int
sp_ztrsv(char *uplo, char *trans, char *diag, SuperMatrix *L,
SuperMatrix *U, doublecomplex *x, int *info)
{
/*
* Purpose
* =======
*
* sp_ztrsv() solves one of the systems of equations
* A*x = b, or A'*x = b,
* where b and x are n element vectors and A is a sparse unit , or
* non-unit, upper or lower triangular matrix.
* No test for singularity or near-singularity is included in this
* routine. Such tests must be performed before calling this routine.
*
* Parameters
* ==========
*
* uplo - (input) char*
* On entry, uplo specifies whether the matrix is an upper or
* lower triangular matrix as follows:
* uplo = 'U' or 'u' A is an upper triangular matrix.
* uplo = 'L' or 'l' A is a lower triangular matrix.
*
* trans - (input) char*
* On entry, trans specifies the equations to be solved as
* follows:
* trans = 'N' or 'n' A*x = b.
* trans = 'T' or 't' A'*x = b.
* trans = 'C' or 'c' A'*x = b.
*
* diag - (input) char*
* On entry, diag specifies whether or not A is unit
* triangular as follows:
* diag = 'U' or 'u' A is assumed to be unit triangular.
* diag = 'N' or 'n' A is not assumed to be unit
* triangular.
*
* L - (input) SuperMatrix*
* The factor L from the factorization Pr*A*Pc=L*U. Use
* compressed row subscripts storage for supernodes,
* i.e., L has types: Stype = SC, Dtype = SLU_Z, Mtype = TRLU.
*
* U - (input) SuperMatrix*
* The factor U from the factorization Pr*A*Pc=L*U.
* U has types: Stype = NC, Dtype = SLU_Z, Mtype = TRU.
*
* x - (input/output) doublecomplex*
* Before entry, the incremented array X must contain the n
* element right-hand side vector b. On exit, X is overwritten
* with the solution vector x.
*
* info - (output) int*
* If *info = -i, the i-th argument had an illegal value.
*
*/
#ifdef _CRAY
_fcd ftcs1 = _cptofcd("L", strlen("L")),
ftcs2 = _cptofcd("N", strlen("N")),
ftcs3 = _cptofcd("U", strlen("U"));
#endif
SCformat *Lstore;
NCformat *Ustore;
doublecomplex *Lval, *Uval;
int incx = 1, incy = 1;
doublecomplex alpha = {1.0, 0.0}, beta = {1.0, 0.0};
doublecomplex comp_zero = {0.0, 0.0};
int nrow;
int fsupc, nsupr, nsupc, luptr, istart, irow;
int i, k, iptr, jcol;
doublecomplex *work;
flops_t solve_ops;
extern SuperLUStat_t SuperLUStat;
/* Test the input parameters */
*info = 0;
if ( !lsame_(uplo,"L") && !lsame_(uplo, "U") ) *info = -1;
else if ( !lsame_(trans, "N") && !lsame_(trans, "T") ) *info = -2;
else if ( !lsame_(diag, "U") && !lsame_(diag, "N") ) *info = -3;
else if ( L->nrow != L->ncol || L->nrow < 0 ) *info = -4;
else if ( U->nrow != U->ncol || U->nrow < 0 ) *info = -5;
if ( *info ) {
i = -(*info);
xerbla_("sp_ztrsv", &i);
return 0;
}
Lstore = L->Store;
Lval = Lstore->nzval;
Ustore = U->Store;
Uval = Ustore->nzval;
solve_ops = 0;
if ( !(work = doublecomplexCalloc(L->nrow)) )
ABORT("Malloc fails for work in sp_ztrsv().");
if ( lsame_(trans, "N") ) { /* Form x := inv(A)*x. */
if ( lsame_(uplo, "L") ) {
/* Form x := inv(L)*x */
if ( L->nrow == 0 ) return 0; /* Quick return */
for (k = 0; k <= Lstore->nsuper; k++) {
fsupc = L_FST_SUPC(k);
istart = L_SUB_START(fsupc);
nsupr = L_SUB_START(fsupc+1) - istart;
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
nrow = nsupr - nsupc;
solve_ops += 4 * nsupc * (nsupc - 1);
solve_ops += 8 * nrow * nsupc;
if ( nsupc == 1 ) {
for (iptr=istart+1; iptr < L_SUB_START(fsupc+1); ++iptr) {
irow = L_SUB(iptr);
++luptr;
zz_mult(&comp_zero, &x[fsupc], &Lval[luptr]);
z_sub(&x[irow], &x[irow], &comp_zero);
}
} else {
#ifdef USE_VENDOR_BLAS
#ifdef _CRAY
CTRSV(ftcs1, ftcs2, ftcs3, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
CGEMV(ftcs2, &nrow, &nsupc, &alpha, &Lval[luptr+nsupc],
&nsupr, &x[fsupc], &incx, &beta, &work[0], &incy);
#else
ztrsv_("L", "N", "U", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
zgemv_("N", &nrow, &nsupc, &alpha, &Lval[luptr+nsupc],
&nsupr, &x[fsupc], &incx, &beta, &work[0], &incy);
#endif
#else
zlsolve ( nsupr, nsupc, &Lval[luptr], &x[fsupc]);
zmatvec ( nsupr, nsupr-nsupc, nsupc, &Lval[luptr+nsupc],
&x[fsupc], &work[0] );
#endif
iptr = istart + nsupc;
for (i = 0; i < nrow; ++i, ++iptr) {
irow = L_SUB(iptr);
z_sub(&x[irow], &x[irow], &work[i]); /* Scatter */
work[i] = comp_zero;
}
}
} /* for k ... */
} else {
/* Form x := inv(U)*x */
if ( U->nrow == 0 ) return 0; /* Quick return */
