1
1
openmpi/ompi/mca/io/ompio/io_ompio_component.c
Edgar Gabriel 3c10ed4ed1 common/ompio: use allocator to manage temporary buffers
use an allocator to manage temporary buffers when copying
unmanaged data from GPU buffer to host. This is necessary,
since the buffers have to be pinned for better performance,
which is an expensive operation.

Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
2018-06-21 09:25:50 -05:00

403 строки
16 KiB
C

/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* 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) 2008-2018 University of Houston. All rights reserved.
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2015-2018 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2016-2017 IBM Corporation. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "opal/class/opal_list.h"
#include "opal/threads/mutex.h"
#include "opal/mca/base/base.h"
#include "ompi/mca/io/io.h"
#include "ompi/mca/fs/base/base.h"
#include "io_ompio.h"
#include "ompi/mca/common/ompio/common_ompio_request.h"
#if OPAL_CUDA_SUPPORT
#include "ompi/mca/common/ompio/common_ompio_cuda.h"
#endif
int mca_io_ompio_cycle_buffer_size = OMPIO_DEFAULT_CYCLE_BUF_SIZE;
int mca_io_ompio_bytes_per_agg = OMPIO_PREALLOC_MAX_BUF_SIZE;
int mca_io_ompio_num_aggregators = -1;
int mca_io_ompio_record_offset_info = 0;
int mca_io_ompio_coll_timing_info = 0;
int mca_io_ompio_max_aggregators_ratio=8;
int mca_io_ompio_aggregators_cutoff_threshold=3;
int mca_io_ompio_overwrite_amode = 1;
int mca_io_ompio_verbose_info_parsing = 0;
int mca_io_ompio_grouping_option=5;
/*
* Private functions
*/
static int register_component(void);
static int open_component(void);
static int close_component(void);
static int init_query(bool enable_progress_threads,
bool enable_mpi_threads);
static const struct mca_io_base_module_2_0_0_t *
file_query (struct ompi_file_t *file,
struct mca_io_base_file_t **private_data,
int *priority);
static int file_unquery(struct ompi_file_t *file,
struct mca_io_base_file_t *private_data);
static int delete_query(const char *filename, struct opal_info_t *info,
struct mca_io_base_delete_t **private_data,
bool *usable, int *priorty);
static int delete_select(const char *filename, struct opal_info_t *info,
struct mca_io_base_delete_t *private_data);
static int register_datarep(const char *,
MPI_Datarep_conversion_function*,
MPI_Datarep_conversion_function*,
MPI_Datarep_extent_function*,
void*);
/*
static int io_progress(void);
*/
/*
* Private variables
*/
static int priority_param = 30;
static int delete_priority_param = 30;
/*
* Global, component-wide OMPIO mutex because OMPIO is not thread safe
*/
opal_mutex_t mca_io_ompio_mutex = {{0}};
/*
* Public string showing this component's version number
*/
const char *mca_io_ompio_component_version_string =
"OMPI/MPI OMPIO io MCA component version " OMPI_VERSION;
mca_io_base_component_2_0_0_t mca_io_ompio_component = {
/* First, the mca_base_component_t struct containing meta information
about the component itself */
.io_version = {
MCA_IO_BASE_VERSION_2_0_0,
.mca_component_name = "ompio",
MCA_BASE_MAKE_VERSION(component, OMPI_MAJOR_VERSION, OMPI_MINOR_VERSION,
OMPI_RELEASE_VERSION),
.mca_open_component = open_component,
.mca_close_component = close_component,
.mca_register_component_params = register_component,
},
.io_data = {
/* The component is checkpoint ready */
MCA_BASE_METADATA_PARAM_CHECKPOINT
},
/* Initial configuration / Open a new file */
.io_init_query = init_query,
.io_file_query = file_query,
.io_file_unquery = file_unquery,
/* Delete a file */
.io_delete_query = delete_query,
.io_delete_select = delete_select,
.io_register_datarep = register_datarep,
};
static int register_component(void)
{
priority_param = 30;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"priority", "Priority of the io ompio component",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&priority_param);
delete_priority_param = 30;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"delete_priority", "Delete priority of the io ompio component",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&delete_priority_param);
mca_io_ompio_record_offset_info = 0;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"record_file_offset_info",
"The information of the file offset/length",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_record_offset_info);
mca_io_ompio_coll_timing_info = 0;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"coll_timing_info",
"Enable collective algorithm timing information",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_coll_timing_info);
mca_io_ompio_cycle_buffer_size = OMPIO_DEFAULT_CYCLE_BUF_SIZE;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"cycle_buffer_size",
"Data size issued by individual reads/writes per call",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_cycle_buffer_size);
mca_io_ompio_bytes_per_agg = OMPIO_PREALLOC_MAX_BUF_SIZE;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"bytes_per_agg",
