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
|
|
|
// errorcheck -0 -m -l
|
|
|
|
|
2016-04-10 21:32:26 +00:00
|
|
|
// Copyright 2015 The Go Authors. All rights reserved.
|
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
|
|
|
// Use of this source code is governed by a BSD-style
|
|
|
|
// license that can be found in the LICENSE file.
|
|
|
|
|
|
|
|
// Test escape analysis for struct function parameters.
|
|
|
|
// Note companion strict_param1 checks *struct function parameters with similar tests.
|
|
|
|
|
|
|
|
package notmain
|
|
|
|
|
|
|
|
var Ssink *string
|
|
|
|
|
|
|
|
type U struct {
|
|
|
|
_sp *string
|
|
|
|
_spp **string
|
|
|
|
}
|
|
|
|
|
|
|
|
type V struct {
|
|
|
|
_u U
|
|
|
|
_up *U
|
|
|
|
_upp **U
|
|
|
|
}
|
|
|
|
|
|
|
|
func (u U) SP() *string { // ERROR "leaking param: u to result ~r0 level=0$"
|
|
|
|
return u._sp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (u U) SPP() **string { // ERROR "leaking param: u to result ~r0 level=0$"
|
|
|
|
return u._spp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (u U) SPPi() *string { // ERROR "leaking param: u to result ~r0 level=1$"
|
|
|
|
return *u._spp
|
|
|
|
}
|
|
|
|
|
|
|
|
func tSPPi() {
|
2020-09-09 04:09:01 +00:00
|
|
|
s := "cat" // ERROR "moved to heap: s$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps := &s
|
|
|
|
pps := &ps
|
2020-09-09 04:09:01 +00:00
|
|
|
pu := &U{ps, pps} // ERROR "&U{...} 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.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720
Fixes #10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
|
|
|
Ssink = pu.SPPi()
|
|
|
|
}
|
|
|
|
|
|
|
|
func tiSPP() {
|
2020-09-09 04:09:01 +00:00
|
|
|
s := "cat" // ERROR "moved to heap: s$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps := &s
|
|
|
|
pps := &ps
|
2020-09-09 04:09:01 +00:00
|
|
|
pu := &U{ps, pps} // ERROR "&U{...} 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.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720
Fixes #10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
|
|
|
Ssink = *pu.SPP()
|
|
|
|
}
|
|
|
|
|
|
|
|
// BAD: need fine-grained analysis to avoid spurious escape of ps
|
|
|
|
func tSP() {
|
2020-09-09 04:09:01 +00:00
|
|
|
s := "cat" // ERROR "moved to heap: s$"
|
|
|
|
ps := &s // ERROR "moved to heap: ps$"
|
2019-04-01 18:58:33 +00:00
|
|
|
pps := &ps
|
2020-09-09 04:09:01 +00:00
|
|
|
pu := &U{ps, pps} // ERROR "&U{...} does not escape$"
|
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
|
|
|
Ssink = pu.SP()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) u() U { // ERROR "leaking param: v to result ~r0 level=0$"
|
|
|
|
return v._u
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UP() *U { // ERROR "leaking param: v to result ~r0 level=0$"
|
|
|
|
return v._up
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPP() **U { // ERROR "leaking param: v to result ~r0 level=0$"
|
|
|
|
return v._upp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPPia() *U { // ERROR "leaking param: v to result ~r0 level=1$"
|
|
|
|
return *v._upp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPPib() *U { // ERROR "leaking param: v to result ~r0 level=1$"
|
|
|
|
return *v.UPP()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) USPa() *string { // ERROR "leaking param: v to result ~r0 level=0$"
|
|
|
|
return v._u._sp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) USPb() *string { // ERROR "leaking param: v to result ~r0 level=0$"
|
|
|
|
return v.u()._sp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) USPPia() *string { // ERROR "leaking param: v to result ~r0 level=1$"
|
|
|
|
return *v._u._spp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) USPPib() *string { // ERROR "leaking param: v to result ~r0 level=1$"
