> From: Daniel Dickman <[email protected]>
> Date: Sun, 9 Jan 2022 16:36:33 -0500
> 
> On Sun, Jan 9, 2022 at 4:18 PM Mark Kettenis <[email protected]> wrote:
> >
> > > From: Greg Steuck <[email protected]>
> > > Date: Sun, 09 Jan 2022 12:47:14 -0800
> > >
> > > Greg Steuck <[email protected]> writes:
> > >
> > > > This was reduced from a ghc test. The results of the program differ
> > > > between OpenBSD 7.0-current-amd64 and a couple of other systems:
> > >
> > > Thanks to phessler@ for testing on arm64 where the bug doesn't happen.
> > > This patch makes OpenBSD-amd64 work the rest of the systems. I added
> > > i386 as it was also similarly broken (but didn't test the change yet).
> >
> > As I said on icb, NetBSD removed all of the x86 assembly sin/cos/tan
> > implementations because:
> >
> >   "The x87 hardware uses a bad approximation to pi for argument reduction"
> >
> > So I think we should use the software fallbacks for all of these
> > functions (and remove the broken assembly implementations).
> >
> > I don't think it makes sense to just remove tanf() and leave the
> > others in place.
> >
> >
> 
> Here's the link to the commit Mark referenced:
> https://github.com/NetBSD/src/commit/4f9e11b0dddf04640fe0553a9133a471af613627
> 
> And then the actual implementations were removed in this commit:
> https://github.com/NetBSD/src/commit/870f792ccadb412e522f37caec6028b0076a871b
> 
> So I guess this is the list of functions to remove, Mark?

Yes.

> I'm testing this on i386 with numpy to see if regress tests improve.
> 
> s_cos.S
> s_cosf.S
> s_sin.S
> s_sinf.S
> s_tan.S
> s_tanf.S
> 

Reply via email to