[PATCH v3 2/2] x86_64: roundeven with fpu and sse4.1 support

[Joseph Myers]

Generally such architecture-specific function versions should be justified 
with benchmark results showing a performance improvement.

That is, first you should add benchmark inputs for the roundeven functions 
in benchtests/.  Then patches adding architecture-specific roundeven 
implementations should include before-and-after benchmark results in the 
commit message to demonstrate that the architecture-specific 
implementation does indeed result in a performance improvement.

Note also the need for an FSF copyright assignment for significant 
contributions: 
https://sourceware.org/glibc/wiki/Contribution%20checklist#FSF_copyright_Assignment

-- 
Joseph S. Myers
joseph@codesourcery.com