2018-11-02 03:09:30 -06:00
|
|
|
// +build !nacl,!js,!aix,!gcflags_noopt
|
2015-02-03 11:52:18 -07:00
|
|
|
// run
|
2014-05-20 10:10:19 -06:00
|
|
|
|
2016-04-10 15:32:26 -06:00
|
|
|
// Copyright 2014 The Go Authors. All rights reserved.
|
2014-04-16 20:08:00 -06:00
|
|
|
// Use of this source code is governed by a BSD-style
|
|
|
|
// license that can be found in the LICENSE file.
|
|
|
|
|
|
|
|
package main
|
|
|
|
|
|
|
|
import (
|
|
|
|
"bytes"
|
|
|
|
"fmt"
|
|
|
|
"io/ioutil"
|
2014-09-16 15:39:55 -06:00
|
|
|
"log"
|
2014-04-16 20:08:00 -06:00
|
|
|
"os"
|
|
|
|
"os/exec"
|
|
|
|
"path/filepath"
|
|
|
|
"regexp"
|
|
|
|
"runtime"
|
|
|
|
"strconv"
|
|
|
|
"strings"
|
|
|
|
)
|
|
|
|
|
|
|
|
var tests = `
|
|
|
|
# These are test cases for the linker analysis that detects chains of
|
|
|
|
# nosplit functions that would cause a stack overflow.
|
|
|
|
#
|
|
|
|
# Lines beginning with # are comments.
|
|
|
|
#
|
|
|
|
# Each test case describes a sequence of functions, one per line.
|
|
|
|
# Each function definition is the function name, then the frame size,
|
|
|
|
# then optionally the keyword 'nosplit', then the body of the function.
|
|
|
|
# The body is assembly code, with some shorthands.
|
|
|
|
# The shorthand 'call x' stands for CALL x(SB).
|
|
|
|
# The shorthand 'callind' stands for 'CALL R0', where R0 is a register.
|
|
|
|
# Each test case must define a function named main, and it must be first.
|
|
|
|
# That is, a line beginning "main " indicates the start of a new test case.
|
|
|
|
# Within a stanza, ; can be used instead of \n to separate lines.
|
|
|
|
#
|
|
|
|
# After the function definition, the test case ends with an optional
|
|
|
|
# REJECT line, specifying the architectures on which the case should
|
|
|
|
# be rejected. "REJECT" without any architectures means reject on all architectures.
|
|
|
|
# The linker should accept the test case on systems not explicitly rejected.
|
|
|
|
#
|
|
|
|
# 64-bit systems do not attempt to execute test cases with frame sizes
|
|
|
|
# that are only 32-bit aligned.
|
|
|
|
|
|
|
|
# Ordinary function should work
|
|
|
|
main 0
|
|
|
|
|
|
|
|
# Large frame marked nosplit is always wrong.
|
|
|
|
main 10000 nosplit
|
|
|
|
REJECT
|
|
|
|
|
|
|
|
# Calling a large frame is okay.
|
|
|
|
main 0 call big
|
|
|
|
big 10000
|
|
|
|
|
|
|
|
# But not if the frame is nosplit.
|
|
|
|
main 0 call big
|
|
|
|
big 10000 nosplit
|
|
|
|
REJECT
|
|
|
|
|
|
|
|
# Recursion is okay.
|
|
|
|
main 0 call main
|
|
|
|
|
|
|
|
# Recursive nosplit runs out of space.
|
|
|
|
main 0 nosplit call main
|
|
|
|
REJECT
|
|
|
|
|
|
|
|
# Chains of ordinary functions okay.
|
|
|
|
main 0 call f1
|
|
|
|
f1 80 call f2
|
|
|
|
f2 80
|
|
|
|
|
|
|
|
# Chains of nosplit must fit in the stack limit, 128 bytes.
|
|
|
|
main 0 call f1
|
|
|
|
f1 80 nosplit call f2
|
|
