unresolved i386_return_pops_args in tree

[Robert Lipe]

> > quit building becuase config/i386/sco5.h calls i386_return_pops_args.
> 
> It changed to ix86_return_pops_args with the new backend.
> 
> Fixed.

Thanx!   It now dies in a different place when building libgcc2.

_floatdidf
_floatdisf
_fixunsdfsi
_fixunssfsi
_fixunsdfdi
/usr/tmp/ccYy2gsa.s:71:invalid operand combination: fstp
/usr/tmp/ccYy2gsa.s:77:invalid operand combination: fld
make[3]: *** [libgcc2.a] Error 1
make[2]: *** [stmp-multilib-sub] Error 2
make[1]: *** [stmp-multilib] Error 1
make: *** [bootstrap3] Error 2
(robertl) rjlhome:/play/negcs/gcc

I remember digging into this once and the System V assemblers didn't
like the "st" operands preferring instead "st(0)".  I don't speak FP
on IA32 enough to know they're equivalent, but that's what I recall
concluding just staring at diffs of the the assembly output. I thought
I even had a crude patch at one point but I can't seem to spot it right
now.

Do you need a copy of the generated .s file to spot this?

RJL