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

Convert to the GM fragment header (a normal frag header + 64 bits to hold the registered memory).

Просмотр исходного кода 
This commit was SVN r3930.
Этот коммит содержится в:
George Bosilca 2005-01-10 08:43:51 +00:00
родитель 884c54decc
Коммит 7cf3f77b43
2 изменённых файлов: 33 добавлений и 32 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

53
src/mca/ptl/gm/src/ptl_gm_priv.c
Просмотреть файл

@ -86,12 +86,11 @@ int mca_ptl_gm_send_next_long_segment( mca_ptl_gm_send_frag_t* frag,
int flags )
{
struct mca_ptl_gm_peer_t* ptl_peer;
ompi_ptr_t* local_address;
gm_status_t status;
char* pointer;
uint64_t length;
int32_t hdr_flags = 0;
mca_ptl_base_header_t* hdr;
mca_ptl_gm_frag_header_t* hdr;
ptl_peer = (struct mca_ptl_gm_peer_t*)frag->send_frag.frag_base.frag_peer;
@ -108,22 +107,20 @@ int mca_ptl_gm_send_next_long_segment( mca_ptl_gm_send_frag_t* frag,
int32_t rc;
OMPI_FREE_LIST_GET( &(ptl_peer->peer_ptl->gm_send_dma_frags), item, rc );
hdr = (mca_ptl_base_header_t*)item;
hdr = (mca_ptl_gm_frag_header_t*)item;
hdr->hdr_frag.hdr_common.hdr_type = MCA_PTL_HDR_TYPE_FRAG;
hdr->hdr_frag.hdr_common.hdr_type = MCA_PTL_HDR_TYPE_FRAG;
hdr->hdr_frag.hdr_common.hdr_flags = frag->send_frag.frag_base.frag_header.hdr_common.hdr_flags | hdr_flags;
hdr->hdr_frag.hdr_src_ptr.lval = 0L; /* for VALGRIND/PURIFY - REPLACE WITH MACRO */
hdr->hdr_frag.hdr_src_ptr.pval = frag;
hdr->hdr_frag.hdr_dst_ptr = frag->send_frag.frag_base.frag_header.hdr_ack.hdr_dst_match;
hdr->hdr_frag.hdr_frag_offset = frag->frag_bytes_processed;
hdr->hdr_frag.hdr_frag_length = length;
local_address = (ompi_ptr_t*)((char*)hdr + sizeof(mca_ptl_base_frag_header_t));
local_address->lval = 0;
local_address->pval = pointer;
hdr->hdr_frag.hdr_src_ptr.lval = 0L; /* for VALGRIND/PURIFY - REPLACE WITH MACRO */
hdr->hdr_frag.hdr_src_ptr.pval = frag;
hdr->hdr_frag.hdr_dst_ptr = frag->send_frag.frag_base.frag_header.hdr_ack.hdr_dst_match;
hdr->hdr_frag.hdr_frag_offset = frag->frag_bytes_processed;
hdr->hdr_frag.hdr_frag_length = length;
hdr->registered_memory.lval = 0L;
hdr->registered_memory.pval = pointer;
gm_send_to_peer_with_callback( ptl_peer->peer_ptl->gm_port, hdr,
GM_SIZE, sizeof(mca_ptl_base_frag_header_t) + sizeof(ompi_ptr_t),
GM_SIZE, sizeof(mca_ptl_gm_frag_header_t),
GM_LOW_PRIORITY, ptl_peer->local_id,
send_continue_callback, (void*)hdr );
frag->frag_bytes_processed += length;
@ -152,10 +149,9 @@ int mca_ptl_gm_peer_send_continue( mca_ptl_gm_peer_t *ptl_peer,
size_t *size,
int flags )
{
mca_ptl_base_header_t* hdr;
mca_ptl_gm_frag_header_t* hdr;
uint64_t remaining_bytes;
ompi_list_item_t *item;
ompi_ptr_t* local_address;
int rc = 0;
fragment->frag_offset = offset;
