1
1
openmpi/opal/mca/btl/vader/btl_vader_component.c
Ralph Castain aec5cd08bd Per the PMIx RFC:
WHAT:    Merge the PMIx branch into the devel repo, creating a new
               OPAL “lmix” framework to abstract PMI support for all RTEs.
               Replace the ORTE daemon-level collectives with a new PMIx
               server and update the ORTE grpcomm framework to support
               server-to-server collectives

WHY:      We’ve had problems dealing with variations in PMI implementations,
               and need to extend the existing PMI definitions to meet exascale
               requirements.

WHEN:   Mon, Aug 25

WHERE:  https://github.com/rhc54/ompi-svn-mirror.git

Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding.

All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level.

Accordingly, we have:

* created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations.

* Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported.

* Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint

* removed the prior OMPI/OPAL modex code

* added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform.

* retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand

This commit was SVN r32570.
2014-08-21 18:56:47 +00:00

430 строки
17 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) 2004-2005 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) 2006-2007 Voltaire. All rights reserved.
* Copyright (c) 2009-2010 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2010-2014 Los Alamos National Security, LLC.
* All rights reserved.
* Copyright (c) 2011 NVIDIA Corporation. All rights reserved.
* Copyright (c) 2014 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "opal_config.h"
#include "opal/util/output.h"
#include "opal/mca/btl/base/btl_base_error.h"
#include "opal/mca/pmix/pmix.h"
#include "btl_vader.h"
#include "btl_vader_frag.h"
#include "btl_vader_fifo.h"
#include "btl_vader_fbox.h"
#include <sys/mman.h>
static int mca_btl_vader_component_progress (void);
static int mca_btl_vader_component_open(void);
static int mca_btl_vader_component_close(void);
static int mca_btl_vader_component_register(void);
static mca_btl_base_module_t** mca_btl_vader_component_init(int *num_btls,
bool enable_progress_threads,
bool enable_mpi_threads);
/*
* Shared Memory (VADER) component instance.
*/
mca_btl_vader_component_t mca_btl_vader_component = {
.super = {
/* First, the mca_base_component_t struct containing meta information
about the component itself */
.btl_version = {
MCA_BTL_DEFAULT_VERSION("vader"),
.mca_open_component = mca_btl_vader_component_open,
.mca_close_component = mca_btl_vader_component_close,
.mca_register_component_params = mca_btl_vader_component_register,
},
.btl_data = {
/* The component is checkpoint ready */
.param_field = MCA_BASE_METADATA_PARAM_CHECKPOINT
},
.btl_init = mca_btl_vader_component_init,
.btl_progress = mca_btl_vader_component_progress,
} /* end super */
};
static int mca_btl_vader_component_register (void)
{
(void) mca_base_var_group_component_register(&mca_btl_vader_component.super.btl_version,
"XPMEM shared memory byte transport later");
/* register VADER component variables */
mca_btl_vader_component.vader_free_list_num = 8;
(void) mca_base_component_var_register(&mca_btl_vader_component.super.btl_version,
"free_list_num", "Initial number of fragments "
