#ifndef HSA_RUNTIME_EXT_IMAGE_UTIL_H
#define HSA_RUNTIME_EXT_IMAGE_UTIL_H

#include <assert.h>
#include <stdint.h>

#include "inc/hsa.h"

// A macro to disallow the copy and move constructor and operator= functions
// This should be used in the private: declarations for a class
#define DISALLOW_COPY_AND_ASSIGN(TypeName) \
  TypeName(const TypeName&);               \
  TypeName(TypeName&&);                    \
  void operator=(const TypeName&);         \
  void operator=(TypeName&&);

#if defined(_MSC_VER)
#define ALIGNED_(x) __declspec(align(x))
#else
#if defined(__GNUC__)
#define ALIGNED_(x) __attribute__ ((aligned(x)))
#endif // __GNUC__
#endif // _MSC_VER

#define MULTILINE(...) # __VA_ARGS__

#if defined(__GNUC__)
#include "mm_malloc.h"
#if defined(__i386__) || defined(__x86_64__)
#include <x86intrin.h>
#else
#error \
    "Processor not identified.  " \
          "Need to provide a lightweight approximate clock interface (aka __rdtsc())."
#endif

namespace ext_image {
#define __forceinline __inline__ __attribute__((always_inline))
static __forceinline void __debugbreak() { __builtin_trap(); }
#define __declspec(x) __attribute__((x))
#undef __stdcall
#define __stdcall  // __attribute__((__stdcall__))
#define __ALIGNED__(x) __attribute__((aligned(x)))

static __forceinline void* _aligned_malloc(size_t size, size_t alignment) {
  return _mm_malloc(size, alignment);
}
static __forceinline void _aligned_free(void* ptr) { return _mm_free(ptr); }
#elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_X64))
#include "intrin.h"
#define __ALIGNED__(x) __declspec(align(x))
namespace ext_image {
#else
#error "Compiler and/or processor not identified."
#endif

/// @brief: Checks if a value is power of two, if it is, return true. Be careful
/// when passing 0.
/// @param: val(Input), the data to be checked.
/// @return: bool.
template <typename T>
static __forceinline bool IsPowerOfTwo(T val) {
  return (val & (val - 1)) == 0;
}

/// @brief: Calculates the floor value aligned based on parameter of alignment.
/// If value is at the boundary of alignment, it is unchanged.
/// @param: value(Input), value to be calculated.
/// @param: alignment(Input), alignment value.
/// @return: T.
template <typename T>
static __forceinline T AlignDown(T value, size_t alignment) {
  assert(IsPowerOfTwo(alignment));
  return (T)(value & ~(alignment - 1));
}

/// @brief: Same as previous one, but first parameter becomes pointer, for more
/// info, see the previous desciption.
/// @param: value(Input), pointer to type T.
/// @param: alignment(Input), alignment value.
/// @return: T*, pointer to type T.
template <typename T>
static __forceinline T* AlignDown(T* value, size_t alignment) {
  return (T*)AlignDown((intptr_t)value, alignment);
}

/// @brief: Calculates the ceiling value aligned based on parameter of
/// alignment.
/// If value is at the boundary of alignment, it is unchanged.
/// @param: value(Input), value to be calculated.
/// @param: alignment(Input), alignment value.
/// @param: T.
template <typename T>
static __forceinline T AlignUp(T value, size_t alignment) {
  return AlignDown((T)(value + alignment - 1), alignment);
}

/// @brief: Same as previous one, but first parameter becomes pointer, for more
/// info, see the previous desciption.
/// @param: value(Input), pointer to type T.
/// @param: alignment(Input), alignment value.
/// @return: T*, pointer to type T.
template <typename T>
static __forceinline T* AlignUp(T* value, size_t alignment) {
  return (T*)AlignDown((intptr_t)((uint8_t*)value + alignment - 1), alignment);
}

