void
RMtx4GetColumn( RVec4* out, const RMtx4* in, int col )
{
RVec4SetX( out, GET_ELEMENT( in, 0, col ) );
RVec4SetY( out, GET_ELEMENT( in, 1, col ) );
RVec4SetZ( out, GET_ELEMENT( in, 2, col ) );
RVec4SetW( out, GET_ELEMENT( in, 3, col ) );
}
void
RMtx4GetRow( RVec4* out, const RMtx4* in, int row )
{
RVec4SetX( out, GET_ELEMENT( in, row, 0 ) );
RVec4SetY( out, GET_ELEMENT( in, row, 1 ) );
RVec4SetZ( out, GET_ELEMENT( in, row, 2 ) );
RVec4SetW( out, GET_ELEMENT( in, row, 3 ) );
}
void
RMtx4Sub( RMtx4* out, const RMtx4* m1, const RMtx4* m2 )
{
int row, col;
for ( row = 0; row < 4; ++row )
for ( col = 0; col < 4; ++col )
SET_ELEMENT( out, row, col,
GET_ELEMENT( m1, row, col ) - GET_ELEMENT( m2, row, col ) );
}
void
RMtx4GetUpper3x3( RMtx3* out, const RMtx4* in )
{
RMtx3SetElements( out,
GET_ELEMENT(in,0,0), GET_ELEMENT(in,0,1), GET_ELEMENT(in,0,2),
GET_ELEMENT(in,1,0), GET_ELEMENT(in,1,1), GET_ELEMENT(in,1,2),
GET_ELEMENT(in,2,0), GET_ELEMENT(in,2,1), GET_ELEMENT(in,2,2) );
}
void
RMtx4Add( RMtx4* out, const RMtx4* m1, const RMtx4* m2 )
{
int row, col;
RMtx4 tmp;
for ( row = 0; row < 4; ++row )
for ( col = 0; col < 4; ++col )
SET_ELEMENT( &tmp, row, col,
GET_ELEMENT( m1, row, col ) + GET_ELEMENT( m2, row, col ) );
RMtx4Copy( out, &tmp );
}
element_type& operator()(size_t r, size_t c) noexcept {
assert(r < nrows_);
assert(c < nrows_);
assert(data_);
return GET_ELEMENT(data_, nrows_, nrows_, r, c);
}
void
RMtx4Scale( RMtx4* out, const RMtx4* m, rmReal f )
{
int row, col;
for ( row = 0; row < 4; ++row )
for ( col = 0; col < 4; ++col )
SET_ELEMENT( out, row, col,
GET_ELEMENT( m, row, col ) * f );
}
void
RMtx4Transpose( RMtx4* out, const RMtx4* in )
{
int row, col;
RMtx4 tmp;
for ( row = 0; row < 4; ++row )
for ( col = 0; col < 4; ++col )
SET_ELEMENT( &tmp, row, col, GET_ELEMENT( in, col, row ) );
RMtx4Copy( out, &tmp );
}
void
RMtx4Mul( RMtx4* out, const RMtx4* m1, const RMtx4* m2 )
{
int row, col;
RMtx4 tmp;
for ( row = 0; row < 4; ++row )
{
for ( col = 0; col < 4; ++col )
{
int i;
rmReal sum = 0.0f;
for ( i = 0; i < 4; ++i )
{
sum += GET_ELEMENT( m1, row, i ) * GET_ELEMENT( m2, i, col );
}
SET_ELEMENT( &tmp, row, col, sum );
}
}
RMtx4Copy( out, &tmp );
}
/*
* === FUNCTION ======================================================================
* Name: OS_SemPost
* Description: This functions release the semaphore and wakes up the task
* being blocked on it. A task can be also be waiting for a time event, so this
* function vefiries whether the relased task is also waiting for a time event
* or not. In case it is, the task is also removed from the time queue.
* =====================================================================================
*/
int OS_SemPost(OS_SemID_t _sem)
{
OS_task_t *task;
OS_Node_t *n;
OS_SemID_t id_sem = _sem;
KTRACE_EVENT_SEM(KEVENT_UNLOCK, SEM(id_sem)->sem_id);
if(OS_ListGetNum(&(SEM(id_sem)->list)) != 0)
{
/* Outs the task from the semaphore queue */
n = OS_ListGetNode(&(SEM(id_sem)->list));
task = GET_ELEMENT(OS_task_t, n, resource_node);
/* Store what is the blocked thread in the semaphore structure */
SEM(id_sem)->tid_in = task->tid;
KTRACE_EVENT_SEM_TID(KEVENT_LOCK_CATCHED, SEM(id_sem)->sem_id, task->tid);
/*
* Verifies if the thread is blocked in the time queue
* due to a TimedWait system call
*/
if(task->status | TS_TIME_BLOCKED) {
OS_TimeRemoveFromQueue(task);
}
/* Reset the wait_sem value */
task->wait_sem = 0;
/* Add the blocked task to the ready task queue */
OS_SchedTaskReady(task->tid);
}
else {
/* Increment the semaphore counter */
SEM(id_sem)->sem_value = 1;
OS_ListReset(&(SEM(id_sem)->list));
}
return 0;
}
GLOBAL void UMF_garbage_collection
(
NumericType *Numeric,
WorkType *Work,
Int drnew, /* compact current front to drnew-by-dcnew */
Int dcnew,
Int do_Fcpos
)
{
/* ---------------------------------------------------------------------- */
/* local variables */
/* ---------------------------------------------------------------------- */
Int size, e, n_row, n_col, nrows, ncols, nrowsleft, ncolsleft, prevsize,
csize, size2, i2, j2, i, j, cdeg, rdeg, *E, row, col,
*Rows, *Cols, *Rows2, *Cols2, nel, e2, *Row_tuples, *Col_tuples,
*Row_degree, *Col_degree ;
Entry *C, *C1, *C3, *C2 ;
Unit *psrc, *pdest, *p, *pnext ;
Element *epsrc, *epdest ;
#ifndef NDEBUG
Int nmark ;
#endif
/* ---------------------------------------------------------------------- */
/* get parameters */
/* ---------------------------------------------------------------------- */
Col_degree = Numeric->Cperm ; /* for NON_PIVOTAL_COL macro */
Row_degree = Numeric->Rperm ; /* for NON_PIVOTAL_ROW macro */
Row_tuples = Numeric->Uip ;
Col_tuples = Numeric->Lip ;
E = Work->E ;
n_row = Work->n_row ;
n_col = Work->n_col ;
/* note that the tuple lengths (Col_tlen and Row_tlen) are updated, but */
/* the tuple lists themselves are stale and are about to be destroyed */
/* and recreated. Do not attempt to scan them until they are recreated. */
#ifndef NDEBUG
DEBUGm1 (("::::GARBAGE COLLECTION::::\n")) ;
UMF_dump_memory (Numeric) ;
#endif
Numeric->ngarbage++ ;
/* ---------------------------------------------------------------------- */
/* delete the tuple lists by marking the blocks as free */
/* ---------------------------------------------------------------------- */
/* do not modify Row_tlen and Col_tlen */
/* those are needed for UMF_build_tuples */
for (row = 0 ; row < n_row ; row++)
{
if (NON_PIVOTAL_ROW (row) && Row_tuples [row])
{
DEBUG2 (("row "ID" tuples "ID"\n", row, Row_tuples [row])) ;
p = Numeric->Memory + Row_tuples [row] - 1 ;
DEBUG2 (("Freeing tuple list row "ID", p-S "ID", size "ID"\n",
