go/test/escape2.go

1827 lines
40 KiB
Go
Raw Normal View History

// errorcheck -0 -m -l
// Copyright 2010 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// Test, using compiler diagnostic flags, that the escape analysis is working.
// Compiles but does not run. Inlining is disabled.
// escape2n.go contains all the same tests but compiles with -N.
package foo
import (
"fmt"
"unsafe"
)
var gxx *int
func foo1(x int) { // ERROR "moved to heap: x$"
gxx = &x // ERROR "&x escapes to heap$"
}
func foo2(yy *int) { // ERROR "leaking param: yy$"
gxx = yy
}
func foo3(x int) *int { // ERROR "moved to heap: x$"
return &x // ERROR "&x escapes to heap$"
}
type T *T
func foo3b(t T) { // ERROR "leaking param: t$"
*t = t
}
// xx isn't going anywhere, so use of yy is ok
func foo4(xx, yy *int) { // ERROR "foo4 xx does not escape$" "foo4 yy does not escape$"
xx = yy
}
// xx isn't going anywhere, so taking address of yy is ok
func foo5(xx **int, yy *int) { // ERROR "foo5 xx does not escape$" "foo5 yy does not escape$"
xx = &yy // ERROR "foo5 &yy does not escape$"
}
func foo6(xx **int, yy *int) { // ERROR "foo6 xx does not escape$" "leaking param: yy$"
*xx = yy
}
func foo7(xx **int, yy *int) { // ERROR "foo7 xx does not escape$" "foo7 yy does not escape$"
**xx = *yy
}
func foo8(xx, yy *int) int { // ERROR "foo8 xx does not escape$" "foo8 yy does not escape$"
xx = yy
return *xx
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo9(xx, yy *int) *int { // ERROR "leaking param: xx to result ~r2 level=0$" "leaking param: yy to result ~r2 level=0$"
xx = yy
return xx
}
func foo10(xx, yy *int) { // ERROR "foo10 xx does not escape$" "foo10 yy does not escape$"
*xx = *yy
}
func foo11() int {
x, y := 0, 42
xx := &x // ERROR "foo11 &x does not escape$"
yy := &y // ERROR "foo11 &y does not escape$"
*xx = *yy
return x
}
var xxx **int
func foo12(yyy **int) { // ERROR "leaking param: yyy$"
xxx = yyy
}
// Must treat yyy as leaking because *yyy leaks, and the escape analysis
// summaries in exported metadata do not distinguish these two cases.
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo13(yyy **int) { // ERROR "leaking param content: yyy$"
*xxx = *yyy
}
func foo14(yyy **int) { // ERROR "foo14 yyy does not escape$"
**xxx = **yyy
}
func foo15(yy *int) { // ERROR "moved to heap: yy$"
xxx = &yy // ERROR "&yy escapes to heap$"
}
func foo16(yy *int) { // ERROR "leaking param: yy$"
*xxx = yy
}
func foo17(yy *int) { // ERROR "foo17 yy does not escape$"
**xxx = *yy
}
func foo18(y int) { // ERROR "moved to heap: y$"
*xxx = &y // ERROR "&y escapes to heap$"
}
func foo19(y int) {
**xxx = y
}
type Bar struct {
i int
ii *int
}
func NewBar() *Bar {
return &Bar{42, nil} // ERROR "&Bar literal escapes to heap$"
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func NewBarp(x *int) *Bar { // ERROR "leaking param: x to result ~r1 level=-1$"
return &Bar{42, x} // ERROR "&Bar literal escapes to heap$"
}
func NewBarp2(x *int) *Bar { // ERROR "NewBarp2 x does not escape$"
return &Bar{*x, nil} // ERROR "&Bar literal escapes to heap$"
}
func (b *Bar) NoLeak() int { // ERROR "\(\*Bar\).NoLeak b does not escape$"
return *(b.ii)
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (b *Bar) Leak() *int { // ERROR "leaking param: b to result ~r0 level=0$"
return &b.i // ERROR "&b.i escapes to heap$"
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (b *Bar) AlsoNoLeak() *int { // ERROR "leaking param: b to result ~r0 level=1$"
return b.ii
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (b Bar) AlsoLeak() *int { // ERROR "leaking param: b to result ~r0 level=0$"
return b.ii
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (b Bar) LeaksToo() *int { // ERROR "leaking param: b to result ~r0 level=0$"
v := 0 // ERROR "moved to heap: v$"
b.ii = &v // ERROR "&v escapes to heap$"
return b.ii
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (b *Bar) LeaksABit() *int { // ERROR "leaking param: b to result ~r0 level=1$"
v := 0 // ERROR "moved to heap: v$"
b.ii = &v // ERROR "&v escapes to heap$"
return b.ii
}
func (b Bar) StillNoLeak() int { // ERROR "Bar.StillNoLeak b does not escape$"
v := 0
b.ii = &v // ERROR "Bar.StillNoLeak &v does not escape$"
return b.i
}
func goLeak(b *Bar) { // ERROR "leaking param: b$"
go b.NoLeak()
}
type Bar2 struct {
i [12]int
ii []int
}
func NewBar2() *Bar2 {
return &Bar2{[12]int{42}, nil} // ERROR "&Bar2 literal escapes to heap$"
}
func (b *Bar2) NoLeak() int { // ERROR "\(\*Bar2\).NoLeak b does not escape$"
return b.i[0]
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (b *Bar2) Leak() []int { // ERROR "leaking param: b to result ~r0 level=0$"
return b.i[:] // ERROR "b.i escapes to heap$"
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (b *Bar2) AlsoNoLeak() []int { // ERROR "leaking param: b to result ~r0 level=1$"
return b.ii[0:1]
}
func (b Bar2) AgainNoLeak() [12]int { // ERROR "Bar2.AgainNoLeak b does not escape$"
return b.i
}
func (b *Bar2) LeakSelf() { // ERROR "leaking param: b$"
b.ii = b.i[0:4] // ERROR "b.i escapes to heap$"
}
func (b *Bar2) LeakSelf2() { // ERROR "leaking param: b$"
var buf []int
buf = b.i[0:] // ERROR "b.i escapes to heap$"
b.ii = buf
}
func foo21() func() int {
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
x := 42
return func() int { // ERROR "func literal escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
return x
}
}
func foo21a() func() int {
x := 42 // ERROR "moved to heap: x$"
return func() int { // ERROR "func literal escapes to heap$"
x++ // ERROR "&x escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
return x
}
}
func foo22() int {
x := 42
return func() int { // ERROR "foo22 func literal does not escape$"
return x
}()
}
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
func foo23(x int) func() int {
return func() int { // ERROR "func literal escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
return x
}
}
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
func foo23a(x int) func() int {
f := func() int { // ERROR "func literal escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
return x
}
return f
}
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
func foo23b(x int) *(func() int) {
f := func() int { return x } // ERROR "func literal escapes to heap$" "moved to heap: f$"
return &f // ERROR "&f escapes to heap$"
}
func foo23c(x int) func() int { // ERROR "moved to heap: x$"
return func() int { // ERROR "func literal escapes to heap$"
x++ // ERROR "&x escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
return x
}
}
func foo24(x int) int {
return func() int { // ERROR "foo24 func literal does not escape$"
return x
}()
}
var x *int
func fooleak(xx *int) int { // ERROR "leaking param: xx$"
x = xx
return *x
}
func foonoleak(xx *int) int { // ERROR "foonoleak xx does not escape$"
return *x + *xx
}
func foo31(x int) int { // ERROR "moved to heap: x$"
return fooleak(&x) // ERROR "&x escapes to heap$"
}
func foo32(x int) int {
return foonoleak(&x) // ERROR "foo32 &x does not escape$"
}
type Foo struct {
xx *int
x int
}
var F Foo
var pf *Foo
func (f *Foo) fooleak() { // ERROR "leaking param: f$"
pf = f
}
func (f *Foo) foonoleak() { // ERROR "\(\*Foo\).foonoleak f does not escape$"
F.x = f.x
}
func (f *Foo) Leak() { // ERROR "leaking param: f$"
f.fooleak()
}
func (f *Foo) NoLeak() { // ERROR "\(\*Foo\).NoLeak f does not escape$"
f.foonoleak()
}
func foo41(x int) { // ERROR "moved to heap: x$"
F.xx = &x // ERROR "&x escapes to heap$"
}
func (f *Foo) foo42(x int) { // ERROR "\(\*Foo\).foo42 f does not escape$" "moved to heap: x$"
f.xx = &x // ERROR "&x escapes to heap$"
}
func foo43(f *Foo, x int) { // ERROR "foo43 f does not escape$" "moved to heap: x$"
f.xx = &x // ERROR "&x escapes to heap$"
}
func foo44(yy *int) { // ERROR "leaking param: yy$"
F.xx = yy
}
func (f *Foo) foo45() { // ERROR "\(\*Foo\).foo45 f does not escape$"
F.x = f.x
}
// See foo13 above for explanation of why f leaks.