"to allocate for shared memory communication.",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_vader_component.vader_free_list_num);
mca_btl_vader_component.vader_free_list_max = 16384;
(void) mca_base_component_var_register(&mca_btl_vader_component.super.btl_version,
"free_list_max", "Maximum number of fragments "
"to allocate for shared memory communication.",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_vader_component.vader_free_list_max);
mca_btl_vader_component.vader_free_list_inc = 64;
(void) mca_base_component_var_register(&mca_btl_vader_component.super.btl_version,
"free_list_inc", "Number of fragments to create "
"on each allocation.", MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_vader_component.vader_free_list_inc);
mca_btl_vader_component.memcpy_limit = 524288;
(void) mca_base_component_var_register(&mca_btl_vader_component.super.btl_version,
"memcpy_limit", "Message size to switch from using "
"memove to memcpy. The relative speed of these two "
"routines can vary by size.", MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_vader_component.memcpy_limit);
mca_btl_vader_component.log_attach_align = 21;
(void) mca_base_component_var_register(&mca_btl_vader_component.super.btl_version,
"log_align", "Log base 2 of the alignment to use for xpmem "
"segments (default: 21, minimum: 12, maximum: 25)",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_vader_component.log_attach_align);
#if OPAL_BTL_VADER_HAVE_XPMEM && 64 == MCA_BTL_VADER_BITNESS
mca_btl_vader_component.segment_size = 1 << 24;
#else
mca_btl_vader_component.segment_size = 1 << 22;
#endif
(void) mca_base_component_var_register(&mca_btl_vader_component.super.btl_version,
"segment_size", "Maximum size of all shared "
#if OPAL_BTL_VADER_HAVE_XPMEM && 64 == MCA_BTL_VADER_BITNESS
"memory buffers (default: 16M)",
#else
"memory buffers (default: 4M)",
#endif
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_vader_component.segment_size);
mca_btl_vader_component.max_inline_send = 256;
(void) mca_base_component_var_register(&mca_btl_vader_component.super.btl_version,
"max_inline_send", "Maximum size to transfer "
"using copy-in copy-out semantics",
MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_vader_component.max_inline_send);
mca_btl_vader.super.btl_exclusivity = MCA_BTL_EXCLUSIVITY_HIGH;
#if OPAL_BTL_VADER_HAVE_XPMEM
mca_btl_vader.super.btl_eager_limit = 32 * 1024;
mca_btl_vader.super.btl_rndv_eager_limit = mca_btl_vader.super.btl_eager_limit;
mca_btl_vader.super.btl_max_send_size = mca_btl_vader.super.btl_eager_limit;
mca_btl_vader.super.btl_min_rdma_pipeline_size = mca_btl_vader.super.btl_eager_limit;
#else
mca_btl_vader.super.btl_eager_limit = 4 * 1024;
mca_btl_vader.super.btl_rndv_eager_limit = 32 * 1024;
mca_btl_vader.super.btl_max_send_size = 32 * 1024;
mca_btl_vader.super.btl_min_rdma_pipeline_size = 32 * 1024;
#endif
mca_btl_vader.super.btl_rdma_pipeline_send_length = mca_btl_vader.super.btl_eager_limit;
mca_btl_vader.super.btl_rdma_pipeline_frag_size = mca_btl_vader.super.btl_eager_limit;
#if OPAL_BTL_VADER_HAVE_XPMEM || OPAL_BTL_VADER_HAVE_CMA
mca_btl_vader.super.btl_flags = MCA_BTL_FLAGS_RDMA | MCA_BTL_FLAGS_SEND_INPLACE;
#else
mca_btl_vader.super.btl_flags = MCA_BTL_FLAGS_SEND_INPLACE;
#endif
mca_btl_vader.super.btl_seg_size = sizeof (mca_btl_base_segment_t);
#if OPAL_BTL_VADER_HAVE_XPMEM || OPAL_BTL_VADER_HAVE_CMA
mca_btl_vader.super.btl_bandwidth = 40000; /* Mbs */
