c++/src/capnp/endian.h

// Copyright (c) 2013-2014 Sandstorm Development Group, Inc. and contributors
// Licensed under the MIT License:
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.

#pragma once

#include "common.h"
#include <inttypes.h>
#include <string.h>  // memcpy

CAPNP_BEGIN_HEADER

namespace capnp {
namespace _ {  // private

// WireValue
//
// Wraps a primitive value as it appears on the wire.  Namely, values are little-endian on the
// wire, because little-endian is the most common endianness in modern CPUs.
//
// Note:  In general, code that depends cares about byte ordering is bad.  See:
//     http://commandcenter.blogspot.com/2012/04/byte-order-fallacy.html
//   Cap'n Proto is special because it is essentially doing compiler-like things, fussing over
//   allocation and layout of memory, in order to squeeze out every last drop of performance.

#if _MSC_VER
// Assume Windows is little-endian.
//
// TODO(msvc): This is ugly. Maybe refactor later checks to be based on CAPNP_BYTE_ORDER or
//   CAPNP_SWAP_BYTES or something, and define that in turn based on _MSC_VER or the GCC
//   intrinsics.

#ifndef __ORDER_BIG_ENDIAN__
#define __ORDER_BIG_ENDIAN__ 4321
#endif
#ifndef __ORDER_LITTLE_ENDIAN__
#define __ORDER_LITTLE_ENDIAN__ 1234
#endif
#ifndef __BYTE_ORDER__
#define __BYTE_ORDER__ __ORDER_LITTLE_ENDIAN__
#endif
#endif

#if CAPNP_REVERSE_ENDIAN
#define CAPNP_WIRE_BYTE_ORDER __ORDER_BIG_ENDIAN__
#define CAPNP_OPPOSITE_OF_WIRE_BYTE_ORDER __ORDER_LITTLE_ENDIAN__
#else
#define CAPNP_WIRE_BYTE_ORDER __ORDER_LITTLE_ENDIAN__
#define CAPNP_OPPOSITE_OF_WIRE_BYTE_ORDER __ORDER_BIG_ENDIAN__
#endif

#if defined(__BYTE_ORDER__) && \
    __BYTE_ORDER__ == CAPNP_WIRE_BYTE_ORDER && \
    !CAPNP_DISABLE_ENDIAN_DETECTION
// CPU is little-endian.  We can just read/write the memory directly.

template <typename T>
class DirectWireValue {
public:
  KJ_ALWAYS_INLINE(T get() const) { return value; }
  KJ_ALWAYS_INLINE(void set(T newValue)) { value = newValue; }

private:
  T value;
};

template <typename T>
using WireValue = DirectWireValue<T>;
// To prevent ODR problems when endian-test, endian-reverse-test, and endian-fallback-test are
// linked together, we define each implementation with a different name and define an alias to the
// one we want to use.

#elif defined(__BYTE_ORDER__) && \
      __BYTE_ORDER__ == CAPNP_OPPOSITE_OF_WIRE_BYTE_ORDER && \
      defined(__GNUC__) && !CAPNP_DISABLE_ENDIAN_DETECTION
// Big-endian, but GCC's __builtin_bswap() is available.

// TODO(perf):  Use dedicated instructions to read little-endian data on big-endian CPUs that have
//   them.

// TODO(perf):  Verify that this code optimizes reasonably.  In particular, ensure that the
//   compiler optimizes away the memcpy()s and keeps everything in registers.

template <typename T, size_t size = sizeof(T)>
class SwappingWireValue;

template <typename T>
class SwappingWireValue<T, 1> {
public:
  KJ_ALWAYS_INLINE(T get() const) { return value; }
  KJ_ALWAYS_INLINE(void set(T newValue)) { value = newValue; }

private:
  T value;
};

template <typename T>
class SwappingWireValue<T, 2> {
public:
  KJ_ALWAYS_INLINE(T get() const) {
    // Not all platforms have __builtin_bswap16() for some reason.  In particular, it is missing
    // on gcc-4.7.3-cygwin32 (but present on gcc-4.8.1-cygwin64).
    uint16_t swapped = (value << 8) | (value >> 8);
    T result;
    memcpy(&result, &swapped, sizeof(T));
    return result;
  }
  KJ_ALWAYS_INLINE(void set(T newValue)) {
    uint16_t raw;
    memcpy(&raw, &newValue, sizeof(T));
    // Not all platforms have __builtin_bswap16() for some reason.  In particular, it is missing
    // on gcc-4.7.3-cygwin32 (but present on gcc-4.8.1-cygwin64).
    value = (raw << 8) | (raw >> 8);
  }

private:
  uint16_t value;
};

template <typename T>
class SwappingWireValue<T, 4> {
public:
  KJ_ALWAYS_INLINE(T get() const) {
    uint32_t swapped = __builtin_bswap32(value);
    T result;
    memcpy(&result, &swapped, sizeof(T));
    return result;
  }
  KJ_ALWAYS_INLINE(void set(T newValue)) {
    uint32_t raw;
    memcpy(&raw, &newValue, sizeof(T));
    value = __builtin_bswap32(raw);
  }

