[Bug modula2/111956] New: Many powerpc platforms do _not_ have support for IEEE754 long double

[Bug modula2/111956] New: Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

            Bug ID: 111956
           Summary: Many powerpc platforms do _not_ have support for
                    IEEE754 long double
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: modula2
          Assignee: gaius at gcc dot gnu.org
          Reporter: gaius at gcc dot gnu.org
  Target Milestone: ---

The recent change to use -mabi=ieeelongdouble on powerpc64le has resulted in a
misconfigure on other powerpc targets.

cc1gm2: error: ‘-mabi=ieeelongdouble’ requires full ISA 2.06 support
compiler exited with status 1
FAIL: gm2/complex/run/pass/arith3.mod compilation,  -O -g

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |macro at orcam dot me.uk

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 112091 has been marked as a duplicate of this bug. ***

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[tschwinge at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

Thomas Schwinge <tschwinge at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |build
                 CC|                            |egallager at gcc dot gnu.org,
                   |                            |tschwinge at gcc dot gnu.org
   Last reconfirmed|                            |2023-10-31
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW

--- Comment #2 from Thomas Schwinge <tschwinge at gcc dot gnu.org> ---
While you (Gaius) here report test failure, similar to what Maciej had reported
in PR112091 "rs6000: redefinition of 'constexpr long double std::abs(long
double)', while building libgm2", I can't even build libgm2 anymore.

Similar had reported Eric (CCed here) on GCC IRC, 2023-10-20:

    <ericgallager> ok now this is an error I haven't seen before
    <ericgallager>
/home/egallager/gcc/.cfarm135_build.build/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/std_abs.h:137:3:
error: redefinition of 'constexpr long double std::abs(long double)'
    <ericgallager> seen on cfarm135
    <ericgallager> I think I remember something about __float128 being weird on
certain powerpc systems? Is there a configure flag that affects it?
    <ericgallager> I'll open a bug about it...
    <ericgallager> although actually wait... I'm not quite sure the component,
though... it's an error from a libstdc++ header, but it shows up while building
KeyBoardLEDs.lo for libm2cor in libgm2, but only for a specific target...

For example, I've got on powerpc64le-unknown-linux-gnu on indeed old Ubuntu
14.04 "trusty":

    +#define M2C_LONGREAL_FLOAT128 1
    +#define M2C_LONGREAL_PPC64LE 1

..., and this fails to build as follows:

    libtool: compile:  [...]/build-gcc/./gcc/xg++ -B[...]/build-gcc/./gcc/
-nostdinc++ -nostdinc++
-I[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/powerpc64le-unknown-linux-gnu
-I[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include
-I[...]/source-gcc/libstdc++-v3/libsupc++
-I[...]/source-gcc/libstdc++-v3/include/backward
-I[...]/source-gcc/libstdc++-v3/testsuite/util
-L[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/src
-L[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/src/.libs
-L[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/libsupc++/.libs
-B/powerpc64le-unknown-linux-gnu/bin/ -B/powerpc64le-unknown-linux-gnu/lib/
-isystem /powerpc64le-unknown-linux-gnu/include -isystem
/powerpc64le-unknown-linux-gnu/sys-include -c
-I[...]/source-gcc/libgm2/libm2cor -g -O2 -D_GNU_SOURCE -g -O2 -I. -I..
-I[...]/source-gcc/gcc/m2/gm2-libs -I[...]/source-gcc/gcc/m2/gm2-libs-iso
-DBUILD_GM2_LIBS -I[...]/source-gcc/libgm2/libm2cor/../
-I[...]/source-gcc/libgm2/libm2cor/../libm2iso -mabi=ieeelongdouble
[...]/source-gcc/libgm2/libm2cor/KeyBoardLEDs.cc  -fPIC -DPIC -o
.libs/KeyBoardLEDs.o
    cc1plus: warning: Using IEEE extended precision ‘long double’ [-Wpsabi]
    In file included from
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/cstdlib:81,
                     from
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/stdlib.h:36,
                     from [...]/source-gcc/libgm2/libm2cor/KeyBoardLEDs.cc:43:
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/std_abs.h:137:3:
error: redefinition of ‘constexpr long double std::abs(long double)’
      137 |   abs(__float128 __x)
          |   ^~~
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/std_abs.h:79:3:
note: ‘constexpr long double std::abs(long double)’ previously defined here
       79 |   abs(long double __x)
          |   ^~~
    make[5]: *** [KeyBoardLEDs.lo] Error 1