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (f *Foo) foo46() { // ERROR "leaking param content: f$"
F.xx = f.xx
}
func (f *Foo) foo47() { // ERROR "leaking param: f$"
f.xx = &f.x // ERROR "&f.x escapes to heap$"
}
var ptrSlice []*int
func foo50(i *int) { // ERROR "leaking param: i$"
ptrSlice[0] = i
}
var ptrMap map[*int]*int
func foo51(i *int) { // ERROR "leaking param: i$"
ptrMap[i] = i
}
func indaddr1(x int) *int { // ERROR "moved to heap: x$"
return &x // ERROR "&x escapes to heap$"
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func indaddr2(x *int) *int { // ERROR "leaking param: x to result ~r1 level=0$"
return *&x // ERROR "indaddr2 &x does not escape$"
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func indaddr3(x *int32) *int { // ERROR "leaking param: x to result ~r1 level=0$"
return *(**int)(unsafe.Pointer(&x)) // ERROR "indaddr3 &x does not escape$"
}
// From package math:
func Float32bits(f float32) uint32 {
return *(*uint32)(unsafe.Pointer(&f)) // ERROR "Float32bits &f does not escape$"
}
func Float32frombits(b uint32) float32 {
return *(*float32)(unsafe.Pointer(&b)) // ERROR "Float32frombits &b does not escape$"
}
func Float64bits(f float64) uint64 {
return *(*uint64)(unsafe.Pointer(&f)) // ERROR "Float64bits &f does not escape$"
}
func Float64frombits(b uint64) float64 {
return *(*float64)(unsafe.Pointer(&b)) // ERROR "Float64frombits &b does not escape$"
}
// contrast with
func float64bitsptr(f float64) *uint64 { // ERROR "moved to heap: f$"
return (*uint64)(unsafe.Pointer(&f)) // ERROR "&f escapes to heap$"
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func float64ptrbitsptr(f *float64) *uint64 { // ERROR "leaking param: f to result ~r1 level=0$"
return (*uint64)(unsafe.Pointer(f))
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func typesw(i interface{}) *int { // ERROR "leaking param: i to result ~r1 level=0$"
switch val := i.(type) {
case *int:
return val
case *int8:
v := int(*val) // ERROR "moved to heap: v$"
return &v // ERROR "&v escapes to heap$"
}
return nil
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func exprsw(i *int) *int { // ERROR "leaking param: i to result ~r1 level=0$"
switch j := i; *j + 110 {
case 12:
return j
case 42:
return nil
}
return nil
}
// assigning to an array element is like assigning to the array
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo60(i *int) *int { // ERROR "leaking param: i to result ~r1 level=0$"
var a [12]*int
a[0] = i
return a[1]
}
func foo60a(i *int) *int { // ERROR "foo60a i does not escape$"
var a [12]*int
a[0] = i
return nil
}
// assigning to a struct field is like assigning to the struct
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo61(i *int) *int { // ERROR "leaking param: i to result ~r1 level=0$"
type S struct {
a, b *int
}
var s S
s.a = i
return s.b
}
func foo61a(i *int) *int { // ERROR "foo61a i does not escape$"
type S struct {
a, b *int
}
var s S
s.a = i
return nil
}
// assigning to a struct field is like assigning to the struct but
// here this subtlety is lost, since s.a counts as an assignment to a
// track-losing dereference.
func foo62(i *int) *int { // ERROR "leaking param: i$"
type S struct {
a, b *int
}
s := new(S) // ERROR "foo62 new\(S\) does not escape$"
s.a = i
return nil // s.b
}
type M interface {
M()
}
func foo63(m M) { // ERROR "foo63 m does not escape$"
}
func foo64(m M) { // ERROR "leaking param: m$"
m.M()
}
func foo64b(m M) { // ERROR "leaking param: m$"
defer m.M()
}
type MV int
func (MV) M() {}
func foo65() {
var mv MV
foo63(&mv) // ERROR "foo65 &mv does not escape$"
}
func foo66() {
var mv MV // ERROR "moved to heap: mv$"
foo64(&mv) // ERROR "&mv escapes to heap$"
}
func foo67() {
var mv MV
foo63(mv) // ERROR "foo67 mv does not escape$"
}
func foo68() {
var mv MV
// escapes but it's an int so irrelevant
foo64(mv) // ERROR "mv escapes to heap$"
}
func foo69(m M) { // ERROR "leaking param: m$"
foo64(m)
}
func foo70(mv1 *MV, m M) { // ERROR "leaking param: m$" "leaking param: mv1$"
m = mv1 // ERROR "mv1 escapes to heap$"
foo64(m)
}
func foo71(x *int) []*int { // ERROR "leaking param: x$"
var y []*int
y = append(y, x)
return y
}
func foo71a(x int) []*int { // ERROR "moved to heap: x$"
var y []*int
y = append(y, &x) // ERROR "&x escapes to heap$"
return y
}
func foo72() {
var x int
var y [1]*int
y[0] = &x // ERROR "foo72 &x does not escape$"
}
func foo72aa() [10]*int {
var x int // ERROR "moved to heap: x$"
var y [10]*int
y[0] = &x // ERROR "&x escapes to heap$"
return y
}
func foo72a() {
var y [10]*int
for i := 0; i < 10; i++ {
// escapes its scope
x := i // ERROR "moved to heap: x$"
y[i] = &x // ERROR "&x escapes to heap$"
}
return
}
func foo72b() [10]*int {
var y [10]*int
for i := 0; i < 10; i++ {
x := i // ERROR "moved to heap: x$"
y[i] = &x // ERROR "&x escapes to heap$"
}
return y
}
// issue 2145
func foo73() {
s := []int{3, 2, 1} // ERROR "foo73 \[\]int literal does not escape$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
for _, v := range s {
vv := v
// actually just escapes its scope
defer func() { // ERROR "func literal escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
println(vv)
}()
}
}
func foo731() {
s := []int{3, 2, 1} // ERROR "foo731 \[\]int literal does not escape$"
for _, v := range s {
vv := v // ERROR "moved to heap: vv$"
// actually just escapes its scope
defer func() { // ERROR "func literal escapes to heap$"
vv = 42 // ERROR "&vv escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
println(vv)
}()
}
}
func foo74() {
s := []int{3, 2, 1} // ERROR "foo74 \[\]int literal does not escape$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
for _, v := range s {
vv := v
// actually just escapes its scope
fn := func() { // ERROR "func literal escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
println(vv)
}
defer fn()
}
}
func foo74a() {
s := []int{3, 2, 1} // ERROR "foo74a \[\]int literal does not escape$"
for _, v := range s {
vv := v // ERROR "moved to heap: vv$"
// actually just escapes its scope
fn := func() { // ERROR "func literal escapes to heap$"
vv += 1 // ERROR "&vv escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
println(vv)
}
defer fn()
}
}
// issue 3975
func foo74b() {
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
var array [3]func()
s := []int{3, 2, 1} // ERROR "foo74b \[\]int literal does not escape$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
for i, v := range s {
vv := v
// actually just escapes its scope
array[i] = func() { // ERROR "func literal escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
println(vv)
}
}
}
func foo74c() {
var array [3]func()
s := []int{3, 2, 1} // ERROR "foo74c \[\]int literal does not escape$"
for i, v := range s {
vv := v // ERROR "moved to heap: vv$"
// actually just escapes its scope
array[i] = func() { // ERROR "func literal escapes to heap$"
println(&vv) // ERROR "&vv escapes to heap$" "foo74c.func1 &vv does not escape$"
}
}
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func myprint(y *int, x ...interface{}) *int { // ERROR "leaking param: y to result ~r2 level=0$" "myprint x does not escape$"
return y
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func myprint1(y *int, x ...interface{}) *interface{} { // ERROR "leaking param: x to result ~r2 level=0$" "myprint1 y does not escape$"
return &x[0] // ERROR "&x\[0\] escapes to heap$"
}
func foo75(z *int) { // ERROR "foo75 z does not escape$"
myprint(z, 1, 2, 3) // ERROR "1 does not escape" "2 does not escape" "3 does not escape" "foo75 ... argument does not escape$"
}
func foo75a(z *int) { // ERROR "foo75a z does not escape$"
myprint1(z, 1, 2, 3) // ERROR "1 does not escape" "2 does not escape" "3 does not escape" "foo75a ... argument does not escape$"
}
func foo75esc(z *int) { // ERROR "leaking param: z$"
gxx = myprint(z, 1, 2, 3) // ERROR "1 does not escape" "2 does not escape" "3 does not escape" "foo75esc ... argument does not escape$"
}
func foo75aesc(z *int) { // ERROR "foo75aesc z does not escape$"
var ppi **interface{} // assignments to pointer dereferences lose track
*ppi = myprint1(z, 1, 2, 3) // ERROR "... argument escapes to heap$" "1 escapes to heap$" "2 escapes to heap$" "3 escapes to heap$"
}
func foo75aesc1(z *int) { // ERROR "foo75aesc1 z does not escape$"
sink = myprint1(z, 1, 2, 3) // ERROR "... argument escapes to heap$" "1 escapes to heap$" "2 escapes to heap$" "3 escapes to heap$" "myprint1\(z, 1, 2, 3\) escapes to heap$"
}
func foo76(z *int) { // ERROR "z does not escape"
myprint(nil, z) // ERROR "foo76 ... argument does not escape$" "z does not escape"
}
func foo76a(z *int) { // ERROR "z does not escape"
myprint1(nil, z) // ERROR "foo76a ... argument does not escape$" "z does not escape"
}
func foo76b() {
myprint(nil, 1, 2, 3) // ERROR "1 does not escape" "2 does not escape" "3 does not escape" "foo76b ... argument does not escape$"
}
func foo76c() {
myprint1(nil, 1, 2, 3) // ERROR "1 does not escape" "2 does not escape" "3 does not escape" "foo76c ... argument does not escape$"
}
func foo76d() {
defer myprint(nil, 1, 2, 3) // ERROR "1 does not escape" "2 does not escape" "3 does not escape" "foo76d ... argument does not escape$"
}
func foo76e() {
defer myprint1(nil, 1, 2, 3) // ERROR "1 does not escape" "2 does not escape" "3 does not escape" "foo76e ... argument does not escape$"
}
func foo76f() {
for {
// TODO: This one really only escapes its scope, but we don't distinguish yet.
