ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Re-worked the implementation of the LAM-like coord component.

Просмотр исходного кода 
It's a bit longer, but much more clear in it's implementation I believe.

Fundamentally it is the same, but is much more solid in the implementation.
I created quite a few directed tests that this version of the implementation 
now passes.

This commit was SVN r14457.
Этот коммит содержится в:
Josh Hursey 2007-04-21 20:35:01 +00:00
родитель f0dd3e329c
Коммит 2af38229c1
7 изменённых файлов: 4435 добавлений и 3500 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

2
ompi/mca/crcp/coord/Makefile.am
Просмотреть файл

@ -19,10 +19,8 @@ dist_pkgdata_DATA = help-ompi-crcp-coord.txt
sources = \
crcp_coord.h \
crcp_coord_pml.h \
crcp_coord_btl.h \
crcp_coord_component.c \
crcp_coord_module.c \
crcp_coord_btl.c \
crcp_coord_pml.c
# Make the output library in this directory, and name it either

3
ompi/mca/crcp/coord/crcp_coord.h
Просмотреть файл

@ -59,9 +59,6 @@ extern "C" {
int ompi_crcp_coord_pml_init(void);
int ompi_crcp_coord_pml_finalize(void);
int ompi_crcp_coord_btl_init(void);
int ompi_crcp_coord_btl_finalize(void);
#if defined(c_plusplus) || defined(__cplusplus)
}
#endif

4
ompi/mca/crcp/coord/crcp_coord_component.c
Просмотреть файл

@ -40,7 +40,7 @@ ompi_crcp_coord_component_t mca_crcp_coord_component = {
/* First do the base component stuff */
{
/* Handle the general mca_component_t struct containing
* meta information about the component itcoord
* meta information about the component
*/
{
OMPI_CRCP_BASE_VERSION_1_0_0,
@ -68,7 +68,7 @@ ompi_crcp_coord_component_t mca_crcp_coord_component = {
/* opal_output handler */
-1,
/* Default priority */
20
10
}
};

19
ompi/mca/crcp/coord/crcp_coord_module.c
Просмотреть файл

@ -37,7 +37,6 @@
#include "crcp_coord.h"
#include "crcp_coord_pml.h"
#include "crcp_coord_btl.h"
/*
* Coord module
@ -97,20 +96,6 @@ static ompi_crcp_base_module_t loc_module = {
/************************************
* Locally Global vars & functions :)
************************************/
opal_mutex_t ompi_crcp_coord_ft_global_cs_lock;
opal_condition_t ompi_crcp_coord_ft_global_cs_cond;
int ompi_crcp_coord_ft_global_cs_count;
bool ompi_crcp_coord_ft_global_cs_block;
opal_mutex_t ompi_crcp_coord_ft_send_cs_lock;
opal_condition_t ompi_crcp_coord_ft_send_cs_cond;
int ompi_crcp_coord_ft_send_cs_count;
bool ompi_crcp_coord_ft_send_cs_block;
opal_mutex_t ompi_crcp_coord_ft_recv_cs_lock;
opal_condition_t ompi_crcp_coord_ft_recv_cs_cond;
int ompi_crcp_coord_ft_recv_cs_count;
bool ompi_crcp_coord_ft_recv_cs_block;
/************************
* Function Definitions
@ -136,8 +121,6 @@ int ompi_crcp_coord_module_init(void)
ompi_crcp_coord_pml_init();
ompi_crcp_coord_btl_init();
return OMPI_SUCCESS;
}
@ -148,8 +131,6 @@ int ompi_crcp_coord_module_finalize(void)
ompi_crcp_coord_pml_finalize();
ompi_crcp_coord_btl_finalize();
return OMPI_SUCCESS;
}

7322
ompi/mca/crcp/coord/crcp_coord_pml.c
Просмотреть файл

Разница между файлами не показана из-за своего большого размера Загрузить разницу

