diff --git a/compute/zgesvd.c b/compute/zgesvd.c
index be88952272af125224d03b30bc4050d3e5f2be26..db664159170531eb47cc0b066a745fb21102b40f 100644
--- a/compute/zgesvd.c
+++ b/compute/zgesvd.c
@@ -610,6 +610,14 @@ int MORSE_zgesvd_Tile_Async( MORSE_enum jobu, MORSE_enum jobvt,
}
morse_sequence_wait( morse, sequence );
+ /* Cleanup the temporary data */
+ if ( jobu != MorseNoVec ) {
+ morse_ztile2lap_cleanup( morse, &descUl, &descUt );
+ }
+ if ( jobvt != MorseNoVec ) {
+ morse_ztile2lap_cleanup( morse, &descVTl, &descVTt );
+ }
+
/* Solve the bidiagonal SVD problem */
/* On exit, U and VT are updated with bidiagonal matrix singular vectors */
#if !defined(CHAMELEON_SIMULATION)
diff --git a/compute/zheevd.c b/compute/zheevd.c
index 68768489e6f10dac7b3e5ceb9eeb2d146b19187e..d630088170a245ff20b8fb934907cc59b7072650 100644
--- a/compute/zheevd.c
+++ b/compute/zheevd.c
@@ -335,12 +335,12 @@ int MORSE_zheevd_Tile_Async( MORSE_enum jobz, MORSE_enum uplo,
MORSE_desc_t descA;
MORSE_desc_t descT;
MORSE_desc_t D, *Dptr = NULL;
- MORSE_Complex64_t *Q2;
- int N, status;
+ MORSE_Complex64_t *Q2 = NULL;
+ int N, NB, status;
double *E;
MORSE_Complex64_t *V;
- MORSE_desc_t descQ2;
- MORSE_desc_t descV;
+ MORSE_desc_t descQ2l, descQ2t;
+ MORSE_desc_t descVl, descVt;
MORSE_desc_t *subA, *subQ, *subT;
morse = morse_context_self();
@@ -395,7 +395,8 @@ int MORSE_zheevd_Tile_Async( MORSE_enum jobz, MORSE_enum uplo,
return morse_request_fail(sequence, request, MORSE_ERR_ILLEGAL_VALUE);
}
- N = descA.m;
+ N = descA.m;
+ NB = descA.mb;
/* Allocate data structures for reduction to tridiagonal form */
E = malloc( (N - 1) * sizeof(double) );
@@ -463,10 +464,12 @@ int MORSE_zheevd_Tile_Async( MORSE_enum jobz, MORSE_enum uplo,
/* Q from MORSE_zhetrd refers to Q2 (lapack layout) */
/* V from LAPACKE_zstedc refers to V (lapack layout) */
/* The final eigenvectors are (Q1 Q2 V) or (Q1^h Q2 V) */
- /* morse_zooplap2tile( descQ2, Q2, NB, NB, N, N, 0, 0, N, N, sequence, request, */
- /* morse_desc_mat_free( &descQ2 ) ); */
- /* morse_zooplap2tile( descV, V, NB, NB, N, N, 0, 0, N, N, sequence, request, */
- /* morse_desc_mat_free(&(descQ2)); morse_desc_mat_free(&(descV)) ); */
+ morse_zlap2tile( morse, &descQ2l, &descQ2t, MorseDescInput, MorseUpperLower,
+ Q2, NB, NB, N, N, N, N, sequence, request );
+
+ morse_zlap2tile( morse, &descVl, &descVt, MorseDescInput, MorseUpperLower,
+ V, NB, NB, N, N, N, N, sequence, request );
+
if (uplo == MorseLower)
{
#if defined(CHAMELEON_COPY_DIAG)
@@ -476,9 +479,9 @@ int MORSE_zheevd_Tile_Async( MORSE_enum jobz, MORSE_enum uplo,
Dptr = &D;
}
#endif
- subA = morse_desc_submatrix(&descA, descA.mb, 0, descA.m -descA.mb, descA.n-descA.nb);
- subQ = morse_desc_submatrix(&descQ2, descQ2.mb, 0, descQ2.m-descQ2.mb, descQ2.n );
- subT = morse_desc_submatrix(&descT, descT.mb, 0, descT.m -descT.mb, descT.n-descT.nb);
+ subA = morse_desc_submatrix(&descA, descA.mb, 0, descA.m -descA.mb, descA.n-descA.nb);
