Fix the NUMA detection for the smcuda BTL.

[bosilca]

Code cleanup initiated by a discussion on Slack. If we fail to identify the process NUMA node, or if the process is associated with more than one, print a warning to inform the user about the potential performance implication.

[edgargabriel]

@bosilca, just curious, is this a bug fix or code cleanup?

[bosilca]

It all started from a discussion on Slack about a warning in the smcuda BTL pointed out by @wenduwan.

[wenduwan]

Looks good to me. Just 1 minor comment.

[wenduwan]

I tried this patch but still see the warning

btl_smcuda.c: In function ‘smcuda_btl_first_time_init’:
btl_smcuda.c:280:27: warning: ‘i’ may be used uninitialized in this function [-Wmaybe-uninitialized]
  280 |             for (w = 0; w < i; w++) {
      |                         ~~^~~
  CCLD     mca_btl_smcuda.la

[bosilca]

Fixed, please try again.

[wenduwan]

@bosilca Thank you. The warning is fixed.

Now it revealed something else. int i is not useful at all. On L427 I see

 427     i = opal_free_list_init(&mca_btl_smcuda_component.sm_frags_eager, length, opal_cache_line_size,
 428                             OBJ_CLASS(mca_btl_smcuda_frag1_t), length_payload, opal_cache_line_size,
 429                             mca_btl_smcuda_component.sm_free_list_num,
 430                             mca_btl_smcuda_component.sm_free_list_max,
 431                             mca_btl_smcuda_component.sm_free_list_inc,
 432                             mca_btl_smcuda_component.sm_mpool, 0, NULL, NULL, NULL);
 433     if (OPAL_SUCCESS != i)
 434         return i;

i is used instead of rc.

[wenduwan]

Thanks @bosilca . I will run some additional tests in the background(I don't expect any issue).

Feel free to merge otherwise.