> 	-msoft-float _shouldnt_ be needed, as there should not be
> 	any code that uses floating point (except for the aforementioned
> 	specifically written assembler), but its there as a safety net.

So, on some ports, -msoft-float or a similar fp-disabling switch
(-mno-fp-regs on the Alpha) is, in fact, needed.

The port that i'm thinking of is the Alpha.  8-)


On the Alpha, the varargs calling conventions are such that unless
programs are compiled with -mno-fp-regs, varargs functions use a few
of the FP regs.  Also, performance optimizations (e.g. fast structure
copies) generated by the compiler sometimes use floating point
registers unless that's specifically disabled.  The latter, in
particular, can clobber registers, and so must be disabled.



cgd