/*
---------------------------------------------------------------
purpose -- fill dvec[J] with the stack storage to solve for J
in a forward solve
created -- 97nov30, cca
---------------------------------------------------------------
*/
void
ETree_forwSolveProfile (
ETree *etree,
double dvec[]
) {
int I, J, maxstack, nDJ, nfront, nUJ, stack ;
int *bndwghts, *fch, *nodwghts, *sib ;
Tree *tree ;
/*
---------------
check the input
---------------
*/
if ( etree == NULL || dvec == NULL ) {
fprintf(stderr,
"\n fatal error in ETree_forwSolveProfile(%p,%p)"
"\n bad input\n", etree, dvec) ;
exit(-1) ;
}
tree = ETree_tree(etree) ;
nodwghts = ETree_nodwghts(etree) ;
bndwghts = ETree_bndwghts(etree) ;
nfront = ETree_nfront(etree) ;
fch = ETree_fch(etree) ;
sib = ETree_sib(etree) ;
/*
---------------------------------------------
loop over the nodes in a post-order traversal
---------------------------------------------
*/
maxstack = stack = 0 ;
for ( J = Tree_postOTfirst(tree) ;
J != -1 ;
J = Tree_postOTnext(tree, J) ) {
nDJ = nodwghts[J] ;
nUJ = bndwghts[J] ;
stack += nDJ + nUJ ;
dvec[J] = stack ;
if ( maxstack < stack ) {
maxstack = stack ;
}
#if MYDEBUG > 0
fprintf(stdout,
"\n working on front %d, nD = %d, nU = %d, stack = %d",
J, nDJ, nUJ, stack) ;
#endif
for ( I = fch[J] ; I != -1 ; I = sib[I] ) {
stack -= bndwghts[I] ;
}
stack -= nDJ ;
}
#if MYDEBUG >= 0
fprintf(stdout,
"\n forward solve : final stack = %d, max stack = %d",
stack, maxstack) ;
#endif
return ; }
/*--------------------------------------------------------------------*/
int
main ( int argc, char *argv[] )
/*
------------------------------------------------------
(1) read in an ETree object.
(2) read in an Graph object.
(3) find the optimal domain/schur complement partition
for a semi-implicit factorization
created -- 96oct03, cca
------------------------------------------------------
*/
{
char *inETreeFileName, *inGraphFileName, *outIVfileName ;
double alpha, nA21, nfent1, nfops1, nL11, nL22, nPhi, nV, t1, t2 ;
Graph *graph ;
int ii, inside, J, K, msglvl, nfind1, nfront, nJ, nleaves1,
nnode1, nvtx, rc, sizeJ, totalgain, vsize, v, w ;
int *adjJ, *compids, *nodwghts, *vadj, *vtxToFront, *vwghts ;
IV *compidsIV ;
IVL *symbfacIVL ;
ETree *etree ;
FILE *msgFile ;
Tree *tree ;
if ( argc != 7 ) {
fprintf(stdout,
"\n\n usage : %s msglvl msgFile inETreeFile inGraphFile alpha"
"\n outIVfile "
"\n msglvl -- message level"
"\n msgFile -- message file"
"\n inETreeFile -- input file, must be *.etreef or *.etreeb"
"\n inGraphFile -- input file, must be *.graphf or *.graphb"
"\n alpha -- weight parameter"
"\n alpha = 0 --> minimize storage"
"\n alpha = 1 --> minimize solve ops"
"\n outIVfile -- output file for oldToNew vector,"
"\n must be *.ivf or *.ivb"
"\n", argv[0]) ;
return(0) ;
}
msglvl = atoi(argv[1]) ;
if ( strcmp(argv[2], "stdout") == 0 ) {
msgFile = stdout ;
} else if ( (msgFile = fopen(argv[2], "a")) == NULL ) {
fprintf(stderr, "\n fatal error in %s"
"\n unable to open file %s\n",
argv[0], argv[2]) ;
return(-1) ;
}
inETreeFileName = argv[3] ;
inGraphFileName = argv[4] ;
alpha = atof(argv[5]) ;
outIVfileName = argv[6] ;
fprintf(msgFile,
"\n %s "
"\n msglvl -- %d"
"\n msgFile -- %s"
"\n inETreeFile -- %s"
"\n inGraphFile -- %s"
"\n alpha -- %f"
"\n outIVfile -- %s"
"\n",
argv[0], msglvl, argv[2],
inETreeFileName, inGraphFileName, alpha, outIVfileName) ;
fflush(msgFile) ;
/*
------------------------
read in the ETree object
------------------------
*/
if ( strcmp(inETreeFileName, "none") == 0 ) {
fprintf(msgFile, "\n no file to read from") ;
spoolesFatal();
}
etree = ETree_new() ;
MARKTIME(t1) ;
rc = ETree_readFromFile(etree, inETreeFileName) ;
MARKTIME(t2) ;
fprintf(msgFile, "\n CPU %9.5f : read in etree from file %s",
t2 - t1, inETreeFileName) ;
if ( rc != 1 ) {
fprintf(msgFile, "\n return value %d from ETree_readFromFile(%p,%s)",
rc, etree, inETreeFileName) ;
spoolesFatal();
}
ETree_leftJustify(etree) ;
fprintf(msgFile, "\n\n after reading ETree object from file %s",
inETreeFileName) ;
if ( msglvl > 2 ) {
ETree_writeForHumanEye(etree, msgFile) ;
} else {
ETree_writeStats(etree, msgFile) ;
}
fflush(msgFile) ;
nfront = ETree_nfront(etree) ;
tree = ETree_tree(etree) ;
nodwghts = ETree_nodwghts(etree) ;
vtxToFront = ETree_vtxToFront(etree) ;
/*
------------------------
read in the Graph object
------------------------
*/
if ( strcmp(inGraphFileName, "none") == 0 ) {
fprintf(msgFile, "\n no file to read from") ;
spoolesFatal();
}
graph = Graph_new() ;
MARKTIME(t1) ;
rc = Graph_readFromFile(graph, inGraphFileName) ;
nvtx = graph->nvtx ;
vwghts = graph->vwghts ;
MARKTIME(t2) ;
fprintf(msgFile, "\n CPU %9.5f : read in graph from file %s",
t2 - t1, inGraphFileName) ;
if ( rc != 1 ) {
fprintf(msgFile, "\n return value %d from Graph_readFromFile(%p,%s)",
rc, graph, inGraphFileName) ;
spoolesFatal();
}
fprintf(msgFile, "\n\n after reading Graph object from file %s",
inGraphFileName) ;
if ( msglvl > 2 ) {
Graph_writeForHumanEye(graph, msgFile) ;
} else {
Graph_writeStats(graph, msgFile) ;
}
fflush(msgFile) ;
/*
----------------------
compute the statistics
----------------------
*/
nnode1 = etree->tree->n ;
nfind1 = ETree_nFactorIndices(etree) ;
nfent1 = ETree_nFactorEntries(etree, SPOOLES_SYMMETRIC) ;
