diff --git a/compute/pzunmlq.c b/compute/pzunmlq.c
index 0aa1054dce5da9ffd09a83d7a36ecb0f25aea1bf..ac2cb0aa76116c1cf70f6cd3efaa0f9621ff5b9b 100644
--- a/compute/pzunmlq.c
+++ b/compute/pzunmlq.c
@@ -27,7 +27,7 @@
#include "control/common.h"
#define A(m,n) A, m, n
-#define B(m,n) B, m, n
+#define C(m,n) C, m, n
#define T(m,n) T, m, n
#define D(k) D, k, k
@@ -35,7 +35,7 @@
* Parallel application of Q using tile V - LQ factorization - dynamic scheduling
*/
void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
- CHAM_desc_t *A, CHAM_desc_t *B, CHAM_desc_t *T, CHAM_desc_t *D,
+ CHAM_desc_t *A, CHAM_desc_t *C, CHAM_desc_t *T, CHAM_desc_t *D,
RUNTIME_sequence_t *sequence, RUNTIME_request_t *request )
{
CHAM_context_t *chamctxt;
@@ -44,9 +44,9 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
size_t ws_host = 0;
int k, m, n;
- int ldak, ldbk, ldbm, lddk;
- int tempmm, tempnn, tempkn, tempkm, tempkmin;
- int ib, minMT, minM;
+ int ldak, ldck, ldcm, lddk;
+ int tempkm, tempkn, tempkmin, tempmm, tempnn;
+ int ib, KT, K;
chamctxt = chameleon_context_self();
if (sequence->status != CHAMELEON_SUCCESS) {
@@ -57,11 +57,11 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
ib = CHAMELEON_IB;
if (A->m > A->n) {
- minM = A->n;
- minMT = A->nt;
+ KT = A->nt;
+ K = A->n;
} else {
- minM = A->m;
- minMT = A->mt;
+ KT = A->mt;
+ K = A->m;
}
if ( D == NULL ) {
@@ -94,13 +94,14 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
/*
* ChamLeft / ChamNoTrans
*/
- for (k = 0; k < minMT; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
- tempkm = k == B->mt-1 ? B->m-k*B->mb : B->mb;
- tempkmin = k == minMT-1 ? minM-k*A->nb : A->nb;
+ tempkm = k == C->mt - 1 ? C->m - k * C->mb : C->mb;
+ tempkmin = k == KT - 1 ? K - k * A->nb : A->nb;
+
ldak = BLKLDD(A, k);
- ldbk = BLKLDD(B, k);
+ ldck = BLKLDD(C, k);
lddk = BLKLDD(D, k);
if ( genD ) {
@@ -118,28 +119,28 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
D(k), lddk );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zunmlq(
&options,
side, trans,
tempkm, tempnn, tempkmin, ib, T->nb,
D(k), lddk,
T(k, k), T->mb,
- B(k, n), ldbk);
+ C(k, n), ldck);
}
RUNTIME_data_flush( sequence, D(k) );
RUNTIME_data_flush( sequence, T(k, k) );
- for (m = k+1; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (m = k+1; 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;
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, m, n ) );
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, m, n ) );
/* TS kernel */
INSERT_TASK_ztpmlqt(
@@ -148,8 +149,8 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
tempmm, tempnn, tempkmin, 0, ib, T->nb,
A(k, m), ldak,
T(k, m), T->mb,
- B(k, n), ldbk,
- B(m, n), ldbm);
+ C(k, n), ldck,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, m) );
@@ -157,9 +158,9 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
}
/* Restore the original location of the tiles */
- for (n = 0; n < B->nt; n++) {
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, k, n ) );
+ for (n = 0; n < C->nt; n++) {
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, k, n ) );
}
RUNTIME_iteration_pop(chamctxt);
@@ -169,23 +170,24 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
* ChamLeft / ChamConjTrans
*/
else {
- for (k = minMT-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
- tempkm = k == B->mt-1 ? B->m-k*B->mb : B->mb;
- tempkmin = k == minMT-1 ? minM-k*A->nb : A->nb;
+ tempkm = k == C->mt - 1 ? C->m - k * C->mb : C->mb;
+ tempkmin = k == KT - 1 ? K - k * A->nb : A->nb;
+
ldak = BLKLDD(A, k);
- ldbk = BLKLDD(B, k);
+ ldck = BLKLDD(C, k);
lddk = BLKLDD(D, k);
- for (m = B->mt-1; m > k; m--) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (m = C->mt-1; m > k; 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;
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, m, n ) );
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, m, n ) );
/* TS kernel */
INSERT_TASK_ztpmlqt(
@@ -194,13 +196,14 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
tempmm, tempnn, tempkmin, 0, ib, T->nb,
A(k, m), ldak,
T(k, m), T->mb,
- B(k, n), ldbk,
- B(m, n), ldbm);
+ C(k, n), ldck,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, m) );
RUNTIME_data_flush( sequence, T(k, m) );
}
+
if ( genD ) {
int tempDkn = k == D->nt-1 ? D->n-k*D->nb : D->nb;
INSERT_TASK_zlacpy(
@@ -216,11 +219,11 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
D(k), lddk );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, k, n ) );
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, k, n ) );
INSERT_TASK_zunmlq(
&options,
@@ -228,7 +231,7 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
tempkm, tempnn, tempkmin, ib, T->nb,
D(k), lddk,
T(k, k), T->mb,
- B(k, n), ldbk);
+ C(k, n), ldck);
}
RUNTIME_data_flush( sequence, D(k) );
RUNTIME_data_flush( sequence, T(k, k) );
@@ -241,22 +244,22 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
*/
else {
if (trans == ChamNoTrans) {
- for (k = minMT-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
- tempkn = k == B->nt - 1 ? B->n - k * B->nb : B->nb;
- tempkmin = k == minMT - 1 ? minM - k * A->nb : A->nb;
+ tempkn = k == C->nt - 1 ? C->n - k * C->nb : C->nb;
+ tempkmin = k == KT - 1 ? K - k * A->nb : A->nb;
ldak = BLKLDD(A, k);
lddk = BLKLDD(D, k);
- for (n = B->nt-1; n > k; n--) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (n = C->nt-1; n > k; n--) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, n ) );
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, n ) );
/* TS kernel */
INSERT_TASK_ztpmlqt(
@@ -265,13 +268,14 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
tempmm, tempnn, tempkmin, 0, ib, T->nb,
A(k, n), ldak,
T(k, n), T->mb,
- B(m, k), ldbm,
- B(m, n), ldbm);
+ C(m, k), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, n) );
RUNTIME_data_flush( sequence, T(k, n) );
}
+
if ( genD ) {
int tempDkn = k == D->nt-1 ? D->n-k*D->nb : D->nb;
INSERT_TASK_zlacpy(
@@ -287,12 +291,12 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
D(k), lddk );
#endif
}
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, k ) );
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, k ) );
INSERT_TASK_zunmlq(
&options,
@@ -300,7 +304,7 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
tempmm, tempkn, tempkmin, ib, T->nb,
D(k), lddk,
T(k, k), T->mb,
- B(m, k), ldbm);
+ C(m, k), ldcm);
}
RUNTIME_data_flush( sequence, D(k) );
@@ -313,16 +317,17 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
* ChamRight / ChamConjTrans
*/
else {
- for (k = 0; k < minMT; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
- tempkn = k == B->nt-1 ? B->n-k*B->nb : B->nb;
- tempkmin = k == minMT-1 ? minM-k*A->mb : A->mb;
+ tempkn = k == C->nt - 1 ? C->n - k * C->nb : C->nb;
+ tempkmin = k == KT - 1 ? K - k * A->nb : A->nb;
ldak = BLKLDD(A, k);
lddk = BLKLDD(D, k);
if ( genD ) {
int tempDkn = k == D->nt-1 ? D->n-k*D->nb : D->nb;
+
INSERT_TASK_zlacpy(
&options,
ChamUpper, tempkmin, tempDkn, A->nb,
@@ -336,29 +341,29 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
D(k), lddk );
#endif
}
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
INSERT_TASK_zunmlq(
&options,
side, trans,
tempmm, tempkn, tempkmin, ib, T->nb,
D(k), lddk,
T(k, k), T->mb,
- B(m, k), ldbm);
+ C(m, k), ldcm);
}
RUNTIME_data_flush( sequence, D(k) );
RUNTIME_data_flush( sequence, T(k, k) );
- for (n = k+1; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (n = k+1; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, n ) );
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, n ) );
/* TS kernel */
INSERT_TASK_ztpmlqt(
@@ -367,8 +372,8 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
tempmm, tempnn, tempkmin, 0, ib, T->nb,
A(k, n), ldak,
T(k, n), T->mb,
- B(m, k), ldbm,
- B(m, n), ldbm);
+ C(m, k), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, n) );
@@ -376,9 +381,9 @@ void chameleon_pzunmlq( int genD, cham_side_t side, cham_trans_t trans,
}
/* Restore the original location of the tiles */
- for (m = 0; m < B->mt; m++) {
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, k ) );
+ for (m = 0; m < C->mt; m++) {
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, k ) );
}
RUNTIME_iteration_pop(chamctxt);
diff --git a/compute/pzunmlq_param.c b/compute/pzunmlq_param.c
index fe122e0b4798aa795b881a94173c82df250d9336..6e27ca3f610cbb139e0f37edc127e69097fcedab 100644
--- a/compute/pzunmlq_param.c
+++ b/compute/pzunmlq_param.c
@@ -22,7 +22,7 @@
#include <stdlib.h>
#define A(m,n) A, m, n
-#define B(m,n) B, m, n
+#define C(m,n) C, m, n
#define T(m,n) T, m, n
#define D(m,n) D, m, n
@@ -31,7 +31,7 @@
*/
void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
cham_side_t side, cham_trans_t trans,
- CHAM_desc_t *A, CHAM_desc_t *B,
+ CHAM_desc_t *A, CHAM_desc_t *C,
CHAM_desc_t *TS, CHAM_desc_t *TT, CHAM_desc_t *D,
RUNTIME_sequence_t *sequence, RUNTIME_request_t *request )
{
@@ -42,9 +42,9 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
size_t ws_host = 0;
int k, m, n, i, p;
- int ldbm, ldak, ldbp, lddk;
- int tempnn, temppn, tempkmin, tempmm, tempkm;
- int ib, K, L;
+ int ldak, lddk, ldcp, ldcm;
+ int temppm, temppn, tempmm, tempnn, tempkm,tempkmin;
+ int ib, KT, L;
int node, nbtiles, *tiles;
chamctxt = chameleon_context_self();
