cmd/compile: index line number tables by source file to improve sparsity

This reduces allocations and also resolves some
lurking inliner/inlinee line-number match problems.
However, it does add about 1.5% to compile time.

This fixes compiler OOMs seen compiling some large protobuf-
derived inputs.  For compiling the compiler itself,

compilebench -pkg cmd/compile/internal/ssa -memprofile withcl.prof

the numberlines-related memory consumption is reduced from 129MB
to 29MB (about a 5% overall reduction in allocation).

Additionally modified after going over changes with Austin
to remove unused code (nobody called size()) and correct
the cache-clearing code.

I've attempted to speed this up by not using maps, and have
not succeeded.  I'd rather get correct code in now, speed it
up later if I can.

Updates #27739.
Fixes #29279.

Change-Id: I098005de4e45196a5f5b10c0886a49f88e9f8fd5
Reviewed-on: https://go-review.googlesource.com/c/go/+/154617
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

src/cmd/compile/internal/ssa/biasedsparsemap.go

@@ -29,7 +29,7 @@ func newBiasedSparseMap(first, last int) *biasedSparseMap {
 
 // cap returns one more than the largest key valid for s
 func (s *biasedSparseMap) cap() int {
-	if s.s == nil {
+	if s == nil || s.s == nil {
 		return 0
 	}
 	return s.s.cap() + int(s.first)
@@ -37,7 +37,7 @@ func (s *biasedSparseMap) cap() int {
 
 // size returns the number of entries stored in s
 func (s *biasedSparseMap) size() int {
-	if s.s == nil {
+	if s == nil || s.s == nil {
 		return 0
 	}
 	return s.s.size()
@@ -45,7 +45,7 @@ func (s *biasedSparseMap) size() int {
 
 // contains reports whether x is a key in s
 func (s *biasedSparseMap) contains(x uint) bool {
-	if s.s == nil {
+	if s == nil || s.s == nil {
 		return false
 	}
 	if int(x) < s.first {
@@ -60,7 +60,7 @@ func (s *biasedSparseMap) contains(x uint) bool {
 // get returns the value s maps for key x, or -1 if
 // x is not mapped or is out of range for s.
 func (s *biasedSparseMap) get(x uint) int32 {
-	if s.s == nil {
+	if s == nil || s.s == nil {
 		return -1
 	}
 	if int(x) < s.first {

src/cmd/compile/internal/ssa/deadcode.go

@@ -258,7 +258,7 @@ func deadcode(f *Func) {
 			if !live[v.ID] {
 				v.resetArgs()
 				if v.Pos.IsStmt() == src.PosIsStmt && reachable[b.ID] {
-					pendingLines.set(v.Pos.Line(), int32(i)) // TODO could be more than one pos for a line
+					pendingLines.set(v.Pos, int32(i)) // TODO could be more than one pos for a line
 				}
 			}
 		}
@@ -267,20 +267,19 @@ func deadcode(f *Func) {
 	// Find new homes for lost lines -- require earliest in data flow with same line that is also in same block
 	for i := len(order) - 1; i >= 0; i-- {
 		w := order[i]
-		if j := pendingLines.get(w.Pos.Line()); j > -1 && f.Blocks[j] == w.Block {
+		if j := pendingLines.get(w.Pos); j > -1 && f.Blocks[j] == w.Block {
 			w.Pos = w.Pos.WithIsStmt()
-			pendingLines.remove(w.Pos.Line())
+			pendingLines.remove(w.Pos)
 		}
 	}
 
 	// Any boundary that failed to match a live value can move to a block end
-	for i := 0; i < pendingLines.size(); i++ {
-		l, bi := pendingLines.getEntry(i)
+	pendingLines.foreachEntry(func(j int32, l uint, bi int32) {
 		b := f.Blocks[bi]
-		if b.Pos.Line() == l {
+		if b.Pos.Line() == l && b.Pos.FileIndex() == j {
 			b.Pos = b.Pos.WithIsStmt()
 		}
-	}
+	})
 
 	// Remove dead values from blocks' value list. Return dead
 	// values to the allocator.

