1
1
openmpi/ompi/mca/sharedfp/sm/sharedfp_sm_read.c
Edgar Gabriel bc0f60dfd9 sharedfp/all components: revamp internal operations
this commit revamps the internal operations of the sharedfp components.
Specifically, it is focused around removing the second file_open
operation for shared file pointers. This makes the code more efficient.
Because of that, there is no necessity anymore for the sharedfp_lazy_open
mca parameter.

Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
2018-06-18 14:34:05 -05:00

183 строки
6.4 KiB
C

/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2017 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) 2013-2018 University of Houston. All rights reserved.
* Copyright (c) 2018 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "sharedfp_sm.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/mca/sharedfp/sharedfp.h"
#include "ompi/mca/sharedfp/base/base.h"
int mca_sharedfp_sm_read ( ompio_file_t *fh,
void *buf, int count, MPI_Datatype datatype, MPI_Status *status)
{
int ret = OMPI_SUCCESS;
OMPI_MPI_OFFSET_TYPE offset = 0;
long bytesRequested = 0;
size_t numofBytes;
if( NULL == fh->f_sharedfp_data){
opal_output(ompi_sharedfp_base_framework.framework_output,
"sharedfp_sm_read - module not initialized \n");
return OMPI_ERROR;
}
/* Calculate the number of bytes to write */
opal_datatype_type_size ( &datatype->super, &numofBytes);
bytesRequested = count * numofBytes;
if ( mca_sharedfp_sm_verbose ) {
opal_output(ompi_sharedfp_base_framework.framework_output,
"sharedfp_sm_read: Bytes Requested is %ld\n",bytesRequested);
}
/*Request the offset to write bytesRequested bytes*/
ret = mca_sharedfp_sm_request_position(fh,bytesRequested,&offset);
offset /= fh->f_etype_size;
if ( -1 != ret ) {
if ( mca_sharedfp_sm_verbose ) {
opal_output(ompi_sharedfp_base_framework.framework_output,
"sharedfp_sm_read: Offset received is %lld\n",offset);
}
/* Read the file */
ret = mca_common_ompio_file_read_at(fh,offset,buf,count,datatype,status);
}
return ret;
}
int mca_sharedfp_sm_read_ordered (ompio_file_t *fh,
void *buf,
int count,
struct ompi_datatype_t *datatype,
ompi_status_public_t *status)
{
int ret = OMPI_SUCCESS;
OMPI_MPI_OFFSET_TYPE offset = 0;
long sendBuff = 0;
long *buff=NULL;
long offsetBuff;
OMPI_MPI_OFFSET_TYPE offsetReceived = 0;
long bytesRequested = 0;
int recvcnt = 1, sendcnt = 1;
size_t numofBytes;
int i;
if ( NULL == fh->f_sharedfp_data){
opal_output(ompi_sharedfp_base_framework.framework_output,
"sharedfp_sm_read_ordered: module not initialized \n");
return OMPI_ERROR;
}
/* Calculate the number of bytes to read*/
opal_datatype_type_size ( &datatype->super, &numofBytes);
sendBuff = count * numofBytes;
if ( 0 == fh->f_rank ) {
buff = (long*)malloc(sizeof(long) * fh->f_size);
if ( NULL == buff )
return OMPI_ERR_OUT_OF_RESOURCE;
}
ret = fh->f_comm->c_coll->coll_gather ( &sendBuff,
sendcnt,
OMPI_OFFSET_DATATYPE,
buff,
recvcnt,
OMPI_OFFSET_DATATYPE,
0,
fh->f_comm,
fh->f_comm->c_coll->coll_gather_module );
if( OMPI_SUCCESS != ret){
goto exit;
}
/* All the counts are present now in the recvBuff.
** The size of recvBuff is sizeof_newComm
*/
if ( 0 == fh->f_rank ) {
for (i = 0; i < fh->f_size ; i ++) {
bytesRequested += buff[i];
if ( mca_sharedfp_sm_verbose ) {
opal_output(ompi_sharedfp_base_framework.framework_output,
"mca_sharedfp_sm_read_ordered: Bytes requested are %ld\n",bytesRequested);
}
}
/* Request the offset to read bytesRequested bytes
** only the root process needs to do the request,
** since the root process will then tell the other
** processes at what offset they should read their
** share of the data.
*/
ret = mca_sharedfp_sm_request_position(fh,bytesRequested,&offsetReceived);
if( OMPI_SUCCESS != ret){
goto exit;
}
if ( mca_sharedfp_sm_verbose ) {
opal_output(ompi_sharedfp_base_framework.framework_output,
"mca_sharedfp_sm_read_ordered: Offset received is %lld\n",offsetReceived);
}
buff[0] += offsetReceived;
for (i = 1 ; i < fh->f_size; i++) {
buff[i] += buff[i-1];
}
}
/* Scatter the results to the other processes*/
ret = fh->f_comm->c_coll->coll_scatter ( buff,
sendcnt,
OMPI_OFFSET_DATATYPE,
&offsetBuff,
recvcnt,
OMPI_OFFSET_DATATYPE,
0,
fh->f_comm,
fh->f_comm->c_coll->coll_scatter_module );
if( OMPI_SUCCESS != ret){
goto exit;
}
/*Each process now has its own individual offset in recvBUFF*/
offset = offsetBuff - sendBuff;
offset /= fh->f_etype_size;
if ( mca_sharedfp_sm_verbose ) {
opal_output(ompi_sharedfp_base_framework.framework_output,
"mca_sharedfp_sm_read_ordered: Offset returned is %lld\n",offset);
}
/* read to the file */
ret = mca_common_ompio_file_read_at_all(fh,offset,buf,count,datatype,status);
exit:
if ( NULL != buff ) {
free ( buff );
}
return ret;
}