void PT::vec_cb_norm(int n, IFloat **vout, IFloat **vin, const int *dir,int parity, IFloat * gauge)
{
//List of the different directions
int wire[n];
int i;
// int j,d,s,k;
//SCUDirArgs for sending and receiving in the n directions
SCUDirArgIR *SCUarg_p[2*non_local_dirs];
//SCUDirArgIR *SCUarg_p[2*n];
SCUDirArgMulti SCUmulti;
static int call_num = 0;
int vlen = VECT_LEN;
int vlen2 = VECT_LEN;
// printf("gauge=%p\n",gauge);
call_num++;
//Name our function
// char *fname="pt_1vec_cb_norm()";
//Set the transfer directions
//If wire[i] is even, then we have communication in the negative direction
//If wire[i] is odd, then we have communication in the positive direction
for(i=0;i<n;i++)
wire[i]=dir[i];
Float dtime;
//If wire[i] is odd, then we have parallel transport in the
//positive direction. In this case, the matrix multiplication is
//done before the field is transferred over to the adjacent node
//
//If we have transfer in the negative T direction (wire[i] = 6) then
//we have to copy the appropriate fields into the send buffer
if(conjugated)
for(i=0;i<n;i++)
{
if(!local[wire[i]/2])
{
if(wire[i]%2)
{
//printf("dir = %d, pre-mulitply\n", wire[i]);
#ifdef PROFILE
dtime = - dclock();
#endif
partrans_cmv(non_local_chi_cb[wire[i]]/2,uc_nl_cb_pre[parity][wire[i]/2],gauge,vin[i],snd_buf_cb[wire[i]/2]);
#ifdef PROFILE
dtime +=dclock();
print_flops("",fname,66*non_local_chi_cb[wire[i]],dtime);
#endif
}
else if((wire[i] == 6))
{
#ifdef PROFILE
dtime = - dclock();
#endif
#if 1
pt_copy_buffer(non_local_chi_cb[6],(long)vin[i],(long)snd_buf_t_cb,(long)Toffset[parity]);
#else
for(j = 0; j < non_local_chi_cb[6];j++)
for(k = 0; k < VECT_LEN;k++)
*(snd_buf_t_cb+j*VECT_LEN+k) = *(vin[i] + *(Toffset[parity]+j)+ k);
//moveMem(snd_buf_t_cb + j*VECT_LEN,vin[i] + *(Toffset[parity]+j)*vlen,VECT_LEN*sizeof(IFloat));
#endif
#ifdef PROFILE
dtime +=dclock();
print_flops(fname,"pt_copy_buffer()",0,dtime);
#endif
}
}
}
else
for(i=0;i<n;i++)
{
if(!local[wire[i]/2])
{
if(wire[i]%2)
{
#ifdef PROFILE
dtime = - dclock();
#endif
partrans_cmv_dag(non_local_chi_cb[wire[i]]/2,uc_nl_cb_pre[parity][wire[i]/2],gauge,vin[i],snd_buf_cb[wire[i]/2]);
#ifdef PROFILE
dtime +=dclock();
print_flops("",fname,66*non_local_chi_cb[wire[i]],dtime);
#endif
}
else if(wire[i] == 6)
{
#ifdef PROFILE
dtime = - dclock();
#endif
#if 1
pt_copy_buffer(non_local_chi_cb[6],(long)vin[i],(long)snd_buf_t_cb,(long)Toffset[parity]);
#else
for(j = 0; j < non_local_chi_cb[6];j++)
for(k = 0; k < VECT_LEN;k++)
*(snd_buf_t_cb+j*VECT_LEN+k) = *(vin[i] + *(Toffset[parity]+j)+ k);
// moveMem(snd_buf_t_cb + j*VECT_LEN,vin[i] + *(Toffset[parity]+j),VECT_LEN*sizeof(IFloat));
#endif
#ifdef PROFILE
dtime +=dclock();
print_flops(fname,"pt_copy_buffer()",0,dtime);
#endif
}
}
}
#ifdef PROFILE
dtime = - dclock();
#endif
#if 1
int comms = 0;
for(i=0;i<n;i++)
{
if(!local[wire[i]/2])
{
//Calculate the starting address for the data to be sent
IFloat *addr = vin[i] + VECT_LEN * offset_cb[wire[i]];
//This points to the appropriate SCUDirArg for receiving
SCUarg_p[2*comms] = SCUarg_cb[2*wire[i]];
//This points to the appropriate SCUDirArg for sending
SCUarg_p[2*comms+1] = SCUarg_cb[2*wire[i]+1];