defer myprint(nil, 1, 2, 3) // ERROR "... argument escapes to heap$" "1 escapes to heap$" "2 escapes to heap$" "3 escapes to heap$"
}
}
func foo76g() {
for {
defer myprint1(nil, 1, 2, 3) // ERROR "... argument escapes to heap$" "1 escapes to heap$" "2 escapes to heap$" "3 escapes to heap$"
}
}
func foo77(z []interface{}) { // ERROR "foo77 z does not escape$"
myprint(nil, z...) // z does not escape
}
func foo77a(z []interface{}) { // ERROR "foo77a z does not escape$"
myprint1(nil, z...)
}
func foo77b(z []interface{}) { // ERROR "leaking param: z$"
var ppi **interface{}
*ppi = myprint1(nil, z...)
}
func foo77c(z []interface{}) { // ERROR "leaking param: z$"
sink = myprint1(nil, z...) // ERROR "myprint1\(nil, z...\) escapes to heap$"
}
func dotdotdot() {
i := 0
myprint(nil, &i) // ERROR "&i does not escape" "dotdotdot ... argument does not escape$"
j := 0
myprint1(nil, &j) // ERROR "&j does not escape" "dotdotdot ... argument does not escape$"
}
func foo78(z int) *int { // ERROR "moved to heap: z$"
return &z // ERROR "&z escapes to heap$"
}
func foo78a(z int) *int { // ERROR "moved to heap: z$"
y := &z // ERROR "&z escapes to heap$"
x := &y // ERROR "foo78a &y does not escape$"
return *x // really return y
}
func foo79() *int {
return new(int) // ERROR "new\(int\) escapes to heap$"
}
func foo80() *int {
var z *int
for {
// Really just escapes its scope but we don't distinguish
z = new(int) // ERROR "new\(int\) escapes to heap$"
}
_ = z
return nil
}
func foo81() *int {
for {
z := new(int) // ERROR "foo81 new\(int\) does not escape$"
_ = z
}
return nil
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func tee(p *int) (x, y *int) { return p, p } // ERROR "leaking param: p to result x level=0$" "leaking param: p to result y level=0$"
func noop(x, y *int) {} // ERROR "noop x does not escape$" "noop y does not escape$"
func foo82() {
var x, y, z int // ERROR "moved to heap: x$" "moved to heap: y$" "moved to heap: z$"
go noop(tee(&z)) // ERROR "&z escapes to heap$"
go noop(&x, &y) // ERROR "&x escapes to heap$" "&y escapes to heap$"
for {
var u, v, w int // ERROR "moved to heap: u$" "moved to heap: v$" "moved to heap: w$"
defer noop(tee(&u)) // ERROR "&u escapes to heap$"
defer noop(&v, &w) // ERROR "&v escapes to heap$" "&w escapes to heap$"
}
}
type Fooer interface {
Foo()
}
type LimitedFooer struct {
Fooer
N int64
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func LimitFooer(r Fooer, n int64) Fooer { // ERROR "leaking param: r to result ~r2 level=-1$"
return &LimitedFooer{r, n} // ERROR "&LimitedFooer literal escapes to heap$"
}
func foo90(x *int) map[*int]*int { // ERROR "leaking param: x$"
return map[*int]*int{nil: x} // ERROR "map\[\*int\]\*int literal escapes to heap$"
}
func foo91(x *int) map[*int]*int { // ERROR "leaking param: x$"
return map[*int]*int{x: nil} // ERROR "map\[\*int\]\*int literal escapes to heap$"
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo92(x *int) [2]*int { // ERROR "leaking param: x to result ~r1 level=0$"
return [2]*int{x, nil}
}
// does not leak c
func foo93(c chan *int) *int { // ERROR "foo93 c does not escape$"
for v := range c {
return v
}
return nil
}
// does not leak m
func foo94(m map[*int]*int, b bool) *int { // ERROR "leaking param: m to result ~r2 level=1"
for k, v := range m {
if b {
return k
}
return v
}
return nil
}
// does leak x
func foo95(m map[*int]*int, x *int) { // ERROR "foo95 m does not escape$" "leaking param: x$"
m[x] = x
}
// does not leak m but does leak content
func foo96(m []*int) *int { // ERROR "leaking param: m to result ~r1 level=1"
return m[0]
}
// does leak m
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo97(m [1]*int) *int { // ERROR "leaking param: m to result ~r1 level=0$"
return m[0]
}
// does not leak m
func foo98(m map[int]*int) *int { // ERROR "foo98 m does not escape$"
return m[0]
}
// does leak m
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo99(m *[1]*int) []*int { // ERROR "leaking param: m to result ~r1 level=0$"
return m[:]
}
// does not leak m
func foo100(m []*int) *int { // ERROR "leaking param: m to result ~r1 level=1"
for _, v := range m {
return v
}
return nil
}
// does leak m
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo101(m [1]*int) *int { // ERROR "leaking param: m to result ~r1 level=0$"
for _, v := range m {
return v
}
return nil
}
// does not leak m
func foo101a(m [1]*int) *int { // ERROR "foo101a m does not escape$"
for i := range m { // ERROR "moved to heap: i$"
return &i // ERROR "&i escapes to heap$"
}
return nil
}
// does leak x
func foo102(m []*int, x *int) { // ERROR "foo102 m does not escape$" "leaking param: x$"
m[0] = x
}
// does not leak x
func foo103(m [1]*int, x *int) { // ERROR "foo103 m does not escape$" "foo103 x does not escape$"
m[0] = x
}
var y []*int
// does not leak x but does leak content
func foo104(x []*int) { // ERROR "leaking param content: x"
copy(y, x)
}
// does not leak x but does leak content
func foo105(x []*int) { // ERROR "leaking param content: x"
_ = append(y, x...)
