static int lua_likwid_finalize(lua_State* L)
{
if (topology_isInitialized == 1)
{
topology_finalize();
topology_isInitialized = 0;
cputopo = NULL;
cpuinfo = NULL;
}
if (numa_isInitialized == 1)
{
numa_finalize();
numa_isInitialized = 0;
numainfo = NULL;
}
if (affinity_isInitialized == 1)
{
affinity_finalize();
affinity_isInitialized = 0;
affinity = NULL;
}
if (perfmon_isInitialized == 1)
{
perfmon_finalize();
perfmon_isInitialized = 0;
}
if (config_isInitialized == 1)
{
destroy_configuration();
config_isInitialized = 0;
configfile = NULL;
}
return 0;
}
int test_affinityinit()
{
int i = 0;
topology_init();
CpuTopology_t cputopo = get_cpuTopology();
numa_init();
affinity_init();
AffinityDomains_t doms = get_affinityDomains();
if (doms == NULL)
goto fail;
if (doms->numberOfSocketDomains != cputopo->numSockets)
goto fail;
if (doms->numberOfNumaDomains == 0)
goto fail;
if (doms->numberOfProcessorsPerSocket == 0)
goto fail;
if (doms->numberOfAffinityDomains == 0)
goto fail;
if (doms->numberOfCacheDomains == 0)
goto fail;
if (doms->numberOfCoresPerCache == 0)
goto fail;
if (doms->numberOfProcessorsPerCache == 0)
goto fail;
if (doms->numberOfProcessorsPerCache < doms->numberOfCoresPerCache)
goto fail;
if (doms->domains == NULL)
goto fail;
for (i = 0; i < doms->numberOfAffinityDomains; i++)
{
if (doms->domains[i].numberOfProcessors == 0)
goto fail;
if (doms->domains[i].numberOfCores == 0)
goto fail;
if (doms->domains[i].numberOfProcessors < doms->domains[i].numberOfCores)
goto fail;
if (doms->domains[i].processorList == NULL)
goto fail;
}
affinity_finalize();
topology_finalize();
return 1;
fail:
affinity_finalize();
topology_finalize();
return 0;
}
static int lua_likwid_putAffinityInfo(lua_State* L)
{
if (affinity_isInitialized)
{
affinity_finalize();
affinity_isInitialized = 0;
affinity = NULL;
}
return 0;
}
int test_perfmoninit_valid()
{
int cpu = 0;
topology_init();
affinity_init();
int ret = perfmon_init(1, &cpu);
if (ret != 0)
goto fail;
if (perfmon_getNumberOfGroups() != 0)
goto fail;
if (perfmon_getNumberOfThreads() != 1)
goto fail;
perfmon_finalize();
affinity_finalize();
topology_finalize();
return 1;
fail:
perfmon_finalize();
affinity_finalize();
topology_finalize();
return 0;
}
int test_perfmoninit()
{
int cpu = 0;
int i;
topology_init();
affinity_init();
for(i=0;i<10;i++)
{
perfmon_init(1, &cpu);
perfmon_finalize();
}
affinity_finalize();
topology_finalize();
return 1;
}
int test_perfmonperfgroup()
{
CpuInfo_t cpuinfo;
int i;
int cpu = 0;
topology_init();
cpuinfo = get_cpuInfo();
int ret = perfmon_init(1, &cpu);
if (ret != 0) {
printf("Perfmon init failed\n");
goto fail;
}
char** glist = NULL;
char** slist = NULL;
char** llist = NULL;
ret = perfmon_getGroups(&glist, &slist, &llist);
if (ret <= 0)
{
goto fail;
}
ret = perfmon_addEventSet(glist[0]);
if (ret != 0) {
printf("Perfmon addEventSet(%s) failed\n", glist[0]);
goto fail;
}
if (perfmon_getNumberOfEvents(ret) == 0) {
printf("Perfmon number of events == 0\n");
goto fail;
}
if (perfmon_getNumberOfMetrics(ret) == 0) {
printf("Perfmon number of metrics == 0\n");
goto fail;
}
for (i=0; i<perfmon_getNumberOfEvents(ret); i++) {
if (strcmp(perfmon_getEventName(ret, i), "") == 0)
goto fail;
if (strcmp(perfmon_getCounterName(ret, i), "") == 0)
goto fail;
}
if (strcmp(perfmon_getGroupName(ret), "Custom") == 0)
{
goto fail;
}
if (strcmp(perfmon_getGroupInfoShort(ret), "Custom") == 0)
{
goto fail;
}
if (strcmp(perfmon_getGroupInfoLong(ret), "Custom") == 0)
{
goto fail;
}
if (perfmon_getLastTimeOfGroup(ret) != 0)
{
goto fail;
}
if (perfmon_getTimeOfGroup(ret) != 0)
{
goto fail;
}
for (i=0; i<perfmon_getNumberOfMetrics(ret); i++) {
if (strcmp(perfmon_getMetricName(ret, i), "") == 0)
goto fail;
if (perfmon_getMetric(ret, i, 0) != 0.0)
goto fail;
}
free(glist);
free(slist);
free(llist);
perfmon_finalize();
affinity_finalize();
topology_finalize();
return 1;
fail:
if (glist)
free(glist);
if (slist)
free(slist);
if (llist)
free(llist);
perfmon_finalize();
affinity_finalize();
topology_finalize();
return 0;
}
int test_perfmoncustomgroup()