for (k = Lstore->nsuper; k >= 0; k--) {
fsupc = L_FST_SUPC(k);
nsupr = L_SUB_START(fsupc+1) - L_SUB_START(fsupc);
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
solve_ops += 4 * nsupc * (nsupc + 1);
if ( nsupc == 1 ) {
z_div(&x[fsupc], &x[fsupc], &Lval[luptr]);
for (i = U_NZ_START(fsupc); i < U_NZ_START(fsupc+1); ++i) {
irow = U_SUB(i);
zz_mult(&comp_zero, &x[fsupc], &Uval[i]);
z_sub(&x[irow], &x[irow], &comp_zero);
}
} else {
#ifdef USE_VENDOR_BLAS
#ifdef _CRAY
CTRSV(ftcs3, ftcs2, ftcs2, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#else
ztrsv_("U", "N", "N", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#endif
#else
zusolve ( nsupr, nsupc, &Lval[luptr], &x[fsupc] );
#endif
for (jcol = fsupc; jcol < L_FST_SUPC(k+1); jcol++) {
solve_ops += 8*(U_NZ_START(jcol+1) - U_NZ_START(jcol));
for (i = U_NZ_START(jcol); i < U_NZ_START(jcol+1);
i++) {
irow = U_SUB(i);
zz_mult(&comp_zero, &x[jcol], &Uval[i]);
z_sub(&x[irow], &x[irow], &comp_zero);
}
}
}
} /* for k ... */
}
} else { /* Form x := inv(A')*x */
if ( lsame_(uplo, "L") ) {
/* Form x := inv(L')*x */
if ( L->nrow == 0 ) return 0; /* Quick return */
for (k = Lstore->nsuper; k >= 0; --k) {
fsupc = L_FST_SUPC(k);
istart = L_SUB_START(fsupc);
nsupr = L_SUB_START(fsupc+1) - istart;
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
solve_ops += 8 * (nsupr - nsupc) * nsupc;
for (jcol = fsupc; jcol < L_FST_SUPC(k+1); jcol++) {
iptr = istart + nsupc;
for (i = L_NZ_START(jcol) + nsupc;
i < L_NZ_START(jcol+1); i++) {
irow = L_SUB(iptr);
zz_mult(&comp_zero, &x[irow], &Lval[i]);
z_sub(&x[jcol], &x[jcol], &comp_zero);
iptr++;
}
}
if ( nsupc > 1 ) {
solve_ops += 4 * nsupc * (nsupc - 1);
#ifdef _CRAY
ftcs1 = _cptofcd("L", strlen("L"));
ftcs2 = _cptofcd("T", strlen("T"));
ftcs3 = _cptofcd("U", strlen("U"));
CTRSV(ftcs1, ftcs2, ftcs3, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#else
ztrsv_("L", "T", "U", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#endif
}
}
} else {
/* Form x := inv(U')*x */
if ( U->nrow == 0 ) return 0; /* Quick return */
for (k = 0; k <= Lstore->nsuper; k++) {
fsupc = L_FST_SUPC(k);
nsupr = L_SUB_START(fsupc+1) - L_SUB_START(fsupc);
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
for (jcol = fsupc; jcol < L_FST_SUPC(k+1); jcol++) {
solve_ops += 8*(U_NZ_START(jcol+1) - U_NZ_START(jcol));
for (i = U_NZ_START(jcol); i < U_NZ_START(jcol+1); i++) {
irow = U_SUB(i);
zz_mult(&comp_zero, &x[irow], &Uval[i]);
z_sub(&x[jcol], &x[jcol], &comp_zero);
}
}
solve_ops += 4 * nsupc * (nsupc + 1);
if ( nsupc == 1 ) {
z_div(&x[fsupc], &x[fsupc], &Lval[luptr]);
} else {
#ifdef _CRAY
ftcs1 = _cptofcd("U", strlen("U"));
ftcs2 = _cptofcd("T", strlen("T"));
ftcs3 = _cptofcd("N", strlen("N"));
CTRSV( ftcs1, ftcs2, ftcs3, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#else
ztrsv_("U", "T", "N", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#endif
}
} /* for k ... */
}
}
SuperLUStat.ops[SOLVE] += solve_ops;
SUPERLU_FREE(work);
return 0;
}
/*! \brief Solves one of the systems of equations A*x = b, or A'*x = b
*
* <pre>
* Purpose
* =======
*
* sp_ztrsv() solves one of the systems of equations
* A*x = b, or A'*x = b,
* where b and x are n element vectors and A is a sparse unit , or
* non-unit, upper or lower triangular matrix.
* No test for singularity or near-singularity is included in this
* routine. Such tests must be performed before calling this routine.
*
* Parameters
* ==========
*
* uplo - (input) char*
* On entry, uplo specifies whether the matrix is an upper or
* lower triangular matrix as follows:
* uplo = 'U' or 'u' A is an upper triangular matrix.
* uplo = 'L' or 'l' A is a lower triangular matrix.
*
* trans - (input) char*
* On entry, trans specifies the equations to be solved as
* follows:
* trans = 'N' or 'n' A*x = b.
* trans = 'T' or 't' A'*x = b.
* trans = 'C' or 'c' A^H*x = b.
*
* diag - (input) char*
* On entry, diag specifies whether or not A is unit
* triangular as follows:
* diag = 'U' or 'u' A is assumed to be unit triangular.
* diag = 'N' or 'n' A is not assumed to be unit
* triangular.
*
* L - (input) SuperMatrix*
* The factor L from the factorization Pr*A*Pc=L*U. Use
* compressed row subscripts storage for supernodes,
* i.e., L has types: Stype = SC, Dtype = SLU_Z, Mtype = TRLU.
