int chpl_launch(int argc, char* argv[], int32_t numLocales) {
int len = strlen(WRAP_TO_STR(LAUNCH_PATH)) + strlen("gasnetrun_ibv") + 1;
char *cmd = chpl_mem_allocMany(len, sizeof(char), CHPL_RT_MD_COMMAND_BUFFER, -1, 0);
sprintf(cmd, "%sgasnetrun_ibv", WRAP_TO_STR(LAUNCH_PATH));
return chpl_launch_using_exec(cmd,
chpl_launch_create_argv(cmd, argc, argv,
numLocales),
argv[0]);
}
int chpl_launch(int argc, char* argv[], int32_t numLocales) {
int len = strlen(CHPL_THIRD_PARTY) + strlen(WRAP_TO_STR(LAUNCH_PATH)) + strlen("amudprun") + 2;
char *cmd = chpl_mem_allocMany(len, sizeof(char), CHPL_RT_MD_COMMAND_BUFFER, -1, 0);
snprintf(cmd, len, "%s/%samudprun", CHPL_THIRD_PARTY, WRAP_TO_STR(LAUNCH_PATH));
return chpl_launch_using_exec(cmd,
chpl_launch_create_argv(cmd, argc, argv,
numLocales),
argv[0]);
}
static char* chpl_launch_create_command(int argc, char* argv[],
int32_t numLocales) {
int i;
int size;
char baseCommand[256];
char* command;
FILE* slurmFile, *expectFile;
char* projectString = getenv(launcherAccountEnvvar);
char* constraint = getenv("CHPL_LAUNCHER_CONSTRAINT");
char* walltime = getenv("CHPL_LAUNCHER_WALLTIME");
char* outputfn = getenv("CHPL_LAUNCHER_SLURM_OUTPUT_FILENAME");
char* basenamePtr = strrchr(argv[0], '/');
pid_t mypid;
if (basenamePtr == NULL) {
basenamePtr = argv[0];
} else {
basenamePtr++;
}
chpl_compute_real_binary_name(argv[0]);
#ifndef DEBUG_LAUNCH
mypid = getpid();
#else
mypid = 0;
#endif
sprintf(sysFilename, "%s%d", baseSysFilename, (int)mypid);
sprintf(expectFilename, "%s%d", baseExpectFilename, (int)mypid);
sprintf(slurmFilename, "%s%d", baseSBATCHFilename, (int)mypid);
if (getenv("CHPL_LAUNCHER_USE_SBATCH") != NULL) {
slurmFile = fopen(slurmFilename, "w");
fprintf(slurmFile, "#!/bin/sh\n\n");
fprintf(slurmFile, "#SBATCH -J Chpl-%.10s\n", basenamePtr);
genNumLocalesOptions(slurmFile, determineQsubVersion(), numLocales, getNumCoresPerLocale());
if (projectString && strlen(projectString) > 0)
fprintf(slurmFile, "#SBATCH -A %s\n", projectString);
if (getenv("CHPL_LAUNCHER_USE_SBATCH") != NULL) {
// fprintf(slurmFile, "#SBATCH -joe\n");
if (outputfn!=NULL)
fprintf(slurmFile, "#SBATCH -o %s.%%j.out\n", outputfn);
else
fprintf(slurmFile, "#SBATCH -o %s.%%j.out\n", argv[0]);
// fprintf(slurmFile, "cd $SBATCH_O_WORKDIR\n");
fprintf(slurmFile, "%s/gasnetrun_ibv -n %d %s ",
WRAP_TO_STR(LAUNCH_PATH), numLocales, chpl_get_real_binary_name());
for (i=1; i<argc; i++) {
fprintf(slurmFile, " '%s'", argv[i]);
}
fprintf(slurmFile, "\n");
}
fclose(slurmFile);
chmod( slurmFilename, 0755);
}
if (getenv("CHPL_LAUNCHER_USE_SBATCH") == NULL) {
expectFile = fopen(expectFilename, "w");
if (verbosity < 2) {
// fprintf(expectFile, "log_user 0\n");
}
fprintf(expectFile, "set timeout -1\n");
// fprintf(expectFile, "chmod +x %s\n",slurmFilename);