|
|
f2 80 nosplit
|
|
|
|
REJECT
|
|
|
|
|
|
|
|
# Larger chains.
|
|
|
|
main 0 call f1
|
|
|
|
f1 16 call f2
|
|
|
|
f2 16 call f3
|
|
|
|
f3 16 call f4
|
|
|
|
f4 16 call f5
|
|
|
|
f5 16 call f6
|
|
|
|
f6 16 call f7
|
|
|
|
f7 16 call f8
|
|
|
|
f8 16 call end
|
|
|
|
end 1000
|
|
|
|
|
|
|
|
main 0 call f1
|
|
|
|
f1 16 nosplit call f2
|
|
|
|
f2 16 nosplit call f3
|
|
|
|
f3 16 nosplit call f4
|
|
|
|
f4 16 nosplit call f5
|
|
|
|
f5 16 nosplit call f6
|
|
|
|
f6 16 nosplit call f7
|
|
|
|
f7 16 nosplit call f8
|
|
|
|
f8 16 nosplit call end
|
|
|
|
end 1000
|
|
|
|
REJECT
|
|
|
|
|
|
|
|
# Test cases near the 128-byte limit.
|
|
|
|
|
|
|
|
# Ordinary stack split frame is always okay.
|
|
|
|
main 112
|
|
|
|
main 116
|
|
|
|
main 120
|
|
|
|
main 124
|
|
|
|
main 128
|
|
|
|
main 132
|
|
|
|
main 136
|
|
|
|
|
|
|
|
# A nosplit leaf can use the whole 128-CallSize bytes available on entry.
|
2017-10-05 14:14:14 -06:00
|
|
|
# (CallSize is 32 on ppc64, 8 on amd64 for frame pointer.)
|
2015-10-29 17:47:24 -06:00
|
|
|
main 96 nosplit
|
|
|
|
main 100 nosplit; REJECT ppc64 ppc64le
|
2018-08-29 00:55:03 -06:00
|
|
|
main 104 nosplit; REJECT ppc64 ppc64le arm64
|
2015-10-29 17:47:24 -06:00
|
|
|
main 108 nosplit; REJECT ppc64 ppc64le
|
2018-08-29 00:55:03 -06:00
|
|
|
main 112 nosplit; REJECT ppc64 ppc64le arm64
|
2015-10-29 17:47:24 -06:00
|
|
|
main 116 nosplit; REJECT ppc64 ppc64le
|
2018-08-29 00:55:03 -06:00
|
|
|
main 120 nosplit; REJECT ppc64 ppc64le amd64 arm64
|
2017-10-05 14:14:14 -06:00
|
|
|
main 124 nosplit; REJECT ppc64 ppc64le amd64
|
2014-04-16 20:08:00 -06:00
|
|
|
main 128 nosplit; REJECT
|
|
|
|
main 132 nosplit; REJECT
|
|
|
|
main 136 nosplit; REJECT
|
|
|
|
|
|
|
|
# Calling a nosplit function from a nosplit function requires
|
|
|
|
# having room for the saved caller PC and the called frame.
|
|
|
|
# Because ARM doesn't save LR in the leaf, it gets an extra 4 bytes.
|
2015-10-29 17:47:24 -06:00
|
|
|
# Because arm64 doesn't save LR in the leaf, it gets an extra 8 bytes.
|
2017-10-05 14:14:14 -06:00
|
|
|
# ppc64 doesn't save LR in the leaf, but CallSize is 32, so it gets 24 bytes.
|
|
|
|
# Because AMD64 uses frame pointer, it has 8 fewer bytes.
|
2015-10-29 17:47:24 -06:00
|
|
|
main 96 nosplit call f; f 0 nosplit
|
|
|
|
main 100 nosplit call f; f 0 nosplit; REJECT ppc64 ppc64le
|
2018-08-29 00:55:03 -06:00
|
|
|
main 104 nosplit call f; f 0 nosplit; REJECT ppc64 ppc64le arm64
|
2015-10-29 17:47:24 -06:00
|
|
|
main 108 nosplit call f; f 0 nosplit; REJECT ppc64 ppc64le
|
2018-08-29 00:55:03 -06:00
|
|
|
main 112 nosplit call f; f 0 nosplit; REJECT ppc64 ppc64le amd64 arm64
|
2017-10-05 14:14:14 -06:00
|
|
|
main 116 nosplit call f; f 0 nosplit; REJECT ppc64 ppc64le amd64
|
2018-08-29 12:49:42 -06:00
|
|
|
main 120 nosplit call f; f 0 nosplit; REJECT ppc64 ppc64le amd64 arm64
