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.
425 строки
12 KiB
C
425 строки
12 KiB
C
/*
|
|
* 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.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
#include "ompi_config.h"
|
|
|
|
#include "mpi.h"
|
|
#include "opal/mca/mca.h"
|
|
#include "ompi/mca/coll/coll.h"
|
|
#include "ompi/request/request.h"
|
|
#include "ompi/constants.h"
|
|
#include "ompi/datatype/datatype.h"
|
|
#include "ompi/communicator/communicator.h"
|
|
#include "coll_tuned.h"
|
|
|
|
/* need to include our own topo prototypes so we can malloc data on the comm correctly */
|
|
#include "coll_tuned_topo.h"
|
|
|
|
/* also need the dynamic rule structures */
|
|
#include "coll_tuned_dynamic_rules.h"
|
|
|
|
#include <stdlib.h>
|
|
#include <stdio.h>
|
|
|
|
#include "coll_tuned_util.h"
|
|
|
|
|
|
ompi_coll_alg_rule_t* ompi_coll_tuned_mk_alg_rules (int n_alg)
|
|
{
|
|
int i;
|
|
ompi_coll_alg_rule_t* alg_rules;
|
|
|
|
alg_rules = (ompi_coll_alg_rule_t *) calloc (n_alg, sizeof (ompi_coll_alg_rule_t));
|
|
if (!alg_rules) return (alg_rules);
|
|
|
|
/* set all we can at this point */
|
|
for (i=0;i<n_alg;i++) {
|
|
alg_rules[i].alg_rule_id = i;
|
|
}
|
|
return (alg_rules);
|
|
}
|
|
|
|
|
|
ompi_coll_com_rule_t* ompi_coll_tuned_mk_com_rules (int n_com_rules, int alg_rule_id)
|
|
{
|
|
int i;
|
|
ompi_coll_com_rule_t * com_rules;
|
|
|
|
com_rules = (ompi_coll_com_rule_t *) calloc (n_com_rules, sizeof (ompi_coll_com_rule_t));
|
|
if (!com_rules) return (com_rules);
|
|
|
|
for (i=0;i<n_com_rules;i++) {
|
|
com_rules[i].mpi_comsize = 0; /* unknown */
|
|
com_rules[i].alg_rule_id = alg_rule_id;
|
|
com_rules[i].com_rule_id = i;
|
|
com_rules[i].n_msg_sizes = 0; /* unknown */
|
|
com_rules[i].msg_rules = (ompi_coll_msg_rule_t *) NULL;
|
|
}
|
|
return (com_rules);
|
|
}
|
|
|
|
|
|
ompi_coll_msg_rule_t* ompi_coll_tuned_mk_msg_rules (int n_msg_rules, int alg_rule_id, int com_rule_id, int mpi_comsize)
|
|
{
|
|
int i;
|
|
ompi_coll_msg_rule_t *msg_rules;
|
|
|
|
msg_rules = (ompi_coll_msg_rule_t *) calloc (n_msg_rules, sizeof (ompi_coll_msg_rule_t));
|
|
if (!msg_rules) return (msg_rules);
|
|
|
|
for (i=0;i<n_msg_rules;i++) {
|
|
msg_rules[i].mpi_comsize = mpi_comsize;
|
|
msg_rules[i].alg_rule_id = alg_rule_id;
|
|
msg_rules[i].com_rule_id = com_rule_id;
|
|
msg_rules[i].msg_rule_id = i;
|
|
msg_rules[i].msg_size = 0; /* unknown */
|
|
msg_rules[i].result_alg = 0; /* unknown */
|
|
msg_rules[i].result_topo_faninout = 0; /* unknown */
|
|
msg_rules[i].result_segsize = 0; /* unknown */
|
|
msg_rules[i].result_max_requests = 0; /* unknown & default */
|
|
}
|
|
return (msg_rules);
|
|
}
|
|
|
|
|
|
/*
|
|
* Debug / IO routines
|
|
*
|
|
*/
|
|
|
|
|
|
int ompi_coll_tuned_dump_msg_rule (ompi_coll_msg_rule_t* msg_p)
|
|
{
|
|
if (!msg_p) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"Message rule was a NULL ptr?!\n"));
|
|
return (-1);
|
|
}
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"alg_id %3d\tcom_id %3d\tcom_size %3d\tmsg_id %3d\t", msg_p->alg_rule_id,
|
|
msg_p->com_rule_id, msg_p->mpi_comsize, msg_p->msg_rule_id));
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"msg_size %6d -> algorithm %2d\ttopo in/out %2d\tsegsize %5ld\tmax_requests %4d\n",
|
|
msg_p->msg_size, msg_p->result_alg, msg_p->result_topo_faninout, msg_p->result_segsize,
|
|
msg_p->result_max_requests));
|
|
|
|
return (0);
|
|
}
|
|
|
|
|
|