323
ompi/mca/crcp/coord/crcp_coord_pml.h
Просмотреть файл

@ -33,6 +33,7 @@
#include "opal/runtime/opal_cr.h"
#include "opal/threads/mutex.h"
#include "opal/threads/condition.h"
#include "ompi/class/ompi_free_list.h"
#include "ompi/mca/crcp/coord/crcp_coord.h"
@ -128,35 +129,24 @@ extern "C" {
int ompi_crcp_coord_request_complete(struct ompi_request_t *request);
/***********************************
* Globally Defined Variables
* Globally Defined Structures
***********************************/
/*
* List of Messages received from ANY_SOURCES
* -- ompi_crcp_coord_pml_message_ref_t
* Types of Messages
*/
OMPI_MODULE_DECLSPEC extern opal_list_t unknown_recv_from_list;
OMPI_MODULE_DECLSPEC extern opal_list_t unknown_persist_recv_list;
enum ompi_crcp_coord_pml_message_type_t {
COORD_MSG_TYPE_UNKNOWN, /* 0 Unknown type */
COORD_MSG_TYPE_B_SEND, /* 1 Blocking Send */
COORD_MSG_TYPE_I_SEND, /* 2 Non-Blocking Send */
COORD_MSG_TYPE_P_SEND, /* 3 Persistent Send */
COORD_MSG_TYPE_B_RECV, /* 4 Blocking Recv */
COORD_MSG_TYPE_I_RECV, /* 5 Non-Blocking Recv */
COORD_MSG_TYPE_P_RECV /* 6 Persistent Recv */
};
typedef enum ompi_crcp_coord_pml_message_type_t ompi_crcp_coord_pml_message_type_t;
/*
* List of pending ACKs to drained messages
* -- ompi_crcp_coord_pml_drain_msg_ack_ref_t
*/
OMPI_MODULE_DECLSPEC extern opal_list_t drained_msg_ack_list;
/*
* List of drained messages to match against
* -- ompi_crcp_coord_pml_message_ref_t
*/
OMPI_MODULE_DECLSPEC extern opal_list_t drained_msg_list;
/*
* List of processes known
* -- ompi_crcp_coord_pml_bookmark_proc_t
*/
OMPI_MODULE_DECLSPEC extern opal_list_t ompi_crcp_coord_pml_procs;
/*
* Message reference
* Message Reference
*/
struct ompi_crcp_coord_pml_message_ref_t {
/** This is a list object */
@ -165,6 +155,10 @@ extern "C" {
/** Sequence Number of this message */
uint64_t msg_id;
/** Type of message this references */
ompi_crcp_coord_pml_message_type_t msg_type;
/** Buffer for data */
void * buffer;
@ -186,18 +180,13 @@ extern "C" {
/** Communicator pointer */
ompi_communicator_t* comm;
/** Message Mode */
mca_pml_base_send_mode_t mode;
/* Is this an asynchronous message */
bool async;
/** Receive Request */
ompi_request_t *request;
/** Status */
ompi_status_public_t status;
/** Peer which we received from */
orte_process_name_t proc_name;
@ -247,7 +236,6 @@ extern "C" {
* drain it to a specific peer
*/
int suggested_rank;
};
typedef struct ompi_crcp_coord_pml_message_ref_t ompi_crcp_coord_pml_message_ref_t;
@ -262,20 +250,13 @@ extern "C" {
* - List of received message from this peer
* - Message totals
*/
struct ompi_crcp_coord_pml_bookmark_proc_t {
struct ompi_crcp_coord_pml_peer_ref_t {
/** This is a list object */
opal_list_item_t super;
/** Name of peer */
orte_process_name_t proc_name;
/*
* Just to control concurrent access to some of these counters,
* and the PML
*/
opal_mutex_t lock;
opal_condition_t cond;
/** List of messages sent to this peer */
opal_list_t send_list; /**< pml_send */
opal_list_t isend_list; /**< pml_isend */
@ -318,262 +299,36 @@ extern "C" {
