1
1
openmpi/ompi/mca/pml/ob1/pml_ob1.h

384 строки
14 KiB
C
Исходник Обычный вид История

/* -*- 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) 2004-2013 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
2015-06-24 06:59:57 +03:00
* 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) 2010 Oracle and/or its affiliates. All rights reserved
* Copyright (c) 2011 Sandia National Laboratories. All rights reserved.
* Copyright (c) 2012-2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
2015-06-24 06:59:57 +03:00
*
* Additional copyrights may follow
2015-06-24 06:59:57 +03:00
*
* $HEADER$
*/
/**
2015-06-24 06:59:57 +03:00
* @file
*/
2005-05-24 02:22:20 +04:00
#ifndef MCA_PML_OB1_H
#define MCA_PML_OB1_H
#include "ompi_config.h"
#include "opal/class/opal_free_list.h"
#include "ompi/request/request.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/mca/pml/base/pml_base_request.h"
#include "ompi/mca/pml/base/pml_base_bsend.h"
#include "ompi/mca/pml/base/pml_base_sendreq.h"
- Split the datatype engine into two parts: an MPI specific part in OMPI and a language agnostic part in OPAL. The convertor is completely moved into OPAL. This offers several benefits as described in RFC http://www.open-mpi.org/community/lists/devel/2009/07/6387.php namely: - Fewer basic types (int* and float* types, boolean and wchar - Fixing naming scheme to ompi-nomenclature. - Usability outside of the ompi-layer. - Due to the fixed nature of simple opal types, their information is completely known at compile time and therefore constified - With fewer datatypes (22), the actual sizes of bit-field types may be reduced from 64 to 32 bits, allowing reorganizing the opal_datatype structure, eliminating holes and keeping data required in convertor (upon send/recv) in one cacheline... This has implications to the convertor-datastructure and other parts of the code. - Several performance tests have been run, the netpipe latency does not change with this patch on Linux/x86-64 on the smoky cluster. - Extensive tests have been done to verify correctness (no new regressions) using: 1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and ompi-ddt: a. running both trunk and ompi-ddt resulted in no differences (except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run correctly). b. with --enable-memchecker and running under valgrind (one buglet when run with static found in test-suite, commited) 2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt: all passed (except for the dynamic/ tests failed!! as trunk/MTT) 3. compilation and usage of HDF5 tests on Jaguar using PGI and PathScale compilers. 4. compilation and usage on Scicortex. - Please note, that for the heterogeneous case, (-m32 compiled binaries/ompi), neither ompi-trunk, nor ompi-ddt branch would successfully launch. This commit was SVN r21641.
2009-07-13 08:56:31 +04:00
#include "ompi/datatype/ompi_datatype.h"
#include "pml_ob1_hdr.h"
#include "ompi/mca/bml/base/base.h"
#include "ompi/proc/proc.h"
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
#include "opal/mca/allocator/base/base.h"
BEGIN_C_DECLS
/**
2005-05-24 02:22:20 +04:00
* OB1 PML module
*/
struct mca_pml_ob1_t {
2015-06-24 06:59:57 +03:00
mca_pml_base_module_t super;
int priority;
int free_list_num; /* initial size of free list */
int free_list_max; /* maximum size of free list */
int free_list_inc; /* number of elements to grow free list */
size_t send_pipeline_depth;
size_t recv_pipeline_depth;
size_t rdma_retries_limit;
int max_rdma_per_request;
int max_send_per_range;
2015-06-24 06:59:57 +03:00
bool leave_pinned;
int leave_pinned_pipeline;
2015-06-24 06:59:57 +03:00
/* lock queue access */
opal_mutex_t lock;
/* free lists */
opal_free_list_t rdma_frags;
opal_free_list_t recv_frags;
opal_free_list_t pending_pckts;
opal_free_list_t buffers;
opal_free_list_t send_ranges;
/* list of pending operations */
opal_list_t pckt_pending;
opal_list_t send_pending;
opal_list_t recv_pending;
opal_list_t rdma_pending;
/* List of pending fragments without a matching communicator */
opal_list_t non_existing_communicator_pending;
2015-06-24 06:59:57 +03:00
bool enabled;
char* allocator_name;
2015-06-24 06:59:57 +03:00
mca_allocator_base_module_t* allocator;
MCA/base: Add new MCA variable system Features: - Support for an override parameter file (openmpi-mca-param-override.conf). Variable values in this file can not be overridden by any file or environment value. - Support for boolean, unsigned, and unsigned long long variables. - Support for true/false values. - Support for enumerations on integer variables. - Support for MPIT scope, verbosity, and binding. - Support for command line source. - Support for setting variable source via the environment using OMPI_MCA_SOURCE_<var name>=source (either command or file:filename) - Cleaner API. - Support for variable groups (equivalent to MPIT categories). Notes: - Variables must be created with a backing store (char **, int *, or bool *) that must live at least as long as the variable. - Creating a variable with the MCA_BASE_VAR_FLAG_SETTABLE enables the use of mca_base_var_set_value() to change the value. - String values are duplicated when the variable is registered. It is up to the caller to free the original value if necessary. The new value will be freed by the mca_base_var system and must not be freed by the user. - Variables with constant scope may not be settable. - Variable groups (and all associated variables) are deregistered when the component is closed or the component repository item is freed. This prevents a segmentation fault from accessing a variable after its component is unloaded. - After some discussion we decided we should remove the automatic registration of component priority variables. Few component actually made use of this feature. - The enumerator interface was updated to be general enough to handle future uses of the interface. - The code to generate ompi_info output has been moved into the MCA variable system. See mca_base_var_dump(). opal: update core and components to mca_base_var system orte: update core and components to mca_base_var system ompi: update core and components to mca_base_var system This commit also modifies the rmaps framework. The following variables were moved from ppr and lama: rmaps_base_pernode, rmaps_base_n_pernode, rmaps_base_n_persocket. Both lama and ppr create synonyms for these variables. This commit was SVN r28236.
