e7ecd56bd2
such, the commit message back to the master SVN repository is fairly
long.
= ORTE Job-Level Output Messages =
Add two new interfaces that should be used for all new code throughout
the ORTE and OMPI layers (we already make the search-and-replace on
the existing ORTE / OMPI layers):
* orte_output(): (and corresponding friends ORTE_OUTPUT,
orte_output_verbose, etc.) This function sends the output directly
to the HNP for processing as part of a job-specific output
channel. It supports all the same outputs as opal_output()
(syslog, file, stdout, stderr), but for stdout/stderr, the output
is sent to the HNP for processing and output. More on this below.
* orte_show_help(): This function is a drop-in-replacement for
opal_show_help(), with two differences in functionality:
1. the rendered text help message output is sent to the HNP for
display (rather than outputting directly into the process' stderr
stream)
1. the HNP detects duplicate help messages and does not display them
(so that you don't see the same error message N times, once from
each of your N MPI processes); instead, it counts "new" instances
of the help message and displays a message every ~5 seconds when
there are new ones ("I got X new copies of the help message...")
opal_show_help and opal_output still exist, but they only output in
the current process. The intent for the new orte_* functions is that
they can apply job-level intelligence to the output. As such, we
recommend that all new ORTE and OMPI code use the new orte_*
functions, not thei opal_* functions.
=== New code ===
For ORTE and OMPI programmers, here's what you need to do differently
in new code:
* Do not include opal/util/show_help.h or opal/util/output.h.
Instead, include orte/util/output.h (this one header file has
declarations for both the orte_output() series of functions and
orte_show_help()).
* Effectively s/opal_output/orte_output/gi throughout your code.
Note that orte_output_open() takes a slightly different argument
list (as a way to pass data to the filtering stream -- see below),
so you if explicitly call opal_output_open(), you'll need to
slightly adapt to the new signature of orte_output_open().
* Literally s/opal_show_help/orte_show_help/. The function signature
is identical.
=== Notes ===
* orte_output'ing to stream 0 will do similar to what
opal_output'ing did, so leaving a hard-coded "0" as the first
argument is safe.
* For systems that do not use ORTE's RML or the HNP, the effect of
orte_output_* and orte_show_help will be identical to their opal
counterparts (the additional information passed to
orte_output_open() will be lost!). Indeed, the orte_* functions
simply become trivial wrappers to their opal_* counterparts. Note
that we have not tested this; the code is simple but it is quite
possible that we mucked something up.
= Filter Framework =
Messages sent view the new orte_* functions described above and
messages output via the IOF on the HNP will now optionally be passed
through a new "filter" framework before being output to
stdout/stderr. The "filter" OPAL MCA framework is intended to allow
preprocessing to messages before they are sent to their final
destinations. The first component that was written in the filter
framework was to create an XML stream, segregating all the messages
into different XML tags, etc. This will allow 3rd party tools to read
the stdout/stderr from the HNP and be able to know exactly what each
text message is (e.g., a help message, another OMPI infrastructure
message, stdout from the user process, stderr from the user process,
etc.).
Filtering is not active by default. Filter components must be
specifically requested, such as:
{{{
$ mpirun --mca filter xml ...
}}}
There can only be one filter component active.
= New MCA Parameters =
The new functionality described above introduces two new MCA
parameters:
* '''orte_base_help_aggregate''': Defaults to 1 (true), meaning that
help messages will be aggregated, as described above. If set to 0,
all help messages will be displayed, even if they are duplicates
(i.e., the original behavior).
* '''orte_base_show_output_recursions''': An MCA parameter to help
debug one of the known issues, described below. It is likely that
this MCA parameter will disappear before v1.3 final.
= Known Issues =
* The XML filter component is not complete. The current output from
this component is preliminary and not real XML. A bit more work
needs to be done to configure.m4 search for an appropriate XML
library/link it in/use it at run time.
* There are possible recursion loops in the orte_output() and
orte_show_help() functions -- e.g., if RML send calls orte_output()
or orte_show_help(). We have some ideas how to fix these, but
figured that it was ok to commit before feature freeze with known
issues. The code currently contains sub-optimal workarounds so
that this will not be a problem, but it would be good to actually
solve the problem rather than have hackish workarounds before v1.3 final.
This commit was SVN r18434.
