Objet : Developers list for StarPU
Archives de la liste
[Starpu-devel] [PATCH v2 4/6] heft.c: Remove the "bundle" parameter from compute_all_performance_predictions().
Chronologique Discussions
- From: Cyril Roelandt <cyril.roelandt@inria.fr>
- To: starpu-devel@lists.gforge.inria.fr
- Subject: [Starpu-devel] [PATCH v2 4/6] heft.c: Remove the "bundle" parameter from compute_all_performance_predictions().
- Date: Mon, 22 Oct 2012 04:56:34 +0200
- List-archive: <http://lists.gforge.inria.fr/pipermail/starpu-devel>
- List-id: "Developers list. For discussion of new features, code changes, etc." <starpu-devel.lists.gforge.inria.fr>
Motivations:
* compute_all_performance_predictions() has one less parameter.
* one less local variable in _heft_push_task().
* will be easier to merge heft and dm*.
* we may need to dereference "task" a little bit more often to get
"task->bundle", but this should not be so bad.
Signed-off-by: Cyril Roelandt <cyril.roelandt@inria.fr>
---
trunk/src/sched_policies/heft.c | 14 ++++++--------
1 files changed, 6 insertions(+), 8 deletions(-)
diff --git a/trunk/src/sched_policies/heft.c b/trunk/src/sched_policies/heft.c
index db91a4d..d2d74cc 100644
--- a/trunk/src/sched_policies/heft.c
+++ b/trunk/src/sched_policies/heft.c
@@ -243,8 +243,7 @@ static void compute_all_performance_predictions(struct
starpu_task *task,
double *best_exp_endp,
double
local_data_penalty[STARPU_NMAXWORKERS][STARPU_MAXIMPLEMENTATIONS],
double
local_power[STARPU_NMAXWORKERS][STARPU_MAXIMPLEMENTATIONS],
- int *forced_worker, int *forced_impl,
- starpu_task_bundle_t bundle)
+ int *forced_worker, int *forced_impl)
{
int calibrating = 0;
double max_exp_end = DBL_MIN;
@@ -259,6 +258,8 @@ static void compute_all_performance_predictions(struct
starpu_task *task,
unsigned nimpl;
+ starpu_task_bundle_t bundle = task->bundle;
+
for (worker = 0; worker < nworkers; worker++)
{
struct _starpu_fifo_taskq *fifo = queue_array[worker];
@@ -410,13 +411,10 @@ static int _heft_push_task(struct starpu_task *task,
unsigned prio)
* and detect if there is some calibration that needs to be done.
*/
- starpu_task_bundle_t bundle = task->bundle;
-
compute_all_performance_predictions(task, local_task_length, exp_end,
&max_exp_end, &best_exp_end,
local_data_penalty,
- local_power, &forced_worker,
&forced_impl,
- bundle);
+ local_power, &forced_worker,
&forced_impl);
/* If there is no prediction available for that task with that arch we
* want to speed-up calibration time so we force this measurement */
@@ -483,7 +481,7 @@ static int _heft_push_task(struct starpu_task *task,
unsigned prio)
/* we should now have the best worker in variable "best" */
double model_best, transfer_model_best;
- if (bundle)
+ if (task->bundle)
{
/* If we have a task bundle, we have computed the expected
* length for the entire bundle, but not for the task alone.
*/
@@ -495,7 +493,7 @@ static int _heft_push_task(struct starpu_task *task,
unsigned prio)
/* Remove the task from the bundle since we have made a
* decision for it, and that other tasks should not consider
it
* anymore. */
- starpu_task_bundle_remove(bundle, task);
+ starpu_task_bundle_remove(task->bundle, task);
}
else
{
--
1.7.9
- Re: [Starpu-devel] [PATCH 6/7] Reindent _dmda_push_task()., (suite)
- Re: [Starpu-devel] [PATCH 6/7] Reindent _dmda_push_task()., Samuel Thibault, 19/10/2012
- [Starpu-devel] [PATCH 7/7] Handle bundles in deque_modeling_policy_data_aware.c., Cyril Roelandt, 16/10/2012
- Re: [Starpu-devel] [PATCH 7/7] Handle bundles in deque_modeling_policy_data_aware.c., Samuel Thibault, 19/10/2012
- [Starpu-devel] [PATCH 1/7] src/sched_policies/heft.c: use an array of _starpu_fifo_taskq* rather than multiple global variables., Cyril Roelandt, 16/10/2012
- Re: [Starpu-devel] [PATCH 0/7][Review needed] A first step towards the factorization of heft and dm*., Samuel Thibault, 19/10/2012
- Re: [Starpu-devel] [PATCH 0/7][Review needed] A first step towards the factorization of heft and dm*., Cyril Roelandt, 22/10/2012
- [Starpu-devel] [PATCH v2 1/6] src/sched_policies/heft.c: use an array of _starpu_fifo_taskq* rather than multiple global variables., Cyril Roelandt, 22/10/2012
- [Starpu-devel] [PATCH v2 2/6] _starpu_fifo_push_task(): Fix "TODO: if prio, put at back"., Cyril Roelandt, 22/10/2012
- [Starpu-devel] [PATCH v2 3/6] Heft: use a scheduler-specific fifo rather than pushing tasks directly to the workers' local tasks., Cyril Roelandt, 22/10/2012
- [Starpu-devel] [PATCH v2 4/6] heft.c: Remove the "bundle" parameter from compute_all_performance_predictions()., Cyril Roelandt, 22/10/2012
- [Starpu-devel] [PATCH v2 5/6] src/sched_policies/deque_modeling_policy_data_aware.c: add a compute_all_performance_predictions() function., Cyril Roelandt, 22/10/2012
- [Starpu-devel] [PATCH v2 6/6] Reindent _dmda_push_task()., Cyril Roelandt, 22/10/2012
- Re: [Starpu-devel] [PATCH 0/7][Review needed] A first step towards the factorization of heft and dm*., Samuel Thibault, 22/10/2012
- Re: [Starpu-devel] [PATCH 0/7][Review needed] A first step towards the factorization of heft and dm*., Cyril Roelandt, 22/10/2012
- [Starpu-devel] [PATCH v2 1/6] src/sched_policies/heft.c: use an array of _starpu_fifo_taskq* rather than multiple global variables., Cyril Roelandt, 22/10/2012
- Re: [Starpu-devel] [PATCH 0/7][Review needed] A first step towards the factorization of heft and dm*., Cyril Roelandt, 22/10/2012
Archives gérées par MHonArc 2.6.19+.