Hi Peter, Running your code locally (Go 1.8.1 / linux / amd64), I don't get a statistically significant difference between the two. The benchmark does generate a ton of garbage so the GC causes the benchmark runs to have a pretty decent variance.
To check this kind of thing, you'll want to run the benchmarks many times and then compare the distributions using a tool like benchstat: https://godoc.org/golang.org/x/perf/cmd/benchstat For example: go test -bench . -count 10 | tee old.txt # make changes go test -bench . -count 10 | tee new.txt benchstat old.txt new.txt When I've seen effects like this before, one explanation I've seen is that the code alignment can change performance characteristics (this only applies to the tightest of loops, IIRC). Too lazy to look up references right now, but I've definitely seen this discussed on this mailing list as well as various issues. -Caleb On Tue, Apr 25, 2017 at 4:09 PM, Peter Fröhlich <peter.hans.froehl...@gmail.com> wrote: > Hi all, > > Sorry if this is an FAQ, but then my Google-foo has failed me. :-/ > > It seems to me that the test functions in a bla_test.go file influence > the benchmark functions in the same file. In the snippet below, if I > comment out the body of the test function, I get better benchmark > numbers than when I leave it in. Suspecting some GC "leftovers" from > the test, I've tried adding "runtime.GC(); b.ResetTimer()" to the > benchmark but that doesn't actually help. > > Before I dig through the source for Go's testing module itself, I > wanted to see if this is a well-known issue, maybe with a well-known > workaround? > > Best, > Peter > > -----CUT----- > $ go version > go version go1.7.5 linux/amd64 > $ cat bla_test.go > package main > > import "testing" > import "container/list" > import "runtime" > > const size = 2270 > > func TestPushBack(t *testing.T) { > // /* > var q list.List > for i := 0; i < size; i++ { > q.PushBack(i) > } > for i := 0; q.Len() > 0; i++ { > e := q.Front() > x := e.Value.(int) > q.Remove(e) > if x != i { > t.Errorf("q.PopFront() = %d, want %d", x, i) > } > } > // */ > } > > func BenchmarkPushBack(b *testing.B) { > runtime.GC() > b.ResetTimer() > for i := 0; i < b.N; i++ { > var q list.List > for n := 0; n < size; n++ { > q.PushBack(n) > } > } > } > $ go test -bench . > BenchmarkPushBack-2 5000 378367 ns/op > PASS > ok X 1.937s > $ go test -bench . > BenchmarkPushBack-2 5000 378196 ns/op > PASS > ok X 1.936s > $ vi bla_test.go # comment out test code > $ go test -bench . > BenchmarkPushBack-2 5000 356310 ns/op > PASS > ok X 1.825s > $ go test -bench . > BenchmarkPushBack-2 5000 354530 ns/op > PASS > ok X 1.816s > > -- > You received this message because you are subscribed to the Google Groups > "golang-nuts" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to golang-nuts+unsubscr...@googlegroups.com. > For more options, visit https://groups.google.com/d/optout. -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