+ subQ = morse_desc_submatrix(&descQ2t, descQ2t.mb, 0, descQ2t.m-descQ2t.mb, descQ2t.n );
+ subT = morse_desc_submatrix(&descT, descT.mb, 0, descT.m -descT.mb, descT.n-descT.nb);
/* Compute Q2 = Q1 * Q2 */
morse_pzunmqr( MorseLeft, MorseNoTrans,
@@ -487,7 +490,7 @@ int MORSE_zheevd_Tile_Async( MORSE_enum jobz, MORSE_enum uplo,
/* Compute the final eigenvectors A = (Q1 * Q2) * V */
morse_pzgemm( MorseNoTrans, MorseNoTrans,
- 1.0, &descQ2, &descV,
+ 1.0, &descQ2t, &descVt,
0.0, &descA,
sequence, request );
@@ -500,9 +503,9 @@ int MORSE_zheevd_Tile_Async( MORSE_enum jobz, MORSE_enum uplo,
Dptr = &D;
}
#endif
- subA = morse_desc_submatrix(&descA, 0, descA.nb, descA.m -descA.mb, descA.n -descA.nb );
- subQ = morse_desc_submatrix(&descQ2, descQ2.mb, 0, descQ2.m-descQ2.mb, descQ2.n );
- subT = morse_desc_submatrix(&descT, 0, descT.nb, descT.m -descT.mb, descT.n -descT.nb );
+ subA = morse_desc_submatrix(&descA, 0, descA.nb, descA.m -descA.mb, descA.n -descA.nb );
+ subQ = morse_desc_submatrix(&descQ2t, descQ2t.mb, 0, descQ2t.m-descQ2t.mb, descQ2t.n );
+ subT = morse_desc_submatrix(&descT, 0, descT.nb, descT.m -descT.mb, descT.n -descT.nb );
/* Compute Q2 = Q1^h * Q2 */
morse_pzunmlq( MorseLeft, MorseConjTrans,
@@ -511,19 +514,24 @@ int MORSE_zheevd_Tile_Async( MORSE_enum jobz, MORSE_enum uplo,
/* Compute the final eigenvectors A = (Q1^h * Q2) * V */
morse_pzgemm( MorseNoTrans, MorseNoTrans,
- 1.0, &descQ2, &descV,
+ 1.0, &descQ2t, &descVt,
0.0, &descA,
sequence, request );
}
+ morse_ztile2lap( morse, &descQ2l, &descQ2t,
+ MorseDescInput, MorseUpperLower, sequence, request );
+ morse_ztile2lap( morse, &descVl, &descVt,
+ MorseDescInput, MorseUpperLower, sequence, request );
+
morse_sequence_wait( morse, sequence );
+ /* Cleanup the temporary data */
+ morse_ztile2lap_cleanup( morse, &descQ2l, &descQ2t );
+ morse_ztile2lap_cleanup( morse, &descVl, &descVt );
+
free(subA); free(subQ); free(subT);
- morse_desc_mat_free( &descQ2 );
free(Q2);
-
- /* Cleanup the temporary data */
- morse_desc_mat_free( &descV );
free(V);
free(E);
if (Dptr != NULL) {
diff --git a/coreblas/compute/core_zhe2ge.c b/coreblas/compute/core_zhe2ge.c
index c4b5f785a8934f16f3623150cdc2d15871ccc2a8..df2d6225acedb0ae7332e1c7ec5d8ec0bdc6b500 100644
--- a/coreblas/compute/core_zhe2ge.c
+++ b/coreblas/compute/core_zhe2ge.c
@@ -39,9 +39,8 @@ void CORE_zhe2ge(MORSE_enum uplo, int M, int N,
MORSE_Complex64_t *Bptr, *BTptr;
int i, j;
- Aptr = A;
- Bptr = B;
- BTptr = B;
+ Aptr = A;
+ Bptr = B;
if (uplo == MorseLower){
for (j = 0; j < N; j++){
@@ -57,23 +56,10 @@ void CORE_zhe2ge(MORSE_enum uplo, int M, int N,
}
Aptr += (LDA - i + j + 1);
Bptr += (LDB - i + j + 1);
-
- /* Bptr[ j * LDB + j ] = A[ j * LDA + j ]; */
-
- /* for (i = j+1; i < M; i++) { */
- /* Bptr [ j * LDB + i ] = A[ j * LDA + i ]; */
- /* BTptr[ i * LDB + j ] = conj(A[ j * LDA + i ]); */
- /* } */
}
}
else{
