Labrador/AVR_Code/USB_BULK_TEST/src/ASF/xmega/utils/compiler.h

1190 lines
32 KiB
C

/**
* \file
*
* \brief Commonly used includes, types and macros.
*
* Copyright (c) 2010-2015 Atmel Corporation. All rights reserved.
*
* \asf_license_start
*
* \page License
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. The name of Atmel may not be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* 4. This software may only be redistributed and used in connection with an
* Atmel microcontroller product.
*
* THIS SOFTWARE IS PROVIDED BY ATMEL "AS IS" AND ANY EXPRESS OR IMPLIED
* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT ARE
* EXPRESSLY AND SPECIFICALLY DISCLAIMED. IN NO EVENT SHALL ATMEL BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
* STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
* \asf_license_stop
*
*/
/*
* Support and FAQ: visit <a href="http://www.atmel.com/design-support/">Atmel Support</a>
*/
#ifndef UTILS_COMPILER_H
#define UTILS_COMPILER_H
/**
* \defgroup group_xmega_utils XMEGA compiler driver
*
* Compiler abstraction layer and code utilities for 8-bit AVR.
* This module provides various abstraction layers and utilities to make code compatible between different compilers.
*
* \{
*/
#if defined(__GNUC__)
# include <avr/io.h>
# include <avr/builtins.h>
#elif defined(__ICCAVR__)
# include <ioavr.h>
# include <intrinsics.h>
#else
# error Unsupported compiler.
#endif
#include <stdbool.h>
#include <stdint.h>
#include <stddef.h>
#include <stdlib.h>
#include <parts.h>
#ifdef __ICCAVR__
/*! \name Compiler Keywords
*
* Port of some keywords from GCC to IAR Embedded Workbench.
*/
//! @{
#define __asm__ asm
#define __inline__ inline
#define __volatile__
//! @}
#endif
/**
* \def UNUSED
* \brief Marking \a v as a unused parameter or value.
*/
#define UNUSED(v) (void)(v)
/**
* \def unused
* \brief Marking \a v as a unused parameter or value.
*/
#define unused(v) do { (void)(v); } while(0)
/**
* \def barrier
* \brief Memory barrier
*/
#ifdef __GNUC__
# define barrier() asm volatile("" ::: "memory")
#else
# define barrier() asm ("")
#endif
/**
* \brief Emit the compiler pragma \a arg.
*
* \param arg The pragma directive as it would appear after \e \#pragma
* (i.e. not stringified).
*/
#define COMPILER_PRAGMA(arg) _Pragma(#arg)
/*
* AVR arch does not care about alignment anyway.
*/
#define COMPILER_PACK_RESET(alignment)
#define COMPILER_PACK_SET(alignment)
/**
* \brief Set aligned boundary.
*/
#if (defined __GNUC__)
#define COMPILER_ALIGNED(a) __attribute__((__aligned__(a)))
#elif (defined __ICCAVR__)
#define COMPILER_ALIGNED(a) COMPILER_PRAGMA(data_alignment = a)
#endif
/**
* \brief Set word-aligned boundary.
*/
#if (defined __GNUC__)
#define COMPILER_WORD_ALIGNED __attribute__((__aligned__(2)))
#elif (defined __ICCAVR__)
#define COMPILER_WORD_ALIGNED COMPILER_PRAGMA(data_alignment = 2)
#endif
/**
* \name Tag functions as deprecated
*
* Tagging a function as deprecated will produce a warning when and only
* when the function is called.
*
* Usage is to add the __DEPRECATED__ symbol before the function definition.
* E.g.:
* __DEPRECATED__ uint8_t some_deprecated_function (void)
* {
* ...
* }
*
* \note Only supported by GCC 3.1 and above, no IAR support
* @{
*/
#if ((defined __GNUC__) && (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >=1)))
#define __DEPRECATED__ __attribute__((__deprecated__))
#else
#define __DEPRECATED__
#endif
//! @}
/*! \name Usual Types
*/
//! @{
typedef unsigned char Bool; //!< Boolean.
#ifndef __cplusplus
#if !defined(__bool_true_false_are_defined)
typedef unsigned char bool; //!< Boolean.
