mirror of
https://github.com/golang/go.git
synced 2025-12-08 06:10:04 +00:00
cmd/internal/obj: remove Follow pass
The Follow pass in the assembler backend reorders and copies instructions. This even applies to hand-written assembly code, which in many cases don't want to be reordered. Now that the SSA compiler does a good job for laying out instructions, the benefit of this pass is very little: AMD64: (old = with Follow, new = without Follow) name old time/op new time/op delta BinaryTree17-12 2.78s ± 1% 2.79s ± 1% +0.44% (p=0.000 n=20+19) Fannkuch11-12 3.11s ± 0% 3.31s ± 1% +6.16% (p=0.000 n=19+19) FmtFprintfEmpty-12 50.9ns ± 1% 51.6ns ± 3% +1.40% (p=0.000 n=17+20) FmtFprintfString-12 127ns ± 0% 128ns ± 1% +0.88% (p=0.000 n=17+17) FmtFprintfInt-12 122ns ± 0% 123ns ± 1% +0.76% (p=0.000 n=20+19) FmtFprintfIntInt-12 185ns ± 1% 186ns ± 1% +0.65% (p=0.000 n=20+19) FmtFprintfPrefixedInt-12 192ns ± 1% 202ns ± 1% +4.99% (p=0.000 n=20+19) FmtFprintfFloat-12 284ns ± 0% 288ns ± 0% +1.33% (p=0.000 n=15+19) FmtManyArgs-12 807ns ± 0% 804ns ± 0% -0.44% (p=0.000 n=16+18) GobDecode-12 7.23ms ± 1% 7.21ms ± 1% ~ (p=0.052 n=20+20) GobEncode-12 6.09ms ± 1% 6.12ms ± 1% +0.41% (p=0.002 n=19+19) Gzip-12 253ms ± 1% 255ms ± 1% +0.95% (p=0.000 n=18+20) Gunzip-12 38.4ms ± 0% 38.5ms ± 0% +0.34% (p=0.000 n=17+17) HTTPClientServer-12 95.4µs ± 2% 96.1µs ± 1% +0.78% (p=0.002 n=19+19) JSONEncode-12 16.5ms ± 1% 16.6ms ± 1% +1.17% (p=0.000 n=19+19) JSONDecode-12 54.6ms ± 1% 55.3ms ± 1% +1.23% (p=0.000 n=18+18) Mandelbrot200-12 4.47ms ± 0% 4.47ms ± 0% +0.06% (p=0.000 n=18+18) GoParse-12 3.47ms ± 1% 3.47ms ± 1% ~ (p=0.583 n=20+20) RegexpMatchEasy0_32-12 84.8ns ± 1% 85.2ns ± 2% +0.51% (p=0.022 n=20+20) RegexpMatchEasy0_1K-12 206ns ± 1% 206ns ± 1% ~ (p=0.770 n=20+20) RegexpMatchEasy1_32-12 82.8ns ± 1% 83.4ns ± 1% +0.64% (p=0.000 n=20+19) RegexpMatchEasy1_1K-12 363ns ± 1% 361ns ± 1% -0.48% (p=0.007 n=20+20) RegexpMatchMedium_32-12 126ns ± 1% 126ns ± 0% +0.72% (p=0.000 n=20+20) RegexpMatchMedium_1K-12 39.1µs ± 1% 39.8µs ± 0% +1.73% (p=0.000 n=19+19) RegexpMatchHard_32-12 1.97µs ± 0% 1.98µs ± 1% +0.29% (p=0.005 n=18+20) RegexpMatchHard_1K-12 59.5µs ± 1% 59.8µs ± 1% +0.36% (p=0.000 n=18+20) Revcomp-12 442ms ± 1% 445ms ± 2% +0.67% (p=0.000 n=19+20) Template-12 58.0ms ± 1% 57.5ms ± 1% -0.85% (p=0.000 n=19+19) TimeParse-12 311ns ± 0% 314ns ± 0% +0.94% (p=0.000 n=20+18) TimeFormat-12 350ns ± 3% 346ns ± 0% ~ (p=0.076 n=20+19) [Geo mean] 55.9µs 56.4µs +0.80% ARM32: name old time/op new time/op delta BinaryTree17-4 30.4s ± 0% 30.1s ± 0% -1.14% (p=0.000 n=10+8) Fannkuch11-4 13.7s ± 0% 13.6s ± 0% -0.75% (p=0.000 n=10+10) FmtFprintfEmpty-4 