[Bug libgcc/97543] New: powerpc64le: libgcc has unexpected long double in .gnu_attribute

[Bug libgcc/97543] New: powerpc64le: libgcc has unexpected long double in .gnu_attribute

[tuliom at ascii dot art.br]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

            Bug ID: 97543
           Summary: powerpc64le: libgcc has unexpected long double in
                    .gnu_attribute
           Product: gcc
           Version: 10.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libgcc
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tuliom at ascii dot art.br
  Target Milestone: ---

Alpine Linux uses musl as C Library.
musl only supports long double as IEEE binary64.
It does not support IBM long double.

In this scenario, building gcc with --disable-multilib --with-long-double-64
still creates a libgcc with a .gnu_attribute pointing to IBM long double,
causing unnecessary link time warnings.

Reproduced as:

$ cat test.c
#include <stdio.h>
#include <stdint.h>
int
main()
{
  long double a = 2.0;
  __int128 b = 8;
  printf("sizeof(long double) = %ld\n", sizeof(long double));
  printf("a * b = %Lf\n", a * (long double) b);
  return 0;
}
$ gcc test.c && ./a.out
/usr/lib/gcc/powerpc64le-alpine-linux-musl/10.2.0/../../../../powerpc64le-alpine-linux-musl/bin/ld:
/tmp/ccmCAapA.o uses 64-bit long double,
/usr/lib/gcc/powerpc64le-alpine-linux-musl/10.2.0/../../../../lib/libgcc_s.so.1
uses 128-bit long double
/usr/lib/gcc/powerpc64le-alpine-linux-musl/10.2.0/../../../../powerpc64le-alpine-linux-musl/bin/ld:
/tmp/ccmCAapA.o uses 64-bit long double,
/usr/lib/gcc/powerpc64le-alpine-linux-musl/10.2.0/../../../../lib/libgcc_s.so.1
uses 128-bit long double
sizeof(long double) = 8
a * b = 16.000000
$ gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/powerpc64le-alpine-linux-musl/10.2.0/lto-wrapper
Target: powerpc64le-alpine-linux-musl
Configured with: /home/devel/aports/main/gcc/src/gcc-10.2.0/configure
--prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info
--build=powerpc64le-alpine-linux-musl --host=powerpc64le-alpine-linux-musl
--target=powerpc64le-alpine-linux-musl --with-pkgversion='Alpine 10.2.0'
--enable-checking=release --disable-fixed-point --disable-libstdcxx-pch
--disable-multilib --disable-nls --disable-werror --disable-symvers
--enable-__cxa_atexit --enable-default-pie --enable-default-ssp
--enable-cloog-backend --enable-languages=c,c++,objc,go,fortran,ada
--with-abi=elfv2 --enable-secureplt --enable-decimal-float=no
--enable-targets=powerpcle-linux --with-long-double-64 --disable-libquadmath
--disable-libssp --disable-libmpx --disable-libmudflap --disable-libsanitizer
--enable-shared --enable-threads --enable-tls --with-system-zlib
--with-linker-hash-style=gnu
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 10.2.0 (Alpine 10.2.0)

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[gustavowalbon at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

Gustavo Walbon <gustavowalbon at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gustavowalbon at gmail dot com

--- Comment #1 from Gustavo Walbon <gustavowalbon at gmail dot com> ---
I have seeing a bunch of packages failing because this warning of linker when I
build several packages on Alpine for PowerPc.

I have a alpine container running on a PowerPC 8, and I rebuilt the all gcc
packages with the flags --with-long-double-64 and --without-long-double-128 to
avoid this error, but that still happens 

I used the same code as Tulio to reproduce that warning.

~ $ gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/powerpc64le-alpine-linux-musl/10.2.0/lto-wrapper
Target: powerpc64le-alpine-linux-musl
Configured with: /home/devel/aports/main/gcc/src/gcc-10.2.0/configure
--prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info
--build=powerpc64le-alpine-linux-musl --host=powerpc64le-alpine-linux-musl
--target=powerpc64le-alpine-linux-musl --with-pkgversion='Alpine 10.2.0'
--enable-checking=release --disable-fixed-point --disable-libstdcxx-pch
--disable-multilib --disable-nls --disable-werror --disable-symvers
--enable-__cxa_atexit --enable-default-pie --enable-default-ssp
--enable-cloog-backend --enable-languages=c,c++,objc,go,fortran,ada
--with-abi=elfv2 --enable-secureplt --enable-decimal-float=no
--enable-targets=powerpcle-linux --with-long-double-64
--without-long-double-128 --disable-libquadmath --disable-libssp
--disable-libmpx --disable-libmudflap --disable-libsanitizer --enable-shared
--enable-threads --enable-tls --with-system-zlib --with-linker-hash-style=gnu
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 10.2.0 (Alpine 10.2.0)

