Ejemplo n.º 1
0
int test_perfmonstart_noadd()
{
    CpuInfo_t cpuinfo;
    int cpu = 0;
    topology_init();
    cpuinfo = get_cpuInfo();
    if (cpuinfo->isIntel == 0)
    {
        topology_finalize();
        return 1;
    }
    int ret = perfmon_init(1, &cpu);
    if (ret != 0)
        goto fail;
    ret = perfmon_startCounters();
    if (ret == 0)
        goto fail;
    perfmon_finalize();
    topology_finalize();
    return 1;
fail:
    perfmon_finalize();
    topology_finalize();
    return 0;
}
Ejemplo n.º 2
0
int test_perfmonstart()
{
    CpuInfo_t cpuinfo;
    int group1, group2;
    int cpu = 0;
    topology_init();
    cpuinfo = get_cpuInfo();
    if (cpuinfo->isIntel == 0)
    {
        topology_finalize();
        return 1;
    }
    int ret = perfmon_init(1, &cpu);
    if (ret != 0)
        goto fail;
    ret = perfmon_addEventSet(eventset_ok);
    if (ret != 0)
        goto fail;
    group1 = ret;
    ret = perfmon_setupCounters(group1);
    if (ret != 0)
        goto fail;
    if (perfmon_getIdOfActiveGroup() != group1)
        goto fail;
    ret = perfmon_startCounters();
    if (ret != 0)
        goto fail;
    perfmon_finalize();
    topology_finalize();
    return 1;
fail:
    perfmon_finalize();
    topology_finalize();
    return 0;
}
Ejemplo n.º 3
0
void
multiplex_start()
{
    struct itimerval val;
    struct sigaction sa;

//    multiplex_useMarker = useMarker;

    val.it_interval.tv_sec = 0;
    val.it_interval.tv_usec = 500;
    val.it_value.tv_sec = 0; 
    val.it_value.tv_usec = 100;

    sa.sa_handler = multiplex_printCounters;
    sigemptyset(&sa.sa_mask);
    sa.sa_flags = SA_RESTART;
    if (sigaction(SIGALRM, &sa, NULL) == -1)
    {
        /* Handle error */;
        perror("sigaction");
    }

    perfmon_startCounters();
    setitimer(ITIMER_REAL, &val,0);
    timer_startCycles(&timeData);
}
Ejemplo n.º 4
0
int main(int argc, char* argv[])
{
    int i;
    int* cpus;
    int gid;
    double result = 0.0;

    // Load the topology module and print some values.
    topology_init();
    // CpuInfo_t contains global information like name, CPU family, ...
    CpuInfo_t info = get_cpuInfo();
    // CpuTopology_t contains information about the topology of the CPUs.
    CpuTopology_t topo = get_cpuTopology();
    printf("Likwid example on a %s with %d CPUs\n", info->name, topo->numHWThreads);

    cpus = malloc(topo->numHWThreads * sizeof(int));
    if (!cpus)
        return 1;

    for (i=0;i<topo->numHWThreads;i++)
    {
        cpus[i] = topo->threadPool[i].apicId;
    }

    // Must be called before perfmon_init() but only if you want to use another
    // access mode as the pre-configured one. For direct access (0) you have to
    // be root.
    //accessClient_setaccessmode(0);

    // Initialize the perfmon module.
    perfmon_init(topo->numHWThreads, cpus);

    // Add eventset string to the perfmon module.
    gid = perfmon_addEventSet(EVENTSET);

    // Setup the eventset identified by group ID (gid).
    perfmon_setupCounters(gid);
    // Start all counters in the previously set up event set.
    perfmon_startCounters();
    // Perform something
    sleep(2);
    // Stop all counters in the previously started event set.
    perfmon_stopCounters();


    // Print the result of every thread/CPU.
    for (i = 0;i < topo->numHWThreads; i++)
    {
        result = perfmon_getResult(gid, 0, i);
        printf("Measurement result for event set %s at CPU %d: %f\n", EVENTSET, cpus[i], result);
    }