@@ -55,8 +55,6 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
ib = CHAMELEON_IB;
- K = chameleon_min(A->mt, A->nt);
-
if ( D == NULL ) {
D = A;
genD = 0;
@@ -85,15 +83,16 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
/* Initialisation of tiles */
tiles = (int*)calloc( qrtree->mt, sizeof(int) );
+ KT = chameleon_min( A->mt, A->nt );
if (side == ChamLeft ) {
if (trans == ChamNoTrans) {
/*
* ChamLeft / ChamNoTrans
*/
- for (k = 0; k < K; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
- tempkm = k == A->mt-1 ? A->m-k*A->mb : A->mb;
+ tempkm = k == A->mt - 1 ? A->m - k * A->mb : A->mb;
ldak = BLKLDD(A, k);
lddk = BLKLDD(D, k);
@@ -101,12 +100,14 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
for (i = 0; i < qrtree->getnbgeqrf(qrtree, k); i++) {
p = qrtree->getm(qrtree, k, i);
- temppn = p == A->nt-1 ? A->n-p*A->nb : A->nb;
- tempkmin = chameleon_min(tempkm, temppn);
- ldbp = BLKLDD(B, p);
+ temppm = p == C->mt-1 ? C->m - p * C->mb : C->mb;
+ tempkmin = chameleon_min( temppm, tempkm );
+
+ ldcp = BLKLDD(C, p);
if ( genD ) {
int tempDpn = p == D->nt-1 ? D->n-p*D->nb : D->nb;
+
INSERT_TASK_zlacpy(
&options,
ChamUpper, tempkmin, tempDpn, A->nb,
@@ -120,17 +121,15 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
D(k, p), lddk );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zunmlq(
- &options,
- side, trans,
- temppn, tempnn, tempkmin, ib, T->nb,
+ &options, side, trans,
+ temppm, tempnn, tempkmin, ib, T->nb,
D(k, p), lddk,
T(k, p), T->mb,
- B(p, n), ldbp);
+ C(p, n), ldcp);
}
-
RUNTIME_data_flush( sequence, D(k, p) );
RUNTIME_data_flush( sequence, T(k, p) );
}
@@ -142,9 +141,9 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
m = tiles[i];
p = qrtree->currpiv(qrtree, k, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbp = BLKLDD(B, p);
- ldbm = BLKLDD(B, m);
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+ ldcp = BLKLDD(C, p);
if( qrtree->gettype(qrtree, k, m) == LIBHQR_KILLED_BY_TS ) {
/* TS kernel */
@@ -156,30 +155,29 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
L = A->nb;
T = TT;
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(p, n), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
INSERT_TASK_ztpmlqt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkm, chameleon_min( L, tempnn ), ib, T->nb,
A(k, m), ldak,
T(k, m), T->mb,
- B(p, n), ldbp,
- B(m, n), ldbm);
+ C(p, n), ldcp,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, m) );
RUNTIME_data_flush( sequence, T(k, m) );
}
/* Restore the original location of the tiles */
- for (n = 0; n < B->nt; n++) {
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, k, n ) );
+ for (n = 0; n < C->nt; n++) {
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, k, n ) );
}
RUNTIME_iteration_pop(chamctxt);
@@ -189,7 +187,7 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
* ChamLeft / ChamConjTrans
*/
else {
- for (k = K-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
tempkm = k == A->mt-1 ? A->m-k*A->mb : A->mb;
@@ -203,9 +201,9 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
m = tiles[i];
p = qrtree->currpiv(qrtree, k, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbp = BLKLDD(B, p);
- ldbm = BLKLDD(B, m);
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcp = BLKLDD(C, p);
+ ldcm = BLKLDD(C, m);
if( qrtree->gettype(qrtree, k, m) == LIBHQR_KILLED_BY_TS ) {
/* TS kernel */
@@ -217,21 +215,20 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
L = A->nb;
T = TT;
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(p, n), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
INSERT_TASK_ztpmlqt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkm, chameleon_min(L, tempnn), ib, T->nb,
A(k, m), ldak,
T(k, m), T->mb,
- B(p, n), ldbp,
- B(m, n), ldbm);
+ C(p, n), ldcp,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, m) );
RUNTIME_data_flush( sequence, T(k, m) );
@@ -241,12 +238,14 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
for (i = 0; i < qrtree->getnbgeqrf(qrtree, k); i++) {
p = qrtree->getm(qrtree, k, i);
- temppn = p == A->nt-1 ? A->n-p*A->nb : A->nb;
- tempkmin = chameleon_min(tempkm, temppn);
- ldbp = BLKLDD(B, p);
+ temppm = p == C->mt-1 ? C->m-p*C->mb : C->mb;
+ tempkmin = chameleon_min( temppm, tempkm );
+
+ ldcp = BLKLDD(C, p);
if ( genD ) {
int tempDpn = p == D->nt-1 ? D->n-p*D->nb : D->nb;
+
INSERT_TASK_zlacpy(
&options,
ChamUpper, tempkmin, tempDpn, A->nb,
@@ -260,25 +259,24 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
D(k, p), lddk );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- RUNTIME_data_migrate( sequence, B(p, n),
- B->get_rankof( B, p, n ) );
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ RUNTIME_data_migrate( sequence, C(p, n),
+ C->get_rankof( C, p, n ) );
INSERT_TASK_zunmlq(
- &options,
- side, trans,
- temppn, tempnn, tempkmin, ib, T->nb,
+ &options, side, trans,
+ temppm, tempnn, tempkmin, ib, T->nb,
D(k, p), lddk,
T(k, p), T->mb,
- B(p, n), ldbp);
+ C(p, n), ldcp);
}
RUNTIME_data_flush( sequence, D(k, p) );
RUNTIME_data_flush( sequence, T(k, p) );
}
-
RUNTIME_iteration_pop(chamctxt);
}
}
@@ -288,7 +286,7 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
*/
else {
if (trans == ChamNoTrans) {
- for (k = K-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
tempkm = k == A->mt-1 ? A->m-k*A->mb : A->mb;
@@ -302,7 +300,7 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
n = tiles[i];
p = qrtree->currpiv(qrtree, k, n);
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
if( qrtree->gettype(qrtree, k, n) == LIBHQR_KILLED_BY_TS ) {
/* TS kernel */
@@ -314,22 +312,22 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
L = tempnn;
T = TT;
}
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(m, p), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
INSERT_TASK_ztpmlqt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkm, L, ib, T->nb,
A(k, n), ldak,
T(k, n), T->mb,
- B(m, p), ldbm,
- B(m, n), ldbm);
+ C(m, p), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, n) );
RUNTIME_data_flush( sequence, T(k, n) );
@@ -339,11 +337,12 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
for (i = 0; i < qrtree->getnbgeqrf(qrtree, k); i++) {
p = qrtree->getm(qrtree, k, i);
- temppn = p == A->nt-1 ? A->n-p*A->nb : A->nb;
- tempkmin = chameleon_min(tempkm, temppn);
+ temppn = p == C->nt-1 ? C->n - p * C->nb : C->nb;
+ tempkmin = chameleon_min( temppn, tempkm );
if ( genD ) {
int tempDpn = p == D->nt-1 ? D->n-p*D->nb : D->nb;
+
INSERT_TASK_zlacpy(
&options,
ChamUpper, tempkmin, tempDpn, A->nb,
@@ -357,26 +356,24 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
D(k, p), lddk );
#endif
}
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- RUNTIME_data_migrate( sequence, B(m, p),
- B->get_rankof( B, m, p ) );
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+
+ RUNTIME_data_migrate( sequence, C(m, p),
+ C->get_rankof( C, m, p ) );
INSERT_TASK_zunmlq(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, temppn, tempkmin, ib, T->nb,
D(k, p), lddk,
T(k, p), T->mb,
- B(m, p), ldbm);
+ C(m, p), ldcm);
}
-
RUNTIME_data_flush( sequence, D(k, p) );
RUNTIME_data_flush( sequence, T(k, p) );
}
-
RUNTIME_iteration_pop(chamctxt);
}
}
@@ -384,7 +381,7 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
* ChamRight / ChamConjTrans
*/
else {
- for (k = 0; k < K; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
tempkm = k == A->mt-1 ? A->m-k*A->mb : A->mb;
@@ -395,11 +392,12 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
for (i = 0; i < qrtree->getnbgeqrf(qrtree, k); i++) {
p = qrtree->getm(qrtree, k, i);
- temppn = p == A->nt-1 ? A->n-p*A->nb : A->nb;
- tempkmin = chameleon_min(tempkm, temppn);
+ temppn = p == C->nt - 1 ? C->n - p * C->nb : C->nb;
+ tempkmin = chameleon_min( temppn, tempkm );
if ( genD ) {
int tempDpn = p == D->nt-1 ? D->n-p*D->nb : D->nb;
+
INSERT_TASK_zlacpy(
&options,
ChamUpper, tempkmin, tempDpn, A->nb,
@@ -413,18 +411,17 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
D(k, p), lddk );
#endif
}
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+
+ for (m = 0; m < C->mt; m++) {
+ ldcm = BLKLDD(C, m);
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
INSERT_TASK_zunmlq(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, temppn, tempkmin, ib, T->nb,
D(k, p), lddk,
T(k, p), TS->mb,
- B(m, p), ldbm);
+ C(m, p), ldcm);
}
-
RUNTIME_data_flush( sequence, D(k, p) );
RUNTIME_data_flush( sequence, T(k, p) );
}
@@ -436,7 +433,7 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
n = tiles[i];
p = qrtree->currpiv(qrtree, k, n);
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
if( qrtree->gettype(qrtree, k, n) == LIBHQR_KILLED_BY_TS ) {
/* TS kernel */
@@ -449,31 +446,30 @@ void chameleon_pzunmlq_param( int genD, const libhqr_tree_t *qrtree,
T = TT;
}
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(m, p), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
INSERT_TASK_ztpmlqt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkm, L, ib, T->nb,
A(k, n), ldak,
T(k, n), T->mb,
- B(m, p), ldbm,
- B(m, n), ldbm);
+ C(m, p), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, n) );
RUNTIME_data_flush( sequence, T(k, n) );
}
/* Restore the original location of the tiles */
