1
1
openmpi/ompi/mca/pml/ob1/pml_ob1_isend.c
David Eberius d377a6b6f4 Added Software-based Performance Counters driver code along with several counters.
This code is the implementation of Software-base Performance Counters as described in the paper 'Using Software-Base Performance Counters to Expose Low-Level Open MPI Performance Information' in EuroMPI/USA '17 (http://icl.cs.utk.edu/news_pub/submissions/software-performance-counters.pdf).  More practical usage information can be found here: https://github.com/davideberius/ompi/wiki/How-to-Use-Software-Based-Performance-Counters-(SPCs)-in-Open-MPI.

All software events functions are put in macros that become no-ops when SOFTWARE_EVENTS_ENABLE is not defined.  The internal timer units have been changed to cycles to avoid division operations which was a large source of overhead as discussed in the paper.  Added a --with-spc configure option to enable SPCs in the Open MPI build.  This defines SOFTWARE_EVENTS_ENABLE.  Added an MCA parameter, mpi_spc_enable, for turning on specific counters.  Added an MCA parameter, mpi_spc_dump_enabled, for turning on and off dumping SPC counters in MPI_Finalize.  Added an SPC test and example.

Signed-off-by: David Eberius <deberius@vols.utk.edu>
2018-06-11 22:48:16 -04:00

294 строки
11 KiB
C

/* -*- 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
* Corporation. All rights reserved.
* Copyright (c) 2004-2018 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2007-2016 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2014 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "pml_ob1.h"
#include "pml_ob1_sendreq.h"
#include "pml_ob1_recvreq.h"
#include "ompi/peruse/peruse-internal.h"
#include "ompi/runtime/ompi_spc.h"
/**
* Single usage request. As we allow recursive calls (as an
* example from the request completion callback), we cannot rely
* on using a global request. Thus, once a send acquires ownership
* of this global request, it should set it to NULL to prevent
* the reuse until the first user completes.
*/
mca_pml_ob1_send_request_t *mca_pml_ob1_sendreq = NULL;
int mca_pml_ob1_isend_init(const void *buf,
size_t count,
ompi_datatype_t * datatype,
int dst,
int tag,
mca_pml_base_send_mode_t sendmode,
ompi_communicator_t * comm,
ompi_request_t ** request)
{
mca_pml_ob1_send_request_t *sendreq = NULL;
MCA_PML_OB1_SEND_REQUEST_ALLOC(comm, dst, sendreq);
if (NULL == sendreq)
return OMPI_ERR_OUT_OF_RESOURCE;
MCA_PML_OB1_SEND_REQUEST_INIT(sendreq,
buf,
count,
datatype,
dst, tag,
comm, sendmode, true);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&(sendreq)->req_send.req_base,
PERUSE_SEND);
/* Work around a leak in start by marking this request as complete. The
* problem occured because we do not have a way to differentiate an
* inital request and an incomplete pml request in start. This line
* allows us to detect this state. */
sendreq->req_send.req_base.req_pml_complete = true;
*request = (ompi_request_t *) sendreq;
