1
1
openmpi/ompi/mca/coll/ml/coll_ml_config.c

630 строки
17 KiB
C
Исходник Обычный вид История

/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2009-2012 Oak Ridge National Laboratory. All rights reserved.
* Copyright (c) 2009-2012 Mellanox Technologies. All rights reserved.
* Copyright (c) 2013-2014 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include <string.h>
#include <ctype.h>
#include <stdlib.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include "coll_ml.h"
#include "coll_ml_inlines.h"
#include "coll_ml_config.h"
#include "coll_ml_lex.h"
static char *key_buffer = NULL;
static size_t key_buffer_len = 0;
typedef struct section_config_t {
char *section_name;
int section_id;
per_collective_configuration_t config;
} section_config_t;
typedef struct coll_config_t {
char *coll_name;
int coll_id;
section_config_t section;
} coll_config_t;
static int algorithm_name_to_id(char *name)
{
assert (NULL != name);
if (!strcasecmp(name,"ML_BCAST_SMALL_DATA_KNOWN"))
return ML_BCAST_SMALL_DATA_KNOWN;
if (!strcasecmp(name,"ML_BCAST_SMALL_DATA_UNKNOWN"))
return ML_BCAST_SMALL_DATA_UNKNOWN;
if (!strcasecmp(name,"ML_BCAST_SMALL_DATA_SEQUENTIAL"))
return ML_BCAST_SMALL_DATA_SEQUENTIAL;
if (!strcasecmp(name,"ML_BCAST_LARGE_DATA_KNOWN"))
return ML_BCAST_LARGE_DATA_KNOWN;
if (!strcasecmp(name,"ML_BCAST_LARGE_DATA_UNKNOWN"))
return ML_BCAST_LARGE_DATA_UNKNOWN;
if (!strcasecmp(name,"ML_BCAST_LARGE_DATA_SEQUENTIAL"))
return ML_BCAST_LARGE_DATA_SEQUENTIAL;
if (!strcasecmp(name,"ML_N_DATASIZE_BINS"))
return ML_N_DATASIZE_BINS;
if (!strcasecmp(name,"ML_NUM_BCAST_FUNCTIONS"))
return ML_NUM_BCAST_FUNCTIONS;
if (!strcasecmp(name,"ML_SCATTER_SMALL_DATA_KNOWN"))
return ML_SCATTER_SMALL_DATA_KNOWN;
if (!strcasecmp(name,"ML_SCATTER_N_DATASIZE_BINS"))
return ML_SCATTER_N_DATASIZE_BINS;
if (!strcasecmp(name,"ML_SCATTER_SMALL_DATA_UNKNOWN"))
return ML_SCATTER_SMALL_DATA_UNKNOWN;
if (!strcasecmp(name,"ML_SCATTER_SMALL_DATA_SEQUENTIAL"))
return ML_SCATTER_SMALL_DATA_SEQUENTIAL;
if (!strcasecmp(name,"ML_NUM_SCATTER_FUNCTIONS"))
return ML_NUM_SCATTER_FUNCTIONS;
if (!strcasecmp(name,"ML_SMALL_DATA_ALLREDUCE"))
return ML_SMALL_DATA_ALLREDUCE;
if (!strcasecmp(name,"ML_LARGE_DATA_ALLREDUCE"))
return ML_LARGE_DATA_ALLREDUCE;
coll/ml: add support for blocking and non-blocking allreduce, reduce, and allgather. The new collectives provide a signifigant performance increase over tuned for small and medium messages. We are initially setting the priority lower than tuned until this has had some time to soak in the trunk. Please set coll_ml_priority to 90 for MTT runs. Credit for this work goes to Manjunath Gorentla Venkata (ORNL), Pavel Shamis (ORNL), and Nathan Hjelm (LANL). Commit details (for reference): Import ORNL's collectives for MPI_Allreduce, MPI_Reduce, and MPI_Allgather. We need to take the basesmuma header into account when calculating the ptpcoll small message thresholds. Add a define to bcol.h indicating the maximum header size so we can take the header into account while not making ptpcoll dependent on information from basesmuma. This resolves an issue with allreduce where ptpcoll overwrites the header of the next buffer in the basesmuma bank. Fix reduce and make a sequential collective launcher in coll_ml_inlines.h The root calculation for reduce was wrong for any root != 0. There are four possibilities for the root: - The root is not the current process but is in the current hierarchy. In this case the root is the index of the global root as specified in the root vector. - The root is not the current process and is not in the next level of the hierarchy. In this case 0 must be the local root since this process will never communicate with the real root. - The root is not the current process but will be in next level of the hierarchy. In this case the current process