diff --git a/compute/CMakeLists.txt b/compute/CMakeLists.txt
index 3f1dc31dd9b88c61b4f923edf02afcc02538ec7e..efada065d0c7bde35a821c17744bc259aecdb754 100644
--- a/compute/CMakeLists.txt
+++ b/compute/CMakeLists.txt
@@ -116,9 +116,7 @@ set(ZSRC
pzgetrf_nopiv.c
pzlacpy.c
pzlange.c
- pzlanhe.c
pzlansy.c
- pzlantr.c
pzlaset2.c
pzlaset.c
pzlauum.c
diff --git a/compute/pzlange.c b/compute/pzlange.c
index 5a6b15bbad1c39f75ed295f2b17ba159df04959d..94200a379d9facdb2b13f1715512d4e640dc420b 100644
--- a/compute/pzlange.c
+++ b/compute/pzlange.c
@@ -26,323 +26,411 @@
//WS_ADD : A->mb + A->nb
#include "control/common.h"
-#define A(m, n) A, m, n
-#define VECNORMS_STEP1(m, n) VECNORMS_STEP1, m, n
-#define VECNORMS_STEP2(m, n) VECNORMS_STEP2, m, n
-#define RESULT(m, n) RESULT, m, n
-
-/**
- *
- */
-void chameleon_pzlange( cham_normtype_t norm, CHAM_desc_t *A, double *result,
- RUNTIME_sequence_t *sequence, RUNTIME_request_t *request )
+#define A(m, n) A, (m), (n)
+#define Wcol(m, n) Wcol, (m), (n)
+#define Welt(m, n) Welt, (m), (n)
+
+static inline void
+chameleon_pzlange_one( cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A,
+ CHAM_desc_t *Wcol, CHAM_desc_t *Welt,
+ RUNTIME_option_t *options)
{
- CHAM_desc_t *VECNORMS_STEP1 = NULL;
- CHAM_desc_t *VECNORMS_STEP2 = NULL;
- CHAM_desc_t *RESULT = NULL;
- CHAM_context_t *chamctxt;
- RUNTIME_option_t options;
-
- int workm, workn;
- int tempkm, tempkn;
- int ldam;
int m, n;
+ int minMNT = chameleon_min( A->mt, A->nt );
+ int minMN = chameleon_min( A->m, A->n );
+ int MT = (uplo == ChamUpper) ? minMNT : A->mt;
+ int NT = (uplo == ChamLower) ? minMNT : A->nt;
+ int M = (uplo == ChamUpper) ? minMN : A->m;
+ int N = (uplo == ChamLower) ? minMN : A->n;
+ int P = Welt->p;
+ int Q = Welt->q;
+
+ /**
+ * Step 1:
+ * For j in [1,P], W(i, n) = reduce( A(i+k*P, n) )
+ */
+ for(n = 0; n < NT; n++) {
+ int mmin = ( uplo == ChamLower ) ? n : 0;
+ int mmax = ( uplo == ChamUpper ) ? chameleon_min(n+1, MT) : MT;
+
+ int tempnn = ( n == (NT-1) ) ? N - n * A->nb : A->nb;
+
+ for(m = mmin; m < mmax; m++) {
+ int tempmm = ( m == (MT-1) ) ? M - m * A->mb : A->mb;
+ int ldam = BLKLDD( A, m );
+
+ if ( (n == m) && (uplo != ChamUpperLower) ) {
+ INSERT_TASK_ztrasm(
+ options,
+ ChamColumnwise, uplo, diag, tempmm, tempnn,
+ A(m, n), ldam, Wcol(m, n) );
+ }
+ else {
+ INSERT_TASK_dzasum(
+ options,
+ ChamColumnwise, ChamUpperLower, tempmm, tempnn,
+ A(m, n), ldam, Wcol(m, n) );
+ }
- chamctxt = chameleon_context_self();
- if (sequence->status != CHAMELEON_SUCCESS)
- return;
- RUNTIME_options_init(&options, chamctxt, sequence, request);
+ if ( m >= P ) {
+ INSERT_TASK_dgeadd(
+ options,
+ ChamNoTrans, tempnn, 1, A->nb,
+ 1.0, Wcol(m, n), tempnn,
+ 1.0, Wcol(m%P, n), tempnn );
+ }
+ }
- *result = 0.0;
- switch ( norm ) {
- /*
- * ChamOneNorm
+ /**
+ * Step 2:
+ * For each i, W(i, n) = reduce( W(0..P-1, n) )
*/
- case ChamOneNorm:
- /* Init workspace handle for the call to zlange but unused */
- RUNTIME_options_ws_alloc( &options, 1, 0 );
-
- workm = chameleon_max( A->mt, A->p );
- workn = A->n;
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, 1, A->nb, A->nb,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
+ for(m = 1; m < P; n++) {
+ INSERT_TASK_dgeadd(
+ options,
+ ChamNoTrans, tempnn, 1, A->nb,
+ 1.0, Wcol(m, n), tempnn,
+ 1.0, Wcol(0, n), tempnn );
+ }
- CHAMELEON_Desc_Create(&(VECNORMS_STEP2), NULL, ChamRealDouble, 1, A->nb, A->nb,
- 1, workn, 0, 0, 1, workn, A->p, A->q);
+ INSERT_TASK_dlange(
+ options,
+ ChamMaxNorm, tempnn, 1, A->nb,
+ Wcol(0, n), tempnn, Welt(0, n));
+ }
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- 1, 1, 0, 0, 1, 1, 1, 1);
+ /**
+ * Step 3:
+ * For n in 0..Q-1, W(m, n) = max( W(m, n..nt[Q] ) )
+ */
+ for(n = Q; n < NT; n++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(0, n), Welt(0, n%Q) );
+ }
- for(n = A->myrank % A->q; n < A->nt; n+=A->q) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
+ /**
+ * Step 4:
+ * For each i, Welt(i, n) = max( Welt(0..P-1, n) )
+ */
+ for(n = 1; n < Q; n++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(0, n), Welt(0, 0) );
+ }
+}
- /* Zeroes my intermediate vectors */
- for(m = (A->myrank / A->q); m < workm; m+=A->p) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, tempkn,
- 0., 0.,
- VECNORMS_STEP1(m, n), 1);
+static inline void
+chameleon_pzlange_inf( cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A,
+ CHAM_desc_t *Wcol, CHAM_desc_t *Welt,
+ RUNTIME_option_t *options)
+{
+ int m, n;
+ int minMNT = chameleon_min( A->mt, A->nt );
+ int minMN = chameleon_min( A->m, A->n );
+ int MT = (uplo == ChamUpper) ? minMNT : A->mt;
+ int NT = (uplo == ChamLower) ? minMNT : A->nt;
+ int M = (uplo == ChamUpper) ? minMN : A->m;
+ int N = (uplo == ChamLower) ? minMN : A->n;
+ int P = Welt->p;
+ int Q = Welt->q;
+
+ /**
+ * Step 1:
+ * For j in [1,Q], Wcol(m, j) = reduce( A(m, j+k*Q) )
+ */
+ for(m = 0; m < MT; m++) {
+ int nmin = ( uplo == ChamUpper ) ? m : 0;
+ int nmax = ( uplo == ChamLower ) ? chameleon_min(m+1, NT) : NT;
+
+ int tempmm = ( m == (MT-1) ) ? M - m * A->mb : A->mb;
+ int ldam = BLKLDD( A, m );
+
+ for(n = nmin; n < nmax; n++) {
+ int tempnn = ( n == (NT-1) ) ? N - n * A->nb : A->nb;
+
+ if ( (n == m) && (uplo != ChamUpperLower) ) {
+ INSERT_TASK_ztrasm(
+ options,
+ ChamRowwise, uplo, diag, tempmm, tempnn,
+ A(m, n), ldam, Wcol(m, n) );
}
-
- /* compute sums of absolute values on columns of each tile */
- for(m = (A->myrank / A->q); m < A->mt; m+=A->p) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
+ else {
INSERT_TASK_dzasum(
- &options,
- ChamColumnwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(m, n));
+ options,
+ ChamRowwise, ChamUpperLower, tempmm, tempnn,
+ A(m, n), ldam, Wcol(m, n) );
}
- /* Zeroes the second intermediate vector */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, tempkn,
- 0., 0.,
- VECNORMS_STEP2(0, n), 1);
-
- /* Compute vector sums between tiles in columns */
- for(m = 0; m < A->mt; m++) {
+ if ( n >= Q ) {
INSERT_TASK_dgeadd(
- &options,
- ChamNoTrans, 1, tempkn, A->mb,
- 1.0, VECNORMS_STEP1(m, n), 1,
- 1.0, VECNORMS_STEP2(0, n), 1);
+ options,
+ ChamNoTrans, tempmm, 1, A->mb,
+ 1.0, Wcol(m, n ), tempmm,
+ 1.0, Wcol(m, n%Q), tempmm );
}
+ }
- /*
- * Compute max norm of each segment of the final vector in the
- * previous workspace
- */
- INSERT_TASK_dlange(
- &options,
- ChamMaxNorm, 1, tempkn, A->nb,
- VECNORMS_STEP2(0, n), 1,
- VECNORMS_STEP1(0, n));
+ /**
+ * Step 2:
+ * For each j, W(m, j) = reduce( Wcol(m, 0..Q-1) )
+ */
+ for(n = 1; n < Q; n++) {
+ INSERT_TASK_dgeadd(
+ options,
+ ChamNoTrans, tempmm, 1, A->mb,
+ 1.0, Wcol(m, n), tempmm,
+ 1.0, Wcol(m, 0), tempmm );
}
- /* Initialize RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(0,0), 1);
+ INSERT_TASK_dlange(
+ options,
+ ChamMaxNorm, tempmm, 1, A->nb,
+ Wcol(m, 0), 1, Welt(m, 0));
+ }
+
+ /**
+ * Step 3:
+ * For m in 0..P-1, Welt(m, n) = max( Wcol(m..mt[P], n ) )
+ */
+ for(m = P; m < MT; m++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, 0), Welt(m%P, 0) );
+ }
+
+ /**
+ * Step 4:
+ * For each i, Welt(i, n) = max( Welt(0..P-1, n) )
+ */
+ for(m = 1; m < P; m++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, 0), Welt(0, 0) );
+ }
+}
+
+static inline void
+chameleon_pzlange_max( cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A, CHAM_desc_t *Welt,
+ RUNTIME_option_t *options)
+{
+ int m, n;
+ int minMNT = chameleon_min( A->mt, A->nt );
+ int minMN = chameleon_min( A->m, A->n );
+ int MT = (uplo == ChamUpper) ? minMNT : A->mt;
+ int NT = (uplo == ChamLower) ? minMNT : A->nt;
+ int M = (uplo == ChamUpper) ? minMN : A->m;
+ int N = (uplo == ChamLower) ? minMN : A->n;
+ int P = Welt->p;
+ int Q = Welt->q;
+
+ /**
+ * Step 1:
+ * For j in [1,Q], Welt(m, j) = reduce( A(m, j+k*Q) )
+ */
+ for(m = 0; m < MT; m++) {
+ int nmin = ( uplo == ChamUpper ) ? m : 0;
+ int nmax = ( uplo == ChamLower ) ? chameleon_min(m+1, NT) : NT;
+