- for (m = 0; m < B->mt; m++) {
- RUNTIME_data_migrate( sequence, B( m, k ),
- B->get_rankof( B, m, k ) );
+ for (m = 0; m < C->mt; m++) {
+ RUNTIME_data_migrate( sequence, C( m, k ),
+ C->get_rankof( C, m, k ) );
}
RUNTIME_iteration_pop(chamctxt);
diff --git a/compute/pzunmlqrh.c b/compute/pzunmlqrh.c
index c3b7d0c1d0f36636d42c1d2fa67296c283df3181..8f2931ccdba43f7f7d89f8f5f1b962a28567c4bf 100644
--- a/compute/pzunmlqrh.c
+++ b/compute/pzunmlqrh.c
@@ -27,7 +27,7 @@
#include "control/common.h"
#define A(m,n) A, (m), (n)
-#define B(m,n) B, (m), (n)
+#define C(m,n) C, (m), (n)
#define T(m,n) T, (m), (n)
#define T2(m,n) T, (m), ((n)+A->nt)
#define D(m,n) D, (m), (n)
@@ -37,7 +37,7 @@
* Householder) - dynamic scheduling
*/
void chameleon_pzunmlqrh( int genD, int BS, cham_side_t side, cham_trans_t trans,
- CHAM_desc_t *A, CHAM_desc_t *B, CHAM_desc_t *T, CHAM_desc_t *D,
+ CHAM_desc_t *A, CHAM_desc_t *C, CHAM_desc_t *T, CHAM_desc_t *D,
RUNTIME_sequence_t *sequence, RUNTIME_request_t *request )
{
CHAM_context_t *chamctxt;
@@ -45,10 +45,11 @@ void chameleon_pzunmlqrh( int genD, int BS, cham_side_t side, cham_trans_t trans
size_t ws_worker = 0;
size_t ws_host = 0;
- int k, m, n;
- int K, N, RD, lastRD;
- int ldak, lddk, ldbN, ldbm, ldbNRD;
- int tempNn, tempkm, tempnn, tempmm, tempNRDn, tempkmin;
+ int k, m, n, p;
+ int KT, RD, lastRD;
+ int ldak, lddk;
+ int ldcp, ldcm;
+ int temppm, temppn, tempkm, tempnn, tempmm, tempkmin;
int ib, node;
chamctxt = chameleon_context_self();
@@ -84,62 +85,66 @@ void chameleon_pzunmlqrh( int genD, int BS, cham_side_t side, cham_trans_t trans
RUNTIME_options_ws_alloc( &options, ws_worker, ws_host );
- K = chameleon_min(A->mt, A->nt);
+ KT = chameleon_min( A->mt, A->nt );
if (side == ChamLeft ) {
if (trans == ChamNoTrans) {
/*
* ChamLeft / ChamNoTrans
*/
- for (k = 0; k < K; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
- tempkm = k == A->mt-1 ? A->m-k*A->mb : A->mb;
+ tempkm = k == A->mt - 1 ? A->m - k * A->mb : A->mb;
+
ldak = BLKLDD(A, k);
lddk = BLKLDD(D, k);
- for (N = k; N < A->nt; N += BS) {
- tempNn = N == A->nt-1 ? A->n-N*A->nb : A->nb;
- tempkmin = chameleon_min(tempkm,tempNn);
- ldbN = BLKLDD(B, N);
+ for (p = k; p < C->mt; p += BS) {
+
+ temppm = p == C->mt-1 ? C->m - p * C->mb : C->mb;
+ tempkmin = chameleon_min( temppm, tempkm );
+
+ ldcp = BLKLDD(C, p);
+
if ( genD ) {
- int tempDNn = N == D->nt-1 ? D->n-N*D->nb : D->nb;
+ int tempDpn = p == D->nt-1 ? D->n-p*D->nb : D->nb;
INSERT_TASK_zlacpy(
&options,
- ChamUpper, tempkmin, tempDNn, A->nb,
- A(k, N), ldak,
- D(k, N), lddk );
+ ChamUpper, tempkmin, tempDpn, A->nb,
+ A(k, p), ldak,
+ D(k, p), lddk );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamLower, tempkmin, tempDNn,
+ ChamLower, tempkmin, tempDpn,
0., 1.,
- D(k, N), lddk );
+ D(k, p), lddk );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zunmlq(
&options,
side, trans,
- tempNn, tempnn,
- tempkmin, ib, T->nb,
- D(k, N), lddk,
- T(k, N), T->mb,
- B(N, n), ldbN);
+ temppm, tempnn, tempkmin, ib, T->nb,
+ D(k, p), lddk,
+ T(k, p), T->mb,
+ C(p, n), ldcp);
}
- RUNTIME_data_flush( sequence, D(k, N) );
- RUNTIME_data_flush( sequence, T(k, N) );
+ RUNTIME_data_flush( sequence, D(k, p) );
+ RUNTIME_data_flush( sequence, T(k, p) );
- for (m = N+1; m < chameleon_min(N+BS, A->nt); m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (m = p+1; m < chameleon_min(p+BS, C->mt); m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(N, n), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TS kernel */
INSERT_TASK_ztpmlqt(
@@ -147,350 +152,369 @@ void chameleon_pzunmlqrh( int genD, int BS, cham_side_t side, cham_trans_t trans
tempmm, tempnn, tempkm, 0, ib, T->nb,
A(k, m), ldak,
T(k, m), T->mb,
- B(N, n), ldbN,
- B(m, n), ldbm);
+ C(p, n), ldcp,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, m) );
RUNTIME_data_flush( sequence, T(k, m) );
}
}
- for (RD = BS; RD < A->nt-k; RD *= 2) {
- for (N = k; N+RD < A->nt; N += 2*RD) {
- tempNRDn = N+RD == A->nt-1 ? A->n-(N+RD)*A->nb : A->nb;
- ldbN = BLKLDD(B, N );
- ldbNRD = BLKLDD(B, N+RD);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
-
- node = B->get_rankof( B, N+RD, n );
- RUNTIME_data_migrate( sequence, B(N, n), node );
- RUNTIME_data_migrate( sequence, B(N+RD, n), node );
+ for (RD = BS; RD < C->mt-k; RD *= 2) {
+ for (p = k; p+RD < C->mt; p += 2*RD) {
+ m = p+RD;
+
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+ ldcp = BLKLDD(C, p);
+
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TT kernel */
INSERT_TASK_ztpmlqt(
&options,
side, trans,
- tempNRDn, tempnn, tempkm, tempnn, ib, T->nb,
- A (k, N+RD), ldak,
- T2(k, N+RD), T->mb,
- B (N, n), ldbN,
- B (N+RD, n), ldbNRD);
+ tempmm, tempnn, tempkm, tempnn, ib, T->nb,
+ A (k, m), ldak,
+ T2(k, m), T->mb,
+ C (p, n), ldcp,
+ C (m, n), ldcm);
}
- RUNTIME_data_flush( sequence, A (k, N+RD) );
- RUNTIME_data_flush( sequence, T2(k, N+RD) );
+ RUNTIME_data_flush( sequence, A (k, m) );
+ RUNTIME_data_flush( sequence, T2(k, m) );
}
}
/* Restore the original location of the tiles */
- for (n = 0; n < B->nt; n++) {
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, k, n ) );
+ for (n = 0; n < C->nt; n++) {
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, k, n ) );
}
RUNTIME_iteration_pop(chamctxt);
}
- } else {
- /*
- * ChamLeft / ChamConjTrans
- */
- for (k = K-1; k >= 0; k--) {
+ }
+ /*
+ * ChamLeft / ChamConjTrans
+ */
+ else {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
tempkm = k == A->mt-1 ? A->m-k*A->mb : A->mb;
ldak = BLKLDD(A, k);
lddk = BLKLDD(D, k);
+
lastRD = 0;
- for (RD = BS; RD < A->nt-k; RD *= 2)
+ for (RD = BS; RD < C->mt-k; RD *= 2)
lastRD = RD;
for (RD = lastRD; RD >= BS; RD /= 2) {
- for (N = k; N+RD < A->nt; N += 2*RD) {
- tempNRDn = N+RD == A->nt-1 ? A->n-(N+RD)*A->nb : A->nb;
- ldbN = BLKLDD(B, N );
- ldbNRD = BLKLDD(B, N+RD);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (p = k; p+RD < C->mt; p += 2*RD) {
+ m = p+RD;
- node = B->get_rankof( B, N+RD, n );
- RUNTIME_data_migrate( sequence, B(N, n), node );
- RUNTIME_data_migrate( sequence, B(N+RD, n), node );
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+ ldcp = BLKLDD(C, p);
+
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TT kernel */
INSERT_TASK_ztpmlqt(
- &options,
- side, trans,
- tempNRDn, tempnn, tempkm, tempnn, ib, T->nb,
- A (k, N+RD), ldak,
- T2(k, N+RD), T->mb,
- B (N, n), ldbN,
- B (N+RD, n), ldbNRD);
+ &options, side, trans,
+ tempmm, tempnn, tempkm, tempnn, ib, T->nb,
+ A (k, m), ldak,
+ T2(k, m), T->mb,
+ C (p, n), ldcp,
+ C (m, n), ldcm);
}
- RUNTIME_data_flush( sequence, A (k, N+RD) );
- RUNTIME_data_flush( sequence, T2(k, N+RD) );
+ RUNTIME_data_flush( sequence, A (k, m) );
+ RUNTIME_data_flush( sequence, T2(k, m) );
}
}
- for (N = k; N < A->nt; N += BS) {
- tempNn = N == A->nt-1 ? A->n-N*A->nb : A->nb;
- tempkmin = chameleon_min(tempkm,tempNn);
- ldbN = BLKLDD(B, N);
- for (m = chameleon_min(N+BS, A->nt)-1; m > N; m--) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
-
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(N, n), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ for (p = k; p < C->mt; p += BS) {
+ ldcp = BLKLDD(C, p);
+
+ for (m = chameleon_min(p+BS, C->mt)-1; m > p; 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;
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TS kernel */
INSERT_TASK_ztpmlqt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkm, 0, ib, T->nb,
A(k, m), ldak,
T(k, m), T->mb,
- B(N, n), ldbN,
- B(m, n), ldbm);
+ C(p, n), ldcp,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, m) );
RUNTIME_data_flush( sequence, T(k, m) );
}
+
+ temppm = p == C->mt-1 ? C->m-p*C->mb : C->mb;
+ tempkmin = chameleon_min( temppm, tempkm );
+
if ( genD ) {
- int tempDNn = N == D->nt-1 ? D->n-N*D->nb : D->nb;
+ int tempDpn = p == D->nt-1 ? D->n-p*D->nb : D->nb;
+
INSERT_TASK_zlacpy(
&options,
- ChamUpper, tempkmin, tempDNn, A->nb,
- A(k, N), ldak,
- D(k, N), lddk );
+ ChamUpper, tempkmin, tempDpn, A->nb,
+ A(k, p), ldak,
+ D(k, p), lddk );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamLower, tempkmin, tempDNn,
+ ChamLower, tempkmin, tempDpn,
0., 1.,
- D(k, N), lddk );
+ D(k, p), lddk );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- RUNTIME_data_migrate( sequence, B(N, n),
- B->get_rankof( B, N, n ) );
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ RUNTIME_data_migrate( sequence, C(p, n),
+ C->get_rankof( C, p, n ) );
INSERT_TASK_zunmlq(
- &options,
- side, trans,
- tempNn, tempnn,
- tempkmin, ib, T->nb,
- D(k, N), lddk,
- T(k, N), T->mb,
- B(N, n), ldbN);
+ &options, side, trans,
+ temppm, tempnn, tempkmin, ib, T->nb,
+ D(k, p), lddk,
+ T(k, p), T->mb,
+ C(p, n), ldcp);
}
- RUNTIME_data_flush( sequence, D(k, N) );
- RUNTIME_data_flush( sequence, T(k, N) );
+ RUNTIME_data_flush( sequence, D(k, p) );
+ RUNTIME_data_flush( sequence, T(k, p) );
}
RUNTIME_iteration_pop(chamctxt);
}
}
}
+ /*
+ * ChamRight / ChamNoTrans
+ */
else {
if (trans == ChamNoTrans) {
- /*
- * ChamRight / ChamNoTrans
- */
- for (k = K-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
tempkm = k == A->mt-1 ? A->m-k*A->mb : A->mb;
ldak = BLKLDD(A, k);
lddk = BLKLDD(D, k);
lastRD = 0;
- for (RD = BS; RD < A->nt-k; RD *= 2)