nfops1 = ETree_nFactorOps(etree, SPOOLES_REAL, SPOOLES_SYMMETRIC) ;
nleaves1 = Tree_nleaves(etree->tree) ;
fprintf(stdout, "\n root front %d has %d vertices",
etree->tree->root,
etree->nodwghtsIV->vec[etree->tree->root]) ;
/*
---------------------------------
create the symbolic factorization
---------------------------------
*/
symbfacIVL = SymbFac_initFromGraph(etree, graph) ;
if ( msglvl > 2 ) {
IVL_writeForHumanEye(symbfacIVL, msgFile) ;
} else {
IVL_writeStats(symbfacIVL, msgFile) ;
}
fflush(msgFile) ;
/*
--------------------------
find the optimal partition
--------------------------
*/
compidsIV = ETree_optPart(etree, graph, symbfacIVL, alpha,
&totalgain, msglvl, msgFile) ;
if ( msglvl > 2 ) {
IV_writeForHumanEye(compidsIV, msgFile) ;
} else {
IV_writeStats(compidsIV, msgFile) ;
}
fflush(msgFile) ;
compids = IV_entries(compidsIV) ;
/*
------------------------------------------------------
compute the number of vertices in the schur complement
------------------------------------------------------
*/
for ( J = 0, nPhi = nV = 0. ; J < nfront ; J++ ) {
if ( compids[J] == 0 ) {
nPhi += nodwghts[J] ;
}
nV += nodwghts[J] ;
}
/*
--------------------------------------------
compute the number of entries in L11 and L22
--------------------------------------------
*/
nL11 = nL22 = 0 ;
for ( J = Tree_postOTfirst(tree) ;
J != -1 ;
J = Tree_postOTnext(tree, J) ) {
nJ = nodwghts[J] ;
if ( msglvl > 3 ) {
fprintf(msgFile, "\n\n front %d, nJ = %d", J, nJ) ;
}
IVL_listAndSize(symbfacIVL, J, &sizeJ, &adjJ) ;
for ( ii = 0, inside = 0 ; ii < sizeJ ; ii++ ) {
w = adjJ[ii] ;
K = vtxToFront[w] ;
if ( msglvl > 3 ) {
fprintf(msgFile, "\n w = %d, K = %d", w, K) ;
}
if ( K > J && compids[K] == compids[J] ) {
inside += (vwghts == NULL) ? 1 : vwghts[w] ;
if ( msglvl > 3 ) {
fprintf(msgFile, ", inside") ;
}
}
}
if ( compids[J] != 0 ) {
if ( msglvl > 3 ) {
fprintf(msgFile, "\n inside = %d, adding %d to L11",
inside, nJ*nJ + 2*nJ*inside) ;
}
nL11 += (nJ*(nJ+1))/2 + nJ*inside ;
} else {
if ( msglvl > 3 ) {
fprintf(msgFile, "\n inside = %d, adding %d to L22",
inside, (nJ*(nJ+1))/2 + nJ*inside) ;
}
nL22 += (nJ*(nJ+1))/2 + nJ*inside ;
}
}
if ( msglvl > 0 ) {
fprintf(msgFile, "\n |L| = %.0f, |L11| = %.0f, |L22| = %.0f",
nfent1, nL11, nL22) ;
}
/*
------------------------------------
compute the number of entries in A21
------------------------------------
*/
nA21 = 0 ;
if ( vwghts != NULL ) {
for ( v = 0 ; v < nvtx ; v++ ) {
J = vtxToFront[v] ;
if ( compids[J] != 0 ) {
Graph_adjAndSize(graph, v, &vsize, &vadj) ;
for ( ii = 0 ; ii < vsize ; ii++ ) {
w = vadj[ii] ;
K = vtxToFront[w] ;
if ( compids[K] == 0 ) {
if ( msglvl > 3 ) {
fprintf(msgFile, "\n A21 : v = %d, w = %d", v, w) ;
}
nA21 += vwghts[v] * vwghts[w] ;
}
}
}
}
} else {
for ( v = 0 ; v < nvtx ; v++ ) {
J = vtxToFront[v] ;
if ( compids[J] != 0 ) {
Graph_adjAndSize(graph, v, &vsize, &vadj) ;
for ( ii = 0 ; ii < vsize ; ii++ ) {
w = vadj[ii] ;
K = vtxToFront[w] ;
if ( compids[K] == 0 ) {
if ( msglvl > 3 ) {
fprintf(msgFile, "\n A21 : v = %d, w = %d", v, w) ;
}
nA21++ ;
}
}
}
}
}
if ( msglvl > 0 ) {
fprintf(msgFile,
"\n |L| = %.0f, |L11| = %.0f, |L22| = %.0f, |A21| = %.0f",
nfent1, nL11, nL22, nA21) ;
fprintf(msgFile,
"\n storage: explicit = %.0f, semi-implicit = %.0f, ratio = %.3f"
"\n opcount: explicit = %.0f, semi-implicit = %.0f, ratio = %.3f",
nfent1, nL11 + nA21 + nL22,
nfent1/(nL11 + nA21 + nL22),
2*nfent1, 4*nL11 + 2*nA21 + 2*nL22,
2*nfent1/(4*nL11 + 2*nA21 + 2*nL22)) ;
fprintf(msgFile, "\n ratios %8.3f %8.3f %8.3f",
nPhi/nV,
nfent1/(nL11 + nA21 + nL22),
2*nfent1/(4*nL11 + 2*nA21 + 2*nL22)) ;
}
/*
----------------
free the objects
----------------
*/
ETree_free(etree) ;
Graph_free(graph) ;
IVL_free(symbfacIVL) ;
fprintf(msgFile, "\n") ;
fclose(msgFile) ;
return(1) ; }
/*
---------------------------------------------
purpose -- to broadcast a front tree object
from one process to all the others
created -- 98may21, cca
---------------------------------------------
*/
ETree *
ETree_MPI_Bcast (
ETree *etree,
int root,
int msglvl,
FILE *msgFile,
MPI_Comm comm
) {
int myid, nvtx, nfront, nint ;
int *buffer ;
/*
-------------
find identity
-------------
*/
MPI_Comm_rank(comm, &myid) ;
if ( myid == root ) {
/*
--------------------------------------------
this process owns the front tree, allocate a
continuous buffer and load the data into it.
--------------------------------------------
*/
nfront = ETree_nfront(etree) ;
nvtx = ETree_nvtx(etree) ;
nint = 3 + 5*nfront + nvtx ;
buffer = IVinit(nint, -1) ;
buffer[0] = nfront ;
buffer[1] = nvtx ;
buffer[2] = ETree_root(etree) ;
IVcopy(nfront, buffer + 3, ETree_par(etree)) ;
IVcopy(nfront, buffer + 3 + nfront, ETree_fch(etree)) ;
IVcopy(nfront, buffer + 3 + 2*nfront, ETree_sib(etree)) ;
IVcopy(nfront, buffer + 3 + 3*nfront, ETree_nodwghts(etree)) ;
IVcopy(nfront, buffer + 3 + 4*nfront, ETree_bndwghts(etree)) ;
IVcopy(nvtx, buffer + 3 + 5*nfront, ETree_vtxToFront(etree)) ;
/*
------------------------------------
send the size of the buffer and then
the buffer to the other processors
------------------------------------
*/
MPI_Bcast(&nint, 1, MPI_INT, root, comm) ;
MPI_Bcast(buffer, nint, MPI_INT, root, comm) ;
} else {
/*
--------------------------------------------
this process will receive the front tree.