~ $ gcc -O0 -mlong-double-64 test-ldouble64.c && ./a.out
/usr/lib/gcc/powerpc64le-alpine-linux-musl/10.2.0/../../../../powerpc64le-alpine-linux-musl/bin/ld:
/tmp/ccHidlkA.o uses 64-bit long double,
/usr/lib/gcc/powerpc64le-alpine-linux-musl/10.2.0/../../../../lib/libgcc_s.so.1
uses 128-bit long double
/usr/lib/gcc/powerpc64le-alpine-linux-musl/10.2.0/../../../../powerpc64le-alpine-linux-musl/bin/ld:
/tmp/ccHidlkA.o uses 64-bit long double,
/usr/lib/gcc/powerpc64le-alpine-linux-musl/10.2.0/../../../../lib/libgcc_s.so.1
uses 128-bit long double
sizeof(long double) = 8
a * b = 16.000000

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[bergner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

Peter Bergner <bergner at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amodra at gcc dot gnu.org,
                   |                            |bergner at gcc dot gnu.org,
                   |                            |meissner at gcc dot gnu.org,
                   |                            |segher at gcc dot gnu.org,
                   |                            |seurer at gcc dot gnu.org

--- Comment #2 from Peter Bergner <bergner at gcc dot gnu.org> ---
There is no --with-long-double-64 configure option (for power).  The configure
script will just silently ignore it.  That said, you mentioned you used
--without-long-double-128 which is the correct way to get a 64-bit long double,
so there does seem to be a problem here.

I'm adding Alan and Mike to the CC list here for their input on how this is all
supposed to work.  Given you disabled multilib and forced 64-bit long double, I
don't know why libgcc has any 128-bit long double usage.

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[segher at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

--- Comment #3 from Segher Boessenkool <segher at gcc dot gnu.org> ---
This part of the attribute (all but the low 2 bits) is not documented
in the as manual, btw.

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[bergner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

--- Comment #4 from Peter Bergner <bergner at gcc dot gnu.org> ---
Created attachment 49436
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=49436&action=edit
patch file

So libgcc compiles are explicitly using -mlong-double-128, which doesn't seem
right when GCC is configured with --disable-multilib --without-long-double-128,
which seems to come from:

libgcc/config/rs6000/t-linux:HOST_LIBGCC2_CFLAGS += -mlong-double-128

Shouldn't we just use the compiler default for long double (ie, don't use that
option at all)?

Something like the attached patch?

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[meissner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

--- Comment #5 from Michael Meissner <meissner at gcc dot gnu.org> ---
One of my patches for adding IEEE 128-bit long double may help with this
situation.  The ibm-ldouble.c module was not being compiled with
-mno-gnu-attributes would affect things if a different long double type was
used:
https://gcc.gnu.org/pipermail/gcc-patches/2020-October/556863.html

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[amodra at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

Alan Modra <amodra at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amodra at gmail dot com

--- Comment #6 from Alan Modra <amodra at gmail dot com> ---
(In reply to Peter Bergner from comment #4)
> Something like the attached patch?
No.  -mlong-double-128 overrides the configured --without-long-double-128. 
--without-long-double-128 doesn't *disable* 128-bit long doubles, it just sets
the default to 64-bit long double.

(In reply to Michael Meissner from comment #5)
Yes, -mno-gnu-attribute makes sense, however it should only apply to files
linked into the shared libgcc, not IMO to files in libgcc.a.

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[meissner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

--- Comment #7 from Michael Meissner <meissner at gcc dot gnu.org> ---
Just to be clear, my patch only turns on -mno-gnu-attributes on compiling
ibm-ldouble.c.  That is the module that has the extended IBM 128-bit support in
it.

However, I believe if any module in libgcc_s.so has the IBM extended long
double attributes set, it essentially taints the entire shared library.

