/***********************************************************************************
* This function determines the cost of moving data between the two meshes assuming
* that a good matching between the two partitions was done!
************************************************************************************/
int ComputeMapCost(idxtype nvtxs, idxtype nparts, idxtype *fepart, idxtype *cpart)
{
idxtype i, j, k, n, ncomm;
KeyValueType cand[nparts*nparts];
idxtype fmatched[nparts], cmatched[nparts];
/* Compute the overlap */
for (i=0; i<nparts; i++) {
for (j=0; j<nparts; j++) {
cand[i*nparts+j].key = 0;
cand[i*nparts+j].val = i*nparts+j;
}
}
for (k=0, i=0; i<nvtxs; i++) {
if (cpart[i] >= 0) {
cand[(fepart[i]-1)*nparts+(cpart[i]-1)].key++;
k++;
}
}
mprintf("Contact points: %D\n", k);
ikeysort(nparts*nparts, cand);
iset(nparts, -1, fmatched);
iset(nparts, -1, cmatched);
for (ncomm=0, k=nparts*nparts-1; k>=0; k--) {
i = cand[k].val/nparts;
j = cand[k].val%nparts;
if (fmatched[i] == -1 && cmatched[j] == -1) {
fmatched[i] = j;
cmatched[j] = i;
}
else
ncomm += cand[k].key;
}
mprintf("Ncomm: %D\n", ncomm);
return ncomm;
}
/*************************************************************************
* This function computes the subdomain graph
**************************************************************************/
void EliminateSubDomainEdges(CtrlType *ctrl, GraphType *graph, int nparts, float *tpwgts)
{
int i, ii, j, k, me, other, nvtxs, total, max, avg, totalout, nind, ncand, ncand2, target, target2, nadd;
int min, move, cpwgt, tvwgt;
idxtype *xadj, *adjncy, *vwgt, *adjwgt, *pwgts, *where, *maxpwgt, *pmat, *ndoms, *mypmat, *otherpmat, *ind;
KeyValueType *cand, *cand2;
nvtxs = graph->nvtxs;
xadj = graph->xadj;
adjncy = graph->adjncy;
vwgt = graph->vwgt;
adjwgt = graph->adjwgt;
where = graph->where;
pwgts = graph->pwgts; /* We assume that this is properly initialized */
maxpwgt = idxwspacemalloc(ctrl, nparts);
ndoms = idxwspacemalloc(ctrl, nparts);
otherpmat = idxwspacemalloc(ctrl, nparts);
ind = idxwspacemalloc(ctrl, nvtxs);
pmat = ctrl->wspace.pmat;
cand = (KeyValueType *)GKmalloc(nparts*sizeof(KeyValueType), "EliminateSubDomainEdges: cand");
cand2 = (KeyValueType *)GKmalloc(nparts*sizeof(KeyValueType), "EliminateSubDomainEdges: cand");
/* Compute the pmat matrix and ndoms */
ComputeSubDomainGraph(graph, nparts, pmat, ndoms);
/* Compute the maximum allowed weight for each domain */
tvwgt = idxsum(nparts, pwgts);
for (i=0; i<nparts; i++)
maxpwgt[i] = 1.25*tpwgts[i]*tvwgt;
/* Get into the loop eliminating subdomain connections */
for (;;) {
total = idxsum(nparts, ndoms);
avg = total/nparts;
max = ndoms[idxamax(nparts, ndoms)];
/* printf("Adjacent Subdomain Stats: Total: %3d, Max: %3d, Avg: %3d [%5d]\n", total, max, avg, idxsum(nparts*nparts, pmat)); */
if (max < 1.4*avg)
break;
me = idxamax(nparts, ndoms);
mypmat = pmat + me*nparts;
totalout = idxsum(nparts, mypmat);
/*printf("Me: %d, TotalOut: %d,\n", me, totalout);*/
/* Sort the connections according to their cut */
for (ncand2=0, i=0; i<nparts; i++) {
if (mypmat[i] > 0) {
cand2[ncand2].key = mypmat[i];
cand2[ncand2++].val = i;
}
}
ikeysort(ncand2, cand2);
move = 0;
for (min=0; min<ncand2; min++) {
if (cand2[min].key > totalout/(2*ndoms[me]))
break;
other = cand2[min].val;
/*printf("\tMinOut: %d to %d\n", mypmat[other], other);*/
idxset(nparts, 0, otherpmat);
/* Go and find the vertices in 'other' that are connected in 'me' */
for (nind=0, i=0; i<nvtxs; i++) {
if (where[i] == other) {
for (j=xadj[i]; j<xadj[i+1]; j++) {
if (where[adjncy[j]] == me) {
ind[nind++] = i;
break;
}
}
}
}
/* Go and construct the otherpmat to see where these nind vertices are connected to */
for (cpwgt=0, ii=0; ii<nind; ii++) {
i = ind[ii];
cpwgt += vwgt[i];
for (j=xadj[i]; j<xadj[i+1]; j++)
otherpmat[where[adjncy[j]]] += adjwgt[j];
}
otherpmat[other] = 0;
for (ncand=0, i=0; i<nparts; i++) {
if (otherpmat[i] > 0) {
cand[ncand].key = -otherpmat[i];
cand[ncand++].val = i;
}
}
ikeysort(ncand, cand);
/*
* Go through and the select the first domain that is common with 'me', and
* does not increase the ndoms[target] higher than my ndoms, subject to the
* maxpwgt constraint. Traversal is done from the mostly connected to the least.
*/
target = target2 = -1;
for (i=0; i<ncand; i++) {
k = cand[i].val;
if (mypmat[k] > 0) {
if (pwgts[k] + cpwgt > maxpwgt[k]) /* Check if balance will go off */
continue;
for (j=0; j<nparts; j++) {
if (otherpmat[j] > 0 && ndoms[j] >= ndoms[me]-1 && pmat[nparts*j+k] == 0)
break;
}
if (j == nparts) { /* No bad second level effects */
for (nadd=0, j=0; j<nparts; j++) {
if (otherpmat[j] > 0 && pmat[nparts*k+j] == 0)
nadd++;
}
/*printf("\t\tto=%d, nadd=%d, %d\n", k, nadd, ndoms[k]);*/
if (target2 == -1 && ndoms[k]+nadd < ndoms[me]) {
target2 = k;
}
if (nadd == 0) {
target = k;
break;
}
}
}
}
if (target == -1 && target2 != -1)
target = target2;
if (target == -1) {
/* printf("\t\tCould not make the move\n");*/
continue;
}
/*printf("\t\tMoving to %d\n", target);*/
/* Update the partition weights */
INC_DEC(pwgts[target], pwgts[other], cpwgt);
MoveGroupMConn(ctrl, graph, ndoms, pmat, nparts, target, nind, ind);
move = 1;
break;
}
if (move == 0)
break;
}
idxwspacefree(ctrl, nparts);
idxwspacefree(ctrl, nparts);
idxwspacefree(ctrl, nparts);
idxwspacefree(ctrl, nvtxs);
GKfree(&cand, &cand2, LTERM);
}
/*************************************************************************
* This function compresses a graph by merging identical vertices
* The compression should lead to at least 10% reduction.
**************************************************************************/
void CompressGraph(CtrlType *ctrl, GraphType *graph, int nvtxs, idxtype *xadj, idxtype *adjncy, idxtype *cptr, idxtype *cind)
{
int i, ii, iii, j, jj, k, l, cnvtxs, cnedges;
idxtype *cxadj, *cadjncy, *cvwgt, *mark, *map;
KeyValueType *keys;
mark = idxsmalloc(nvtxs, -1, "CompressGraph: mark");
map = idxsmalloc(nvtxs, -1, "CompressGraph: map");
keys = (KeyValueType *)GKmalloc(nvtxs*sizeof(KeyValueType), "CompressGraph: keys");
/* Compute a key for each adjacency list */
for (i=0; i<nvtxs; i++) {
k = 0;
for (j=xadj[i]; j<xadj[i+1]; j++)
k += adjncy[j];
keys[i].key = k+i; /* Add the diagonal entry as well */
keys[i].val = i;
}
ikeysort(nvtxs, keys);
l = cptr[0] = 0;
for (cnvtxs=i=0; i<nvtxs; i++) {
ii = keys[i].val;
if (map[ii] == -1) {
mark[ii] = i; /* Add the diagonal entry */
for (j=xadj[ii]; j<xadj[ii+1]; j++)
mark[adjncy[j]] = i;
cind[l++] = ii;
map[ii] = cnvtxs;
for (j=i+1; j<nvtxs; j++) {
iii = keys[j].val;
if (keys[i].key != keys[j].key || xadj[ii+1]-xadj[ii] != xadj[iii+1]-xadj[iii])
break; /* Break if keys or degrees are different */
if (map[iii] == -1) { /* Do a comparison if iii has not been mapped */
for (jj=xadj[iii]; jj<xadj[iii+1]; jj++) {
if (mark[adjncy[jj]] != i)
break;
}
if (jj == xadj[iii+1]) { /* Identical adjacency structure */
map[iii] = cnvtxs;
cind[l++] = iii;
}
}
}
cptr[++cnvtxs] = l;
}
}
/* printf("Original: %6d, Compressed: %6d\n", nvtxs, cnvtxs); */
InitGraph(graph);
if (cnvtxs >= COMPRESSION_FRACTION*nvtxs) {
graph->nvtxs = nvtxs;
graph->nedges = xadj[nvtxs];
graph->ncon = 1;
graph->xadj = xadj;
graph->adjncy = adjncy;
graph->gdata = idxmalloc(3*nvtxs+graph->nedges, "CompressGraph: gdata");
graph->vwgt = graph->gdata;
graph->adjwgtsum = graph->gdata+nvtxs;
graph->cmap = graph->gdata+2*nvtxs;
graph->adjwgt = graph->gdata+3*nvtxs;
idxset(nvtxs, 1, graph->vwgt);
idxset(graph->nedges, 1, graph->adjwgt);
for (i=0; i<nvtxs; i++)
graph->adjwgtsum[i] = xadj[i+1]-xadj[i];
graph->label = idxmalloc(nvtxs, "CompressGraph: label");
for (i=0; i<nvtxs; i++)
graph->label[i] = i;
}
else { /* Ok, form the compressed graph */
cnedges = 0;
for (i=0; i<cnvtxs; i++) {
ii = cind[cptr[i]];
cnedges += xadj[ii+1]-xadj[ii];
}
/* Allocate memory for the compressed graph*/
graph->gdata = idxmalloc(4*cnvtxs+1 + 2*cnedges, "CompressGraph: gdata");
cxadj = graph->xadj = graph->gdata;
cvwgt = graph->vwgt = graph->gdata + cnvtxs+1;
graph->adjwgtsum = graph->gdata + 2*cnvtxs+1;