    // Uninitialize the perfmon module.
    perfmon_finalize();
    // Uninitialize the topology module.
    topology_finalize();
    return 0;
}
Ejemplo n.º 5
0
int test_perfmonstart_noinit()
{
    int ret = perfmon_startCounters();
    if (ret == 0)
        goto fail;
    return 1;
fail:
    return 0;
}
Ejemplo n.º 6
0
static int lua_likwid_startCounters(lua_State* L)
{
    int ret;
    if (perfmon_isInitialized == 0)
    {
        return 0;
    }
    ret = perfmon_startCounters();
    lua_pushnumber(L,ret);
    return 1;
}
Ejemplo n.º 7
0
void
daemon_interrupt(int sig)
{
    if (daemon_run)
    {
        perfmon_stopCounters();
        daemon_run = 0;
        printf("DAEMON:  STOP on %d\n",sig);
    }
    else
    {
        perfmon_setupEventSet(eventString, NULL);
        perfmon_startCounters();
        daemon_run = 1;
        printf("DAEMON:  START\n");
    }
}
Ejemplo n.º 8
0
void
daemon_start(struct timespec interval)
{
    daemon_run = 1;
    perfmon_startCounters();
    timer_start(&timeData);

    while (1)
    {
        if (daemon_run)
        {
            timer_stop(&timeData);
            perfmon_readCounters();
            perfmon_logCounterResults( timer_print(&timeData) );
            timer_start(&timeData);
        }
        nanosleep( &interval, NULL);
    }
}
Ejemplo n.º 9
0
int test_perfmonresult()
{
    CpuInfo_t cpuinfo;
    int cpu = 0;
    int group;
    topology_init();
    cpuinfo = get_cpuInfo();
    if (cpuinfo->isIntel == 0)
    {
        topology_finalize();
        return 1;
    }
    int ret = perfmon_init(1, &cpu);
    if (ret != 0)
        goto fail;
    ret = perfmon_addEventSet(eventset_ok);
    if (ret != 0)
        goto fail;
    group = ret;
    ret = perfmon_setupCounters(group);
    if (ret != 0)
        goto fail;

    ret = perfmon_startCounters();
    if (ret != 0)
        goto fail;
    sleep(1);
    ret = perfmon_stopCounters();
    if (ret != 0)
        goto fail;
    if ((perfmon_getResult(group,0,0) == 0)||(perfmon_getResult(group,1,0) == 0))
        goto fail;
    if (perfmon_getTimeOfGroup(group) == 0)
        goto fail;
    perfmon_finalize();
    topology_finalize();
    return 1;
fail:
    perfmon_finalize();
    topology_finalize();
    return 0;
}
Ejemplo n.º 10
0
int test_perfmonresult_nostop()
{
    CpuInfo_t cpuinfo;
    int cpu = 0;
    int group;
    topology_init();
    cpuinfo = get_cpuInfo();
    if (cpuinfo->isIntel == 0)
    {
        topology_finalize();
        return 1;
    }
    int ret = perfmon_init(1, &cpu);
    if (ret != 0)
        goto fail;
    ret = perfmon_addEventSet(eventset_ok);
    if (ret != 0)
        goto fail;
    group = ret;
    ret = perfmon_setupCounters(group);
    if (ret != 0)
        goto fail;
    ret = perfmon_startCounters();
    if (ret != 0)
        goto fail;
    double result = perfmon_getResult(group,0,0);
    if (result != 0)
        goto fail;
    perfmon_finalize();
    topology_finalize();
    return 1;
fail:
    perfmon_finalize();
    topology_finalize();
    return 0;
}
Ejemplo n.º 11
0
void
likwid_markerInit(void)
{
    int i;
    int verbosity;
    int setinit = 0;
    bstring bThreadStr;
    bstring bEventStr;
    struct bstrList* threadTokens;
    struct bstrList* eventStrings;
    char* modeStr = getenv("LIKWID_MODE");
    char* eventStr = getenv("LIKWID_EVENTS");
    char* cThreadStr = getenv("LIKWID_THREADS");
    char* filepath = getenv("LIKWID_FILEPATH");
    char* perfpid = getenv("LIKWID_PERF_EXECPID");
    char execpid[20];
    /* Dirty hack to avoid nonnull warnings */
    int (*ownatoi)(const char*);
    ownatoi = &atoi;

    if ((modeStr != NULL) && (filepath != NULL) && (eventStr != NULL) && (cThreadStr != NULL) && likwid_init == 0)
    {
        setinit = 1;
    }
    else if (likwid_init == 0)
    {
        fprintf(stderr, "Running without Marker API. Activate Marker API with -m on commandline.\n");
        return;
    }
    else
    {
        return;
    }

    if (!lock_check())
    {
        fprintf(stderr,"Access to performance counters is locked.\n");
        exit(EXIT_FAILURE);
    }

    topology_init();
    numa_init();
    affinity_init();
    hashTable_init();