2013-03-28 01:09:41 +04:00
unsigned int unexpected_limit;
};
2015-06-24 06:59:57 +03:00
typedef struct mca_pml_ob1_t mca_pml_ob1_t;
extern mca_pml_ob1_t mca_pml_ob1;
extern int mca_pml_ob1_output;
/*
* PML interface functions.
*/
extern int mca_pml_ob1_add_comm(
struct ompi_communicator_t* comm
);
extern int mca_pml_ob1_del_comm(
struct ompi_communicator_t* comm
);
extern int mca_pml_ob1_add_procs(
struct ompi_proc_t **procs,
size_t nprocs
);
extern int mca_pml_ob1_del_procs(
struct ompi_proc_t **procs,
size_t nprocs
);
extern int mca_pml_ob1_enable( bool enable );
extern int mca_pml_ob1_progress(void);
extern int mca_pml_ob1_iprobe( int dst,
int tag,
struct ompi_communicator_t* comm,
int *matched,
ompi_status_public_t* status );
extern int mca_pml_ob1_probe( int dst,
int tag,
struct ompi_communicator_t* comm,
ompi_status_public_t* status );
extern int mca_pml_ob1_improbe( int dst,
int tag,
struct ompi_communicator_t* comm,
int *matched,
struct ompi_message_t **message,
ompi_status_public_t* status );
extern int mca_pml_ob1_mprobe( int dst,
int tag,
struct ompi_communicator_t* comm,
struct ompi_message_t **message,
ompi_status_public_t* status );
extern 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 mode,
struct ompi_communicator_t* comm,
struct ompi_request_t **request );
extern 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 mode,
struct ompi_communicator_t* comm,
struct ompi_request_t **request );
extern 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 mode,
struct ompi_communicator_t* comm );
extern int mca_pml_ob1_irecv_init( void *buf,
size_t count,
ompi_datatype_t *datatype,
int src,
int tag,
struct ompi_communicator_t* comm,
struct ompi_request_t **request );
extern int mca_pml_ob1_irecv( void *buf,
size_t count,
ompi_datatype_t *datatype,
int src,
int tag,
struct ompi_communicator_t* comm,
struct ompi_request_t **request );
extern int mca_pml_ob1_recv( void *buf,
size_t count,
ompi_datatype_t *datatype,
int src,
int tag,
struct ompi_communicator_t* comm,
ompi_status_public_t* status );
extern int mca_pml_ob1_imrecv( void *buf,
size_t count,
ompi_datatype_t *datatype,
struct ompi_message_t **message,
struct ompi_request_t **request );
extern int mca_pml_ob1_mrecv( void *buf,
size_t count,
ompi_datatype_t *datatype,
struct ompi_message_t **message,
ompi_status_public_t* status );
extern int mca_pml_ob1_dump( struct ompi_communicator_t* comm,
int verbose );
extern int mca_pml_ob1_start( size_t count,
ompi_request_t** requests );
extern int mca_pml_ob1_ft_event( int state );
/**
* We will use these requests to hold on a traditionally allocated
* requests in order to allow the parallel debugger full access to the
* message queues (instead of allocating the requests on the stack).