+ for (RD = BS; RD < C->nt-k; RD *= 2)
lastRD = RD;
for (RD = lastRD; RD >= BS; RD /= 2) {
- for (N = k; N+RD < A->nt; N += 2*RD) {
- tempNRDn = N+RD == A->nt-1 ? A->n-(N+RD)*A->nb : A->nb;
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+ for (p = k; p+RD < C->nt; p += 2*RD) {
+ n = p+RD;
+
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, N+RD );
- RUNTIME_data_migrate( sequence, B(m, N), node );
- RUNTIME_data_migrate( sequence, B(m, N+RD), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TT kernel */
INSERT_TASK_ztpmlqt(
- &options,
- side, trans,
- tempmm, tempNRDn, tempkm, tempNRDn, ib, T->nb,
- A (k, N+RD), ldak,
- T2(k, N+RD), T->mb,
- B (m, N ), ldbm,
- B (m, N+RD), ldbm);
+ &options, side, trans,
+ tempmm, tempnn, tempkm, tempnn, ib, T->nb,
+ A (k, n), ldak,
+ T2(k, n), T->mb,
+ C (m, p), ldcm,
+ C (m, n), ldcm);
}
- RUNTIME_data_flush( sequence, A (k, N+RD) );
- RUNTIME_data_flush( sequence, T2(k, N+RD) );
+ RUNTIME_data_flush( sequence, A (k, n) );
+ RUNTIME_data_flush( sequence, T2(k, n) );
}
}
- for (N = k; N < A->nt; N += BS) {
- tempNn = N == A->nt-1 ? A->n-N*A->nb : A->nb;
- tempkmin = chameleon_min(tempkm,tempNn);
- for (n = chameleon_min(N+BS, A->nt)-1; n > N; n--) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
-
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(m, N), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ for (p = k; p < C->nt; p += BS) {
+
+ for (n = chameleon_min(p+BS, C->nt)-1; n > p; n--) {
+
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TS kernel */
INSERT_TASK_ztpmlqt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkm, 0, ib, T->nb,
A(k, n), ldak,
T(k, n), T->mb,
- B(m, N), ldbm,
- B(m, n), ldbm);
+ C(m, p), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, n) );
RUNTIME_data_flush( sequence, T(k, n) );
}
+
+ temppn = p == C->nt-1 ? C->n - p * C->nb : C->nb;
+ tempkmin = chameleon_min( temppn, tempkm );
+
if ( genD ) {
- int tempDNn = N == D->nt-1 ? D->n-N*D->nb : D->nb;
+ int tempDpn = p == D->nt-1 ? D->n-p*D->nb : D->nb;
+
INSERT_TASK_zlacpy(
&options,
- ChamUpper, tempkmin, tempDNn, A->nb,
- A(k, N), ldak,
- D(k, N), lddk );
+ ChamUpper, tempkmin, tempDpn, A->nb,
+ A(k, p), ldak,
+ D(k, p), lddk );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamLower, tempkmin, tempDNn,
+ ChamLower, tempkmin, tempDpn,
0., 1.,
- D(k, N), lddk );
+ D(k, p), lddk );
#endif
}
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
- RUNTIME_data_migrate( sequence, B(m, N),
- B->get_rankof( B, m, N ) );
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+
+ RUNTIME_data_migrate( sequence, C(m, p),
+ C->get_rankof( C, m, p ) );
INSERT_TASK_zunmlq(
- &options,
- side, trans,
- tempmm, tempNn,
- tempkmin, ib, T->nb,
- D(k, N), lddk,
- T(k, N), T->mb,
- B(m, N), ldbm);
+ &options, side, trans,
+ tempmm, temppn, tempkmin, ib, T->nb,
+ D(k, p), lddk,
+ T(k, p), T->mb,
+ C(m, p), ldcm);
}
- RUNTIME_data_flush( sequence, D(k, N) );
- RUNTIME_data_flush( sequence, T(k, N) );
+ RUNTIME_data_flush( sequence, D(k, p) );
+ RUNTIME_data_flush( sequence, T(k, p) );
}
-
RUNTIME_iteration_pop(chamctxt);
}
- } else {
- /*
- * ChamRight / ChamConjTrans
- */
- for (k = 0; k < K; k++) {
+ }
+ /*
+ * ChamRight / ChamConjTrans
+ */
+ else {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
tempkm = k == A->mt-1 ? A->m-k*A->mb : A->mb;
ldak = BLKLDD(A, k);
lddk = BLKLDD(D, k);
- for (N = k; N < A->nt; N += BS) {
- tempNn = N == A->nt-1 ? A->n-N*A->nb : A->nb;
- tempkmin = chameleon_min(tempkm,tempNn);
+
+ for (p = k; p < C->nt; p += BS) {
+ temppn = p == C->nt - 1 ? C->n - p * C->nb : C->nb;
+ tempkmin = chameleon_min( temppn, tempkm );
+
if ( genD ) {
- int tempDNn = N == D->nt-1 ? D->n-N*D->nb : D->nb;
+ int tempDpn = p == D->nt-1 ? D->n-p*D->nb : D->nb;
+
INSERT_TASK_zlacpy(
&options,
- ChamUpper, tempkmin, tempDNn, A->nb,
- A(k, N), ldak,
- D(k, N), lddk );
+ ChamUpper, tempkmin, tempDpn, A->nb,
+ A(k, p), ldak,
+ D(k, p), lddk );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamLower, tempkmin, tempDNn,
+ ChamLower, tempkmin, tempDpn,
0., 1.,
- D(k, N), lddk );
+ D(k, p), lddk );
#endif
}
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+
+ for (m = 0; m < C->mt; m++) {
+ ldcm = BLKLDD(C, m);
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
INSERT_TASK_zunmlq(
- &options,
- side, trans,
- tempmm, tempNn,
- tempkmin, ib, T->nb,
- D(k, N), lddk,
- T(k, N), T->mb,
- B(m, N), ldbm);
+ &options, side, trans,
+ tempmm, temppn, tempkmin, ib, T->nb,
+ D(k, p), lddk,
+ T(k, p), T->mb,
+ C(m, p), ldcm);
}
- RUNTIME_data_flush( sequence, D(k, N) );
- RUNTIME_data_flush( sequence, T(k, N) );
+ RUNTIME_data_flush( sequence, D(k, p) );
+ RUNTIME_data_flush( sequence, T(k, p) );
- for (n = N+1; n < chameleon_min(N+BS, A->nt); n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (n = p+1; n < chameleon_min(p+BS, C->nt); n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(m, N), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TS kernel */
INSERT_TASK_ztpmlqt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkm, 0, ib, T->nb,
A(k, n), ldak,
T(k, n), T->mb,
- B(m, N), ldbm,
- B(m, n), ldbm);
+ C(m, p), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(k, n) );
RUNTIME_data_flush( sequence, T(k, n) );
}
}
- for (RD = BS; RD < A->nt-k; RD *= 2) {
- for (N = k; N+RD < A->nt; N += 2*RD) {
- tempNRDn = N+RD == A->nt-1 ? A->n-(N+RD)*A->nb : A->nb;
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (RD = BS; RD < C->nt-k; RD *= 2) {
+ for (p = k; p+RD < C->nt; p += 2*RD) {
+ n = p + RD;
+ tempnn = n == C->mt-1 ? C->m-n*C->mb : C->mb;
+
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, N+RD );
- RUNTIME_data_migrate( sequence, B(m, N), node );
- RUNTIME_data_migrate( sequence, B(m, N+RD), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TT kernel */
INSERT_TASK_ztpmlqt(
&options,
side, trans,
- tempmm, tempNRDn, tempkm, tempNRDn, ib, T->nb,
- A (k, N+RD), ldak,
- T2(k, N+RD), T->mb,
- B (m, N ), ldbm,
- B (m, N+RD), ldbm);
+ tempmm, tempnn, tempkm, tempnn, ib, T->nb,
+ A (k, n), ldak,
+ T2(k, n), T->mb,
+ C (m, p), ldcm,
+ C (m, n), ldcm);
}
- RUNTIME_data_flush( sequence, A (k, N+RD) );
- RUNTIME_data_flush( sequence, T2(k, N+RD) );
+ RUNTIME_data_flush( sequence, A (k, n) );
+ RUNTIME_data_flush( sequence, T2(k, n) );
}
}
/* Restore the original location of the tiles */
- for (m = 0; m < B->mt; m++) {
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, k ) );
+ for (m = 0; m < C->mt; m++) {
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, k ) );
}
RUNTIME_iteration_pop(chamctxt);
diff --git a/compute/pzunmqr.c b/compute/pzunmqr.c
index c872ea0de59ea6122a7cd358cbd210d147c5da90..3a9e93bbee000aab330e562ee82bb6befd44d5db 100644
--- a/compute/pzunmqr.c
+++ b/compute/pzunmqr.c
@@ -27,7 +27,7 @@
#include "control/common.h"
#define A(m,n) A, m, n
-#define B(m,n) B, m, n
+#define C(m,n) C, m, n
#define T(m,n) T, m, n
#define D(k) D, k, k
@@ -35,7 +35,7 @@
* Parallel application of Q using tile V - QR factorization - dynamic scheduling
*/
void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
- CHAM_desc_t *A, CHAM_desc_t *B, CHAM_desc_t *T, CHAM_desc_t *D,
+ CHAM_desc_t *A, CHAM_desc_t *C, CHAM_desc_t *T, CHAM_desc_t *D,
RUNTIME_sequence_t *sequence, RUNTIME_request_t *request )
{
CHAM_context_t *chamctxt;
@@ -44,9 +44,9 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
size_t ws_host = 0;
int k, m, n;
- int ldak, ldbk, ldam, ldan, ldbm, lddk;
- int tempkm, tempnn, tempkmin, tempmm, tempkn;
- int ib, minMT, minM;
+ int ldak, ldck, ldam, ldan, ldcm, lddk;
+ int tempkm, tempkn, tempkmin, tempmm, tempnn;
+ int ib, KT, K;
chamctxt = chameleon_context_self();
if (sequence->status != CHAMELEON_SUCCESS) {
@@ -57,11 +57,11 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
ib = CHAMELEON_IB;
if (A->m > A->n) {
- minM = A->n;
- minMT = A->nt;
+ KT = A->nt;
+ K = A->n;
} else {
- minM = A->m;
- minMT = A->mt;
+ KT = A->mt;
+ K = A->m;
}
if ( D == NULL ) {
@@ -94,14 +94,16 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
/*
* ChamLeft / ChamConjTrans
*/
- for (k = 0; k < minMT; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
- tempkm = k == B->mt-1 ? B->m-k*B->mb : B->mb;
- tempkmin = k == minMT-1 ? minM-k*A->nb : A->nb;
+ tempkm = k == C->mt - 1 ? C->m - k * C->mb : C->mb;
+ tempkmin = k == KT - 1 ? K - k * A->nb : A->nb;
+
ldak = BLKLDD(A, k);
+ ldck = BLKLDD(C, k);
lddk = BLKLDD(D, k);
- ldbk = BLKLDD(B, k);
+
if ( genD ) {
int tempDkm = k == D->mt-1 ? D->m-k*D->mb : D->mb;
@@ -118,29 +120,29 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
D(k), lddk );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zunmqr(
&options,
side, trans,
tempkm, tempnn, tempkmin, ib, T->nb,
D(k), lddk,
T(k, k), T->mb,
- B(k, n), ldbk);
+ C(k, n), ldck);
}
RUNTIME_data_flush( sequence, D(k) );
RUNTIME_data_flush( sequence, T(k, k) );
- for (m = k+1; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+ for (m = k+1; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
ldam = BLKLDD(A, m);
- ldbm = BLKLDD(B, m);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ ldcm = BLKLDD(C, m);
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, m, n ) );
