Re: [PATCH v2 07/11] math64: New DIV_U64_ROUND_CLOSEST helper

[Pali Rohár]

On Friday 25 June 2021 19:39:10 Geert Uytterhoeven wrote:
> Hi Willy,
> 
> On Fri, Jun 25, 2021 at 5:50 PM Willy Tarreau <w@xxxxxx> wrote:
> > On Fri, Jun 25, 2021 at 05:38:03PM +0200, Pali Rohár wrote:
> > > On Friday 25 June 2021 17:22:31 Geert Uytterhoeven wrote:
> > > > > +/*
> > > > > + * DIV_U64_ROUND_CLOSEST - unsigned 64bit divide with 32bit divisor rounded to nearest integer
> > > > > + * @dividend: unsigned 64bit dividend
> > > > > + * @divisor: unsigned 32bit divisor
> > > > > + *
> > > > > + * Divide unsigned 64bit dividend by unsigned 32bit divisor
> > > > > + * and round to closest integer.
> > > > > + *
> > > > > + * Return: dividend / divisor rounded to nearest integer
> > > > > + */
> > > > > +#define DIV_U64_ROUND_CLOSEST(dividend, divisor)       \
> > > > > +       ({ u32 _tmp = (divisor); div_u64((u64)(dividend) + _tmp / 2, _tmp); })
> > > >
> > > > Given "dividend" should already be an unsigned 64-bit value, I don't
> > > > think the cast to "u64" is needed. Similar macros in this file also
> > > > don't have the cast.
> > >
> > > It is just to ensure that plus operation between dividend and _tmp is
> > > evaluated in 64-bit context to prevent overflow. Just a case when user
> > > calls this macro with 32-bit dividend param. As it is a macro (and not
> > > inline function) type is not automatically enforced.
> >
> > I agree, a large u32 argument added to _tmp/2 could overflow and remain
> > 32 bits, yielding an incorrect result. The cast is mandatory here (and
> > will either emit no code, or be useful).
> 
> Fair enough.
> So we want to add a cast to DIV64_U64_ROUND_CLOSEST() above, too?

For DIV64_U64_ROUND_CLOSEST() it is not needed. divisor is copied into
u64 _tmp variable and therefore "(dividend) + _tmp / 2" is already
evaluated in 64-bit context even when dividend is only 32-bit.

The only trap is that negative value as written below.

> > The only trap I'm seeing is if a negative signed int is passed in dividend,
> > it will be sign-extended and will give a large u64 value. A preliminary
> > u32 cast could avoid this but would break valid u64 arguments, and I'd
> > claim we never know what the user wants if this happens in the first place.
> 
> Yep.
> 
> Gr{oetje,eeting}s,
> 
>                         Geert
> 
> -- 
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
> 
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                 -- Linus Torvalds