|
|
|
|
return v._u.SPPi()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPiSPa() *string { // ERROR "leaking param: v to result ~r0 level=1$"
|
|
|
|
return v._up._sp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPiSPb() *string { // ERROR "leaking param: v to result ~r0 level=1$"
|
|
|
|
return v._up.SP()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPiSPc() *string { // ERROR "leaking param: v to result ~r0 level=1$"
|
|
|
|
return v.UP()._sp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPiSPd() *string { // ERROR "leaking param: v to result ~r0 level=1$"
|
|
|
|
return v.UP().SP()
|
|
|
|
}
|
|
|
|
|
|
|
|
// BAD: need fine-grained (field-sensitive) analysis to avoid spurious escape of all but &s3
|
|
|
|
func tUPiSPa() {
|
|
|
|
s1 := "ant"
|
2020-09-09 04:09:01 +00:00
|
|
|
s2 := "bat" // ERROR "moved to heap: s2$"
|
|
|
|
s3 := "cat" // ERROR "moved to heap: s3$"
|
|
|
|
s4 := "dog" // ERROR "moved to heap: s4$"
|
|
|
|
s5 := "emu" // ERROR "moved to heap: s5$"
|
|
|
|
s6 := "fox" // ERROR "moved to heap: s6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps2 := &s2
|
2020-09-09 04:09:01 +00:00
|
|
|
ps4 := &s4 // ERROR "moved to heap: ps4$"
|
|
|
|
ps6 := &s6 // ERROR "moved to heap: ps6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
u1 := U{&s1, &ps2}
|
2020-09-09 04:09:01 +00:00
|
|
|
u2 := &U{&s3, &ps4} // ERROR "&U{...} does not escape$"
|
|
|
|
u3 := &U{&s5, &ps6} // ERROR "&U{...} escapes to heap$"
|
|
|
|
v := &V{u1, u2, &u3} // ERROR "&V{...} 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.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720
Fixes #10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
|
|
|
Ssink = v.UPiSPa() // Ssink = &s3 (only &s3 really escapes)
|
|
|
|
}
|
|
|
|
|
|
|
|
// BAD: need fine-grained (field-sensitive) analysis to avoid spurious escape of all but &s3
|
|
|
|
func tUPiSPb() {
|
|
|
|
s1 := "ant"
|
2020-09-09 04:09:01 +00:00
|
|
|
s2 := "bat" // ERROR "moved to heap: s2$"
|
|
|
|
s3 := "cat" // ERROR "moved to heap: s3$"
|
|
|
|
s4 := "dog" // ERROR "moved to heap: s4$"
|
|
|
|
s5 := "emu" // ERROR "moved to heap: s5$"
|
|
|
|
s6 := "fox" // ERROR "moved to heap: s6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps2 := &s2
|
2020-09-09 04:09:01 +00:00
|
|
|
ps4 := &s4 // ERROR "moved to heap: ps4$"
|
|
|
|
ps6 := &s6 // ERROR "moved to heap: ps6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
u1 := U{&s1, &ps2}
|
2020-09-09 04:09:01 +00:00
|
|
|
u2 := &U{&s3, &ps4} // ERROR "&U{...} does not escape$"
|
|
|
|
u3 := &U{&s5, &ps6} // ERROR "&U{...} escapes to heap$"
|
|
|
|
v := &V{u1, u2, &u3} // ERROR "&V{...} 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.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720
Fixes #10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
|
|
|
Ssink = v.UPiSPb() // Ssink = &s3 (only &s3 really escapes)
|
|
|
|
}
|
|
|
|
|
|
|
|
// BAD: need fine-grained (field-sensitive) analysis to avoid spurious escape of all but &s3
|
|
|
|
func tUPiSPc() {
|
|
|
|
s1 := "ant"
|
2020-09-09 04:09:01 +00:00
|
|
|
s2 := "bat" // ERROR "moved to heap: s2$"
|
|
|
|
s3 := "cat" // ERROR "moved to heap: s3$"
|
|
|
|
s4 := "dog" // ERROR "moved to heap: s4$"
|
|
|
|
s5 := "emu" // ERROR "moved to heap: s5$"
|
|
|
|
s6 := "fox" // ERROR "moved to heap: s6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps2 := &s2
|
2020-09-09 04:09:01 +00:00
|
|
|
ps4 := &s4 // ERROR "moved to heap: ps4$"
|
|
|
|
ps6 := &s6 // ERROR "moved to heap: ps6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
u1 := U{&s1, &ps2}
|
2020-09-09 04:09:01 +00:00
|
|
|
u2 := &U{&s3, &ps4} // ERROR "&U{...} does not escape$"
|
|
|
|
u3 := &U{&s5, &ps6} // ERROR "&U{...} escapes to heap$"
