P.S. to: Call for testers: SH optimized software floating point

[amylaar@spamcop.net]

P.S.: there is a backwards compatibility issue with truncdfsf2 / extendsfdf2
for m2e / m3e / m4-single-only multilibs, since the single precision floating
point values are passed in r4 / r0.

I could add code inside the library functions to be fully compatible with the
original ABI - i.e. float argument in fr4, float result in fr0.
The downside would be that that gives poor scheduling.

Or I could pick a new name for the functions with the altered ABI.

Which solution would people prefer?