mpirun.cpp

MPICH是MPI的重要研究,提供了一系列的接口函数,为并行计算的实现提供了编程环境.

    for (int i=n+1; i<=argc; i++)
    {
	argv[i-n] = argv[i];
    }
    argc -= n;
}

static bool isnumber(char *str)
{
    int i, n = strlen(str);
    for (i=0; i<n; i++)
    {
	if (!isdigit(str[i]))
	    return false;
    }
    return true;
}

bool ReadMPDDefault(char *str)
{
    DWORD length = 100;
    char value[100] = "no";

    if (ReadMPDRegistry(str, value, &length))
    {
	if ((stricmp(value, "yes") == 0) ||
	    (stricmp(value, "y") == 0) ||
	    (stricmp(value, "1") == 0))
	    return true;
    }

    return false;
}

bool CreateShmCliqueString(HostNode *pHosts, char *str)
{
    int i, iProc, iterProc, nProc = 0;
    HostNode *n, *iter;
    bool *pDone;
    char *strOrig = str;

    str[0] = '\0';

    if (pHosts == NULL)
    {
	char temp[20];
	nProc = g_nHosts;
	for (i=0; i<nProc; i++)
	{
	    sprintf(temp, "(%d)", i);
	    strcat(str, temp);
	}
	return true;
    }

    n = pHosts;
    while (n)
    {
	nProc += n->nSMPProcs;
	n = n->next;
    }

    if (nProc < 1)
	return false;

    pDone = new bool[nProc];
    for (i=0; i<nProc; i++)
	pDone[i] = false;

    n = pHosts;
    iProc = 0;
    while (n)
    {
	if (!pDone[iProc])
	{
	    pDone[iProc] = true;
	    str += sprintf(str, "(%d", iProc);
	    iProc++;
	    for (i=1; i<n->nSMPProcs; i++)
	    {
		pDone[iProc] = true;
		str += sprintf(str, ",%d", iProc);
		iProc++;
	    }
	    iter = n->next;
	    iterProc = iProc;
	    while (iter)
	    {
		if (stricmp(iter->host, n->host) == 0)
		{
		    pDone[iterProc] = true;
		    str += sprintf(str, ",%d", iterProc);
		    iterProc++;
		    for (i=1; i<iter->nSMPProcs; i++)
		    {
			pDone[iterProc] = true;
			str += sprintf(str, ",%d", iterProc);
			iterProc++;
		    }
		}
		else
		{
		    for (i=0; i<iter->nSMPProcs; i++)
			iterProc++;
		}
		iter = iter->next;
	    }
	    str += sprintf(str, ")");
	}
	else
	{
	    for (i=0; i<n->nSMPProcs; i++)
		iProc++;
	}
	n = n->next;
    }

    return true;
}

void CreateSingleShmCliqueString(int nCliqueCount, int *pMembers, char *pszSingleShmCliqueString)
{
    int i;
    char *str = pszSingleShmCliqueString;

    if (nCliqueCount < 1)
	return;

    str += sprintf(str, "(");
    str += sprintf(str, "%d", pMembers[0]);
    for (i=1; i<nCliqueCount; i++)
    {
	str += sprintf(str, ",%d", pMembers[i]);
    }
    sprintf(str, ")");
    //printf("CreateSingleShmCliqueString produced: %s\n", pszSingleShmCliqueString);
    //exit(0);
}

void SetupTimeouts()
{
    char pszTimeout[20];
    DWORD length;
    char *pszEnvVariable;

