Interface single precision routines

[dpo]

We should be able to do something like

for (jtype, ctype) in ((:Float32, Cfloat), (:Float64, Cdouble))
  @eval begin
    function NLPModels.objgrad!(nlp :: CUTEstModel, x :: Vector{$jtype}, g :: Vector{$jtype})
      nvar = nlp.meta.nvar
      ncon = nlp.meta.ncon
      f = $ctype[0]
      io_err = Cint[0]
      get_grad = 1
      if ncon > 0
        ccall(dlsym(cutest_lib, "cutest_cofg_"), Nothing,
            (Ptr{Int32}, Ref{Int32}, Ptr{$jtype}, Ptr{$jtype}, Ptr{$jtype}, Ref{Int32}),
                 io_err,       nvar,          x,            f,           g,   get_grad);
      else
        ccall(dlsym(cutest_lib, "cutest_uofg_"), Nothing,
            (Ptr{Int32}, Ref{Int32}, Ptr{$jtype}, Ptr{$jtype}, Ptr{$jtype}, Ref{Int32}),
                 io_err,       nvar,           x,           f,           g,   get_grad);
      end
      nlp.counters.neval_obj += 1
      nlp.counters.neval_grad += 1
      @cutest_error

      return f[1], g
    end
  end
end

See #177

[amontoison]

Single and quadruple precision routines are now interfaced.