return OMPI_SUCCESS;
}
/* try to get a small message out on to the wire quickly */
static inline int mca_pml_ob1_send_inline (const void *buf, size_t count,
ompi_datatype_t * datatype,
int dst, int tag, int16_t seqn,
ompi_proc_t *dst_proc, mca_bml_base_endpoint_t* endpoint,
ompi_communicator_t * comm)
{
mca_pml_ob1_match_hdr_t match;
mca_bml_base_btl_t *bml_btl;
opal_convertor_t convertor;
size_t size;
int rc;
bml_btl = mca_bml_base_btl_array_get_next(&endpoint->btl_eager);
if( NULL == bml_btl->btl->btl_sendi)
return OMPI_ERR_NOT_AVAILABLE;
ompi_datatype_type_size (datatype, &size);
if ((size * count) > 256) { /* some random number */
return OMPI_ERR_NOT_AVAILABLE;
}
if (count > 0) {
/* initialize just enough of the convertor to avoid a SEGV in opal_convertor_cleanup */
OBJ_CONSTRUCT(&convertor, opal_convertor_t);
/* We will create a convertor specialized for the */
/* remote architecture and prepared with the datatype. */
opal_convertor_copy_and_prepare_for_send (dst_proc->super.proc_convertor,
(const struct opal_datatype_t *) datatype,
count, buf, 0, &convertor);
opal_convertor_get_packed_size (&convertor, &size);
} else {
size = 0;
}
mca_pml_ob1_match_hdr_prepare (&match, MCA_PML_OB1_HDR_TYPE_MATCH, 0,
comm->c_contextid, comm->c_my_rank,
tag, seqn);
ob1_hdr_hton(&match, MCA_PML_OB1_HDR_TYPE_MATCH, dst_proc);
/* try to send immediately */
rc = mca_bml_base_sendi (bml_btl, &convertor, &match, OMPI_PML_OB1_MATCH_HDR_LEN,
size, MCA_BTL_NO_ORDER, MCA_BTL_DES_FLAGS_PRIORITY | MCA_BTL_DES_FLAGS_BTL_OWNERSHIP,
MCA_PML_OB1_HDR_TYPE_MATCH, NULL);
/* This #if is required due to an issue that arises with the IBM CI (XL Compiler).
* The compiler doesn't seem to like having a compiler hint attached to an if
* statement that only has a no-op inside and has the following error:
*
* 1586-494 (U) INTERNAL COMPILER ERROR: Signal 11.
*/
#if SPC_ENABLE == 1
if(OPAL_LIKELY(rc == OPAL_SUCCESS)) {
SPC_USER_OR_MPI(tag, (ompi_spc_value_t)size, OMPI_SPC_BYTES_SENT_USER, OMPI_SPC_BYTES_SENT_MPI);
}
#endif
if (count > 0) {
opal_convertor_cleanup (&convertor);
}
if (OPAL_UNLIKELY(OMPI_SUCCESS != rc)) {
return rc;
}
return (int) size;
}
int mca_pml_ob1_isend(const void *buf,
size_t count,
ompi_datatype_t * datatype,
int dst,
int tag,
mca_pml_base_send_mode_t sendmode,
ompi_communicator_t * comm,
ompi_request_t ** request)
{
mca_pml_ob1_comm_proc_t *ob1_proc = mca_pml_ob1_peer_lookup (comm, dst);
mca_pml_ob1_send_request_t *sendreq = NULL;
ompi_proc_t *dst_proc = ob1_proc->ompi_proc;
mca_bml_base_endpoint_t* endpoint = mca_bml_base_get_endpoint (dst_proc);
int16_t seqn = 0;
int rc;
if (OPAL_UNLIKELY(NULL == endpoint)) {
return OMPI_ERR_UNREACH;
}
if (!OMPI_COMM_CHECK_ASSERT_ALLOW_OVERTAKE(comm)) {
seqn = (uint16_t) OPAL_THREAD_ADD_FETCH32(&ob1_proc->send_sequence, 1);
}
if (MCA_PML_BASE_SEND_SYNCHRONOUS != sendmode) {
rc = mca_pml_ob1_send_inline (buf, count, datatype, dst, tag, seqn, dst_proc,
endpoint, comm);
if (OPAL_LIKELY(0 <= rc)) {