    [...]

    libtool: compile:  [...]/build-gcc/./gcc/xg++ -B[...]/build-gcc/./gcc/
-nostdinc++ -nostdinc++
-I[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/powerpc64le-unknown-linux-gnu
-I[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include
-I[...]/source-gcc/libstdc++-v3/libsupc++
-I[...]/source-gcc/libstdc++-v3/include/backward
-I[...]/source-gcc/libstdc++-v3/testsuite/util
-L[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/src
-L[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/src/.libs
-L[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/libsupc++/.libs
-B/powerpc64le-unknown-linux-gnu/bin/ -B/powerpc64le-unknown-linux-gnu/lib/
-isystem /powerpc64le-unknown-linux-gnu/include -isystem
/powerpc64le-unknown-linux-gnu/sys-include -c
-I[...]/source-gcc/libgm2/libm2pim -g -O2 -D_GNU_SOURCE -g -O2 -I. -I..
-I[...]/source-gcc/gcc/m2/gm2-libs -I[...]/source-gcc/gcc/m2/gm2-libs-iso
-DBUILD_GM2_LIBS -I[...]/source-gcc/libgm2/libm2pim/../
-I[...]/source-gcc/libgm2/libm2pim/../libm2iso -mabi=ieeelongdouble
[...]/source-gcc/libgm2/libm2pim/wrapc.cc  -fPIC -DPIC -o .libs/wrapc.o
    cc1plus: warning: Using IEEE extended precision ‘long double’ [-Wpsabi]
    In file included from
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/cmath:49,
                     from
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/math.h:36,
                     from [...]/source-gcc/libgm2/libm2pim/wrapc.cc:35:
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/std_abs.h:137:3:
error: redefinition of ‘constexpr long double std::abs(long double)’
      137 |   abs(__float128 __x)
          |   ^~~
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/std_abs.h:79:3:
note: ‘constexpr long double std::abs(long double)’ previously defined here
       79 |   abs(long double __x)
          |   ^~~
    In file included from
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/stl_pair.h:60,
                     from
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/stl_algobase.h:64,
                     from
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/specfun.h:43,
                     from
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/cmath:3699:
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/type_traits:514:12:
error: redefinition of ‘struct std::__is_floating_point_helper<long double>’
      514 |     struct __is_floating_point_helper<__float128>
          |            ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/type_traits:479:12:
note: previous definition of ‘struct std::__is_floating_point_helper<long
double>’
      479 |     struct __is_floating_point_helper<long double>
          |            ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/stl_algobase.h:1060:3:
error: redefinition of ‘constexpr long long int std::__size_to_integer(long
double)’
     1060 |   __size_to_integer(__float128 __n) { return (long long)__n; }
          |   ^~~~~~~~~~~~~~~~~
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/stl_algobase.h:1057:3:
note: ‘constexpr long long int std::__size_to_integer(long double)’ previously
defined here
     1057 |   __size_to_integer(long double __n) { return (long long)__n; }
          |   ^~~~~~~~~~~~~~~~~
    In file included from
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/bits/specfun.h:44:
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/limits:2089:12:
error: redefinition of ‘struct std::numeric_limits<long double>’
     2089 |     struct numeric_limits<__float128>
          |            ^~~~~~~~~~~~~~~~~~~~~~~~~~
   
[...]/build-gcc/powerpc64le-unknown-linux-gnu/libstdc++-v3/include/limits:1820:12:
note: previous definition of ‘struct std::numeric_limits<long double>’
     1820 |     struct numeric_limits<long double>
          |            ^~~~~~~~~~~~~~~~~~~~~~~~~~~
    make[5]: *** [wrapc.lo] Error 1

If I to revert commit r14-4149-g81d5ca0b9b8431f1bd7a5ec8a2c94f04bb0cf032 "PR
108143/modula2 LONGREAL and powerpc64le-linux" (..., which by now means
resolving conflicts due to subsequent changes), then I get clean build and test
results.

