cmd/compile: walk progs to generate debug_lines data

Walking the progs is simpler than using the is_stmt symbol shenanigans.
This is a reinstatement of CL 196661, which was rolled back due to tests
failing. Unlike that original CL, this change should output the same
debug_lines data the original approach wrote.

The stats for JUST this CLC, note teh small speedup in compilation, and
the lack of difference in binary size.

name                      old time/op       new time/op       delta
Template                        229ms ± 4%        218ms ± 1%  -4.95%  (p=0.000 n=10+8)
Unicode                        92.6ms ± 9%       88.6ms ±13%    ~     (p=0.089 n=10+10)
GoTypes                         850ms ± 2%        831ms ± 4%  -2.23%  (p=0.009 n=10+10)
Compiler                        3.99s ± 1%        3.93s ± 1%  -1.29%  (p=0.000 n=10+9)
SSA                             13.7s ± 1%        13.7s ± 1%    ~     (p=0.912 n=10+10)
Flate                           140ms ± 3%        138ms ± 3%  -1.90%  (p=0.009 n=10+10)
GoParser                        172ms ± 2%        169ms ± 4%    ~     (p=0.095 n=9+10)
Reflect                         530ms ± 3%        516ms ± 5%    ~     (p=0.052 n=10+10)
Tar                             202ms ± 1%        196ms ± 3%  -2.83%  (p=0.002 n=9+10)
XML                             280ms ± 3%        270ms ± 4%  -3.48%  (p=0.009 n=10+10)
LinkCompiler                    927ms ± 2%        907ms ± 4%    ~     (p=0.052 n=10+10)
ExternalLinkCompiler            1.97s ± 2%        1.97s ± 3%    ~     (p=0.853 n=10+10)
LinkWithoutDebugCompiler        549ms ± 3%        543ms ± 5%    ~     (p=0.481 n=10+10)
StdCmd                          12.0s ± 1%        12.0s ± 1%    ~     (p=0.905 n=9+10)

name                      old user-time/op  new user-time/op  delta
Template                        372ms ±18%        344ms ±11%    ~     (p=0.190 n=10+10)
Unicode                         264ms ±23%        241ms ±43%    ~     (p=0.315 n=8+10)
GoTypes                         1.56s ±22%        1.68s ± 5%    ~     (p=0.237 n=10+8)
Compiler                        7.41s ± 2%        7.31s ± 3%    ~     (p=0.123 n=10+10)
SSA                             24.5s ± 2%        24.7s ± 1%    ~     (p=0.133 n=10+9)
Flate                           199ms ± 6%        188ms ±28%    ~     (p=0.353 n=10+10)
GoParser                        243ms ±11%        240ms ± 6%    ~     (p=0.968 n=10+9)
Reflect                         929ms ±21%        862ms ±35%    ~     (p=0.190 n=10+10)
Tar                             284ms ± 9%        296ms ±17%    ~     (p=0.497 n=9+10)
XML                             386ms ±21%        398ms ±28%    ~     (p=1.000 n=9+10)
LinkCompiler                    1.13s ± 9%        1.12s ± 8%    ~     (p=0.546 n=9+9)
ExternalLinkCompiler            2.37s ±15%        2.30s ± 9%    ~     (p=0.549 n=10+9)
LinkWithoutDebugCompiler        646ms ±10%        642ms ±13%    ~     (p=0.853 n=10+10)

name                      old alloc/op      new alloc/op      delta
Template                       36.5MB ± 0%       36.5MB ± 0%  -0.11%  (p=0.000 n=10+9)
Unicode                        28.5MB ± 0%       28.5MB ± 0%    ~     (p=0.190 n=10+10)
GoTypes                         121MB ± 0%        121MB ± 0%  -0.10%  (p=0.000 n=9+10)
Compiler                        549MB ± 0%        549MB ± 0%  -0.10%  (p=0.000 n=9+10)
SSA                            1.92GB ± 0%       1.92GB ± 0%  -0.13%  (p=0.000 n=10+10)
Flate                          23.0MB ± 0%       23.0MB ± 0%  -0.07%  (p=0.000 n=10+10)
GoParser                       27.9MB ± 0%       27.9MB ± 0%  -0.09%  (p=0.000 n=10+10)
Reflect                        77.9MB ± 0%       77.8MB ± 0%  -0.13%  (p=0.000 n=9+10)
Tar                            34.5MB ± 0%       34.4MB ± 0%  -0.09%  (p=0.000 n=10+10)
XML                            44.3MB ± 0%       44.3MB ± 0%  -0.08%  (p=0.000 n=10+10)
LinkCompiler                    229MB ± 0%        225MB ± 0%  -1.74%  (p=0.000 n=10+10)
ExternalLinkCompiler            233MB ± 0%        242MB ± 0%  +3.81%  (p=0.000 n=10+10)
LinkWithoutDebugCompiler        156MB ± 0%        152MB ± 0%  -2.29%  (p=0.000 n=10+9)

