e627c91227
This commit adds support for placing the send memory segment in a traditional shared memory segment when XPMEM is not available. The current default is to reserve 4MB for shared memory on each process. The latest benchmarks show vader performing better than sm on both Intel and AMD CPUs. For large messages vader will now use CMA if it is available (and XPMEM is not). cmr=v1.7.5:reviewer=jsquyres This commit was SVN r30123.
99 строки
3.4 KiB
C
99 строки
3.4 KiB
C
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
|
|
/*
|
|
* Copyright (c) 2004-2011 The Trustees of Indiana University and Indiana
|
|
* University Research and Technology
|
|
* Corporation. All rights reserved.
|
|
* Copyright (c) 2004-2009 The University of Tennessee and The University
|
|
* of Tennessee Research Foundation. All rights
|
|
* reserved.
|
|
* Copyright (c) 2004-2007 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) 2006-2007 Voltaire. All rights reserved.
|
|
* Copyright (c) 2009 Cisco Systems, Inc. All rights reserved.
|
|
* Copyright (c) 2010-2013 Los Alamos National Security, LLC.
|
|
* All rights reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
#include "ompi_config.h"
|
|
|
|
#include "btl_vader.h"
|
|
#include "btl_vader_frag.h"
|
|
#include "btl_vader_fifo.h"
|
|
|
|
#include "btl_vader_fbox.h"
|
|
|
|
/**
|
|
* Initiate an inline send to the peer.
|
|
*
|
|
* @param btl (IN) BTL module
|
|
* @param peer (IN) BTL peer addressing
|
|
*/
|
|
int mca_btl_vader_sendi (struct mca_btl_base_module_t *btl,
|
|
struct mca_btl_base_endpoint_t *endpoint,
|
|
struct opal_convertor_t *convertor,
|
|
void *header, size_t header_size,
|
|
size_t payload_size, uint8_t order,
|
|
uint32_t flags, mca_btl_base_tag_t tag,
|
|
mca_btl_base_descriptor_t **descriptor)
|
|
{
|
|
mca_btl_vader_frag_t *frag;
|
|
void *data_ptr = NULL;
|
|
size_t length;
|
|
|
|
if (payload_size) {
|
|
opal_convertor_get_current_pointer (convertor, &data_ptr);
|
|
}
|
|
|
|
if (!(payload_size && opal_convertor_need_buffers (convertor)) &&
|
|
mca_btl_vader_fbox_sendi (endpoint, tag, header, header_size, data_ptr, payload_size)) {
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
|
|
length = header_size + payload_size;
|
|
|
|
/* allocate a fragment, giving up if we can't get one */
|
|
frag = (mca_btl_vader_frag_t *) mca_btl_vader_alloc (btl, endpoint, order, length,
|
|
flags | MCA_BTL_DES_FLAGS_BTL_OWNERSHIP);
|
|
if (OPAL_UNLIKELY(NULL == frag)) {
|
|
*descriptor = NULL;
|
|
|
|
return OMPI_ERR_OUT_OF_RESOURCE;
|
|
}
|
|
|
|
/* fill in fragment fields */
|
|
frag->hdr->len = length;
|
|
frag->hdr->tag = tag;
|
|
|
|
/* write the match header (with MPI comm/tag/etc. info) */
|
|
memcpy (frag->segments[0].seg_addr.pval, header, header_size);
|
|
|
|
/* write the message data if there is any */
|
|
/* we can't use single-copy semantics here since as caller will consider the send
|
|
complete when we return */
|
|
if (payload_size) {
|
|
uint32_t iov_count = 1;
|
|
struct iovec iov;
|
|
|
|
/* pack the data into the supplied buffer */
|
|
iov.iov_base = (IOVBASE_TYPE *)((uintptr_t)frag->segments[0].seg_addr.pval + header_size);
|
|
iov.iov_len = length = payload_size;
|
|
|
|
(void) opal_convertor_pack (convertor, &iov, &iov_count, &length);
|
|
|
|
assert (length == payload_size);
|
|
}
|
|
|
|
/* write the fragment pointer to peer's the FIFO. the progress function will return the fragment */
|
|
vader_fifo_write_ep (frag->hdr, endpoint);
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|