1
1
openmpi/ompi/mca/bml/bml.h

582 строки
20 KiB
C
Исходник Обычный вид История

/*
* Copyright (c) 2004-2007 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) 2004-2006 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2008 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2012 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file
*
* BML Management Layer (BML)
*
*/
#ifndef MCA_BML_H
#define MCA_BML_H
#include "ompi_config.h"
#include "opal/mca/mca.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 "opal/datatype/opal_convertor.h"
#include "opal/mca/crs/crs.h"
#include "opal/mca/crs/base/base.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/btl/btl.h"
#include "ompi/mca/bml/base/bml_base_btl.h"
#include "ompi/types.h"
#include "ompi/constants.h"
#define OPAL_ENABLE_DEBUG_RELIABILITY 0
/*
* BML types
*/
struct ompi_proc_t;
struct mca_bml_base_module_t;
struct mca_bml_base_endpoint_t;
struct mca_mpool_base_resources_t;
/*
* Cached set of information for each btl
*/
struct mca_bml_base_btl_t {
uint32_t btl_flags; /**< support for put/get? */
float btl_weight; /**< BTL weight for scheduling */
struct mca_btl_base_module_t *btl; /**< BTL module */
struct mca_btl_base_endpoint_t* btl_endpoint; /**< BTL addressing info */
};
typedef struct mca_bml_base_btl_t mca_bml_base_btl_t;
/**
* A dynamically growable array of mca_bml_base_btl_t instances.
* Maintains an index into the array that is used for round-robin
* scheduling across contents.
*/
struct mca_bml_base_btl_array_t {
opal_object_t super;
size_t arr_size; /**< number available */
size_t arr_reserve; /**< size of allocated btl_proc array */
size_t arr_index; /**< last used index*/
mca_bml_base_btl_t* bml_btls; /**< array of bml btl's */
};
typedef struct mca_bml_base_btl_array_t mca_bml_base_btl_array_t;
OMPI_DECLSPEC OBJ_CLASS_DECLARATION(mca_bml_base_btl_array_t);
/**
* If required, reallocate (grow) the array to the indicate size.
*
* @param array (IN)
* @param size (IN)
*/
static inline size_t mca_bml_base_btl_array_get_size(mca_bml_base_btl_array_t* array)
{
return array->arr_size;
}
/**
* Grow the array if required, and set the size.
*
* @param array (IN)
* @param size (IN)
*/
static inline void mca_bml_base_btl_array_set_size(mca_bml_base_btl_array_t* array, size_t size)
{
if(array->arr_size > array->arr_reserve)
mca_bml_base_btl_array_reserve(array, size);
array->arr_size = size;
}
/**
* Grow the array size by one and return the item at that index.
*
* @param array (IN)
*/
static inline mca_bml_base_btl_t* mca_bml_base_btl_array_insert(mca_bml_base_btl_array_t* array)
{
#if OPAL_ENABLE_DEBUG
if(array->arr_size >= array->arr_reserve) {
opal_output(0, "mca_bml_base_btl_array_insert: invalid array index %lu >= %lu",
(unsigned long)array->arr_size, (unsigned long)array->arr_reserve);
return 0;
}
#endif
return &array->bml_btls[array->arr_size++];
}
/**
* Remove a btl from a bml_btl
*
* @param array (IN)
* @param btl (IN)
*/
static inline bool mca_bml_base_btl_array_remove( mca_bml_base_btl_array_t* array,
struct mca_btl_base_module_t* btl )
{
size_t i = 0;
/* find the btl */
for( i = 0; i < array->arr_size; i++ ) {
if( array->bml_btls[i].btl == btl ) {
/* make sure not to go out of bounds */
for( ; i < array->arr_size-1; i++ ) {
/* move all btl's back by 1, so the found
btl is "removed" */
array->bml_btls[i] = array->bml_btls[(i+1)];
}
array->arr_size--;
array->arr_index = 0;
return true;
}
}
return false;
}
/**
* Return an array item at the specified index.
*
* @param array (IN)
* @param item_index (IN)
*/
static inline mca_bml_base_btl_t* mca_bml_base_btl_array_get_index(mca_bml_base_btl_array_t* array, size_t item_index)
{
#if OPAL_ENABLE_DEBUG
if(item_index >= array->arr_size) {
opal_output(0, "mca_bml_base_btl_array_get_index: invalid array index %lu >= %lu",
(unsigned long)item_index, (unsigned long)array->arr_size);
return 0;
}
#endif
return &array->bml_btls[item_index];
}
/**
* Return the next LRU index in the array.
