[3/3] sparc32/sparcv9: add a VIS3 version of fdim

On 08/03/2016 08:09 PM, Aurelien Jarno wrote:
> diff --git a/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> new file mode 100644
> index 0000000..c9c7704
> --- /dev/null
> +++ b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> @@ -0,0 +1,32 @@
> +/* Compute positive difference, sparc 32-bit.
> +   Copyright (C) 2016 Free Software Foundation, Inc.
> +   This file is part of the GNU C Library.
> +
> +   The GNU C Library is free software; you can redistribute it and/or
> +   modify it under the terms of the GNU Lesser General Public
> +   License as published by the Free Software Foundation; either
> +   version 2.1 of the License, or (at your option) any later version.
> +
> +   The GNU C Library is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> +   Lesser General Public License for more details.
> +
> +   You should have received a copy of the GNU Lesser General Public
> +   License along with the GNU C Library; if not, see
> +   <http://www.gnu.org/licenses/>.  */
> +
> +#ifdef HAVE_AS_VIS3_SUPPORT
> +# include <sparc-ifunc.h>
> +# include <math.h>
> +
> +extern double __fdim_vis3 (double, double);
> +extern double __fdim_generic (double, double);
> +
> +sparc_libm_ifunc(__fdim, hwcap & HWCAP_SPARC_VIS3 ? __fdim_vis3 : __fdim_generic);
> +weak_alias (__fdim, fdim)
> +
> +# define __fdim __fdim_generic
> +#endif
> +
> +#include <math/s_fdim.c>

Hi Aurelien,

I was working on a patch to use the new generic libm infrastructure on fdim*.
While testing the sparc changes, I noticed the fdiml@GLIBC_2.1 symbol is
missing.  I built this from a recent copy of master.  The include of s_fdim.c
seems to pick up the copy in math/, not in sysdeps/ieee754/ldbl-opt/ which
provides the extra compat symbols.  I filed BZ 20517, feel free to close if
this is a mistake.

I built using the F22 sparc cross-toolchain via a command similar to:

../configure CXX=broken CC="sparc64-linux-gnu-gcc -m32" --prefix=/usr/     \
             --with-headers=/path/to/generic/sparc/headers/                \
             --host=sparcv9-linux-gnu libc_cv_ssp=no libc_cv_ssp_strong=no

Likewise, the following seemed to restore it:


Thanks,
Paul

On 08/26/2016 09:25 AM, Paul E. Murphy wrote:
> diff --git a/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> index c9c7704..9666741 100644
> --- a/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> +++ b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> @@ -29,4 +29,4 @@ weak_alias (__fdim, fdim)
>  # define __fdim __fdim_generic
>  #endif
> 
> -#include <math/s_fdim.c>
> +#include <ldbl-opt/s_fdim.c>

Note, Joseph mentioned in BZ this is used for 64b builds,
which does not use ldbl-opt. So, this is not a correct fix.

On Fri, 26 Aug 2016, Paul E. Murphy wrote:

> On 08/26/2016 09:25 AM, Paul E. Murphy wrote:
> > diff --git a/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> > index c9c7704..9666741 100644
> > --- a/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> > +++ b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> > @@ -29,4 +29,4 @@ weak_alias (__fdim, fdim)
> >  # define __fdim __fdim_generic
> >  #endif
> > 
> > -#include <math/s_fdim.c>
> > +#include <ldbl-opt/s_fdim.c>
> 
> Note, Joseph mentioned in BZ this is used for 64b builds,

As it's in sysdeps/sparc/sparc32/sparcv9 it shouldn't be used for 64b 
builds....

On 2016-08-26 09:34, Paul E. Murphy wrote:
> 
> 
> On 08/26/2016 09:25 AM, Paul E. Murphy wrote:
> > diff --git a/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> > index c9c7704..9666741 100644
> > --- a/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> > +++ b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
> > @@ -29,4 +29,4 @@ weak_alias (__fdim, fdim)
> >  # define __fdim __fdim_generic
> >  #endif
> > 
> > -#include <math/s_fdim.c>
> > +#include <ldbl-opt/s_fdim.c>
> 
> Note, Joseph mentioned in BZ this is used for 64b builds,
> which does not use ldbl-opt. So, this is not a correct fix.

I actually think your fix is the correct one. I have found an old build
log and this is the file used when building with --disable-multiarch. I
will run a build to confirm that.

I do wonder if simply using "#include_next <s_fdim.c>" would prevent
this kind of error. It's a GNU extension, but it's seems to be used in
many places already, so it's probably acceptable.

Aurelien

From: Joseph Myers <joseph@codesourcery.com>
Date: Fri, 26 Aug 2016 16:02:11 +0000

> On Fri, 26 Aug 2016, Paul E. Murphy wrote:
> 
>> On 08/26/2016 09:25 AM, Paul E. Murphy wrote:
>> > diff --git a/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
>> > index c9c7704..9666741 100644
>> > --- a/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
>> > +++ b/sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c
>> > @@ -29,4 +29,4 @@ weak_alias (__fdim, fdim)
>> >  # define __fdim __fdim_generic
>> >  #endif
>> > 
>> > -#include <math/s_fdim.c>
>> > +#include <ldbl-opt/s_fdim.c>
>> 
>> Note, Joseph mentioned in BZ this is used for 64b builds,
> 
> As it's in sysdeps/sparc/sparc32/sparcv9 it shouldn't be used for 64b 
> builds....

Right, that directory is for 32-bit v9 builds.