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[egallager at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #3 from Eric Gallager <egallager at gcc dot gnu.org> ---
(In reply to Thomas Schwinge from comment #2)
> While you (Gaius) here report test failure, similar to what Maciej had
> reported in PR112091 "rs6000: redefinition of 'constexpr long double
> std::abs(long double)', while building libgm2", I can't even build libgm2
> anymore.
> 
> Similar had reported Eric (CCed here) on GCC IRC, 2023-10-20:
> 
>     <ericgallager> ok now this is an error I haven't seen before
>     <ericgallager>
> /home/egallager/gcc/.cfarm135_build.build/powerpc64le-unknown-linux-gnu/
> libstdc++-v3/include/bits/std_abs.h:137:3: error: redefinition of 'constexpr
> long double std::abs(long double)'
>     <ericgallager> seen on cfarm135
>     <ericgallager> I think I remember something about __float128 being weird
> on certain powerpc systems? Is there a configure flag that affects it?
>     <ericgallager> I'll open a bug about it...
>     <ericgallager> although actually wait... I'm not quite sure the
> component, though... it's an error from a libstdc++ header, but it shows up
> while building KeyBoardLEDs.lo for libm2cor in libgm2, but only for a
> specific target...

Ah, thanks for the cc! I'd meant to look further into this, but got distracted
by other things...

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

Gaius Mulley <gaius at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gaius at gcc dot gnu.org

--- Comment #4 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Created attachment 56508
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56508&action=edit
Proposed fix

Here is a proposed patch which detects whether the target_cpu is power10 and
target is ppc64le.  I've seen it build successfully on gcc135 (without
bootstrap) and bootstrap on gcc120  (I'm currently running a bootstrap on
gcc135).

I'm unsure whether these libgm2/configure.ac and gcc/m2/configure.ac tests are
sufficient - and suspect more combinations should be checked (ie the shell case
statements might need expanding).

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[macro at orcam dot me.uk]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #5 from Maciej W. Rozycki <macro at orcam dot me.uk> ---
Hmm, this seems awkward to me.  It won't work AFAICT with the usual
native bootstrap environment, where you just run:

$ /path/to/configure && make bootstrap

and it won't work where you have an old version of glibc installed
and build with:

$ /path/to/configure --with-cpu=power10 && make bootstrap

either.  Shouldn't this just use the same check that libstdc++ does?
See the piece starting with:

# Check for IEEE128 support in libm.

in libstdc++-v3/configure.ac.  Maybe you just need to check for the
presence of `__frexpieee128' and ignore the rest, as you don't seem
to be willing to provide support for the IBM long double format in
libgm2, or do you?

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

Gaius Mulley <gaius at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #56508|0                           |1
        is obsolete|                            |

--- Comment #6 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Created attachment 56522
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56522&action=edit
Proposed fix v5

Many thanks for the hint about the __frexpieee128 libm test.

Here is the latest patch which passes all tests on gcc120 and has 96 failures
on gcc135.  It was configured with --with-long-double-format=ieee on gcc120.

It has separate checks (or switch) for host and target long double 128 support.

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

Gaius Mulley <gaius at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED

--- Comment #7 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Adding another patch using git diff -w to reduce patch clutter for review.

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #8 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Created attachment 56524
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56524&action=edit
Proposed fix v5 (git diff -w)

Here is the same patch as v5 but generated using git diff -w.

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[macro at orcam dot me.uk]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #9 from Maciej W. Rozycki <macro at orcam dot me.uk> ---
Hmm, the host check for `__frexpieee128' in gcc/ will surely not do
what's intended: even if the host is `powerpc*-*-linux*', the target
will often be something else and vice versa (libgm2's host is GCC's
target).

I think there is no way to verify target C library features in gcc/
at `configure' time, because at this point we may not yet have a target
compiler.  I haven't dealt with such a situation before, but AFAICS
people have used GCC_GLIBC_VERSION_GTE_IFELSE to explicitly check for
the glibc version required instead.  There's a relevant case for
TARGET_DEFAULT_LONG_DOUBLE_128 you can use as an example.

I'm not sure if such a check is needed though, unless perhaps for
sanity, as you only define TARGET_LIBM_PROVIDES_LONG_DOUBLE_IEEE128 if
--with-long-double-format=ieee has been explicitly given.