for (j = 0; j < N; j++){
- /* for (i = 0; i < j; i++) { */
- /* Bptr [ j * LDB + i ] = A[ j * LDA + i ]; */
- /* BTptr[ i * LDB + j ] = conj(A[ j * LDA + i ]); */
- /* } */
- /* Bptr[ j * LDB + j ] = A[ j * LDA + j ]; */
-
BTptr = B + j;
for (i = 0; i < j; i++, Bptr++, Aptr++, BTptr += LDB) {
*Bptr = *Aptr;
diff --git a/runtime/starpu/codelets/codelet_zgemm.c b/runtime/starpu/codelets/codelet_zgemm.c
index 7fa98e57072324ed6261b601ba94e108d2919c2d..56445309d66f1591da48dceb4690116a5c3f860c 100644
--- a/runtime/starpu/codelets/codelet_zgemm.c
+++ b/runtime/starpu/codelets/codelet_zgemm.c
@@ -41,35 +41,11 @@ void MORSE_TASK_zgemm(const MORSE_option_t *options,
(void)nb;
struct starpu_codelet *codelet = &cl_zgemm;
void (*callback)(void*) = options->profiling ? cl_zgemm_callback : NULL;
- int sizeA = lda*k;
- int sizeB = ldb*n;
- int sizeC = ldc*n;
- int execution_rank = C->get_rankof( C, Cm, Cn );
- int rank_changed=0;
- (void)execution_rank;
-
- /* force execution on the rank owning the largest data (tile) */
- int threshold;
- char* env = getenv("MORSE_COMM_FACTOR_THRESHOLD");
-
- if (env != NULL)
- threshold = (unsigned)atoi(env);
- else
- threshold = 10;
- if ( sizeA > threshold*sizeC ){
- execution_rank = A->get_rankof( A, Am, An );
- rank_changed = 1;
- }else if( sizeB > threshold*sizeC ){
- execution_rank = B->get_rankof( B, Bm, Bn );
- rank_changed = 1;
- }
MORSE_BEGIN_ACCESS_DECLARATION;
MORSE_ACCESS_R(A, Am, An);
MORSE_ACCESS_R(B, Bm, Bn);
MORSE_ACCESS_RW(C, Cm, Cn);
- if (rank_changed)
- MORSE_RANK_CHANGED(execution_rank);
MORSE_END_ACCESS_DECLARATION;
starpu_insert_task(
@@ -89,9 +65,6 @@ void MORSE_TASK_zgemm(const MORSE_option_t *options,
STARPU_VALUE, &ldc, sizeof(int),
STARPU_PRIORITY, options->priority,
STARPU_CALLBACK, callback,
-#if defined(CHAMELEON_USE_MPI)
- STARPU_EXECUTE_ON_NODE, execution_rank,
-#endif
#if defined(CHAMELEON_CODELETS_HAVE_NAME)
STARPU_NAME, "zgemm",
#endif
diff --git a/runtime/starpu/codelets/codelet_ztrsm.c b/runtime/starpu/codelets/codelet_ztrsm.c
index 779597aaf1c823dd60da28ed4b7667bf827e1656..aa6a9bb2a8dc5b6bc22455396b261e87333c9ff4 100644
--- a/runtime/starpu/codelets/codelet_ztrsm.c
+++ b/runtime/starpu/codelets/codelet_ztrsm.c
@@ -40,29 +40,10 @@ void MORSE_TASK_ztrsm(const MORSE_option_t *options,
(void)nb;
struct starpu_codelet *codelet = &cl_ztrsm;
void (*callback)(void*) = options->profiling ? cl_ztrsm_callback : NULL;
- int sizeA = lda*m;
- int sizeB = ldb*n;
- int execution_rank = B->get_rankof( B, Bm, Bn );
- int rank_changed=0;
- (void)execution_rank;
-
- /* force execution on the rank owning the largest data (tile) */
- int threshold;
- char* env = getenv("MORSE_COMM_FACTOR_THRESHOLD");
- if (env != NULL)
- threshold = (unsigned)atoi(env);
- else
- threshold = 10;
- if ( sizeA > threshold*sizeB ){
- execution_rank = A->get_rankof( A, Am, An );
- rank_changed=1;
- }
MORSE_BEGIN_ACCESS_DECLARATION;
MORSE_ACCESS_R(A, Am, An);
