LocalTimeStepWithEnclaveTasking.py

Go to the documentation of this file.

# This file is part of the ExaHyPE2 project. For conditions of distribution and
# use, please see the copyright notice at www.peano-framework.org
from exahype2.solvers.PDETerms import PDETerms
from exahype2.solvers.fv.EnclaveTasking import EnclaveTasking
 
import jinja2
 
from .kernels import create_abstract_solver_declarations
from .kernels import create_abstract_solver_definitions
from .kernels import create_solver_declarations
from .kernels import create_solver_definitions
from .kernels import create_rusanov_kernel_definitions
from .kernels import create_compute_Riemann_kernel_for_Rusanov_dsl
 
from .kernels import create_compute_time_step_size_kernel_for_local_time_stepping
from .kernels import create_compute_new_time_step_size_kernel_for_local_time_stepping
from .kernels import create_abstract_solver_user_declarations_for_local_time_stepping
from .kernels import create_abstract_solver_user_definitions_for_local_time_stepping
 
from .kernels import SolverVariant
from .kernels import KernelVariant
 
from exahype2.solvers.fv.kernels import create_halo_layer_construction_with_interpolation_for_reconstructed_patch
from exahype2.solvers.kernels import create_abstract_solver_user_declarations_for_fixed_time_stepping
from exahype2.solvers.fv.kernels import create_finish_time_step_implementation_for_local_time_stepping
from exahype2.solvers.fv.kernels import create_start_time_step_implementation_for_local_time_stepping
 
from .kernels import create_constructor_implementation_for_local_time_stepping
from .kernels import create_finish_time_step_implementation_for_local_time_stepping
 
import math
 
 
class LocalTimeStepWithEnclaveTasking(EnclaveTasking):
  def __init__(self,
    name, patch_size, unknowns, auxiliary_variables, min_volume_h, max_volume_h, time_step_relaxation,
    flux=PDETerms.User_Defined_Implementation,
    ncp=PDETerms.None_Implementation,
    eigenvalues=PDETerms.User_Defined_Implementation,
    boundary_conditions=PDETerms.User_Defined_Implementation,
    refinement_criterion=PDETerms.Empty_Implementation,
    initial_conditions=PDETerms.User_Defined_Implementation,
    source_term=PDETerms.None_Implementation,
    plot_grid_properties=False,
    interpolate_linearly_in_time=True,
    pde_terms_without_state=False,
    avoid_staircase_effect=True,
    discretisation_steps=8, overlap=1
  ):
    """
    time_step_size: Float
      This is the normalised time step size w.r.t. the coarsest admissible h value. If
      the code employs AMR on top of it and refines further, it will automatically
      downscale the time step size accordingly. So hand in a valid time step size w.r.t.
      to max_volume_h.
 
    avoid_staircase_effect: Boolean
      Please consult create_compute_new_time_step_size_kernel_for_local_time_stepping() for a
      discussion of this flag.
 
    discretisation_steps: Float
      This routine discretises (buckets) the time step sizes. You find more information
      in the C++ code in exahype2::TimeStepping.
    """
    self._interpolate_linearly_in_time = interpolate_linearly_in_time
 
    super(LocalTimeStepWithEnclaveTasking,self).__init__(name,
                                                         patch_size,
                                                         overlap,
                                                         unknowns,
                                                         auxiliary_variables,
                                                         min_volume_h,
                                                         max_volume_h,
                                                         plot_grid_properties,
                                                         pde_terms_without_state,
                                                         kernel_namespace="rusanov")
    self._time_step_relaxation = time_step_relaxation
 
    self._flux_implementation_flux_implementation                 = PDETerms.None_Implementation
    self._ncp_implementation_ncp_implementation                  = PDETerms.None_Implementation
    self._eigenvalues_implementation_eigenvalues_implementation          = PDETerms.None_Implementation
    self._source_term_implementation_source_term_implementation          = PDETerms.None_Implementation
 