}
// does leak x
func foo106(x *int) { // ERROR "leaking param: x$"
_ = append(y, x)
}
func foo107(x *int) map[*int]*int { // ERROR "leaking param: x$"
return map[*int]*int{x: nil} // ERROR "map\[\*int\]\*int literal escapes to heap$"
}
func foo108(x *int) map[*int]*int { // ERROR "leaking param: x$"
return map[*int]*int{nil: x} // ERROR "map\[\*int\]\*int literal escapes to heap$"
}
func foo109(x *int) *int { // ERROR "leaking param: x$"
m := map[*int]*int{x: nil} // ERROR "foo109 map\[\*int\]\*int literal does not escape$"
for k, _ := range m {
return k
}
return nil
}
func foo110(x *int) *int { // ERROR "leaking param: x$"
m := map[*int]*int{nil: x} // ERROR "foo110 map\[\*int\]\*int literal does not escape$"
return m[nil]
}
func foo111(x *int) *int { // ERROR "leaking param: x to result ~r1 level=0"
m := []*int{x} // ERROR "foo111 \[\]\*int literal does not escape$"
return m[0]
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo112(x *int) *int { // ERROR "leaking param: x to result ~r1 level=0$"
m := [1]*int{x}
return m[0]
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo113(x *int) *int { // ERROR "leaking param: x to result ~r1 level=0$"
m := Bar{ii: x}
return m.ii
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo114(x *int) *int { // ERROR "leaking param: x to result ~r1 level=0$"
m := &Bar{ii: x} // ERROR "foo114 &Bar literal does not escape$"
return m.ii
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo115(x *int) *int { // ERROR "leaking param: x to result ~r1 level=0$"
return (*int)(unsafe.Pointer(uintptr(unsafe.Pointer(x)) + 1))
}
func foo116(b bool) *int {
if b {
x := 1 // ERROR "moved to heap: x$"
return &x // ERROR "&x escapes to heap$"
} else {
y := 1 // ERROR "moved to heap: y$"
return &y // ERROR "&y escapes to heap$"
}
return nil
}
func foo117(unknown func(interface{})) { // ERROR "foo117 unknown does not escape$"
x := 1 // ERROR "moved to heap: x$"
unknown(&x) // ERROR "&x escapes to heap$"
}
func foo118(unknown func(*int)) { // ERROR "foo118 unknown does not escape$"
x := 1 // ERROR "moved to heap: x$"
unknown(&x) // ERROR "&x escapes to heap$"
}
func external(*int)
func foo119(x *int) { // ERROR "leaking param: x$"
external(x)
}
func foo120() {
// formerly exponential time analysis
L1:
L2:
L3:
L4:
L5:
L6:
L7:
L8:
L9:
L10:
L11:
L12:
L13:
L14:
L15:
L16:
L17:
L18:
L19:
L20:
L21:
L22:
L23:
L24:
L25:
L26:
L27:
L28:
L29:
L30:
L31:
L32:
L33:
L34:
L35:
L36:
L37:
L38:
L39:
L40:
L41:
L42:
L43:
L44:
L45:
L46:
L47:
L48:
L49:
L50:
L51:
L52:
L53:
L54:
L55:
L56:
L57:
L58:
L59:
L60:
L61:
L62:
L63:
L64:
L65:
L66:
L67:
L68:
L69:
L70:
L71:
L72:
L73:
L74:
L75:
L76:
L77:
L78:
L79:
L80:
L81:
L82:
L83:
L84:
L85:
L86:
L87:
L88:
L89:
L90:
L91:
L92:
L93:
L94:
L95:
L96:
L97:
L98:
L99:
L100:
// use the labels to silence compiler errors
goto L1
goto L2
goto L3
goto L4
goto L5
goto L6
goto L7
goto L8
goto L9
goto L10
goto L11
goto L12
goto L13
goto L14
goto L15
goto L16
goto L17
goto L18
goto L19
goto L20
goto L21
goto L22
goto L23
goto L24
goto L25
goto L26
goto L27
goto L28
goto L29
goto L30
goto L31
goto L32
goto L33
goto L34
goto L35
goto L36
goto L37
goto L38
goto L39
goto L40
goto L41
goto L42
goto L43
goto L44
goto L45
goto L46
goto L47
goto L48
goto L49
goto L50
goto L51
goto L52
goto L53
goto L54
goto L55
goto L56
goto L57
goto L58
goto L59
goto L60
goto L61
goto L62
goto L63
goto L64
goto L65
goto L66
goto L67
goto L68
goto L69
goto L70
goto L71
goto L72
goto L73
goto L74
goto L75
goto L76
goto L77
goto L78
goto L79
goto L80
goto L81
goto L82
goto L83
goto L84
goto L85
goto L86
goto L87
goto L88
goto L89
goto L90
goto L91
goto L92
goto L93
goto L94
goto L95
goto L96
goto L97
goto L98
goto L99
goto L100
}
func foo121() {
for i := 0; i < 10; i++ {
defer myprint(nil, i) // ERROR "... argument escapes to heap$" "i escapes to heap$"
go myprint(nil, i) // ERROR "... argument escapes to heap$" "i escapes to heap$"
}
}
// same as foo121 but check across import
func foo121b() {
for i := 0; i < 10; i++ {
defer fmt.Printf("%d", i) // ERROR "... argument escapes to heap$" "i escapes to heap$"
go fmt.Printf("%d", i) // ERROR "... argument escapes to heap$" "i escapes to heap$"
}
}
// a harmless forward jump
func foo122() {
var i *int
goto L1
L1:
i = new(int) // ERROR "foo122 new\(int\) does not escape$"
_ = i
}
// a backward jump, increases loopdepth
func foo123() {
var i *int
L1:
i = new(int) // ERROR "new\(int\) escapes to heap$"
goto L1
_ = i
}
func foo124(x **int) { // ERROR "foo124 x does not escape$"
var i int // ERROR "moved to heap: i$"
p := &i // ERROR "&i escapes to heap$"
func() { // ERROR "foo124 func literal does not escape$"
*x = p // ERROR "leaking closure reference p$"
}()
}
func foo125(ch chan *int) { // ERROR "foo125 ch does not escape$"
var i int // ERROR "moved to heap: i$"
p := &i // ERROR "&i escapes to heap$"
func() { // ERROR "foo125 func literal does not escape$"
ch <- p // ERROR "leaking closure reference p$"
}()
}
func foo126() {
var px *int // loopdepth 0
for {
// loopdepth 1
var i int // ERROR "moved to heap: i$"
func() { // ERROR "foo126 func literal does not escape$"
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 21:44:20 +00:00
px = &i // ERROR "&i escapes to heap$" "leaking closure reference i"
}()
}
_ = px
}
var px *int
func foo127() {
var i int // ERROR "moved to heap: i$"
p := &i // ERROR "&i escapes to heap$"
q := p
px = q
}
func foo128() {
var i int
p := &i // ERROR "foo128 &i does not escape$"
q := p
_ = q
}
func foo129() {
var i int // ERROR "moved to heap: i$"
p := &i // ERROR "&i escapes to heap$"
func() { // ERROR "foo129 func literal does not escape$"
q := p // ERROR "leaking closure reference p$"
func() { // ERROR "foo129.func1 func literal does not escape$"
r := q // ERROR "leaking closure reference q$"
px = r
}()
}()
}
func foo130() {
for {
var i int // ERROR "moved to heap: i$"
func() { // ERROR "foo130 func literal does not escape$"
px = &i // ERROR "&i escapes to heap$" "leaking closure reference i$"
}()
}
}
func foo131() {
var i int // ERROR "moved to heap: i$"
func() { // ERROR "foo131 func literal does not escape$"
px = &i // ERROR "&i escapes to heap$" "leaking closure reference i$"
}()
}
func foo132() {
var i int // ERROR "moved to heap: i$"
go func() { // ERROR "func literal escapes to heap$"
px = &i // ERROR "&i escapes to heap$" "leaking closure reference i$"
}()
}
func foo133() {
var i int // ERROR "moved to heap: i$"
defer func() { // ERROR "foo133 func literal does not escape$"
px = &i // ERROR "&i escapes to heap$" "leaking closure reference i$"
}()
}
func foo134() {
var i int
p := &i // ERROR "foo134 &i does not escape$"
func() { // ERROR "foo134 func literal does not escape$"
q := p
func() { // ERROR "foo134.func1 func literal does not escape$"
r := q
_ = r
}()
}()
}
func foo135() {
var i int // ERROR "moved to heap: i$"
p := &i // ERROR "&i escapes to heap$"
go func() { // ERROR "func literal escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
q := p
func() { // ERROR "foo135.func1 func literal does not escape$"
r := q
_ = r
}()
}()
}
func foo136() {
var i int // ERROR "moved to heap: i$"
p := &i // ERROR "&i escapes to heap$"
go func() { // ERROR "func literal escapes to heap$"
q := p // ERROR "leaking closure reference p$"
func() { // ERROR "foo136.func1 func literal does not escape$"
r := q // ERROR "leaking closure reference q$"
px = r
}()
}()
}
func foo137() {
var i int // ERROR "moved to heap: i$"
p := &i // ERROR "&i escapes to heap$"
func() { // ERROR "foo137 func literal does not escape$"
q := p // ERROR "leaking closure reference p$"
go func() { // ERROR "func literal escapes to heap$"
cmd/gc: capture variables by value Language specification says that variables are captured by reference. And that is what gc compiler does. However, in lots of cases it is possible to capture variables by value under the hood without affecting visible behavior of programs. For example, consider the following typical pattern: func (o *Obj) requestMany(urls []string) []Result { wg := new(sync.WaitGroup) wg.Add(len(urls)) res := make([]Result, len(urls)) for i := range urls { i := i go func() { res[i] = o.requestOne(urls[i]) wg.Done() }() } wg.Wait() return res } Currently o, wg, res, and i are captured by reference causing 3+len(urls) allocations (e.g. PPARAM o is promoted to PPARAMREF and moved to heap). But all of them can be captured by value without changing behavior. This change implements simple strategy for capturing by value: if a captured variable is not addrtaken and never assigned to, then it is captured by value (it is effectively const). This simple strategy turned out to be very effective: ~80% of all captures in std lib are turned into value captures. The remaining 20% are mostly in defers and non-escaping closures, that is, they do not cause allocations anyway. benchmark old allocs new allocs delta BenchmarkCompressedZipGarbage 153 126 -17.65% BenchmarkEncodeDigitsSpeed1e4 91 69 -24.18% BenchmarkEncodeDigitsSpeed1e5 178 129 -27.53% BenchmarkEncodeDigitsSpeed1e6 1510 1051 -30.40% BenchmarkEncodeDigitsDefault1e4 100 75 -25.00% BenchmarkEncodeDigitsDefault1e5 193 139 -27.98% BenchmarkEncodeDigitsDefault1e6 1420 985 -30.63% BenchmarkEncodeDigitsCompress1e4 100 75 -25.00% BenchmarkEncodeDigitsCompress1e5 193 139 -27.98% BenchmarkEncodeDigitsCompress1e6 1420 985 -30.63% BenchmarkEncodeTwainSpeed1e4 109 81 -25.69% BenchmarkEncodeTwainSpeed1e5 211 151 -28.44% BenchmarkEncodeTwainSpeed1e6 1588 1097 -30.92% BenchmarkEncodeTwainDefault1e4 103 77 -25.24% BenchmarkEncodeTwainDefault1e5 199 143 -28.14% BenchmarkEncodeTwainDefault1e6 1324 917 -30.74% BenchmarkEncodeTwainCompress1e4 103 77 -25.24% BenchmarkEncodeTwainCompress1e5 190 137 -27.89% BenchmarkEncodeTwainCompress1e6 1327 919 -30.75% BenchmarkConcurrentDBExec 16223 16220 -0.02% BenchmarkConcurrentStmtQuery 17687 16182 -8.51% BenchmarkConcurrentStmtExec 5191 5186 -0.10% BenchmarkConcurrentTxQuery 17665 17661 -0.02% BenchmarkConcurrentTxExec 15154 15150 -0.03% BenchmarkConcurrentTxStmtQuery 17661 16157 -8.52% BenchmarkConcurrentTxStmtExec 3677 3673 -0.11% BenchmarkConcurrentRandom 14000 13614 -2.76% BenchmarkManyConcurrentQueries 25 22 -12.00% BenchmarkDecodeComplex128Slice 318 252 -20.75% BenchmarkDecodeFloat64Slice 318 252 -20.75% BenchmarkDecodeInt32Slice 318 252 -20.75% BenchmarkDecodeStringSlice 2318 2252 -2.85% BenchmarkDecode 11 8 -27.27% BenchmarkEncodeGray 64 56 -12.50% BenchmarkEncodeNRGBOpaque 64 56 -12.50% BenchmarkEncodeNRGBA 67 58 -13.43% BenchmarkEncodePaletted 68 60 -11.76% BenchmarkEncodeRGBOpaque 64 56 -12.50% BenchmarkGoLookupIP 153 139 -9.15% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServer 62 59 -4.84% BenchmarkClientServerParallel4 62 59 -4.84% BenchmarkClientServerParallel64 62 59 -4.84% BenchmarkClientServerParallelTLS4 79 76 -3.80% BenchmarkClientServerParallelTLS64 112 109 -2.68% BenchmarkCreateGoroutinesCapture 10 6 -40.00% BenchmarkAfterFunc 1006 1005 -0.10% Fixes #6632. Change-Id: I0cd51e4d356331d7f3c5f447669080cd19b0d2ca Reviewed-on: https://go-review.googlesource.com/3166 Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-19 19:59:58 +00:00
r := q
_ = r
}()
}()
}
func foo138() *byte {
type T struct {
x [1]byte
}
t := new(T) // ERROR "new\(T\) escapes to heap$"
return &t.x[0] // ERROR "&t.x\[0\] escapes to heap$"
}
func foo139() *byte {
type T struct {
x struct {
y byte
}
}
t := new(T) // ERROR "new\(T\) escapes to heap$"
return &t.x.y // ERROR "&t.x.y escapes to heap$"
}
// issue 4751
func foo140() interface{} {
type T struct {
X string
}
type U struct {
X string
T *T
}
t := &T{} // ERROR "&T literal escapes to heap$"
return U{ // ERROR "U literal escapes to heap$"
X: t.X,
T: t,
}
}
//go:noescape
func F1([]byte)
func F2([]byte)
//go:noescape
func F3(x []byte) // ERROR "F3 x does not escape$"
func F4(x []byte)
func G() {
var buf1 [10]byte
F1(buf1[:]) // ERROR "G buf1 does not escape$"
var buf2 [10]byte // ERROR "moved to heap: buf2$"
F2(buf2[:]) // ERROR "buf2 escapes to heap$"
var buf3 [10]byte
F3(buf3[:]) // ERROR "G buf3 does not escape$"
var buf4 [10]byte // ERROR "moved to heap: buf4$"
F4(buf4[:]) // ERROR "buf4 escapes to heap$"
}
type Tm struct {
x int
}
func (t *Tm) M() { // ERROR "\(\*Tm\).M t does not escape$"
}
func foo141() {
var f func()
t := new(Tm) // ERROR "new\(Tm\) escapes to heap$"
f = t.M // ERROR "foo141 t.M does not escape$"
_ = f
}
var gf func()
func foo142() {
t := new(Tm) // ERROR "new\(Tm\) escapes to heap$"
gf = t.M // ERROR "t.M escapes to heap$"
}
// issue 3888.
func foo143() {
for i := 0; i < 1000; i++ {
func() { // ERROR "foo143 func literal does not escape$"
for i := 0; i < 1; i++ {
var t Tm
t.M() // ERROR "foo143.func1 t does not escape$"
}
}()
}
}
// issue 5773
// Check that annotations take effect regardless of whether they
// are before or after the use in the source code.
//go:noescape
func foo144a(*int)
func foo144() {
var x int
foo144a(&x) // ERROR "foo144 &x does not escape$"
var y int
foo144b(&y) // ERROR "foo144 &y does not escape$"
}
//go:noescape
func foo144b(*int)
// issue 7313: for loop init should not be treated as "in loop"
type List struct {
Next *List
}
func foo145(l List) { // ERROR "foo145 l does not escape$"
var p *List
for p = &l; p.Next != nil; p = p.Next { // ERROR "foo145 &l does not escape$"
}
}
func foo146(l List) { // ERROR "foo146 l does not escape$"
var p *List
p = &l // ERROR "foo146 &l does not escape$"
for ; p.Next != nil; p = p.Next {
}
}
func foo147(l List) { // ERROR "foo147 l does not escape$"
var p *List
p = &l // ERROR "foo147 &l does not escape$"
for p.Next != nil {
p = p.Next
}
}
func foo148(l List) { // ERROR "foo148 l does not escape$"
for p := &l; p.Next != nil; p = p.Next { // ERROR "foo148 &l does not escape$"
}
}
// related: address of variable should have depth of variable, not of loop
func foo149(l List) { // ERROR "foo149 l does not escape$"
var p *List
for {
for p = &l; p.Next != nil; p = p.Next { // ERROR "foo149 &l does not escape$"
}
}
}
// issue 7934: missed ... if element type had no pointers
var save150 []byte
func foo150(x ...byte) { // ERROR "leaking param: x$"
save150 = x
}
func bar150() {
foo150(1, 2, 3) // ERROR "... argument escapes to heap$"
}
// issue 7931: bad handling of slice of array
var save151 *int
func foo151(x *int) { // ERROR "leaking param: x$"
save151 = x
}
func bar151() {
var a [64]int // ERROR "moved to heap: a$"
a[4] = 101
foo151(&(&a)[4:8][0]) // ERROR "&\(&a\)\[4:8\]\[0\] escapes to heap$" "&a escapes to heap$"
}
func bar151b() {
var a [10]int // ERROR "moved to heap: a$"
b := a[:] // ERROR "a escapes to heap$"
foo151(&b[4:8][0]) // ERROR "&b\[4:8\]\[0\] escapes to heap$"
}
func bar151c() {
var a [64]int // ERROR "moved to heap: a$"
a[4] = 101
foo151(&(&a)[4:8:8][0]) // ERROR "&\(&a\)\[4:8:8\]\[0\] escapes to heap$" "&a escapes to heap$"
}
func bar151d() {
var a [10]int // ERROR "moved to heap: a$"
b := a[:] // ERROR "a escapes to heap$"
foo151(&b[4:8:8][0]) // ERROR "&b\[4:8:8\]\[0\] escapes to heap$"
}
cmd/gc: fix escape analysis of func returning indirect of parameter I introduced this bug when I changed the escape analysis to run in phases based on call graph dependency order, in order to be more precise about inputs escaping back to outputs (functions returning their arguments). Given func f(z **int) *int { return *z } we were tagging the function as 'z does not escape and is not returned', which is all true, but not enough information. If used as: var x int p := &x q := &p leak(f(q)) then the compiler might try to keep x, p, and q all on the stack, since (according to the recorded information) nothing interesting ends up being passed to leak. In fact since f returns *q = p, &x is passed to leak and x needs to be heap allocated. To trigger the bug, you need a chain that the compiler wants to keep on the stack (like x, p, q above), and you need a function that returns an indirect of its argument, and you need to pass the head of the chain to that function. This doesn't come up very often: this bug has been present since June 2012 (between Go 1 and Go 1.1) and we haven't seen it until now. It helps that most functions that return indirects are getters that are simple enough to be inlined, avoiding the bug. Earlier versions of Go also had the benefit that if &x really wasn't used beyond x's lifetime, nothing broke if you put &x in a heap-allocated structure accidentally. With the new stack copying, though, heap-allocated structures containing &x are not updated when the stack is copied and x moves, leading to crashes in Go 1.3 that were not crashes in Go 1.2 or Go 1.1. The fix is in two parts. First, in the analysis of a function, recognize when a value obtained via indirect of a parameter ends up being returned. Mark those parameters as having content escape back to the return results (but we don't bother to write down which result). Second, when using the analysis to analyze, say, f(q), mark parameters with content escaping as having any indirections escape to the heap. (We don't bother trying to match the content to the return value.) The fix could be less precise (simpler). In the first part we might mark all content-escaping parameters as plain escaping, and then the second part could be dropped. Or we might assume that when calling f(q) all the things pointed at by q escape always (for any f and q). The fix could also be more precise (more complex). We might record the specific mapping from parameter to result along with the number of indirects from the parameter to the thing being returned as the result, and then at the call sites we could set up exactly the right graph for the called function. That would make notleaks(f(q)) be able to keep x on the stack, because the reuslt of f(q) isn't passed to anything that leaks it. The less precise the fix, the more stack allocations become heap allocations. This fix is exactly as precise as it needs to be so that none of the current stack allocations in the standard library turn into heap allocations. Fixes #8120. LGTM=iant R=golang-codereviews, iant CC=golang-codereviews, khr, r https://golang.org/cl/102040046
2014-06-03 15:35:59 +00:00
// issue 8120
type U struct {
s *string
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (u *U) String() *string { // ERROR "leaking param: u to result ~r0 level=1$"
