double Atomtype::CalcPE(int frame_i, const Trajectory &trj, const coordinates &rand_xyz, const cubicbox_m256 &box, double vol) const
{
float pe = 0.0;
int atom_i = 0;
/* BEGIN SIMD SECTION */
// This performs the exact same calculation after the SIMD section
// but doing it on 8 atoms at a time using SIMD instructions.
coordinates8 rand_xyz8(rand_xyz), atom_xyz;
__m256 r2_8, mask, r6, ri6, pe_tmp;
__m256 pe_sum = _mm256_setzero_ps();
float result[n] __attribute__((aligned (16)));
for (; atom_i < this->n-8; atom_i+=8)
{
atom_xyz = trj.GetXYZ8(frame_i, this->name, atom_i);
r2_8 = distance2(atom_xyz, rand_xyz8, box);
mask = _mm256_cmp_ps(r2_8, rcut2_8, _CMP_LT_OS);
r6 = _mm256_and_ps(mask, _mm256_mul_ps(_mm256_mul_ps(r2_8, r2_8), r2_8));
ri6 = _mm256_and_ps(mask, _mm256_rcp_ps(r6));
pe_tmp = _mm256_and_ps(mask, _mm256_mul_ps(ri6, _mm256_sub_ps(_mm256_mul_ps(c12_8, ri6), c6_8)));
pe_sum = _mm256_add_ps(pe_tmp, pe_sum);
}
_mm256_store_ps(result, pe_sum);
for (int i = 0; i < 8; i++)
{
pe += result[i];
}
/* END SIMD SECTION */
for (; atom_i < this->n; atom_i++)
{
coordinates atom_xyz = trj.GetXYZ(frame_i, this->name, atom_i);
float r2 = distance2(atom_xyz, rand_xyz, cubicbox(box));
if (r2 < this->rcut2)
{
float ri6 = 1.0/(pow(r2,3));
pe += ri6*(this->c12*ri6 - this->c6);
}
}
pe += this->n/vol * this->tail_factor;;
return pe;
}
void Clusters::do_clustering(int frame, Trajectory &traj, string group, double rcut2)
{
coordinates atom_j_vec;
double r2;
int atom_counter_i;
int atom_counter_j;
int atom_i;
int atom_j;
int mol_i;
int mol_j;
int c_i;
int c_j;
triclinicbox box;
vector <coordinates> atom_i_vec(atoms_per_mol);
box = traj.GetBox(frame);
initialize();
for (mol_i = 0; mol_i < this->mol_n; mol_i++)
{
c_i = this->index.at(mol_i);
/* Go ahead and save the atoms from this molecule to a vector so that we
* don't have to keep accessing it later below. */
atom_counter_i = mol_i * atoms_per_mol;
for (atom_i = 0; atom_i < this->atoms_per_mol; atom_i++)
{
atom_i_vec.at(atom_i) = traj.GetXYZ(frame, group, atom_counter_i);
atom_counter_i++;
}
for (mol_j = 0; mol_j < this->mol_n; mol_j++)
{
c_j = this->index.at(mol_j);
if (c_i != c_j)
{
/* This compares every atom on the two molecules. If one pair is
* within the cutoff range then the two molecules should be
* added to the same cluster and no more checking is necessary
* for these two molecules. */
for (atom_i = 0; atom_i < this->atoms_per_mol; atom_i++)
{
atom_counter_j = mol_j * atoms_per_mol;
for (atom_j = 0; atom_j < this->atoms_per_mol; atom_j++)
{
atom_j_vec = traj.GetXYZ(frame, group, atom_counter_j);
atom_counter_j++;
r2 = distance2(atom_i_vec.at(atom_i), atom_j_vec, box);
if (r2 < rcut2)
{
add(c_i, c_j);
goto nextmol;
}
}
}
}
nextmol:
continue;
}
}
return;
}