Re: The arithmetic support of SCmode and DCmode.

[craig at jcb-sc dot com]

>In general, that is so; however, on systems which support IEEE P754/854 range 
>widening, such as x86 and 68k, the faster in-line code would be preferable.  
>It's also possible to take an intermediate position, by setting up c_div.c 
>and z_div.c so that the code generated depends on whether <float.h> shows 
>that widening the operands produces the necessary widened range.

Yes, ideally, the back end would make sure it got this "right", using
open-coding of range-widening operations, algorithms like z_div's, or
just calling a library function.

In the meantime, is it indeed the case that c_div.c could just use
`double' for intermediate computations using the "direct, fast" approach
gcc uses in optabs.c, and that would solve all the problems?

In that case, we should write up a patch that does that, perhaps
conditional on the machine having `double' and the relevant conversions
be (nearly) as fast as `float', and send it to the libf2c maintainer.

        tq vm, (burley)