg, instr *instruction) {
+	m := f.m
+	m.insertReloadRegisterAt(v, instr, true)
+}
+
+// InsertMoveBefore implements regalloc.Function.
+func (f *regAllocFn) InsertMoveBefore(dst, src regalloc.VReg, instr *instruction) {
+	f.m.insertMoveBefore(dst, src, instr)
+}
+
+// LoopNestingForestChild implements regalloc.Function.
+func (f *regAllocFn) LoopNestingForestChild(pos *labelPosition, i int) *labelPosition {
+	childSB := pos.sb.LoopNestingForestChildren()[i]
+	return f.m.getOrAllocateSSABlockLabelPosition(childSB)
+}
+
+// Succ implements regalloc.Block.
+func (f *regAllocFn) Succ(pos *labelPosition, i int) *labelPosition {
+	succSB := pos.sb.Succ(i)
+	if succSB.ReturnBlock() {
+		return nil
+	}
+	return f.m.getOrAllocateSSABlockLabelPosition(succSB)
+}
+
+// Pred implements regalloc.Block.
+func (f *regAllocFn) Pred(pos *labelPosition, i int) *labelPosition {
+	predSB := pos.sb.Pred(i)
+	return f.m.getOrAllocateSSABlockLabelPosition(predSB)
+}
+
+// BlockParams implements regalloc.Function.
+func (f *regAllocFn) BlockParams(pos *labelPosition, regs *[]regalloc.VReg) []regalloc.VReg {
+	c := f.m.compiler
+	*regs = (*regs)[:0]
+	for i := 0; i < pos.sb.Params(); i++ {
+		v := c.VRegOf(pos.sb.Param(i))
+		*regs = append(*regs, v)
+	}
+	return *regs
+}
+
+// ID implements regalloc.Block.
+func (pos *labelPosition) ID() int32 {
+	return int32(pos.sb.ID())
+}
+
+// InstrIteratorBegin implements regalloc.Block.
+func (pos *labelPosition) InstrIteratorBegin() *instruction {
+	ret := pos.begin
+	pos.cur = ret
+	return ret
+}
+
+// InstrIteratorNext implements regalloc.Block.
+func (pos *labelPosition) InstrIteratorNext() *instruction {
+	for {
+		if pos.cur == pos.end {
+			return nil
+		}
+		instr := pos.cur.next
+		pos.cur = instr
+		if instr == nil {
+			return nil
+		} else if instr.addedBeforeRegAlloc {
+			// Only concerned about the instruction added before regalloc.
+			return instr
+		}
+	}
+}
+
+// InstrRevIteratorBegin implements regalloc.Block.
+func (pos *labelPosition) InstrRevIteratorBegin() *instruction {
+	pos.cur = pos.end
+	return pos.cur
+}
+
+// InstrRevIteratorNext implements regalloc.Block.
+func (pos *labelPosition) InstrRevIteratorNext() *instruction {
+	for {
+		if pos.cur == pos.begin {
+			return nil
+		}
+		instr := pos.cur.prev
+		pos.cur = instr
+		if instr == nil {
+			return nil
+		} else if instr.addedBeforeRegAlloc {
+			// Only concerned about the instruction added before regalloc.
+			return instr
+		}
+	}
+}
+
+// FirstInstr implements regalloc.Block.
+func (pos *labelPosition) FirstInstr() *instruction { return pos.begin }
+
+// LastInstrForInsertion implements regalloc.Block.
+func (pos *labelPosition) LastInstrForInsertion() *instruction {
+	return lastInstrForInsertion(pos.begin, pos.end)
+}
+
+// Preds implements regalloc.Block.
+func (pos *labelPosition) Preds() int { return pos.sb.Preds() }
+
+// Entry implements regalloc.Block.
+func (pos *labelPosition) Entry() bool { return pos.sb.EntryBlock() }
+
+// Succs implements regalloc.Block.
+func (pos *labelPosition) Succs() int { return pos.sb.Succs() }
+
+// LoopHeader implements regalloc.Block.
+func (pos *labelPosition) LoopHeader() bool { return pos.sb.LoopHeader() }
+
+// LoopNestingForestChildren implements regalloc.Block.
+func (pos *labelPosition) LoopNestingForestChildren() int {
+	return len(pos.sb.LoopNestingForestChildren())
+}
+
+func (m *machine) swap(cur *instruction, x1, x2, tmp regalloc.VReg) {
 	prevNext := cur.next
 	var mov1, mov2, mov3 *instruction
 	if x1.RegType() == regalloc.RegTypeInt {
@@ -32,12 +240,12 @@ func (m *machine) Swap(cur *instruction, x1, x2, tmp regalloc.VReg) {
 		if !tmp.Valid() {
 			r2 := x2.RealReg()
 			// Temporarily spill x1 to stack.
-			cur = m.InsertStoreRegisterAt(x1, cur, true).prev
+			cur = m.insertStoreRegisterAt(x1, cur, true).prev
 			// Then move x2 to x1.
 			cur = linkInstr(cur, m.allocateInstr().asFpuMov128(x1, x2))
 			linkInstr(cur, prevNext)
 			// Then reload the original value on x1 from stack to r2.
-			m.InsertReloadRegisterAt(x1.SetRealReg(r2), cur, true)
+			m.insertReloadRegisterAt(x1.SetRealReg(r2), cur, true)
 		} else {
 			mov1 = m.allocateInstr().asFpuMov128(tmp, x1)
 			mov2 = m.allocateInstr().asFpuMov128(x1, x2)
@@ -50,8 +258,7 @@ func (m *machine) Swap(cur *instruction, x1, x2, tmp regalloc.VReg) {
 	}
 }
 
