Re: Large inlines in include/linux/skbuff.h
From: Eric Dumazet
Date: Thu Apr 22 2004 - 14:25:39 EST
Denis Vlasenko wrote:
If this a specialized machine, with a small program that mostly uses
recv() & send() syscalls, then, inlining functions is a gain, since
icache may have a 100% hit ratio. Optimization guidelines are good for
the common cases, not every cases.
And if it is NOT a specialized machine? icache miss will nullify
any speed advantages, while you still pay space penalty.
We don't need to pay for at least ~250 kbytes wasted overall
in allyesconfig kernel for each and every specialized
setup conceivable, IMHO.
The point is : if this IS a specialized machine, then the kernel is
custom one, not allyesconfig.
This is imho what I do for specialized machines, and yes, I even inline
some specific functions, like fget() and others.
But I didnt asked to not doing the un-inlining, I was just reminding
that some guys (like me) are doing micro-optimizations that 'seem' to go
against Optimizations guidelines from intel or AMD.
Eric
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/