moebiusband/MD-Bench
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add version iterating most internal loop multiple times

Browse Source 
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
This commit is contained in:
Rafael Ravedutti
2021-05-05 03:04:41 +02:00
parent faf1e2ae85
commit 15de65303e
7 changed files with 247 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
src/force.c
View File

@@ -27,12 +27,12 @@
#include <parameter.h>
#include <atom.h>
double computeForce(
Parameter *param,
Atom *atom,
Neighbor *neighbor,
int profile)
{
// Number of times to compute the most internal loop
#ifndef INTERNAL_LOOP_NTIMES
#define INTERNAL_LOOP_NTIMES 1
#endif
double computeForce(Parameter *param, Atom *atom, Neighbor *neighbor) {
int Nlocal = atom->Nlocal;
int* neighs;
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
@@ -47,10 +47,6 @@ double computeForce(
fz[i] = 0.0;
}
if(profile) {
// LIKWID_MARKER_START("force");
}
#pragma omp parallel for
for(int i = 0; i < Nlocal; i++) {
neighs = &neighbor->neighbors[i * neighbor->maxneighs];
@@ -64,20 +60,22 @@ double computeForce(
// printf("%d: %d\n", i, numneighs);
for(int k = 0; k < numneighs; k++) {
int j = neighs[k];
MD_FLOAT delx = xtmp - atom_x(j);
MD_FLOAT dely = ytmp - atom_y(j);
MD_FLOAT delz = ztmp - atom_z(j);
MD_FLOAT rsq = delx * delx + dely * dely + delz * delz;
for(int n = 0; n < INTERNAL_LOOP_NTIMES; n++) {
for(int k = 0; k < numneighs; k++) {
int j = neighs[k];
MD_FLOAT delx = xtmp - atom_x(j);
MD_FLOAT dely = ytmp - atom_y(j);
MD_FLOAT delz = ztmp - atom_z(j);
MD_FLOAT rsq = delx * delx + dely * dely + delz * delz;
if(rsq < cutforcesq) {
MD_FLOAT sr2 = 1.0 / rsq;
MD_FLOAT sr6 = sr2 * sr2 * sr2 * sigma6;
MD_FLOAT force = 48.0 * sr6 * (sr6 - 0.5) * sr2 * epsilon;
fix += delx * force;
fiy += dely * force;
fiz += delz * force;
if(rsq < cutforcesq) {
MD_FLOAT sr2 = 1.0 / rsq;
MD_FLOAT sr6 = sr2 * sr2 * sr2 * sigma6;
MD_FLOAT force = 48.0 * sr6 * (sr6 - 0.5) * sr2 * epsilon;
fix += delx * force;
fiy += dely * force;
fiz += delz * force;
}
}
}
@@ -86,9 +84,5 @@ double computeForce(
fz[i] += fiz;
}
if(profile) {
// LIKWID_MARKER_STOP("force");
}
return 0.0;
}

10
src/main-stub.c
View File

@@ -16,7 +16,7 @@
#define LATTICE_DISTANCE 10.0
#define NEIGH_DISTANCE 1.0
extern double computeForce( Parameter*, Atom*, Neighbor*, int);
extern double computeForce( Parameter*, Atom*, Neighbor*);
void init(Parameter *param) {
param->epsilon = 1.0;
@@ -188,19 +188,19 @@ int main(int argc, const char *argv[]) {
DEBUG("Building neighbor lists...\n");
buildNeighbor(atom, &neighbor);
DEBUG("Computing forces...\n");
computeForce(&param, atom, &neighbor, 0);
computeForce(&param, atom, &neighbor);
double S, E;
S = getTimeStamp();
LIKWID_MARKER_START("force");
for(int i = 0; i < param.ntimes; i++) {
computeForce(&param, atom, &neighbor, 1);
computeForce(&param, atom, &neighbor);
}
LIKWID_MARKER_STOP("force");
E = getTimeStamp();
double T_accum = E-S;
const double atoms_updates_per_sec = atom->Nlocal * param.ntimes / T_accum;
const double cycles_per_atom = T_accum * freq / (atom->Nlocal * param.ntimes);
const double atoms_updates_per_sec = (double)(atom->Nlocal * INTERNAL_LOOP_NTIMES * param.ntimes) / T_accum;
const double cycles_per_atom = T_accum * freq / (double)(atom->Nlocal * param.ntimes * INTERNAL_LOOP_NTIMES);
const double cycles_per_neigh = cycles_per_atom / (double)(atoms_per_unit_cell - 1);
if(!csv) {

6
src/main.c
View File

@@ -47,7 +47,7 @@ typedef enum {
NUMTIMER
} timertype;
extern double computeForce( Parameter*, Atom*, Neighbor*, int);
extern double computeForce( Parameter*, Atom*, Neighbor*);
void init(Parameter *param)
{
@@ -205,7 +205,7 @@ int main (int argc, char** argv)
setup(&param, &atom, &neighbor);
computeThermo(0, &param, &atom);
computeForce(&param, &atom, &neighbor, 1);
computeForce(&param, &atom, &neighbor);
timer[FORCE] = 0.0;
timer[NEIGH] = 0.0;
@@ -221,7 +221,7 @@ int main (int argc, char** argv)
timer[NEIGH] += reneighbour(&param, &atom, &neighbor);
}
timer[FORCE] += computeForce(&param, &atom, &neighbor, 1);
timer[FORCE] += computeForce(&param, &atom, &neighbor);
finalIntegrate(&param, &atom);
if(!((n + 1) % param.nstat) && (n+1) < param.ntimes) {