Re: PATCH: Fix the MIPS ISA ELF setting (Re: RedHat 7.1/mipsupdate)

[cgd at broadcom dot com]

"H . J . Lu" <hjl@lucon.org> writes:
> What are you talking about? The problem we have is
> 
> -march=r4100 -mips2
> 
> doesn't work.

Right.  So, you'll note that Eric said:

> > > -march=r4100 -mips2
> > >
> > > which means to use instructions from mips2 plus those from r4100.
> > >
> > >
> 
> I don't think that this is going to work in the compiler any time
> soon.
> 
> >
> > Any comments on this patch?
> 
> I'd like to hold off until the above works. [ ... ]


You didn't seem to like that, or at least you said:

> I don't think compiler will generate those by itself. They are from
> the asm statement.

In the absence of that (treating -march=r4100 as MIPS2 + 4100 in the
compiler, and passing that to the assembler), I was stating what I
think is the right interim solution for handling cpu-specific code in
asms.


So, my point is, you want one to be able to do one of two things:

* compile with -mips2 and put .set push/.set pop in your asms.  (This
  has drawbacks, as stated, but it's a decent solution.)

* compile with -march=r4100 and have The Right Thing happen.

When the options are done being munged (the way I see it), -march and
-mipsN should be mutually exclusive.  i.e. you should be able to say
-mips2, -march=mips2, or -march=r4100 (but not -mips2 -march=4100).

Explicitly supporting what you suggest right now ... will cause great
pain later, because then it'll have to be desupported "again."



chris