+ int tempmm = ( m == (MT-1) ) ? M - m * A->mb : A->mb;
+ int ldam = BLKLDD( A, m );
+
+ for(n = nmin; n < nmax; n++) {
+ int tempnn = ( n == (NT-1) ) ? N - n * A->nb : A->nb;
+
+ if ( (n == m) && (uplo != ChamUpperLower) ) {
+ INSERT_TASK_zlantr(
+ options,
+ ChamMaxNorm, uplo, diag, tempmm, tempnn, A->nb,
+ A(m, n), ldam, Welt(m, n));
+ }
+ else {
+ INSERT_TASK_zlange(
+ options,
+ ChamMaxNorm, tempmm, tempnn, A->nb,
+ A(m, n), ldam, Welt(m, n));
+ }
- /* Compute max norm between tiles in the row */
- if (A->myrank < A->q) {
- for(n = 0; n < A->nt; n++) {
+ if ( n >= Q ) {
INSERT_TASK_dlange_max(
- &options,
- VECNORMS_STEP1(0, n),
- RESULT(0,0));
+ options,
+ Welt(m, n), Welt(m, n%Q) );
}
}
- /* Scatter norm over processus */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
+ /**
+ * Step 2:
+ * For each j, W(m, j) = reduce( Welt(m, 0..Q-1) )
+ */
+ for(n = 1; n < Q; n++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, n), Welt(m, 0) );
+ }
+ }
+
+ /**
+ * Step 3:
+ * For m in 0..P-1, Welt(m, n) = max( Welt(m..mt[P], n ) )
+ */
+ for(m = P; m < MT; m++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, 0), Welt(m%P, 0) );
+ }
+
+ /**
+ * Step 4:
+ * For each i, Welt(i, n) = max( Welt(0..P-1, n) )
+ */
+ for(m = 1; m < P; m++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, 0), Welt(0, 0) );
+ }
+}
+
+static inline void
+chameleon_pzlange_frb( cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A, CHAM_desc_t *Welt,
+ RUNTIME_option_t *options)
+{
+ int m, n;
+ int minMNT = chameleon_min( A->mt, A->nt );
+ int minMN = chameleon_min( A->m, A->n );
+ int MT = (uplo == ChamUpper) ? minMNT : A->mt;
+ int NT = (uplo == ChamLower) ? minMNT : A->nt;
+ int M = (uplo == ChamUpper) ? minMN : A->m;
+ int N = (uplo == ChamLower) ? minMN : A->n;
+ int P = Welt->p;
+ int Q = Welt->q;
+
+ /**
+ * Step 1:
+ * For j in [1,Q], Welt(m, j) = reduce( A(m, j+k*Q) )
+ */
+ for(m = 0; m < MT; m++) {
+ int nmin = ( uplo == ChamUpper ) ? m : 0;
+ int nmax = ( uplo == ChamLower ) ? chameleon_min(m+1, NT) : NT;
+
+ int tempmm = ( m == (MT-1) ) ? M - m * A->mb : A->mb;
+ int ldam = BLKLDD( A, m );
+
+ for(n = nmin; n < nmax; n++) {
+ int tempnn = ( n == (NT-1) ) ? N - n * A->nb : A->nb;
+
+ if ( (n == m) && (uplo != ChamUpperLower) ) {
+ INSERT_TASK_ztrssq(
+ options,
+ uplo, diag, tempmm, tempnn,
+ A(m, n), ldam, Welt(m, n) );
+ }
+ else {
+ INSERT_TASK_zgessq(
+ options,
+ tempmm, tempnn,
+ A(m, n), ldam, Welt(m, n) );
+ }
+
+ if ( n >= Q ) {
+ INSERT_TASK_dplssq(
+ options, Welt(m, n), Welt(m, n%Q) );
}
}
- CHAMELEON_Desc_Flush( VECNORMS_STEP2, sequence );
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- CHAMELEON_Desc_Destroy( &(VECNORMS_STEP2) );
- break;
- /*
- * ChamInfNorm
+ /**
+ * Step 2:
+ * For each j, W(m, j) = reduce( Welt(m, 0..Q-1) )
*/
- case ChamInfNorm:
- /* Init workspace handle for the call to zlange */
- RUNTIME_options_ws_alloc( &options, A->mb, 0 );
+ for(n = 1; n < Q; n++) {
+ INSERT_TASK_dplssq(
+ options, Welt(m, n), Welt(m, 0) );
+ }
+ }
- workm = A->m;
- workn = chameleon_max( A->nt, A->q );
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, A->mb, 1, A->mb,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
+ /**
+ * Step 3:
+ * For m in 0..P-1, Welt(m, n) = max( Welt(m..mt[P], n ) )
+ */
+ for(m = P; m < MT; m++) {
+ INSERT_TASK_dplssq(
+ options, Welt(m, 0), Welt(m%P, 0) );
+ }
- CHAMELEON_Desc_Create(&(VECNORMS_STEP2), NULL, ChamRealDouble, A->mb, 1, A->mb,
- workm, 1, 0, 0, workm, 1, A->p, A->q);
+ /**
+ * Step 4:
+ * For each i, Welt(i, n) = max( Welt(0..P-1, n) )
+ */
+ for(m = 1; m < P; m++) {
+ INSERT_TASK_dplssq(
+ options, Welt(m, 0), Welt(0, 0) );
+ }
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- A->p, A->q, 0, 0, A->p, A->q, A->p, A->q);
+ INSERT_TASK_dplssq2(
+ options, Welt(0, 0) );
+}
- for(m = (A->myrank / A->q); m < A->mt; m+=A->p) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
+/**
+ *
+ */
+void chameleon_pzlange_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_diag_t diag,
+ CHAM_desc_t *A, double *result,
+ RUNTIME_sequence_t *sequence, RUNTIME_request_t *request )
+{
+ CHAM_context_t *chamctxt;
+ RUNTIME_option_t options;
+ CHAM_desc_t *Wcol = NULL;
+ CHAM_desc_t *Welt = NULL;
+ double alpha = 0.0;
+ double beta = 0.0;
- /* Zeroes my intermediate vectors */
- for(n = A->myrank % A->q; n < workn; n+=A->q) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- VECNORMS_STEP1(m, n), 1);
- }
+ int workn, workmt, worknt;
+ int m, n;
- /* compute sums of absolute values on rows of each tile */
- for(n = A->myrank % A->q; n < A->nt; n+=A->q) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_dzasum(
- &options,
- ChamRowwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(m, n));
- }
+ chamctxt = chameleon_context_self();
+ if (sequence->status != CHAMELEON_SUCCESS)
+ return;
+ RUNTIME_options_init(&options, chamctxt, sequence, request);
- /* Zeroes the second intermediate vector */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- VECNORMS_STEP2(m, 0), 1);
+ *result = 0.0;
- /* compute vector sums between tiles in rows locally on each rank */
- for(n = A->myrank % A->q + A->q; n < A->nt; n+=A->q) {
- INSERT_TASK_dgeadd(
- &options,
- ChamNoTrans, tempkm, 1, A->mb,
- 1.0, VECNORMS_STEP1(m, n), tempkm,
- 1.0, VECNORMS_STEP1(m, A->myrank % A->q), tempkm);
- }
+ workmt = chameleon_max( A->mt, A->p );
+ worknt = chameleon_max( A->nt, A->q );
+ workn = chameleon_max( A->n, A->q );
- /* compute vector sums between tiles in rows between ranks */
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dgeadd(
- &options,
- ChamNoTrans, tempkm, 1, A->mb,
- 1.0, VECNORMS_STEP1(m, n), tempkm,
- 1.0, VECNORMS_STEP2(m, 0), tempkm);
- }
+ switch ( norm ) {
+ case ChamOneNorm:
+ RUNTIME_options_ws_alloc( &options, 1, 0 );
- /*
- * Compute max norm of each segment of the final vector in the
- * previous workspace
- */
- INSERT_TASK_dlange(
- &options,
- ChamMaxNorm, tempkm, 1, A->nb,
- VECNORMS_STEP2(m, 0), tempkm,
- VECNORMS_STEP1(m, 0));
- }
+ CHAMELEON_Desc_Create( &Wcol, NULL, ChamRealDouble, 1, A->nb, A->nb,
+ workmt, worknt * A->nb, 0, 0, workmt, worknt * A->nb, A->p, A->q );
- /* Initialize RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(A->myrank / A->q, A->myrank % A->q), 1);
+ CHAMELEON_Desc_Create( &Welt, NULL, ChamRealDouble, 1, 1, 1,
+ A->p, worknt, 0, 0, A->p, worknt, A->p, A->q );
- /* compute max norm between tiles in the column locally on each rank */
- if (A->myrank % A->q == 0) {
- for(m = (A->myrank / A->q); m < A->mt; m+=A->p) {
- INSERT_TASK_dlange_max(
- &options,
- VECNORMS_STEP1(m, 0),
- RESULT(A->myrank / A->q, A->myrank % A->q));
- }
- }
+ break;
- /* compute max norm between tiles in the column between ranks */
- if (A->myrank % A->q == 0) {
- for(m = 0; m < A->p; m++) {
- INSERT_TASK_dlange_max(
- &options,
- RESULT(m,0),
- RESULT(0,0));
- }
- }
+ /*
+ * ChamInfNorm
+ */
+ case ChamInfNorm:
+ RUNTIME_options_ws_alloc( &options, A->mb, 0 );
- /* Scatter norm over processus */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
- }
- }
- CHAMELEON_Desc_Flush( VECNORMS_STEP2, sequence );
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- CHAMELEON_Desc_Destroy( &(VECNORMS_STEP2) );
+ CHAMELEON_Desc_Create( &Wcol, NULL, ChamRealDouble, A->mb, 1, A->mb,
+ workmt * A->mb, worknt, 0, 0, workmt * A->mb, worknt, A->p, A->q );
+
+ CHAMELEON_Desc_Create( &Welt, NULL, ChamRealDouble, 1, 1, 1,
+ workmt, A->q, 0, 0, workmt, A->q, A->p, A->q );
break;
/*
* ChamFrobeniusNorm
*/
case ChamFrobeniusNorm:
+ RUNTIME_options_ws_alloc( &options, 1, 0 );
- workm = chameleon_max( A->mt, A->p );
- workn = chameleon_max( A->nt, A->q );
-
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, 1, 2, 2,
- workm, 2*workn, 0, 0, workm, 2*workn, A->p, A->q);
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 2, 2,
- 1, 2, 0, 0, 1, 2, 1, 1);
-
- /* Compute local norm to each tile */
- for(m = 0; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
