core/RunCommand.h

// -*- tab-width: 2; indent-tabs-mode: nil; coding: utf-8-with-signature -*-
//-----------------------------------------------------------------------------
// Copyright 2000-2025 CEA (www.cea.fr) IFPEN (www.ifpenergiesnouvelles.com)
// See the top-level COPYRIGHT file for details.
// SPDX-License-Identifier: Apache-2.0
//-----------------------------------------------------------------------------
/*---------------------------------------------------------------------------*/
/* RunCommand.h                                                (C) 2000-2025 */
/*                                                                           */
/* Gestion d'une commande sur accélérateur.                                  */
/*---------------------------------------------------------------------------*/
#ifndef ARCANE_ACCELERATOR_CORE_RUNCOMMAND_H
#define ARCANE_ACCELERATOR_CORE_RUNCOMMAND_H
/*---------------------------------------------------------------------------*/
/*---------------------------------------------------------------------------*/
 
#include "arcane/accelerator/core/AcceleratorCoreGlobal.h"
 
/*---------------------------------------------------------------------------*/
/*---------------------------------------------------------------------------*/
 
namespace Arcane::Accelerator
{
namespace impl
{
extern "C++" ARCANE_ACCELERATOR_CORE_EXPORT IReduceMemoryImpl*
internalGetOrCreateReduceMemoryImpl(RunCommand* command);
}
 
/*---------------------------------------------------------------------------*/
/*---------------------------------------------------------------------------*/
class ARCANE_ACCELERATOR_CORE_EXPORT RunCommand
{
  friend impl::IReduceMemoryImpl* impl::internalGetOrCreateReduceMemoryImpl(RunCommand* command);
  friend impl::RunCommandLaunchInfo;
  friend impl::RunQueueImpl;
  friend class ViewBuildInfo;
  template<typename T> friend class RunCommandLocalMemory;
 
  friend RunCommand makeCommand(const RunQueue& run_queue);
  friend RunCommand makeCommand(const RunQueue* run_queue);
 
 public:
 
  ~RunCommand();
 
 protected:
 
  explicit RunCommand(const RunQueue& run_queue);
 
 public:
 
  RunCommand(RunCommand&& command) = delete;
  RunCommand(const RunCommand&) = delete;
  RunCommand& operator=(const RunCommand&) = delete;
  RunCommand& operator=(RunCommand&&) = delete;
 
 public:

  RunCommand& addTraceInfo(const TraceInfo& ti);

  RunCommand& addKernelName(const String& v);

  RunCommand& addNbThreadPerBlock(Int32 v);

  const TraceInfo& traceInfo() const;

  const String& kernelName() const;
 
  /*
   * \brief Nombre de threads par bloc ou 0 pour la valeur par défaut.
   *
   * Cette valeur est utilisée uniquement si on s'exécute sur accélérateur.
   */
  Int32 nbThreadPerBlock() const;

  void setParallelLoopOptions(const ParallelLoopOptions& opt);

  const ParallelLoopOptions& parallelLoopOptions() const;

  friend ARCANE_ACCELERATOR_CORE_EXPORT RunCommand&
  operator<<(RunCommand& command, const TraceInfo& trace_info);
 
 private:
 
  // Pour RunCommandLaunchInfo
  void _internalNotifyBeginLaunchKernel();
  void _internalNotifyEndLaunchKernel();
  void _internalNotifyBeginLaunchKernelSyclEvent(void* sycl_event_ptr);
  ForLoopOneExecStat* _internalCommandExecStat();
 
 private:

  impl::RunQueueImpl* _internalQueueImpl() const;
  impl::NativeStream _internalNativeStream() const;
  static impl::RunCommandImpl* _internalCreateImpl(impl::RunQueueImpl* queue);
  static void _internalDestroyImpl(impl::RunCommandImpl* p);
  void _addSharedMemory(Int32 size);
  Int32 _sharedMemory() const;
 
 private:
 
  void _allocateReduceMemory(Int32 nb_grid);
 
 private:
 
  impl::RunCommandImpl* m_p;
};
 
/*---------------------------------------------------------------------------*/
/*---------------------------------------------------------------------------*/
 
} // End namespace Arcane::Accelerator
 
/*---------------------------------------------------------------------------*/
/*---------------------------------------------------------------------------*/
 
#endif