row, (Int) (p-Numeric->Memory), p->header.size)) ;
ASSERT (p->header.size > 0) ;
ASSERT (p >= Numeric->Memory + Numeric->itail) ;
ASSERT (p < Numeric->Memory + Numeric->size) ;
p->header.size = -p->header.size ;
Row_tuples [row] = 0 ;
}
}
for (col = 0 ; col < n_col ; col++)
{
if (NON_PIVOTAL_COL (col) && Col_tuples [col])
{
DEBUG2 (("col "ID" tuples "ID"\n", col, Col_tuples [col])) ;
p = Numeric->Memory + Col_tuples [col] - 1 ;
DEBUG2 (("Freeing tuple list col "ID", p-S "ID", size "ID"\n",
col, (Int) (p-Numeric->Memory), p->header.size)) ;
ASSERT (p->header.size > 0) ;
ASSERT (p >= Numeric->Memory + Numeric->itail) ;
ASSERT (p < Numeric->Memory + Numeric->size) ;
p->header.size = -p->header.size ;
Col_tuples [col] = 0 ;
}
}
/* ---------------------------------------------------------------------- */
/* mark the elements, and compress the name space */
/* ---------------------------------------------------------------------- */
nel = Work->nel ;
ASSERT (nel < Work->elen) ;
#ifndef NDEBUG
nmark = 0 ;
UMF_dump_current_front (Numeric, Work, FALSE) ;
DEBUGm1 (("E [0] "ID" \n", E [0])) ;
ASSERT (IMPLIES (E [0],
Work->Flublock == (Entry *) (Numeric->Memory + E [0]))) ;
ASSERT (IMPLIES (Work->Flublock,
Work->Flublock == (Entry *) (Numeric->Memory + E [0]))) ;
ASSERT ((E [0] != 0) == (Work->Flublock != (Entry *) NULL)) ;
#endif
e2 = 0 ;
for (e = 0 ; e <= nel ; e++) /* for all elements in order of creation */
{
if (E [e])
{
psrc = Numeric->Memory + E [e] ;
psrc-- ; /* get the header of this block */
if (e > 0)
{
e2++ ; /* do not renumber element zero */
}
ASSERT (psrc->header.size > 0) ;
psrc->header.size = e2 ; /* store the new name in the header */
#ifndef NDEBUG
nmark++ ;
#endif
DEBUG7 ((ID":: Mark e "ID" at psrc-S "ID", new e "ID"\n",
nmark, e, (Int) (psrc-Numeric->Memory), e2)) ;
E [e] = 0 ;
if (e == Work->prior_element)
{
Work->prior_element = e2 ;
}
}
}
/* all 1..e2 are now in use (element zero may or may not be in use) */
Work->nel = e2 ;
nel = Work->nel ;
#ifndef NDEBUG
for (e = 0 ; e < Work->elen ; e++)
{
ASSERT (!E [e]) ;
}
#endif
/* ---------------------------------------------------------------------- */
/* compress the elements */
/* ---------------------------------------------------------------------- */
/* point to tail marker block of size 1 + header */
psrc = Numeric->Memory + Numeric->size - 2 ;
pdest = psrc ;
prevsize = psrc->header.prevsize ;
DEBUG7 (("Starting the compression:\n")) ;
while (prevsize > 0)
{
/* ------------------------------------------------------------------ */
/* move up to the next element above the current header, and */
/* get the element name and size */
/* (if it is an element, the name will be positive) */
/* ------------------------------------------------------------------ */
size = prevsize ;
psrc -= (size + 1) ;
e = psrc->header.size ;
prevsize = psrc->header.prevsize ;
/* top block at tail has prevsize of 0 */
/* a free block will have a negative size, so skip it */
/* otherwise, if size >= 0, it holds the element name, not the size */
DEBUG8 (("psrc-S: "ID" prevsize: "ID" size: "ID,
(Int) (psrc-Numeric->Memory), prevsize, size)) ;
if (e == 0)
{
/* -------------------------------------------------------------- */
/* this is the current frontal matrix */
/* -------------------------------------------------------------- */
Entry *F1, *F2, *Fsrc, *Fdst ;
Int c, r, k, dr, dc, gap, gap1, gap2, nb ;
/* shift the frontal matrix down */
F1 = (Entry *) (psrc + 1) ;
/* get the size of the current front. r and c could be zero */
k = Work->fnpiv ;
dr = Work->fnr_curr ;
dc = Work->fnc_curr ;
r = Work->fnrows ;
c = Work->fncols ;
nb = Work->nb ;
ASSERT ((dr >= 0 && (dr % 2) == 1) || dr == 0) ;
ASSERT (drnew >= 0) ;
if (drnew % 2 == 0)
{
/* make sure leading frontal matrix dimension is always odd */
drnew++ ;
}
drnew = MIN (dr, drnew) ;
ASSERT ((drnew >= 0 && (drnew % 2) == 1) || drnew == 0) ;
pnext = pdest ;
#ifndef NDEBUG
DEBUGm2 (("move front: dr "ID" dc "ID" r "ID" drnew "ID" c "ID
" dcnew " ID" k "ID"\n", dr, dc, r, drnew, c, dcnew, k)) ;
DEBUG7 (("\n")) ;
DEBUG7 ((ID":: Move current frontal matrix from: psrc-S: "ID" \n",
nmark, (Int) (psrc-Numeric->Memory))) ;
nmark-- ;
ASSERT (E [e] == 0) ;
ASSERT (Work->Flublock == F1) ;
ASSERT (Work->Flblock == Work->Flublock + nb*nb) ;
ASSERT (Work->Fublock == Work->Flblock + dr*nb) ;
ASSERT (Work->Fcblock == Work->Fublock + nb*dc) ;
DEBUG7 (("C block: ")) ;
UMF_dump_dense (Work->Fcblock, dr, r, c) ;
DEBUG7 (("L block: ")) ;
UMF_dump_dense (Work->Flblock, dr, r, k);
DEBUG7 (("U' block: ")) ;
UMF_dump_dense (Work->Fublock, dc, c, k) ;
DEBUG7 (("LU block: ")) ;
UMF_dump_dense (Work->Flublock, nb, k, k) ;
ASSERT (r <= drnew && c <= dcnew && drnew <= dr && dcnew <= dc) ;
#endif
/* compact frontal matrix to drnew-by-dcnew before moving it */
/* do not compact the LU block (nb-by-nb) */
/* compact the columns of L (from dr-by-nb to drnew-by-nb) */
Fsrc = Work->Flblock ;
Fdst = Work->Flblock ;
ASSERT (Fdst == F1 + nb*nb) ;
gap1 = dr - r ;
gap2 = drnew - r ;
ASSERT (gap1 >= 0) ;
for (j = 0 ; j < k ; j++)
{
for (i = 0 ; i < r ; i++)
{
*Fdst++ = *Fsrc++ ;
}
Fsrc += gap1 ;
Fdst += gap2 ;
}
ASSERT (Fdst == F1 + nb*nb + drnew*k) ;
Fdst += drnew * (nb - k) ;
/* compact the rows of U (U' from dc-by-nb to dcnew-by-nb) */
Fsrc = Work->Fublock ;
ASSERT (Fdst == F1 + nb*nb + drnew*nb) ;
gap1 = dc - c ;
gap2 = dcnew - c ;
for (i = 0 ; i < k ; i++)
{
for (j = 0 ; j < c ; j++)
{
*Fdst++ = *Fsrc++ ;
}
Fsrc += gap1 ;
Fdst += gap2 ;
}
ASSERT (Fdst == F1 + nb*nb + drnew*nb + dcnew*k) ;
