void SIMIX_network_init(void)
{
- rdv_points = xbt_dict_new();
+ rdv_points = xbt_dict_new_homogeneous(SIMIX_rdv_free);
}
void SIMIX_network_exit(void)
rdv->comm_fifo = xbt_fifo_new();
if (rdv->name)
- xbt_dict_set(rdv_points, rdv->name, rdv, SIMIX_rdv_free);
+ xbt_dict_set(rdv_points, rdv->name, rdv, NULL);
}
return rdv;
}
void SIMIX_rdv_free(void *data)
{
smx_rdv_t rdv = (smx_rdv_t) data;
- if (rdv->name)
- xbt_free(rdv->name);
+ xbt_free(rdv->name);
xbt_fifo_free(rdv->comm_fifo);
xbt_free(rdv);
}
xbt_fifo_item_t item;
void* req_data = NULL;
- xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){
+ xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t) {
if (action->comm.type == SIMIX_COMM_SEND) {
req_data = action->comm.src_data;
} else if (action->comm.type == SIMIX_COMM_RECEIVE) {
smx_action_t action;
xbt_fifo_item_t item;
- xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t){
+ xbt_fifo_foreach(rdv->comm_fifo, item, action, smx_action_t) {
if (action->comm.type == SIMIX_COMM_RECEIVE
&& (!match_fun || match_fun(data, action->comm.dst_data))) {
XBT_DEBUG("Found a matching communication action %p", action);
/* alloc structures */
act = xbt_mallocator_get(simix_global->action_mallocator);
+
act->type = SIMIX_ACTION_COMMUNICATE;
act->state = SIMIX_WAITING;
*/
void SIMIX_comm_destroy(smx_action_t action)
{
- XBT_DEBUG("Destroy action %p (refcount:%d)", action, action->comm.refcount);
+ XBT_DEBUG("Destroy action %p (refcount: %d), state: %d",
+ action, action->comm.refcount, action->state);
- if (action->comm.refcount <= 0)
+ if (action->comm.refcount <= 0) {
+ xbt_backtrace_display_current();
xbt_die("the refcount of comm %p is already 0 before decreasing it. "
"That's a bug!", action);
-
+ }
action->comm.refcount--;
if (action->comm.refcount > 0)
return;
if (action->comm.detached && action->state != SIMIX_DONE) {
/* the communication has failed and was detached:
* we have to free the buffer */
- ((void_f_pvoid_t) action->comm.src_data)(action->comm.src_buff);
+ action->comm.clean_fun(action->comm.src_buff);
+ action->comm.src_buff = NULL;
}
xbt_mallocator_release(simix_global->action_mallocator, action);
smx_action_t SIMIX_comm_isend(smx_process_t src_proc, smx_rdv_t rdv,
double task_size, double rate,
void *src_buff, size_t src_buff_size,
- int (*match_fun)(void *, void *), void *data,
+ int (*match_fun)(void *, void *),
+ void (*clean_fun)(void *), // used to free the action in case of problem after a detached send
+ void *data,
int detached)
{
smx_action_t action;
action->state = SIMIX_READY;
action->comm.type = SIMIX_COMM_READY;
}
+ xbt_fifo_push(src_proc->comms, action);
- /* If the communication action is detached then decrease the refcount
- * by one, so it will be eliminated by the receivers destroy call */
+ /* if the communication action is detached then decrease the refcount
+ * by one, so it will be eliminated by the receiver's destroy call */
if (detached) {
action->comm.detached = 1;
action->comm.refcount--;
+ action->comm.clean_fun = clean_fun;
+ } else {
+ action->comm.clean_fun = NULL;
}
/* Setup the communication request */
}
SIMIX_comm_start(action);
- return action;
+ return (detached ? NULL : action);
}
smx_action_t SIMIX_comm_irecv(smx_process_t dst_proc, smx_rdv_t rdv,
action->state = SIMIX_READY;
action->comm.type = SIMIX_COMM_READY;
}
+ xbt_fifo_push(dst_proc->comms, action);
/* Setup communication request */
action->comm.dst_proc = dst_proc;
void SIMIX_pre_comm_wait(smx_req_t req, smx_action_t action, double timeout, int idx)
{
+
/* the request may be a wait, a send or a recv */
surf_action_t sleep;
SIMIX_comm_finish(action);
return;
}
+ XBT_INFO("Comm_wait. state:%d; I'm %s",action->state,
+ req->issuer == action->comm.src_proc?"sender":"receiver");
/* If the action has already finish perform the error handling, */
/* otherwise set up a waiting timeout on the right side */
if (action->state != SIMIX_WAITING && action->state != SIMIX_RUNNING) {
SIMIX_comm_finish(action);
} else { /* if (timeout >= 0) { we need a surf sleep action even when there is no timeout, otherwise surf won't tell us when the host fails */
+ XBT_INFO("Not done, we need a sleep action");
sleep = surf_workstation_model->extension.workstation.sleep(req->issuer->smx_host->host, timeout);
surf_workstation_model->action_data_set(sleep, action);
{
/* If both the sender and the receiver are already there, start the communication */
if (action->state == SIMIX_READY) {
+
smx_host_t sender = action->comm.src_proc->smx_host;
smx_host_t receiver = action->comm.dst_proc->smx_host;
}
}
+/**
+ * \brief Answers the SIMIX requests associated to a communication action.