name                      old allocs/op     new allocs/op     delta
Template                         373k ± 0%         373k ± 0%  -0.21%  (p=0.000 n=10+10)
Unicode                          340k ± 0%         340k ± 0%  -0.04%  (p=0.000 n=10+10)
GoTypes                         1.33M ± 0%        1.33M ± 0%  -0.20%  (p=0.000 n=10+9)
Compiler                        5.39M ± 0%        5.38M ± 0%  -0.16%  (p=0.000 n=10+10)
SSA                             18.3M ± 0%        18.2M ± 0%  -0.15%  (p=0.000 n=10+10)
Flate                            235k ± 0%         234k ± 0%  -0.23%  (p=0.000 n=10+7)
GoParser                         309k ± 0%         308k ± 0%  -0.20%  (p=0.000 n=10+10)
Reflect                          970k ± 0%         968k ± 0%  -0.30%  (p=0.000 n=10+10)
Tar                              347k ± 0%         347k ± 0%  -0.22%  (p=0.000 n=10+10)
XML                              425k ± 0%         424k ± 0%  -0.16%  (p=0.000 n=10+10)
LinkCompiler                     602k ± 0%         601k ± 0%  -0.03%  (p=0.000 n=9+10)
ExternalLinkCompiler            1.65M ± 0%        1.65M ± 0%  -0.02%  (p=0.000 n=10+10)
LinkWithoutDebugCompiler         220k ± 0%         220k ± 0%  -0.03%  (p=0.016 n=10+9)

name                      old object-bytes  new object-bytes  delta
Template                        553kB ± 0%        553kB ± 0%  -0.01%  (p=0.000 n=10+10)
Unicode                         215kB ± 0%        215kB ± 0%    ~     (all equal)
GoTypes                        2.02MB ± 0%       2.02MB ± 0%  -0.00%  (p=0.000 n=10+10)
Compiler                       7.98MB ± 0%       7.98MB ± 0%  -0.01%  (p=0.000 n=10+10)
SSA                            27.1MB ± 0%       27.1MB ± 0%  -0.00%  (p=0.000 n=10+10)
Flate                           340kB ± 0%        340kB ± 0%  -0.01%  (p=0.000 n=10+10)
GoParser                        434kB ± 0%        434kB ± 0%  -0.00%  (p=0.000 n=10+10)
Reflect                        1.34MB ± 0%       1.34MB ± 0%  -0.01%  (p=0.000 n=10+10)
Tar                             479kB ± 0%        479kB ± 0%  -0.00%  (p=0.000 n=10+10)
XML                             618kB ± 0%        618kB ± 0%  -0.01%  (p=0.000 n=10+10)

name                      old export-bytes  new export-bytes  delta
Template                       20.4kB ± 0%       20.4kB ± 0%    ~     (all equal)
Unicode                        8.21kB ± 0%       8.21kB ± 0%    ~     (all equal)
GoTypes                        36.6kB ± 0%       36.6kB ± 0%    ~     (all equal)
Compiler                        116kB ± 0%        116kB ± 0%  +0.00%  (p=0.000 n=10+10)
SSA                             141kB ± 0%        141kB ± 0%  +0.00%  (p=0.000 n=10+10)
Flate                          5.10kB ± 0%       5.10kB ± 0%    ~     (all equal)
GoParser                       8.92kB ± 0%       8.92kB ± 0%    ~     (all equal)
Reflect                        11.8kB ± 0%       11.8kB ± 0%    ~     (all equal)
Tar                            10.9kB ± 0%       10.9kB ± 0%    ~     (all equal)
XML                            17.4kB ± 0%       17.4kB ± 0%    ~     (all equal)

name                      old text-bytes    new text-bytes    delta
HelloSize                       742kB ± 0%        742kB ± 0%    ~     (all equal)
CmdGoSize                      10.6MB ± 0%       10.6MB ± 0%    ~     (all equal)

name                      old data-bytes    new data-bytes    delta
HelloSize                      10.7kB ± 0%       10.7kB ± 0%    ~     (all equal)
CmdGoSize                       312kB ± 0%        312kB ± 0%    ~     (all equal)

name                      old bss-bytes     new bss-bytes     delta
HelloSize                       122kB ± 0%        122kB ± 0%    ~     (all equal)
CmdGoSize                       146kB ± 0%        146kB ± 0%    ~     (all equal)

name                      old exe-bytes     new exe-bytes     delta
HelloSize                      1.10MB ± 0%       1.10MB ± 0%    ~     (all equal)
CmdGoSize                      14.9MB ± 0%       14.9MB ± 0%  -0.03%  (p=0.000 n=10+10)

