Please include an indication of the performance difference (with glibc's benchtests) when proposing a performance patch. (In this case, that means for both the fma and the non-fma cases.) -- Joseph S. Myers joseph@codesourcery.com