I am unable to reproduce your results on AMD64. I don't see any regression.
Run your benchmarks with option -benchmem.
Version: devel go1.20-1eeb257b88 Tue Sep 20 02:58:09 2022 +0000
Sizeof(Array{}): 16
Sizeof(int(0)): 8
goarch: amd64
BenchmarkSliceOfInt-4 391646 2583 ns/op 8192 B/op 1 allocs/op
BenchmarkSliceOfArray-4 236977 4869 ns/op 16384 B/op 1 allocs/op
Version: go1.17.13
Sizeof(Array{}): 16
Sizeof(int(0)): 8
goarch: amd64
BenchmarkSliceOfInt-4 426343 2383 ns/op 8192 B/op 1 allocs/op
BenchmarkSliceOfArray-4 233817 4771 ns/op 16384 B/op 1 allocs/op
The difference is type int is 8 bytes and type [2]int (type Array) is 16
bytes.
Change the type of Array to 8 bytes (type [1]int).
Version: devel go1.20-1eeb257b88 Tue Sep 20 02:58:09 2022 +0000
Sizeof(Array{}): 8
Sizeof(int(0)): 8
goarch: amd64
BenchmarkSliceOfInt-4 405793 2562 ns/op 8192 B/op 1 allocs/op
BenchmarkSliceOfArray-4 469999 2540 ns/op 8192 B/op 1 allocs/op
Version: go1.17.13
Sizeof(Array{}): 8
Sizeof(int(0)): 8
goarch: amd64
BenchmarkSliceOfInt-4 440340 2378 ns/op 8192 B/op 1 allocs/op
BenchmarkSliceOfArray-4 481071 2400 ns/op 8192 B/op 1 allocs/op
There is no significant difference.
peter
On Monday, September 19, 2022 at 10:01:37 PM UTC-4 jpsw...@gmail.com wrote:
> I stumbled across a performance regression on ARM64 for go 1.18.6 and
> 1.19.1 that wasn't present in earlier releases. In the following benchmark,
> you can see `BenchmarkSliceOfArray` experiences a 70% increase in
>
> execution time while `BenchmarkSliceOfInt` remains unchanged.
>
>
> BenchmarkSliceOfArray 413121 2855 ns/op
>
> BenchmarkSliceOfArray 216478 4881 ns/op
>
>
> Through the use of `pprof` and `GOSSAFUNC go build`, it was observed that
> the latest release employs `runtime.memmove` and accounts for the change in
> performance. I don't see the invocation of `memmove` or the perfromance
> degradation on AMD64.
>
>
> I believe this can be traced back to -
> https://go-review.googlesource.com/c/go/+/425234.
>
>
> Is this new behavior in this scenario correct or unindented?
>
>
> Should I open an issue for this?
>
>
> I am a newbie, so forgive me if there are errors in my approach.
>
>
> ----
>
> go version go1.19 darwin/arm64
>
> % GOMAXPROCS=1 go1.19 test -cpuprofile cpu.prof -bench .
>
> goos: darwin
>
> goarch: arm64
>
> pkg: foo/bar
>
> BenchmarkSliceOfInt 572226 2066 ns/op
>
> BenchmarkSliceOfArray 413121 2855 ns/op
>
> PASS
>
> ok foo/bar 3.488s
>
> ----
>
> go version go1.19.1 darwin/arm64
>
> % GOMAXPROCS=1 go test -cpuprofile cpu.prof -bench .
>
> goos: darwin
>
> goarch: arm64
>
> pkg: foo/bar
>
> BenchmarkSliceOfInt 527084 2065 ns/op
>
> BenchmarkSliceOfArray 216478 4881 ns/op
>
> PASS
>
> ok foo/bar 2.468s
>
> ok foo/bar 3.538s
>
> ----
>
> package datamove
>
>
> type Array [2]int
>
>
> func moveSliceOfArrayData(s []Array) []Array {
>
> for i := 1; i < len(s); i++ {
>
> s[i-1], s[i] = s[i], s[i-1]
>
> }
>
> return s
>
> }
>
>
> func moveSliceOfIntData(s []int) []int {
>
> for i := 1; i < len(s); i++ {
>
> s[i-1], s[i] = s[i], s[i-1]
>
> }
>
> return s
>
> }
>
> ----
>
> package datamove
>
>
> import (
>
> "testing"
>
> )
>
>
> var resultInt []int
>
> var resultArray []Array
>
>
> func BenchmarkSliceOfInt(b *testing.B) {
>
> var r []int
>
> for n := 0; n < b.N; n++ {
>
> s := make([]int, 1000)
>
> r = moveSliceOfIntData(s)
>
> }
>
> resultInt = r
>
> }
>
>
> func BenchmarkSliceOfArray(b *testing.B) {
>
> var r []Array
>
> for n := 0; n < b.N; n++ {
>
> s := make([]Array, 1000)
>
> r = moveSliceOfArrayData(s)
>
> }
>
> resultArray = r
>
> }
>
>
--
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.
To view this discussion on the web visit
https://groups.google.com/d/msgid/golang-nuts/73f49ef0-5e4e-4bb8-9976-534894ffde38n%40googlegroups.com.