- for(n = 0; n < A->nt; n++) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- VECNORMS_STEP1(m,n), 1);
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- }
- }
-
- /* Initialize arrays */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- RESULT(0,0), 1);
-
- /* Compute accumulation of scl and ssq */
- for(m = 0; m < A->mt; m++) {
- for(n = 0; n < A->nt; n++) {
- INSERT_TASK_dplssq(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
- }
- }
- /* Compute scl * sqrt(ssq) */
- INSERT_TASK_dplssq2(
- &options,
- RESULT(0,0));
-
- /* Copy max norm in tiles to dispatch on every nodes */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
- }
- }
-
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
+ alpha = 1.;
+ CHAMELEON_Desc_Create( &Welt, NULL, ChamRealDouble, 2, 1, 2,
+ workmt*2, workn, 0, 0, workmt*2, workn, A->p, A->q );
break;
/*
@@ -350,68 +438,81 @@ void chameleon_pzlange( cham_normtype_t norm, CHAM_desc_t *A, double *result,
*/
case ChamMaxNorm:
default:
- /* Init workspace handle for the call to zlange but unused */
RUNTIME_options_ws_alloc( &options, 1, 0 );
- workm = chameleon_max( A->mt, A->p );
- workn = chameleon_max( A->nt, A->q );
-
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, 1, 1, 1,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- 1, 1, 0, 0, 1, 1, 1, 1);
+ CHAMELEON_Desc_Create( &Welt, NULL, ChamRealDouble, 1, 1, 1,
+ workmt, workn, 0, 0, workmt, workn, A->p, A->q );
+ }
- /* Compute local maximum to each tile */
- for(m = 0; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
- for(n = 0; n < A->nt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_zlange(
+ /* Initialize workspaces */
+ if ( (norm == ChamInfNorm) ||
+ (norm == ChamOneNorm) )
+ {
+ /* Initialize Wcol tile */
+ for(m = 0; m < Wcol->mt; m++) {
+ for(n = 0; n < Wcol->nt; n++) {
+ INSERT_TASK_dlaset(
&options,
- ChamMaxNorm, tempkm, tempkn, A->nb,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
+ ChamUpperLower, Wcol->mb, Wcol->nb,
+ alpha, beta,
+ Wcol(m,n), Wcol->mb );
}
}
+ }
+ for(m = 0; m < Welt->mt; m++) {
+ for(n = 0; n < Welt->nt; n++) {
+ INSERT_TASK_dlaset(
+ &options,
+ ChamUpperLower, Welt->mb, Welt->nb,
+ alpha, beta,
+ Welt(m,n), Welt->mb );
+ }
+ }
- /* Initialize RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(0,0), 1);
+ switch ( norm ) {
+ case ChamOneNorm:
+ chameleon_pzlange_one( uplo, diag, A, Wcol, Welt, &options );
+ CHAMELEON_Desc_Flush( Wcol, sequence );
+ break;
- /* Compute max norm between tiles */
- for(m = 0; m < A->mt; m++) {
- for(n = 0; n < A->nt; n++) {
- INSERT_TASK_dlange_max(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
- }
- }
+ case ChamInfNorm:
+ chameleon_pzlange_inf( uplo, diag, A, Wcol, Welt, &options );
+ CHAMELEON_Desc_Flush( Wcol, sequence );
+ break;
- /* Copy max norm in tiles to dispatch on every nodes */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
+ case ChamFrobeniusNorm:
+ chameleon_pzlange_frb( uplo, diag, A, Welt, &options );
+ break;
+
+ case ChamMaxNorm:
+ default:
+ chameleon_pzlange_max( uplo, diag, A, Welt, &options );
+ }
+
+ /**
+ * Broadcast the result
+ */
+ for(m = 0; m < A->p; m++) {
+ for(n = 0; n < A->q; n++) {
+ if ( (m != 0) && (n != 0) ) {
INSERT_TASK_dlacpy(
&options,
ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
+ Welt(0,0), 1, Welt(m, n), 1);
}
}
-
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
}
- *result = *(double *)VECNORMS_STEP1->get_blkaddr(VECNORMS_STEP1, A->myrank / A->q, A->myrank % A->q );
+ CHAMELEON_Desc_Flush( Welt, sequence );
+ RUNTIME_sequence_wait(chamctxt, sequence);
+
+ *result = *(double *)Welt->get_blkaddr(Welt, A->myrank / A->q, A->myrank % A->q );
+
+ if ( Wcol != NULL ) {
+ CHAMELEON_Desc_Destroy( &Wcol );
+ }
+ CHAMELEON_Desc_Destroy( &Welt );
- CHAMELEON_Desc_Destroy( &(VECNORMS_STEP1) );
- CHAMELEON_Desc_Destroy( &(RESULT) );
RUNTIME_options_ws_free(&options);
RUNTIME_options_finalize(&options, chamctxt);
}
diff --git a/compute/pzlanhe.c b/compute/pzlanhe.c
deleted file mode 100644
index e7c181296209393fd2b940d92a44238cc72c663f..0000000000000000000000000000000000000000
--- a/compute/pzlanhe.c
+++ /dev/null
@@ -1,460 +0,0 @@
-/**
- *
- * @file pzlanhe.c
- *
- * @copyright 2009-2014 The University of Tennessee and The University of
- * Tennessee Research Foundation. All rights reserved.
- * @copyright 2012-2014 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
- * Univ. Bordeaux. All rights reserved.
- *
- ***
- *
- * @brief Chameleon zlanhe parallel algorithm
- *
- * @version 1.0.0
- * @comment This file has been automatically generated
- * from Plasma 2.6.0 for CHAMELEON 1.0.0
- * @author Emmanuel Agullo
- * @author Mathieu Faverge
- * @date 2010-11-15
- * @precisions normal z -> c
- *
- */
-//ALLOC_WS : A->mb
-//#include <stdlib.h>
-//#include <math.h>
-//WS_ADD : A->mb
-#include "control/common.h"
-
-#define A(m, n) A, m, n
-#define VECNORMS_STEP1(m, n) VECNORMS_STEP1, m, n
-#define VECNORMS_STEP2(m, n) VECNORMS_STEP2, m, n
-#define RESULT(m, n) RESULT, m, n
-/**
- *
- */
-/**
- *
- */
-void chameleon_pzlanhe(cham_normtype_t norm, cham_uplo_t uplo, CHAM_desc_t *A, double *result,
- RUNTIME_sequence_t *sequence, RUNTIME_request_t *request)
-{
- CHAM_desc_t *VECNORMS_STEP1 = NULL;
- CHAM_desc_t *VECNORMS_STEP2 = NULL;
- CHAM_desc_t *RESULT = NULL;
- CHAM_context_t *chamctxt;
- RUNTIME_option_t options;
-
- int workm, workn;
- int tempkm, tempkn;
- int ldam;
- int m, n;
- /* int part_p, part_q; */
-
- /* part_p = A->myrank / A->q; */
- /* part_q = A->myrank % A->q; */
-
- chamctxt = chameleon_context_self();
- if (sequence->status != CHAMELEON_SUCCESS)
- return;
- RUNTIME_options_init(&options, chamctxt, sequence, request);
-
- *result = 0.0;
- switch ( norm ) {
- /*
- * ChamOneNorm / ChamInfNorm
- */
- case ChamOneNorm:
- case ChamInfNorm:
- /* Init workspace handle for the call to zlanhe */
- RUNTIME_options_ws_alloc( &options, A->mb, 0 );
-
- workm = A->m;
- workn = chameleon_max( A->nt, A->q );
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, A->mb, 1, A->mb,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
-
- CHAMELEON_Desc_Create(&(VECNORMS_STEP2), NULL, ChamRealDouble, A->mb, 1, A->mb,
- workm, 1, 0, 0, workm, 1, A->p, A->q);
-
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- 1, 1, 0, 0, 1, 1, 1, 1);
-
- /* Zeroes my intermediate vectors */
- for(m = 0; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- for(n = 0; n < workn; n++) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- VECNORMS_STEP1(m, n), 1);
- }
- }
-
- for(m = (A->myrank / A->q); m < A->mt; m+=A->p) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
-
- /* compute sums of absolute values on diagonal tile m */
- INSERT_TASK_dzasum(
- &options,
- ChamRowwise, uplo, tempkm, tempkm,
- A(m, m), ldam, VECNORMS_STEP1(m, m));
-
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n < m; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- /* compute sums of absolute values on rows of tile m */
- INSERT_TASK_dzasum(
- &options,
- ChamRowwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(m, n));
- /* same operation on the symmetric part */
- INSERT_TASK_dzasum(
- &options,
- ChamColumnwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(n, m));
- }
- }
- /*
- * ChamUpper
- */
- else {
-// for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
-// n < A->mt; n+=A->q) {
- for(n = m+1; n < A->mt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- /* compute sums of absolute values on rows of tile m */
- INSERT_TASK_dzasum(
- &options,
- ChamRowwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(m, n));
- /* same operation on the symmetric part */
- INSERT_TASK_dzasum(
- &options,
- ChamColumnwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(n, m));
- }
- }
- }
-
- /* compute vector sum between tiles in rows */
- for(m = (A->myrank / A->q); m < A->mt; m+=A->p) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- VECNORMS_STEP2(m, 0), 1);