Fdst += dcnew * (nb - k) ;
/* compact the columns of C (from dr-by-dc to drnew-by-dcnew) */
Fsrc = Work->Fcblock ;
ASSERT (Fdst == F1 + nb*nb + drnew*nb + nb*dcnew) ;
gap1 = dr - r ;
gap2 = drnew - r ;
for (j = 0 ; j < c ; j++)
{
for (i = 0 ; i < r ; i++)
{
*Fdst++ = *Fsrc++ ;
}
Fsrc += gap1 ;
Fdst += gap2 ;
}
ASSERT (Fdst == F1 + nb*nb + drnew*nb + nb*dcnew + drnew*c) ;
/* recompute Fcpos, if necessary */
if (do_Fcpos)
{
Int *Fcols, *Fcpos ;
Fcols = Work->Fcols ;
Fcpos = Work->Fcpos ;
for (j = 0 ; j < c ; j++)
{
col = Fcols [j] ;
ASSERT (col >= 0 && col < Work->n_col) ;
ASSERT (Fcpos [col] == j * dr) ;
Fcpos [col] = j * drnew ;
}
#ifndef NDEBUG
{
Int cnt = 0 ;
for (j = 0 ; j < Work->n_col ; j++)
{
if (Fcpos [j] != EMPTY) cnt++ ;
}
DEBUGm2 (("Recompute Fcpos cnt "ID" c "ID"\n", cnt, c)) ;
ASSERT (cnt == c) ;
}
#endif
}
#ifndef NDEBUG
DEBUGm2 (("Compacted front, drnew "ID" dcnew "ID"\n", drnew, dcnew)) ;
DEBUG7 (("C block: ")) ;
UMF_dump_dense (F1 + nb*nb + drnew*nb + nb*dcnew, drnew, r, c) ;
DEBUG7 (("L block: ")) ;
UMF_dump_dense (F1 + nb*nb, drnew, r, k) ;
DEBUG7 (("U block: ")) ;
UMF_dump_dense (F1 + nb*nb + drnew*nb, nb, k, c) ;
DEBUG7 (("LU block: ")) ;
UMF_dump_dense (F1, nb, k, k) ;
#endif
/* Compacted dimensions of the new frontal matrix. */
Work->fnr_curr = drnew ;
Work->fnc_curr = dcnew ;
Work->fcurr_size = (drnew + nb) * (dcnew + nb) ;
size = UNITS (Entry, Work->fcurr_size) ;
/* make sure the object doesn't evaporate. The front can have
* zero size (Work->fcurr_size = 0), but the size of the memory
* block containing it cannot have zero size. */
size = MAX (1, size) ;
/* get the destination of frontal matrix */
pnext->header.prevsize = size ;
pdest -= (size + 1) ;
F2 = (Entry *) (pdest + 1) ;
ASSERT ((unsigned Int) psrc + 1 + size <= (unsigned Int) pnext) ;
ASSERT (psrc <= pdest) ;
ASSERT (F1 <= F2) ;
/* move the C block first */
Fsrc = F1 + nb*nb + drnew*nb + nb*dcnew + drnew*c ;
Fdst = F2 + nb*nb + drnew*nb + nb*dcnew + drnew*c ;
gap = drnew - r ;
for (j = c-1 ; j >= 0 ; j--)
{
Fsrc -= gap ;
Fdst -= gap ;
/* move column j of C */
for (i = r-1 ; i >= 0 ; i--)
{
*--Fdst = *--Fsrc ;
}
}
ASSERT (Fsrc == F1 + nb*nb + drnew*nb + nb*dcnew) ;
ASSERT (Fdst == F2 + nb*nb + drnew*nb + nb*dcnew) ;
/* move the U block */
Fsrc -= dcnew * (nb - k) ;
Fdst -= dcnew * (nb - k) ;
ASSERT (Fsrc == F1 + nb*nb + drnew*nb + dcnew*k) ;
ASSERT (Fdst == F2 + nb*nb + drnew*nb + dcnew*k) ;
gap = dcnew - c ;
for (i = k-1 ; i >= 0 ; i--)
{
Fsrc -= gap ;
Fdst -= gap ;
for (j = c-1 ; j >= 0 ; j--)
{
*--Fdst = *--Fsrc ;
}
}
ASSERT (Fsrc == F1 + nb*nb + drnew*nb) ;
ASSERT (Fdst == F2 + nb*nb + drnew*nb) ;
/* move the L block */
Fsrc -= drnew * (nb - k) ;
Fdst -= drnew * (nb - k) ;
ASSERT (Fsrc == F1 + nb*nb + drnew*k) ;
ASSERT (Fdst == F2 + nb*nb + drnew*k) ;
gap = drnew - r ;
for (j = k-1 ; j >= 0 ; j--)
{
Fsrc -= gap ;
Fdst -= gap ;
for (i = r-1 ; i >= 0 ; i--)
{
*--Fdst = *--Fsrc ;
}
}
ASSERT (Fsrc == F1 + nb*nb) ;
ASSERT (Fdst == F2 + nb*nb) ;
/* move the LU block */
Fsrc -= nb * (nb - k) ;
Fdst -= nb * (nb - k) ;
ASSERT (Fsrc == F1 + nb*k) ;
ASSERT (Fdst == F2 + nb*k) ;
gap = nb - k ;
for (j = k-1 ; j >= 0 ; j--)
{
Fsrc -= gap ;
Fdst -= gap ;
for (i = k-1 ; i >= 0 ; i--)
{
*--Fdst = *--Fsrc ;
}
}
ASSERT (Fsrc == F1) ;
ASSERT (Fdst == F2) ;
E [0] = (pdest + 1) - Numeric->Memory ;
Work->Flublock = (Entry *) (Numeric->Memory + E [0]) ;
ASSERT (Work->Flublock == F2) ;
Work->Flblock = Work->Flublock + nb * nb ;
Work->Fublock = Work->Flblock + drnew * nb ;
Work->Fcblock = Work->Fublock + nb * dcnew ;
pdest->header.prevsize = 0 ;
pdest->header.size = size ;
#ifndef NDEBUG
DEBUG7 (("After moving compressed current frontal matrix:\n")) ;
DEBUG7 (("C block: ")) ;
UMF_dump_dense (Work->Fcblock, drnew, r, c) ;
DEBUG7 (("L block: ")) ;
UMF_dump_dense (Work->Flblock, drnew, r, k);
DEBUG7 (("U' block: ")) ;
UMF_dump_dense (Work->Fublock, dcnew, c, k) ;
DEBUG7 (("LU block: ")) ;
UMF_dump_dense (Work->Flublock, nb, k, k) ;
#endif
}
else if (e > 0)
{
/* -------------------------------------------------------------- */
/* this is an element, compress and move from psrc down to pdest */
/* -------------------------------------------------------------- */
#ifndef NDEBUG
DEBUG7 (("\n")) ;
DEBUG7 ((ID":: Move element "ID": from: "ID" \n",
nmark, e, (Int) (psrc-Numeric->Memory))) ;
nmark-- ;
ASSERT (e <= nel) ;
ASSERT (E [e] == 0) ;
#endif
/* -------------------------------------------------------------- */
/* get the element scalars, and pointers to C, Rows, and Cols: */
/* -------------------------------------------------------------- */
p = psrc + 1 ;
GET_ELEMENT (epsrc, p, Cols, Rows, ncols, nrows, C) ;
nrowsleft = epsrc->nrowsleft ;
ncolsleft = epsrc->ncolsleft ;
cdeg = epsrc->cdeg ;
rdeg = epsrc->rdeg ;
#ifndef NDEBUG
DEBUG7 ((" nrows "ID" nrowsleft "ID"\n", nrows, nrowsleft)) ;
DEBUG7 ((" ncols "ID" ncolsleft "ID"\n", ncols, ncolsleft)) ;
DEBUG8 ((" Rows:")) ;
for (i = 0 ; i < nrows ; i++) DEBUG8 ((" "ID, Rows [i])) ;
DEBUG8 (("\n Cols:")) ;
for (j = 0 ; j < ncols ; j++) DEBUG8 ((" "ID, Cols [j])) ;
DEBUG8 (("\n")) ;
#endif
/* -------------------------------------------------------------- */
/* determine the layout of the new element */
/* -------------------------------------------------------------- */
csize = nrowsleft * ncolsleft ;
size2 = UNITS (Element, 1)
+ UNITS (Int, nrowsleft + ncolsleft)
+ UNITS (Entry, csize) ;
DEBUG7 (("Old size "ID" New size "ID"\n", size, size2)) ;
pnext = pdest ;
pnext->header.prevsize = size2 ;
pdest -= (size2 + 1) ;
ASSERT (size2 <= size) ;
