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

btl: change argument type of BTL receive callbacks

Просмотр исходного кода 
This commit updates the btl interface to change the parameters
passed to receive callbacks. The interface used to pass the tag,
a btl base descriptor, and the callback context. Most of the
values in the btl base descriptor were unused and only helped
simplify the callbacks from the self btl. All of the arguments
have now been replaced with a single receive callback descriptor.
This descriptor contains the incoming endpoint, data segment(s),
tag, and callback context. All btls have been updated to use
the new callback and the btl interface version has been bumped
to v3.2.0.

As part of this change the descriptor argument (and the segments
contained within it) have been marked as const. The were treated
as const before but this change could allow the compiler to make
better optimization decisions and will enforce that the callback
does not attempt to change the data in the descriptor.

Signed-off-by: Nathan Hjelm <hjelmn@google.com>
Этот коммит содержится в:
Nathan Hjelm 2020-06-22 15:43:19 -06:00
родитель dbc56758b6
Коммит 88f51fbb8e
22 изменённых файлов: 262 добавлений и 259 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

6
ompi/mca/pml/ob1/pml_ob1.h
Просмотреть файл

@ -311,8 +311,8 @@ void mca_pml_ob1_process_pending_rdma(void);
* Compute the total number of bytes on supplied descriptor
*/
static inline size_t
mca_pml_ob1_compute_segment_length_base(mca_btl_base_segment_t *segments,
size_t count, size_t hdrlen)
mca_pml_ob1_compute_segment_length_base (const mca_btl_base_segment_t *segments,
size_t count, size_t hdrlen)
{
size_t i, length = 0;
@ -323,7 +323,7 @@ mca_pml_ob1_compute_segment_length_base(mca_btl_base_segment_t *segments,
}
static inline size_t
mca_pml_ob1_compute_segment_length_remote (size_t seg_size, void *segments,
mca_pml_ob1_compute_segment_length_remote (size_t seg_size, const void *segments,
size_t count, ompi_proc_t *rem_proc)
{
mca_btl_base_segment_t *segment = (mca_btl_base_segment_t *) segments;

