You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
qmk_firmware/LUFA/Common/Endianness.h

477 lines
17 KiB

/*
LUFA Library
Copyright (C) Dean Camera, 2011.
dean [at] fourwalledcubicle [dot] com
www.lufa-lib.org
*/
/*
Copyright 2011 Dean Camera (dean [at] fourwalledcubicle [dot] com)
Permission to use, copy, modify, distribute, and sell this
software and its documentation for any purpose is hereby granted
without fee, provided that the above copyright notice appear in
all copies and that both that the copyright notice and this
permission notice and warranty disclaimer appear in supporting
documentation, and that the name of the author not be used in
advertising or publicity pertaining to distribution of the
software without specific, written prior permission.
The author disclaim all warranties with regard to this
software, including all implied warranties of merchantability
and fitness. In no event shall the author be liable for any
special, indirect or consequential damages or any damages
whatsoever resulting from loss of use, data or profits, whether
in an action of contract, negligence or other tortious action,
arising out of or in connection with the use or performance of
this software.
*/
/** \file
* \brief Endianness and Byte Ordering macros and functions.
*
* \copydetails Group_Endianness
*/
/** \ingroup Group_Endianness
* \defgroup Group_ByteSwapping Byte Reordering
* \brief Macros and functions for forced byte reordering.
*/
/** \ingroup Group_Endianness
* \defgroup Group_EndianConversion Endianness Conversion
* \brief Macros and functions for automatic endianness conversion.
*/
/** \ingroup Group_Common
* \defgroup Group_Endianness Endianness and Byte Ordering
* \brief Convenience macros and functions relating to byte (re-)ordering
*
* Common library convenience macros and functions relating to byte (re-)ordering.
*
* @{
*/
#ifndef __LUFA_ENDIANNESS_H__
#define __LUFA_ENDIANNESS_H__
/* Enable C linkage for C++ Compilers: */
#if defined(__cplusplus)
extern "C" {
#endif
/* Preprocessor Checks: */
#if !defined(__INCLUDE_FROM_COMMON_H)
#error Do not include this file directly. Include LUFA/Common/Common.h instead to gain this functionality.
#endif
#if !(defined(ARCH_BIG_ENDIAN) || defined(ARCH_LITTLE_ENDIAN))
#error ARCH_BIG_ENDIAN or ARCH_LITTLE_ENDIAN not set for the specified architecture.
#endif
/* Public Interface - May be used in end-application: */
/* Macros: */
/** Swaps the byte ordering of a 16-bit value at compile-time. Do not use this macro for swapping byte orderings
* of dynamic values computed at runtime, use \ref SwapEndian_16() instead. The result of this macro can be used
* inside struct or other variable initializers outside of a function, something that is not possible with the
* inline function variant.
*
* \ingroup Group_ByteSwapping
*
* \param[in] x 16-bit value whose byte ordering is to be swapped.
*
* \return Input value with the byte ordering reversed.
*/
#define SWAPENDIAN_16(x) (uint16_t)((((x) & 0xFF00) >> 8) | (((x) & 0x00FF) << 8))
/** Swaps the byte ordering of a 32-bit value at compile-time. Do not use this macro for swapping byte orderings
* of dynamic values computed at runtime- use \ref SwapEndian_32() instead. The result of this macro can be used
* inside struct or other variable initializers outside of a function, something that is not possible with the
* inline function variant.
*
* \ingroup Group_ByteSwapping
*
* \param[in] x 32-bit value whose byte ordering is to be swapped.
*
* \return Input value with the byte ordering reversed.
*/
#define SWAPENDIAN_32(x) (uint32_t)((((x) & 0xFF000000UL) >> 24UL) | (((x) & 0x00FF0000UL) >> 8UL) | \
(((x) & 0x0000FF00UL) << 8UL) | (((x) & 0x000000FFUL) << 24UL))
#if defined(ARCH_BIG_ENDIAN) && !defined(le16_to_cpu)
#define le16_to_cpu(x) SwapEndian_16(x)
#define le32_to_cpu(x) SwapEndian_32(x)
#define be16_to_cpu(x) x
#define be32_to_cpu(x) x
#define cpu_to_le16(x) SwapEndian_16(x)
#define cpu_to_le32(x) SwapEndian_32(x)
#define cpu_to_be16(x) x
#define cpu_to_be32(x) x
#define LE16_TO_CPU(x) SWAPENDIAN_16(x)
#define LE32_TO_CPU(x) SWAPENDIAN_32(x)
#define BE16_TO_CPU(x) x
#define BE32_TO_CPU(x) x
#define CPU_TO_LE16(x) SWAPENDIAN_16(x)
#define CPU_TO_LE32(x) SWAPENDIAN_32(x)
#define CPU_TO_BE16(x) x
#define CPU_TO_BE32(x) x
#elif !defined(le16_to_cpu)
/** \name Run-time endianness conversion */
//@{
/** Performs a conversion between a Little Endian encoded 16-bit piece of data and the
* Endianness of the currently selected CPU architecture.
