Exactly so. ‘-Ofast -fno-finite-math-only’ gives correct results. What’s going on here? I don’t expect this has anything to do with NaN or Inf.
> On Apr 13, 2016, at 2:13 PM, Jay Foad <jay.f...@gmail.com> wrote: > > I'd guess -ffinite-math-only. > > Jay. > > On 13 April 2016 at 19:36, Xiao-Yong Jin <jinxiaoy...@gmail.com> wrote: >> Looks like a gcc optimization problem (clang works correctly). >> Using ‘-Ofast’ with gcc 5.3 reproduces the bug. >> Using ‘-O3’ works fine. >> >> So one of the options down here caused the bug: >> -fassociative-math >> -fcx-limited-range >> -ffinite-math-only >> -fmath-errno >> -freciprocal-math >> -fsigned-zeros >> -ftrapping-math >> -funsafe-math-optimizations >> >> I’ll see if I can pin down exactly which one. Do you have any idea what >> could be the culprit? >> >> - Xiao-Yong >> >> >>> On Apr 13, 2016, at 1:01 PM, Juergen Sauermann >>> <juergen.sauerm...@t-online.de> wrote: >>> >>> Hi Xiao, >>> >>> interesting. I am still getting this in *SVN**719*: >>> >>> * 8 ¯8⋆÷3 >>> 2 1J1.732050808* >>> >>> /// Jürgwn >>> >>> >>> On 04/13/2016 03:42 AM, Xiao-Yong Jin wrote: >>>> Hi, >>>> >>>> With the svn version 719, >>>> >>>> 8 ¯8⋆÷3 >>>> 2 0.000000000 >>>> >>>> but with the release version 1.5, >>>> >>>> 8 ¯8⋆÷3 >>>> 2 1J1.732050808 >>>> >>>> Best, >>>> Xiao-Yong >>>> >>>> >>>> >>> >> >>
