/*
* Byte Swapping Operations
* (C) 1999-2011 Jack Lloyd
* (C) 2007 Yves Jerschow
*
* Distributed under the terms of the Botan license
*/

#ifndef BOTAN_BYTE_SWAP_H__
#define BOTAN_BYTE_SWAP_H__

#include <botan/types.h>
#include <botan/rotate.h>

#if defined(BOTAN_TARGET_CPU_HAS_SSE2) && !defined(BOTAN_NO_SSE_INTRINSICS)
  #include <emmintrin.h>
#endif

namespace Botan {

/**
* Swap a 16 bit integer
*/
inline u16bit reverse_bytes(u16bit val)
   {
   return rotate_left(val, 8);
   }

/**
* Swap a 32 bit integer
*/
inline u32bit reverse_bytes(u32bit val)
   {
#if BOTAN_GCC_VERSION >= 430 && !defined(BOTAN_TARGET_CPU_IS_ARM_FAMILY)
   /*
   GCC intrinsic added in 4.3, works for a number of CPUs

   However avoid under ARM, as it branches to a function in libgcc
   instead of generating inline asm, so slower even than the generic
   rotate version below.
   */
   return __builtin_bswap32(val);

#elif BOTAN_USE_GCC_INLINE_ASM && defined(BOTAN_TARGET_CPU_IS_X86_FAMILY)

   // GCC-style inline assembly for x86 or x86-64
   asm("bswapl %0" : "=r" (val) : "0" (val));
   return val;

#elif BOTAN_USE_GCC_INLINE_ASM && defined(BOTAN_TARGET_CPU_IS_ARM_FAMILY)

   asm ("eor r3, %1, %1, ror #16\n\t"
        "bic r3, r3, #0x00FF0000\n\t"
        "mov %0, %1, ror #8\n\t"
        "eor %0, %0, r3, lsr #8"
        : "=r" (val)
        : "0" (val)
        : "r3", "cc");

   return val;

#elif defined(_MSC_VER) && defined(BOTAN_TARGET_ARCH_IS_X86_32)

   // Visual C++ inline asm for 32-bit x86, by Yves Jerschow
   __asm mov eax, val;
   __asm bswap eax;

#else

   // Generic implementation
   return (rotate_right(val, 8) & 0xFF00FF00) |
          (rotate_left (val, 8) & 0x00FF00FF);

#endif
   }

/**
* Swap a 64 bit integer
*/
inline u64bit reverse_bytes(u64bit val)
   {
#if BOTAN_GCC_VERSION >= 430

   // GCC intrinsic added in 4.3, works for a number of CPUs
   return __builtin_bswap64(val);

#elif BOTAN_USE_GCC_INLINE_ASM && defined(BOTAN_TARGET_ARCH_IS_X86_64)
   // GCC-style inline assembly for x86-64
   asm("bswapq %0" : "=r" (val) : "0" (val));
   return val;

#else
   /* Generic implementation. Defined in terms of 32-bit bswap so any
    * optimizations in that version can help here (particularly
    * useful for 32-bit x86).
    */

   u32bit hi = static_cast<u32bit>(val >> 32);
   u32bit lo = static_cast<u32bit>(val);

   hi = reverse_bytes(hi);
   lo = reverse_bytes(lo);

   return (static_cast<u64bit>(lo) << 32) | hi;
#endif
   }

/**
* Swap 4 Ts in an array
*/
template<typename T>
inline void bswap_4(T x[4])
   {
   x[0] = reverse_bytes(x[0]);
   x[1] = reverse_bytes(x[1]);
   x[2] = reverse_bytes(x[2]);
   x[3] = reverse_bytes(x[3]);
   }

#if defined(BOTAN_TARGET_CPU_HAS_SSE2) && !defined(BOTAN_NO_SSE_INTRINSICS)

/**
* Swap 4 u32bits in an array using SSE2 shuffle instructions
*/
template<>
inline void bswap_4(u32bit x[4])
   {
   __m128i T = _mm_loadu_si128(reinterpret_cast<const __m128i*>(x));

   T = _mm_shufflehi_epi16(T, _MM_SHUFFLE(2, 3, 0, 1));
   T = _mm_shufflelo_epi16(T, _MM_SHUFFLE(2, 3, 0, 1));

   T =  _mm_or_si128(_mm_srli_epi16(T, 8), _mm_slli_epi16(T, 8));

   _mm_storeu_si128(reinterpret_cast<__m128i*>(x), T);
   }

#endif

}

#endif