int CORE_dormqr(PLASMA_enum side, PLASMA_enum trans,
int M, int N, int K, int IB,
const double *A, int LDA,
const double *T, int LDT,
double *C, int LDC,
double *WORK, int LDWORK)
{
int i, kb;
int i1, i3;
int nq, nw;
int ic = 0;
int jc = 0;
int ni = N;
int mi = M;
/* Check input arguments */
if ((side != PlasmaLeft) && (side != PlasmaRight)) {
coreblas_error(1, "Illegal value of side");
return -1;
}
/*
* NQ is the order of Q and NW is the minimum dimension of WORK
*/
if (side == PlasmaLeft) {
nq = M;
nw = N;
}
else {
nq = N;
nw = M;
}
if ((trans != PlasmaNoTrans) && (trans != PlasmaTrans)) {
coreblas_error(2, "Illegal value of trans");
return -2;
}
if (M < 0) {
coreblas_error(3, "Illegal value of M");
return -3;
}
if (N < 0) {
coreblas_error(4, "Illegal value of N");
return -4;
}
if ((K < 0) || (K > nq)) {
coreblas_error(5, "Illegal value of K");
return -5;
}
if ((IB < 0) || ( (IB == 0) && ((M > 0) && (N > 0)) )) {
coreblas_error(6, "Illegal value of IB");
return -6;
}
if ((LDA < max(1,nq)) && (nq > 0)) {
coreblas_error(8, "Illegal value of LDA");
return -8;
}
if ((LDC < max(1,M)) && (M > 0)) {
coreblas_error(12, "Illegal value of LDC");
return -12;
}
if ((LDWORK < max(1,nw)) && (nw > 0)) {
coreblas_error(14, "Illegal value of LDWORK");
return -14;
}
/* Quick return */
if ((M == 0) || (N == 0) || (K == 0))
return PLASMA_SUCCESS;
if (((side == PlasmaLeft) && (trans != PlasmaNoTrans))
|| ((side == PlasmaRight) && (trans == PlasmaNoTrans))) {
i1 = 0;
i3 = IB;
}
else {
i1 = ( ( K-1 ) / IB )*IB;
i3 = -IB;
}
for(i = i1; (i >- 1) && (i < K); i+=i3 ) {
kb = min(IB, K-i);
if (side == PlasmaLeft) {
/*
* H or H' is applied to C(i:m,1:n)
*/
mi = M - i;
ic = i;
}
else {
/*
* H or H' is applied to C(1:m,i:n)
*/
ni = N - i;
jc = i;
}
/*
* Apply H or H'
*/
LAPACKE_dlarfb_work(LAPACK_COL_MAJOR,
lapack_const(side),
lapack_const(trans),
lapack_const(PlasmaForward),
lapack_const(PlasmaColumnwise),
mi, ni, kb,
&A[LDA*i+i], LDA,
&T[LDT*i], LDT,
&C[LDC*jc+ic], LDC,
WORK, LDWORK);
}
return PLASMA_SUCCESS;
}
int CORE_dgeqrt(int M, int N, int IB,
double *A, int LDA,
double *T, int LDT,
double *TAU,
double *WORK)
{
int i, k, sb;
int iinfo = 0;
/* Check input arguments */
if (M < 0) {
coreblas_error(1, "Illegal value of M");
return -1;
}
if (N < 0) {
coreblas_error(2, "Illegal value of N");
return -2;
}
if ((IB < 0) || ( (IB == 0) && ((M > 0) && (N > 0)) )) {
coreblas_error(3, "Illegal value of IB");
return -3;
}
if ((LDA < max(1,M)) && (M > 0)) {
coreblas_error(5, "Illegal value of LDA");
return -5;
}
if ((LDT < max(1,IB)) && (IB > 0)) {
coreblas_error(7, "Illegal value of LDT");
return -7;
}
/* Quick return */
if ((M == 0) || (N == 0) || (IB == 0))
return PLASMA_SUCCESS;
k = min(M, N);
for(i = 0; i < k; i += IB) {
sb = min(IB, k-i);
iinfo = LAPACKE_dgeqr2_work(LAPACK_COL_MAJOR, M-i, sb,
&A[LDA*i+i], LDA, &TAU[i], WORK);
LAPACKE_dlarft_work(LAPACK_COL_MAJOR,
lapack_const(PlasmaForward),
lapack_const(PlasmaColumnwise),
M-i, sb,
&A[LDA*i+i], LDA, &TAU[i],
&T[LDT*i], LDT);
if (N > i+sb) {
LAPACKE_dlarfb_work(
LAPACK_COL_MAJOR,
lapack_const(PlasmaLeft),
lapack_const(PlasmaTrans),
lapack_const(PlasmaForward),
lapack_const(PlasmaColumnwise),
M-i, N-i-sb, sb,
&A[LDA*i+i], LDA,
&T[LDT*i], LDT,
&A[LDA*(i+sb)+i], LDA,
WORK, N-i-sb);
}
}
return PLASMA_SUCCESS;
}
