Re: [PATCH v18 3/4] pwm: Add support for RZ/G2L GPT

From: Geert Uytterhoeven <geert@linux-m68k.org>
To: "Uwe Kleine-König" <u.kleine-koenig@pengutronix.de>
Cc: Biju Das <biju.das.jz@bp.renesas.com>,
	Thierry Reding <thierry.reding@gmail.com>,
	 Philipp Zabel <p.zabel@pengutronix.de>,
	Fabrizio Castro <fabrizio.castro.jz@renesas.com>,
	 Magnus Damm <magnus.damm@gmail.com>,
	linux-pwm@vger.kernel.org,  linux-renesas-soc@vger.kernel.org,
	 Prabhakar Mahadev Lad <prabhakar.mahadev-lad.rj@bp.renesas.com>,
	Biju Das <biju.das.au@gmail.com>,
	 kernel@pengutronix.de
Subject: Re: [PATCH v18 3/4] pwm: Add support for RZ/G2L GPT
Date: Tue, 12 Mar 2024 09:11:42 +0100	[thread overview]
Message-ID: <CAMuHMdV8SnMgawrMspemJNfsAHW-wSboXeEOgZ6F37QqrmiLSA@mail.gmail.com> (raw)
In-Reply-To: <hy5crf2leuvewkn5omgrk2bmkndivwmhst4yrefnd3mepy4nzd@xw3rtkxdnb2g>

Hi Uwe,

On Tue, Mar 12, 2024 at 8:20 AM Uwe Kleine-König
<u.kleine-koenig@pengutronix.de> wrote:
> On Tue, Feb 20, 2024 at 07:43:17PM +0000, Biju Das wrote:
> > --- /dev/null
> > +++ b/drivers/pwm/pwm-rzg2l-gpt.c
> > +static u64 calculate_period_or_duty(struct rzg2l_gpt_chip *rzg2l_gpt, u32 val, u8 prescale)
> > +{
> > +     u64 tmp, d;
> > +
> > +     /*
> > +      * Rate is in MHz and is always integer for peripheral clk
> > +      * 2^32 * 2^10 (prescalar) * 10^9 > 2^64
> > +      * 2^32 * 2^10 (prescalar) * 10^6 < 2^64
> > +      * Multiply val with prescalar first, if the result is less than
> > +      * 2^34, then multiply by 10^9. Otherwise divide nr and dr by 10^3
> > +      * so that it will never overflow.
> > +      */
> > +
> > +     tmp = (u64)val << (2 * prescale);
> > +     if (tmp <= (1ULL << 34)) {
>
> I would have written that as:
>
>         if (tmp >> 34 == 0)
>
> (which implements tmp < (1ULL << 34), which doesn't matter much).
>
> > +             tmp *= NSEC_PER_SEC;
> > +             d = rzg2l_gpt->rate;
> > +     } else {
> > +             tmp *= div64_u64(NSEC_PER_SEC, KILO);
>
> I don't know if the compiler is clever enough to not calculate that
> every time?

Not on 32-bit when written that way.

> Also using div64_u64 is too heavy given that both values fit
> into an u32.

Indeed, so "NSEC_PER_SEC / KILO" should be fine.
I guess NSEC_PER_MSEC would be too obfuscating?

>
> > +             d = div64_u64(rzg2l_gpt->rate, KILO);
>
> At first I thought you could better use 1024 as the common divisor here
> as it could be implemented using a shift operation. But I understood
> with the comment above that we're not losing precision here as both
> NSEC_PER_SEC and rate are a multiple of 1000.
>
> Maybe s/Rate is in MHz and is always integer for peripheral clk/Rate is
> a multiple of 1000000, and so dividing by 1000 is an exact operation./ ?
>
> > +     }
> > +
> > +     return DIV64_U64_ROUND_UP(tmp, d);
> > +}

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

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