MPFR binding for Fortran

[Thomas Koenig]

Hi,

I've been working on a little pet project of mine, an MPFR binding
for Fortran. I have now released a very first version github,
at https://github.com/tkoenig1/FMPFR .

So, if anybody wants to try out some arbitrary-precision floating
point calculations in Fortran, give it a spin.

Suggestions for improvements are also welcome.

A little code sample:

program memain
use fmpfr_oper
implicit none
type (fmpfr), dimension(:), allocatable :: a
call set_default_prec (128)
allocate (a(2))
a(1) = fmpfr("1.3")
a(2) = a(1) + 2
print *,a
print *,sin(a)
end program memain

[feanor]

Thomas Koenig schrieb am Montag, 18. Juli 2022 um 23:21:11 UTC+2:
> Hi,
>
> I've been working on a little pet project of mine, an MPFR binding
> for Fortran. I have now released a very first version github,
> at https://github.com/tkoenig1/FMPFR .
>
> So, if anybody wants to try out some arbitrary-precision floating
> point calculations in Fortran, give it a spin.

Nice. Thanks for the effort. I tried it out immediately. I removed the symlinks for the autotools files and replaced them with copies
from my own project. Then I compiled/link your example program as
gfortran foo.f90 -L/home/reuter/local/lib -lfmpfr -I../src -lmpfr
But both on Ubuntu 18.04 and on Apple Darwin x86_64 I got
NaN NaN
NaN NaN
The dependencies look ok to me;
$ ldd a.out
linux-vdso.so.1 (0x00007fff175d5000)
libmpfr.so.6 => /usr/lib/x86_64-linux-gnu/libmpfr.so.6 (0x00007f522538e000)
libgfortran.so.4 => /usr/lib/x86_64-linux-gnu/libgfortran.so.4 (0x00007f5224faf000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f5224bbe000)
libgmp.so.10 => /usr/lib/x86_64-linux-gnu/libgmp.so.10 (0x00007f522493d000)
/lib64/ld-linux-x86-64.so.2 (0x00007f5225819000)
libquadmath.so.0 => /usr/lib/x86_64-linux-gnu/libquadmath.so.0 (0x00007f52246fd000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f522435f000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007f5224147000)

Any idea what goes wrong here?
Cheers,
JRR

[Thomas Koenig]

feanor <juergen...@desy.de> schrieb:

> Thomas Koenig schrieb am Montag, 18. Juli 2022 um 23:21:11 UTC+2:
>> Hi,
>>
>> I've been working on a little pet project of mine, an MPFR binding
>> for Fortran. I have now released a very first version github,
>> at https://github.com/tkoenig1/FMPFR .
>>
>> So, if anybody wants to try out some arbitrary-precision floating
>> point calculations in Fortran, give it a spin.
>
> Nice. Thanks for the effort. I tried it out immediately. I removed the symlinks for the autotools files and replaced them with copies
> from my own project.

Ah, something I will have to do as well.

> Then I compiled/link your example program as
> gfortran foo.f90 -L/home/reuter/local/lib -lfmpfr -I../src -lmpfr
> But both on Ubuntu 18.04 and on Apple Darwin x86_64 I got
> NaN NaN
> NaN NaN

Definitely seems wrong - maybe a too-early version of gfortran?
That would be a bummer.

What does "gfortran -v" give you on your two systems?

[Thomas Koenig]

Thomas Koenig <tko...@netcologne.de> schrieb:

Never mind that - I have found the bug, and I will fix it.

[Thomas Koenig]

Thomas Koenig <tko...@netcologne.de> schrieb:

OK, fixed now. I fell into a trap that is easy to fall into
when doing C-style rc = "some pure function", this
was optimized away (and I hadn't tested it with -O2 before,
it worked with -O0).

[feanor]

Thomas Koenig schrieb am Dienstag, 19. Juli 2022 um 22:26:42 UTC+2:

> > Never mind that - I have found the bug, and I will fix it.
> OK, fixed now. I fell into a trap that is easy to fall into
> when doing C-style rc = "some pure function", this
> was optimized away (and I hadn't tested it with -O2 before,
> it worked with -O0).

Now I cannot link anymore, during compilation I get the warning:
mpfr_glue.c:56:3: warning: implicit declaration of function ‘mpfr_set_float128’; did you mean ‘fmpfr_set_float128’? [-Wimplicit-function-declaration]
56 | mpfr_set_float128 (rop, op, rnd);
| ^~~~~~~~~~~~~~~~~
| fmpfr_set_float128
and at linking the test program I get:
Undefined symbols for architecture x86_64:
"_mpfr_set_float128", referenced from:
_fmpfr_set_float128 in libfmpfr.a(fmpfr_glue.o)
ld: symbol(s) not found for architecture x86_64

mpfr_set_float128 appears in mpfr.h as
__MPFR_DECLSPEC mpfr_set_float128
this seems for Windows DLL !?
On Ubuntu 18.04 compilation and linking works now (with gcc/gfortran 7.5.0), on Darwin x86_64 (Intel, not M1 arm64) it doesn't

[Thomas Koenig]

feanor <juergen...@desy.de> schrieb:

Fixed, thanks for the reports!

Hope you can do a bit more with it now :-)

[JRR]

Am 20.07.22 um 19:36 schrieb Thomas Koenig:

Thanks, Thomas. I still have an issue on Darwin x86_64:

Undefined symbols for architecture x86_64:

"_fmpfr_set_ld", referenced from:
___fmpfr_oper_MOD_ass_set_ld in libfmpfr.a(fmpfr_oper.o)
___fmpfr_oper_MOD_fun_set_ld in libfmpfr.a(fmpfr_oper.o)

ld: symbol(s) not found for architecture x86_64

I think the C compiler also needs the preprocessor flags that you
only hand over to the Fortran code, I hackfixed this by

AM_CFLAGS = = @fc_preprocessor_flag@ @fc_preflags@

in the src/Makefile.am
but there is probably a more elegant way. Then it also works on Darwin
x86_64. I haven't tried the arm64 yet.


--
Juergen Reuter
Theoretical Particle Physics
Deutsches Elektronen-Synchrotron (DESY)
Hamburg, Germany

[Thomas Koenig]

JRR <juergen...@invalid.com> schrieb:

Again, thanks for the testing!

I have just added the -D flags to configure.ac and (which is also
important) added a #define MPFR_WANT_FLOAT128 if float128 is supported.

I hope it works now.

Best regards

Thomas