clear its data, receive the number of int's,
then receive the buffer
--------------------------------------------
*/
if ( etree != NULL ) {
ETree_free(etree) ;
}
MPI_Bcast(&nint, 1, MPI_INT, root, comm) ;
buffer = IVinit(nint, -1) ;
MPI_Bcast(buffer, nint, MPI_INT, root, comm) ;
/*
----------------------------------------
create an ETree object and fill its data
----------------------------------------
*/
etree = ETree_new() ;
nfront = buffer[0] ;
nvtx = buffer[1] ;
ETree_init1(etree, nfront, nvtx) ;
etree->tree->n = nfront ;
etree->tree->root = buffer[2] ;
IVcopy(nfront, ETree_par(etree), buffer + 3) ;
IVcopy(nfront, ETree_fch(etree), buffer + 3 + nfront) ;
IVcopy(nfront, ETree_sib(etree), buffer + 3 + 2*nfront) ;
IVcopy(nfront, ETree_nodwghts(etree), buffer + 3 + 3*nfront) ;
IVcopy(nfront, ETree_bndwghts(etree), buffer + 3 + 4*nfront) ;
IVcopy(nvtx, ETree_vtxToFront(etree), buffer + 3 + 5*nfront) ;
}
/*
---------------
free the buffer
---------------
*/
IVfree(buffer) ;
return(etree) ; }
/*
--------------------------------------------------------------
purpose -- generate and return some statistics
about the factor and solve
type -- type of entries
SPOOLES_REAL or SPOOLES_COMPLEX
symmetryflag -- symmetry type
SPOOLES_SYMMETRIC, SPOOLES_HERMITIAN or SPOOLES_NONSYMMETRIC
on return ---
*pnfront -- # of fronts
*pnfactorind -- # of factor indices
*pnfactorent -- # of factor entries
*pnsolveops -- # of solve operations
*pnfactorops -- # of factor operations
return values --
1 -- normal return
-1 -- bridge is NULL
-2 -- type is bad, must be SPOOLES_REAL or SPOOLES_COMPLEX
-3 -- symmetryflag is bad, must be SPOOLES_SYMMETRIC,
SPOOLES_HERMITIAN or SPOOLES_NONSYMMETRIC
-4 -- type and symmetryflag mismatch
-5 -- front tree is not present
-6 -- pnfront is NULL
-7 -- pnfactorind is NULL
-8 -- pnfactorent is NULL
-9 -- pnsolveops is NULL
-10 -- pnfactorops is NULL
created -- 98oct01, cca
--------------------------------------------------------------
*/
int
Bridge_factorStats (
Bridge *bridge,
int type,
int symmetryflag,
int *pnfront,
int *pnfactorind,
int *pnfactorent,
int *pnsolveops,
double *pnfactorops
) {
ETree *etree ;
int nentD, nentU ;
/*
---------------
check the input
---------------
*/
if ( bridge == NULL ) {
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n bridge is NULL\n") ;
return(-1) ;
}
switch ( type ) {
case SPOOLES_REAL :
case SPOOLES_COMPLEX :
break ;
default :
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n bad type %d\n", type) ;
return(-3) ;
break ;
}
switch ( symmetryflag ) {
case SPOOLES_SYMMETRIC :
case SPOOLES_HERMITIAN :
case SPOOLES_NONSYMMETRIC :
break ;
default :
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n bad symmetryflag %d\n", symmetryflag) ;
return(-3) ;
break ;
}
if ( type == SPOOLES_REAL && symmetryflag == SPOOLES_HERMITIAN ) {
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n type %d, symmetryflag %d, mismatch\n",
type, symmetryflag) ;
return(-4) ;
}
if ( (etree = bridge->frontETree) == NULL ) {
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n front tree is not present\n") ;
return(-5) ;
}
if ( pnfront == NULL ) {
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n pnfront is NULL\n") ;
return(-6) ;
}
if ( pnfactorind == NULL ) {
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n pnfactorind is NULL\n") ;
return(-7) ;
}
if ( pnfactorent == NULL ) {
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n pnfactorent is NULL\n") ;
return(-8) ;
}
if ( pnsolveops == NULL ) {
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n pnsolveops is NULL\n") ;
return(-9) ;
}
if ( pnfactorops == NULL ) {
fprintf(stderr, "\n error in Bridge_factorStats()"
"\n pnfactorops is NULL\n") ;
return(-10) ;
}
*pnfront = ETree_nfront(etree) ;
*pnfactorind = ETree_nFactorIndices(etree) ;
*pnfactorent = ETree_nFactorEntries(etree, symmetryflag) ;
*pnfactorops = ETree_nFactorOps(etree, type, symmetryflag) ;
nentD = etree->nvtx ;
nentU = *pnfactorent - nentD ;
switch ( type ) {
case SPOOLES_REAL :
*pnsolveops = 4*nentU + nentD ;
break ;
case SPOOLES_COMPLEX :
*pnsolveops = 16*nentU + 8*nentD ;
break ;
}
return(1) ; }
/*
---------------------------------------------------------------
purpose -- fill dvec[J] with the active storage to eliminate J
using the left-looking general sparse method
symflag -- symmetry flag, one of SPOOLES_SYMMETRIC,
SPOOLES_HERMITIAN or SPOOLES_NONSYMMETRIC
created -- 97may21, cca
---------------------------------------------------------------
*/
void
ETree_GSstorageProfile (
ETree *etree,
int symflag,
IVL *symbfacIVL,
int *vwghts,
double dvec[]
) {
int count, ii, I, J, K, nDJ, nfront, nUJ, sizeI, sizeJ, storage, v ;
int *bndwghts, *head, *indI, *indJ,
*link, *nodwghts, *offsets, *vtxToFront ;
Tree *tree ;
/*
---------------
check the input
---------------
*/
if ( etree == NULL || symbfacIVL == NULL || dvec == NULL ) {
fprintf(stderr,
"\n fatal error in ETree_GSstorageProfile(%p,%p,%p,%p)"
"\n bad input\n", etree, symbfacIVL, vwghts, dvec) ;
exit(-1) ;
}
tree = ETree_tree(etree) ;
nodwghts = ETree_nodwghts(etree) ;
bndwghts = ETree_bndwghts(etree) ;
vtxToFront = ETree_vtxToFront(etree) ;
nfront = ETree_nfront(etree) ;
head = IVinit(nfront, -1) ;
link = IVinit(nfront, -1) ;
offsets = IVinit(nfront, 0) ;
/*
---------------------------------------------
loop over the nodes in a post-order traversal
---------------------------------------------
*/
storage = 0 ;
for ( J = Tree_postOTfirst(tree) ;
J != -1 ;
J = Tree_postOTnext(tree, J) ) {
nDJ = nodwghts[J] ;
nUJ = bndwghts[J] ;
if ( symflag == SPOOLES_SYMMETRIC || symflag == SPOOLES_HERMITIAN ) {
storage += (nDJ*(nDJ + 1))/2 + nDJ*nUJ ;
} else if ( symflag == SPOOLES_NONSYMMETRIC ) {
storage += nDJ*nDJ + 2*nDJ*nUJ ;
}
dvec[J] = storage ;
#if MYDEBUG > 0
fprintf(stdout,
"\n working on front %d, nD = %d, nU = %d, storage = %d",
J, nDJ, nUJ, storage) ;
#endif
/*
-----------------------------
loop over the updating fronts
-----------------------------
*/
while ( (I = head[J]) != -1 ) {
head[J] = link[I] ;
IVL_listAndSize(symbfacIVL, I, &sizeI, &indI) ;
