/***************************************************************************//**
* Parallel tile matrix-matrix multiplication - dynamic scheduling
**/
void plasma_pzgemm_quark(PLASMA_enum transA, PLASMA_enum transB,
PLASMA_Complex64_t alpha, PLASMA_desc A, PLASMA_desc B,
PLASMA_Complex64_t beta, PLASMA_desc C,
PLASMA_sequence *sequence, PLASMA_request *request)
{
plasma_context_t *plasma;
Quark_Task_Flags task_flags = Quark_Task_Flags_Initializer;
int m, n, k;
int ldam, ldak, ldbn, ldbk, ldcm;
int tempmm, tempnn, tempkn, tempkm;
PLASMA_Complex64_t zbeta;
PLASMA_Complex64_t zone = (PLASMA_Complex64_t)1.0;
plasma = plasma_context_self();
if (sequence->status != PLASMA_SUCCESS)
return;
QUARK_Task_Flag_Set(&task_flags, TASK_SEQUENCE, (intptr_t)sequence->quark_sequence);
for (m = 0; m < C.mt; m++) {
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
ldcm = BLKLDD(C, m);
for (n = 0; n < C.nt; n++) {
tempnn = n == C.nt-1 ? C.n-n*C.nb : C.nb;
/*
* A: PlasmaNoTrans / B: PlasmaNoTrans
*/
if (transA == PlasmaNoTrans) {
ldam = BLKLDD(A, m);
if (transB == PlasmaNoTrans) {
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
ldbk = BLKLDD(B, k);
zbeta = k == 0 ? beta : zone;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
transA, transB,
tempmm, tempnn, tempkn, A.mb,
alpha, A(m, k), ldam, /* lda * Z */
B(k, n), ldbk, /* ldb * Y */
zbeta, C(m, n), ldcm); /* ldc * Y */
}
}
/*
* A: PlasmaNoTrans / B: Plasma[Conj]Trans
*/
else {
ldbn = BLKLDD(B, n);
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
zbeta = k == 0 ? beta : zone;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
transA, transB,
tempmm, tempnn, tempkn, A.mb,
alpha, A(m, k), ldam, /* lda * Z */
B(n, k), ldbn, /* ldb * Z */
zbeta, C(m, n), ldcm); /* ldc * Y */
}
}
}
/*
* A: Plasma[Conj]Trans / B: PlasmaNoTrans
*/
else {
if (transB == PlasmaNoTrans) {
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
ldbk = BLKLDD(B, k);
zbeta = k == 0 ? beta : zone;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
transA, transB,
tempmm, tempnn, tempkm, A.mb,
alpha, A(k, m), ldak, /* lda * X */
B(k, n), ldbk, /* ldb * Y */
zbeta, C(m, n), ldcm); /* ldc * Y */
}
}
/*
* A: Plasma[Conj]Trans / B: Plasma[Conj]Trans
*/
else {
ldbn = BLKLDD(B, n);
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
zbeta = k == 0 ? beta : zone;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
transA, transB,
tempmm, tempnn, tempkm, A.mb,
alpha, A(k, m), ldak, /* lda * X */
B(n, k), ldbn, /* ldb * Z */
zbeta, C(m, n), ldcm); /* ldc * Y */
}
}
}
}
}
}
/***************************************************************************//**
* Parallel tile triangular matrix inverse - dynamic scheduling
**/
void plasma_pztrtri_quark(PLASMA_enum uplo, PLASMA_enum diag, PLASMA_desc A,
PLASMA_sequence *sequence, PLASMA_request *request)
{
plasma_context_t *plasma;
Quark_Task_Flags task_flags = Quark_Task_Flags_Initializer;
int k, m, n;
int ldam, ldan;
int tempkn, tempmm, tempnn;
PLASMA_Complex64_t zone = (PLASMA_Complex64_t) 1.0;
PLASMA_Complex64_t mzone = (PLASMA_Complex64_t)-1.0;
plasma = plasma_context_self();
if (sequence->status != PLASMA_SUCCESS)
return;
QUARK_Task_Flag_Set(&task_flags, TASK_SEQUENCE, (intptr_t)sequence->quark_sequence);
/*
* PlasmaLower
*/
if (uplo == PlasmaLower) {
for (n = 0; n < A.nt; n++) {
tempnn = n == A.nt-1 ? A.n-n*A.nb : A.nb;
ldan = BLKLDD(A, n);
for (m = n+1; m < A.mt; m++) {