must be the root. - I am the root. The root is my index. Tested with IMB which rotates the root on every call to MPI_Reduce. Consider IMB the reproducer for the issue this commit solves. Make the bcast algorithm decision an enumerated variable Resolve various asset failures when destructing coll ml requests. Two issues: - Always reset the request to be invalid before returning it to the free list. This will avoid an asset in ompi_request_t's destructor. OMPI_REQUEST_FINI does this (and also releases the fortran handle index). - Never explicitly construct or destruct the superclass of an opal object. This screws up the class function tables and will cause either an assert failure or a segmentation fault when destructing coll ml requests. Cleanup allgather. I removed the duplicate non-blocking and blocking functions and modeled the cleanup after what I found in allreduce. Also cleaned up the code somewhat. Don't bother copying from the send to the recieve buffer in bcol_basesmuma_allreduce_intra_fanin_fanout if the pointers are the same. The eliminates a warning about memcpy and aliasing and avoids an unnecessary call to memcpy. Alwasy call CHECK_AND_RELEASE on memsync collectives. There was a call to OBJ_RELEASE on the collective communicator but because CHECK_AND_RECYLCE was never called there was not matching call to OBJ_RELEASE. This caused coll ml to leak communicators. Make allreduce use the sequential collective launcher in coll_ml_inlines.h Just launch the next collective in the component progress. I am a little unsure about this patch. There appears to be some sort of race between collectives that causes buffer exhaustion in some cases (IMB Allreduce is a reproducer). Changing progress to only launch the next bcol seems to resolve the issue but might not be the best fix. Note that I see little-no performance penalty for this change. Fix allreduce when there are extra sources. There was an issue with the buffer offset calculation when there are extra sources. In the case of extra sources == 1 the offset was set to buffer_size (just past the header of the next buffer). I adjusted the buffer size to take into accoun the maximum header size (see the earlier commit that added this) and simplified the offset calculation. Make reduce/allreduce non-blocking. This is required for MPI_Comm_idup to work correctly. This has been tested with various layouts using the ibm testsuite and imb and appears to have the same performance as the old blocking version. Fix allgather for non-contiguous layouts and simplify parsing the topology. Some things in this patch: - There were several comments to the effect that level 0 of the hierarchy MUST contain all of the ranks. At least one function made this assumption but it was not true. I changed the sbgp components and the coll ml initization code to enforce this requirement. - Ensure that hierarchy level 0 has the ranks in the correct scatter gather order. This removes the need for a separate sort list and fixes the offset calculation for allgather. - There were several passes over the hierarchy to determine properties of the hierarchy. I eliminated these extra passes and the memory allocation associated with them and calculate the tree properties on the fly. The same DFS recursion also handles the re-order of level 0. All these changes have been verified with MPI_Allreduce, MPI_Reduce, and MPI_Allgather. All functions now pass all IBM/Open MPI, and IMB tests. coll/ml: correct pointer usage for MPI_BOTTOM Since contiguous datatypes are copied via memcpy (bypassing the convertor) we need to adjust for the lb of the datatype. This corrects problems found testing code that uses MPI_BOTTOM (NULL) as the send pointer. Add fallback collectives for allreduce and reduce. cmr=v1.7.5:reviewer=pasha This commit was SVN r30363.