#if MYDEBUG > 0
fprintf(stdout,
"\n updating front %d, offset = %d, sizeI = %d",
I, offsets[I], sizeI) ;
IVfprintf(stdout, sizeI, indI) ;
#endif
for ( ii = offsets[I], count = 0, K = -1 ; ii < sizeI ; ii++ ) {
v = indI[ii] ;
#if MYDEBUG > 0
fprintf(stdout, "\n ii = %d, v = %d, K = %d",
ii, v, vtxToFront[v]) ;
fflush(stdout) ;
#endif
K = vtxToFront[v] ;
if ( K < 0 || K >= nfront ) {
fprintf(stderr, "\n\n fatal error"
"\n ii = %d, v = %d, K = %d", ii, v, K) ;
exit(-1) ;
}
if ( (K = vtxToFront[v]) != J ) {
#if MYDEBUG > 0
fprintf(stdout, "\n linking to next ancestor %d", K) ;
#endif
link[I] = head[K] ;
head[K] = I ;
offsets[I] = ii ;
break ;
}
count += (vwghts == NULL) ? 1 : vwghts[v] ;
#if MYDEBUG > 0
fprintf(stdout, "\n count = %d", count) ;
fflush(stdout) ;
#endif
}
if ( symflag == SPOOLES_SYMMETRIC
|| symflag == SPOOLES_HERMITIAN ) {
storage -= count*nodwghts[I] ;
} else if ( symflag == SPOOLES_NONSYMMETRIC ) {
storage -= 2*count*nodwghts[I] ;
}
}
if ( symflag == SPOOLES_SYMMETRIC || symflag == SPOOLES_HERMITIAN ) {
storage -= (nDJ*(nDJ+1))/2 ;
} else if ( symflag == SPOOLES_NONSYMMETRIC ) {
storage -= nDJ*nDJ ;
}
if ( nUJ > 0 ) {
IVL_listAndSize(symbfacIVL, J, &sizeJ, &indJ) ;
for ( ii = 0 ; ii < sizeJ ; ii++ ) {
v = indJ[ii] ;
if ( (K = vtxToFront[v]) != J ) {
break ;
}
}
offsets[J] = ii ;
#if MYDEBUG > 0
fprintf(stdout, "\n linking to next ancestor %d", K) ;
#endif
IVL_listAndSize(symbfacIVL, J, &sizeJ, &indJ) ;
link[J] = head[K] ;
head[K] = J ;
}
#if MYDEBUG > 0
fprintf(stdout, "\n at end of step %d, storage = %d",
J, storage) ;
#endif
}
#if MYDEBUG >= 0
fprintf(stdout, "\n GS: final storage = %d", storage) ;
#endif
IVfree(head) ;
IVfree(link) ;
IVfree(offsets) ;
return ; }
/*
---------------------------------------------------------------
purpose -- fill dvec[J] with the active storage to eliminate J
using the right-looking general sparse method
symflag -- symmetry flag, one of SPOOLES_SYMMETRIC,
SPOOLES_HERMITIAN or SPOOLES_NONSYMMETRIC
created -- 98dec19, cca
---------------------------------------------------------------
*/
void
ETree_FSstorageProfile (
ETree *etree,
int symflag,
IVL *symbfacIVL,
double dvec[]
) {
char *incore ;
int ii, J, K, nDJ, nfront, nUJ, sizeJ, storage ;
int *bndwghts, *indJ, *mark, *nodwghts, *stor, *vtxToFront ;
Tree *tree ;
/*
---------------
check the input
---------------
*/
if ( etree == NULL || symbfacIVL == NULL || dvec == NULL ) {
fprintf(stderr,
"\n fatal error in ETree_FSstorageProfile(%p,%p,%p)"
"\n bad input\n", etree, symbfacIVL, dvec) ;
exit(-1) ;
}
tree = ETree_tree(etree) ;
nodwghts = ETree_nodwghts(etree) ;
bndwghts = ETree_bndwghts(etree) ;
vtxToFront = ETree_vtxToFront(etree) ;
nfront = ETree_nfront(etree) ;
incore = CVinit(nfront, 'F') ;
stor = IVinit(nfront, 0) ;
mark = IVinit(nfront, -1) ;
/*
--------------------------------------------
compute the storage for each front's chevron
--------------------------------------------
*/
if ( symflag == SPOOLES_SYMMETRIC || symflag == SPOOLES_HERMITIAN ) {
for ( J = 0 ; J < nfront ; J++ ) {
nDJ = nodwghts[J] ;
nUJ = bndwghts[J] ;
stor[J] = (nDJ*(nDJ+1))/2 + nDJ*nUJ ;
}
} else {
for ( J = 0 ; J < nfront ; J++ ) {
nDJ = nodwghts[J] ;
nUJ = bndwghts[J] ;
stor[J] = nDJ*nDJ + 2*nDJ*nUJ ;
}
}
/*
---------------------------------------------
loop over the nodes in a post-order traversal
---------------------------------------------
*/
storage = 0 ;
for ( J = Tree_postOTfirst(tree) ;
J != -1 ;
J = Tree_postOTnext(tree, J) ) {
if ( incore[J] == 'F' ) {
storage += stor[J] ;
incore[J] = 'T' ;
}
IVL_listAndSize(symbfacIVL, J, &sizeJ, &indJ) ;
mark[J] = J ;
for ( ii = 0 ; ii < sizeJ ; ii++ ) {
K = vtxToFront[indJ[ii]] ;
if ( mark[K] != J ) {
mark[K] = J ;
if ( incore[K] == 'F' ) {
storage += stor[K] ;
incore[K] = 'T' ;
}
}
}
dvec[J] = storage ;
storage -= stor[J] ;
}
IVfree(mark) ;
IVfree(stor) ;
CVfree(incore) ;
return ; }
/*
-----------------------------------------------------------
purpose -- to initialize subtree with the subtree
of the front tree using nodes in nodeidsIV.
vtxIV is filled with the vertices in the subtree
return values ---
1 -- normal return
-1 -- subtree is NULL
-2 -- nodeidsIV is NULL
-3 -- etree is NULL
-4 -- nodeidsIV is invalid
-5 -- vtxIV is NULL
created -- 98oct15, cca
-----------------------------------------------------------
*/
int
ETree_initFromSubtree (
ETree *subtree,
IV *nodeidsIV,
ETree *etree,
IV *vtxIV
) {
int J, Jsub, nfrontInETree, nfrontInSubtree,
nvtxInETree, nvtxInSubtree, v, vSub ;
int *bndwghts, *bndwghtsSub, *localmap, *nodwghts, *nodwghtsSub,
*subtreeNodes, *vtxInSubtree, *vtxToFront, *vtxToFrontSub ;
/*
---------------
check the input
---------------
*/
if ( subtree == NULL ) {
fprintf(stderr, "\n\n error in ETree_initFromSubtree()"
"\n subtree is NULL\n") ;
return(-1) ;
}
if ( nodeidsIV == NULL ) {
fprintf(stderr, "\n\n error in ETree_initFromSubtree()"
"\n nodeidsIV is NULL\n") ;
return(-2) ;
}
if ( etree == NULL ) {
fprintf(stderr, "\n\n error in ETree_initFromSubtree()"
"\n etree is NULL\n") ;
return(-3) ;
}
nfrontInETree = ETree_nfront(etree) ;
IV_sizeAndEntries(nodeidsIV, &nfrontInSubtree, &subtreeNodes) ;
if ( nfrontInSubtree < 0 || nfrontInSubtree >= nfrontInETree ) {
fprintf(stderr, "\n\n error in ETree_initFromSubtree()"
"\n nfrontInETree = %d, nfrontInSubtree = %d\n",
nfrontInETree, nfrontInSubtree) ;
return(-4) ;
}
for ( Jsub = 0 ; Jsub < nfrontInSubtree ; Jsub++ ) {
J = subtreeNodes[Jsub] ;
if ( J < 0 || J >= nfrontInETree ) {
fprintf(stderr, "\n\n error in ETree_initFromSubtree()"
"\n nfrontInETree = %d, subtreeNodes[%d] = %d\n",
nfrontInETree, Jsub, subtreeNodes[Jsub]) ;
return(-4) ;
}
}
if ( vtxIV == NULL ) {
fprintf(stderr, "\n\n error in ETree_initFromSubtree()"
"\n vtxIV is NULL\n") ;
return(-5) ;
}
nvtxInETree = ETree_nvtx(etree) ;
vtxToFront = ETree_vtxToFront(etree) ;
/*
----------------------------
create a global-to-local map
----------------------------
*/
localmap = IVinit(nfrontInETree, -1) ;
for ( Jsub = 0 ; Jsub < nfrontInSubtree ; Jsub++ ) {
J = subtreeNodes[Jsub] ;
localmap[J] = Jsub ;
