Re: Synchronizing Bit operations V2

[Russell King <rmk+lkml@arm.linux.org.uk>]

On Thu, Mar 30, 2006 at 04:58:25PM -0800, Christoph Lameter wrote:
> On Thu, 30 Mar 2006, Luck, Tony wrote:
> 
> > > Also some higher level functions may want to have the mode passed to them 
> > > as parameters. See f.e. include/linux/buffer_head.h. Without the 
> > > parameters you will have to maintain farms of definitions for all cases.
> > 
> > But if any part of the call chain from those higher level functions
> > down to these low level functions is not inline, then the compiler
> > won't be able to collapse out the "switch (mode)" ... so we'd end up
> > with a ton of extra object code.
> 
> Correct. But such bitops are typically defined to be inline.

That's doesn't seem to be the point that Tony was making.  To illustrate
it let's add a practical example:

static inline void clear_bit_mode(int bit, unsigned long *ptr, int mode)
{
	case (mode) {
	...
	}
}

void foo(blah blah, int mode)
{
	... complex function ...
	clear_bit_mode(bit, ptr, mode);
	...
}

void bar(blah blah)
{
	foo(blah, MODE_BARRIER);
}

In this case, the compiler can not optimise the unnecessary code from
the clear_bit_mode because the mode argument quite definitely is not a
constant known at compile time.  Only if 'foo' was a static inline
would it be known.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core