- for(n = 0; n < A->nt; n++) {
- INSERT_TASK_dgeadd(
- &options,
- ChamNoTrans, tempkm, 1, A->mb,
- 1.0, VECNORMS_STEP1(m, n), tempkm,
- 1.0, VECNORMS_STEP2(m, 0), tempkm);
- }
- /*
- * Compute max norm of each segment of the final vector in the
- * previous workspace
- */
- INSERT_TASK_dlange(
- &options,
- ChamMaxNorm, tempkm, 1, A->nb,
- VECNORMS_STEP2(m, 0), tempkm,
- VECNORMS_STEP1(m, 0));
- }
-
- /* Initialize RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(0,0), 1);
-
- /* compute max norm between tiles in the column */
- if (A->myrank % A->q == 0) {
- for(m = 0; m < A->mt; m++) {
- INSERT_TASK_dlange_max(
- &options,
- VECNORMS_STEP1(m, 0),
- RESULT(0,0));
- }
- }
-
- /* Scatter norm over processus */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
- }
- }
- CHAMELEON_Desc_Flush( VECNORMS_STEP2, sequence );
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- *result = *(double *)VECNORMS_STEP1->get_blkaddr(VECNORMS_STEP1, A->myrank / A->q, A->myrank % A->q );
- CHAMELEON_Desc_Destroy( &(VECNORMS_STEP1) );
- CHAMELEON_Desc_Destroy( &(VECNORMS_STEP2) );
- CHAMELEON_Desc_Destroy( &(RESULT) );
- break;
- /*
- * ChamFrobeniusNorm
- */
- case ChamFrobeniusNorm:
- workm = chameleon_max( A->mt, A->p );
- workn = chameleon_max( A->nt, A->q );
-
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, 1, 2, 2,
- workm, 2*workn, 0, 0, workm, 2*workn, A->p, A->q);
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 2, 2,
- 1, 2, 0, 0, 1, 2, 1, 1);
-
- /* Compute local norm to each tile */
- for(m = (A->myrank / A->q); m < A->mt; m+=A->p) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
-
- /* Zeroes my intermediate vectors */
- for(n = A->myrank % A->q; n < workn; n+=A->q) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- VECNORMS_STEP1(m,n), 1);
- }
-
- /* compute norm on diagonal tile m */
- INSERT_TASK_zhessq(
- &options,
- uplo, tempkm,
- A(m, m), ldam,
- VECNORMS_STEP1(m, m));
-
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n < m; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- /* compute norm on the lower part */
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- /* same operation on the symmetric part */
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- }
- }
- /*
- * ChamUpper
- */
- else {
-// for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
-// n < A->mt; n+=A->q) {
- for(n = m+1; n < A->mt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- /* compute norm on the lower part */
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- /* same operation on the symmetric part */
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- }
- }
- }
-
- /* Initialize arrays */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- RESULT(0,0), 1);
-
- /* Compute accumulation of scl and ssq */
- for(m = (A->myrank / A->q); m < A->mt; m+=A->p) {
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n <= m; n++) {
- INSERT_TASK_dplssq(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
- }
- }
- /*
- * ChamUpper
- */
- else {
-// for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
-// n < A->mt; n+=A->q) {
- for(n = m; n < A->mt; n++) {
- INSERT_TASK_dplssq(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
- }
- }
- }
-
- /* Compute scl * sqrt(ssq) */
- INSERT_TASK_dplssq2(
- &options,
- RESULT(0,0));
-
- /* Copy max norm in tiles to dispatch on every nodes */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
- }
- }
-
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- *result = *(double *)VECNORMS_STEP1->get_blkaddr(VECNORMS_STEP1, A->myrank / A->q, A->myrank % A->q );
- CHAMELEON_Desc_Destroy( &(VECNORMS_STEP1) );
- CHAMELEON_Desc_Destroy( &(RESULT) );
- break;
-
- /*
- * ChamMaxNorm
- */
- case ChamMaxNorm:
- default:
- /* Init workspace handle for the call to zlange but unused */
- RUNTIME_options_ws_alloc( &options, 1, 0 );
-
- workm = chameleon_max( A->mt, A->p );
- workn = chameleon_max( A->nt, A->q );
-
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, 1, 1, 1,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- 1, 1, 0, 0, 1, 1, 1, 1);
-
- /* Compute local maximum to each tile */
- for(m = 0; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
-
- INSERT_TASK_zlanhe(
- &options,
- ChamMaxNorm, uplo, tempkm, A->nb,
- A(m, m), ldam,
- VECNORMS_STEP1(m, m));
-
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n < m; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_zlange(
- &options,
- ChamMaxNorm, tempkm, tempkn, A->nb,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- }
- }
- /*
- * ChamUpper
- */
- else {
- //for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
- // n < A->mt; n+=A->q) {
- for(n = m+1; n < A->mt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_zlange(
- &options,
- ChamMaxNorm, tempkm, tempkn, A->nb,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- }
- }
- }
-
- /* Initialize RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(0,0), 1);
-
- /* Compute max norm between tiles */
- for(m = 0; m < A->mt; m++) {
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n <= m; n++) {
- INSERT_TASK_dlange_max(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
- }
- }
- /*
- * ChamUpper
- */
- else {
- //for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
- // n < A->mt; n+=A->q) {
- for(n = m; n < A->mt; n++) {
- INSERT_TASK_dlange_max(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
- }
- }
- }
-
- /* Copy max norm in tiles to dispatch on every nodes */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
- }
- }
-
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- *result = *(double *)VECNORMS_STEP1->get_blkaddr(VECNORMS_STEP1, A->myrank / A->q, A->myrank % A->q );
- CHAMELEON_Desc_Destroy( &(VECNORMS_STEP1) );
- CHAMELEON_Desc_Destroy( &(RESULT) );
- }
- RUNTIME_options_ws_free(&options);
- RUNTIME_options_finalize(&options, chamctxt);
-}
diff --git a/compute/pzlansy.c b/compute/pzlansy.c
index 96411e26c4946c157bec2218e334b5f0bb373ad9..1df96420f3ecf199123a4b54643145b5a314a408 100644
--- a/compute/pzlansy.c
+++ b/compute/pzlansy.c
@@ -21,445 +21,415 @@
*
*/
//ALLOC_WS : A->mb
+//WS_ADD : A->mb
+
#include <stdlib.h>
#include <math.h>
-//WS_ADD : A->mb
#include "control/common.h"
-#define A(m, n) A, m, n
-#define VECNORMS_STEP1(m, n) VECNORMS_STEP1, m, n
-#define VECNORMS_STEP2(m, n) VECNORMS_STEP2, m, n
-#define RESULT(m, n) RESULT, m, n
-/**
- *
- */
-/**
- *
- */
-void chameleon_pzlansy(cham_normtype_t norm, cham_uplo_t uplo, CHAM_desc_t *A, double *result,
- RUNTIME_sequence_t *sequence, RUNTIME_request_t *request)
-{
- CHAM_desc_t *VECNORMS_STEP1 = NULL;
- CHAM_desc_t *VECNORMS_STEP2 = NULL;
- CHAM_desc_t *RESULT = NULL;
- CHAM_context_t *chamctxt;
- RUNTIME_option_t options;
+#define A(m, n) A, (m), (n)
+#define Wcol(m, n) Wcol, (m), (n)
+#define Welt(m, n) Welt, (m), (n)
- int workm, workn;
- int tempkm, tempkn;
- int ldam;
+static inline void
+chameleon_pzlansy_inf( cham_uplo_t uplo, CHAM_desc_t *A,
+ CHAM_desc_t *Wcol, CHAM_desc_t *Welt,
+ RUNTIME_option_t *options)
+{
int m, n;
- /* int part_p, part_q; */
-
- /* part_p = A->myrank / A->q; */
- /* part_q = A->myrank % A->q; */
+ int MT = A->mt;
+ int NT = A->nt;
+ int M = A->m;
+ int N = A->n;
+ int P = Welt->p;
+ int Q = Welt->q;
+
+ /**
+ * Step 1:
+ * For j in [1,Q], Wcol(m, j) = reduce( A(m, j+k*Q) )
+ */
+ for(m = 0; m < MT; m++) {
+ int nmin = ( uplo == ChamUpper ) ? m : 0;
+ int nmax = ( uplo == ChamLower ) ? chameleon_min(m+1, NT) : NT;
- chamctxt = chameleon_context_self();
- if (sequence->status != CHAMELEON_SUCCESS)
- return;
- RUNTIME_options_init(&options, chamctxt, sequence, request);
+ int tempmm = ( m == (MT-1) ) ? M - m * A->mb : A->mb;
+ int ldam = BLKLDD( A, m );
- *result = 0.0;
- switch ( norm ) {
- /*
- * ChamOneNorm / ChamInfNorm
- */
- case ChamOneNorm:
- case ChamInfNorm:
- /* Init workspace handle for the call to zlange */
- RUNTIME_options_ws_alloc( &options, A->mb, 0 );