664ns ± 1% 651ns ± 1% -1.96% (p=0.000 n=7+8) FmtFprintfString-4 1.83µs ± 2% 1.77µs ± 2% -3.21% (p=0.000 n=10+10) FmtFprintfInt-4 1.57µs ± 2% 1.54µs ± 2% -2.25% (p=0.007 n=10+10) FmtFprintfIntInt-4 2.37µs ± 2% 2.31µs ± 1% -2.68% (p=0.000 n=10+10) FmtFprintfPrefixedInt-4 2.14µs ± 2% 2.10µs ± 1% -1.83% (p=0.006 n=10+10) FmtFprintfFloat-4 3.69µs ± 2% 3.74µs ± 1% +1.60% (p=0.000 n=10+10) FmtManyArgs-4 9.43µs ± 1% 9.17µs ± 1% -2.70% (p=0.000 n=10+10) GobDecode-4 76.3ms ± 1% 75.5ms ± 1% -1.14% (p=0.003 n=10+10) GobEncode-4 70.7ms ± 2% 69.0ms ± 1% -2.36% (p=0.000 n=10+10) Gzip-4 2.64s ± 1% 2.65s ± 0% +0.59% (p=0.002 n=10+10) Gunzip-4 402ms ± 0% 398ms ± 0% -1.11% (p=0.000 n=10+9) HTTPClientServer-4 458µs ± 0% 457µs ± 0% ~ (p=0.247 n=10+10) JSONEncode-4 171ms ± 0% 172ms ± 0% +0.56% (p=0.000 n=10+10) JSONDecode-4 672ms ± 1% 668ms ± 1% ~ (p=0.105 n=10+10) Mandelbrot200-4 33.5ms ± 0% 33.5ms ± 0% ~ (p=0.156 n=9+10) GoParse-4 33.9ms ± 0% 34.0ms ± 0% +0.36% (p=0.031 n=9+9) RegexpMatchEasy0_32-4 823ns ± 1% 835ns ± 1% +1.49% (p=0.000 n=8+8) RegexpMatchEasy0_1K-4 3.99µs ± 0% 4.02µs ± 1% +0.92% (p=0.000 n=8+10) RegexpMatchEasy1_32-4 877ns ± 3% 904ns ± 2% +3.07% (p=0.012 n=10+10) RegexpMatchEasy1_1K-4 5.99µs ± 0% 5.97µs ± 1% -0.38% (p=0.023 n=8+8) RegexpMatchMedium_32-4 1.40µs ± 2% 1.40µs ± 2% ~ (p=0.590 n=10+9) RegexpMatchMedium_1K-4 357µs ± 0% 355µs ± 1% -0.72% (p=0.000 n=7+8) RegexpMatchHard_32-4 22.3µs ± 0% 22.1µs ± 0% -0.49% (p=0.000 n=8+7) RegexpMatchHard_1K-4 661µs ± 0% 658µs ± 0% -0.42% (p=0.000 n=8+7) Revcomp-4 46.3ms ± 0% 46.3ms ± 0% ~ (p=0.393 n=10+10) Template-4 753ms ± 1% 750ms ± 0% ~ (p=0.211 n=10+9) TimeParse-4 4.28µs ± 1% 4.22µs ± 1% -1.34% (p=0.000 n=8+10) TimeFormat-4 9.00µs ± 0% 9.05µs ± 0% +0.59% (p=0.000 n=10+10) [Geo mean] 538µs 535µs -0.55% ARM64: name old time/op new time/op delta BinaryTree17-8 8.39s ± 0% 8.39s ± 0% ~ (p=0.684 n=10+10) Fannkuch11-8 5.95s ± 0% 5.99s ± 0% +0.63% (p=0.000 n=10+10) FmtFprintfEmpty-8 116ns ± 0% 116ns ± 0% ~ (all equal) FmtFprintfString-8 361ns ± 0% 360ns ± 0% -0.31% (p=0.003 n=8+6) FmtFprintfInt-8 290ns ± 0% 290ns ± 0% ~ (p=0.620 n=9+9) FmtFprintfIntInt-8 476ns ± 1% 469ns ± 0% -1.47% (p=0.000 n=10+6) FmtFprintfPrefixedInt-8 412ns ± 2% 417ns ± 2% +1.39% (p=0.006 n=9+10) FmtFprintfFloat-8 652ns ± 1% 652ns ± 0% ~ (p=0.161 n=10+8) FmtManyArgs-8 1.94µs ± 0% 1.94µs ± 2% ~ (p=0.781 n=10+10) GobDecode-8 17.7ms ± 1% 17.7ms ± 0% ~ (p=0.962 n=10+7) GobEncode-8 15.6ms ± 0% 15.6ms ± 1% ~ (p=0.063 n=10+10) Gzip-8 786ms ± 0% 787ms ± 0% ~ (p=0.356 n=10+9) Gunzip-8 127ms ± 0% 127ms ± 0% +0.08% (p=0.028 n=10+9) HTTPClientServer-8 198µs ± 6% 198µs ± 7% ~ (p=0.796 n=10+10) JSONEncode-8 