src/cmd/compile/internal/ssa/func.go

@@ -69,7 +69,7 @@ type Func struct {
 	cachedIdom       []*Block   // cached immediate dominators
 	cachedSdom       SparseTree // cached dominator tree
 	cachedLoopnest   *loopnest  // cached loop nest information
-	cachedLineStarts *biasedSparseMap // cached map/set of line numbers to integers
+	cachedLineStarts *xposmap   // cached map/set of xpos to integers
 
 	auxmap    auxmap             // map from aux values to opaque ids used by CSE
 	constants map[int64][]*Value // constants cache, keyed by constant value; users must check value's Op and Type

src/cmd/compile/internal/ssa/func_test.go

@@ -152,7 +152,7 @@ func (c *Conf) Fun(entry string, blocs ...bloc) fun {
 	// But not both.
 	f.Cache = new(Cache)
 	f.pass = &emptyPass
-	f.cachedLineStarts = newBiasedSparseMap(0, 100)
+	f.cachedLineStarts = newXposmap(map[int]lineRange{0: {0, 100}, 1: {0, 100}, 2: {0, 100}, 3: {0, 100}, 4: {0, 100}})
 
 	blocks := make(map[string]*Block)
 	values := make(map[string]*Value)

src/cmd/compile/internal/ssa/nilcheck.go

@@ -124,7 +124,7 @@ func nilcheckelim(f *Func) {
 					ptr := v.Args[0]
 					if nonNilValues[ptr.ID] {
 						if v.Pos.IsStmt() == src.PosIsStmt { // Boolean true is a terrible statement boundary.
-							pendingLines.add(v.Pos.Line())
+							pendingLines.add(v.Pos)
 							v.Pos = v.Pos.WithNotStmt()
 						}
 						// This is a redundant explicit nil check.
@@ -141,7 +141,7 @@ func nilcheckelim(f *Func) {
 							f.Warnl(v.Pos, "removed nil check")
 						}
 						if v.Pos.IsStmt() == src.PosIsStmt { // About to lose a statement boundary
-							pendingLines.add(v.Pos.Line())
+							pendingLines.add(v.Pos)
 						}
 						v.reset(OpUnknown)
 						f.freeValue(v)
@@ -154,15 +154,15 @@ func nilcheckelim(f *Func) {
 					work = append(work, bp{op: ClearPtr, ptr: ptr})
 					fallthrough // a non-eliminated nil check might be a good place for a statement boundary.
 				default:
-					if pendingLines.contains(v.Pos.Line()) && v.Pos.IsStmt() != src.PosNotStmt {
+					if pendingLines.contains(v.Pos) && v.Pos.IsStmt() != src.PosNotStmt {
 						v.Pos = v.Pos.WithIsStmt()
-						pendingLines.remove(v.Pos.Line())
+						pendingLines.remove(v.Pos)
 					}
 				}
 			}
-			if pendingLines.contains(b.Pos.Line()) {
+			if pendingLines.contains(b.Pos) {
 				b.Pos = b.Pos.WithIsStmt()
-				pendingLines.remove(b.Pos.Line())
+				pendingLines.remove(b.Pos)
 			}
 			for j := i; j < len(b.Values); j++ {
 				b.Values[j] = nil
@@ -212,7 +212,7 @@ func nilcheckelim2(f *Func) {
 					f.Warnl(v.Pos, "removed nil check")
 				}
 				if v.Pos.IsStmt() == src.PosIsStmt {
-					pendingLines.add(v.Pos.Line())
+					pendingLines.add(v.Pos)
 				}
 				v.reset(OpUnknown)
 				firstToRemove = i
@@ -273,16 +273,16 @@ func nilcheckelim2(f *Func) {
 		for j := i; j < len(b.Values); j++ {
 			v := b.Values[j]
 			if v.Op != OpUnknown {
-				if v.Pos.IsStmt() != src.PosNotStmt && pendingLines.contains(v.Pos.Line()) {
+				if v.Pos.IsStmt() != src.PosNotStmt && pendingLines.contains(v.Pos) {
 					v.Pos = v.Pos.WithIsStmt()
-					pendingLines.remove(v.Pos.Line())
+					pendingLines.remove(v.Pos)
 				}
 				b.Values[i] = v
 				i++
 			}
 		}
 
