-malign-*=0 vs 2

-malign-*=0 vs 2

[David Ronis]

Several days ago I asked what the correct value for the -malign-* options should 
be.  The unanimous answer was 0 on a pentium.  Fine, but now I'm trying to 
use egcs to recompile my kernel (linux 2.0.32) and notice that Linus has hard
coded -malign-*=2 for the 586 architecture.  Is this correct for gcc2 but
not for egcs?  Is this correct for the kernel?  For now, I've changed the values
to zero (for my pentium), but would like someone to comment on why this is 
correct, and why Linus is wrong.


David Ronis

Re: -malign-*=0 vs 2

[Jeffrey A Law]

  In message < 199712081630.LAA19482@onsager.chem.mcgill.ca. >you write:
  > to zero (for my pentium), but would like someone to comment on why this is 
  > correct, and why Linus is wrong.
I think you would have to ask Linux why he did things like that.

jeff

Re: -malign-*=0 vs 2

[Marc Lehmann]

On Mon, Dec 08, 1997 at 11:55:23PM -0700, Jeffrey A Law wrote:
>   In message < 199712081630.LAA19482@onsager.chem.mcgill.ca. >you write:
>   > to zero (for my pentium), but would like someone to comment on why this is 
>   > correct, and why Linus is wrong.
> I think you would have to ask Linux why he did things like that.

Linus found out that the 16 byte alignment was simply excessive and not necessary.
He did not try to use zero alignment, which should be as fast as 16 byte alignment,
so he stuck with the (IMHO) 4 byte alignment.

Anyhow, kernel speed wasn't that much of an issue, since time-critical paths
are full of hand-optimized assembler anyway, and the memory loss
due to the big alignment was considerable.

      -----==-                                              |
      ----==-- _                                            |
      ---==---(_)__  __ ____  __       Marc Lehmann       +--
      --==---/ / _ \/ // /\ \/ /       pcg@goof.com       |e|
      -=====/_/_//_/\_,_/ /_/\_\                          --+
    The choice of a GNU generation                        |
                                                          |