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

[Michael Eager <eager at mvista dot com>]

"Eric M. Christopher" wrote:
> 
> > Just to clarify, I am talking about gas, not gcc. There is no need to pass
> > -m4100 to gcc.
> 
> Yes.  I understand.  Though I don't know why you aren't passing
> -march=vr4100 to gcc (or at least -mtune=vr4100).  You don't want 4100
> tuned code?
> 
> > Any solution should take into acount of existing kernel source and previous
> > binutils which have been working fine until now. If you change kernel, now
> > you force everyone to use the new binutils. The problem is the current gas
> > won't allow "-m4100 -mips2" anymore. Why not fix that?
> 
> Because -m4100 -mips2 is wrong.  -mips2 is simply an alias for a
> processor that implements the standard mips2 ISA. Here you are
> specifying two architectures on the same command line.  Why would you
> think this would work (other than it worked in the past)?

IMO, processor specification for MIPS in both binutils and gcc is baroque.  
The result is this discussion of exactly what should result when you specify
apparently incompatible options.  MIPS has a jumble of processor, 
architecture, and cpu options, which are confusing and interact in 
ways that are confusing rather than useful.

I think that you should be able to specify exactly one of arch, isa, 
or cpu.   It makes no sense to me to have both -m4100 and -mcpu=4100.
The 4100 is a processor, not an architecture.  -mcpu=4100 should be
a complete specification.  -mips2 is an architecture.  Specifying either 
one should be a complete specification, with nothing additional necessary.
Or permitted.  They are not orthogonal specifications.


--
Michael Eager     eager@mvista.com	408-328-8426	
MontaVista Software, Inc. 1237 E. Arques Ave., Sunnyvale, CA  94085