void SmearJacobireference(sFloat *res, gFloat **gaugeFull, sFloat *spinorField, double r, int steps) {
sFloat *tmpSpinor = (sFloat *)malloc(V*spinorSiteSize*sizeof(sFloat));
sFloat *spinorIn, *spinorOut, *tmp;
//Copy the contents of original spinor into tmpSpinor
xeqy(tmpSpinor, spinorField, V*spinorSiteSize);
spinorIn = tmpSpinor;
spinorOut = res;
gFloat *gaugeEven[4], *gaugeOdd[4];
for (int dir = 0; dir < 4; dir++) {
gaugeEven[dir] = gaugeFull[dir];
gaugeOdd[dir] = gaugeFull[dir]+Vh*gaugeSiteSize;
}
for(int iter = 0; iter < steps; iter++) {
xeqay(spinorOut, 1/(1+6*r), spinorIn, V*spinorSiteSize);
for (int oddBit = 0; oddBit < 2; oddBit++) {
for (int i = 0; i < Vh; i++) {
int fullindex = oddBit*Vh + i;
//Spatial smearing only
for (int dir = 0; dir < 6; dir++) {
gFloat *gauge = gaugeLink(i, dir, oddBit, gaugeEven, gaugeOdd, 1);
sFloat *spinor = spinorNeighborFullLattice(fullindex, dir, spinorIn, 1);
sFloat gaugedSpinor[4*3*2];
for (int s = 0; s < 4; s++) {
if (dir % 2 == 0) su3Mul(&gaugedSpinor[s*(3*2)], gauge, &spinor[s*(3*2)]);
else su3Tmul(&gaugedSpinor[s*(3*2)], gauge, &spinor[s*(3*2)]);
}
//Accumulate result from gaugedSpinor into spinorOut
xpeqay(&spinorOut[fullindex*(4*3*2)], r/(1+6*r), gaugedSpinor, 4*3*2);
}
}
}
//Swap the pointers
tmp = spinorIn;
spinorIn = spinorOut;
spinorOut = tmp;
}
//Copy the contents of spinorIn into res, unless res already points to
//spinorIn
if(spinorIn != res) {
xeqy(res, spinorIn, V*spinorSiteSize);
}
free(tmpSpinor);
}
/*
* Solves lapl(u) x = b, for x, given b, using Conjugate Gradient
*/
void
cg(latparams lp, field **x, field **b, link **g)
{
size_t L = lp.L;
int max_iter = 100;
float tol = 1e-9;
/* Temporary fields needed for CG */
field **r = new_field(lp);
field **p = new_field(lp);
field **Ap = new_field(lp);
/* Initial residual and p-vector */
lapl(lp, r, x, g);
xmy(lp, b, r);
xeqy(lp, p, r);
/* Initial r-norm and b-norm */
float rr = xdotx(lp, r);
float bb = xdotx(lp, b);
double t_lapl = 0;
int iter = 0;
for(iter=0; iter<max_iter; iter++) {
printf(" %6d, res = %+e\n", iter, rr/bb);
if(sqrt(rr/bb) < tol)
break;
double t = stop_watch(0);
lapl(lp, Ap, p, g);
t_lapl += stop_watch(t);
float pAp = xdoty(lp, p, Ap);
float alpha = rr/pAp;
axpy(lp, alpha, p, x);
axpy(lp, -alpha, Ap, r);
float r1r1 = xdotx(lp, r);
float beta = r1r1/rr;
xpay(lp, r, beta, p);
rr = r1r1;
}
/* Recompute residual after convergence */
lapl(lp, r, x, g);
xmy(lp, b, r);
rr = xdotx(lp, r);
double beta_fp = 50*((double)L*L*L)/(t_lapl/(double)iter)*1e-9;
double beta_io = 40*((double)L*L*L)/(t_lapl/(double)iter)*1e-9;
printf(" Converged after %6d iterations, res = %+e\n", iter, rr/bb);
printf(" Time in lapl(): %+6.3e sec/call, %4.2e Gflop/s, %4.2e GB/s\n",
t_lapl/(double)iter, beta_fp, beta_io);
del_field(r);
del_field(p);
del_field(Ap);
return;
}
/*
* Solves lapl(u) x = b, for x, given b, using Conjugate Gradient
*/
void
cg(size_t L, _Complex float *x, _Complex float *b, _Complex float *u)
{
int max_iter = 100;
float tol = 1e-6;
/* Temporary fields needed for CG */
_Complex float *r = new_field(L);
_Complex float *p = new_field(L);
_Complex float *Ap = new_field(L);
/* Initial residual and p-vector */
lapl(L, r, x, u);
xmy(L, b, r);
xeqy(L, p, r);
/* Initial r-norm and b-norm */
float rr = xdotx(L, r);
float bb = xdotx(L, b);
double t_lapl = 0;
int iter = 0;
for(iter=0; iter<max_iter; iter++) {
printf(" %6d, res = %+e\n", iter, rr/bb);
if(sqrt(rr/bb) < tol)
break;
double t = stop_watch(0);
lapl(L, Ap, p, u);
t_lapl += stop_watch(t);
float pAp = xdoty(L, p, Ap);
float alpha = rr/pAp;
axpy(L, alpha, p, x);
axpy(L, -alpha, Ap, r);
float r1r1 = xdotx(L, r);
float beta = r1r1/rr;
xpay(L, r, beta, p);
rr = r1r1;
}
/* Recompute residual after convergence */
lapl(L, r, x, u);
xmy(L, b, r);
rr = xdotx(L, r);
double beta_fp = 34*L*L/(t_lapl/(double)iter)*1e-9;
double beta_io = 32*L*L/(t_lapl/(double)iter)*1e-9;
printf(" Converged after %6d iterations, res = %+e\n", iter, rr/bb);
printf(" Time in lapl(): %+6.3e sec/call, %4.2e Gflop/s, %4.2e GB/s\n",
t_lapl/(double)iter, beta_fp, beta_io);
free(r);
free(p);
free(Ap);
return;
}