//#ifndef LIKWID_USE_PERFEVENT
    HPMmode(atoi(modeStr));
//#endif
    if (getenv("LIKWID_DEBUG") != NULL)
    {
        perfmon_verbosity = atoi(getenv("LIKWID_DEBUG"));
        verbosity = perfmon_verbosity;
    }

    bThreadStr = bfromcstr(cThreadStr);
    threadTokens = bsplit(bThreadStr,',');
    num_cpus = threadTokens->qty;
    for (i=0; i<num_cpus; i++)
    {
        threads2Cpu[i] = ownatoi(bdata(threadTokens->entry[i]));
    }
    bdestroy(bThreadStr);
    bstrListDestroy(threadTokens);

    if (getenv("LIKWID_PIN") != NULL)
    {
        likwid_pinThread(threads2Cpu[0]);
        if (getenv("OMP_NUM_THREADS") != NULL)
        {
            if (ownatoi(getenv("OMP_NUM_THREADS")) > num_cpus)
            {
                use_locks = 1;
            }
        }
        if (getenv("CILK_NWORKERS") != NULL)
        {
            if (ownatoi(getenv("CILK_NWORKERS")) > num_cpus)
            {
                use_locks = 1;
            }
        }
    }
#ifdef LIKWID_USE_PERFEVENT
    if (perfpid != NULL)
    {
        snprintf(execpid, 19, "%d", getpid());
        setenv("LIKWID_PERF_PID", execpid, 1);
        char* perfflags = getenv("LIKWID_PERF_FLAGS");
        if (perfflags)
        {
            setenv("LIKWID_PERF_FLAGS", getenv("LIKWID_PERF_FLAGS"), 1);
        }
    }
#endif

    i = perfmon_init(num_cpus, threads2Cpu);
    if (i<0)
    {
        //fprintf(stderr,"Failed to initialize LIKWID perfmon library.\n");
        return;
    }

    bEventStr = bfromcstr(eventStr);
    eventStrings = bsplit(bEventStr,'|');
    numberOfGroups = eventStrings->qty;
    groups = malloc(numberOfGroups * sizeof(int));
    if (!groups)
    {
        fprintf(stderr,"Cannot allocate space for group handling.\n");
        bstrListDestroy(eventStrings);
        exit(EXIT_FAILURE);
    }
    for (i=0; i<eventStrings->qty; i++)
    {
        groups[i] = perfmon_addEventSet(bdata(eventStrings->entry[i]));
    }
    bstrListDestroy(eventStrings);
    bdestroy(bEventStr);

    for (i=0; i<num_cpus; i++)
    {
        hashTable_initThread(threads2Cpu[i]);
        for(int j=0; j<groupSet->groups[groups[0]].numberOfEvents;j++)
        {
            groupSet->groups[groups[0]].events[j].threadCounter[i].init = TRUE;
            groupSet->groups[groups[0]].state = STATE_START;
        }
    }
    if (setinit)
    {
        likwid_init = 1;
    }
    threads2Pthread[registered_cpus] = pthread_self();
    registered_cpus++;

    groupSet->activeGroup = 0;

    perfmon_setupCounters(groupSet->activeGroup);
    perfmon_startCounters();
}
Ejemplo n.º 12
0
int main(int argn, char** argc)
{
    int err, i ,j;
    int numCPUs = 0;
    int gid;
    DATATYPE *a,*b,*c,*d;
    TimeData timer;
    double triad_time, copy_time, scale_time, stream_time;
    char estr[1024];
    double result, scalar = 3.0;
    char* ptr;

    if (argn != 3)
    {
        printf("Usage: %s <cpustr> <events>\n", argc[0]);
        return 1;
    }

    strcpy(estr, argc[2]);

