cross-compiled floating point constants broken

[Stephen L Moshier]

>  I found two files that use
> %d for printing inside ASM_OUTPUT_DOUBLE et al.  We can document that use
> of %d is wrong though.

On i386 the output_move_double function generates pieces in the form
of 32-bit decimal immediate integer operands.  The assembler (gas in
this instance) accepts either unsigned from the alpha hosted cross
compiler or signed natively.

The case that really causes trouble on a 32-bit target machine is
those 16-digit hex numbers.