Re: [go-nuts] Congrats to the Go team

Yes. It is this https://github.com/robaho/fixed/blob/5493e8737df761ffcf6be9441b4b8ae41fcf5da4/fixed_bench_test.go#L10On Apr 27, 2024, at 10:31 AM, Steven Hartland wrote:Do you have the test code for that specific test?That would allow others to have a look at it and also confirm if the test is so

Re: [go-nuts] Congrats to the Go team

Do you have the test code for that specific test? That would allow others to have a look at it and also confirm if the test is somehow optimising the function call away. On Sat, 27 Apr 2024 at 05:12, robert engels wrote: > Apologies. The assembly files were reversed. But the timings remain the

Re: [go-nuts] Congrats to the Go team

Why would it be optimized away in 1.12 and not optimized away in 1.21?I could have made a mistake in my recording so I’ll test it again tomorrow. On Apr 26, 2024, at 10:03 PM, 'Keith Randall' via golang-nuts wrote:Isn't that assembly exactly the opposite? The code that is passing in registers sho

Re: [go-nuts] Congrats to the Go team

I agree but in this case it is very consistent. Even if that were the case, wouldn’t that mean that 1.12 had better optimization in this case?I will dig in today and report back with the generated code. On Apr 26, 2024, at 12:17 AM, 'Keith Randall' via golang-nuts wrote:> There is a pretty signif

Re: [go-nuts] Congrats to the Go team

> There is a pretty significant degradation in AddFixed() which may be concerning to the Go team What is the benchmark for this? I am usually suspicious of sub-nanosecond benchmark times. Generally that indicates that the benchmark completely optimized away and all you are measuring is an empty

Re: [go-nuts] Congrats to the Go team

Thanks. I noticed those but didn’t look into how to address. Appreciate it. I will rerun. > On Apr 25, 2024, at 12:23 PM, Steven Hartland > wrote: > > Thanks for these, not sure if you noticed the notes from each run e.g. need > >= 4 samples to detect a difference at alpha level 0.05. > > Ba

Re: [go-nuts] Congrats to the Go team

Thanks for these, not sure if you noticed the notes from each run e.g. need >= 4 samples to detect a difference at alpha level 0.05. Basically run the benchmark with a -count=6 or more and then run the tool, and you get a the comparison values which are typically the interesting bit On Thu, 25 Ap

Re: [go-nuts] Congrats to the Go team

And for the 1.12 vs 1.22: │ /Users/robertengels/go1.12.17.txt │ /Users/robertengels/go1.22.2.txt│ │ sec/op │sec/op vs base │ AddFixed-8 0.5900n ± ∞ ¹ 0.7931n ± ∞ ¹~

Re: [go-nuts] Congrats to the Go team

│ /Users/robertengels/go1.21.5.txt │ /Users/robertengels/go1.22.2.txt│ │

Re: [go-nuts] Congrats to the Go team

What’s it look like when your run it through https://pkg.go.dev/golang.org/x/perf/cmd/benchstat which will provide a nice side by side comparison? On Wed, 24 Apr 2024 at 19:26, 'Robert Engels' via golang-nuts < golang-nuts@googlegroups.com> wrote: > I have a fairly stable project github.com/robah

Re: [go-nuts] Congrats to the Go team

Rough guess, it seems about the same. 1.22.2: BenchmarkAddFixed-8 10 0.7931 ns/op 0 B/op 0 allocs/op

[go-nuts] Congrats to the Go team

I have a fairly stable project github.com/robaho/fixed which is almost 100% cpu bound. It doesn’t change so it makes a great way to compare the performance of different Go versions using the same hardware. I took the time to re-run the tests today. Using 1.21.17