install header about galario

[fredRos]

It's a bit more complicated. If we allow full flexibility for the user, we have to declare both the single and double precision versions of all function. Either replication or macros like in fftw3.h. This requires more work, either renaming functions as in FFTW shift -> shiftf etc or allow only function that have a data type so we can overload. This is currently not possible for example for C_acc_fft(int nx, void* data). It is not possible to define data's type from that declaration.

So I'd prefer to force users to decide at compile time between single and double precision. If somebody needs more later on, we can discuss. For now, that's too much work.

[fredRos]

Here is a minimal program compile-test.cxx

#include "galario.hpp"
#include <complex>
#include <iostream>
#include <vector>

int main()
{
    constexpr int nx = 5;
    std::vector<dcomplex> data(nx*nx);

    std::cout << "Size of data type: " << sizeof(dcomplex) << std::endl;
    C_acc_shift(nx, &data[0]);
    return 0;
}

and how to run it with double precision

export GALARIODIR=$SWMOD_INST_PREFIX
$ g++ -std=c++14 -fopenmp  -I$GALARIODIR/include/ compile-test.cxx -L$GALARIODIR/lib/ -DDOUBLE_PRECISION -lgalario -o compile-test && ./compile-test
Size of data type: 16

and single precision

$ g++ -std=c++14 -fopenmp  -I$GALARIODIR/include/ compile-test.cxx -L$GALARIODIR/lib/ -lgalario_single -o compile-test && ./compile-test
Size of data type: 8

It is clunky and error prone that you one has to define a preprocessor symbol for double with -DDOUBLE_PRECISION and the matching library name but for single, only the library name is needed.

I think for external users there is a potential for a name conflict for DOUBLE_PRECISION. We'd better name it GALARIO_DOUBLE. In addition, we have to make up our mind if single or double should be the default. Sticking with single would mean less work for the moment