graph->cmap = graph->gdata + 3*cnvtxs+1;
cadjncy = graph->adjncy = graph->gdata + 4*cnvtxs+1;
graph->adjwgt = graph->gdata + 4*cnvtxs+1 + cnedges;
/* Now go and compress the graph */
idxset(nvtxs, -1, mark);
l = cxadj[0] = 0;
for (i=0; i<cnvtxs; i++) {
cvwgt[i] = cptr[i+1]-cptr[i];
mark[i] = i; /* Remove any dioganal entries in the compressed graph */
for (j=cptr[i]; j<cptr[i+1]; j++) {
ii = cind[j];
for (jj=xadj[ii]; jj<xadj[ii+1]; jj++) {
k = map[adjncy[jj]];
if (mark[k] != i)
cadjncy[l++] = k;
mark[k] = i;
}
}
cxadj[i+1] = l;
}
graph->nvtxs = cnvtxs;
graph->nedges = l;
graph->ncon = 1;
idxset(graph->nedges, 1, graph->adjwgt);
for (i=0; i<cnvtxs; i++)
graph->adjwgtsum[i] = cxadj[i+1]-cxadj[i];
graph->label = idxmalloc(cnvtxs, "CompressGraph: label");
for (i=0; i<cnvtxs; i++)
graph->label[i] = i;
}
GKfree(&keys, &map, &mark, LTERM);
}
/*************************************************************************
* This function performs an edge-based FM refinement
**************************************************************************/
void Mc_Serial_Balance2Way(GraphType *graph, float *tpwgts, float lbfactor)
{
int i, ii, j, k, kwgt, nvtxs, ncon, nbnd, nswaps, from, to, limit, tmp, cnum;
idxtype *xadj, *adjncy, *adjwgt, *where, *id, *ed, *bndptr, *bndind;
idxtype *moved, *swaps, *qnum;
float *nvwgt, *npwgts, mindiff[MAXNCON], origbal, minbal, newbal;
FPQueueType parts[MAXNCON][2];
int higain, oldgain, mincut, newcut, mincutorder;
int qsizes[MAXNCON][2];
KeyValueType *cand;
nvtxs = graph->nvtxs;
ncon = graph->ncon;
xadj = graph->xadj;
nvwgt = graph->nvwgt;
adjncy = graph->adjncy;
adjwgt = graph->adjwgt;
where = graph->where;
id = graph->sendind;
ed = graph->recvind;
npwgts = graph->gnpwgts;
bndptr = graph->sendptr;
bndind = graph->recvptr;
moved = idxmalloc(nvtxs, "moved");
swaps = idxmalloc(nvtxs, "swaps");
qnum = idxmalloc(nvtxs, "qnum");
cand = (KeyValueType *)GKmalloc(nvtxs*sizeof(KeyValueType), "cand");
limit = amin(amax(0.01*nvtxs, 15), 100);
/* Initialize the queues */
for (i=0; i<ncon; i++) {
FPQueueInit(&parts[i][0], nvtxs);
FPQueueInit(&parts[i][1], nvtxs);
qsizes[i][0] = qsizes[i][1] = 0;
}
for (i=0; i<nvtxs; i++) {
qnum[i] = samax(ncon, nvwgt+i*ncon);
qsizes[qnum[i]][where[i]]++;
}
for (from=0; from<2; from++) {
for (j=0; j<ncon; j++) {
if (qsizes[j][from] == 0) {
for (i=0; i<nvtxs; i++) {
if (where[i] != from)
continue;
k = samax2(ncon, nvwgt+i*ncon);
if (k == j &&
qsizes[qnum[i]][from] > qsizes[j][from] &&
nvwgt[i*ncon+qnum[i]] < 1.3*nvwgt[i*ncon+j]) {
qsizes[qnum[i]][from]--;
qsizes[j][from]++;
qnum[i] = j;
}
}
}
}
}
for (i=0; i<ncon; i++)
mindiff[i] = fabs(tpwgts[i]-npwgts[i]);
minbal = origbal = Serial_Compute2WayHLoadImbalance(ncon, npwgts, tpwgts);
newcut = mincut = graph->mincut;
mincutorder = -1;
idxset(nvtxs, -1, moved);
/* Insert all nodes in the priority queues */
nbnd = graph->gnvtxs;
for (i=0; i<nvtxs; i++) {
cand[i].key = id[i]-ed[i];
cand[i].val = i;
}
ikeysort(nvtxs, cand);
for (ii=0; ii<nvtxs; ii++) {
i = cand[ii].val;
FPQueueInsert(&parts[qnum[i]][where[i]], i, (float)(ed[i]-id[i]));
}
for (nswaps=0; nswaps<nvtxs; nswaps++) {
if (minbal < lbfactor)
break;
Serial_SelectQueue(ncon, npwgts, tpwgts, &from, &cnum, parts);
to = (from+1)%2;
if (from == -1 || (higain = FPQueueGetMax(&parts[cnum][from])) == -1)
break;
saxpy2(ncon, 1.0, nvwgt+higain*ncon, 1, npwgts+to*ncon, 1);
saxpy2(ncon, -1.0, nvwgt+higain*ncon, 1, npwgts+from*ncon, 1);
newcut -= (ed[higain]-id[higain]);
newbal = Serial_Compute2WayHLoadImbalance(ncon, npwgts, tpwgts);
if (newbal < minbal || (newbal == minbal &&
(newcut < mincut || (newcut == mincut &&
Serial_BetterBalance(ncon, npwgts, tpwgts, mindiff))))) {
mincut = newcut;
minbal = newbal;
mincutorder = nswaps;
for (i=0; i<ncon; i++)
mindiff[i] = fabs(tpwgts[i]-npwgts[i]);
}
else if (nswaps-mincutorder > limit) { /* We hit the limit, undo last move */
newcut += (ed[higain]-id[higain]);
saxpy2(ncon, 1.0, nvwgt+higain*ncon, 1, npwgts+from*ncon, 1);
saxpy2(ncon, -1.0, nvwgt+higain*ncon, 1, npwgts+to*ncon, 1);
break;
}
where[higain] = to;
moved[higain] = nswaps;
swaps[nswaps] = higain;
/**************************************************************
* Update the id[i]/ed[i] values of the affected nodes
***************************************************************/
SWAP(id[higain], ed[higain], tmp);
if (ed[higain] == 0 && bndptr[higain] != -1 && xadj[higain] < xadj[higain+1])
BNDDelete(nbnd, bndind, bndptr, higain);
if (ed[higain] > 0 && bndptr[higain] == -1)
BNDInsert(nbnd, bndind, bndptr, higain);
for (j=xadj[higain]; j<xadj[higain+1]; j++) {
k = adjncy[j];
oldgain = ed[k]-id[k];
kwgt = (to == where[k] ? adjwgt[j] : -adjwgt[j]);
INC_DEC(id[k], ed[k], kwgt);
/* Update the queue position */
if (moved[k] == -1)
FPQueueUpdate(&parts[qnum[k]][where[k]], k, (float)(oldgain), (float)(ed[k]-id[k]));
/* Update its boundary information */
if (ed[k] == 0 && bndptr[k] != -1)
BNDDelete(nbnd, bndind, bndptr, k);
else if (ed[k] > 0 && bndptr[k] == -1)
BNDInsert(nbnd, bndind, bndptr, k);
}
}
/****************************************************************
* Roll back computations
*****************************************************************/
for (nswaps--; nswaps>mincutorder; nswaps--) {
higain = swaps[nswaps];
to = where[higain] = (where[higain]+1)%2;
SWAP(id[higain], ed[higain], tmp);
if (ed[higain] == 0 && bndptr[higain] != -1 && xadj[higain] < xadj[higain+1])
BNDDelete(nbnd, bndind, bndptr, higain);
else if (ed[higain] > 0 && bndptr[higain] == -1)
BNDInsert(nbnd, bndind, bndptr, higain);
saxpy2(ncon, 1.0, nvwgt+higain*ncon, 1, npwgts+to*ncon, 1);
saxpy2(ncon, -1.0, nvwgt+higain*ncon, 1, npwgts+((to+1)%2)*ncon, 1);
for (j=xadj[higain]; j<xadj[higain+1]; j++) {
k = adjncy[j];
kwgt = (to == where[k] ? adjwgt[j] : -adjwgt[j]);
INC_DEC(id[k], ed[k], kwgt);
if (bndptr[k] != -1 && ed[k] == 0)
BNDDelete(nbnd, bndind, bndptr, k);
if (bndptr[k] == -1 && ed[k] > 0)
BNDInsert(nbnd, bndind, bndptr, k);
}
}
graph->mincut = mincut;
graph->gnvtxs = nbnd;
for (i=0; i<ncon; i++) {
FPQueueFree(&parts[i][0]);
FPQueueFree(&parts[i][1]);
}
GKfree((void **)&cand, (void **)&qnum, (void **)&moved, (void **)&swaps, LTERM);
return;
}
/*************************************************************************
* This function performs an edge-based FM refinement
**************************************************************************/
void Mc_Serial_FM_2WayRefine(GraphType *graph, float *tpwgts, int npasses)
{
int i, ii, j, k;
int kwgt, nvtxs, ncon, nbnd, nswaps, from, to, pass, limit, tmp, cnum;
idxtype *xadj, *adjncy, *adjwgt, *where, *id, *ed, *bndptr, *bndind;
idxtype *moved, *swaps, *qnum;
float *nvwgt, *npwgts, mindiff[MAXNCON], origbal, minbal, newbal;
FPQueueType parts[MAXNCON][2];
int higain, oldgain, mincut, initcut, newcut, mincutorder;
float rtpwgts[MAXNCON*2];
KeyValueType *cand;
int mype;
MPI_Comm_rank(MPI_COMM_WORLD, &mype);
nvtxs = graph->nvtxs;
ncon = graph->ncon;
xadj = graph->xadj;
nvwgt = graph->nvwgt;
adjncy = graph->adjncy;
adjwgt = graph->adjwgt;
where = graph->where;
id = graph->sendind;
ed = graph->recvind;
npwgts = graph->gnpwgts;
bndptr = graph->sendptr;
bndind = graph->recvptr;
moved = idxmalloc(nvtxs, "moved");
swaps = idxmalloc(nvtxs, "swaps");
qnum = idxmalloc(nvtxs, "qnum");
cand = (KeyValueType *)GKmalloc(nvtxs*sizeof(KeyValueType), "cand");
limit = amin(amax(0.01*nvtxs, 25), 150);
/* Initialize the queues */
for (i=0; i<ncon; i++) {
FPQueueInit(&parts[i][0], nvtxs);
FPQueueInit(&parts[i][1], nvtxs);
}
for (i=0; i<nvtxs; i++)
qnum[i] = samax(ncon, nvwgt+i*ncon);
origbal = Serial_Compute2WayHLoadImbalance(ncon, npwgts, tpwgts);
for (i=0; i<ncon; i++) {
rtpwgts[i] = origbal*tpwgts[i];
rtpwgts[ncon+i] = origbal*tpwgts[ncon+i];
}
idxset(nvtxs, -1, moved);
for (pass=0; pass<npasses; pass++) { /* Do a number of passes */
for (i=0; i<ncon; i++) {
FPQueueReset(&parts[i][0]);
FPQueueReset(&parts[i][1]);
}
mincutorder = -1;
newcut = mincut = initcut = graph->mincut;
for (i=0; i<ncon; i++)
mindiff[i] = fabs(tpwgts[i]-npwgts[i]);