fprintf(expectFile, "set prompt \"(%%|#|\\\\$|>) $\"\n");
// fprintf(expectFile, "spawn sbatch ");
fprintf(expectFile, "spawn -noecho salloc ");
fprintf(expectFile, "-J %.10s ",basenamePtr); // pass
fprintf(expectFile, "-N %d ",numLocales);
fprintf(expectFile, "--ntasks-per-node=1 ",numLocales);
fprintf(expectFile, "--exclusive "); // give exclusive access to the nodes
fprintf(expectFile, "--time=%s ",walltime);
if (constraint) {
fprintf(expectFile, " -C %s", constraint);
}
// fprintf(expectFile, "-I %s ", slurmFilename);
fprintf(expectFile, " %s/gasnetrun_ibv -n %d %s ",
WRAP_TO_STR(LAUNCH_PATH), numLocales, chpl_get_real_binary_name());
for (i=1; i<argc; i++) {
fprintf(expectFile, " %s", argv[i]);
}
// fprintf(expectFile, "\\n\"\n");
fprintf(expectFile, "\n\n");
// fprintf(expectFile, "expect -re $prompt\n");
// fprintf(expectFile, "send \"cd \\$SBATCH_O_WORKDIR\\n\"\n");
// fprintf(expectFile, "expect -re $prompt\n");
// fprintf(expectFile, "sleep 10\n");
// fprintf(expectFile, "interact -o -re $prompt {return}\n");
// fprintf(expectFile, "send_user \"\\n\"\n");
// fprintf(expectFile, "send \"exit\\n\"\n");
fprintf(expectFile, "interact -o -re $prompt {return}\n");
fclose(expectFile);
sprintf(baseCommand, "expect %s", expectFilename);
} else {
// sprintf(baseCommand, "sbatch %s\n", slurmFilename);
sprintf(baseCommand, "sbatch %s\n", slurmFilename);
}
size = strlen(baseCommand) + 1;
command = chpl_mem_allocMany(size, sizeof(char), CHPL_RT_MD_COMMAND_BUFFER, -1, "");
sprintf(command, "%s", baseCommand);
if (strlen(command)+1 > size) {
chpl_internal_error("buffer overflow");
}
return command;
}
static char* chpl_launch_create_command(int argc, char* argv[],
int32_t numLocales) {
int i;
int size;
char baseCommand[256];
char* command;
FILE* pbsFile, *expectFile;
char* projectString = getenv(launcherAccountEnvvar);
char* basenamePtr = strrchr(argv[0], '/');
pid_t mypid;
if (basenamePtr == NULL) {
basenamePtr = argv[0];
} else {
basenamePtr++;
}
chpl_compute_real_binary_name(argv[0]);
#ifndef DEBUG_LAUNCH
mypid = getpid();
#else
mypid = 0;
#endif
sprintf(sysFilename, "%s%d", baseSysFilename, (int)mypid);
sprintf(expectFilename, "%s%d", baseExpectFilename, (int)mypid);
sprintf(pbsFilename, "%s%d", basePBSFilename, (int)mypid);
pbsFile = fopen(pbsFilename, "w");
fprintf(pbsFile, "#!/bin/sh\n\n");
fprintf(pbsFile, "#PBS -N Chpl-%.10s\n", basenamePtr);
genNumLocalesOptions(pbsFile, determineQsubVersion(), numLocales, getNumCoresPerLocale());
if (projectString && strlen(projectString) > 0)
fprintf(pbsFile, "#PBS -A %s\n", projectString);
fclose(pbsFile);
expectFile = fopen(expectFilename, "w");
if (verbosity < 2) {
fprintf(expectFile, "log_user 0\n");
}
fprintf(expectFile, "set timeout -1\n");
fprintf(expectFile, "set prompt \"(%%|#|\\\\$|>) $\"\n");
fprintf(expectFile, "spawn qsub -z ");