*
* @param array (IN)
*
* @param index (OUT)
*/
static inline mca_bml_base_btl_t* mca_bml_base_btl_array_get_next(mca_bml_base_btl_array_t* array)
{
#if OPAL_ENABLE_DEBUG
if(array->arr_size == 0) {
opal_output(0, "mca_bml_base_btl_array_get_next: invalid array size");
return 0;
}
#endif
if( 1 == array->arr_size ) {
return &array->bml_btls[0]; /* force the return to avoid a jump */
} else {
size_t current_position = array->arr_index; /* force to always start from zero */
if( (current_position + 1) == array->arr_size ) {
array->arr_index = 0; /* next time serve from the beginning */
} else {
array->arr_index = current_position + 1; /* continue */
}
return &array->bml_btls[current_position];
}
}
/**
* Locate an element in the array
*
* @param array (IN)
* @param index (IN)
*/
static inline mca_bml_base_btl_t* mca_bml_base_btl_array_find(
mca_bml_base_btl_array_t* array, struct mca_btl_base_module_t* btl)
{
size_t i=0;
for(i=0; i<array->arr_size; i++) {
if(array->bml_btls[i].btl == btl) {
return &array->bml_btls[i];
}
}
return NULL;
}
/**
* Structure associated w/ ompi_proc_t that contains the set
* of BTLs used to reach a destination
*/
struct mca_bml_base_endpoint_t {
opal_list_item_t super; /**< base_endpoint is a list item */
struct ompi_proc_t* btl_proc; /**< backpointer to target ompi_proc_t */
size_t btl_pipeline_send_length; /**< max of pipeline send_length of available BTLs */
size_t btl_send_limit; /**< max of min rdma pipeline for available rmda btls */
size_t btl_max_send_size; /**< min of max send size for available send btls */
mca_bml_base_btl_array_t btl_eager; /**< array of btls to use for first fragments */
mca_bml_base_btl_array_t btl_send; /**< array of btls to use for remaining fragments */
mca_bml_base_btl_array_t btl_rdma; /**< array of btls that support (prefer) rdma */
size_t btl_rdma_index; /**< index of last used BTL for RDMA */
uint32_t btl_flags_or; /**< the bitwise OR of the btl flags */
};
typedef struct mca_bml_base_endpoint_t mca_bml_base_endpoint_t;
OMPI_DECLSPEC OBJ_CLASS_DECLARATION(mca_bml_base_endpoint_t);
static inline void mca_bml_base_alloc( mca_bml_base_btl_t* bml_btl,
mca_btl_base_descriptor_t** des,
uint8_t order, size_t size, uint32_t flags )
{
mca_btl_base_module_t* btl = bml_btl->btl;
*des = btl->btl_alloc(btl, bml_btl->btl_endpoint, order, size, flags);
}
static inline void mca_bml_base_free( mca_bml_base_btl_t* bml_btl,
mca_btl_base_descriptor_t* des )
{
mca_btl_base_module_t* btl = bml_btl->btl;
btl->btl_free( btl, des );
/* The previous function is supposed to release the des object
* so we should not touch it anymore.
*/
}
#if OPAL_ENABLE_DEBUG_RELIABILITY
int mca_bml_base_send( mca_bml_base_btl_t* bml_btl,
mca_btl_base_descriptor_t* des,
mca_btl_base_tag_t tag );
#else
static inline int mca_bml_base_send( mca_bml_base_btl_t* bml_btl,
mca_btl_base_descriptor_t* des,
mca_btl_base_tag_t tag )
{
int rc;
mca_btl_base_module_t* btl = bml_btl->btl;
des->des_context = (void*) bml_btl;
rc = btl->btl_send(btl, bml_btl->btl_endpoint, des, tag);
if (rc == OMPI_ERR_RESOURCE_BUSY)
rc = OMPI_SUCCESS;
return rc;
}
#endif
static inline int mca_bml_base_send_status( mca_bml_base_btl_t* bml_btl,
mca_btl_base_descriptor_t* des,
mca_btl_base_tag_t tag )
{
mca_btl_base_module_t* btl = bml_btl->btl;
des->des_context = (void*) bml_btl;
return btl->btl_send(btl, bml_btl->btl_endpoint, des, tag);
}
static inline int mca_bml_base_sendi( mca_bml_base_btl_t* bml_btl,
- 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
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_base_module_t* btl = bml_btl->btl;
return btl->btl_sendi(btl, bml_btl->btl_endpoint,
convertor, header, header_size,
payload_size, order, flags, tag, descriptor);
}
static inline int mca_bml_base_put( mca_bml_base_btl_t* bml_btl,
mca_btl_base_descriptor_t* des)
{
mca_btl_base_module_t* btl = bml_btl->btl;
des->des_context = (void*) bml_btl;
return btl->btl_put( btl, bml_btl->btl_endpoint, des );
}
static inline int mca_bml_base_get( mca_bml_base_btl_t* bml_btl,
mca_btl_base_descriptor_t* des)
{
mca_btl_base_module_t* btl = bml_btl->btl;
des->des_context = (void*) bml_btl;
return btl->btl_get( btl, bml_btl->btl_endpoint, des );
}
static inline void mca_bml_base_prepare_src(mca_bml_base_btl_t* bml_btl,
mca_mpool_base_registration_t* reg,
- 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
struct opal_convertor_t* conv,
uint8_t order,
size_t reserve,
size_t *size,
uint32_t flags,
mca_btl_base_descriptor_t** des)
{
mca_btl_base_module_t* btl = bml_btl->btl;
*des = btl->btl_prepare_src( btl, bml_btl->btl_endpoint, reg, conv,
order, reserve, size, flags );
if( OPAL_LIKELY((*des) != NULL) ) {
(*des)->des_context = (void*) bml_btl;
}
}
static inline void mca_bml_base_prepare_dst(mca_bml_base_btl_t* bml_btl,
mca_mpool_base_registration_t* reg,
struct opal_convertor_t* conv,
uint8_t order,
size_t reserve,
size_t *size,
uint32_t flags,
mca_btl_base_descriptor_t** des)
{
2014-10-15 00:25:28 +04:00
mca_btl_base_module_t* btl = bml_btl->btl;
*des = btl->btl_prepare_dst( btl, bml_btl->btl_endpoint, reg, conv,
order, reserve, size, flags );
if( OPAL_LIKELY((*des) != NULL) ) {
(*des)->des_context = (void*) bml_btl;
}
}
/*
* BML component interface functions and datatype.
*/
/**
* MCA->BML Initializes the BML component and creates specific BML
* module(s).
*
* @param num_bmls (OUT) Returns the number of bml modules created, or 0
* if the transport is not available.
*
* @param enable_progress_threads (IN) Whether this component is
* allowed to run a hidden/progress thread or not.
*
* @param enable_mpi_threads (IN) Whether support for multiple MPI
* threads is enabled or not (i.e., MPI_THREAD_MULTIPLE), which
* indicates whether multiple threads may invoke this component
* simultaneously or not.
*
* @return Array of pointers to BML modules, or NULL if the transport
* is not available.
*
* During component initialization, the BML component should discover
* the physical devices that are available for the given transport,
* and create a BML module to represent each device. Any addressing
* information required by peers to reach the device should be published
* during this function via the mca_base_modex_send() interface.
*
*/
typedef struct mca_bml_base_module_t* (*mca_bml_base_component_init_fn_t)(
int* priority,
bool enable_progress_threads,
bool enable_mpi_threads
);
/**
* BML component descriptor. Contains component version information
* and component open/close/init functions.
*/
struct mca_bml_base_component_2_0_0_t {
mca_base_component_t bml_version;
mca_base_component_data_t bml_data;
mca_bml_base_component_init_fn_t bml_init;
};
typedef struct mca_bml_base_component_2_0_0_t mca_bml_base_component_2_0_0_t;
typedef struct mca_bml_base_component_2_0_0_t mca_bml_base_component_t;
/*
* BML module interface functions and datatype.
*/
/**
* MCA->BML Clean up any resources held by BML module
* before the module is unloaded.
*
* @param bml (IN) BML module.
*
* Prior to unloading a BML module, the MCA framework will call
* the BML finalize method of the module. Any resources held by
* the BML should be released and if required the memory corresponding
* to the BML module freed.
*
*/
typedef int (*mca_bml_base_module_finalize_fn_t)( void );
/**
* PML->BML notification of change in the process list.
*
* @param nprocs (IN) Number of processes
* @param procs (IN) Set of processes
* @param reachable (OUT) Bitmask indicating set of peer processes that are reachable by this BML.
* @return OMPI_SUCCESS or error status on failure.
*
* The mca_bml_base_module_add_procs_fn_t() is called by the PML to
* determine the set of BMLs that should be used to reach each process.
* Any addressing information exported by the peer via the mca_base_modex_send()
* function should be available during this call via the corresponding
* mca_base_modex_recv() function. The BML may utilize this information to
* determine reachability of each peer process.
*
* For each process that is reachable by the BML, the bit corresponding to the index
* into the proc array (nprocs) should be set in the reachable bitmask. The PML
* provides the BML the option to return a pointer to a data structure defined
* by the BML that is returned to the BML on subsequent calls to the BML data
* transfer functions (e.g bml_send). This may be used by the BML to cache any addressing
* or connection information (e.g. TCP socket, IP queue pair).
*
* \note This function will return OMPI_ERR_UNREACH if one or more
* processes can not be reached by the currently active BTLs. This is
* not a fatal error, and the calling layer is free to continue using
* the BML interface.
*/
typedef int (*mca_bml_base_module_add_procs_fn_t)(
size_t nprocs,
struct ompi_proc_t** procs,
struct opal_bitmap_t* reachable
);
/**
* Notification of change to the process list.
*
* @param nprocs (IN) Number of processes
* @param proc (IN) Set of processes
* @return Status indicating if cleanup was successful
*
* When the process list changes, the PML notifies the BML of the
* change, to provide the opportunity to cleanup or release any
* resources associated with the peer.
*/
typedef int (*mca_bml_base_module_del_procs_fn_t)(
size_t nprocs,
struct ompi_proc_t** procs
);
/**
* Notification of change to the btl list.
*
* @param bml (IN) BTL module
* @return Status indicating if cleanup was successful
*
* On recovery of a btl, add it to the set of forwarding
* entries used by the BML.
*/
typedef int (*mca_bml_base_module_add_btl_fn_t)( struct mca_btl_base_module_t* );
/**
* Notification of change to the btl list.
*
* @param bml (IN) BTL module
* @return Status indicating if cleanup was successful
*
* On failure of a btl, remove it from the set of forwarding
* entries used by the BML.
*/
typedef int (*mca_bml_base_module_del_btl_fn_t)( struct mca_btl_base_module_t* );
/**
* Notification of change to the btl list.
*
* @param bml (IN) BTL module
* @return Status indicating if cleanup was successful
*
* On failure of a btl, remove it from the set of forwarding
* entries used by the BML.
*/
typedef int (*mca_bml_base_module_del_proc_btl_fn_t)(
struct ompi_proc_t*,
struct mca_btl_base_module_t* );
/**
* Register a callback function that is called on receipt
* of a fragment.
*
* @param bml (IN) BML module
* @return Status indicating if cleanup was successful
*
* When the process list changes, the PML notifies the BML of the
* change, to provide the opportunity to cleanup or release any
* resources associated with the peer.
*/
typedef int (*mca_bml_base_module_register_fn_t)(
mca_btl_base_tag_t tag,
mca_btl_base_module_recv_cb_fn_t cbfunc,
void* cbdata
);
/**
* Register a callback function that is called of error.
*
* @param bml (IN) BML module
* @return Status indicating if cleanup was successful
*
*/
typedef int (*mca_bml_base_module_register_error_cb_fn_t)(
mca_btl_base_module_error_cb_fn_t cbfunc
);
/**
* Fault Tolerance Event Notification Function
* @param status Checkpoint Status
* @return OMPI_SUCCESS or failure status
*/
typedef int (*mca_bml_base_module_ft_event_fn_t)(int status);
/**
* BML module interface functions and attributes.
*/
struct mca_bml_base_module_t {
/* BML common attributes */
mca_bml_base_component_t* bml_component; /**< pointer back to the BML component structure */
/* BML function table */
mca_bml_base_module_add_procs_fn_t bml_add_procs;
mca_bml_base_module_del_procs_fn_t bml_del_procs;
mca_bml_base_module_add_btl_fn_t bml_add_btl;
mca_bml_base_module_del_btl_fn_t bml_del_btl;
mca_bml_base_module_del_proc_btl_fn_t bml_del_proc_btl;
mca_bml_base_module_register_fn_t bml_register;
mca_bml_base_module_register_error_cb_fn_t bml_register_error;
mca_bml_base_module_finalize_fn_t bml_finalize;
mca_bml_base_module_ft_event_fn_t bml_ft_event;
};
typedef struct mca_bml_base_module_t mca_bml_base_module_t;
/*
* Macro for use in modules that are of type bml
*/
#define MCA_BML_BASE_VERSION_2_0_0 \
MCA_BASE_VERSION_2_0_0, \
"bml", 2, 0, 0
#endif /* OMPI_MCA_BML_H */