Also ISTM you can omit the target check for `powerpc64le-*-linux*' here
keeping $with_long_double_format=ieee check only and get support for the
relevant `powerpc64-*-linux*' targets too, as -with-long-double-format=
will already have verified correct usage.

Finally you may or may not have to check for $gcc_cv_target_ldbl128
equal to "yes" too, in case someone has used --without-long-double-128
(I'm not sure what the consequences would be, but it has caught my
attention, so please double-check).

Overall please refer to someone more familiar with POWER GCC targets as
I can only comment based on what I can see in the scripts and my general
experience.  The best course of action might be submitting the patch to
gcc-patches for review, cc-ing RS6000 port maintainers, as the most
relevant people may not be reading this bug report.

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[tschwinge at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #10 from Thomas Schwinge <tschwinge at gcc dot gnu.org> ---
In addition to what Maciej said (..., and similarly, I don't have any proper
knowledge about PowerPC details):

(In reply to Gaius Mulley from comment #6)
> Created attachment 56522 [details]
> Proposed fix v5

Thanks for looking into this!

> Here is the latest patch which [...] 96
> failures on gcc135.  [...]

With no special 'configure' flags, I'm seeing (presumably) those, too.

I noticed in 'build-gcc/gcc/m2/config-make' (generated from
'gcc/m2/config-make.in'):

    # Does the target have -mabi=ieeelongdouble support in libm?  (yes/no).
    HAVE_TARGET_LONG_DOUBLE_IEEE = @have_target_long_double_ieee@

..., so missing 'AC_SUBST' or similar -- but is that actually unused?

I further noticed the following delta when regenerating 'libgm2/configure':

     case "$target" in
        powerpc*-*-linux*)
    -     LONG_DOUBLE_COMPAT_FLAGS="$LONG_DOUBLE_COMPAT_FLAGS
-mno-gnu-attribute"
          # Check for IEEE128 support in libm:


(In reply to Gaius Mulley from comment #8)
> Here is the same patch as v5 but generated using git diff -w.

Please don't include unrelated changes (here: whitespace cleanup); handle that
separately (if you must).  ;-)

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[tkoenig at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

Thomas Koenig <tkoenig at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tkoenig at gcc dot gnu.org

--- Comment #11 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
A remark - gfortran handles 128-bit reals on POWER as well, it might be a good
idea to look into libgfortran's configure scripts.

[Bug modula2/111956] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #12 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Many thanks for the feedback and testing - I'll rework the patch and post it to
gcc-patches as suggested.

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |14.0
                 CC|                            |rguenth at gcc dot gnu.org
             Target|                            |powerpc*-*-*
            Summary|Many powerpc platforms do   |[14 Regression] Many
                   |_not_ have support for      |powerpc platforms do _not_
                   |IEEE754 long double         |have support for IEEE754
                   |                            |long double

--- Comment #13 from Richard Biener <rguenth at gcc dot gnu.org> ---
It's a regression btw, GCC 13 was fine with this configuration.  I'm also
seeing this.

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #14 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Ah apologies, is it best that I revert:

https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;h=81d5ca0b9b8431f1bd7a5ec8a2c94f04bb0cf032

happy to do this in the morning.

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Gaius Mulley from comment #14)
> Ah apologies, is it best that I revert:
> 
> https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;
> h=81d5ca0b9b8431f1bd7a5ec8a2c94f04bb0cf032
> 
> happy to do this in the morning.

I think it might be better to define M2C_LONGREAL_FLOAT128/M2C_LONGREAL_PPC64LE
(whatever they exactly indicate) in terms of the existing

--with-long-double-128
--with-long-double-format

aka the TARGET_DEFAULT_LONG_DOUBLE_128 that's put into the config plus
with_long_double_format (I think that causes TARGET_IEEEQUAD_DEFAULT
to be defined to 1/0, but only for ppc, via config.gcc and tm_defines).

I can't say whether it's better to revert or disable/fix as I can't say
how this for example affects the M2 ABI (like if there was any 'long double'
before this change and what effective type this used).