fprintf(expectFile, "-V "); // pass through all environment variables
fprintf(expectFile, "-I %s\n", pbsFilename);
fprintf(expectFile, "expect -re $prompt\n");
fprintf(expectFile, "send \"cd \\$PBS_O_WORKDIR\\n\"\n");
fprintf(expectFile, "expect -re $prompt\n");
fprintf(expectFile, "send \"%s/%s/gasnetrun_ibv -n %d -N %d",
CHPL_THIRD_PARTY, WRAP_TO_STR(LAUNCH_PATH), numLocales, numLocales);
propagate_charset_environment(expectFile);
fprintf(expectFile, " %s ", chpl_get_real_binary_name());
for (i=1; i<argc; i++) {
fprintf(expectFile, " '%s'", argv[i]);
}
fprintf(expectFile, "\\n\"\n");
fprintf(expectFile, "interact -o -re $prompt {return}\n");
fprintf(expectFile, "send_user \"\\n\"\n");
fprintf(expectFile, "send \"exit\\n\"\n");
fclose(expectFile);
sprintf(baseCommand, "expect %s", expectFilename);
size = strlen(baseCommand) + 1;
command = chpl_mem_allocMany(size, sizeof(char), CHPL_RT_MD_COMMAND_BUFFER, -1, 0);
sprintf(command, "%s", baseCommand);
if (strlen(command)+1 > size) {
chpl_internal_error("buffer overflow");
}
return command;
}
static char* chpl_launch_create_command(int argc, char* argv[],
int32_t numLocales) {
int i;
int size;
char baseCommand[2*FILENAME_MAX];
char* command;
FILE* slurmFile, *expectFile;
char* projectString = getenv(launcherAccountEnvvar);
char* constraint = getenv("CHPL_LAUNCHER_CONSTRAINT");
char* outputfn = getenv("CHPL_LAUNCHER_SLURM_OUTPUT_FILENAME");
char* basenamePtr = strrchr(argv[0], '/');
char* nodeAccessEnv = NULL;
pid_t mypid;
if (basenamePtr == NULL) {
basenamePtr = argv[0];
} else {
basenamePtr++;
}
chpl_compute_real_binary_name(argv[0]);
// command line walltime takes precedence over env var
if (!walltime) {
walltime = getenv("CHPL_LAUNCHER_WALLTIME");
}
// command line partition takes precedence over env var
if (!partition) {
partition = getenv("CHPL_LAUNCHER_PARTITION");
}
// command line exclude list takes precedence over env var
if (!exclude) {
exclude = getenv("CHPL_LAUNCHER_EXCLUDE");
}
// request exclusive node access by default, but allow user to override
nodeAccessEnv = getenv("CHPL_LAUNCHER_NODE_ACCESS");
if (nodeAccessEnv == NULL || strcmp(nodeAccessEnv, "exclusive") == 0) {
nodeAccessStr = "exclusive";
} else if (strcmp(nodeAccessEnv, "shared") == 0 ||
strcmp(nodeAccessEnv, "share") == 0 ||
strcmp(nodeAccessEnv, "oversubscribed") == 0 ||
strcmp(nodeAccessEnv, "oversubscribe") == 0) {
nodeAccessStr = "share";
} else if (strcmp(nodeAccessEnv, "unset") == 0) {
nodeAccessStr = NULL;
} else {
chpl_warning("unsupported 'CHPL_LAUNCHER_NODE_ACCESS' option", 0, 0);
nodeAccessStr = "exclusive";
}
if (debug) {
mypid = 0;
} else {
mypid = getpid();
}
sprintf(expectFilename, "%s%d", baseExpectFilename, (int)mypid);
sprintf(slurmFilename, "%s%d", baseSBATCHFilename, (int)mypid);