739 строки
27 KiB
C
739 строки
27 KiB
C
/*
|
|
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
|
|
* University Research and Technology
|
|
* Corporation. All rights reserved.
|
|
* Copyright (c) 2004-2005 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 Cisco Systems, Inc. All rights reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
|
|
#include "ompi_config.h"
|
|
#include "ompi/constants.h"
|
|
#include "opal/event/event.h"
|
|
#include "opal/util/if.h"
|
|
#include "opal/util/argv.h"
|
|
#include "orte/util/output.h"
|
|
#include "ompi/mca/pml/pml.h"
|
|
#include "ompi/mca/btl/btl.h"
|
|
#include "ompi/request/request.h"
|
|
|
|
#ifdef HAVE_INTTYPES_H
|
|
#include <inttypes.h>
|
|
#endif
|
|
|
|
#include "opal/mca/base/mca_base_param.h"
|
|
#include "orte/mca/errmgr/errmgr.h"
|
|
#include "ompi/mca/mpool/base/base.h"
|
|
#include "ompi/mca/mpool/rdma/mpool_rdma.h"
|
|
#include "btl_gm.h"
|
|
#include "btl_gm_frag.h"
|
|
#include "btl_gm_endpoint.h"
|
|
#include "ompi/mca/btl/base/base.h"
|
|
#include "ompi/mca/btl/base/btl_base_error.h"
|
|
#include "ompi/datatype/convertor.h"
|
|
#include "btl_gm_endpoint.h"
|
|
#include "orte/util/proc_info.h"
|
|
#include "orte/runtime/orte_globals.h"
|
|
#include "orte/util/name_fns.h"
|
|
#include "ompi/runtime/ompi_module_exchange.h"
|
|
|
|
|
|
#if OMPI_ENABLE_PROGRESS_THREADS
|
|
static void* mca_btl_gm_progress_thread( opal_object_t* arg );
|
|
#endif
|
|
static int gm_reg_mr(void *reg_data, void *base, size_t size,
|
|
mca_mpool_base_registration_t *reg);
|
|
static int gm_dereg_mr(void *reg_data, mca_mpool_base_registration_t *reg);
|
|
|
|
|
|
mca_btl_gm_component_t mca_btl_gm_component = {
|
|
{
|
|
/* First, the mca_base_component_t struct containing meta information
|
|
about the component itself */
|
|
|
|
{
|
|
/* Indicate that we are a pml v1.0.0 component (which also implies a
|
|
specific MCA version) */
|
|
|
|
MCA_BTL_BASE_VERSION_1_0_1,
|
|
|
|
"gm", /* MCA component name */
|
|
OMPI_MAJOR_VERSION, /* MCA component major version */
|
|
OMPI_MINOR_VERSION, /* MCA component minor version */
|
|
OMPI_RELEASE_VERSION, /* MCA component release version */
|
|
mca_btl_gm_component_open, /* component open */
|
|
mca_btl_gm_component_close /* component close */
|
|
},
|
|
|
|
/* Next the MCA v1.0.0 component meta data */
|
|
|
|
{
|
|
/* The component is not checkpoint ready */
|
|
MCA_BASE_METADATA_PARAM_NONE
|
|
},
|
|
|
|
mca_btl_gm_component_init,
|
|
mca_btl_gm_component_progress,
|
|
}
|
|
};
|
|
|
|
|
|
/*
|
|
* utility routines for parameter registration
|
|
*/
|
|
|
|
static inline char* mca_btl_gm_param_register_string(
|
|
const char* param_name,
|
|
const char* default_value)
|
|
{
|
|
char *param_value;
|
|
int id = mca_base_param_register_string("btl","gm",param_name,NULL,default_value);
|
|
mca_base_param_lookup_string(id, ¶m_value);
|
|
return param_value;
|
|
}
|
|
|
|
static inline int mca_btl_gm_param_register_int(
|
|
const char* param_name,
|
|
int default_value)
|
|
{
|
|
int id = mca_base_param_register_int("btl","gm",param_name,NULL,default_value);
|
|
int param_value = default_value;
|
|
mca_base_param_lookup_int(id,¶m_value);
|
|
return param_value;
|
|
}
|
|
|
|
/*
|
|
* Called by MCA framework to open the component, registers
|
|
* component parameters.
