1
1
openmpi/ompi/tools/ompi-profiler/ompi-profiler.c

462 строки
15 KiB
C
Исходник Обычный вид История

Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
/*
* Copyright (c) 2004-2007 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) 2007 Cisco Systems, Inc. All rights reserved.
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
* Copyright (c) 2007 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "opal_config.h"
#include "opal/constants.h"
#include "opal/version.h"
#ifdef HAVE_STRING_H
#include <string.h>
#endif
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
#include <stdio.h>
#include <stdlib.h>
#include <ctype.h>
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#ifdef HAVE_SYS_STAT_H
#include <sys/stat.h>
#endif
#include <fcntl.h>
#include <errno.h>
#include <signal.h>
#include "opal/dss/dss.h"
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
#include "opal/class/opal_list.h"
#include "opal/util/argv.h"
#include "opal/runtime/opal.h"
#include "opal/util/cmd_line.h"
#include "opal/mca/base/base.h"
#include "opal/util/opal_environ.h"
#include "opal/util/show_help.h"
/*
* Globals
*/
typedef struct {
opal_list_item_t item;
char *name;
opal_list_t frameworks;
} orte_profile_node_t;
static void node_constructor(orte_profile_node_t *ptr)
{
ptr->name = NULL;
OBJ_CONSTRUCT(&ptr->frameworks, opal_list_t);
}
static void node_destructor(orte_profile_node_t *ptr)
{
if (NULL != ptr->name) {
free(ptr->name);
}
OBJ_DESTRUCT(&ptr->frameworks);
}
OBJ_CLASS_INSTANCE(orte_profile_node_t,
opal_list_item_t,
node_constructor,
node_destructor);
typedef struct {
opal_list_item_t item;
int num_nodes;
char *framework;
char *component;
char *params;
} orte_profile_t;
static void profile_constructor(orte_profile_t *ptr)
{
ptr->num_nodes = 0;
ptr->framework = NULL;
ptr->component = NULL;
ptr->params = NULL;
}
static void profile_destructor(orte_profile_t *ptr)
{
if (NULL != ptr->framework) {
free(ptr->framework);
}
if (NULL != ptr->component) {
free(ptr->component);
}
if (NULL != ptr->params) {
free(ptr->params);
}
}
OBJ_CLASS_INSTANCE(orte_profile_t,
opal_list_item_t,
profile_constructor,
profile_destructor);
static void read_file(opal_list_t *nodes, FILE *fp);
/* global variables */
static bool help = false;
static bool version = false;
static bool verbose = false;
static bool configout = false;
static char *profilefile = NULL;
static bool report = false;
static opal_cmd_line_init_t cmd_line_init[] = {
/* Various "obvious" options */
{ NULL, NULL, NULL, 'h', "help", "help", 0,
&help, OPAL_CMD_LINE_TYPE_BOOL,
"This help message" },
{ NULL, NULL, NULL, 'V', "version", "version", 0,
&version, OPAL_CMD_LINE_TYPE_BOOL,
"Print version and exit" },
{ NULL, NULL, NULL, '\0', "verbose", "verbose", 0,
&verbose, OPAL_CMD_LINE_TYPE_BOOL,
"Print version and exit" },
{ NULL, NULL, NULL, '\0', "config", "config", 0,
&configout, OPAL_CMD_LINE_TYPE_BOOL,
"Print framework/component usage" },
{ NULL, NULL, NULL, '\0', "profile", "profile", 1,
&profilefile, OPAL_CMD_LINE_TYPE_STRING,
"File to update with system profile parameters" },
{ NULL, NULL, NULL, '\0', "report", "report", 0,
&report, OPAL_CMD_LINE_TYPE_BOOL,
"Print out a report of the data in the given profile file" },
/* End of list */
{ NULL, NULL, NULL, '\0', NULL, NULL, 0,
NULL, OPAL_CMD_LINE_TYPE_NULL, NULL }
};
int main(int argc, char *argv[])
{
opal_list_t nodes, frames;
opal_list_item_t *item, *item2, *itemold;
orte_profile_node_t *node;
orte_profile_t *profile, *frame;
int ret;
int len;
opal_cmd_line_t cmd_line;
char *args = NULL;
char *configfile=NULL;
char *cmd;
opal_cmd_line_create(&cmd_line, cmd_line_init);
mca_base_cmd_line_setup(&cmd_line);
if (OPAL_SUCCESS != (ret = opal_cmd_line_parse(&cmd_line, true,
argc, argv)) ) {
return ret;
}
/* init enough of opal to use a few utilities */
if (OPAL_SUCCESS != opal_init_util()) {
fprintf(stderr, "OPAL failed to initialize -- ompi-profiler aborting\n");
exit(1);
}
/* check for some simple options */
if (help) {
args = opal_cmd_line_get_usage_msg(&cmd_line);
opal_show_help("help-ompi-profiler.txt", "ompi-profiler:usage", false,
OPAL_VERSION, args, PACKAGE_BUGREPORT);
