Skip to content

G
go
Commit fe5be5ac authored by Shahar Kohanim's avatar Shahar Kohanim Committed by Brad Fitzpatrick
Browse files
Options

cmd/link: more idiomatic object reader 

name       old secs    new secs    delta
LinkCmdGo   0.52 ± 3%   0.52 ± 7%    ~     (p=0.325 n=93+100)

name       old MaxRSS  new MaxRSS  delta
LinkCmdGo   120k ± 1%   118k ± 4%  -1.10%   (p=0.000 n=87+96)

Change-Id: I967660b8dc6036d28eeea1b6b30f400fadd57b05
Reviewed-on: https://go-review.googlesource.com/21372
Run-TryBot: Shahar Kohanim <skohanim@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: default avatarBrad Fitzpatrick <bradfitz@golang.org>
parent 7a8caf7d
Show whitespace changes
Inline Side-by-side
Showing with 194 additions and 156 deletions
src/cmd/link/internal/ld/lib.go
View file @ fe5be5ac
...@@ -1347,7 +1347,7 @@ func ldobj(f *obj.Biobuf, pkg string, length int64, pn string, file string, when ...@@ -1347,7 +1347,7 @@ func ldobj(f *obj.Biobuf, pkg string, length int64, pn string, file string, when
ldpkg(f, pkg, import1-import0-2, pn, whence) // -2 for !\n ldpkg(f, pkg, import1-import0-2, pn, whence) // -2 for !\n
obj.Bseek(f, import1, 0) obj.Bseek(f, import1, 0)
ldobjfile(Ctxt, f, pkg, eof-obj.Boffset(f), pn) LoadObjFile(Ctxt, f, pkg, eof-obj.Boffset(f), pn)
return nil return nil
} }
......
src/cmd/link/internal/ld/link.go
View file @ fe5be5ac
...@@ -189,7 +189,6 @@ type Link struct { ...@@ -189,7 +189,6 @@ type Link struct {
Filesyms *LSym Filesyms *LSym
Moduledata *LSym Moduledata *LSym
LSymBatch []LSym LSymBatch []LSym
CurRefs []*LSym // List of symbol references for the file being read.
} }
// The smallest possible offset from the hardware stack pointer to a local // The smallest possible offset from the hardware stack pointer to a local
......
src/cmd/link/internal/ld/objfile.go
View file @ fe5be5ac
...@@ -108,8 +108,10 @@ package ld ...@@ -108,8 +108,10 @@ package ld
// - There are SymID in the object file that should really just be strings. // - There are SymID in the object file that should really just be strings.
import ( import (
"bufio"
"bytes" "bytes"
"cmd/internal/obj" "cmd/internal/obj"
"io"
"log" "log"
"strconv" "strconv"
"strings" "strings"
...@@ -120,111 +122,143 @@ const ( ...@@ -120,111 +122,143 @@ const (
endmagic = "\xff\xffgo13ld" endmagic = "\xff\xffgo13ld"
) )
func ldobjfile(ctxt *Link, f *obj.Biobuf, pkg string, length int64, pn string) { var emptyPkg = []byte(`"".`)
// objReader reads Go object files.
type objReader struct {
rd *bufio.Reader
ctxt *Link
pkg string
pn string
// List of symbol references for the file being read.
dupSym *LSym
// rdBuf is used by readString and readSymName as scratch for reading strings.
