Passing references to array elements to GSL?

  • Charlie Zender
    Charlie Zender

    Hi Henry,

    A user here at UCI is starting to delve into ncap2/GSL, and who has
    run into questions that I can't answer.

    The first is can ncap2 use pass array elements by reference?
    For example, in passing GSL parameters whose values are to be returned
    she would like to do this:

          gsl_fit_linear(time,1,snc(area_idx,exp_idx,:),1,$time.size, &c0(area_idx,exp_idx), &tc1, &cov00, &cov01,&cov11,&sumsq);

    The fourth parameter, &c0(area_idx,exp_idx), is intended to be a
    reference to an array element. But this does not seem to work.
    Instead she has to pass a reference to a scalar, &tc0:

          gsl_fit_linear(time,1,snc(area_idx,exp_idx,:),1,$time.size, &tc0,&tc1,&cov00,&cov01,&cov11,&sumsq);

    and then manually copy the returned value into the array element that
    she wished to pass. Is this because the GSL routine wants a full
    variable? Or…? What is the best strategy for allowing this?
    e.g., can we overload these GSL prototypes to accept different types
    of references, i.e., references to simple "data slots" (float *) as
    well as refereces to full variables (var *)? Thoughts?


  • Charlie Zender
    Charlie Zender

    Also, with regards to

          gsl_fit_linear(time,1,snc(area_idx,exp_idx,:),1,$time.size, &tc0,&tc1,&cov00,&cov01,&cov11,&sumsq);

    she noted that the variables cov00 and cov01 did not contain the
    covariances on return. they contained something, just not the
    covariances. is it possible that the NCO prototype was somehow
    mangles? she was able to manually compute the covariances with


    she also noted that the gsl_stats_covariance() and correlation
    functions have slightly different prototypes than the GSL functions
    themselves. Apparently the NCO versions require extra parameters to
    specify the size of the input arrays, whereas the GSL routines figure
    that out automatically aor make assumptions.