|
|
*/
|
|
|
|
int mca_btl_gm_component_open(void)
|
|
{
|
|
/* initialize state */
|
|
mca_btl_gm_component.gm_num_btls=0;
|
|
mca_btl_gm_component.gm_btls=NULL;
|
|
|
|
/* initialize objects */
|
|
OBJ_CONSTRUCT(&mca_btl_gm_component.gm_procs, opal_list_t);
|
|
OBJ_CONSTRUCT(&mca_btl_gm_component.gm_lock, opal_mutex_t);
|
|
|
|
/* register GM component parameters */
|
|
mca_btl_gm_component.gm_free_list_num =
|
|
mca_btl_gm_param_register_int ("free_list_num", 8);
|
|
mca_btl_gm_component.gm_free_list_max =
|
|
mca_btl_gm_param_register_int ("free_list_max", -1);
|
|
mca_btl_gm_component.gm_free_list_inc =
|
|
mca_btl_gm_param_register_int ("free_list_inc", 8);
|
|
mca_btl_gm_component.gm_debug =
|
|
mca_btl_gm_param_register_int("debug", 0);
|
|
mca_btl_gm_component.gm_mpool_name =
|
|
mca_btl_gm_param_register_string("mpool", "rdma");
|
|
mca_btl_gm_component.gm_max_ports =
|
|
mca_btl_gm_param_register_int("max_ports", 16);
|
|
mca_btl_gm_component.gm_max_boards =
|
|
mca_btl_gm_param_register_int("max_boards", 4);
|
|
mca_btl_gm_component.gm_max_btls =
|
|
mca_btl_gm_param_register_int("max_modules", 4);
|
|
mca_btl_gm_component.gm_num_high_priority =
|
|
mca_btl_gm_param_register_int("num_high_priority", 8);
|
|
mca_btl_gm_component.gm_num_repost =
|
|
mca_btl_gm_param_register_int("num_repost", 4);
|
|
mca_btl_gm_component.gm_port_name=
|
|
mca_btl_gm_param_register_string("port_name", "OMPI");
|
|
|
|
/* register gm module parameters */
|
|
mca_btl_gm_module.super.btl_exclusivity = MCA_BTL_EXCLUSIVITY_DEFAULT;
|
|
mca_btl_gm_module.super.btl_eager_limit = 32*1024;
|
|
mca_btl_gm_module.super.btl_rndv_eager_limit = 32*1024;
|
|
mca_btl_gm_module.super.btl_max_send_size = 64*1024;
|
|
mca_btl_gm_module.super.btl_rdma_pipeline_send_length = 512*1024;
|
|
mca_btl_gm_module.super.btl_rdma_pipeline_frag_size = 128*1024;
|
|
mca_btl_gm_module.super.btl_min_rdma_pipeline_size = 128*1024;
|
|
#if OMPI_MCA_BTL_GM_HAVE_RDMA_PUT
|
|
mca_btl_gm_module.super.btl_flags = MCA_BTL_FLAGS_PUT |
|
|
MCA_BTL_FLAGS_NEED_ACK | MCA_BTL_FLAGS_NEED_CSUM;
|
|
#else
|
|
mca_btl_gm_module.super.btl_flags = MCA_BTL_FLAGS_SEND;
|
|
#endif
|
|
mca_btl_gm_module.super.btl_bandwidth = 250;
|
|
mca_btl_gm_module.super.btl_latency = 0;
|
|
mca_btl_base_param_register(&mca_btl_gm_component.super.btl_version,
|
|
&mca_btl_gm_module.super);
|
|
|
|
/* compute the eager frag size */
|
|
mca_btl_gm_component.gm_eager_frag_size =
|
|
gm_min_size_for_length(mca_btl_gm_module.super.btl_eager_limit) - 1;
|
|
mca_btl_gm_module.super.btl_eager_limit =
|
|
gm_max_length_for_size(mca_btl_gm_component.gm_eager_frag_size) -
|
|
sizeof(mca_btl_base_header_t);
|
|
|
|
/* compute the max frag size */
|
|
mca_btl_gm_component.gm_max_frag_size =
|
|
gm_min_size_for_length(mca_btl_gm_module.super.btl_max_send_size) - 1;
|
|
mca_btl_gm_module.super.btl_max_send_size =
|
|
gm_max_length_for_size(mca_btl_gm_component.gm_max_frag_size) -
|
|
sizeof(mca_btl_base_header_t);
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
|
|
/*
|
|
* component cleanup - sanity checking of queue lengths
|
|
*/
|
|
|
|
int mca_btl_gm_component_close(void)
|
|
{
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
static int gm_reg_mr(void *reg_data, void *base, size_t size,
|
|
mca_mpool_base_registration_t *reg)
|
|
{
|
|
struct gm_port *port = (struct gm_port*)reg_data;
|
|
int rc;
|
|
|
|
rc = gm_register_memory(port, base, size);
|
|
|
|
if(rc != GM_SUCCESS)
|
|
return OMPI_ERR_OUT_OF_RESOURCE;
|
|
|
|
return MPI_SUCCESS;
|
|
}
|
|
|
|
static int gm_dereg_mr(void *reg_data, mca_mpool_base_registration_t *reg)
|
|
{
|
|
struct gm_port *port = (struct gm_port*)reg_data;
|
|
int rc;
|
|
|
|
rc = gm_deregister_memory(port, reg->base, reg->bound - reg->base + 1);
