Adding necessary if-statement to avoid automatic model loading

Author: ChSonnabend

GPU/GPUTracking/Definitions/GPUSettingsList.h

@@ -229,7 +229,7 @@ AddOption(nnInferenceDeviceId, unsigned int, 0, "", 0, "(unsigned int) Specify i
 AddOption(nnInferenceAllocateDevMem, int, 0, "", 0, "(bool, default = 0), if the device memory should be allocated for inference")
 AddOption(nnInferenceDtype, std::string, "fp32", "", 0, "(std::string) Specify the datatype for which inference is performed (fp32: default, fp16)") // fp32 or fp16
 AddOption(nnInferenceThreadsPerNN, int, 0, "", 0, "Number of threads used to evaluate one neural network")
-AddOption(nnInferenceEnableOrtOptimization, unsigned int, 1, "", 0, "Enables graph optimizations in ONNX Runtime. Can be greater than 1!")
+AddOption(nnInferenceEnableOrtOptimization, unsigned int, 99, "", 0, "Enables graph optimizations in ONNX Runtime. Can be [0, 1, 2, 99] -> see https://github.com/microsoft/onnxruntime/blob/3f71d637a83dc3540753a8bb06740f67e926dc13/include/onnxruntime/core/session/onnxruntime_c_api.h#L347")
 AddOption(nnInferenceOrtProfiling, int, 0, "", 0, "Enables profiling of model execution in ONNX Runtime")
 AddOption(nnInferenceOrtProfilingPath, std::string, ".", "", 0, "If nnInferenceOrtProfiling is set, the path to store the profiling data")
 AddOption(nnInferenceVerbosity, int, 1, "", 0, "0: No messages; 1: Warnings; 2: Warnings + major debugs; >3: All debugs")

GPU/GPUTracking/Global/GPUChainTrackingClusterizer.cxx

@@ -613,32 +613,34 @@ int32_t GPUChainTracking::RunTPCClusterizer(bool synchronizeOutput)
   }
 
 #ifdef GPUCA_HAS_ONNX
-  uint32_t maxClusters = -1;
-  for (uint32_t iSector = 0; iSector < NSECTORS; iSector++) {
-    maxClusters = std::max(maxClusters, processors()->tpcClusterer[iSector].mNMaxClusters);
-  }
-  for (uint32_t iSector = 0; iSector < NSECTORS; iSector++) {
-    GPUTPCNNClusterizer& clustererNN = processors()->tpcNNClusterer[iSector];
-    const GPUSettingsProcessingNNclusterizer& nn_settings = GetProcessingSettings().nn;
-    clustererNN.nnClusterizerUseCfRegression = nn_settings.nnClusterizerUseCfRegression;
-    clustererNN.nnClusterizerSizeInputRow = nn_settings.nnClusterizerSizeInputRow;
-    clustererNN.nnClusterizerSizeInputPad = nn_settings.nnClusterizerSizeInputPad;
-    clustererNN.nnClusterizerSizeInputTime = nn_settings.nnClusterizerSizeInputTime;
-    clustererNN.nnClusterizerAddIndexData = nn_settings.nnClusterizerAddIndexData;
-    clustererNN.nnClusterizerElementSize = ((2 * nn_settings.nnClusterizerSizeInputRow + 1) * (2 * nn_settings.nnClusterizerSizeInputPad + 1) * (2 * nn_settings.nnClusterizerSizeInputTime + 1)) + (nn_settings.nnClusterizerAddIndexData ? 3 : 0);
-    clustererNN.nnClusterizerBatchedMode = nn_settings.nnClusterizerBatchedMode;
-    clustererNN.nnClusterizerBoundaryFillValue = nn_settings.nnClusterizerBoundaryFillValue;
-    clustererNN.nnClusterizerTotalClusters = maxClusters;
-    clustererNN.nnClassThreshold = nn_settings.nnClassThreshold;
-    clustererNN.nnSigmoidTrafoClassThreshold = nn_settings.nnSigmoidTrafoClassThreshold;
-    if (nn_settings.nnClusterizerVerbosity < 0) {
-      clustererNN.nnClusterizerVerbosity = nn_settings.nnInferenceVerbosity;
-    } else {
-      clustererNN.nnClusterizerVerbosity = nn_settings.nnClusterizerVerbosity;
+  if (GetProcessingSettings().nn.applyNNclusterizer) {
+    uint32_t maxClusters = -1;
+    for (uint32_t iSector = 0; iSector < NSECTORS; iSector++) {
+      maxClusters = std::max(maxClusters, processors()->tpcClusterer[iSector].mNMaxClusters);
+    }
+    for (uint32_t iSector = 0; iSector < NSECTORS; iSector++) {
+      GPUTPCNNClusterizer& clustererNN = processors()->tpcNNClusterer[iSector];
+      const GPUSettingsProcessingNNclusterizer& nn_settings = GetProcessingSettings().nn;
+      clustererNN.nnClusterizerUseCfRegression = nn_settings.nnClusterizerUseCfRegression;
+      clustererNN.nnClusterizerSizeInputRow = nn_settings.nnClusterizerSizeInputRow;
+      clustererNN.nnClusterizerSizeInputPad = nn_settings.nnClusterizerSizeInputPad;
+      clustererNN.nnClusterizerSizeInputTime = nn_settings.nnClusterizerSizeInputTime;
+      clustererNN.nnClusterizerAddIndexData = nn_settings.nnClusterizerAddIndexData;
+      clustererNN.nnClusterizerElementSize = ((2 * nn_settings.nnClusterizerSizeInputRow + 1) * (2 * nn_settings.nnClusterizerSizeInputPad + 1) * (2 * nn_settings.nnClusterizerSizeInputTime + 1)) + (nn_settings.nnClusterizerAddIndexData ? 3 : 0);
+      clustererNN.nnClusterizerBatchedMode = nn_settings.nnClusterizerBatchedMode;
+      clustererNN.nnClusterizerBoundaryFillValue = nn_settings.nnClusterizerBoundaryFillValue;
+      clustererNN.nnClusterizerTotalClusters = maxClusters;
+      clustererNN.nnClassThreshold = nn_settings.nnClassThreshold;
+      clustererNN.nnSigmoidTrafoClassThreshold = nn_settings.nnSigmoidTrafoClassThreshold;
+      if (nn_settings.nnClusterizerVerbosity < 0) {
+        clustererNN.nnClusterizerVerbosity = nn_settings.nnInferenceVerbosity;
+      } else {
+        clustererNN.nnClusterizerVerbosity = nn_settings.nnClusterizerVerbosity;
+      }
+      clustererNN.nnClusterizerDtype = nn_settings.nnInferenceDtype.find("32") != std::string::npos;
+      GPUTPCNNClusterizerHost nnApplication(nn_settings, clustererNN);
+      AllocateRegisteredMemory(clustererNN.mMemoryId);
     }
-    clustererNN.nnClusterizerDtype = nn_settings.nnInferenceDtype.find("32") != std::string::npos;
-    GPUTPCNNClusterizerHost nnApplication(nn_settings, clustererNN);
-    AllocateRegisteredMemory(clustererNN.mMemoryId);
   }
 #endif