|
2015-10-29 17:47:24 -06:00
|
|
|
main 124 nosplit call f; f 0 nosplit; REJECT ppc64 ppc64le amd64 386
|
2014-04-16 20:08:00 -06:00
|
|
|
main 128 nosplit call f; f 0 nosplit; REJECT
|
|
|
|
main 132 nosplit call f; f 0 nosplit; REJECT
|
|
|
|
main 136 nosplit call f; f 0 nosplit; REJECT
|
|
|
|
|
|
|
|
# Calling a splitting function from a nosplit function requires
|
|
|
|
# having room for the saved caller PC of the call but also the
|
2014-08-14 13:29:37 -06:00
|
|
|
# saved caller PC for the call to morestack.
|
2017-10-05 14:14:14 -06:00
|
|
|
# Architectures differ in the same way as before.
|
2015-10-29 17:47:24 -06:00
|
|
|
main 96 nosplit call f; f 0 call f
|
|
|
|
main 100 nosplit call f; f 0 call f; REJECT ppc64 ppc64le
|
2018-08-29 00:55:03 -06:00
|
|
|
main 104 nosplit call f; f 0 call f; REJECT ppc64 ppc64le amd64 arm64
|
2017-10-05 14:14:14 -06:00
|
|
|
main 108 nosplit call f; f 0 call f; REJECT ppc64 ppc64le amd64
|
2018-08-29 00:55:03 -06:00
|
|
|
main 112 nosplit call f; f 0 call f; REJECT ppc64 ppc64le amd64 arm64
|
2015-10-29 17:47:24 -06:00
|
|
|
main 116 nosplit call f; f 0 call f; REJECT ppc64 ppc64le amd64
|
2018-08-29 00:55:03 -06:00
|
|
|
main 120 nosplit call f; f 0 call f; REJECT ppc64 ppc64le amd64 386 arm64
|
2015-10-29 17:47:24 -06:00
|
|
|
main 124 nosplit call f; f 0 call f; REJECT ppc64 ppc64le amd64 386
|
2014-04-16 20:08:00 -06:00
|
|
|
main 128 nosplit call f; f 0 call f; REJECT
|
|
|
|
main 132 nosplit call f; f 0 call f; REJECT
|
|
|
|
main 136 nosplit call f; f 0 call f; REJECT
|
|
|
|
|
|
|
|
# Indirect calls are assumed to be splitting functions.
|
2015-10-29 17:47:24 -06:00
|
|
|
main 96 nosplit callind
|
|
|
|
main 100 nosplit callind; REJECT ppc64 ppc64le
|
2018-08-29 00:55:03 -06:00
|
|
|
main 104 nosplit callind; REJECT ppc64 ppc64le amd64 arm64
|
2017-10-05 14:14:14 -06:00
|
|
|
main 108 nosplit callind; REJECT ppc64 ppc64le amd64
|
2018-08-29 00:55:03 -06:00
|
|
|
main 112 nosplit callind; REJECT ppc64 ppc64le amd64 arm64
|
2015-10-29 17:47:24 -06:00
|
|
|
main 116 nosplit callind; REJECT ppc64 ppc64le amd64
|
2018-08-29 00:55:03 -06:00
|
|
|
main 120 nosplit callind; REJECT ppc64 ppc64le amd64 386 arm64
|
2015-10-29 17:47:24 -06:00
|
|
|
main 124 nosplit callind; REJECT ppc64 ppc64le amd64 386
|
2014-04-16 20:08:00 -06:00
|
|
|
main 128 nosplit callind; REJECT
|
|
|
|
main 132 nosplit callind; REJECT
|
|
|
|
main 136 nosplit callind; REJECT
|
|
|
|
|
|
|
|
# Issue 7623
|
|
|
|
main 0 call f; f 112
|
|
|
|
main 0 call f; f 116
|
|
|
|
main 0 call f; f 120
|
|
|
|
main 0 call f; f 124
|
|
|
|
main 0 call f; f 128
|
|
|
|
main 0 call f; f 132
|
|
|
|
main 0 call f; f 136
|
|
|
|
`
|
|
|
|
|
|
|
|
var (
|
|
|
|
commentRE = regexp.MustCompile(`(?m)^#.*`)
|
|
|
|
rejectRE = regexp.MustCompile(`(?s)\A(.+?)((\n|; *)REJECT(.*))?\z`)
|
|
|
|