*
* On little endian architectures, this macro does nothing.
*
* \note This macro is designed for run-time conversion of data - for compile-time endianness
* conversion, use \ref LE16_TO_CPU instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define le16_to_cpu(x) x
/** Performs a conversion between a Little Endian encoded 32-bit piece of data and the
* Endianness of the currently selected CPU architecture.
*
* On little endian architectures, this macro does nothing.
*
* \note This macro is designed for run-time conversion of data - for compile-time endianness
* conversion, use \ref LE32_TO_CPU instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define le32_to_cpu(x) x
/** Performs a conversion between a Big Endian encoded 16-bit piece of data and the
* Endianness of the currently selected CPU architecture.
*
* On big endian architectures, this macro does nothing.
*
* \note This macro is designed for run-time conversion of data - for compile-time endianness
* conversion, use \ref BE16_TO_CPU instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define be16_to_cpu(x) SwapEndian_16(x)
/** Performs a conversion between a Big Endian encoded 32-bit piece of data and the
* Endianness of the currently selected CPU architecture.
*
* On big endian architectures, this macro does nothing.
*
* \note This macro is designed for run-time conversion of data - for compile-time endianness
* conversion, use \ref BE32_TO_CPU instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define be32_to_cpu(x) SwapEndian_32(x)
/** Performs a conversion on a natively encoded 16-bit piece of data to ensure that it
* is in Little Endian format regardless of the currently selected CPU architecture.
*
* On little endian architectures, this macro does nothing.
*
* \note This macro is designed for run-time conversion of data - for compile-time endianness
* conversion, use \ref CPU_TO_LE16 instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define cpu_to_le16(x) x
/** Performs a conversion on a natively encoded 32-bit piece of data to ensure that it
* is in Little Endian format regardless of the currently selected CPU architecture.
*
* On little endian architectures, this macro does nothing.
*
* \note This macro is designed for run-time conversion of data - for compile-time endianness
* conversion, use \ref CPU_TO_LE32 instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define cpu_to_le32(x) x
/** Performs a conversion on a natively encoded 16-bit piece of data to ensure that it
* is in Big Endian format regardless of the currently selected CPU architecture.
*
* On big endian architectures, this macro does nothing.
*
* \note This macro is designed for run-time conversion of data - for compile-time endianness
* conversion, use \ref CPU_TO_BE16 instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define cpu_to_be16(x) SwapEndian_16(x)
/** Performs a conversion on a natively encoded 32-bit piece of data to ensure that it
* is in Big Endian format regardless of the currently selected CPU architecture.
*
* On big endian architectures, this macro does nothing.
*
* \note This macro is designed for run-time conversion of data - for compile-time endianness
* conversion, use \ref CPU_TO_BE32 instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define cpu_to_be32(x) SwapEndian_32(x)
//@}
/** \name Compile-time endianness conversion */
//@{
/** Performs a conversion between a Little Endian encoded 16-bit piece of data and the
* Endianness of the currently selected CPU architecture.
*
* On little endian architectures, this macro does nothing.
*
* \note This macro is designed for compile-time conversion of data - for run time endianness
* conversion, use \ref le16_to_cpu instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define LE16_TO_CPU(x) x
/** Performs a conversion between a Little Endian encoded 32-bit piece of data and the
* Endianness of the currently selected CPU architecture.
*
* On little endian architectures, this macro does nothing.
*
* \note This macro is designed for compile-time conversion of data - for run time endianness
* conversion, use \ref le32_to_cpu instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define LE32_TO_CPU(x) x
/** Performs a conversion between a Big Endian encoded 16-bit piece of data and the
* Endianness of the currently selected CPU architecture.
*
* On big endian architectures, this macro does nothing.
*
* \note This macro is designed for compile-time conversion of data - for run-time endianness
* conversion, use \ref be16_to_cpu instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define BE16_TO_CPU(x) SWAPENDIAN_16(x)
/** Performs a conversion between a Big Endian encoded 32-bit piece of data and the
* Endianness of the currently selected CPU architecture.
