Opcm push 202011

cpucounters.cpp

@@ -2061,6 +2061,7 @@ bool PCM::checkModel()
     if (cpu_model == BROADWELL_XEON_E3) cpu_model = BROADWELL;
     if (cpu_model == SKL_UY) cpu_model = SKL;
     if (cpu_model == KBL_1) cpu_model = KBL;
+    if (cpu_model == CML) cpu_model = KBL;
 
     if(!isCPUModelSupported((int)cpu_model))
     {

cpucounters.h

@@ -1243,6 +1243,7 @@ class PCM_API PCM
         SKL_UY = 78,
         KBL = 158,
         KBL_1 = 142,
+        CML = 166,
         ICL = 126,
         BDX = 79,
         KNL = 87,

pcm-core.cpp

@@ -43,7 +43,6 @@
 #include <vector>
 #define PCM_DELAY_DEFAULT 1.0 // in seconds
 #define PCM_DELAY_MIN 0.015 // 15 milliseconds is practical on most modern CPUs
-#define PCM_CALIBRATION_INTERVAL 50 // calibrate clock only every 50th iteration
 #define MAX_CORES 4096
 
 using namespace std;
@@ -136,6 +135,7 @@ void print_usage(const string progname)
     cerr << "  [-e event1] [-e event2] [-e event3] .. => optional list of custom events to monitor\n";
 	cerr << "  event description example: cpu/umask=0x01,event=0x05,name=MISALIGN_MEM_REF.LOADS/ \n";
 	cerr << "  -yc   | --yescores  | /yc              => enable specific cores to output\n";
+	cerr << "  -i[=number] | /i[=number]              => allow to determine number of iterations\n";
     print_help_force_rtm_abort_mode(41);
 	cerr << " Examples:\n";
 	cerr << "  " << progname << " 1                   => print counters every second without core and socket output\n";
@@ -292,10 +292,8 @@ int main(int argc, char * argv[])
 	char **sysArgv = NULL;
 	uint32 cur_event = 0;
 	bool csv = false;
-	long diff_usec = 0; // deviation of clock is useconds between measurements
 	uint64 txn_rate = 1;
-	int calibrated = PCM_CALIBRATION_INTERVAL - 2; // keeps track is the clock calibration needed
-	unsigned int numberOfIterations = 0; // number of iterations
+	MainLoop mainLoop;
 	string program = string(argv[0]);
 	EventSelectRegister regs[PERF_MAX_COUNTERS];
 	PCM::ExtendedCustomCoreEventDescription conf;
@@ -335,17 +333,8 @@ int main(int argc, char * argv[])
 			continue;
 		}
 		else
-		if (strncmp(*argv, "-i", 2) == 0 ||
-			strncmp(*argv, "/i", 2) == 0)
+		if (mainLoop.parseArg(*argv))
 		{
-			string cmd = string(*argv);
-			size_t found = cmd.find('=', 2);
-			if (found != string::npos) {
-				string tmp = cmd.substr(found + 1);
-				if (!tmp.empty()) {
-					numberOfIterations = (unsigned int)atoi(tmp.c_str());
-				}
-			}
 			continue;
 		}
 		else if (strncmp(*argv, "-c",2) == 0 ||
@@ -515,38 +504,12 @@ int main(int argc, char * argv[])
 	}
 
 
-	unsigned int ic = 1;
-
-	while ((ic <= numberOfIterations) || (numberOfIterations == 0))
+	mainLoop([&]()
 	{
 		if(!csv) cout << std::flush;
-		int delay_ms = int(delay * 1000);
-		int calibrated_delay_ms = delay_ms;
-#ifdef _MSC_VER
-		// compensate slow Windows console output
-		if(AfterTime) delay_ms -= (int)(m->getTickCount() - BeforeTime);
-		if(delay_ms < 0) delay_ms = 0;
-#else
-		// compensation of delay on Linux/UNIX
-		// to make the samling interval as monotone as possible
-		struct timeval start_ts, end_ts;
-		if(calibrated == 0) {
-			gettimeofday(&end_ts, NULL);
-			diff_usec = (end_ts.tv_sec-start_ts.tv_sec)*1000000.0+(end_ts.tv_usec-start_ts.tv_usec);
-			calibrated_delay_ms = delay_ms - diff_usec/1000.0;
-		}
-#endif
-		if (sysCmd == NULL || numberOfIterations != 0 || m->isBlocked() == false)
-		{
-			MySleepMs(calibrated_delay_ms);
-		}
 
