accelerator.h

Go to the documentation of this file.

// This file is part of the Peano project. For conditions of distribution and
// use, please see the copyright notice at www.peano-framework.org
#pragma once
 
#include "Device.h"
 
#include "tarch/multicore/multicore.h"
#include "tarch/compiler/CompilerSpecificSettings.h"
 
#if !defined(GPUOffloadingOMP) and !defined(GPUOffloadingSYCL) and !defined(GPUOffloadingCPP)
#define GPUOffloadingOff
#endif
 
#include "tarch/accelerator/sycl/Device.h"
 
namespace tarch {
  #if defined(GPUOffloadingHIP)
    #define GPUCallableMethod __host__ __device__
  #elif defined(GPUOffloadingSYCL) or defined(SharedSYCL)
    #define GPUCallableMethod SYCL_EXTERNAL
  #elif defined(GPUOffloadingCPP) && defined(_NVHPC_STDPAR_GPU)
    #define GPUCallableMethod __host__ __device__
  #else
    #define GPUCallableMethod
  #endif
 
  #if defined(GPUOffloadingHIP)
    #define GPUCallableInlineMethod __host__ __device__ __forceinline__
  #elif defined(GPUOffloadingSYCL) or defined(SharedSYCL)
    #define GPUCallableInlineMethod inline
  #elif defined(GPUOffloadingCPP) && defined(_NVHPC_STDPAR_GPU)
    #define GPUCallableInlineMethod __host__ __device__ __forceinline__
  #else
    #define GPUCallableInlineMethod //inline
  #endif
 
#if defined(GPUOffloadingOMP)
#pragma omp declare target
  void gpuAbort();
#pragma omp end declare target
#elif defined(GPUOffloadingSYCL) or defined(SharedSYCL)
  GPUCallableMethod void gpuAbort();
#elif defined(GPUOffloadingCPP)
  GPUCallableMethod void gpuAbort();
#else
  void gpuAbort();
#endif
 
  enum class MemoryLocation {
    Heap,
    ManagedSharedAcceleratorDeviceMemory
  };
 
  std::string toString(MemoryLocation value);
 
  namespace internal {
    void* allocateRawData(
      std::size_t    size,
      MemoryLocation location,
      int            device
    );
  } // namespace internal
 
  template <class T = double>
  T* allocateMemory(
    std::size_t count,
    MemoryLocation location,
    int device = accelerator::Device::HostDevice
) {
    return reinterpret_cast<T*>(internal::allocateRawData(
      count * sizeof(T),
      location,
      device
    ));
  }
 
  void freeMemory(
    void* data,
    MemoryLocation location,
    int device = accelerator::Device::HostDevice
  );
 
  std::size_t padSizeToAlignment(std::size_t size, std::size_t alignment);
 
  #if defined(GPUOffloadingOMP)
  #pragma omp declare target
  #endif
  double* memset(double* dest, double ch, size_t byteCount);
  #if defined(GPUOffloadingOMP)
  #pragma omp end declare target
  #endif
} // namespace tarch