tempmm = m == A.mt-1 ? A.m-m*A.mb : A.mb;
ldam = BLKLDD(A, m);
QUARK_CORE_ztrsm(
plasma->quark, &task_flags,
PlasmaRight, uplo, PlasmaNoTrans, diag,
tempmm, tempnn, A.mb,
mzone, A(n, n), ldan,
A(m, n), ldam);
}
for (m = n+1; m < A.mt; m++) {
tempmm = m == A.mt-1 ? A.m-m*A.mb : A.mb;
ldam = BLKLDD(A, m);
for (k = 0; k < n; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
PlasmaNoTrans, PlasmaNoTrans,
tempmm, tempkn, tempnn, A.mb,
zone, A(m, n), ldam,
A(n, k), ldan,
zone, A(m, k), ldam);
}
}
for (m = 0; m < n; m++) {
tempmm = m == A.mt-1 ? A.m-m*A.mb : A.mb;
QUARK_CORE_ztrsm(
plasma->quark, &task_flags,
PlasmaLeft, uplo, PlasmaNoTrans, diag,
tempnn, tempmm, A.mb,
zone, A(n, n), ldan,
A(n, m), ldan);
}
QUARK_CORE_ztrtri(
plasma->quark, &task_flags,
uplo, diag,
tempnn, A.mb,
A(n, n), ldan,
sequence, request, A.nb*n);
}
}
/*
* PlasmaUpper
*/
else {
for (m = 0; m < A.mt; m++) {
tempmm = m == A.mt-1 ? A.m-m*A.mb : A.mb;
ldam = BLKLDD(A, m);
for (n = m+1; n < A.nt; n++) {
tempnn = n == A.nt-1 ? A.n-n*A.nb : A.nb;
QUARK_CORE_ztrsm(
plasma->quark, &task_flags,
PlasmaLeft, uplo, PlasmaNoTrans, diag,
tempmm, tempnn, A.mb,
mzone, A(m, m), ldam,
A(m, n), ldam);
}
for (n = 0; n < m; n++) {
tempnn = n == A.nt-1 ? A.n-n*A.nb : A.nb;
ldan = BLKLDD(A, n);
for (k = m+1; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
PlasmaNoTrans, PlasmaNoTrans,
tempnn, tempkn, tempmm, A.mb,
zone, A(n, m), ldan,
A(m, k), ldam,
zone, A(n, k), ldan);
}
QUARK_CORE_ztrsm(
plasma->quark, &task_flags,
PlasmaRight, uplo, PlasmaNoTrans, diag,
tempnn, tempmm, A.mb,
zone, A(m, m), ldam,
A(n, m), ldan);
}
QUARK_CORE_ztrtri(
plasma->quark, &task_flags,
uplo, diag,
tempmm, A.mb,
A(m, m), ldam,
sequence, request, A.mb*m);
}
}
}
/***************************************************************************//**
* Parallel tile Hermitian rank-k update - dynamic scheduling
**/
void plasma_pzherk_quark(PLASMA_enum uplo, PLASMA_enum trans,
double alpha, PLASMA_desc A,
double beta, PLASMA_desc C,
PLASMA_sequence *sequence, PLASMA_request *request)
{
plasma_context_t *plasma;
Quark_Task_Flags task_flags = Quark_Task_Flags_Initializer;
int m, n, k;
int ldak, ldam, ldan, ldcm, ldcn;
int tempnn, tempmm, tempkn, tempkm;
PLASMA_Complex64_t zone = (PLASMA_Complex64_t)1.0;
PLASMA_Complex64_t zalpha = (PLASMA_Complex64_t)alpha;
PLASMA_Complex64_t zbeta;
double dbeta;
plasma = plasma_context_self();
if (sequence->status != PLASMA_SUCCESS)
return;
QUARK_Task_Flag_Set(&task_flags, TASK_SEQUENCE, (intptr_t)sequence->quark_sequence);
for (n = 0; n < C.nt; n++) {
tempnn = n == C.nt-1 ? C.n-n*C.nb : C.nb;
ldan = BLKLDD(A, n);
ldcn = BLKLDD(C, n);
/*
* PlasmaNoTrans
*/
if (trans == PlasmaNoTrans) {
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
dbeta = k == 0 ? beta : 1.0;
QUARK_CORE_zherk(
plasma->quark, &task_flags,
uplo, trans,
tempnn, tempkn, A.mb,
alpha, A(n, k), ldan, /* ldan * K */
dbeta, C(n, n), ldcn); /* ldc * N */
}
/*
* PlasmaNoTrans / PlasmaLower
*/
if (uplo == PlasmaLower) {
for (m = n+1; m < C.mt; m++) {
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
ldam = BLKLDD(A, m);
ldcm = BLKLDD(C, m);
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
zbeta = k == 0 ? (PLASMA_Complex64_t)beta : zone;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
trans, PlasmaConjTrans,
tempmm, tempnn, tempkn, A.mb,
zalpha, A(m, k), ldam, /* ldam * K */