    self._compute_eigenvalue_compute_eigenvalue                  = True
 
    self._compute_time_step_size_compute_time_step_size              = create_compute_time_step_size_kernel_for_local_time_stepping( name, time_step_relaxation )
    self._compute_new_time_step_size_compute_new_time_step_size          = create_compute_new_time_step_size_kernel_for_local_time_stepping(time_step_relaxation,avoid_staircase_effect,discretisation_steps)
 
    self.set_implementationset_implementation(flux=flux,
      ncp=ncp,
      eigenvalues=eigenvalues,
      boundary_conditions=boundary_conditions,
      refinement_criterion=refinement_criterion,
      initial_conditions=initial_conditions,
      source_term=source_term)
 
 
  def set_implementation(self,
    flux=None,ncp=None,
    eigenvalues=None,
    boundary_conditions=None,refinement_criterion=None,initial_conditions=None,source_term=None,
    memory_location         = None,
    use_split_loop          = False,
    additional_action_set_includes = "",
    additional_user_includes       = ""
  ):
    """
    If you pass in User_Defined, then the generator will create C++ stubs
    that you have to befill manually. If you pass in None_Implementation, it
    will create nop, i.e., no implementation or defaults. Any other string
    is copied 1:1 into the implementation. If you pass in None, then the
    set value so far won't be overwritten.
 
    Please note that not all options are supported by all solvers.
 
    This routine should be the very last invoked by the constructor.
    """
    if flux                 is not None:  self._flux_implementation_flux_implementation                       = flux
    if ncp                  is not None:  self._ncp_implementation_ncp_implementation                        = ncp
    if eigenvalues          is not None:  self._eigenvalues_implementation_eigenvalues_implementation                = eigenvalues
    if source_term          is not None:  self._source_term_implementation_source_term_implementation                = source_term
 
    create_rusanov_kernel_definitions()
 
    self._compute_kernel_call_compute_kernel_call = create_compute_Riemann_kernel_for_Rusanov_dsl(
      self._flux_implementation_flux_implementation,
      self._ncp_implementation_ncp_implementation,
      self._source_term_implementation_source_term_implementation,
      compute_max_eigenvalue_of_next_time_step = True,
      solver_variant = SolverVariant.WithVirtualFunctions,
      kernel_variant = KernelVariant.PatchWiseAoS
    )
 
    self._compute_kernel_call_stateless_compute_kernel_call_stateless = create_compute_Riemann_kernel_for_Rusanov_dsl(
      self._flux_implementation_flux_implementation,
      self._ncp_implementation_ncp_implementation,
      self._source_term_implementation_source_term_implementation,
      compute_max_eigenvalue_of_next_time_step = True,
      solver_variant = SolverVariant.Stateless,
      kernel_variant = KernelVariant.PatchWiseAoS
    )
 
    self._fused_compute_kernel_call_stateless_cpu_fused_compute_kernel_call_stateless_cpu = create_compute_Riemann_kernel_for_Rusanov_dsl(
      self._flux_implementation_flux_implementation,
      self._ncp_implementation_ncp_implementation,
      self._source_term_implementation_source_term_implementation,
      compute_max_eigenvalue_of_next_time_step = True,
      solver_variant = SolverVariant.Stateless,
      kernel_variant = KernelVariant.PatchWiseAoS
    )
 
    self._fused_compute_kernel_call_stateless_gpu_fused_compute_kernel_call_stateless_gpu = create_compute_Riemann_kernel_for_Rusanov_dsl(
      self._flux_implementation_flux_implementation,
      self._ncp_implementation_ncp_implementation,
      self._source_term_implementation_source_term_implementation,
      compute_max_eigenvalue_of_next_time_step = True,
      solver_variant = SolverVariant.Accelerator,
      kernel_variant = KernelVariant.PatchWiseAoS
    )
 
