Re: [PATCH v2 01/15] audio: replace open-coded buffer arithmetic

[Volker Rümelin]

Am 26.01.22 um 09:48 schrieb Gerd Hoffmann:
>    Hi,
>
> > > > --- a/audio/audio_int.h
> > > > +++ b/audio/audio_int.h
> > > > @@ -266,6 +266,12 @@ static inline size_t audio_ring_dist(size_t dst, size_t 
> > > > src, size_t len)
> > > >        return (dst >= src) ? (dst - src) : (len - src + dst);
> > > >    }
> > > You haven't touched this function, but while I am looking at it, all function
> > > arguments are unsigned. So probably modulo operator might be used to get rid
> > > of a branch here.
> > That would be "return (len - dist + pos) % len;" but on my x86_64 system I
> > always prefer a conditional move instruction to a 64 bit integer division
> > instruction.
> Why?  Performance?
>
> Don't underestimate the optimizer of a modern compiler.  In many cases
> it simply isn't worth the effort.  Better optimize the code for humans,
> i.e. make it easy to read and understand.

I know this micro optimization is useless. I will not try to defend it 
because there are no good reasons for it. But I can't see that "return 
(len - dist + pos) % len;" is more readable than "return pos >= dist ? 
pos - dist : len - dist + pos;".

With best regards
Volker

> take care,
>    Gerd