free(args);
/* If someone asks for help, that should be all we do */
exit(0);
}
if (version) {
opal_show_help("help-ompi-profiler.txt", "ompi-profiler:version", false,
OPAL_VERSION, PACKAGE_BUGREPORT);
exit(0);
}
if (report) {
int fd;
int32_t num_bytes, cnt;
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
char *nodename, *attr;
opal_byte_object_t bo;
opal_buffer_t buf;
uint8_t *dump;
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
/* just read the given file and print out a report */
if (NULL == profilefile) {
opal_show_help("help-ompi-profiler.txt", "ompi-profiler:report-wo-file", false);
exit(1);
}
fd = open(profilefile, O_RDONLY);
if (fd < 0) {
opal_show_help("help-ompi-profiler.txt", "ompi-profiler:report-file-not-found", false);
exit(1);
}
/* loop through file until end */
while (0 < read(fd, &bo.size, sizeof(bo.size))) {
/* this is the number of bytes in the byte object */
bo.bytes = malloc(bo.size);
if (0 > read(fd, bo.bytes, bo.size)) {
fprintf(stderr, "ompi-profiler: unable to read file\n");
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
close(fd);
exit(1);
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
}
/* setup to unpack the object */
OBJ_CONSTRUCT(&buf, opal_buffer_t);
opal_dss.load(&buf, bo.bytes, bo.size);
/* unpack the nodename */
cnt = 1;
if (OPAL_SUCCESS != opal_dss.unpack(&buf, &nodename, &cnt, OPAL_STRING)) {
fprintf(stderr, "ompi-profiler: could not unpack node name\n");
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
close(fd);
exit(1);
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
}
/* loop through the rest of the object to unpack the attr's themselves */
while (OPAL_SUCCESS == opal_dss.unpack(&buf, &attr, &cnt, OPAL_STRING)) {
/* read the number of bytes in the blob */
cnt = 1;
if (OPAL_SUCCESS != opal_dss.unpack(&buf, &num_bytes, &cnt, OPAL_INT32)) {
fprintf(stderr, "ompi-profiler: data size not found\n");
close(fd);
exit(0);
}
/* unpack the bytes just so we can dump them */
dump = malloc(num_bytes);
if (OPAL_SUCCESS != opal_dss.unpack(&buf, dump, &num_bytes, OPAL_BYTE)) {
fprintf(stderr, "ompi-profiler: data not found\n");
close(fd);
exit(0);
}
free(dump);
/* report the results */
fprintf(stdout, "Node %s reported %d bytes for attribute %s\n",
nodename, num_bytes, attr);
free(attr);
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
}
free(nodename);
OBJ_DESTRUCT(&buf);
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
}
close(fd);
Enable modex-less launch. Consists of: 1. minor modification to include two new opal MCA params: (a) opal_profile: outputs what components were selected by each framework currently enabled for most, but not all, frameworks (b) opal_profile_file: name of file that contains profile info required for modex 2. introduction of two new tools: (a) ompi-probe: MPI process that simply calls MPI_Init/Finalize with opal_profile set. Also reports back the rml IP address for all interfaces on the node (b) ompi-profiler: uses ompi-probe to create the profile_file, also reports out a summary of what framework components are actually being used to help with configuration options 3. modification of the grpcomm basic component to utilize the profile file in place of the modex where possible 4. modification of orterun so it properly sees opal mca params and handles opal_profile correctly to ensure we don't get its profile 5. similar mod to orted as for orterun 6. addition of new test that calls orte_init followed by calls to grpcomm.barrier This is all completely benign unless actively selected. At the moment, it only supports modex-less launch for openib-based systems. Minor mod to the TCP btl would be required to enable it as well, if people are interested. Similarly, anyone interested in enabling other BTL's for modex-less operation should let me know and I'll give you the magic details. This seems to significantly improve scalability provided the file can be locally located on the nodes. I'm looking at an alternative means of disseminating the info (perhaps in launch message) as an option for removing that constraint. This commit was SVN r20098.