int ompi_coll_tuned_dump_com_rule (ompi_coll_com_rule_t* com_p)
|
|
{
|
|
int i;
|
|
|
|
if (!com_p) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"Com rule was a NULL ptr?!\n"));
|
|
return (-1);
|
|
}
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream, "alg_id %3d\tcom_id %3d\tcom_size %3d\t", com_p->alg_rule_id, com_p->com_rule_id, com_p->mpi_comsize));
|
|
|
|
if (!com_p->n_msg_sizes) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"no msgsizes defined\n"));
|
|
return (0);
|
|
}
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"number of message sizes %3d\n", com_p->n_msg_sizes));
|
|
|
|
for (i=0;i<com_p->n_msg_sizes;i++) {
|
|
ompi_coll_tuned_dump_msg_rule (&(com_p->msg_rules[i]));
|
|
}
|
|
|
|
return (0);
|
|
}
|
|
|
|
|
|
int ompi_coll_tuned_dump_alg_rule (ompi_coll_alg_rule_t* alg_p)
|
|
{
|
|
int i;
|
|
|
|
if (!alg_p) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"Algorithm rule was a NULL ptr?!\n"));
|
|
return (-1);
|
|
}
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"alg_id %3d\t", alg_p->alg_rule_id));
|
|
|
|
if (!alg_p->n_com_sizes) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"no coms defined\n"));
|
|
return (0);
|
|
}
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"number of com sizes %3d\n", alg_p->n_com_sizes));
|
|
|
|
for (i=0;i<alg_p->n_com_sizes;i++) {
|
|
ompi_coll_tuned_dump_com_rule (&(alg_p->com_rules[i]));
|
|
}
|
|
|
|
return (0);
|
|
}
|
|
|
|
|
|
int ompi_coll_tuned_dump_all_rules (ompi_coll_alg_rule_t* alg_p, int n_rules)
|
|
{
|
|
int i;
|
|
|
|
if (!alg_p) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"Algorithm rule was a NULL ptr?!\n"));
|
|
return (-1);
|
|
}
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"Number of algorithm rules %3d\n", n_rules));
|
|
|
|
for (i=0;i<n_rules;i++) {
|
|
ompi_coll_tuned_dump_alg_rule (&(alg_p[i]));
|
|
}
|
|
|
|
return (0);
|
|
}
|
|
|
|
|
|
/*
|
|
* Memory free routines
|
|
*
|
|
*/
|
|
|
|
|
|
int ompi_coll_tuned_free_msg_rules_in_com_rule (ompi_coll_com_rule_t* com_p)
|
|
{
|
|
int rc=0;
|
|
ompi_coll_msg_rule_t* msg_p;
|
|
|
|
if (!com_p) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"attempt to free NULL com_rule ptr\n"));
|
|
return (-1);
|
|
}
|
|
|
|
if (com_p->n_msg_sizes) {
|
|
msg_p = com_p->msg_rules;
|
|
|
|
if (!msg_p) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"attempt to free NULL msg_rules when msg count was %d\n", com_p->n_msg_sizes));
|
|
rc = -1; /* some error */
|
|
}
|
|
else {
|
|
/* ok, memory exists for the msg rules so free that first */
|
|
free (com_p->msg_rules);
|
|
com_p->msg_rules = (ompi_coll_msg_rule_t*) NULL;
|
|
}
|
|
|
|
} /* if we have msg rules to free as well */
|
|
|
|
return (rc);
|
|
}
|
|
|
|
|
|
|
|
int ompi_coll_tuned_free_coms_in_alg_rule (ompi_coll_alg_rule_t* alg_p)
|
|
{
|
|
int rc=0;
|
|
int i;
|
|
|
|
ompi_coll_com_rule_t* com_p;
|
|
|
|
if (!alg_p) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"attempt to free NULL alg_rule ptr\n"));
|
|
return (-1);
|
|
}
|
|
|
|
if (alg_p->n_com_sizes) {
|
|
com_p = alg_p->com_rules;
|
|
|
|
if (!com_p) {
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"attempt to free NULL com_rules when com count was %d\n", alg_p->n_com_sizes));
|
|
}
|
|
else {
|
|
/* ok, memory exists for the com rules so free their message rules first */
|
|
for (i=0;i<alg_p->n_com_sizes;i++) {
|
|
com_p = &(alg_p->com_rules[i]);
|
|
