Re: [ros-dev] [ros-diffs] [tkreuzer] 67709: [D3DX9_36/WINED3D] - Remove useless copysignf invocation and the wrong defines for it - BrokenHacksOnTopOfBrokenCode--; - Should be sent upstream

[Thomas Faber]

On 2015-05-14 06:00, [e-mail] wrote:
> int sign = (copysignf(1, in) < 0);
> int sign = (in < 0);
> if (copysignf(1.0f, value) < 0.0f)
> if (value < 0.0f)
>      ++idx;

I believe the behavior would be different here for negative zero:
copysignf(1.0f, -0.0f) should be < 0.0f
-0.0f should be == 0.0f

Maybe that's the reason for having these calls?

[David Quintana (gigaherz)]

Judging by the involved macro, if -0.0 is not < 0, then the old code
had exactly the same bug.

On 14 May 2015 at 14:50, Thomas Faber [e-mail] wrote:
> On 2015-05-14 06:00, [e-mail] wrote:
> > int sign = (copysignf(1, in) < 0);
> > int sign = (in < 0);
> > if (copysignf(1.0f, value) < 0.0f)
> > if (value < 0.0f)
> >      ++idx;
>
> I believe the behavior would be different here for negative zero:
> copysignf(1.0f, -0.0f) should be < 0.0f
> -0.0f should be == 0.0f
>
> Maybe that's the reason for having these calls?
>
> Ros-dev mailing list
> [e-mail]
> http://www.reactos.org/mailman/listinfo/ros-dev

[Thomas Faber]

Yes it did. Except that macro was a ROS-Diff in the first place ;)

On 2015-05-14 11:12, David Quintana (gigaherz) wrote:
> Judging by the involved macro, if -0.0 is not < 0, then the old code
> had exactly the same bug.
>
> On 14 May 2015 at 14:50, Thomas Faber [e-mail] wrote:
> > On 2015-05-14 06:00, [e-mail] wrote:
> > > int sign = (copysignf(1, in) < 0);
> > > int sign = (in < 0);
> > > if (copysignf(1.0f, value) < 0.0f)
> > > if (value < 0.0f)
> > >       ++idx;
> >
> > I believe the behavior would be different here for negative zero:
> > copysignf(1.0f, -0.0f) should be < 0.0f
> > -0.0f should be == 0.0f
> >
> > Maybe that's the reason for having these calls?

[Timo Kreuzer]

ok, I'll bring it back after some cleanup.

Am 14.05.2015 um 18:18 schrieb Thomas Faber:
> Yes it did. Except that macro was a ROS-Diff in the first place ;)
>
> On 2015-05-14 11:12, David Quintana (gigaherz) wrote:
> > Judging by the involved macro, if -0.0 is not < 0, then the old code
> > had exactly the same bug.
> >
> > On 14 May 2015 at 14:50, Thomas Faber [e-mail] wrote:
> > > On 2015-05-14 06:00, [e-mail] wrote:
> > > > int sign = (copysignf(1, in) < 0);
> > > > int sign = (in < 0);
> > > > if (copysignf(1.0f, value) < 0.0f)
> > > > if (value < 0.0f)
> > > >       ++idx;
> > >
> > > I believe the behavior would be different here for negative zero:
> > > copysignf(1.0f, -0.0f) should be < 0.0f
> > > -0.0f should be == 0.0f
> > >
> > > Maybe that's the reason for having these calls?
>
> Ros-dev mailing list
> [e-mail]
> http://www.reactos.org/mailman/listinfo/ros-dev