212
ompi/mca/pml/ob1/pml_ob1_recvfrag.c
Просмотреть файл

@ -19,6 +19,7 @@
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 Sandia National Laboratories
* All rights reserved.
* Copyright (c) 2020 Google, LLC. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -75,7 +76,7 @@ OBJ_CLASS_INSTANCE( mca_pml_ob1_recv_frag_t,
static void
append_frag_to_list(opal_list_t *queue, mca_btl_base_module_t *btl,
mca_pml_ob1_match_hdr_t *hdr, mca_btl_base_segment_t* segments,
const mca_pml_ob1_match_hdr_t *hdr, const mca_btl_base_segment_t *segments,
size_t num_segments, mca_pml_ob1_recv_frag_t* frag)
{
if(NULL == frag) {
@ -89,7 +90,7 @@ append_frag_to_list(opal_list_t *queue, mca_btl_base_module_t *btl,
static void
append_frag_to_umq(custom_match_umq *queue, mca_btl_base_module_t *btl,
mca_pml_ob1_match_hdr_t *hdr, mca_btl_base_segment_t* segments,
const mca_pml_ob1_match_hdr_t *hdr, const mca_btl_base_segment_t *segments,
size_t num_segments, mca_pml_ob1_recv_frag_t* frag)
{
if(NULL == frag) {
@ -113,20 +114,18 @@ append_frag_to_umq(custom_match_umq *queue, mca_btl_base_module_t *btl,
* messages. On the vertical layer, messages with contiguous sequence
* number organize themselves in a way to minimize the search space.
*/
void
append_frag_to_ordered_list(mca_pml_ob1_recv_frag_t** queue,
mca_pml_ob1_recv_frag_t *frag,
uint16_t seq)
void append_frag_to_ordered_list (mca_pml_ob1_recv_frag_t **queue,
mca_pml_ob1_recv_frag_t *frag,
uint16_t seq)
{
mca_pml_ob1_recv_frag_t *prior, *next;
mca_pml_ob1_match_hdr_t *hdr;
const mca_pml_ob1_match_hdr_t *hdr = &frag->hdr.hdr_match;
frag->super.super.opal_list_next = (opal_list_item_t*)frag;
frag->super.super.opal_list_prev = (opal_list_item_t*)frag;
frag->range = NULL;
hdr = &frag->hdr.hdr_match;
if( NULL == *queue ) { /* no pending fragments yet */
if (NULL == *queue) { /* no pending fragments yet */
*queue = frag;
return;
}
@ -292,11 +291,11 @@ remove_head_from_ordered_list(mca_pml_ob1_recv_frag_t** queue)
* @param type (IN) Type of the message header.
* @return OMPI_SUCCESS or error status on failure.
*/
static int mca_pml_ob1_recv_frag_match( mca_btl_base_module_t *btl,
mca_pml_ob1_match_hdr_t *hdr,
mca_btl_base_segment_t* segments,
static int mca_pml_ob1_recv_frag_match (mca_btl_base_module_t *btl,
const mca_pml_ob1_match_hdr_t *hdr,
const mca_btl_base_segment_t *segments,
size_t num_segments,
int type );
int type);
/**
* Match incoming frags against posted receives. If frag is not NULL then we assume
@ -312,24 +311,23 @@ static int mca_pml_ob1_recv_frag_match( mca_btl_base_module_t *btl,
* @return OMPI_SUCCESS or error status on failure.
*/
static int
mca_pml_ob1_recv_frag_match_proc( mca_btl_base_module_t *btl,
ompi_communicator_t* comm_ptr,
mca_pml_ob1_recv_frag_match_proc (mca_btl_base_module_t *btl,
ompi_communicator_t *comm_ptr,
mca_pml_ob1_comm_proc_t *proc,
mca_pml_ob1_match_hdr_t *hdr,
mca_btl_base_segment_t* segments,
const mca_pml_ob1_match_hdr_t *hdr,
const mca_btl_base_segment_t *segments,
size_t num_segments,
int type,
mca_pml_ob1_recv_frag_t* frag );
mca_pml_ob1_recv_frag_t *frag);
static mca_pml_ob1_recv_request_t*
match_one(mca_btl_base_module_t *btl,
mca_pml_ob1_match_hdr_t *hdr, mca_btl_base_segment_t* segments,
size_t num_segments, ompi_communicator_t *comm_ptr,
mca_pml_ob1_comm_proc_t *proc,
mca_pml_ob1_recv_frag_t* frag);
static mca_pml_ob1_recv_request_t *match_one (mca_btl_base_module_t *btl,
const mca_pml_ob1_match_hdr_t *hdr,
const mca_btl_base_segment_t *segments,
size_t num_segments, ompi_communicator_t *comm_ptr,
mca_pml_ob1_comm_proc_t *proc,
mca_pml_ob1_recv_frag_t *frag);
mca_pml_ob1_recv_frag_t*
check_cantmatch_for_match(mca_pml_ob1_comm_proc_t *proc)
mca_pml_ob1_recv_frag_t *check_cantmatch_for_match (mca_pml_ob1_comm_proc_t *proc)
{
mca_pml_ob1_recv_frag_t *frag = proc->frags_cant_match;
@ -339,23 +337,21 @@ check_cantmatch_for_match(mca_pml_ob1_comm_proc_t *proc)
return NULL;
}
void mca_pml_ob1_recv_frag_callback_match(mca_btl_base_module_t* btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* des,
void* cbdata )
void mca_pml_ob1_recv_frag_callback_match (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor)
{
mca_btl_base_segment_t* segments = des->des_segments;
mca_pml_ob1_match_hdr_t* hdr = (mca_pml_ob1_match_hdr_t*)segments->seg_addr.pval;
const mca_btl_base_segment_t *segments = descriptor->des_segments;
const mca_pml_ob1_match_hdr_t *hdr = (const mca_pml_ob1_match_hdr_t *) segments->seg_addr.pval;
ompi_communicator_t *comm_ptr;
mca_pml_ob1_recv_request_t *match = NULL;
mca_pml_ob1_comm_t *comm;
mca_pml_ob1_comm_proc_t *proc;
size_t num_segments = des->des_segment_count;
size_t num_segments = descriptor->des_segment_count;
size_t bytes_received = 0;
assert(num_segments <= MCA_BTL_DES_MAX_SEGMENTS);
if( OPAL_UNLIKELY(segments->seg_len < OMPI_PML_OB1_MATCH_HDR_LEN) ) {
if (OPAL_UNLIKELY(segments->seg_len < OMPI_PML_OB1_MATCH_HDR_LEN)) {
return;
}
ob1_hdr_ntoh(((mca_pml_ob1_hdr_t*) hdr), MCA_PML_OB1_HDR_TYPE_MATCH);
@ -513,63 +509,51 @@ void mca_pml_ob1_recv_frag_callback_match(mca_btl_base_module_t* btl,
OB1_MATCHING_UNLOCK(&comm->matching_lock);
}
}
return;
}
void mca_pml_ob1_recv_frag_callback_rndv(mca_btl_base_module_t* btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* des,
void* cbdata )
void mca_pml_ob1_recv_frag_callback_rndv (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor)
{
mca_btl_base_segment_t* segments = des->des_segments;
mca_pml_ob1_hdr_t* hdr = (mca_pml_ob1_hdr_t*)segments->seg_addr.pval;
const mca_btl_base_segment_t *segments = descriptor->des_segments;
const mca_pml_ob1_hdr_t *hdr = (mca_pml_ob1_hdr_t *) segments->seg_addr.pval;
if( OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_common_hdr_t)) ) {
return;
}
ob1_hdr_ntoh(hdr, MCA_PML_OB1_HDR_TYPE_RNDV);
mca_pml_ob1_recv_frag_match(btl, &hdr->hdr_match, segments,
des->des_segment_count, MCA_PML_OB1_HDR_TYPE_RNDV);
return;
descriptor->des_segment_count, MCA_PML_OB1_HDR_TYPE_RNDV);
}
void mca_pml_ob1_recv_frag_callback_rget(mca_btl_base_module_t* btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* des,
void* cbdata )
void mca_pml_ob1_recv_frag_callback_rget (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor)
{
mca_btl_base_segment_t* segments = des->des_segments;
mca_pml_ob1_hdr_t* hdr = (mca_pml_ob1_hdr_t*)segments->seg_addr.pval;
const mca_btl_base_segment_t *segments = descriptor->des_segments;
const mca_pml_ob1_hdr_t *hdr = (mca_pml_ob1_hdr_t *) segments->seg_addr.pval;
if( OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_common_hdr_t)) ) {
return;
}
ob1_hdr_ntoh(hdr, MCA_PML_OB1_HDR_TYPE_RGET);
mca_pml_ob1_recv_frag_match(btl, &hdr->hdr_match, segments,
des->des_segment_count, MCA_PML_OB1_HDR_TYPE_RGET);
return;
descriptor->des_segment_count, MCA_PML_OB1_HDR_TYPE_RGET);
}
void mca_pml_ob1_recv_frag_callback_ack(mca_btl_base_module_t* btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* des,
void* cbdata )
void mca_pml_ob1_recv_frag_callback_ack (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor)
{
mca_btl_base_segment_t* segments = des->des_segments;
mca_pml_ob1_hdr_t* hdr = (mca_pml_ob1_hdr_t*)segments->seg_addr.pval;
const mca_btl_base_segment_t *segments = descriptor->des_segments;
const mca_pml_ob1_hdr_t *hdr = (mca_pml_ob1_hdr_t *) segments->seg_addr.pval;
mca_pml_ob1_send_request_t* sendreq;
size_t size;
if( OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_common_hdr_t)) ) {
if (OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_common_hdr_t))) {
return;
}
ob1_hdr_ntoh(hdr, MCA_PML_OB1_HDR_TYPE_ACK);
sendreq = (mca_pml_ob1_send_request_t*)hdr->hdr_ack.hdr_src_req.pval;
sendreq = (mca_pml_ob1_send_request_t *) hdr->hdr_ack.hdr_src_req.pval;
sendreq->req_recv = hdr->hdr_ack.hdr_dst_req;
/* if the request should be delivered entirely by copy in/out
@ -615,21 +599,18 @@ void mca_pml_ob1_recv_frag_callback_ack(mca_btl_base_module_t* btl,
}
#endif /* OPAL_CUDA_SUPPORT */
if(send_request_pml_complete_check(sendreq) == false)
if (send_request_pml_complete_check(sendreq) == false)
mca_pml_ob1_send_request_schedule(sendreq);
return;
}
void mca_pml_ob1_recv_frag_callback_frag(mca_btl_base_module_t* btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* des,
void* cbdata ) {
mca_btl_base_segment_t* segments = des->des_segments;
mca_pml_ob1_hdr_t* hdr = (mca_pml_ob1_hdr_t*)segments->seg_addr.pval;
void mca_pml_ob1_recv_frag_callback_frag (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor)
{
const mca_btl_base_segment_t *segments = descriptor->des_segments;
const mca_pml_ob1_hdr_t *hdr = (mca_pml_ob1_hdr_t *) segments->seg_addr.pval;
mca_pml_ob1_recv_request_t* recvreq;
if( OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_common_hdr_t)) ) {
if (OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_common_hdr_t))) {
return;
}
@ -643,54 +624,47 @@ void mca_pml_ob1_recv_frag_callback_frag(mca_btl_base_module_t* btl,
assert(btl->btl_flags & MCA_BTL_FLAGS_CUDA_COPY_ASYNC_RECV);
/* This will trigger the opal_convertor_pack to start asynchronous copy. */
mca_pml_ob1_recv_request_frag_copy_start(recvreq,btl,segments,des->des_segment_count,des);
mca_pml_ob1_recv_request_frag_copy_start(recvreq,btl,segments,descriptor->des_segment_count,des);
/* Let BTL know that it CANNOT free the frag */
des->des_flags |= MCA_BTL_DES_FLAGS_CUDA_COPY_ASYNC;
descriptor->des_flags |= MCA_BTL_DES_FLAGS_CUDA_COPY_ASYNC;
return;
}
#endif /* OPAL_CUDA_SUPPORT */
mca_pml_ob1_recv_request_progress_frag(recvreq,btl,segments,des->des_segment_count);
return;
mca_pml_ob1_recv_request_progress_frag(recvreq,btl,segments,descriptor->des_segment_count);
}
void mca_pml_ob1_recv_frag_callback_put(mca_btl_base_module_t* btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* des,
void* cbdata ) {
mca_btl_base_segment_t* segments = des->des_segments;
mca_pml_ob1_hdr_t* hdr = (mca_pml_ob1_hdr_t*)segments->seg_addr.pval;
mca_pml_ob1_send_request_t* sendreq;
void mca_pml_ob1_recv_frag_callback_put (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor)
{
const mca_btl_base_segment_t *segments = descriptor->des_segments;
const mca_pml_ob1_hdr_t *hdr = (mca_pml_ob1_hdr_t *) segments->seg_addr.pval;
mca_pml_ob1_send_request_t *sendreq;
if( OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_common_hdr_t)) ) {
if (OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_common_hdr_t))) {
return;
}
ob1_hdr_ntoh(hdr, MCA_PML_OB1_HDR_TYPE_PUT);
sendreq = (mca_pml_ob1_send_request_t*)hdr->hdr_rdma.hdr_req.pval;
mca_pml_ob1_send_request_put(sendreq,btl,&hdr->hdr_rdma);
return;
ob1_hdr_ntoh (hdr, MCA_PML_OB1_HDR_TYPE_PUT);
sendreq = (mca_pml_ob1_send_request_t *) hdr->hdr_rdma.hdr_req.pval;
mca_pml_ob1_send_request_put (sendreq, btl, &hdr->hdr_rdma);
}
void mca_pml_ob1_recv_frag_callback_fin(mca_btl_base_module_t* btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* des,
void* cbdata ) {
mca_btl_base_segment_t* segments = des->des_segments;
mca_pml_ob1_fin_hdr_t* hdr = (mca_pml_ob1_fin_hdr_t *) segments->seg_addr.pval;
void mca_pml_ob1_recv_frag_callback_fin(mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor)
{
const mca_btl_base_segment_t *segments = descriptor->des_segments;
const mca_pml_ob1_fin_hdr_t *hdr = (mca_pml_ob1_fin_hdr_t *) segments->seg_addr.pval;
mca_pml_ob1_rdma_frag_t *frag;
if( OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_fin_hdr_t)) ) {
if (OPAL_UNLIKELY(segments->seg_len < sizeof(mca_pml_ob1_fin_hdr_t))) {
return;
}
ob1_hdr_ntoh((union mca_pml_ob1_hdr_t *)hdr, MCA_PML_OB1_HDR_TYPE_FIN);
ob1_hdr_ntoh ((union mca_pml_ob1_hdr_t *) hdr, MCA_PML_OB1_HDR_TYPE_FIN);
frag = (mca_pml_ob1_rdma_frag_t *) hdr->hdr_frag.pval;
frag->cbfunc (frag, hdr->hdr_size);
}
@ -719,9 +693,9 @@ static inline mca_pml_ob1_recv_request_t* get_next_posted_recv(
return (mca_pml_ob1_recv_request_t*)i;
}
static mca_pml_ob1_recv_request_t *match_incomming(
mca_pml_ob1_match_hdr_t *hdr, mca_pml_ob1_comm_t *comm,
mca_pml_ob1_comm_proc_t *proc)
static mca_pml_ob1_recv_request_t *match_incomming(const mca_pml_ob1_match_hdr_t *hdr,
mca_pml_ob1_comm_t *comm,
mca_pml_ob1_comm_proc_t *proc)
{
#if !MCA_PML_OB1_CUSTOM_MATCH
mca_pml_ob1_recv_request_t *specific_recv, *wild_recv;
@ -772,9 +746,9 @@ static mca_pml_ob1_recv_request_t *match_incomming(
}
#if !MCA_PML_OB1_CUSTOM_MATCH
static mca_pml_ob1_recv_request_t *match_incomming_no_any_source (
mca_pml_ob1_match_hdr_t *hdr, mca_pml_ob1_comm_t *comm,
mca_pml_ob1_comm_proc_t *proc)
static mca_pml_ob1_recv_request_t *match_incomming_no_any_source (const mca_pml_ob1_match_hdr_t *hdr,
mca_pml_ob1_comm_t *comm,
mca_pml_ob1_comm_proc_t *proc)
{
mca_pml_ob1_recv_request_t *recv_req;
int tag = hdr->hdr_tag;
@ -794,12 +768,12 @@ static mca_pml_ob1_recv_request_t *match_incomming_no_any_source (
}
#endif
static mca_pml_ob1_recv_request_t*
match_one(mca_btl_base_module_t *btl,
mca_pml_ob1_match_hdr_t *hdr, mca_btl_base_segment_t* segments,
size_t num_segments, ompi_communicator_t *comm_ptr,
mca_pml_ob1_comm_proc_t *proc,
mca_pml_ob1_recv_frag_t* frag)
static mca_pml_ob1_recv_request_t *match_one (mca_btl_base_module_t *btl,
const mca_pml_ob1_match_hdr_t *hdr,
const mca_btl_base_segment_t *segments,
size_t num_segments, ompi_communicator_t *comm_ptr,
mca_pml_ob1_comm_proc_t *proc,
mca_pml_ob1_recv_frag_t* frag)
{
#if SPC_ENABLE == 1
opal_timer_t timer = 0;
@ -905,9 +879,9 @@ match_one(mca_btl_base_module_t *btl,
* - fragments may be corrupt
* - this routine may be called simultaneously by more than one thread
*/
static int mca_pml_ob1_recv_frag_match( mca_btl_base_module_t *btl,
mca_pml_ob1_match_hdr_t *hdr,
mca_btl_base_segment_t* segments,
static int mca_pml_ob1_recv_frag_match (mca_btl_base_module_t *btl,
const mca_pml_ob1_match_hdr_t *hdr,
const mca_btl_base_segment_t *segments,
size_t num_segments,
int type)
{
@ -989,17 +963,17 @@ static int mca_pml_ob1_recv_frag_match( mca_btl_base_module_t *btl,
* ATTENTION: THIS FUNCTION MUST BE CALLED WITH COMMUNICATOR LOCK HELD.
* THE LOCK WILL BE RELEASED UPON RETURN. USE WITH CARE. */
static int
mca_pml_ob1_recv_frag_match_proc( mca_btl_base_module_t *btl,
mca_pml_ob1_recv_frag_match_proc (mca_btl_base_module_t *btl,
ompi_communicator_t* comm_ptr,
mca_pml_ob1_comm_proc_t *proc,
mca_pml_ob1_match_hdr_t *hdr,
mca_btl_base_segment_t* segments,
const mca_pml_ob1_match_hdr_t *hdr,
const mca_btl_base_segment_t *segments,
size_t num_segments,
int type,
mca_pml_ob1_recv_frag_t* frag )
mca_pml_ob1_recv_frag_t *frag)
{
/* local variables */
mca_pml_ob1_comm_t* comm = (mca_pml_ob1_comm_t *)comm_ptr->c_pml_comm;
mca_pml_ob1_comm_t *comm = (mca_pml_ob1_comm_t *)comm_ptr->c_pml_comm;
mca_pml_ob1_recv_request_t *match = NULL;
/* If we are here, this is the sequence number we were expecting,

