src/cmd/compile/internal/ssa/schedule.go - go - Git at Google

 // Copyright 2015 The Go Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.

 package ssa

 import "container/heap"

 const (
 	ScorePhi = iota // towards top of block
 	ScoreNilCheck
 	ScoreReadTuple
 	ScoreVarDef
 	ScoreMemory
 	ScoreDefault
 	ScoreFlags
 	ScoreControl // towards bottom of block
 )

 type ValHeap struct {
 	a     []*Value
 	score []int8
 }

 func (h ValHeap) Len() int      { return len(h.a) }
 func (h ValHeap) Swap(i, j int) { a := h.a; a[i], a[j] = a[j], a[i] }

 func (h *ValHeap) Push(x interface{}) {
 	// Push and Pop use pointer receivers because they modify the slice's length,
 	// not just its contents.
 	v := x.(*Value)
 	h.a = append(h.a, v)
 }
 func (h *ValHeap) Pop() interface{} {
 	old := h.a
 	n := len(old)
 	x := old[n-1]
 	h.a = old[0 : n-1]
 	return x
 }
 func (h ValHeap) Less(i, j int) bool {
 	x := h.a[i]
 	y := h.a[j]
 	sx := h.score[x.ID]
 	sy := h.score[y.ID]
 	if c := sx - sy; c != 0 {
 		return c > 0 // higher score comes later.
 	}
 	if x.Pos != y.Pos { // Favor in-order line stepping
 		return x.Pos.After(y.Pos)
 	}
 	if x.Op != OpPhi {
 		if c := len(x.Args) - len(y.Args); c != 0 {
 			return c < 0 // smaller args comes later
 		}
 	}
 	return x.ID > y.ID
 }

 // Schedule the Values in each Block. After this phase returns, the
 // order of b.Values matters and is the order in which those values
 // will appear in the assembly output. For now it generates a
 // reasonable valid schedule using a priority queue. TODO(khr):
 // schedule smarter.
 func schedule(f *Func) {
 	// For each value, the number of times it is used in the block
 	// by values that have not been scheduled yet.
 	uses := make([]int32, f.NumValues())

 	// reusable priority queue
 	priq := new(ValHeap)

 	// "priority" for a value
 	score := make([]int8, f.NumValues())

 	// scheduling order. We queue values in this list in reverse order.
 	var order []*Value

 	// maps mem values to the next live memory value
 	nextMem := make([]*Value, f.NumValues())
 	// additional pretend arguments for each Value. Used to enforce load/store ordering.
 	additionalArgs := make([][]*Value, f.NumValues())

 	for _, b := range f.Blocks {
 		// Compute score. Larger numbers are scheduled closer to the end of the block.
 		for _, v := range b.Values {
 			switch {
 			case v.Op == OpAMD64LoweredGetClosurePtr || v.Op == OpPPC64LoweredGetClosurePtr ||
 				v.Op == OpARMLoweredGetClosurePtr || v.Op == OpARM64LoweredGetClosurePtr ||
 				v.Op == Op386LoweredGetClosurePtr || v.Op == OpMIPS64LoweredGetClosurePtr ||
 				v.Op == OpS390XLoweredGetClosurePtr || v.Op == OpMIPSLoweredGetClosurePtr:
 				// We also score GetLoweredClosurePtr as early as possible to ensure that the
 				// context register is not stomped. GetLoweredClosurePtr should only appear
 				// in the entry block where there are no phi functions, so there is no
 				// conflict or ambiguity here.
 				if b != f.Entry {
 					f.Fatalf("LoweredGetClosurePtr appeared outside of entry block, b=%s", b.String())
 				}
 				score[v.ID] = ScorePhi
 			case v.Op == OpAMD64LoweredNilCheck || v.Op == OpPPC64LoweredNilCheck ||
 				v.Op == OpARMLoweredNilCheck || v.Op == OpARM64LoweredNilCheck ||
 				v.Op == Op386LoweredNilCheck || v.Op == OpMIPS64LoweredNilCheck ||
 				v.Op == OpS390XLoweredNilCheck || v.Op == OpMIPSLoweredNilCheck:
 				// Nil checks must come before loads from the same address.
 				score[v.ID] = ScoreNilCheck
 			case v.Op == OpPhi:
 				// We want all the phis first.
 				score[v.ID] = ScorePhi
 			case v.Op == OpVarDef:
 				// We want all the vardefs next.
 				score[v.ID] = ScoreVarDef
 			case v.Type.IsMemory():
 				// Schedule stores as early as possible. This tends to
 				// reduce register pressure. It also helps make sure
 				// VARDEF ops are scheduled before the corresponding LEA.
 				score[v.ID] = ScoreMemory
 			case v.Op == OpSelect0 || v.Op == OpSelect1:
 				// Schedule the pseudo-op of reading part of a tuple
 				// immediately after the tuple-generating op, since
 				// this value is already live. This also removes its
 				// false dependency on the other part of the tuple.
 				// Also ensures tuple is never spilled.
 				score[v.ID] = ScoreReadTuple
 			case v.Type.IsFlags() || v.Type.IsTuple():
 				// Schedule flag register generation as late as possible.
 				// This makes sure that we only have one live flags
 				// value at a time.
 				score[v.ID] = ScoreFlags
 			default:
 				score[v.ID] = ScoreDefault
 			}
 		}
 	}

