mirror of https://github.com/golang/go.git
2 Commits
| Author | SHA1 | Message | Date |
|---|---|---|---|
David Chase
|
7fbb1b36c3 |
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>
|
|
Dmitry Vyukov
|
878a86a129 |
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> |