MORSE_ACCESS_RW(B, Bm, Bn);
- if (rank_changed)
- MORSE_RANK_CHANGED(execution_rank);
MORSE_END_ACCESS_DECLARATION;
starpu_insert_task(
@@ -80,9 +61,6 @@ void MORSE_TASK_ztrsm(const MORSE_option_t *options,
STARPU_VALUE, &ldb, sizeof(int),
STARPU_PRIORITY, options->priority,
STARPU_CALLBACK, callback,
-#if defined(CHAMELEON_USE_MPI)
- STARPU_EXECUTE_ON_NODE, execution_rank,
-#endif
#if defined(CHAMELEON_CODELETS_HAVE_NAME)
STARPU_NAME, "ztrsm",
#endif
diff --git a/runtime/starpu/codelets/codelet_zunmqr.c b/runtime/starpu/codelets/codelet_zunmqr.c
index fe740bd88f0667d1edbdd6b1b64ba7e8dfd45785..7beabcf9e08ca6f306ff472d424445709088186c 100644
--- a/runtime/starpu/codelets/codelet_zunmqr.c
+++ b/runtime/starpu/codelets/codelet_zunmqr.c
@@ -120,34 +120,11 @@ void MORSE_TASK_zunmqr(const MORSE_option_t *options,
{
struct starpu_codelet *codelet = &cl_zunmqr;
void (*callback)(void*) = options->profiling ? cl_zunmqr_callback : NULL;
- int sizeA = lda*k;
- int sizeT = ldt*n;
- int sizeC = ldc*n;
- int execution_rank = C->get_rankof( C, Cm, Cn );
- int rank_changed=0;
- (void)execution_rank;
-
- /* force execution on the rank owning the largest data (tile) */
- int threshold;
- char* env = getenv("MORSE_COMM_FACTOR_THRESHOLD");
- if (env != NULL)
- threshold = (unsigned)atoi(env);
- else
- threshold = 10;
- if ( sizeA > threshold*sizeC ){
- execution_rank = A->get_rankof( A, Am, An );
- rank_changed = 1;
- }else if( sizeT > threshold*sizeC ){
- execution_rank = T->get_rankof( T, Tm, Tn );
- rank_changed = 1;
- }
MORSE_BEGIN_ACCESS_DECLARATION;
MORSE_ACCESS_R(A, Am, An);
MORSE_ACCESS_R(T, Tm, Tn);
MORSE_ACCESS_RW(C, Cm, Cn);
- if (rank_changed)
- MORSE_RANK_CHANGED(execution_rank);
MORSE_END_ACCESS_DECLARATION;
starpu_insert_task(
@@ -169,9 +146,6 @@ void MORSE_TASK_zunmqr(const MORSE_option_t *options,
STARPU_VALUE, &nb, sizeof(int),
STARPU_PRIORITY, options->priority,
STARPU_CALLBACK, callback,
-#if defined(CHAMELEON_USE_MPI)
- STARPU_EXECUTE_ON_NODE, execution_rank,
-#endif
#if defined(CHAMELEON_CODELETS_HAVE_NAME)
STARPU_NAME, "zunmqr",
#endif
diff --git a/runtime/starpu/control/runtime_descriptor.c b/runtime/starpu/control/runtime_descriptor.c
index bd804a02c784f65e89bfc29847ad188b072a2ea9..8c3990103c34208a297e2fc5e6da78a14920a745 100644
--- a/runtime/starpu/control/runtime_descriptor.c
+++ b/runtime/starpu/control/runtime_descriptor.c
@@ -145,8 +145,6 @@ void RUNTIME_desc_create( MORSE_desc_t *desc )
{
int64_t lmt = desc->lmt;
int64_t lnt = desc->lnt;
- starpu_data_handle_t *tiles;
- (void)tiles;
desc->occurences = 1;
@@ -156,7 +154,6 @@ void RUNTIME_desc_create( MORSE_desc_t *desc )
*/
desc->schedopt = (void*)calloc(lnt*lmt,sizeof(starpu_data_handle_t));
assert(desc->schedopt);
- tiles = (starpu_data_handle_t*)(desc->schedopt);
#if defined(CHAMELEON_USE_CUDA) && !defined(CHAMELEON_SIMULATION)
/*
diff --git a/timing/time_zgemm_tile.c b/timing/time_zgemm_tile.c