Change-Id: Ie078a42b29353b96654fa1f0f47d600b5a53762d
Reviewed-on: https://go-review.googlesource.com/c/go/+/200017
Reviewed-by: Jeremy Faller <jeremy@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Run-TryBot: Jeremy Faller <jeremy@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/cmd/internal/obj/dwarf.go

@@ -8,6 +8,7 @@ package obj
 
 import (
 	"cmd/internal/dwarf"
+	"cmd/internal/src"
 	"fmt"
 )
 
@@ -48,78 +49,45 @@ func (ctxt *Link) generateDebugLinesSymbol(s, lines *LSym) {
 	// generating the line table. See below.
 	// TODO: Once delve can support multiple DW_LNS_end_statements, we don't have
 	// to do this.
-	is_stmt := uint8(1)
+	stmt := true
+	line := int64(1)
 	pc := s.Func.Text.Pc
-	line := 1
+	name := ""
-	file := 1
+	prologue, wrotePrologue := false, false
+	// Walk the progs, generating the DWARF table.
+	for p := s.Func.Text; p != nil; p = p.Link {
+		prologue = prologue || (p.Pos.Xlogue() == src.PosPrologueEnd)
+		// If we're not at a real instruction, keep looping!
+		if p.Pos.Line() == 0 || (p.Link != nil && p.Link.Pc == p.Pc) {
+			continue
+		}
+		newStmt := p.Pos.IsStmt() != src.PosNotStmt
+		newName, newLine := linkgetlineFromPos(ctxt, p.Pos)
 
-	// The linker will insert the DW_LNE_set_address once determined; therefore,
+		// Output debug info.
-	// it's omitted here.
+		wrote := false
-
+		if name != newName {
-	// Generate the actual line information.
+			newFile := ctxt.PosTable.FileIndex(newName) + 1 // 1 indexing for the table.
-	// We use the pcline and pcfile to generate this section, and it's suboptimal.
-	// Likely better would be to generate this dirrectly from the progs and not
-	// parse those tables.
-	// TODO: Generate from the progs if it's faster.
-	pcfile := NewPCIter(uint32(ctxt.Arch.Arch.MinLC))
-	pcline := NewPCIter(uint32(ctxt.Arch.Arch.MinLC))
-	pcstmt := NewPCIter(uint32(ctxt.Arch.Arch.MinLC))
-	pcfile.Init(s.Func.Pcln.Pcfile.P)
-	pcline.Init(s.Func.Pcln.Pcline.P)
-	var pctostmtData Pcdata
-	funcpctab(ctxt, &pctostmtData, s, "pctostmt", pctostmt, nil)
-	pcstmt.Init(pctostmtData.P)
-	var thispc uint32
-
-	for !pcfile.Done && !pcline.Done {
-		// Only changed if it advanced
-		if int32(file) != pcfile.Value {
 			dctxt.AddUint8(lines, dwarf.DW_LNS_set_file)
-			dwarf.Uleb128put(dctxt, lines, fileNums[pcfile.Value])
+			dwarf.Uleb128put(dctxt, lines, int64(newFile))
-			file = int(pcfile.Value)
+			name = newName
+			wrote = true
+		}
+		if prologue && !wrotePrologue {
+			dctxt.AddUint8(lines, uint8(dwarf.DW_LNS_set_prologue_end))
+			wrotePrologue = true
+			wrote = true
+		}
+		if stmt != newStmt {
+			dctxt.AddUint8(lines, uint8(dwarf.DW_LNS_negate_stmt))
+			stmt = newStmt
+			wrote = true
 		}
 