|
|
|
|
if(rc != GM_SUCCESS) {
|
|
orte_output(0, "%s: error unpinning gm memory errno says %s\n",
|
|
__func__, strerror(errno));
|
|
return OMPI_ERROR;
|
|
}
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
/**
|
|
* Initialize module instance
|
|
*/
|
|
|
|
static int
|
|
mca_btl_gm_module_init (mca_btl_gm_module_t * btl)
|
|
{
|
|
mca_mpool_base_resources_t resources;
|
|
int32_t num_high_priority;
|
|
int32_t i;
|
|
int rc;
|
|
|
|
/* initialize objects */
|
|
OBJ_CONSTRUCT(&btl->gm_frag_eager, ompi_free_list_t);
|
|
OBJ_CONSTRUCT(&btl->gm_frag_max, ompi_free_list_t);
|
|
OBJ_CONSTRUCT(&btl->gm_frag_user, ompi_free_list_t);
|
|
OBJ_CONSTRUCT(&btl->gm_pending, opal_list_t);
|
|
OBJ_CONSTRUCT(&btl->gm_repost, opal_list_t);
|
|
#if OMPI_ENABLE_PROGRESS_THREADS
|
|
OBJ_CONSTRUCT(&btl->gm_thread, opal_thread_t);
|
|
#endif
|
|
|
|
btl->error_cb = NULL;
|
|
/* query nic tokens */
|
|
btl->gm_num_send_tokens = gm_num_send_tokens (btl->port);
|
|
btl->gm_max_send_tokens = btl->gm_num_send_tokens;
|
|
btl->gm_num_recv_tokens = gm_num_receive_tokens (btl->port);
|
|
btl->gm_max_recv_tokens = btl->gm_num_recv_tokens;
|
|
|
|
/* dont allow high priority to exceed 1/2 of available recv tokens */
|
|
num_high_priority = mca_btl_gm_component.gm_num_high_priority;
|
|
if(num_high_priority > (btl->gm_num_recv_tokens >> 1)) {
|
|
num_high_priority = btl->gm_num_recv_tokens >> 1;
|
|
}
|
|
if(mca_btl_gm_component.gm_num_repost > num_high_priority >> 1) {
|
|
btl->gm_num_repost = (num_high_priority >> 1);
|
|
} else {
|
|
btl->gm_num_repost = mca_btl_gm_component.gm_num_repost;
|
|
}
|
|
|
|
/* initialize memory pool */
|
|
resources.reg_data = (void*)btl->port;
|
|
resources.sizeof_reg = sizeof(mca_mpool_base_registration_t);
|
|
resources.register_mem = gm_reg_mr;
|
|
resources.deregister_mem = gm_dereg_mr;
|
|
btl->super.btl_mpool = mca_mpool_base_module_create(
|
|
mca_btl_gm_component.gm_mpool_name,
|
|
&btl->super,
|
|
&resources);
|
|
if(NULL == btl->super.btl_mpool) {
|
|
orte_output (0, "[%s:%d] unable to initialize mpool", __FILE__, __LINE__);
|
|
return OMPI_ERROR;
|
|
}
|
|
|
|
/* initialize free lists */
|
|
ompi_free_list_init_new( &btl->gm_frag_eager,
|
|
sizeof (mca_btl_gm_frag_eager_t),
|
|
CACHE_LINE_SIZE,
|
|
OBJ_CLASS (mca_btl_gm_frag_eager_t),
|
|
(1 << mca_btl_gm_component.gm_eager_frag_size) + sizeof (uintptr_t),
|
|
CACHE_LINE_SIZE,
|
|
btl->gm_max_send_tokens,
|
|
mca_btl_gm_component.gm_free_list_max,
|
|
mca_btl_gm_component.gm_free_list_inc,
|
|
btl->super.btl_mpool );
|
|
|
|
ompi_free_list_init_new( &btl->gm_frag_max,
|
|
sizeof (mca_btl_gm_frag_max_t),
|
|
CACHE_LINE_SIZE,
|
|
OBJ_CLASS (mca_btl_gm_frag_max_t),
|
|
(1 << mca_btl_gm_component.gm_max_frag_size) + sizeof (uintptr_t),
|
|
CACHE_LINE_SIZE,
|
|
btl->gm_max_recv_tokens,
|
|
mca_btl_gm_component.gm_free_list_max,
|
|
mca_btl_gm_component.gm_free_list_inc,
|
|
btl->super.btl_mpool );
|
|
|
|
ompi_free_list_init_new( &btl->gm_frag_user,
|
|
sizeof (mca_btl_gm_frag_user_t),
|
|
CACHE_LINE_SIZE,
|
|
OBJ_CLASS (mca_btl_gm_frag_user_t),
|
|
0,CACHE_LINE_SIZE,
|
|
mca_btl_gm_component.gm_free_list_num,
|
|
mca_btl_gm_component.gm_free_list_max,
|
|
mca_btl_gm_component.gm_free_list_inc,
|
|
NULL );
|
|
|
|
|
|
/* post receive buffers */
|
|
for(i=0; i<num_high_priority; i++) {
|
|
mca_btl_gm_frag_t* frag;
|
|
MCA_BTL_GM_FRAG_ALLOC_EAGER(btl, frag, rc);
|
|
if(NULL == frag) {
|
|
return rc;
|
|
}
|
|
frag->type = MCA_BTL_GM_EAGER;
|
|
frag->base.des_src = NULL;
|
|
frag->base.des_src_cnt = 0;
|
|
frag->base.des_dst = &frag->segment;
|
|
frag->base.des_dst_cnt = 1;
|
|
frag->priority = GM_HIGH_PRIORITY;
|
|
gm_provide_receive_buffer(btl->port, frag->hdr, frag->size, frag->priority);