ompi_coll_tuned_free_msg_rules_in_com_rule (com_p);
|
|
}
|
|
/* we are now free to free the com rules themselives */
|
|
free (alg_p->com_rules);
|
|
alg_p->com_rules = (ompi_coll_com_rule_t*) NULL;
|
|
}
|
|
|
|
} /* if we have msg rules to free as well */
|
|
|
|
return (rc);
|
|
}
|
|
|
|
|
|
int ompi_coll_tuned_free_all_rules (ompi_coll_alg_rule_t* alg_p, int n_algs)
|
|
{
|
|
int i;
|
|
int rc = 0;
|
|
|
|
for(i=0;i<n_algs;i++) {
|
|
rc += ompi_coll_tuned_free_coms_in_alg_rule (&(alg_p[i]));
|
|
}
|
|
|
|
free (alg_p);
|
|
|
|
return (rc);
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
* query functions
|
|
* i.e. the functions that get me the algorithm, topo fanin/out and segment size fast
|
|
* and also get the rules that are needed by each communicator as needed
|
|
*
|
|
*/
|
|
|
|
/*
|
|
* This function is used to get the pointer to the nearest (less than or equal)
|
|
* com rule for this MPI collective (alg_id) for a given
|
|
* MPI communicator size. The complete rule base must be presented.
|
|
*
|
|
* If no rule exits returns NULL, else the com rule ptr
|
|
* (which can be used in the coll_tuned_get_target_method_params() call)
|
|
*
|
|
*/
|
|
ompi_coll_com_rule_t* ompi_coll_tuned_get_com_rule_ptr (ompi_coll_alg_rule_t* rules, int alg_id, int mpi_comsize)
|
|
{
|
|
ompi_coll_alg_rule_t* alg_p = (ompi_coll_alg_rule_t*) NULL;
|
|
ompi_coll_com_rule_t* com_p = (ompi_coll_com_rule_t*) NULL;
|
|
ompi_coll_com_rule_t* best_com_p = (ompi_coll_com_rule_t*) NULL;
|
|
int i, best;
|
|
|
|
if (!rules) { /* no rule base no resulting com rule */
|
|
return ((ompi_coll_com_rule_t*)NULL);
|
|
}
|
|
|
|
alg_p = &(rules[alg_id]); /* get the algorithm rule pointer */
|
|
|
|
if (!alg_p->n_com_sizes) { /* check for count of communicator sizes */
|
|
return ((ompi_coll_com_rule_t*)NULL); /* no com sizes so no rule */
|
|
}
|
|
|
|
/* ok have some com sizes, now to find the one closest to my mpi_comsize */
|
|
|
|
/* make a copy of the first com rule */
|
|
best_com_p = com_p = alg_p->com_rules;
|
|
i = best = 0;
|
|
|
|
while (i<alg_p->n_com_sizes) {
|
|
/* ORTE_OUTPUT((ompi_coll_tuned_stream,"checking comsize %d against alg_id %d com_id %d index %d com_size %d", */
|
|
/* mpi_comsize, com_p->alg_rule_id, com_p->com_rule_id, i, com_p->mpi_comsize)); */
|
|
if (com_p->mpi_comsize <= mpi_comsize) {
|
|
best = i;
|
|
best_com_p = com_p;
|
|
/* ORTE_OUTPUT((ompi_coll_tuned_stream(":ok\n")); */
|
|
}
|
|
else {
|
|
/* ORTE_OUTPUT((ompi_coll_tuned_stream(":nop\n")); */
|
|
break;
|
|
}
|
|
/* go to the next entry */
|
|
com_p++;
|
|
i++;
|
|
}
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"Selected the following com rule id %d\n", best_com_p->com_rule_id));
|
|
ompi_coll_tuned_dump_com_rule (best_com_p);
|
|
|
|
return (best_com_p);
|
|
}
|
|
|
|
/*
|
|
* This function takes a com_rule ptr (from the communicators coll tuned data structure)
|
|
* (Which is chosen for a particular MPI collective)
|
|
* and a (total_)msg_size and it returns (0) and a algorithm to use and a recommended topo faninout and segment size
|
|
* all based on the user supplied rules
|
|
*
|
|
* Just like the above functions it uses a less than or equal msg size
|
|
* (hense config file must have a default defined for '0' if we reach this point)
|
|
* else if no rules match we return '0' + '0,0' or used fixed decision table with no topo chand and no segmentation
|
|
* of users data.. shame.