A(n, k), ldan, /* ldan * K */
zbeta, C(m, n), ldcm); /* ldc * N */
}
}
}
/*
* PlasmaNoTrans / PlasmaUpper
*/
else {
for (m = n+1; m < C.mt; m++) {
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
ldam = BLKLDD(A, m);
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
zbeta = k == 0 ? (PLASMA_Complex64_t)beta : zone;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
trans, PlasmaConjTrans,
tempnn, tempmm, tempkn, A.mb,
zalpha, A(n, k), ldan, /* ldan * K */
A(m, k), ldam, /* ldam * M */
zbeta, C(n, m), ldcn); /* ldc * M */
}
}
}
}
/*
* Plasma[Conj]Trans
*/
else {
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
dbeta = k == 0 ? beta : 1.0;
QUARK_CORE_zherk(
plasma->quark, &task_flags,
uplo, trans,
tempnn, tempkm, A.mb,
alpha, A(k, n), ldak, /* lda * N */
dbeta, C(n, n), ldcn); /* ldc * N */
}
/*
* Plasma[Conj]Trans / PlasmaLower
*/
if (uplo == PlasmaLower) {
for (m = n+1; m < C.mt; m++) {
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
ldcm = BLKLDD(C, m);
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
zbeta = k == 0 ? (PLASMA_Complex64_t)beta : zone;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
trans, PlasmaNoTrans,
tempmm, tempnn, tempkm, A.mb,
zalpha, A(k, m), ldak, /* lda * M */
A(k, n), ldak, /* lda * N */
zbeta, C(m, n), ldcm); /* ldc * N */
}
}
}
/*
* Plasma[Conj]Trans / PlasmaUpper
*/
else {
for (m = n+1; m < C.mt; m++) {
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
zbeta = k == 0 ? (PLASMA_Complex64_t)beta : zone;
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
trans, PlasmaNoTrans,
tempnn, tempmm, tempkm, A.mb,
zalpha, A(k, n), ldak, /* lda * K */
A(k, m), ldak, /* lda * M */
zbeta, C(n, m), ldcn); /* ldc * M */
}
}
}
}
}
}
/***************************************************************************//**
* Parallel tile LU factorization with no pivoting - dynamic scheduling
**/
void plasma_pzgetrf_nopiv_quark(PLASMA_desc A,
PLASMA_sequence *sequence,
PLASMA_request *request)
{
plasma_context_t *plasma;
Quark_Task_Flags task_flags = Quark_Task_Flags_Initializer;
int k, m, n, ib;
int ldak, ldam;
int tempkm, tempkn, tempmm, tempnn;
PLASMA_Complex64_t zone = (PLASMA_Complex64_t) 1.0;
PLASMA_Complex64_t mzone = (PLASMA_Complex64_t)-1.0;
plasma = plasma_context_self();
if (sequence->status != PLASMA_SUCCESS)
return;
QUARK_Task_Flag_Set(&task_flags, TASK_SEQUENCE, (intptr_t)sequence->quark_sequence);
ib = PLASMA_IB;
for (k = 0; k < min(A.mt, A.nt); k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
ldak = BLKLDD(A, k);
QUARK_CORE_zgetrf_nopiv(
plasma->quark, &task_flags,
tempkm, tempkn, ib, A.mb,
A(k, k), ldak,
sequence, request, A.mb*k);
for (m = k+1; m < A.mt; m++) {
tempmm = m == A.mt-1 ? A.m-m*A.mb : A.mb;
ldam = BLKLDD(A, m);
QUARK_CORE_ztrsm(
plasma->quark, &task_flags,
PlasmaRight, PlasmaUpper, PlasmaNoTrans, PlasmaNonUnit,
tempmm, tempkn, A.mb,
zone, A(k, k), ldak,
A(m, k), ldam);
}
for (n = k+1; n < A.nt; n++) {
tempnn = n == A.nt-1 ? A.n-n*A.nb : A.nb;
QUARK_CORE_ztrsm(
plasma->quark, &task_flags,
PlasmaLeft, PlasmaLower, PlasmaNoTrans, PlasmaUnit,
tempkm, tempnn, A.mb,
zone, A(k, k), ldak,
A(k, n), ldak);
for (m = k+1; m < A.mt; m++) {
tempmm = m == A.mt-1 ? A.m-m*A.mb : A.mb;
ldam = BLKLDD(A, m);
QUARK_CORE_zgemm(
plasma->quark, &task_flags,
PlasmaNoTrans, PlasmaNoTrans,
tempmm, tempnn, A.mb, A.mb,
mzone, A(m, k), ldam,
A(k, n), ldak,
zone, A(m, n), ldam);
}
}
}
}
