Re: [Dri-devel] Re: [Dri-devel][patch] 3DNow normalization bug

[Felix <fx...@gm...>]

On Thu, 30 Jan 2003 00:14:43 +0100
Dieter Nützel <Die...@ha...> wrote:

> Am Mittwoch, 29. Januar 2003 22:30 schrieb Felix Kühling:
> > On Wed, 29 Jan 2003 07:59:30 -0700
> >
> > Brian Paul <br...@tu...> wrote:
> > > >>> Since these functions are globally exported, it might be worth it to
> > > >>> write a quick test that calls the various
> > > >>> _transform_normalize_normals functions to make sure that they all
> > > >>> produces the same (or close enough) results.
> > > >>
> > > >> And:
> > > >> _transform_normalize_normals_no_rot
> > > >> _transform_rescale_normals_no_rot
> > > >> _transform_rescale_normals
> > > >> _transform_normals_no_rot
> > > >> _transform_normals
> > > >> _normalize_normals
> > > >> _rescale_normals
> > > >>
> > > >> These should be tested too, while we're at it.
> > > >
> > > > Agreed.
> > > >
> > > > Brian: If such tests do get written, where should they live in the
> > > > tree?
> > >
> > > There's a number of test routines in the src/math/m_debug_*.c files.  I
> > > think that would be the place to add more if needed.
> >
> > It's all there. Even benchmarking is included. It just needs to be
> > recompiled with DEBUG and RUN_DEBUG_BENCHMARK defined.
> >
> > I downloaded Mesa CVS and configured it with --enable-debug and
> > --enable-profile. When I started a GL app with the correct library path
> > I got this:
> >
> > cpu vendor: AuthenticAMD
> > cpu name: AMD Duron(tm) processor
> > MMX cpu detected.
> > 3DNow! cpu detected.
> > -----------------------------
> > (i = 0, j = 0)
> > 1.177931         -0.033552       [ratio = -2.848425e-02 - 29 bit missed]
> > 0.841573         0.728316        [ratio = 8.654217e-01 - 21 bit missed]
> > 0.735570         -0.684420       [ratio = -9.304624e-01 - 25 bit missed]
> > Mesa implementation error: _mesa_normal_tab[0][NORM_NORMALIZE] failed test
> > (3DNow!) Please report to the Mesa bug database at www.mesa3d.org
> > Mesa: Mesa DEBUG build Jan 29 2003 21:43:51
> 
> Mine is _longer_ (dual Athlon MP 1900+;-)
> 
> Mesa/demos> ./glinfo
> cpu vendor: AuthenticAMD
> cpu name: AMD Athlon(tm) MP
> MMX cpu detected.
> 3DNow! cpu detected.
> -----------------------------
> (i = 0, j = 0)
> 1.177931         -0.033552       [ratio = -2.848425e-02 - 29 bit missed]
> 0.841573         0.728316        [ratio = 8.654216e-01 - 21 bit missed]
> 0.735570         -0.684420       [ratio = -9.304624e-01 - 25 bit missed]
> Mesa implementation error: _mesa_normal_tab[0][NORM_NORMALIZE] failed test 
> (3DNow!)
> Please report to the Mesa bug database at www.mesa3d.org
> Testing OS support for SSE... yes.
> Testing OS support for SSE unmasked exceptions... SIGFPE, yes.
> Tests of OS support for SSE passed.
> SSE cpu detected.
> -----------------------------
> (i = 0, j = 0)
> 0.356922         -0.121018       [ratio = -3.390600e-01 - 26 bit missed]
> 0.288519         0.972743        [ratio = 3.371504e+00 - 25 bit missed]
> -0.547221        0.197804        [ratio = -3.614698e-01 - 26 bit missed]
> Mesa implementation error: _mesa_normal_tab[0][NORM_NORMALIZE] failed test 
> (SSE)
> Please report to the Mesa bug database at www.mesa3d.org
> Mesa: Mesa DEBUG build Jan 29 2003 23:50:36
> GL_VERSION: 1.4 Mesa 5.1
> 
> > A patch to fix this is attached. Basically the same kind of problem as
> > before. Now it reports no more problems :) in any of the math functions
> > tested on my Duron. This should be about everything except SSE.
> 
> _BOTH_ are fixed through your patch.

That's because there is no such SSE function. So after installing SSE
functions the same 3dnow normalize was tested again. Thanks for testing
on an SSE machine.

Felix

               __\|/__    ___     ___     ___
__Tschüß_______\_6 6_/___/__ \___/__ \___/___\___You can do anything,___
_____Felix_______\Ä/\ \_____\ \_____\ \______U___just not everything____
  fx...@gm...    >o<__/   \___/   \___/        at the same time!