Experiment

Author: Meinersbur

offload/DeviceRTL/src/Kernel.cpp

@@ -82,7 +82,7 @@ extern "C" {
 int32_t __kmpc_target_init(KernelEnvironmentTy &KernelEnvironment,
                            KernelLaunchEnvironmentTy &KernelLaunchEnvironment) {
   ConfigurationEnvironmentTy &Configuration = KernelEnvironment.Configuration;
-  bool IsSPMD = Configuration.ExecMode & OMP_TGT_EXEC_MODE_SPMD;
+  bool IsSPMD = true;
   bool UseGenericStateMachine = Configuration.UseGenericStateMachine;
   if (IsSPMD) {
     inititializeRuntime(/*IsSPMD=*/true, KernelEnvironment,

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

@@ -633,7 +633,7 @@ uint32_t GenericKernelTy::getNumThreads(GenericDeviceTy &GenericDevice,
   assert(ThreadLimitClause[1] == 1 && ThreadLimitClause[2] == 1 &&
          "Multi dimensional launch not supported yet.");
 
-  if (ThreadLimitClause[0] > 0 && isGenericMode())
+  if (ThreadLimitClause[0] > 0 && false)
     ThreadLimitClause[0] += GenericDevice.getWarpSize();
 
   return std::min(MaxNumThreads, (ThreadLimitClause[0] > 0)