This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] math: Update i686 ulps
On Fri, 1 Jun 2018, Florian Weimer wrote:
> > How do they compare to ulps for these functions on e.g. x86_64? My
> > presumption is that they arise from use of -mfpmath=sse (thus, meaning no
> > excess precision, so results should mostly be the same as on other
> > architectures).
>
> Hmm. Right. sysdeps/x86_64/fpu/libm-test-ulps gas this:
>
> Function: "y1_upward":
> double: 7
> float: 2
> float128: 5
> idouble: 7
> ifloat: 2
> ifloat128: 5
> ildouble: 7
> ldouble: 7
>
> So they are not substantially worse.
In that case I think you should commit the patch.
> (If I copy over the x86-64 ulp baseline to the i686 directory, I still get
> many testsuite failures, though.)
x86_64 and i686 both have some architecture-specific function
implementations, so one can't expect to have exactly the same ulps for
them.
--
Joseph S. Myers
joseph@codesourcery.com