{
CpuInfo_t cpuinfo;
int cpu = 0;
topology_init();
cpuinfo = get_cpuInfo();
int ret = perfmon_init(1, &cpu);
if (ret != 0) {
printf("Perfmon init failed\n");
goto fail;
}
ret = perfmon_addEventSet(eventset_ok);
if (ret != 0) {
printf("Perfmon addEventSet(ok) failed\n");
goto fail;
}
if (perfmon_getNumberOfEvents(ret) != 3) {
printf("Perfmon number of events != 3\n");
goto fail;
}
if (perfmon_getNumberOfMetrics(ret) != 0) {
printf("Perfmon number of metrics != 0\n");
goto fail;
}
if (strcmp(perfmon_getEventName(ret, 0), event1_ok) != 0)
{
goto fail;
}
if (strcmp(perfmon_getEventName(ret, 1), event2_ok) != 0)
{
goto fail;
}
if (strcmp(perfmon_getEventName(ret, 2), event3_ok) != 0)
{
goto fail;
}
if (strcmp(perfmon_getCounterName(ret, 0), ctr1_ok) != 0)
{
goto fail;
}
if (strcmp(perfmon_getCounterName(ret, 1), ctr2_ok) != 0)
{
goto fail;
}
if (strcmp(perfmon_getCounterName(ret, 2), ctr3_ok) != 0)
{
goto fail;
}
if (strcmp(perfmon_getGroupName(ret), "Custom") != 0)
{
goto fail;
}
if (strcmp(perfmon_getGroupInfoShort(ret), "Custom") != 0)
{
goto fail;
}
if (strcmp(perfmon_getGroupInfoLong(ret), "Custom") != 0)
{
goto fail;
}
if (perfmon_getLastTimeOfGroup(ret) != 0)
{
goto fail;
}
perfmon_finalize();
affinity_finalize();
topology_finalize();
return 1;
fail:
perfmon_finalize();
affinity_finalize();
topology_finalize();
return 0;
}
int test_perfmonaddeventset()
{
char eventset_fail1[] = "INSTR_RETIRED.ANY:FIXC0";
char eventset_fail2[] = "INSTR_RETIRED-ANY:FIXC0";
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) {
printf("Perfmon init failed\n");
goto fail;
}
if (perfmon_getNumberOfGroups() != 0) {
printf("Perfmon number of groups != 0\n");
goto fail;
}
if (perfmon_getNumberOfThreads() != 1) {
printf("Perfmon number of threads != 1\n");
goto fail;
}
if (perfmon_getIdOfActiveGroup() != -1) {
printf("Perfmon id of active group != -1\n");
goto fail;
}
ret = perfmon_addEventSet(eventset_ok);
if (ret != 0) {
printf("Perfmon addEventSet(ok) failed\n");
goto fail;
}
if (perfmon_getNumberOfGroups() != 1) {
printf("Perfmon number of groups != 1\n");
goto fail;
}
if (perfmon_getNumberOfEvents(ret) != 3) {
printf("Perfmon number of events != 3\n");
goto fail;
}
if (perfmon_getIdOfActiveGroup() != -1) {
printf("Perfmon id of active group != -1\n");
goto fail;
}
ret = perfmon_addEventSet(eventset_option);
if (ret != 1) {
printf("Perfmon addEventSet(options) failed\n");
goto fail;
}
if (perfmon_getNumberOfGroups() != 2) {
printf("Perfmon number of groups != 2\n");
goto fail;
}
if (perfmon_getNumberOfEvents(ret) != 3) {
printf("Perfmon number of events != 3\n");
goto fail;
}
if (perfmon_getIdOfActiveGroup() != -1) {
printf("Perfmon id of active group != -1\n");
goto fail;
}
ret = perfmon_addEventSet(eventset_fail1);
if (ret >= 0) {
printf("Perfmon addEventSet(fail1) failed\n");
goto fail;
}
if (perfmon_getNumberOfGroups() != 2) {
printf("Perfmon number of groups != 2\n");
goto fail;
}
ret = perfmon_addEventSet(eventset_fail2);
if (ret >= 0) {
printf("Perfmon addEventSet(fail2) failed\n");
goto fail;
}
if (perfmon_getNumberOfGroups() != 2) {
printf("Perfmon number of groups != 2\n");
goto fail;
}
if (perfmon_getIdOfActiveGroup() != -1) {
printf("Perfmon id of active group != -1\n");
goto fail;
}
perfmon_finalize();
affinity_finalize();
topology_finalize();
return 1;
fail:
perfmon_finalize();
affinity_finalize();
topology_finalize();
return 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;
}
int main(int argc, char* argv[])
{
int i, j;
int err;
int* cpus;
int gid;
double result = 0.0;
char estr[] = "L2_LINES_IN_ALL:PMC0,L2_TRANS_L2_WB:PMC1";
//perfmon_setVerbosity(3);
// 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(10);
// 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, i);
printf("Measurement result for event set %s at CPU %d: %f\n", ptr, cpus[i], result);
}
ptr = strtok(NULL,",");
j++;
}
free(cpus);
// Uninitialize the perfmon module.
perfmon_finalize();
affinity_finalize();
// Uninitialize the topology module.
topology_finalize();
return 0;
}
