void pcmInit() {
std::cout << "Initializing PCM" << std::endl;
PCM* m = PCM::getInstance();
m->disableJKTWorkaround();
switch( m->program() ) {
case PCM::Success:
std::cout << "PCM Initialized" << std::endl;
return;
case PCM::PMUBusy:
std::cout << "PCM::PMU Busy!" << std::endl;
m->resetPMU();
return;
default:
return;
}
}
int main(int argc, char * argv[])
{
std::cout << "\n Intel(r) Performance Counter Monitor " << INTEL_PCM_VERSION << std::endl;
std::cout << "\n Power Monitoring Utility\n Copyright (c) 2011-2012 Intel Corporation\n";
int imc_profile = 0;
int pcu_profile = 0;
int delay = -1;
char * ext_program = NULL;
freq_band[0] = default_freq_band[0];
freq_band[1] = default_freq_band[1];
freq_band[2] = default_freq_band[2];
int my_opt = -1;
while ((my_opt = getopt(argc, argv, "m:p:a:b:c:")) != -1)
{
switch(my_opt)
{
case 'm':
imc_profile = atoi(optarg);
break;
case 'p':
pcu_profile = atoi(optarg);
break;
case 'a':
freq_band[0] = atoi(optarg);
break;
case 'b':
freq_band[1] = atoi(optarg);
break;
case 'c':
freq_band[2] = atoi(optarg);
break;
default:
print_usage(argv[0]);
return -1;
}
}
if (optind >= argc)
{
print_usage(argv[0]);
return -1;
}
delay = atoi(argv[optind]);
if(delay == 0)
ext_program = argv[optind];
else
delay = (delay<0)?1:delay;
#ifdef _MSC_VER
// Increase the priority a bit to improve context switching delays on Windows
SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_ABOVE_NORMAL);
TCHAR driverPath[1024];
GetCurrentDirectory(1024, driverPath);
wcscat(driverPath, L"\\msr.sys");
// WARNING: This driver code (msr.sys) is only for testing purposes, not for production use
Driver drv;
// drv.stop(); // restart driver (usually not needed)
if (!drv.start(driverPath))
{
std::cout << "Can not access CPU performance counters" << std::endl;
std::cout << "You must have signed msr.sys driver in your current directory and have administrator rights to run this program" << std::endl;
return -1;
}
#endif
PCM * m = PCM::getInstance();
m->disableJKTWorkaround();
if(!(m->hasPCICFGUncore()))
{
std::cout <<"Unsupported processor model ("<<m->getCPUModel()<<"). Only models "<<PCM::JAKETOWN<<" (JAKETOWN), " << PCM::IVYTOWN<< " (IVYTOWN) are supported."<< std::endl;
return -1;
}
if(PCM::Success != m->programServerUncorePowerMetrics(imc_profile,pcu_profile,freq_band))
{
#ifdef _MSC_VER
std::cout << "You must have signed msr.sys driver in your current directory and have administrator rights to run this program" << std::endl;
#elif defined(__linux__)
std::cout << "You need to be root and loaded 'msr' Linux kernel module to execute the program. You may load the 'msr' module with 'modprobe msr'. \n";
#endif
return -1;
}
ServerUncorePowerState * BeforeState = new ServerUncorePowerState[m->getNumSockets()];
ServerUncorePowerState * AfterState = new ServerUncorePowerState[m->getNumSockets()];
uint64 BeforeTime = 0, AfterTime = 0;
std::cout << std::dec << std::endl;
std::cout.precision(2);
std::cout << std::fixed;
std::cout << "\nMC counter group: "<<imc_profile << std::endl;
std::cout << "PCU counter group: "<<pcu_profile << std::endl;
if(pcu_profile == 0)
std::cout << "Freq bands [0/1/2]: "<<freq_band[0]*100 << " MHz; "<< freq_band[1]*100 << " MHz; "<<freq_band[2]*100 << " MHz; "<<std::endl;
if(!ext_program)
std::cout << "Update every "<<delay<<" seconds"<< std::endl;
uint32 i = 0;
BeforeTime = m->getTickCount();
for(i=0; i<m->getNumSockets(); ++i)
BeforeState[i] = m->getServerUncorePowerState(i);
while(1)
{
std::cout << "----------------------------------------------------------------------------------------------"<<std::endl;
#ifdef _MSC_VER
int delay_ms = delay * 1000;
// compensate slow Windows console output
if(AfterTime) delay_ms -= (int)(m->getTickCount() - BeforeTime);
if(delay_ms < 0) delay_ms = 0;
#else
int delay_ms = delay * 1000;
#endif
if(ext_program)
MySystem(ext_program);
else
MySleepMs(delay_ms);
AfterTime = m->getTickCount();
for(i=0; i<m->getNumSockets(); ++i)
AfterState[i] = m->getServerUncorePowerState(i);
std::cout << "Time elapsed: "<<AfterTime-BeforeTime<<" ms\n";
std::cout << "Called sleep function for "<<delay_ms<<" ms\n";
for(uint32 socket=0;socket<m->getNumSockets();++socket)
{
for(uint32 port=0;port<m->getQPILinksPerSocket();++port)
{
std::cout << "S"<<socket<<"P"<<port
<< "; QPIClocks: "<< getQPIClocks(port,BeforeState[socket],AfterState[socket])
<< "; L0p Tx Cycles: "<< 100.*getNormalizedQPIL0pTxCycles(port,BeforeState[socket],AfterState[socket])<< "%"
<< "; L1 Cycles: " << 100.*getNormalizedQPIL1Cycles(port,BeforeState[socket],AfterState[socket])<< "%"
<< "\n";
}
for(uint32 channel=0;channel<m->getMCChannelsPerSocket();++channel)
{
if(imc_profile <= 3 && imc_profile >= 0)
{
std::cout << "S"<<socket<<"CH"<<channel <<"; DRAMClocks: "<< getDRAMClocks(channel,BeforeState[socket],AfterState[socket])
<< "; Rank"<<getFirstRank(imc_profile)<<" CKE Off Residency: "<< std::setw(3) <<
100.*getCKEOffResidency(channel,getFirstRank(imc_profile),BeforeState[socket],AfterState[socket])<<"%"
<< "; Rank"<<getFirstRank(imc_profile)<<" CKE Off Average Cycles: "<<
getCKEOffAverageCycles(channel,getFirstRank(imc_profile),BeforeState[socket],AfterState[socket])
<< "; Rank"<<getFirstRank(imc_profile)<<" Cycles per transition: "<<
getCyclesPerTransition(channel,getFirstRank(imc_profile),BeforeState[socket],AfterState[socket])
<< "\n";
std::cout << "S"<<socket<<"CH"<<channel <<"; DRAMClocks: "<< getDRAMClocks(channel,BeforeState[socket],AfterState[socket])
<< "; Rank"<<getSecondRank(imc_profile)<<" CKE Off Residency: "<< std::setw(3) <<
100.*getCKEOffResidency(channel,getSecondRank(imc_profile),BeforeState[socket],AfterState[socket])<<"%"
<< "; Rank"<<getSecondRank(imc_profile)<<" CKE Off Average Cycles: "<<
getCKEOffAverageCycles(channel,getSecondRank(imc_profile),BeforeState[socket],AfterState[socket])
<< "; Rank"<<getSecondRank(imc_profile)<<" Cycles per transition: "<<
getCyclesPerTransition(channel,getSecondRank(imc_profile),BeforeState[socket],AfterState[socket])
<< "\n";
} else if(imc_profile == 4)
{
std::cout << "S"<<socket<<"CH"<<channel
<< "; DRAMClocks: "<< getDRAMClocks(channel,BeforeState[socket],AfterState[socket])
<< "; Self-refresh cycles: "<< getSelfRefreshCycles(channel,BeforeState[socket],AfterState[socket])
<< "; Self-refresh transitions: "<< getSelfRefreshTransitions(channel,BeforeState[socket],AfterState[socket])
<< "\n";
}
}
switch(pcu_profile)
{
case 0:
std::cout << "S"<<socket
<< "; PCUClocks: "<< getPCUClocks(BeforeState[socket],AfterState[socket])
<< "; Freq band 0/1/2 cycles: "<< 100.*getNormalizedPCUCounter(1,BeforeState[socket],AfterState[socket])<<"%"
<< "; "<< 100.*getNormalizedPCUCounter(2,BeforeState[socket],AfterState[socket])<<"%"
<< "; "<< 100.*getNormalizedPCUCounter(3,BeforeState[socket],AfterState[socket])<<"%"
<< "\n";
break;
case 1:
std::cout << "S"<<socket
<< "; PCUClocks: "<< getPCUClocks(BeforeState[socket],AfterState[socket])
<< "; core C0/C3/C6-state residency: "<< getNormalizedPCUCounter(1,BeforeState[socket],AfterState[socket])
<< "; "<< getNormalizedPCUCounter(2,BeforeState[socket],AfterState[socket])
<< "; "<< getNormalizedPCUCounter(3,BeforeState[socket],AfterState[socket])
<< "\n";
break;
case 2:
std::cout << "S"<<socket
<< "; PCUClocks: "<< getPCUClocks(BeforeState[socket],AfterState[socket])
<< "; Internal prochot cycles: "<< getNormalizedPCUCounter(1,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "; External prochot cycles:" << getNormalizedPCUCounter(2,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "; Thermal freq limit cycles:" << getNormalizedPCUCounter(3,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "\n";
break;
case 3:
std::cout << "S"<<socket
<< "; PCUClocks: "<< getPCUClocks(BeforeState[socket],AfterState[socket])
<< "; Thermal freq limit cycles: "<< getNormalizedPCUCounter(1,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "; Power freq limit cycles:" << getNormalizedPCUCounter(2,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "; Clipped freq limit cycles:" << getNormalizedPCUCounter(3,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "\n";
break;
case 4:
std::cout << "S"<<socket
<< "; PCUClocks: "<< getPCUClocks(BeforeState[socket],AfterState[socket])
<< "; OS freq limit cycles: "<< getNormalizedPCUCounter(1,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "; Power freq limit cycles:" << getNormalizedPCUCounter(2,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "; Clipped freq limit cycles:" << getNormalizedPCUCounter(3,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "\n";
break;
case 5:
std::cout << "S"<<socket
<< "; PCUClocks: "<< getPCUClocks(BeforeState[socket],AfterState[socket])
<< "; Frequency transition count: "<< getPCUCounter(1,BeforeState[socket],AfterState[socket]) <<" "
<< "; Cycles spent changing frequency: " << getNormalizedPCUCounter(2,BeforeState[socket],AfterState[socket])*100. <<" %"
<< "\n";
break;
}
std::cout << "S"<<socket
<< "; Consumed energy units: "<< getConsumedEnergy(BeforeState[socket],AfterState[socket])
<< "; Consumed Joules: "<< getConsumedJoules(BeforeState[socket],AfterState[socket])
<< "; Watts: "<< 1000.*getConsumedJoules(BeforeState[socket],AfterState[socket])/double(AfterTime-BeforeTime)
<< "; Thermal headroom below TjMax: " << AfterState[socket].getPackageThermalHeadroom()
<< "\n";
std::cout << "S"<<socket
<< "; Consumed DRAM energy units: "<< getDRAMConsumedEnergy(BeforeState[socket],AfterState[socket])
<< "; Consumed DRAM Joules: "<< getDRAMConsumedJoules(BeforeState[socket],AfterState[socket])
<< "; DRAM Watts: "<< 1000.*getDRAMConsumedJoules(BeforeState[socket],AfterState[socket])/double(AfterTime-BeforeTime)
<< "\n";
}
std::swap(BeforeState,AfterState);
std::swap(BeforeTime,AfterTime);
if(ext_program)
{
std::cout << "----------------------------------------------------------------------------------------------"<<std::endl;
break;
}
}
delete [] BeforeState;
delete [] AfterState;
}