-#ifndef _MSC_VER
-		calibrated = (calibrated + 1) % PCM_CALIBRATION_INTERVAL;
-		if(calibrated == 0) {
-			gettimeofday(&start_ts, NULL);
-		}
-#endif
+		calibratedSleep(delay, sysCmd, mainLoop, m);
+
 		AfterTime = m->getTickCount();
 		m->getAllCounterStates(SysAfterState, DummySocketStates, AfterState);
 
@@ -611,9 +574,9 @@ int main(int argc, char * argv[])
 
 		if ( m->isBlocked() ) {
 			// in case PCM was blocked after spawning child application: break monitoring loop here
-			break;
+			return false;
 		}
-		++ic;
-	}
+		return true;
+	});
 	exit(EXIT_SUCCESS);
 }

pcm-latency.cpp

@@ -449,13 +449,13 @@ void build_registers(PCM *m, PCM::ExtendedCustomCoreEventDescription conf, bool
     m->programServerUncoreLatencyMetrics(enable_pmm);
 }
 
-void collect_data(PCM *m, bool enable_pmm, bool enable_verbose, int delay_ms)
+void collect_data(PCM *m, bool enable_pmm, bool enable_verbose, int delay_ms, MainLoop & mainLoop)
 {
 
     BeforeState = new ServerUncoreCounterState[m->getNumSockets()];
     AfterState = new ServerUncoreCounterState[m->getNumSockets()];
 
-    while (1)
+    mainLoop([&]()
     {
         collect_beforestate_uncore(m);
         collect_beforestate_core(m);
@@ -470,7 +470,8 @@ void collect_data(PCM *m, bool enable_pmm, bool enable_verbose, int delay_ms)
 
         print_all_stats(m, enable_pmm, enable_verbose);
         std::cout << std::flush;
-    }
+        return true;
+    });
 
     delete[] BeforeState;
     delete[] AfterState;
@@ -479,9 +480,10 @@ void collect_data(PCM *m, bool enable_pmm, bool enable_verbose, int delay_ms)
 void print_usage()
 {
     cerr << "\nUsage: \n";
-    cerr << " -h | --help | /h        => Print this help and exit\n";
-    cerr << " --PMM | -pmm            => to enable PMM (Default DDR uncore latency)\n";
-    cerr << " -v | --verbose          => Verbose Output\n";
+    cerr << " -h | --help | /h          => print this help and exit\n";
+    cerr << " --PMM | -pmm              => to enable PMM (Default DDR uncore latency)\n";
+    cerr << " -i[=number] | /i[=number] => allow to determine number of iterations\n";
+    cerr << " -v | --verbose            => verbose Output\n";
     cerr << "\n";
 }
 
@@ -493,6 +495,7 @@ int main(int argc, char * argv[])
     bool enable_pmm = false;
     bool enable_verbose = false;
     int delay_ms = 1000;
+    MainLoop mainLoop;
     if(argc > 1) do
     {
         argv++;
@@ -505,6 +508,10 @@ int main(int argc, char * argv[])
             print_usage();
             exit(EXIT_FAILURE);
         }
+        else if (mainLoop.parseArg(*argv))
+        {
+            continue;
+        }
         else if (strncmp(*argv, "--PMM",6) == 0 || strncmp(*argv, "-pmm", 5) == 0)
         {
             argv++;
@@ -528,7 +535,7 @@ int main(int argc, char * argv[])
     PCM * m = PCM::getInstance();
 
     build_registers(m, conf, enable_pmm, enable_verbose);
-    collect_data(m, enable_pmm, enable_verbose, delay_ms);
+    collect_data(m, enable_pmm, enable_verbose, delay_ms, mainLoop);
 
     exit(EXIT_SUCCESS);
 }

pcm-memory.cpp

@@ -39,7 +39,6 @@
 
 #define PCM_DELAY_DEFAULT 1.0 // in seconds
 #define PCM_DELAY_MIN 0.015 // 15 milliseconds is practical on most modern CPUs
-#define PCM_CALIBRATION_INTERVAL 50 // calibrate clock only every 50th iteration
 