if (getenv("CHPL_LAUNCHER_USE_SBATCH") != NULL) {
slurmFile = fopen(slurmFilename, "w");
fprintf(slurmFile, "#!/bin/sh\n\n");
fprintf(slurmFile, "#SBATCH -J Chpl-%.10s\n", basenamePtr);
genNumLocalesOptions(slurmFile, determineSlurmVersion(), numLocales, getNumCoresPerLocale());
if (projectString && strlen(projectString) > 0)
fprintf(slurmFile, "#SBATCH -A %s\n", projectString);
if (getenv("CHPL_LAUNCHER_USE_SBATCH") != NULL) {
// fprintf(slurmFile, "#SBATCH -joe\n");
if (outputfn!=NULL)
fprintf(slurmFile, "#SBATCH -o %s\n", outputfn);
else
fprintf(slurmFile, "#SBATCH -o %s.%%j.out\n", argv[0]);
// fprintf(slurmFile, "cd $SBATCH_O_WORKDIR\n");
fprintf(slurmFile, "%s/%s/gasnetrun_ibv -n %d -N %d",
CHPL_THIRD_PARTY, WRAP_TO_STR(LAUNCH_PATH), numLocales, numLocales);
propagate_environment(slurmFile);
fprintf(slurmFile, " %s ", chpl_get_real_binary_name());
for (i=1; i<argc; i++) {
fprintf(slurmFile, " '%s'", argv[i]);
}
fprintf(slurmFile, "\n");
}
fclose(slurmFile);
chmod( slurmFilename, 0755);
}
if (getenv("CHPL_LAUNCHER_USE_SBATCH") == NULL) {
expectFile = fopen(expectFilename, "w");
if (verbosity < 2) {
// fprintf(expectFile, "log_user 0\n");
}
fprintf(expectFile, "set timeout -1\n");
// fprintf(expectFile, "chmod +x %s\n",slurmFilename);
fprintf(expectFile, "set prompt \"(%%|#|\\\\$|>) $\"\n");
// fprintf(expectFile, "spawn sbatch ");
fprintf(expectFile, "spawn -noecho salloc --quiet ");
fprintf(expectFile, "-J %.10s ",basenamePtr); // pass
fprintf(expectFile, "-N %d ",numLocales);
fprintf(expectFile, "--ntasks-per-node=1 ");
if (nodeAccessStr != NULL)
fprintf(expectFile, "--%s ", nodeAccessStr);
if (walltime)
fprintf(expectFile, "--time=%s ",walltime);
if(partition)
fprintf(expectFile, "--partition=%s ",partition);
if(exclude)
fprintf(expectFile, "--exclude=%s ",exclude);
if (constraint) {
fprintf(expectFile, " -C %s", constraint);
}
// fprintf(expectFile, "-I %s ", slurmFilename);
fprintf(expectFile, " %s/%s/gasnetrun_ibv -n %d -N %d",
CHPL_THIRD_PARTY, WRAP_TO_STR(LAUNCH_PATH), numLocales, numLocales);
propagate_environment(expectFile);
fprintf(expectFile, " %s ", chpl_get_real_binary_name());
for (i=1; i<argc; i++) {
fprintf(expectFile, " %s", argv[i]);
}
// fprintf(expectFile, "\\n\"\n");
fprintf(expectFile, "\n\n");
// fprintf(expectFile, "expect -re $prompt\n");
// fprintf(expectFile, "send \"cd \\$SBATCH_O_WORKDIR\\n\"\n");
// fprintf(expectFile, "expect -re $prompt\n");
// fprintf(expectFile, "sleep 10\n");
// fprintf(expectFile, "interact -o -re $prompt {return}\n");
// fprintf(expectFile, "send_user \"\\n\"\n");
// fprintf(expectFile, "send \"exit\\n\"\n");
fprintf(expectFile, "interact -o -re $prompt {return}\n");
fclose(expectFile);
sprintf(baseCommand, "expect %s", expectFilename);
} else {
// sprintf(baseCommand, "sbatch %s\n", slurmFilename);