*
* U - (input) SuperMatrix*
* The factor U from the factorization Pr*A*Pc=L*U.
* U has types: Stype = NC, Dtype = SLU_Z, Mtype = TRU.
*
* x - (input/output) doublecomplex*
* Before entry, the incremented array X must contain the n
* element right-hand side vector b. On exit, X is overwritten
* with the solution vector x.
*
* info - (output) int*
* If *info = -i, the i-th argument had an illegal value.
* </pre>
*/
int
sp_ztrsv(char *uplo, char *trans, char *diag, SuperMatrix *L,
SuperMatrix *U, doublecomplex *x, SuperLUStat_t *stat, int *info)
{
#ifdef _CRAY
_fcd ftcs1 = _cptofcd("L", strlen("L")),
ftcs2 = _cptofcd("N", strlen("N")),
ftcs3 = _cptofcd("U", strlen("U"));
#endif
SCformat *Lstore;
NCformat *Ustore;
doublecomplex *Lval, *Uval;
int incx = 1, incy = 1;
doublecomplex temp;
doublecomplex alpha = {1.0, 0.0}, beta = {1.0, 0.0};
doublecomplex comp_zero = {0.0, 0.0};
int nrow;
int fsupc, nsupr, nsupc, luptr, istart, irow;
int i, k, iptr, jcol;
doublecomplex *work;
flops_t solve_ops;
/* Test the input parameters */
*info = 0;
if ( !lsame_(uplo,"L") && !lsame_(uplo, "U") ) *info = -1;
else if ( !lsame_(trans, "N") && !lsame_(trans, "T") &&
!lsame_(trans, "C")) *info = -2;
else if ( !lsame_(diag, "U") && !lsame_(diag, "N") ) *info = -3;
else if ( L->nrow != L->ncol || L->nrow < 0 ) *info = -4;
else if ( U->nrow != U->ncol || U->nrow < 0 ) *info = -5;
if ( *info ) {
i = -(*info);
xerbla_("sp_ztrsv", &i);
return 0;
}
Lstore = L->Store;
Lval = Lstore->nzval;
Ustore = U->Store;
Uval = Ustore->nzval;
solve_ops = 0;
if ( !(work = doublecomplexCalloc(L->nrow)) )
ABORT("Malloc fails for work in sp_ztrsv().");
if ( lsame_(trans, "N") ) { /* Form x := inv(A)*x. */
if ( lsame_(uplo, "L") ) {
/* Form x := inv(L)*x */
if ( L->nrow == 0 ) return 0; /* Quick return */
for (k = 0; k <= Lstore->nsuper; k++) {
fsupc = L_FST_SUPC(k);
istart = L_SUB_START(fsupc);
nsupr = L_SUB_START(fsupc+1) - istart;
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
nrow = nsupr - nsupc;
/* 1 z_div costs 10 flops */
solve_ops += 4 * nsupc * (nsupc - 1) + 10 * nsupc;
solve_ops += 8 * nrow * nsupc;
if ( nsupc == 1 ) {
for (iptr=istart+1; iptr < L_SUB_START(fsupc+1); ++iptr) {
irow = L_SUB(iptr);
++luptr;
zz_mult(&comp_zero, &x[fsupc], &Lval[luptr]);
z_sub(&x[irow], &x[irow], &comp_zero);
}
} else {
#ifdef USE_VENDOR_BLAS
#ifdef _CRAY
CTRSV(ftcs1, ftcs2, ftcs3, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
CGEMV(ftcs2, &nrow, &nsupc, &alpha, &Lval[luptr+nsupc],
&nsupr, &x[fsupc], &incx, &beta, &work[0], &incy);
#else
ztrsv_("L", "N", "U", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
zgemv_("N", &nrow, &nsupc, &alpha, &Lval[luptr+nsupc],
&nsupr, &x[fsupc], &incx, &beta, &work[0], &incy);
#endif
#else
zlsolve ( nsupr, nsupc, &Lval[luptr], &x[fsupc]);
zmatvec ( nsupr, nsupr-nsupc, nsupc, &Lval[luptr+nsupc],
&x[fsupc], &work[0] );
#endif
iptr = istart + nsupc;
for (i = 0; i < nrow; ++i, ++iptr) {
irow = L_SUB(iptr);
z_sub(&x[irow], &x[irow], &work[i]); /* Scatter */
work[i] = comp_zero;
}
}
} /* for k ... */
} else {
/* Form x := inv(U)*x */
if ( U->nrow == 0 ) return 0; /* Quick return */
for (k = Lstore->nsuper; k >= 0; k--) {
fsupc = L_FST_SUPC(k);
nsupr = L_SUB_START(fsupc+1) - L_SUB_START(fsupc);
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
/* 1 z_div costs 10 flops */
solve_ops += 4 * nsupc * (nsupc + 1) + 10 * nsupc;
if ( nsupc == 1 ) {
z_div(&x[fsupc], &x[fsupc], &Lval[luptr]);
for (i = U_NZ_START(fsupc); i < U_NZ_START(fsupc+1); ++i) {
irow = U_SUB(i);
zz_mult(&comp_zero, &x[fsupc], &Uval[i]);
z_sub(&x[irow], &x[irow], &comp_zero);
}
} else {
#ifdef USE_VENDOR_BLAS
#ifdef _CRAY
CTRSV(ftcs3, ftcs2, ftcs2, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#else
ztrsv_("U", "N", "N", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#endif
#else
zusolve ( nsupr, nsupc, &Lval[luptr], &x[fsupc] );
#endif
for (jcol = fsupc; jcol < L_FST_SUPC(k+1); jcol++) {
solve_ops += 8*(U_NZ_START(jcol+1) - U_NZ_START(jcol));
for (i = U_NZ_START(jcol); i < U_NZ_START(jcol+1);
i++) {
irow = U_SUB(i);
zz_mult(&comp_zero, &x[jcol], &Uval[i]);
z_sub(&x[irow], &x[irow], &comp_zero);
}
}
}
} /* for k ... */
}
} else if ( lsame_(trans, "T") ) { /* Form x := inv(A')*x */
if ( lsame_(uplo, "L") ) {
/* Form x := inv(L')*x */
