cmd/compile: add debugging and stats output to numberlines

This is useful for debugging line number assignment and
also for making sense of pathological line number inputs.

Activated with
-gcflags=-d=ssa/number_lines/stats=1 (the bit matters)
-gcflags=-d=ssa/number_lines/debug

Stats:
"SUM_LINE_RANGE",
   SUM for f in files {MAX line in f {line}-MIN line in f {line}}
"MAXMIN_LINE_RANGE",
   MAX for f in files {MAX line in f {line}} -
   MIN for f in files {MIN line in f {line}}
"MAXFILE", maxfile,
   MAX for f in files {f}
"NFILES", len(entries)
   | files |

Change-Id: I8a7336e6370452fe2e3a62de17606db9bd6a6fd3
Reviewed-on: https://go-review.googlesource.com/c/go/+/174947
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/numberlines.go

@@ -7,6 +7,8 @@ package ssa
 import (
 	"cmd/internal/obj"
 	"cmd/internal/src"
+	"fmt"
+	"sort"
 )
 
 func isPoorStatementOp(op Op) bool {
@@ -82,6 +84,32 @@ func (b *Block) FirstPossibleStmtValue() *Value {
 	return nil
 }
 
+func flc(p src.XPos) string {
+	if p == src.NoXPos {
+		return "none"
+	}
+	return fmt.Sprintf("(%d):%d:%d", p.FileIndex(), p.Line(), p.Col())
+}
+
+type fileAndPair struct {
+	f  int32
+	lp lineRange
+}
+
+type fileAndPairs []fileAndPair
+
+func (fap fileAndPairs) Len() int {
+	return len(fap)
+}
+func (fap fileAndPairs) Less(i, j int) bool {
+	return fap[i].f < fap[j].f
+}
+func (fap fileAndPairs) Swap(i, j int) {
+	fap[i], fap[j] = fap[j], fap[i]
+}
+
+// -d=ssa/number_lines/stats=1 (that bit) for line and file distribution statistics
+// -d=ssa/number_lines/debug for information about why particular values are marked as statements.
 func numberLines(f *Func) {
 	po := f.Postorder()
 	endlines := make(map[ID]src.XPos)
@@ -131,6 +159,9 @@ func numberLines(f *Func) {
 			if b.Pos.IsStmt() != src.PosNotStmt {
 				b.Pos = b.Pos.WithIsStmt()
 				endlines[b.ID] = b.Pos
+				if f.pass.debug > 0 {
+					fmt.Printf("Mark stmt effectively-empty-block %s %s %s\n", f.Name, b, flc(b.Pos))
+				}
 				continue
 			}
 			line := src.NoXPos
@@ -153,11 +184,18 @@ func numberLines(f *Func) {
 		// check predecessors for any difference; if firstPos differs, then it is a boundary.
 		if len(b.Preds) == 0 { // Don't forget the entry block
 			b.Values[firstPosIndex].Pos = firstPos.WithIsStmt()
-		} else {
+			if f.pass.debug > 0 {
+				fmt.Printf("Mark stmt entry-block %s %s %s %s\n", f.Name, b, b.Values[firstPosIndex], flc(firstPos))
+			}
+		} else { // differing pred
 			for _, p := range b.Preds {
 				pbi := p.Block().ID
 				if endlines[pbi].Line() != firstPos.Line() || !endlines[pbi].SameFile(firstPos) {
 					b.Values[firstPosIndex].Pos = firstPos.WithIsStmt()
+					if f.pass.debug > 0 {
+						fmt.Printf("Mark stmt differing-pred %s %s %s %s, different=%s ending %s\n",
+							f.Name, b, b.Values[firstPosIndex], flc(firstPos), p.Block(), flc(endlines[pbi]))
+					}
 					break
 				}
 			}
@@ -173,6 +211,9 @@ func numberLines(f *Func) {
 			i = nextGoodStatementIndex(v, i, b)
 			v = b.Values[i]
 			if v.Pos.Line() != firstPos.Line() || !v.Pos.SameFile(firstPos) {
+				if f.pass.debug > 0 {
+					fmt.Printf("Mark stmt new line %s %s %s %s prev pos = %s\n", f.Name, b, v, flc(v.Pos), flc(firstPos))
+				}
 				firstPos = v.Pos
 				v.Pos = v.Pos.WithIsStmt()
 			} else {
@@ -180,11 +221,43 @@ func numberLines(f *Func) {
 			}
 		}
 		if b.Pos.IsStmt() != src.PosNotStmt && (b.Pos.Line() != firstPos.Line() || !b.Pos.SameFile(firstPos)) {
+			if f.pass.debug > 0 {
+				fmt.Printf("Mark stmt end of block differs %s %s %s prev pos = %s\n", f.Name, b, flc(b.Pos), flc(firstPos))
+			}
 			b.Pos = b.Pos.WithIsStmt()
 			firstPos = b.Pos
 		}
 		endlines[b.ID] = firstPos
 	}
+	if f.pass.stats&1 != 0 {
+		// Report summary statistics on the shape of the sparse map about to be constructed
+		// TODO use this information to make sparse maps faster.
+		var entries fileAndPairs
+		for k, v := range ranges {
+			entries = append(entries, fileAndPair{int32(k), v})
+		}
+		sort.Sort(entries)
+		total := uint64(0)            // sum over files of maxline(file) - minline(file)
+		maxfile := int32(0)           // max(file indices)
+		minline := uint32(0xffffffff) // min over files of minline(file)
+		maxline := uint32(0)          // max over files of maxline(file)
+		for _, v := range entries {
+			if f.pass.stats > 1 {
+				f.LogStat("file", v.f, "low", v.lp.first, "high", v.lp.last)
+			}
+			total += uint64(v.lp.last - v.lp.first)
+			if maxfile < v.f {
+				maxfile = v.f
+			}
+			if minline > v.lp.first {
+				minline = v.lp.first
+			}
+			if maxline < v.lp.last {
+				maxline = v.lp.last
+			}
+		}
+		f.LogStat("SUM_LINE_RANGE", total, "MAXMIN_LINE_RANGE", maxline-minline, "MAXFILE", maxfile, "NFILES", len(entries))
+	}
 	// cachedLineStarts is an empty sparse map for values that are included within ranges.
 	f.cachedLineStarts = newXposmap(ranges)
 }