lkml.org 
[lkml]   [2022]   [Sep]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH] minmax: clamp more efficiently by avoiding extra comparison
Hey again,

On Fri, Sep 23, 2022 at 12:40 PM Jason A. Donenfeld <Jason@zx2c4.com> wrote:
>
> Hi Andy,
>
> On Fri, Sep 23, 2022 at 12:36 PM Andy Shevchenko
> <andriy.shevchenko@linux.intel.com> wrote:
> >
> > On Fri, Sep 23, 2022 at 12:06:21PM +0200, Jason A. Donenfeld wrote:
> > > Currently the clamp algorithm does:
> > >
> > > if (val > hi)
> > > val = hi;
> > > if (val < lo)
> > > val = lo;
> > >
> > > But since hi > lo by definition, this can be made more efficient with:
> >
> > It's strongly speaking, but we have to proof that, right?
> > So, while I haven't checked the code, this change should also
> > include (does it?) the corresponding compile-time checks (for
> > constant arguments) in similar way how it's done for GENMASK().
> >
> > Otherwise I have no objections.
>
> I think most cases are with compile time constants, but some cases are
> with variables. What should we do in that case? Checking variables at
> runtime incurs the same cost as the old code. I guess we could do this
> fast thing for constants and the slower old thing for non-constants?
> Or not do either, keep this commit as is, and just accept that if you
> pass bogus bounds to clamp, you're going to end up with something
> weird, which is already the case now so not a big deal?

Actually, yea, I think we should keep this commit as-is and not add
additional checking becauseeeee not only is hi>lo by definition, but
both for the old code and for the new code, the result of lo>hi is
total nonsense:

Assuming hi > lo, these snippets all yield the same result:

if (val > hi)
val = hi;
if (val < lo)
val = lo;

if (val > hi)
val = hi;
else if (val < lo)
val = lo;

if (val < lo)
val = lo;
if (val > hi)
val = hi;

if (val < lo)
val = lo;
else if (val > hi)
val = hi;

Assuming lo > hi, and the first condition triggers, these snippets all
yield different results, all of which are undefined nonsense:

if (val > hi)
val = hi;
if (val < lo)
val = lo;
--> val is lo

if (val > hi)
val = hi;
else if (val < lo)
val = lo;
--> val is hi

if (val < lo)
val = lo;
if (val > hi)
val = hi;
--> val is hi

if (val < lo)
val = lo;
else if (val > hi)
val = hi;
--> val is lo

Jason

\
 
 \ /
  Last update: 2022-09-23 12:50    [W:0.075 / U:0.660 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site