This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: Reorganize -ffast-math code.



> Why is flag_fast_math checked before
> generating builtins?  Do these builtins incompletely implement
> these functions for all IEEE arguments and results?

The sine and cosine builtins were relegated to fast-math because
of their range reduction error, not merely because of problems
with IEEE special numbers.  There were user complaints about
the sizable errors committed and no PLOSS or TLOSS warnings.

One could raise a similar question about the builtin cabs function.
Currently one has a choice between a method with no overflow
protection and a variable-accuracy implementation that has larger
error than the fdlibm library routine.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]