minbal = Serial_Compute2WayHLoadImbalance(ncon, npwgts, tpwgts);
/* Insert boundary nodes in the priority queues */
nbnd = graph->gnvtxs;
for (i=0; i<nbnd; i++) {
cand[i].key = id[i]-ed[i];
cand[i].val = i;
}
ikeysort(nbnd, cand);
for (ii=0; ii<nbnd; ii++) {
i = bndind[cand[ii].val];
FPQueueInsert(&parts[qnum[i]][where[i]], i, (float)(ed[i]-id[i]));
}
for (nswaps=0; nswaps<nvtxs; nswaps++) {
Serial_SelectQueue(ncon, npwgts, rtpwgts, &from, &cnum, parts);
to = (from+1)%2;
if (from == -1 || (higain = FPQueueGetMax(&parts[cnum][from])) == -1)
break;
saxpy2(ncon, 1.0, nvwgt+higain*ncon, 1, npwgts+to*ncon, 1);
saxpy2(ncon, -1.0, nvwgt+higain*ncon, 1, npwgts+from*ncon, 1);
newcut -= (ed[higain]-id[higain]);
newbal = Serial_Compute2WayHLoadImbalance(ncon, npwgts, tpwgts);
if ((newcut < mincut && newbal-origbal <= .00001) ||
(newcut == mincut && (newbal < minbal ||
(newbal == minbal && Serial_BetterBalance(ncon, npwgts, tpwgts, mindiff))))) {
mincut = newcut;
minbal = newbal;
mincutorder = nswaps;
for (i=0; i<ncon; i++)
mindiff[i] = fabs(tpwgts[i]-npwgts[i]);
}
else if (nswaps-mincutorder > limit) { /* We hit the limit, undo last move */
newcut += (ed[higain]-id[higain]);
saxpy2(ncon, 1.0, nvwgt+higain*ncon, 1, npwgts+from*ncon, 1);
saxpy2(ncon, -1.0, nvwgt+higain*ncon, 1, npwgts+to*ncon, 1);
break;
}
where[higain] = to;
moved[higain] = nswaps;
swaps[nswaps] = higain;
/**************************************************************
* Update the id[i]/ed[i] values of the affected nodes
***************************************************************/
SWAP(id[higain], ed[higain], tmp);
if (ed[higain] == 0 && xadj[higain] < xadj[higain+1])
BNDDelete(nbnd, bndind, bndptr, higain);
for (j=xadj[higain]; j<xadj[higain+1]; j++) {
k = adjncy[j];
oldgain = ed[k]-id[k];
kwgt = (to == where[k] ? adjwgt[j] : -adjwgt[j]);
INC_DEC(id[k], ed[k], kwgt);
/* Update its boundary information and queue position */
if (bndptr[k] != -1) { /* If k was a boundary vertex */
if (ed[k] == 0) { /* Not a boundary vertex any more */
BNDDelete(nbnd, bndind, bndptr, k);
if (moved[k] == -1) /* Remove it if in the queues */
FPQueueDelete(&parts[qnum[k]][where[k]], k);
}
else { /* If it has not been moved, update its position in the queue */
if (moved[k] == -1)
FPQueueUpdate(&parts[qnum[k]][where[k]], k, (float)oldgain, (float)(ed[k]-id[k]));
}
}
else {
if (ed[k] > 0) { /* It will now become a boundary vertex */
BNDInsert(nbnd, bndind, bndptr, k);
if (moved[k] == -1)
FPQueueInsert(&parts[qnum[k]][where[k]], k, (float)(ed[k]-id[k]));
}
}
}
}
/****************************************************************
* Roll back computations
*****************************************************************/
for (i=0; i<nswaps; i++)
moved[swaps[i]] = -1; /* reset moved array */
for (nswaps--; nswaps>mincutorder; nswaps--) {
higain = swaps[nswaps];
to = where[higain] = (where[higain]+1)%2;
SWAP(id[higain], ed[higain], tmp);
if (ed[higain] == 0 && bndptr[higain] != -1 && xadj[higain] < xadj[higain+1])
BNDDelete(nbnd, bndind, bndptr, higain);
else if (ed[higain] > 0 && bndptr[higain] == -1)
BNDInsert(nbnd, bndind, bndptr, higain);
saxpy2(ncon, 1.0, nvwgt+higain*ncon, 1, npwgts+to*ncon, 1);
saxpy2(ncon, -1.0, nvwgt+higain*ncon, 1, npwgts+((to+1)%2)*ncon, 1);
for (j=xadj[higain]; j<xadj[higain+1]; j++) {
k = adjncy[j];
kwgt = (to == where[k] ? adjwgt[j] : -adjwgt[j]);
INC_DEC(id[k], ed[k], kwgt);
if (bndptr[k] != -1 && ed[k] == 0)
BNDDelete(nbnd, bndind, bndptr, k);
if (bndptr[k] == -1 && ed[k] > 0)
BNDInsert(nbnd, bndind, bndptr, k);
}
}
graph->mincut = mincut;
graph->gnvtxs = nbnd;
if (mincutorder == -1 || mincut == initcut)
break;
}
for (i=0; i<ncon; i++) {
FPQueueFree(&parts[i][0]);
FPQueueFree(&parts[i][1]);
}
GKfree((void **)&cand, (void **)&qnum, (void **)&moved, (void **)&swaps, LTERM);
return;
}
/**************************************************************
* This subroutine remaps a partitioning on a single processor
**************************************************************/
void SerialRemap(GraphType *graph, int nparts, idxtype *base, idxtype *scratch,
idxtype *remap, float *tpwgts)
{
int i, ii, j, k;
int nvtxs, nmapped, max_mult;
int from, to, current_from, smallcount, bigcount;
KeyValueType *flowto, *bestflow;
KeyKeyValueType *sortvtx;
idxtype *vsize, *htable, *map, *rowmap;
nvtxs = graph->nvtxs;
vsize = graph->vsize;
max_mult = amin(MAX_NPARTS_MULTIPLIER, nparts);
sortvtx = (KeyKeyValueType *)GKmalloc(nvtxs*sizeof(KeyKeyValueType), "sortvtx");
flowto = (KeyValueType *)GKmalloc((nparts*max_mult+nparts)*sizeof(KeyValueType), "flowto");
bestflow = flowto+nparts;
map = htable = idxsmalloc(nparts*2, -1, "htable");
rowmap = map+nparts;
for (i=0; i<nvtxs; i++) {
sortvtx[i].key1 = base[i];
sortvtx[i].key2 = vsize[i];
sortvtx[i].val = i;
}
qsort((void *)sortvtx, (size_t)nvtxs, (size_t)sizeof(KeyKeyValueType), SSMIncKeyCmp);
for (j=0; j<nparts; j++) {
flowto[j].key = 0;
flowto[j].val = j;
}
/* this step has nparts*nparts*log(nparts) computational complexity */
bigcount = smallcount = current_from = 0;
for (ii=0; ii<nvtxs; ii++) {
i = sortvtx[ii].val;
from = base[i];
to = scratch[i];
if (from > current_from) {
/* reset the hash table */
for (j=0; j<smallcount; j++)
htable[flowto[j].val] = -1;
ASSERTS(idxsum(nparts, htable) == -nparts);
ikeysort(smallcount, flowto);
for (j=0; j<amin(smallcount, max_mult); j++, bigcount++) {
bestflow[bigcount].key = flowto[j].key;
bestflow[bigcount].val = current_from*nparts+flowto[j].val;
}
smallcount = 0;
current_from = from;
}
if (htable[to] == -1) {
htable[to] = smallcount;
flowto[smallcount].key = -vsize[i];
flowto[smallcount].val = to;
smallcount++;
}
else {
flowto[htable[to]].key += -vsize[i];
}
}
/* reset the hash table */
for (j=0; j<smallcount; j++)
htable[flowto[j].val] = -1;
ASSERTS(idxsum(nparts, htable) == -nparts);
ikeysort(smallcount, flowto);
for (j=0; j<amin(smallcount, max_mult); j++, bigcount++) {
bestflow[bigcount].key = flowto[j].key;
bestflow[bigcount].val = current_from*nparts+flowto[j].val;
}
ikeysort(bigcount, bestflow);
ASSERTS(idxsum(nparts, map) == -nparts);
ASSERTS(idxsum(nparts, rowmap) == -nparts);
nmapped = 0;
/* now make as many assignments as possible */
for (ii=0; ii<bigcount; ii++) {
i = bestflow[ii].val;
j = i % nparts; /* to */
k = i / nparts; /* from */
if (map[j] == -1 && rowmap[k] == -1 && SimilarTpwgts(tpwgts, graph->ncon, j, k)) {
map[j] = k;
rowmap[k] = j;
nmapped++;
}
if (nmapped == nparts)
break;
}
/* remap the rest */
/* it may help try remapping to the same label first */
if (nmapped < nparts) {
for (j=0; j<nparts && nmapped<nparts; j++) {
if (map[j] == -1) {
for (ii=0; ii<nparts; ii++) {
i = (j+ii) % nparts;
if (rowmap[i] == -1 && SimilarTpwgts(tpwgts, graph->ncon, i, j)) {
map[j] = i;
rowmap[i] = j;
nmapped++;
break;
}
}
}
}
}
/* check to see if remapping fails (due to dis-similar tpwgts) */
/* if remapping fails, revert to original mapping */
if (nmapped < nparts)
for (i=0; i<nparts; i++)
map[i] = i;
for (i=0; i<nvtxs; i++)
remap[i] = map[remap[i]];
GKfree((void **)&sortvtx, (void **)&flowto, (void **)&htable, LTERM);
}
/*************************************************************************
* This function performs k-way refinement
**************************************************************************/
void Mc_SerialKWayAdaptRefine(GraphType *graph, int nparts, idxtype *home,
float *orgubvec, int npasses)
{
int i, ii, iii, j, k;
int nvtxs, ncon, pass, nmoves, myndegrees;
int from, me, myhome, to, oldcut, gain, tmp;
idxtype *xadj, *adjncy, *adjwgt;
idxtype *where;
EdgeType *mydegrees;
RInfoType *rinfo, *myrinfo;
float *npwgts, *nvwgt, *minwgt, *maxwgt, ubvec[MAXNCON];
int gain_is_greater, gain_is_same, fit_in_to, fit_in_from, going_home;
int zero_gain, better_balance_ft, better_balance_tt;
KeyValueType *cand;
int mype;
MPI_Comm_rank(MPI_COMM_WORLD, &mype);
nvtxs = graph->nvtxs;
ncon = graph->ncon;
xadj = graph->xadj;
adjncy = graph->adjncy;
adjwgt = graph->adjwgt;
where = graph->where;
rinfo = graph->rinfo;
npwgts = graph->gnpwgts;