 	for _, b := range f.Blocks {
 		// Find store chain for block.
 		// Store chains for different blocks overwrite each other, so
 		// the calculated store chain is good only for this block.
 		for _, v := range b.Values {
 			if v.Op != OpPhi && v.Type.IsMemory() {
 				for _, w := range v.Args {
 					if w.Type.IsMemory() {
 						nextMem[w.ID] = v
 					}
 				}
 			}
 		}

 		// Compute uses.
 		for _, v := range b.Values {
 			if v.Op == OpPhi {
 				// If a value is used by a phi, it does not induce
 				// a scheduling edge because that use is from the
 				// previous iteration.
 				continue
 			}
 			for _, w := range v.Args {
 				if w.Block == b {
 					uses[w.ID]++
 				}
 				// Any load must come before the following store.
 				if !v.Type.IsMemory() && w.Type.IsMemory() {
 					// v is a load.
 					s := nextMem[w.ID]
 					if s == nil || s.Block != b {
 						continue
 					}
 					additionalArgs[s.ID] = append(additionalArgs[s.ID], v)
 					uses[v.ID]++
 				}
 			}
 		}

 		if b.Control != nil && b.Control.Op != OpPhi {
 			// Force the control value to be scheduled at the end,
 			// unless it is a phi value (which must be first).
 			score[b.Control.ID] = ScoreControl

 			// Schedule values dependent on the control value at the end.
 			// This reduces the number of register spills. We don't find
 			// all values that depend on the control, just values with a
 			// direct dependency. This is cheaper and in testing there
 			// was no difference in the number of spills.
 			for _, v := range b.Values {
 				if v.Op != OpPhi {
 					for _, a := range v.Args {
 						if a == b.Control {
 							score[v.ID] = ScoreControl
 						}
 					}
 				}
 			}
 		}

 		// To put things into a priority queue
 		// The values that should come last are least.
 		priq.score = score
 		priq.a = priq.a[:0]

 		// Initialize priority queue with schedulable values.
 		for _, v := range b.Values {
 			if uses[v.ID] == 0 {
 				heap.Push(priq, v)
 			}
 		}

 		// Schedule highest priority value, update use counts, repeat.
 		order = order[:0]
 		tuples := make(map[ID][]*Value)
 		for {
 			// Find highest priority schedulable value.
 			// Note that schedule is assembled backwards.

 			if priq.Len() == 0 {
 				break
 			}

 			v := heap.Pop(priq).(*Value)