//Set the send address
if(wire[i]%2)
SCUarg_p[2*comms+1]->Addr((void *)snd_buf_cb[wire[i]/2]);
else if(wire[i] == 6)
SCUarg_p[2*comms+1]->Addr((void *)snd_buf_t_cb);
else
SCUarg_p[2*comms+1]->Addr((void *)addr);
comms++;
}
}
#endif
if(comms){
SCUmulti.Init(SCUarg_p,2*comms);
//Begin transmission
SCUmulti.SlowStartTrans();
}
//Do local calculations
if(conjugated)
{
for(i=0;i<n;i++)
{
#ifdef PROFILE
dtime = - dclock();
#endif
if(wire[i]%2)
partrans_cmv(local_chi_cb[wire[i]]/2,uc_l_cb[parity][wire[i]],gauge,vin[i],vout[i]);
else
partrans_cmv_dag(local_chi_cb[wire[i]]/2,uc_l_cb[parity][wire[i]],gauge,vin[i],vout[i]);
#ifdef PROFILE
dtime +=dclock();
print_flops("",fname,66*local_chi_cb[wire[i]],dtime);
#endif
}
}
else
{
for(i=0;i<n;i++)
{
#ifdef PROFILE
dtime = - dclock();
#endif
if(!(wire[i]%2))
partrans_cmv(local_chi_cb[wire[i]]/2,uc_l_cb[parity][wire[i]],gauge,vin[i],vout[i]);
else
partrans_cmv_dag(local_chi_cb[wire[i]]/2,uc_l_cb[parity][wire[i]],gauge,vin[i],vout[i]);
#ifdef PROFILE
dtime +=dclock();
print_flops("",fname,66*local_chi_cb[wire[i]],dtime);
#endif
}
}
//End transmission
if(comms){ SCUmulti.TransComplete(); }
//If wire[i] is even, then we have transport in the negative direction.
//In this case, the vector field is multiplied by the SU(3) link matrix
//after all communication is complete
IFloat *fp0, *fp1;
for(i=0;i<n;i++)
{
if(!local[wire[i]/2])
{
if(!(wire[i]%2))
{
#ifdef PROFILE
dtime = - dclock();
#endif
if(conjugated)
partrans_cmv_dag(non_local_chi_cb[wire[i]]/2,uc_nl_cb[parity][wire[i]],gauge,rcv_buf[wire[i]],vout[i]);
else
partrans_cmv(non_local_chi_cb[wire[i]]/2,uc_nl_cb[parity][wire[i]],gauge,rcv_buf[wire[i]],vout[i]);
#ifdef PROFILE
dtime +=dclock();
print_flops("",fname,66*non_local_chi_cb[wire[i]],dtime);
#endif
}
//Otherwise we have parallel transport in the positive direction.
//In this case, the received data has already been pre-multiplied
//All we need to do is to put the transported field in the correct place
else
{
#ifdef PROFILE
dtime = - dclock();
#endif
#if 1
pt_copy(non_local_chi_cb[wire[i]]/2,uc_nl_cb[parity][wire[i]],rcv_buf[wire[i]],vout[i]);
#else
//Place the data in the receive buffer into the result vector
for(s=0;s<non_local_chi_cb[wire[i]];s++)
{
fp0 = (IFloat *)((long)rcv_buf[wire[i]]+uc_nl_cb[parity][wire[i]][s].src);
fp1 = (IFloat *)((long)vout[i]+uc_nl_cb[parity][wire[i]][s].dest);
for(d = 0;d<VECT_LEN;d++)
*(fp1+d) = *(fp0+d);
//moveMem(fp1,fp0,VECT_LEN*sizeof(IFloat));
}
#endif
#ifdef PROFILE
dtime +=dclock();
print_flops(fname,"pt_copy()",0,dtime);
#endif
}
}
}
// ParTrans::PTflops +=33*n*vol;
}
/*! Take a top-object parse-tree (pt0), and expand all tree references one level.
*
* One level only. Parse-tree is expanded itself (not copy).
*
* @param[in] h Handle needed to resolve tree-references (\@tree)
* @param[in] coprev Parent, if any
* @param[in,out] pt0 parse-tree to expand. In: original, out: expanded
* @note The loop may need to suboptimal iterations since after every time you
* find a tree reference, you add new elements to the list and re-iterates
* from start. Since the expanded elements may be references,...