lineRE = regexp.MustCompile(`(\w+) (\d+)( nosplit)?(.*)`)
|
|
|
|
callRE = regexp.MustCompile(`\bcall (\w+)\b`)
|
|
|
|
callindRE = regexp.MustCompile(`\bcallind\b`)
|
|
|
|
)
|
|
|
|
|
|
|
|
func main() {
|
|
|
|
goarch := os.Getenv("GOARCH")
|
|
|
|
if goarch == "" {
|
|
|
|
goarch = runtime.GOARCH
|
|
|
|
}
|
|
|
|
|
2015-05-21 11:28:13 -06:00
|
|
|
version, err := exec.Command("go", "tool", "compile", "-V").Output()
|
2015-01-14 09:09:50 -07:00
|
|
|
if err != nil {
|
|
|
|
bug()
|
2015-05-27 13:33:43 -06:00
|
|
|
fmt.Printf("running go tool compile -V: %v\n", err)
|
2015-01-14 09:09:50 -07:00
|
|
|
return
|
|
|
|
}
|
2017-10-05 14:14:14 -06:00
|
|
|
if s := string(version); goarch == "amd64" && strings.Contains(s, "X:") && !strings.Contains(s, "framepointer") {
|
|
|
|
// Skip this test if framepointer is NOT enabled on AMD64
|
2015-01-14 09:09:50 -07:00
|
|
|
return
|
|
|
|
}
|
|
|
|
|
2014-04-16 20:08:00 -06:00
|
|
|
dir, err := ioutil.TempDir("", "go-test-nosplit")
|
|
|
|
if err != nil {
|
|
|
|
bug()
|
|
|
|
fmt.Printf("creating temp dir: %v\n", err)
|
|
|
|
return
|
|
|
|
}
|
|
|
|
defer os.RemoveAll(dir)
|
2019-03-12 14:28:03 -06:00
|
|
|
os.Setenv("GOPATH", filepath.Join(dir, "_gopath"))
|
2014-04-16 20:08:00 -06:00
|
|
|
|
2019-02-22 13:58:15 -07:00
|
|
|
if err := ioutil.WriteFile(filepath.Join(dir, "go.mod"), []byte("module go-test-nosplit\n"), 0666); err != nil {
|
|
|
|
log.Panic(err)
|
|
|
|
}
|
|
|
|
|
2014-04-16 20:08:00 -06:00
|
|
|
tests = strings.Replace(tests, "\t", " ", -1)
|
|
|
|
tests = commentRE.ReplaceAllString(tests, "")
|
|
|
|
|
|
|
|
nok := 0
|
|
|
|
nfail := 0
|
|
|
|
TestCases:
|
|
|
|
for len(tests) > 0 {
|
|
|
|
var stanza string
|
|
|
|
i := strings.Index(tests, "\nmain ")
|
|
|
|
if i < 0 {
|
|
|
|
stanza, tests = tests, ""
|
|
|
|
} else {
|
|
|
|
stanza, tests = tests[:i], tests[i+1:]
|
|
|
|
}
|
|
|
|
|
|
|
|
m := rejectRE.FindStringSubmatch(stanza)
|
|
|
|
if m == nil {
|
|
|
|
bug()
|
|
|
|
fmt.Printf("invalid stanza:\n\t%s\n", indent(stanza))
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
lines := strings.TrimSpace(m[1])
|
|
|
|
reject := false
|
|
|
|
if m[2] != "" {
|
|
|
|
if strings.TrimSpace(m[4]) == "" {
|
|
|
|
reject = true
|
|
|
|
} else {
|
|
|
|
for _, rej := range strings.Fields(m[4]) {
|
|
|
|
if rej == goarch {
|
|
|
|
reject = true
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if lines == "" && !reject {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
|
2014-09-16 15:39:55 -06:00
|
|
|
var gobuf bytes.Buffer
|
|
|
|
fmt.Fprintf(&gobuf, "package main\n")
|
|
|
|
|
2014-04-16 20:08:00 -06:00
|
|
|
var buf bytes.Buffer
|
2014-08-14 11:59:58 -06:00
|
|
|
ptrSize := 4
|
|
|
|
switch goarch {
|
2016-12-15 10:05:26 -07:00
|
|
|
case "mips", "mipsle":
|
2017-10-05 14:14:14 -06:00
|
|
|
fmt.Fprintf(&buf, "#define REGISTER (R0)\n")
|
2015-09-28 15:30:32 -06:00