/* Setup the weight intervals of the various subdomains */
cand = (KeyValueType *)GKmalloc(nvtxs*sizeof(KeyValueType), "cand");
minwgt = fmalloc(nparts*ncon, "minwgt");
maxwgt = fmalloc(nparts*ncon, "maxwgt");
ComputeHKWayLoadImbalance(ncon, nparts, npwgts, ubvec);
for (i=0; i<ncon; i++)
ubvec[i] = amax(ubvec[i], orgubvec[i]);
for (i=0; i<nparts; i++) {
for (j=0; j<ncon; j++) {
maxwgt[i*ncon+j] = ubvec[j]/(float)nparts;
minwgt[i*ncon+j] = ubvec[j]*(float)nparts;
}
}
for (pass=0; pass<npasses; pass++) {
oldcut = graph->mincut;
for (i=0; i<nvtxs; i++) {
cand[i].key = rinfo[i].id-rinfo[i].ed;
cand[i].val = i;
}
ikeysort(nvtxs, cand);
nmoves = 0;
for (iii=0; iii<nvtxs; iii++) {
i = cand[iii].val;
myrinfo = rinfo+i;
if (myrinfo->ed >= myrinfo->id) {
from = where[i];
myhome = home[i];
nvwgt = graph->nvwgt+i*ncon;
if (myrinfo->id > 0 &&
AreAllHVwgtsBelow(ncon, 1.0, npwgts+from*ncon, -1.0, nvwgt, minwgt+from*ncon))
continue;
mydegrees = myrinfo->degrees;
myndegrees = myrinfo->ndegrees;
for (k=0; k<myndegrees; k++) {
to = mydegrees[k].edge;
gain = mydegrees[k].ewgt - myrinfo->id;
if (gain >= 0 &&
(AreAllHVwgtsBelow(ncon, 1.0, npwgts+to*ncon, 1.0, nvwgt, maxwgt+to*ncon) ||
IsHBalanceBetterFT(ncon,npwgts+from*ncon,npwgts+to*ncon,nvwgt,ubvec))) {
break;
}
}
/* break out if you did not find a candidate */
if (k == myndegrees)
continue;
for (j=k+1; j<myndegrees; j++) {
to = mydegrees[j].edge;
going_home = (myhome == to);
gain_is_same = (mydegrees[j].ewgt == mydegrees[k].ewgt);
gain_is_greater = (mydegrees[j].ewgt > mydegrees[k].ewgt);
fit_in_to = AreAllHVwgtsBelow(ncon,1.0,npwgts+to*ncon,1.0,nvwgt,maxwgt+to*ncon);
better_balance_ft = IsHBalanceBetterFT(ncon,npwgts+from*ncon,
npwgts+to*ncon,nvwgt,ubvec);
better_balance_tt = IsHBalanceBetterTT(ncon,npwgts+mydegrees[k].edge*ncon,
npwgts+to*ncon,nvwgt,ubvec);
if (
(gain_is_greater &&
(fit_in_to ||
better_balance_ft)
)
||
(gain_is_same &&
(
(fit_in_to &&
going_home)
||
better_balance_tt
)
)
) {
k = j;
}
}
to = mydegrees[k].edge;
going_home = (myhome == to);
zero_gain = (mydegrees[k].ewgt == myrinfo->id);
fit_in_from = AreAllHVwgtsBelow(ncon,1.0,npwgts+from*ncon,0.0,npwgts+from*ncon,
maxwgt+from*ncon);
better_balance_ft = IsHBalanceBetterFT(ncon,npwgts+from*ncon,
npwgts+to*ncon,nvwgt,ubvec);
if (zero_gain &&
!going_home &&
!better_balance_ft &&
fit_in_from)
continue;
/*=====================================================================
* If we got here, we can now move the vertex from 'from' to 'to'
*======================================================================*/
graph->mincut -= mydegrees[k].ewgt-myrinfo->id;
/* Update where, weight, and ID/ED information of the vertex you moved */
saxpy2(ncon, 1.0, nvwgt, 1, npwgts+to*ncon, 1);
saxpy2(ncon, -1.0, nvwgt, 1, npwgts+from*ncon, 1);
where[i] = to;
myrinfo->ed += myrinfo->id-mydegrees[k].ewgt;
SWAP(myrinfo->id, mydegrees[k].ewgt, tmp);
if (mydegrees[k].ewgt == 0) {
myrinfo->ndegrees--;
mydegrees[k].edge = mydegrees[myrinfo->ndegrees].edge;
mydegrees[k].ewgt = mydegrees[myrinfo->ndegrees].ewgt;
}
else
mydegrees[k].edge = from;
/* Update the degrees of adjacent vertices */
for (j=xadj[i]; j<xadj[i+1]; j++) {
ii = adjncy[j];
me = where[ii];
myrinfo = rinfo+ii;
mydegrees = myrinfo->degrees;
if (me == from) {
INC_DEC(myrinfo->ed, myrinfo->id, adjwgt[j]);
}
else {
if (me == to) {
INC_DEC(myrinfo->id, myrinfo->ed, adjwgt[j]);
}
}
/* Remove contribution of the ed from 'from' */
if (me != from) {
for (k=0; k<myrinfo->ndegrees; k++) {
if (mydegrees[k].edge == from) {
if (mydegrees[k].ewgt == adjwgt[j]) {
myrinfo->ndegrees--;
mydegrees[k].edge = mydegrees[myrinfo->ndegrees].edge;
mydegrees[k].ewgt = mydegrees[myrinfo->ndegrees].ewgt;
}
else
mydegrees[k].ewgt -= adjwgt[j];
break;
}
}
}
/* Add contribution of the ed to 'to' */
if (me != to) {
for (k=0; k<myrinfo->ndegrees; k++) {
if (mydegrees[k].edge == to) {
mydegrees[k].ewgt += adjwgt[j];
break;
}
}
if (k == myrinfo->ndegrees) {
mydegrees[myrinfo->ndegrees].edge = to;
mydegrees[myrinfo->ndegrees++].ewgt = adjwgt[j];
}
}
}
nmoves++;
}
}
if (graph->mincut == oldcut)
break;
}
GKfree((void **)&minwgt, (void **)&maxwgt, (void **)&cand, LTERM);
return;
}
/*************************************************************************
* This function balances two partitions by moving the highest gain
* (including negative gain) vertices to the other domain.
* It is used only when tha unbalance is due to non contigous
* subdomains. That is, the are no boundary vertices.
* It moves vertices from the domain that is overweight to the one that
* is underweight.
**************************************************************************/
void Mc_Serial_Init2WayBalance(GraphType *graph, float *tpwgts)
{
int i, ii, j, k;
int kwgt, nvtxs, nbnd, ncon, nswaps, from, to, cnum, tmp;
idxtype *xadj, *adjncy, *adjwgt, *where, *id, *ed, *bndptr, *bndind;
idxtype *qnum;
float *nvwgt, *npwgts;
FPQueueType parts[MAXNCON][2];
int higain, oldgain, mincut;
KeyValueType *cand;
nvtxs = graph->nvtxs;
ncon = graph->ncon;
xadj = graph->xadj;
adjncy = graph->adjncy;
nvwgt = graph->nvwgt;
adjwgt = graph->adjwgt;
where = graph->where;
id = graph->sendind;
ed = graph->recvind;
npwgts = graph->gnpwgts;
bndptr = graph->sendptr;
bndind = graph->recvptr;
qnum = idxmalloc(nvtxs, "qnum");
cand = (KeyValueType *)GKmalloc(nvtxs*sizeof(KeyValueType), "cand");
/* This is called for initial partitioning so we know from where to pick nodes */
from = 1;
to = (from+1)%2;
for (i=0; i<ncon; i++) {
FPQueueInit(&parts[i][0], nvtxs);
FPQueueInit(&parts[i][1], nvtxs);
}
/* Compute the queues in which each vertex will be assigned to */
for (i=0; i<nvtxs; i++)
qnum[i] = samax(ncon, nvwgt+i*ncon);
for (i=0; i<nvtxs; i++) {
cand[i].key = id[i]-ed[i];
cand[i].val = i;
}
ikeysort(nvtxs, cand);
/* Insert the nodes of the proper partition in the appropriate priority queue */
for (ii=0; ii<nvtxs; ii++) {
i = cand[ii].val;
if (where[i] == from) {
if (ed[i] > 0)
FPQueueInsert(&parts[qnum[i]][0], i, (float)(ed[i]-id[i]));
else
FPQueueInsert(&parts[qnum[i]][1], i, (float)(ed[i]-id[i]));
}
}
mincut = graph->mincut;
nbnd = graph->gnvtxs;
for (nswaps=0; nswaps<nvtxs; nswaps++) {
if (Serial_AreAnyVwgtsBelow(ncon, 1.0, npwgts+from*ncon, 0.0, nvwgt, tpwgts+from*ncon))
break;
if ((cnum = Serial_SelectQueueOneWay(ncon, npwgts, tpwgts, from, parts)) == -1)
break;
if ((higain = FPQueueGetMax(&parts[cnum][0])) == -1)
higain = FPQueueGetMax(&parts[cnum][1]);
mincut -= (ed[higain]-id[higain]);
saxpy2(ncon, 1.0, nvwgt+higain*ncon, 1, npwgts+to*ncon, 1);
saxpy2(ncon, -1.0, nvwgt+higain*ncon, 1, npwgts+from*ncon, 1);
where[higain] = to;
/**************************************************************
* Update the id[i]/ed[i] values of the affected nodes
***************************************************************/
SWAP(id[higain], ed[higain], tmp);
if (ed[higain] == 0 && bndptr[higain] != -1 && xadj[higain] < xadj[higain+1])
BNDDelete(nbnd, bndind, bndptr, higain);
if (ed[higain] > 0 && bndptr[higain] == -1)
BNDInsert(nbnd, bndind, bndptr, higain);
for (j=xadj[higain]; j<xadj[higain+1]; j++) {
k = adjncy[j];
oldgain = ed[k]-id[k];
kwgt = (to == where[k] ? adjwgt[j] : -adjwgt[j]);
INC_DEC(id[k], ed[k], kwgt);
/* Update the queue position */
if (where[k] == from) {
if (ed[k] > 0 && bndptr[k] == -1) { /* It moves in boundary */
FPQueueDelete(&parts[qnum[k]][1], k);
FPQueueInsert(&parts[qnum[k]][0], k, (float)(ed[k]-id[k]));
}
else { /* It must be in the boundary already */
FPQueueUpdate(&parts[qnum[k]][0], k, (float)(oldgain), (float)(ed[k]-id[k]));
}
}
/* Update its boundary information */
if (ed[k] == 0 && bndptr[k] != -1)
BNDDelete(nbnd, bndind, bndptr, k);
else if (ed[k] > 0 && bndptr[k] == -1)
BNDInsert(nbnd, bndind, bndptr, k);
}
}
graph->mincut = mincut;
graph->gnvtxs = nbnd;
for (i=0; i<ncon; i++) {
FPQueueFree(&parts[i][0]);
FPQueueFree(&parts[i][1]);
}
GKfree((void **)&cand, (void **)&qnum, LTERM);
}
/*************************************************************************
* This function computes the assignment using the the objective the