|
|
|
|
v := &V{u1, u2, &u3} // ERROR "&V{...} 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.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720
Fixes #10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
|
|
|
Ssink = v.UPiSPc() // Ssink = &s3 (only &s3 really escapes)
|
|
|
|
}
|
|
|
|
|
|
|
|
// BAD: need fine-grained (field-sensitive) analysis to avoid spurious escape of all but &s3
|
|
|
|
func tUPiSPd() {
|
|
|
|
s1 := "ant"
|
2020-09-09 04:09:01 +00:00
|
|
|
s2 := "bat" // ERROR "moved to heap: s2$"
|
|
|
|
s3 := "cat" // ERROR "moved to heap: s3$"
|
|
|
|
s4 := "dog" // ERROR "moved to heap: s4$"
|
|
|
|
s5 := "emu" // ERROR "moved to heap: s5$"
|
|
|
|
s6 := "fox" // ERROR "moved to heap: s6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps2 := &s2
|
2020-09-09 04:09:01 +00:00
|
|
|
ps4 := &s4 // ERROR "moved to heap: ps4$"
|
|
|
|
ps6 := &s6 // ERROR "moved to heap: ps6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
u1 := U{&s1, &ps2}
|
2020-09-09 04:09:01 +00:00
|
|
|
u2 := &U{&s3, &ps4} // ERROR "&U{...} does not escape$"
|
|
|
|
u3 := &U{&s5, &ps6} // ERROR "&U{...} escapes to heap$"
|
|
|
|
v := &V{u1, u2, &u3} // ERROR "&V{...} 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.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720
Fixes #10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
|
|
|
Ssink = v.UPiSPd() // Ssink = &s3 (only &s3 really escapes)
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPiSPPia() *string { // ERROR "leaking param: v to result ~r0 level=2$"
|
|
|
|
return *v._up._spp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPiSPPib() *string { // ERROR "leaking param: v to result ~r0 level=2$"
|
|
|
|
return v._up.SPPi()
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPiSPPic() *string { // ERROR "leaking param: v to result ~r0 level=2$"
|
|
|
|
return *v.UP()._spp
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPiSPPid() *string { // ERROR "leaking param: v to result ~r0 level=2$"
|
|
|
|
return v.UP().SPPi()
|
|
|
|
}
|
|
|
|
|
|
|
|
// BAD: need fine-grained (field-sensitive) analysis to avoid spurious escape of all but &s4
|
|
|
|
func tUPiSPPia() {
|
|
|
|
s1 := "ant"
|
|
|
|
s2 := "bat"
|
|
|
|
s3 := "cat"
|
2020-09-09 04:09:01 +00:00
|
|
|
s4 := "dog" // ERROR "moved to heap: s4$"
|
|
|
|
s5 := "emu" // ERROR "moved to heap: s5$"
|
|
|
|
s6 := "fox" // ERROR "moved to heap: s6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps2 := &s2
|
|
|
|
ps4 := &s4
|
2020-09-09 04:09:01 +00:00
|
|
|
ps6 := &s6 // ERROR "moved to heap: ps6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
u1 := U{&s1, &ps2}
|
2020-09-09 04:09:01 +00:00
|
|
|
u2 := &U{&s3, &ps4} // ERROR "&U{...} does not escape$"
|
|
|
|
u3 := &U{&s5, &ps6} // ERROR "&U{...} does not escape$"
|
|
|
|
v := &V{u1, u2, &u3} // ERROR "&V{...} does not escape$"
|
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
|
|
|
Ssink = v.UPiSPPia() // Ssink = *&ps4 = &s4 (only &s4 really escapes)
|
|
|
|
}
|
|
|
|
|
|
|
|
// BAD: need fine-grained (field-sensitive) analysis to avoid spurious escape of all but &s4
|
|
|
|
func tUPiSPPib() {
|
|
|
|
s1 := "ant"
|
|
|
|
s2 := "bat"
|
|
|
|
s3 := "cat"
|
2020-09-09 04:09:01 +00:00
|
|
|
s4 := "dog" // ERROR "moved to heap: s4$"
|
|
|
|
s5 := "emu" // ERROR "moved to heap: s5$"
|
|
|
|
s6 := "fox" // ERROR "moved to heap: s6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps2 := &s2
|
|
|
|
ps4 := &s4
|
2020-09-09 04:09:01 +00:00
|
|
|
ps6 := &s6 // ERROR "moved to heap: ps6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
u1 := U{&s1, &ps2}
|
2020-09-09 04:09:01 +00:00
|
|
|
u2 := &U{&s3, &ps4} // ERROR "&U{...} does not escape$"
|
|
|
|
u3 := &U{&s5, &ps6} // ERROR "&U{...} does not escape$"