* @see pt_expand_treeref_cleanup
*/
int
pt_expand_treeref(cligen_handle h,
cg_obj *coprev,
parse_tree *pt0)
{
int i, k;
cg_obj *co;
parse_tree *ptref; /* tree referenced by pt0 orig */
parse_tree pt1ref = {0, }; /* tree referenced by pt0 resolved */
cg_obj *cot; /* treeref object */
char *treename;
cg_obj *coprev2;
if (pt0->pt_vec == NULL)
return 0;
again:
for (i=0; i<pt0->pt_len; i++){ /* */
if ((co = pt0->pt_vec[i]) == NULL)
continue;
/* XXX remove reference */
if (co->co_type == CO_REFERENCE && !co->co_refdone){
/* Expansion is made in-line so we need to know if already
expanded */
treename = co->co_command;
/* Find the referring tree */
if ((ptref = cligen_tree_find(h, treename)) == NULL){
fprintf(stderr, "CLIgen subtree '%s' not found\n",
treename);
return -1;
}
/* make a copy of ptref -> pt1ref */
coprev2 = co_up(co);
if (pt_copy(*ptref, coprev2, &pt1ref) < 0){ /* From ptref -> pt1ref */
fprintf(stderr, "%s: Copying parse-tree\n", __FUNCTION__);
return -1;
}
/* Recursively add extra NULLs in non-terminals */
if (co->co_hide && /* XXX: hide to trunk? */
pt_reference_trunc(pt1ref) < 0)
return -1;
/* Recursively install callback all through the referenced tree */
if (co->co_callbacks &&
pt_callback_reference(pt1ref, co->co_callbacks) < 0)
return -1;
/* Copy top-levels into original parse-tree */
for (k=0; k<pt1ref.pt_len; k++)
if ((cot = pt1ref.pt_vec[k]) != NULL){
cot->co_treeref++; /* Mark as expanded referenced tree */
if (co_insert(pt0, cot) == NULL) /* XXX alphabetically */
return -1;
}
/* Due to loop above, all co in vec should be moved, it should
be safe to remove */
free(pt1ref.pt_vec);
if (coprev && coprev->co_ref) /* coprev2 ? */
coprev->co_ref->co_pt = coprev->co_pt;
co->co_refdone = 1;
goto again;
}
}
return 0;
}
/* Merge and validate the PrintTicket part according to the current scope. */
Bool pt_mandv(
/*@in@*/ /*@notnull@*/
xmlGFilter* filter,
/*@in@*/ /*@notnull@*/
XPS_PT* pt,
/*@in@*/ /*@notnull@*/
xps_partname_t* part)
{
Bool status;
OBJECT part_file = OBJECT_NOTVM_NOTHING;
uint8* pt_uri;
uint32 pt_uri_len;
xmlGIStr *pt_mimetype ;
static XPS_CONTENT_TYPES pt_content_types[] = {
{ XML_INTERN(mimetype_printing_printticket) },
XPS_CONTENT_TYPES_END
} ;
HQASSERT((pt != NULL),
"pt_mandv: state pointer NULL");
HQASSERT((pt->device == NULL ||
(pt->scope >= PT_SCOPE_JOB && pt->scope <= PT_SCOPE_PAGE)),
"pt_mandv: PT has invalid scope");
/* Catch absence of any PT device */
if ( pt->device == NULL ) {
return TRUE ;
}
/* Open a PS file on the PrintTicket part for read. */
if ( !xps_open_file_from_partname(filter, part, &part_file,
XML_INTERN(rel_xps_2005_06_printticket),
pt_content_types,
&pt_mimetype, FALSE)) {
return FALSE ;
}
status = FALSE;
/* Open the scope start file and copy the print ticket part contents to it.
* Note: the scope start file gets closed by reading back the PS config and
* that this will happen before the next print ticket is seen.
*/
if ( pt_open_config_file(pt, (uint8*)"S", SW_RDWR, RW_FLAG, &pt->scope_file) ) {
status = pt_copy(oFile(part_file), oFile(pt->scope_file));
if ( !status ) {
pt_close_start_file(pt);
if ( !hqn_uri_get_field(part->uri, &pt_uri, &pt_uri_len, HQN_URI_NAME)) {
HQFAIL("Unable to get PT URI") ;
pt_uri_len = 0;
}
(void)pt_error_handler(pt, pt_uri, pt_uri_len);
}
}
/* Failure reading from the PT part will have already been caught so don't
* really care what happens when closing here.
*/
(void)xml_file_close(&part_file);
return status ;
} /* pt_mandv */