|
|
*
|
|
* On error return 0 so we default to fixed rules anyway :)
|
|
*
|
|
*/
|
|
|
|
int ompi_coll_tuned_get_target_method_params (ompi_coll_com_rule_t* base_com_rule, int mpi_msgsize, int *result_topo_faninout,
|
|
int* result_segsize, int* max_requests)
|
|
{
|
|
ompi_coll_msg_rule_t* msg_p = (ompi_coll_msg_rule_t*) NULL;
|
|
ompi_coll_msg_rule_t* best_msg_p = (ompi_coll_msg_rule_t*) NULL;
|
|
int i, best;
|
|
|
|
if (!base_com_rule) {
|
|
return (0);
|
|
}
|
|
|
|
if (!result_topo_faninout) {
|
|
return (0);
|
|
}
|
|
|
|
if (!result_segsize) {
|
|
return (0);
|
|
}
|
|
|
|
if (!max_requests) {
|
|
return (0);
|
|
}
|
|
|
|
if (!base_com_rule->n_msg_sizes) { /* check for count of message sizes */
|
|
return (0); /* no msg sizes so no rule */
|
|
}
|
|
|
|
/* ok have some msg sizes, now to find the one closest to my mpi_msgsize */
|
|
|
|
/* make a copy of the first msg rule */
|
|
best_msg_p = msg_p = base_com_rule->msg_rules;
|
|
i = best = 0;
|
|
|
|
while (i<base_com_rule->n_msg_sizes) {
|
|
/* ORTE_OUTPUT((ompi_coll_tuned_stream,"checking mpi_msgsize %d against com_id %d msg_id %d index %d msg_size %d", */
|
|
/* mpi_msgsize, msg_p->com_rule_id, msg_p->msg_rule_id, i, msg_p->msg_size)); */
|
|
if (msg_p->msg_size <= mpi_msgsize) {
|
|
best = i;
|
|
best_msg_p = msg_p;
|
|
/* ORTE_OUTPUT((ompi_coll_tuned_stream(":ok\n")); */
|
|
}
|
|
else {
|
|
/* ORTE_OUTPUT((ompi_coll_tuned_stream(":nop\n")); */
|
|
break;
|
|
}
|
|
/* go to the next entry */
|
|
msg_p++;
|
|
i++;
|
|
}
|
|
|
|
ORTE_OUTPUT((ompi_coll_tuned_stream,"Selected the following msg rule id %d\n", best_msg_p->msg_rule_id));
|
|
ompi_coll_tuned_dump_msg_rule (best_msg_p);
|
|
|
|
/* return the segment size */
|
|
*result_topo_faninout = best_msg_p->result_topo_faninout;
|
|
|
|
/* return the segment size */
|
|
*result_segsize = best_msg_p->result_segsize;
|
|
|
|
/* return the maximum requests */
|
|
*max_requests = best_msg_p->result_max_requests;
|
|
|
|
/* return the algorithm/method to use */
|
|
return (best_msg_p->result_alg);
|
|
}
|
|
|