2008-12-10 02:49:02 +03:00
exit(0);
}
/* do a quick sanity check - since they didn't want a report, see if they don't
* want -anything-
*/
if (!configout && NULL == profilefile) {
/* save us the annoyance - you have to want -something-! */
fprintf(stderr, "ompi-profiler: no options specified - aborting\n");
exit(1);
}
/* setup the cmd to execute */
if (configout) {
asprintf(&configfile, "profiler.%d", getpid());
} else {
configfile = strdup("/dev/null");
}
if (NULL != profilefile) {
asprintf(&cmd, "mpirun -pernode -mca opal_profile 1 -mca opal_profile_file %s -mca grpcomm basic ompi-probe >& %s",
profilefile, configfile);
} else {
asprintf(&cmd, "mpirun -pernode -mca opal_profile 1 -mca grpcomm basic ompi-probe >& %s",
configfile);
}
if (verbose) {
fprintf(stderr, "ompi-profiler: executing %s\n", cmd);
}
/* execute it */
if (0 > system(cmd)) {
fprintf(stderr, "ompi-profiler: could not execute cmd %s\n", cmd);
free(cmd);
goto CLEANUP;
}
free(cmd);
/* did they want the configuration output? */
if (configout) {
FILE *fp;
struct stat buf;
/* does the file already exist? */
if (0 != stat(configfile, &buf)) {
/* file must not have been created */
fprintf(stderr, "Temporary output file %s could not be found - config report cannot be generated\n", configfile);
goto CLEANUP;
}
/* yes - read the info so we can output it */
fp = fopen(configfile, "r");
if (NULL == fp) {
fprintf(stderr, "Impossible to open the file %s in read mode\n", configfile );
goto CLEANUP;
}
OBJ_CONSTRUCT(&nodes, opal_list_t);
read_file(&nodes, fp);
fclose(fp);
/* setup a list of framework info */
OBJ_CONSTRUCT(&frames, opal_list_t);
len = opal_list_get_size(&nodes);
/* convert the results over to the new list */
while (NULL != (item = opal_list_remove_first(&nodes))) {
node = (orte_profile_node_t*)item;
/* loop through this node's frameworks */
item2 = opal_list_get_first(&node->frameworks);
while (item2 != opal_list_get_end(&node->frameworks)) {
profile = (orte_profile_t*)item2;
/* is this framework already in our list? */
for (itemold = opal_list_get_first(&frames);
itemold != opal_list_get_end(&frames);
itemold = opal_list_get_next(itemold)) {
frame = (orte_profile_t*)itemold;
if (0 == strcmp(profile->framework, frame->framework) &&
0 == strcmp(profile->component, frame->component)) {
/* all matches - increment # matches */
frame->num_nodes++;
goto COMPLETE;
}
}
/* get here if the framework/component is new */
frame = OBJ_NEW(orte_profile_t);
frame->num_nodes++;
frame->framework = strdup(profile->framework);
frame->component = strdup(profile->component);
opal_list_append(&frames, &frame->item);
COMPLETE:
item2 = opal_list_get_next(item2);
}
OBJ_RELEASE(item);
}
OBJ_DESTRUCT(&nodes);
/* output the list */
while (NULL != (itemold = opal_list_remove_first(&frames))) {
frame = (orte_profile_t*)itemold;
if (len == frame->num_nodes) {
