ping
At 2025-07-24 00:54:22, "ZAMBAR" <zam...@163.com> wrote:
> This patch is an experimental implementation of multi-target clones
>clones for Fortran, allowing the use of the `TARGET_CLONES` attribute.
>It is designed to work with the syntax like:
>
>`!GCC$ ATTRIBUTES TARGET_CLONES("default", "avx", "avx512f") :: MySub`
>
> It is now worked with both `SUBROUTINE` and `FUNCTION`, using
>infrastructures provided by the GCC. Here is a brief request for
>comments (RFC) on the implementation:
>
>- Which syntax is preferred for the attribute? For now I've
> implemented C-like syntax, which supports single or multiple quotes
> wrapped strings splitted by commas.
>- The difference between Fortran and C/C++ ABI in generating decls.
> Eg. Fortran functions in `CONTAINS` block are not public or
> regarded as in file scope. Which leads to generating counters
> before processing multiple-targets, and leads to
> `somefunc.1.default` like names.
>- The binding result in readelf's symbol table is different.
> While C/C++ generates `WEAK` resolver functions, but Fortran gives
> a `LOCAL` one while in `CONTAINS` block.
>
>If upstreams have any suggestions or comments, please let me know!
>
>ZAMBAR (2):
> Fortran: target clones basic support
> Fortran: removed RTL at function start
>
> gcc/fortran/decl.cc | 116
++++++++++++++++++++++++++++++++++++++
> gcc/fortran/f95-lang.cc | 34 +++++++++++
> gcc/fortran/gfortran.h | 5 ++
> gcc/fortran/symbol.cc | 8 +++
> gcc/fortran/trans-decl.cc | 26 ++++++++-
> 5 files changed, 187 insertions(+), 2 deletions(-)
>
>--
>2.34.1
--
ZAMBAR He <hezam...@outlook.com>
