Given no PGO coverage should have minimal impact and there seems to be limited ways to use the runtime codepaths.
It might be worth collecting average profiles for some go programs, filter for the runtime (or maybe std) and apply it if no profile at all were provided. This wouldn't make your code faster, but might for std and runtime.
Both seems benchmarkable claims, then we could discuss the process costs to collecting and maintaining such profiles.
similar to #70291 this idea came up during a brief PGO discussion at the Go contributor summit at golab.io
Given no PGO coverage should have minimal impact and there seems to be limited ways to use the runtime codepaths.
It might be worth collecting average profiles for some go programs, filter for the runtime (or maybe std) and apply it if no profile at all were provided. This wouldn't make your code faster, but might for std and runtime.
Both seems benchmarkable claims, then we could discuss the process costs to collecting and maintaining such profiles.
similar to #70291 this idea came up during a brief PGO discussion at the Go contributor summit at golab.io