rdBuf []byte
refs []*LSym
data []byte
reloc []Reloc
pcdata []Pcdata
autom []Auto
funcdata []*LSym
funcdataoff []int64
file []*LSym
}
func LoadObjFile(ctxt *Link, f *obj.Biobuf, pkg string, length int64, pn string) {
start := obj.Boffset(f) start := obj.Boffset(f)
ctxt.IncVersion() r := &objReader{
rd: f.Reader(),
pkg: pkg,
ctxt: ctxt,
pn: pn,
dupSym: &LSym{Name: ".dup"},
}
r.loadObjFile()
if obj.Boffset(f) != start+length {
log.Fatalf("%s: unexpected end at %d, want %d", pn, int64(obj.Boffset(f)), int64(start+length))
}
}
func (r *objReader) loadObjFile() {
// Increment context version, versions are used to differentiate static files in different packages
r.ctxt.IncVersion()
// Magic header
var buf [8]uint8 var buf [8]uint8
obj.Bread(f, buf[:]) r.readFull(buf[:])
if string(buf[:]) != startmagic { if string(buf[:]) != startmagic {
log.Fatalf("%s: invalid file start %x %x %x %x %x %x %x %x", pn, buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6], buf[7]) log.Fatalf("%s: invalid file start %x %x %x %x %x %x %x %x", r.pn, buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6], buf[7])
} }
c := obj.Bgetc(f)
if c != 1 { // Version
log.Fatalf("%s: invalid file version number %d", pn, c) c, err := r.rd.ReadByte()
if err != nil || c != 1 {
log.Fatalf("%s: invalid file version number %d", r.pn, c)
} }
var lib string // Autolib
for { for {
lib = rdstring(f) lib := r.readString()
if lib == "" { if lib == "" {
break break
} }
addlib(ctxt, pkg, pn, lib) addlib(r.ctxt, r.pkg, r.pn, lib)
} }
ctxt.CurRefs = []*LSym{nil} // zeroth ref is nil // Symbol references
r.refs = []*LSym{nil} // zeroth ref is nil
for { for {
c, err := f.Peek(1) c, err := r.rd.Peek(1)
if err != nil { if err != nil {
log.Fatalf("%s: peeking: %v", pn, err) log.Fatalf("%s: peeking: %v", r.pn, err)
} }
if c[0] == 0xff { if c[0] == 0xff {
obj.Bgetc(f) r.rd.ReadByte()
break break
} }
readref(ctxt, f, pkg, pn) r.readRef()
} }
sl := rdslices(f) // Lengths
r.readSlices()
obj.Bread(f, sl.data) // Data section
r.readFull(r.data)
// Defined symbols
for { for {
c, err := f.Peek(1) c, err := r.rd.Peek(1)
if err != nil { if err != nil {
log.Fatalf("%s: peeking: %v", pn, err) log.Fatalf("%s: peeking: %v", r.pn, err)
} }
if c[0] == 0xff { if c[0] == 0xff {
break break
} }
readsym(ctxt, f, sl, pkg, pn) r.readSym()
} }
// Magic footer
buf = [8]uint8{} buf = [8]uint8{}
obj.Bread(f, buf[:]) r.readFull(buf[:])
if string(buf[:]) != endmagic { if string(buf[:]) != endmagic {
log.Fatalf("%s: invalid file end", pn) log.Fatalf("%s: invalid file end", r.pn)
}
if obj.Boffset(f) != start+length {
log.Fatalf("%s: unexpected end at %d, want %d", pn, int64(obj.Boffset(f)), int64(start+length))
} }
} }
var dupSym = &LSym{Name: ".dup"} func (r *objReader) readSlices() {
n := r.readInt()
type slices struct { r.data = make([]byte, n)
data []byte n = r.readInt()
reloc []Reloc r.reloc = make([]Reloc, n)
pcdata []Pcdata n = r.readInt()
autom []Auto r.pcdata = make([]Pcdata, n)
funcdata []*LSym n = r.readInt()
funcdataoff []int64 r.autom = make([]Auto, n)
file []*LSym n = r.readInt()
r.funcdata = make([]*LSym, n)
r.funcdataoff = make([]int64, n)
n = r.readInt()
r.file = make([]*LSym, n)
} }
func rdslices(f *obj.Biobuf) *slices { // Symbols are prefixed so their content doesn't get confused with the magic footer.