    allocate_vector(&a, SIZE);
    allocate_vector(&b, SIZE);
    allocate_vector(&c, SIZE);
    allocate_vector(&d, SIZE);

    err = topology_init();
    if (err < 0)
    {
        printf("Failed to initialize LIKWID's topology module\n");
        return 1;
    }
    CpuTopology_t topo = get_cpuTopology();
    affinity_init();
    int* cpus = (int*)malloc(topo->numHWThreads * sizeof(int));
    if (!cpus)
        return 1;
    numCPUs = cpustr_to_cpulist(argc[1], cpus, topo->numHWThreads);
    omp_set_num_threads(numCPUs);
    err = perfmon_init(numCPUs, cpus);
    if (err < 0)
    {
        printf("Failed to initialize LIKWID's performance monitoring module\n");
        affinity_finalize();
        topology_finalize();
        return 1;
    }
    gid = perfmon_addEventSet(estr);
    if (gid < 0)
    {
        printf("Failed to add event string %s to LIKWID's performance monitoring module\n", estr);
        perfmon_finalize();
        affinity_finalize();
        topology_finalize();
        return 1;
    }

    err = perfmon_setupCounters(gid);
    if (err < 0)
    {
        printf("Failed to setup group %d in LIKWID's performance monitoring module\n", gid);
        perfmon_finalize();
        affinity_finalize();
        topology_finalize();
        return 1;
    }

#ifdef _OPENMP
    printf(HLINE);
#pragma omp parallel
    {
#pragma omp master
    {
        printf ("Number of Threads requested = %i\n",omp_get_num_threads());
    }
    likwid_pinThread(cpus[omp_get_thread_num()]);
    printf ("Thread %d running on processor %d ....\n",omp_get_thread_num(),sched_getcpu());
    }
#endif

#pragma omp parallel for
    for (int j=0; j<SIZE; j++) {
        a[j] = 1.0;
        b[j] = 2.0;
        c[j] = 0.0;
        d[j] = 1.0;
    }

    err = perfmon_startCounters();
    if (err < 0)
    {
        printf("Failed to start counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }
    time_start(&timer);
#pragma omp parallel
    {
        for (int k=0; k<ITER; k++)
        {
            LIKWID_MARKER_START("copy");
#pragma omp for
            for (int j=0; j<SIZE; j++)
            {
                c[j] = a[j];
            }
            LIKWID_MARKER_STOP("copy");
        }
    }
    time_stop(&timer);
    err = perfmon_stopCounters();
    copy_time = time_print(&timer)/(double)ITER;
    if (err < 0)
    {
        printf("Failed to stop counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }

    printf("Processed %.1f Mbyte at copy benchmark in %.4f seconds: %.2f MByte/s\n",
                        1E-6*(2*SIZE*sizeof(DATATYPE)),
                        copy_time,
                        1E-6*((2*SIZE*sizeof(DATATYPE))/copy_time));

    ptr = strtok(estr,",");
    j = 0;
    while (ptr != NULL)
    {
        for (i = 0;i < numCPUs; i++)
        {
            result = perfmon_getResult(gid, j, cpus[i]);
            printf("Measurement result for event set %s at CPU %d: %f\n", ptr, cpus[i], result);
        }
        ptr = strtok(NULL,",");
        j++;
    }
    strcpy(estr, argc[2]);
    perfmon_setupCounters(gid);

    err = perfmon_startCounters();
    if (err < 0)
    {
        printf("Failed to start counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }
    time_start(&timer);
#pragma omp parallel
    {
        for (int k=0; k<ITER; k++)
        {
            LIKWID_MARKER_START("scale");
#pragma omp for
            for (int j=0; j<SIZE; j++)
            {
                b[j] = scalar*c[j];
            }
            LIKWID_MARKER_STOP("scale");
        }
    }
    time_stop(&timer);
    err = perfmon_stopCounters();
    scale_time = time_print(&timer)/(double)ITER;
    if (err < 0)
    {
        printf("Failed to stop counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }

    printf("Processed %.1f Mbyte at scale benchmark in %.4f seconds: %.2f MByte/s\n",
                        1E-6*(2*SIZE*sizeof(DATATYPE)),
                        copy_time,
                        1E-6*((2*SIZE*sizeof(DATATYPE))/copy_time));