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, m, n ) );
/* TS kernel */
INSERT_TASK_ztpmqrt(
@@ -149,8 +151,8 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
tempmm, tempnn, tempkmin, 0, ib, T->nb,
A(m, k), ldam,
T(m, k), T->mb,
- B(k, n), ldbk,
- B(m, n), ldbm);
+ C(k, n), ldck,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(m, k) );
@@ -158,9 +160,9 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
}
/* Restore the original location of the tiles */
- for (n = 0; n < B->nt; n++) {
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, k, n ) );
+ for (n = 0; n < C->nt; n++) {
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, k, n ) );
}
RUNTIME_iteration_pop(chamctxt);
@@ -170,23 +172,25 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
* ChamLeft / ChamNoTrans
*/
else {
- for (k = minMT-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
- tempkm = k == B->mt-1 ? B->m-k*B->mb : B->mb;
- tempkmin = k == minMT-1 ? minM-k*A->nb : A->nb;
+ tempkm = k == C->mt - 1 ? C->m - k * C->mb : C->mb;
+ tempkmin = k == KT - 1 ? K - k * A->nb : A->nb;
+
ldak = BLKLDD(A, k);
- ldbk = BLKLDD(B, k);
+ ldck = BLKLDD(C, k);
lddk = BLKLDD(D, k);
- for (m = B->mt-1; m > k; m--) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+
+ for (m = C->mt-1; m > k; m--) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
ldam = BLKLDD(A, m);
- ldbm = BLKLDD(B, m);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ ldcm = BLKLDD(C, m);
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, m, n ) );
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, m, n ) );
/* TS kernel */
INSERT_TASK_ztpmqrt(
@@ -195,32 +199,34 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
tempmm, tempnn, tempkmin, 0, ib, T->nb,
A(m, k), ldam,
T(m, k), T->mb,
- B(k, n), ldbk,
- B(m, n), ldbm);
+ C(k, n), ldck,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(m, k) );
RUNTIME_data_flush( sequence, T(m, k) );
}
if ( genD ) {
+ int tempDkm = k == D->mt-1 ? D->m-k*D->mb : D->mb;
+
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempkm, tempkmin, A->nb,
+ ChamLower, tempDkm, tempkmin, A->nb,
A(k, k), ldak,
D(k), lddk );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempkm, tempkmin,
+ ChamUpper, tempDkm, tempkmin,
0., 1.,
D(k), lddk );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, k, n ) );
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, k, n ) );
INSERT_TASK_zunmqr(
&options,
@@ -228,7 +234,7 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
tempkm, tempnn, tempkmin, ib, T->nb,
D(k), lddk,
T(k, k), T->mb,
- B(k, n), ldbk);
+ C(k, n), ldck);
}
RUNTIME_data_flush( sequence, D(k) );
RUNTIME_data_flush( sequence, T(k, k) );
@@ -241,22 +247,23 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
*/
else {
if (trans == ChamConjTrans) {
- for (k = minMT-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
- tempkn = k == B->nt - 1 ? B->n - k * B->nb : B->nb;
- tempkmin = k == minMT - 1 ? minM - k * A->nb : A->nb;
+ tempkn = k == C->nt - 1 ? C->n - k * C->nb : C->nb;
+ tempkmin = k == KT - 1 ? K - k * A->nb : A->nb;
ldak = BLKLDD(A, k);
lddk = BLKLDD(D, k);
- for (n = B->nt-1; n > k; n--) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+
+ for (n = C->nt-1; n > k; n--) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
ldan = BLKLDD(A, n);
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, n ) );
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, n ) );
/* TS kernel */
INSERT_TASK_ztpmqrt(
@@ -265,41 +272,44 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
tempmm, tempnn, tempkmin, 0, ib, T->nb,
A(n, k), ldan,
T(n, k), T->mb,
- B(m, k), ldbm,
- B(m, n), ldbm);
+ C(m, k), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(n, k) );
RUNTIME_data_flush( sequence, T(n, k) );
}
+
if ( genD ) {
+ int tempDkm = k == D->mt-1 ? D->m-k*D->mb : D->mb;
+
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempkn, tempkmin, A->nb,
+ ChamLower, tempDkm, tempkmin, A->nb,
A(k, k), ldak,
D(k), lddk );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempkn, tempkmin,
+ ChamUpper, tempDkm, tempkmin,
0., 1.,
D(k), lddk );
#endif
}
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, k ) );
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, k ) );
INSERT_TASK_zunmqr(
&options,
side, trans,
tempmm, tempkn, tempkmin, ib, T->nb,
- D(k), lddk,
+ D(k), lddk,
T(k, k), T->mb,
- B(m, k), ldbm);
+ C(m, k), ldcm);
}
RUNTIME_data_flush( sequence, D(k) );
@@ -312,51 +322,54 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
* ChamRight / ChamNoTrans
*/
else {
- for (k = 0; k < minMT; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
- tempkn = k == B->nt-1 ? B->n-k*B->nb : B->nb;
- tempkmin = k == minMT-1 ? minM-k*A->nb : A->nb;
+ tempkn = k == C->nt - 1 ? C->n - k * C->nb : C->nb;
+ tempkmin = k == KT - 1 ? K - k * A->nb : A->nb;
ldak = BLKLDD(A, k);
lddk = BLKLDD(D, k);
+
if ( genD ) {
+ int tempDkm = k == D->mt - 1 ? D->m - k * D->mb : D->mb;
+
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempkn, tempkmin, A->nb,
+ ChamLower, tempDkm, tempkmin, A->nb,
A(k, k), ldak,
- D(k), lddk );
+ D(k), lddk );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempkn, tempkmin,
+ ChamUpper, tempDkm, tempkmin,
0., 1.,
D(k), lddk );
#endif
}
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
INSERT_TASK_zunmqr(
&options,
side, trans,
tempmm, tempkn, tempkmin, ib, T->nb,
- D(k), lddk,
+ D(k), lddk,
T(k, k), T->mb,
- B(m, k), ldbm);
+ C(m, k), ldcm);
}
RUNTIME_data_flush( sequence, D(k) );
RUNTIME_data_flush( sequence, T(k, k) );
- for (n = k+1; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = k+1; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
ldan = BLKLDD(A, n);
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, n ) );
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, n ) );
/* TS kernel */
INSERT_TASK_ztpmqrt(
@@ -365,8 +378,8 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
tempmm, tempnn, tempkmin, 0, ib, T->nb,
A(n, k), ldan,
T(n, k), T->mb,
- B(m, k), ldbm,
- B(m, n), ldbm);
+ C(m, k), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(n, k) );
@@ -374,9 +387,9 @@ void chameleon_pzunmqr( int genD, cham_side_t side, cham_trans_t trans,
}
/* Restore the original location of the tiles */
- for (m = 0; m < B->mt; m++) {
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, k ) );
+ for (m = 0; m < C->mt; m++) {
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, k ) );
}
RUNTIME_iteration_pop(chamctxt);
diff --git a/compute/pzunmqr_param.c b/compute/pzunmqr_param.c
index 923a3f8a73a5dc53f07804fe23e357d3ae14c708..a3905647c0767cbaad4f4a8be98e1397b6996924 100644
--- a/compute/pzunmqr_param.c
+++ b/compute/pzunmqr_param.c
@@ -22,7 +22,7 @@
#include <stdlib.h>
#define A(m,n) A, m, n
-#define B(m,n) B, m, n
+#define C(m,n) C, m, n
#define T(m,n) T, m, n
#define D(m,n) D, m, n
@@ -31,7 +31,7 @@
*/
void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
cham_side_t side, cham_trans_t trans,
- CHAM_desc_t *A, CHAM_desc_t *B,
+ CHAM_desc_t *A, CHAM_desc_t *C,
CHAM_desc_t *TS, CHAM_desc_t *TT, CHAM_desc_t *D,
RUNTIME_sequence_t *sequence, RUNTIME_request_t *request )
{
@@ -42,9 +42,9 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
size_t ws_host = 0;
int k, m, n, i, p;
- int ldam, ldan, ldbm, ldbp, lddn, lddm;
- int tempnn, tempkmin, tempmm, tempkn;
- int ib, K, L;
+ int ldap, ldam, ldan, lddp, ldcp, ldcm;
+ int temppm, temppn, tempmm, tempnn, tempkn,tempkmin;
+ int ib, KT, L;
int node, nbtiles, *tiles;
chamctxt = chameleon_context_self();
@@ -55,8 +55,6 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
ib = CHAMELEON_IB;
- K = chameleon_min(A->mt, A->nt);
-
if ( D == NULL ) {
D = A;
genD = 0;
@@ -71,7 +69,7 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
#if defined(CHAMELEON_USE_CUDA)
/* Worker space
*
- * zunmqr = A->nb * ib
+ * zunmqr = A->nb * ib
* ztpmqrt = 3 * A->nb * ib
*/
ws_worker = chameleon_max( ws_worker, ib * A->nb * 3 );
@@ -85,54 +83,55 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
/* Initialisation of tiles */
tiles = (int*)calloc( qrtree->mt, sizeof(int) );
+ KT = chameleon_min( A->mt, A->nt );
if (side == ChamLeft ) {
if (trans == ChamConjTrans) {
/*
* ChamLeft / ChamConjTrans
*/
- for (k = 0; k < K; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
- tempkn = k == A->nt-1 ? A->n-k*A->nb : A->nb;
+ tempkn = k == A->nt - 1 ? A->n - k * A->nb : A->nb;
T = TS;
for (i = 0; i < qrtree->getnbgeqrf(qrtree, k); i++) {
- m = qrtree->getm(qrtree, k, i);
+ p = qrtree->getm(qrtree, k, i);
- tempmm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- tempkmin = chameleon_min(tempmm, tempkn);
- ldam = BLKLDD(A, m);
- lddm = BLKLDD(D, m);
- ldbm = BLKLDD(B, m);
+ temppm = p == C->mt-1 ? C->m - p * C->mb : C->mb;
+ tempkmin = chameleon_min( temppm, tempkn );
+
+ ldap = BLKLDD(A, p);
+ lddp = BLKLDD(D, p);
+ ldcp = BLKLDD(C, p);
if ( genD ) {
- int tempDmm = m == D->mt-1 ? D->m-m*D->mb : D->mb;
+ int tempDpm = p == D->mt-1 ? D->m-p*D->mb : D->mb;
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempDmm, tempkmin, A->nb,
- A(m, k), ldam,
- D(m, k), lddm );
+ ChamLower, tempDpm, tempkmin, A->nb,
+ A(p, k), ldap,