sl := &slices{} const symPrefix = 0xfe
n := rdint(f)
sl.data = make([]byte, n)
n = rdint(f)
sl.reloc = make([]Reloc, n)
n = rdint(f)
sl.pcdata = make([]Pcdata, n)
n = rdint(f)
sl.autom = make([]Auto, n)
n = rdint(f)
sl.funcdata = make([]*LSym, n)
sl.funcdataoff = make([]int64, n)
n = rdint(f)
sl.file = make([]*LSym, n)
return sl
}
func readsym(ctxt *Link, f *obj.Biobuf, sl *slices, pkg string, pn string) { func (r *objReader) readSym() {
if obj.Bgetc(f) != 0xfe { if c, err := r.rd.ReadByte(); c != symPrefix || err != nil {
log.Fatalln("readsym out of sync") log.Fatalln("readSym out of sync")
} }
t := rdint(f) t := r.readInt()
s := rdsym(ctxt, f, pkg) s := r.readSymIndex()
flags := rdint(f) flags := r.readInt()
dupok := flags&1 != 0 dupok := flags&1 != 0
local := flags&2 != 0 local := flags&2 != 0
size := rdint(f) size := r.readInt()
typ := rdsym(ctxt, f, pkg) typ := r.readSymIndex()
data := rddata(f, &sl.data) data := r.readData()
nreloc := rdint(f) nreloc := r.readInt()
isdup := false isdup := false
var dup *LSym var dup *LSym
...@@ -243,17 +277,17 @@ func readsym(ctxt *Link, f *obj.Biobuf, sl *slices, pkg string, pn string) { ...@@ -243,17 +277,17 @@ func readsym(ctxt *Link, f *obj.Biobuf, sl *slices, pkg string, pn string) {
goto overwrite goto overwrite
} }
if s.Type != obj.SBSS && s.Type != obj.SNOPTRBSS && !dupok && !s.Attr.DuplicateOK() { if s.Type != obj.SBSS && s.Type != obj.SNOPTRBSS && !dupok && !s.Attr.DuplicateOK() {
log.Fatalf("duplicate symbol %s (types %d and %d) in %s and %s", s.Name, s.Type, t, s.File, pn) log.Fatalf("duplicate symbol %s (types %d and %d) in %s and %s", s.Name, s.Type, t, s.File, r.pn)
} }
if len(s.P) > 0 { if len(s.P) > 0 {
dup = s dup = s
s = dupSym s = r.dupSym
isdup = true isdup = true
} }
} }
overwrite: overwrite:
s.File = pkg s.File = r.pkg
if dupok { if dupok {
s.Attr |= AttrDuplicateOK s.Attr |= AttrDuplicateOK
} }
...@@ -261,7 +295,7 @@ overwrite: ...@@ -261,7 +295,7 @@ overwrite:
log.Fatalf("bad sxref") log.Fatalf("bad sxref")
} }
if t == 0 { if t == 0 {
log.Fatalf("missing type for %s in %s", s.Name, pn) log.Fatalf("missing type for %s in %s", s.Name, r.pn)
} }
if t == obj.SBSS && (s.Type == obj.SRODATA || s.Type == obj.SNOPTRBSS) { if t == obj.SBSS && (s.Type == obj.SRODATA || s.Type == obj.SNOPTRBSS) {
t = int(s.Type) t = int(s.Type)
...@@ -279,80 +313,80 @@ overwrite: ...@@ -279,80 +313,80 @@ overwrite:
} }
s.P = data s.P = data
if nreloc > 0 { if nreloc > 0 {
s.R = sl.reloc[:nreloc:nreloc] s.R = r.reloc[:nreloc:nreloc]
if !isdup { if !isdup {
sl.reloc = sl.reloc[nreloc:] r.reloc = r.reloc[nreloc:]
} }
var r *Reloc
for i := 0; i < nreloc; i++ { for i := 0; i < nreloc; i++ {
r = &s.R[i] s.R[i] = Reloc{
r.Off = rdint32(f) Off: r.readInt32(),
r.Siz = rduint8(f) Siz: r.readUint8(),