-		if pendingLines.contains(b.Pos.Line()) {
+		if pendingLines.contains(b.Pos) {
 			b.Pos = b.Pos.WithIsStmt()
 		}
 

src/cmd/compile/internal/ssa/numberlines.go

@@ -7,7 +7,6 @@ package ssa
 import (
 	"cmd/internal/obj"
 	"cmd/internal/src"
-	"math"
 )
 
 func isPoorStatementOp(op Op) bool {
@@ -51,7 +50,7 @@ func nextGoodStatementIndex(v *Value, i int, b *Block) int {
 		if b.Values[j].Pos.IsStmt() == src.PosNotStmt { // ignore non-statements
 			continue
 		}
-		if b.Values[j].Pos.Line() == v.Pos.Line() {
+		if b.Values[j].Pos.Line() == v.Pos.Line() && v.Pos.SameFile(b.Values[j].Pos) {
 			return j
 		}
 		return i
@@ -86,14 +85,22 @@ func (b *Block) FirstPossibleStmtValue() *Value {
 func numberLines(f *Func) {
 	po := f.Postorder()
 	endlines := make(map[ID]src.XPos)
-	last := uint(0)              // uint follows type of XPos.Line()
-	first := uint(math.MaxInt32) // unsigned, but large valid int when cast
-	note := func(line uint) {
-		if line < first {
-			first = line
+	ranges := make(map[int]lineRange)
+	note := func(p src.XPos) {
+		line := uint32(p.Line())
+		i := int(p.FileIndex())
+		lp, found := ranges[i]
+		change := false
+		if line < lp.first || !found {
+			lp.first = line
+			change = true
 		}
-		if line > last {
-			last = line
+		if line > lp.last {
+			lp.last = line
+			change = true
+		}
+		if change {
+			ranges[i] = lp
 		}
 	}
 
@@ -104,12 +111,12 @@ func numberLines(f *Func) {
 		firstPos := src.NoXPos
 		firstPosIndex := -1
 		if b.Pos.IsStmt() != src.PosNotStmt {
-			note(b.Pos.Line())
+			note(b.Pos)
 		}
 		for i := 0; i < len(b.Values); i++ {
 			v := b.Values[i]
 			if v.Pos.IsStmt() != src.PosNotStmt {
-				note(v.Pos.Line())
+				note(v.Pos)
 				// skip ahead to better instruction for this line if possible
 				i = nextGoodStatementIndex(v, i, b)
 				v = b.Values[i]
@@ -161,7 +168,7 @@ func numberLines(f *Func) {
 			if v.Pos.IsStmt() == src.PosNotStmt {
 				continue
 			}
-			note(v.Pos.Line())
+			note(v.Pos)
 			// skip ahead if possible
 			i = nextGoodStatementIndex(v, i, b)
 			v = b.Values[i]
@@ -178,5 +185,6 @@ func numberLines(f *Func) {
 		}
 		endlines[b.ID] = firstPos
 	}
-	f.cachedLineStarts = newBiasedSparseMap(int(first), int(last))
+	// cachedLineStarts is an empty sparse map for values that are included within ranges.
+	f.cachedLineStarts = newXposmap(ranges)
 }

src/cmd/compile/internal/ssa/rewrite.go

@@ -64,7 +64,7 @@ func applyRewrite(f *Func, rb blockRewriter, rv valueRewriter) {
 							// TODO: it's possible (in FOR loops, in particular) for statement boundaries for the same
 							// line to appear in more than one block, but only one block is stored, so if both end
 							// up here, then one will be lost.
-							pendingLines.set(a.Pos.Line(), int32(a.Block.ID))
+							pendingLines.set(a.Pos, int32(a.Block.ID))
 						}
 						a.Pos = a.Pos.WithNotStmt()
 					}
@@ -97,7 +97,7 @@ func applyRewrite(f *Func, rb blockRewriter, rv valueRewriter) {
 	for _, b := range f.Blocks {
 		j := 0
 		for i, v := range b.Values {
-			vl := v.Pos.Line()
+			vl := v.Pos
 			if v.Op == OpInvalid {
 				if v.Pos.IsStmt() == src.PosIsStmt {
 					pendingLines.set(vl, int32(b.ID))
@@ -114,9 +114,9 @@ func applyRewrite(f *Func, rb blockRewriter, rv valueRewriter) {
 			}
 			j++
 		}
-		if pendingLines.get(b.Pos.Line()) == int32(b.ID) {
+		if pendingLines.get(b.Pos) == int32(b.ID) {
 			b.Pos = b.Pos.WithIsStmt()
-			pendingLines.remove(b.Pos.Line())
+			pendingLines.remove(b.Pos)
 		}
 		if j != len(b.Values) {
 			tail := b.Values[j:]