+ for(n = nmin; n < nmax; n++) {
+ int tempnn = ( n == (NT-1) ) ? N - n * A->nb : A->nb;
- workm = A->m;
- workn = chameleon_max( A->nt, A->q );
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, A->mb, 1, A->mb,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
+ if ( n == m ) {
+ INSERT_TASK_dzasum(
+ options,
+ ChamRowwise, uplo, tempmm, tempnn,
+ A(m, n), ldam, Wcol(m, n) );
+ }
+ else {
+ INSERT_TASK_dzasum(
+ options,
+ ChamRowwise, ChamUpperLower, tempmm, tempnn,
+ A(m, n), ldam, Wcol(m, n) );
+
+ INSERT_TASK_dzasum(
+ options,
+ ChamColumnwise, ChamUpperLower, tempmm, tempnn,
+ A(m, n), ldam, Wcol(n, m) );
+ }
+ }
+ }
- CHAMELEON_Desc_Create(&(VECNORMS_STEP2), NULL, ChamRealDouble, A->mb, 1, A->mb,
- workm, 1, 0, 0, workm, 1, A->p, A->q);
+ for(m = 0; m < MT; m++) {
+ int tempmm = ( m == (MT-1) ) ? M - m * A->mb : A->mb;
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- 1, 1, 0, 0, 1, 1, 1, 1);
+ for(n = Q; n < NT; n++) {
+ INSERT_TASK_dgeadd(
+ options,
+ ChamNoTrans, tempmm, 1, A->nb,
+ 1.0, Wcol(m, n ), tempmm,
+ 1.0, Wcol(m, n%Q), tempmm );
+ }
- /* Zeroes my intermediate vectors */
- for(m = 0; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- for(n = 0; n < workn; n++) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- VECNORMS_STEP1(m, n), 1);
- }
+ /**
+ * Step 2:
+ * For each j, W(m, j) = reduce( Wcol(m, 0..Q-1) )
+ */
+ for(n = 1; n < Q; n++) {
+ INSERT_TASK_dgeadd(
+ options,
+ ChamNoTrans, tempmm, 1, A->mb,
+ 1.0, Wcol(m, n), tempmm,
+ 1.0, Wcol(m, 0), tempmm );
}
- for(m = 0; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
+ INSERT_TASK_dlange(
+ options,
+ ChamMaxNorm, tempmm, 1, A->nb,
+ Wcol(m, 0), 1, Welt(m, 0));
+ }
- /* compute sums of absolute values on diagonal tile m */
- INSERT_TASK_dzasum(
- &options,
- ChamRowwise, uplo, tempkm, tempkm,
- A(m, m), ldam, VECNORMS_STEP1(m, m));
-
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n < m; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- /* compute sums of absolute values on rows of tile m */
- INSERT_TASK_dzasum(
- &options,
- ChamRowwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(m, n));
- /* same operation on the symmetric part */
- INSERT_TASK_dzasum(
- &options,
- ChamColumnwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(n, m));
+ /**
+ * Step 3:
+ * For m in 0..P-1, Welt(m, n) = max( Wcol(m..mt[P], n ) )
+ */
+ for(m = P; m < MT; m++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, 0), Welt(m%P, 0) );
+ }
+
+ /**
+ * Step 4:
+ * For each i, Welt(i, n) = max( Welt(0..P-1, n) )
+ */
+ for(m = 1; m < P; m++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, 0), Welt(0, 0) );
+ }
+}
+
+static inline void
+chameleon_pzlansy_max( cham_trans_t trans, cham_uplo_t uplo, CHAM_desc_t *A,
+ CHAM_desc_t *Welt, RUNTIME_option_t *options)
+{
+ int m, n;
+ int MT = A->mt;
+ int NT = A->nt;
+ int M = A->m;
+ int N = A->n;
+ int P = Welt->p;
+ int Q = Welt->q;
+
+ /**
+ * Step 1:
+ * For j in [1,Q], Welt(m, j) = reduce( A(m, j+k*Q) )
+ */
+ for(m = 0; m < MT; m++) {
+ int nmin = (uplo == ChamUpper ) ? m : 0;
+ int nmax = (uplo == ChamLower ) ? chameleon_min(m+1, NT) : NT;
+
+ int tempmm = ( m == (MT-1) ) ? M - m * A->mb : A->mb;
+ int ldam = BLKLDD( A, m );
+
+ for(n = nmin; n < nmax; n++) {
+ int tempnn = ( n == (NT-1) ) ? N - n * A->nb : A->nb;
+
+ if ( n == m ) {
+ if ( trans == ChamConjTrans) {
+ INSERT_TASK_zlanhe(
+ options,
+ ChamMaxNorm, uplo, tempmm, A->nb,
+ A(m, n), ldam, Welt(m, n));
+ }
+ else {
+ INSERT_TASK_zlansy(
+ options,
+ ChamMaxNorm, uplo, tempmm, A->nb,
+ A(m, n), ldam, Welt(m, n));
}
}
- /*
- * ChamUpper
- */
else {
- // for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
- // n < A->mt; n+=A->q) {
- for(n = m+1; n < A->mt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- /* compute sums of absolute values on rows of tile m */
- INSERT_TASK_dzasum(
- &options,
- ChamRowwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(m, n));
- /* same operation on the symmetric part */
- INSERT_TASK_dzasum(
- &options,
- ChamColumnwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, VECNORMS_STEP1(n, m));
- }
+ INSERT_TASK_zlange(
+ options,
+ ChamMaxNorm, tempmm, tempnn, A->nb,
+ A(m, n), ldam, Welt(m, n));
}
- }
- /* compute vector sum between tiles in rows */
- for(m = 0; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- VECNORMS_STEP2(m, 0), 1);
- for(n = 0; n < A->nt; n++) {
- INSERT_TASK_dgeadd(
- &options,
- ChamNoTrans, tempkm, 1, A->mb,
- 1.0, VECNORMS_STEP1(m, n), tempkm,
- 1.0, VECNORMS_STEP2(m, 0), tempkm);
- }
- /*
- * Compute max norm of each segment of the final vector in the
- * previous workspace
- */
- INSERT_TASK_dlange(
- &options,
- ChamMaxNorm, tempkm, 1, A->nb,
- VECNORMS_STEP2(m, 0), tempkm,
- VECNORMS_STEP1(m, 0));
- }
-
- /* Initialize RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(0,0), 1);
-
- /* compute max norm between tiles in the column */
- if (A->myrank % A->q == 0) {
- for(m = 0; m < A->mt; m++) {
+ if ( n >= Q ) {
INSERT_TASK_dlange_max(
- &options,
- VECNORMS_STEP1(m, 0),
- RESULT(0,0));
+ options,
+ Welt(m, n), Welt(m, n%Q) );
}
}
- /* Scatter norm over processus */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
- }
+ /**
+ * Step 2:
+ * For each j, W(m, j) = reduce( Welt(m, 0..Q-1) )
+ */
+ for(n = 1; n < Q; n++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, n), Welt(m, 0) );
}
- CHAMELEON_Desc_Flush( VECNORMS_STEP2, sequence );
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- CHAMELEON_Desc_Destroy( &(VECNORMS_STEP2) );
- break;
- /*
- * ChamFrobeniusNorm
+ }
+
+ /**
+ * Step 3:
+ * For m in 0..P-1, Welt(m, n) = max( Welt(m..mt[P], n ) )
*/
- case ChamFrobeniusNorm:
- workm = chameleon_max( A->mt, A->p );
- workn = chameleon_max( A->nt, A->q );
+ for(m = P; m < MT; m++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, 0), Welt(m%P, 0) );
+ }
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, 1, 2, 2,
- workm, 2*workn, 0, 0, workm, 2*workn, A->p, A->q);
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 2, 2,
- 1, 2, 0, 0, 1, 2, 1, 1);
+ /**
+ * Step 4:
+ * For each i, Welt(i, n) = max( Welt(0..P-1, n) )
+ */
+ for(m = 1; m < P; m++) {
+ INSERT_TASK_dlange_max(
+ options,
+ Welt(m, 0), Welt(0, 0) );
+ }
+}
- /* Compute local norm to each tile */
- for(m = (A->myrank / A->q); m < A->mt; m+=A->p) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
+static inline void
+chameleon_pzlansy_frb( cham_trans_t trans, cham_uplo_t uplo,
+ CHAM_desc_t *A, CHAM_desc_t *Welt,
+ RUNTIME_option_t *options)
+{
+ int m, n;
+ int MT = A->mt;
+ int NT = A->nt;
+ int M = A->m;
+ int N = A->n;
+ int P = Welt->p;
+ int Q = Welt->q;
+
+ /**
+ * Step 1:
+ * For j in [1,Q], Welt(m, j) = reduce( A(m, j+k*Q) )
+ */
+ for(m = 0; m < MT; m++) {
+ int nmin = (uplo == ChamUpper ) ? m : 0;
+ int nmax = (uplo == ChamLower ) ? chameleon_min(m+1, NT) : NT;
- /* Zeroes my intermediate vector */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- VECNORMS_STEP1(m, m), 1);
- /* compute norm on diagonal tile m */
- INSERT_TASK_zsyssq(
- &options,
- uplo, tempkm,
- A(m, m), ldam,
- VECNORMS_STEP1(m, m));
-
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n < m; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- /* Zeroes my intermediate vector */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- VECNORMS_STEP1(m, n), 1);
- /* compute norm on the lower part */
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- /* same operation on the symmetric part */
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- }
- }
- /*
- * ChamUpper
- */
- else {
- // for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
- // n < A->mt; n+=A->q) {
- for(n = m+1; n < A->mt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- /* Zeroes my intermediate vector */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- VECNORMS_STEP1(m, n), 1);
- /* compute norm on the lower part */