ASSERT ((unsigned Int) psrc + 1 + size <= (unsigned Int) pnext) ;
ASSERT (psrc <= pdest) ;
p = pdest + 1 ;
epdest = (Element *) p ;
p += UNITS (Element, 1) ;
Cols2 = (Int *) p ;
Rows2 = Cols2 + ncolsleft ;
p += UNITS (Int, nrowsleft + ncolsleft) ;
C2 = (Entry *) p ;
ASSERT (epdest >= epsrc) ;
ASSERT (Rows2 >= Rows) ;
ASSERT (Cols2 >= Cols) ;
ASSERT (C2 >= C) ;
ASSERT (p + UNITS (Entry, csize) == pnext) ;
/* -------------------------------------------------------------- */
/* move the contribution block */
/* -------------------------------------------------------------- */
/* overlap = psrc + size + 1 > pdest ; */
if (nrowsleft < nrows || ncolsleft < ncols)
{
/* ---------------------------------------------------------- */
/* compress contribution block in place prior to moving it */
/* ---------------------------------------------------------- */
DEBUG7 (("Compress C in place prior to move:\n"));
#ifndef NDEBUG
UMF_dump_dense (C, nrows, nrows, ncols) ;
#endif
C1 = C ;
C3 = C ;
for (j = 0 ; j < ncols ; j++)
{
if (Cols [j] >= 0)
{
for (i = 0 ; i < nrows ; i++)
{
if (Rows [i] >= 0)
{
*C3++ = C1 [i] ;
}
}
}
C1 += nrows ;
}
ASSERT (C3-C == csize) ;
DEBUG8 (("Newly compressed contrib. block (all in use):\n")) ;
#ifndef NDEBUG
UMF_dump_dense (C, nrowsleft, nrowsleft, ncolsleft) ;
#endif
}
/* shift the contribution block down */
C += csize ;
C2 += csize ;
for (i = 0 ; i < csize ; i++)
{
*--C2 = *--C ;
}
/* -------------------------------------------------------------- */
/* move the row indices */
/* -------------------------------------------------------------- */
i2 = nrowsleft ;
for (i = nrows - 1 ; i >= 0 ; i--)
{
ASSERT (Rows2+i2 >= Rows+i) ;
if (Rows [i] >= 0)
{
Rows2 [--i2] = Rows [i] ;
}
}
ASSERT (i2 == 0) ;
j2 = ncolsleft ;
for (j = ncols - 1 ; j >= 0 ; j--)
{
ASSERT (Cols2+j2 >= Cols+j) ;
if (Cols [j] >= 0)
{
Cols2 [--j2] = Cols [j] ;
}
}
ASSERT (j2 == 0) ;
/* -------------------------------------------------------------- */
/* construct the new header */
/* -------------------------------------------------------------- */
/* E [0...e] is now valid */
E [e] = (pdest + 1) - Numeric->Memory ;
epdest = (Element *) (pdest + 1) ;
epdest->next = EMPTY ; /* destroys the son list */
epdest->ncols = ncolsleft ;
epdest->nrows = nrowsleft ;
epdest->ncolsleft = ncolsleft ;
epdest->nrowsleft = nrowsleft ;
epdest->rdeg = rdeg ;
epdest->cdeg = cdeg ;
ASSERT (size2 <= size) ;
pdest->header.prevsize = 0 ;
pdest->header.size = size2 ;
DEBUG7 (("After moving it:\n")) ;
#ifndef NDEBUG
UMF_dump_element (Numeric, Work, e, FALSE) ;
#endif
}
#ifndef NDEBUG
else
{
DEBUG8 ((" free\n")) ;
}
#endif
DEBUG7 (("psrc "ID" tail "ID"\n",
(Int) (psrc-Numeric->Memory), Numeric->itail)) ;
}
ASSERT (psrc == Numeric->Memory + Numeric->itail) ;
ASSERT (nmark == 0) ;
/* ---------------------------------------------------------------------- */
/* final tail pointer */
/* ---------------------------------------------------------------------- */
ASSERT (pdest >= Numeric->Memory + Numeric->itail) ;
Numeric->itail = pdest - Numeric->Memory ;
pdest->header.prevsize = 0 ;
Numeric->ibig = EMPTY ;
Numeric->tail_usage = Numeric->size - Numeric->itail ;
/* ---------------------------------------------------------------------- */
/* clear the unused E [nel+1 .. Work->elen - 1] */
/* ---------------------------------------------------------------------- */
for (e = nel+1 ; e < Work->elen ; e++)
{
E [e] = 0 ;
}
#ifndef NDEBUG
UMF_dump_packed_memory (Numeric, Work) ;
#endif
DEBUG8 (("::::GARBAGE COLLECTION DONE::::\n")) ;
}
/*
* Function: tid_t OS_TaskCreate(uint32_t _stack_size, void *_ip, uint8 _priority)
*
* This routine creates a new thread and returns the TCB created
*/
tid_t OS_TaskCreate(uint32_t *_stackaddr, uint32_t _stacksize, void *_ip, void *_arg, uint8_t _priority)
{
OS_task_t *task;
OS_Node_t *n;
// DEBUG ("(_stackaddr=%p, _stacksize=%p, _ip=%p, _arg=%p, priority=%d)",
// _stackaddr, _stacksize, _ip, _arg, _priority);
ASSERT (_priority <= SCHED_MIN_PRIORITY);
/* get a free thread control block */
n = (OS_Node_t *) OS_ListGetNode (&tcb_list);
/* assert that we really have a free tcb */
ASSERT (n != NULL);
if (n == NULL) return -1;
task = GET_ELEMENT(OS_task_t, n, node);
TRACE (task->tid, "d");
KTRACE (KTRACE_CREATE, task->tid, _priority);
/* Stablish the thread stack */
if (_stackaddr == NULL)
{
if (_stacksize < CONFIG_ERCOS_MIN_STACK) _stacksize = CONFIG_ERCOS_MIN_STACK;
_stackaddr = OS_Malloc (_stacksize);
} else {
ASSERT (_stacksize >= CONFIG_ERCOS_MIN_STACK);
}
task->stack.size = _stacksize;
task->stack.addr = (uint8_t *)_stackaddr;
/* Fill the stack with the 0x5a mark */
#ifndef CONFIG_NDEBUG
// unsigned i;
// DEBUG ("Stack @ %p, %p bytes", _stackaddr, _stacksize);
// for (i = 0; i < _stacksize; i++) ((uint8_t *) _stackaddr)[i] = 0x5a;
#endif
/* Init the new thread context */
ercos_hal_hwcontext_init (&(task->context), task->stack.addr, task->stack.size, 0, OS_ThreadWrapper);
/* Stablish the thread status */
task->status = TS_READY;
/* Stablish the thread entry point */
task->entry_point = _ip;
task->arg = _arg;
/* Init the pointer to the catched mutex */
task->pmutex =(OS_Mutex_t*) NULL;
/* Set the thread base and temporal priorities */
task->priority = _priority;
task->base_priority = _priority;
/* Add the task to the correspond sched queue */
OS_SchedTaskReady(task->tid);
return task->tid;
}
GLOBAL void UMF_assemble
#else
GLOBAL void UMF_assemble_fixq
#endif
(
NumericType *Numeric,
WorkType *Work
)
{
/* ---------------------------------------------------------------------- */
/* local variables */