uint32_t matched_recv_msgs;
uint32_t matched_irecv_msgs;
uint32_t matched_recv_init_msgs;
/** Total Number of messages drained */
uint32_t total_drained_msgs;
};
typedef struct ompi_crcp_coord_pml_bookmark_proc_t ompi_crcp_coord_pml_bookmark_proc_t;
typedef struct ompi_crcp_coord_pml_peer_ref_t ompi_crcp_coord_pml_peer_ref_t;
OBJ_CLASS_DECLARATION(ompi_crcp_coord_pml_bookmark_proc_t);
void ompi_crcp_coord_pml_bookmark_proc_construct(ompi_crcp_coord_pml_bookmark_proc_t *bkm_proc);
void ompi_crcp_coord_pml_bookmark_proc_destruct( ompi_crcp_coord_pml_bookmark_proc_t *bkm_proc);
OBJ_CLASS_DECLARATION(ompi_crcp_coord_pml_peer_ref_t);
void ompi_crcp_coord_pml_peer_ref_construct(ompi_crcp_coord_pml_peer_ref_t *bkm_proc);
void ompi_crcp_coord_pml_peer_ref_destruct( ompi_crcp_coord_pml_peer_ref_t *bkm_proc);
/*
* Local version of the PML state
*/
struct ompi_crcp_coord_pml_state_t {
ompi_crcp_base_pml_state_t super;
ompi_crcp_base_pml_state_t p_super;
ompi_crcp_base_pml_state_t *prev_ptr;
ompi_crcp_coord_pml_bookmark_proc_t *peer_ref;
ompi_crcp_coord_pml_peer_ref_t *peer_ref;
ompi_crcp_coord_pml_message_ref_t *msg_ref;
};
typedef struct ompi_crcp_coord_pml_state_t ompi_crcp_coord_pml_state_t;
OBJ_CLASS_DECLARATION(ompi_crcp_coord_pml_state_t);
/***************
* A bit of locking, it's good for you
***************/
/***********************************
* Globally Defined Variables
***********************************/
/*
* Any thread can call this when entering a critical section
* This in not strictly a critical section, but a protected
* section of code while checkpointing is occuring.
* List of known peers
*/
#if OPAL_ENABLE_FT == 1
#define OMPI_CRCP_COORD_CS_ENTER(bvar, ctr, lock, cond) \
{ \
opal_mutex_lock(&lock); \
while( bvar ) { \
opal_condition_wait(&cond, \
&lock); \
} \
ctr++; \
opal_mutex_unlock(&lock); \
}
#else
#define OMPI_CRCP_COORD_CS_ENTER(bvar, ctr, lock, cond) ;
#endif
/*
* Any thread can call this when exiting a critical section
*/
#if OPAL_ENABLE_FT == 1
#define OMPI_CRCP_COORD_CS_EXIT(ctr, lock, cond) \
{ \
opal_mutex_lock(&lock); \
ctr--; \
opal_condition_signal(&cond); \
opal_mutex_unlock(&lock); \
}
#else
#define OMPI_CRCP_COORD_CS_EXIT(ctr, lock, cond) ;
#endif
/*
* Checkpoint protocol calls this to restrict processes
* from entering a specific critical section.
*/
#if OPAL_ENABLE_FT == 1
#define OMPI_CRCP_COORD_CS_RESTRICT(bvar, ctr, lock, cond, wait, dbg_str) \
{ \
opal_mutex_lock(&lock); \
bvar = true; \
while(ctr > 0 && wait) { \
opal_condition_wait(&cond, \
&lock); \
} \
opal_mutex_unlock(&lock); \
}
#else
#define OMPI_CRCP_COORD_CS_RESTRICT(bvar, ctr, lock, cond, wait, dbg_str) ;
#endif
/*
* Checkpoint protocol calls this to release all the blocking
* threads so that they may enter the critical section.
*/
#if OPAL_ENABLE_FT == 1
#define OMPI_CRCP_COORD_CS_RELEASE(bvar, lock, cond) \
{ \
opal_mutex_lock(&lock); \
bvar = false; \
opal_condition_signal(&cond); \
opal_mutex_unlock(&lock); \
}
#else
#define OMPI_CRCP_COORD_CS_RELEASE(bvar, lock, cond) ;
#endif