*
* On big endian architectures, this macro does nothing.
*
* \note This macro is designed for compile-time conversion of data - for run-time endianness
* conversion, use \ref be32_to_cpu instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define BE32_TO_CPU(x) SWAPENDIAN_32(x)
/** Performs a conversion on a natively encoded 16-bit piece of data to ensure that it
* is in Little Endian format regardless of the currently selected CPU architecture.
*
* On little endian architectures, this macro does nothing.
*
* \note This macro is designed for compile-time conversion of data - for run-time endianness
* conversion, use \ref cpu_to_le16 instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define CPU_TO_LE16(x) x
/** Performs a conversion on a natively encoded 32-bit piece of data to ensure that it
* is in Little Endian format regardless of the currently selected CPU architecture.
*
* On little endian architectures, this macro does nothing.
*
* \note This macro is designed for compile-time conversion of data - for run-time endianness
* conversion, use \ref cpu_to_le32 instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define CPU_TO_LE32(x) x
/** Performs a conversion on a natively encoded 16-bit piece of data to ensure that it
* is in Big Endian format regardless of the currently selected CPU architecture.
*
* On big endian architectures, this macro does nothing.
*
* \note This macro is designed for compile-time conversion of data - for run-time endianness
* conversion, use \ref cpu_to_be16 instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define CPU_TO_BE16(x) SWAPENDIAN_16(x)
/** Performs a conversion on a natively encoded 32-bit piece of data to ensure that it
* is in Big Endian format regardless of the currently selected CPU architecture.
*
* On big endian architectures, this macro does nothing.
*
* \note This macro is designed for compile-time conversion of data - for run-time endianness
* conversion, use \ref cpu_to_be32 instead.
*
* \ingroup Group_EndianConversion
*
* \param[in] x Data to perform the endianness conversion on.
*
* \return Endian corrected version of the input value.
*/
#define CPU_TO_BE32(x) SWAPENDIAN_32(x)
//! @}
#endif
/* Inline Functions: */
/** Function to reverse the byte ordering of the individual bytes in a 16 bit value.
*
* \ingroup Group_ByteSwapping
*
* \param[in] Word Word of data whose bytes are to be swapped.
*/
static inline uint16_t SwapEndian_16(const uint16_t Word) ATTR_WARN_UNUSED_RESULT ATTR_CONST;
static inline uint16_t SwapEndian_16(const uint16_t Word)
{
uint8_t Temp;
union
{
uint16_t Word;
uint8_t Bytes[2];
} Data;
Data.Word = Word;
Temp = Data.Bytes[0];
Data.Bytes[0] = Data.Bytes[1];
Data.Bytes[1] = Temp;
return Data.Word;
}
/** Function to reverse the byte ordering of the individual bytes in a 32 bit value.
*
* \ingroup Group_ByteSwapping
*
* \param[in] DWord Double word of data whose bytes are to be swapped.
*/
static inline uint32_t SwapEndian_32(const uint32_t DWord) ATTR_WARN_UNUSED_RESULT ATTR_CONST;
static inline uint32_t SwapEndian_32(const uint32_t DWord)
{
uint8_t Temp;
union
{
uint32_t DWord;
uint8_t Bytes[4];
} Data;
Data.DWord = DWord;
Temp = Data.Bytes[0];
Data.Bytes[0] = Data.Bytes[3];
Data.Bytes[3] = Temp;
Temp = Data.Bytes[1];
Data.Bytes[1] = Data.Bytes[2];
Data.Bytes[2] = Temp;
return Data.DWord;
}
/** Function to reverse the byte ordering of the individual bytes in a n byte value.
*
* \ingroup Group_ByteSwapping
*
* \param[in,out] Data Pointer to a number containing an even number of bytes to be reversed.
* \param[in] Length Length of the data in bytes.
*/
static inline void SwapEndian_n(void* const Data,
uint8_t Length) ATTR_NON_NULL_PTR_ARG(1);
static inline void SwapEndian_n(void* const Data,
uint8_t Length)
{
uint8_t* CurrDataPos = (uint8_t*)Data;
while (Length > 1)
{
uint8_t Temp = *CurrDataPos;
*CurrDataPos = *(CurrDataPos + Length - 1);
*(CurrDataPos + Length - 1) = Temp;
CurrDataPos++;
Length -= 2;
}
}
/* Disable C linkage for C++ Compilers: */
#if defined(__cplusplus)
}
#endif
#endif
/** @} */