* minimization of the total volume of data that needs to move
**************************************************************************/
void ParallelTotalVReMap(CtrlType *ctrl, idxtype *lpwgts, idxtype *map,
WorkSpaceType *wspace, int npasses, int ncon)
{
int i, ii, j, k, nparts, mype;
int pass, maxipwgt, nmapped, oldwgt, newwgt, done;
idxtype *rowmap, *mylpwgts;
KeyValueType *recv, send;
int nsaved, gnsaved;
mype = ctrl->mype;
nparts = ctrl->nparts;
recv = (KeyValueType *)GKmalloc(sizeof(KeyValueType)*nparts, "remap: recv");
mylpwgts = idxmalloc(nparts, "mylpwgts");
done = nmapped = 0;
idxset(nparts, -1, map);
rowmap = idxset(nparts, -1, wspace->pv3);
idxcopy(nparts, lpwgts, mylpwgts);
for (pass=0; pass<npasses; pass++) {
maxipwgt = idxamax(nparts, mylpwgts);
if (mylpwgts[maxipwgt] > 0 && !done) {
send.key = -mylpwgts[maxipwgt];
send.val = mype*nparts+maxipwgt;
}
else {
send.key = 0;
send.val = -1;
}
/* each processor sends its selection */
MPI_Allgather((void *)&send, 2, IDX_DATATYPE, (void *)recv, 2, IDX_DATATYPE, ctrl->comm);
ikeysort(nparts, recv);
if (recv[0].key == 0)
break;
/* now make as many assignments as possible */
for (ii=0; ii<nparts; ii++) {
i = recv[ii].val;
if (i == -1)
continue;
j = i % nparts;
k = i / nparts;
if (map[j] == -1 && rowmap[k] == -1 && SimilarTpwgts(ctrl->tpwgts, ncon, j, k)) {
map[j] = k;
rowmap[k] = j;
nmapped++;
mylpwgts[j] = 0;
if (mype == k)
done = 1;
}
if (nmapped == nparts)
break;
}
if (nmapped == nparts)
break;
}
/* Map unmapped partitions */
if (nmapped < nparts) {
for (i=j=0; j<nparts && nmapped<nparts; j++) {
if (map[j] == -1) {
for (; i<nparts; i++) {
if (rowmap[i] == -1 && SimilarTpwgts(ctrl->tpwgts, ncon, i, j)) {
map[j] = i;
rowmap[i] = j;
nmapped++;
break;
}
}
}
}
}
/* check to see if remapping fails (due to dis-similar tpwgts) */
/* if remapping fails, revert to original mapping */
if (nmapped < nparts) {
for (i=0; i<nparts; i++)
map[i] = i;
IFSET(ctrl->dbglvl, DBG_REMAP, rprintf(ctrl, "Savings from parallel remapping: %0\n"));
}
else {
/* check for a savings */
oldwgt = lpwgts[mype];
newwgt = lpwgts[rowmap[mype]];
nsaved = newwgt - oldwgt;
gnsaved = GlobalSESum(ctrl, nsaved);
/* undo everything if we don't see a savings */
if (gnsaved <= 0) {
for (i=0; i<nparts; i++)
map[i] = i;
}
IFSET(ctrl->dbglvl, DBG_REMAP, rprintf(ctrl, "Savings from parallel remapping: %d\n", amax(0,gnsaved)));
}
GKfree((void **)&recv, (void **)&mylpwgts, LTERM);
}
/*************************************************************************
* This function tests the repeated shmem_put
**************************************************************************/
void SetUp(CtrlType *ctrl, GraphType *graph, WorkSpaceType *wspace)
{
int i, j, k, islocal, penum, gnvtxs, nvtxs, nlocal, firstvtx, lastvtx, nsend, nrecv, nnbrs, nadj;
int npes=ctrl->npes, mype=ctrl->mype;
idxtype *vtxdist, *xadj, *adjncy;
idxtype *peind, *recvptr, *recvind, *sendptr, *sendind;
idxtype *receive, *pemap, *imap, *lperm;
idxtype *pexadj, *peadjncy, *peadjloc, *startsind;
KeyValueType *recvrequests, *sendrequests, *adjpairs;
IFSET(ctrl->dbglvl, DBG_TIME, MPI_Barrier(ctrl->comm));
IFSET(ctrl->dbglvl, DBG_TIME, starttimer(ctrl->SetupTmr));
gnvtxs = graph->gnvtxs;
nvtxs = graph->nvtxs;
vtxdist = graph->vtxdist;
xadj = graph->xadj;
adjncy = graph->adjncy;
firstvtx = vtxdist[mype];
lastvtx = vtxdist[mype+1];
pemap = wspace->pv1;
idxset(npes, -1, pemap);
lperm = graph->lperm = idxmalloc(nvtxs, "SetUp: graph->lperm");
for (i=0; i<nvtxs; i++)
lperm[i] = i;
/*************************************************************
* Determine what you need to receive
*************************************************************/
receive = wspace->indices; /* Use the large global received array for now */
adjpairs = wspace->pairs;
for (nlocal = nadj = i = 0; i<nvtxs; i++) {
islocal = 1;
for (j=xadj[i]; j<xadj[i+1]; j++) {
k = adjncy[j];
if (k >= firstvtx && k < lastvtx) {
adjncy[j] = k-firstvtx;
continue; /* local vertex */
}
adjpairs[nadj].key = k;
adjpairs[nadj++].val = j;
islocal = 0;
}
if (islocal) {
lperm[i] = lperm[nlocal];
lperm[nlocal++] = i;
}
}
/* Take care the received part now */
ikeysort(nadj, adjpairs);
adjpairs[nadj].key = gnvtxs+1; /* Boundary condition */
for (nrecv=i=0; i<nadj; i++) {
adjncy[adjpairs[i].val] = nvtxs+nrecv;
if (adjpairs[i].key != adjpairs[i+1].key)
receive[nrecv++] = adjpairs[i].key;
}
/* Allocate space for the setup info attached to this level of the graph */
peind = graph->peind = idxmalloc(npes, "SetUp: peind");
recvptr = graph->recvptr = idxmalloc(npes+1, "SetUp: recvptr");
recvind = graph->recvind = idxmalloc(nrecv, "SetUp: recvind");
/* Take care of the received portion */
idxcopy(nrecv, receive, recvind); /* Copy the vertices to be received into recvind */
i = nnbrs = recvptr[0] = 0;
for (penum=0; penum<npes; penum++) {
for (j=i; j<nrecv; j++) {
if (recvind[j] >= vtxdist[penum+1])
break;
}
if (j > i) {
peind[nnbrs] = penum;
recvptr[++nnbrs] = j;
i = j;
}
}
/*************************************************************
* Determine what you need to send
*************************************************************/
/* Tell the other processors what they need to send you */
recvrequests = wspace->pepairs1;
sendrequests = wspace->pepairs2;
for (i=0; i<npes; i++)
recvrequests[i].key = 0;
for (i=0; i<nnbrs; i++) {
recvrequests[peind[i]].key = recvptr[i+1]-recvptr[i];
recvrequests[peind[i]].val = nvtxs+recvptr[i];
}
MPI_Alltoall((void *)recvrequests, 2, IDX_DATATYPE, (void *)sendrequests, 2, IDX_DATATYPE, ctrl->comm);
sendptr = graph->sendptr = idxmalloc(npes+1, "SetUp: sendptr");
startsind = wspace->pv2;
for (j=i=0; i<npes; i++) {
if (sendrequests[i].key > 0) {
sendptr[j] = sendrequests[i].key;
startsind[j] = sendrequests[i].val;
j++;
}
}
ASSERT(ctrl, nnbrs == j);
MAKECSR(i, j, sendptr);
nsend = sendptr[nnbrs];
sendind = graph->sendind = idxmalloc(nsend, "SetUp: sendind");
/* Issue the receives for sendind */
for (i=0; i<nnbrs; i++) {
MPI_Irecv((void *)(sendind+sendptr[i]), sendptr[i+1]-sendptr[i], IDX_DATATYPE,
peind[i], 1, ctrl->comm, ctrl->rreq+i);
}
/* Issue the sends. My recvind[penum] becomes penum's sendind[mype] */
for (i=0; i<nnbrs; i++) {
MPI_Isend((void *)(recvind+recvptr[i]), recvptr[i+1]-recvptr[i], IDX_DATATYPE,
peind[i], 1, ctrl->comm, ctrl->sreq+i);
}
MPI_Waitall(nnbrs, ctrl->rreq, ctrl->statuses);
MPI_Waitall(nnbrs, ctrl->sreq, ctrl->statuses);
/* Create the peadjncy data structure for sparse boundary exchanges */
pexadj = graph->pexadj = idxsmalloc(nvtxs+1, 0, "SetUp: pexadj");
peadjncy = graph->peadjncy = idxmalloc(nsend, "SetUp: peadjncy");
peadjloc = graph->peadjloc = idxmalloc(nsend, "SetUp: peadjloc");
for (i=0; i<nsend; i++) {
ASSERTP(ctrl, sendind[i] >= firstvtx && sendind[i] < lastvtx, (ctrl, "%d %d %d\n", sendind[i], firstvtx, lastvtx));
pexadj[sendind[i]-firstvtx]++;
}
MAKECSR(i, nvtxs, pexadj);
for (i=0; i<nnbrs; i++) {
for (j=sendptr[i]; j<sendptr[i+1]; j++) {
k = pexadj[sendind[j]-firstvtx]++;
peadjncy[k] = i; /* peind[i] is the actual PE number */
peadjloc[k] = startsind[i]++;
}
}
ASSERT(ctrl, pexadj[nvtxs] == nsend);
for (i=nvtxs; i>0; i--)
pexadj[i] = pexadj[i-1];
pexadj[0] = 0;
graph->nnbrs = nnbrs;
graph->nrecv = nrecv;
graph->nsend = nsend;
graph->nlocal = nlocal;
/* Create the inverse map from ladjncy to adjncy */
imap = graph->imap = idxmalloc(nvtxs+nrecv, "SetUp: imap");
for (i=0; i<nvtxs; i++)
imap[i] = firstvtx+i;
for (i=0; i<nrecv; i++)
imap[nvtxs+i] = recvind[i];
/* Check if wspace->nlarge is large enough for nrecv and nsend */
if (wspace->nlarge < nrecv+nsend) {
free(wspace->indices);
free(wspace->pairs);
wspace->nlarge = nrecv+nsend;
wspace->indices = idxmalloc(wspace->nlarge, "SetUp: wspace->indices");
wspace->pairs = (KeyValueType *)GKmalloc(sizeof(KeyValueType)*wspace->nlarge, "SetUp: wspace->pairs");
}
IFSET(ctrl->dbglvl, DBG_TIME, stoptimer(ctrl->SetupTmr));
#ifdef DEBUG_SETUPINFO
rprintf(ctrl, "[%5d %5d] \tl:[%5d %5d] \ts:[%5d, %5d] \tr:[%5d, %5d]\n",
GlobalSEMin(ctrl, nvtxs), GlobalSEMax(ctrl, nvtxs),