#endif
#endif
typedef int8_t S8 ; //!< 8-bit signed integer.
typedef uint8_t U8 ; //!< 8-bit unsigned integer.
typedef int16_t S16; //!< 16-bit signed integer.
typedef uint16_t U16; //!< 16-bit unsigned integer.
typedef uint16_t le16_t;
typedef uint16_t be16_t;
typedef int32_t S32; //!< 32-bit signed integer.
typedef uint32_t U32; //!< 32-bit unsigned integer.
typedef uint32_t le32_t;
typedef uint32_t be32_t;
typedef int64_t S64; //!< 64-bit signed integer.
typedef uint64_t U64; //!< 64-bit unsigned integer.
typedef float F32; //!< 32-bit floating-point number.
typedef double F64; //!< 64-bit floating-point number.
typedef uint16_t iram_size_t;
//! @}
/*! \name Status Types
*/
//! @{
typedef Bool Status_bool_t; //!< Boolean status.
typedef U8 Status_t; //!< 8-bit-coded status.
//! @}
/*! \name Aliasing Aggregate Types
*/
//! @{
//! 16-bit union.
typedef union
{
S16 s16 ;
U16 u16 ;
S8 s8 [2];
U8 u8 [2];
} Union16;
//! 32-bit union.
typedef union
{
S32 s32 ;
U32 u32 ;
S16 s16[2];
U16 u16[2];
S8 s8 [4];
U8 u8 [4];
} Union32;
//! 64-bit union.
typedef union
{
S64 s64 ;
U64 u64 ;
S32 s32[2];
U32 u32[2];
S16 s16[4];
U16 u16[4];
S8 s8 [8];
U8 u8 [8];
} Union64;
//! Union of pointers to 64-, 32-, 16- and 8-bit unsigned integers.
typedef union
{
S64 *s64ptr;
U64 *u64ptr;
S32 *s32ptr;
U32 *u32ptr;
S16 *s16ptr;
U16 *u16ptr;
S8 *s8ptr ;
U8 *u8ptr ;
} UnionPtr;
//! Union of pointers to volatile 64-, 32-, 16- and 8-bit unsigned integers.
typedef union
{
volatile S64 *s64ptr;
volatile U64 *u64ptr;
volatile S32 *s32ptr;
volatile U32 *u32ptr;
volatile S16 *s16ptr;
volatile U16 *u16ptr;
volatile S8 *s8ptr ;
volatile U8 *u8ptr ;
} UnionVPtr;
//! Union of pointers to constant 64-, 32-, 16- and 8-bit unsigned integers.
typedef union
{
const S64 *s64ptr;
const U64 *u64ptr;
const S32 *s32ptr;
const U32 *u32ptr;
const S16 *s16ptr;
const U16 *u16ptr;
const S8 *s8ptr ;
const U8 *u8ptr ;
} UnionCPtr;
//! Union of pointers to constant volatile 64-, 32-, 16- and 8-bit unsigned integers.
typedef union
{
const volatile S64 *s64ptr;
const volatile U64 *u64ptr;
const volatile S32 *s32ptr;
const volatile U32 *u32ptr;
const volatile S16 *s16ptr;
const volatile U16 *u16ptr;
const volatile S8 *s8ptr ;
const volatile U8 *u8ptr ;
} UnionCVPtr;
//! Structure of pointers to 64-, 32-, 16- and 8-bit unsigned integers.
typedef struct
{
S64 *s64ptr;
U64 *u64ptr;
S32 *s32ptr;
U32 *u32ptr;
S16 *s16ptr;
U16 *u16ptr;
S8 *s8ptr ;
U8 *u8ptr ;
} StructPtr;
//! Structure of pointers to volatile 64-, 32-, 16- and 8-bit unsigned integers.
typedef struct
{
volatile S64 *s64ptr;
volatile U64 *u64ptr;
volatile S32 *s32ptr;
volatile U32 *u32ptr;
volatile S16 *s16ptr;
volatile U16 *u16ptr;
volatile S8 *s8ptr ;
volatile U8 *u8ptr ;
} StructVPtr;
//! Structure of pointers to constant 64-, 32-, 16- and 8-bit unsigned integers.