+ * \param action a finished communication action
+ */
void SIMIX_comm_finish(smx_action_t action)
{
- unsigned int destroy_count = 0;
+ volatile unsigned int destroy_count = 0;
smx_req_t req;
while ((req = xbt_fifo_shift(action->request_list))) {
case SIMIX_LINK_FAILURE:
TRY {
- XBT_DEBUG("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p)",
- action, action->comm.src_proc->smx_host->name, action->comm.dst_proc->smx_host->name,
- req->issuer->name, req->issuer);
+ XBT_INFO("Link failure in action %p between '%s' and '%s': posting an exception to the issuer: %s (%p) detached:%d",
+ action,
+ action->comm.src_proc ? action->comm.src_proc->smx_host->name : NULL,
+ action->comm.dst_proc ? action->comm.dst_proc->smx_host->name : NULL,
+ req->issuer->name, req->issuer,action->comm.detached);
+ if (action->comm.src_proc == req->issuer) {
+ XBT_INFO("I'm source");
+ } else if (action->comm.dst_proc == req->issuer) {
+ XBT_INFO("I'm dest");
+ } else {
+ XBT_INFO("I'm neither source nor dest");
+ }
THROWF(network_error, 0, "Link failure");
}
CATCH(req->issuer->running_ctx->exception) {
}
break;
+ case SIMIX_CANCELED:
+ TRY {
+ if (req->issuer == action->comm.dst_proc) {
+ THROWF(cancel_error, 0, "Communication canceled by the sender");
+ }
+ else {
+ THROWF(cancel_error, 0, "Communication canceled by the receiver");
+ }
+ }
+ CATCH(req->issuer->running_ctx->exception) {
+ req->issuer->doexception = 1;
+ }
+ break;
+
default:
- THROW_IMPOSSIBLE;
+ xbt_die("Unexpected action state in SIMIX_comm_finish: %d", action->state);
}
/* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
}
req->issuer->waiting_action = NULL;
+ xbt_fifo_remove(req->issuer->comms, action);
SIMIX_request_answer(req);
destroy_count++;
}
SIMIX_comm_destroy(action);
}
+/**
+ * \brief This function is called when a Surf communication action is finished.
+ * \param action the corresponding Simix communication
+ */
void SIMIX_post_comm(smx_action_t action)
{
/* Update action state */
surf_workstation_model->action_state_get(action->comm.dst_timeout) == SURF_ACTION_FAILED)
action->state = SIMIX_DST_HOST_FAILURE;
else if (action->comm.surf_comm &&
- surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED)
+ surf_workstation_model->action_state_get(action->comm.surf_comm) == SURF_ACTION_FAILED) {
+ XBT_INFO("Puta madre. Surf says that the link broke");
action->state = SIMIX_LINK_FAILURE;
- else
+ } else
action->state = SIMIX_DONE;
- XBT_DEBUG("SIMIX_post_comm: action state = %d", action->state);
+ XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d",
+ action, action->state, action->comm.src_proc, action->comm.dst_proc, action->comm.detached);
- /* After this point the surf actions associated with the simix communicate
- action are no longer needed, thus we delete them. */
+ /* destroy the surf actions associated with the Simix communication */
SIMIX_comm_destroy_internal_actions(action);
- /* If there are requests associated with the action, then answer them */
- if (xbt_fifo_size(action->request_list))
+ /* remove the communication action from the list of pending communications
+ * of both processes (if they still exist) */
+ if (action->comm.src_proc) {
+ xbt_fifo_remove(action->comm.src_proc->comms, action);
+ }
+ if (action->comm.dst_proc) {
+ xbt_fifo_remove(action->comm.dst_proc->comms, action);
+ }
+
+ /* if there are requests associated with the action, then answer them */
+ if (xbt_fifo_size(action->request_list)) {
SIMIX_comm_finish(action);
+ }
}
void SIMIX_comm_cancel(smx_action_t action)
{
- /* If the action is a waiting state means that it is still in a rdv */
+ /* if the action is a waiting state means that it is still in a rdv */
/* so remove from it and delete it */
if (action->state == SIMIX_WAITING) {
SIMIX_rdv_remove(action->comm.rdv, action);
- action->state = SIMIX_FAILED;
- } else {
- /* When running the MC there are no surf actions */
- if(!MC_IS_ENABLED)
- surf_workstation_model->action_cancel(action->comm.surf_comm);
+ action->state = SIMIX_CANCELED;
+ }
+ else if (!MC_IS_ENABLED /* when running the MC there are no surf actions */
+ && (action->state == SIMIX_READY || action->state == SIMIX_RUNNING)) {
+
+ surf_workstation_model->action_cancel(action->comm.surf_comm);
}
}
memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size);
}
+void smpi_comm_copy_data_callback(smx_action_t comm, size_t buff_size)
+{
+ XBT_INFO("Copy the data over");
+ memcpy(comm->comm.dst_buff, comm->comm.src_buff, buff_size);
+ if (comm->comm.detached) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the original buffer available to the application ASAP
+ comm->comm.clean_fun(comm->comm.src_buff);
+ comm->comm.src_buff = NULL;
+ }
+}
+
/**
* \brief Copy the communication data from the sender's buffer to the receiver's one
* \param comm The communication
if (comm->comm.dst_buff_size)
*comm->comm.dst_buff_size = buff_size;
- if (buff_size == 0)
- return;
-
- (*SIMIX_comm_copy_data_callback) (comm, buff_size);
+ if (buff_size > 0)
+ SIMIX_comm_copy_data_callback (comm, buff_size);
/* Set the copied flag so we copy data only once */
/* (this function might be called from both communication ends) */