    ptr = strtok(estr,",");
    j = 0;
    while (ptr != NULL)
    {
        for (i = 0;i < numCPUs; i++)
        {
            result = perfmon_getResult(gid, j, cpus[i]);
            printf("Measurement result for event set %s at CPU %d: %f\n", ptr, cpus[i], result);
        }
        ptr = strtok(NULL,",");
        j++;
    }
    strcpy(estr, argc[2]);
    perfmon_setupCounters(gid);
    err = perfmon_startCounters();
    if (err < 0)
    {
        printf("Failed to start counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }
    time_start(&timer);
#pragma omp parallel
    {
        for (int k=0; k<ITER; k++)
        {
            LIKWID_MARKER_START("stream");
#pragma omp for
            for (int j=0; j<SIZE; j++)
            {
                c[j] = a[j] + b[j];
            }
            LIKWID_MARKER_STOP("stream");
        }
    }
    time_stop(&timer);
    err = perfmon_stopCounters();
    stream_time = time_print(&timer)/(double)ITER;
    if (err < 0)
    {
        printf("Failed to stop counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }

    printf("Processed %.1f Mbyte at stream benchmark in %.4f seconds: %.2f MByte/s\n",
                        1E-6*(2*SIZE*sizeof(DATATYPE)),
                        copy_time,
                        1E-6*((2*SIZE*sizeof(DATATYPE))/copy_time));

    ptr = strtok(estr,",");
    j = 0;
    while (ptr != NULL)
    {
        for (i = 0;i < numCPUs; i++)
        {
            result = perfmon_getResult(gid, j, cpus[i]);
            printf("Measurement result for event set %s at CPU %d: %f\n", ptr, cpus[i], result);
        }
        ptr = strtok(NULL,",");
        j++;
    }
    strcpy(estr, argc[2]);
    perfmon_setupCounters(gid);
    err = perfmon_startCounters();
    if (err < 0)
    {
        printf("Failed to start counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }
    time_start(&timer);
#pragma omp parallel
    {
        for (int k=0; k<ITER; k++)
        {

            LIKWID_MARKER_START("triad");
#pragma omp for
            for (int j=0; j<SIZE; j++)
            {
                a[j] = b[j] +  c[j] * scalar;
            }
            LIKWID_MARKER_STOP("triad");
        }
    }
    time_stop(&timer);
    err = perfmon_stopCounters();
    triad_time = time_print(&timer)/(double)ITER;
    if (err < 0)
    {
        printf("Failed to stop counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }



    printf("Processed %.1f Mbyte at triad benchmark in %.4f seconds: %.2f MByte/s\n",
                        1E-6*(4*SIZE*sizeof(DATATYPE)),
                        triad_time,
                        1E-6*((4*SIZE*sizeof(DATATYPE))/triad_time));
    ptr = strtok(estr,",");
    j = 0;
    while (ptr != NULL)
    {
        for (i = 0;i < numCPUs; i++)
        {
            result = perfmon_getResult(gid, j, cpus[i]);
            printf("Measurement result for event set %s at CPU %d: %f\n", ptr, cpus[i], result);
        }
        ptr = strtok(NULL,",");
        j++;
    }

    perfmon_finalize();
    affinity_finalize();
    topology_finalize();
    return 0;
}
Ejemplo n.º 13
0
int main (int argc, char** argv)
{
    int socket_fd = -1;
    int optInfo = 0;
    int optClock = 0;
    int optStethoscope = 0;
    int optSockets = 0;
    double runtime;
    int hasDRAM = 0;
    int c;
    bstring argString;
    bstring eventString = bfromcstr("CLOCK");
    int numSockets=1;
    int numThreads=0;
    int threadsSockets[MAX_NUM_NODES*2];
    int threads[MAX_NUM_THREADS];

    threadsSockets[0] = 0;
    
    if (argc == 1)
    {
    	HELP_MSG;
    	exit (EXIT_SUCCESS);
    }

    while ((c = getopt (argc, argv, "+c:hiM:ps:v")) != -1)
    {
        switch (c)
        {
            case 'c':
                CHECK_OPTION_STRING;
                numSockets = bstr_to_cpuset_physical((uint32_t*) threadsSockets, argString);
                bdestroy(argString);
                optSockets = 1;
                break;