/// @brief: Checks if the input value is at the boundary of alignment, if it is,
/// @return true.
/// @param: value(Input), value to be checked.
/// @param: alignment(Input), alignment value.
/// @return: bool.
template <typename T>
static __forceinline bool IsMultipleOf(T value, size_t alignment) {
  return (AlignUp(value, alignment) == value);
}

/// @brief: Same as previous one, but first parameter becomes pointer, for more
/// info, see the previous desciption.
/// @param: value(Input), pointer to type T.
/// @param: alignment(Input), alignment value.
/// @return: bool.
template <typename T>
static __forceinline bool IsMultipleOf(T* value, size_t alignment) {
  return (AlignUp(value, alignment) == value);
}

static __forceinline uint32_t NextPow2(uint32_t value) {
  if (value == 0) return 1;
  uint32_t v = value - 1;
  v |= v >> 1;
  v |= v >> 2;
  v |= v >> 4;
  v |= v >> 8;
  v |= v >> 16;
  return v + 1;
}

static __forceinline uint64_t NextPow2(uint64_t value) {
  if (value == 0) return 1;
  uint64_t v = value - 1;
  v |= v >> 1;
  v |= v >> 2;
  v |= v >> 4;
  v |= v >> 8;
  v |= v >> 16;
  v |= v >> 32;
  return v + 1;
}

template<uint32_t lowBit, uint32_t highBit, typename T>
static __forceinline uint32_t BitSelect(T p) {
  static_assert(sizeof(T) <= sizeof(uintptr_t), "Type out of range.");
  static_assert(highBit < sizeof(uintptr_t)*8, "Bit index out of range.");

  uintptr_t ptr = p;
  if(highBit != (sizeof(uintptr_t)*8-1))
    return (uint32_t)((ptr & ((1ull<<(highBit+1))-1)) >> lowBit);
  else
    return (uint32_t)(ptr >> lowBit);
}

inline uint32_t PtrLow16Shift8(const void* p) {
  uintptr_t ptr = reinterpret_cast<uintptr_t>(p);
  return (uint32_t)((ptr & 0xFFFFULL) >> 8);
}

inline uint32_t PtrHigh64Shift16(const void* p) {
  uintptr_t ptr = reinterpret_cast<uintptr_t>(p);
  return (uint32_t)((ptr & 0xFFFFFFFFFFFF0000ULL) >> 16);
}

inline uint32_t PtrLow40Shift8(const void* p) {
  uintptr_t ptr = reinterpret_cast<uintptr_t>(p);
  return (uint32_t)((ptr & 0xFFFFFFFFFFULL) >> 8);
}

inline uint32_t PtrHigh64Shift40(const void* p) {
  uintptr_t ptr = reinterpret_cast<uintptr_t>(p);
  return (uint32_t)((ptr & 0xFFFFFF0000000000ULL) >> 40);
}

inline uint32_t PtrLow32(const void* p) {
  return static_cast<uint32_t>(reinterpret_cast<uintptr_t>(p));
}

inline uint32_t PtrHigh32(const void* p) {
  uint32_t ptr = 0;
#ifdef HSA_LARGE_MODEL
  ptr = static_cast<uint32_t>(reinterpret_cast<uintptr_t>(p) >> 32);
#endif
  return ptr;
}

/**
* Generic functor compatible with the STL algorithms that enables proper
* destruction of a container of pointers. If (for instance), \c v is a vector
* of pointers to objects of type T, then the destructors of the elements in
* \c v are invoked when calling
* \code{std::for_each(v.begin(), v.end(), DeleteObject())}
*
* The original code and further information about this function object can be
* found in "Effective STL", 1st edition, item 7.
*/
struct DeleteObject {
  template<typename T>
  void operator()(const T *ptr) const {
    delete ptr;
  }
};
}  // namespace ext_image

#endif  // HSA_RUNTIME_EXT_IMAGE_UTIL_H