r.Type = rdint32(f) Type: r.readInt32(),
r.Add = rdint64(f) Add: r.readInt64(),
r.Sym = rdsym(ctxt, f, pkg) Sym: r.readSymIndex(),
}
} }
} }
if s.Type == obj.STEXT { if s.Type == obj.STEXT {
s.Args = rdint32(f) s.Args = r.readInt32()
s.Locals = rdint32(f) s.Locals = r.readInt32()
if rduint8(f) != 0 { if r.readUint8() != 0 {
s.Attr |= AttrNoSplit s.Attr |= AttrNoSplit
} }
flags := rdint(f) flags := r.readInt()
if flags&(1<<2) != 0 { if flags&(1<<2) != 0 {
s.Attr |= AttrReflectMethod s.Attr |= AttrReflectMethod
} }
n := rdint(f) n := r.readInt()
s.Autom = sl.autom[:n:n] s.Autom = r.autom[:n:n]
if !isdup { if !isdup {
sl.autom = sl.autom[n:] r.autom = r.autom[n:]
} }
for i := 0; i < n; i++ { for i := 0; i < n; i++ {
s.Autom[i] = Auto{ s.Autom[i] = Auto{
Asym: rdsym(ctxt, f, pkg), Asym: r.readSymIndex(),
Aoffset: rdint32(f), Aoffset: r.readInt32(),
Name: rdint16(f), Name: r.readInt16(),
Gotype: rdsym(ctxt, f, pkg), Gotype: r.readSymIndex(),
} }
} }
s.Pcln = new(Pcln) s.Pcln = new(Pcln)
pc := s.Pcln pc := s.Pcln
pc.Pcsp.P = rddata(f, &sl.data) pc.Pcsp.P = r.readData()
pc.Pcfile.P = rddata(f, &sl.data) pc.Pcfile.P = r.readData()
pc.Pcline.P = rddata(f, &sl.data) pc.Pcline.P = r.readData()
n = rdint(f) n = r.readInt()
pc.Pcdata = sl.pcdata[:n:n] pc.Pcdata = r.pcdata[:n:n]
if !isdup { if !isdup {
sl.pcdata = sl.pcdata[n:] r.pcdata = r.pcdata[n:]
} }
for i := 0; i < n; i++ { for i := 0; i < n; i++ {
pc.Pcdata[i].P = rddata(f, &sl.data) pc.Pcdata[i].P = r.readData()
} }
n = rdint(f) n = r.readInt()
pc.Funcdata = sl.funcdata[:n:n] pc.Funcdata = r.funcdata[:n:n]
pc.Funcdataoff = sl.funcdataoff[:n:n] pc.Funcdataoff = r.funcdataoff[:n:n]
if !isdup { if !isdup {
sl.funcdata = sl.funcdata[n:] r.funcdata = r.funcdata[n:]
sl.funcdataoff = sl.funcdataoff[n:] r.funcdataoff = r.funcdataoff[n:]
} }
for i := 0; i < n; i++ { for i := 0; i < n; i++ {
pc.Funcdata[i] = rdsym(ctxt, f, pkg) pc.Funcdata[i] = r.readSymIndex()
} }
for i := 0; i < n; i++ { for i := 0; i < n; i++ {
pc.Funcdataoff[i] = rdint64(f) pc.Funcdataoff[i] = r.readInt64()
} }
n = rdint(f) n = r.readInt()
pc.File = sl.file[:n:n] pc.File = r.file[:n:n]
if !isdup { if !isdup {
sl.file = sl.file[n:] r.file = r.file[n:]
} }
for i := 0; i < n; i++ { for i := 0; i < n; i++ {
pc.File[i] = rdsym(ctxt, f, pkg) pc.File[i] = r.readSymIndex()
} }
if !isdup { if !isdup {
...@@ -360,30 +394,37 @@ overwrite: ...@@ -360,30 +394,37 @@ overwrite:
log.Fatalf("symbol %s listed multiple times", s.Name) log.Fatalf("symbol %s listed multiple times", s.Name)
} }
s.Attr |= AttrOnList s.Attr |= AttrOnList
if ctxt.Etextp != nil { if r.ctxt.Etextp != nil {
ctxt.Etextp.Next = s r.ctxt.Etextp.Next = s
} else { } else {
ctxt.Textp = s r.ctxt.Textp = s
} }
ctxt.Etextp = s r.ctxt.Etextp = s
} }
} }
} }