46
ompi/mca/pml/ob1/pml_ob1_recvfrag.h
Просмотреть файл

@ -14,6 +14,7 @@
* Copyright (c) 2011 Sandia National Laboratories. All rights reserved.
* Copyright (c) 2012-2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2020 Google, LLC. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -113,60 +114,49 @@ do { \
* Callback from BTL on receipt of a recv_frag (match).
*/
extern void mca_pml_ob1_recv_frag_callback_match( mca_btl_base_module_t *btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* descriptor,
void* cbdata );
extern void mca_pml_ob1_recv_frag_callback_match (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor);
/**
* Callback from BTL on receipt of a recv_frag (rndv).
*/
extern void mca_pml_ob1_recv_frag_callback_rndv( mca_btl_base_module_t *btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* descriptor,
void* cbdata );
extern void mca_pml_ob1_recv_frag_callback_rndv (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor);
/**
* Callback from BTL on receipt of a recv_frag (rget).
*/
extern void mca_pml_ob1_recv_frag_callback_rget( mca_btl_base_module_t *btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* descriptor,
void* cbdata );
extern void mca_pml_ob1_recv_frag_callback_rget (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor);
/**
* Callback from BTL on receipt of a recv_frag (ack).
*/
extern void mca_pml_ob1_recv_frag_callback_ack( mca_btl_base_module_t *btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* descriptor,
void* cbdata );
extern void mca_pml_ob1_recv_frag_callback_ack (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor);
/**
* Callback from BTL on receipt of a recv_frag (frag).
*/
extern void mca_pml_ob1_recv_frag_callback_frag( mca_btl_base_module_t *btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* descriptor,
void* cbdata );
extern void mca_pml_ob1_recv_frag_callback_frag (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor);
/**
* Callback from BTL on receipt of a recv_frag (put).
*/
extern void mca_pml_ob1_recv_frag_callback_put( mca_btl_base_module_t *btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* descriptor,
void* cbdata );
extern void mca_pml_ob1_recv_frag_callback_put (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor);
/**
* Callback from BTL on receipt of a recv_frag (fin).
*/
extern void mca_pml_ob1_recv_frag_callback_fin( mca_btl_base_module_t *btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* descriptor,
void* cbdata );
extern void mca_pml_ob1_recv_frag_callback_fin (mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor);
/**
* Extract the next fragment from the cant_match ordered list. This fragment

11
ompi/mca/pml/ob1/pml_ob1_recvreq.c
Просмотреть файл

@ -20,6 +20,7 @@
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 Sandia National Laboratories
* All rights reserved.
* Copyright (c) 2020 Google, LLC. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -517,7 +518,7 @@ int mca_pml_ob1_recv_request_get_frag (mca_pml_ob1_rdma_frag_t *frag)
void mca_pml_ob1_recv_request_progress_frag( mca_pml_ob1_recv_request_t* recvreq,
mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments )
{
size_t bytes_received, data_offset = 0;
@ -652,7 +653,7 @@ void mca_pml_ob1_recv_request_frag_copy_finished( mca_btl_base_module_t* btl,
void mca_pml_ob1_recv_request_progress_rget( mca_pml_ob1_recv_request_t* recvreq,
mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments )
{
mca_pml_ob1_rget_hdr_t* hdr = (mca_pml_ob1_rget_hdr_t*)segments->seg_addr.pval;
@ -806,7 +807,7 @@ void mca_pml_ob1_recv_request_progress_rget( mca_pml_ob1_recv_request_t* recvreq
void mca_pml_ob1_recv_request_progress_rndv( mca_pml_ob1_recv_request_t* recvreq,
mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments )
{
size_t bytes_received = 0;
@ -876,7 +877,7 @@ void mca_pml_ob1_recv_request_progress_rndv( mca_pml_ob1_recv_request_t* recvreq
*/
void mca_pml_ob1_recv_request_progress_match( mca_pml_ob1_recv_request_t* recvreq,
mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments )
{
size_t bytes_received, data_offset = 0;
@ -932,7 +933,7 @@ void mca_pml_ob1_recv_request_progress_match( mca_pml_ob1_recv_request_t* recvre
void mca_pml_ob1_recv_request_matched_probe( mca_pml_ob1_recv_request_t* recvreq,
mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments )
{
size_t bytes_packed = 0;

15
ompi/mca/pml/ob1/pml_ob1_recvreq.h
Просмотреть файл

@ -15,6 +15,7 @@
* reserved.
* Copyright (c) 2014 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2020 Google, LLC. All rights reserved.
*
* $COPYRIGHT$
*
@ -239,7 +240,7 @@ static inline void prepare_recv_req_converter(mca_pml_ob1_recv_request_t *req)
recv_req_matched(request, hdr)
static inline void recv_req_matched(mca_pml_ob1_recv_request_t *req,
mca_pml_ob1_match_hdr_t *hdr)
const mca_pml_ob1_match_hdr_t *hdr)
{
req->req_recv.req_base.req_ompi.req_status.MPI_SOURCE = hdr->hdr_src;
req->req_recv.req_base.req_ompi.req_status.MPI_TAG = hdr->hdr_tag;
@ -278,7 +279,7 @@ do {
uint32_t iov_count = 0; \
size_t max_data = bytes_received; \
size_t n, offset = seg_offset; \
mca_btl_base_segment_t* segment = segments; \
const mca_btl_base_segment_t *segment = segments; \
\
for( n = 0; n < num_segments; n++, segment++ ) { \
if(offset >= segment->seg_len) { \
@ -314,7 +315,7 @@ do {
void mca_pml_ob1_recv_request_progress_match(
mca_pml_ob1_recv_request_t* req,
struct mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments);
/**
@ -324,7 +325,7 @@ void mca_pml_ob1_recv_request_progress_match(
void mca_pml_ob1_recv_request_progress_frag(
mca_pml_ob1_recv_request_t* req,
struct mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments);
#if OPAL_CUDA_SUPPORT
@ -347,7 +348,7 @@ void mca_pml_ob1_recv_request_frag_copy_finished(struct mca_btl_base_module_t* b
void mca_pml_ob1_recv_request_progress_rndv(
mca_pml_ob1_recv_request_t* req,
struct mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments);
/**
@ -357,7 +358,7 @@ void mca_pml_ob1_recv_request_progress_rndv(
void mca_pml_ob1_recv_request_progress_rget(
mca_pml_ob1_recv_request_t* req,
struct mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments);
/**
@ -367,7 +368,7 @@ void mca_pml_ob1_recv_request_progress_rget(
void mca_pml_ob1_recv_request_matched_probe(
mca_pml_ob1_recv_request_t* req,
struct mca_btl_base_module_t* btl,
mca_btl_base_segment_t* segments,
const mca_btl_base_segment_t* segments,
size_t num_segments);
/**

