diff --git a/compute/map.c b/compute/map.c
index 16f10b5ec47f87dacfaaa9a00dc931865640c1ff..78840d6922d6cbaf6f2b6a7a5454b1cea1d7fa1d 100644
--- a/compute/map.c
+++ b/compute/map.c
@@ -16,6 +16,33 @@
*/
#include "control/common.h"
+struct map_args_s {
+ cham_unary_operator_t function;
+ void *args;
+};
+
+static inline int
+map_cpu( void *op_args,
+ cham_uplo_t uplo, int m, int n, int ndata,
+ const CHAM_desc_t *descA, CHAM_tile_t *tileA, ... )
+{
+ struct map_args_s *options = (struct map_args_s *)op_args;
+
+ if ( ndata > 1 ) {
+ fprintf( stderr, "map_cpu: supports only one piece of data and %d have been given\n", ndata );
+ }
+ options->function( descA, uplo, m, n, tileA, options->args );
+
+ return 0;
+}
+
+static cham_map_operator_t map_op = {
+ .name = "map",
+ .cpufunc = map_cpu,
+ .cudafunc = NULL,
+ .hipfunc = NULL,
+};
+
/**
********************************************************************************
*
@@ -78,8 +105,17 @@ int CHAMELEON_map_Tile( cham_access_t access,
}
chameleon_sequence_create( chamctxt, &sequence );
- CHAMELEON_map_Tile_Async( access, uplo, A, op_fct, op_args, sequence, &request );
-
+ {
+ cham_map_data_t data = {
+ .access = access,
+ .desc = A
+ };
+ struct map_args_s map_args = {
+ .function = op_fct,
+ .args = op_args,
+ };
+ chameleon_pmap( uplo, 1, &data, &map_op, &map_args, sequence, &request );
+ }
CHAMELEON_Desc_Flush( A, sequence );
chameleon_sequence_wait( chamctxt, sequence );
@@ -164,7 +200,20 @@ int CHAMELEON_map_Tile_Async( cham_access_t access,
return CHAMELEON_SUCCESS;
}
- chameleon_pmap( access, uplo, A, op_fct, op_args, sequence, request );
+ {
+ cham_map_data_t data = {
+ .access = access,
+ .desc = A
+ };
+ struct map_args_s map_args = {
+ .function = op_fct,
+ .args = op_args,
+ };
+ chameleon_pmap( uplo, 1, &data, &map_op, &map_args, sequence, request );
+ /* Need to wait to make sure no one access map_args after this function returned. */
+ CHAMELEON_Desc_Flush( A, sequence );
+ chameleon_sequence_wait( chamctxt, sequence );
+ }
return CHAMELEON_SUCCESS;
}
diff --git a/compute/pzlatms.c b/compute/pzlatms.c
index 29cc20fe587cdaaa7731b0931e913187297c8357..8e6ba0f5fc1b85a612ff835b351332d3fe831c44 100644
--- a/compute/pzlatms.c
+++ b/compute/pzlatms.c
@@ -13,7 +13,7 @@
* @version 1.3.0
* @author Mathieu Faverge
* @author Lionel Eyraud-Dubois
- * @date 2023-07-05
+ * @date 2024-03-11
* @precisions normal z -> s d c
*
*/
@@ -33,9 +33,9 @@
static RUNTIME_id_t zlatms_runtime_id = RUNTIME_SCHED_STARPU;
static inline int
-zlaset_diag( const CHAM_desc_t *descA,
- cham_uplo_t uplo, int m, int n,
- CHAM_tile_t *tileA, void *op_args )
+zlaset_diag_cpu( void *op_args,
+ cham_uplo_t uplo, int m, int n, int ndata,
+ const CHAM_desc_t *descA, CHAM_tile_t *tileA, ... )
{
CHAMELEON_Complex64_t *A;
const double *D = (const double *)op_args;
@@ -45,6 +45,10 @@ zlaset_diag( const CHAM_desc_t *descA,
int minmn = chameleon_min( tempmm, tempnn );
int lda, i;
+ if ( ndata > 1 ) {
+ fprintf( stderr, "zlaset_diag_cpu: supports only one piece of data and %d have been given\n", ndata );
+ }
+
if ( zlatms_runtime_id == RUNTIME_SCHED_PARSEC ) {
A = (CHAMELEON_Complex64_t*)tileA;
lda = descA->get_blkldd( descA, m );
@@ -68,6 +72,13 @@ zlaset_diag( const CHAM_desc_t *descA,
return 0;
}
+static cham_map_operator_t zlaset_diag_map = {
+ .name = "zlaset_diag",
+ .cpufunc = zlaset_diag_cpu,
+ .cudafunc = NULL,
+ .hipfunc = NULL,
+};
+
/**
* Parallel scale of a matrix A
*/
@@ -163,11 +174,16 @@ void chameleon_pzlatms( cham_dist_t idist, unsigned long long int seed, cham_sym
#endif
/* Copy D to the diagonal of A */
- for (n = 0; n < kt; n++) {
- INSERT_TASK_map(
- &options,