+ D(p, k), lddp );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempDmm, tempkmin,
+ ChamUpper, tempDpm, tempkmin,
0., 1.,
- D(m, k), lddm );
+ D(p, k), lddp );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zunmqr(
- &options,
- side, trans,
- tempmm, tempnn, tempkmin, ib, T->nb,
- D(m, k), lddm,
- T(m, k), T->mb,
- B(m, n), ldbm);
+ &options, side, trans,
+ temppm, tempnn, tempkmin, ib, T->nb,
+ D(p, k), lddp,
+ T(p, k), T->mb,
+ C(p, n), ldcp);
}
- RUNTIME_data_flush( sequence, D(m, k) );
- RUNTIME_data_flush( sequence, T(m, k) );
+ RUNTIME_data_flush( sequence, D(p, k) );
+ RUNTIME_data_flush( sequence, T(p, k) );
}
/* Setting the order of the tiles*/
@@ -142,10 +141,10 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
m = tiles[i];
p = qrtree->currpiv(qrtree, k, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
ldam = BLKLDD(A, m);
- ldbm = BLKLDD(B, m);
- ldbp = BLKLDD(B, p);
+ ldcp = BLKLDD(C, p);
+ ldcm = BLKLDD(C, m);
if( qrtree->gettype(qrtree, k, m) == LIBHQR_KILLED_BY_TS ) {
/* TS kernel */
@@ -157,30 +156,29 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
L = tempmm;
T = TT;
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(p, n), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
INSERT_TASK_ztpmqrt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkn, L, ib, T->nb,
A(m, k), ldam,
T(m, k), T->mb,
- B(p, n), ldbp,
- B(m, n), ldbm);
+ C(p, n), ldcp,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(m, k) );
RUNTIME_data_flush( sequence, T(m, k) );
}
/* Restore the original location of the tiles */
- for (n = 0; n < B->nt; n++) {
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, k, n ) );
+ for (n = 0; n < C->nt; n++) {
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, k, n ) );
}
RUNTIME_iteration_pop(chamctxt);
@@ -190,7 +188,7 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
* ChamLeft / ChamNoTrans
*/
else {
- for (k = K-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
tempkn = k == A->nt-1 ? A->n-k*A->nb : A->nb;
@@ -198,14 +196,14 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
/* Setting the order of the tiles*/
nbtiles = libhqr_walk_stepk( qrtree, k, tiles );
- for (i = nbtiles-1; i >=0; i--) {
+ for (i = nbtiles-1; i >= 0; i--) {
m = tiles[i];
p = qrtree->currpiv(qrtree, k, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
ldam = BLKLDD(A, m);
- ldbm = BLKLDD(B, m);
- ldbp = BLKLDD(B, p);
+ ldcp = BLKLDD(C, p);
+ ldcm = BLKLDD(C, m);
if( qrtree->gettype(qrtree, k, m) == LIBHQR_KILLED_BY_TS ) {
/* TS kernel */
@@ -217,21 +215,20 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
L = tempmm;
T = TT;
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(p, n), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
INSERT_TASK_ztpmqrt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkn, L, ib, T->nb,
A(m, k), ldam,
T(m, k), T->mb,
- B(p, n), ldbp,
- B(m, n), ldbm);
+ C(p, n), ldcp,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(m, k) );
RUNTIME_data_flush( sequence, T(m, k) );
@@ -239,49 +236,48 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
T = TS;
for (i = 0; i < qrtree->getnbgeqrf(qrtree, k); i++) {
- m = qrtree->getm(qrtree, k, i);
+ p = qrtree->getm(qrtree, k, i);
- tempmm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- tempkmin = chameleon_min(tempmm, tempkn);
- ldam = BLKLDD(A, m);
- lddm = BLKLDD(D, m);
- ldbm = BLKLDD(B, m);
+ temppm = p == C->mt-1 ? C->m-p*C->mb : C->mb;
+ tempkmin = chameleon_min( temppm, tempkn );
+
+ ldap = BLKLDD(A, p);
+ lddp = BLKLDD(D, p);
+ ldcp = BLKLDD(C, p);
if ( genD ) {
- int tempDmm = m == D->mt-1 ? D->m-m*D->mb : D->mb;
+ int tempDpm = p == D->mt-1 ? D->m-p*D->mb : D->mb;
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempDmm, tempkmin, A->nb,
- A(m, k), ldam,
- D(m, k), lddm );
+ ChamLower, tempDpm, tempkmin, A->nb,
+ A(p, k), ldap,
+ D(p, k), lddp );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempDmm, tempkmin,
+ ChamUpper, tempDpm, tempkmin,
0., 1.,
- D(m, k), lddm );
+ D(p, k), lddp );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- RUNTIME_data_migrate( sequence, B(m, n),
- B->get_rankof( B, m, n ) );
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ RUNTIME_data_migrate( sequence, C(p, n),
+ C->get_rankof( C, p, n ) );
INSERT_TASK_zunmqr(
- &options,
- side, trans,
- tempmm, tempnn, tempkmin, ib, T->nb,
- D(m, k), lddm,
- T(m, k), T->mb,
- B(m, n), ldbm);
+ &options, side, trans,
+ temppm, tempnn, tempkmin, ib, T->nb,
+ D(p, k), lddp,
+ T(p, k), T->mb,
+ C(p, n), ldcp);
}
-
- RUNTIME_data_flush( sequence, D(m, k) );
- RUNTIME_data_flush( sequence, T(m, k) );
+ RUNTIME_data_flush( sequence, D(p, k) );
+ RUNTIME_data_flush( sequence, T(p, k) );
}
-
RUNTIME_iteration_pop(chamctxt);
}
}
@@ -291,10 +287,10 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
*/
else {
if (trans == ChamConjTrans) {
- for (k = K-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
- tempkn = k == A->nt-1 ? A->n - k*A->nb : A->nb;
+ tempkn = k == A->nt-1 ? A->n - k * A->nb : A->nb;
/* Setting the order of the tiles*/
nbtiles = libhqr_walk_stepk( qrtree, k, tiles );
@@ -303,7 +299,7 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
n = tiles[i];
p = qrtree->currpiv(qrtree, k, n);
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
ldan = BLKLDD(A, n);
if( qrtree->gettype(qrtree, k, n) == LIBHQR_KILLED_BY_TS ) {
@@ -317,22 +313,21 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
T = TT;
}
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(m, p), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
INSERT_TASK_ztpmqrt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkn, chameleon_min( L, tempmm ), ib, T->nb,
A(n, k), ldan,
T(n, k), T->mb,
- B(m, p), ldbm,
- B(m, n), ldbm);
+ C(m, p), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(n, k) );
RUNTIME_data_flush( sequence, T(n, k) );
@@ -340,46 +335,46 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
T = TS;
for (i = 0; i < qrtree->getnbgeqrf(qrtree, k); i++) {
- n = qrtree->getm(qrtree, k, i);
+ p = qrtree->getm(qrtree, k, i);
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- tempkmin = chameleon_min(tempnn, tempkn);
- ldan = BLKLDD(A, n);
- lddn = BLKLDD(D, n);
+ temppn = p == C->nt-1 ? C->n - p * C->nb : C->nb;
+ tempkmin = chameleon_min(temppn, tempkn);
+ ldap = BLKLDD(A, p);
+ lddp = BLKLDD(D, p);
if ( genD ) {
- int tempDnn = n == D->nt-1 ? D->n-n*D->nb : D->nb;
+ int tempDpm = p == D->mt-1 ? D->m-p*D->mb : D->mb;
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempDnn, tempkmin, A->nb,
- A(n, k), ldan,
- D(n, k), lddn );
+ ChamLower, tempDpm, tempkmin, A->nb,
+ A(p, k), ldap,
+ D(p, k), lddp );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempDnn, tempkmin,
+ ChamUpper, tempDpm, tempkmin,
0., 1.,
- D(n, k), lddn );
+ D(p, k), lddp );
#endif
}
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- RUNTIME_data_migrate( sequence, B(m, n),
- B->get_rankof( B, m, n ) );
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+
+ RUNTIME_data_migrate( sequence, C(m, p),
+ C->get_rankof( C, m, p ) );
INSERT_TASK_zunmqr(
- &options,
- side, trans,
- tempmm, tempnn, tempkmin, ib, T->nb,
- D(n, k), lddn,
- T(n, k), T->mb,
- B(m, n), ldbm);
+ &options, side, trans,
+ tempmm, temppn, tempkmin, ib, T->nb,
+ D(p, k), lddp,
+ T(p, k), T->mb,
+ C(m, p), ldcm);
}
- RUNTIME_data_flush( sequence, D(n, k) );
- RUNTIME_data_flush( sequence, T(n, k) );
+ RUNTIME_data_flush( sequence, D(p, k) );
+ RUNTIME_data_flush( sequence, T(p, k) );
}
RUNTIME_iteration_pop(chamctxt);
}
@@ -388,49 +383,49 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
* ChamRight / ChamNoTrans
*/
else {
- for (k = 0; k < K; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
tempkn = k == A->nt-1 ? A->n-k*A->nb : A->nb;
T = TS;
for (i = 0; i < qrtree->getnbgeqrf(qrtree, k); i++) {
- n = qrtree->getm(qrtree, k, i);
+ p = qrtree->getm(qrtree, k, i);
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- tempkmin = chameleon_min(tempnn, tempkn);
- ldan = BLKLDD(A, n);
- lddn = BLKLDD(D, n);
+ temppn = p == C->nt - 1 ? C->n - p * C->nb : C->nb;
+ tempkmin = chameleon_min( temppn, tempkn );
+ ldap = BLKLDD(A, p);
+ lddp = BLKLDD(D, p);
if ( genD ) {
- int tempDnn = n == D->nt-1 ? D->n-n*D->nb : D->nb;
+ int tempDpm = p == D->mt-1 ? D->m-p*D->mb : D->mb;
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempDnn, tempkmin, A->nb,
- A(n, k), ldan,
- D(n, k), lddn );
+ ChamLower, tempDpm, tempkmin, A->nb,
+ A(p, k), ldap,
+ D(p, k), lddp );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempDnn, tempkmin,
+ ChamUpper, tempDpm, tempkmin,
0., 1.,
- D(n, k), lddn );
+ D(p, k), lddp );
#endif
}
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+
+ for (m = 0; m < C->mt; m++) {
+ ldcm = BLKLDD(C, m);
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
INSERT_TASK_zunmqr(
- &options,
- side, trans,
- tempmm, tempnn, tempkmin, ib, T->nb,
- D(n, k), lddn,
- T(n, k), T->mb,
- B(m, n), ldbm);
+ &options, side, trans,
+ tempmm, temppn, tempkmin, ib, T->nb,
+ D(p, k), lddp,
+ T(p, k), T->mb,
+ C(m, p), ldcm);
}
- RUNTIME_data_flush( sequence, D(n, k) );
- RUNTIME_data_flush( sequence, T(n, k) );
+ RUNTIME_data_flush( sequence, D(p, k) );
+ RUNTIME_data_flush( sequence, T(p, k) );
