Re: [PATCH] Linux Kernel Markers 0.14 for 2.6.17

[Mathieu Desnoyers <compudj@krystal.dyndns.org>]

* Jeremy Fitzhardinge (jeremy@goop.org) wrote:
> Mathieu Desnoyers wrote:
> >Hi,
> >
> >Constructing on Jeremy Fitzhardinge's comments about gcc optimizations, I
> >rewrote (once more) the markers mechanism so that the optimized mode does 
> >not
> >jump between two different inline asm. Instead, the optimized version uses 
> >a
> >load immediate (in assembly) that will be used by a test to decide of a 
> >branch
> >(in C).
> >  
> 
> I should have spelled out my point a bit more.  If you've got a flag 
> you're just testing, couldn't you just do:
> 
> 	if (__mark_enabled_##name)
> 		(*__mark_func)(...);
> 
> and do without the asms or the section?
> 

Because a supplementary memory read is added on the critical path with a normal
flag test. The assembly can provide an immediate value without any need of
memory read from the data section.

To change the behavior of the program, I just have to change the immediate value
in the movb instruction.

However, the non-optimized generic version does exactly this : it simply tests a
flag loaded from memory. It can be very useful on embedded systems where the
code is in read-only memory.

Regards,

Mathieu


OpenPGP public key:              http://krystal.dyndns.org:8080/key/compudj.gpg
Key fingerprint:     8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68