@ -168,7 +164,7 @@ int mca_ptl_gm_peer_send_continue( mca_ptl_gm_peer_t *ptl_peer,
/* The first DMA memory buffer has been alocated in same time as the fragment */
item = (ompi_list_item_t*)fragment->send_buf;
hdr = (mca_ptl_base_header_t*)item;
hdr = (mca_ptl_gm_frag_header_t*)item;
remaining_bytes = fragment->send_frag.frag_base.frag_size - fragment->frag_bytes_processed;
if( remaining_bytes <= mca_ptl_gm_component.gm_eager_limit ) { /* small protocol */
@ -182,7 +178,7 @@ int mca_ptl_gm_peer_send_continue( mca_ptl_gm_peer_t *ptl_peer,
if( NULL == item ) {
OMPI_FREE_LIST_WAIT( &(ptl_peer->peer_ptl->gm_send_dma_frags), item, rc );
ompi_atomic_sub( &(ptl_peer->peer_ptl->num_send_tokens), 1 );
hdr = (mca_ptl_base_header_t*)item;
hdr = (mca_ptl_gm_frag_header_t*)item;
}
iov.iov_base = (char*)item + sizeof(mca_ptl_base_frag_header_t);
iov.iov_len = mca_ptl_gm_component.gm_segment_size - sizeof(mca_ptl_base_frag_header_t);
@ -205,7 +201,7 @@ int mca_ptl_gm_peer_send_continue( mca_ptl_gm_peer_t *ptl_peer,
fragment->frag_bytes_processed += iov.iov_len;
remaining_bytes -= iov.iov_len;
if( remaining_bytes == 0 )
hdr->hdr_common.hdr_flags |= PTL_FLAG_GM_LAST_FRAGMENT;
hdr->hdr_frag.hdr_common.hdr_flags |= PTL_FLAG_GM_LAST_FRAGMENT;
/* for the last piece set the header type to FIN */
gm_send_to_peer_with_callback( ptl_peer->peer_ptl->gm_port, hdr,
@ -234,10 +230,8 @@ int mca_ptl_gm_peer_send_continue( mca_ptl_gm_peer_t *ptl_peer,
hdr->hdr_frag.hdr_dst_ptr = sendreq->req_peer_match;
hdr->hdr_frag.hdr_frag_offset = fragment->frag_offset;
hdr->hdr_frag.hdr_frag_length = *size;
local_address = (ompi_ptr_t*)((char*)hdr + sizeof(mca_ptl_base_frag_header_t));
local_address->lval = 0;
local_address->pval = NULL;
hdr->registered_memory.lval = 0L;
hdr->registered_memory.pval = NULL;
gm_send_to_peer_with_callback( ptl_peer->peer_ptl->gm_port, hdr,
GM_SIZE, sizeof(mca_ptl_base_frag_header_t) + sizeof(ompi_ptr_t),
@ -648,8 +642,8 @@ static void mca_ptl_gm_get_callback( struct gm_port *port, void * context, gm_st
}
static mca_ptl_gm_recv_frag_t*
mca_ptl_gm_recv_frag_frag( struct mca_ptl_gm_module_t *ptl,
mca_ptl_base_header_t *hdr )
mca_ptl_gm_recv_frag_frag( struct mca_ptl_gm_module_t* ptl,
mca_ptl_gm_frag_header_t* hdr )
{
mca_pml_base_recv_request_t *request;
ompi_convertor_t local_convertor, *convertor;
@ -658,7 +652,7 @@ mca_ptl_gm_recv_frag_frag( struct mca_ptl_gm_module_t *ptl,
int32_t freeAfter, rc;
mca_ptl_gm_recv_frag_t* recv_frag;
if( hdr->hdr_common.hdr_flags & PTL_FLAG_GM_HAS_FRAGMENT ) {
if( hdr->hdr_frag.hdr_common.hdr_flags & PTL_FLAG_GM_HAS_FRAGMENT ) {
recv_frag = (mca_ptl_gm_recv_frag_t*)hdr->hdr_frag.hdr_dst_ptr.pval;
request = (mca_pml_base_recv_request_t*)recv_frag->frag_recv.frag_request;