/* ---------------------------------------------------------------------- */
Int e, i, row, col, i2, nrows, ncols, f, tpi, extcdeg, extrdeg, rdeg0,
cdeg0, son_list, next, nrows_to_assemble,
ncols_to_assemble, ngetrows, j, j2,
nrowsleft, /* number of rows remaining in S */
ncolsleft, /* number of columns remaining in S */
prior_Lson, prior_Uson, *E, *Cols, *Rows, *Wm, *Woo,
*Row_tuples, *Row_degree, *Row_tlen,
*Col_tuples, *Col_tlen ;
Unit *Memory, *p ;
Element *ep ;
Tuple *tp, *tp1, *tp2, *tpend ;
Entry
*S, /* a pointer into the contribution block of a son */
*Fcblock, /* current contribution block */
*Fcol ; /* a column of FC */
Int *Frpos,
*Fcpos,
fnrows, /* number of rows in contribution block in F */
fncols ; /* number of columns in contribution block in F */
#if !defined (FIXQ) || !defined (NDEBUG)
Int *Col_degree ;
#endif
#ifndef NDEBUG
Int n_row, n_col ;
n_row = Work->n_row ;
n_col = Work->n_col ;
DEBUG3 (("::Assemble SCANS 1-4\n")) ;
UMF_dump_current_front (Numeric, Work, TRUE) ;
#endif
#if !defined (FIXQ) || !defined (NDEBUG)
Col_degree = Numeric->Cperm ; /* not updated if FIXQ is true */
#endif
/* ---------------------------------------------------------------------- */
/* get parameters */
/* ---------------------------------------------------------------------- */
fncols = Work->fncols ;
fnrows = Work->fnrows ;
Fcpos = Work->Fcpos ;
Frpos = Work->Frpos ;
Row_degree = Numeric->Rperm ;
Row_tuples = Numeric->Uip ;
Row_tlen = Numeric->Uilen ;
Col_tuples = Numeric->Lip ;
Col_tlen = Numeric->Lilen ;
E = Work->E ;
Memory = Numeric->Memory ;
Wm = Work->Wm ;
Woo = Work->Woo ;
rdeg0 = Work->rdeg0 ;
cdeg0 = Work->cdeg0 ;
#ifndef NDEBUG
DEBUG6 (("============================================\n")) ;
DEBUG6 (("Degree update, assembly.\n")) ;
DEBUG6 (("pivot row pattern: fncols="ID"\n", fncols)) ;
for (j = 0 ; j < fncols ; j++)
{
col = Work->Fcols [j] ;
DEBUG6 ((ID" ", col)) ;
ASSERT (Fcpos [col] == j * Work->fnr_curr) ;
ASSERT (NON_PIVOTAL_COL (col)) ;
}
ASSERT (Fcpos [Work->pivcol] >= 0) ;
DEBUG6 (("pivcol: "ID" pos "ID" fnr_curr "ID" fncols "ID"\n",
Work->pivcol, Fcpos [Work->pivcol], Work->fnr_curr, fncols)) ;
ASSERT (Fcpos [Work->pivcol] < fncols * Work->fnr_curr) ;
DEBUG6 (("\npivot col pattern: fnrows="ID"\n", fnrows)) ;
for (i = 0 ; i < fnrows ; i++)
{
row = Work->Frows [i] ;
DEBUG6 ((ID" ", row)) ;
ASSERT (Frpos [row] == i) ;
ASSERT (NON_PIVOTAL_ROW (row)) ;
}
DEBUG6 (("\n")) ;
ASSERT (Frpos [Work->pivrow] >= 0) ;
ASSERT (Frpos [Work->pivrow] < fnrows) ;
ASSERT (Work->Flublock == (Entry *) (Numeric->Memory + E [0])) ;
ASSERT (Work->Fcblock == Work->Flublock + Work->nb *
(Work->nb + Work->fnr_curr + Work->fnc_curr)) ;
#endif
Fcblock = Work->Fcblock ;
/* ---------------------------------------------------------------------- */
/* determine the largest actual frontal matrix size (for Info only) */
/* ---------------------------------------------------------------------- */
ASSERT (fnrows == Work->fnrows_new + 1) ;
ASSERT (fncols == Work->fncols_new + 1) ;
Numeric->maxnrows = MAX (Numeric->maxnrows, fnrows) ;
Numeric->maxncols = MAX (Numeric->maxncols, fncols) ;
/* this is safe from integer overflow, since the current frontal matrix
* is already allocated. */
Numeric->maxfrsize = MAX (Numeric->maxfrsize, fnrows * fncols) ;
/* ---------------------------------------------------------------------- */
/* assemble from prior elements into the current frontal matrix */
/* ---------------------------------------------------------------------- */
DEBUG2 (("New assemble start [prior_element:"ID"\n", Work->prior_element)) ;
/* Currently no rows or columns are marked. No elements are scanned, */
/* that is, (ep->next == EMPTY) is true for all elements */
son_list = 0 ; /* start creating son_list [ */
/* ---------------------------------------------------------------------- */
/* determine if most recent element is Lson or Uson of current front */
/* ---------------------------------------------------------------------- */
if (!Work->do_extend)
{
prior_Uson = ( Work->pivcol_in_front && !Work->pivrow_in_front) ;
prior_Lson = (!Work->pivcol_in_front && Work->pivrow_in_front) ;
if (prior_Uson || prior_Lson)
{
e = Work->prior_element ;
if (e != EMPTY)
{
ASSERT (E [e]) ;
p = Memory + E [e] ;
ep = (Element *) p ;
ep->next = son_list ;
son_list = e ;
#ifndef NDEBUG
DEBUG2 (("e "ID" is Prior son "ID" "ID"\n",
e, prior_Uson, prior_Lson)) ;
UMF_dump_element (Numeric, Work, e, FALSE) ;
#endif
ASSERT (E [e]) ;
}
}
}
Work->prior_element = EMPTY ;
/* ---------------------------------------------------------------------- */
/* SCAN1-row: scan the element lists of each new row in the pivot col */
/* and compute the external column degree for each frontal */
/* ---------------------------------------------------------------------- */
for (i2 = Work->fscan_row ; i2 < fnrows ; i2++)
{
/* Get a row */
row = Work->NewRows [i2] ;
if (row < 0) row = FLIP (row) ;
ASSERT (row >= 0 && row < n_row) ;
DEBUG6 (("SCAN1-row: "ID"\n", row)) ;
#ifndef NDEBUG
UMF_dump_rowcol (0, Numeric, Work, row, FALSE) ;
#endif
ASSERT (NON_PIVOTAL_ROW (row)) ;
tpi = Row_tuples [row] ;
if (!tpi) continue ;
tp = (Tuple *) (Memory + tpi) ;
tp1 = tp ;
tp2 = tp ;
tpend = tp + Row_tlen [row] ;
for ( ; tp < tpend ; tp++)
{
e = tp->e ;
ASSERT (e > 0 && e <= Work->nel) ;
if (!E [e]) continue ; /* element already deallocated */
f = tp->f ;
p = Memory + E [e] ;
ep = (Element *) p ;
p += UNITS (Element, 1) ;
Rows = ((Int *) p) + ep->ncols ;
if (Rows [f] == EMPTY) continue ; /* row already assembled */
ASSERT (row == Rows [f]) ;
