Thank you Volker. I want to say that I was not trying to benchmark the two version of Go (two weeks ago I had go1.15.8, now 1.16). At the beginning I was trying to do some linear algebra with Go as a free “alternative of Matlab”, so I noted the difference between the exe coming from the two versions. Now I have understand that this is not the way to measure the specific performance of my issue. I will learn more and also testing.
Sorry for your time, but thank you very much for your fast help. Il giorno sabato 20 febbraio 2021 alle 17:31:24 UTC+1 Volker Dobler ha scritto: > (premature send, sorry) > > 4. Your matrix multiplication is just a few CPU instructions, it is hard > to measure something tiny reliable. So use larger matrixes. > 5. You are benchmarking also how fast you can format and output > the results to stdout. This might or might not be intentional but > probably is just wrong. > 6. There is no point in comparing Go 1.15 and Go 1.16 with microbenchmarks > which do not properly measure what you think they do. Your code spends > far too much time in a lot of things. Come up with a _proper_ benchmark > based on testing.B and go test. Make sure this benchmark is stable. Make > sure this benchmarks really benchmarks what you are trying to measure. > Then redo the benchmarks 10 times and compare them doing proper > statistics. > > V. > > > On Saturday, 20 February 2021 at 15:48:32 UTC+1 TiT8 wrote: > >> Hello everyone, >> I'm Lorenzo and I'm not a computer scientist, I have used some "Matlab" >> and I am new with Go (surprisingly fast, simple and elegant). >> >> I'm trying to do some basic operation on matrix with the "Gonum" package >> using vectorization. I was curious about the speed of development and the >> performance (repeat... surprisingly for me!). So I've tried to do some >> measures: >> >> *package main* >> >> *import (* >> * "fmt"* >> * "time"* >> >> * "gonum.org/v1/gonum/mat <http://gonum.org/v1/gonum/mat>"* >> *)* >> >> *func main() {* >> * p := mat.NewDense(3, 3, nil)* >> * a := mat.NewDense(3, 3, []float64{* >> * 1, 2, 3, 1, 2, 3, 1, 2, 3,* >> * })* >> >> * start := time.Now()* >> >> * p.Mul(a, a) *// element-wise >> * fmt.Printf("%v\n", mat.Formatted(p, mat.Prefix(""), mat.Squeeze()))* >> * p.Add(a, a) * >> * fmt.Printf("%v\n", mat.Formatted(p, mat.Prefix(""), mat.Squeeze()))* >> * p.MulElem(a, a)* >> * fmt.Printf("%v\n", mat.Formatted(p, mat.Prefix(""), mat.Squeeze()))* >> >> * fmt.Println(time.Since(start))* >> *}* >> >> >> Well... when I run the "go run mat.go" command the result time is about >> 150 microseconds, but when I run "go build" and then execute the binary the >> result time is about 4 *milli*seconds. This happen when I use go1.16. >> >> When I use go1.15.8, on the same code, the exe is faster (about 120 >> microseconds) then the "go run mat.go" (about 180 microseconds). One order >> of magnitude faster than go1.16 exe (and basic Matlab). >> >> So: >> >> - am I in error with the code (sorry, it's all new for me)? If yes, >> where is my fault? >> - if the "benchmark" was correct, why go1.16 build executable is >> slower than go1.15.8 exe and "go run" command? >> - Is it my machine's fault or Gonum package or what? >> - Do you also have these values? >> >> >> My pc: HP ELITEBOOK 8560w Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz, >> 2201 Mhz, 4 Core(s), 8 Logical Processor(s) x64. >> I am on the *Windows Subsystem Linux 2. * >> >> >> *Sorry for my English, I hope you understand my issue, thank you for the >> attention* >> >> >> >> >> >> -- 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/2cadbc63-1b00-487d-a19f-3772d3ae5487n%40googlegroups.com.