|
|
|
case "mips64", "mips64le":
|
|
|
|
ptrSize = 8
|
2017-10-05 14:14:14 -06:00
|
|
|
fmt.Fprintf(&buf, "#define REGISTER (R0)\n")
|
2014-12-05 17:13:20 -07:00
|
|
|
case "ppc64", "ppc64le":
|
2014-08-14 11:59:58 -06:00
|
|
|
ptrSize = 8
|
2017-10-05 14:14:14 -06:00
|
|
|
fmt.Fprintf(&buf, "#define REGISTER (CTR)\n")
|
2014-08-14 11:59:58 -06:00
|
|
|
case "arm":
|
2017-10-05 14:14:14 -06:00
|
|
|
fmt.Fprintf(&buf, "#define REGISTER (R0)\n")
|
2015-03-08 07:27:51 -06:00
|
|
|
case "arm64":
|
|
|
|
ptrSize = 8
|
2017-10-05 14:14:14 -06:00
|
|
|
fmt.Fprintf(&buf, "#define REGISTER (R0)\n")
|
2014-08-14 11:59:58 -06:00
|
|
|
case "amd64":
|
|
|
|
ptrSize = 8
|
|
|
|
fmt.Fprintf(&buf, "#define REGISTER AX\n")
|
2020-01-24 09:10:04 -07:00
|
|
|
case "riscv64":
|
|
|
|
ptrSize = 8
|
|
|
|
fmt.Fprintf(&buf, "#define REGISTER A0\n")
|
2016-03-20 17:41:34 -06:00
|
|
|
case "s390x":
|
|
|
|
ptrSize = 8
|
|
|
|
fmt.Fprintf(&buf, "#define REGISTER R10\n")
|
2014-08-14 11:59:58 -06:00
|
|
|
default:
|
2014-04-16 20:08:00 -06:00
|
|
|
fmt.Fprintf(&buf, "#define REGISTER AX\n")
|
|
|
|
}
|
|
|
|
|
|
|
|
for _, line := range strings.Split(lines, "\n") {
|
|
|
|
line = strings.TrimSpace(line)
|
|
|
|
if line == "" {
|
|
|
|
continue
|
|
|
|
}
|
2014-08-27 12:08:26 -06:00
|
|
|
for i, subline := range strings.Split(line, ";") {
|
2014-04-16 20:08:00 -06:00
|
|
|
subline = strings.TrimSpace(subline)
|
|
|
|
if subline == "" {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
m := lineRE.FindStringSubmatch(subline)
|
|
|
|
if m == nil {
|
|
|
|
bug()
|
|
|
|
fmt.Printf("invalid function line: %s\n", subline)
|
|
|
|
continue TestCases
|
|
|
|
}
|
|
|
|
name := m[1]
|
|
|
|
size, _ := strconv.Atoi(m[2])
|
2014-08-27 12:08:26 -06:00
|
|
|
|
cmd/compile, cmd/link, runtime: make defers low-cost through inline code and extra funcdata
Generate inline code at defer time to save the args of defer calls to unique
(autotmp) stack slots, and generate inline code at exit time to check which defer
calls were made and make the associated function/method/interface calls. We
remember that a particular defer statement was reached by storing in the deferBits
variable (always stored on the stack). At exit time, we check the bits of the
deferBits variable to determine which defer function calls to make (in reverse
order). These low-cost defers are only used for functions where no defers
appear in loops. In addition, we don't do these low-cost defers if there are too
many defer statements or too many exits in a function (to limit code increase).
When a function uses open-coded defers, we produce extra
FUNCDATA_OpenCodedDeferInfo information that specifies the number of defers, and
for each defer, the stack slots where the closure and associated args have been
stored. The funcdata also includes the location of the deferBits variable.