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- /* same operation on the symmetric part */
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- }
- }
- }
+ int tempmm = ( m == (MT-1) ) ? M - m * A->mb : A->mb;
+ int ldam = BLKLDD( A, m );
- /* Zeroes my intermediate vector */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- RESULT(0,0), 1);
-
- /* Compute accumulation of scl and ssq */
- for(m = (A->myrank / A->q); m < A->mt; m+=A->p) {
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n <= m; n++) {
- INSERT_TASK_dplssq(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
+ for(n = nmin; n < nmax; n++) {
+ int tempnn = ( n == (NT-1) ) ? N - n * A->nb : A->nb;
+
+ if ( n == m ) {
+ if ( trans == ChamConjTrans) {
+ INSERT_TASK_zhessq(
+ options, uplo, tempmm,
+ A(m, n), ldam, Welt(m, n) );
+ }
+ else {
+ INSERT_TASK_zsyssq(
+ options, uplo, tempmm,
+ A(m, n), ldam, Welt(m, n) );
}
}
- /*
- * ChamUpper
- */
else {
- // for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
- // n < A->mt; n+=A->q) {
- for(n = m; n < A->mt; n++) {
- INSERT_TASK_dplssq(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
- }
+ INSERT_TASK_zgessq(
+ options, tempmm, tempnn,
+ A(m, n), ldam, Welt(m, n) );
+ INSERT_TASK_zgessq(
+ options, tempmm, tempnn,
+ A(m, n), ldam, Welt(n, m) );
}
}
+ }
- /* Compute scl * sqrt(ssq) */
- INSERT_TASK_dplssq2(
- &options,
- RESULT(0,0));
+ for(m = 0; m < MT; m++) {
+ for(n = Q; n < NT; n++) {
+ INSERT_TASK_dplssq(
+ options, Welt(m, n), Welt(m, n%Q) );
+ }
- /* Copy max norm in tiles to dispatch on every nodes */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
- }
+ /**
+ * Step 2:
+ * For each j, W(m, j) = reduce( Welt(m, 0..Q-1) )
+ */
+ for(n = 1; n < Q; n++) {
+ INSERT_TASK_dplssq(
+ options, Welt(m, n), Welt(m, 0) );
}
+ }
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- break;
+ /**
+ * Step 3:
+ * For m in 0..P-1, Welt(m, n) = max( Welt(m..mt[P], n ) )
+ */
+ for(m = P; m < MT; m++) {
+ INSERT_TASK_dplssq(
+ options, Welt(m, 0), Welt(m%P, 0) );
+ }
- /*
- * ChamMaxNorm
+ /**
+ * Step 4:
+ * For each i, Welt(i, n) = max( Welt(0..P-1, n) )
*/
- case ChamMaxNorm:
- default:
- /* Init workspace handle for the call to zlange but unused */
+ for(m = 1; m < P; m++) {
+ INSERT_TASK_dplssq(
+ options, Welt(m, 0), Welt(0, 0) );
+ }
+
+ INSERT_TASK_dplssq2(
+ options, Welt(0, 0) );
+}
+
+/**
+ *
+ */
+void chameleon_pzlansy_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_trans_t trans,
+ CHAM_desc_t *A, double *result,
+ RUNTIME_sequence_t *sequence, RUNTIME_request_t *request )
+{
+ CHAM_context_t *chamctxt;
+ RUNTIME_option_t options;
+ CHAM_desc_t *Wcol = NULL;
+ CHAM_desc_t *Welt = NULL;
+ double alpha = 0.0;
+ double beta = 0.0;
+
+ int workn, workmt, worknt;
+ int m, n;
+
+ chamctxt = chameleon_context_self();
+ if (sequence->status != CHAMELEON_SUCCESS)
+ return;
+ RUNTIME_options_init(&options, chamctxt, sequence, request);
+
+ *result = 0.0;
+
+ workmt = chameleon_max( A->mt, A->p );
+ worknt = chameleon_max( A->nt, A->q );
+ workn = chameleon_max( A->n, A->q );
+
+ switch ( norm ) {
+ case ChamOneNorm:
+ case ChamInfNorm:
RUNTIME_options_ws_alloc( &options, 1, 0 );
- workm = chameleon_max( A->mt, A->p );
- workn = chameleon_max( A->nt, A->q );
+ CHAMELEON_Desc_Create( &Wcol, NULL, ChamRealDouble, 1, A->nb, A->nb,
+ workmt, worknt * A->nb, 0, 0, workmt, worknt * A->nb, A->p, A->q );
- CHAMELEON_Desc_Create(&(VECNORMS_STEP1), NULL, ChamRealDouble, 1, 1, 1,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- 1, 1, 0, 0, 1, 1, 1, 1);
+ CHAMELEON_Desc_Create( &Welt, NULL, ChamRealDouble, 1, 1, 1,
+ A->p, worknt, 0, 0, A->p, worknt, A->p, A->q );
- /* Compute local maximum to each tile */
- for(m = 0; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
+ break;
- INSERT_TASK_zlansy(
- &options,
- ChamMaxNorm, uplo, tempkm, A->nb,
- A(m, m), ldam,
- VECNORMS_STEP1(m, m));
-
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n < m; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_zlange(
- &options,
- ChamMaxNorm, tempkm, tempkn, A->nb,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- }
- }
- /*
- * ChamUpper
- */
- else {
- //for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
- // n < A->mt; n+=A->q) {
- for(n = m+1; n < A->mt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_zlange(
- &options,
- ChamMaxNorm, tempkm, tempkn, A->nb,
- A(m, n), ldam,
- VECNORMS_STEP1(m, n));
- }
- }
- }
+ /*
+ * ChamFrobeniusNorm
+ */
+ case ChamFrobeniusNorm:
+ RUNTIME_options_ws_alloc( &options, 1, 0 );
- /* Zeroes RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(0,0), 1);
-
- /* Compute max norm between tiles */
- for(m = 0; m < A->mt; m++) {
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- //for(n = A->myrank % A->q; n < m; n+=A->q) {
- for(n = 0; n <= m; n++) {
- INSERT_TASK_dlange_max(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
- }
- }
- /*
- * ChamUpper
- */
- else {
- //for(n = ( part_q > part_p ? (m/part_p)*part_p + part_q : (m/part_p)*part_p + part_q + A->q );
- // n < A->mt; n+=A->q) {
- for(n = m; n < A->mt; n++) {
- INSERT_TASK_dlange_max(
- &options,
- VECNORMS_STEP1(m, n),
- RESULT(0,0));
- }
+ alpha = 1.;
+ CHAMELEON_Desc_Create( &Welt, NULL, ChamRealDouble, 2, 1, 2,
+ workmt*2, workn, 0, 0, workmt*2, workn, A->p, A->q );
+ break;
+
+ /*
+ * ChamMaxNorm
+ */
+ case ChamMaxNorm:
+ default:
+ RUNTIME_options_ws_alloc( &options, 1, 0 );
+
+ CHAMELEON_Desc_Create( &Welt, NULL, ChamRealDouble, 1, 1, 1,
+ workmt, workn, 0, 0, workmt, workn, A->p, A->q );
+ }
+
+ /* Initialize workspaces */
+ if ( (norm == ChamInfNorm) ||
+ (norm == ChamOneNorm) )
+ {
+ /* Initialize Wcol tile */
+ for(m = 0; m < Wcol->mt; m++) {
+ for(n = 0; n < Wcol->nt; n++) {
+ INSERT_TASK_dlaset(
+ &options,
+ ChamUpperLower, Wcol->mb, Wcol->nb,
+ alpha, beta,
+ Wcol(m,n), Wcol->mb );
}
}
+ }
+ for(m = 0; m < Welt->mt; m++) {
+ for(n = 0; n < Welt->nt; n++) {
+ INSERT_TASK_dlaset(
+ &options,
+ ChamUpperLower, Welt->mb, Welt->nb,
+ alpha, beta,
+ Welt(m,n), Welt->mb );
+ }
+ }
+
+ switch ( norm ) {
+ case ChamOneNorm:
+ case ChamInfNorm:
+ chameleon_pzlansy_inf( uplo, A, Wcol, Welt, &options );
+ CHAMELEON_Desc_Flush( Wcol, sequence );
+ break;
+
+ case ChamFrobeniusNorm:
+ chameleon_pzlansy_frb( trans, uplo, A, Welt, &options );
+ break;
+
+ case ChamMaxNorm:
+ default:
+ chameleon_pzlansy_max( trans, uplo, A, Welt, &options );
+ }
- /* Copy max norm in tiles to dispatch on every nodes */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
+ /**
+ * Broadcast the result
+ */
+ for(m = 0; m < A->p; m++) {
+ for(n = 0; n < A->q; n++) {
+ if ( (m != 0) && (n != 0) ) {
INSERT_TASK_dlacpy(
&options,
ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- VECNORMS_STEP1(m, n), 1 );
+ Welt(0,0), 1, Welt(m, n), 1);
}
}
-
- CHAMELEON_Desc_Flush( VECNORMS_STEP1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
}
- *result = *(double *)VECNORMS_STEP1->get_blkaddr(VECNORMS_STEP1, A->myrank / A->q, A->myrank % A->q );
+ CHAMELEON_Desc_Flush( Welt, sequence );
+ RUNTIME_sequence_wait(chamctxt, sequence);
+
+ *result = *(double *)Welt->get_blkaddr(Welt, A->myrank / A->q, A->myrank % A->q );
+
+ if ( Wcol != NULL ) {
+ CHAMELEON_Desc_Destroy( &Wcol );
+ }
+ CHAMELEON_Desc_Destroy( &Welt );
- CHAMELEON_Desc_Destroy( &(VECNORMS_STEP1) );
- CHAMELEON_Desc_Destroy( &(RESULT) );
RUNTIME_options_ws_free(&options);
RUNTIME_options_finalize(&options, chamctxt);
}
diff --git a/compute/pzlantr.c b/compute/pzlantr.c
deleted file mode 100644
index acf27965580c15df4fd4535249a8112e47e07fdb..0000000000000000000000000000000000000000
--- a/compute/pzlantr.c
+++ /dev/null
@@ -1,674 +0,0 @@
-/**
- *
- * @file pzlantr.c
- *
- * @copyright 2009-2014 The University of Tennessee and The University of
- * Tennessee Research Foundation. All rights reserved.
- * @copyright 2012-2014 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
- * Univ. Bordeaux. All rights reserved.
