Re: [PATCH 2/2] Optimize compound_head() by avoiding a shared page flag

[Andrew Morton <akpm@linux-foundation.org>]

On Sat, 7 Apr 2007 15:16:17 -0700 (PDT) Christoph Lameter <clameter@sgi.com> wrote:

> On Fri, 6 Apr 2007, Andrew Morton wrote:
> 
> > Did you investigate
> > 
> > static inline int page_tail(struct page *page)
> > {
> > 	return ((page->flags & (PG_compound|PG_tail)) == (PG_compound|PG_tail));
> > }
> 
> The usual test_bit that we are using there uses a volatile reference 
> so these wont be combined if I check them separately.
> 
> A working example of the above would be much uglier:
> 
> static inline int page_tail(struct page *page)
> {
>  	return ((page->flags & ((1L << PG_compound)|(1L << PG_tail))) == 
> ((1L << PG_compound)|(1L << PG_tail)));
> }
> 
> May be this can be cleaned up somehow.

It might generate better code to do

	unsigned long compound;

	compound = page->flags & (1 << PG_compound);
	if (PG_compound > PG_tail)
		return compound & (page->flags << (PG_compound - PG_tail));
	else
		return compound & (page->flags << (PG_tail - PG_compound));


ie: get the PG_compound flag into `compound', then bitwise-and that with
the PG_tail flag, after shifting it into PG_compound' slot.  The return
value will be zero if either bit is clear, (1<<PG_compound) if both are
set.  The `if (PG_compound > PG_tail)' will be swallowed by the compiler.

The compiler should turn it all into

	(page->flags & N) & (page->flags << M)

Which may or may not be better than (page->flags & N == N), dunno. 
Probably not - if the compiler's any good it won't save a branch, I
suspect.



Which is all a ton of fun, but this subversion of the architecture's
freedom to use volatile, memory barriers etc is a worry.  We do the same in
page_alloc.c, of course...