[Bug go/63269] libgo/math test failures in TestLog2

["ian at airs dot com" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63269

Ian Lance Taylor <ian at airs dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2014-11-05
     Ever confirmed|0                           |1

--- Comment #3 from Ian Lance Taylor <ian at airs dot com> ---
First, let me say that this code is in the Go master library and must be fixed
there.  It might be more effective to discuss it on the Go issue tracker at
http://golang.org/issue.

I don't agree with your argument for item 1.  You say that the wrong result
happens when using a fused multiply-add instruction.  The math library is
compiled with -ffp-contract=off (see MATH_FLAG in configure.ac and
Makefile.am), so the compiler should not be generating a fused multiply-add
instruction.

I'm not entirely persuaded by your argument for item 2.  Zero is a special
value.  When we expect a zero, we should get a zero, not something close to
zero.  I don't think this change is correct in general.  It may be correct for
some specific cases, but then we need to investigate those.

Item 3 is the same sort of thing: when we expect zero, we should, in general,
get exactly zero.