if (ep->cdeg < cdeg0)
{
/* first time seen in scan1-row */
ep->cdeg = ep->nrowsleft + cdeg0 ;
DEBUG6 (("e "ID" First seen: cdeg: "ID" ", e, ep->cdeg-cdeg0)) ;
ASSERT (ep->ncolsleft > 0 && ep->nrowsleft > 0) ;
}
ep->cdeg-- ; /* decrement external column degree */
DEBUG6 (("e "ID" New ext col deg: "ID"\n", e, ep->cdeg - cdeg0)) ;
/* this element is not yet in the new son list */
if (ep->cdeg == cdeg0 && ep->next == EMPTY)
{
/* A new LUson or Uson has been found */
ep->next = son_list ;
son_list = e ;
}
ASSERT (ep->cdeg >= cdeg0) ;
*tp2++ = *tp ; /* leave the tuple in the list */
}
Row_tlen [row] = tp2 - tp1 ;
}
/* ---------------------------------------------------------------------- */
/* SCAN1-col: scan the element lists of each new col in the pivot row */
/* and compute the external row degree for each frontal */
/* ---------------------------------------------------------------------- */
for (j2 = Work->fscan_col ; j2 < fncols ; j2++)
{
/* Get a column */
col = Work->NewCols [j2] ;
if (col < 0) col = FLIP (col) ;
ASSERT (col >= 0 && col < n_col) ;
DEBUG6 (("SCAN 1-col: "ID"\n", col)) ;
#ifndef NDEBUG
UMF_dump_rowcol (1, Numeric, Work, col, FALSE) ;
#endif
ASSERT (NON_PIVOTAL_COL (col)) ;
tpi = Col_tuples [col] ;
if (!tpi) continue ;
tp = (Tuple *) (Memory + tpi) ;
tp1 = tp ;
tp2 = tp ;
tpend = tp + Col_tlen [col] ;
for ( ; tp < tpend ; tp++)
{
e = tp->e ;
ASSERT (e > 0 && e <= Work->nel) ;
if (!E [e]) continue ; /* element already deallocated */
f = tp->f ;
p = Memory + E [e] ;
ep = (Element *) p ;
p += UNITS (Element, 1) ;
Cols = (Int *) p ;
if (Cols [f] == EMPTY) continue ; /* column already assembled */
ASSERT (col == Cols [f]) ;
if (ep->rdeg < rdeg0)
{
/* first time seen in scan1-col */
ep->rdeg = ep->ncolsleft + rdeg0 ;
DEBUG6 (("e "ID" First seen: rdeg: "ID" ", e, ep->rdeg-rdeg0)) ;
ASSERT (ep->ncolsleft > 0 && ep->nrowsleft > 0) ;
}
ep->rdeg-- ; /* decrement external row degree */
DEBUG6 (("e "ID" New ext row degree: "ID"\n", e, ep->rdeg-rdeg0)) ;
if (ep->rdeg == rdeg0 && ep->next == EMPTY)
{
/* A new LUson or Lson has been found */
ep->next = son_list ;
son_list = e ;
}
ASSERT (ep->rdeg >= rdeg0) ;
*tp2++ = *tp ; /* leave the tuple in the list */
}
Col_tlen [col] = tp2 - tp1 ;
}
/* ---------------------------------------------------------------------- */
/* assemble new sons via full scans */
/* ---------------------------------------------------------------------- */
next = EMPTY ;
for (e = son_list ; e > 0 ; e = next)
{
ASSERT (e > 0 && e <= Work->nel && E [e]) ;
p = Memory + E [e] ;
DEBUG2 (("New son: "ID"\n", e)) ;
#ifndef NDEBUG
UMF_dump_element (Numeric, Work, e, FALSE) ;
#endif
GET_ELEMENT (ep, p, Cols, Rows, ncols, nrows, S) ;
nrowsleft = ep->nrowsleft ;
ncolsleft = ep->ncolsleft ;
next = ep->next ;
ep->next = EMPTY ;
extrdeg = (ep->rdeg < rdeg0) ? ncolsleft : (ep->rdeg - rdeg0) ;
extcdeg = (ep->cdeg < cdeg0) ? nrowsleft : (ep->cdeg - cdeg0) ;
ncols_to_assemble = ncolsleft - extrdeg ;
nrows_to_assemble = nrowsleft - extcdeg ;
DEBUG2 (("extrdeg "ID" extcdeg "ID"\n", extrdeg, extcdeg)) ;
if (extrdeg == 0 && extcdeg == 0)
{
/* -------------------------------------------------------------- */
/* this is an LUson - assemble an entire contribution block */
/* -------------------------------------------------------------- */
DEBUG6 (("LUson found: "ID"\n", e)) ;
if (nrows == nrowsleft)
{
/* ---------------------------------------------------------- */
/* no rows assembled out of this LUson yet */
/* ---------------------------------------------------------- */
/* compute the compressed column offset vector*/
/* [ use Wm [0..nrows-1] for offsets */
#pragma ivdep
for (i = 0 ; i < nrows ; i++)
{
row = Rows [i] ;
Row_degree [row] -= ncolsleft ;
Wm [i] = Frpos [row] ;
}
if (ncols == ncolsleft)
{
/* ------------------------------------------------------ */
/* no rows or cols assembled out of LUson yet */
/* ------------------------------------------------------ */
for (j = 0 ; j < ncols ; j++)
{
col = Cols [j] ;
#ifndef FIXQ
Col_degree [col] -= nrowsleft ;
#endif
Fcol = Fcblock + Fcpos [col] ;
#pragma ivdep
for (i = 0 ; i < nrows ; i++)
{
/* Fcol [Wm [i]] += S [i] ; */
ASSEMBLE (Fcol [Wm [i]], S [i]) ;
}
S += nrows ;
}
}
else
{
/* ------------------------------------------------------ */
/* only cols have been assembled out of LUson */
/* ------------------------------------------------------ */
for (j = 0 ; j < ncols ; j++)
{
col = Cols [j] ;
if (col >= 0)
{
#ifndef FIXQ
Col_degree [col] -= nrowsleft ;
#endif
Fcol = Fcblock + Fcpos [col] ;
#pragma ivdep
for (i = 0 ; i < nrows ; i++)
{
/* Fcol [Wm [i]] += S [i] ; */
ASSEMBLE (Fcol [Wm [i]], S [i]) ;
}
}
S += nrows ;
}
}
/* ] done using Wm [0..nrows-1] for offsets */
}
else
{
/* ---------------------------------------------------------- */
/* some rows have been assembled out of this LUson */
/* ---------------------------------------------------------- */
/* compute the compressed column offset vector*/
/* [ use Woo,Wm [0..nrowsleft-1] for offsets */
ngetrows = 0 ;
for (i = 0 ; i < nrows ; i++)
{
row = Rows [i] ;
if (row >= 0)
{
Row_degree [row] -= ncolsleft ;
Woo [ngetrows] = i ;
Wm [ngetrows++] = Frpos [row] ;
}
}
ASSERT (ngetrows == nrowsleft) ;
if (ncols == ncolsleft)
{
/* ------------------------------------------------------ */
/* only rows have been assembled out of this LUson */
/* ------------------------------------------------------ */
for (j = 0 ; j < ncols ; j++)
{
col = Cols [j] ;
#ifndef FIXQ
Col_degree [col] -= nrowsleft ;
#endif
Fcol = Fcblock + Fcpos [col] ;
#pragma ivdep
for (i = 0 ; i < nrowsleft ; i++)