|
|
}
|
|
|
|
for(i=mca_btl_gm_component.gm_num_high_priority; i<btl->gm_max_recv_tokens; i++) {
|
|
mca_btl_gm_frag_t* frag;
|
|
MCA_BTL_GM_FRAG_ALLOC_MAX(btl, frag, rc);
|
|
if(NULL == frag) {
|
|
return rc;
|
|
}
|
|
frag->type = MCA_BTL_GM_SEND;
|
|
frag->base.des_src = NULL;
|
|
frag->base.des_src_cnt = 0;
|
|
frag->base.des_dst = &frag->segment;
|
|
frag->base.des_dst_cnt = 1;
|
|
frag->priority = GM_LOW_PRIORITY;
|
|
gm_provide_receive_buffer(btl->port, frag->hdr, frag->size, frag->priority);
|
|
}
|
|
|
|
/* enable rdma */
|
|
if( GM_SUCCESS != gm_allow_remote_memory_access (btl->port) ) {
|
|
orte_output (0, "[%s:%d] unable to allow remote memory access", __FILE__, __LINE__);
|
|
return OMPI_ERROR;
|
|
}
|
|
|
|
#if OMPI_ENABLE_PROGRESS_THREADS
|
|
/* start progress thread */
|
|
btl->gm_progress = true;
|
|
btl->gm_thread.t_run = mca_btl_gm_progress_thread;
|
|
btl->gm_thread.t_arg = btl;
|
|
if(OPAL_SUCCESS != (rc = opal_thread_start(&btl->gm_thread))) {
|
|
orte_output (0, "[%s:%d] unable to create progress thread, retval=%d", __FILE__, __LINE__, rc);
|
|
return rc;
|
|
}
|
|
#endif
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
|
|
/*
|
|
* Scan all ports on the boards. As it's difficult to find the total number of boards
|
|
* so we use a predefined maximum.
|
|
*/
|
|
static int mca_btl_gm_discover( void )
|
|
{
|
|
uint32_t board_no;
|
|
uint32_t port_no;
|
|
uint32_t node_id;
|
|
struct gm_port* port;
|
|
#if GM_API_VERSION > 0x200
|
|
uint32_t global_id;
|
|
#else
|
|
char global_id[GM_MAX_HOST_NAME_LEN];
|
|
#endif /* GM_API_VERSION > 0x200 */
|
|
int rc;
|
|
|
|
for( board_no = 0; board_no < mca_btl_gm_component.gm_max_boards; board_no++ ) {
|
|
mca_btl_gm_module_t *btl;
|
|
|
|
/* open the first available gm port for this board */
|
|
for( port_no = 4; port_no < mca_btl_gm_component.gm_max_ports; port_no++ ) {
|
|
if (3 == port_no) {
|
|
continue; /* port 0,1,3 reserved */
|
|
} else if (GM_SUCCESS == gm_open(&port, board_no, port_no,
|
|
mca_btl_gm_component.gm_port_name, GM_API_VERSION) ) {
|
|
break;
|
|
}
|
|
}
|
|
if( port_no == mca_btl_gm_component.gm_max_ports ) {
|
|
continue;
|
|
}
|
|
|
|
/* Get node local Id */
|
|
if( GM_SUCCESS != gm_get_node_id( port, &node_id) ) {
|
|
orte_output (0, " failure to get node_id \n");
|
|
continue;
|
|
}
|
|
/* Gather an unique id for the node */
|
|
#if GM_API_VERSION > 0x200
|
|
if (GM_SUCCESS != gm_node_id_to_global_id( port, node_id, &global_id) ) {
|
|
orte_output (0, "[%s:%d] Unable to get my GM global unique id", __FILE__, __LINE__);
|
|
continue;
|
|
}
|
|
#else
|
|
if( GM_SUCCESS != gm_get_host_name( port, global_id ) ) {
|
|
orte_output( 0, "[%s:%d] Unable to get the GM host name\n", __FILE__, __LINE__);
|
|
continue;
|
|
}
|
|
#endif /* GM_API_VERSION > 0x200 */
|
|
|
|
/* create the btl module */
|
|
btl = (mca_btl_gm_module_t *)malloc( sizeof(mca_btl_gm_module_t) );
|
|
if (NULL == btl) {
|
|
orte_output( 0, "[%s:%d] out of resources", __FILE__, __LINE__);
|
|
return OMPI_ERR_OUT_OF_RESOURCE;
|
|
}
|
|
/* copy the basic informations into the new BTL */
|
|
memcpy (btl, &mca_btl_gm_module, sizeof(mca_btl_gm_module_t));
|
|
|
|
/* setup local address */
|
|
btl->port = port;
|
|
btl->gm_addr.port_id = port_no;
|
|
btl->gm_addr.node_id = node_id;
|
|
#if GM_API_VERSION > 0x200
|
|
btl->gm_addr.global_id = global_id;
|
|
#else
|
|
strncpy( btl->gm_addr.global_id, global_id, GM_MAX_HOST_NAME_LEN );
|
|
#endif /* GM_API_VERSION > 0x200 */
|
|
|
|
if(mca_btl_gm_component.gm_debug > 0) {
|
|
orte_output(0,
|
|
"%s gm_port %08lX, "
|
|
"board %" PRIu32 ", global %" PRIu32 " "
|
|
"node %" PRIu32 "port %" PRIu32 "\n",
|
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