 			// Add it to the schedule.
 			// Do not emit tuple-reading ops until we're ready to emit the tuple-generating op.
 			//TODO: maybe remove ReadTuple score above, if it does not help on performance
 			switch {
 			case v.Op == OpSelect0:
 				if tuples[v.Args[0].ID] == nil {
 					tuples[v.Args[0].ID] = make([]*Value, 2)
 				}
 				tuples[v.Args[0].ID][0] = v
 			case v.Op == OpSelect1:
 				if tuples[v.Args[0].ID] == nil {
 					tuples[v.Args[0].ID] = make([]*Value, 2)
 				}
 				tuples[v.Args[0].ID][1] = v
 			case v.Type.IsTuple() && tuples[v.ID] != nil:
 				if tuples[v.ID][1] != nil {
 					order = append(order, tuples[v.ID][1])
 				}
 				if tuples[v.ID][0] != nil {
 					order = append(order, tuples[v.ID][0])
 				}
 				delete(tuples, v.ID)
 				fallthrough
 			default:
 				order = append(order, v)
 			}

 			// Update use counts of arguments.
 			for _, w := range v.Args {
 				if w.Block != b {
 					continue
 				}
 				uses[w.ID]--
 				if uses[w.ID] == 0 {
 					// All uses scheduled, w is now schedulable.
 					heap.Push(priq, w)
 				}
 			}
 			for _, w := range additionalArgs[v.ID] {
 				uses[w.ID]--
 				if uses[w.ID] == 0 {
 					// All uses scheduled, w is now schedulable.
 					heap.Push(priq, w)
 				}
 			}
 		}
 		if len(order) != len(b.Values) {
 			f.Fatalf("schedule does not include all values")
 		}
 		for i := 0; i < len(b.Values); i++ {
 			b.Values[i] = order[len(b.Values)-1-i]
 		}
 	}

 	f.scheduled = true
 }

 // storeOrder orders values with respect to stores. That is,
 // if v transitively depends on store s, v is ordered after s,
 // otherwise v is ordered before s.
 // Specifically, values are ordered like
 //   store1
 //   NilCheck that depends on store1
 //   other values that depends on store1
 //   store2
 //   NilCheck that depends on store2
 //   other values that depends on store2
 //   ...
 // The order of non-store and non-NilCheck values are undefined
 // (not necessarily dependency order). This should be cheaper
 // than a full scheduling as done above.
 // Note that simple dependency order won't work: there is no
 // dependency between NilChecks and values like IsNonNil.
 // Auxiliary data structures are passed in as arguments, so
 // that they can be allocated in the caller and be reused.
 // This function takes care of reset them.
 func storeOrder(values []*Value, sset *sparseSet, storeNumber []int32) []*Value {
 	if len(values) == 0 {
 		return values
 	}

 	f := values[0].Block.Func

 	// find all stores
 	var stores []*Value // members of values that are store values
 	hasNilCheck := false
 	sset.clear() // sset is the set of stores that are used in other values
 	for _, v := range values {
 		if v.Type.IsMemory() {
 			stores = append(stores, v)
 			if v.Op == OpInitMem || v.Op == OpPhi {
 				continue
 			}
 			sset.add(v.MemoryArg().ID) // record that v's memory arg is used
 		}
 		if v.Op == OpNilCheck {
 			hasNilCheck = true
 		}
 	}
 	if len(stores) == 0 || !hasNilCheck && f.pass.name == "nilcheckelim" {
 		// there is no store, the order does not matter
 		return values
 	}

 	// find last store, which is the one that is not used by other stores
 	var last *Value
 	for _, v := range stores {
 		if !sset.contains(v.ID) {
 			if last != nil {
 				f.Fatalf("two stores live simultaneously: %v and %v", v, last)
 			}
 			last = v
 		}
 	}

 	// We assign a store number to each value. Store number is the
 	// index of the latest store that this value transitively depends.
 	// The i-th store in the current block gets store number 3*i. A nil
 	// check that depends on the i-th store gets store number 3*i+1.
 	// Other values that depends on the i-th store gets store number 3*i+2.
 	// Special case: 0 -- unassigned, 1 or 2 -- the latest store it depends
 	// is in the previous block (or no store at all, e.g. value is Const).
 	// First we assign the number to all stores by walking back the store chain,
 	// then assign the number to other values in DFS order.
 	count := make([]int32, 3*(len(stores)+1))
 	sset.clear() // reuse sparse set to ensure that a value is pushed to stack only once
 	for n, w := len(stores), last; n > 0; n-- {
 		storeNumber[w.ID] = int32(3 * n)
 		count[3*n]++
 		sset.add(w.ID)
 		if w.Op == OpInitMem || w.Op == OpPhi {
 			if n != 1 {
 				f.Fatalf("store order is wrong: there are stores before %v", w)
 			}
 			break
 		}
 		w = w.MemoryArg()
 	}
 	var stack []*Value
 	for _, v := range values {
 		if sset.contains(v.ID) {
 			// in sset means v is a store, or already pushed to stack, or already assigned a store number
 			continue
 		}
 		stack = append(stack, v)
 		sset.add(v.ID)