"Size of temporary buffer for collective I/O operations",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_bytes_per_agg);
mca_io_ompio_num_aggregators = -1;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"num_aggregators",
"number of aggregators for collective I/O operations",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_num_aggregators);
mca_io_ompio_grouping_option = 5;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"grouping_option",
"Option for grouping of processes in the aggregator selection "
"1: Data volume based grouping 2: maximizing group size uniformity 3: maximimze "
"data contiguity 4: hybrid optimization 5: simple (default) "
"6: skip refinement step 7: simple+: grouping based on default file view",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_grouping_option);
mca_io_ompio_max_aggregators_ratio = 8;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"max_aggregators_ratio",
"Maximum number of processes that can be an aggregator expressed as "
"the ratio to the number of process used to open the file"
" i.e 1 out of n processes can be an aggregator, with n being specified"
" by this mca parameter.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_max_aggregators_ratio);
mca_io_ompio_aggregators_cutoff_threshold=3;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"aggregators_cutoff_threshold",
"Relativ cutoff threshold for incrementing the number of aggregators "
"in the simple aggregator selection algorithm (5). Lower value "
"for this parameter will lead to higher no. of aggregators.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_aggregators_cutoff_threshold);
mca_io_ompio_overwrite_amode = 1;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"overwrite_amode",
"Overwrite WRONLY amode to RDWR to enable data sieving "
"1: allow overwrite (default) "
"0: do not overwrite amode provided by application ",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_overwrite_amode);
mca_io_ompio_verbose_info_parsing = 0;
(void) mca_base_component_var_register(&mca_io_ompio_component.io_version,
"verbose_info_parsing",
"Provide visual output when parsing info objects "
"0: no verbose output (default) "
"1: verbose output by rank 0 "
"2: verbose output by all ranks ",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&mca_io_ompio_verbose_info_parsing);
return OMPI_SUCCESS;
}
static int open_component(void)
{
/* Create the mutex */
OBJ_CONSTRUCT(&mca_io_ompio_mutex, opal_mutex_t);
mca_common_ompio_request_init ();
return mca_common_ompio_set_callbacks(ompi_io_ompio_generate_current_file_view,
mca_io_ompio_get_mca_parameter_value);
}
static int close_component(void)
{
mca_common_ompio_request_fini ();
#if OPAL_CUDA_SUPPORT
mca_common_ompio_cuda_alloc_fini();
#endif
OBJ_DESTRUCT(&mca_io_ompio_mutex);
return OMPI_SUCCESS;
}
static int init_query(bool enable_progress_threads,
bool enable_mpi_threads)
{
return OMPI_SUCCESS;
}
static const struct mca_io_base_module_2_0_0_t *
file_query(struct ompi_file_t *file,
struct mca_io_base_file_t **private_data,
int *priority)
{
mca_common_ompio_data_t *data;
char *tmp;
int rank;
int is_lustre=0; //false
tmp = strchr (file->f_filename, ':');
rank = ompi_comm_rank ( file->f_comm);
if (!tmp) {
if ( 0 == rank) {
if (LUSTRE == mca_fs_base_get_fstype(file->f_filename)) {
is_lustre = 1; //true
}
}
file->f_comm->c_coll->coll_bcast (&is_lustre,
1,
MPI_INT,
0,
file->f_comm,
file->f_comm->c_coll->coll_bcast_module);
}
else {
if (!strncasecmp(file->f_filename, "lustre:", 7) ) {
is_lustre = 1;
}
}
if (is_lustre) {
*priority = 1;
}
else {
*priority = priority_param;
}
/* Allocate a space for this module to hang private data (e.g.,
the OMPIO file handle) */
data = calloc(1, sizeof(mca_common_ompio_data_t));
if (NULL == data) {
return NULL;
}
*private_data = (struct mca_io_base_file_t*) data;
/* All done */
return &mca_io_ompio_module;
}
static int file_unquery(struct ompi_file_t *file,
struct mca_io_base_file_t *private_data)
{
/* Free the ompio module-specific data that was allocated in
_file_query(), above */
if (NULL != private_data) {
free(private_data);
}
return OMPI_SUCCESS;
}
static int delete_query(const char *filename, struct opal_info_t *info,
struct mca_io_base_delete_t **private_data,
bool *usable, int *priority)
{
*priority = delete_priority_param;
*usable = true;
*private_data = NULL;
return OMPI_SUCCESS;
}
static int delete_select(const char *filename, struct opal_info_t *info,
struct mca_io_base_delete_t *private_data)
{
int ret;
OPAL_THREAD_LOCK (&mca_io_ompio_mutex);
ret = mca_common_ompio_file_delete (filename, info);
OPAL_THREAD_UNLOCK (&mca_io_ompio_mutex);
return ret;
}
static int register_datarep(const char * datarep,
MPI_Datarep_conversion_function* read_fn,
MPI_Datarep_conversion_function* write_fn,
MPI_Datarep_extent_function* extent_fn,
void* state)
{
return OMPI_ERROR;
}
/*
static int io_progress (void)
{
return OMPI_SUCCESS;
}
*/