[Bug target/85657] Make __ibm128 a separate type, even if long double uses the IBM double-double format

[meissner at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85657

--- Comment #7 from Michael Meissner <meissner at gcc dot gnu.org> ---
Author: meissner
Date: Fri Jun 22 14:11:56 2018
New Revision: 261891

URL: https://gcc.gnu.org/viewcvs?rev=261891&root=gcc&view=rev
Log:
[gcc]
2018-06-22  Michael Meissner  <meissner@linux.ibm.com>

        Back port from trunk
        2018-06-08  David Edelsohn  <dje.gcc@gmail.com>

        * config/rs6000/rs6000.c (rs6000_passes_ieee128): Protect with #if
        TARGET_ELF.

        Back port from trunk
        2018-06-04  Michael Meissner  <meissner@linux.ibm.com>

        * config/rs6000/rs6000.c (rs6000_passes_ieee128): New boolean to
        track if we pass or return IEEE 128-bit floating point.
        (ieee128_mangling_gcc_8_1): New boolean to say whether to generate
        C++ mangling that is compatible with GCC 8.1.
        (TARGET_ASM_GLOBALIZE_DECL_NAME): Override target hook.
        (init_cumulative_args): Note if we pass or return IEEE 128-bit
        floating point types.
        (rs6000_function_arg_advance_1): Likewise.
        (rs6000_mangle_type): Optionally generate mangled names that match
        what GCC 8.1 generated for IEEE 128-bit floating point types.
        (rs6000_globalize_decl_name): If we have an external function that
        passes or returns IEEE 128-bit types, generate a weak reference
        from the mangled name used in GCC 8.1 to the current mangled
        name.
        (rs6000_init_builtins): Make __ibm128 use the long double type if
        long double is IBM extended double.  Make __float128 use the long
        double type if long double is IEEE 128-bit.

        PR target/85657
        * config/rs6000/rs6000-builtin.def (BU_IBM128_2): New helper
        macro for __ibm128 built-in functions.
        (PACK_IF): Add __ibm128 pack/unpack functions.
        (UNPACK_IF): Likewise.
        * config/rs6000/rs6000.c (rs6000_builtin_mask_calculate): Do not
        enable long double built-in functions if long double is IEEE
        128-bit floating point.
        (rs6000_invalid_builtin): Update long double built-in function
        error message.
        (rs6000_expand_builtin): For PACK_IF and UNPACK_IF built-in
        functions, adjust the built-in function to use the long double
        built-in function if __ibm128 and long double are the same type.
        * doc/extend.texi (PowerPC builtins): Update documention for
        __builtin_{,un}pack_longdouble.  Add documentation for
        __builtin_{,un}pack_ibm128.

        Back port from trunk
        2018-06-01  Segher Boessenkool  <segher@kernel.crashing.org>

        * config/rs6000/rs6000.c (rs6000_mangle_type): Change the mangling of
        the 128-bit floating point types.  Fix function comment.

[gcc/testsuite]
2018-06-18  Michael Meissner  <meissner@linux.ibm.com>

        Back port from trunk
        2018-06-06  Michael Meissner  <meissner@linux.ibm.com>

        PR target/85657
        * g++.dg/pr85657.C: Only test whether __ibm128 and long double can
        be used in templates.  Don't check for them in overloaded functions.


Added:
    branches/gcc-8-branch/gcc/testsuite/gcc.target/powerpc/pr85657-4.c
    branches/gcc-8-branch/gcc/testsuite/gcc.target/powerpc/pr85657-5.c
    branches/gcc-8-branch/gcc/testsuite/gcc.target/powerpc/pr85657-6.c
Modified:
    branches/gcc-8-branch/gcc/ChangeLog
    branches/gcc-8-branch/gcc/config/rs6000/rs6000-builtin.def
    branches/gcc-8-branch/gcc/config/rs6000/rs6000.c
    branches/gcc-8-branch/gcc/doc/extend.texi
    branches/gcc-8-branch/gcc/testsuite/ChangeLog
    branches/gcc-8-branch/gcc/testsuite/g++.dg/pr85657.C