question about gcc assembly

question about gcc assembly

[Coline Lelong-Pantel]

Hi

I've compiled a short program with two different gcc versions
(3.2.3 and 3.4.3) but it seems that the produced assembly code
isn't the same.

In fact 3.4.3 produce a more important code. For example when
gcc 3.2.3 produces :
movl	$0x405c28f6, -4(%ebp)

gcc 3.4.3 may use:
movl	$0x405c28f6, %eax
movl	%eax, -4(%ebp)

I wonder why it now uses %eax register ? And is there any way
to prevent this, in order to have similar assembly code with
both gcc versions ? Maybe some kind of option ?

Thanks for your answers.

Coline

PS Excuse my poor english, but it isn't my native langage.

Accédez au courrier électronique de La Poste : www.laposte.net
3615 LAPOSTENET (0,34 €/mn)

Re: question about gcc assembly

[Tim Prince]

Coline Lelong-Pantel wrote:
> Hi
> 
> I've compiled a short program with two different gcc versions
> (3.2.3 and 3.4.3) but it seems that the produced assembly code
> isn't the same.
> 
> In fact 3.4.3 produce a more important code. For example when
> gcc 3.2.3 produces :
> movl	$0x405c28f6, -4(%ebp)
> 
> gcc 3.4.3 may use:
> movl	$0x405c28f6, %eax
> movl	%eax, -4(%ebp)
> 
> I wonder why it now uses %eax register ? And is there any way
> to prevent this, in order to have similar assembly code with
> both gcc versions ? Maybe some kind of option ?
> 
> Thanks for your answers.
> 

Both gcc versions you quote are fairly old; if efficiency interests you, 
you might try a current version.  Were you able to show a significant 
difference in performance on a CPU which is important to you?  I would 
guess the most important reason for avoiding use of eax here would be to 
keep that register available for other use.  The optimization could be 
affected by compiler switches (-Os, -O2) or by selection of -march 
option, and many other possibilities.
Newer CPUs have hardware merging of operations, as well as stalls on 
certain immediate data operations, so it is impossible to know the 
performance implications from what you have presented.

Re: question about gcc assembly

[Coline Lelong-Pantel]

> Both gcc versions you quote are fairly old; if efficiency
interests you,
> you might try a current version.  Were you able to show a
significant
> difference in performance on a CPU which is important to you?

I have not checked that, my goal is to produce, if it is
possible, the same assembly code.

I would
> guess the most important reason for avoiding use of eax here
would be to
> keep that register available for other use.  The
optimization could be
> affected by compiler switches (-Os, -O2) or by selection of
-march
> option, and many other possibilities.

If I try using optimisation flags or -march, i get a
completely different code.

> Newer CPUs have hardware merging of operations, as well as
stalls on
> certain immediate data operations, so it is impossible to
know the
> performance implications from what you have presented.

What do you mean by "hardware merging of operations" ?


Accédez au courrier électronique de La Poste : www.laposte.net
3615 LAPOSTENET (0,34 €/mn)