- ChamRW, ChamUpperLower, A(n, n),
- zlaset_diag, D );
+ {
+ cham_map_data_t data = {
+ .access = ChamRW,
+ .desc = A,
+ };
+ for (n = 0; n < kt; n++) {
+ INSERT_TASK_map(
+ &options, ChamUpperLower, n, n,
+ 1, &data, &zlaset_diag_map, D );
+ }
}
/**
diff --git a/compute/zprint.c b/compute/zprint.c
index 53243ff9b41912c49bbc44be76cbc6aa2ef69e50..cc2be40769abf806dd9012e92b705fb297ab9ca7 100644
--- a/compute/zprint.c
+++ b/compute/zprint.c
@@ -12,7 +12,7 @@
* @version 1.3.0
* @author Mathieu Faverge
* @author Matthieu Kuhn
- * @date 2023-07-05
+ * @date 2024-03-11
* @precisions normal z -> s d c
*
*/
@@ -33,9 +33,9 @@ struct zprint_args_s {
};
static inline int
-zprint( const CHAM_desc_t *descA,
- cham_uplo_t uplo, int m, int n,
- CHAM_tile_t *tileA, void *op_args )
+zprint_cpu( void *op_args,
+ cham_uplo_t uplo, int m, int n, int ndata,
+ const CHAM_desc_t *descA, CHAM_tile_t *tileA, ... )
{
CHAMELEON_Complex64_t *A;
struct zprint_args_s *options = (struct zprint_args_s *)op_args;
@@ -44,6 +44,10 @@ zprint( const CHAM_desc_t *descA,
int tempnn = n == descA->nt-1 ? descA->n-n*descA->nb : descA->nb;
int lda;
+ if ( ndata > 1 ) {
+ fprintf( stderr, "zprint_cpu: supports only one piece of data and %d have been given\n", ndata );
+ }
+
if ( zprint_runtime_id == RUNTIME_SCHED_PARSEC ) {
A = (CHAMELEON_Complex64_t*)tileA;
lda = descA->get_blkldd( descA, m );
@@ -61,6 +65,13 @@ zprint( const CHAM_desc_t *descA,
return 0;
}
+static cham_map_operator_t zprint_map = {
+ .name = "zprint",
+ .cpufunc = zprint_cpu,
+ .cudafunc = NULL,
+ .hipfunc = NULL,
+};
+
/**
********************************************************************************
*
@@ -99,12 +110,12 @@ int CHAMELEON_zprint( FILE *file, const char *header,
cham_uplo_t uplo, int M, int N,
CHAMELEON_Complex64_t *A, int LDA )
{
- int NB;
- int status;
- CHAM_context_t *chamctxt;
- RUNTIME_sequence_t *sequence = NULL;
- RUNTIME_request_t request = RUNTIME_REQUEST_INITIALIZER;
- CHAM_desc_t descAl, descAt;
+ int NB, status;
+ CHAM_context_t *chamctxt;
+ RUNTIME_sequence_t *sequence = NULL;
+ RUNTIME_request_t request = RUNTIME_REQUEST_INITIALIZER;
+ CHAM_desc_t descAl, descAt;
+ cham_map_data_t data;
struct zprint_args_s options = {
.file = file,
.header = header,
@@ -152,7 +163,10 @@ int CHAMELEON_zprint( FILE *file, const char *header,
/* Call the tile interface */
zprint_runtime_id = chamctxt->scheduler;
- chameleon_pmap( ChamR, uplo, &descAt, zprint, &options, sequence, &request );
+
+ data.access = ChamR;
+ data.desc = &descAt;
+ chameleon_pmap( uplo, 1, &data, &zprint_map, &options, sequence, &request );
/* Submit the matrix conversion back */
chameleon_ztile2lap( chamctxt, &descAl, &descAt,
@@ -199,9 +213,10 @@ int CHAMELEON_zprint( FILE *file, const char *header,
int CHAMELEON_zprint_Tile( FILE *file, const char *header,
cham_uplo_t uplo, CHAM_desc_t *A )
{
- CHAM_context_t *chamctxt;
- RUNTIME_sequence_t *sequence = NULL;
- RUNTIME_request_t request = RUNTIME_REQUEST_INITIALIZER;
+ CHAM_context_t *chamctxt;
+ RUNTIME_sequence_t *sequence = NULL;
+ RUNTIME_request_t request = RUNTIME_REQUEST_INITIALIZER;
+ cham_map_data_t data;
struct zprint_args_s options = {
.file = file,
.header = header,
@@ -216,7 +231,11 @@ int CHAMELEON_zprint_Tile( FILE *file, const char *header,
chameleon_sequence_create( chamctxt, &sequence );
zprint_runtime_id = chamctxt->scheduler;
- chameleon_pmap( ChamR, uplo, A, zprint, &options, sequence, &request );
+
+ data.access = ChamR;
+ data.desc = A;
+
+ chameleon_pmap( uplo, 1, &data, &zprint_map, &options, sequence, &request );
CHAMELEON_Desc_Flush( A, sequence );
chameleon_sequence_wait( chamctxt, sequence );