1
1
openmpi/ompi/mca/osc/pt2pt/osc_pt2pt_frag.h
Nathan Hjelm 80ed805a16 osc/pt2pt: fix synchronization bugs
The fragment flush code tries to send the active fragment before
sending any queued fragments. This could cause osc messages to arrive
out-of-order at the target (bad). Ensure ordering by alway sending
the active fragment after sending queued fragments.

This commit also fixes a bug when a synchronization message (unlock,
flush, complete) can not be packed at the end of an existing active
fragment. In this case the source process will end up sending 1 more
fragment than claimed in the synchronization message. To fix the issue
a check has been added that fixes the fragment count if this situation
is detected.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2015-04-06 08:39:19 -06:00

144 строки
4.5 KiB
C

/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2012 Sandia National Laboratories. All rights reserved.
* Copyright (c) 2014-2015 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#ifndef OSC_PT2PT_FRAG_H
#define OSC_PT2PT_FRAG_H
#include "ompi/communicator/communicator.h"
#include "osc_pt2pt_header.h"
#include "osc_pt2pt_request.h"
#include "opal/align.h"
/** Communication buffer for packing messages */
struct ompi_osc_pt2pt_frag_t {
opal_free_list_item_t super;
/* target rank of buffer */
int target;
unsigned char *buffer;
/* space remaining in buffer */
size_t remain_len;
/* start of unused space */
char *top;
/* Number of operations which have started writing into the frag, but not yet completed doing so */
int32_t pending;
ompi_osc_pt2pt_frag_header_t *header;
ompi_osc_pt2pt_module_t *module;
};
typedef struct ompi_osc_pt2pt_frag_t ompi_osc_pt2pt_frag_t;
OBJ_CLASS_DECLARATION(ompi_osc_pt2pt_frag_t);
extern int ompi_osc_pt2pt_frag_start(ompi_osc_pt2pt_module_t *module, ompi_osc_pt2pt_frag_t *buffer);
extern int ompi_osc_pt2pt_frag_flush_target(ompi_osc_pt2pt_module_t *module, int target);
extern int ompi_osc_pt2pt_frag_flush_all(ompi_osc_pt2pt_module_t *module);
/*
* Note: module lock must be held during this operation
*/
static inline int ompi_osc_pt2pt_frag_alloc (ompi_osc_pt2pt_module_t *module, int target,
size_t request_len, ompi_osc_pt2pt_frag_t **buffer,
char **ptr)
{
ompi_osc_pt2pt_frag_t *curr;
int ret;
/* osc pt2pt headers can have 64-bit values. these will need to be aligned
* on an 8-byte boundary on some architectures so we up align the allocation
* size here. */
request_len = OPAL_ALIGN(request_len, 8, size_t);
if (request_len > mca_osc_pt2pt_component.buffer_size) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
OPAL_THREAD_LOCK(&module->lock);
curr = module->peers[target].active_frag;
if (NULL == curr || curr->remain_len < request_len) {
opal_free_list_item_t *item = NULL;
if (NULL != curr) {
curr->remain_len = 0;
module->peers[target].active_frag = NULL;
opal_atomic_mb ();
/* If there's something pending, the pending finish will
start the buffer. Otherwise, we need to start it now. */
if (0 == OPAL_THREAD_ADD32(&curr->pending, -1)) {
ret = ompi_osc_pt2pt_frag_start(module, curr);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
return ret;
}
}
}
item = opal_free_list_get (&mca_osc_pt2pt_component.frags);
if (OPAL_UNLIKELY(NULL == item)) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
curr = module->peers[target].active_frag =
(ompi_osc_pt2pt_frag_t*) item;
curr->target = target;
curr->header = (ompi_osc_pt2pt_frag_header_t*) curr->buffer;
curr->top = (char*) (curr->header + 1);
curr->remain_len = mca_osc_pt2pt_component.buffer_size;
curr->module = module;
curr->pending = 1;
curr->header->base.type = OMPI_OSC_PT2PT_HDR_TYPE_FRAG;
curr->header->base.flags = OMPI_OSC_PT2PT_HDR_FLAG_VALID;
if (module->passive_target_access_epoch) {
curr->header->base.flags |= OMPI_OSC_PT2PT_HDR_FLAG_PASSIVE_TARGET;
}
curr->header->source = ompi_comm_rank(module->comm);
curr->header->num_ops = 0;
curr->header->windx = ompi_comm_get_cid(module->comm);
if (curr->remain_len < request_len) {
OPAL_THREAD_UNLOCK(&module->lock);
return OMPI_ERR_TEMP_OUT_OF_RESOURCE;
}
}
*ptr = curr->top;
*buffer = curr;
curr->top += request_len;
curr->remain_len -= request_len;
OPAL_THREAD_UNLOCK(&module->lock);
OPAL_THREAD_ADD32(&curr->pending, 1);
OPAL_THREAD_ADD32(&curr->header->num_ops, 1);
return OMPI_SUCCESS;
}
/*
* Note: module lock must be held for this operation
*/
static inline int ompi_osc_pt2pt_frag_finish(ompi_osc_pt2pt_module_t *module,
ompi_osc_pt2pt_frag_t* buffer)
{
if (0 == OPAL_THREAD_ADD32(&buffer->pending, -1)) {
return ompi_osc_pt2pt_frag_start(module, buffer);
}
return OMPI_SUCCESS;
}
#endif