/* NTH: it is legal to return ompi_request_empty since the only valid
* field in a send completion status is whether or not the send was
* cancelled (which it can't be at this point anyway). */
*request = &ompi_request_empty;
return OMPI_SUCCESS;
}
}
MCA_PML_OB1_SEND_REQUEST_ALLOC(comm, dst, sendreq);
if (NULL == sendreq)
return OMPI_ERR_OUT_OF_RESOURCE;
MCA_PML_OB1_SEND_REQUEST_INIT(sendreq,
buf,
count,
datatype,
dst, tag,
comm, sendmode, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&(sendreq)->req_send.req_base,
PERUSE_SEND);
MCA_PML_OB1_SEND_REQUEST_START_W_SEQ(sendreq, endpoint, seqn, rc);
*request = (ompi_request_t *) sendreq;
return rc;
}
int mca_pml_ob1_send(const void *buf,
size_t count,
ompi_datatype_t * datatype,
int dst,
int tag,
mca_pml_base_send_mode_t sendmode,
ompi_communicator_t * comm)
{
mca_pml_ob1_comm_proc_t *ob1_proc = mca_pml_ob1_peer_lookup (comm, dst);
ompi_proc_t *dst_proc = ob1_proc->ompi_proc;
mca_bml_base_endpoint_t* endpoint = mca_bml_base_get_endpoint (dst_proc);
mca_pml_ob1_send_request_t *sendreq = NULL;
int16_t seqn = 0;
int rc;
if (OPAL_UNLIKELY(NULL == endpoint)) {
return OMPI_ERR_UNREACH;
}
if (OPAL_UNLIKELY(MCA_PML_BASE_SEND_BUFFERED == sendmode)) {
/* large buffered sends *need* a real request so use isend instead */
ompi_request_t *brequest;
rc = mca_pml_ob1_isend (buf, count, datatype, dst, tag, sendmode, comm, &brequest);
if (OPAL_UNLIKELY(OMPI_SUCCESS != rc)) {
return rc;
}
ompi_request_wait_completion (brequest);
ompi_request_free (&brequest);
return OMPI_SUCCESS;
}
if (!OMPI_COMM_CHECK_ASSERT_ALLOW_OVERTAKE(comm)) {
seqn = (uint16_t) OPAL_THREAD_ADD_FETCH32(&ob1_proc->send_sequence, 1);
}
/**
* The immediate send will not have a request, so they are
* intracable from the point of view of any debugger attached to
* the parallel application.
*/
if (MCA_PML_BASE_SEND_SYNCHRONOUS != sendmode) {
rc = mca_pml_ob1_send_inline (buf, count, datatype, dst, tag, seqn, dst_proc,
endpoint, comm);
if (OPAL_LIKELY(0 <= rc)) {
return OMPI_SUCCESS;
}
}
if (OPAL_LIKELY(!ompi_mpi_thread_multiple)) {
sendreq = mca_pml_ob1_sendreq;
mca_pml_ob1_sendreq = NULL;
}
if( OPAL_UNLIKELY(NULL == sendreq) ) {
MCA_PML_OB1_SEND_REQUEST_ALLOC(comm, dst, sendreq);
if (NULL == sendreq)
return OMPI_ERR_TEMP_OUT_OF_RESOURCE;
}
sendreq->req_send.req_base.req_proc = dst_proc;
sendreq->rdma_frag = NULL;
MCA_PML_OB1_SEND_REQUEST_INIT(sendreq,
buf,
count,
datatype,
dst, tag,
comm, sendmode, false);
PERUSE_TRACE_COMM_EVENT (PERUSE_COMM_REQ_ACTIVATE,
&sendreq->req_send.req_base,
PERUSE_SEND);
MCA_PML_OB1_SEND_REQUEST_START_W_SEQ(sendreq, endpoint, seqn, rc);
if (OPAL_LIKELY(rc == OMPI_SUCCESS)) {
ompi_request_wait_completion(&sendreq->req_send.req_base.req_ompi);
rc = sendreq->req_send.req_base.req_ompi.req_status.MPI_ERROR;
}
if (OPAL_UNLIKELY(ompi_mpi_thread_multiple || NULL != mca_pml_ob1_sendreq)) {
MCA_PML_OB1_SEND_REQUEST_RETURN(sendreq);
} else {
mca_pml_ob1_send_request_fini (sendreq);
mca_pml_ob1_sendreq = sendreq;
}
return rc;
}