 #define DEFAULT_DISPLAY_COLUMNS 2
 
@@ -91,6 +90,7 @@ void print_help(const string prog_name)
          << "                                        to a file, in case filename is provided\n";
     cerr << "  -columns=X | /columns=X            => Number of columns to display the NUMA Nodes, defaults to 2.\n";
     cerr << "  -all | /all                        => Display all channels (even with no traffic)\n";
+    cerr << "  -i[=number] | /i[=number]          => allow to determine number of iterations\n";
 #ifdef _MSC_VER
     cerr << "  --uninstallDriver | --installDriver=> (un)install driver\n";
 #endif
@@ -826,12 +826,8 @@ int main(int argc, char * argv[])
     uint32 no_columns = DEFAULT_DISPLAY_COLUMNS; // Default number of columns is 2
     char * sysCmd = NULL;
     char ** sysArgv = NULL;
-#ifndef _MSC_VER
-    long diff_usec = 0; // deviation of clock is useconds between measurements
-    int calibrated = PCM_CALIBRATION_INTERVAL - 2; // keeps track is the clock calibration needed
-#endif
     int rankA = -1, rankB = -1;
-    unsigned int numberOfIterations = 0; // number of iterations
+    MainLoop mainLoop;
 
     string program = string(argv[0]);
 
@@ -867,17 +863,8 @@ int main(int argc, char * argv[])
             continue;
         }
 	else
-        if (strncmp(*argv, "-i", 2) == 0 ||
-            strncmp(*argv, "/i", 2) == 0)
+        if (mainLoop.parseArg(*argv))
         {
-            string cmd = string(*argv);
-            size_t found = cmd.find('=', 2);
-            if (found != string::npos) {
-                string tmp = cmd.substr(found + 1);
-                if (!tmp.empty()) {
-                    numberOfIterations = (unsigned int)atoi(tmp.c_str());
-                }
-            }
             continue;
         }
         else
@@ -1097,39 +1084,11 @@ int main(int argc, char * argv[])
         MySystem(sysCmd, sysArgv);
     }
 
-    unsigned int i = 1;
-
-    while ((i <= numberOfIterations) || (numberOfIterations == 0))
+    mainLoop([&]()
     {
         if(!csv) cout << flush;
-        int delay_ms = int(delay * 1000);
-        int calibrated_delay_ms = delay_ms;
-#ifdef _MSC_VER
-        // compensate slow Windows console output
-        if(AfterTime) delay_ms -= (int)(m->getTickCount() - BeforeTime);
-        if(delay_ms < 0) delay_ms = 0;
-#else
-        // compensation of delay on Linux/UNIX
-        // to make the samling interval as monotone as possible
-        struct timeval start_ts, end_ts;
-        if(calibrated == 0) {
-            gettimeofday(&end_ts, NULL);
-            diff_usec = (end_ts.tv_sec-start_ts.tv_sec)*1000000.0+(end_ts.tv_usec-start_ts.tv_usec);
-            calibrated_delay_ms = delay_ms - diff_usec/1000.0;
-        }
-#endif
-
-        if (sysCmd == NULL || numberOfIterations != 0 || m->isBlocked() == false)
-        {
-            MySleepMs(calibrated_delay_ms);
-        }
 
-#ifndef _MSC_VER
-        calibrated = (calibrated + 1) % PCM_CALIBRATION_INTERVAL;
-        if(calibrated == 0) {
-            gettimeofday(&start_ts, NULL);
-        }
-#endif
+        calibratedSleep(delay, sysCmd, mainLoop, m);
 
