diff options
| author | Ben Schwartz <bemasc@google.com> | 2019-06-10 12:29:23 -0400 |
|---|---|---|
| committer | Ian Lance Taylor <iant@golang.org> | 2020-03-22 20:37:22 +0000 |
| commit | 02e492f1a33033d88c2c307262a56e578939fc60 (patch) | |
| tree | fb89216861a8ea5edffdf7cf2439d1756719136a /src/runtime/chan_test.go | |
| parent | 9a2db7c41ba5b1f9498d5354e6c2570610f626fb (diff) | |
| download | go-02e492f1a33033d88c2c307262a56e578939fc60.tar.gz go-02e492f1a33033d88c2c307262a56e578939fc60.zip | |
runtime: speed up receive on empty closed channel
Currently, nonblocking receive on an open channel is about
700 times faster than nonblocking receive on a closed channel.
This change makes closed channels equally fast.
Fixes #32529. Includes a correction based on #36714.
relevant benchstat output:
name old time/op new time/op delta
MakeChan/Byte-40 140ns ± 4% 137ns ± 7% -2.38% (p=0.023 n=17+19)
MakeChan/Int-40 174ns ± 5% 173ns ± 6% ~ (p=0.437 n=18+19)
MakeChan/Ptr-40 315ns ±15% 301ns ±15% ~ (p=0.051 n=20+20)
MakeChan/Struct/0-40 123ns ± 8% 99ns ±11% -19.18% (p=0.000 n=20+17)
MakeChan/Struct/32-40 297ns ± 8% 241ns ±18% -19.13% (p=0.000 n=20+20)
MakeChan/Struct/40-40 344ns ± 5% 273ns ±23% -20.49% (p=0.000 n=20+20)
ChanNonblocking-40 0.32ns ± 2% 0.32ns ± 2% -1.25% (p=0.000 n=19+18)
SelectUncontended-40 5.72ns ± 1% 5.71ns ± 2% ~ (p=0.326 n=19+19)
SelectSyncContended-40 10.9µs ±10% 10.6µs ± 3% -2.77% (p=0.009 n=20+16)
SelectAsyncContended-40 1.00µs ± 0% 1.10µs ± 0% +10.75% (p=0.000 n=18+19)
SelectNonblock-40 1.22ns ± 2% 1.21ns ± 4% ~ (p=0.141 n=18+19)
ChanUncontended-40 240ns ± 4% 233ns ± 4% -2.82% (p=0.000 n=20+20)
ChanContended-40 86.7µs ± 0% 82.7µs ± 0% -4.64% (p=0.000 n=20+19)
ChanSync-40 294ns ± 7% 284ns ± 9% -3.44% (p=0.006 n=20+20)
ChanSyncWork-40 38.4µs ±19% 34.0µs ± 4% -11.33% (p=0.000 n=20+18)
ChanProdCons0-40 1.50µs ± 1% 1.63µs ± 0% +8.53% (p=0.000 n=19+19)
ChanProdCons10-40 1.17µs ± 0% 1.18µs ± 1% +0.44% (p=0.000 n=19+20)
ChanProdCons100-40 985ns ± 0% 959ns ± 1% -2.64% (p=0.000 n=20+20)
ChanProdConsWork0-40 1.50µs ± 0% 1.60µs ± 2% +6.54% (p=0.000 n=18+20)
ChanProdConsWork10-40 1.26µs ± 0% 1.26µs ± 2% +0.40% (p=0.015 n=20+19)
ChanProdConsWork100-40 1.27µs ± 0% 1.22µs ± 0% -4.15% (p=0.000 n=20+19)
SelectProdCons-40 1.50µs ± 1% 1.53µs ± 1% +1.95% (p=0.000 n=20+20)
ChanCreation-40 82.1ns ± 5% 81.6ns ± 7% ~ (p=0.483 n=19+19)
ChanSem-40 877ns ± 0% 719ns ± 0% -17.98% (p=0.000 n=18+19)
ChanPopular-40 1.75ms ± 2% 1.78ms ± 3% +1.76% (p=0.002 n=20+19)
ChanClosed-40 215ns ± 1% 0ns ± 6% -99.82% (p=0.000 n=20+18)
Previously committed in CL 181543 and reverted in CL 216158.
Change-Id: Ib767b08d724cfad03598d77271dbc1087485feb8
Reviewed-on: https://go-review.googlesource.com/c/go/+/216818
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Diffstat (limited to 'src/runtime/chan_test.go')
| -rw-r--r-- | src/runtime/chan_test.go | 14 |
1 files changed, 14 insertions, 0 deletions
diff --git a/src/runtime/chan_test.go b/src/runtime/chan_test.go index 1180e76fcd..039a086e9b 100644 --- a/src/runtime/chan_test.go +++ b/src/runtime/chan_test.go @@ -1127,6 +1127,20 @@ func BenchmarkChanPopular(b *testing.B) { wg.Wait() } +func BenchmarkChanClosed(b *testing.B) { + c := make(chan struct{}) + close(c) + b.RunParallel(func(pb *testing.PB) { + for pb.Next() { + select { + case <-c: + default: + b.Error("Unreachable") + } + } + }) +} + var ( alwaysFalse = false workSink = 0 |