e7ecd56bd2
such, the commit message back to the master SVN repository is fairly
long.
= ORTE Job-Level Output Messages =
Add two new interfaces that should be used for all new code throughout
the ORTE and OMPI layers (we already make the search-and-replace on
the existing ORTE / OMPI layers):
* orte_output(): (and corresponding friends ORTE_OUTPUT,
orte_output_verbose, etc.) This function sends the output directly
to the HNP for processing as part of a job-specific output
channel. It supports all the same outputs as opal_output()
(syslog, file, stdout, stderr), but for stdout/stderr, the output
is sent to the HNP for processing and output. More on this below.
* orte_show_help(): This function is a drop-in-replacement for
opal_show_help(), with two differences in functionality:
1. the rendered text help message output is sent to the HNP for
display (rather than outputting directly into the process' stderr
stream)
1. the HNP detects duplicate help messages and does not display them
(so that you don't see the same error message N times, once from
each of your N MPI processes); instead, it counts "new" instances
of the help message and displays a message every ~5 seconds when
there are new ones ("I got X new copies of the help message...")
opal_show_help and opal_output still exist, but they only output in
the current process. The intent for the new orte_* functions is that
they can apply job-level intelligence to the output. As such, we
recommend that all new ORTE and OMPI code use the new orte_*
functions, not thei opal_* functions.
=== New code ===
For ORTE and OMPI programmers, here's what you need to do differently
in new code:
* Do not include opal/util/show_help.h or opal/util/output.h.
Instead, include orte/util/output.h (this one header file has
declarations for both the orte_output() series of functions and
orte_show_help()).
* Effectively s/opal_output/orte_output/gi throughout your code.
Note that orte_output_open() takes a slightly different argument
list (as a way to pass data to the filtering stream -- see below),
so you if explicitly call opal_output_open(), you'll need to
slightly adapt to the new signature of orte_output_open().
* Literally s/opal_show_help/orte_show_help/. The function signature
is identical.
=== Notes ===
* orte_output'ing to stream 0 will do similar to what
opal_output'ing did, so leaving a hard-coded "0" as the first
argument is safe.
* For systems that do not use ORTE's RML or the HNP, the effect of
orte_output_* and orte_show_help will be identical to their opal
counterparts (the additional information passed to
orte_output_open() will be lost!). Indeed, the orte_* functions
simply become trivial wrappers to their opal_* counterparts. Note
that we have not tested this; the code is simple but it is quite
possible that we mucked something up.
= Filter Framework =
Messages sent view the new orte_* functions described above and
messages output via the IOF on the HNP will now optionally be passed
through a new "filter" framework before being output to
stdout/stderr. The "filter" OPAL MCA framework is intended to allow
preprocessing to messages before they are sent to their final
destinations. The first component that was written in the filter
framework was to create an XML stream, segregating all the messages
into different XML tags, etc. This will allow 3rd party tools to read
the stdout/stderr from the HNP and be able to know exactly what each
text message is (e.g., a help message, another OMPI infrastructure
message, stdout from the user process, stderr from the user process,
etc.).
Filtering is not active by default. Filter components must be
specifically requested, such as:
{{{
$ mpirun --mca filter xml ...
}}}
There can only be one filter component active.
= New MCA Parameters =
The new functionality described above introduces two new MCA
parameters:
* '''orte_base_help_aggregate''': Defaults to 1 (true), meaning that
help messages will be aggregated, as described above. If set to 0,
all help messages will be displayed, even if they are duplicates
(i.e., the original behavior).
* '''orte_base_show_output_recursions''': An MCA parameter to help
debug one of the known issues, described below. It is likely that
this MCA parameter will disappear before v1.3 final.
= Known Issues =
* The XML filter component is not complete. The current output from
this component is preliminary and not real XML. A bit more work
needs to be done to configure.m4 search for an appropriate XML
library/link it in/use it at run time.
* There are possible recursion loops in the orte_output() and
orte_show_help() functions -- e.g., if RML send calls orte_output()
or orte_show_help(). We have some ideas how to fix these, but
figured that it was ok to commit before feature freeze with known
issues. The code currently contains sub-optimal workarounds so
that this will not be a problem, but it would be good to actually
solve the problem rather than have hackish workarounds before v1.3 final.
This commit was SVN r18434.