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaiusmod2 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #16 from gaiusmod2 at gmail dot com ---
"rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org> writes:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956
>
> --- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> ---
> (In reply to Gaius Mulley from comment #14)
>> Ah apologies, is it best that I revert:
>> 
>> https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;
>> h=81d5ca0b9b8431f1bd7a5ec8a2c94f04bb0cf032
>> 
>> happy to do this in the morning.
>
> I think it might be better to define M2C_LONGREAL_FLOAT128/M2C_LONGREAL_PPC64LE
> (whatever they exactly indicate) in terms of the existing
>
> --with-long-double-128
> --with-long-double-format
>
> aka the TARGET_DEFAULT_LONG_DOUBLE_128 that's put into the config plus
> with_long_double_format (I think that causes TARGET_IEEEQUAD_DEFAULT
> to be defined to 1/0, but only for ppc, via config.gcc and
> tm_defines).

yes thanks for the hints this sounds good.  I'll pursue this line for a fix.

> I can't say whether it's better to revert or disable/fix as I can't say
> how this for example affects the M2 ABI (like if there was any 'long double'
> before this change and what effective type this used).

Prior to the patch it used the default C long double type but many of
the intrinsic functions were broken.  Currently with the patch gcc120
has no regression test failures and gcc135 (yesterday with manual
configure intervention) was at 96 failures in the m2 testsuite.

I suspect configure confusion and hence using
TARGET_DEFAULT_LONG_DOUBLE_128 and TARGET_IEEEQUAD_DEFAULT should
resolve it.

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #17 from Richard Biener <rguenth at gcc dot gnu.org> ---
Note that the default C long double is what is controlled by
--with-long-double-128 and --with-long-double-format, those configs do not tell
you
whether using __ieee128 with -mabi=ieeelongdouble will work.

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #18 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Created attachment 57110
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57110&action=edit
Work in progress fix

Many thanks for the config.gcc hints!  It now builds on gcc120 and gcc135 with
the work in progress patch.

Currently it lacks the -mabi= handling in gm2spec.cc, I've not revisited this
file yet and so any #ifdef M2C_LONGREAL (is now always) undefined.

However cc1gm2 uses the config.gcc and tm_defines and a summary of results on
gcc120:

                === gm2 Summary ===

# of expected passes            12962
# of unexpected failures        24
# of unresolved testcases       6

and on gcc135:

                === gm2 Summary ===

# of expected passes            12974
# of unexpected failures        18

I think the -mabi= should just override the default settings - and hopefully
the results will improve.

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

Gaius Mulley <gaius at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #56522|0                           |1
        is obsolete|                            |
  Attachment #56524|0                           |1
        is obsolete|                            |
  Attachment #57110|0                           |1
        is obsolete|                            |

--- Comment #19 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Created attachment 57122
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57122&action=edit
Proposed fix

Here is the proposed fix - which in addition to the previous work in progress
patch implements -mabi= options.   Many thanks for all the hints.  I'll post it
to gcc patches as suggested above.

Seen it bootstrap on:

power8 
======

# of expected passes            12974
# of unexpected failures        18

power9
======

# of expected passes            12992

x86_64
======

# of expected passes            12992

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #20 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Created attachment 57123
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57123&action=edit
ChangeLog for proposed fix

For completeness here is the changelog.

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

--- Comment #21 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Gaius Mulley <gaius@gcc.gnu.org>:

https://gcc.gnu.org/g:f2872e00f61a5a333fa22e651d37d3b2035ac41f

commit r14-8227-gf2872e00f61a5a333fa22e651d37d3b2035ac41f
Author: Gaius Mulley <gaiusmod2@gmail.com>
Date:   Thu Jan 18 13:06:30 2024 +0000

    PR modula2/111956 Many powerpc platforms do _not_ have support for IEEE754

    This patch corrects commit
    r14-4149-g81d5ca0b9b8431f1bd7a5ec8a2c94f04bb0cf032 which assummed
    all powerpc platforms would have IEEE754 long double.  The patch
    ensures that cc1gm2 obtains the default IEEE754 long double availability
    from the configure generated tm_defines.  The user command
    line switches -mabi=ibmlongdouble and -mabi=ieeelongdouble are implemented
    to override the configuration defaults.

