Add spmd constraint and address comments

Author: Kewen12

offload/plugins-nextgen/amdgpu/src/rtl.cpp

@@ -884,6 +884,11 @@ struct AMDGPUKernelTy : public GenericKernelTy {
   /// Indicates whether or not we need to set up our own private segment size.
   bool usesDynamicStack() const { return DynamicStack; }
 
+  /// Get the execution mode of this kernel.
+  OMPTgtExecModeFlags getExecutionMode() const {
+    return getExecutionModeFlags();
+  }
+
   /// Envar to disable host-exec thread creation.
   BoolEnvar OMPX_DisableHostExec;
 
@@ -2186,7 +2191,8 @@ struct AMDGPUStreamTy {
 
     // If runtime autotuning is enabled, setup the callback functions to process
     // the data after kernel completed.
-    if (Device.enableRuntimeAutotuning()) {
+    if (Device.enableRuntimeAutotuning() &&
+        Kernel.getExecutionMode() == OMP_TGT_EXEC_MODE_SPMD) {
       std::string KernelName(Kernel.getName());
       KernelRunRecord *KernelRecords = Device.getKernelRunRecords();
 

offload/plugins-nextgen/common/include/PluginInterface.h

@@ -1305,7 +1305,7 @@ struct KernelRunRecord {
     uint32_t IdxThread = 0;
     uint32_t IdxCUMultiplier = 0;
     // Run counters.
-    uint32_t RunCounters;
+    uint32_t RunCounters = 0;
     // Entry with minimum running time.
     KernelRunEntry MinEntries;
   };
@@ -1363,10 +1363,19 @@ struct KernelRunRecord {
   }
 
   bool reachedRunLimitForKernel(std::string KernelName) {
+    if (TuningData.count(KernelName) == 0) {
+      // If no record for this kernel.
+      return false;
+    }
+
     return TuningData[KernelName].RunCounters > RunLimiter;
   }
 
   uint32_t getRunCounterForKernel(std::string KernelName) {
+    if (TuningData.count(KernelName) == 0) {
+      return 0;
+    }
+
     return TuningData[KernelName].RunCounters;
   }
 

offload/plugins-nextgen/common/src/PluginInterface.cpp

@@ -734,7 +734,8 @@ Error GenericKernelTy::launch(GenericDeviceTy &GenericDevice, void **ArgPtrs,
     KernelRunCounter = KernelRecord->getRunCounterForKernel(KernelName);
   }
   // If Autotuning is enabled and the kernel is not launched for the first time.
-  if (GenericDevice.enableRuntimeAutotuning() && KernelRunCounter > 0) {
+  if (GenericDevice.enableRuntimeAutotuning() && isSPMDMode() &&
+      KernelRunCounter > 0) {
     assert(KernelRecord &&
            "Autotuning is enabled, but KernelRunRecord is not initialized!");