    length = 20;
    if (ReadMPDRegistry("timeout", pszTimeout, &length))
    {
	g_nLaunchTimeout = atoi(pszTimeout);
	if (g_nLaunchTimeout < 1)
	    g_nLaunchTimeout = MPIRUN_DEFAULT_TIMEOUT;
    }
    length = 20;
    if (ReadMPDRegistry("short_timeout", pszTimeout, &length))
    {
	g_nMPIRUN_SHORT_TIMEOUT = atoi(pszTimeout);
	if (g_nMPIRUN_SHORT_TIMEOUT < 1)
	    g_nMPIRUN_SHORT_TIMEOUT = MPIRUN_SHORT_TIMEOUT;
    }
    length = 20;
    if (ReadMPDRegistry("startup_timeout", pszTimeout, &length))
    {
	g_nMPIRUN_CREATE_PROCESS_TIMEOUT = atoi(pszTimeout);
	if (g_nMPIRUN_CREATE_PROCESS_TIMEOUT < 1)
	    g_nMPIRUN_CREATE_PROCESS_TIMEOUT = MPIRUN_CREATE_PROCESS_TIMEOUT;
    }
    pszEnvVariable = getenv("MPIRUN_SHORT_TIMEOUT");
    if (pszEnvVariable)
    {
	g_nMPIRUN_SHORT_TIMEOUT = atoi(pszEnvVariable);
	if (g_nMPIRUN_SHORT_TIMEOUT < 1)
	    g_nMPIRUN_SHORT_TIMEOUT = MPIRUN_SHORT_TIMEOUT;
    }
    pszEnvVariable = getenv("MPIRUN_STARTUP_TIMEOUT");
    if (pszEnvVariable)
    {
	g_nMPIRUN_CREATE_PROCESS_TIMEOUT = atoi(pszEnvVariable);
	if (g_nMPIRUN_CREATE_PROCESS_TIMEOUT < 1)
	    g_nMPIRUN_CREATE_PROCESS_TIMEOUT = MPIRUN_CREATE_PROCESS_TIMEOUT;
    }
}

// Function name	: main
// Description	    : 
// Return type		: void 
// Argument         : int argc
// Argument         : char *argv[]
int main(int argc, char *argv[])
{
    int i;
    int iproc = 0;
    char pszJobID[100];
    char pszEnv[MAX_CMD_LENGTH] = "";
    HANDLE *pThread = NULL;
    int nShmLow, nShmHigh;
    DWORD dwThreadID;
    bool bLogon = false;
    char pBuffer[MAX_CMD_LENGTH];
    char phrase[MPD_PASSPHRASE_MAX_LENGTH + 1];// = MPD_DEFAULT_PASSPHRASE;
    bool bLogonDots = true;
    HANDLE hStdout;
    char cMapDrive, pszMapShare[MAX_PATH];
    int nArgsToStrip;
    bool bRunLocal;
    char pszMachineFileName[MAX_PATH] = "";
    bool bUseMachineFile;
    bool bDoSMP;
    bool bPhraseNeeded;
    DWORD dwType;
    bool bUsePwdFile = false;
    char pszPwdFileName[MAX_PATH];
    bool bUseDebugFlag = false;
    DWORD length;
    WSADATA wsaData;
    int err;
    bool bNoDriveMapping = false;
    bool bCredentialsPrompt = true;
    bool bUsePriorities = false;
    int nPriorityClass = 1;
    int nPriority = 3;
    char pszShmCliqueString[MAX_CMD_LENGTH];
    char pszSingleShmCliqueString[MAX_CMD_LENGTH];
    int nCliqueCount, *pMembers;

    if (argc < 2)
    {
	PrintOptions();
	return 0;
    }

    SetConsoleCtrlHandler(CtrlHandlerRoutine, TRUE);

    if ((err = WSAStartup( MAKEWORD( 2, 0 ), &wsaData )) != 0)
    {
	printf("Winsock2 dll not initialized, error %d", err);
	switch (err)
	{
	case WSASYSNOTREADY:
	    printf("Indicates that the underlying network subsystem is not ready for network communication.\n");
	    break;
	case WSAVERNOTSUPPORTED:
	    printf("The version of Windows Sockets support requested is not provided by this particular Windows Sockets implementation.\n");
	    break;
	case WSAEINPROGRESS:
	    printf("A blocking Windows Sockets 1.1 operation is in progress.\n");
	    break;
	case WSAEPROCLIM:
	    printf("Limit on the number of tasks supported by the Windows Sockets implementation has been reached.\n");
	    break;
	case WSAEFAULT:
	    printf("The lpWSAData is not a valid pointer.\n");
	    break;
	default:
	    Translate_Error(err, pBuffer);
	    printf("%s\n", pBuffer);
	    break;
	}
	return 0;
    }