2014-01-22 19:39:19 +04:00
if (!strcasecmp(name,"ML_SMALL_DATA_REDUCE"))
return ML_SMALL_DATA_ALLREDUCE;
if (!strcasecmp(name,"ML_LARGE_DATA_REDUCE"))
return ML_LARGE_DATA_ALLREDUCE;
if (!strcasecmp(name,"ML_SMALL_DATA_REDUCE"))
return ML_SMALL_DATA_REDUCE;
if (!strcasecmp(name,"ML_LARGE_DATA_REDUCE"))
return ML_LARGE_DATA_REDUCE;
if (!strcasecmp(name,"ML_NUM_ALLREDUCE_FUNCTIONS"))
return ML_NUM_ALLREDUCE_FUNCTIONS;
if (!strcasecmp(name,"ML_SMALL_DATA_ALLTOALL"))
return ML_SMALL_DATA_ALLTOALL;
if (!strcasecmp(name,"ML_LARGE_DATA_ALLTOALL"))
return ML_LARGE_DATA_ALLTOALL;
if (!strcasecmp(name,"ML_NUM_ALLTOALL_FUNCTIONS"))
return ML_NUM_ALLTOALL_FUNCTIONS;
if (!strcasecmp(name,"ML_SMALL_DATA_ALLGATHER"))
return ML_SMALL_DATA_ALLGATHER;
if (!strcasecmp(name,"ML_LARGE_DATA_ALLGATHER"))
return ML_LARGE_DATA_ALLGATHER;
if (!strcasecmp(name,"ML_NUM_ALLGATHER_FUNCTIONS"))
return ML_NUM_ALLGATHER_FUNCTIONS;
if (!strcasecmp(name,"ML_SMALL_DATA_GATHER"))
return ML_SMALL_DATA_GATHER;
if (!strcasecmp(name,"ML_LARGE_DATA_GATHER"))
return ML_LARGE_DATA_GATHER;
if (!strcasecmp(name,"ML_NUM_GATHER_FUNCTIONS"))
return ML_NUM_GATHER_FUNCTIONS;
if (!strcasecmp(name,"ML_BARRIER_DEFAULT"))
return ML_BARRIER_DEFAULT;
/* ERROR */
return ML_UNDEFINED;
}
static int hierarchy_name_to_id(char *name)
{
assert (NULL != name);
if (!strcasecmp(name, "FULL_HR")) {
return COLL_ML_HR_FULL;
}
if (!strcasecmp(name, "FULL_HR_NO_BASESOCKET")) {
return COLL_ML_HR_NBS;
}
if (!strcasecmp(name, "PTP_ONLY")) {
return COLL_ML_HR_SINGLE_PTP;
}
if (!strcasecmp(name, "IBOFFLOAD_ONLY")) {
return COLL_ML_HR_SINGLE_IBOFFLOAD;
}
/* Error */
return ML_UNDEFINED;
}
static int section_name_to_id(char *name)
{
assert (NULL != name);
if (!strcasecmp(name, "SMALL")) {
return ML_SMALL_MSG;
}
if (!strcasecmp(name, "LARGE")) {
return ML_LARGE_MSG;
}
/* Error */
return ML_UNDEFINED;
}
static int coll_name_to_id(char *name)
{
assert (NULL != name);
if (!strcasecmp(name, "ALLGATHER")) {
return ML_ALLGATHER;
}
if (!strcasecmp(name, "ALLGATHERV")) {
return ML_ALLGATHERV;
}
if (!strcasecmp(name, "ALLREDUCE")) {
return ML_ALLREDUCE;
}
if (!strcasecmp(name, "ALLTOALL")) {
return ML_ALLTOALL;
}
if (!strcasecmp(name, "ALLTOALLV")) {
return ML_ALLTOALLV;
}
if (!strcasecmp(name, "ALLTOALLW")) {
return ML_ALLTOALLW;
}
if (!strcasecmp(name, "ALLTOALLW")) {
return ML_ALLTOALLW;
}
if (!strcasecmp(name, "BARRIER")) {
return ML_BARRIER;
}
if (!strcasecmp(name, "BCAST")) {
return ML_BCAST;
}
if (!strcasecmp(name, "EXSCAN")) {
return ML_EXSCAN;
}
if (!strcasecmp(name, "GATHER")) {
return ML_GATHER;
}
if (!strcasecmp(name, "GATHERV")) {
return ML_GATHERV;
}
if (!strcasecmp(name, "REDUCE")) {
return ML_REDUCE;
}
if (!strcasecmp(name, "REDUCE_SCATTER")) {
return ML_REDUCE_SCATTER;
}
if (!strcasecmp(name, "SCAN")) {
return ML_SCAN;
}
if (!strcasecmp(name, "SCATTER")) {
return ML_SCATTER;
}
if (!strcasecmp(name, "SCATTERV")) {
return ML_SCATTERV;
}
/* nonblocking functions */
if (!strcasecmp(name, "IALLGATHER")) {
return ML_IALLGATHER;
}
if (!strcasecmp(name, "IALLGATHERV")) {
