Re: [PATCH v1 1/1] clk: fractional-divider: eliminate while-loop

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Thu, 2015-03-19 at 23:34 -0700, Stephen Boyd wrote:
> On 03/19, Andy Shevchenko wrote:
> > We may do a calculation of the rounded rate based on (*prate / div) value in
> > case it is not 0. It's as much higher as power of two of nearest value to
> > (*prate / div / maxn). Thus, the patch replaces while-loop by direct
> > calculations.
> > 
> > While here, fix off-by-one error. maxn is the maximum value that can be hold by
> > a register which means all ones in it.
> > 
> > Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> > ---
> >  drivers/clk/clk-fractional-divider.c | 13 ++++++-------
> >  1 file changed, 6 insertions(+), 7 deletions(-)
> > 
> > diff --git a/drivers/clk/clk-fractional-divider.c b/drivers/clk/clk-fractional-divider.c
> > index 6aa72d9..4c0541b 100644
> > --- a/drivers/clk/clk-fractional-divider.c
> > +++ b/drivers/clk/clk-fractional-divider.c
> > @@ -14,6 +14,7 @@
> >  #include <linux/device.h>
> >  #include <linux/slab.h>
> >  #include <linux/gcd.h>
> > +#include <linux/log2.h>
> >  
> >  #define to_clk_fd(_hw) container_of(_hw, struct clk_fractional_divider, hw)
> >  
> > @@ -49,18 +50,16 @@ static long clk_fd_round_rate(struct clk_hw *hw, unsigned long rate,
> >  			      unsigned long *prate)
> >  {
> >  	struct clk_fractional_divider *fd = to_clk_fd(hw);
> > -	unsigned maxn = (fd->nmask >> fd->nshift) + 1;
> > -	unsigned div;
> > +	unsigned maxn = fd->nmask >> fd->nshift;
> > +	unsigned long div, n;
> >  
> >  	if (!rate || rate >= *prate)
> >  		return *prate;
> >  
> >  	div = gcd(*prate, rate);
> > -
> > -	while ((*prate / div) > maxn) {
> > -		div <<= 1;
> > -		rate <<= 1;
> > -	}
> > +	n = *prate / div;
> > +	if (n > maxn)
> > +		return rate * roundup_pow_of_two(n / maxn);
> >  
> >  	return rate;
> >  }
> 
> Is there any reason why this code isn't using
> rational_best_approximation() and mult_frac()? 

For fist glance it should be faster that rational_best_approximation(),
though less precise.

I could switch to them.

-- 
Andy Shevchenko <andriy.shevchenko@intel.com>
Intel Finland Oy