GlobalSEMin(ctrl, nlocal), GlobalSEMax(ctrl, nlocal),
GlobalSEMin(ctrl, nsend), GlobalSEMax(ctrl, nsend),
GlobalSEMin(ctrl, nrecv), GlobalSEMax(ctrl, nrecv));
PrintSetUpInfo(ctrl, graph);
#endif
}
/*************************************************************************
* This function converts a mesh into a dual graph
**************************************************************************/
void ParMETIS_V3_Mesh2Dual(idxtype *elmdist, idxtype *eptr, idxtype *eind,
int *numflag, int *ncommonnodes, idxtype **xadj,
idxtype **adjncy, MPI_Comm *comm)
{
int i, j, jj, k, kk, m;
int npes, mype, pe, count, mask, pass;
int nelms, lnns, my_nns, node;
int firstelm, firstnode, lnode, nrecv, nsend;
int *scounts, *rcounts, *sdispl, *rdispl;
idxtype *nodedist, *nmap, *auxarray;
idxtype *gnptr, *gnind, *nptr, *nind, *myxadj, *myadjncy = NULL;
idxtype *sbuffer, *rbuffer, *htable;
KeyValueType *nodelist, *recvbuffer;
idxtype ind[200], wgt[200];
int gmaxnode, gminnode;
CtrlType ctrl;
SetUpCtrl(&ctrl, -1, 0, *comm);
npes = ctrl.npes;
mype = ctrl.mype;
nelms = elmdist[mype+1]-elmdist[mype];
if (*numflag == 1)
ChangeNumberingMesh2(elmdist, eptr, eind, NULL, NULL, NULL, npes, mype, 1);
mask = (1<<11)-1;
/*****************************/
/* Determine number of nodes */
/*****************************/
gminnode = GlobalSEMin(&ctrl, eind[idxamin(eptr[nelms], eind)]);
for (i=0; i<eptr[nelms]; i++)
eind[i] -= gminnode;
gmaxnode = GlobalSEMax(&ctrl, eind[idxamax(eptr[nelms], eind)]);
/**************************/
/* Check for input errors */
/**************************/
ASSERTS(nelms > 0);
/* construct node distribution array */
nodedist = idxsmalloc(npes+1, 0, "nodedist");
for (nodedist[0]=0, i=0,j=gmaxnode+1; i<npes; i++) {
k = j/(npes-i);
nodedist[i+1] = nodedist[i]+k;
j -= k;
}
my_nns = nodedist[mype+1]-nodedist[mype];
firstnode = nodedist[mype];
nodelist = (KeyValueType *)GKmalloc(eptr[nelms]*sizeof(KeyValueType), "nodelist");
auxarray = idxmalloc(eptr[nelms], "auxarray");
htable = idxsmalloc(amax(my_nns, mask+1), -1, "htable");
scounts = imalloc(4*npes+2, "scounts");
rcounts = scounts+npes;
sdispl = scounts+2*npes;
rdispl = scounts+3*npes+1;
/*********************************************/
/* first find a local numbering of the nodes */
/*********************************************/
for (i=0; i<nelms; i++) {
for (j=eptr[i]; j<eptr[i+1]; j++) {
nodelist[j].key = eind[j];
nodelist[j].val = j;
auxarray[j] = i; /* remember the local element ID that uses this node */
}
}
ikeysort(eptr[nelms], nodelist);
for (count=1, i=1; i<eptr[nelms]; i++) {
if (nodelist[i].key > nodelist[i-1].key)
count++;
}
lnns = count;
nmap = idxmalloc(lnns, "nmap");
/* renumber the nodes of the elements array */
count = 1;
nmap[0] = nodelist[0].key;
eind[nodelist[0].val] = 0;
nodelist[0].val = auxarray[nodelist[0].val]; /* Store the local element ID */
for (i=1; i<eptr[nelms]; i++) {
if (nodelist[i].key > nodelist[i-1].key) {
nmap[count] = nodelist[i].key;
count++;
}
eind[nodelist[i].val] = count-1;
nodelist[i].val = auxarray[nodelist[i].val]; /* Store the local element ID */
}
MPI_Barrier(*comm);
/**********************************************************/
/* perform comms necessary to construct node-element list */
/**********************************************************/
iset(npes, 0, scounts);
for (pe=i=0; i<eptr[nelms]; i++) {
while (nodelist[i].key >= nodedist[pe+1])
pe++;
scounts[pe] += 2;
}
ASSERTS(pe < npes);
MPI_Alltoall((void *)scounts, 1, MPI_INT, (void *)rcounts, 1, MPI_INT, *comm);
icopy(npes, scounts, sdispl);
MAKECSR(i, npes, sdispl);
icopy(npes, rcounts, rdispl);
MAKECSR(i, npes, rdispl);
ASSERTS(sdispl[npes] == eptr[nelms]*2);
nrecv = rdispl[npes]/2;
recvbuffer = (KeyValueType *)GKmalloc(amax(1, nrecv)*sizeof(KeyValueType), "recvbuffer");
MPI_Alltoallv((void *)nodelist, scounts, sdispl, IDX_DATATYPE, (void *)recvbuffer,
rcounts, rdispl, IDX_DATATYPE, *comm);
/**************************************/
/* construct global node-element list */
/**************************************/
gnptr = idxsmalloc(my_nns+1, 0, "gnptr");
for (i=0; i<npes; i++) {
for (j=rdispl[i]/2; j<rdispl[i+1]/2; j++) {
lnode = recvbuffer[j].key-firstnode;
ASSERTS(lnode >= 0 && lnode < my_nns)
gnptr[lnode]++;
}
}
MAKECSR(i, my_nns, gnptr);
gnind = idxmalloc(amax(1, gnptr[my_nns]), "gnind");
for (pe=0; pe<npes; pe++) {
firstelm = elmdist[pe];
for (j=rdispl[pe]/2; j<rdispl[pe+1]/2; j++) {
lnode = recvbuffer[j].key-firstnode;
gnind[gnptr[lnode]++] = recvbuffer[j].val+firstelm;
}
}
SHIFTCSR(i, my_nns, gnptr);
/*********************************************************/
/* send the node-element info to the relevant processors */
/*********************************************************/
iset(npes, 0, scounts);
/* use a hash table to ensure that each node is sent to a proc only once */
for (pe=0; pe<npes; pe++) {
for (j=rdispl[pe]/2; j<rdispl[pe+1]/2; j++) {
lnode = recvbuffer[j].key-firstnode;
if (htable[lnode] == -1) {
scounts[pe] += gnptr[lnode+1]-gnptr[lnode];
htable[lnode] = 1;
}
}
/* now reset the hash table */
for (j=rdispl[pe]/2; j<rdispl[pe+1]/2; j++) {
lnode = recvbuffer[j].key-firstnode;
htable[lnode] = -1;
}
}
MPI_Alltoall((void *)scounts, 1, MPI_INT, (void *)rcounts, 1, MPI_INT, *comm);
icopy(npes, scounts, sdispl);
MAKECSR(i, npes, sdispl);
/* create the send buffer */
nsend = sdispl[npes];
sbuffer = (idxtype *)realloc(nodelist, sizeof(idxtype)*amax(1, nsend));
count = 0;
for (pe=0; pe<npes; pe++) {
for (j=rdispl[pe]/2; j<rdispl[pe+1]/2; j++) {
lnode = recvbuffer[j].key-firstnode;
if (htable[lnode] == -1) {
for (k=gnptr[lnode]; k<gnptr[lnode+1]; k++) {
if (k == gnptr[lnode])
sbuffer[count++] = -1*(gnind[k]+1);
else
sbuffer[count++] = gnind[k];
}
htable[lnode] = 1;
}
}
ASSERTS(count == sdispl[pe+1]);
/* now reset the hash table */
for (j=rdispl[pe]/2; j<rdispl[pe+1]/2; j++) {
lnode = recvbuffer[j].key-firstnode;
htable[lnode] = -1;
}
}
icopy(npes, rcounts, rdispl);
MAKECSR(i, npes, rdispl);
nrecv = rdispl[npes];
rbuffer = (idxtype *)realloc(recvbuffer, sizeof(idxtype)*amax(1, nrecv));
MPI_Alltoallv((void *)sbuffer, scounts, sdispl, IDX_DATATYPE, (void *)rbuffer,
rcounts, rdispl, IDX_DATATYPE, *comm);
k = -1;
nptr = idxsmalloc(lnns+1, 0, "nptr");
nind = rbuffer;
for (pe=0; pe<npes; pe++) {
for (j=rdispl[pe]; j<rdispl[pe+1]; j++) {
if (nind[j] < 0) {
k++;
nind[j] = (-1*nind[j])-1;
}
nptr[k]++;
}
}
MAKECSR(i, lnns, nptr);
ASSERTS(k+1 == lnns);
ASSERTS(nptr[lnns] == nrecv)
myxadj = *xadj = idxsmalloc(nelms+1, 0, "xadj");
idxset(mask+1, -1, htable);
firstelm = elmdist[mype];
/* Two passes -- in first pass, simply find out the memory requirements */
for (pass=0; pass<2; pass++) {
for (i=0; i<nelms; i++) {
for (count=0, j=eptr[i]; j<eptr[i+1]; j++) {
node = eind[j];
for (k=nptr[node]; k<nptr[node+1]; k++) {
if ((kk=nind[k]) == firstelm+i)
continue;
m = htable[(kk&mask)];
if (m == -1) {
ind[count] = kk;
wgt[count] = 1;
htable[(kk&mask)] = count++;
}
else {
if (ind[m] == kk) {
wgt[m]++;
}
else {
for (jj=0; jj<count; jj++) {
if (ind[jj] == kk) {
wgt[jj]++;
break;
}
}
if (jj == count) {
ind[count] = kk;
wgt[count++] = 1;
}
}
}
}
}
for (j=0; j<count; j++) {
htable[(ind[j]&mask)] = -1;
if (wgt[j] >= *ncommonnodes) {
if (pass == 0)
myxadj[i]++;
else
myadjncy[myxadj[i]++] = ind[j];
}
}
}
if (pass == 0) {
MAKECSR(i, nelms, myxadj);
myadjncy = *adjncy = idxmalloc(myxadj[nelms], "adjncy");
}
else {
SHIFTCSR(i, nelms, myxadj);
}
}
/*****************************************/
/* correctly renumber the elements array */
/*****************************************/
for (i=0; i<eptr[nelms]; i++)
eind[i] = nmap[eind[i]] + gminnode;
if (*numflag == 1)
ChangeNumberingMesh2(elmdist, eptr, eind, myxadj, myadjncy, NULL, npes, mype, 0);
/* do not free nodelist, recvbuffer, rbuffer */
GKfree((void **)&scounts, (void **)&nodedist, (void **)&nmap, (void **)&sbuffer,
(void **)&htable, (void **)&nptr, (void **)&nind, (void **)&gnptr,
(void **)&gnind, (void **)&auxarray, LTERM);
FreeCtrl(&ctrl);
return;
}
/*************************************************************************