return ML_IALLGATHERV;
}
if (!strcasecmp(name, "IALLREDUCE")) {
return ML_IALLREDUCE;
}
if (!strcasecmp(name, "IALLTOALL")) {
return ML_IALLTOALL;
}
if (!strcasecmp(name, "IALLTOALLV")) {
return ML_IALLTOALLV;
}
if (!strcasecmp(name, "IALLTOALLW")) {
return ML_IALLTOALLW;
}
if (!strcasecmp(name, "IALLTOALLW")) {
return ML_IALLTOALLW;
}
if (!strcasecmp(name, "IBARRIER")) {
return ML_IBARRIER;
}
if (!strcasecmp(name, "IBCAST")) {
return ML_IBCAST;
}
if (!strcasecmp(name, "IEXSCAN")) {
return ML_IEXSCAN;
}
if (!strcasecmp(name, "IGATHER")) {
return ML_IGATHER;
}
if (!strcasecmp(name, "IGATHERV")) {
return ML_IGATHERV;
}
if (!strcasecmp(name, "IREDUCE")) {
return ML_IREDUCE;
}
if (!strcasecmp(name, "IREDUCE_SCATTER")) {
return ML_IREDUCE_SCATTER;
}
if (!strcasecmp(name, "ISCAN")) {
return ML_ISCAN;
}
if (!strcasecmp(name, "ISCATTER")) {
return ML_ISCATTER;
}
if (!strcasecmp(name, "ISCATTERV")) {
return ML_ISCATTERV;
}
/* Error - collecives name was not matched */
return ML_UNDEFINED;
}
static int set_collective_name(coll_config_t *coll_config)
{
int coll_id =
coll_name_to_id(coll_ml_config_yytext);
if (ML_UNDEFINED == coll_id) {
return OMPI_ERROR;
}
coll_config->coll_id = coll_id;
coll_config->coll_name = strdup(coll_ml_config_yytext);
return OMPI_SUCCESS;
}
static int set_section_name(section_config_t *section_config)
{
int section_id;
section_id = section_name_to_id(coll_ml_config_yytext);
if (ML_UNDEFINED == section_id) {
return OMPI_ERROR;
}
section_config->section_id = section_id;
section_config->section_name = strdup(coll_ml_config_yytext);
return OMPI_SUCCESS;
}
void mca_coll_ml_reset_config(per_collective_configuration_t *config)
{
config->topology_id = ML_UNDEFINED;
config->threshold = ML_UNDEFINED;
config->algorithm_id = ML_UNDEFINED;
config->fragmentation_enabled = ML_UNDEFINED;
}
static void reset_section(section_config_t *section_cf)
{
if (section_cf->section_name) {
free (section_cf->section_name);
section_cf->section_name = NULL;
}
section_cf->section_id = ML_UNDEFINED;
mca_coll_ml_reset_config(&section_cf->config);
}
static void reset_collective(coll_config_t *coll_cf)
{
if (coll_cf->coll_name) {
free (coll_cf->coll_name);
coll_cf->coll_name = NULL;
}
coll_cf->coll_id = ML_UNDEFINED;
reset_section(&coll_cf->section);
}
/*
* String to integer;
*/
static int string_to_int(char *str)
{
while (isspace(*str)) {
++str;
}
/* Nope -- just decimal, so use atoi() */
return atoi(str);
}
static int parse_algorithm_key(section_config_t *section, char *value)
{
int ret;
ret = algorithm_name_to_id(value);
if (ML_UNDEFINED == ret) {
return OMPI_ERROR;
} else {
section->config.algorithm_id = ret;
}
return OMPI_SUCCESS;
}
static int parse_threshold_key(section_config_t *section, char *value)
{
assert (NULL != value);
if(!strcasecmp(value, "unlimited")) {
section->config.threshold = -1;
} else {
section->config.threshold = string_to_int(value);
}
return OMPI_SUCCESS;
}
static int parse_hierarchy_key(section_config_t *section, char *value)
{
int ret;
ret = hierarchy_name_to_id(value);
if (ML_UNDEFINED == ret) {