if ( L->nrow == 0 ) return 0; /* Quick return */
for (k = Lstore->nsuper; k >= 0; --k) {
fsupc = L_FST_SUPC(k);
istart = L_SUB_START(fsupc);
nsupr = L_SUB_START(fsupc+1) - istart;
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
solve_ops += 8 * (nsupr - nsupc) * nsupc;
for (jcol = fsupc; jcol < L_FST_SUPC(k+1); jcol++) {
iptr = istart + nsupc;
for (i = L_NZ_START(jcol) + nsupc;
i < L_NZ_START(jcol+1); i++) {
irow = L_SUB(iptr);
zz_mult(&comp_zero, &x[irow], &Lval[i]);
z_sub(&x[jcol], &x[jcol], &comp_zero);
iptr++;
}
}
if ( nsupc > 1 ) {
solve_ops += 4 * nsupc * (nsupc - 1);
#ifdef _CRAY
ftcs1 = _cptofcd("L", strlen("L"));
ftcs2 = _cptofcd("T", strlen("T"));
ftcs3 = _cptofcd("U", strlen("U"));
CTRSV(ftcs1, ftcs2, ftcs3, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#else
ztrsv_("L", "T", "U", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#endif
}
}
} else {
/* Form x := inv(U')*x */
if ( U->nrow == 0 ) return 0; /* Quick return */
for (k = 0; k <= Lstore->nsuper; k++) {
fsupc = L_FST_SUPC(k);
nsupr = L_SUB_START(fsupc+1) - L_SUB_START(fsupc);
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
for (jcol = fsupc; jcol < L_FST_SUPC(k+1); jcol++) {
solve_ops += 8*(U_NZ_START(jcol+1) - U_NZ_START(jcol));
for (i = U_NZ_START(jcol); i < U_NZ_START(jcol+1); i++) {
irow = U_SUB(i);
zz_mult(&comp_zero, &x[irow], &Uval[i]);
z_sub(&x[jcol], &x[jcol], &comp_zero);
}
}
/* 1 z_div costs 10 flops */
solve_ops += 4 * nsupc * (nsupc + 1) + 10 * nsupc;
if ( nsupc == 1 ) {
z_div(&x[fsupc], &x[fsupc], &Lval[luptr]);
} else {
#ifdef _CRAY
ftcs1 = _cptofcd("U", strlen("U"));
ftcs2 = _cptofcd("T", strlen("T"));
ftcs3 = _cptofcd("N", strlen("N"));
CTRSV( ftcs1, ftcs2, ftcs3, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#else
ztrsv_("U", "T", "N", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#endif
}
} /* for k ... */
}
} else { /* Form x := conj(inv(A'))*x */
if ( lsame_(uplo, "L") ) {
/* Form x := conj(inv(L'))*x */
if ( L->nrow == 0 ) return 0; /* Quick return */
for (k = Lstore->nsuper; k >= 0; --k) {
fsupc = L_FST_SUPC(k);
istart = L_SUB_START(fsupc);
nsupr = L_SUB_START(fsupc+1) - istart;
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
solve_ops += 8 * (nsupr - nsupc) * nsupc;
for (jcol = fsupc; jcol < L_FST_SUPC(k+1); jcol++) {
iptr = istart + nsupc;
for (i = L_NZ_START(jcol) + nsupc;
i < L_NZ_START(jcol+1); i++) {
irow = L_SUB(iptr);
zz_conj(&temp, &Lval[i]);
zz_mult(&comp_zero, &x[irow], &temp);
z_sub(&x[jcol], &x[jcol], &comp_zero);
iptr++;
}
}
if ( nsupc > 1 ) {
solve_ops += 4 * nsupc * (nsupc - 1);
#ifdef _CRAY
ftcs1 = _cptofcd("L", strlen("L"));
ftcs2 = _cptofcd(trans, strlen("T"));
ftcs3 = _cptofcd("U", strlen("U"));
ZTRSV(ftcs1, ftcs2, ftcs3, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#else
ztrsv_("L", trans, "U", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#endif
}
}
} else {
/* Form x := conj(inv(U'))*x */
if ( U->nrow == 0 ) return 0; /* Quick return */
for (k = 0; k <= Lstore->nsuper; k++) {
fsupc = L_FST_SUPC(k);
nsupr = L_SUB_START(fsupc+1) - L_SUB_START(fsupc);
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
for (jcol = fsupc; jcol < L_FST_SUPC(k+1); jcol++) {
solve_ops += 8*(U_NZ_START(jcol+1) - U_NZ_START(jcol));
for (i = U_NZ_START(jcol); i < U_NZ_START(jcol+1); i++) {
irow = U_SUB(i);
zz_conj(&temp, &Uval[i]);
zz_mult(&comp_zero, &x[irow], &temp);
z_sub(&x[jcol], &x[jcol], &comp_zero);
}
}
/* 1 z_div costs 10 flops */
solve_ops += 4 * nsupc * (nsupc + 1) + 10 * nsupc;
if ( nsupc == 1 ) {
zz_conj(&temp, &Lval[luptr]);
z_div(&x[fsupc], &x[fsupc], &temp);
} else {
#ifdef _CRAY
ftcs1 = _cptofcd("U", strlen("U"));
ftcs2 = _cptofcd(trans, strlen("T"));
ftcs3 = _cptofcd("N", strlen("N"));
ZTRSV( ftcs1, ftcs2, ftcs3, &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#else
ztrsv_("U", trans, "N", &nsupc, &Lval[luptr], &nsupr,
&x[fsupc], &incx);
#endif
}
} /* for k ... */
}
}
stat->ops[SOLVE] += solve_ops;
SUPERLU_FREE(work);
return 0;
}
void
zgstrs (trans_t trans, SuperMatrix *L, SuperMatrix *U,
int *perm_c, int *perm_r, SuperMatrix *B,
SuperLUStat_t *stat, int *info)
{
/*
* Purpose
* =======
*
* ZGSTRS solves a system of linear equations A*X=B or A'*X=B
* with A sparse and B dense, using the LU factorization computed by
* ZGSTRF.