private:
  uint32_t value;
};

template <typename T>
class SwappingWireValue<T, 8> {
public:
  KJ_ALWAYS_INLINE(T get() const) {
    uint64_t swapped = __builtin_bswap64(value);
    T result;
    memcpy(&result, &swapped, sizeof(T));
    return result;
  }
  KJ_ALWAYS_INLINE(void set(T newValue)) {
    uint64_t raw;
    memcpy(&raw, &newValue, sizeof(T));
    value = __builtin_bswap64(raw);
  }

private:
  uint64_t value;
};

template <typename T>
using WireValue = SwappingWireValue<T>;
// To prevent ODR problems when endian-test, endian-reverse-test, and endian-fallback-test are
// linked together, we define each implementation with a different name and define an alias to the
// one we want to use.

#else
// Unknown endianness.  Fall back to bit shifts.

#if !CAPNP_DISABLE_ENDIAN_DETECTION
#if _MSC_VER
#pragma message("Couldn't detect endianness of your platform.  Using unoptimized fallback implementation.")
#pragma message("Consider changing this code to detect your platform and send us a patch!")
#else
#warning "Couldn't detect endianness of your platform.  Using unoptimized fallback implementation."
#warning "Consider changing this code to detect your platform and send us a patch!"
#endif
#endif  // !CAPNP_DISABLE_ENDIAN_DETECTION

template <typename T, size_t size = sizeof(T)>
class ShiftingWireValue;

template <typename T>
class ShiftingWireValue<T, 1> {
public:
  KJ_ALWAYS_INLINE(T get() const) { return value; }
  KJ_ALWAYS_INLINE(void set(T newValue)) { value = newValue; }

private:
  T value;
};

template <typename T>
class ShiftingWireValue<T, 2> {
public:
  KJ_ALWAYS_INLINE(T get() const) {
    uint16_t raw = (static_cast<uint16_t>(bytes[0])     ) |
                   (static_cast<uint16_t>(bytes[1]) << 8);
    T result;
    memcpy(&result, &raw, sizeof(T));
    return result;
  }
  KJ_ALWAYS_INLINE(void set(T newValue)) {
    uint16_t raw;
    memcpy(&raw, &newValue, sizeof(T));
    bytes[0] = raw;
    bytes[1] = raw >> 8;
  }

private:
  union {
    byte bytes[2];
    uint16_t align;
  };
};

template <typename T>
class ShiftingWireValue<T, 4> {
public:
  KJ_ALWAYS_INLINE(T get() const) {
    uint32_t raw = (static_cast<uint32_t>(bytes[0])      ) |
                   (static_cast<uint32_t>(bytes[1]) <<  8) |
                   (static_cast<uint32_t>(bytes[2]) << 16) |
                   (static_cast<uint32_t>(bytes[3]) << 24);
    T result;
    memcpy(&result, &raw, sizeof(T));
    return result;
  }
  KJ_ALWAYS_INLINE(void set(T newValue)) {
    uint32_t raw;
    memcpy(&raw, &newValue, sizeof(T));
    bytes[0] = raw;
    bytes[1] = raw >> 8;
    bytes[2] = raw >> 16;
    bytes[3] = raw >> 24;
  }

private:
  union {
    byte bytes[4];
    uint32_t align;
  };
};

template <typename T>
class ShiftingWireValue<T, 8> {
public:
  KJ_ALWAYS_INLINE(T get() const) {
    uint64_t raw = (static_cast<uint64_t>(bytes[0])      ) |
                   (static_cast<uint64_t>(bytes[1]) <<  8) |
                   (static_cast<uint64_t>(bytes[2]) << 16) |
                   (static_cast<uint64_t>(bytes[3]) << 24) |
                   (static_cast<uint64_t>(bytes[4]) << 32) |
                   (static_cast<uint64_t>(bytes[5]) << 40) |
                   (static_cast<uint64_t>(bytes[6]) << 48) |
                   (static_cast<uint64_t>(bytes[7]) << 56);
    T result;
    memcpy(&result, &raw, sizeof(T));
    return result;
  }
  KJ_ALWAYS_INLINE(void set(T newValue)) {
    uint64_t raw;
    memcpy(&raw, &newValue, sizeof(T));
    bytes[0] = raw;
    bytes[1] = raw >> 8;
    bytes[2] = raw >> 16;
    bytes[3] = raw >> 24;
    bytes[4] = raw >> 32;
    bytes[5] = raw >> 40;
    bytes[6] = raw >> 48;
    bytes[7] = raw >> 56;
  }

private:
  union {
    byte bytes[8];
    uint64_t align;
  };
};

template <typename T>
using WireValue = ShiftingWireValue<T>;
// To prevent ODR problems when endian-test, endian-reverse-test, and endian-fallback-test are
// linked together, we define each implementation with a different name and define an alias to the
// one we want to use.

#endif

}  // namespace _ (private)
}  // namespace capnp

CAPNP_END_HEADER