fprintf(stderr, "All nodes use framework %s component %s\n", frame->framework, frame->component);
} else {
fprintf(stderr, "%d nodes use framework %s component %s\n", frame->num_nodes, frame->framework, frame->component);
}
OBJ_RELEASE(frame);
}
OBJ_DESTRUCT(&frames);
}
CLEANUP:
/* remove the file */
if (NULL != configfile) {
if (0 != strcmp("/dev/null", configfile)) {
unlink(configfile);
}
free(configfile);
}
return 0;
}
static void read_file(opal_list_t *nodes, FILE *fp)
{
char line[1024];
char *endprefix, *endnodename, *data, *nodename;
char **inputs;
opal_list_item_t *item;
orte_profile_node_t *node;
orte_profile_t *profile;
int len;
memset(line, 0, sizeof(line));
while (NULL != fgets(line, sizeof(line), fp)) {
/* get the length of the line */
len = strlen(line);
/* remove any trailing newline */
if (line[len-1] == '\n') {
line[len-1] = '\0';
}
if ('[' != line[0]) {
/* indicates empty line - ignore it */
continue;
}
if (NULL == (endprefix = strchr(line, ']'))) {
fprintf(stderr, "ompi-profiler: read bad input for ] %s\n", line);
continue;
}
*endprefix = '\0';
/* break the prefix at the colon - we don't need the pid */
if (NULL == (endnodename = strchr(line, ':'))) {
fprintf(stderr, "ompi-profiler: read bad input for : %s\n", line);
continue;
}
*endnodename = '\0';
nodename = &line[1];
/* is this node already in our list */
for (item = opal_list_get_first(nodes);
item != opal_list_get_end(nodes);
item = opal_list_get_next(item)) {
node = (orte_profile_node_t*)item;
if (0 == strcmp(node->name, nodename)) {
/* already present - just add to it */
goto PROCESS;
}
}
/* if we got here, then it wasn't found */
node = OBJ_NEW(orte_profile_node_t);
node->name = strdup(nodename);
opal_list_append(nodes, &node->item);
PROCESS:
/* point to the rest of the data */
data = endprefix;
data += 2; /* get past space */
/* use an opal utility to parse it */
if ((NULL == (inputs = opal_argv_split(data, ':'))) ||
opal_argv_count(inputs) < 2) {
fprintf(stderr, "ompi-profiler: read bad input for second : %s\n", data);
opal_argv_free(inputs);
continue;
}
/* first entry must be the framework - see if we already have it */
for (item = opal_list_get_first(&node->frameworks);
item != opal_list_get_end(&node->frameworks);
item = opal_list_get_next(item)) {
profile = (orte_profile_t*)item;
if (0 == strcmp(inputs[0], profile->framework)) {
/* this will happen if mpirun outputs some of the
* frameworks AND a proc is local to mpirun
*/
goto SKIP;
}
}
profile = OBJ_NEW(orte_profile_t);
profile->framework = strdup(inputs[0]);
/* second entry is component */
profile->component = strdup(inputs[1]);
/* if there is anything more, just save it */
if (NULL != inputs[2]) {
profile->params = opal_argv_join(&inputs[2], ':');
}
opal_list_append(&node->frameworks, &profile->item);
SKIP:
opal_argv_free(inputs);
memset(line, 0, sizeof(line));
}
}