*
* See supermatrix.h for the definition of 'SuperMatrix' structure.
*
* Arguments
* =========
*
* trans (input) trans_t
* Specifies the form of the system of equations:
* = NOTRANS: A * X = B (No transpose)
* = TRANS: A'* X = B (Transpose)
* = CONJ: A**H * X = B (Conjugate transpose)
*
* L (input) SuperMatrix*
* The factor L from the factorization Pr*A*Pc=L*U as computed by
* zgstrf(). Use compressed row subscripts storage for supernodes,
* i.e., L has types: Stype = SLU_SC, Dtype = SLU_Z, Mtype = SLU_TRLU.
*
* U (input) SuperMatrix*
* The factor U from the factorization Pr*A*Pc=L*U as computed by
* zgstrf(). Use column-wise storage scheme, i.e., U has types:
* Stype = SLU_NC, Dtype = SLU_Z, Mtype = SLU_TRU.
*
* perm_c (input) int*, dimension (L->ncol)
* Column permutation vector, which defines the
* permutation matrix Pc; perm_c[i] = j means column i of A is
* in position j in A*Pc.
*
* perm_r (input) int*, dimension (L->nrow)
* Row permutation vector, which defines the permutation matrix Pr;
* perm_r[i] = j means row i of A is in position j in Pr*A.
*
* B (input/output) SuperMatrix*
* B has types: Stype = SLU_DN, Dtype = SLU_Z, Mtype = SLU_GE.
* On entry, the right hand side matrix.
* On exit, the solution matrix if info = 0;
*
* stat (output) SuperLUStat_t*
* Record the statistics on runtime and floating-point operation count.
* See util.h for the definition of 'SuperLUStat_t'.
*
* info (output) int*
* = 0: successful exit
* < 0: if info = -i, the i-th argument had an illegal value
*
*/
#ifdef _CRAY
_fcd ftcs1, ftcs2, ftcs3, ftcs4;
#endif
int incx = 1, incy = 1;
#ifdef USE_VENDOR_BLAS
doublecomplex alpha = {1.0, 0.0}, beta = {1.0, 0.0};
doublecomplex *work_col;
#endif
doublecomplex temp_comp;
DNformat *Bstore;
doublecomplex *Bmat;
SCformat *Lstore;
NCformat *Ustore;
doublecomplex *Lval, *Uval;
int fsupc, nrow, nsupr, nsupc, luptr, istart, irow;
int i, j, k, iptr, jcol, n, ldb, nrhs;
doublecomplex *work, *rhs_work, *soln;
flops_t solve_ops;
void zprint_soln();
/* Test input parameters ... */
*info = 0;
Bstore = B->Store;
ldb = Bstore->lda;
nrhs = B->ncol;
if ( trans != NOTRANS && trans != TRANS && trans != CONJ ) *info = -1;
else if ( L->nrow != L->ncol || L->nrow < 0 ||
L->Stype != SLU_SC || L->Dtype != SLU_Z || L->Mtype != SLU_TRLU )
*info = -2;
else if ( U->nrow != U->ncol || U->nrow < 0 ||
U->Stype != SLU_NC || U->Dtype != SLU_Z || U->Mtype != SLU_TRU )
*info = -3;
else if ( ldb < SUPERLU_MAX(0, L->nrow) ||
B->Stype != SLU_DN || B->Dtype != SLU_Z || B->Mtype != SLU_GE )
*info = -6;
if ( *info ) {
i = -(*info);
xerbla_("zgstrs", &i);
return;
}
n = L->nrow;
work = doublecomplexCalloc(n * nrhs);
if ( !work ) ABORT("Malloc fails for local work[].");
soln = doublecomplexMalloc(n);
if ( !soln ) ABORT("Malloc fails for local soln[].");
Bmat = Bstore->nzval;
Lstore = L->Store;
Lval = Lstore->nzval;
Ustore = U->Store;
Uval = Ustore->nzval;
solve_ops = 0;
if ( trans == NOTRANS ) {
/* Permute right hand sides to form Pr*B */
for (i = 0; i < nrhs; i++) {
rhs_work = &Bmat[i*ldb];
for (k = 0; k < n; k++) soln[perm_r[k]] = rhs_work[k];
for (k = 0; k < n; k++) rhs_work[k] = soln[k];
}
/* Forward solve PLy=Pb. */
for (k = 0; k <= Lstore->nsuper; k++) {
fsupc = L_FST_SUPC(k);
istart = L_SUB_START(fsupc);
nsupr = L_SUB_START(fsupc+1) - istart;
nsupc = L_FST_SUPC(k+1) - fsupc;
nrow = nsupr - nsupc;
solve_ops += 4 * nsupc * (nsupc - 1) * nrhs;
solve_ops += 8 * nrow * nsupc * nrhs;
if ( nsupc == 1 ) {
for (j = 0; j < nrhs; j++) {
rhs_work = &Bmat[j*ldb];
luptr = L_NZ_START(fsupc);
for (iptr=istart+1; iptr < L_SUB_START(fsupc+1); iptr++){
irow = L_SUB(iptr);
++luptr;
zz_mult(&temp_comp, &rhs_work[fsupc], &Lval[luptr]);
z_sub(&rhs_work[irow], &rhs_work[irow], &temp_comp);
}
}
} else {
luptr = L_NZ_START(fsupc);