src/cmd/compile/internal/ssa/xposmap.go

+// Copyright 2019 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 (
+	"cmd/internal/src"
+	"fmt"
+)
+
+type lineRange struct {
+	first, last uint32
+}
+
+// An xposmap is a map from fileindex and line of src.XPos to int32,
+// implemented sparsely to save space (column and statement status are ignored).
+// The sparse skeleton is constructed once, and then reused by ssa phases
+// that (re)move values with statements attached.
+type xposmap struct {
+	// A map from file index to maps from line range to integers (block numbers)
+	maps map[int32]*biasedSparseMap
+	// The next two fields provide a single-item cache for common case of repeated lines from same file.
+	lastIndex int32            // -1 means no entry in cache
+	lastMap   *biasedSparseMap // map found at maps[lastIndex]
+}
+
+// newXposmap constructs an xposmap valid for inputs which have a file index in the keys of x,
+// and line numbers in the range x[file index].
+// The resulting xposmap will panic if a caller attempts to set or add an XPos not in that range.
+func newXposmap(x map[int]lineRange) *xposmap {
+	maps := make(map[int32]*biasedSparseMap)
+	for i, p := range x {
+		maps[int32(i)] = newBiasedSparseMap(int(p.first), int(p.last))
+	}
+	return &xposmap{maps: maps, lastIndex: -1} // zero for the rest is okay
+}
+
+// clear removes data from the map but leaves the sparse skeleton.
+func (m *xposmap) clear() {
+	for _, l := range m.maps {
+		if l != nil {
+			l.clear()
+		}
+	}
+	m.lastIndex = -1
+	m.lastMap = nil
+}
+
+// mapFor returns the line range map for a given file index.
+func (m *xposmap) mapFor(index int32) *biasedSparseMap {
+	if index == m.lastIndex {
+		return m.lastMap
+	}
+	mf := m.maps[index]
+	m.lastIndex = index
+	m.lastMap = mf
+	return mf
+}
+
+// set inserts p->v into the map.
+// If p does not fall within the set of fileindex->lineRange used to construct m, this will panic.
+func (m *xposmap) set(p src.XPos, v int32) {
+	s := m.mapFor(p.FileIndex())
+	if s == nil {
+		panic(fmt.Sprintf("xposmap.set(%d), file index not found in map\n", p.FileIndex()))
+	}
+	s.set(p.Line(), v)
+}
+
+// get returns the int32 associated with the file index and line of p.
+func (m *xposmap) get(p src.XPos) int32 {
+	s := m.mapFor(p.FileIndex())
+	if s == nil {
+		return -1
+	}
+	return s.get(p.Line())
+}
+
+// add adds p to m, treating m as a set instead of as a map.
+// If p does not fall within the set of fileindex->lineRange used to construct m, this will panic.
+// Use clear() in between set/map interpretations of m.
+func (m *xposmap) add(p src.XPos) {
+	m.set(p, 0)
+}
+
+// contains returns whether the file index and line of p are in m,
+// treating m as a set instead of as a map.
+func (m *xposmap) contains(p src.XPos) bool {
+	s := m.mapFor(p.FileIndex())
+	if s == nil {
+		return false
+	}
+	return s.contains(p.Line())
+}
+
+// remove removes the file index and line for p from m,
+// whether m is currently treated as a map or set.
+func (m *xposmap) remove(p src.XPos) {
+	s := m.mapFor(p.FileIndex())
+	if s == nil {
+		return
+	}
+	s.remove(p.Line())
+}
+
+// foreachEntry applies f to each (fileindex, line, value) triple in m.
+func (m *xposmap) foreachEntry(f func(j int32, l uint, v int32)) {
+	for j, mm := range m.maps {
+		s := mm.size()
+		for i := 0; i < s; i++ {
+			l, v := mm.getEntry(i)
+			f(j, l, v)
+		}
+	}
+}

src/cmd/internal/src/xpos.go

@@ -76,6 +76,7 @@ func (p XPos) WithXlogue(x PosXlogue) XPos {
 	return p
 }
 
+// LineNumber returns a string for the line number, "?" if it is not known.
 func (p XPos) LineNumber() string {
 	if !p.IsKnown() {
 		return "?"
@@ -83,6 +84,13 @@ func (p XPos) LineNumber() string {
 	return p.lico.lineNumber()
 }
 
+// FileIndex returns a smallish non-negative integer corresponding to the
+// file for this source position.  Smallish is relative; it can be thousands
+// large, but not millions.
+func (p XPos) FileIndex() int32 {
+	return p.index
+}
+
 func (p XPos) LineNumberHTML() string {
 	if !p.IsKnown() {
 		return "?"