Re: [PATCH 3/5] xen/sort: Switch to an extern inline implementation

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Thu, 11 Nov 2021, Julien Grall wrote:
> On 11/11/2021 17:57, Andrew Cooper wrote:
> > There are exactly 3 callers of sort() in the hypervisor.
> > 
> > Both arm callers pass in NULL for the swap function.  While this might seem
> > like an attractive option at first, it causes generic_swap() to be used
> > which
> > forced a byte-wise copy.  Provide real swap functions which the compiler can
> > optimise sensibly.
> 
> I understand the theory, but none of the two calls are in hot paths or deal
> with large set on Arm. So I am rather hesitant to introduce specialised swap
> in each case as it doesn't seem we will gain much from this change.

While I like Andrew's optimization, like Julien, I would also rather not
have to introduce specialized swap functions any time a sort() is
called. Why not keeping around generic_swap as extern gnu_inline in
sort.h as well so that the ARM callers can simply:

    sort(bootinfo.mem.bank, bootinfo.mem.nr_banks, sizeof(struct membank),
         cmp_memory_node, generic_swap);

?

That looks like a good option, although it would still result in a
small increase in bloat.