[PATCH] Add FMA_EXPR, un-cse multiplications before expansion

[Geert Bosch]

On Oct 21, 2010, at 20:12, Joseph S. Myers wrote:
> You have working fma functions right now in git glibc, except for:
> 
> * fmal for IBM long double (double double)
> 
> * targets without exceptions and rounding mode support (generally using 
> software floating point), where I suspect a pure soft-fp implementation 
> may be best.

This is really great, but leads us back to the same 
issue that it won't do anything for non-glibc systems.

  -Geert