|
|
(unsigned long) port, board_no, global_id, node_id, port_no);
|
|
}
|
|
|
|
if((rc = mca_btl_gm_module_init(btl)) != OMPI_SUCCESS) {
|
|
orte_output(0, "[%s:%d] unable to initialze gm port", __FILE__, __LINE__);
|
|
return rc;
|
|
}
|
|
|
|
/* everything is OK let's mark it as usable and go to the next one */
|
|
mca_btl_gm_component.gm_btls[mca_btl_gm_component.gm_num_btls] = btl;
|
|
if(++mca_btl_gm_component.gm_num_btls >= mca_btl_gm_component.gm_max_btls ) {
|
|
break;
|
|
}
|
|
}
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
/*
|
|
* Register GM component addressing information. The MCA framework
|
|
* will make this available to all peers.
|
|
*/
|
|
|
|
static int
|
|
mca_btl_gm_modex_send(void)
|
|
{
|
|
int rc;
|
|
size_t i;
|
|
size_t size;
|
|
mca_btl_gm_addr_t *addrs = NULL;
|
|
|
|
size = mca_btl_gm_component.gm_num_btls * sizeof (mca_btl_gm_addr_t);
|
|
if (0 != size) {
|
|
addrs = (mca_btl_gm_addr_t *)malloc (size);
|
|
if (NULL == addrs) {
|
|
return OMPI_ERR_OUT_OF_RESOURCE;
|
|
}
|
|
|
|
for (i = 0; i < mca_btl_gm_component.gm_num_btls; i++) {
|
|
mca_btl_gm_module_t *btl = mca_btl_gm_component.gm_btls[i];
|
|
addrs[i] = btl->gm_addr;
|
|
MCA_BTL_GM_ADDR_HTON(addrs[i]);
|
|
}
|
|
}
|
|
rc = ompi_modex_send (&mca_btl_gm_component.super.btl_version, addrs, size);
|
|
if (NULL != addrs) {
|
|
free (addrs);
|
|
}
|
|
return rc;
|
|
}
|
|
|
|
|
|
/*
|
|
* Initialize the GM component,
|
|
* check how many boards are available and open ports on them.
|
|
*/
|
|
|
|
mca_btl_base_module_t **
|
|
mca_btl_gm_component_init (int *num_btl_modules,
|
|
bool enable_progress_threads,
|
|
bool enable_mpi_threads)
|
|
{
|
|
mca_btl_base_module_t **btls;
|
|
*num_btl_modules = 0;
|
|
|
|
OPAL_THREAD_LOCK(&mca_btl_gm_component.gm_lock);
|
|
|
|
/* try to initialize GM */
|
|
if( GM_SUCCESS != gm_init() ) {
|
|
orte_output( 0, "[%s:%d] error in initializing the gm library\n", __FILE__, __LINE__ );
|
|
mca_btl_gm_component.gm_num_btls = 0;
|
|
mca_btl_gm_modex_send();
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
return NULL;
|
|
}
|
|
|
|
/* First discover all available boards. For each board we create a unique BTL */
|
|
mca_btl_gm_component.gm_btls = malloc( mca_btl_gm_component.gm_max_btls * sizeof (mca_btl_gm_module_t *));
|
|
if (NULL == mca_btl_gm_component.gm_btls) {
|
|
orte_output( 0, "[%s:%d] out of resources.", __FILE__, __LINE__ );
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
return NULL;
|
|
}
|
|
|
|
/* initialize gm */
|
|
if (OMPI_SUCCESS != mca_btl_gm_discover()) {
|
|
mca_btl_base_error_no_nics("Myrinet/GM", "NIC");
|
|
mca_btl_gm_component.gm_num_btls = 0;
|
|
mca_btl_gm_modex_send();
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
return NULL;
|
|
}
|
|
if (mca_btl_gm_component.gm_num_btls == 0) {
|
|
mca_btl_base_error_no_nics("Myrinet/GM", "NIC");
|
|
mca_btl_gm_component.gm_num_btls = 0;
|
|
mca_btl_gm_modex_send();
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
return NULL;
|
|
}
|
|
|
|
/* publish GM parameters with the MCA framework */
|
|
if (OMPI_SUCCESS != mca_btl_gm_modex_send()) {
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
return NULL;
|
|
}
|
|
|
|
/* return array of BTLs */
|
|
btls = (mca_btl_base_module_t**) malloc (
|
|
mca_btl_gm_component.gm_num_btls * sizeof(mca_btl_base_module_t *));
|
|
if (NULL == btls) {
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
return NULL;
|
|
}
|
|
|
|
memcpy(btls, mca_btl_gm_component.gm_btls,
|
|
mca_btl_gm_component.gm_num_btls * sizeof(mca_btl_gm_module_t *));
|
|
*num_btl_modules = mca_btl_gm_component.gm_num_btls;
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
return btls;
|
|
}
|
|
|
|
|
|
/*
|
|
* GM component progress.