-		// Only changed if it advanced
+		if line != int64(newLine) || wrote {
-		if is_stmt != uint8(pcstmt.Value) {
+			pcdelta := p.Pc - pc
-			new_stmt := uint8(pcstmt.Value)
+			putpclcdelta(ctxt, dctxt, lines, uint64(pcdelta), int64(newLine)-line)
-			switch new_stmt &^ 1 {
+			line, pc = int64(newLine), p.Pc
-			case PrologueEnd:
-				dctxt.AddUint8(lines, uint8(dwarf.DW_LNS_set_prologue_end))
-			case EpilogueBegin:
-				// TODO if there is a use for this, add it.
-				// Don't forget to increase OPCODE_BASE by 1 and add entry for standard_opcode_lengths[11]
-				panic("unsupported EpilogueBegin")
-			}
-			new_stmt &= 1
-			if is_stmt != new_stmt {
-				is_stmt = new_stmt
-				dctxt.AddUint8(lines, uint8(dwarf.DW_LNS_negate_stmt))
-			}
-		}
-
-		// putpcldelta makes a row in the DWARF matrix, always, even if line is unchanged.
-		putpclcdelta(ctxt, dctxt, lines, uint64(s.Func.Text.Pc+int64(thispc)-pc), int64(pcline.Value)-int64(line))
-
-		pc = s.Func.Text.Pc + int64(thispc)
-		line = int(pcline.Value)
-
-		// Take the minimum step forward for the three iterators
-		thispc = pcfile.NextPC
-		if pcline.NextPC < thispc {
-			thispc = pcline.NextPC
-		}
-		if !pcstmt.Done && pcstmt.NextPC < thispc {
-			thispc = pcstmt.NextPC
-		}
-
-		if pcfile.NextPC == thispc {
-			pcfile.Next()
-		}
-		if !pcstmt.Done && pcstmt.NextPC == thispc {
-			pcstmt.Next()
-		}
-		if pcline.NextPC == thispc {
-			pcline.Next()
 		}
 	}
 
@@ -129,16 +97,16 @@ func (ctxt *Link) generateDebugLinesSymbol(s, lines *LSym) {
 	//   file = 1
 	//   line = 1
 	//   column = 0
-	//   is_stmt = set in header, we assume true
+	//   stmt = set in header, we assume true
 	//   basic_block = false
 	// Careful readers of the DWARF specification will note that we don't reset
 	// the address of the state machine -- but this will happen at the beginning
-	// of the NEXT block of opcodes. (See the SetAddress call above.)
+	// of the NEXT block of opcodes.
 	dctxt.AddUint8(lines, dwarf.DW_LNS_set_file)
 	dwarf.Uleb128put(dctxt, lines, 1)
 	dctxt.AddUint8(lines, dwarf.DW_LNS_advance_line)
 	dwarf.Sleb128put(dctxt, lines, int64(1-line))
-	if is_stmt != 1 {
+	if !stmt {
 		dctxt.AddUint8(lines, dwarf.DW_LNS_negate_stmt)
 	}
 	dctxt.AddUint8(lines, dwarf.DW_LNS_copy)

src/cmd/internal/obj/pcln.go

@@ -5,16 +5,10 @@
 package obj
 
 import (
-	"cmd/internal/src"
 	"encoding/binary"
 	"log"
 )
 
-const (
-	PrologueEnd   = 2 + iota // overload "is_stmt" to include prologue_end
-	EpilogueBegin            // overload "is_stmt" to include epilogue_end
-)
-
 // funcpctab writes to dst a pc-value table mapping the code in func to the values
 // returned by valfunc parameterized by arg. The invocation of valfunc to update the
 // current value is, for each p,
@@ -249,34 +243,6 @@ func pctospadj(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg in
 	return oldval + p.Spadj
 }
 
-// pctostmt returns either,
-// if phase==0, then whether the current instruction is a step-target (Dwarf is_stmt)
-//     bit-or'd with whether the current statement is a prologue end or epilogue begin
-// else (phase == 1), zero.
-//
-func pctostmt(ctxt *Link, sym *LSym, oldval int32, p *Prog, phase int32, arg interface{}) int32 {
-	if phase == 1 {
-		return 0 // Ignored; also different from initial value of -1, if that ever matters.
-	}
-	s := p.Pos.IsStmt()
-	l := p.Pos.Xlogue()
-
-	var is_stmt int32
-
-	// PrologueEnd, at least, is passed to the next instruction
-	switch l {
-	case src.PosPrologueEnd:
-		is_stmt = PrologueEnd
-	case src.PosEpilogueBegin:
-		is_stmt = EpilogueBegin
-	}
-
-	if s != src.PosNotStmt {
-		is_stmt |= 1 // either PosDefaultStmt from asm, or PosIsStmt from go
-	}
-	return is_stmt
-}
-
 // pctopcdata computes the pcdata value in effect at p.
 // A PCDATA instruction sets the value in effect at future
 // non-PCDATA instructions.