#ifdef USE_VENDOR_BLAS
#ifdef _CRAY
ftcs1 = _cptofcd("L", strlen("L"));
ftcs2 = _cptofcd("N", strlen("N"));
ftcs3 = _cptofcd("U", strlen("U"));
CTRSM( ftcs1, ftcs1, ftcs2, ftcs3, &nsupc, &nrhs, &alpha,
&Lval[luptr], &nsupr, &Bmat[fsupc], &ldb);
CGEMM( ftcs2, ftcs2, &nrow, &nrhs, &nsupc, &alpha,
&Lval[luptr+nsupc], &nsupr, &Bmat[fsupc], &ldb,
&beta, &work[0], &n );
#else
ztrsm_("L", "L", "N", "U", &nsupc, &nrhs, &alpha,
&Lval[luptr], &nsupr, &Bmat[fsupc], &ldb);
zgemm_( "N", "N", &nrow, &nrhs, &nsupc, &alpha,
&Lval[luptr+nsupc], &nsupr, &Bmat[fsupc], &ldb,
&beta, &work[0], &n );
#endif
for (j = 0; j < nrhs; j++) {
rhs_work = &Bmat[j*ldb];
work_col = &work[j*n];
iptr = istart + nsupc;
for (i = 0; i < nrow; i++) {
irow = L_SUB(iptr);
z_sub(&rhs_work[irow], &rhs_work[irow], &work_col[i]);
work_col[i].r = 0.0;
work_col[i].i = 0.0;
iptr++;
}
}
#else
for (j = 0; j < nrhs; j++) {
rhs_work = &Bmat[j*ldb];
zlsolve (nsupr, nsupc, &Lval[luptr], &rhs_work[fsupc]);
zmatvec (nsupr, nrow, nsupc, &Lval[luptr+nsupc],
&rhs_work[fsupc], &work[0] );
iptr = istart + nsupc;
for (i = 0; i < nrow; i++) {
irow = L_SUB(iptr);
z_sub(&rhs_work[irow], &rhs_work[irow], &work[i]);
work[i].r = 0.;
work[i].i = 0.;
iptr++;
}
}
#endif
} /* else ... */
} /* for L-solve */
#ifdef DEBUG
printf("After L-solve: y=\n");
zprint_soln(n, nrhs, Bmat);
#endif
/*
* Back solve Ux=y.
*/
for (k = Lstore->nsuper; k >= 0; k--) {
fsupc = L_FST_SUPC(k);
istart = L_SUB_START(fsupc);
nsupr = L_SUB_START(fsupc+1) - istart;
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
solve_ops += 4 * nsupc * (nsupc + 1) * nrhs;
if ( nsupc == 1 ) {
rhs_work = &Bmat[0];
for (j = 0; j < nrhs; j++) {
z_div(&rhs_work[fsupc], &rhs_work[fsupc], &Lval[luptr]);
rhs_work += ldb;
}
} else {
#ifdef USE_VENDOR_BLAS
#ifdef _CRAY
ftcs1 = _cptofcd("L", strlen("L"));
ftcs2 = _cptofcd("U", strlen("U"));
ftcs3 = _cptofcd("N", strlen("N"));
CTRSM( ftcs1, ftcs2, ftcs3, ftcs3, &nsupc, &nrhs, &alpha,
&Lval[luptr], &nsupr, &Bmat[fsupc], &ldb);
#else
ztrsm_("L", "U", "N", "N", &nsupc, &nrhs, &alpha,
&Lval[luptr], &nsupr, &Bmat[fsupc], &ldb);
#endif
#else
for (j = 0; j < nrhs; j++)
zusolve ( nsupr, nsupc, &Lval[luptr], &Bmat[fsupc+j*ldb] );
#endif
}
for (j = 0; j < nrhs; ++j) {
rhs_work = &Bmat[j*ldb];
for (jcol = fsupc; jcol < fsupc + nsupc; jcol++) {
solve_ops += 8*(U_NZ_START(jcol+1) - U_NZ_START(jcol));
for (i = U_NZ_START(jcol); i < U_NZ_START(jcol+1); i++ ){
irow = U_SUB(i);
zz_mult(&temp_comp, &rhs_work[jcol], &Uval[i]);
z_sub(&rhs_work[irow], &rhs_work[irow], &temp_comp);
}
}
}
} /* for U-solve */
#ifdef DEBUG
printf("After U-solve: x=\n");
zprint_soln(n, nrhs, Bmat);
#endif
/* Compute the final solution X := Pc*X. */
for (i = 0; i < nrhs; i++) {
rhs_work = &Bmat[i*ldb];
for (k = 0; k < n; k++) soln[k] = rhs_work[perm_c[k]];
for (k = 0; k < n; k++) rhs_work[k] = soln[k];
}
stat->ops[SOLVE] = solve_ops;
} else { /* Solve A'*X=B */
/* Permute right hand sides to form Pc'*B. */
for (i = 0; i < nrhs; i++) {
rhs_work = &Bmat[i*ldb];
for (k = 0; k < n; k++) soln[perm_c[k]] = rhs_work[k];
for (k = 0; k < n; k++) rhs_work[k] = soln[k];
}
stat->ops[SOLVE] = 0;
if (trans == TRANS) {
for (k = 0; k < nrhs; ++k) {
/* Multiply by inv(U'). */
sp_ztrsv("U", "T", "N", L, U, &Bmat[k*ldb], stat, info);
/* Multiply by inv(L'). */
sp_ztrsv("L", "T", "U", L, U, &Bmat[k*ldb], stat, info);
}
}
else {
for (k = 0; k < nrhs; ++k) {
/* Multiply by inv(U'). */
sp_ztrsv("U", "C", "N", L, U, &Bmat[k*ldb], stat, info);
/* Multiply by inv(L'). */
sp_ztrsv("L", "C", "U", L, U, &Bmat[k*ldb], stat, info);
}
}
/* Compute the final solution X := Pr'*X (=inv(Pr)*X) */
for (i = 0; i < nrhs; i++) {
rhs_work = &Bmat[i*ldb];
for (k = 0; k < n; k++) soln[k] = rhs_work[perm_r[k]];
for (k = 0; k < n; k++) rhs_work[k] = soln[k];
}
}
SUPERLU_FREE(work);