*/
extern struct mca_pml_ob1_recv_request_t *mca_pml_ob1_recvreq;
extern struct mca_pml_ob1_send_request_t *mca_pml_ob1_sendreq;
END_C_DECLS
struct mca_pml_ob1_pckt_pending_t {
opal_free_list_item_t super;
ompi_proc_t* proc;
mca_pml_ob1_hdr_t hdr;
struct mca_bml_base_btl_t *bml_btl;
uint8_t order;
int status;
};
typedef struct mca_pml_ob1_pckt_pending_t mca_pml_ob1_pckt_pending_t;
OBJ_CLASS_DECLARATION(mca_pml_ob1_pckt_pending_t);
#define MCA_PML_OB1_PCKT_PENDING_ALLOC(pckt) \
do { \
pckt = (mca_pml_ob1_pckt_pending_t *) \
opal_free_list_get (&mca_pml_ob1.pending_pckts); \
} while (0)
#define MCA_PML_OB1_PCKT_PENDING_RETURN(pckt) \
do { \
/* return packet */ \
opal_free_list_return (&mca_pml_ob1.pending_pckts, \
(opal_free_list_item_t*)pckt); \
} while(0)
#define MCA_PML_OB1_ADD_FIN_TO_PENDING(P, D, Sz, B, O, S) \
do { \
mca_pml_ob1_pckt_pending_t *_pckt; \
\
MCA_PML_OB1_PCKT_PENDING_ALLOC(_pckt); \
mca_pml_ob1_fin_hdr_prepare (&_pckt->hdr.hdr_fin, 0, \
(D).lval, (Sz)); \
_pckt->proc = (P); \
_pckt->bml_btl = (B); \
_pckt->order = (O); \
_pckt->status = (S); \
OPAL_THREAD_LOCK(&mca_pml_ob1.lock); \
opal_list_append(&mca_pml_ob1.pckt_pending, \
(opal_list_item_t*)_pckt); \
OPAL_THREAD_UNLOCK(&mca_pml_ob1.lock); \
} while(0)
2015-06-24 06:59:57 +03:00
int mca_pml_ob1_send_fin(ompi_proc_t* proc, mca_bml_base_btl_t* bml_btl,
opal_ptr_t hdr_frag, uint64_t size, uint8_t order, int status);
/* This function tries to resend FIN/ACK packets from pckt_pending queue.
* Packets are added to the queue when sending of FIN or ACK is failed due to
* resource unavailability. bml_btl passed to the function doesn't represents
* packet's destination, it represents BTL on which resource was freed, so only
* this BTL should be considered for resending packets */
void mca_pml_ob1_process_pending_packets(mca_bml_base_btl_t* bml_btl);
/* This function retries failed PUT/GET operations on frag. When RDMA operation
* cannot be accomplished for some reason, frag is put on the rdma_pending list.
* Later the operation is retried. The destination of RDMA operation is stored
* inside the frag structure */
void mca_pml_ob1_process_pending_rdma(void);
#define MCA_PML_OB1_PROGRESS_PENDING(bml_btl) \
do { \
if(opal_list_get_size(&mca_pml_ob1.pckt_pending)) \
mca_pml_ob1_process_pending_packets(bml_btl); \
if(opal_list_get_size(&mca_pml_ob1.recv_pending)) \
mca_pml_ob1_recv_request_process_pending(); \
if(opal_list_get_size(&mca_pml_ob1.send_pending)) \
mca_pml_ob1_send_request_process_pending(bml_btl); \
if(opal_list_get_size(&mca_pml_ob1.rdma_pending)) \
mca_pml_ob1_process_pending_rdma(); \
} while (0)
/*
* 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)
{
size_t i, length = 0;
for (i = 0; i < count ; ++i) {
length += segments[i].seg_len;
}
return (length - hdrlen);
}
static inline size_t
mca_pml_ob1_compute_segment_length_remote (size_t seg_size, void *segments,
size_t count, ompi_proc_t *rem_proc)
{
mca_btl_base_segment_t *segment = (mca_btl_base_segment_t *) segments;
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
ompi_proc_t *local_proc = ompi_proc_local();
#endif
size_t i, length = 0;
for (i = 0 ; i < count ; ++i) {
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 04:47:28 +04:00
if ((rem_proc->super.proc_arch & OPAL_ARCH_ISBIGENDIAN) !=
(local_proc->super.proc_arch & OPAL_ARCH_ISBIGENDIAN))
/* NTH: seg_len is always 64-bit so use swap_bytes8 */
length += opal_swap_bytes8(segment->seg_len);
else
#endif
length += segment->seg_len;
segment = (mca_btl_base_segment_t *)((char *)segment + seg_size);
}
return length;
}
/* represent BTL chosen for sending request */
struct mca_pml_ob1_com_btl_t {
mca_bml_base_btl_t *bml_btl;
struct mca_btl_base_registration_handle_t *btl_reg;
size_t length;
};
typedef struct mca_pml_ob1_com_btl_t mca_pml_ob1_com_btl_t;
int mca_pml_ob1_com_btl_comp(const void *v1, const void *v2);
/* Calculate what percentage of a message to send through each BTL according to
* relative weight */
static inline void
mca_pml_ob1_calc_weighted_length( mca_pml_ob1_com_btl_t *btls, int num_btls, size_t size,
double weight_total )
{
int i;
size_t length_left;
/* shortcut for common case for only one BTL */
if( OPAL_LIKELY(1 == num_btls) ) {
btls[0].length = size;
return;
}
/* sort BTLs according of their weights so BTLs with smaller weight will
* not hijack all of the traffic */
qsort( btls, num_btls, sizeof(mca_pml_ob1_com_btl_t),
mca_pml_ob1_com_btl_comp );
for(length_left = size, i = 0; i < num_btls; i++) {
mca_bml_base_btl_t* bml_btl = btls[i].bml_btl;
size_t length = 0;
if( OPAL_UNLIKELY(0 != length_left) ) {
length = (length_left > bml_btl->btl->btl_eager_limit)?
((size_t)(size * (bml_btl->btl_weight / weight_total))) :
length_left;
if(length > length_left)
length = length_left;
length_left -= length;
}
btls[i].length = length;
}
/* account for rounding errors */
btls[0].length += length_left;
}
#endif