return OMPI_ERROR;
}
section->config.topology_id = ret;
return OMPI_SUCCESS;
}
static int parse_fragmentation_key(section_config_t *section, char *value)
{
assert (NULL != value);
if(!strcasecmp(value, "enable")) {
section->config.fragmentation_enabled = 1;
} else if (!strcasecmp(value, "disable")) {
section->config.fragmentation_enabled = 0;
} else {
ML_ERROR(("Line %d, unexpected fragmentation value %s. Legal values are: enable/disable",
coll_ml_config_yynewlines, value));
return OMPI_ERROR;
}
return OMPI_SUCCESS;
}
/* Save configuration that have been collected so far */
static int save_settings(coll_config_t *coll_config)
{
per_collective_configuration_t *cf;
if (ML_UNDEFINED == coll_config->coll_id || ML_UNDEFINED == coll_config->section.section_id) {
return OMPI_ERROR;
}
cf = &mca_coll_ml_component.coll_config[coll_config->coll_id][coll_config->section.section_id];
cf->topology_id = coll_config->section.config.topology_id;
cf->threshold = coll_config->section.config.threshold;
cf->algorithm_id = coll_config->section.config.algorithm_id;
cf->fragmentation_enabled = coll_config->section.config.fragmentation_enabled;
return OMPI_SUCCESS;
}
/*
* Parse a single line
*/
static int parse_line(section_config_t *section)
{
int val, ret = OMPI_SUCCESS;
char *value = NULL;
/* Save the name name */
if (key_buffer_len < strlen(coll_ml_config_yytext) + 1) {
char *tmp;
key_buffer_len = strlen(coll_ml_config_yytext) + 1;
tmp = (char *) realloc(key_buffer, key_buffer_len);
if (NULL == tmp) {
free(key_buffer);
key_buffer_len = 0;
key_buffer = NULL;
return OMPI_ERR_TEMP_OUT_OF_RESOURCE;
}
key_buffer = tmp;
}
strncpy(key_buffer, coll_ml_config_yytext, key_buffer_len);
/* The first thing we have to see is an "=" */
val = coll_ml_config_yylex();
if (coll_ml_config_parse_done || COLL_ML_CONFIG_PARSE_EQUAL != val) {
ML_ERROR(("Line %d, expected = before key: %s",
coll_ml_config_yynewlines,
key_buffer));
return OMPI_ERROR;
}
/* Next we get the value */
val = coll_ml_config_yylex();
if (COLL_ML_CONFIG_PARSE_SINGLE_WORD == val ||
COLL_ML_CONFIG_PARSE_VALUE == val) {
value = strdup(coll_ml_config_yytext);
/* Now we need to see the newline */
val = coll_ml_config_yylex();
if (COLL_ML_CONFIG_PARSE_NEWLINE != val &&
COLL_ML_CONFIG_PARSE_DONE != val) {
ML_ERROR(("Line %d, expected new line after %s",
coll_ml_config_yynewlines,
key_buffer));
free(value);
return OMPI_ERROR;
}
}
/* If we did not get EOL or EOF, something is wrong */
else if (COLL_ML_CONFIG_PARSE_DONE != val &&
COLL_ML_CONFIG_PARSE_NEWLINE != val) {
ML_ERROR(("Line %d, expected new line or end of line",
coll_ml_config_yynewlines));
ret = OMPI_ERROR;
goto Error;
}
/* Line parsing is done, read the values */
if (!strcasecmp(key_buffer, "algorithm")) {
ret = parse_algorithm_key(section, value);
if (OMPI_SUCCESS != ret) {
goto Error;
}
}
else if (!strcasecmp(key_buffer, "threshold")) {
ret = parse_threshold_key(section, value);
if (OMPI_SUCCESS != ret) {
goto Error;
}
}
else if (!strcasecmp(key_buffer, "hierarchy")) {
ret = parse_hierarchy_key(section, value);
if (OMPI_SUCCESS != ret) {