|
|
*/
|
|
|
|
|
|
int mca_btl_gm_component_progress()
|
|
{
|
|
static int32_t inprogress = 0;
|
|
int count = 0;
|
|
size_t i;
|
|
|
|
/* could get into deadlock in this case as we post recvs after callback completes */
|
|
if(OPAL_THREAD_ADD32(&inprogress, 1) > 1) {
|
|
OPAL_THREAD_ADD32(&inprogress, -1);
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
OPAL_THREAD_LOCK(&mca_btl_gm_component.gm_lock);
|
|
for( i = 0; i < mca_btl_gm_component.gm_num_btls; ) {
|
|
mca_btl_gm_module_t* btl = mca_btl_gm_component.gm_btls[i];
|
|
gm_recv_event_t* event = gm_receive(btl->port);
|
|
|
|
/* If there are no receive events just skip the function call */
|
|
switch(gm_ntohc(event->recv.type)) {
|
|
case GM_FAST_RECV_EVENT:
|
|
case GM_FAST_PEER_RECV_EVENT:
|
|
case GM_FAST_HIGH_RECV_EVENT:
|
|
case GM_FAST_HIGH_PEER_RECV_EVENT:
|
|
{
|
|
unsigned char* buffer = (unsigned char*)gm_ntohp(event->recv.buffer);
|
|
mca_btl_gm_frag_t* frag = (mca_btl_gm_frag_t*)*((uintptr_t*)(buffer - sizeof(uintptr_t)));
|
|
mca_btl_base_header_t* hdr = (mca_btl_base_header_t *)gm_ntohp(event->recv.message);
|
|
mca_btl_active_message_callback_t* reg;
|
|
frag->segment.seg_addr.pval = (hdr+1);
|
|
frag->segment.seg_len = gm_ntohl(event->recv.length) - sizeof(mca_btl_base_header_t);
|
|
reg = mca_btl_base_active_message_trigger + hdr->tag;
|
|
|
|
/* cbfunc may be null if interface goes down.. */
|
|
if(reg->cbfunc) {
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
reg->cbfunc(&btl->super, hdr->tag, &frag->base, reg->cbdata);
|
|
OPAL_THREAD_LOCK(&mca_btl_gm_component.gm_lock);
|
|
} else {
|
|
btl->error_cb(&btl->super,
|
|
MCA_BTL_ERROR_FLAGS_FATAL);
|
|
return 0;
|
|
}
|
|
MCA_BTL_GM_FRAG_POST(btl,frag);
|
|
count++;
|
|
break;
|
|
}
|
|
case GM_RECV_EVENT:
|
|
case GM_PEER_RECV_EVENT:
|
|
case GM_HIGH_RECV_EVENT:
|
|
case GM_HIGH_PEER_RECV_EVENT:
|
|
{
|
|
unsigned char* buffer = (unsigned char*)gm_ntohp(event->recv.buffer);
|
|
mca_btl_gm_frag_t* frag = (mca_btl_gm_frag_t*)*((uintptr_t*)(buffer - sizeof(uintptr_t)));
|
|
mca_btl_base_header_t* hdr = (mca_btl_base_header_t*)buffer;
|
|
mca_btl_active_message_callback_t* reg;
|
|
frag->segment.seg_addr.pval = (hdr+1);
|
|
frag->segment.seg_len = gm_ntohl(event->recv.length) - sizeof(mca_btl_base_header_t);
|
|
reg = mca_btl_base_active_message_trigger + hdr->tag;
|
|
if(reg->cbfunc) {
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
reg->cbfunc(&btl->super, hdr->tag, &frag->base, reg->cbdata);
|
|
OPAL_THREAD_LOCK(&mca_btl_gm_component.gm_lock);
|
|
|
|
MCA_BTL_GM_FRAG_POST(btl,frag);
|
|
} else {
|
|
btl->error_cb(&btl->super,
|
|
MCA_BTL_ERROR_FLAGS_FATAL);
|
|
return 0;
|
|
}
|
|
count++;
|
|
break;
|
|
}
|
|
case GM_NO_RECV_EVENT:
|
|
i++;
|
|
break;
|
|
default:
|
|
gm_unknown(btl->port, event);
|
|
break;
|
|
}
|
|
}
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
OPAL_THREAD_ADD32(&inprogress, -1);
|
|
return count;
|
|
}
|
|
|
|
|
|
#if OMPI_ENABLE_PROGRESS_THREADS
