Re: [patch] epoll use a single inode ...

[Valdis . Kletnieks]

On Thu, 08 Mar 2007 08:24:04 +0100, Eric Dumazet said:
>
> But what is the cost of the conditional branch you added in prefetch(x) ?
> 
> if (!x) return;
> 
> (correctly predicted or not, but do powerPC have a BTB ?)
> 
> About the NULL 'potential problem', maybe we could use a dummy nil (but 
> mapped) object, and use its address in lists, ie compare for &nil instead of 
> NULL. This would avoid :
> 
> - The conditional test in some prefetch() implementations
> - The potential TLB problem with the NULL value.

You avoid those two, but add the very high likelyhood that a statement of the
form 'if (!x) {....}' will creep back in and bug out.  I doubt that changing
the form of a basic C idiom to save a few cycles is worth it, especially when
the (!x) form can be tested without a memory fetch, but (x != nil) may require
fetching 'nil' (remember - the x86 is a very popular chipset, but register
starved - if the loop is any size at all, 'nil' may require a reload each
time around, costing you the win you thought you had....

Attachment: pgp00000.pgp
Description: PGP signature