    // Set defaults
    g_bDoMultiColorOutput = !ReadMPDDefault("nocolor");
    bRunLocal = false;
    g_bNoMPI = false;
    bLogon = false;
    bLogonDots = !ReadMPDDefault("nodots");
    GetCurrentDirectory(MAX_PATH, g_pszDir);
    bUseMachineFile = false;
    bDoSMP = true;
    phrase[0] = '\0';
    bPhraseNeeded = true;
    g_nHosts = 0;
    g_pHosts = NULL;
    bNoDriveMapping = ReadMPDDefault("nomapping");
    g_bOutputExitCodes = ReadMPDDefault("exitcodes");
    if (ReadMPDDefault("nopopup_debug"))
    {
	SetErrorMode(SEM_FAILCRITICALERRORS | SEM_NOGPFAULTERRORBOX);
    }
    if (ReadMPDDefault("usejobhost"))
    {
	length = MAX_HOST_LENGTH;
	if (ReadMPDRegistry("jobhost", g_pszJobHost, &length))
	{
	    g_bUseJobHost = true;
	    length = 100;
	    if (ReadMPDRegistry("jobhostpwd", g_pszJobHostMPDPwd, &length))
	    {
		g_bUseJobMPDPwd = true;
	    }
	}
    }
    bUseDebugFlag = ReadMPDDefault("dbg");
    SetupTimeouts();


