Fixes issue with rvs_normal where it lacks the elemental property

[14NGiestas]

As noted in PR#662 there's an error in the demo program given for rvs_normal where it lacks the shown elemental property.

example/demo_normal_rvs.f90:29:12:

   29 |     print *, norm(a, b)         ! a rank 3 random variates array
      |            1
Error: There is no specific function for the generic ‘rvs_normal’ at (1)
<ERROR> Compilation failed for object " example_demo_normal_rvs.f90.o "
<ERROR>stopping due to failed compilation

[jvdp1]

After reading the code again, I wonder if this function could be elemental. For example, if an array v(2) is passed as argument, will the function provided the same result if result(2) is computed before result(1), instead of computing result(1) before result(2)? I don't think that the standard implies any order of the computations.

[14NGiestas]

I think it's just a matter of batch calling the function over an array without having to deal with do loops, if you take a look at L185 you will see the pattern, 2 arguments -> impure elemental.

But I can see there is a problem with reproducible "randomness": if the order of the computations is changed (by doing it in parallel it's known that it happens), and since it depends on some stateful routines (there are some variables with the save attribute) I guess the result will be different, but I don't think the distribution will be changed.

[jvdp1]

But I can see there is a problem with reproducible "randomness": if the order of the computations is changed (by doing it in parallel it's known that it happens), and since it depends on some stateful routines (there are some variables with the save attribute) I guess the result will be different, but I don't think the distribution will be changed.

Indeed, I refered to the problem of reproducible randomness. I agree with you that it won't affect the generated distribution.
Furthermore, the equivalent function in stdlib_stats_distribution_uniform is already impure elemental. Therefore, it would be logical that the equivalent fiunctions are impure elemental in both stdlib_stats_normal and in stdlib_stats_distribution_exponential.

The specs should be adapted to reflect these changes too.

[jvdp1]

LGTM. Thank you.

[jvdp1]

Thank you. I will merge it to move further wiuth other PRs.