7. Running MirgeCOM

• 7.1. Running with large numbers of ranks and nodes
  • 7.1.1. Avoiding the startup overhead of Python
  • 7.1.2. Avoiding errors and overheads due to caching of kernels
• 7.2. OpenCL device selection
• 7.3. Profiling and logging
  • 7.3.1. Profiling memory consumption with memray
    • 7.3.1.1. Common issues
  • 7.3.2. Profiling kernel execution
    • PyOpenCLProfilingArrayContext
      • PyOpenCLProfilingArrayContext.tabulate_profiling_data()
      • PyOpenCLProfilingArrayContext.call_loopy()
      • PyOpenCLProfilingArrayContext.get_profiling_data_for_kernel()
      • PyOpenCLProfilingArrayContext.reset_profiling_data_for_kernel()
    • SingleCallKernelProfile
    • MultiCallKernelProfile
  • 7.3.3. Time series logging
    • StateConsumer
      • StateConsumer.__init__()
      • StateConsumer.set_state_vars()
    • DiscretizationBasedQuantity
    • KernelProfile
    • PythonMemoryUsage
    • DeviceMemoryUsage
    • initialize_logmgr()
    • logmgr_add_cl_device_info()
    • logmgr_add_device_memory_usage()
    • logmgr_add_many_discretization_quantities()
    • logmgr_add_mempool_usage()
    • add_package_versions()
    • set_sim_state()
    • logmgr_set_time()
• 7.4. Scripting launches
• 7.5. Running on specific systems
  • 7.5.1. General
  • 7.5.2. Quartz
  • 7.5.3. Lassen
• 7.6. OpenCL kernel caching
  • 7.6.1. Loopy
  • 7.6.2. PyOpenCL
  • 7.6.3. PoCL
  • 7.6.4. CUDA