https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55916
--- Comment #13 from Jouko Orava <jouko.orava at iki dot fi> ---
(In reply to Joost VandeVondele from comment #12)
> I'm wondering if you had a working patch at some point.
Not a comprehensive one, no. I recall modifying libgfortran/runtime/memory.c
was not sufficient.
I ended up going pretty deep into how GCC generates the code in an effort to
find out how to replace *all* calls to malloc(), free(), calloc(), and
realloc() with fortran_malloc(), fortran_free(), fortran_calloc(), and
fortran_realloc(), respectively, that GCC generates for Fortran code, with
libgfortran initially providing wrappers around the respective C library calls.
That way it would be possible to automagically choose whether the C library
standard allocator was sufficient at the GCC compile time (based on maximum
alignment versus largest non-vector type supported by the architecture), or if
a higher-alignment allocator (memalign(), posix_memalign()) would be used
instead. Or, better yet, even replace the C library allocator with a custom
one, one that better caters to Fortran use cases.
At some point I realized there was about zero percent chance such invasive
changes would be accepted upstream, so I gave up.
The workaround I showed in #11,
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55916#c11
suffices for me, after all; it basically does the replacement by interposing
those functions on the C library side instead. It can be used either at compile
time (by linking the C code into the final binary), or at run time for
already-compiled binaries (by using LD_PRELOAD).