 		for len(stack) > 0 {
 			w := stack[len(stack)-1]
 			if storeNumber[w.ID] != 0 {
 				stack = stack[:len(stack)-1]
 				continue
 			}
 			if w.Op == OpPhi {
 				// Phi value doesn't depend on store in the current block.
 				// Do this early to avoid dependency cycle.
 				storeNumber[w.ID] = 2
 				count[2]++
 				stack = stack[:len(stack)-1]
 				continue
 			}

 			max := int32(0) // latest store dependency
 			argsdone := true
 			for _, a := range w.Args {
 				if a.Block != w.Block {
 					continue
 				}
 				if !sset.contains(a.ID) {
 					stack = append(stack, a)
 					sset.add(a.ID)
 					argsdone = false
 					break
 				}
 				if storeNumber[a.ID]/3 > max {
 					max = storeNumber[a.ID] / 3
 				}
 			}
 			if !argsdone {
 				continue
 			}

 			n := 3*max + 2
 			if w.Op == OpNilCheck {
 				n = 3*max + 1
 			}
 			storeNumber[w.ID] = n
 			count[n]++
 			stack = stack[:len(stack)-1]
 		}
 	}

 	// convert count to prefix sum of counts: count'[i] = sum_{j<=i} count[i]
 	for i := range count {
 		if i == 0 {
 			continue
 		}
 		count[i] += count[i-1]
 	}
 	if count[len(count)-1] != int32(len(values)) {
 		f.Fatalf("storeOrder: value is missing, total count = %d, values = %v", count[len(count)-1], values)
 	}

 	// place values in count-indexed bins, which are in the desired store order
 	order := make([]*Value, len(values))
 	for _, v := range values {
 		s := storeNumber[v.ID]
 		order[count[s-1]] = v
 		count[s-1]++
 	}

 	return order
 }
	// Copyright 2015 The Go Authors. All rights reserved.
	// Use of this source code is governed by a BSD-style
	// license that can be found in the LICENSE file.

	package ssa

	import "container/heap"

	const (
	ScorePhi = iota // towards top of block
	ScoreNilCheck
	ScoreReadTuple
	ScoreVarDef
	ScoreMemory
	ScoreDefault
	ScoreFlags
	ScoreControl // towards bottom of block
	)

	type ValHeap struct {
	a []*Value
	score []int8
	}

	func (h ValHeap) Len() int { return len(h.a) }
	func (h ValHeap) Swap(i, j int) { a := h.a; a[i], a[j] = a[j], a[i] }

	func (h *ValHeap) Push(x interface{}) {
	// Push and Pop use pointer receivers because they modify the slice's length,
	// not just its contents.
	v := x.(*Value)
	h.a = append(h.a, v)
	}
	func (h *ValHeap) Pop() interface{} {
	old := h.a
	n := len(old)
	x := old[n-1]
	h.a = old[0 : n-1]
	return x
	}
	func (h ValHeap) Less(i, j int) bool {
	x := h.a[i]
	y := h.a[j]
	sx := h.score[x.ID]
	sy := h.score[y.ID]
	if c := sx - sy; c != 0 {
	return c > 0 // higher score comes later.
	}
	if x.Pos != y.Pos { // Favor in-order line stepping
	return x.Pos.After(y.Pos)
	}
	if x.Op != OpPhi {
	if c := len(x.Args) - len(y.Args); c != 0 {
	return c < 0 // smaller args comes later
	}
	}
	return x.ID > y.ID
	}