            case 'h':
                HELP_MSG;
                exit (EXIT_SUCCESS);
            case 'i':
                optInfo = 1;
                break;
            case 'M':  /* Set MSR Access mode */
                CHECK_OPTION_STRING;
                accessClient_setaccessmode(str2int((char*) argString->data));
                bdestroy(argString);
                break;
            case 'p':
                optClock = 1;
                break;
            case 's':
                CHECK_OPTION_STRING;
                optStethoscope = str2int((char*) argString->data);
                bdestroy(argString);
                break;
            case 'v':
                VERSION_MSG;
                exit (EXIT_SUCCESS);
            case '?':
            	if (optopt == 's' || optopt == 'M' || optopt == 'c')
            	{
            		HELP_MSG;
            	}
                else if (isprint (optopt))
                {
                    fprintf (stderr, "Unknown option `-%c'.\n", optopt);
                }
                else
                {
                    fprintf (stderr,
                            "Unknown option character `\\x%x'.\n",
                            optopt);
                }
                exit( EXIT_FAILURE);
            default:
                HELP_MSG;
                exit (EXIT_SUCCESS);
        }
    }

    if (!lock_check())
    {
        fprintf(stderr,"Access to performance counters is locked.\n");
        exit(EXIT_FAILURE);
    }
    
    if (optClock && optind == argc)
    {
    	fprintf(stderr,"Commandline option -p requires an executable.\n");
    	exit(EXIT_FAILURE);
    }
    if (optSockets && !optStethoscope && optind == argc)
    {
    	fprintf(stderr,"Commandline option -c requires an executable if not used in combination with -s.\n");
    	exit(EXIT_FAILURE);
    }

    if (cpuid_init() == EXIT_FAILURE)
    {
        fprintf(stderr, "CPU not supported\n");
        exit(EXIT_FAILURE);
    }
    
    if (numSockets > cpuid_topology.numSockets)
    {
    	fprintf(stderr, "System has only %d sockets but %d are given on commandline\n",
    			cpuid_topology.numSockets, numSockets);
    	exit(EXIT_FAILURE);
    }

    numa_init(); /* consider NUMA node as power unit for the moment */
    accessClient_init(&socket_fd);
    msr_init(socket_fd);
    timer_init();

    /* check for supported processors */
    if ((cpuid_info.model == SANDYBRIDGE_EP) ||
            (cpuid_info.model == SANDYBRIDGE) ||
            (cpuid_info.model == IVYBRIDGE) ||
            (cpuid_info.model == IVYBRIDGE_EP) ||
            (cpuid_info.model == HASWELL) ||
            (cpuid_info.model == NEHALEM_BLOOMFIELD) ||
            (cpuid_info.model == NEHALEM_LYNNFIELD) ||
            (cpuid_info.model == NEHALEM_WESTMERE))
    {
        power_init(numa_info.nodes[0].processors[0]);
    }
    else
    {
        fprintf (stderr, "Query Turbo Mode only supported on Intel Nehalem/Westmere/SandyBridge/IvyBridge/Haswell processors!\n");
        exit(EXIT_FAILURE);
    }

    double clock = (double) timer_getCpuClock();

    printf(HLINE);
    printf("CPU name:\t%s \n",cpuid_info.name);
    printf("CPU clock:\t%3.2f GHz \n",  (float) clock * 1.E-09);
    printf(HLINE);

    if (optInfo)
    {
        if (power_info.turbo.numSteps != 0)
        {
            printf("Base clock:\t%.2f MHz \n",  power_info.baseFrequency );
            printf("Minimal clock:\t%.2f MHz \n",  power_info.minFrequency );
            printf("Turbo Boost Steps:\n");
            for (int i=0; i < power_info.turbo.numSteps; i++ )
            {
                printf("C%d %.2f MHz \n",i+1,  power_info.turbo.steps[i] );
            }
        }
        printf(HLINE);
    }