goto Error;
}
}
else if (!strcasecmp(key_buffer, "fragmentation")) {
ret = parse_fragmentation_key(section, value);
if (OMPI_SUCCESS != ret) {
goto Error;
}
/* Failed to parse the key */
} else {
ML_ERROR(("Line %d, unknown key %s",
coll_ml_config_yynewlines, key_buffer));
}
/* All done */
Error:
if (NULL != value) {
free(value);
}
return ret;
}
/**************************************************************************/
/*
* Parse a single file
*/
static int parse_file(char *filename)
{
int val;
int ret = OMPI_SUCCESS;
bool first_section = true, first_coll = true;
coll_config_t coll_config;
memset (&coll_config, 0, sizeof (coll_config));
reset_collective(&coll_config);
/* Open the file */
coll_ml_config_yyin = fopen(filename, "r");
if (NULL == coll_ml_config_yyin) {
ML_ERROR(("Failed to open config file %s", filename));
ret = OMPI_ERR_NOT_FOUND;
goto cleanup;
}
/* Do the parsing */
coll_ml_config_parse_done = false;
coll_ml_config_yynewlines = 1;
coll_ml_config_init_buffer(coll_ml_config_yyin);
while (!coll_ml_config_parse_done) {
val = coll_ml_config_yylex();
switch (val) {
case COLL_ML_CONFIG_PARSE_DONE:
case COLL_ML_CONFIG_PARSE_NEWLINE:
break;
case COLL_ML_CONFIG_PARSE_COLLECTIVE:
/* dump all the information to last section that was defined */
if (!first_coll) {
ret = save_settings(&coll_config);
if (OMPI_SUCCESS != ret) {
ML_ERROR(("Error in syntax for collective %s", coll_config.coll_name));
goto cleanup;
}
}
/* reset collective config */
reset_collective(&coll_config);
first_coll = false;
first_section = true;
ret = set_collective_name(&coll_config);
if (OMPI_SUCCESS != ret) {
goto cleanup;
}
break;
case COLL_ML_CONFIG_PARSE_SECTION:
if (ML_UNDEFINED == coll_config.coll_id) {
ML_ERROR(("Collective section wasn't defined !"));
ret = OMPI_ERROR;
goto cleanup;
}
if (!first_section) {
/* dump all the information to last section that was defined */
ret = save_settings(&coll_config);
if (OMPI_SUCCESS != ret) {
ML_ERROR(("Error in syntax for collective %s section %s", coll_config.coll_name,
coll_config.section.section_name));
goto cleanup;
}
}
first_section = false;
/* reset all section values */
reset_section(&coll_config.section);
/* set new section name */
ret = set_section_name(&coll_config.section);
if (OMPI_SUCCESS != ret) {
goto cleanup;
}
break;
case COLL_ML_CONFIG_PARSE_SINGLE_WORD:
if (ML_UNDEFINED == coll_config.coll_id ||
ML_UNDEFINED == coll_config.section.section_id) {
ML_ERROR(("Collective section or sub-section was not defined !"));
ret = OMPI_ERROR;
goto cleanup;
} else {
parse_line(&coll_config.section);
}
break;
default:
/* anything else is an error */
ML_ERROR(("Unexpected token!"));
ret = OMPI_ERROR;
goto cleanup;
break;
}
}
save_settings(&coll_config);
fclose(coll_ml_config_yyin);
coll_ml_config_yylex_destroy ();
ret = OMPI_SUCCESS;
cleanup:
reset_collective(&coll_config);
if (NULL != key_buffer) {
free(key_buffer);
key_buffer = NULL;
key_buffer_len = 0;
}
return ret;
}
int mca_coll_ml_config_file_init(void)
{
return parse_file(mca_coll_ml_component.config_file_name);
}