6
ompi/mca/pml/ob1/pml_ob1_sendreq.c
Просмотреть файл

@ -1216,9 +1216,9 @@ int mca_pml_ob1_send_request_put_frag( mca_pml_ob1_rdma_frag_t *frag )
* (3) Queue the RDMA put
*/
void mca_pml_ob1_send_request_put( mca_pml_ob1_send_request_t* sendreq,
mca_btl_base_module_t* btl,
mca_pml_ob1_rdma_hdr_t* hdr )
void mca_pml_ob1_send_request_put (mca_pml_ob1_send_request_t *sendreq,
mca_btl_base_module_t *btl,
const mca_pml_ob1_rdma_hdr_t *hdr)
{
mca_bml_base_endpoint_t *bml_endpoint = sendreq->req_endpoint;
mca_pml_ob1_rdma_frag_t* frag;

6
ompi/mca/pml/ob1/pml_ob1_sendreq.h
Просмотреть файл

@ -512,9 +512,9 @@ mca_pml_ob1_send_request_start( mca_pml_ob1_send_request_t* sendreq )
* Initiate a put scheduled by the receiver.
*/
void mca_pml_ob1_send_request_put( mca_pml_ob1_send_request_t* sendreq,
mca_btl_base_module_t* btl,
mca_pml_ob1_rdma_hdr_t* hdr );
void mca_pml_ob1_send_request_put (mca_pml_ob1_send_request_t *sendreq,
mca_btl_base_module_t *btl,
const mca_pml_ob1_rdma_hdr_t *hdr);
int mca_pml_ob1_send_request_put_frag(mca_pml_ob1_rdma_frag_t* frag);

57
opal/mca/btl/btl.h
Просмотреть файл

@ -450,13 +450,12 @@ typedef struct mca_btl_base_segment_t mca_btl_base_segment_t;
#define MCA_BTL_BASE_SEGMENT_NTOH(s)
#endif
/**
* A descriptor that holds the parameters to a send/put/get
* @brief Descriptor to hold outgoing send messages
*
* This descriptor holds the parameters to perform a send
* operation along w/ a callback routine that is called on
* completion of the request.
* Note: receive callbacks will store the incomming data segments in
* des_segments
*/
struct mca_btl_base_descriptor_t {
opal_free_list_item_t super;
mca_btl_base_segment_t *des_segments; /**< local segments */
@ -519,6 +518,28 @@ OPAL_DECLSPEC OBJ_CLASS_DECLARATION(mca_btl_base_descriptor_t);
*/
#define MCA_BTL_REG_HANDLE_MAX_SIZE 256
/**
* @brief Descriptor for incoming messages
*
* This descriptor holds the information on incoming messages. The
* data this descriptor describes is only valid during a callback.
*/
struct mca_btl_base_receive_descriptor_t {
/** Incoming endpoint. May be NULL if the endpoint is not readily
* available for this BTL. */
struct mca_btl_base_endpoint_t *endpoint;
/** Local segment data. Copy this data if it is needed after the
* receive callback returns. **/
const mca_btl_base_segment_t *des_segments;
/** Number of segment is des_segments */
size_t des_segment_count;
mca_btl_base_tag_t tag;
/** Callback data supplied at callback registration time. */
void *cbdata;
};
typedef struct mca_btl_base_receive_descriptor_t mca_btl_base_receive_descriptor_t;
/*
* BTL base header, stores the tag at a minimum
*/
@ -581,7 +602,7 @@ typedef int (*mca_btl_base_component_progress_fn_t)(void);
/**
* Callback function that is called asynchronously on receipt
* of data by the transport layer.
* Note that the the mca_btl_base_descriptor_t is only valid within the
* Note that the the mca_btl_base_receive_descriptor_t is only valid within the
* completion function, this implies that all data payload in the
* mca_btl_base_descriptor_t must be copied out within this callback or
* forfeited back to the BTL.
@ -589,17 +610,12 @@ typedef int (*mca_btl_base_component_progress_fn_t)(void);
* segments for all callbacks.
*
* @param[IN] btl BTL module
* @param[IN] tag The active message receive callback tag value
* @param[IN] descriptor The BTL descriptor (contains the receive payload)
* @param[IN] cbdata Opaque callback data
*/
typedef void (*mca_btl_base_module_recv_cb_fn_t)(
struct mca_btl_base_module_t* btl,
mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t* descriptor,
void* cbdata
);
struct mca_btl_base_module_t *btl,
const mca_btl_base_receive_descriptor_t *descriptor);
typedef struct mca_btl_active_message_callback_t {
mca_btl_base_module_recv_cb_fn_t cbfunc;
@ -1262,14 +1278,21 @@ struct mca_btl_base_module_t {
};
typedef struct mca_btl_base_module_t mca_btl_base_module_t;
#define MCA_BTL_BASE_MAJOR_VERSION 3
#define MCA_BTL_BASE_MINOR_VERSION 2
#define MCA_BTL_BASE_PATCH_VERSION 0
/*
* Macro for use in modules that are of type btl v3.1.0
* Macro for use in modules that are of type btl v3.2.0
*/
#define MCA_BTL_BASE_VERSION_3_1_0 \
OPAL_MCA_BASE_VERSION_2_1_0("btl", 3, 1, 0)
#define MCA_BTL_BASE_VERSION_3_2_0 \
OPAL_MCA_BASE_VERSION_2_1_0("btl", \
MCA_BTL_BASE_MAJOR_VERSION, \
MCA_BTL_BASE_MINOR_VERSION, \
MCA_BTL_BASE_PATCH_VERSION)
#define MCA_BTL_DEFAULT_VERSION(name) \
MCA_BTL_BASE_VERSION_3_1_0, \
MCA_BTL_BASE_VERSION_3_2_0, \
.mca_component_name = name, \
MCA_BASE_MAKE_VERSION(component, OPAL_MAJOR_VERSION, OPAL_MINOR_VERSION, \
OPAL_RELEASE_VERSION)
@ -1277,7 +1300,7 @@ typedef struct mca_btl_base_module_t mca_btl_base_module_t;
/**
* Convinience macro for detecting the BTL interface version.
*/
#define BTL_VERSION 310
#define BTL_VERSION 320
END_C_DECLS