SUPERLU_FREE(soln);
}
void
zgstrs (trans_t trans, SuperMatrix *L, SuperMatrix *U,
int *perm_c, int *perm_r, SuperMatrix *B,
SuperLUStat_t *stat, int *info)
{
#ifdef _CRAY
_fcd ftcs1, ftcs2, ftcs3, ftcs4;
#endif
int incx = 1, incy = 1;
#ifdef USE_VENDOR_BLAS
doublecomplex alpha = {1.0, 0.0}, beta = {1.0, 0.0};
doublecomplex *work_col;
#endif
doublecomplex temp_comp;
DNformat *Bstore;
doublecomplex *Bmat;
SCformat *Lstore;
NCformat *Ustore;
doublecomplex *Lval, *Uval;
int fsupc, nrow, nsupr, nsupc, luptr, istart, irow;
int i, j, k, iptr, jcol, n, ldb, nrhs;
doublecomplex *work, *rhs_work, *soln;
flops_t solve_ops;
void zprint_soln();
/* Test input parameters ... */
*info = 0;
Bstore = B->Store;
ldb = Bstore->lda;
nrhs = B->ncol;
if ( trans != NOTRANS && trans != TRANS && trans != CONJ ) *info = -1;
else if ( L->nrow != L->ncol || L->nrow < 0 ||
L->Stype != SLU_SC || L->Dtype != SLU_Z || L->Mtype != SLU_TRLU )
*info = -2;
else if ( U->nrow != U->ncol || U->nrow < 0 ||
U->Stype != SLU_NC || U->Dtype != SLU_Z || U->Mtype != SLU_TRU )
*info = -3;
else if ( ldb < SUPERLU_MAX(0, L->nrow) ||
B->Stype != SLU_DN || B->Dtype != SLU_Z || B->Mtype != SLU_GE )
*info = -6;
if ( *info ) {
i = -(*info);
input_error("zgstrs", &i);
return;
}
n = L->nrow;
work = doublecomplexCalloc(n * nrhs);
if ( !work ) ABORT("Malloc fails for local work[].");
soln = doublecomplexMalloc(n);
if ( !soln ) ABORT("Malloc fails for local soln[].");
Bmat = Bstore->nzval;
Lstore = L->Store;
Lval = Lstore->nzval;
Ustore = U->Store;
Uval = Ustore->nzval;
solve_ops = 0;
if ( trans == NOTRANS ) {
/* Permute right hand sides to form Pr*B */
for (i = 0; i < nrhs; i++) {
rhs_work = &Bmat[i*ldb];
for (k = 0; k < n; k++) soln[perm_r[k]] = rhs_work[k];
for (k = 0; k < n; k++) rhs_work[k] = soln[k];
}
/* Forward solve PLy=Pb. */
for (k = 0; k <= Lstore->nsuper; k++) {
fsupc = L_FST_SUPC(k);
istart = L_SUB_START(fsupc);
nsupr = L_SUB_START(fsupc+1) - istart;
nsupc = L_FST_SUPC(k+1) - fsupc;
nrow = nsupr - nsupc;
solve_ops += 4 * nsupc * (nsupc - 1) * nrhs;
solve_ops += 8 * nrow * nsupc * nrhs;
if ( nsupc == 1 ) {
for (j = 0; j < nrhs; j++) {
rhs_work = &Bmat[j*ldb];
luptr = L_NZ_START(fsupc);
for (iptr=istart+1; iptr < L_SUB_START(fsupc+1); iptr++){
irow = L_SUB(iptr);
++luptr;
zz_mult(&temp_comp, &rhs_work[fsupc], &Lval[luptr]);
z_sub(&rhs_work[irow], &rhs_work[irow], &temp_comp);
}
}
} else {
luptr = L_NZ_START(fsupc);
#ifdef USE_VENDOR_BLAS
#ifdef _CRAY
ftcs1 = _cptofcd("L", strlen("L"));
ftcs2 = _cptofcd("N", strlen("N"));
ftcs3 = _cptofcd("U", strlen("U"));
CTRSM( ftcs1, ftcs1, ftcs2, ftcs3, &nsupc, &nrhs, &alpha,
&Lval[luptr], &nsupr, &Bmat[fsupc], &ldb);
CGEMM( ftcs2, ftcs2, &nrow, &nrhs, &nsupc, &alpha,
&Lval[luptr+nsupc], &nsupr, &Bmat[fsupc], &ldb,
&beta, &work[0], &n );
#else
ztrsm_("L", "L", "N", "U", &nsupc, &nrhs, &alpha,
&Lval[luptr], &nsupr, &Bmat[fsupc], &ldb);
zgemm_( "N", "N", &nrow, &nrhs, &nsupc, &alpha,
&Lval[luptr+nsupc], &nsupr, &Bmat[fsupc], &ldb,
&beta, &work[0], &n );
#endif
for (j = 0; j < nrhs; j++) {
rhs_work = &Bmat[j*ldb];
work_col = &work[j*n];
iptr = istart + nsupc;
for (i = 0; i < nrow; i++) {
irow = L_SUB(iptr);
z_sub(&rhs_work[irow], &rhs_work[irow], &work_col[i]);
work_col[i].r = 0.0;
work_col[i].i = 0.0;
iptr++;
}
}
#else
for (j = 0; j < nrhs; j++) {
rhs_work = &Bmat[j*ldb];
zlsolve (nsupr, nsupc, &Lval[luptr], &rhs_work[fsupc]);
zmatvec (nsupr, nrow, nsupc, &Lval[luptr+nsupc],
&rhs_work[fsupc], &work[0] );
iptr = istart + nsupc;
for (i = 0; i < nrow; i++) {
irow = L_SUB(iptr);
z_sub(&rhs_work[irow], &rhs_work[irow], &work[i]);
work[i].r = 0.;
work[i].i = 0.;
iptr++;
}
}
#endif
} /* else ... */
} /* for L-solve */
#ifdef DEBUG
printf("After L-solve: y=\n");
zprint_soln(n, nrhs, Bmat);
#endif
/*
* Back solve Ux=y.