|
|
static void* mca_btl_gm_progress_thread( opal_object_t* arg )
|
|
{
|
|
opal_thread_t* thread = (opal_thread_t*)arg;
|
|
mca_btl_gm_module_t* btl = thread->t_arg;
|
|
|
|
/* This thread enter in a cancel enabled state */
|
|
pthread_setcancelstate( PTHREAD_CANCEL_ENABLE, NULL );
|
|
pthread_setcanceltype( PTHREAD_CANCEL_ASYNCHRONOUS, NULL );
|
|
|
|
OPAL_THREAD_LOCK(&mca_btl_gm_component.gm_lock);
|
|
while(btl->gm_progress) {
|
|
gm_recv_event_t* event;
|
|
|
|
/* dont process events while the app is in the library */
|
|
while(opal_progress_threads()) {
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
while(opal_progress_threads())
|
|
sched_yield();
|
|
usleep(100); /* give app a chance to re-enter library */
|
|
OPAL_THREAD_LOCK(&mca_btl_gm_component.gm_lock);
|
|
}
|
|
|
|
/* otherwise processes any pending events */
|
|
event = gm_blocking_receive_no_spin(btl->port);
|
|
switch(gm_ntohc(event->recv.type)) {
|
|
case GM_FAST_RECV_EVENT:
|
|
case GM_FAST_PEER_RECV_EVENT:
|
|
case GM_FAST_HIGH_RECV_EVENT:
|
|
case GM_FAST_HIGH_PEER_RECV_EVENT:
|
|
{
|
|
unsigned char* buffer = (unsigned char*)gm_ntohp(event->recv.buffer);
|
|
mca_btl_gm_frag_t* frag = (mca_btl_gm_frag_t*)*((uintptr_t*)(buffer - sizeof(uintptr_t)));
|
|
mca_btl_base_header_t* hdr = (mca_btl_base_header_t *)gm_ntohp(event->recv.message);
|
|
mca_btl_active_message_callback_t* reg;
|
|
frag->segment.seg_addr.pval = (hdr+1);
|
|
frag->segment.seg_len = gm_ntohl(event->recv.length) - sizeof(mca_btl_base_header_t);
|
|
reg = mca_btl_base_active_message_trigger + hdr->tag;
|
|
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
reg->cbfunc(&btl->super, hdr->tag, &frag->base, reg->cbdata);
|
|
OPAL_THREAD_LOCK(&mca_btl_gm_component.gm_lock);
|
|
|
|
MCA_BTL_GM_FRAG_POST(btl,frag);
|
|
break;
|
|
}
|
|
case GM_RECV_EVENT:
|
|
case GM_PEER_RECV_EVENT:
|
|
case GM_HIGH_RECV_EVENT:
|
|
case GM_HIGH_PEER_RECV_EVENT:
|
|
{
|
|
unsigned char* buffer = (unsigned char*)gm_ntohp(event->recv.buffer);
|
|
mca_btl_gm_frag_t* frag = (mca_btl_gm_frag_t*)*((uintptr_t*)(buffer - sizeof(uintptr_t)));
|
|
mca_btl_base_header_t* hdr = (mca_btl_base_header_t*)buffer;
|
|
mca_btl_active_message_callback_t* reg;
|
|
frag->segment.seg_addr.pval = (hdr+1);
|
|
frag->segment.seg_len = gm_ntohl(event->recv.length) - sizeof(mca_btl_base_header_t);
|
|
reg = mca_btl_base_active_message_trigger + hdr->tag;
|
|
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
reg->cbfunc(&btl->super, hdr->tag, &frag->base, reg->cbdata);
|
|
OPAL_THREAD_LOCK(&mca_btl_gm_component.gm_lock);
|
|
|
|
MCA_BTL_GM_FRAG_POST(btl,frag);
|
|
break;
|
|
}
|
|
case _GM_SLEEP_EVENT:
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
gm_unknown(btl->port, event);
|
|
OPAL_THREAD_LOCK(&mca_btl_gm_component.gm_lock);
|
|
break;
|
|
default:
|
|
gm_unknown(btl->port, event);
|
|
break;
|
|
}
|
|
}
|
|
OPAL_THREAD_UNLOCK(&mca_btl_gm_component.gm_lock);
|
|
return PTHREAD_CANCELED;
|
|
}
|
|
#endif
|
|
|