Re: [RFC][PATCH] introduce ptr_diff()

[Matthew Wilcox <matthew@wil.cx>]

On Fri, Aug 20, 2010 at 01:07:37PM +0100, Al Viro wrote:
> On Thu, Aug 19, 2010 at 02:23:09PM +0200, Andi Kleen wrote:
> > IMHO it would be better to simply disable the warning in sparse instead
> > of uglying the code just to work around sparse bogosity. It doesnt' seem
> > to make much sense. A subtraction followed by a shift is not expensive.
> 
> "Bogosity" in question is quite real and proposed "fix" actually makes the
> things even worse.  At least gcc bothers to optimize the division in there...
> 
> Basically, sparse warns about pointer subtractions that divide by constants
> that are not powers of two.  Which _does_ have non-trivial price, even with
> optimizations done by gcc.

I'm not sure the price is so high.  I googled around and came across
http://mdfs.net/Docs/Comp/ARM/Cookbook/cook3 (section 3.3).  Dividing by
2^n, (2^n + 2^m) or (2^n - 2^m) can be done using a small series of adds
and subtractions (important on ARM as it had no divide instruction at
the time).  Most structures are going to be of one of these sizes ...
and in particular, struct page is 56 bytes in my config, which is 64 - 8.
Maybe sparse needs to be taught that dividing by 2^n [+-] 2^m is cheap
enough to not warn about.

-- 
Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."