    // Parse mpirun options
    while (argv[1] && (argv[1][0] == '-' || argv[1][0] == '/'))
    {
	nArgsToStrip = 1;
	if (stricmp(&argv[1][1], "np") == 0)
	{
	    if (argc < 3)
	    {
		printf("Error: no number specified after -np option.\n");
		return 0;
	    }
	    g_nHosts = atoi(argv[2]);
	    if (g_nHosts < 1)
	    {
		printf("Error: must specify a number greater than 0 after the -np option\n");
		return 0;
	    }
	    nArgsToStrip = 2;
	}
	else if (stricmp(&argv[1][1], "localonly") == 0)
	{
	    bRunLocal = true;
	    if (argc > 2)
	    {
		if (isnumber(argv[2]))
		{
		    g_nHosts = atoi(argv[2]);
		    if (g_nHosts < 1)
		    {
			printf("Error: If you specify a number after -localonly option,\n        it must be greater than 0.\n");
			return 0;
		    }
		    nArgsToStrip = 2;
		}
	    }
	}
	else if (stricmp(&argv[1][1], "machinefile") == 0)
	{
	    if (argc < 3)
	    {
		printf("Error: no filename specified after -machinefile option.\n");
		return 0;
	    }
	    strcpy(pszMachineFileName, argv[2]);
	    bUseMachineFile = true;
	    nArgsToStrip = 2;
	}
	else if (stricmp(&argv[1][1], "map") == 0)
	{
	    if (argc < 3)
	    {
		printf("Error: no drive specified after -map option.\n");
		return 0;
	    }
	    if ((strlen(argv[2]) > 2) && argv[2][1] == ':')
	    {
		MapDriveNode *pNode = new MapDriveNode;
		pNode->cDrive = argv[2][0];
		strcpy(pNode->pszShare, &argv[2][2]);
		pNode->pNext = g_pDriveMapList;
		g_pDriveMapList = pNode;
	    }
	    nArgsToStrip = 2;
	}
	else if (stricmp(&argv[1][1], "dir") == 0)
	{
	    if (argc < 3)
	    {
		printf("Error: no directory after -dir option\n");
		return 0;
	    }
	    strcpy(g_pszDir, argv[2]);
	    nArgsToStrip = 2;
	}
	else if (stricmp(&argv[1][1], "env") == 0)
	{
	    if (argc < 3)
	    {
		printf("Error: no environment variables after -env option\n");
		return 0;
	    }
	    strncpy(g_pszEnv, argv[2], MAX_CMD_LENGTH);
	    g_pszEnv[MAX_CMD_LENGTH-1] = '\0';
	    if (strlen(argv[2]) >= MAX_CMD_LENGTH)
	    {
		printf("Warning: environment variables truncated.\n");
	    }
	    nArgsToStrip = 2;
	}
	else if (stricmp(&argv[1][1], "logon") == 0)
	{
	    bLogon = true;
	}
	else if (stricmp(&argv[1][1], "noprompt") == 0)
	{
	    bCredentialsPrompt = false;
	}
	else if (stricmp(&argv[1][1], "dbg") == 0)
	{
	    bUseDebugFlag = true;
	}
	else if (stricmp(&argv[1][1], "pwdfile") == 0)
	{
	    bUsePwdFile = true;
	    if (argc < 3)
	    {
		printf("Error: no filename specified after -pwdfile option\n");
		return 0;
	    }
	    strncpy(pszPwdFileName, argv[2], MAX_PATH);
	    pszPwdFileName[MAX_PATH-1] = '\0';
	    nArgsToStrip = 2;
	}
	else if (stricmp(&argv[1][1], "mpduser") == 0)
	{
	    g_bUseMPDUser = true;
	}
	else if (stricmp(&argv[1][1], "hosts") == 0)
	{
	    if (g_nHosts != 0)
	    {
		printf("Error: only one option is allowed to determine the number of processes.\n");
		printf("       -hosts cannot be used with -np or -localonly\n");
		return 0;
	    }
	    if (argc > 2)
	    {
		if (isnumber(argv[2]))
		{
		    g_nHosts = atoi(argv[2]);
		    if (g_nHosts < 1)
		    {
			printf("Error: You must specify a number greater than 0 after -hosts.\n");
			return 0;
		    }
		    nArgsToStrip = 2 + g_nHosts;
		    int index = 3;
		    for (i=0; i<g_nHosts; i++)
		    {
			if (index >= argc)
			{
			    printf("Error: missing host name after -hosts option.\n");
			    return 0;
			}
			HostNode *pNode = new HostNode;
			pNode->next = NULL;
			pNode->nSMPProcs = 1;
			pNode->exe[0] = '\0';
			strcpy(pNode->host, argv[index]);
			index++;
			if (argc > index)
			{
			    if (isnumber(argv[index]))
			    {
				pNode->nSMPProcs = atoi(argv[index]);
				index++;
				nArgsToStrip++;
			    }
			}
			if (g_pHosts == NULL)
			{
			    g_pHosts = pNode;
			}
			else
			{
			    HostNode *pIter = g_pHosts;
			    while (pIter->next)
				pIter = pIter->next;
			    pIter->next = pNode;
			}
		    }
		}
		else
		{
		    printf("Error: You must specify the number of hosts after the -hosts option.\n");
		    return 0;
		}
	    }
	    else
	    {
		printf("Error: not enough arguments.\n");
		return 0;
	    }
	}
	else if (stricmp(&argv[1][1], "tcp") == 0)
	{
	    bDoSMP = false;
	}
	else if (stricmp(&argv[1][1], "getphrase") == 0)
	{
	    GetMPDPassPhrase(phrase);
	    bPhraseNeeded = false;
	}
	else if (stricmp(&argv[1][1], "nocolor") == 0)
	{
	    g_bDoMultiColorOutput = false;
	}
	else if (stricmp(&argv[1][1], "nompi") == 0)
	{
	    g_bNoMPI = true;
	}
	else if (stricmp(&argv[1][1], "nodots") == 0)
	{
	    bLogonDots = false;
	}
	else if (stricmp(&argv[1][1], "nomapping") == 0)
	{
	    bNoDriveMapping = true;
	}
	else if (stricmp(&argv[1][1], "nopopup_debug") == 0)
	{
	    SetErrorMode(SEM_FAILCRITICALERRORS | SEM_NOGPFAULTERRORBOX);
	}
	else if (stricmp(&argv[1][1], "help") == 0 || argv[1][1] == '?')
	{
	    PrintOptions();
	    return 0;
	}
	else if (stricmp(&argv[1][1], "help2") == 0)
	{
	    PrintExtraOptions();
	    return 0;
	}
	else if (stricmp(&argv[1][1], "jobhost") == 0)
	{
	    g_bUseJobHost = true;
	    if (argc < 3)
	    {
		printf("Error: no host name specified after -jobhost option\n");
		return 0;
	    }
	    strncpy(g_pszJobHost, argv[2], MAX_HOST_LENGTH);
	    g_pszJobHost[MAX_HOST_LENGTH-1] = '\0';
	    nArgsToStrip = 2;
	}
	else if (stricmp(&argv[1][1], "jobhostmpdpwd") == 0)
	{
	    g_bUseJobMPDPwd = true;
	    if (argc < 3)
	    {
		printf("Error: no passphrase specified after -jobhostmpdpwd option\n");
		return 0;
	    }
	    strncpy(g_pszJobHostMPDPwd, argv[2], 100);
	    g_pszJobHostMPDPwd[99] = '\0';
	    nArgsToStrip = 2;
	}
	else if (stricmp(&argv[1][1], "exitcodes") == 0)
	{