-// InsertMoveBefore implements backend.RegAllocFunctionMachine.
-func (m *machine) InsertMoveBefore(dst, src regalloc.VReg, instr *instruction) {
+func (m *machine) insertMoveBefore(dst, src regalloc.VReg, instr *instruction) {
 	typ := src.RegType()
 	if typ != dst.RegType() {
 		panic("BUG: src and dst must have the same type")
@@ -70,13 +277,7 @@ func (m *machine) InsertMoveBefore(dst, src regalloc.VReg, instr *instruction) {
 	linkInstr(cur, prevNext)
 }
 
-// SSABlockLabel implements backend.RegAllocFunctionMachine.
-func (m *machine) SSABlockLabel(id ssa.BasicBlockID) backend.Label {
-	return m.executableContext.SsaBlockIDToLabels[id]
-}
-
-// InsertStoreRegisterAt implements backend.RegAllocFunctionMachine.
-func (m *machine) InsertStoreRegisterAt(v regalloc.VReg, instr *instruction, after bool) *instruction {
+func (m *machine) insertStoreRegisterAt(v regalloc.VReg, instr *instruction, after bool) *instruction {
 	if !v.IsRealReg() {
 		panic("BUG: VReg must be backed by real reg to be stored")
 	}
@@ -100,8 +301,7 @@ func (m *machine) InsertStoreRegisterAt(v regalloc.VReg, instr *instruction, aft
 	return linkInstr(cur, prevNext)
 }
 
-// InsertReloadRegisterAt implements backend.RegAllocFunctionMachine.
-func (m *machine) InsertReloadRegisterAt(v regalloc.VReg, instr *instruction, after bool) *instruction {
+func (m *machine) insertReloadRegisterAt(v regalloc.VReg, instr *instruction, after bool) *instruction {
 	if !v.IsRealReg() {
 		panic("BUG: VReg must be backed by real reg to be stored")
 	}
@@ -134,8 +334,7 @@ func (m *machine) InsertReloadRegisterAt(v regalloc.VReg, instr *instruction, af
 	return linkInstr(cur, prevNext)
 }
 
-// LastInstrForInsertion implements backend.RegAllocFunctionMachine.
-func (m *machine) LastInstrForInsertion(begin, end *instruction) *instruction {
+func lastInstrForInsertion(begin, end *instruction) *instruction {
 	cur := end
 	for cur.kind == nop0 {
 		cur = cur.prev
diff --git a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/unwind_stack.go b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/unwind_stack.go
index edb0e36e3..a72b86f6b 100644
--- a/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/unwind_stack.go
+++ b/vendor/github.com/tetratelabs/wazero/internal/engine/wazevo/backend/isa/arm64/unwind_stack.go
@@ -14,7 +14,7 @@ func UnwindStack(sp, _, top uintptr, returnAddresses []uintptr) []uintptr {
 
 	var stackBuf []byte
 	{
-		// TODO: use unsafe.Slice after floor version is set to Go 1.20.
+		//nolint:staticcheck
 		hdr := (*reflect.SliceHeader)(unsafe.Pointer(&stackBuf))
 		hdr.Data = sp
 		hdr.Len = l
@@ -78,13 +78,7 @@ func GoCallStackView(stackPointerBeforeGoCall *uint64) []uint64 {
 	//              +-----------------+ <---- stackPointerBeforeGoCall
 	//                 (low address)
 	ptr := unsafe.Pointer(stackPointerBeforeGoCall)
+	data := (*uint64)(unsafe.Add(ptr, 16)) // skips the (frame_size, sliceSize).
 	size := *(*uint64)(unsafe.Add(ptr, 8))
-	var view []uint64
-	{
-		sh := (*reflect.SliceHeader)(unsafe.Pointer(&view))
-		sh.Data = uintptr(unsafe.Add(ptr, 16)) // skips the (frame_size, sliceSize).
-		sh.Len = int(size)
-		sh.Cap = int(size)
-	}
-	return view
+	return unsafe.Slice(data, size)
 }