18
opal/mca/btl/ofi/btl_ofi_frag.c
Просмотреть файл

@ -4,6 +4,7 @@
* Copyright (c) 2018 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2018 Intel Inc. All rights reserved
* Copyright (c) 2020 Google, LLC. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -138,17 +139,18 @@ int mca_btl_ofi_recv_frag (mca_btl_ofi_module_t *ofi_btl,
mca_btl_ofi_base_frag_t *frag)
{
int rc;
mca_btl_active_message_callback_t *reg;
mca_btl_active_message_callback_t *reg = mca_btl_base_active_message_trigger + frag->hdr.tag;
mca_btl_base_segment_t segment = {.seg_addr.pval = (void *)(frag + 1),
.seg_len = frag->hdr.len};
/* Tell PML where the payload is */
frag->base.des_segments = frag->segments;
frag->segments[0].seg_addr.pval = frag+1;
frag->segments[0].seg_len = frag->hdr.len;
frag->base.des_segment_count = 1;
mca_btl_base_receive_descriptor_t recv_desc = {.endpoint = endpoint,
.des_segments = &segment,
.des_segment_count = 1,
.tag = frag->hdr.tag,
.cbdata = reg->cbdata};
/* call the callback */
reg = mca_btl_base_active_message_trigger + frag->hdr.tag;
reg->cbfunc (&ofi_btl->super, frag->hdr.tag, &frag->base, reg->cbdata);
reg->cbfunc (&ofi_btl->super, &recv_desc);
mca_btl_ofi_frag_complete(frag, OPAL_SUCCESS);
/* repost the recv */

14
opal/mca/btl/portals4/btl_portals4_component.c
Просмотреть файл

@ -15,6 +15,7 @@
* Copyright (c) 2014 Bull SAS. All rights reserved.
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2020 Google, LLC. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -572,8 +573,8 @@ mca_btl_portals4_component_progress(void)
static ptl_event_t ev;
unsigned int which;
mca_btl_active_message_callback_t* reg;
mca_btl_base_segment_t seg[2];
mca_btl_base_descriptor_t btl_base_descriptor;
mca_btl_base_segment_t seg[1];
mca_btl_base_receive_descriptor_t recv_descriptor = {.des_segments = seg, .des_segment_count = 1};
while (true) {
ret = PtlEQPoll(mca_btl_portals4_component.eqs_h, mca_btl_portals4_component.num_btls, 0, &ev, &which);
@ -658,15 +659,18 @@ mca_btl_portals4_component_progress(void)
tag = (unsigned char) (ev.hdr_data);
btl_base_descriptor.des_segments = seg;
btl_base_descriptor.des_segment_count = 1;
seg[0].seg_addr.pval = ev.start;
seg[0].seg_len = ev.mlength;
reg = mca_btl_base_active_message_trigger + tag;
recv_descriptor.endpoint = NULL;
recv_descriptor.tag = tag;
recv_descriptor.cbdata = reg->cbdata;
OPAL_OUTPUT_VERBOSE((50, opal_btl_base_framework.framework_output,
"PTL_EVENT_PUT: tag=%x base_descriptor=%p cbfunc: %lx\n", tag, (void*)&btl_base_descriptor, (uint64_t)reg->cbfunc));
reg->cbfunc(&portals4_btl->super, tag, &btl_base_descriptor, reg->cbdata);
reg->cbfunc(&portals4_btl->super, &recv_descriptor);
goto done;
break;

11
opal/mca/btl/self/btl_self.c
Просмотреть файл

@ -15,6 +15,7 @@
* reserved.
* Copyright (c) 2016 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2020 Google, LLC. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -204,12 +205,16 @@ static int mca_btl_self_send (struct mca_btl_base_module_t *btl,
struct mca_btl_base_descriptor_t *des,
mca_btl_base_tag_t tag)
{
mca_btl_active_message_callback_t* reg;
mca_btl_active_message_callback_t *reg = mca_btl_base_active_message_trigger + tag;
mca_btl_base_receive_descriptor_t recv_desc = {.endpoint = endpoint,
.des_segments = des->des_segments,
.des_segment_count = des->des_segment_count,
.tag = tag,
.cbdata = reg->cbdata};
int btl_ownership = (des->des_flags & MCA_BTL_DES_FLAGS_BTL_OWNERSHIP);
/* upcall */
reg = mca_btl_base_active_message_trigger + tag;
reg->cbfunc( btl, tag, des, reg->cbdata );
reg->cbfunc (btl, &recv_desc);
/* send completion */
if( des->des_flags & MCA_BTL_DES_SEND_ALWAYS_CALLBACK ) {

16
opal/mca/btl/sm/btl_sm_component.c
Просмотреть файл

@ -634,18 +634,16 @@ static mca_btl_base_module_t **mca_btl_sm_component_init (int *num_btls,
void mca_btl_sm_poll_handle_frag (mca_btl_sm_hdr_t *hdr, struct mca_btl_base_endpoint_t *endpoint)
{
mca_btl_base_segment_t segments[2];
mca_btl_base_descriptor_t frag = {.des_segments = segments, .des_segment_count = 1};
const mca_btl_active_message_callback_t *reg;
if (hdr->flags & MCA_BTL_SM_FLAG_COMPLETE) {
mca_btl_sm_frag_complete (hdr->frag);
return;
}
reg = mca_btl_base_active_message_trigger + hdr->tag;
segments[0].seg_addr.pval = (void *) (hdr + 1);
segments[0].seg_len = hdr->len;
const mca_btl_active_message_callback_t *reg = mca_btl_base_active_message_trigger + hdr->tag;
mca_btl_base_segment_t segments[2] = {[0] = {.seg_addr.pval = (void *) (hdr + 1), .seg_len = hdr->len}};
mca_btl_base_receive_descriptor_t frag = {.endpoint = endpoint, .des_segments = segments,
.des_segment_count = 1, .tag = hdr->frag,
.cbdata = reg->cbdata};
if (hdr->flags & MCA_BTL_SM_FLAG_SINGLE_COPY) {
mca_rcache_base_registration_t *xpmem_reg;
@ -659,10 +657,10 @@ void mca_btl_sm_poll_handle_frag (mca_btl_sm_hdr_t *hdr, struct mca_btl_base_end
frag.des_segment_count = 2;
/* recv upcall */
reg->cbfunc(&mca_btl_sm.super, hdr->tag, &frag, reg->cbdata);
reg->cbfunc(&mca_btl_sm.super, &frag);
sm_return_registration (xpmem_reg, endpoint);
} else {
reg->cbfunc(&mca_btl_sm.super, hdr->tag, &frag, reg->cbdata);
reg->cbfunc(&mca_btl_sm.super, &frag);
}
if (OPAL_UNLIKELY(MCA_BTL_SM_FLAG_SETUP_FBOX & hdr->flags)) {