}
/*
---------------------------------------------
compute the number of vertices in the subtree
---------------------------------------------
*/
nvtxInSubtree = 0 ;
for ( v = 0 ; v < nvtxInETree ; v++ ) {
J = vtxToFront[v] ;
if ( (Jsub = localmap[J]) != -1 ) {
nvtxInSubtree++ ;
}
}
/*
----------------------
initialize the subtree
----------------------
*/
ETree_init1(subtree, nfrontInSubtree, nvtxInSubtree) ;
/*
-----------------------------
initialize the subtree's tree
-----------------------------
*/
Tree_initFromSubtree(subtree->tree, nodeidsIV, etree->tree) ;
/*
-----------------------------------
set the nodwght and bndwght vectors
-----------------------------------
*/
nodwghts = ETree_nodwghts(etree) ;
bndwghts = ETree_bndwghts(etree) ;
nodwghtsSub = ETree_nodwghts(subtree) ;
bndwghtsSub = ETree_bndwghts(subtree) ;
for ( Jsub = 0 ; Jsub < nfrontInSubtree ; Jsub++ ) {
J = subtreeNodes[Jsub] ;
nodwghtsSub[Jsub] = nodwghts[J] ;
bndwghtsSub[Jsub] = bndwghts[J] ;
}
/*
-------------------------------------
set the subtree's vtxToFront[] vector
and fill vtxIV with the vertices
-------------------------------------
*/
IV_init(vtxIV, nvtxInSubtree, NULL) ;
vtxInSubtree = IV_entries(vtxIV) ;
vtxToFrontSub = ETree_vtxToFront(subtree) ;
for ( v = vSub = 0 ; v < nvtxInETree ; v++ ) {
J = vtxToFront[v] ;
if ( (Jsub = localmap[J]) != -1 ) {
vtxInSubtree[vSub] = v ;
vtxToFrontSub[vSub] = Jsub ;
vSub++ ;
}
}
/*
------------------------
free the working storage
------------------------
*/
IVfree(localmap) ;
return(1) ; }
/*--------------------------------------------------------------------*/
int
main ( int argc, char *argv[] )
/*
---------------------------------------------------------------
read in a ETree object, create an IV object with the same size,
mark the vertices in the top level separator(s), write the IV
object to a file
created -- 96may02, cca
---------------------------------------------------------------
*/
{
char *inETreeFileName, *outIVfileName ;
double t1, t2 ;
int msglvl, rc, J, K, ncomp, nfront, nvtx, v ;
int *bndwghts, *compids, *fch, *map, *nodwghts,
*par, *sib, *vtxToFront ;
IV *compidsIV, *mapIV ;
ETree *etree ;
FILE *msgFile ;
Tree *tree ;
if ( argc != 5 ) {
fprintf(stdout,
"\n\n usage : %s msglvl msgFile inETreeFile outIVfile"
"\n msglvl -- message level"
"\n msgFile -- message file"
"\n inETreeFile -- input file, must be *.etreef or *.etreeb"
"\n outIVfile -- output file, must be *.ivf or *.ivb"
"\n", argv[0]) ;
return(0) ;
}
msglvl = atoi(argv[1]) ;
if ( strcmp(argv[2], "stdout") == 0 ) {
msgFile = stdout ;
} else if ( (msgFile = fopen(argv[2], "a")) == NULL ) {
fprintf(stderr, "\n fatal error in %s"
"\n unable to open file %s\n",
argv[0], argv[2]) ;
return(-1) ;
}
inETreeFileName = argv[3] ;
outIVfileName = argv[4] ;
fprintf(msgFile,
"\n %s "
"\n msglvl -- %d"
"\n msgFile -- %s"
"\n inETreeFile -- %s"
"\n outIVfile -- %s"
"\n",
argv[0], msglvl, argv[2], inETreeFileName, outIVfileName) ;
fflush(msgFile) ;
/*
------------------------
read in the ETree object
------------------------
*/
if ( strcmp(inETreeFileName, "none") == 0 ) {
fprintf(msgFile, "\n no file to read from") ;
exit(0) ;
}
etree = ETree_new() ;
MARKTIME(t1) ;
rc = ETree_readFromFile(etree, inETreeFileName) ;
MARKTIME(t2) ;
fprintf(msgFile, "\n CPU %9.5f : read in etree from file %s",
t2 - t1, inETreeFileName) ;
if ( rc != 1 ) {
fprintf(msgFile, "\n return value %d from ETree_readFromFile(%p,%s)",
rc, etree, inETreeFileName) ;
exit(-1) ;
}
fprintf(msgFile, "\n\n after reading ETree object from file %s",
inETreeFileName) ;
if ( msglvl > 2 ) {
ETree_writeForHumanEye(etree, msgFile) ;
} else {
ETree_writeStats(etree, msgFile) ;
}
fflush(msgFile) ;
nfront = ETree_nfront(etree) ;
nvtx = ETree_nvtx(etree) ;
bndwghts = ETree_bndwghts(etree) ;
vtxToFront = ETree_vtxToFront(etree) ;
nodwghts = ETree_nodwghts(etree) ;
par = ETree_par(etree) ;
fch = ETree_fch(etree) ;
sib = ETree_sib(etree) ;
tree = ETree_tree(etree) ;
/*
-----------------------------------------
create the map from fronts to components,
top level separator(s) are component zero
-----------------------------------------
*/
mapIV = IV_new() ;
IV_init(mapIV, nfront, NULL) ;
map = IV_entries(mapIV) ;
ncomp = 0 ;
for ( J = Tree_preOTfirst(tree) ;
J != -1 ;
J = Tree_preOTnext(tree, J) ) {
if ( (K = par[J]) == -1 ) {
map[J] = 0 ;
} else if ( map[K] != 0 ) {
map[J] = map[K] ;
} else if ( J == fch[K] && sib[J] == -1
&& bndwghts[J] == nodwghts[K] + bndwghts[K] ) {
map[J] = 0 ;
} else {
map[J] = ++ncomp ;
}
}
fprintf(msgFile, "\n\n mapIV object") ;
if ( msglvl > 2 ) {
IV_writeForHumanEye(mapIV, msgFile) ;
} else {
IV_writeStats(mapIV, msgFile) ;
}
/*
----------------------------------------
fill the map from vertices to components
----------------------------------------
*/
compidsIV = IV_new() ;
IV_init(compidsIV, nvtx, NULL) ;
compids = IV_entries(compidsIV) ;
for ( v = 0 ; v < nvtx ; v++ ) {
compids[v] = map[vtxToFront[v]] ;
}
fprintf(msgFile, "\n\n compidsIV object") ;
if ( msglvl > 2 ) {
IV_writeForHumanEye(compidsIV, msgFile) ;
} else {
IV_writeStats(compidsIV, msgFile) ;
}
fflush(msgFile) ;
/*
-----------------------
write out the IV object
-----------------------
*/
if ( strcmp(outIVfileName, "none") != 0 ) {
MARKTIME(t1) ;
rc = IV_writeToFile(compidsIV, outIVfileName) ;
MARKTIME(t2) ;
fprintf(msgFile, "\n CPU %9.5f : write etree to file %s",
t2 - t1, outIVfileName) ;
}
if ( rc != 1 ) {
fprintf(msgFile, "\n return value %d from IV_writeToFile(%p,%s)",
rc, compidsIV, outIVfileName) ;
}
/*
----------------
free the objects
----------------
*/
ETree_free(etree) ;
IV_free(mapIV) ;
IV_free(compidsIV) ;
fprintf(msgFile, "\n") ;
fclose(msgFile) ;
return(1) ; }
/*--------------------------------------------------------------------*/
int
main ( int argc, char *argv[] )
/*
------------------------------------------------------------
make ETree objects for nested dissection on a regular grid
1 -- vertex elimination tree
2 -- fundamental supernode front tree
3 -- merge only children if possible
