Re: [PATCH v2] Work around the NOP issue of Loongson2F

["Maciej W. Rozycki" <macro at linux-mips dot org>]

On Mon, 16 Nov 2009, Richard Sandiford wrote:

> >  Per Richard's suggestion (which I second) please name the option 
> > -mfix-loongson2f and adjust all the corresponding variables, etc. 
> > accordingly.
> 
> Wu Zhangjin was right; I was actually suggesting -mfix-loongson2f-nop.
> I've no problem subdividing -mfix-* options based on specific errata
> like this, especially given that new versions of the processor are
> free from it.

 Ah, I misunderstood, sorry.

 I'd be a little bit worried about proliferating options to handle 
processor errata and actually I take the fact that current steppings have 
been fixed as an argument against it (because you need no -mfix-* option 
for these to run code correctly).  In particular -- if we get more such 
specific options, will one have to specify all of them explicitly to 
gurantee predictable behaviour of all the members of the CPU family, or 
will there be a global -mfix-loongson2f setting to cover all of of the 
errata?  The former will not scale -- you'll have to tweak Makefiles, etc. 
whenever a new erratum is discovered.  The latter has a problem with 
negatives that'll require complex handling.  I'm using loongson2f as an 
example here of course -- that's a general concern of any such option.

 I'm not going to defend my stand fiercely though -- I just wanted to make 
sure such concerns were taken into account -- and you are the maintainer 
to decide, after all.

  Maciej