Currently our LaunchConfig only covers bare minimum (functionally equivalent to what cuLaunchKernel can do). However, the purpose of moving to use cuLaunchKernelEx was to be able to use advanced features, and supporting thread-block clusters is one goal.

xref:

• https://docs.nvidia.com/cuda/cuda-driver-api/unionCUlaunchAttributeValue.html#unionCUlaunchAttributeValue
• CU_LAUNCH_ATTRIBUTE_CLUSTER_DIMENSION
• https://docs.nvidia.com/cuda/cuda-c-programming-guide/#thread-block-clusters