cmd/gc: fix escape analysis of func returning indirect of parameter I introduced this bug when I changed the escape analysis to run in phases based on call graph dependency order, in order to be more precise about inputs escaping back to outputs (functions returning their arguments). Given func f(z **int) *int { return *z } we were tagging the function as 'z does not escape and is not returned', which is all true, but not enough information. If used as: var x int p := &x q := &p leak(f(q)) then the compiler might try to keep x, p, and q all on the stack, since (according to the recorded information) nothing interesting ends up being passed to leak. In fact since f returns *q = p, &x is passed to leak and x needs to be heap allocated. To trigger the bug, you need a chain that the compiler wants to keep on the stack (like x, p, q above), and you need a function that returns an indirect of its argument, and you need to pass the head of the chain to that function. This doesn't come up very often: this bug has been present since June 2012 (between Go 1 and Go 1.1) and we haven't seen it until now. It helps that most functions that return indirects are getters that are simple enough to be inlined, avoiding the bug. Earlier versions of Go also had the benefit that if &x really wasn't used beyond x's lifetime, nothing broke if you put &x in a heap-allocated structure accidentally. With the new stack copying, though, heap-allocated structures containing &x are not updated when the stack is copied and x moves, leading to crashes in Go 1.3 that were not crashes in Go 1.2 or Go 1.1. The fix is in two parts. First, in the analysis of a function, recognize when a value obtained via indirect of a parameter ends up being returned. Mark those parameters as having content escape back to the return results (but we don't bother to write down which result). Second, when using the analysis to analyze, say, f(q), mark parameters with content escaping as having any indirections escape to the heap. (We don't bother trying to match the content to the return value.) The fix could be less precise (simpler). In the first part we might mark all content-escaping parameters as plain escaping, and then the second part could be dropped. Or we might assume that when calling f(q) all the things pointed at by q escape always (for any f and q). The fix could also be more precise (more complex). We might record the specific mapping from parameter to result along with the number of indirects from the parameter to the thing being returned as the result, and then at the call sites we could set up exactly the right graph for the called function. That would make notleaks(f(q)) be able to keep x on the stack, because the reuslt of f(q) isn't passed to anything that leaks it. The less precise the fix, the more stack allocations become heap allocations. This fix is exactly as precise as it needs to be so that none of the current stack allocations in the standard library turn into heap allocations. Fixes #8120. LGTM=iant R=golang-codereviews, iant CC=golang-codereviews, khr, r https://golang.org/cl/102040046
2014-06-03 15:35:59 +00:00
return u.s
}
type V struct {
s *string
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
// BAD -- level of leak ought to be 0
func NewV(u U) *V { // ERROR "leaking param: u to result ~r1 level=-1"
return &V{u.String()} // ERROR "&V literal escapes to heap$" "NewV u does not escape"
cmd/gc: fix escape analysis of func returning indirect of parameter I introduced this bug when I changed the escape analysis to run in phases based on call graph dependency order, in order to be more precise about inputs escaping back to outputs (functions returning their arguments). Given func f(z **int) *int { return *z } we were tagging the function as 'z does not escape and is not returned', which is all true, but not enough information. If used as: var x int p := &x q := &p leak(f(q)) then the compiler might try to keep x, p, and q all on the stack, since (according to the recorded information) nothing interesting ends up being passed to leak. In fact since f returns *q = p, &x is passed to leak and x needs to be heap allocated. To trigger the bug, you need a chain that the compiler wants to keep on the stack (like x, p, q above), and you need a function that returns an indirect of its argument, and you need to pass the head of the chain to that function. This doesn't come up very often: this bug has been present since June 2012 (between Go 1 and Go 1.1) and we haven't seen it until now. It helps that most functions that return indirects are getters that are simple enough to be inlined, avoiding the bug. Earlier versions of Go also had the benefit that if &x really wasn't used beyond x's lifetime, nothing broke if you put &x in a heap-allocated structure accidentally. With the new stack copying, though, heap-allocated structures containing &x are not updated when the stack is copied and x moves, leading to crashes in Go 1.3 that were not crashes in Go 1.2 or Go 1.1. The fix is in two parts. First, in the analysis of a function, recognize when a value obtained via indirect of a parameter ends up being returned. Mark those parameters as having content escape back to the return results (but we don't bother to write down which result). Second, when using the analysis to analyze, say, f(q), mark parameters with content escaping as having any indirections escape to the heap. (We don't bother trying to match the content to the return value.) The fix could be less precise (simpler). In the first part we might mark all content-escaping parameters as plain escaping, and then the second part could be dropped. Or we might assume that when calling f(q) all the things pointed at by q escape always (for any f and q). The fix could also be more precise (more complex). We might record the specific mapping from parameter to result along with the number of indirects from the parameter to the thing being returned as the result, and then at the call sites we could set up exactly the right graph for the called function. That would make notleaks(f(q)) be able to keep x on the stack, because the reuslt of f(q) isn't passed to anything that leaks it. The less precise the fix, the more stack allocations become heap allocations. This fix is exactly as precise as it needs to be so that none of the current stack allocations in the standard library turn into heap allocations. Fixes #8120. LGTM=iant R=golang-codereviews, iant CC=golang-codereviews, khr, r https://golang.org/cl/102040046
2014-06-03 15:35:59 +00:00
}
func foo152() {
a := "a" // ERROR "moved to heap: a$"
u := U{&a} // ERROR "&a escapes to heap$"
cmd/gc: fix escape analysis of func returning indirect of parameter I introduced this bug when I changed the escape analysis to run in phases based on call graph dependency order, in order to be more precise about inputs escaping back to outputs (functions returning their arguments). Given func f(z **int) *int { return *z } we were tagging the function as 'z does not escape and is not returned', which is all true, but not enough information. If used as: var x int p := &x q := &p leak(f(q)) then the compiler might try to keep x, p, and q all on the stack, since (according to the recorded information) nothing interesting ends up being passed to leak. In fact since f returns *q = p, &x is passed to leak and x needs to be heap allocated. To trigger the bug, you need a chain that the compiler wants to keep on the stack (like x, p, q above), and you need a function that returns an indirect of its argument, and you need to pass the head of the chain to that function. This doesn't come up very often: this bug has been present since June 2012 (between Go 1 and Go 1.1) and we haven't seen it until now. It helps that most functions that return indirects are getters that are simple enough to be inlined, avoiding the bug. Earlier versions of Go also had the benefit that if &x really wasn't used beyond x's lifetime, nothing broke if you put &x in a heap-allocated structure accidentally. With the new stack copying, though, heap-allocated structures containing &x are not updated when the stack is copied and x moves, leading to crashes in Go 1.3 that were not crashes in Go 1.2 or Go 1.1. The fix is in two parts. First, in the analysis of a function, recognize when a value obtained via indirect of a parameter ends up being returned. Mark those parameters as having content escape back to the return results (but we don't bother to write down which result). Second, when using the analysis to analyze, say, f(q), mark parameters with content escaping as having any indirections escape to the heap. (We don't bother trying to match the content to the return value.) The fix could be less precise (simpler). In the first part we might mark all content-escaping parameters as plain escaping, and then the second part could be dropped. Or we might assume that when calling f(q) all the things pointed at by q escape always (for any f and q). The fix could also be more precise (more complex). We might record the specific mapping from parameter to result along with the number of indirects from the parameter to the thing being returned as the result, and then at the call sites we could set up exactly the right graph for the called function. That would make notleaks(f(q)) be able to keep x on the stack, because the reuslt of f(q) isn't passed to anything that leaks it. The less precise the fix, the more stack allocations become heap allocations. This fix is exactly as precise as it needs to be so that none of the current stack allocations in the standard library turn into heap allocations. Fixes #8120. LGTM=iant R=golang-codereviews, iant CC=golang-codereviews, khr, r https://golang.org/cl/102040046
2014-06-03 15:35:59 +00:00
v := NewV(u)
println(v)
}
// issue 8176 - &x in type switch body not marked as escaping
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func foo153(v interface{}) *int { // ERROR "leaking param: v to result ~r1 level=-1$"
switch x := v.(type) {
case int: // ERROR "moved to heap: x$"
return &x // ERROR "&x escapes to heap$"
}
panic(0)
}
// issue 8185 - &result escaping into result
func f() (x int, y *int) { // ERROR "moved to heap: x$"
y = &x // ERROR "&x escapes to heap$"
return
}
func g() (x interface{}) { // ERROR "moved to heap: x$"
x = &x // ERROR "&x escapes to heap$"
return
}
var sink interface{}
type Lit struct {
p *int
}
func ptrlitNoescape() {
// Both literal and element do not escape.
