int test_perfmonresult_noadd()
{
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;
double result = perfmon_getResult(0,0,0);
if (result != 0)
goto fail;
perfmon_finalize();
topology_finalize();
return 1;
fail:
perfmon_finalize();
topology_finalize();
return 0;
}
int test_perfmonresult_nostart()
{
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;
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;
}
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;
}
int test_perfmonresult_noinit()
{
double result = perfmon_getResult(0,0,0);
if (result != 0)
goto fail;
return 1;
fail:
return 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;
}
static int lua_likwid_getResult(lua_State* L)
{
int groupId, eventId, threadId;
double result = 0;
groupId = lua_tonumber(L,1);
eventId = lua_tonumber(L,2);
threadId = lua_tonumber(L,3);
result = perfmon_getResult(groupId-1, eventId-1, threadId-1);
lua_pushnumber(L,result);
return 1;
}
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[] = "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;
}