At least on 64-bit little endian Linux, it does not use long double at all.  It
uses __ibm128 explicitly.

However, t-linux and t-ibm-ldouble both explicitly set -mlong-double-128. 
Since long double is explicitly set to 128-bit, it sets the gnu attributes for
__ibm128 (since internally __ibm128 is mapped to long double in this case).

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[meissner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

--- Comment #8 from Michael Meissner <meissner at gcc dot gnu.org> ---
In addition to ibm-ldouble.c, the following functions set the gnu attribute #4
to 5 (i.e. pass/use IBM extended double as long double):

_divtc3
_fixtfdi
_fixunstfdi
_floatditf
_floatunditf
_multc3
_powitf2
ppc64-fp

This means we have to change the definition of TFmode and TCtype from:

typedef float TFtype __attribute__ ((mode (TF)));
typedef _Complex float TCtype __attribute__ ((mode (TC)));

to:

typedef float TFtype __attribute__ ((mode (IF)));
typedef _Complex float TCtype __attribute__ ((mode (IC)));

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[segher at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

--- Comment #9 from Segher Boessenkool <segher at gcc dot gnu.org> ---
Yes, that looks correct.

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

--- Comment #10 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Michael Meissner <meissner@gcc.gnu.org>:

https://gcc.gnu.org/g:9f1a6501994a2d18ec4fe2a6664637f48021b210

commit r11-5728-g9f1a6501994a2d18ec4fe2a6664637f48021b210
Author: Michael Meissner <meissner@linux.ibm.com>
Date:   Thu Dec 3 14:50:26 2020 -0500

    PowerPC: PR libgcc/97543 and libgcc/97643, fix long double issues

    If you use a compiler with long double defaulting to 64-bit instead of
128-bit
    with IBM extended double, you get linker warnings about mis-matches in the
gnu
    attributes for long double (PR libgcc/97543).  Even if the compiler is
    configured to have long double be 64 bit as the default with the
configuration
    option '--without-long-double-128' you get the warnings.

    You also get the same issues if you use a compiler with long double
defaulting
    to IEEE 128-bit instead of IBM extended double (PR libgcc/97643).

    The issue is the way libgcc.a/libgcc.so is built.  Right now when building
    libgcc under Linux, the long double size is set to 128-bits when building
    libgcc.  However, the gnu attributes are set, leading to the warnings.

    One feature of the current GNU attribute implementation is if you have a
shared
    library (such as libgcc_s.so), the GNU attributes for the shared library is
an
    inclusive OR of all of the objects within the library.  This means if any
    object file that uses the -mlong-double-128 option and uses long double,
the GNU
    attributes for the library will indicate that it uses 128-bit IBM long
    doubles.  If you have a static library, you will get the warning only if
you
    actually reference an object file  with the attribute set.

    This patch does two things:

        1)  All of the object files that support IBM 128-bit long doubles
            explicitly set the ABI to IBM extended double.

        2)  I turned off GNU attributes for building the shared library or for
            building the IBM 128-bit long double support.

    libgcc/
    2020-12-03  Michael Meissner  <meissner@linux.ibm.com>

            PR libgcc/97543
            PR libgcc/97643
            * config/rs6000/t-linux (IBM128_STATIC_OBJS): New make variable.
            (IBM128_SHARED_OBJS): New make variable.
            (IBM128_OBJS): New make variable.  Set all objects to use the
            explicit IBM format, and disable gnu attributes.
            (IBM128_CFLAGS): New make variable.
            (gcc_s_compile): Add -mno-gnu-attribute to all shared library
            modules.

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[gustavowalbon at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

--- Comment #11 from Gustavo Walbon <gustavowalbon at gmail dot com> ---
Michael,

Works with GCC v10.
I backported this patch for the Alpine[1] gcc v10 and then I used the libgcc
package to test. I have used the binutils v2.35.1 with the warning of the GNU
Attribute and the patch doesn't show the warning.


1 -
https://github.com/walbon/aports/commit/f188c0b799df8550c595d6a94ffc6dab0254d11a

Thanks Michael

[Bug libgcc/97543] powerpc64le: libgcc has unexpected long double in .gnu_attribute

[meissner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97543

Michael Meissner <meissner at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |FIXED

--- Comment #12 from Michael Meissner <meissner at gcc dot gnu.org> ---
Back port to GCC 10 done on January 20th, 2021.  Bug closed.