324 строки
13 KiB
C
324 строки
13 KiB
C
/*
|
|
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
|
|
* University Research and Technology
|
|
* Corporation. All rights reserved.
|
|
* Copyright (c) 2004-2006 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 Sun Microsystems, Inc. All rights reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
#include "ompi_config.h"
|
|
|
|
#include "mpi.h"
|
|
#include "ompi/constants.h"
|
|
#include "ompi/datatype/datatype.h"
|
|
#include "ompi/communicator/communicator.h"
|
|
#include "ompi/mca/coll/coll.h"
|
|
#include "ompi/mca/coll/base/coll_tags.h"
|
|
#include "ompi/mca/pml/pml.h"
|
|
#include "ompi/op/op.h"
|
|
#include "coll_tuned.h"
|
|
#include "coll_tuned_topo.h"
|
|
#include "coll_tuned_util.h"
|
|
|
|
int
|
|
ompi_coll_tuned_alltoallv_intra_pairwise(void *sbuf, int *scounts, int *sdisps,
|
|
struct ompi_datatype_t *sdtype,
|
|
void* rbuf, int *rcounts, int *rdisps,
|
|
struct ompi_datatype_t *rdtype,
|
|
struct ompi_communicator_t *comm,
|
|
struct mca_coll_base_module_1_1_0_t *module)
|
|
{
|
|
int line = -1, err = 0;
|
|
int rank, size, step;
|
|
int sendto, recvfrom;
|
|
void *psnd, *prcv;
|
|
ptrdiff_t sext, rext;
|
|
|
|
size = ompi_comm_size(comm);
|
|
rank = ompi_comm_rank(comm);
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,
|
|
"coll:tuned:alltoallv_intra_pairwise rank %d", rank));
|
|
|
|
ompi_ddt_type_extent(sdtype, &sext);
|
|
ompi_ddt_type_extent(rdtype, &rext);
|
|
|
|
psnd = ((char *) sbuf) + (sdisps[rank] * sext);
|
|
prcv = ((char *) rbuf) + (rdisps[rank] * rext);
|
|
|
|
if (0 != scounts[rank]) {
|
|
err = ompi_ddt_sndrcv(psnd, scounts[rank], sdtype,
|
|
prcv, rcounts[rank], rdtype);
|
|
if (MPI_SUCCESS != err) {
|
|
return err;
|
|
}
|
|
}
|
|
|
|
/* If only one process, we're done. */
|
|
if (1 == size) {
|
|
return MPI_SUCCESS;
|
|
}
|
|
|
|
/* Perform pairwise exchange starting from 1 since local exhange is done */
|
|
for (step = 1; step < size + 1; step++) {
|
|
|
|
/* Determine sender and receiver for this step. */
|
|
sendto = (rank + step) % size;
|
|
recvfrom = (rank + size - step) % size;
|
|
|
|
/* Determine sending and receiving locations */
|
|
psnd = (char*)sbuf + sdisps[sendto] * sext;
|
|
prcv = (char*)rbuf + rdisps[recvfrom] * rext;
|
|
|
|
/* send and receive */
|
|
err = ompi_coll_tuned_sendrecv( psnd, scounts[sendto], sdtype, sendto,
|
|
MCA_COLL_BASE_TAG_ALLTOALLV,
|
|
prcv, rcounts[recvfrom], rdtype, recvfrom,
|
|
MCA_COLL_BASE_TAG_ALLTOALLV,
|
|
comm, MPI_STATUS_IGNORE, rank);
|
|
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl; }
|
|
}
|
|
|
|
return MPI_SUCCESS;
|
|
|
|
err_hndl:
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,
|
|
"%s:%4d\tError occurred %d, rank %2d", __FILE__, line,
|
|
err, rank));
|
|
return err;
|
|
}
|
|
|
|
/*
|
|
* Linear functions are copied from the basic coll module. For
|
|
* some small number of nodes and/or small data sizes they are just as
|
|
* fast as tuned/tree based segmenting operations and as such may be
|
|
* selected by the decision functions. These are copied into this module
|
|
* due to the way we select modules in V1. i.e. in V2 we will handle this
|
|
* differently and so will not have to duplicate code.
|
|
* GEF Oct05 after asking Jeff.
