Print Memory Peak Working Set Usage (#168)

* print peak working set usage

* pass in optional second value for parameter -SaveTensorData and fix test cases to compare csv values correctly

* refactor working set print

* add extra line

Author: yskim1501

Tools/WinMLRunner/src/OutputHelper.h

@@ -200,20 +200,24 @@ class OutputHelper
         double firstLoadWorkingSetMemoryUsage = profiler[LOAD_MODEL].GetAverage(CounterType::WORKING_SET_USAGE);
         double firstLoadSharedMemoryUsage = profiler[LOAD_MODEL].GetAverage(CounterType::GPU_SHARED_MEM_USAGE);
         double firstLoadDedicatedMemoryUsage = profiler[LOAD_MODEL].GetAverage(CounterType::GPU_DEDICATED_MEM_USAGE);
+        double firstLoadPeakWorkingSetUsage = profiler[LOAD_MODEL].GetAverage(CounterType::PEAK_WORKING_SET_USAGE);
 
         double firstSessionCreationWorkingSetMemoryUsage = profiler[CREATE_SESSION].GetAverage(CounterType::WORKING_SET_USAGE);
         double firstSessionCreationSharedMemoryUsage = profiler[CREATE_SESSION].GetAverage(CounterType::GPU_SHARED_MEM_USAGE);
         double firstSessionCreationDedicatedMemoryUsage = profiler[CREATE_SESSION].GetAverage(CounterType::GPU_DEDICATED_MEM_USAGE);
+        double firstSessionPeakWorkingSetUsage = profiler[CREATE_SESSION].GetAverage(CounterType::PEAK_WORKING_SET_USAGE);
 
         double averageBindMemoryUsage = profiler[BIND_VALUE].GetAverage(CounterType::WORKING_SET_USAGE);
         double minBindMemoryUsage = profiler[BIND_VALUE].GetMin(CounterType::WORKING_SET_USAGE);
         double maxBindMemoryUsage = profiler[BIND_VALUE].GetMax(CounterType::WORKING_SET_USAGE);
         double firstBindMemoryUsage = profiler[BIND_VALUE_FIRST_RUN].GetAverage(CounterType::WORKING_SET_USAGE);
+        double firstBindPeakMemoryUsage = profiler[BIND_VALUE_FIRST_RUN].GetAverage(CounterType::PEAK_WORKING_SET_USAGE);
 
         double averageEvalMemoryUsage = profiler[EVAL_MODEL].GetAverage(CounterType::WORKING_SET_USAGE);
         double minEvalMemoryUsage = profiler[EVAL_MODEL].GetMin(CounterType::WORKING_SET_USAGE);
         double maxEvalMemoryUsage = profiler[EVAL_MODEL].GetMax(CounterType::WORKING_SET_USAGE);
         double firstEvalMemoryUsage = profiler[EVAL_MODEL_FIRST_RUN].GetAverage(CounterType::WORKING_SET_USAGE);
+        double firstEvalPeakMemoryUsage = profiler[EVAL_MODEL_FIRST_RUN].GetAverage(CounterType::PEAK_WORKING_SET_USAGE);
 
         double averageBindDedicatedMemoryUsage = profiler[BIND_VALUE].GetAverage(CounterType::GPU_DEDICATED_MEM_USAGE);
         double minBindDedicatedMemoryUsage = profiler[BIND_VALUE].GetMin(CounterType::GPU_DEDICATED_MEM_USAGE);
@@ -253,6 +257,8 @@ class OutputHelper
             profiler[BIND_VALUE_FIRST_RUN].GetAverage(CounterType::GPU_DEDICATED_MEM_USAGE) +
             profiler[EVAL_MODEL_FIRST_RUN].GetAverage(CounterType::GPU_DEDICATED_MEM_USAGE);
 
+        double firstIterationPeakWorkingSet = firstLoadPeakWorkingSetUsage + firstSessionPeakWorkingSetUsage + firstBindPeakMemoryUsage + firstEvalPeakMemoryUsage;
+
         printf("\nResults (device = %s, numIterations = %d, inputBinding = %s, inputDataType = %s, deviceCreationLocation = %s):\n",
             TypeHelper::Stringify(deviceType).c_str(),
             numIterations,
@@ -280,6 +286,17 @@ class OutputHelper
         std::cout << "  Working Set Memory usage (evaluate): " << firstEvalMemoryUsage << " MB" << std::endl;
         std::cout << "  Working Set Memory usage (load, bind, session creation, and evaluate): " << firstIterationWorkingSetMemoryUsage << " MB" << std::endl;
 
+        if (isPerformanceConsoleOutputVerbose)
+        {
+            std::cout << std::endl;
+            std::cout << "  Peak Working Set Memory Difference (from start to load): " << firstLoadPeakWorkingSetUsage << " MB" << std::endl;
+            std::cout << "  Peak Working Set Memory Difference (from model load to session creation): " << firstSessionPeakWorkingSetUsage << " MB" << std::endl;
+            std::cout << "  Peak Working Set Memory Difference (from session to bind): " << firstBindPeakMemoryUsage << " MB" << std::endl;
+            std::cout << "  Peak Working Set Memory Difference (from bind to evaluate): " << firstEvalPeakMemoryUsage << " MB" << std::endl;
+        }
+
+        std::cout << "  Peak Working Set Memory Difference (load, bind, session creation, and evaluate): " << firstIterationPeakWorkingSet << " MB" << std::endl;
+
         if (isPerformanceConsoleOutputVerbose)
         {
             std::cout << "\n  Dedicated Memory usage (load): " << firstLoadDedicatedMemoryUsage << " MB" << std::endl;

Tools/WinMLRunner/src/TimerHelper.h

@@ -129,7 +129,8 @@ class CpuPerfCounter
 
 private:
 
-    bool GetProcessMemoryCounters(DWORD pid, ULONG& pageFaultCount, SIZE_T& pageFileUsage, SIZE_T& peakPageFileUsage, SIZE_T& workingSetSize, SIZE_T& peakWorkingSetSize)
+    bool GetProcessMemoryCounters(DWORD pid, ULONG& pageFaultCount, SIZE_T& pageFileUsage, SIZE_T& peakPageFileUsage,
+        SIZE_T& workingSetSize, SIZE_T& peakWorkingSetSize)
     {
         HANDLE hProcess = NULL;