{
/* Fcol [Wm [i]] += S [Woo [i]] ; */
ASSEMBLE (Fcol [Wm [i]], S [Woo [i]]) ;
}
S += nrows ;
}
}
else
{
/* ------------------------------------------------------ */
/* both rows and columns have been assembled out of LUson */
/* ------------------------------------------------------ */
for (j = 0 ; j < ncols ; j++)
{
col = Cols [j] ;
if (col >= 0)
{
#ifndef FIXQ
Col_degree [col] -= nrowsleft ;
#endif
Fcol = Fcblock + Fcpos [col] ;
#pragma ivdep
for (i = 0 ; i < nrowsleft ; i++)
{
/* Fcol [Wm [i]] += S [Woo [i]] ; */
ASSEMBLE (Fcol [Wm [i]], S [Woo [i]]) ;
}
}
S += nrows ;
}
}
/* ] done using Woo,Wm [0..nrowsleft-1] */
}
/* deallocate the element: remove from ordered list */
UMF_mem_free_tail_block (Numeric, E [e]) ;
E [e] = 0 ;
}
else if (extcdeg == 0)
{
/* -------------------------------------------------------------- */
/* this is a Uson - assemble all possible columns */
/* -------------------------------------------------------------- */
DEBUG6 (("New USON: "ID"\n", e)) ;
ASSERT (extrdeg > 0) ;
DEBUG6 (("New uson "ID" cols to do "ID"\n", e, ncols_to_assemble)) ;
if (ncols_to_assemble > 0)
{
Int skip = FALSE ;
if (ncols_to_assemble * 16 < ncols && nrows == 1)
{
/* this is a tall and thin frontal matrix consisting of
* only one column (most likely an original column). Do
* not assemble it. It cannot be the pivot column, since
* the pivot column element would be an LU son, not an Lson,
* of the current frontal matrix. */
ASSERT (nrowsleft == 1) ;
ASSERT (Rows [0] >= 0 && Rows [0] < Work->n_row) ;
skip = TRUE ;
Work->any_skip = TRUE ;
}
if (!skip)
{
if (nrows == nrowsleft)
{
/* -------------------------------------------------- */
/* no rows have been assembled out of this Uson yet */
/* -------------------------------------------------- */
/* compute the compressed column offset vector */
/* [ use Wm [0..nrows-1] for offsets */
#pragma ivdep
for (i = 0 ; i < nrows ; i++)
{
row = Rows [i] ;
ASSERT (row >= 0 && row < n_row) ;
Row_degree [row] -= ncols_to_assemble ;
Wm [i] = Frpos [row] ;
}
for (j = 0 ; j < ncols ; j++)
{
col = Cols [j] ;
if ((col >= 0) && (Fcpos [col] >= 0))
{
#ifndef FIXQ
Col_degree [col] -= nrowsleft ;
#endif
Fcol = Fcblock + Fcpos [col] ;
#pragma ivdep
for (i = 0 ; i < nrows ; i++)
{
/* Fcol [Wm [i]] += S [i] ; */
ASSEMBLE (Fcol [Wm [i]], S [i]) ;
}
/* flag the column as assembled from Uson */
Cols [j] = EMPTY ;
}
S += nrows ;
}
/* ] done using Wm [0..nrows-1] for offsets */
}
else
{
/* -------------------------------------------------- */
/* some rows have been assembled out of this Uson */
/* -------------------------------------------------- */
/* compute the compressed column offset vector*/
/* [ use Woo,Wm [0..nrows-1] for offsets */
ngetrows = 0 ;
for (i = 0 ; i < nrows ; i++)
{
row = Rows [i] ;
if (row >= 0)
{
Row_degree [row] -= ncols_to_assemble ;
ASSERT (row < n_row && Frpos [row] >= 0) ;
Woo [ngetrows] = i ;
Wm [ngetrows++] = Frpos [row] ;
}
}
ASSERT (ngetrows == nrowsleft) ;
for (j = 0 ; j < ncols ; j++)
{
col = Cols [j] ;
if ((col >= 0) && (Fcpos [col] >= 0))
{
#ifndef FIXQ
Col_degree [col] -= nrowsleft ;
#endif
Fcol = Fcblock + Fcpos [col] ;
#pragma ivdep
for (i = 0 ; i < nrowsleft ; i++)
{
/* Fcol [Wm [i]] += S [Woo [i]] ; */
ASSEMBLE (Fcol [Wm [i]], S [Woo [i]]) ;
}
/* flag the column as assembled from Uson */
Cols [j] = EMPTY ;
}
S += nrows ;
}
/* ] done using Woo,Wm */
}
ep->ncolsleft = extrdeg ;
}
}
}
else
{
/* -------------------------------------------------------------- */
/* this is an Lson - assemble all possible rows */
/* -------------------------------------------------------------- */
DEBUG6 (("New LSON: "ID"\n", e)) ;
ASSERT (extrdeg == 0 && extcdeg > 0) ;
DEBUG6 (("New lson "ID" rows to do "ID"\n", e, nrows_to_assemble)) ;
if (nrows_to_assemble > 0)
{
Int skip = FALSE ;
if (nrows_to_assemble * 16 < nrows && ncols == 1)
{
/* this is a tall and thin frontal matrix consisting of
* only one column (most likely an original column). Do
* not assemble it. It cannot be the pivot column, since
* the pivot column element would be an LU son, not an Lson,
* of the current frontal matrix. */
ASSERT (ncolsleft == 1) ;
ASSERT (Cols [0] >= 0 && Cols [0] < Work->n_col) ;
Work->any_skip = TRUE ;
skip = TRUE ;
}
if (!skip)
{
/* compute the compressed column offset vector */
/* [ use Woo,Wm [0..nrows-1] for offsets */
ngetrows = 0 ;
for (i = 0 ; i < nrows ; i++)
{
row = Rows [i] ;
if ((row >= 0) && (Frpos [row] >= 0))
{
ASSERT (row < n_row) ;
Row_degree [row] -= ncolsleft ;
Woo [ngetrows] = i ;
Wm [ngetrows++] = Frpos [row] ;
/* flag the row as assembled from the Lson */
Rows [i] = EMPTY ;
}
}
ASSERT (nrowsleft - ngetrows == extcdeg) ;
ASSERT (ngetrows == nrows_to_assemble) ;
if (ncols == ncolsleft)
{
/* -------------------------------------------------- */
/* no columns assembled out this Lson yet */
/* -------------------------------------------------- */
for (j = 0 ; j < ncols ; j++)
{
col = Cols [j] ;
ASSERT (col >= 0 && col < n_col) ;
#ifndef FIXQ
Col_degree [col] -= nrows_to_assemble ;
#endif
Fcol = Fcblock + Fcpos [col] ;
#pragma ivdep
for (i = 0 ; i < nrows_to_assemble ; i++)
{
/* Fcol [Wm [i]] += S [Woo [i]] ; */
ASSEMBLE (Fcol [Wm [i]], S [Woo [i]]) ;
}
S += nrows ;
}
}
else
{
/* -------------------------------------------------- */
/* some columns have been assembled out of this Lson */
/* -------------------------------------------------- */
for (j = 0 ; j < ncols ; j++)
{
col = Cols [j] ;
ASSERT (col < n_col) ;
if (col >= 0)
{
#ifndef FIXQ
Col_degree [col] -= nrows_to_assemble ;
#endif
Fcol = Fcblock + Fcpos [col] ;
