This is the mail archive of the
gcc-help@gcc.gnu.org
mailing list for the GCC project.
Re: float behaivior on ARMv8-A
- From: Andrew Haley <aph at redhat dot com>
- To: N30dG <christian dot dreihsig at t-online dot de>, gcc-help at gcc dot gnu dot org
- Date: Sun, 29 May 2016 09:54:30 +0100
- Subject: Re: float behaivior on ARMv8-A
- Authentication-results: sourceware.org; auth=none
- References: <1462124797268-1259502 dot post at n5 dot nabble dot com> <57272096 dot 5000003 at redhat dot com> <1462298729580-1260215 dot post at n5 dot nabble dot com> <5728EE8E dot 7020207 at redhat dot com> <1464421039509-1267857 dot post at n5 dot nabble dot com>
On 28/05/16 08:37, N30dG wrote:
> I was able to locate the Problem(s).
> And I think this maybe could be a gcc-bug. It seem's that sin() and sqrt()
> works in less precision when compiled for AARCH64 (no ffast-math or other
> "unsafe" float-options enabled). I've worked that around by using sinl() and
> sqrtl() and now everythink work's like expected. I'm note sure if this is a
> bug or a normal behavior when compiling for different architectures.
>
> If this could be a bug and you need additional information feel free to
> write.
Sure. If there's anything we can do to zoom in on the problem that
would help. Significant inaccuracy in trig functions probably is a
bug.
Having said that, Standard C does not guarantee any particular
accuracy for these funtions. Also, this is probably not a bug in GCC
itself but in whatever C library you're using.
Andrew.