|
|
|
|
v := &V{u1, u2, &u3} // ERROR "&V{...} 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.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720
Fixes #10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
|
|
|
Ssink = v.UPiSPPib() // Ssink = *&ps4 = &s4 (only &s4 really escapes)
|
|
|
|
}
|
|
|
|
|
|
|
|
// BAD: need fine-grained (field-sensitive) analysis to avoid spurious escape of all but &s4
|
|
|
|
func tUPiSPPic() {
|
|
|
|
s1 := "ant"
|
|
|
|
s2 := "bat"
|
|
|
|
s3 := "cat"
|
2020-09-09 04:09:01 +00:00
|
|
|
s4 := "dog" // ERROR "moved to heap: s4$"
|
|
|
|
s5 := "emu" // ERROR "moved to heap: s5$"
|
|
|
|
s6 := "fox" // ERROR "moved to heap: s6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps2 := &s2
|
|
|
|
ps4 := &s4
|
2020-09-09 04:09:01 +00:00
|
|
|
ps6 := &s6 // ERROR "moved to heap: ps6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
u1 := U{&s1, &ps2}
|
2020-09-09 04:09:01 +00:00
|
|
|
u2 := &U{&s3, &ps4} // ERROR "&U{...} does not escape$"
|
|
|
|
u3 := &U{&s5, &ps6} // ERROR "&U{...} does not escape$"
|
|
|
|
v := &V{u1, u2, &u3} // ERROR "&V{...} does not escape$"
|
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
|
|
|
Ssink = v.UPiSPPic() // Ssink = *&ps4 = &s4 (only &s4 really escapes)
|
|
|
|
}
|
|
|
|
|
|
|
|
// BAD: need fine-grained (field-sensitive) analysis to avoid spurious escape of all but &s4
|
|
|
|
func tUPiSPPid() {
|
|
|
|
s1 := "ant"
|
|
|
|
s2 := "bat"
|
|
|
|
s3 := "cat"
|
2020-09-09 04:09:01 +00:00
|
|
|
s4 := "dog" // ERROR "moved to heap: s4$"
|
|
|
|
s5 := "emu" // ERROR "moved to heap: s5$"
|
|
|
|
s6 := "fox" // ERROR "moved to heap: s6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps2 := &s2
|
|
|
|
ps4 := &s4
|
2020-09-09 04:09:01 +00:00
|
|
|
ps6 := &s6 // ERROR "moved to heap: ps6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
u1 := U{&s1, &ps2}
|
2020-09-09 04:09:01 +00:00
|
|
|
u2 := &U{&s3, &ps4} // ERROR "&U{...} does not escape$"
|
|
|
|
u3 := &U{&s5, &ps6} // ERROR "&U{...} does not escape$"
|
|
|
|
v := &V{u1, u2, &u3} // ERROR "&V{...} 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.${i}.prof ;
done
showed a 15% reduction in allocations performed by the compiler.
Update #3753
Update #4720
Fixes #10466
Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432
Reviewed-on: https://go-review.googlesource.com/8202
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 20:36:15 +00:00
|
|
|
Ssink = v.UPiSPPid() // Ssink = *&ps4 = &s4 (only &s4 really escapes)
|
|
|
|
}
|
|
|
|
|
|
|
|
func (v V) UPPiSPPia() *string { // ERROR "leaking param: v to result ~r0 level=3$"
|
|
|
|
return *(*v._upp)._spp
|
|
|
|
}
|
|
|
|
|
|
|
|
// This test isolates the one value that needs to escape, not because
|
|
|
|
// it distinguishes fields but because it knows that &s6 is the only
|
|
|
|
// value reachable by two indirects from v.
|
|
|
|
// The test depends on the level cap in the escape analysis tags
|
|
|
|
// being able to encode that fact.
|
|
|
|
func tUPPiSPPia() { // This test is sensitive to the level cap in function summary results.
|
|
|
|
s1 := "ant"
|
|
|
|
s2 := "bat"
|
|
|
|
s3 := "cat"
|
|
|
|
s4 := "dog"
|
|
|
|
s5 := "emu"
|
2020-09-09 04:09:01 +00:00
|
|
|
s6 := "fox" // ERROR "moved to heap: s6$"
|
2019-04-01 18:58:33 +00:00
|
|
|
ps2 := &s2
|
|
|
|
ps4 := &s4
|
|
|
|
ps6 := &s6
|
|
|
|
u1 := U{&s1, &ps2}
|
2020-09-09 04:09:01 +00:00
|
|
|
u2 := &U{&s3, &ps4} // ERROR "&U{...} does not escape$"
|
|
|
|
u3 := &U{&s5, &ps6} // ERROR "&U{...} does not escape$"
|
|
|
|
v := &V{u1, u2, &u3} // ERROR "&V{...} does not escape$"
|
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
|
|
|
Ssink = v.UPPiSPPia() // Ssink = *&ps6 = &s6 (only &s6 really escapes)
|
|
|
|
}
|