#pragma ivdep
for (i = 0 ; i < nrows_to_assemble ; i++)
{
/* Fcol [Wm [i]] += S [Woo [i]] ; */
ASSEMBLE (Fcol [Wm [i]], S [Woo [i]]) ;
}
}
S += nrows ;
}
}
/* ] done using Woo,Wm */
ep->nrowsleft = extcdeg ;
}
}
}
}
/* Note that garbage collection, and build tuples */
/* both destroy the son list. */
/* ] son_list now empty */
/* ---------------------------------------------------------------------- */
/* If frontal matrix extended, assemble old L/Usons from new rows/cols */
/* ---------------------------------------------------------------------- */
/* ---------------------------------------------------------------------- */
/* SCAN2-row: assemble rows of old Lsons from the new rows */
/* ---------------------------------------------------------------------- */
#ifndef NDEBUG
DEBUG7 (("Current frontal matrix: (prior to scan2-row)\n")) ;
UMF_dump_current_front (Numeric, Work, TRUE) ;
#endif
/* rescan the pivot row */
if (Work->any_skip)
{
row_assemble (Work->pivrow, Numeric, Work) ;
}
if (Work->do_scan2row)
{
for (i2 = Work->fscan_row ; i2 < fnrows ; i2++)
{
/* Get a row */
row = Work->NewRows [i2] ;
if (row < 0) row = FLIP (row) ;
ASSERT (row >= 0 && row < n_row) ;
if (!(row == Work->pivrow && Work->any_skip))
{
/* assemble it */
row_assemble (row, Numeric, Work) ;
}
}
}
/* ---------------------------------------------------------------------- */
/* SCAN2-col: assemble columns of old Usons from the new columns */
/* ---------------------------------------------------------------------- */
#ifndef NDEBUG
DEBUG7 (("Current frontal matrix: (prior to scan2-col)\n")) ;
UMF_dump_current_front (Numeric, Work, TRUE) ;
#endif
/* rescan the pivot col */
if (Work->any_skip)
{
col_assemble (Work->pivcol, Numeric, Work) ;
}
if (Work->do_scan2col)
{
for (j2 = Work->fscan_col ; j2 < fncols ; j2++)
{
/* Get a column */
col = Work->NewCols [j2] ;
if (col < 0) col = FLIP (col) ;
ASSERT (col >= 0 && col < n_col) ;
if (!(col == Work->pivcol && Work->any_skip))
{
/* assemble it */
col_assemble (col, Numeric, Work) ;
}
}
}
/* ---------------------------------------------------------------------- */
/* done. the remainder of this routine is used only when in debug mode */
/* ---------------------------------------------------------------------- */
#ifndef NDEBUG
/* ---------------------------------------------------------------------- */
/* when debugging: make sure the assembly did everything that it could */
/* ---------------------------------------------------------------------- */
DEBUG3 (("::Assemble done\n")) ;
for (i2 = 0 ; i2 < fnrows ; i2++)
{
/* Get a row */
row = Work->Frows [i2] ;
ASSERT (row >= 0 && row < n_row) ;
DEBUG6 (("DEBUG SCAN 1: "ID"\n", row)) ;
UMF_dump_rowcol (0, Numeric, Work, row, TRUE) ;
ASSERT (NON_PIVOTAL_ROW (row)) ;
tpi = Row_tuples [row] ;
if (!tpi) continue ;
tp = (Tuple *) (Memory + tpi) ;
tpend = tp + Row_tlen [row] ;
for ( ; tp < tpend ; tp++)
{
e = tp->e ;
ASSERT (e > 0 && e <= Work->nel) ;
if (!E [e]) continue ; /* element already deallocated */
f = tp->f ;
p = Memory + E [e] ;
ep = (Element *) p ;
p += UNITS (Element, 1) ;
Cols = (Int *) p ;
Rows = ((Int *) p) + ep->ncols ;
if (Rows [f] == EMPTY) continue ; /* row already assembled */
ASSERT (row == Rows [f]) ;
extrdeg = (ep->rdeg < rdeg0) ? ep->ncolsleft : (ep->rdeg - rdeg0) ;
extcdeg = (ep->cdeg < cdeg0) ? ep->nrowsleft : (ep->cdeg - cdeg0) ;
DEBUG6 ((
"e "ID" After assembly ext row deg: "ID" ext col degree "ID"\n",
e, extrdeg, extcdeg)) ;
if (Work->any_skip)
{
/* no Lsons in any row, except for very tall and thin ones */
ASSERT (extrdeg >= 0) ;
if (extrdeg == 0)
{
/* this is an unassemble Lson */
ASSERT (ep->ncols == 1) ;
ASSERT (ep->ncolsleft == 1) ;
col = Cols [0] ;
ASSERT (col != Work->pivcol) ;
}
}
else
{
/* no Lsons in any row */
ASSERT (extrdeg > 0) ;
/* Uson external row degree is = number of cols left */
ASSERT (IMPLIES (extcdeg == 0, extrdeg == ep->ncolsleft)) ;
}
}
}
/* ---------------------------------------------------------------------- */
for (j2 = 0 ; j2 < fncols ; j2++)
{
/* Get a column */
col = Work->Fcols [j2] ;
ASSERT (col >= 0 && col < n_col) ;
DEBUG6 (("DEBUG SCAN 2: "ID"\n", col)) ;
#ifndef FIXQ
UMF_dump_rowcol (1, Numeric, Work, col, TRUE) ;
#else
UMF_dump_rowcol (1, Numeric, Work, col, FALSE) ;
#endif
ASSERT (NON_PIVOTAL_COL (col)) ;
tpi = Col_tuples [col] ;
if (!tpi) continue ;
tp = (Tuple *) (Memory + tpi) ;
tpend = tp + Col_tlen [col] ;
for ( ; tp < tpend ; tp++)
{
e = tp->e ;
ASSERT (e > 0 && e <= Work->nel) ;
if (!E [e]) continue ; /* element already deallocated */
f = tp->f ;
p = Memory + E [e] ;
ep = (Element *) p ;
p += UNITS (Element, 1) ;
Cols = (Int *) p ;
Rows = ((Int *) p) + ep->ncols ;
if (Cols [f] == EMPTY) continue ; /* column already assembled */
ASSERT (col == Cols [f]) ;
extrdeg = (ep->rdeg < rdeg0) ? ep->ncolsleft : (ep->rdeg - rdeg0) ;
extcdeg = (ep->cdeg < cdeg0) ? ep->nrowsleft : (ep->cdeg - cdeg0) ;
DEBUG6 (("e "ID" After assembly ext col deg: "ID"\n", e, extcdeg)) ;
if (Work->any_skip)
{
/* no Usons in any column, except for very tall and thin ones */
ASSERT (extcdeg >= 0) ;
if (extcdeg == 0)
{
/* this is an unassemble Uson */
ASSERT (ep->nrows == 1) ;
ASSERT (ep->nrowsleft == 1) ;
row = Rows [0] ;
ASSERT (row != Work->pivrow) ;
}
}
else
{
/* no Usons in any column */
ASSERT (extcdeg > 0) ;
/* Lson external column degree is = number of rows left */
ASSERT (IMPLIES (extrdeg == 0, extcdeg == ep->nrowsleft)) ;
}
}
}
#endif /* NDEBUG */
}
rmReal
RMtx4GetElement( const RMtx4* in, int row, int col )
{
return GET_ELEMENT( in, row, col );
}