Missing support when task class execution scope from a .. b where b < a

[abouteiller]

Original report by Qinglei Cao (Bitbucket: Qinglei_Cao, ).

---

If the execution scope of a task class is defined like this:

 281 potrf_dtrsm(m, k) [high_priority = on]
 282 
 283 // Execution space
 284 k = 0 .. descA->mt-2
 285 m = [ mi = 0 .. %{ return analysis->trsm_num[k]-1; %} ] %{ return analysis->trsm[k][mi]; %}

where analysis->trsm_num[k] could be 0 (mi = 0 .. -1), and in this case no task exist for this iteration k.

However, it seems it’s not the case, at least in the internal_init when counting the number of local tasks. For instance,

static int HiCMA_dpotrf_L_2flow_sparse_potrf_dtrsm_internal_init(parsec_execution_stream_t * es,
                                                                 __parsec_HiCMA_dpotrf_L_2flow_sparse_potrf_dtrsm_task_t
                                                                 * this_task)
{
...........

    for (k = __jdf2c_k_start; k <= __jdf2c_k_end; k += __jdf2c_k_inc) {
        assignments.k.value = k;
        {                       /* block for the local variable 'mi' */
            int __jdf2c__loop_mi = 1;
            int mi;
            for (assignments.ldef[0].value = mi = 0; __jdf2c__loop_mi == 1; assignments.ldef[0].value = mi += 1) {      /* Arbitrary iterator on mi */
                if (mi == HiCMA_dpotrf_L_2flow_sparse_potrf_dtrsm_inline_c_expr26_line_285(__parsec_tp, &assignments))
                    __jdf2c__loop_mi = 0;       /* Execute once only when reaching the end */
...........
}

BTW, this issue also happens in the startup for each taskclass.