Therefore, for panics, we can use this funcdata to determine exactly which defers
are active, and call the appropriate functions/methods/closures with the correct
arguments for each active defer.
In order to unwind the stack correctly after a recover(), we need to add an extra
code segment to functions with open-coded defers that simply calls deferreturn()
and returns. This segment is not reachable by the normal function, but is returned
to by the runtime during recovery. We set the liveness information of this
deferreturn() to be the same as the liveness at the first function call during the
last defer exit code (so all return values and all stack slots needed by the defer
calls will be live).
I needed to increase the stackguard constant from 880 to 896, because of a small
amount of new code in deferreturn().
The -N flag disables open-coded defers. '-d defer' prints out the kind of defer
being used at each defer statement (heap-allocated, stack-allocated, or
open-coded).
Cost of defer statement [ go test -run NONE -bench BenchmarkDefer$ runtime ]
With normal (stack-allocated) defers only: 35.4 ns/op
With open-coded defers: 5.6 ns/op
Cost of function call alone (remove defer keyword): 4.4 ns/op
Text size increase (including funcdata) for go binary without/with open-coded defers: 0.09%
The average size increase (including funcdata) for only the functions that use
open-coded defers is 1.1%.
The cost of a panic followed by a recover got noticeably slower, since panic
processing now requires a scan of the stack for open-coded defer frames. This scan
is required, even if no frames are using open-coded defers:
Cost of panic and recover [ go test -run NONE -bench BenchmarkPanicRecover runtime ]
Without open-coded defers: 62.0 ns/op
With open-coded defers: 255 ns/op
A CGO Go-to-C-to-Go benchmark got noticeably faster because of open-coded defers:
CGO Go-to-C-to-Go benchmark [cd misc/cgo/test; go test -run NONE -bench BenchmarkCGoCallback ]
Without open-coded defers: 443 ns/op
With open-coded defers: 347 ns/op
Updates #14939 (defer performance)
Updates #34481 (design doc)
Change-Id: I63b1a60d1ebf28126f55ee9fd7ecffe9cb23d1ff
Reviewed-on: https://go-review.googlesource.com/c/go/+/202340
Reviewed-by: Austin Clements <austin@google.com>
2019-06-24 13:59:22 -06:00
|
|
|
// The limit was originally 128 but is now 768 (896-128).
|
2014-08-27 12:08:26 -06:00
|
|
|
// Instead of rewriting the test cases above, adjust
|
2015-05-01 10:36:18 -06:00
|
|
|
// the first stack frame to use up the extra bytes.
|
2014-08-27 12:08:26 -06:00
|
|
|
if i == 0 {
|
cmd/compile, cmd/link, runtime: make defers low-cost through inline code and extra funcdata
Generate inline code at defer time to save the args of defer calls to unique
(autotmp) stack slots, and generate inline code at exit time to check which defer
calls were made and make the associated function/method/interface calls. We
remember that a particular defer statement was reached by storing in the deferBits
variable (always stored on the stack). At exit time, we check the bits of the
deferBits variable to determine which defer function calls to make (in reverse
order). These low-cost defers are only used for functions where no defers
appear in loops. In addition, we don't do these low-cost defers if there are too
many defer statements or too many exits in a function (to limit code increase).
When a function uses open-coded defers, we produce extra
FUNCDATA_OpenCodedDeferInfo information that specifies the number of defers, and
for each defer, the stack slots where the closure and associated args have been
stored. The funcdata also includes the location of the deferBits variable.
Therefore, for panics, we can use this funcdata to determine exactly which defers
are active, and call the appropriate functions/methods/closures with the correct
arguments for each active defer.
In order to unwind the stack correctly after a recover(), we need to add an extra
code segment to functions with open-coded defers that simply calls deferreturn()
and returns. This segment is not reachable by the normal function, but is returned
to by the runtime during recovery. We set the liveness information of this
deferreturn() to be the same as the liveness at the first function call during the
last defer exit code (so all return values and all stack slots needed by the defer
calls will be live).
I needed to increase the stackguard constant from 880 to 896, because of a small
amount of new code in deferreturn().