*/
for (k = Lstore->nsuper; k >= 0; k--) {
fsupc = L_FST_SUPC(k);
istart = L_SUB_START(fsupc);
nsupr = L_SUB_START(fsupc+1) - istart;
nsupc = L_FST_SUPC(k+1) - fsupc;
luptr = L_NZ_START(fsupc);
solve_ops += 4 * nsupc * (nsupc + 1) * nrhs;
if ( nsupc == 1 ) {
rhs_work = &Bmat[0];
for (j = 0; j < nrhs; j++) {
z_div(&rhs_work[fsupc], &rhs_work[fsupc], &Lval[luptr]);
rhs_work += ldb;
}
} else {
#ifdef USE_VENDOR_BLAS
#ifdef _CRAY
ftcs1 = _cptofcd("L", strlen("L"));
ftcs2 = _cptofcd("U", strlen("U"));
ftcs3 = _cptofcd("N", strlen("N"));
CTRSM( ftcs1, ftcs2, ftcs3, ftcs3, &nsupc, &nrhs, &alpha,
&Lval[luptr], &nsupr, &Bmat[fsupc], &ldb);
#else
ztrsm_("L", "U", "N", "N", &nsupc, &nrhs, &alpha,
&Lval[luptr], &nsupr, &Bmat[fsupc], &ldb);
#endif
#else
for (j = 0; j < nrhs; j++)
zusolve ( nsupr, nsupc, &Lval[luptr], &Bmat[fsupc+j*ldb] );
#endif
}
for (j = 0; j < nrhs; ++j) {
rhs_work = &Bmat[j*ldb];
for (jcol = fsupc; jcol < fsupc + nsupc; jcol++) {
solve_ops += 8*(U_NZ_START(jcol+1) - U_NZ_START(jcol));
for (i = U_NZ_START(jcol); i < U_NZ_START(jcol+1); i++ ){
irow = U_SUB(i);
zz_mult(&temp_comp, &rhs_work[jcol], &Uval[i]);
z_sub(&rhs_work[irow], &rhs_work[irow], &temp_comp);
}
}
}
} /* for U-solve */
#ifdef DEBUG
printf("After U-solve: x=\n");
zprint_soln(n, nrhs, Bmat);
#endif
/* Compute the final solution X := Pc*X. */
for (i = 0; i < nrhs; i++) {
rhs_work = &Bmat[i*ldb];
for (k = 0; k < n; k++) soln[k] = rhs_work[perm_c[k]];
for (k = 0; k < n; k++) rhs_work[k] = soln[k];
}
stat->ops[SOLVE] = solve_ops;
} else { /* Solve A'*X=B or CONJ(A)*X=B */
/* Permute right hand sides to form Pc'*B. */
for (i = 0; i < nrhs; i++) {
rhs_work = &Bmat[i*ldb];
for (k = 0; k < n; k++) soln[perm_c[k]] = rhs_work[k];
for (k = 0; k < n; k++) rhs_work[k] = soln[k];
}
stat->ops[SOLVE] = 0;
if (trans == TRANS) {
for (k = 0; k < nrhs; ++k) {
/* Multiply by inv(U'). */
sp_ztrsv("U", "T", "N", L, U, &Bmat[k*ldb], stat, info);
/* Multiply by inv(L'). */
sp_ztrsv("L", "T", "U", L, U, &Bmat[k*ldb], stat, info);
}
} else { /* trans == CONJ */
for (k = 0; k < nrhs; ++k) {
/* Multiply by conj(inv(U')). */
sp_ztrsv("U", "C", "N", L, U, &Bmat[k*ldb], stat, info);
/* Multiply by conj(inv(L')). */
sp_ztrsv("L", "C", "U", L, U, &Bmat[k*ldb], stat, info);
}
}
/* Compute the final solution X := Pr'*X (=inv(Pr)*X) */
for (i = 0; i < nrhs; i++) {
rhs_work = &Bmat[i*ldb];
for (k = 0; k < n; k++) soln[k] = rhs_work[perm_r[k]];
for (k = 0; k < n; k++) rhs_work[k] = soln[k];
}
}
SUPERLU_FREE(work);
SUPERLU_FREE(soln);
}