func readref(ctxt *Link, f *obj.Biobuf, pkg string, pn string) { func (r *objReader) readFull(b []byte) {
if obj.Bgetc(f) != 0xfe { _, err := io.ReadFull(r.rd, b)
log.Fatalf("readsym out of sync") if err != nil {
log.Fatalf("%s: error reading %s", r.pn, err)
}
}
func (r *objReader) readRef() {
if c, err := r.rd.ReadByte(); c != symPrefix || err != nil {
log.Fatalf("readSym out of sync")
} }
name := rdsymName(f, pkg) name := r.readSymName()
v := rdint(f) v := r.readInt()
if v != 0 && v != 1 { if v != 0 && v != 1 {
log.Fatalf("invalid symbol version %d", v) log.Fatalf("invalid symbol version %d", v)
} }
if v == 1 { if v == 1 {
v = ctxt.Version v = r.ctxt.Version
} }
s := Linklookup(ctxt, name, v) s := Linklookup(r.ctxt, name, v)
ctxt.CurRefs = append(ctxt.CurRefs, s) r.refs = append(r.refs, s)
if s == nil || v != 0 { if s == nil || v != 0 {
return return
...@@ -400,9 +441,9 @@ func readref(ctxt *Link, f *obj.Biobuf, pkg string, pn string) { ...@@ -400,9 +441,9 @@ func readref(ctxt *Link, f *obj.Biobuf, pkg string, pn string) {
if uint64(uint32(x)) != x { if uint64(uint32(x)) != x {
log.Panicf("$-symbol %s too large: %d", s.Name, x) log.Panicf("$-symbol %s too large: %d", s.Name, x)
} }
Adduint32(ctxt, s, uint32(x)) Adduint32(r.ctxt, s, uint32(x))
case "$f64.", "$i64.": case "$f64.", "$i64.":
Adduint64(ctxt, s, x) Adduint64(r.ctxt, s, x)
default: default:
log.Panicf("unrecognized $-symbol: %s", s.Name) log.Panicf("unrecognized $-symbol: %s", s.Name)
} }
...@@ -413,14 +454,13 @@ func readref(ctxt *Link, f *obj.Biobuf, pkg string, pn string) { ...@@ -413,14 +454,13 @@ func readref(ctxt *Link, f *obj.Biobuf, pkg string, pn string) {
} }
} }
func rdint64(f *obj.Biobuf) int64 { func (r *objReader) readInt64() int64 {
r := f.Reader()
uv := uint64(0) uv := uint64(0)
for shift := uint(0); ; shift += 7 { for shift := uint(0); ; shift += 7 {
if shift >= 64 { if shift >= 64 {
log.Fatalf("corrupt input") log.Fatalf("corrupt input")
} }
c, err := r.ReadByte() c, err := r.rd.ReadByte()
if err != nil { if err != nil {
log.Fatalln("error reading input: ", err) log.Fatalln("error reading input: ", err)
} }
...@@ -433,63 +473,61 @@ func rdint64(f *obj.Biobuf) int64 { ...@@ -433,63 +473,61 @@ func rdint64(f *obj.Biobuf) int64 {
return int64(uv>>1) ^ (int64(uint64(uv)<<63) >> 63) return int64(uv>>1) ^ (int64(uint64(uv)<<63) >> 63)
} }
func rdint(f *obj.Biobuf) int { func (r *objReader) readInt() int {
n := rdint64(f) n := r.readInt64()
if int64(int(n)) != n { if int64(int(n)) != n {
log.Panicf("%v out of range for int", n) log.Panicf("%v out of range for int", n)
} }
return int(n) return int(n)
} }
func rdint32(f *obj.Biobuf) int32 { func (r *objReader) readInt32() int32 {
n := rdint64(f) n := r.readInt64()
if int64(int32(n)) != n { if int64(int32(n)) != n {
log.Panicf("%v out of range for int32", n) log.Panicf("%v out of range for int32", n)