    self._abstract_solver_user_declarations_abstract_solver_user_declarations  = create_abstract_solver_declarations(self._flux_implementation_flux_implementation, self._ncp_implementation_ncp_implementation, self._eigenvalues_implementation_eigenvalues_implementation, self._source_term_implementation_source_term_implementation, self._pde_terms_without_state)
    self._abstract_solver_user_definitions_abstract_solver_user_definitions   = create_abstract_solver_definitions(self._flux_implementation_flux_implementation, self._ncp_implementation_ncp_implementation, self._eigenvalues_implementation_eigenvalues_implementation, self._source_term_implementation_source_term_implementation, self._pde_terms_without_state)
 
    self._constructor_implementation_constructor_implementation         = create_constructor_implementation_for_local_time_stepping()
 
    self._abstract_solver_user_declarations_abstract_solver_user_declarations += create_abstract_solver_user_declarations_for_local_time_stepping()
    self._abstract_solver_user_definitions_abstract_solver_user_definitions  += create_abstract_solver_user_definitions_for_local_time_stepping()
 
    self._solver_user_declarations_solver_user_declarations           = create_solver_declarations(self._flux_implementation_flux_implementation, self._ncp_implementation_ncp_implementation, self._eigenvalues_implementation_eigenvalues_implementation, self._source_term_implementation_source_term_implementation, self._pde_terms_without_state)
    self._solver_user_definitions_solver_user_definitions            = create_solver_definitions(self._flux_implementation_flux_implementation, self._ncp_implementation_ncp_implementation, self._eigenvalues_implementation_eigenvalues_implementation, self._source_term_implementation_source_term_implementation, self._pde_terms_without_state)
 
    self._start_time_step_implementation_start_time_step_implementation     = create_start_time_step_implementation_for_local_time_stepping(True)
    self._finish_time_step_implementation_finish_time_step_implementation    = create_finish_time_step_implementation_for_local_time_stepping()
 
    super(LocalTimeStepWithEnclaveTasking,self).set_implementation(boundary_conditions, refinement_criterion, initial_conditions, memory_location, use_split_loop, additional_action_set_includes, additional_user_includes)
 
 
  def create_action_sets(self):
    """
    The actual action sets all are created by the superclass. So nothing
    is to be done here. But we want to reset the actual updates and
    projection, and these only happen if we are allowed to update
    indeed.
    """
    super(LocalTimeStepWithEnclaveTasking, self).create_action_sets()
 
    update_cell_guard  = "::exahype2::runTimeStepOnCell( fineGridCell" + self._name_name + "CellLabel, fineGridFaces" + self._name_name + "FaceLabel, repositories::getMinTimeStamp())"
    updated_cell_guard = "fineGridCell" + self._name_name + "CellLabel.getSemaphoreNumber()!=::exahype2::EnclaveBookkeeping::NoEnclaveTaskNumber"
    self._action_set_update_cell_action_set_update_cell.guard                += " and " + update_cell_guard
    self._action_set_merge_enclave_task_outcome.guard  = updated_cell_guard
 
    if self._interpolate_linearly_in_time:
      self._action_set_update_cell_action_set_update_cell._Template_TouchCellFirstTime_Fill_Halos = create_halo_layer_construction_with_interpolation_for_reconstructed_patch(self._name_name)
 
    self._action_set_AMR.event_lifetime = 2*3*int(self._max_volume_h/self._min_volume_h+1)
 
 
  @property
  def user_action_set_includes(self):
    return super(LocalTimeStepWithEnclaveTasking, self).user_action_set_includes + """
#include "exahype2/TimeStepping.h"
#include "kernels/rusanov.h"
"""
 
 
  def create_data_structures(self):
    super(LocalTimeStepWithEnclaveTasking, self).create_data_structures()
 
    self._patch_overlap_old.generator.send_condition               = self._patch_overlap_new.generator.send_condition
    self._patch_overlap_old.generator.receive_and_merge_condition  = self._patch_overlap_new.generator.receive_and_merge_condition