* This function finds a matching
**************************************************************************/
void Moc_GlobalMatch_Balance(CtrlType *ctrl, GraphType *graph, WorkSpaceType *wspace)
{
int h, i, ii, j, k;
int nnbrs, nvtxs, ncon, cnvtxs, firstvtx, lastvtx, maxi, maxidx, nkept;
int otherlastvtx, nrequests, nchanged, pass, nmatched, wside;
idxtype *xadj, *ladjncy, *adjwgt, *vtxdist, *home, *myhome, *shome, *rhome;
idxtype *match, *rmatch, *smatch;
idxtype *peind, *sendptr, *recvptr;
idxtype *perm, *iperm, *nperm, *changed;
floattype *nvwgt, maxnvwgt;
int *nreqs_pe;
KeyValueType *match_requests, *match_granted, *pe_requests;
maxnvwgt = 1.0/((floattype)(ctrl->nparts)*MAXNVWGT_FACTOR);
graph->match_type = MATCH_GLOBAL;
IFSET(ctrl->dbglvl, DBG_TIME, MPI_Barrier(ctrl->comm));
IFSET(ctrl->dbglvl, DBG_TIME, starttimer(ctrl->MatchTmr));
nvtxs = graph->nvtxs;
ncon = graph->ncon;
xadj = graph->xadj;
ladjncy = graph->adjncy;
adjwgt = graph->adjwgt;
home = graph->home;
nvwgt = graph->nvwgt;
vtxdist = graph->vtxdist;
firstvtx = vtxdist[ctrl->mype];
lastvtx = vtxdist[ctrl->mype+1];
match = graph->match = idxsmalloc(nvtxs+graph->nrecv, UNMATCHED, "HEM_Match: match");
myhome = idxsmalloc(nvtxs+graph->nrecv, UNMATCHED, "HEM_Match: myhome");
/*------------------------------------------------------------
/ Send/Receive the home information of interface vertices
/------------------------------------------------------------*/
if (ctrl->partType == ADAPTIVE_PARTITION || ctrl->partType == REFINE_PARTITION) {
idxcopy(nvtxs, home, myhome);
shome = wspace->indices;
rhome = myhome + nvtxs;
CommInterfaceData(ctrl, graph, myhome, shome, rhome);
}
nnbrs = graph->nnbrs;
peind = graph->peind;
sendptr = graph->sendptr;
recvptr = graph->recvptr;
/* Use wspace->indices as the tmp space for matching info of the boundary
* vertices that are sent and received */
rmatch = match + nvtxs;
smatch = wspace->indices;
changed = smatch+graph->nsend;
/* Use wspace->indices as the tmp space for match requests of the boundary
* vertices that are sent and received */
match_requests = wspace->pairs;
match_granted = match_requests + graph->nsend;
nreqs_pe = ismalloc(nnbrs, 0, "Match_HEM: nreqs_pe");
nkept = graph->gnvtxs/ctrl->npes - nvtxs;
perm = (idxtype *)wspace->degrees;
iperm = perm + nvtxs;
FastRandomPermute(nvtxs, perm, 1);
for (i=0; i<nvtxs; i++)
iperm[perm[i]] = i;
nperm = iperm + nvtxs;
for (i=0; i<nnbrs; i++)
nperm[i] = i;
/*************************************************************
* Go now and find a matching by doing multiple iterations
*************************************************************/
/* First nullify the heavy vertices */
for (nchanged=i=0; i<nvtxs; i++) {
for (h=0; h<ncon; h++)
if (nvwgt[i*ncon+h] > maxnvwgt) {
break;
}
if (h != ncon) {
match[i] = TOO_HEAVY;
nchanged++;
}
}
if (GlobalSESum(ctrl, nchanged) > 0) {
IFSET(ctrl->dbglvl, DBG_PROGRESS,
rprintf(ctrl, "We found %d heavy vertices!\n", GlobalSESum(ctrl, nchanged)));
CommInterfaceData(ctrl, graph, match, smatch, rmatch);
}
for (nmatched=pass=0; pass<NMATCH_PASSES; pass++) {
wside = (graph->level+pass)%2;
nchanged = nrequests = 0;
for (ii=nmatched; ii<nvtxs; ii++) {
i = perm[ii];
if (match[i] == UNMATCHED) { /* Unmatched */
maxidx = i;
maxi = -1;
/* Find a heavy-edge matching */
for (j=xadj[i]; j<xadj[i+1]; j++) {
k = ladjncy[j];
if (match[k] == UNMATCHED &&
myhome[k] == myhome[i] &&
(maxi == -1 ||
adjwgt[maxi] < adjwgt[j] ||
(maxidx < nvtxs &&
k < nvtxs &&
adjwgt[maxi] == adjwgt[j] &&
BetterVBalance(ncon,nvwgt+i*ncon,nvwgt+maxidx*ncon,nvwgt+k*ncon) >= 0))) {
maxi = j;
maxidx = k;
}
}
if (maxi != -1) {
k = ladjncy[maxi];
if (k < nvtxs) { /* Take care the local vertices first */
/* Here we give preference the local matching by granting it right away */
if (i <= k) {
match[i] = firstvtx+k + KEEP_BIT;
match[k] = firstvtx+i;
}
else {
match[i] = firstvtx+k;
match[k] = firstvtx+i + KEEP_BIT;
}
changed[nchanged++] = i;
changed[nchanged++] = k;
}
else { /* Take care any remote boundary vertices */
match[k] = MAYBE_MATCHED;
/* Alternate among which vertices will issue the requests */
if ((wside ==0 && firstvtx+i < graph->imap[k]) || (wside == 1 && firstvtx+i > graph->imap[k])) {
match[i] = MAYBE_MATCHED;
match_requests[nrequests].key = graph->imap[k];
match_requests[nrequests].val = firstvtx+i;
nrequests++;
}
}
}
}
}
#ifdef DEBUG_MATCH
PrintVector2(ctrl, nvtxs, firstvtx, match, "Match1");
myprintf(ctrl, "[c: %2d] Nlocal: %d, Nrequests: %d\n", c, nlocal, nrequests);
#endif
/***********************************************************
* Exchange the match_requests, requests for me are stored in
* match_granted
************************************************************/
/* Issue the receives first. Note that from each PE can receive a maximum
of the interface node that it needs to send it in the case of a mat-vec */
for (i=0; i<nnbrs; i++) {
MPI_Irecv((void *)(match_granted+recvptr[i]), 2*(recvptr[i+1]-recvptr[i]), IDX_DATATYPE,
peind[i], 1, ctrl->comm, ctrl->rreq+i);
}
/* Issue the sends next. This needs some work */
ikeysort(nrequests, match_requests);
for (j=i=0; i<nnbrs; i++) {
otherlastvtx = vtxdist[peind[i]+1];
for (k=j; k<nrequests && match_requests[k].key < otherlastvtx; k++);
MPI_Isend((void *)(match_requests+j), 2*(k-j), IDX_DATATYPE, peind[i], 1, ctrl->comm, ctrl->sreq+i);
j = k;
}
/* OK, now get into the loop waiting for the operations to finish */
MPI_Waitall(nnbrs, ctrl->rreq, ctrl->statuses);
for (i=0; i<nnbrs; i++) {
MPI_Get_count(ctrl->statuses+i, IDX_DATATYPE, nreqs_pe+i);
nreqs_pe[i] = nreqs_pe[i]/2; /* Adjust for pairs of IDX_DATATYPE */
}
MPI_Waitall(nnbrs, ctrl->sreq, ctrl->statuses);
/***********************************************************
* Now, go and service the requests that you received in
* match_granted
************************************************************/
RandomPermute(nnbrs, nperm, 0);
for (ii=0; ii<nnbrs; ii++) {
i = nperm[ii];
pe_requests = match_granted+recvptr[i];
for (j=0; j<nreqs_pe[i]; j++) {
k = pe_requests[j].key;
ASSERTP(ctrl, k >= firstvtx && k < lastvtx, (ctrl, "%d %d %d %d %d\n", firstvtx, lastvtx, k, j, peind[i]));
/* myprintf(ctrl, "Requesting a match %d %d\n", pe_requests[j].key, pe_requests[j].val); */
if (match[k-firstvtx] == UNMATCHED) { /* Bingo, lets grant this request */
changed[nchanged++] = k-firstvtx;
if (nkept >= 0) { /* Flip a coin for who gets it */
match[k-firstvtx] = pe_requests[j].val + KEEP_BIT;
nkept--;
}
else {
match[k-firstvtx] = pe_requests[j].val;
pe_requests[j].key += KEEP_BIT;
nkept++;
}
/* myprintf(ctrl, "Request from pe:%d (%d %d) granted!\n", peind[i], pe_requests[j].val, pe_requests[j].key); */
}
else { /* We are not granting the request */
/* myprintf(ctrl, "Request from pe:%d (%d %d) not granted!\n", peind[i], pe_requests[j].val, pe_requests[j].key); */
pe_requests[j].key = UNMATCHED;
}
}
}
/***********************************************************
* Exchange the match_granted information. It is stored in
* match_requests
************************************************************/
/* Issue the receives first. Note that from each PE can receive a maximum
of the interface node that it needs to send during the case of a mat-vec */
for (i=0; i<nnbrs; i++) {
MPI_Irecv((void *)(match_requests+sendptr[i]), 2*(sendptr[i+1]-sendptr[i]), IDX_DATATYPE,
peind[i], 1, ctrl->comm, ctrl->rreq+i);
}
/* Issue the sends next. */
for (i=0; i<nnbrs; i++) {
MPI_Isend((void *)(match_granted+recvptr[i]), 2*nreqs_pe[i], IDX_DATATYPE,
peind[i], 1, ctrl->comm, ctrl->sreq+i);
}
/* OK, now get into the loop waiting for the operations to finish */
MPI_Waitall(nnbrs, ctrl->rreq, ctrl->statuses);
for (i=0; i<nnbrs; i++) {
MPI_Get_count(ctrl->statuses+i, IDX_DATATYPE, nreqs_pe+i);
nreqs_pe[i] = nreqs_pe[i]/2; /* Adjust for pairs of IDX_DATATYPE */
}
MPI_Waitall(nnbrs, ctrl->sreq, ctrl->statuses);
/***********************************************************
* Now, go and through the match_requests and update local
* match information for the matchings that were granted.