#else
mca_btl_vader.super.btl_bandwidth = 10000; /* Mbs */
#endif
mca_btl_vader.super.btl_latency = 1; /* Microsecs */
/* Call the BTL based to register its MCA params */
mca_btl_base_param_register(&mca_btl_vader_component.super.btl_version,
&mca_btl_vader.super);
return OPAL_SUCCESS;
}
/*
* Called by MCA framework to open the component, registers
* component parameters.
*/
static int mca_btl_vader_component_open(void)
{
/* initialize objects */
OBJ_CONSTRUCT(&mca_btl_vader_component.vader_frags_eager, ompi_free_list_t);
OBJ_CONSTRUCT(&mca_btl_vader_component.vader_frags_user, ompi_free_list_t);
#if !OPAL_BTL_VADER_HAVE_XPMEM
OBJ_CONSTRUCT(&mca_btl_vader_component.vader_frags_max_send, ompi_free_list_t);
#endif
return OPAL_SUCCESS;
}
/*
* component cleanup - sanity checking of queue lengths
*/
static int mca_btl_vader_component_close(void)
{
OBJ_DESTRUCT(&mca_btl_vader_component.vader_frags_eager);
OBJ_DESTRUCT(&mca_btl_vader_component.vader_frags_user);
#if !OPAL_BTL_VADER_HAVE_XPMEM
OBJ_DESTRUCT(&mca_btl_vader_component.vader_frags_max_send);
#endif
if (NULL != mca_btl_vader_component.my_segment) {
munmap (mca_btl_vader_component.my_segment, mca_btl_vader_component.segment_size);
}
return OPAL_SUCCESS;
}
static int mca_btl_base_vader_modex_send (void)
{
struct vader_modex_t modex;
int modex_size;
int rc;
#if OPAL_BTL_VADER_HAVE_XPMEM
modex.seg_id = mca_btl_vader_component.my_seg_id;
modex.segment_base = mca_btl_vader_component.my_segment;
modex_size = sizeof (modex);
#else
modex_size = opal_shmem_sizeof_shmem_ds (&mca_btl_vader_component.seg_ds);
memmove (&modex.seg_ds, &mca_btl_vader_component.seg_ds, modex_size);
#endif
OPAL_MODEX_SEND(rc, PMIX_ASYNC_RDY, PMIX_LOCAL,
&mca_btl_vader_component.super.btl_version,
&modex, modex_size);
return rc;
}
/*
* VADER component initialization
*/
static mca_btl_base_module_t **mca_btl_vader_component_init (int *num_btls,
bool enable_progress_threads,
bool enable_mpi_threads)
{
mca_btl_vader_component_t *component = &mca_btl_vader_component;
mca_btl_base_module_t **btls = NULL;
int rc;
*num_btls = 0;
/* disable if there are no local peers */
if (0 == MCA_BTL_VADER_NUM_LOCAL_PEERS) {
BTL_VERBOSE(("No peers to communicate with. Disabling vader."));
return NULL;
}
/* limit segment alignment to be between 4k and 16M */
if (mca_btl_vader_component.log_attach_align < 12) {
mca_btl_vader_component.log_attach_align = 12;
} else if (mca_btl_vader_component.log_attach_align > 25) {
mca_btl_vader_component.log_attach_align = 25;
}
btls = (mca_btl_base_module_t **) calloc (1, sizeof (mca_btl_base_module_t *));
if (NULL == btls) {
return NULL;
}
/* ensure a sane segment size */
if (mca_btl_vader_component.segment_size < (2 << 20)) {
mca_btl_vader_component.segment_size = (2 << 20);
}
if (mca_btl_vader_component.segment_size > (1ul << MCA_BTL_VADER_OFFSET_BITS)) {
mca_btl_vader_component.segment_size = 2ul << MCA_BTL_VADER_OFFSET_BITS;
}
#if OPAL_BTL_VADER_HAVE_XPMEM
/* create an xpmem segment for the entire memory space */
component->my_seg_id = xpmem_make (0, VADER_MAX_ADDRESS, XPMEM_PERMIT_MODE, (void *)0666);
if (-1 == component->my_seg_id) {
BTL_VERBOSE(("Could not create xpmem segment"));
free (btls);
return NULL;
}
component->my_segment = mmap (NULL, mca_btl_vader_component.segment_size, PROT_READ |
PROT_WRITE, MAP_ANONYMOUS | MAP_SHARED, -1, 0);
if ((void *)-1 == component->my_segment) {