|
|
*/
|
|
int
|
|
ompi_coll_tuned_alltoallv_intra_basic_linear(void *sbuf, int *scounts, int *sdisps,
|
|
struct ompi_datatype_t *sdtype,
|
|
void *rbuf, int *rcounts, int *rdisps,
|
|
struct ompi_datatype_t *rdtype,
|
|
struct ompi_communicator_t *comm,
|
|
struct mca_coll_base_module_1_1_0_t *module)
|
|
{
|
|
int i, size, rank, err;
|
|
char *psnd, *prcv;
|
|
int nreqs;
|
|
ptrdiff_t sext, rext;
|
|
MPI_Request *preq;
|
|
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
|
|
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,
|
|
"coll:tuned:alltoallv_intra_basic_linear rank %d", rank));
|
|
|
|
size = ompi_comm_size(comm);
|
|
rank = ompi_comm_rank(comm);
|
|
|
|
ompi_ddt_type_extent(sdtype, &sext);
|
|
ompi_ddt_type_extent(rdtype, &rext);
|
|
|
|
/* Simple optimization - handle send to self first */
|
|
psnd = ((char *) sbuf) + (sdisps[rank] * sext);
|
|
prcv = ((char *) rbuf) + (rdisps[rank] * rext);
|
|
if (0 != scounts[rank]) {
|
|
err = ompi_ddt_sndrcv(psnd, scounts[rank], sdtype,
|
|
prcv, rcounts[rank], rdtype);
|
|
if (MPI_SUCCESS != err) {
|
|
return err;
|
|
}
|
|
}
|
|
|
|
/* If only one process, we're done. */
|
|
if (1 == size) {
|
|
return MPI_SUCCESS;
|
|
}
|
|
|
|
/* Now, initiate all send/recv to/from others. */
|
|
nreqs = 0;
|
|
preq = data->mcct_reqs;
|
|
|
|
/* Post all receives first */
|
|
for (i = 0; i < size; ++i) {
|
|
if (i == rank || 0 == rcounts[i]) {
|
|
continue;
|
|
}
|
|
|
|
prcv = ((char *) rbuf) + (rdisps[i] * rext);
|
|
err = MCA_PML_CALL(irecv_init(prcv, rcounts[i], rdtype,
|
|
i, MCA_COLL_BASE_TAG_ALLTOALLV, comm,
|
|
preq++));
|
|
++nreqs;
|
|
if (MPI_SUCCESS != err) {
|
|
ompi_coll_tuned_free_reqs(data->mcct_reqs, nreqs);
|
|
return err;
|
|
}
|
|
}
|
|
|
|
/* Now post all sends */
|
|
for (i = 0; i < size; ++i) {
|
|
if (i == rank || 0 == scounts[i]) {
|
|
continue;
|
|
}
|
|
|
|
psnd = ((char *) sbuf) + (sdisps[i] * sext);
|
|
err = MCA_PML_CALL(isend_init(psnd, scounts[i], sdtype,
|
|
i, MCA_COLL_BASE_TAG_ALLTOALLV,
|
|
MCA_PML_BASE_SEND_STANDARD, comm,
|
|
preq++));
|
|
++nreqs;
|
|
if (MPI_SUCCESS != err) {
|
|
ompi_coll_tuned_free_reqs(data->mcct_reqs, nreqs);
|
|
return err;
|
|
}
|
|
}
|
|
|
|
/* Start your engines. This will never return an error. */
|
|
MCA_PML_CALL(start(nreqs, data->mcct_reqs));
|
|
|
|
/* Wait for them all. If there's an error, note that we don't care
|
|
* what the error was -- just that there *was* an error. The PML
|
|
* will finish all requests, even if one or more of them fail.
|
|
* i.e., by the end of this call, all the requests are free-able.
|
|
* So free them anyway -- even if there was an error, and return the
|
|
* error after we free everything. */
|
|
err = ompi_request_wait_all(nreqs, data->mcct_reqs,
|
|
MPI_STATUSES_IGNORE);
|
|
|
|
/* Free the requests. */
|
|
ompi_coll_tuned_free_reqs(data->mcct_reqs, nreqs);
|
|
|
|
return err;
|
|
}
|
|
|
|
/*
|
|
* The following are used by dynamic and forced rules. Publish
|
|
* details of each algorithm and if its forced/fixed/locked in as you add
|
|
* methods/algorithms you must update this and the query/map routines.
|
|
* This routine is called by the component only. This makes sure that
|
|
* the mca parameters are set to their initial values and perms.
|
|
* Module does not call this. They call the forced_getvalues routine
|
|
* instead.