	// Schedule the Values in each Block. After this phase returns, the
	// order of b.Values matters and is the order in which those values
	// will appear in the assembly output. For now it generates a
	// reasonable valid schedule using a priority queue. TODO(khr):
	// schedule smarter.
	func schedule(f *Func) {
	// For each value, the number of times it is used in the block
	// by values that have not been scheduled yet.
	uses := make([]int32, f.NumValues())

	// reusable priority queue
	priq := new(ValHeap)

	// "priority" for a value
	score := make([]int8, f.NumValues())

	// scheduling order. We queue values in this list in reverse order.
	var order []*Value

	// maps mem values to the next live memory value
	nextMem := make([]*Value, f.NumValues())
	// additional pretend arguments for each Value. Used to enforce load/store ordering.
	additionalArgs := make([][]*Value, f.NumValues())

	for _, b := range f.Blocks {
	// Compute score. Larger numbers are scheduled closer to the end of the block.
	for _, v := range b.Values {
	switch {
	case v.Op == OpAMD64LoweredGetClosurePtr \|\| v.Op == OpPPC64LoweredGetClosurePtr \|\|
	v.Op == OpARMLoweredGetClosurePtr \|\| v.Op == OpARM64LoweredGetClosurePtr \|\|
	v.Op == Op386LoweredGetClosurePtr \|\| v.Op == OpMIPS64LoweredGetClosurePtr \|\|
	v.Op == OpS390XLoweredGetClosurePtr \|\| v.Op == OpMIPSLoweredGetClosurePtr:
	// We also score GetLoweredClosurePtr as early as possible to ensure that the
	// context register is not stomped. GetLoweredClosurePtr should only appear
	// in the entry block where there are no phi functions, so there is no
	// conflict or ambiguity here.
	if b != f.Entry {
	f.Fatalf("LoweredGetClosurePtr appeared outside of entry block, b=%s", b.String())
	}
	score[v.ID] = ScorePhi
	case v.Op == OpAMD64LoweredNilCheck \|\| v.Op == OpPPC64LoweredNilCheck \|\|
	v.Op == OpARMLoweredNilCheck \|\| v.Op == OpARM64LoweredNilCheck \|\|
	v.Op == Op386LoweredNilCheck \|\| v.Op == OpMIPS64LoweredNilCheck \|\|
	v.Op == OpS390XLoweredNilCheck \|\| v.Op == OpMIPSLoweredNilCheck:
	// Nil checks must come before loads from the same address.
	score[v.ID] = ScoreNilCheck
	case v.Op == OpPhi:
	// We want all the phis first.
	score[v.ID] = ScorePhi
	case v.Op == OpVarDef:
	// We want all the vardefs next.
	score[v.ID] = ScoreVarDef
	case v.Type.IsMemory():
	// Schedule stores as early as possible. This tends to
	// reduce register pressure. It also helps make sure
	// VARDEF ops are scheduled before the corresponding LEA.
	score[v.ID] = ScoreMemory
	case v.Op == OpSelect0 \|\| v.Op == OpSelect1:
	// Schedule the pseudo-op of reading part of a tuple
	// immediately after the tuple-generating op, since
	// this value is already live. This also removes its
	// false dependency on the other part of the tuple.
	// Also ensures tuple is never spilled.
	score[v.ID] = ScoreReadTuple
	case v.Type.IsFlags() \|\| v.Type.IsTuple():
	// Schedule flag register generation as late as possible.
	// This makes sure that we only have one live flags
	// value at a time.
	score[v.ID] = ScoreFlags
	default:
	score[v.ID] = ScoreDefault
	}
	}
	}

	for _, b := range f.Blocks {
	// Find store chain for block.
	// Store chains for different blocks overwrite each other, so
	// the calculated store chain is good only for this block.
	for _, v := range b.Values {
	if v.Op != OpPhi && v.Type.IsMemory() {
	for _, w := range v.Args {
	if w.Type.IsMemory() {
	nextMem[w.ID] = v
	}
	}
	}
	}