BTL_VERBOSE(("Could not create anonymous memory segment"));
free (btls);
return NULL;
}
#else
{
char *sm_file;
rc = asprintf(&sm_file, "%s" OPAL_PATH_SEP "vader_segment.%s.%d", opal_process_info.proc_session_dir,
opal_process_info.nodename, MCA_BTL_VADER_LOCAL_RANK);
if (0 > rc) {
free (btls);
return NULL;
}
rc = opal_shmem_segment_create (&mca_btl_vader_component.seg_ds, sm_file, mca_btl_vader_component.segment_size);
free (sm_file);
if (OPAL_SUCCESS != rc) {
BTL_VERBOSE(("Could not create shared memory segment"));
free (btls);
return NULL;
}
component->my_segment = opal_shmem_segment_attach (&mca_btl_vader_component.seg_ds);
if (NULL == component->my_segment) {
BTL_VERBOSE(("Could not attach to just created shared memory segment"));
goto failed;
}
}
#endif
component->segment_offset = 0;
memset (component->my_segment + MCA_BTL_VADER_FIFO_SIZE, 0, MCA_BTL_VADER_NUM_LOCAL_PEERS *
MCA_BTL_VADER_FBOX_PEER_SIZE);
/* initialize my fifo */
rc = vader_fifo_init ((struct vader_fifo_t *) component->my_segment);
if (OPAL_SUCCESS != rc) {
BTL_VERBOSE(("Error initializing FIFO"));
goto failed;
}
rc = mca_btl_base_vader_modex_send ();
if (OPAL_SUCCESS != rc) {
BTL_VERBOSE(("Error sending modex"));
goto failed;
}
*num_btls = 1;
/* get pointer to the btls */
btls[0] = (mca_btl_base_module_t *) &mca_btl_vader;
/* set flag indicating btl not inited */
mca_btl_vader.btl_inited = false;
return btls;
failed:
#if OPAL_BTL_VADER_HAVE_XPMEM
munmap (component->my_segment, mca_btl_vader_component.segment_size);
#else
opal_shmem_unlink (&mca_btl_vader_component.seg_ds);
#endif
if (btls) {
free (btls);
}
return NULL;
}
static inline int mca_btl_vader_poll_fifo (void)
{
const mca_btl_active_message_callback_t *reg;
struct mca_btl_base_endpoint_t *endpoint;
mca_btl_vader_hdr_t *hdr;
/* poll the fifo until it is empty or a limit has been hit (8 is arbitrary) */
for (int fifo_count = 0 ; fifo_count < 16 ; ++fifo_count) {
mca_btl_vader_frag_t frag = {.base = {.des_local = frag.segments, .des_local_count = 1}};
hdr = vader_fifo_read (mca_btl_vader_component.my_fifo, &endpoint);
if (NULL == hdr) {
return fifo_count;
}
if (hdr->flags & MCA_BTL_VADER_FLAG_COMPLETE) {
mca_btl_vader_frag_complete (hdr->frag);
continue;
}
reg = mca_btl_base_active_message_trigger + hdr->tag;
frag.segments[0].seg_addr.pval = (void *) (hdr + 1);
frag.segments[0].seg_len = hdr->len;
if (hdr->flags & MCA_BTL_VADER_FLAG_SINGLE_COPY) {
mca_mpool_base_registration_t *xpmem_reg;
xpmem_reg = vader_get_registation (endpoint, hdr->sc_iov.iov_base,
hdr->sc_iov.iov_len, 0,
&frag.segments[1].seg_addr.pval);
frag.segments[1].seg_len = hdr->sc_iov.iov_len;
/* recv upcall */
frag.base.des_local_count = 2;
reg->cbfunc(&mca_btl_vader.super, hdr->tag, &(frag.base), reg->cbdata);
vader_return_registration (xpmem_reg, endpoint);
} else {
reg->cbfunc(&mca_btl_vader.super, hdr->tag, &(frag.base), reg->cbdata);
}
/* return the fragment */
hdr->flags = MCA_BTL_VADER_FLAG_COMPLETE;
vader_fifo_write_back (hdr, endpoint);
}
return 1;
}
static int mca_btl_vader_component_progress (void)
{
bool fboxed;
/* check for messages in fast boxes */
for (int spin_count = 5 ; spin_count ; --spin_count) {
fboxed = (int) mca_btl_vader_check_fboxes ();
if (fboxed) {
break;
}
}
if (VADER_FIFO_FREE == mca_btl_vader_component.my_fifo->fifo_head) {
return (int) fboxed;
}
return mca_btl_vader_poll_fifo () + (int) fboxed;
}