    if (cpuid_info.model == SANDYBRIDGE_EP)
    {
        hasDRAM = 1;
    }
    else if ((cpuid_info.model != SANDYBRIDGE) &&
            (cpuid_info.model != SANDYBRIDGE_EP)  &&
            (cpuid_info.model != IVYBRIDGE)  &&
            (cpuid_info.model != IVYBRIDGE_EP)  &&
            (cpuid_info.model != HASWELL))
    {
        fprintf (stderr, "RAPL not supported on this processor!\n");
        exit(EXIT_FAILURE);
    }

    if (optInfo)
    {
        printf("Thermal Spec Power: %g Watts \n", power_info.tdp );
        printf("Minimum  Power: %g Watts \n", power_info.minPower);
        printf("Maximum  Power: %g Watts \n", power_info.maxPower);
        printf("Maximum  Time Window: %g micro sec \n", power_info.maxTimeWindow);
        printf(HLINE);
        exit(EXIT_SUCCESS);
    }

    if (optClock)
    {
        affinity_init();
        argString = bformat("S%u:0-%u", threadsSockets[0], cpuid_topology.numCoresPerSocket-1);
        for (int i=1; i<numSockets; i++)
        {
            bstring tExpr = bformat("@S%u:0-%u", threadsSockets[i], cpuid_topology.numCoresPerSocket-1);
            bconcat(argString, tExpr);
        }
        numThreads = bstr_to_cpuset(threads, argString);
        bdestroy(argString);
        perfmon_init(numThreads, threads, stdout);
        perfmon_setupEventSet(eventString, NULL);
    }

    {
        PowerData pDataPkg[MAX_NUM_NODES*2];
        PowerData pDataDram[MAX_NUM_NODES*2];
        printf("Measure on sockets: %d", threadsSockets[0]);
        for (int i=1; i<numSockets; i++)
        {
            printf(", %d", threadsSockets[i]);
        }
        printf("\n");

        if (optStethoscope)
        {
            if (optClock)
            {
                perfmon_startCounters();
            }
            else
            {
                for (int i=0; i<numSockets; i++)
                {
                    int cpuId = numa_info.nodes[threadsSockets[i]].processors[0];
                    if (hasDRAM) power_start(pDataDram+i, cpuId, DRAM);
                    power_start(pDataPkg+i, cpuId, PKG);
                }
            }
            sleep(optStethoscope);

            if (optClock)
            {
                perfmon_stopCounters();
                perfmon_printCounterResults();
                perfmon_finalize();
            }
            else
            {
                for (int i=0; i<numSockets; i++)
                {
                    int cpuId = numa_info.nodes[threadsSockets[i]].processors[0];
                    power_stop(pDataPkg+i, cpuId, PKG);
                    if (hasDRAM) power_stop(pDataDram+i, cpuId, DRAM);
                }
            }
            runtime = (double) optStethoscope;
        }
        else
        {
            TimerData time;
            argv +=  optind;
            bstring exeString = bfromcstr(argv[0]);

            for (int i=1; i<(argc-optind); i++)
            {
                bconchar(exeString, ' ');
                bcatcstr(exeString, argv[i]);
            }
            printf("%s\n",bdata(exeString));


            if (optClock)
            {
                perfmon_startCounters();
            }
            else
            {
                for (int i=0; i<numSockets; i++)
                {
                    int cpuId = numa_info.nodes[threadsSockets[i]].processors[0];
                    if (hasDRAM) power_start(pDataDram+i, cpuId, DRAM);
                    power_start(pDataPkg+i, cpuId, PKG);
                }

                timer_start(&time);
            }

            if (system(bdata(exeString)) == EOF)
            {
                fprintf(stderr, "Failed to execute %s!\n", bdata(exeString));
                exit(EXIT_FAILURE);
            }

            if (optClock)
            {
                perfmon_stopCounters();
                perfmon_printCounterResults();
                perfmon_finalize();
            }
            else
            {
                timer_stop(&time);

                for (int i=0; i<numSockets; i++)
                {
                    int cpuId = numa_info.nodes[threadsSockets[i]].processors[0];
                    power_stop(pDataPkg+i, cpuId, PKG);
                    if (hasDRAM) power_stop(pDataDram+i, cpuId, DRAM);
                }
                runtime = timer_print(&time);
            }
        }

        if (!optClock)
        {
            printf("Runtime: %g second \n",runtime);
            printf(HLINE);
            for (int i=0; i<numSockets; i++)
            {
                printf("Socket %d\n",threadsSockets[i]);
                printf("Domain: PKG \n");
                printf("Energy consumed: %g Joules \n", power_printEnergy(pDataPkg+i));
                printf("Power consumed: %g Watts \n", power_printEnergy(pDataPkg+i) / runtime );
                if (hasDRAM)
                {
                    printf("Domain: DRAM \n");
                    printf("Energy consumed: %g Joules \n", power_printEnergy(pDataDram+i));
                    printf("Power consumed: %g Watts \n", power_printEnergy(pDataDram+i) / runtime );
                }
                printf("\n");
            }
        }
    }

