David Chase
David Chase
@davidmdm I work on the Go compiler, sometimes the runtime, sometimes the libraries. When we change the compiler, we like to be sure that we are making it better for...
I think quality of `wasip1` is a different benchmark, arguably I need to fiddle with the benchmark runner itself so that it can do that or has an option to...
Provisionally assigned to @qmuntal, not clear if this is a problem at tip or not.
The problem with backward is that there's an inlining that doesn't happen, I think the Backward iterator has cost 81.
With three recent CLs, there are some improvements: ``` BenchmarkSliceFunctions/AllForLoop-10-8 318656155 3.549 ns/op 0 B/op 0 allocs/op BenchmarkSliceFunctions/All-10-8 284950826 4.232 ns/op 0 B/op 0 allocs/op BenchmarkSliceFunctions/BackwardForLoop-10-8 344576744 3.479 ns/op 0...
The intent was to help with rangefunc performance. If you find it necessary to play these games for code that matters (production code, etc), please let us know, because otherwise...
PS - on a weekend lark, did a crude test of "if a function is package-private and only called once..." and the crude test is definitely not ready for prime...
As of 1.24 freeze, the differences observed in this benchmark appear to be caused by whether the benchmarked function gets inlined into the BenchmarkXXX function. This is still a rangefunc-related...
Related CL (that needs to mention this bug): https://go.dev/cl/603335
cc @golang/tools-team