- *
- ***
- *
- * @brief Chameleon zlantr parallel algorithm
- *
- * @version 1.0.0
- * @comment This file has been automatically generated
- * from Plasma 2.6.0 for CHAMELEON 1.0.0
- * @author Mathieu Faverge
- * @date 2010-11-15
- * @precisions normal z -> c d s
- *
- */
-#include <stdlib.h>
-#include <math.h>
-#include "control/common.h"
-
-#define A(m, n) A, m, n
-#define W1(m, n) W1, m, n
-#define W2(m, n) W2, m, n
-#define RESULT(m, n) RESULT, m, n
-/**
- *
- */
-void chameleon_pzlantr(cham_normtype_t norm, cham_uplo_t uplo, cham_diag_t diag,
- CHAM_desc_t *A, double *result,
- RUNTIME_sequence_t *sequence, RUNTIME_request_t *request)
-{
- CHAM_desc_t *W1 = NULL;
- CHAM_desc_t *W2 = NULL;
- CHAM_desc_t *RESULT = NULL;
- CHAM_context_t *chamctxt;
- RUNTIME_option_t options;
-
- int workm, workn;
- int tempkm, tempkn;
- int ldam, ldan;
- int m, n, minMNT;
- /* int part_p, part_q; */
-
- minMNT = chameleon_min( A->mt, A->nt );
-
- /* part_p = A->myrank / A->q; */
- /* part_q = A->myrank % A->q; */
-
- chamctxt = chameleon_context_self();
- if (sequence->status != CHAMELEON_SUCCESS)
- return;
- RUNTIME_options_init(&options, chamctxt, sequence, request);
-
- *result = 0.0;
- switch ( norm ) {
- /*
- * ChamOneNorm
- */
- case ChamOneNorm:
- /* Init workspace handle for the call to zlange but unused */
- RUNTIME_options_ws_alloc( &options, 1, 0 );
-
- workm = chameleon_max( A->mt, A->p );
- workn = ( uplo == ChamLower ) ? chameleon_min( A->m, A->n ) : A->n;
-
- CHAMELEON_Desc_Create(&(W1), NULL, ChamRealDouble, 1, A->nb, A->nb,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
-
- CHAMELEON_Desc_Create(&(W2), NULL, ChamRealDouble, 1, A->nb, A->nb,
- 1, workn, 0, 0, 1, workn, A->p, A->q);
-
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- 1, 1, 0, 0, 1, 1, 1, 1);
-
- /*
- * ChamUpper
- */
- if (uplo == ChamUpper) {
- /* Zeroes intermediate vector */
- for(n = 0; n < W2->nt; n++) {
- tempkn = n == W2->nt-1 ? W2->n-n*W2->nb : W2->nb;
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, tempkn,
- 0., 0.,
- W2(0, n), 1);
- }
- for(m = 0; m < minMNT; m++) {
- /* Zeroes intermediate vectors */
- for(n = m; n < W1->nt; n++) {
- tempkn = n == W1->nt-1 ? W1->n-n*W1->nb : W1->nb;
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, tempkn,
- 0., 0.,
- W1(m, n), 1);
- }
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- tempkn = m == A->nt-1 ? A->n-m*A->nb : A->nb;
- ldam = BLKLDD(A, m);
- /* compute sums of absolute values on columns of diag tile */
- INSERT_TASK_ztrasm(
- &options,
- ChamColumnwise, uplo, diag, tempkm, tempkn,
- A(m, m), ldam,
- W1(m, m));
-
- /* compute sums of absolute values on columns of each tile */
- for(n = m+1; n < A->nt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_dzasum(
- &options,
- ChamColumnwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, W1(m, n));
- }
-
- /* Compute vector sums between tiles in columns */
- for(n = m; n < W1->nt; n++) {
- tempkn = n == W1->nt-1 ? W1->n-n*W1->nb : W1->nb;
- INSERT_TASK_dgeadd(
- &options,
- ChamNoTrans, 1, tempkn, W1->mb,
- 1.0, W1(m, n), 1,
- 1.0, W2(0, n), 1);
- }
- }
- }
- /*
- * ChamLower
- */
- else {
- for(n = 0; n < minMNT; n++) {
- tempkm = n == A->mt-1 ? A->m-n*A->mb : A->mb;
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- ldan = BLKLDD(A, n);
- /* Zeroes intermediate vectors */
- for(m = n; m < A->mt; m++) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, tempkn,
- 0., 0.,
- W1(m, n), 1);
- }
- /* Zeroes the second intermediate vector */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, tempkn,
- 0., 0.,
- W2(0, n), 1);
-
- /* compute sums of absolute values on columns of diag tile */
- INSERT_TASK_ztrasm(
- &options,
- ChamColumnwise, uplo, diag, tempkm, tempkn,
- A(n, n), ldan,
- W1(n, n));
-
- /* compute sums of absolute values on columns of each tile */
- for(m = n+1; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
- INSERT_TASK_dzasum(
- &options,
- ChamColumnwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, W1(m, n));
- }
-
- /* Compute vector sums between tiles in columns */
- for(m = n; m < A->mt; m++) {
- INSERT_TASK_dgeadd(
- &options,
- ChamNoTrans, 1, tempkn, A->mb,
- 1.0, W1(m, n), 1,
- 1.0, W2(0, n), 1);
- }
- }
- }
-
- /*
- * Compute max norm of each segment of the final vector in the
- * previous workspace
- */
- for(n = 0; n < A->nt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_dlange(
- &options,
- ChamMaxNorm, 1, tempkn, A->nb,
- W2(0, n), 1,
- W1(0, n));
- }
-
- /* Initialize RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(0,0), 1);
-
- /* Compute max norm between tiles in the row */
- if (A->myrank < A->q) {
- for(n = 0; n < A->nt; n++) {
- INSERT_TASK_dlange_max(
- &options,
- W1(0, n),
- RESULT(0,0));
- }
- }
-
- /* Scatter norm over processus */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- W1(m, n), 1 );
- }
- }
- CHAMELEON_Desc_Flush( W2, sequence );
- CHAMELEON_Desc_Flush( W1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- *result = *(double *)W1->get_blkaddr(W1, A->myrank / A->q, A->myrank % A->q );
- CHAMELEON_Desc_Destroy( &(W1) );
- CHAMELEON_Desc_Destroy( &(W2) );
- CHAMELEON_Desc_Destroy( &(RESULT) );
- break;
- /*
- * ChamInfNorm
- */
- case ChamInfNorm:
- /* Init workspace handle for the call to zlange */
- RUNTIME_options_ws_alloc( &options, A->mb, 0 );
-
- workm = ( uplo == ChamUpper ) ? chameleon_min( A->m, A->n ) : A->m;
- workn = chameleon_max( A->nt, A->q );
- CHAMELEON_Desc_Create(&(W1), NULL, ChamRealDouble, A->mb, 1, A->mb,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
-
- CHAMELEON_Desc_Create(&(W2), NULL, ChamRealDouble, A->mb, 1, A->mb,
- workm, 1, 0, 0, workm, 1, A->p, A->q);
-
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- 1, 1, 0, 0, 1, 1, 1, 1);
-
- /*
- * ChamUpper
- */
- if (uplo == ChamUpper) {
- for(m = 0; m < minMNT; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- tempkn = m == A->nt-1 ? A->n-m*A->nb : A->nb;
- ldam = BLKLDD(A, m);
- /* Zeroes intermediate vectors */
- for(n = m; n < A->nt; n++) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- W1(m, n), 1);
- }
- /* Zeroes intermediate vector */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- W2(m, 0), 1);
-
- /* compute sums of absolute values on rows of diag tile */
- INSERT_TASK_ztrasm(
- &options,
- ChamRowwise, uplo, diag, tempkm, tempkn,
- A(m, m), ldam,
- W1(m, m));
-
- /* compute sums of absolute values on rows of each tile */
- for(n = m+1; n < A->nt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_dzasum(
- &options,
- ChamRowwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, W1(m, n));
- }
-
- /* Compute vector sums between tiles in rows */
- for(n = m; n < A->nt; n++) {
- INSERT_TASK_dgeadd(
- &options,
- ChamNoTrans, tempkm, 1, A->mb,
- 1.0, W1(m, n), tempkm,
- 1.0, W2(m, 0), tempkm);
- }
-
- }
- }
- /*
- * ChamLower
- */
- else {
- /* Zeroes intermediate vector */
- for(m = 0; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- W2(m, 0), 1);
- }
- for(n = 0; n < minMNT; n++) {
- /* Zeroes intermediate vectors */
- for(m = n; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, tempkm, 1,
- 0., 0.,
- W1(m, n), tempkm);
- }
- tempkm = n == A->mt-1 ? A->m-n*A->mb : A->mb;
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- ldan = BLKLDD(A, n);
- /* compute sums of absolute values on rows of diag tile */
- INSERT_TASK_ztrasm(
- &options,
- ChamRowwise, uplo, diag, tempkm, tempkn,
- A(n, n), ldan,
- W1(n, n));
-
- /* compute sums of absolute values on rows of each tile */
- for(m = n+1; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
- INSERT_TASK_dzasum(
- &options,
- ChamRowwise, ChamUpperLower, tempkm, tempkn,
- A(m, n), ldam, W1(m, n));
- }
-
- /* Compute vector sums between tiles in rows */
- for(m = n; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- INSERT_TASK_dgeadd(
- &options,
- ChamNoTrans, tempkm, 1, A->mb,
- 1.0, W1(m, n), tempkm,
- 1.0, W2(m, 0), tempkm);
- }
- }
- }
-
- /*
- * Compute max norm of each segment of the final vector in the
- * previous workspace
- */
- for(m = 0; m < W1->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- INSERT_TASK_dlange(
- &options,
- ChamMaxNorm, tempkm, 1, A->nb,
- W2(m, 0), 1,
- W1(m, 0));
- }
-
- /* Initialize RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(0,0), 1);
-
- /* compute max norm between tiles in the column */
- if (A->myrank % A->q == 0) {
- for(m = 0; m < W1->mt; m++) {
- INSERT_TASK_dlange_max(
- &options,
- W1(m, 0),
- RESULT(0,0));
- }
- }
-
- /* Scatter norm over processus */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- W1(m, n), 1 );
- }
- }
- CHAMELEON_Desc_Flush( W2, sequence );
- CHAMELEON_Desc_Flush( W1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- *result = *(double *)W1->get_blkaddr(W1, A->myrank / A->q, A->myrank % A->q );
- CHAMELEON_Desc_Destroy( &(W1) );
- CHAMELEON_Desc_Destroy( &(W2) );
- CHAMELEON_Desc_Destroy( &(RESULT) );
- break;
- /*
- * ChamFrobeniusNorm
- */
- case ChamFrobeniusNorm:
- workm = chameleon_max( A->mt, A->p );
- workn = chameleon_max( A->nt, A->q );
-
- CHAMELEON_Desc_Create(&(W1), NULL, ChamRealDouble, 1, 2, 2,
- workm, 2*workn, 0, 0, workm, 2*workn, A->p, A->q);
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 2, 2,
- 1, 2, 0, 0, 1, 2, 1, 1);
-
