ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

io/ompio: add a new grouping option avoiding communication

Просмотр исходного кода 
the new grouping option simple+ performs all calculations used
for the aggregator selection as if the default file view would be used,
thus avoiding communication in file_set_view all together. This mode
is useful for applications that do not set a file view, but use
explicit offset operations on the default file view.

Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
Этот коммит содержится в:
Edgar Gabriel 2017-09-18 12:30:34 -05:00
родитель 5cb5eb68f5
Коммит 76a8c67575
4 изменённых файлов: 73 добавлений и 59 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

123
ompi/mca/common/ompio/common_ompio_file_view.c
Просмотреть файл

@ -30,7 +30,7 @@
#include "ompi/mca/fcoll/base/base.h"
#include "ompi/mca/topo/topo.h"
static OMPI_MPI_OFFSET_TYPE get_contiguous_chunk_size (mca_io_ompio_file_t *);
static OMPI_MPI_OFFSET_TYPE get_contiguous_chunk_size (mca_io_ompio_file_t *, int flag);
static int datatype_duplicate (ompi_datatype_t *oldtype, ompi_datatype_t **newtype );
static int datatype_duplicate (ompi_datatype_t *oldtype, ompi_datatype_t **newtype )
{
@ -139,7 +139,13 @@ int mca_common_ompio_set_view (mca_io_ompio_file_t *fh,
// in orig_file type, No need to set args on this one.
ompi_datatype_duplicate (newfiletype, &fh->f_filetype);
fh->f_cc_size = get_contiguous_chunk_size (fh);
if( SIMPLE_PLUS == mca_io_ompio_grouping_option ) {
fh->f_cc_size = get_contiguous_chunk_size (fh, 1);
}
else {
fh->f_cc_size = get_contiguous_chunk_size (fh, 0);
}
if (opal_datatype_is_contiguous_memory_layout(&etype->super,1)) {
if (opal_datatype_is_contiguous_memory_layout(&filetype->super,1) &&
@ -166,7 +172,7 @@ int mca_common_ompio_set_view (mca_io_ompio_file_t *fh,
}
}
if ( SIMPLE != mca_io_ompio_grouping_option ) {
if ( SIMPLE != mca_io_ompio_grouping_option || SIMPLE_PLUS != mca_io_ompio_grouping_option ) {
ret = mca_io_ompio_fview_based_grouping(fh,
&num_groups,
@ -179,6 +185,7 @@ int mca_common_ompio_set_view (mca_io_ompio_file_t *fh,
else {
int done=0;
int ndims;
if ( fh->f_comm->c_flags & OMPI_COMM_CART ){
ret = fh->f_comm->c_topo->topo.cart.cartdim_get( fh->f_comm, &ndims);
if ( OMPI_SUCCESS != ret ){
@ -253,7 +260,7 @@ exit:
return ret;
}
OMPI_MPI_OFFSET_TYPE get_contiguous_chunk_size (mca_io_ompio_file_t *fh)
OMPI_MPI_OFFSET_TYPE get_contiguous_chunk_size (mca_io_ompio_file_t *fh, int flag)
{
int uniform = 0;
OMPI_MPI_OFFSET_TYPE avg[3] = {0,0,0};
@ -268,60 +275,66 @@ OMPI_MPI_OFFSET_TYPE get_contiguous_chunk_size (mca_io_ompio_file_t *fh)
** 2. each section in the file view has exactly the same size
*/
for (i=0 ; i<(int)fh->f_iov_count ; i++) {
avg[0] += fh->f_decoded_iov[i].iov_len;
if (i && 0 == uniform) {
if (fh->f_decoded_iov[i].iov_len != fh->f_decoded_iov[i-1].iov_len) {
uniform = 1;
}
}
}
if ( 0 != fh->f_iov_count ) {
avg[0] = avg[0]/fh->f_iov_count;
}
avg[1] = (OMPI_MPI_OFFSET_TYPE) fh->f_iov_count;
avg[2] = (OMPI_MPI_OFFSET_TYPE) uniform;
fh->f_comm->c_coll->coll_allreduce (avg,
global_avg,
3,
OMPI_OFFSET_DATATYPE,
MPI_SUM,
fh->f_comm,
fh->f_comm->c_coll->coll_allreduce_module);
global_avg[0] = global_avg[0]/fh->f_size;
global_avg[1] = global_avg[1]/fh->f_size;
#if 0
/* Disabling the feature since we are not using it anyway. Saves us one allreduce operation. */
int global_uniform=0;
if ( global_avg[0] == avg[0] &&
global_avg[1] == avg[1] &&
0 == avg[2] &&
0 == global_avg[2] ) {
uniform = 0;
if ( flag ) {
global_avg[0] = MCA_IO_DEFAULT_FILE_VIEW_SIZE;
}
else {
uniform = 1;
}
/* second confirmation round to see whether all processes agree
** on having a uniform file view or not
*/
fh->f_comm->c_coll->coll_allreduce (&uniform,
&global_uniform,
1,
MPI_INT,
MPI_MAX,
fh->f_comm,
fh->f_comm->c_coll->coll_allreduce_module);
if ( 0 == global_uniform ){
/* yes, everybody agrees on having a uniform file view */
fh->f_flags |= OMPIO_UNIFORM_FVIEW;
}
for (i=0 ; i<(int)fh->f_iov_count ; i++) {
avg[0] += fh->f_decoded_iov[i].iov_len;
if (i && 0 == uniform) {
if (fh->f_decoded_iov[i].iov_len != fh->f_decoded_iov[i-1].iov_len) {
uniform = 1;
}
}
}
if ( 0 != fh->f_iov_count ) {
avg[0] = avg[0]/fh->f_iov_count;
}
avg[1] = (OMPI_MPI_OFFSET_TYPE) fh->f_iov_count;
avg[2] = (OMPI_MPI_OFFSET_TYPE) uniform;
fh->f_comm->c_coll->coll_allreduce (avg,
global_avg,
3,
OMPI_OFFSET_DATATYPE,
MPI_SUM,
fh->f_comm,
fh->f_comm->c_coll->coll_allreduce_module);
global_avg[0] = global_avg[0]/fh->f_size;
global_avg[1] = global_avg[1]/fh->f_size;
#if 0
/* Disabling the feature since we are not using it anyway. Saves us one allreduce operation. */
int global_uniform=0;
if ( global_avg[0] == avg[0] &&
global_avg[1] == avg[1] &&
0 == avg[2] &&
0 == global_avg[2] ) {
uniform = 0;
}
else {
uniform = 1;
}
/* second confirmation round to see whether all processes agree
** on having a uniform file view or not
*/
fh->f_comm->c_coll->coll_allreduce (&uniform,
&global_uniform,
1,
MPI_INT,
MPI_MAX,
fh->f_comm,
fh->f_comm->c_coll->coll_allreduce_module);
if ( 0 == global_uniform ){
/* yes, everybody agrees on having a uniform file view */
fh->f_flags |= OMPIO_UNIFORM_FVIEW;
}
#endif
}
return global_avg[0];
}