i := 0
x := &Lit{&i} // ERROR "ptrlitNoescape &Lit literal does not escape$" "ptrlitNoescape &i does not escape$"
_ = x
}
func ptrlitNoEscape2() {
// Literal does not escape, but element does.
i := 0 // ERROR "moved to heap: i$"
x := &Lit{&i} // ERROR "&i escapes to heap$" "ptrlitNoEscape2 &Lit literal does not escape$"
sink = *x // ERROR "\*x escapes to heap$"
}
func ptrlitEscape() {
// Both literal and element escape.
i := 0 // ERROR "moved to heap: i$"
x := &Lit{&i} // ERROR "&Lit literal escapes to heap$" "&i escapes to heap$"
sink = x // ERROR "x escapes to heap$"
}
// self-assignments
type Buffer struct {
arr [64]byte
buf1 []byte
buf2 []byte
str1 string
str2 string
}
func (b *Buffer) foo() { // ERROR "\(\*Buffer\).foo b does not escape$"
b.buf1 = b.buf1[1:2] // ERROR "\(\*Buffer\).foo ignoring self-assignment to b.buf1$"
b.buf1 = b.buf1[1:2:3] // ERROR "\(\*Buffer\).foo ignoring self-assignment to b.buf1$"
b.buf1 = b.buf2[1:2] // ERROR "\(\*Buffer\).foo ignoring self-assignment to b.buf1$"
b.buf1 = b.buf2[1:2:3] // ERROR "\(\*Buffer\).foo ignoring self-assignment to b.buf1$"
}
func (b *Buffer) bar() { // ERROR "leaking param: b$"
b.buf1 = b.arr[1:2] // ERROR "b.arr escapes to heap$"
}
func (b *Buffer) baz() { // ERROR "\(\*Buffer\).baz b does not escape$"
b.str1 = b.str1[1:2] // ERROR "\(\*Buffer\).baz ignoring self-assignment to b.str1$"
b.str1 = b.str2[1:2] // ERROR "\(\*Buffer\).baz ignoring self-assignment to b.str1$"
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func (b *Buffer) bat() { // ERROR "leaking param content: b$"
o := new(Buffer) // ERROR "new\(Buffer\) escapes to heap$"
o.buf1 = b.buf1[1:2]
sink = o // ERROR "o escapes to heap$"
}
func quux(sp *string, bp *[]byte) { // ERROR "quux bp does not escape$" "quux sp does not escape$"
*sp = (*sp)[1:2] // ERROR "quux ignoring self-assignment to \*sp$"
*bp = (*bp)[1:2] // ERROR "quux ignoring self-assignment to \*bp$"
}
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
type StructWithString struct {
p *int
s string
}
// This is escape analysis false negative.
// We assign the pointer to x.p but leak x.s. Escape analysis coarsens flows
// to just x, and thus &i looks escaping.
func fieldFlowTracking() {
var x StructWithString
i := 0 // ERROR "moved to heap: i$"
x.p = &i // ERROR "&i escapes to heap$"
sink = x.s // ERROR "x.s escapes to heap$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
}
// String operations.
func slicebytetostring0() {
b := make([]byte, 20) // ERROR "slicebytetostring0 make\(\[\]byte, 20\) does not escape$"
s := string(b) // ERROR "slicebytetostring0 string\(b\) does not escape$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
_ = s
}
func slicebytetostring1() {
b := make([]byte, 20) // ERROR "slicebytetostring1 make\(\[\]byte, 20\) does not escape$"
s := string(b) // ERROR "slicebytetostring1 string\(b\) does not escape$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
s1 := s[0:1]
_ = s1
}
func slicebytetostring2() {
b := make([]byte, 20) // ERROR "slicebytetostring2 make\(\[\]byte, 20\) does not escape$"
s := string(b) // ERROR "string\(b\) escapes to heap$"
s1 := s[0:1] // ERROR "moved to heap: s1$"
sink = &s1 // ERROR "&s1 escapes to heap$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
}
func slicebytetostring3() {
b := make([]byte, 20) // ERROR "slicebytetostring3 make\(\[\]byte, 20\) does not escape$"
s := string(b) // ERROR "string\(b\) escapes to heap$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
s1 := s[0:1]
sink = s1 // ERROR "s1 escapes to heap$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
}
func addstr0() {
s0 := "a"
s1 := "b"
s := s0 + s1 // ERROR "addstr0 s0 \+ s1 does not escape$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
_ = s
}
func addstr1() {
s0 := "a"
s1 := "b"
s := "c"
s += s0 + s1 // ERROR "addstr1 s0 \+ s1 does not escape$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
_ = s
}
func addstr2() {
b := make([]byte, 20) // ERROR "addstr2 make\(\[\]byte, 20\) does not escape$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
s0 := "a"
s := string(b) + s0 // ERROR "addstr2 string\(b\) \+ s0 does not escape$" "addstr2 string\(b\) does not escape$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
_ = s
}
func addstr3() {
s0 := "a"
s1 := "b"
s := s0 + s1 // ERROR "s0 \+ s1 escapes to heap$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
s2 := s[0:1]
sink = s2 // ERROR "s2 escapes to heap$"
cmd/gc: allocate buffers for non-escaped strings on stack Currently we always allocate string buffers in heap. For example, in the following code we allocate a temp string just for comparison: if string(byteSlice) == "abc" { ... } This change extends escape analysis to cover []byte->string conversions and string concatenation. If the result of operations does not escape, compiler allocates a small buffer on stack and passes it to slicebytetostring and concatstrings. Then runtime uses the buffer if the result fits into it. Size of the buffer is 32 bytes. There is no fundamental theory behind this number. Just an observation that on std lib tests/benchmarks frequency of string allocation is inversely proportional to string length; and there is significant number of allocations up to length 32. benchmark old allocs new allocs delta BenchmarkFprintfBytes 2 1 -50.00% BenchmarkDecodeComplex128Slice 318 316 -0.63% BenchmarkDecodeFloat64Slice 318 316 -0.63% BenchmarkDecodeInt32Slice 318 316 -0.63% BenchmarkDecodeStringSlice 2318 2316 -0.09% BenchmarkStripTags 11 5 -54.55% BenchmarkDecodeGray 111 102 -8.11% BenchmarkDecodeNRGBAGradient 200 188 -6.00% BenchmarkDecodeNRGBAOpaque 165 152 -7.88% BenchmarkDecodePaletted 319 309 -3.13% BenchmarkDecodeRGB 166 157 -5.42% BenchmarkDecodeInterlacing 279 268 -3.94% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% BenchmarkGoLookupIPWithBrokenNameServer 245 226 -7.76% BenchmarkClientServerParallel4 62 61 -1.61% BenchmarkClientServerParallel64 62 61 -1.61% BenchmarkClientServerParallelTLS4 79 78 -1.27% BenchmarkClientServerParallelTLS64 112 111 -0.89% benchmark old ns/op new ns/op delta BenchmarkFprintfBytes 381 311 -18.37% BenchmarkStripTags 2615 2351 -10.10% BenchmarkDecodeNRGBAGradient 3715887 3635096 -2.17% BenchmarkDecodeNRGBAOpaque 3047645 2928644 -3.90% BenchmarkGoLookupIP 153 135 -11.76% BenchmarkGoLookupIPNoSuchHost 508 466 -8.27% Change-Id: I9ec01da816945c3329d7be3c7794b520418c3f99 Reviewed-on: https://go-review.googlesource.com/3120 Reviewed-by: Keith Randall <khr@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-01-21 14:37:59 +00:00
}
func intstring0() bool {
// string does not escape
x := '0'
s := string(x) // ERROR "intstring0 string\(x\) does not escape$"
return s == "0"
}
func intstring1() string {
// string does not escape, but the buffer does
x := '0'
s := string(x) // ERROR "string\(x\) escapes to heap$"
return s
}
func intstring2() {
// string escapes to heap
x := '0'
s := string(x) // ERROR "moved to heap: s$" "string\(x\) escapes to heap$"
sink = &s // ERROR "&s escapes to heap$"
}
func stringtoslicebyte0() {
s := "foo"
x := []byte(s) // ERROR "stringtoslicebyte0 \(\[\]byte\)\(s\) does not escape$"
_ = x
}
func stringtoslicebyte1() []byte {
s := "foo"
return []byte(s) // ERROR "\(\[\]byte\)\(s\) escapes to heap$"
}
func stringtoslicebyte2() {
s := "foo"
sink = []byte(s) // ERROR "\(\[\]byte\)\(s\) escapes to heap$"
}
func stringtoslicerune0() {
s := "foo"
x := []rune(s) // ERROR "stringtoslicerune0 \(\[\]rune\)\(s\) does not escape$"
_ = x
}
func stringtoslicerune1() []rune {
s := "foo"
return []rune(s) // ERROR "\(\[\]rune\)\(s\) escapes to heap$"
}
func stringtoslicerune2() {
s := "foo"
sink = []rune(s) // ERROR "\(\[\]rune\)\(s\) escapes to heap$"
}
func slicerunetostring0() {
r := []rune{1, 2, 3} // ERROR "slicerunetostring0 \[\]rune literal does not escape$"
s := string(r) // ERROR "slicerunetostring0 string\(r\) does not escape$"
