cmd/compile: bounds check not needed

[KaurkerDevourer]

https://godbolt.org/z/EW16jnEG7
As you can see, there is panicIndex, on s[I + j]. But in golang, for i := 0; I < len(s); I++ doesn't go into panicIndex. Why does

for i := 0; i < len(s) - len(t); i++ {
    for j := 0; j < len(t); j++ {
         s[i + j]
    }
}

calls panicIndex? There is no need in this. Compiler sees that i + j is always in range [0, len(s)). Easy optimisation, probably not only in this case.

[carlmjohnson]

This doesn’t need to be a proposal. If you can write the optimization for the compiler and prove it works, that’s sufficient.

[KaurkerDevourer]

This doesn’t need to be a proposal. If you can write the optimization for the compiler and prove it works, that’s sufficient.

Hm, okey =)
Let me try this on a weekend. Can you please help me, im newbie in golang core code - where can I find anything about panicIndex and its optimisation (It is not callable in for I = 0; I < len(s); I++ f.e.)?
If anyone has an idea, help me with this please.

[randall77]

Optimizations like this are generally done in the prove pass, in cmd/compile/internal/ssa/prove.go.
Modifying that code is not for the feint of heart. There are lots of tricky details like getting the behavior on overflow/underflow correct.

[KaurkerDevourer]

I have candidates for what I need, but im not sure.
https://github.com/golang/go/blob/master/src/cmd/compile/internal/ssa/prove.go#L800
and
https://github.com/golang/go/blob/master/src/cmd/compile/internal/ssa/writebarrier.go
But where can I find any docs about this exact files. Comments do not provide needed information. Maybe someone just knows and can point me at the code that removes redundant panicIndex?