	// Compute uses.
	for _, v := range b.Values {
	if v.Op == OpPhi {
	// If a value is used by a phi, it does not induce
	// a scheduling edge because that use is from the
	// previous iteration.
	continue
	}
	for _, w := range v.Args {
	if w.Block == b {
	uses[w.ID]++
	}
	// Any load must come before the following store.
	if !v.Type.IsMemory() && w.Type.IsMemory() {
	// v is a load.
	s := nextMem[w.ID]
	if s == nil \|\| s.Block != b {
	continue
	}
	additionalArgs[s.ID] = append(additionalArgs[s.ID], v)
	uses[v.ID]++
	}
	}
	}

	if b.Control != nil && b.Control.Op != OpPhi {
	// Force the control value to be scheduled at the end,
	// unless it is a phi value (which must be first).
	score[b.Control.ID] = ScoreControl

	// Schedule values dependent on the control value at the end.
	// This reduces the number of register spills. We don't find
	// all values that depend on the control, just values with a
	// direct dependency. This is cheaper and in testing there
	// was no difference in the number of spills.
	for _, v := range b.Values {
	if v.Op != OpPhi {
	for _, a := range v.Args {
	if a == b.Control {
	score[v.ID] = ScoreControl
	}
	}
	}
	}
	}

	// To put things into a priority queue
	// The values that should come last are least.
	priq.score = score
	priq.a = priq.a[:0]

	// Initialize priority queue with schedulable values.
	for _, v := range b.Values {
	if uses[v.ID] == 0 {
	heap.Push(priq, v)
	}
	}

	// Schedule highest priority value, update use counts, repeat.
	order = order[:0]
	tuples := make(map[ID][]*Value)
	for {
	// Find highest priority schedulable value.
	// Note that schedule is assembled backwards.

	if priq.Len() == 0 {
	break
	}

	v := heap.Pop(priq).(*Value)

	// Add it to the schedule.
	// Do not emit tuple-reading ops until we're ready to emit the tuple-generating op.
	//TODO: maybe remove ReadTuple score above, if it does not help on performance
	switch {
	case v.Op == OpSelect0:
	if tuples[v.Args[0].ID] == nil {
	tuples[v.Args[0].ID] = make([]*Value, 2)
	}
	tuples[v.Args[0].ID][0] = v
	case v.Op == OpSelect1:
	if tuples[v.Args[0].ID] == nil {
	tuples[v.Args[0].ID] = make([]*Value, 2)
	}
	tuples[v.Args[0].ID][1] = v
	case v.Type.IsTuple() && tuples[v.ID] != nil:
	if tuples[v.ID][1] != nil {
	order = append(order, tuples[v.ID][1])
	}
	if tuples[v.ID][0] != nil {
	order = append(order, tuples[v.ID][0])
	}
	delete(tuples, v.ID)
	fallthrough
	default:
	order = append(order, v)
	}

	// Update use counts of arguments.
	for _, w := range v.Args {
	if w.Block != b {
	continue
	}
	uses[w.ID]--
	if uses[w.ID] == 0 {
	// All uses scheduled, w is now schedulable.
	heap.Push(priq, w)
	}
	}
	for _, w := range additionalArgs[v.ID] {
	uses[w.ID]--
	if uses[w.ID] == 0 {
	// All uses scheduled, w is now schedulable.
	heap.Push(priq, w)
	}
	}
	}
	if len(order) != len(b.Values) {
	f.Fatalf("schedule does not include all values")
	}
	for i := 0; i < len(b.Values); i++ {
	b.Values[i] = order[len(b.Values)-1-i]
	}
	}

	f.scheduled = true
	}

	// storeOrder orders values with respect to stores. That is,
	// if v transitively depends on store s, v is ordered after s,
	// otherwise v is ordered before s.
	// Specifically, values are ordered like
	// store1
	// NilCheck that depends on store1
	// other values that depends on store1
	// store2
	// NilCheck that depends on store2
	// other values that depends on store2
	// ...
	// The order of non-store and non-NilCheck values are undefined
	// (not necessarily dependency order). This should be cheaper
	// than a full scheduling as done above.
	// Note that simple dependency order won't work: there is no
	// dependency between NilChecks and values like IsNonNil.
	// Auxiliary data structures are passed in as arguments, so
	// that they can be allocated in the caller and be reused.
	// This function takes care of reset them.
	func storeOrder(values []Value, sset sparseSet, storeNumber []int32) []*Value {
	if len(values) == 0 {
	return values
	}