42.5ms ± 0% 42.2ms ± 0% -0.73% (p=0.000 n=9+8) JSONDecode-8 158ms ± 1% 162ms ± 0% +2.28% (p=0.000 n=10+9) Mandelbrot200-8 10.1ms ± 0% 10.1ms ± 0% -0.01% (p=0.000 n=10+9) GoParse-8 8.54ms ± 1% 8.63ms ± 1% +1.06% (p=0.000 n=10+9) RegexpMatchEasy0_32-8 231ns ± 1% 225ns ± 0% -2.52% (p=0.000 n=9+10) RegexpMatchEasy0_1K-8 1.63µs ± 0% 1.63µs ± 0% ~ (p=0.170 n=10+10) RegexpMatchEasy1_32-8 253ns ± 0% 249ns ± 0% -1.41% (p=0.000 n=9+10) RegexpMatchEasy1_1K-8 2.08µs ± 0% 2.08µs ± 0% -0.32% (p=0.000 n=9+10) RegexpMatchMedium_32-8 355ns ± 1% 351ns ± 0% -1.04% (p=0.007 n=10+7) RegexpMatchMedium_1K-8 104µs ± 0% 104µs ± 0% ~ (p=0.148 n=10+10) RegexpMatchHard_32-8 5.79µs ± 0% 5.79µs ± 0% ~ (p=0.578 n=10+10) RegexpMatchHard_1K-8 176µs ± 0% 176µs ± 0% ~ (p=0.137 n=10+10) Revcomp-8 1.37s ± 1% 1.36s ± 1% -0.26% (p=0.023 n=10+10) Template-8 151ms ± 1% 154ms ± 1% +2.14% (p=0.000 n=9+10) TimeParse-8 723ns ± 2% 721ns ± 1% ~ (p=0.592 n=10+10) TimeFormat-8 804ns ± 2% 798ns ± 3% ~ (p=0.344 n=10+10) [Geo mean] 154µs 154µs -0.02% Therefore remove this pass. Also reduce text size by 0.5~2%. Comment out some dead code in runtime/sys_nacl_amd64p32.s which contains undefined symbols. Change-Id: I1473986fe5b18b3d2554ce96cdc6f0999b8d955d Reviewed-on: https://go-review.googlesource.com/36205 Run-TryBot: Cherry Zhang <cherryyz@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
This commit is contained in:
Cherry Zhang
parent
76d4274491
commit
bed8129ee6
9 changed files with 34 additions and 1126 deletions
|
|
@ -1418,151 +1418,10 @@ func compound(ctxt *obj.Link, p *obj.Prog) bool {
|
|||
return false
|
||||
}
|
||||
|
||||
func follow(ctxt *obj.Link, s *obj.LSym) {
|
||||
ctxt.Cursym = s
|
||||
|
||||
firstp := ctxt.NewProg()
|
||||
lastp := firstp
|
||||
xfol(ctxt, s.Text, &lastp)
|
||||
lastp.Link = nil
|
||||
s.Text = firstp.Link
|
||||
}
|
||||
|
||||
func xfol(ctxt *obj.Link, p *obj.Prog, last **obj.Prog) {
|
||||
var q *obj.Prog
|
||||
var r *obj.Prog
|
||||
var i int
|
||||
|
||||
loop:
|
||||
if p == nil {
|
||||
return
|
||||
}
|
||||
a := p.As
|
||||
if a == AJMP {
|
||||
q = p.Pcond
|
||||
if (p.Mark&NOSCHED != 0) || q != nil && (q.Mark&NOSCHED != 0) {
|
||||
p.Mark |= FOLL
|
||||
(*last).Link = p
|
||||
*last = p
|
||||
p = p.Link
|
||||
xfol(ctxt, p, last)
|
||||
p = q
|
||||
if p != nil && p.Mark&FOLL == 0 {
|
||||
goto loop
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
if q != nil {
|
||||
p.Mark |= FOLL
|
||||
p = q
|
||||
if p.Mark&FOLL == 0 {
|
||||
goto loop
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if p.Mark&FOLL != 0 {
|
||||
i = 0
|
||||
q = p
|
||||
for ; i < 4; i, q = i+1, q.Link {
