inlining reduces performance compared to manual inlining #35687
Description
$ go version 1.13.4
Does this issue reproduce with the latest release?
yes
What operating system and processor architecture are you using (go env)?
go env Output
$ go env GO111MODULE="" GOARCH="amd64" GOBIN="" GOCACHE="/home/jaap/.cache/go-build" GOENV="/home/jaap/.config/go/env" GOEXE="" GOFLAGS="" GOHOSTARCH="amd64" GOHOSTOS="linux" GONOPROXY="" GONOSUMDB="" GOOS="linux" GOPATH="/home/jaap/go" GOPRIVATE="" GOPROXY="https://proxy.golang.org,direct" GOROOT="/usr/lib/go" GOSUMDB="sum.golang.org" GOTMPDIR="" GOTOOLDIR="/usr/lib/go/pkg/tool/linux_amd64" GCCGO="gccgo" AR="ar" CC="gcc" CXX="g++" CGO_ENABLED="1" GOMOD="/home/jaap/go-projects/generator/go.mod" CGO_CFLAGS="-g -O2" CGO_CPPFLAGS="" CGO_CXXFLAGS="-g -O2" CGO_FFLAGS="-g -O2" CGO_LDFLAGS="-g -O2" PKG_CONFIG="pkg-config" GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build995427981=/tmp/go-build -gno-record-gcc-switches"
What did you do?
package main
type Week [5][9]timeslot
type timeslot struct {
Score float32 `json:"score"`
Percentage float32 `json:"percentage"`
Reasons []int `json:"reasons"`
}
type timeid struct {
ScheduleID string `json:"schedule_id"`
Week uint8 `json:"week"`
Day int `json:"day"`
Period int `json:"period"`
Lesson int `json:"lesson"`
}
func (w Week) getPeriod(time timeid) timeslot {
return w[time.Day][time.Period]
}
func main() {
week := Week{}
time := timeid{}
averageScore := float32(5)
totalDistanceBottom := float32(5)
for day := range week {
time.Day = day
for period := range week[day] {
time.Period = period
score := week[time.Day][time.Period].Score
//score := week.getPeriod(time).Score
if score < averageScore {
distance := averageScore - score
week[day][period].Percentage = -distance / totalDistanceBottom
}
}
}
}
What did you expect to see?
A manually inlined function call without a lot of moves.
Note that this does have bounds checks later on in the asm.
0x0069 00105 ($PWD/main.go:32) MOVQ "".time+40(SP), DX
0x006e 00110 ($PWD/main.go:32) CMPQ DX, $5
0x0072 00114 ($PWD/main.go:32) JCC 254
0x0078 00120 ($PWD/main.go:32) LEAQ (DX)(DX*8), DX
0x007c 00124 ($PWD/main.go:32) SHLQ $5, DX
0x0080 00128 ($PWD/main.go:32) PCDATA $0, $2
0x0080 00128 ($PWD/main.go:32) LEAQ "".week+64(SP)(DX*1), DX
0x0085 00133 ($PWD/main.go:32) MOVQ CX, BX
0x0088 00136 ($PWD/main.go:32) SHLQ $5, CX
0x008c 00140 ($PWD/main.go:32) PCDATA $0, $0
0x008c 00140 ($PWD/main.go:32) MOVSS (DX)(CX*1), X0
What did you see instead?
An automatically inlined function with a lot of moves
0x006d 00109 ($PWD/main.go:33) PCDATA $0, $1
0x006d 00109 ($PWD/main.go:33) PCDATA $1, $3
0x006d 00109 ($PWD/main.go:33) LEAQ "".w+1552(SP), DI
0x0075 00117 ($PWD/main.go:33) PCDATA $0, $2
0x0075 00117 ($PWD/main.go:33) LEAQ "".week+112(SP), SI
0x007a 00122 ($PWD/main.go:29) MOVQ CX, DX
0x007d 00125 ($PWD/main.go:33) MOVL $180, CX
0x0082 00130 ($PWD/main.go:33) PCDATA $0, $0
0x0082 00130 ($PWD/main.go:33) REP
0x0083 00131 ($PWD/main.go:33) MOVSQ
0x0085 00133 ($PWD/main.go:33) PCDATA $1, $4
0x0085 00133 ($PWD/main.go:33) MOVUPS "".time+64(SP), X0
0x008a 00138 ($PWD/main.go:33) MOVUPS X0, "".time+16(SP)
0x008f 00143 ($PWD/main.go:33) MOVUPS "".time+80(SP), X0
0x0094 00148 ($PWD/main.go:33) MOVUPS X0, "".time+32(SP)
0x0099 00153 ($PWD/main.go:33) MOVUPS "".time+96(SP), X0
0x009e 00158 ($PWD/main.go:33) MOVUPS X0, "".time+48(SP)