4 -- merge all children if possible
5 -- split large non-leaf fronts
created -- 98feb05, cca
------------------------------------------------------------
*/
{
char *outETreeFileName ;
double ops[6] ;
double t1, t2 ;
EGraph *egraph ;
ETree *etree0, *etree1, *etree2, *etree3, *etree4, *etree5 ;
FILE *msgFile ;
Graph *graph ;
int nfronts[6], nfind[6], nzf[6] ;
int maxsize, maxzeros, msglvl, n1, n2, n3, nvtx, rc, v ;
int *newToOld, *oldToNew ;
IV *nzerosIV ;
if ( argc != 9 ) {
fprintf(stdout,
"\n\n usage : %s msglvl msgFile n1 n2 n3 maxzeros maxsize outFile"
"\n msglvl -- message level"
"\n msgFile -- message file"
"\n n1 -- number of points in the first direction"
"\n n2 -- number of points in the second direction"
"\n n3 -- number of points in the third direction"
"\n maxzeros -- number of points in the third direction"
"\n maxsize -- maximum number of vertices in a front"
"\n outFile -- output file, must be *.etreef or *.etreeb"
"\n", argv[0]) ;
return(0) ;
}
msglvl = atoi(argv[1]) ;
if ( strcmp(argv[2], "stdout") == 0 ) {
msgFile = stdout ;
} else if ( (msgFile = fopen(argv[2], "a")) == NULL ) {
fprintf(stderr, "\n fatal error in %s"
"\n unable to open file %s\n",
argv[0], argv[2]) ;
return(-1) ;
}
n1 = atoi(argv[3]) ;
n2 = atoi(argv[4]) ;
n3 = atoi(argv[5]) ;
maxzeros = atoi(argv[6]) ;
maxsize = atoi(argv[7]) ;
outETreeFileName = argv[8] ;
fprintf(msgFile,
"\n %s "
"\n msglvl -- %d"
"\n msgFile -- %s"
"\n n1 -- %d"
"\n n2 -- %d"
"\n n3 -- %d"
"\n maxzeros -- %d"
"\n maxsize -- %d"
"\n outFile -- %s"
"\n",
argv[0], msglvl, argv[2], n1, n2, n3,
maxzeros, maxsize, outETreeFileName) ;
fflush(msgFile) ;
/*
----------------------------
create the grid graph object
----------------------------
*/
if ( n1 == 1 ) {
egraph = EGraph_make9P(n2, n3, 1) ;
} else if ( n2 == 1 ) {
egraph = EGraph_make9P(n1, n3, 1) ;
} else if ( n3 == 1 ) {
egraph = EGraph_make9P(n1, n2, 1) ;
} else {
egraph = EGraph_make27P(n1, n2, n3, 1) ;
}
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n %d x %d x %d grid EGraph", n1, n2, n3) ;
EGraph_writeForHumanEye(egraph, msgFile) ;
fflush(msgFile) ;
}
graph = EGraph_mkAdjGraph(egraph) ;
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n %d x %d x %d grid Graph", n1, n2, n3) ;
Graph_writeForHumanEye(graph, msgFile) ;
fflush(msgFile) ;
}
/*
----------------------------------
get the nested dissection ordering
----------------------------------
*/
nvtx = n1*n2*n3 ;
newToOld = IVinit(nvtx, -1) ;
oldToNew = IVinit(nvtx, -1) ;
mkNDperm(n1, n2, n3, newToOld, 0, n1-1, 0, n2-1, 0, n3-1) ;
for ( v = 0 ; v < nvtx ; v++ ) {
oldToNew[newToOld[v]] = v ;
}
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n %d x %d x %d nd ordering", n1, n2, n3) ;
IVfprintf(msgFile, nvtx, oldToNew) ;
fflush(msgFile) ;
}
/*
------------------------------------------
create the vertex elimination ETree object
------------------------------------------
*/
etree0 = ETree_new() ;
ETree_initFromGraphWithPerms(etree0, graph, newToOld, oldToNew) ;
nfronts[0] = ETree_nfront(etree0) ;
nfind[0] = ETree_nFactorIndices(etree0) ;
nzf[0] = ETree_nFactorEntries(etree0, SPOOLES_SYMMETRIC) ;
ops[0] = ETree_nFactorOps(etree0, SPOOLES_REAL, SPOOLES_SYMMETRIC) ;
fprintf(msgFile,
"\n vtx tree : %8d fronts, %8d indices, %8d |L|, %12.0f ops",
nfronts[0], nfind[0], nzf[0], ops[0]) ;
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n vertex elimination tree") ;
ETree_writeForHumanEye(etree0, msgFile) ;
fflush(msgFile) ;
}
/*
---------------------------------------------
create the fundamental supernode ETree object
---------------------------------------------
*/
nzerosIV = IV_new() ;
IV_init(nzerosIV, nvtx, NULL) ;
IV_fill(nzerosIV, 0) ;
etree1 = ETree_mergeFrontsOne(etree0, 0, nzerosIV) ;
nfronts[1] = ETree_nfront(etree1) ;
nfind[1] = ETree_nFactorIndices(etree1) ;
nzf[1] = ETree_nFactorEntries(etree1, SPOOLES_SYMMETRIC) ;
ops[1] = ETree_nFactorOps(etree1, SPOOLES_REAL, SPOOLES_SYMMETRIC) ;
fprintf(msgFile,
"\n fs tree : %8d fronts, %8d indices, %8d |L|, %12.0f ops",
nfronts[1], nfind[1], nzf[1], ops[1]) ;
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n fundamental supernode front tree") ;
ETree_writeForHumanEye(etree1, msgFile) ;
fprintf(msgFile, "\n\n nzerosIV") ;
IV_writeForHumanEye(nzerosIV, msgFile) ;
fflush(msgFile) ;
}
/*
---------------------------
try to absorb only children
---------------------------
*/
etree2 = ETree_mergeFrontsOne(etree1, maxzeros, nzerosIV) ;
nfronts[2] = ETree_nfront(etree2) ;
nfind[2] = ETree_nFactorIndices(etree2) ;
nzf[2] = ETree_nFactorEntries(etree2, SPOOLES_SYMMETRIC) ;
ops[2] = ETree_nFactorOps(etree2, SPOOLES_REAL, SPOOLES_SYMMETRIC) ;
fprintf(msgFile,
"\n merge one : %8d fronts, %8d indices, %8d |L|, %12.0f ops",
nfronts[2], nfind[2], nzf[2], ops[2]) ;
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n front tree after mergeOne") ;
ETree_writeForHumanEye(etree2, msgFile) ;
fprintf(msgFile, "\n\n nzerosIV") ;
IV_writeForHumanEye(nzerosIV, msgFile) ;
fflush(msgFile) ;
}
/*
--------------------------
try to absorb all children
--------------------------
*/
etree3 = ETree_mergeFrontsAll(etree2, maxzeros, nzerosIV) ;
nfronts[3] = ETree_nfront(etree3) ;
nfind[3] = ETree_nFactorIndices(etree3) ;
nzf[3] = ETree_nFactorEntries(etree3, SPOOLES_SYMMETRIC) ;
ops[3] = ETree_nFactorOps(etree3, SPOOLES_REAL, SPOOLES_SYMMETRIC) ;
fprintf(msgFile,
"\n merge all : %8d fronts, %8d indices, %8d |L|, %12.0f ops",
nfronts[3], nfind[3], nzf[3], ops[3]) ;
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n front tree after mergeAll") ;
ETree_writeForHumanEye(etree3, msgFile) ;
fprintf(msgFile, "\n\n nzerosIV") ;
IV_writeForHumanEye(nzerosIV, msgFile) ;
fflush(msgFile) ;
}
/*
--------------------------------
try to absorb any other children
--------------------------------
*/
etree4 = etree3 ;
/*
etree4 = ETree_mergeFrontsAny(etree3, maxzeros, nzerosIV) ;
nfronts[4] = ETree_nfront(etree4) ;
nfind[4] = ETree_nFactorIndices(etree4) ;
nzf[4] = ETree_nFactorEntries(etree4, SPOOLES_SYMMETRIC) ;
ops[4] = ETree_nFactorOps(etree4, SPOOLES_REAL, SPOOLES_SYMMETRIC) ;