	f := values[0].Block.Func

	// find all stores
	var stores []*Value // members of values that are store values
	hasNilCheck := false
	sset.clear() // sset is the set of stores that are used in other values
	for _, v := range values {
	if v.Type.IsMemory() {
	stores = append(stores, v)
	if v.Op == OpInitMem \|\| v.Op == OpPhi {
	continue
	}
	sset.add(v.MemoryArg().ID) // record that v's memory arg is used
	}
	if v.Op == OpNilCheck {
	hasNilCheck = true
	}
	}
	if len(stores) == 0 \|\| !hasNilCheck && f.pass.name == "nilcheckelim" {
	// there is no store, the order does not matter
	return values
	}

	// find last store, which is the one that is not used by other stores
	var last *Value
	for _, v := range stores {
	if !sset.contains(v.ID) {
	if last != nil {
	f.Fatalf("two stores live simultaneously: %v and %v", v, last)
	}
	last = v
	}
	}

	// We assign a store number to each value. Store number is the
	// index of the latest store that this value transitively depends.
	// The i-th store in the current block gets store number 3*i. A nil
	// check that depends on the i-th store gets store number 3*i+1.
	// Other values that depends on the i-th store gets store number 3*i+2.
	// Special case: 0 -- unassigned, 1 or 2 -- the latest store it depends
	// is in the previous block (or no store at all, e.g. value is Const).
	// First we assign the number to all stores by walking back the store chain,
	// then assign the number to other values in DFS order.
	count := make([]int32, 3*(len(stores)+1))
	sset.clear() // reuse sparse set to ensure that a value is pushed to stack only once
	for n, w := len(stores), last; n > 0; n-- {
	storeNumber[w.ID] = int32(3 * n)
	count[3*n]++
	sset.add(w.ID)
	if w.Op == OpInitMem \|\| w.Op == OpPhi {
	if n != 1 {
	f.Fatalf("store order is wrong: there are stores before %v", w)
	}
	break
	}
	w = w.MemoryArg()
	}
	var stack []*Value
	for _, v := range values {
	if sset.contains(v.ID) {
	// in sset means v is a store, or already pushed to stack, or already assigned a store number
	continue
	}
	stack = append(stack, v)
	sset.add(v.ID)

	for len(stack) > 0 {
	w := stack[len(stack)-1]
	if storeNumber[w.ID] != 0 {
	stack = stack[:len(stack)-1]
	continue
	}
	if w.Op == OpPhi {
	// Phi value doesn't depend on store in the current block.
	// Do this early to avoid dependency cycle.
	storeNumber[w.ID] = 2
	count[2]++
	stack = stack[:len(stack)-1]
	continue
	}

	max := int32(0) // latest store dependency
	argsdone := true
	for _, a := range w.Args {
	if a.Block != w.Block {
	continue
	}
	if !sset.contains(a.ID) {
	stack = append(stack, a)
	sset.add(a.ID)
	argsdone = false
	break
	}
	if storeNumber[a.ID]/3 > max {
	max = storeNumber[a.ID] / 3
	}
	}
	if !argsdone {
	continue
	}

	n := 3*max + 2
	if w.Op == OpNilCheck {
	n = 3*max + 1
	}
	storeNumber[w.ID] = n
	count[n]++
	stack = stack[:len(stack)-1]
	}
	}

	// convert count to prefix sum of counts: count'[i] = sum_{j<=i} count[i]
	for i := range count {
	if i == 0 {
	continue
	}
	count[i] += count[i-1]
	}
	if count[len(count)-1] != int32(len(values)) {
	f.Fatalf("storeOrder: value is missing, total count = %d, values = %v", count[len(count)-1], values)
	}

	// place values in count-indexed bins, which are in the desired store order
	order := make([]*Value, len(values))
	for _, v := range values {
	s := storeNumber[v.ID]
	order[count[s-1]] = v
	count[s-1]++
	}

	return order
	}