|
||||
if q == *last || (q.Mark&NOSCHED != 0) {
|
||||
break
|
||||
}
|
||||
a = q.As
|
||||
if a == obj.ANOP {
|
||||
i--
|
||||
continue
|
||||
}
|
||||
|
||||
if a == AJMP || a == ARET || a == ARFE {
|
||||
goto copy
|
||||
}
|
||||
if q.Pcond == nil || (q.Pcond.Mark&FOLL != 0) {
|
||||
continue
|
||||
}
|
||||
if a != ABEQ && a != ABNE {
|
||||
continue
|
||||
}
|
||||
|
||||
copy:
|
||||
for {
|
||||
r = ctxt.NewProg()
|
||||
*r = *p
|
||||
if r.Mark&FOLL == 0 {
|
||||
fmt.Printf("can't happen 1\n")
|
||||
}
|
||||
r.Mark |= FOLL
|
||||
if p != q {
|
||||
p = p.Link
|
||||
(*last).Link = r
|
||||
*last = r
|
||||
continue
|
||||
}
|
||||
|
||||
(*last).Link = r
|
||||
*last = r
|
||||
if a == AJMP || a == ARET || a == ARFE {
|
||||
return
|
||||
}
|
||||
r.As = ABNE
|
||||
if a == ABNE {
|
||||
r.As = ABEQ
|
||||
}
|
||||
r.Pcond = p.Link
|
||||
r.Link = p.Pcond
|
||||
if r.Link.Mark&FOLL == 0 {
|
||||
xfol(ctxt, r.Link, last)
|
||||
}
|
||||
if r.Pcond.Mark&FOLL == 0 {
|
||||
fmt.Printf("can't happen 2\n")
|
||||
}
|
||||
return
|
||||
}
|
||||
}
|
||||
|
||||
a = AJMP
|
||||
q = ctxt.NewProg()
|
||||
q.As = a
|
||||
q.Pos = p.Pos
|
||||
q.To.Type = obj.TYPE_BRANCH
|
||||
q.To.Offset = p.Pc
|
||||
q.Pcond = p
|
||||
p = q
|
||||
}
|
||||
|
||||
p.Mark |= FOLL
|
||||
(*last).Link = p
|
||||
*last = p
|
||||
if a == AJMP || a == ARET || a == ARFE {
|
||||
if p.Mark&NOSCHED != 0 {
|
||||
p = p.Link
|
||||
goto loop
|
||||
}
|
||||
|
||||
return
|
||||
}
|
||||
|
||||
if p.Pcond != nil {
|
||||
if a != AJAL && p.Link != nil {
|
||||
xfol(ctxt, p.Link, last)
|
||||
p = p.Pcond
|
||||
if p == nil || (p.Mark&FOLL != 0) {
|
||||
return
|
||||
}
|
||||
goto loop
|
||||
}
|
||||
}
|
||||
|
||||
p = p.Link
|
||||
goto loop
|
||||
}
|
||||
|
||||
var Linkmips64 = obj.LinkArch{
|
||||
Arch: sys.ArchMIPS64,
|
||||
Preprocess: preprocess,
|
||||
Assemble: span0,
|
||||
Follow: follow,
|
||||
Progedit: progedit,
|
||||
}
|
||||
|
||||
|
|
@ -1570,7 +1429,6 @@ var Linkmips64le = obj.LinkArch{
|
|||
Arch: sys.ArchMIPS64LE,
|
||||
Preprocess: preprocess,
|
||||
Assemble: span0,
|
||||
Follow: follow,
|
||||
Progedit: progedit,
|
||||
}
|
||||
|
||||
|
|
@ -1578,7 +1436,6 @@ var Linkmips = obj.LinkArch{
|
|||
Arch: sys.ArchMIPS,
|
||||
Preprocess: preprocess,
|
||||
Assemble: span0,
|
||||
Follow: follow,
|
||||
Progedit: progedit,
|
||||
}
|
||||
|
||||
|
|
@ -1586,6 +1443,5 @@ var Linkmipsle = obj.LinkArch{
|
|||
Arch: sys.ArchMIPSLE,
|
||||
Preprocess: preprocess,
|
||||
Assemble: span0,
|
||||
Follow: follow,
|
||||
Progedit: progedit,
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue