Skip to content
GitLab
Menu
Projects
Groups
Snippets
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
Iker Martín Álvarez
Proteo
Commits
3e176fda
Commit
3e176fda
authored
Jun 22, 2021
by
iker_martin
Browse files
Anadida funcion para realizar comunicaciones en segundo plano por parte de hebras
parent
d79aefaf
Changes
3
Hide whitespace changes
Inline
Side-by-side
Codes/Main/Main.c
View file @
3e176fda
...
@@ -3,6 +3,7 @@
...
@@ -3,6 +3,7 @@
#include <mpi.h>
#include <mpi.h>
#include <fcntl.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <sys/stat.h>
#include <pthread.h>
#include "../IOcodes/read_ini.h"
#include "../IOcodes/read_ini.h"
#include "../IOcodes/results.h"
#include "../IOcodes/results.h"
#include "../malleability/ProcessDist.h"
#include "../malleability/ProcessDist.h"
...
@@ -17,6 +18,11 @@ int checkpoint(int iter, int state, MPI_Request **comm_req);
...
@@ -17,6 +18,11 @@ int checkpoint(int iter, int state, MPI_Request **comm_req);
void
TC
(
int
numS
);
void
TC
(
int
numS
);
int
start_redistribution
(
int
numS
,
MPI_Request
**
comm_req
);
int
start_redistribution
(
int
numS
,
MPI_Request
**
comm_req
);
int
check_redistribution
(
int
iter
,
MPI_Request
**
comm_req
);
int
check_redistribution
(
int
iter
,
MPI_Request
**
comm_req
);
int
end_redistribution
(
int
iter
);
int
thread_creation
();
int
thread_check
();
void
*
thread_async_work
(
void
*
void_arg
);
void
iterate
(
double
*
matrix
,
int
n
,
int
async_comm
);
void
iterate
(
double
*
matrix
,
int
n
,
int
async_comm
);
void
computeMatrix
(
double
*
matrix
,
int
n
);
void
computeMatrix
(
double
*
matrix
,
int
n
);
...
@@ -38,16 +44,26 @@ typedef struct {
...
@@ -38,16 +44,26 @@ typedef struct {
int
iter_start
;
int
iter_start
;
int
argc
;
int
argc
;
int
numS
;
// Cantidad de procesos hijos
int
commAsync
;
MPI_Comm
children
,
parents
;
MPI_Comm
children
,
parents
;
char
**
argv
;
char
**
argv
;
char
*
sync_array
,
*
async_array
;
char
*
sync_array
,
*
async_array
;
}
group_data
;
}
group_data
;
typedef
struct
{
int
myId
,
numP
,
numS
,
adr
;
MPI_Comm
children
;
char
*
sync_array
;
}
thread_data
;
configuration
*
config_file
;
configuration
*
config_file
;
group_data
*
group
;
group_data
*
group
;
results_data
*
results
;
results_data
*
results
;
int
run_id
=
0
;
// Utilizado para diferenciar más fácilmente ejecuciones en el análisis
int
run_id
=
0
;
// Utilizado para diferenciar más fácilmente ejecuciones en el análisis
pthread_t
async_thread
;
// TODO Cambiar de sitio?
int
main
(
int
argc
,
char
*
argv
[])
{
int
main
(
int
argc
,
char
*
argv
[])
{
int
numP
,
myId
,
res
;
int
numP
,
myId
,
res
;
int
req
;
int
req
;
...
@@ -57,7 +73,13 @@ int main(int argc, char *argv[]) {
...
@@ -57,7 +73,13 @@ int main(int argc, char *argv[]) {
MPI_Comm_rank
(
MPI_COMM_WORLD
,
&
myId
);
MPI_Comm_rank
(
MPI_COMM_WORLD
,
&
myId
);
init_group_struct
(
argv
,
argc
,
myId
,
numP
);
init_group_struct
(
argv
,
argc
,
myId
,
numP
);
init_application
();
MPI_Comm_get_parent
(
&
(
group
->
parents
));
if
(
group
->
parents
==
MPI_COMM_NULL
)
{
// Si son el primer grupo de procesos, recogen la configuracion inicial
init_application
();
}
else
{
// Si son procesos hijos deben comunicarse con las padres
Sons_init
();
}
if
(
group
->
grp
==
0
)
{
if
(
group
->
grp
==
0
)
{
MPI_Barrier
(
MPI_COMM_WORLD
);
MPI_Barrier
(
MPI_COMM_WORLD
);
...
@@ -70,6 +92,7 @@ int main(int argc, char *argv[]) {
...
@@ -70,6 +92,7 @@ int main(int argc, char *argv[]) {
MPI_Barrier
(
MPI_COMM_WORLD
);
MPI_Barrier
(
MPI_COMM_WORLD
);
results
->
exec_time
=
MPI_Wtime
()
-
results
->
exec_start
;
results
->
exec_time
=
MPI_Wtime
()
-
results
->
exec_start
;
}
}
print_final_results
();
print_final_results
();
free_application_data
();
free_application_data
();
...
@@ -139,16 +162,18 @@ int checkpoint(int iter, int state, MPI_Request **comm_req) {
...
@@ -139,16 +162,18 @@ int checkpoint(int iter, int state, MPI_Request **comm_req) {
// Comprobar si se tiene que realizar un redimensionado
// Comprobar si se tiene que realizar un redimensionado
if
(
config_file
->
iters
[
group
->
grp
]
>
iter
||
config_file
->
resizes
==
group
->
grp
+
1
)
{
return
MAL_COMM_UNINITIALIZED
;}
if
(
config_file
->
iters
[
group
->
grp
]
>
iter
||
config_file
->
resizes
==
group
->
grp
+
1
)
{
return
MAL_COMM_UNINITIALIZED
;}
int
numS
=
config_file
->
procs
[
group
->
grp
+
1
];
group
->
numS
=
config_file
->
procs
[
group
->
grp
+
1
];
results
->
spawn_start
=
MPI_Wtime
();
results
->
spawn_start
=
MPI_Wtime
();
TC
(
numS
);
TC
(
group
->
numS
);
results
->
spawn_time
[
group
->
grp
]
=
MPI_Wtime
()
-
results
->
spawn_start
;
results
->
spawn_time
[
group
->
grp
]
=
MPI_Wtime
()
-
results
->
spawn_start
;
state
=
start_redistribution
(
numS
,
comm_req
);
state
=
start_redistribution
(
group
->
numS
,
comm_req
);
}
else
if
(
state
==
MAL_ASYNC_PENDING
)
{
}
else
if
(
state
==
MAL_ASYNC_PENDING
)
{
state
=
check_redistribution
(
iter
,
comm_req
);
state
=
thread_check
();
if
(
state
==
MAL_COMM_COMPLETED
)
end_redistribution
(
iter
);
//state = check_redistribution(iter, comm_req);
}
}
return
state
;
return
state
;
...
@@ -195,20 +220,57 @@ int start_redistribution(int numS, MPI_Request **comm_req) {
...
@@ -195,20 +220,57 @@ int start_redistribution(int numS, MPI_Request **comm_req) {
if
(
config_file
->
adr
>
0
)
{
if
(
config_file
->
adr
>
0
)
{
results
->
async_start
=
MPI_Wtime
();
results
->
async_start
=
MPI_Wtime
();
send_async
(
group
->
async_array
,
config_file
->
adr
,
group
->
myId
,
group
->
numP
,
ROOT
,
group
->
children
,
numS
,
comm_req
,
config_file
->
aib
);
//send_async(group->async_array, config_file->adr, group->myId, group->numP, ROOT, group->children, numS, comm_req, config_file->aib);
return
MAL_ASYNC_PENDING
;
return
thread_creation
();
//return MAL_ASYNC_PENDING;
}
}
if
(
config_file
->
sdr
>
0
)
{
return
end_redistribution
(
0
);
results
->
sync_start
=
MPI_Wtime
();
}
send_sync
(
group
->
sync_array
,
config_file
->
sdr
,
group
->
myId
,
group
->
numP
,
ROOT
,
group
->
children
,
numS
);
/*
* Crea una hebra para ejecutar una comunicación en segundo plano.
*/
int
thread_creation
()
{
if
(
pthread_create
(
&
async_thread
,
NULL
,
thread_async_work
,
NULL
))
{
printf
(
"Error al crear el hilo
\n
"
);
MPI_Abort
(
MPI_COMM_WORLD
,
-
1
);
return
-
1
;
}
}
return
MAL_ASYNC_PENDING
;
}
/*
send_results
(
results
,
rootBcast
,
config_file
->
resizes
,
group
->
children
);
* Comprobación por parte de una hebra maestra que indica
// Desconectar intercomunicador con los hijos
* si una hebra esclava ha terminado su comunicación en segundo plano.
MPI_Comm_disconnect
(
&
(
group
->
children
));
*
* El estado de la comunicación es devuelto al finalizar la función.
*/
int
thread_check
()
{
if
(
group
->
commAsync
==
MAL_COMM_COMPLETED
)
{
if
(
pthread_join
(
async_thread
,
NULL
))
{
printf
(
"Error al esperar al hilo
\n
"
);
MPI_Abort
(
MPI_COMM_WORLD
,
-
1
);
return
-
2
;
}
return
MAL_COMM_COMPLETED
;
}
return
MAL_COMM_COMPLETED
;
return
MAL_ASYNC_PENDING
;
}
/*
* Función ejecutada por una hebra.
* Ejecuta una comunicación síncrona con los hijos que
* para el usuario se puede considerar como en segundo plano.
*
* Cuando termina la comunicación la hebra maestra puede comprobarlo
* por el valor "commAsync".
*/
void
*
thread_async_work
(
void
*
void_arg
)
{
send_sync
(
group
->
async_array
,
config_file
->
adr
,
group
->
myId
,
group
->
numP
,
ROOT
,
group
->
children
,
group
->
numS
);
group
->
commAsync
=
MAL_COMM_COMPLETED
;
pthread_exit
(
NULL
);
}
}
/*
/*
...
@@ -225,11 +287,8 @@ int start_redistribution(int numS, MPI_Request **comm_req) {
...
@@ -225,11 +287,8 @@ int start_redistribution(int numS, MPI_Request **comm_req) {
* los hijos han terminado de recibir.
* los hijos han terminado de recibir.
*/
*/
int
check_redistribution
(
int
iter
,
MPI_Request
**
comm_req
)
{
int
check_redistribution
(
int
iter
,
MPI_Request
**
comm_req
)
{
int
completed
,
all_completed
,
test_err
,
iter_send
;
int
completed
,
all_completed
,
test_err
;
int
numS
=
config_file
->
procs
[
group
->
grp
+
1
];
int
rootBcast
=
MPI_PROC_NULL
;
MPI_Request
*
req_completed
;
MPI_Request
*
req_completed
;
if
(
group
->
myId
==
ROOT
)
rootBcast
=
MPI_ROOT
;
if
(
config_file
->
aib
==
MAL_USE_NORMAL
)
{
if
(
config_file
->
aib
==
MAL_USE_NORMAL
)
{
req_completed
=
&
(
*
comm_req
)[
0
];
req_completed
=
&
(
*
comm_req
)[
0
];
...
@@ -254,18 +313,34 @@ int check_redistribution(int iter, MPI_Request **comm_req) {
...
@@ -254,18 +313,34 @@ int check_redistribution(int iter, MPI_Request **comm_req) {
//Para la desconexión de ambos grupos de procesos es necesario indicar a MPI que esta
//Para la desconexión de ambos grupos de procesos es necesario indicar a MPI que esta
//ha terminado, aunque solo se pueda llegar a este punto cuando ha terminado
//ha terminado, aunque solo se pueda llegar a este punto cuando ha terminado
}
}
free
(
*
comm_req
);
iter_send
=
iter
;
return
end_redistribution
(
iter
);
MPI_Bcast
(
&
iter_send
,
1
,
MPI_INT
,
rootBcast
,
group
->
children
);
}
/*
* Termina la redistribución de los datos con los hijos, comprobando
* si se han realizado iteraciones con comunicaciones en segundo plano
* y enviando cuantas iteraciones se han realizado a los hijos.
*
* Además se realizan las comunicaciones síncronas se las hay.
* Finalmente termina enviando los datos temporales a los hijos.
*/
int
end_redistribution
(
int
iter
)
{
int
rootBcast
=
MPI_PROC_NULL
;
if
(
group
->
myId
==
ROOT
)
rootBcast
=
MPI_ROOT
;
if
(
iter
>
0
)
{
// Mandar a los hijos iteracion en la que comenzar
MPI_Bcast
(
&
iter
,
1
,
MPI_INT
,
rootBcast
,
group
->
children
);
}
if
(
config_file
->
sdr
>
0
)
{
// Realizar envio sincrono
if
(
config_file
->
sdr
>
0
)
{
// Realizar envio sincrono
results
->
sync_start
=
MPI_Wtime
();
results
->
sync_start
=
MPI_Wtime
();
send_sync
(
group
->
sync_array
,
config_file
->
sdr
,
group
->
myId
,
group
->
numP
,
ROOT
,
group
->
children
,
numS
);
send_sync
(
group
->
sync_array
,
config_file
->
sdr
,
group
->
myId
,
group
->
numP
,
ROOT
,
group
->
children
,
group
->
numS
);
}
}
send_results
(
results
,
rootBcast
,
config_file
->
resizes
,
group
->
children
);
send_results
(
results
,
rootBcast
,
config_file
->
resizes
,
group
->
children
);
// Desconectar intercomunicador con los hijos
// Desconectar intercomunicador con los hijos
MPI_Comm_disconnect
(
&
(
group
->
children
));
MPI_Comm_disconnect
(
&
(
group
->
children
));
free
(
*
comm_req
);
return
MAL_COMM_COMPLETED
;
return
MAL_COMM_COMPLETED
;
}
}
...
@@ -287,7 +362,7 @@ void Sons_init() {
...
@@ -287,7 +362,7 @@ void Sons_init() {
init_results_data
(
&
results
,
config_file
->
resizes
-
1
,
config_file
->
iters
[
group
->
grp
]);
init_results_data
(
&
results
,
config_file
->
resizes
-
1
,
config_file
->
iters
[
group
->
grp
]);
if
(
config_file
->
adr
>
0
)
{
// Recibir datos asincronos
if
(
config_file
->
adr
>
0
)
{
// Recibir datos asincronos
recv_
a
sync
(
&
(
group
->
async_array
),
config_file
->
adr
,
group
->
myId
,
group
->
numP
,
ROOT
,
group
->
parents
,
numP_parents
,
config_file
->
aib
);
recv_sync
(
&
(
group
->
async_array
),
config_file
->
adr
,
group
->
myId
,
group
->
numP
,
ROOT
,
group
->
parents
,
numP_parents
);
results
->
async_time
[
group
->
grp
]
=
MPI_Wtime
();
results
->
async_time
[
group
->
grp
]
=
MPI_Wtime
();
MPI_Bcast
(
&
(
group
->
iter_start
),
1
,
MPI_INT
,
ROOT
,
group
->
parents
);
MPI_Bcast
(
&
(
group
->
iter_start
),
1
,
MPI_INT
,
ROOT
,
group
->
parents
);
}
}
...
@@ -489,6 +564,7 @@ void init_group_struct(char *argv[], int argc, int myId, int numP) {
...
@@ -489,6 +564,7 @@ void init_group_struct(char *argv[], int argc, int myId, int numP) {
group
->
numP
=
numP
;
group
->
numP
=
numP
;
group
->
grp
=
0
;
group
->
grp
=
0
;
group
->
iter_start
=
0
;
group
->
iter_start
=
0
;
group
->
commAsync
=
MAL_COMM_UNINITIALIZED
;
group
->
argc
=
argc
;
group
->
argc
=
argc
;
group
->
argv
=
argv
;
group
->
argv
=
argv
;
}
}
...
@@ -503,42 +579,35 @@ void init_group_struct(char *argv[], int argc, int myId, int numP) {
...
@@ -503,42 +579,35 @@ void init_group_struct(char *argv[], int argc, int myId, int numP) {
* se comunican con los padres para inicializar sus datos.
* se comunican con los padres para inicializar sus datos.
*/
*/
void
init_application
()
{
void
init_application
()
{
if
(
group
->
argc
<
2
)
{
printf
(
"Falta el fichero de configuracion. Uso:
\n
./programa config.ini id
\n
El argumento numerico id es opcional
\n
"
);
exit
(
0
);
}
if
(
group
->
argc
>
2
)
{
run_id
=
atoi
(
group
->
argv
[
2
]);
}
MPI_Comm_get_parent
(
&
(
group
->
parents
));
config_file
=
read_ini_file
(
group
->
argv
[
1
]);
if
(
group
->
parents
!=
MPI_COMM_NULL
)
{
// Si son procesos hijos deben comunicarse con las padres
init_results_data
(
&
results
,
config_file
->
resizes
,
config_file
->
iters
[
group
->
grp
]);
Sons_init
();
if
(
config_file
->
sdr
>
0
)
{
}
else
{
// Si son el primer grupo de procesos, recogen la configuracion inicial
malloc_comm_array
(
&
(
group
->
sync_array
),
config_file
->
sdr
,
group
->
myId
,
group
->
numP
);
}
if
(
group
->
argc
<
2
)
{
if
(
config_file
->
adr
>
0
)
{
printf
(
"Falta el fichero de configuracion. Uso:
\n
./programa config.ini id
\n
El argumento numerico id es opcional
\n
"
);
malloc_comm_array
(
&
(
group
->
async_array
),
config_file
->
adr
,
group
->
myId
,
group
->
numP
);
exit
(
0
);
}
}
if
(
group
->
argc
>
2
)
{
run_id
=
atoi
(
group
->
argv
[
2
]);
}
config_file
=
read_ini_file
(
group
->
argv
[
1
]);
init_results_data
(
&
results
,
config_file
->
resizes
,
config_file
->
iters
[
group
->
grp
]);
if
(
config_file
->
sdr
>
0
)
{
malloc_comm_array
(
&
(
group
->
sync_array
),
config_file
->
sdr
,
group
->
myId
,
group
->
numP
);
}
if
(
config_file
->
adr
>
0
)
{
malloc_comm_array
(
&
(
group
->
async_array
),
config_file
->
adr
,
group
->
myId
,
group
->
numP
);
}
double
result
,
start_time
=
MPI_Wtime
();
double
result
,
start_time
=
MPI_Wtime
();
int
i
;
int
i
;
result
=
0
;
result
=
0
;
for
(
i
=
0
;
i
<
10000
;
i
++
)
{
for
(
i
=
0
;
i
<
20000
;
i
++
)
{
result
+=
computePiSerial
(
config_file
->
matrix_tam
);
result
+=
computePiSerial
(
config_file
->
matrix_tam
);
}
printf
(
"Creado Top con valor %lf
\n
"
,
result
);
fflush
(
stdout
);
config_file
->
Top
=
(
MPI_Wtime
()
-
start_time
)
/
10000
;
//Tiempo de una iteracion
MPI_Bcast
(
&
(
config_file
->
Top
),
1
,
MPI_DOUBLE
,
ROOT
,
MPI_COMM_WORLD
);
}
}
printf
(
"Creado Top con valor %lf
\n
"
,
result
);
fflush
(
stdout
);
config_file
->
Top
=
(
MPI_Wtime
()
-
start_time
)
/
20000
;
//Tiempo de una iteracion
MPI_Bcast
(
&
(
config_file
->
Top
),
1
,
MPI_DOUBLE
,
ROOT
,
MPI_COMM_WORLD
);
}
}
/*
/*
...
...
Codes/malleability/CommDist.c
View file @
3e176fda
...
@@ -113,7 +113,7 @@ void recv_sync(char **array, int qty, int myId, int numP, int root, MPI_Comm int
...
@@ -113,7 +113,7 @@ void recv_sync(char **array, int qty, int myId, int numP, int root, MPI_Comm int
// Obtener distribución para este hijo
// Obtener distribución para este hijo
get_dist
(
qty
,
myId
,
numP
,
&
dist_data
);
get_dist
(
qty
,
myId
,
numP
,
&
dist_data
);
*
array
=
malloc
(
dist_data
.
tamBl
*
sizeof
(
char
));
*
array
=
malloc
(
dist_data
.
tamBl
*
sizeof
(
char
));
(
*
array
)[
dist_data
.
tamBl
]
=
'\0'
;
//
(*array)[dist_data.tamBl] = '\0';
dist_data
.
intercomm
=
intercomm
;
dist_data
.
intercomm
=
intercomm
;
/* PREPARAR DATOS DE RECEPCION SOBRE VECTOR*/
/* PREPARAR DATOS DE RECEPCION SOBRE VECTOR*/
...
...
Codes/malleability/CommDist.h
View file @
3e176fda
...
@@ -4,7 +4,7 @@
...
@@ -4,7 +4,7 @@
#include <string.h>
#include <string.h>
#define MAL_COMM_COMPLETED 0
#define MAL_COMM_COMPLETED 0
#define MAL_COMM_UNINITIALIZED
0
#define MAL_COMM_UNINITIALIZED
2
#define MAL_ASYNC_PENDING 1
#define MAL_ASYNC_PENDING 1
#define MAL_USE_NORMAL 0
#define MAL_USE_NORMAL 0
...
...
Write
Preview
Supports
Markdown
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment