int test_perfmonsetup()
{
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;
if (perfmon_getNumberOfGroups() != 0)
goto fail;
if (perfmon_getNumberOfThreads() != 1)
goto fail;
ret = perfmon_addEventSet(eventset_ok);
if (ret != 0)
goto fail;
group1 = ret;
if (perfmon_getNumberOfGroups() != 1)
goto fail;
if (perfmon_getNumberOfEvents(group1) != 3)
goto fail;
ret = perfmon_setupCounters(group1);
if (ret != 0)
goto fail;
if (perfmon_getIdOfActiveGroup() != group1)
goto fail;
ret = perfmon_addEventSet(eventset_option);
if (ret != 1)
goto fail;
group2 = ret;
if (perfmon_getIdOfActiveGroup() != group1)
goto fail;
if (perfmon_getNumberOfGroups() != 2)
goto fail;
if (perfmon_getNumberOfEvents(group1) != 3)
goto fail;
if (perfmon_getNumberOfEvents(group2) != 3)
goto fail;
perfmon_finalize();
topology_finalize();
return 1;
fail:
perfmon_finalize();
topology_finalize();
return 0;
}
static int lua_likwid_getNumberOfGroups(lua_State* L)
{
int number;
if (perfmon_isInitialized == 0)
{
return 0;
}
number = perfmon_getNumberOfGroups();
lua_pushnumber(L,number);
return 1;
}
int main(int argc, char* argv[])
{
int i, g;
int nevents = 10;
double events[10];
double time;
int count;
// Init Marker API in serial region once in the beginning
LIKWID_MARKER_INIT;
#pragma omp parallel
{
// Each thread must add itself to the Marker API, therefore must be
// in parallel region
LIKWID_MARKER_THREADINIT;
// Optional. Register region name
LIKWID_MARKER_REGISTER("example");
}
// perfmon_getNumberOfGroups is not part of the MarkerAPI,
// it belongs to the normal LIKWID API. But the MarkerAPI
// has no function to get the number of configured groups.
for (g=0;g < perfmon_getNumberOfGroups(); g++)
{
#pragma omp parallel
{
printf("Thread %d sleeps now for %d seconds\n", omp_get_thread_num(), SLEEPTIME);
// Start measurements inside a parallel region
LIKWID_MARKER_START("example");
// Insert your code here.
// Often contains an OpenMP for pragma. Regions can be nested.
sleep(SLEEPTIME);
// Stop measurements inside a parallel region
LIKWID_MARKER_STOP("example");
printf("Thread %d wakes up again\n", omp_get_thread_num());
// If you need the performance data inside your application, use
LIKWID_MARKER_GET("example", &nevents, events, &time, &count);
// where events is an array of doubles with nevents entries,
// time is a double* and count an int*.
printf("Region example measures %d events, total measurement time is %f\n", nevents, time);
printf("The region was called %d times\n", count);
for (i = 0; i < nevents; i++)
{
printf("Event %d: %f\n", i, events[i]);
}
// If multiple groups given, you can switch to the next group
LIKWID_MARKER_SWITCH;
}
}
// Close Marker API and write results to file for further evaluation done
// by likwid-perfctr
LIKWID_MARKER_CLOSE;
return 0;
}
static int lua_likwid_switchGroup(lua_State* L)
{
int ret = -1;
int newgroup = lua_tonumber(L,1)-1;
if (perfmon_isInitialized == 0)
{
return 0;
}
if (newgroup >= perfmon_getNumberOfGroups())
{
newgroup = 0;
}
if (newgroup == perfmon_getIdOfActiveGroup())
{
lua_pushinteger(L, ret);
return 1;
}
ret = perfmon_switchActiveGroup(newgroup);
lua_pushinteger(L, ret);
return 1;
}
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_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;
}
