Re: [PATCH] mul_u64_u64_div_u64: make it precise always

From: Andrew Morton
Date: Fri Jun 28 2024 - 18:16:01 EST


On Fri, 28 Jun 2024 17:46:33 -0400 (EDT) Nicolas Pitre <npitre@xxxxxxxxxxxx> wrote:

> On Fri, 28 Jun 2024, Andrew Morton wrote:
>
> > On Fri, 28 Jun 2024 15:06:20 -0400 (EDT) Nicolas Pitre <npitre@xxxxxxxxxxxx> wrote:
> >
> > > Library facilities must always return exact results. If the caller may
> > > be contented with approximations then it should do the approximation on
> > > its own.
> > >
> > > In this particular case the comment in the code says "the algorithm
> > > ... might lose some precision". Well, if you try it with e.g.:
> > >
> > > a = 18446462598732840960
> > > b = 18446462598732840960
> > > c = 18446462598732840961
> > >
> > > then the produced answer is 0 whereas the exact answer should be
> > > 18446462598732840959. This is _some_ precision loss indeed!
> > >
> > > Let's reimplement this function so it always produces the exact result
> > > regardless of its inputs while preserving existing fast paths
> > > when possible.
> >
> > I assume this was tested with some userspace harness? It would be
> > interesting to see that so that reviewers can see that suitable cases
> > have been covered.
>
> I do have a user space test tool but it isn't pretty looking.
> How should this be acceptably presented for public consumption?

Well, the fancy way would be as a new lib/test_XXX.c. There's a ton of
precedent for that. A less fancy way would be to paste the code into
this email thread ;)