The -N flag disables open-coded defers. '-d defer' prints out the kind of defer
being used at each defer statement (heap-allocated, stack-allocated, or
open-coded).
Cost of defer statement [ go test -run NONE -bench BenchmarkDefer$ runtime ]
With normal (stack-allocated) defers only: 35.4 ns/op
With open-coded defers: 5.6 ns/op
Cost of function call alone (remove defer keyword): 4.4 ns/op
Text size increase (including funcdata) for go binary without/with open-coded defers: 0.09%
The average size increase (including funcdata) for only the functions that use
open-coded defers is 1.1%.
The cost of a panic followed by a recover got noticeably slower, since panic
processing now requires a scan of the stack for open-coded defer frames. This scan
is required, even if no frames are using open-coded defers:
Cost of panic and recover [ go test -run NONE -bench BenchmarkPanicRecover runtime ]
Without open-coded defers: 62.0 ns/op
With open-coded defers: 255 ns/op
A CGO Go-to-C-to-Go benchmark got noticeably faster because of open-coded defers:
CGO Go-to-C-to-Go benchmark [cd misc/cgo/test; go test -run NONE -bench BenchmarkCGoCallback ]
Without open-coded defers: 443 ns/op
With open-coded defers: 347 ns/op
Updates #14939 (defer performance)
Updates #34481 (design doc)
Change-Id: I63b1a60d1ebf28126f55ee9fd7ecffe9cb23d1ff
Reviewed-on: https://go-review.googlesource.com/c/go/+/202340
Reviewed-by: Austin Clements <austin@google.com>
2019-06-24 13:59:22 -06:00
|
|
|
size += (896 - 128) - 128
|
2015-05-01 10:36:18 -06:00
|
|
|
// Noopt builds have a larger stackguard.
|
2015-10-19 09:36:07 -06:00
|
|
|
// See ../src/cmd/dist/buildruntime.go:stackGuardMultiplier
|
2017-04-18 13:53:25 -06:00
|
|
|
// This increase is included in objabi.StackGuard
|
2015-05-01 10:36:18 -06:00
|
|
|
for _, s := range strings.Split(os.Getenv("GO_GCFLAGS"), " ") {
|
|
|
|
if s == "-N" {
|
cmd/compile, cmd/link, runtime: make defers low-cost through inline code and extra funcdata
Generate inline code at defer time to save the args of defer calls to unique
(autotmp) stack slots, and generate inline code at exit time to check which defer
calls were made and make the associated function/method/interface calls. We
remember that a particular defer statement was reached by storing in the deferBits
variable (always stored on the stack). At exit time, we check the bits of the
deferBits variable to determine which defer function calls to make (in reverse
order). These low-cost defers are only used for functions where no defers
appear in loops. In addition, we don't do these low-cost defers if there are too
many defer statements or too many exits in a function (to limit code increase).
When a function uses open-coded defers, we produce extra
FUNCDATA_OpenCodedDeferInfo information that specifies the number of defers, and
for each defer, the stack slots where the closure and associated args have been
stored. The funcdata also includes the location of the deferBits variable.
Therefore, for panics, we can use this funcdata to determine exactly which defers
are active, and call the appropriate functions/methods/closures with the correct
arguments for each active defer.
In order to unwind the stack correctly after a recover(), we need to add an extra
code segment to functions with open-coded defers that simply calls deferreturn()
and returns. This segment is not reachable by the normal function, but is returned
to by the runtime during recovery. We set the liveness information of this
deferreturn() to be the same as the liveness at the first function call during the
last defer exit code (so all return values and all stack slots needed by the defer
calls will be live).
I needed to increase the stackguard constant from 880 to 896, because of a small
amount of new code in deferreturn().
The -N flag disables open-coded defers. '-d defer' prints out the kind of defer
being used at each defer statement (heap-allocated, stack-allocated, or
open-coded).
Cost of defer statement [ go test -run NONE -bench BenchmarkDefer$ runtime ]
With normal (stack-allocated) defers only: 35.4 ns/op
With open-coded defers: 5.6 ns/op
Cost of function call alone (remove defer keyword): 4.4 ns/op
Text size increase (including funcdata) for go binary without/with open-coded defers: 0.09%
The average size increase (including funcdata) for only the functions that use
open-coded defers is 1.1%.