/* here we can have a synchronisation problem if several threads work in same time
@ -705,13 +699,12 @@ mca_ptl_gm_recv_frag_frag( struct mca_ptl_gm_module_t *ptl,
ptl->super.ptl_recv_progress( (mca_ptl_base_module_t*)ptl, request, max_data, max_data );
} else {
gm_status_t status;
ompi_ptr_t* remote_memory = (ompi_ptr_t*)((char*)hdr + sizeof(mca_ptl_base_frag_header_t));
mca_ptl_gm_peer_t* peer = (mca_ptl_gm_peer_t*)recv_frag->frag_recv.frag_base.frag_peer;
char* pointer = (char*)request->req_base.req_addr + recv_frag->frag_offset;
uint64_t length = mca_ptl_gm_component.gm_rdma_frag_size;
recv_frag->frag_recv.frag_base.frag_header.hdr_frag = hdr->hdr_frag;
if( NULL == remote_memory->pval ) { /* first round of the local rendez-vous protocol */
if( NULL == hdr->registered_memory.pval ) { /* first round of the local rendez-vous protocol */
/* send an ack message to the sender ... quick hack (TODO) */
recv_frag->frag_recv.frag_base.frag_header.hdr_frag.hdr_frag_length = 0;
mca_ptl_gm_send_quick_fin_message( (mca_ptl_gm_peer_t*)recv_frag->frag_recv.frag_base.frag_peer,
@ -727,7 +720,7 @@ mca_ptl_gm_recv_frag_frag( struct mca_ptl_gm_module_t *ptl,
* get a fragment with the remote_memory field not NULL it can start getting the data.
*/
pointer += hdr->hdr_frag.hdr_frag_offset;
gm_get( ptl->gm_port, remote_memory->lval, pointer, hdr->hdr_frag.hdr_frag_length,
gm_get( ptl->gm_port, hdr->registered_memory.lval, pointer, hdr->hdr_frag.hdr_frag_length,
GM_LOW_PRIORITY, peer->local_id, peer->port_number, mca_ptl_gm_get_callback, recv_frag );
pointer += hdr->hdr_frag.hdr_frag_length;
length = recv_frag->frag_recv.frag_base.frag_size - recv_frag->frag_bytes_processed;
@ -832,7 +825,7 @@ frag_management_fct_t* frag_management_fct[MCA_PTL_HDR_TYPE_MAX] = {
NULL,
mca_ptl_gm_recv_frag_match,
mca_ptl_gm_recv_frag_match,
mca_ptl_gm_recv_frag_frag,
(frag_management_fct_t*)mca_ptl_gm_recv_frag_frag, /* force the conversion to remove a warning */
mca_ptl_gm_ctrl_frag,
mca_ptl_gm_ctrl_frag,
NULL,

12
src/mca/ptl/gm/src/ptl_gm_sendfrag.h
Просмотреть файл

@ -46,11 +46,19 @@ extern "C" {
/* specific header for GM rendezvous protocol. It will be filled up by the sender
* and should be ab;e to hold a pointer to the last registered memory location.
*/
struct mca_ptl_gm_rdv_header_t {
struct mca_ptl_gm_frag_header_t {
mca_ptl_base_frag_header_t hdr_frag;
ompi_ptr_t registered_memory;
};
typedef struct mca_ptl_gm_rdv_header_t mca_ptl_gm_rdv_header_t;
typedef struct mca_ptl_gm_frag_header_t mca_ptl_gm_frag_header_t;
struct mca_ptl_gm_pipeline_info_t {
uint64_t offset;
uint64_t length;
ompi_ptr_t local_memory;
ompi_ptr_t remote_memory;
};
typedef struct mca_ptl_gm_pipeline_info_t mca_ptl_gm_pipeline_info_t;
struct mca_ptl_gm_peer_t;