************************************************************/
for (i=0; i<nnbrs; i++) {
pe_requests = match_requests+sendptr[i];
for (j=0; j<nreqs_pe[i]; j++) {
match[pe_requests[j].val-firstvtx] = pe_requests[j].key;
if (pe_requests[j].key != UNMATCHED)
changed[nchanged++] = pe_requests[j].val-firstvtx;
}
}
for (i=0; i<nchanged; i++) {
ii = iperm[changed[i]];
perm[ii] = perm[nmatched];
iperm[perm[nmatched]] = ii;
nmatched++;
}
CommChangedInterfaceData(ctrl, graph, nchanged, changed, match, match_requests, match_granted, wspace->pv4);
}
/* Traverse the vertices and those that were unmatched, match them with themselves */
cnvtxs = 0;
for (i=0; i<nvtxs; i++) {
if (match[i] == UNMATCHED || match[i] == TOO_HEAVY) {
match[i] = (firstvtx+i) + KEEP_BIT;
cnvtxs++;
}
else if (match[i] >= KEEP_BIT) { /* A matched vertex which I get to keep */
cnvtxs++;
}
}
if (ctrl->dbglvl&DBG_MATCHINFO) {
PrintVector2(ctrl, nvtxs, firstvtx, match, "Match");
myprintf(ctrl, "Cnvtxs: %d\n", cnvtxs);
rprintf(ctrl, "Done with matching...\n");
}
GKfree((void **)(&myhome), (void **)(&nreqs_pe), LTERM);
IFSET(ctrl->dbglvl, DBG_TIME, MPI_Barrier(ctrl->comm));
IFSET(ctrl->dbglvl, DBG_TIME, stoptimer(ctrl->MatchTmr));
IFSET(ctrl->dbglvl, DBG_TIME, starttimer(ctrl->ContractTmr));
Moc_Global_CreateCoarseGraph(ctrl, graph, wspace, cnvtxs);
IFSET(ctrl->dbglvl, DBG_TIME, MPI_Barrier(ctrl->comm));
IFSET(ctrl->dbglvl, DBG_TIME, stoptimer(ctrl->ContractTmr));
}
/*************************************************************************
* This function sorts a distributed list of KeyValueType in increasing
* order, and uses it to compute a partition. It uses samplesort.
**************************************************************************/
void PartSort(CtrlType *ctrl, GraphType *graph, KeyValueType *elmnts, WorkSpaceType *wspace)
{
int i, j, k, nvtxs, nrecv, npes=ctrl->npes, mype=ctrl->mype, firstvtx, lastvtx;
idxtype *scounts, *rcounts, *vtxdist, *perm;
KeyValueType *relmnts, *mypicks, *allpicks;
nvtxs = graph->nvtxs;
vtxdist = graph->vtxdist;
scounts = wspace->pv1;
rcounts = wspace->pv2;
/* Allocate memory for the splitters */
mypicks = (KeyValueType *)GKmalloc(sizeof(KeyValueType)*(npes+1), "ParSort: mypicks");
allpicks = (KeyValueType *)GKmalloc(sizeof(KeyValueType)*npes*npes, "ParSort: allpicks");
/* Sort the local elements */
ikeysort(nvtxs, elmnts);
/* Select the local npes-1 equally spaced elements */
for (i=1; i<npes; i++) {
mypicks[i-1].key = elmnts[i*(nvtxs/npes)].key;
mypicks[i-1].val = elmnts[i*(nvtxs/npes)].val;
}
/* PrintPairs(ctrl, npes-1, mypicks, "Mypicks"); */
/* Gather the picks to all the processors */
MPI_Allgather((void *)mypicks, 2*(npes-1), IDX_DATATYPE, (void *)allpicks, 2*(npes-1), IDX_DATATYPE, ctrl->comm);
/* PrintPairs(ctrl, npes*(npes-1), allpicks, "Allpicks"); */
/* Sort all the picks */
ikeyvalsort(npes*(npes-1), allpicks);
/* PrintPairs(ctrl, npes*(npes-1), allpicks, "Allpicks"); */
/* Select the final splitters. Set the boundaries to simplify coding */
for (i=1; i<npes; i++)
mypicks[i] = allpicks[i*(npes-1)];
mypicks[0].key = MIN_INT;
mypicks[npes].key = MAX_INT;
/* PrintPairs(ctrl, npes+1, mypicks, "Mypicks"); */
/* Compute the number of elements that belong to each bucket */
idxset(npes, 0, scounts);
for (j=i=0; i<nvtxs; i++) {
if (elmnts[i].key < mypicks[j+1].key || (elmnts[i].key == mypicks[j+1].key && elmnts[i].val < mypicks[j+1].val))
scounts[j]++;
else
scounts[++j]++;
}
MPI_Alltoall(scounts, 1, IDX_DATATYPE, rcounts, 1, IDX_DATATYPE, ctrl->comm);
/*
PrintVector(ctrl, npes, 0, scounts, "Scounts");
PrintVector(ctrl, npes, 0, rcounts, "Rcounts");
*/
/* Allocate memory for sorted elements and receive them */
MAKECSR(i, npes, scounts);
MAKECSR(i, npes, rcounts);
nrecv = rcounts[npes];
if (wspace->nlarge >= nrecv)
relmnts = (KeyValueType *)wspace->pairs;
else
relmnts = (KeyValueType *)GKmalloc(sizeof(KeyValueType)*nrecv, "ParSort: relmnts");
/* Issue the receives first */
for (i=0; i<npes; i++)
MPI_Irecv((void *)(relmnts+rcounts[i]), 2*(rcounts[i+1]-rcounts[i]), IDX_DATATYPE, i, 1, ctrl->comm, ctrl->rreq+i);
/* Issue the sends next */
for (i=0; i<npes; i++)
MPI_Isend((void *)(elmnts+scounts[i]), 2*(scounts[i+1]-scounts[i]), IDX_DATATYPE, i, 1, ctrl->comm, ctrl->sreq+i);
MPI_Waitall(npes, ctrl->rreq, ctrl->statuses);
MPI_Waitall(npes, ctrl->sreq, ctrl->statuses);
/* OK, now do the local sort of the relmnts. Use perm to keep track original order */
perm = idxmalloc(nrecv, "ParSort: perm");
for (i=0; i<nrecv; i++) {
perm[i] = relmnts[i].val;
relmnts[i].val = i;
}
ikeysort(nrecv, relmnts);
/* Compute what needs to be shifted */
MPI_Scan((void *)(&nrecv), (void *)(&lastvtx), 1, MPI_INT, MPI_SUM, ctrl->comm);
firstvtx = lastvtx-nrecv;
/*myprintf(ctrl, "first, last: %d %d\n", firstvtx, lastvtx); */
for (j=0, i=0; i<npes; i++) {
if (vtxdist[i+1] > firstvtx) { /* Found the first PE that is passed me */
if (vtxdist[i+1] >= lastvtx) {
/* myprintf(ctrl, "Shifting %d elements to processor %d\n", lastvtx-firstvtx, i); */
for (k=0; k<lastvtx-firstvtx; k++, j++)
relmnts[relmnts[j].val].key = i;
}
else {
/* myprintf(ctrl, "Shifting %d elements to processor %d\n", vtxdist[i+1]-firstvtx, i); */
for (k=0; k<vtxdist[i+1]-firstvtx; k++, j++)
relmnts[relmnts[j].val].key = i;
firstvtx = vtxdist[i+1];
}
}
if (vtxdist[i+1] >= lastvtx)
break;
}
/* Reverse the ordering on the relmnts[].val */
for (i=0; i<nrecv; i++) {
ASSERTP(ctrl, relmnts[i].key>=0 && relmnts[i].key<npes, (ctrl, "%d %d\n", i, relmnts[i].key));
relmnts[i].val = perm[i];
}
/* OK, now sent it back */
/* Issue the receives first */
for (i=0; i<npes; i++)
MPI_Irecv((void *)(elmnts+scounts[i]), 2*(scounts[i+1]-scounts[i]), IDX_DATATYPE, i, 1, ctrl->comm, ctrl->rreq+i);
/* Issue the sends next */
for (i=0; i<npes; i++)
MPI_Isend((void *)(relmnts+rcounts[i]), 2*(rcounts[i+1]-rcounts[i]), IDX_DATATYPE, i, 1, ctrl->comm, ctrl->sreq+i);
MPI_Waitall(npes, ctrl->rreq, ctrl->statuses);
MPI_Waitall(npes, ctrl->sreq, ctrl->statuses);
/* Construct a partition for the graph */
graph->where = idxmalloc(graph->nvtxs+graph->nrecv, "PartSort: graph->where");
firstvtx = vtxdist[mype];
for (i=0; i<nvtxs; i++) {
ASSERTP(ctrl, elmnts[i].key>=0 && elmnts[i].key<npes, (ctrl, "%d %d\n", i, elmnts[i].key));
ASSERTP(ctrl, elmnts[i].val>=vtxdist[mype] && elmnts[i].val<vtxdist[mype+1], (ctrl, "%d %d %d %d\n", i, vtxdist[mype], vtxdist[mype+1], elmnts[i].val));
graph->where[elmnts[i].val-firstvtx] = elmnts[i].key;
}
GKfree((void **)&mypicks, (void **)&allpicks, (void **)&perm, LTERM);
if (wspace->nlarge < nrecv)
free(relmnts);
}