fprintf(msgFile,
"\n merge any : %8d fronts, %8d indices, %8d |L|, %12.0f ops",
nfronts[4], nfind[4], nzf[4], ops[4]) ;
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n front tree after mergeAny") ;
ETree_writeForHumanEye(etree3, msgFile) ;
fprintf(msgFile, "\n\n nzerosIV") ;
IV_writeForHumanEye(nzerosIV, msgFile) ;
fflush(msgFile) ;
}
*/
/*
--------------------
split the front tree
--------------------
*/
etree5 = ETree_splitFronts(etree4, NULL, maxsize, 0) ;
nfronts[5] = ETree_nfront(etree5) ;
nfind[5] = ETree_nFactorIndices(etree5) ;
nzf[5] = ETree_nFactorEntries(etree5, SPOOLES_SYMMETRIC) ;
ops[5] = ETree_nFactorOps(etree5, SPOOLES_REAL, SPOOLES_SYMMETRIC) ;
fprintf(msgFile,
"\n split : %8d fronts, %8d indices, %8d |L|, %12.0f ops",
nfronts[5], nfind[5], nzf[5], ops[5]) ;
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n front tree after split") ;
ETree_writeForHumanEye(etree4, msgFile) ;
fflush(msgFile) ;
}
fprintf(msgFile, "\n\n complex symmetric ops %.0f",
ETree_nFactorOps(etree5, SPOOLES_COMPLEX, SPOOLES_SYMMETRIC)) ;
/*
--------------------------
write out the ETree object
--------------------------
*/
if ( strcmp(outETreeFileName, "none") != 0 ) {
MARKTIME(t1) ;
rc = ETree_writeToFile(etree5, outETreeFileName) ;
MARKTIME(t2) ;
fprintf(msgFile, "\n CPU %9.5f : write etree to file %s",
t2 - t1, outETreeFileName) ;
if ( rc != 1 ) {
fprintf(msgFile,
"\n return value %d from ETree_writeToFile(%p,%s)",
rc, etree5, outETreeFileName) ;
}
}
/*
----------------
free the objects
----------------
*/
ETree_free(etree0) ;
ETree_free(etree1) ;
ETree_free(etree2) ;
ETree_free(etree3) ;
/*
ETree_free(etree4) ;
*/
ETree_free(etree5) ;
EGraph_free(egraph) ;
Graph_free(graph) ;
IVfree(newToOld) ;
IVfree(oldToNew) ;
IV_free(nzerosIV) ;
fprintf(msgFile, "\n") ;
fclose(msgFile) ;
return(1) ; }
//static void factor_MT(struct factorinfo *pfi, InpMtx *mtxA, int size, FILE *msgFile, int symmetryflag)
void factor_MT(struct factorinfo *pfi, InpMtx *mtxA, int size, FILE *msgFile, int symmetryflag)
{
Graph *graph;
IV *ownersIV;
IVL *symbfacIVL;
Chv *rootchv;
/* Initialize pfi: */
pfi->size = size;
pfi->msgFile = msgFile;
DVfill(10, pfi->cpus, 0.0);
/*
* STEP 1 : find a low-fill ordering
* (1) create the Graph object
*/
ssolve_creategraph(&graph, &pfi->frontETree, mtxA, size, msgFile);
/*
* STEP 2: get the permutation, permute the matrix and
* front tree and get the symbolic factorization
*/
ssolve_permuteA(&pfi->oldToNewIV, &pfi->newToOldIV, &symbfacIVL, pfi->frontETree,
mtxA, msgFile, symmetryflag);
/*
* STEP 3: Prepare distribution to multiple threads/cpus
*/
{
DV *cumopsDV;
int nfront;
nfront = ETree_nfront(pfi->frontETree);
pfi->nthread = num_cpus;
if (pfi->nthread > nfront)
pfi->nthread = nfront;
cumopsDV = DV_new();
DV_init(cumopsDV, pfi->nthread, NULL);
ownersIV = ETree_ddMap(pfi->frontETree, SPOOLES_REAL, symmetryflag,
cumopsDV, 1. / (2. * pfi->nthread));
if (DEBUG_LVL > 1) {
fprintf(msgFile,
"\n\n map from fronts to threads");
IV_writeForHumanEye(ownersIV, msgFile);
fprintf(msgFile,
"\n\n factor operations for each front");
DV_writeForHumanEye(cumopsDV, msgFile);
fflush(msgFile);
} else {
fprintf(msgFile, "\n\n Using %d threads\n",
pfi->nthread);
}
DV_free(cumopsDV);
}
/*
* STEP 4: initialize the front matrix object
*/
{
pfi->frontmtx = FrontMtx_new();
pfi->mtxmanager = SubMtxManager_new();
SubMtxManager_init(pfi->mtxmanager, LOCK_IN_PROCESS, 0);
FrontMtx_init(pfi->frontmtx, pfi->frontETree, symbfacIVL, SPOOLES_REAL,
symmetryflag, FRONTMTX_DENSE_FRONTS,
SPOOLES_PIVOTING, LOCK_IN_PROCESS, 0, NULL,
pfi->mtxmanager, DEBUG_LVL, pfi->msgFile);
}
/*
* STEP 5: compute the numeric factorization in parallel
*/
{
ChvManager *chvmanager;
int stats[20];
int error;
chvmanager = ChvManager_new();
ChvManager_init(chvmanager, LOCK_IN_PROCESS, 1);
IVfill(20, stats, 0);
rootchv = FrontMtx_MT_factorInpMtx(pfi->frontmtx, mtxA, MAGIC_TAU, MAGIC_DTOL,
chvmanager, ownersIV, 0,
&error, pfi->cpus, stats, DEBUG_LVL,
pfi->msgFile);
ChvManager_free(chvmanager);
if (DEBUG_LVL > 1) {
fprintf(msgFile, "\n\n factor matrix");
FrontMtx_writeForHumanEye(pfi->frontmtx, pfi->msgFile);
fflush(pfi->msgFile);
}
if (rootchv != NULL) {
fprintf(pfi->msgFile, "\n\n matrix found to be singular\n");
exit(-1);
}
if (error >= 0) {
fprintf(pfi->msgFile, "\n\n fatal error at front %d", error);
exit(-1);
}
}
/*
* STEP 6: post-process the factorization
*/
ssolve_postfactor(pfi->frontmtx, pfi->msgFile);
/*
* STEP 7: get the solve map object for the parallel solve
*/
{
pfi->solvemap = SolveMap_new();
SolveMap_ddMap(pfi->solvemap, symmetryflag,
FrontMtx_upperBlockIVL(pfi->frontmtx),
FrontMtx_lowerBlockIVL(pfi->frontmtx), pfi->nthread, ownersIV,
FrontMtx_frontTree(pfi->frontmtx), RNDSEED, DEBUG_LVL,
pfi->msgFile);
}
/* cleanup: */
InpMtx_free(mtxA);
IVL_free(symbfacIVL);
Graph_free(graph);
IV_free(ownersIV);
}
/*--------------------------------------------------------------------*/
int
main ( int argc, char *argv[] )
/*
----------------------------------------
get statistics for a semi-implicit solve
created -- 97dec11, cca
----------------------------------------
*/
{
char *inGraphFileName, *inETreeFileName, *inMapFileName ;
double nA21, nL, nL11, nL22, nPhi, nV, t1, t2 ;
ETree *etree ;
int ii, inside, J, K, msglvl, nfront, nJ,
nvtx, rc, sizeJ, v, vsize, w ;
int *adjJ, *frontmap, *map, *nodwghts,
*vadj, *vtxToFront, *vwghts ;
IV *mapIV ;
IVL *symbfacIVL ;
Graph *graph ;
FILE *msgFile ;
Tree *tree ;
if ( argc != 6 ) {
fprintf(stdout,
"\n\n usage : %s msglvl msgFile GraphFile ETreeFile mapFile "
"\n msglvl -- message level"
"\n msgFile -- message file"
"\n GraphFile -- input graph file, must be *.graphf or *.graphb"
"\n ETreeFile -- input ETree file, must be *.etreef or *.etreeb"
"\n mapFile -- input map IV file, must be *.ivf or *.ivb"
"\n",
argv[0]) ;
return(0) ;
}
msglvl = atoi(argv[1]) ;
if ( strcmp(argv[2], "stdout") == 0 ) {
msgFile = stdout ;
} else if ( (msgFile = fopen(argv[2], "a")) == NULL ) {
fprintf(stderr, "\n fatal error in %s"
"\n unable to open file %s\n",
argv[0], argv[2]) ;
return(-1) ;
}
