From 8c9825d5cdf91145d317446254bf34098e32ab63 Mon Sep 17 00:00:00 2001
From: Mathieu Faverge <mathieu.faverge@inria.fr>
Date: Thu, 17 Oct 2024 11:54:05 +0200
Subject: [PATCH] context: Add an optlacpy context option
---
control/context.c | 3 ++-
include/chameleon/struct_context.h | 17 +++++++++--------
runtime/starpu/control/runtime_options.c | 2 +-
3 files changed, 12 insertions(+), 10 deletions(-)
diff --git a/control/context.c b/control/context.c
index cfbe13a71..5bcd6d598 100644
--- a/control/context.c
+++ b/control/context.c
@@ -21,7 +21,7 @@
* @author Matthieu Kuhn
* @author Loris Lucido
* @author Terry Cojean
- * @date 2023-09-11
+ * @date 2024-10-17
*
***
*
@@ -140,6 +140,7 @@ CHAM_context_t *chameleon_context_create()
chamctxt->progress_enabled = chameleon_env_on_off( "CHAMELEON_PROGRESS", CHAMELEON_FALSE );
chamctxt->generic_enabled = chameleon_env_on_off( "CHAMELEON_GENERIC", CHAMELEON_FALSE );
chamctxt->autominmax_enabled = chameleon_env_on_off( "CHAMELEON_AUTOMINMAX", CHAMELEON_TRUE );
+ chamctxt->optlacpy_enabled = chameleon_env_on_off( "CHAMELEON_OPTIMIZED_LACPY", CHAMELEON_TRUE );
chamctxt->runtime_paused = CHAMELEON_FALSE;
diff --git a/include/chameleon/struct_context.h b/include/chameleon/struct_context.h
index 471dcfc11..b66621cca 100644
--- a/include/chameleon/struct_context.h
+++ b/include/chameleon/struct_context.h
@@ -14,7 +14,7 @@
* @version 1.3.0
* @author Mathieu Faverge
* @author Florent Pruvost
- * @date 2024-03-16
+ * @date 2024-10-17
*
*/
#ifndef _struct_context_h_
@@ -43,18 +43,19 @@ typedef struct chameleon_context_s {
cham_bool_t progress_enabled;
cham_bool_t generic_enabled;
cham_bool_t autominmax_enabled;
+ cham_bool_t optlacpy_enabled; /**< Enable runtime cpy instead of lacpy kernel */
cham_bool_t runtime_paused;
- cham_householder_t householder; /**> "domino" (flat) or tree-based (reduction) Householder */
- cham_translation_t translation; /**> In place or Out of place layout conversion */
+ cham_householder_t householder; /**< "domino" (flat) or tree-based (reduction) Householder */
+ cham_translation_t translation; /**< In place or Out of place layout conversion */
int nb;
int ib;
- int rhblock; /**> block size for tree-based (reduction) Householder */
- int lookahead; /**> depth of the look ahead in algorithms */
- void *schedopt; /**> structure for runtimes */
- int mpi_outer_init; /**> MPI has been initialized outside our functions */
- MPI_Comm comm; /**> MPI communicator */
+ int rhblock; /**< block size for tree-based (reduction) Householder */
+ int lookahead; /**< depth of the look ahead in algorithms */
+ void *schedopt; /**< structure for runtimes */
+ int mpi_outer_init; /**< MPI has been initialized outside our functions */
+ MPI_Comm comm; /**< MPI communicator */
} CHAM_context_t;
END_C_DECLS
diff --git a/runtime/starpu/control/runtime_options.c b/runtime/starpu/control/runtime_options.c
index 64f3b569e..8423d9d7d 100644
--- a/runtime/starpu/control/runtime_options.c
+++ b/runtime/starpu/control/runtime_options.c
@@ -32,7 +32,7 @@ void RUNTIME_options_init( RUNTIME_option_t *options, CHAM_context_t *chamctxt,
options->priority = RUNTIME_PRIORITY_MIN;
options->workerid = (schedopt == NULL) ? -1 : schedopt->workerid;
options->forcesub = 0;
- options->withlacpy = 0;
+ options->withlacpy = !(chamctxt->optlacpy_enabled);
options->withcuda = (chamctxt->ncudas > 0);
options->ws_wsize = 0;
options->ws_hsize = 0;
--
GitLab