} }
return int32(n) return int32(n)
} }
func rdint16(f *obj.Biobuf) int16 { func (r *objReader) readInt16() int16 {
n := rdint64(f) n := r.readInt64()
if int64(int16(n)) != n { if int64(int16(n)) != n {
log.Panicf("%v out of range for int16", n) log.Panicf("%v out of range for int16", n)
} }
return int16(n) return int16(n)
} }
func rduint8(f *obj.Biobuf) uint8 { func (r *objReader) readUint8() uint8 {
n := rdint64(f) n := r.readInt64()
if int64(uint8(n)) != n { if int64(uint8(n)) != n {
log.Panicf("%v out of range for uint8", n) log.Panicf("%v out of range for uint8", n)
} }
return uint8(n) return uint8(n)
} }
// rdBuf is used by rdstring and rdsymName as scratch for reading strings. func (r *objReader) readString() string {
var rdBuf []byte n := r.readInt()
var emptyPkg = []byte(`"".`) if len(r.rdBuf) < n {
r.rdBuf = make([]byte, n)
func rdstring(f *obj.Biobuf) string {
n := rdint(f)
if len(rdBuf) < n {
rdBuf = make([]byte, n)
} }
obj.Bread(f, rdBuf[:n]) r.readFull(r.rdBuf[:n])
return string(rdBuf[:n]) return string(r.rdBuf[:n])
} }
func rddata(f *obj.Biobuf, buf *[]byte) []byte { func (r *objReader) readData() []byte {
n := rdint(f) n := r.readInt()
p := (*buf)[:n:n] p := r.data[:n:n]
*buf = (*buf)[n:] r.data = r.data[n:]
return p return p
} }
// rdsymName reads a symbol name, replacing all "". with pkg. // readSymName reads a symbol name, replacing all "". with pkg.
func rdsymName(f *obj.Biobuf, pkg string) string { func (r *objReader) readSymName() string {
n := rdint(f) rdBuf := r.rdBuf
pkg := r.pkg
n := r.readInt()
if n == 0 { if n == 0 {
rdint64(f) r.readInt64()
return "" return ""
} }
...@@ -497,7 +535,7 @@ func rdsymName(f *obj.Biobuf, pkg string) string { ...@@ -497,7 +535,7 @@ func rdsymName(f *obj.Biobuf, pkg string) string {
rdBuf = make([]byte, n, 2*n) rdBuf = make([]byte, n, 2*n)
} }
origName := rdBuf[:n] origName := rdBuf[:n]
obj.Bread(f, origName) r.readFull(origName)
adjName := rdBuf[n:n] adjName := rdBuf[n:n]
for { for {
i := bytes.Index(origName, emptyPkg) i := bytes.Index(origName, emptyPkg)
...@@ -512,12 +550,13 @@ func rdsymName(f *obj.Biobuf, pkg string) string { ...@@ -512,12 +550,13 @@ func rdsymName(f *obj.Biobuf, pkg string) string {
} }
name := string(adjName) name := string(adjName)
if len(adjName) > len(rdBuf) { if len(adjName) > len(rdBuf) {
rdBuf = adjName // save the larger buffer for reuse r.rdBuf = adjName // save the larger buffer for reuse
} }
return name return name
} }
func rdsym(ctxt *Link, f *obj.Biobuf, pkg string) *LSym { // Reads the index of a symbol reference and resolves it to a symbol
i := rdint(f) func (r *objReader) readSymIndex() *LSym {
return ctxt.CurRefs[i] i := r.readInt()
return r.refs[i]
} }
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7