[PATCH] Add support for MIPS64r6

[Andrew Bennett]

> From: Richard Sandiford [mailto:rdsandiford@googlemail.com]
> Sent: 13 September 2014 09:45
> To: Andrew Bennett
> Cc: binutils@sourceware.org; Rich Fuhler; Matthew Fortune; Saeed Ghazanfar
> Subject: Re: [PATCH] Add support for MIPS64r6
> 
> Andrew Bennett <Andrew.Bennett@imgtec.com> writes:
> > +/* OP_NON_ZERO_REG matcher.  */
> > +
> > +static bfd_boolean
> > +match_non_zero_reg_operand (struct mips_arg_info *arg, const struct
> mips_operand *operand)
> 
> Long line.
> 
> > @@ -11044,8 +11294,11 @@ macro (struct mips_cl_insn *ip, char *str)
> >        goto ld_st;
> >      case M_LWC2_AB:
> >        s = "lwc2";
> > -      fmt = COP12_FMT;
> > -      offbits = (mips_opts.micromips ? 12 : 16);
> > +      fmt = (ISA_IS_R6 (mips_opts.isa) ? "E,+:(d)"
> > +	     : COP12_FMT);
> 
> You do this for all uses of COP12_FMT, so please do it in COP12_FMT itself.
> 
> > @@ -11103,13 +11359,19 @@ macro (struct mips_cl_insn *ip, char *str)
> >        goto ld_st;
> >      case M_LL_AB:
> >        s = "ll";
> > -      fmt = MEM12_FMT;
> > -      offbits = (mips_opts.micromips ? 12 : 16);
> > +      fmt = (ISA_IS_R6 (mips_opts.isa) ? "t,+j(b)"
> > +	     : MEM12_FMT);
> > +      offbits = (mips_opts.micromips ? 12
> > +		 : ISA_IS_R6 (mips_opts.isa) ? 9
> > +		 : 16);
> 
> And here I think we want to put the fmt ?: in a new LL_SC_FMT macro,
> for all LL and SC variants.
> 
> > @@ -16018,10 +16366,16 @@ s_nan (int ignore ATTRIBUTE_UNUSED)
> >
> >    if (i == sizeof (str_2008) - 1
> >        && memcmp (input_line_pointer, str_2008, i) == 0)
> > -    mips_flag_nan2008 = TRUE;
> > +    mips_nan2008 = 1;
> >    else if (i == sizeof (str_legacy) - 1
> >  	   && memcmp (input_line_pointer, str_legacy, i) == 0)
> > -    mips_flag_nan2008 = FALSE;
> > +    {
> > +      if (ISA_HAS_LEGACY_NAN (file_mips_opts.isa))
> > +	mips_nan2008 = 0;
> > +      else
> > +	as_fatal (_("`%s' does not support legacy NaN"),
> > +	          mips_cpu_info_from_isa (file_mips_opts.isa)->name);
> 
> This should be as_bad rather than as_fatal.  Fatal errors are good for
> bogus command lines, but for directives we should try to continue.
> 
> OK with those changes, thanks.


I have now made these changes, and are ready to commit the patch, but I
forgot to mention that this patch relies on the soft float/single float
patch that Matthew submitted to have clean testsuite results.  Are you 
happy for me to still commit the patch, or shall I wait until Matthew 
has committed his work first?

Many thanks,



Andrew