6
opal/mca/btl/sm/btl_sm_fbox.h
Просмотреть файл

@ -211,9 +211,11 @@ static inline bool mca_btl_sm_check_fboxes (void)
/* the 0xff tag indicates we should skip the rest of the buffer */
if (OPAL_LIKELY((0xfe & hdr.data.tag) != 0xfe)) {
mca_btl_base_segment_t segment;
mca_btl_base_descriptor_t desc = {.des_segments = &segment, .des_segment_count = 1};
const mca_btl_active_message_callback_t *reg =
mca_btl_base_active_message_trigger + hdr.data.tag;
mca_btl_base_receive_descriptor_t desc = {.endpoint = ep, .des_segments = &segment,
.des_segment_count = 1, .tag = hdr.data.tag,
.cbdata = reg->cbdata};
/* fragment fits entirely in the remaining buffer space. some
* btl users do not handle fragmented data so we can't split
@ -224,7 +226,7 @@ static inline bool mca_btl_sm_check_fboxes (void)
segment.seg_addr.pval = (void *) (ep->fbox_in.buffer + start + sizeof (hdr));
/* call the registered callback function */
reg->cbfunc(&mca_btl_sm.super, hdr.data.tag, &desc, reg->cbdata);
reg->cbfunc(&mca_btl_sm.super, &desc);
} else if (OPAL_LIKELY(0xfe == hdr.data.tag)) {
/* process fragment header */
fifo_value_t *value = (fifo_value_t *)(ep->fbox_in.buffer + start + sizeof (hdr));

8
opal/mca/btl/tcp/btl_tcp.h
Просмотреть файл

@ -68,14 +68,6 @@ extern opal_event_base_t* mca_btl_tcp_event_base;
MCA_BTL_TCP_FRAG_RETURN(frag); \
} \
} while (0)
#define MCA_BTL_TCP_RECV_TRIGGER_CB(frag) \
do { \
if( MCA_BTL_TCP_HDR_TYPE_SEND == frag->hdr.type ) { \
mca_btl_active_message_callback_t* reg; \
reg = mca_btl_base_active_message_trigger + frag->hdr.base.tag; \
reg->cbfunc(&frag->endpoint->endpoint_btl->super, frag->hdr.base.tag, &frag->base, reg->cbdata); \
} \
} while (0)
extern opal_list_t mca_btl_tcp_ready_frag_pending_queue;
extern opal_mutex_t mca_btl_tcp_ready_frag_mutex;

14
opal/mca/btl/tcp/btl_tcp_endpoint.c
Просмотреть файл

@ -1,3 +1,4 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
@ -15,6 +16,7 @@
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 Amazon.com, Inc. or its affiliates. All Rights reserved.
* Copyright (c) 2020 Google, LLC. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -1016,9 +1018,15 @@ static void mca_btl_tcp_endpoint_recv_handler(int sd, short flags, void* user)
} else {
btl_endpoint->endpoint_recv_frag = NULL;
if( MCA_BTL_TCP_HDR_TYPE_SEND == frag->hdr.type ) {
mca_btl_active_message_callback_t* reg;
reg = mca_btl_base_active_message_trigger + frag->hdr.base.tag;
reg->cbfunc(&frag->btl->super, frag->hdr.base.tag, &frag->base, reg->cbdata);
mca_btl_active_message_callback_t *reg =
mca_btl_base_active_message_trigger + frag->hdr.base.tag;
const mca_btl_base_receive_descriptor_t desc =
{.endpoint = btl_endpoint,
.des_segments = frag->base.des_segments,
.des_segment_count = frag->base.des_segment_count,
.tag = frag->hdr.base.tag,
.cbdata = reg->cbdata};
reg->cbfunc(&frag->btl->super, &desc);
}
#if MCA_BTL_TCP_ENDPOINT_CACHE
if( 0 != btl_endpoint->endpoint_cache_length ) {

12
opal/mca/btl/uct/btl_uct_component.c
Просмотреть файл

@ -17,7 +17,7 @@
* Copyright (c) 2018 Amazon.com, Inc. or its affiliates. All Rights reserved.
* Copyright (c) 2018 Triad National Security, LLC. All rights
* reserved.
* Copyright (c) 2019 Google, LLC. All rights reserved.
* Copyright (c) 2019-2020 Google, LLC. All rights reserved.
* Copyright (c) 2019 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
@ -301,17 +301,15 @@ ucs_status_t mca_btl_uct_am_handler (void *arg, void *data, size_t length, unsig
mca_btl_uct_device_context_t *tl_context = (mca_btl_uct_device_context_t *) arg;
mca_btl_uct_module_t *uct_btl = tl_context->uct_btl;
mca_btl_uct_am_header_t *header = (mca_btl_uct_am_header_t *) data;
mca_btl_active_message_callback_t *reg;
mca_btl_active_message_callback_t *reg = mca_btl_base_active_message_trigger + header->data.tag;
mca_btl_base_segment_t seg = {.seg_addr = {.pval = (void *) ((intptr_t) data + sizeof (*header))},
.seg_len = length - sizeof (*header)};
mca_btl_uct_base_frag_t frag = {.base = {.des_segments = &seg, .des_segment_count = 1}};
mca_btl_base_receive_descriptor_t desc = {.endpoint = NULL, .des_segments = &seg, .des_segment_count = 1,
.tag = header->data.tag, .cbdata = reg->cbdata};
/* prevent recursion */
tl_context->in_am_callback = true;
reg = mca_btl_base_active_message_trigger + header->data.tag;
reg->cbfunc (&uct_btl->super, header->data.tag, &frag.base, reg->cbdata);
reg->cbfunc (&uct_btl->super, &desc);
tl_context->in_am_callback = false;
return UCS_OK;

4
opal/mca/btl/ugni/btl_ugni.h
Просмотреть файл

@ -303,10 +303,10 @@ OPAL_MODULE_DECLSPEC extern mca_btl_ugni_module_t mca_btl_ugni_module;
static inline uint32_t mca_btl_ugni_ep_get_device_index (mca_btl_ugni_module_t *ugni_module)
{
static volatile uint32_t device_index = (uint32_t) 0;
static opal_atomic_int64_t device_index = 0;
/* don't really care if the device index is atomically updated */
return opal_atomic_fetch_add_32 ((volatile int32_t *) &device_index, 1) % mca_btl_ugni_component.virtual_device_count;
return (uint32_t) (opal_atomic_fetch_add_64 (&device_index, 1) % mca_btl_ugni_component.virtual_device_count);
}
/**

