EAGLE_25 Bug with runner_dopair1_density_vec
The output of EAGLE_25 when the code is vectorised does not agree with the un-vectorised code after 815 steps. The timesteps change, but the code continues to completion anyway.
I noticed that when I find the first pi and last pj particle I was using their own smoothing lengths to calculate the limits:
int active_id = ci->count - 1;
for (int k = ci->count - 1; k >= 0; k--) {
p = &parts_i[sort_i[k].i];
h = p->h;
d = sort_i[k].d + h * kernel_gamma + dx_max - rshift;
max_di[k] = d;
/* If the particle is out of range set the index to
* the last active particle within range. */
if (d < dj_min) {
//first_pi = active_id;
first_pi = k;
break;
} else {
if (part_is_active(p, e)) active_id = k;
}
}
instead of hi_max and hj_max:
int active_id = ci->count - 1;
for (int k = ci->count - 1; k >= 0; k--) {
p = &parts_i[sort_i[k].i];
d = sort_i[k].d + hi_max + dx_max;
max_di[k] = d;
/* If the particle is out of range set the index to
* the last active particle within range. */
if (d < dj_min) {
//first_pi = active_id;
first_pi = k;
break;
} else {
if (part_is_active(p, e)) active_id = k;
}
}