index 4936a1a1ff14e8a0ee469d53dccc1d65d536de06..ae504b8bca796cb07c95b82ab0c6f0f629754b74 100644
--- a/timing/time_zgemm_tile.c
+++ b/timing/time_zgemm_tile.c
@@ -48,6 +48,9 @@ RunTest(int *iparam, double *dparam, morse_time_t *t_)
#if !defined(CHAMELEON_SIMULATION)
LAPACKE_zlarnv_work(1, ISEED, 1, &alpha);
LAPACKE_zlarnv_work(1, ISEED, 1, &beta);
+#else
+ alpha = 1.5;
+ beta = -2.3;
#endif
/* Save C for check */
diff --git a/timing/time_zgesvd_tile.c b/timing/time_zgesvd_tile.c
index b7c8f2e86f4942208f8591b1961994da99d3feee..4f207aeb6a802fe79f4db848cf5b1ac44a87d562 100644
--- a/timing/time_zgesvd_tile.c
+++ b/timing/time_zgesvd_tile.c
@@ -36,7 +36,7 @@ RunTest(int *iparam, double *dparam, morse_time_t *t_)
/* Allocate Data */
PASTE_CODE_ALLOCATE_MATRIX_TILE( descA, 1, MORSE_Complex64_t, MorseComplexDouble, LDA, M, N );
PASTE_CODE_ALLOCATE_MATRIX( VT, (jobvt == MorseVec), MORSE_Complex64_t, N, N );
- PASTE_CODE_ALLOCATE_MATRIX( U, (jobu == MorseVec), MORSE_Complex64_t, M, M );
+ PASTE_CODE_ALLOCATE_MATRIX( U, (jobu == MorseVec), MORSE_Complex64_t, M, M );
PASTE_CODE_ALLOCATE_MATRIX( S, 1, double, N, 1 );
/* Initialiaze Data */
@@ -46,7 +46,7 @@ RunTest(int *iparam, double *dparam, morse_time_t *t_)
MORSE_Alloc_Workspace_zgesvd(N, N, &descT, 1, 1);
if ( jobu == MorseVec ) {
- LAPACKE_zlaset_work(LAPACK_COL_MAJOR, 'A', M, M, 0., 1., U, M);
+ LAPACKE_zlaset_work(LAPACK_COL_MAJOR, 'A', M, M, 0., 1., U, M);
}
if ( jobvt == MorseVec ) {
LAPACKE_zlaset_work(LAPACK_COL_MAJOR, 'A', N, N, 0., 1., VT, N);
@@ -56,18 +56,18 @@ RunTest(int *iparam, double *dparam, morse_time_t *t_)
INFO = MORSE_zgesvd_Tile(jobu, jobvt, descA, S, descT, U, M, VT, N);
STOP_TIMING();
- if(INFO!=0){
- printf(" ERROR OCCURED INFO %d\n",INFO);
+ if( INFO != 0 ) {
+ printf(" ERROR OCCURED INFO %d\n",INFO);
}
/* DeAllocate Workspace */
MORSE_Dealloc_Workspace(&descT);
- if (jobu == MorseVec) {
- free( U );
+ if ( U != NULL ) {
+ free( U );
}
- if (jobvt == MorseVec) {
- free( VT );
+ if ( VT != NULL) {
+ free( VT );
}
PASTE_CODE_FREE_MATRIX( descA );
free( S );
diff --git a/timing/time_zlange_tile.c b/timing/time_zlange_tile.c
index 4ffe9e77a1768e282577f9816e333a667de77326..d52d1898614918ab42f444fed5c804e641be4fbe 100644
--- a/timing/time_zlange_tile.c
+++ b/timing/time_zlange_tile.c
@@ -77,7 +77,8 @@ RunTest(int *iparam, double *dparam, morse_time_t *t_)
free( A );
}
#endif
- PASTE_CODE_FREE_MATRIX( descA );
+ PASTE_CODE_FREE_MATRIX( descA );
+ (void)normmorse;
return 0;
}
diff --git a/timing/timing.c b/timing/timing.c
index 093ade83cec701cb3880aaa4ee36e91fe161880f..38c89fbfc4cfd80c5045cb759fca8780b004dd6b 100644
--- a/timing/timing.c
+++ b/timing/timing.c
@@ -160,7 +160,6 @@ Test(int64_t n, int *iparam) {
fadds = (double)(_FADDS);
fmuls = (double)(_FMULS);
flops = 1e-9 * (fmuls * fp_per_mul + fadds * fp_per_add);
- gflops = 0.0;
if ( iparam[IPARAM_WARMUP] ) {
int status = RunTest( iparam, dparam, &(t[0]));