sprintf(baseCommand, "sbatch %s\n", slurmFilename);
}
size = strlen(baseCommand) + 1;
command = chpl_mem_allocMany(size, sizeof(char), CHPL_RT_MD_COMMAND_BUFFER, -1, 0);
sprintf(command, "%s", baseCommand);
if (strlen(command)+1 > size) {
chpl_internal_error("buffer overflow");
}
return command;
}
static char* chpl_launch_create_command(int argc, char* argv[],
int32_t numLocales) {
int i;
int size;
char baseCommand[256];
char* command;
FILE* llFile;
// char* projectString = getenv(launcherAccountEnvvar);
char* basenamePtr = strrchr(argv[0], '/');
pid_t mypid;
if (!walltime) {
chpl_error("You must specify the wall clock time limit of your job using --walltime\n"
"or CHPL_LAUNCHER_WALLTIME (HH:MM:SS)", 0, NULL);
}
if (basenamePtr == NULL) {
basenamePtr = argv[0];
} else {
basenamePtr++;
}
chpl_compute_real_binary_name(argv[0]);
if (debug) {
mypid = 0;
} else {
mypid = getpid();
}
sprintf(llFilename, "%s%d", baseLLFilename, (int)mypid);
llFile = fopen(llFilename, "w");
fprintf(llFile, "#!/bin/bash\n");
if (queue) {
fprintf(llFile, "# @ class = %s\n", queue);
}
fprintf(llFile, "# @ job_name = %s\n", basenamePtr);
fprintf(llFile, "# @ initialdir = .\n");
fprintf(llFile, "# @ output = %s_%%j.out\n", basenamePtr);
fprintf(llFile, "# @ error = %s_%%j.out\n", basenamePtr);
fprintf(llFile, "# @ total_tasks = %d\n", numLocales);
fprintf(llFile, "# @ cpus_per_task = 4\n");
fprintf(llFile, "# @ tasks_per_node = 1\n");
fprintf(llFile, "# @ wall_clock_limit = %s\n", walltime);
fprintf(llFile, "\n");
#if CHPL_COMM_SUBSTRATE == udp
fprintf(llFile, "MLIST=$(/opt/perf/bin/sl_get_machine_list -j=\\$SLURM_JOB_ID )\n");
fprintf(llFile, "\n");
fprintf(llFile, "export -n SSH_SERVERS\n");
fprintf(llFile, "for i in $MLIST ; do\n");
fprintf(llFile, " export SSH_SERVERS=\"$SSH_SERVERS $i\" ;\n");
fprintf(llFile, "done\n");
if (debug) {
fprintf(llFile, "echo $SSH_SERVERS\n");
}
fprintf(llFile, "\n");
fprintf(llFile, "%samudprun ", WRAP_TO_STR(LAUNCH_PATH));
#elif CHPL_COMM_SUBSTRATE == mpi
fprintf(llFile, "export MPIRUN_CMD='srun --kill-on-bad-exit %%C'\n");
fprintf(llFile, "export MPIRUN_CMD_OK=true\n");
fprintf(llFile, "\n");
fprintf(llFile, "%sgasnetrun_mpi ", WRAP_TO_STR(LAUNCH_PATH));
#elif CHPL_COMM_SUBSTRATE == none
#else
#error "Unknown CHPL_COMM_SUBSTRATE"
#endif
#if CHPL_COMM_SUBSTRATE != none
fprintf(llFile, "-n %d ", numLocales);
#endif
fprintf(llFile, "%s", chpl_get_real_binary_name());
for (i=1; i<argc; i++) {
fprintf(llFile, " '%s'", argv[i]);
}
fprintf(llFile, " || echo -n \"\"\n");
fclose(llFile);
sprintf(baseCommand, "mnsubmit %s", llFilename);
size = strlen(baseCommand) + 1;
command = chpl_mem_allocMany(size, sizeof(char), CHPL_RT_MD_COMMAND_BUFFER, -1, 0);
sprintf(command, "%s", baseCommand);
if (strlen(command)+1 > size) {
chpl_internal_error("buffer overflow");
}
return command;
}