|
|
*/
|
|
int ompi_coll_tuned_alltoallv_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t
|
|
*mca_param_indices)
|
|
{
|
|
int max_alg = 2, requested_alg, max_requests;
|
|
|
|
ompi_coll_tuned_forced_max_algorithms[ALLTOALLV] = max_alg;
|
|
|
|
mca_base_param_reg_int (&mca_coll_tuned_component.super.collm_version,
|
|
"alltoallv_algorithm_count",
|
|
"Number of alltoallv algorithms available",
|
|
false, true, max_alg, NULL);
|
|
|
|
mca_param_indices->algorithm_param_index
|
|
= mca_base_param_reg_int(&mca_coll_tuned_component.super.collm_version,
|
|
"alltoallv_algorithm",
|
|
"Which alltoallv algorithm is used. "
|
|
"Can be locked down to choice of: 0 ignore, "
|
|
"1 basic linear, 2 pairwise.",
|
|
false, false, 0, NULL);
|
|
mca_base_param_lookup_int(mca_param_indices->algorithm_param_index, &(requested_alg));
|
|
if (requested_alg > max_alg) {
|
|
if (0 == ompi_comm_rank( MPI_COMM_WORLD)) {
|
|
orte_output(0, "Alltoallv algorithm #%d is not available (range [0..%d]). "
|
|
"Switching back to ignore(0)\n",
|
|
requested_alg, max_alg );
|
|
}
|
|
mca_base_param_set_int(mca_param_indices->algorithm_param_index, 0);
|
|
}
|
|
|
|
return (MPI_SUCCESS);
|
|
}
|
|
|
|
|
|
|
|
int ompi_coll_tuned_alltoallv_intra_do_forced(void *sbuf, int *scounts, int *sdisps,
|
|
struct ompi_datatype_t *sdtype,
|
|
void* rbuf, int *rcounts, int *rdisps,
|
|
struct ompi_datatype_t *rdtype,
|
|
struct ompi_communicator_t *comm,
|
|
struct mca_coll_base_module_1_1_0_t *module)
|
|
{
|
|
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
|
|
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,
|
|
"coll:tuned:alltoallv_intra_do_forced selected algorithm %d",
|
|
data->user_forced[ALLTOALLV].algorithm));
|
|
|
|
switch (data->user_forced[ALLTOALLV].algorithm) {
|
|
case (0):
|
|
return ompi_coll_tuned_alltoallv_intra_dec_fixed(sbuf, scounts, sdisps, sdtype,
|
|
rbuf, rcounts, rdisps, rdtype,
|
|
comm, module);
|
|
case (1):
|
|
return ompi_coll_tuned_alltoallv_intra_basic_linear(sbuf, scounts, sdisps, sdtype,
|
|
rbuf, rcounts, rdisps, rdtype,
|
|
comm, module);
|
|
case (2):
|
|
return ompi_coll_tuned_alltoallv_intra_pairwise(sbuf, scounts, sdisps, sdtype,
|
|
rbuf, rcounts, rdisps, rdtype,
|
|
comm, module);
|
|
default:
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,
|
|
"coll:tuned:alltoallv_intra_do_forced attempt to "
|
|
"select algorithm %d when only 0-%d is valid.",
|
|
data->user_forced[ALLTOALLV].algorithm,
|
|
ompi_coll_tuned_forced_max_algorithms[ALLTOALLV]));
|
|
return (MPI_ERR_ARG);
|
|
}
|
|
}
|
|
|
|
/* If the user selects dynamic rules and specifies the algorithm to
|
|
* use, then this function is called. */
|
|
int ompi_coll_tuned_alltoallv_intra_do_this(void *sbuf, int *scounts, int *sdisps,
|
|
struct ompi_datatype_t *sdtype,
|
|
void* rbuf, int *rcounts, int *rdisps,
|
|
struct ompi_datatype_t *rdtype,
|
|
struct ompi_communicator_t *comm,
|
|
struct mca_coll_base_module_1_1_0_t *module,
|
|
int algorithm)
|
|
{
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,
|
|
"coll:tuned:alltoallv_intra_do_this selected algorithm %d ",
|
|
algorithm));
|
|
|
|
switch (algorithm) {
|
|
case (0):
|
|
return ompi_coll_tuned_alltoallv_intra_dec_fixed(sbuf, scounts, sdisps, sdtype,
|
|
rbuf, rcounts, rdisps, rdtype,
|
|
comm, module);
|
|
case (1):
|
|
return ompi_coll_tuned_alltoallv_intra_basic_linear(sbuf, scounts, sdisps, sdtype,
|
|
rbuf, rcounts, rdisps, rdtype,
|
|
comm, module);
|
|
case (2):
|
|
return ompi_coll_tuned_alltoallv_intra_pairwise(sbuf, scounts, sdisps, sdtype,
|
|
rbuf, rcounts, rdisps, rdtype,
|
|
comm, module);
|
|
default:
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,
|
|
"coll:tuned:alltoall_intra_do_this attempt to select "
|
|
"algorithm %d when only 0-%d is valid.",
|
|
algorithm, ompi_coll_tuned_forced_max_algorithms[ALLTOALLV]));
|
|
return (MPI_ERR_ARG);
|
|
}
|
|
}
|