void bfmcommIroIro<Float>::comm (int result_cb, Fermion_t psi,int dag)
{
comm_start(result_cb,psi,dag);
comm_complete(result_cb,psi);
}
void wfm::dslash(Float *chi,
Float *u,
Float *psi,
int cb,
int dag)
{
/*
* To a first approximation, we simply
* remap the arguments into a form acceptable
* to the assembler, then call it
*/
/*
*Pull in the first Psi to cache early
*/
cache_touch(psi);
cache_touch(psi+4);
cache_touch(psi+8);
cache_touch(psi+12);
cache_touch(psi+16);
cache_touch(psi+20);
decom(psi,u,cb,dag);
#ifdef DEBUG_BENCHMARK_COMMS
double ndata = 2*2*allbound * 12 * TwoSpinSize() * 1.0E-6 * 100;
struct timeval start,stop,delta;
gettimeofday(&start,NULL);
for(int i=0;i<100;i++) {
#endif
comm_start(cb);
/*
* Hackers: you could split here and do something else...
* Such as DWF fith dimension, or a clover term etc...
* Might as well pull in a few sites worth of pointer table
* while we're waiting for the comms
*/
comm_complete(cb);
#ifdef DEBUG_BENCHMARK_COMMS
}
gettimeofday(&stop,NULL);
timersub(&stop,&start,&delta);
double seconds = delta.tv_usec * 1.0E-6 + delta.tv_sec;
if ( isBoss() ) printf("Comms %le MB in %le seconds = %le MB/s\n",ndata,seconds,ndata/seconds);
ndata = 2*2*allbound * 12 * TwoSpinSize() ;
if ( isBoss() ) printf("ndata = %d \n",ndata);
#endif
#ifdef DEBUG_OUTPUT_VECTORS
static int file_cnt;
{
char buf[256];
sprintf(buf,"2spin.%d.%d",UniqueID(),file_cnt);
FILE *fp = fopen(buf,"w");
for(int i=0;i<vol;i++) {
for(int pm=0;pm<2;pm++){
for(int mu=0;mu<4;mu++){
int offset = interleave_site(pm,mu,i);
for(int s=0;s<2;s++){
for(int c=0;c<3;c++){
for(int r=0;r<2;r++){
int scri;
if ( WFM_BGL ) scri = r + s*6+c*2;
else scri = r + s*2+c*4;
int gbl[4];
site_to_global(cb, i, gbl, local_latt );
if ( SloppyPrecision ) {
float * pointer = (float *) two_spinor;
fprintf(fp,"%d %d %d %d %d %d %d %d %d %e\n",gbl[0],gbl[1],gbl[2],gbl[3],
pm,mu,s,c,r,pointer[PAD_HALF_SPINOR_SIZE*offset+scri]);
} else {
Float * pointer = (Float *) two_spinor;
fprintf(fp,"%d %d %d %d %d %d %d %d %d %e\n",gbl[0],gbl[1],gbl[2],gbl[3],
pm,mu,s,c,r,pointer[PAD_HALF_SPINOR_SIZE*offset+scri]);
}
}}}
}
}
}
fclose(fp);}
#endif
cache_touch(two_spinor);
cache_touch(two_spinor+4);
cache_touch(two_spinor+8);
recon(chi,u,cb,dag);
#ifdef DEBUG_OUTPUT_VECTORS
{
char buf[256];
sprintf(buf,"recon.%d.%d",UniqueID(),file_cnt++);
FILE *fp = fopen(buf,"w");
for(int i=0;i<vol;i++) {
for(int pm=0;pm<2;pm++){
for(int mu=0;mu<4;mu++){
int offset = interleave_site(pm,mu,i);
for(int s=0;s<4;s++){
for(int c=0;c<3;c++){
for(int r=0;r<2;r++){
int scri;
scri = r + s*6+c*2;
Float * pointer = (Float *) chi;
int gbl[4];
site_to_global(cb, i, gbl, local_latt );
fprintf(fp,"%d %d %d %d %d %d %d %d %d %d %e\n",gbl[0],gbl[1],gbl[2],gbl[3],
i,pm,mu,s,c,r,pointer[SPINOR_SIZE*offset+scri]);
}}}
}
}
}
fclose(fp);}
exit(0);
#endif
return;
}
