On a smaller example I can see that it is indeed is because floats are not
double floats.
It's seem to be the statement
implicit double precision (a-h,o-z)
in Fortran code that throws things upside down. I'm not so familiar with
the Fortran standard and especially with gfortran extensions, but I would
think if anything, this statement would force the arguments to be double
float (real*8) precision. Also, I put "-fdefault-real-8" option to the
compiler...
But thanks for the tip, perhaps I would just comment this statement out in
Fortran.
On Wednesday, November 9, 2016 at 3:49:55 PM UTC+2, Ángel de Vicente wrote:
>
> Hi,
>
> Alexey Cherkaev <alexey....@gmail.com <javascript:>> writes:
> > To cut long story short, I have a problem of passing floating-point
> numbers from
> > Julia to Fortran subroutine. Documentation suggests, that each parameter
> to
> > Fortran's program needs "Ref{T}" wrapper as they are passed by reference
> (I've
> > seen that older code also uses Ptr{T} with "&" at variable names). It
> works fine
> > for integers (I even can see the difference if "FInt" alias is changed
> between
> > "Int32" and "Int64" and if "-fdefault-integer-8" in compiling option is
> present
> > or omitted). But I don't see a single floating point value passed
> correctly
> > (including floating-point arrays). I've also tried, as an extreme
> measure, to
> > pass a floating point parameter as a vector (of size 1) and have a
> corresponding
> > type in ccall to be "Ptr{Float64}"
>
> Perhaps in Fortran you have 32bits floats? I just tried with this, and I
> have no problem passing floats to Fortran:
>
> ,----
> | testdp(ar) = ccall((:__tests_MOD_testdp, "./testmodgfort"),Float64,
> (Ptr{Float64},), &ar)
> | println("Result of function call to Fortran ", testdp(5.0))
> `----
>
> ,----
> | DOUBLE PRECISION FUNCTION testdp (a)
> | DOUBLE PRECISION :: a
> | testdp = 5.0*a
> | END FUNCTION testdp
> `----
>
> ,----
> | julia> include("julia-fortran.jl")
> | Result of function call to Fortran 25.0
> `----
>
> --
> Ángel de Vicente
> http://www.iac.es/galeria/angelv/
>