/*
* Some short cuts.
*/
OPAL_DECLSPEC extern opal_mutex_t ompi_crcp_coord_ft_global_cs_lock;
OPAL_DECLSPEC extern opal_condition_t ompi_crcp_coord_ft_global_cs_cond;
OPAL_DECLSPEC extern int ompi_crcp_coord_ft_global_cs_count;
OPAL_DECLSPEC extern bool ompi_crcp_coord_ft_global_cs_block;
OPAL_DECLSPEC extern opal_mutex_t ompi_crcp_coord_ft_send_cs_lock;
OPAL_DECLSPEC extern opal_condition_t ompi_crcp_coord_ft_send_cs_cond;
OPAL_DECLSPEC extern int ompi_crcp_coord_ft_send_cs_count;
OPAL_DECLSPEC extern bool ompi_crcp_coord_ft_send_cs_block;
OPAL_DECLSPEC extern opal_mutex_t ompi_crcp_coord_ft_recv_cs_lock;
OPAL_DECLSPEC extern opal_condition_t ompi_crcp_coord_ft_recv_cs_cond;
OPAL_DECLSPEC extern int ompi_crcp_coord_ft_recv_cs_count;
OPAL_DECLSPEC extern bool ompi_crcp_coord_ft_recv_cs_block;
#if OPAL_ENABLE_FT == 1 && OPAL_ENABLE_FT_THREAD == 1
/* Global stuff */
#define OMPI_CRCP_COORD_FT_GLOBAL_INIT() \
{ \
OBJ_CONSTRUCT(&ompi_crcp_coord_ft_global_cs_lock, opal_mutex_t); \
OBJ_CONSTRUCT(&ompi_crcp_coord_ft_global_cs_cond, opal_condition_t); \
ompi_crcp_coord_ft_global_cs_count = 0; \
ompi_crcp_coord_ft_global_cs_block = false; \
}
#define OMPI_CRCP_COORD_FT_GLOBAL_FINALIZE() \
{ \
OBJ_DESTRUCT(&ompi_crcp_coord_ft_global_cs_lock); \
OBJ_DESTRUCT(&ompi_crcp_coord_ft_global_cs_cond); \
ompi_crcp_coord_ft_global_cs_count = 0; \
ompi_crcp_coord_ft_global_cs_block = false; \
}
#define OMPI_CRCP_COORD_FT_GLOBAL_CS_ENTER() \
OMPI_CRCP_COORD_CS_ENTER(ompi_crcp_coord_ft_global_cs_block, \
ompi_crcp_coord_ft_global_cs_count, \
ompi_crcp_coord_ft_global_cs_lock, \
ompi_crcp_coord_ft_global_cs_cond);
#define OMPI_CRCP_COORD_FT_GLOBAL_CS_EXIT() \
OMPI_CRCP_COORD_CS_EXIT(ompi_crcp_coord_ft_global_cs_count, \
ompi_crcp_coord_ft_global_cs_lock, \
ompi_crcp_coord_ft_global_cs_cond);
#define OMPI_CRCP_COORD_FT_GLOBAL_CS_RESTRICT(wait) \
OMPI_CRCP_COORD_CS_RESTRICT(ompi_crcp_coord_ft_global_cs_block, \
ompi_crcp_coord_ft_global_cs_count, \
ompi_crcp_coord_ft_global_cs_lock, \
ompi_crcp_coord_ft_global_cs_cond, \
wait, \
"CRCP GLOBAL");
#define OMPI_CRCP_COORD_FT_GLOBAL_CS_RELEASE() \
OMPI_CRCP_COORD_CS_RELEASE(ompi_crcp_coord_ft_global_cs_block, \
ompi_crcp_coord_ft_global_cs_lock, \
ompi_crcp_coord_ft_global_cs_cond);
/* Send stuff */
#define OMPI_CRCP_COORD_FT_SEND_INIT() \
{ \
OBJ_CONSTRUCT(&ompi_crcp_coord_ft_send_cs_lock, opal_mutex_t); \
OBJ_CONSTRUCT(&ompi_crcp_coord_ft_send_cs_cond, opal_condition_t); \
ompi_crcp_coord_ft_send_cs_count = 0; \
ompi_crcp_coord_ft_send_cs_block = false; \
}
#define OMPI_CRCP_COORD_FT_SEND_FINALIZE() \
{ \
OBJ_DESTRUCT(&ompi_crcp_coord_ft_send_cs_lock); \
OBJ_DESTRUCT(&ompi_crcp_coord_ft_send_cs_cond); \
ompi_crcp_coord_ft_send_cs_count = 0; \
ompi_crcp_coord_ft_send_cs_block = false; \
}
#define OMPI_CRCP_COORD_FT_SEND_CS_ENTER() \
OMPI_CRCP_COORD_CS_ENTER(ompi_crcp_coord_ft_send_cs_block, \
ompi_crcp_coord_ft_send_cs_count, \
ompi_crcp_coord_ft_send_cs_lock, \
ompi_crcp_coord_ft_send_cs_cond);
#define OMPI_CRCP_COORD_FT_SEND_CS_EXIT() \
OMPI_CRCP_COORD_CS_EXIT(ompi_crcp_coord_ft_send_cs_count, \