2
ompi/mca/io/ompio/io_ompio.h
Просмотреть файл

@ -111,7 +111,7 @@ OMPI_DECLSPEC extern int mca_io_ompio_coll_timing_info;
#define OPTIMIZE_GROUPING 4
#define SIMPLE 5
#define NO_REFINEMENT 6
#define SIMPLE_PLUS 7
#define OMPIO_UNIFORM_DIST_THRESHOLD 0.5
#define OMPIO_CONTG_THRESHOLD 1048576

5
ompi/mca/io/ompio/io_ompio_aggregators.c
Просмотреть файл

@ -497,8 +497,9 @@ int mca_io_ompio_set_aggregator_props (struct mca_io_ompio_file_t *fh,
fh->f_flags |= OMPIO_AGGREGATOR_IS_SET;
if (-1 == num_aggregators) {
if ( SIMPLE == mca_io_ompio_grouping_option ||
NO_REFINEMENT == mca_io_ompio_grouping_option ) {
if ( SIMPLE == mca_io_ompio_grouping_option ||
NO_REFINEMENT == mca_io_ompio_grouping_option ||
SIMPLE_PLUS == mca_io_ompio_grouping_option ) {
fh->f_aggregator_index = 0;
fh->f_final_num_aggrs = fh->f_init_num_aggrs;
fh->f_procs_per_group = fh->f_init_procs_per_group;

2
ompi/mca/io/ompio/io_ompio_component.c
Просмотреть файл

@ -212,7 +212,7 @@ static int register_component(void)
"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",
"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,