         AfterTime = m->getTickCount();
         for(uint32 i=0; i<m->getNumSockets(); ++i)
@@ -1150,10 +1109,10 @@ int main(int argc, char * argv[])
 
         if ( m->isBlocked() ) {
         // in case PCM was blocked after spawning child application: break monitoring loop here
-            break;
+            return false;
         }
-        ++i;
-    }
+        return true;
+    });
 
     delete[] BeforeState;
     delete[] AfterState;

pcm-numa.cpp

@@ -41,7 +41,6 @@
 #include <vector>
 #define PCM_DELAY_DEFAULT 1.0       // in seconds
 #define PCM_DELAY_MIN 0.015         // 15 milliseconds is practical on most modern CPUs
-#define PCM_CALIBRATION_INTERVAL 50 // calibrate clock only every 50th iteration
 
 using namespace std;
 using namespace pcm;
@@ -57,6 +56,7 @@ void print_usage(const string progname)
     cerr << "  -h    | --help  | /h               => print this help and exit\n";
     cerr << "  -csv[=file.csv] | /csv[=file.csv]  => output compact CSV format to screen or\n"
          << "                                        to a file, in case filename is provided\n";
+    cerr << "  -i[=number] | /i[=number]          => allow to determine number of iterations\n";
     cerr << " Examples:\n";
     cerr << "  " << progname << " 1                  => print counters every second without core and socket output\n";
     cerr << "  " << progname << " 0.5 -csv=test.log  => twice a second save counter values to test.log in CSV format\n";
@@ -111,8 +111,7 @@ int main(int argc, char * argv[])
     char * sysCmd = NULL;
     char ** sysArgv = NULL;
     bool csv = false;
-    long diff_usec = 0;                            // deviation of clock is useconds between measurements
-    int calibrated = PCM_CALIBRATION_INTERVAL - 2; // keeps track is the clock calibration needed
+    MainLoop mainLoop;
     string program = string(argv[0]);
 
     PCM * m = PCM::getInstance();
@@ -142,6 +141,10 @@ int main(int argc, char * argv[])
                 }
                 continue;
             }
+            else if (mainLoop.parseArg(*argv))
+            {
+                continue;
+            }
             else if (strncmp(*argv, "--", 2) == 0)
             {
                 argv++;
@@ -256,36 +259,12 @@ int main(int argc, char * argv[])
         MySystem(sysCmd, sysArgv);
     }
 
-    while (1)
+    mainLoop([&]()
     {
         if (!csv) cout << flush;
-        int delay_ms = int(delay * 1000);
-        int calibrated_delay_ms = delay_ms;
-#ifdef _MSC_VER
-        // compensate slow Windows console output
-        if (AfterTime) delay_ms -= (int)(m->getTickCount() - BeforeTime);
-        if (delay_ms < 0) delay_ms = 0;
-#else
-        // compensation of delay on Linux/UNIX
-        // to make the samling interval as monotone as possible
-        struct timeval start_ts, end_ts;
-        if (calibrated == 0) {
-            gettimeofday(&end_ts, NULL);
-            diff_usec = (end_ts.tv_sec - start_ts.tv_sec) * 1000000.0 + (end_ts.tv_usec - start_ts.tv_usec);
-            calibrated_delay_ms = delay_ms - diff_usec / 1000.0;
-        }
-#endif
-        if (sysCmd == NULL || m->isBlocked() == false)
-        {
-            MySleepMs(calibrated_delay_ms);
-        }
 
-#ifndef _MSC_VER
-        calibrated = (calibrated + 1) % PCM_CALIBRATION_INTERVAL;
-        if (calibrated == 0) {
-            gettimeofday(&start_ts, NULL);
-        }
-#endif
+        calibratedSleep(delay, sysCmd, mainLoop, m);
+
         AfterTime = m->getTickCount();
         m->getAllCounterStates(SysAfterState, DummySocketStates, AfterState);
 
@@ -326,9 +305,10 @@ int main(int argc, char * argv[])
 
         if (m->isBlocked()) {
             // in case PCM was blocked after spawning child application: break monitoring loop here
-            break;
+            return false;
         }
-    }
+        return true;
+    });
 
     exit(EXIT_SUCCESS);
 }