#if 0
    if ( cpuid_hasFeature(TM2) )
    {
        thermal_init(0);
        printf("Current core temperatures:\n");

        for (uint32_t i = 0; i < cpuid_topology.numCoresPerSocket; i++ )
        {
            printf("Core %d: %u C\n",
                    numa_info.nodes[socketId].processors[i],
                    thermal_read(numa_info.nodes[socketId].processors[i]));
        }
    }
#endif

    msr_finalize();
    return EXIT_SUCCESS;
}
Ejemplo n.º 14
0
int main(int argc, char* argv[])
{
    int i, j;
    int err;
    int* cpus;
    int gid;
    double result = 0.0;
    char estr[] = "INSTR_RETIRED_ANY:FIXC0,CPU_CLK_UNHALTED_CORE:FIXC1,CPU_CLK_UNHALTED_REF:FIXC2,TEMP_CORE:TMP0";
    // Load the topology module and print some values.
    err = topology_init();
    if (err < 0)
    {
        printf("Failed to initialize LIKWID's topology module\n");
        return 1;
    }
    // CpuInfo_t contains global information like name, CPU family, ...
    CpuInfo_t info = get_cpuInfo();
    // CpuTopology_t contains information about the topology of the CPUs.
    CpuTopology_t topo = get_cpuTopology();
    // Create affinity domains. Commonly only needed when reading Uncore counters
    //affinity_init();

    printf("Likwid example on a %s with %d CPUs\n", info->name, topo->numHWThreads);

    cpus = (int*)malloc(topo->numHWThreads * sizeof(int));
    if (!cpus)
        return 1;

    for (i=0;i<topo->numHWThreads;i++)
    {
        cpus[i] = topo->threadPool[i].apicId;
    }

    // Must be called before perfmon_init() but only if you want to use another
    // access mode as the pre-configured one. For direct access (0) you have to
    // be root.
    //accessClient_setaccessmode(0);

    // Initialize the perfmon module.
    err = perfmon_init(topo->numHWThreads, cpus);
    if (err < 0)
    {
        printf("Failed to initialize LIKWID's performance monitoring module\n");
        topology_finalize();
        return 1;
    }

    // Add eventset string to the perfmon module.
    gid = perfmon_addEventSet(estr);
    if (gid < 0)
    {
        printf("Failed to add event string %s to LIKWID's performance monitoring module\n", estr);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }

    // Setup the eventset identified by group ID (gid).
    err = perfmon_setupCounters(gid);
    if (err < 0)
    {
        printf("Failed to setup group %d in LIKWID's performance monitoring module\n", gid);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }
    // Start all counters in the previously set up event set.
    err = perfmon_startCounters();
    if (err < 0)
    {
        printf("Failed to start counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }
    // Perform something
    sleep(2);
    // Stop all counters in the previously started event set.
    err = perfmon_stopCounters();
    if (err < 0)
    {
        printf("Failed to stop counters for group %d for thread %d\n",gid, (-1*err)-1);
        perfmon_finalize();
        topology_finalize();
        return 1;
    }


    // Print the result of every thread/CPU for all events in estr.
    char* ptr = strtok(estr,",");
    j = 0;
    while (ptr != NULL)
    {
        for (i = 0;i < topo->numHWThreads; i++)
        {
            result = perfmon_getResult(gid, j, cpus[i]);
            printf("Measurement result for event set %s at CPU %d: %f\n", ptr, cpus[i], result);
        }
        ptr = strtok(NULL,",");
        j++;
    }

    // Uninitialize the perfmon module.
    perfmon_finalize();
    // Uninitialize the topology module.
    topology_finalize();
    return 0;
}