_ = s
}
func slicerunetostring1() string {
r := []rune{1, 2, 3} // ERROR "slicerunetostring1 \[\]rune literal does not escape$"
return string(r) // ERROR "string\(r\) escapes to heap$"
}
func slicerunetostring2() {
r := []rune{1, 2, 3} // ERROR "slicerunetostring2 \[\]rune literal does not escape$"
sink = string(r) // ERROR "string\(r\) escapes to heap$"
}
func makemap0() {
m := make(map[int]int) // ERROR "makemap0 make\(map\[int\]int\) does not escape$"
m[0] = 0
m[1]++
delete(m, 1)
sink = m[0] // ERROR "m\[0\] escapes to heap$"
}
func makemap1() map[int]int {
return make(map[int]int) // ERROR "make\(map\[int\]int\) escapes to heap$"
}
func makemap2() {
m := make(map[int]int) // ERROR "make\(map\[int\]int\) escapes to heap$"
sink = m // ERROR "m escapes to heap$"
}
func nonescapingEface(m map[interface{}]bool) bool { // ERROR "nonescapingEface m does not escape$"
return m["foo"] // ERROR "nonescapingEface .foo. does not escape$"
}
func nonescapingIface(m map[M]bool) bool { // ERROR "nonescapingIface m does not escape$"
return m[MV(0)] // ERROR "nonescapingIface MV\(0\) does not escape$"
}
cmd/gc: fix escape analysis of closures Fixes #10353 See test/escape2.go:issue10353. Previously new(int) did not escape to heap, and so heap-allcated closure was referencing a stack var. This breaks the invariant that heap must not contain pointers to stack. Look at the following program: package main func main() { foo(new(int)) bar(new(int)) } func foo(x *int) func() { return func() { println(*x) } } // Models what foo effectively does. func bar(x *int) *C { return &C{x} } type C struct { x *int } Without this patch escape analysis works as follows: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:0 depth:1 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x to result ~r1 escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:5: new(int) escapes to heap /tmp/live2.go:4: main new(int) does not escape new(int) does not escape while being captured by the closure. With this patch escape analysis of foo and bar works similarly: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 &(func literal)( l(9)) scope:foo[0] escwalk: level:-1 depth:1 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:-1 depth:2 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:4: new(int) escapes to heap /tmp/live2.go:5: new(int) escapes to heap Change-Id: Ifd14b7ae3fc11820e3b5eb31eb07f35a22ed0932 Reviewed-on: https://go-review.googlesource.com/8408 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Dmitry Vyukov <dvyukov@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-04-06 15:17:20 +00:00
func issue10353() {
x := new(int) // ERROR "new\(int\) escapes to heap$"
cmd/gc: fix escape analysis of closures Fixes #10353 See test/escape2.go:issue10353. Previously new(int) did not escape to heap, and so heap-allcated closure was referencing a stack var. This breaks the invariant that heap must not contain pointers to stack. Look at the following program: package main func main() { foo(new(int)) bar(new(int)) } func foo(x *int) func() { return func() { println(*x) } } // Models what foo effectively does. func bar(x *int) *C { return &C{x} } type C struct { x *int } Without this patch escape analysis works as follows: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:0 depth:1 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x to result ~r1 escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:5: new(int) escapes to heap /tmp/live2.go:4: main new(int) does not escape new(int) does not escape while being captured by the closure. With this patch escape analysis of foo and bar works similarly: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 &(func literal)( l(9)) scope:foo[0] escwalk: level:-1 depth:1 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:-1 depth:2 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:4: new(int) escapes to heap /tmp/live2.go:5: new(int) escapes to heap Change-Id: Ifd14b7ae3fc11820e3b5eb31eb07f35a22ed0932 Reviewed-on: https://go-review.googlesource.com/8408 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Dmitry Vyukov <dvyukov@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-04-06 15:17:20 +00:00
issue10353a(x)()
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
func issue10353a(x *int) func() { // ERROR "leaking param: x to result ~r1 level=-1$"
return func() { // ERROR "func literal escapes to heap$"
cmd/gc: fix escape analysis of closures Fixes #10353 See test/escape2.go:issue10353. Previously new(int) did not escape to heap, and so heap-allcated closure was referencing a stack var. This breaks the invariant that heap must not contain pointers to stack. Look at the following program: package main func main() { foo(new(int)) bar(new(int)) } func foo(x *int) func() { return func() { println(*x) } } // Models what foo effectively does. func bar(x *int) *C { return &C{x} } type C struct { x *int } Without this patch escape analysis works as follows: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:0 depth:1 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x to result ~r1 escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:5: new(int) escapes to heap /tmp/live2.go:4: main new(int) does not escape new(int) does not escape while being captured by the closure. With this patch escape analysis of foo and bar works similarly: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 &(func literal)( l(9)) scope:foo[0] escwalk: level:-1 depth:1 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:-1 depth:2 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:4: new(int) escapes to heap /tmp/live2.go:5: new(int) escapes to heap Change-Id: Ifd14b7ae3fc11820e3b5eb31eb07f35a22ed0932 Reviewed-on: https://go-review.googlesource.com/8408 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Dmitry Vyukov <dvyukov@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-04-06 15:17:20 +00:00
println(*x)
}
}
func issue10353b() {
var f func()
for {
x := new(int) // ERROR "new\(int\) escapes to heap$"
f = func() { // ERROR "func literal escapes to heap$"
cmd/gc: fix escape analysis of closures Fixes #10353 See test/escape2.go:issue10353. Previously new(int) did not escape to heap, and so heap-allcated closure was referencing a stack var. This breaks the invariant that heap must not contain pointers to stack. Look at the following program: package main func main() { foo(new(int)) bar(new(int)) } func foo(x *int) func() { return func() { println(*x) } } // Models what foo effectively does. func bar(x *int) *C { return &C{x} } type C struct { x *int } Without this patch escape analysis works as follows: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:0 depth:1 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x to result ~r1 escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:5: new(int) escapes to heap /tmp/live2.go:4: main new(int) does not escape new(int) does not escape while being captured by the closure. With this patch escape analysis of foo and bar works similarly: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 &(func literal)( l(9)) scope:foo[0] escwalk: level:-1 depth:1 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:-1 depth:2 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:4: new(int) escapes to heap /tmp/live2.go:5: new(int) escapes to heap Change-Id: Ifd14b7ae3fc11820e3b5eb31eb07f35a22ed0932 Reviewed-on: https://go-review.googlesource.com/8408 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Dmitry Vyukov <dvyukov@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-04-06 15:17:20 +00:00
println(*x)
}
}
_ = f
}
func issue11387(x int) func() int {
f := func() int { return x } // ERROR "func literal escapes to heap"
slice1 := []func() int{f} // ERROR "\[\].* does not escape"
slice2 := make([]func() int, 1) // ERROR "make\(.*\) does not escape"
copy(slice2, slice1)
return slice2[0]
}