}
/* Setting the order of tiles */
@@ -440,7 +435,7 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
n = tiles[i];
p = qrtree->currpiv(qrtree, k, n);
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
ldan = BLKLDD(A, n);
if( qrtree->gettype(qrtree, k, n) == LIBHQR_KILLED_BY_TS ) {
@@ -454,31 +449,30 @@ void chameleon_pzunmqr_param( int genD, const libhqr_tree_t *qrtree,
T = TT;
}
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(m, p), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
INSERT_TASK_ztpmqrt(
- &options,
- side, trans,
+ &options, side, trans,
tempmm, tempnn, tempkn, chameleon_min( L, tempmm ), ib, T->nb,
A(n, k), ldan,
T(n, k), T->mb,
- B(m, p), ldbm,
- B(m, n), ldbm);
+ C(m, p), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(n, k) );
RUNTIME_data_flush( sequence, T(n, k) );
}
/* Restore the original location of the tiles */
- for (m = 0; m < B->mt; m++) {
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, k ) );
+ for (m = 0; m < C->mt; m++) {
+ RUNTIME_data_migrate( sequence, C( m, k ),
+ C->get_rankof( C, m, k ) );
}
RUNTIME_iteration_pop(chamctxt);
diff --git a/compute/pzunmqrrh.c b/compute/pzunmqrrh.c
index 7ba65c846a48672612bc5554cd6aa8ca7b367b94..1d4500f849c3072d7bbf9e54e0f25c72363a5fbd 100644
--- a/compute/pzunmqrrh.c
+++ b/compute/pzunmqrrh.c
@@ -27,7 +27,7 @@
#include "control/common.h"
#define A(m,n) A, (m), (n)
-#define B(m,n) B, (m), (n)
+#define C(m,n) C, (m), (n)
#define T(m,n) T, (m), (n)
#define T2(m,n) T, (m), ((n)+A->nt)
#define D(m,n) D, (m), (n)
@@ -37,7 +37,7 @@
* Householder) - dynamic scheduling
*/
void chameleon_pzunmqrrh( int genD, int BS, cham_side_t side, cham_trans_t trans,
- CHAM_desc_t *A, CHAM_desc_t *B, CHAM_desc_t *T, CHAM_desc_t *D,
+ CHAM_desc_t *A, CHAM_desc_t *C, CHAM_desc_t *T, CHAM_desc_t *D,
RUNTIME_sequence_t *sequence, RUNTIME_request_t *request )
{
CHAM_context_t *chamctxt;
@@ -45,11 +45,12 @@ void chameleon_pzunmqrrh( int genD, int BS, cham_side_t side, cham_trans_t trans
size_t ws_worker = 0;
size_t ws_host = 0;
- int k, m, n;
- int K, M, RD, lastRD;
- int ldaM, ldam, ldan, ldaMRD, lddM;
- int ldbM, ldbm, ldbMRD;
- int tempMm, tempkn, tempnn, tempmm, tempMRDm, tempkmin;
+ int k, m, n, p;
+ int KT, RD, lastRD;
+ int ldap, ldam, ldan;
+ int ldcp, ldcm;
+ int lddp;
+ int temppm, temppn, tempkn, tempnn, tempmm, tempkmin;
int ib, node;
chamctxt = chameleon_context_self();
@@ -85,61 +86,66 @@ void chameleon_pzunmqrrh( int genD, int BS, cham_side_t side, cham_trans_t trans
RUNTIME_options_ws_alloc( &options, ws_worker, ws_host );
- K = chameleon_min(A->mt, A->nt);
+ KT = chameleon_min( A->mt, A->nt );
if (side == ChamLeft ) {
if (trans == ChamConjTrans) {
/*
* ChamLeft / ChamConjTrans
*/
- for (k = 0; k < K; k++) {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
- tempkn = k == A->nt-1 ? A->n-k*A->nb : A->nb;
- for (M = k; M < A->mt; M += BS) {
- tempMm = M == A->mt-1 ? A->m-M*A->mb : A->mb;
- tempkmin = chameleon_min(tempMm, tempkn);
- ldaM = BLKLDD(A, M);
- lddM = BLKLDD(D, M);
- ldbM = BLKLDD(B, M);
+ tempkn = k == A->nt - 1 ? A->n - k * A->nb : A->nb;
+
+ for (p = k; p < C->mt; p += BS) {
+
+ temppm = p == C->mt-1 ? C->m - p * C->mb : C->mb;
+ tempkmin = chameleon_min( temppm, tempkn );
+
+ ldap = BLKLDD(A, p);
+ lddp = BLKLDD(D, p);
+ ldcp = BLKLDD(C, p);
+
if ( genD ) {
- int tempDMm = M == D->mt-1 ? D->m-M*D->mb : D->mb;
+ int tempDpm = p == D->mt-1 ? D->m-p*D->mb : D->mb;
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempDMm, tempkmin, A->nb,
- A(M, k), ldaM,
- D(M, k), lddM );
+ ChamLower, tempDpm, tempkmin, A->nb,
+ A(p, k), ldap,
+ D(p, k), lddp );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempDMm, tempkmin,
+ ChamUpper, tempDpm, tempkmin,
0., 1.,
- D(M, k), lddM );
+ D(p, k), lddp );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
INSERT_TASK_zunmqr(
&options,
side, trans,
- tempMm, tempnn, tempkmin, ib, T->nb,
- D(M, k), lddM,
- T(M, k), T->mb,
- B(M, n), ldbM);
+ temppm, tempnn, tempkmin, ib, T->nb,
+ D(p, k), lddp,
+ T(p, k), T->mb,
+ C(p, n), ldcp);
}
- RUNTIME_data_flush( sequence, D(M, k) );
- RUNTIME_data_flush( sequence, T(M, k) );
+ RUNTIME_data_flush( sequence, D(p, k) );
+ RUNTIME_data_flush( sequence, T(p, k) );
- for (m = M+1; m < chameleon_min(M+BS, A->mt); m++) {
- tempmm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldbm = BLKLDD(B, m);
+ for (m = p+1; m < chameleon_min(p+BS, C->mt); m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
ldam = BLKLDD(A, m);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ ldcm = BLKLDD(C, m);
+
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(M, n), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TS kernel */
INSERT_TASK_ztpmqrt(
@@ -147,101 +153,108 @@ void chameleon_pzunmqrrh( int genD, int BS, cham_side_t side, cham_trans_t trans
tempmm, tempnn, tempkn, 0, ib, T->nb,
A(m, k), ldam,
T(m, k), T->mb,
- B(M, n), ldbM,
- B(m, n), ldbm);
+ C(p, n), ldcp,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(m, k) );
RUNTIME_data_flush( sequence, T(m, k) );
}
}
- for (RD = BS; RD < A->mt-k; RD *= 2) {
- for (M = k; M+RD < A->mt; M += 2*RD) {
- tempMRDm = M+RD == A->mt-1 ? A->m-(M+RD)*A->mb : A->mb;
- ldbM = BLKLDD(B, M );
- ldbMRD = BLKLDD(B, M+RD);
- ldaMRD = BLKLDD(A, M+RD);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
-
- node = B->get_rankof( B, M+RD, n );
- RUNTIME_data_migrate( sequence, B(M, n), node );
- RUNTIME_data_migrate( sequence, B(M+RD, n), node );
+ for (RD = BS; RD < C->mt-k; RD *= 2) {
+ for (p = k; p+RD < C->mt; p += 2*RD) {
+ m = p+RD;
+
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldam = BLKLDD(A, m);
+ ldcm = BLKLDD(C, m);
+ ldcp = BLKLDD(C, p);
+
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TT kernel */
INSERT_TASK_ztpmqrt(
&options, side, trans,
- tempMRDm, tempnn, tempkn, tempMRDm, ib, T->nb,
- A (M+RD, k), ldaMRD,
- T2(M+RD, k), T->mb,
- B (M, n), ldbM,
- B (M+RD, n), ldbMRD);
+ tempmm, tempnn, tempkn, tempmm, ib, T->nb,
+ A (m, k), ldam,
+ T2(m, k), T->mb,
+ C (p, n), ldcp,
+ C (m, n), ldcm);
}
- RUNTIME_data_flush( sequence, A (M+RD, k) );
- RUNTIME_data_flush( sequence, T2(M+RD, k) );
+ RUNTIME_data_flush( sequence, A (m, k) );
+ RUNTIME_data_flush( sequence, T2(m, k) );
}
}
/* Restore the original location of the tiles */
- for (n = 0; n < B->nt; n++) {
- RUNTIME_data_migrate( sequence, B(k, n),
- B->get_rankof( B, k, n ) );
+ for (n = 0; n < C->nt; n++) {
+ RUNTIME_data_migrate( sequence, C(k, n),
+ C->get_rankof( C, k, n ) );
}
RUNTIME_iteration_pop(chamctxt);
}
- } else {
- /*
- * ChamLeft / ChamNoTrans
- */
- for (k = K-1; k >= 0; k--) {
+ }
+ /*
+ * ChamLeft / ChamNoTrans
+ */
+ else {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
tempkn = k == A->nt-1 ? A->n-k*A->nb : A->nb;
lastRD = 0;
- for (RD = BS; RD < A->mt-k; RD *= 2)
+ for (RD = BS; RD < C->mt-k; RD *= 2)
lastRD = RD;
for (RD = lastRD; RD >= BS; RD /= 2) {
- for (M = k; M+RD < A->mt; M += 2*RD) {
- tempMRDm = M+RD == A->mt-1 ? A->m-(M+RD)*A->mb : A->mb;
- ldbM = BLKLDD(B, M );
- ldbMRD = BLKLDD(B, M+RD);
- ldaMRD = BLKLDD(A, M+RD);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
-
- node = B->get_rankof( B, M+RD, n );
- RUNTIME_data_migrate( sequence, B(M, n), node );
- RUNTIME_data_migrate( sequence, B(M+RD, n), node );
+ for (p = k; p+RD < C->mt; p += 2*RD) {
+ m = p+RD;
+
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldam = BLKLDD(A, m);
+ ldcm = BLKLDD(C, m);
+ ldcp = BLKLDD(C, p);
+
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TT kernel */
INSERT_TASK_ztpmqrt(
&options, side, trans,
- tempMRDm, tempnn, tempkn, tempMRDm, ib, T->nb,
- A (M+RD, k), ldaMRD,
- T2(M+RD, k), T->mb,
- B (M, n), ldbM,
- B (M+RD, n), ldbMRD);
+ tempmm, tempnn, tempkn, tempmm, ib, T->nb,
+ A (m, k), ldam,
+ T2(m, k), T->mb,
+ C (p, n), ldcp,
+ C (m, n), ldcm);
}
- RUNTIME_data_flush( sequence, A (M+RD, k) );
- RUNTIME_data_flush( sequence, T2(M+RD, k) );
+ RUNTIME_data_flush( sequence, A (m, k) );
+ RUNTIME_data_flush( sequence, T2(m, k) );
}
}
- for (M = k; M < A->mt; M += BS) {
- tempMm = M == A->mt-1 ? A->m-M*A->mb : A->mb;
- tempkmin = chameleon_min(tempMm, tempkn);
- ldaM = BLKLDD(A, M);
- lddM = BLKLDD(D, M);
- ldbM = BLKLDD(B, M);
- for (m = chameleon_min(M+BS, A->mt)-1; m > M; m--) {
- tempmm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldbm = BLKLDD(B, m);
+ for (p = k; p < C->mt; p += BS) {
+ ldap = BLKLDD(A, p);
+ lddp = BLKLDD(D, p);
+ ldcp = BLKLDD(C, p);
+
+ for (m = chameleon_min(p+BS, C->mt)-1; m > p; m--) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
ldam = BLKLDD(A, m);
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(M, n), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(p, n), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TS kernel */
INSERT_TASK_ztpmqrt(
@@ -249,100 +262,108 @@ void chameleon_pzunmqrrh( int genD, int BS, cham_side_t side, cham_trans_t trans
tempmm, tempnn, tempkn, 0, ib, T->nb,
A(m, k), ldam,
T(m, k), T->mb,
- B(M, n), ldbM,
- B(m, n), ldbm);
+ C(p, n), ldcp,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(m, k) );
RUNTIME_data_flush( sequence, T(m, k) );
}
+
+ temppm = p == C->mt-1 ? C->m-p*C->mb : C->mb;