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- /* Compute local maximum to each tile */
- for(n = 0; n < minMNT; n++) {
- tempkm = n == A->mt-1 ? A->m-n*A->mb : A->mb;
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- ldan = BLKLDD(A, n);
- /* Zeroes my intermediate vectors */
- for(m = n; m < A->mt; m++) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- W1(m,n), 1);
- }
- /* Compute local norm of the diagonal tile */
- INSERT_TASK_ztrssq(
- &options,
- uplo, diag, tempkm, tempkn,
- A(n, n), ldan,
- W1(n, n));
- /* Compute local norm to each tile */
- for(m = n+1; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- W1(m, n));
- }
- }
- }
- /*
- * ChamUpper
- */
- else {
- /* Compute local maximum to each tile */
- for(m = 0; m < minMNT; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- tempkn = m == A->nt-1 ? A->n-m*A->nb : A->nb;
- ldam = BLKLDD(A, m);
- /* Zeroes my intermediate vectors */
- for(n = m; n < A->nt; n++) {
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- W1(m,n), 1);
- }
- /* Compute local norm of the diagonal tile */
- INSERT_TASK_ztrssq(
- &options,
- uplo, diag, tempkm, tempkn,
- A(m, m), ldam,
- W1(m, m));
- /* Compute local norm to each tile */
- for(n = m+1; n < A->nt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_zgessq(
- &options,
- tempkm, tempkn,
- A(m, n), ldam,
- W1(m, n));
- }
- }
- }
-
- /* Initialize arrays */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 2,
- 1., 0.,
- RESULT(0,0), 1);
-
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- /* Compute accumulation of scl and ssq */
- for(n = 0; n < minMNT; n++) {
- for(m = n; m < A->mt; m++) {
- INSERT_TASK_dplssq(
- &options,
- W1(m, n),
- RESULT(0,0));
- }
- }
- }
- /*
- * ChamUpper
- */
- else {
- /* Compute accumulation of scl and ssq */
- for(m = 0; m < minMNT; m++) {
- for(n = m; n < A->nt; n++) {
- INSERT_TASK_dplssq(
- &options,
- W1(m, n),
- RESULT(0,0));
- }
- }
- }
-
- /* Compute scl * sqrt(ssq) */
- INSERT_TASK_dplssq2(
- &options,
- RESULT(0,0));
-
- /* Copy max norm in tiles to dispatch on every nodes */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- W1(m, n), 1 );
- }
- }
-
- CHAMELEON_Desc_Flush( W1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- *result = *(double *)W1->get_blkaddr(W1, A->myrank / A->q, A->myrank % A->q );
- CHAMELEON_Desc_Destroy( &(W1) );
- CHAMELEON_Desc_Destroy( &(RESULT) );
- break;
-
- /*
- * ChamMaxNorm
- */
- case ChamMaxNorm:
- default:
- /* Init workspace handle for the call to zlange but unused */
- RUNTIME_options_ws_alloc( &options, 1, 0 );
-
- workm = chameleon_max( A->mt, A->p );
- workn = chameleon_max( A->nt, A->q );
-
- CHAMELEON_Desc_Create(&(W1), NULL, ChamRealDouble, 1, 1, 1,
- workm, workn, 0, 0, workm, workn, A->p, A->q);
- CHAMELEON_Desc_Create(&(RESULT), NULL, ChamRealDouble, 1, 1, 1,
- 1, 1, 0, 0, 1, 1, 1, 1);
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- /* Compute local maximum to each tile */
- for(n = 0; n < minMNT; n++) {
- tempkm = n == A->mt-1 ? A->m-n*A->mb : A->mb;
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- ldan = BLKLDD(A, n);
-
- INSERT_TASK_zlantr(
- &options,
- ChamMaxNorm, uplo, diag,
- tempkm, tempkn, A->nb,
- A(n, n), ldan,
- W1(n, n));
-
- for(m = n+1; m < A->mt; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- ldam = BLKLDD(A, m);
- INSERT_TASK_zlange(
- &options,
- ChamMaxNorm, tempkm, tempkn, A->nb,
- A(m, n), ldam,
- W1(m, n));
- }
- }
- }
- /*
- * ChamUpper
- */
- else {
- /* Compute local maximum to each tile */
- for(m = 0; m < minMNT; m++) {
- tempkm = m == A->mt-1 ? A->m-m*A->mb : A->mb;
- tempkn = m == A->nt-1 ? A->n-m*A->nb : A->nb;
- ldam = BLKLDD(A, m);
-
- INSERT_TASK_zlantr(
- &options,
- ChamMaxNorm, uplo, diag,
- tempkm, tempkn, A->nb,
- A(m, m), ldam,
- W1(m, m));
-
- for(n = m+1; n < A->nt; n++) {
- tempkn = n == A->nt-1 ? A->n-n*A->nb : A->nb;
- INSERT_TASK_zlange(
- &options,
- ChamMaxNorm, tempkm, tempkn, A->nb,
- A(m, n), ldam,
- W1(m, n));
- }
- }
- }
-
- /* Initialize RESULT array */
- INSERT_TASK_dlaset(
- &options,
- ChamUpperLower, 1, 1,
- 0., 0.,
- RESULT(0,0), 1);
-
- /*
- * ChamLower
- */
- if (uplo == ChamLower) {
- /* Compute max norm between tiles */
- for(n = 0; n < minMNT; n++) {
- for(m = n; m < A->mt; m++) {
- INSERT_TASK_dlange_max(
- &options,
- W1(m, n),
- RESULT(0,0));
- }
- }
- }
- /*
- * ChamUpper
- */
- else {
- /* Compute max norm between tiles */
- for(m = 0; m < minMNT; m++) {
- for(n = m; n < A->nt; n++) {
- INSERT_TASK_dlange_max(
- &options,
- W1(m, n),
- RESULT(0,0));
- }
- }
- }
-
- /* Copy max norm in tiles to dispatch on every nodes */
- for(m = 0; m < A->p; m++) {
- for(n = 0; n < A->q; n++) {
- INSERT_TASK_dlacpy(
- &options,
- ChamUpperLower, 1, 1, 1,
- RESULT(0,0), 1,
- W1(m, n), 1 );
- }
- }
-
- CHAMELEON_Desc_Flush( W1, sequence );
- CHAMELEON_Desc_Flush( RESULT, sequence );
- RUNTIME_sequence_wait(chamctxt, sequence);
- *result = *(double *)W1->get_blkaddr(W1, A->myrank / A->q, A->myrank % A->q );
- CHAMELEON_Desc_Destroy( &(W1) );
- CHAMELEON_Desc_Destroy( &(RESULT) );
- }
- RUNTIME_options_ws_free(&options);
- RUNTIME_options_finalize(&options, chamctxt);
-}
diff --git a/compute/zlange.c b/compute/zlange.c
index 82968715d9ef59ab1e7e75b38443153ca1788b70..e807b403f7d8f544657d3524c16337107ef7062c 100644
--- a/compute/zlange.c
+++ b/compute/zlange.c
@@ -287,7 +287,7 @@ int CHAMELEON_zlange_Tile_Async( cham_normtype_t norm, CHAM_desc_t *A, double *v
return CHAMELEON_SUCCESS;
}
- chameleon_pzlange( norm, A, value, sequence, request );
+ chameleon_pzlange_generic( norm, ChamUpperLower, ChamNonUnit, A, value, sequence, request );
return CHAMELEON_SUCCESS;
}
diff --git a/compute/zlanhe.c b/compute/zlanhe.c
index 5297f8eabb7cec89bf327809d580c7298f3a0a86..4eff40c2907f80ab7278519a532c5394c317f68e 100644
--- a/compute/zlanhe.c
+++ b/compute/zlanhe.c
@@ -295,7 +295,7 @@ int CHAMELEON_zlanhe_Tile_Async( cham_normtype_t norm, cham_uplo_t uplo, CHAM_de
return CHAMELEON_SUCCESS;
}
- chameleon_pzlanhe( norm, uplo, A, value, sequence, request );
+ chameleon_pzlansy_generic( norm, uplo, ChamConjTrans, A, value, sequence, request );
return CHAMELEON_SUCCESS;
}
diff --git a/compute/zlansy.c b/compute/zlansy.c
index 995f21c80d0e1d4a13316f7a768a757cd80c72ec..53b4e088d417f345a089c35b3904cbc86e50dd65 100644
--- a/compute/zlansy.c
+++ b/compute/zlansy.c
@@ -295,7 +295,7 @@ int CHAMELEON_zlansy_Tile_Async( cham_normtype_t norm, cham_uplo_t uplo, CHAM_de
return CHAMELEON_SUCCESS;
}
- chameleon_pzlansy( norm, uplo, A, value, sequence, request );
+ chameleon_pzlansy_generic( norm, uplo, ChamTrans, A, value, sequence, request );
return CHAMELEON_SUCCESS;
}
diff --git a/compute/zlantr.c b/compute/zlantr.c
index fb78aa2e8d1954ff79db4c326b6a246b2788d9b6..82a372e48dab76e541522f7b5607e487bb79f718 100644
--- a/compute/zlantr.c
+++ b/compute/zlantr.c
@@ -323,7 +323,7 @@ int CHAMELEON_zlantr_Tile_Async( cham_normtype_t norm, cham_uplo_t uplo, cham_di
return CHAMELEON_SUCCESS;
}
- chameleon_pzlantr( norm, uplo, diag, A, value, sequence, request );
+ chameleon_pzlange_generic( norm, uplo, diag, A, value, sequence, request );
return CHAMELEON_SUCCESS;
}
diff --git a/control/compute_z.h b/control/compute_z.h
index 0f5c0c7f77bead5bf041f9157111d8b25edc56b5..e99d83e57f2deb08c4a9099b8a959cb4571ea7c9 100644
--- a/control/compute_z.h
+++ b/control/compute_z.h
@@ -82,10 +82,11 @@ void chameleon_pzher2k(cham_uplo_t uplo, cham_trans_t trans, CHAMELEON_Complex64
void chameleon_pzhetrd_he2hb(cham_uplo_t uplo, CHAM_desc_t *A, CHAM_desc_t *T, CHAM_desc_t *E, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);
void chameleon_pzlacpy(cham_uplo_t uplo, CHAM_desc_t *A, CHAM_desc_t *B, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);
void chameleon_pzlag2c(CHAM_desc_t *A, CHAM_desc_t *SB, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);
-void chameleon_pzlange(cham_normtype_t norm, CHAM_desc_t *A, double *result, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);
-void chameleon_pzlanhe(cham_normtype_t norm, cham_uplo_t uplo, CHAM_desc_t *A, double *result, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);
-void chameleon_pzlansy(cham_normtype_t norm, cham_uplo_t uplo, CHAM_desc_t *A, double *result, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);
-void chameleon_pzlantr(cham_normtype_t norm, cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A, double *result, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);
+void chameleon_pzlange_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_diag_t diag, CHAM_desc_t *A,
+ double *result, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request );
+void chameleon_pzlansy_generic( cham_normtype_t norm, cham_uplo_t uplo, cham_trans_t trans,
+ CHAM_desc_t *A, double *result,
+ RUNTIME_sequence_t *sequence, RUNTIME_request_t *request );
void chameleon_pzlascal(cham_uplo_t uplo, CHAMELEON_Complex64_t alpha, CHAM_desc_t *A, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);
void chameleon_pzlaset( cham_uplo_t uplo, CHAMELEON_Complex64_t alpha, CHAMELEON_Complex64_t beta, CHAM_desc_t *A, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);
void chameleon_pzlaset2(cham_uplo_t uplo, CHAMELEON_Complex64_t alpha, CHAM_desc_t *A, RUNTIME_sequence_t *sequence, RUNTIME_request_t *request);