    Bootstrapped on power8 and power9 machines.

    gcc/m2/ChangeLog:

            PR modula2/111956
            * Make-lang.in (host_mc_longreal): Remove.
            * configure: Regenerate.
            * configure.ac (M2C_LONGREAL_FLOAT128): Remove.
            (M2C_LONGREAL_PPC64LE): Remove.
            * gm2-compiler/M2Options.def (SetIBMLongDouble): New procedure.
            (GetIBMLongDouble): New procedure function.
            (SetIEEELongDouble): New procedure.
            (GetIEEELongDouble): New procedure function.
            * gm2-compiler/M2Options.mod (SetIBMLongDouble): New procedure.
            (GetIBMLongDouble): New procedure function.
            (SetIEEELongDouble): New procedure.
            (GetIEEELongDouble): New procedure function.
            (InitializeLongDoubleFlags): New procedure called during
            module block initialization.
            * gm2-gcc/m2configure.cc: Remove duplicate includes.
            (m2configure_M2CLongRealFloat128): Remove.
            (m2configure_M2CLongRealIBM128): Remove.
            (m2configure_M2CLongRealLongDouble): Remove.
            (m2configure_M2CLongRealLongDoublePPC64LE): Remove.
            (m2configure_TargetIEEEQuadDefault): New function.
            * gm2-gcc/m2configure.def (M2CLongRealFloat128): Remove.
            (M2CLongRealIBM128): Remove.
            (M2CLongRealLongDouble): Remove.
            (M2CLongRealLongDoublePPC64LE): Remove.
            (TargetIEEEQuadDefault): New function.
            * gm2-gcc/m2configure.h (m2configure_M2CLongRealFloat128): Remove.
            (m2configure_M2CLongRealIBM128): Remove.
            (m2configure_M2CLongRealLongDouble): Remove.
            (m2configure_M2CLongRealLongDoublePPC64LE): Remove.
            (m2configure_TargetIEEEQuadDefault): New function.
            * gm2-gcc/m2options.h (M2Options_SetIBMLongDouble): New prototype.
            (M2Options_GetIBMLongDouble): New prototype.
            (M2Options_SetIEEELongDouble): New prototype.
            (M2Options_GetIEEELongDouble): New prototype.
            * gm2-gcc/m2type.cc (build_m2_long_real_node): Re-implement using
            results of M2Options_GetIBMLongDouble and
M2Options_GetIEEELongDouble.
            * gm2-lang.cc (gm2_langhook_handle_option): Add case
            OPT_mabi_ibmlongdouble and call M2Options_SetIBMLongDouble.
            Add case OPT_mabi_ieeelongdouble and call
M2Options_SetIEEELongDouble.
            * gm2config.aci.in: Regenerate.
            * gm2spec.cc (lang_specific_driver): Remove block defined by
            M2C_LONGREAL_PPC64LE.
            Remove case OPT_mabi_ibmlongdouble.
            Remove case OPT_mabi_ieeelongdouble.

    libgm2/ChangeLog:

            PR modula2/111956
            * Makefile.am (TARGET_LONGDOUBLE_ABI): Remove.
            * Makefile.in: Regenerate.
            * libm2cor/Makefile.am (TARGET_LONGDOUBLE_ABI): Remove.
            * libm2cor/Makefile.in: Regenerate.
            * libm2iso/Makefile.am (TARGET_LONGDOUBLE_ABI): Remove.
            * libm2iso/Makefile.in: Regenerate.
            * libm2log/Makefile.am (TARGET_LONGDOUBLE_ABI): Remove.
            * libm2log/Makefile.in: Regenerate.
            * libm2min/Makefile.am (TARGET_LONGDOUBLE_ABI): Remove.
            * libm2min/Makefile.in: Regenerate.
            * libm2pim/Makefile.am (TARGET_LONGDOUBLE_ABI): Remove.
            * libm2pim/Makefile.in: Regenerate.

    Signed-off-by: Gaius Mulley <gaiusmod2@gmail.com>

[Bug modula2/111956] [14 Regression] Many powerpc platforms do _not_ have support for IEEE754 long double

[gaius at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111956

Gaius Mulley <gaius at gcc dot gnu.org> changed:

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

--- Comment #22 from Gaius Mulley <gaius at gcc dot gnu.org> ---
Closing now that the patch has been applied.

Also acknowledging that this fix doesn't solve the libm2 ABI compatibility.
I suspect a new PR should be opened for the ABI issue - maybe referring back
here.
But closing this PR as the regression has gone.