+ tempkmin = chameleon_min( temppm, tempkn );
+
if ( genD ) {
- int tempDMm = M == D->mt-1 ? D->m-M*D->mb : D->mb;
+ int tempDpm = p == D->mt-1 ? D->m-p*D->mb : D->mb;
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempDMm, tempkmin, A->nb,
- A(M, k), ldaM,
- D(M, k), lddM );
+ ChamLower, tempDpm, tempkmin, A->nb,
+ A(p, k), ldap,
+ D(p, k), lddp );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempDMm, tempkmin,
+ ChamUpper, tempDpm, tempkmin,
0., 1.,
- D(M, k), lddM );
+ D(p, k), lddp );
#endif
}
- for (n = 0; n < B->nt; n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- RUNTIME_data_migrate( sequence, B(M, n),
- B->get_rankof( B, M, n ) );
+ for (n = 0; n < C->nt; n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+
+ RUNTIME_data_migrate( sequence, C(p, n),
+ C->get_rankof( C, p, n ) );
INSERT_TASK_zunmqr(
&options, side, trans,
- tempMm, tempnn, tempkmin, ib, T->nb,
- D(M, k), lddM,
- T(M, k), T->mb,
- B(M, n), ldbM);
+ temppm, tempnn, tempkmin, ib, T->nb,
+ D(p, k), lddp,
+ T(p, k), T->mb,
+ C(p, n), ldcp);
}
- RUNTIME_data_flush( sequence, D(M, k) );
- RUNTIME_data_flush( sequence, T(M, k) );
+ RUNTIME_data_flush( sequence, D(p, k) );
+ RUNTIME_data_flush( sequence, T(p, k) );
}
RUNTIME_iteration_pop(chamctxt);
}
}
}
+ /*
+ * ChamRight / ChamConjTrans
+ */
else {
if (trans == ChamConjTrans) {
- /*
- * ChamRight / ChamConjTrans
- */
- for (k = K-1; k >= 0; k--) {
+ for (k = KT-1; k >= 0; k--) {
RUNTIME_iteration_push(chamctxt, k);
- tempkn = k == A->nt-1 ? A->n-k*A->nb : A->nb;
+ tempkn = k == A->nt-1 ? A->n - k * A->nb : A->nb;
+
lastRD = 0;
- for (RD = BS; RD < A->mt-k; RD *= 2)
+ for (RD = BS; RD < C->nt-k; RD *= 2)
lastRD = RD;
for (RD = lastRD; RD >= BS; RD /= 2) {
- for (M = k; M+RD < A->mt; M += 2*RD) {
- tempMRDm = M+RD == A->mt-1 ? A->m-(M+RD)*A->mb : A->mb;
- ldaMRD = BLKLDD(A, M+RD);
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+ for (p = k; p+RD < C->nt; p += 2*RD) {
+ n = p+RD;
+
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
+ ldan = BLKLDD(A, n);
+
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, M+RD );
- RUNTIME_data_migrate( sequence, B(m, M), node );
- RUNTIME_data_migrate( sequence, B(m, M+RD), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TT kernel */
INSERT_TASK_ztpmqrt(
&options, side, trans,
- tempmm, tempMRDm, tempkn, tempmm, ib, T->nb,
- A (M+RD, k), ldaMRD,
- T2(M+RD, k), T->mb,
- B (m, M), ldbm,
- B (m, M+RD), ldbm);
+ tempmm, tempnn, tempkn, tempmm, ib, T->nb,
+ A (n, k), ldan,
+ T2(n, k), T->mb,
+ C (m, p), ldcm,
+ C (m, n), ldcm);
}
- RUNTIME_data_flush( sequence, A (M+RD, k) );
- RUNTIME_data_flush( sequence, T2(M+RD, k) );
+ RUNTIME_data_flush( sequence, A (n, k) );
+ RUNTIME_data_flush( sequence, T2(n, k) );
}
}
- for (M = k; M < A->mt; M += BS) {
- tempMm = M == A->mt-1 ? A->m-M*A->mb : A->mb;
- tempkmin = chameleon_min(tempMm, tempkn);
- ldaM = BLKLDD(A, M);
- lddM = BLKLDD(D, M);
- for (n = chameleon_min(M+BS, A->mt)-1; n > M; n--) {
+ for (p = k; p < C->nt; p += BS) {
+
+ for (n = chameleon_min(p+BS, C->nt)-1; n > p; n--) {
+
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
ldan = BLKLDD(A, n);
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(m, M), node );
- RUNTIME_data_migrate( sequence, B(m, m), node );
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TS kernel */
INSERT_TASK_ztpmqrt(
@@ -350,102 +371,111 @@ void chameleon_pzunmqrrh( int genD, int BS, cham_side_t side, cham_trans_t trans
tempmm, tempnn, tempkn, 0, ib, T->nb,
A(n, k), ldan,
T(n, k), T->mb,
- B(m, M), ldbm,
- B(m, n), ldbm);
+ C(m, p), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(n, k) );
RUNTIME_data_flush( sequence, T(n, k) );
}
+
+ temppn = p == C->nt-1 ? C->n - p * C->nb : C->nb;
+ tempkmin = chameleon_min( temppn, tempkn );
+ ldap = BLKLDD(A, p);
+ lddp = BLKLDD(D, p);
+
if ( genD ) {
- int tempDMm = M == D->mt-1 ? D->m-M*D->mb : D->mb;
+ int tempDpm = p == D->mt-1 ? D->m-p*D->mb : D->mb;
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempDMm, tempkmin, A->nb,
- A(M, k), ldaM,
- D(M, k), lddM );
+ ChamLower, tempDpm, tempkmin, A->nb,
+ A(p, k), ldap,
+ D(p, k), lddp );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempDMm, tempkmin,
+ ChamUpper, tempDpm, tempkmin,
0., 1.,
- D(M, k), lddM );
+ D(p, k), lddp );
#endif
}
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- RUNTIME_data_migrate( sequence, B(m, M),
- B->get_rankof( B, m, M ) );
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+
+ RUNTIME_data_migrate( sequence, C(m, p),
+ C->get_rankof( C, m, p ) );
INSERT_TASK_zunmqr(
- &options,
- side, trans,
- tempmm, tempMm, tempkmin, ib, T->nb,
- D(M, k), lddM,
- T(M, k), T->mb,
- B(m, M), ldbm);
+ &options, side, trans,
+ tempmm, temppn, tempkmin, ib, T->nb,
+ D(p, k), lddp,
+ T(p, k), T->mb,
+ C(m, p), ldcm);
}
- RUNTIME_data_flush( sequence, D(M, k) );
- RUNTIME_data_flush( sequence, T(M, k) );
+ RUNTIME_data_flush( sequence, D(p, k) );
+ RUNTIME_data_flush( sequence, T(p, k) );
}
-
RUNTIME_iteration_pop(chamctxt);
}
- } else {
- /*
- * ChamRight / ChamNoTrans
- */
- for (k = 0; k < K; k++) {
+ }
+ /*
+ * ChamRight / ChamNoTrans
+ */
+ else {
+ for (k = 0; k < KT; k++) {
RUNTIME_iteration_push(chamctxt, k);
tempkn = k == A->nt-1 ? A->n-k*A->nb : A->nb;
- for (M = k; M < A->mt; M += BS) {
- tempMm = M == A->mt-1 ? A->m-M*A->mb : A->mb;
- tempkmin = chameleon_min(tempMm, tempkn);
- ldaM = BLKLDD(A, M);
- lddM = BLKLDD(D, M);
+
+ for (p = k; p < C->nt; p += BS) {
+ temppn = p == C->nt - 1 ? C->n - p * C->nb : C->nb;
+ tempkmin = chameleon_min( temppn, tempkn );
+
+ ldap = BLKLDD(A, p);
+ lddp = BLKLDD(D, p);
+
if ( genD ) {
- int tempDMm = M == D->mt-1 ? D->m-M*D->mb : D->mb;
+ int tempDpm = p == D->mt-1 ? D->m-p*D->mb : D->mb;
INSERT_TASK_zlacpy(
&options,
- ChamLower, tempDMm, tempkmin, A->nb,
- A(M, k), ldaM,
- D(M, k), lddM );
+ ChamLower, tempDpm, tempkmin, A->nb,
+ A(p, k), ldap,
+ D(p, k), lddp );
#if defined(CHAMELEON_USE_CUDA)
INSERT_TASK_zlaset(
&options,
- ChamUpper, tempDMm, tempkmin,
+ ChamUpper, tempDpm, tempkmin,
0., 1.,
- D(M, k), lddM );
+ D(p, k), lddp );
#endif
}
- for (m = 0; m < B->mt; m++) {
- ldbm = BLKLDD(B, m);
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
+
+ for (m = 0; m < C->mt; m++) {
+ ldcm = BLKLDD(C, m);
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
INSERT_TASK_zunmqr(
- &options,
- side, trans,
- tempmm, tempMm, tempkmin, ib, T->nb,
- D(M, k), lddM,
- T(M, k), T->mb,
- B(m, M), ldbm);
+ &options, side, trans,
+ tempmm, temppn, tempkmin, ib, T->nb,
+ D(p, k), lddp,
+ T(p, k), T->mb,
+ C(m, p), ldcm);
}
- RUNTIME_data_flush( sequence, D(M, k) );
- RUNTIME_data_flush( sequence, T(M, k) );
+ RUNTIME_data_flush( sequence, D(p, k) );
+ RUNTIME_data_flush( sequence, T(p, k) );
- for (n = M+1; n < chameleon_min(M+BS, A->mt); n++) {
- tempnn = n == B->nt-1 ? B->n-n*B->nb : B->nb;
+ for (n = p+1; n < chameleon_min(p+BS, C->nt); n++) {
+ tempnn = n == C->nt-1 ? C->n-n*C->nb : C->nb;
ldan = BLKLDD(A, n);
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
- node = B->get_rankof( B, m, n );
- RUNTIME_data_migrate( sequence, B(m, M), node );
- RUNTIME_data_migrate( sequence, B(m, n), node );
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TS kernel */
INSERT_TASK_ztpmqrt(
@@ -453,43 +483,45 @@ void chameleon_pzunmqrrh( int genD, int BS, cham_side_t side, cham_trans_t trans
tempmm, tempnn, tempkn, 0, ib, T->nb,
A(n, k), ldan,
T(n, k), T->mb,
- B(m, M), ldbm,
- B(m, n), ldbm);
+ C(m, p), ldcm,
+ C(m, n), ldcm);
}
RUNTIME_data_flush( sequence, A(n, k) );
RUNTIME_data_flush( sequence, T(n, k) );
}
}
- for (RD = BS; RD < A->mt-k; RD *= 2) {
- for (M = k; M+RD < A->mt; M += 2*RD) {
- tempMRDm = M+RD == A->mt-1 ? A->m-(M+RD)*A->mb : A->mb;
- ldaMRD = BLKLDD(A, M+RD);
- for (m = 0; m < B->mt; m++) {
- tempmm = m == B->mt-1 ? B->m-m*B->mb : B->mb;
- ldbm = BLKLDD(B, m);
-
- node = B->get_rankof( B, m, M+RD );
- RUNTIME_data_migrate( sequence, B(m, M), node );
- RUNTIME_data_migrate( sequence, B(m, M+RD), node );
+ for (RD = BS; RD < C->nt-k; RD *= 2) {
+ for (p = k; p+RD < C->nt; p += 2*RD) {
+ n = p + RD;
+ tempnn = n == C->mt-1 ? C->m-n*C->mb : C->mb;
+ ldan = BLKLDD(A, n);
+
+ for (m = 0; m < C->mt; m++) {
+ tempmm = m == C->mt-1 ? C->m-m*C->mb : C->mb;
+ ldcm = BLKLDD(C, m);
+
+ node = C->get_rankof( C, m, n );
+ RUNTIME_data_migrate( sequence, C(m, p), node );
+ RUNTIME_data_migrate( sequence, C(m, n), node );
/* TT kernel */
INSERT_TASK_ztpmqrt(
&options, side, trans,
- tempmm, tempMRDm, tempkn, tempmm, ib, T->nb,
- A (M+RD, k), ldaMRD,
- T2(M+RD, k), T->mb,
- B (m, M ), ldbm,
- B (m, M+RD), ldbm);
+ tempmm, tempnn, tempkn, tempmm, ib, T->nb,
+ A (n, k), ldan,
+ T2(n, k), T->mb,
+ C (m, p), ldcm,
+ C (m, n), ldcm);
}
- RUNTIME_data_flush( sequence, A (M+RD, k) );
- RUNTIME_data_flush( sequence, T2(M+RD, k) );
+ RUNTIME_data_flush( sequence, A (n, k) );
+ RUNTIME_data_flush( sequence, T2(n, k) );
}
}
/* Restore the original location of the tiles */
- for (m = 0; m < B->mt; m++) {
- RUNTIME_data_migrate( sequence, B(m, k),
- B->get_rankof( B, m, k ) );
+ for (m = 0; m < C->mt; m++) {
+ RUNTIME_data_migrate( sequence, C(m, k),
+ C->get_rankof( C, m, k ) );
}
RUNTIME_iteration_pop(chamctxt);