The cost of a panic followed by a recover got noticeably slower, since panic
processing now requires a scan of the stack for open-coded defer frames. This scan
is required, even if no frames are using open-coded defers:
Cost of panic and recover [ go test -run NONE -bench BenchmarkPanicRecover runtime ]
Without open-coded defers: 62.0 ns/op
With open-coded defers: 255 ns/op
A CGO Go-to-C-to-Go benchmark got noticeably faster because of open-coded defers:
CGO Go-to-C-to-Go benchmark [cd misc/cgo/test; go test -run NONE -bench BenchmarkCGoCallback ]
Without open-coded defers: 443 ns/op
With open-coded defers: 347 ns/op
Updates #14939 (defer performance)
Updates #34481 (design doc)
Change-Id: I63b1a60d1ebf28126f55ee9fd7ecffe9cb23d1ff
Reviewed-on: https://go-review.googlesource.com/c/go/+/202340
Reviewed-by: Austin Clements <austin@google.com>
2019-06-24 13:59:22 -06:00
|
|
|
size += 896
|
2015-05-01 10:36:18 -06:00
|
|
|
}
|
|
|
|
}
|
2014-08-27 12:08:26 -06:00
|
|
|
}
|
|
|
|
|
2018-08-29 00:55:03 -06:00
|
|
|
if size%ptrSize == 4 {
|
2014-04-16 20:08:00 -06:00
|
|
|
continue TestCases
|
|
|
|
}
|
|
|
|
nosplit := m[3]
|
|
|
|
body := m[4]
|
|
|
|
|
|
|
|
if nosplit != "" {
|
|
|
|
nosplit = ",7"
|
|
|
|
} else {
|
|
|
|
nosplit = ",0"
|
|
|
|
}
|
|
|
|
body = callRE.ReplaceAllString(body, "CALL ·$1(SB);")
|
|
|
|
body = callindRE.ReplaceAllString(body, "CALL REGISTER;")
|
|
|
|
|
2014-09-16 15:39:55 -06:00
|
|
|
fmt.Fprintf(&gobuf, "func %s()\n", name)
|
2014-04-16 20:08:00 -06:00
|
|
|
fmt.Fprintf(&buf, "TEXT ·%s(SB)%s,$%d-0\n\t%s\n\tRET\n\n", name, nosplit, size, body)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-09-16 15:39:55 -06:00
|
|
|
if err := ioutil.WriteFile(filepath.Join(dir, "asm.s"), buf.Bytes(), 0666); err != nil {
|
|
|
|
log.Fatal(err)
|
|
|
|
}
|
|
|
|
if err := ioutil.WriteFile(filepath.Join(dir, "main.go"), gobuf.Bytes(), 0666); err != nil {
|
|
|
|
log.Fatal(err)
|
|
|
|
}
|
2014-04-16 20:08:00 -06:00
|
|
|
|
|
|
|
cmd := exec.Command("go", "build")
|
|
|
|
cmd.Dir = dir
|
|
|
|
output, err := cmd.CombinedOutput()
|
|
|
|
if err == nil {
|
|
|
|
nok++
|
|
|
|
if reject {
|
|
|
|
bug()
|
|
|
|
fmt.Printf("accepted incorrectly:\n\t%s\n", indent(strings.TrimSpace(stanza)))
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
nfail++
|
|
|
|
if !reject {
|
|
|
|
bug()
|
|
|
|
fmt.Printf("rejected incorrectly:\n\t%s\n", indent(strings.TrimSpace(stanza)))
|
|
|
|
fmt.Printf("\n\tlinker output:\n\t%s\n", indent(string(output)))
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if !bugged && (nok == 0 || nfail == 0) {
|
|
|
|
bug()
|
|
|
|
fmt.Printf("not enough test cases run\n")
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
func indent(s string) string {
|
|
|
|
return strings.Replace(s, "\n", "\n\t", -1)
|
|
|
|
}
|
|
|
|
|
|
|
|
var bugged = false
|
|
|
|
|
|
|
|
func bug() {
|
|
|
|
if !bugged {
|
|
|
|
bugged = true
|
|
|
|
fmt.Printf("BUG\n")
|
|
|
|
}
|
|
|
|
}
|