1
1
openmpi/ompi/mca/coll/inter/coll_inter_gatherv.c

145 строки
4.2 KiB
C
Исходник Обычный вид История

/*
* Copyright (c) 2004-2005 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) 2006-2007 University of Houston. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "coll_inter.h"
#include "mpi.h"
#include "ompi/constants.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 "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
/*
* gatherv_inter
*
* Function: - gatherv operation using a local gather on c_local_comm
* Accepts: - same arguments as MPI_Gatherv()
* Returns: - MPI_SUCCESS or error code
*/
int
mca_coll_inter_gatherv_inter(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void *rbuf, int *rcounts, int *disps,
struct ompi_datatype_t *rdtype, int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
int i, rank, size, size_local, total=0, err;
int *count=NULL,*displace=NULL;
char *ptmp=NULL;
MPI_Aint incr;
MPI_Aint extent;
MPI_Aint lb;
ompi_datatype_t *ndtype;
size = ompi_comm_remote_size(comm);
rank = ompi_comm_rank(comm);
size_local = ompi_comm_size(comm);
if (MPI_PROC_NULL == root) {
/* do nothing */
err = OMPI_SUCCESS;
} else if (MPI_ROOT != root) {
if (0 == rank) {
count = (int *)malloc(sizeof(int) * size_local);
if (NULL == count) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
displace = (int *)malloc(sizeof(int) * size_local);
if (NULL == displace) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
}
err = comm->c_local_comm->c_coll.coll_gather(&scount, 1, MPI_INT,
count, 1, MPI_INT,
0, comm->c_local_comm,
comm->c_local_comm->c_coll.coll_gather_module);
if (OMPI_SUCCESS != err) {
return err;
}
if(0 == rank) {
displace[0] = 0;
for (i = 1; i < size_local; i++) {
displace[i] = displace[i-1] + count[i-1];
}
/* Perform the gatherv locally with the first process as root */
- 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
err = ompi_datatype_get_extent(sdtype, &lb, &extent);
if (OMPI_SUCCESS != err) {
return OMPI_ERROR;
}
incr = 0;
for (i = 0; i < size_local; i++) {
incr = incr + extent*count[i];
}
ptmp = (char*)malloc(incr);
if (NULL == ptmp) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
}
err = comm->c_local_comm->c_coll.coll_gatherv(sbuf, scount, sdtype,
ptmp, count, displace,
sdtype,0, comm->c_local_comm,
comm->c_local_comm->c_coll.coll_gatherv_module);
if (OMPI_SUCCESS != err) {
return err;
}
if (0 == rank) {
for (i = 0; i < size_local; i++) {
total = total + count[i];
}
/* First process sends data to the root */
err = MCA_PML_CALL(send(ptmp, total, sdtype, root,
MCA_COLL_BASE_TAG_GATHERV,
MCA_PML_BASE_SEND_STANDARD, comm));
if (OMPI_SUCCESS != err) {
return err;
}
}
if (NULL != ptmp) {
free(ptmp);
}
if (NULL != displace) {
free(displace);
}
if (NULL != count) {
free(count);
}
} else {
/* I am the root, loop receiving the data. */
- 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
ompi_datatype_create_indexed(size,rcounts,disps,rdtype,&ndtype);
ompi_datatype_commit(&ndtype);
err = MCA_PML_CALL(recv(rbuf, 1, ndtype, 0,
MCA_COLL_BASE_TAG_GATHERV,
comm, MPI_STATUS_IGNORE));
if (OMPI_SUCCESS != err) {
return err;
}
- 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
ompi_datatype_destroy(&ndtype);
}
/* All done */
return err;
}