14
opal/mca/btl/ugni/btl_ugni_get.c
Просмотреть файл

@ -84,30 +84,26 @@ static void mca_btl_ugni_callback_eager_get (struct mca_btl_base_module_t *btl,
uint8_t tag = frag->hdr.eager.send.lag >> 24;
size_t payload_len = frag->hdr.eager.size;
size_t hdr_len = len - payload_len;
mca_btl_active_message_callback_t *reg;
mca_btl_active_message_callback_t *reg = mca_btl_base_active_message_trigger + tag;
mca_btl_base_segment_t segs[2];
mca_btl_ugni_base_frag_t tmp;
const mca_btl_base_receive_descriptor_t tmp = {.endpoint = endpoint, .des_segments = segs,
.des_segment_count = hdr_len ? 2 : 1,
.tag = tag, .cbdata = reg->cbdata};
int rc;
BTL_VERBOSE(("eager get for rem_ctx %p complete", frag->hdr.eager.ctx))
tmp.base.des_segments = segs;
if (hdr_len) {
tmp.base.des_segment_count = 2;
segs[0].seg_addr.pval = frag->hdr.eager_ex.pml_header;
segs[0].seg_len = hdr_len;
segs[1].seg_addr.pval = local_address;
segs[1].seg_len = payload_len;
} else {
tmp.base.des_segment_count = 1;
segs[0].seg_addr.pval = local_address;
segs[0].seg_len = payload_len;
}
reg = mca_btl_base_active_message_trigger + tag;
reg->cbfunc(&ugni_module->super, tag, &(tmp.base), reg->cbdata);
reg->cbfunc(&ugni_module->super, &tmp);
/* fill in the response header */
frag->hdr.rdma.ctx = frag->hdr.eager.ctx;

17
opal/mca/btl/ugni/btl_ugni_smsg.c
Просмотреть файл

@ -57,7 +57,6 @@ int mca_btl_ugni_smsg_process (mca_btl_base_endpoint_t *ep)
mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_ep_btl (ep);
mca_btl_active_message_callback_t *reg;
mca_btl_ugni_base_frag_t frag;
mca_btl_base_segment_t seg;
bool disconnect = false;
int32_t _tmp_value = 0;
uintptr_t data_ptr;
@ -95,6 +94,7 @@ int mca_btl_ugni_smsg_process (mca_btl_base_endpoint_t *ep)
switch (tag) {
case MCA_BTL_UGNI_TAG_SEND:
frag.hdr.send = ((mca_btl_ugni_send_frag_hdr_t *) data_ptr)[0];
tag = frag.hdr.send.lag >> 24;
@ -103,15 +103,14 @@ int mca_btl_ugni_smsg_process (mca_btl_base_endpoint_t *ep)
BTL_VERBOSE(("received smsg fragment. hdr = {len = %u, tag = %d}", len, tag));
reg = mca_btl_base_active_message_trigger + tag;
frag.base.des_segments = &seg;
frag.base.des_segment_count = 1;
seg.seg_addr.pval = (void *)((uintptr_t)data_ptr + sizeof (mca_btl_ugni_send_frag_hdr_t));
seg.seg_len = len;
assert (NULL != reg->cbfunc);
reg->cbfunc(&ugni_module->super, tag, &(frag.base), reg->cbdata);
mca_btl_base_segment_t seg =
{.seg_addr.pval = (void *)((uintptr_t)data_ptr + sizeof (mca_btl_ugni_send_frag_hdr_t)),
seg.seg_len = len};
const mca_btl_base_receive_descriptor_t desc = {.endpoint = ep, .des_segments = &seg,
.des_segment_count = 1, .tag = tag,
.cbdata = reg->cbdata};
reg->cbfunc(&ugni_module->super, &desc);
break;
case MCA_BTL_UGNI_TAG_GET_INIT:

2
opal/mca/btl/usnic/btl_usnic_frag.h
Просмотреть файл

@ -199,7 +199,7 @@ struct opal_btl_usnic_endpoint_t;
*/
typedef struct opal_btl_usnic_recv_segment_t {
opal_btl_usnic_segment_t rs_base;
mca_btl_base_descriptor_t rs_desc;
mca_btl_base_receive_descriptor_t rs_desc;
mca_btl_base_segment_t rs_segment;
/* receive segments have protocol header prepended */

14
opal/mca/btl/usnic/btl_usnic_recv.c
Просмотреть файл

@ -1,3 +1,4 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2008 The Trustees of Indiana University and Indiana
* University Research and Technology
@ -142,7 +143,10 @@ void opal_btl_usnic_recv_call(opal_btl_usnic_module_t *module,
(unsigned)bseg->us_btl_header->payload_len,
(int)bseg->us_btl_header->tag);
#endif
reg->cbfunc(&module->super, hdr->tag, &seg->rs_desc, reg->cbdata);
seg->rs_desc.endpoint = endpoint;
seg->rs_desc.tag = hdr->tag;
seg->rs_desc.cbdata = reg->cbdata;
reg->cbfunc(&module->super, &seg->rs_desc);
/*
* If this is a PUT, need to copy it to user buffer
@ -271,7 +275,7 @@ void opal_btl_usnic_recv_call(opal_btl_usnic_module_t *module,
fip->rfi_bytes_left -= chunk_hdr->ch_hdr.payload_len;
if (0 == fip->rfi_bytes_left) {
mca_btl_base_descriptor_t desc;
mca_btl_base_receive_descriptor_t desc;
mca_btl_base_segment_t segment;
segment.seg_addr.pval = fip->rfi_data;
@ -293,8 +297,10 @@ void opal_btl_usnic_recv_call(opal_btl_usnic_module_t *module,
chunk_hdr->ch_hdr.tag;
/* mca_pml_ob1_recv_frag_callback_frag() */
reg->cbfunc(&module->super, chunk_hdr->ch_hdr.tag,
&desc, reg->cbdata);
desc.endpoint = endpoint;
desc.tag = chunk_hdr->ch_hdr.tag;
desc.cbdata = reg->cbdata;
reg->cbfunc(&module->super, &desc);
/* free temp buffer for non-put */
if (fip->rfi_data_in_pool) {

12
opal/mca/btl/usnic/btl_usnic_recv.h
Просмотреть файл

@ -1,3 +1,4 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2013-2017 Cisco Systems, Inc. All rights reserved
* $COPYRIGHT$
@ -329,8 +330,10 @@ opal_btl_usnic_dump_hex(15, USNIC_OUT, bseg->us_btl_header, bseg->us_btl_header-
*/
reg = mca_btl_base_active_message_trigger + bseg->us_btl_header->tag;
seg->rs_segment.seg_len = bseg->us_btl_header->payload_len;
reg->cbfunc(&module->super, bseg->us_btl_header->tag,
&seg->rs_desc, reg->cbdata);
seg->rs_desc.endpoint = endpoint;
seg->rs_desc.tag = bseg->us_btl_header->tag;
seg->rs_desc.cbdata = reg->cbdata;
reg->cbfunc(&module->super, &seg->rs_desc);
drop:
channel->chan_deferred_recv = seg;
@ -437,8 +440,9 @@ opal_btl_usnic_recv(opal_btl_usnic_module_t *module,
*/
reg = mca_btl_base_active_message_trigger + bseg->us_btl_header->tag;
seg->rs_segment.seg_len = bseg->us_btl_header->payload_len;
reg->cbfunc(&module->super, bseg->us_btl_header->tag,
&seg->rs_desc, reg->cbdata);
seg->rs_desc.tag = bseg->us_btl_header->tag;
seg->rs_desc.cbdata = reg->cbdata;
reg->cbfunc(&module->super, &seg->rs_desc);
}