ompi_crcp_coord_ft_send_cs_lock, \
ompi_crcp_coord_ft_send_cs_cond);
#define OMPI_CRCP_COORD_FT_SEND_CS_RESTRICT(wait) \
OMPI_CRCP_COORD_CS_RESTRICT(ompi_crcp_coord_ft_send_cs_block, \
ompi_crcp_coord_ft_send_cs_count, \
ompi_crcp_coord_ft_send_cs_lock, \
ompi_crcp_coord_ft_send_cs_cond, \
wait, \
"CRCP SEND");
#define OMPI_CRCP_COORD_FT_SEND_CS_RELEASE() \
OMPI_CRCP_COORD_CS_RELEASE(ompi_crcp_coord_ft_send_cs_block, \
ompi_crcp_coord_ft_send_cs_lock, \
ompi_crcp_coord_ft_send_cs_cond);
/* Receive stuff */
#define OMPI_CRCP_COORD_FT_RECV_INIT() \
{ \
OBJ_CONSTRUCT(&ompi_crcp_coord_ft_recv_cs_lock, opal_mutex_t); \
OBJ_CONSTRUCT(&ompi_crcp_coord_ft_recv_cs_cond, opal_condition_t); \
ompi_crcp_coord_ft_recv_cs_count = 0; \
ompi_crcp_coord_ft_recv_cs_block = false; \
}
#define OMPI_CRCP_COORD_FT_RECV_FINALIZE() \
{ \
OBJ_DESTRUCT(&ompi_crcp_coord_ft_recv_cs_lock); \
OBJ_DESTRUCT(&ompi_crcp_coord_ft_recv_cs_cond); \
ompi_crcp_coord_ft_recv_cs_count = 0; \
ompi_crcp_coord_ft_recv_cs_block = false; \
}
#define OMPI_CRCP_COORD_FT_RECV_CS_ENTER() \
OMPI_CRCP_COORD_CS_ENTER(ompi_crcp_coord_ft_recv_cs_block, \
ompi_crcp_coord_ft_recv_cs_count, \
ompi_crcp_coord_ft_recv_cs_lock, \
ompi_crcp_coord_ft_recv_cs_cond);
#define OMPI_CRCP_COORD_FT_RECV_CS_EXIT() \
OMPI_CRCP_COORD_CS_EXIT(ompi_crcp_coord_ft_recv_cs_count, \
ompi_crcp_coord_ft_recv_cs_lock, \
ompi_crcp_coord_ft_recv_cs_cond);
#define OMPI_CRCP_COORD_FT_RECV_CS_RESTRICT(wait) \
OMPI_CRCP_COORD_CS_RESTRICT(ompi_crcp_coord_ft_recv_cs_block, \
ompi_crcp_coord_ft_recv_cs_count, \
ompi_crcp_coord_ft_recv_cs_lock, \
ompi_crcp_coord_ft_recv_cs_cond, \
wait, \
"CRCP RECV");
#define OMPI_CRCP_COORD_FT_RECV_CS_RELEASE() \
OMPI_CRCP_COORD_CS_RELEASE(ompi_crcp_coord_ft_recv_cs_block, \
ompi_crcp_coord_ft_recv_cs_lock, \
ompi_crcp_coord_ft_recv_cs_cond);
#else
#define OMPI_CRCP_COORD_FT_GLOBAL_INIT() ;
#define OMPI_CRCP_COORD_FT_GLOBAL_FINALIZE() ;
#define OMPI_CRCP_COORD_FT_GLOBAL_CS_ENTER() ;
#define OMPI_CRCP_COORD_FT_GLOBAL_CS_EXIT() ;
#define OMPI_CRCP_COORD_FT_GLOBAL_CS_RESTRICT(wait) ;
#define OMPI_CRCP_COORD_FT_GLOBAL_CS_RELEASE() ;
#define OMPI_CRCP_COORD_FT_SEND_INIT() ;
#define OMPI_CRCP_COORD_FT_SEND_FINALIZE() ;
#define OMPI_CRCP_COORD_FT_SEND_CS_ENTER() ;
#define OMPI_CRCP_COORD_FT_SEND_CS_EXIT() ;
#define OMPI_CRCP_COORD_FT_SEND_CS_RESTRICT(wait) ;
#define OMPI_CRCP_COORD_FT_SEND_CS_RELEASE() ;
#define OMPI_CRCP_COORD_FT_RECV_INIT() ;
#define OMPI_CRCP_COORD_FT_RECV_FINALIZE() ;
#define OMPI_CRCP_COORD_FT_RECV_CS_ENTER() ;
#define OMPI_CRCP_COORD_FT_RECV_CS_EXIT() ;
#define OMPI_CRCP_COORD_FT_RECV_CS_RESTRICT(wait) ;
#define OMPI_CRCP_COORD_FT_RECV_CS_RELEASE() ;
#endif /* ENABLE_FT */
extern opal_list_t ompi_crcp_coord_pml_peer_refs;
#if defined(c_plusplus) || defined(__cplusplus)
}

2
ompi/mca/crcp/coord/help-ompi-crcp-coord.txt
Просмотреть файл

@ -16,5 +16,5 @@
#
# $HEADER$
#
# This is the US/English general help file for ORTE SnapC framework.
# This is the US/English general help file for CRCP Coord framework.
#