int mat_invert_uml(field_offset src, field_offset dest, field_offset temp,
double mass ){
int cgn;
double finalrsq;
register int i;
register site *s;
if( src==temp ){
printf("BOTCH\n"); exit(0);
}
/* multiply by U - even sites only */
dslash( src, F_OFFSET(ttt), EVEN);
scalar_mult_add_latvec( F_OFFSET(ttt), src,
-2.0*mass, temp, EVEN);
scalar_mult_latvec( temp, -1.0, temp, EVEN);
/* invert with M_adj M even */
cgn = ks_congrad( temp, dest, mass, niter, rsqprop,
EVEN, &finalrsq );
/* multiply by (1/2m)L, does nothing to even sites */
/* fix up odd sites , 1/2m (Dslash_oe*dest_e + phi_odd) */
dslash( dest, F_OFFSET(ttt), ODD );
FORODDSITES(i,s){
sub_su3_vector( (su3_vector *)F_PT(s,src), &(s->ttt),
(su3_vector *)F_PT(s,dest) );
scalar_mult_su3_vector( (su3_vector *)F_PT(s,dest), 1.0/(2.0*mass),
(su3_vector *)F_PT(s,dest) );
}
// -----------------------------------------------------------------
// dest <-- M^(-1) src
int mat_invert_uml(field_offset src, field_offset dest,
field_offset temp, Real mass) {
register int i;
register site *s;
int cgn;
Real norm = 1.0 / (2.0 * mass);
if (src == temp) {
printf("MAT_INVERT_UML: src = temp\n");
exit(0);
}
// "Precondition" both even and odd sites
// temp <-- M^dag src
dslash(src, F_OFFSET(ttt), EVENANDODD);
scalar_mult_add_latvec(F_OFFSET(ttt), src,
-2.0 * mass, temp, EVENANDODD);
scalar_mult_latvec(temp, -1.0, temp, EVENANDODD);
// dest_e <-- (M^dag M)^-1 temp_e (even sites only)
cgn = ks_congrad(temp, dest, mass, EVEN);
// Reconstruct odd site solution
// dest_o <-- 1/2m (Dslash_oe * dest_e + src_o)
dslash(dest, F_OFFSET(ttt), ODD);
FORODDSITES(i, s) {
sub_vector((vector *)F_PT(s, src), &(s->ttt),
(vector *)F_PT(s, dest));
scalar_mult_vector((vector *)F_PT(s, dest), norm,
(vector *)F_PT(s, dest));
}
void dslashRef() {
// FIXME: remove once reference clover is finished
if (inv_param.matpc_type == QUDA_MATPC_EVEN_EVEN_ASYMMETRIC) {
inv_param.matpc_type = QUDA_MATPC_EVEN_EVEN;
} else if (inv_param.matpc_type == QUDA_MATPC_ODD_ODD_ASYMMETRIC) {
inv_param.matpc_type = QUDA_MATPC_ODD_ODD;
}
// compare to dslash reference implementation
printf("Calculating reference implementation...");
fflush(stdout);
switch (test_type) {
case 0:
dslash(spinorRef->V(), hostGauge, spinor->V(), parity, dagger,
inv_param.cpu_prec, gauge_param.cpu_prec, inv_param.mass);
break;
case 1:
matpc(spinorRef->V(), hostGauge, spinor->V(), kappa5, inv_param.matpc_type, dagger,
inv_param.cpu_prec, gauge_param.cpu_prec, inv_param.mass);
break;
case 2:
mat(spinorRef->V(), hostGauge, spinor->V(), kappa5, dagger,
inv_param.cpu_prec, gauge_param.cpu_prec, inv_param.mass);
break;
default:
printf("Test type not defined\n");
exit(-1);
}
printf("done.\n");
}
/* Compute M^-1 * phi, answer in dest
Uses phi, ttt, resid, xxx, and cg_p as workspace */
int mat_invert_cg( field_offset src, field_offset dest, field_offset temp,
double mass ){
int cgn;
double finalrsq;
clear_latvec( dest, EVENANDODD );
cgn = ks_congrad( src, dest, mass,
niter,rsqprop,EVENANDODD,&finalrsq);
/* Multiply by Madjoint */
dslash( dest, F_OFFSET(ttt), EVENANDODD);
scalar_mult_add_latvec( F_OFFSET(ttt), dest,
-2.0*mass, F_OFFSET(ttt), EVENANDODD);
scalar_mult_latvec( F_OFFSET(ttt), -1.0, dest, EVENANDODD );
return(cgn);
}
/* Before calling checkmul() you should call grsource(EVENANDODD) and
congrad(...,EVENANDODD) */
void checkmul() {
register int i,j;
register site *s;
dslash( F_OFFSET(xxx), F_OFFSET(ttt), EVENANDODD);
scalar_mult_add_latvec( F_OFFSET(ttt), F_OFFSET(xxx), 2.0*mass,
F_OFFSET(ttt), EVENANDODD );
FORALLSITESDOMAIN(i,s){
printf("Site %d %d %d %d\n",s->x,s->y,s->z,s->t);
for(j=0;j<3;j++){
printf("%d %d\t%e\t%e\t%e\n",i,j,(double)s->g_rand.c[j].real,
(double)s->ttt.c[j].real,(double)s->g_rand.c[j].real - (double)s->ttt.c[j].real);
printf("%d %d\t%e\t%e\t%e\n",i,j,(double)s->g_rand.c[j].imag,
(double)s->ttt.c[j].imag,(double)s->g_rand.c[j].imag - (double)s->ttt.c[j].imag);
}
printf("\n");
}
// -----------------------------------------------------------------
// For single pseudofermion or inner Hasenbusch pseudofermion
// dest = M^dag g_rand
void grsource_imp(field_offset dest, Real M, int parity) {
register int i, j;
register site *s;
FORALLSITES(i, s) {
for (j = 0; j < 3; j++) {
#ifdef SITERAND
s->g_rand.c[j].real = gaussian_rand_no(&(s->site_prn));
s->g_rand.c[j].imag = gaussian_rand_no(&(s->site_prn));
#else
s->g_rand.c[j].real = gaussian_rand_no(&node_prn);
s->g_rand.c[j].imag = gaussian_rand_no(&node_prn);
#endif
}
}
// Hit g_rand with M^dag
dslash(F_OFFSET(g_rand), dest, parity);
scalar_mult_latvec(dest, -1.0, dest, parity);
scalar_mult_add_latvec(dest, F_OFFSET(g_rand), 2.0 * M, dest, parity);
}
/* "parity" is EVEN, ODD, or EVENANDODD. The parity is the parity at
which phi is computed. g_rand must always be computed at all sites. */
void grsource(int parity) {
register int i,j;
register site *s;
FORALLSITES(i,s){
#ifdef SCHROED_FUN
if(s->t > 0){
#endif
for(j=0;j<3;j++){
#ifdef SITERAND
s->g_rand.c[j].real = gaussian_rand_no(&(s->site_prn));
s->g_rand.c[j].imag = gaussian_rand_no(&(s->site_prn));
#else
s->g_rand.c[j].real = gaussian_rand_no(&node_prn);
s->g_rand.c[j].imag = gaussian_rand_no(&node_prn);
#endif
}
#ifdef SCHROED_FUN
}
else{ /* Set all fermion vectors to zero at t=0 */
for(j=0;j<3;j++){
s->phi.c[j].real = s->phi.c[j].imag = 0.0;
s->resid.c[j].real = s->resid.c[j].imag = 0.0;
s->cg_p.c[j].real = s->cg_p.c[j].imag = 0.0;
s->xxx.c[j].real = s->xxx.c[j].imag = 0.0;
s->ttt.c[j].real = s->ttt.c[j].imag = 0.0;
s->g_rand.c[j].real = s->g_rand.c[j].imag = 0.0;
}
}
#endif
}
clear_latvec( F_OFFSET(xxx), EVENANDODD );
dslash( F_OFFSET(g_rand), F_OFFSET(phi), parity);
scalar_mult_latvec( F_OFFSET(phi), -1.0, F_OFFSET(phi), parity );
scalar_mult_add_latvec( F_OFFSET(phi), F_OFFSET(g_rand), 2.0*mass,
F_OFFSET(phi), parity );
}/* grsource */