inGraphFileName = argv[3] ;
inETreeFileName = argv[4] ;
inMapFileName = argv[5] ;
fprintf(msgFile,
"\n %s "
"\n msglvl -- %d"
"\n msgFile -- %s"
"\n GraphFile -- %s"
"\n ETreeFile -- %s"
"\n mapFile -- %s"
"\n",
argv[0], msglvl, argv[2],
inGraphFileName, inETreeFileName, inMapFileName) ;
fflush(msgFile) ;
/*
------------------------
read in the Graph object
------------------------
*/
graph = Graph_new() ;
if ( strcmp(inGraphFileName, "none") == 0 ) {
fprintf(msgFile, "\n no file to read from") ;
exit(0) ;
}
MARKTIME(t1) ;
rc = Graph_readFromFile(graph, inGraphFileName) ;
MARKTIME(t2) ;
fprintf(msgFile, "\n CPU %9.5f : read in graph from file %s",
t2 - t1, inGraphFileName) ;
nvtx = graph->nvtx ;
vwghts = graph->vwghts ;
if ( rc != 1 ) {
fprintf(msgFile, "\n return value %d from Graph_readFromFile(%p,%s)",
rc, graph, inGraphFileName) ;
exit(-1) ;
}
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n after reading Graph object from file %s",
inGraphFileName) ;
Graph_writeForHumanEye(graph, msgFile) ;
fflush(msgFile) ;
}
/*
------------------------
read in the ETree object
------------------------
*/
etree = ETree_new() ;
if ( strcmp(inETreeFileName, "none") == 0 ) {
fprintf(msgFile, "\n no file to read from") ;
exit(0) ;
}
MARKTIME(t1) ;
rc = ETree_readFromFile(etree, inETreeFileName) ;
MARKTIME(t2) ;
fprintf(msgFile, "\n CPU %9.5f : read in etree from file %s",
t2 - t1, inETreeFileName) ;
nfront = ETree_nfront(etree) ;
tree = ETree_tree(etree) ;
vtxToFront = ETree_vtxToFront(etree) ;
nodwghts = ETree_nodwghts(etree) ;
nL = ETree_nFactorEntries(etree, 2) ;
if ( rc != 1 ) {
fprintf(msgFile, "\n return value %d from ETree_readFromFile(%p,%s)",
rc, etree, inETreeFileName) ;
exit(-1) ;
}
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n after reading ETree object from file %s",
inETreeFileName) ;
ETree_writeForHumanEye(etree, msgFile) ;
fflush(msgFile) ;
}
/*
-------------------------
read in the map IV object
-------------------------
*/
mapIV = IV_new() ;
if ( strcmp(inMapFileName, "none") == 0 ) {
fprintf(msgFile, "\n no file to read from") ;
exit(0) ;
}
MARKTIME(t1) ;
rc = IV_readFromFile(mapIV, inMapFileName) ;
MARKTIME(t2) ;
fprintf(msgFile, "\n CPU %9.5f : read in mapIV from file %s",
t2 - t1, inMapFileName) ;
map = IV_entries(mapIV) ;
if ( rc != 1 ) {
fprintf(msgFile, "\n return value %d from IV_readFromFile(%p,%s)",
rc, mapIV, inMapFileName) ;
exit(-1) ;
}
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n after reading IV object from file %s",
inMapFileName) ;
IV_writeForHumanEye(mapIV, msgFile) ;
fflush(msgFile) ;
}
nV = nPhi = 0 ;
if ( vwghts == NULL ) {
for ( v = 0 ; v < nvtx ; v++ ) {
nV++ ;
if ( map[v] == 0 ) {
nPhi++ ;
}
}
} else {
for ( v = 0 ; v < nvtx ; v++ ) {
nV += vwghts[v] ;
if ( map[v] == 0 ) {
nPhi += vwghts[v] ;
}
}
}
fprintf(msgFile, "\n nPhi = %.0f, nV = %.0f", nPhi, nV) ;
/*
-------------------------
get the frontmap[] vector
-------------------------
*/
frontmap = IVinit(nfront, -1) ;
for ( v = 0 ; v < nvtx ; v++ ) {
J = vtxToFront[v] ;
if ( frontmap[J] == -1 ) {
frontmap[J] = map[v] ;
} else if ( frontmap[J] != map[v] ) {
fprintf(msgFile, "\n\n error, frontmap[%d] = %d, map[%d] = %d",
J, frontmap[J], v, map[v]) ;
}
}
/*
----------------------------------
compute the symbolic factorization
----------------------------------
*/
symbfacIVL = SymbFac_initFromGraph(etree, graph) ;
if ( msglvl > 2 ) {
fprintf(msgFile, "\n\n symbolic factorization") ;
IVL_writeForHumanEye(symbfacIVL, msgFile) ;
fflush(msgFile) ;
}
/*
--------------------------------------------
compute the number of entries in L11 and L22
--------------------------------------------
*/
nL11 = nL22 = 0 ;
for ( J = Tree_postOTfirst(tree) ;
J != -1 ;
J = Tree_postOTnext(tree, J) ) {
nJ = nodwghts[J] ;
if ( msglvl > 3 ) {
fprintf(msgFile, "\n\n front %d, nJ = %d", J, nJ) ;
}
IVL_listAndSize(symbfacIVL, J, &sizeJ, &adjJ) ;
for ( ii = 0, inside = 0 ; ii < sizeJ ; ii++ ) {
w = adjJ[ii] ;
K = vtxToFront[w] ;
if ( msglvl > 3 ) {
fprintf(msgFile, "\n w = %d, K = %d", w, K) ;
}
if ( K > J && frontmap[K] == frontmap[J] ) {
inside += (vwghts == NULL) ? 1 : vwghts[w] ;
if ( msglvl > 3 ) {
fprintf(msgFile, ", inside") ;
}
}
}
if ( frontmap[J] != 0 ) {
if ( msglvl > 3 ) {
fprintf(msgFile, "\n inside = %d, adding %d to L11",
inside, nJ*nJ + 2*nJ*inside) ;
}
nL11 += nJ*nJ + 2*nJ*inside ;
} else {
if ( msglvl > 3 ) {
fprintf(msgFile, "\n inside = %d, adding %d to L22",
inside, nJ*nJ + 2*nJ*inside) ;
}
nL22 += nJ*nJ + 2*nJ*inside ;
}
}
if ( msglvl > 0 ) {
fprintf(msgFile, "\n |L| = %.0f, |L11| = %.0f, |L22| = %.0f",
nL, nL11, nL22) ;
}
/*
------------------------------------
compute the number of entries in A21
------------------------------------
*/
nA21 = 0 ;
if ( vwghts != NULL ) {
for ( v = 0 ; v < nvtx ; v++ ) {
if ( map[v] == 0 ) {
Graph_adjAndSize(graph, v, &vsize, &vadj) ;
for ( ii = 0 ; ii < vsize ; ii++ ) {
w = vadj[ii] ;
if ( map[v] != map[w] ) {
if ( msglvl > 3 ) {
fprintf(msgFile, "\n A21 : v = %d, w = %d", v, w) ;
}
nA21 += vwghts[v] * vwghts[w] ;
}
}
}
}
} else {
for ( v = 0 ; v < nvtx ; v++ ) {
if ( map[v] == 0 ) {
Graph_adjAndSize(graph, v, &vsize, &vadj) ;
for ( ii = 0 ; ii < vsize ; ii++ ) {
w = vadj[ii] ;
if ( map[v] != map[w] ) {
if ( msglvl > 3 ) {
fprintf(msgFile, "\n A21 : v = %d, w = %d", v, w) ;
}
nA21++ ;
}
}
}
}
}
if ( msglvl > 0 ) {
fprintf(msgFile,
"\n |L| = %.0f, |L11| = %.0f, |L22| = %.0f, |A21| = %.0f",
nL, nL11, nL22, nA21) ;
fprintf(msgFile,
"\n storage: explicit = %.0f, semi-implicit = %.0f, ratio = %.3f"
"\n opcount: explicit = %.0f, semi-implicit = %.0f, ratio = %.3f",
nL, nL11 + nA21 + nL22,
nL/(nL11 + nA21 + nL22),
2*nL, 4*nL11 + 2*nA21 + 2*nL22,
2*nL/(4*nL11 + 2*nA21 + 2*nL22)) ;
fprintf(msgFile, "\n ratios %8.3f %8.3f %8.3f",
nPhi/nV,
nL/(nL11 + nA21 + nL22),
2*nL/(4*nL11 + 2*nA21 + 2*nL22)) ;
}
/*
------------------------
free the working storage
------------------------
*/
Graph_free(graph) ;
ETree_free(etree) ;
IV_free(mapIV) ;
IVL_free(symbfacIVL) ;
IVfree(frontmap) ;
fprintf(msgFile, "\n") ;
fclose(msgFile) ;
return(1) ; }