typedef struct
{
const S64 *s64ptr;
const U64 *u64ptr;
const S32 *s32ptr;
const U32 *u32ptr;
const S16 *s16ptr;
const U16 *u16ptr;
const S8 *s8ptr ;
const U8 *u8ptr ;
} StructCPtr;
//! Structure of pointers to constant volatile 64-, 32-, 16- and 8-bit unsigned integers.
typedef struct
{
const volatile S64 *s64ptr;
const volatile U64 *u64ptr;
const volatile S32 *s32ptr;
const volatile U32 *u32ptr;
const volatile S16 *s16ptr;
const volatile U16 *u16ptr;
const volatile S8 *s8ptr ;
const volatile U8 *u8ptr ;
} StructCVPtr;
//! @}
//_____ M A C R O S ________________________________________________________
/*! \name Usual Constants
*/
//! @{
#define DISABLE 0
#define ENABLE 1
#ifndef __cplusplus
#if !defined(__bool_true_false_are_defined)
#define false 0
#define true 1
#endif
#endif
#define PASS 0
#define FAIL 1
#define LOW 0
#define HIGH 1
//! @}
//! \name Compile time error handling
//@{
/**
* \internal
* \def ERROR_FUNC(name, msg)
* \brief Fail compilation if function call isn't eliminated
*
* If the compiler fails to optimize away all calls to the function \a
* name, terminate compilation and display \a msg to the user.
*
* \note Not all compilers support this, so this is best-effort only.
* Sometimes, there may be a linker error instead, and when optimization
* is disabled, this mechanism will be completely disabled.
*/
#ifndef ERROR_FUNC
# define ERROR_FUNC(name, msg) \
extern int name(void)
#endif
//@}
//! \name Function call demultiplexing
//@{
//! Error function for failed demultiplexing.
ERROR_FUNC(compiler_demux_bad_size, "Invalid parameter size");
/**
* \internal
* \brief Demultiplex function call based on size of datatype
*
* Evaluates to a function call to a function name with suffix 8, 16 or 32
* depending on the size of the datatype. Any number of parameters can be
* passed to the function.
*
* Usage:
* \code
void foo8(uint8_t a, void *b);
void foo16(uint16_t a, void *b);
void foo32(uint32_t a, void *b);
#define foo(x, y) compiler_demux_size(sizeof(x), foo, x, y)
\endcode
*
* \param size Size of the datatype.
* \param func Base function name.
* \param ... List of parameters to pass to the function.
*/
#define compiler_demux_size(size, func, ...) \
(((size) == 1) ? func##8(__VA_ARGS__) : \
((size) == 2) ? func##16(__VA_ARGS__) : \
((size) == 4) ? func##32(__VA_ARGS__) : \
compiler_demux_bad_size())
//@}
/**
* \def __always_inline
* \brief The function should always be inlined.
*
* This annotation instructs the compiler to ignore its inlining
* heuristics and inline the function no matter how big it thinks it
* becomes.
*/
#if (defined __GNUC__)
#define __always_inline inline __attribute__((__always_inline__))
#elif (defined __ICCAVR__)
#define __always_inline _Pragma("inline=forced")
#endif
//! \name Optimization Control
//@{
/**
* \def __always_optimize
* \brief The function should always be optimized.
*
* This annotation instructs the compiler to ignore global optimization
* settings and always compile the function with a high level of
* optimization.
*/
#if (defined __GNUC__)
#define __always_optimize __attribute__((optimize(3)))
#elif (defined __ICCAVR__)
#define __always_optimize _Pragma("optimize=high")
#endif
/**
* \def likely(exp)
* \brief The expression \a exp is likely to be true
*/
#ifndef likely
# define likely(exp) (exp)
#endif
/**
* \def unlikely(exp)
* \brief The expression \a exp is unlikely to be true
*/
#ifndef unlikely
# define unlikely(exp) (exp)
#endif
/**
* \def is_constant(exp)
* \brief Determine if an expression evaluates to a constant value.
*
* \param exp Any expression
*
* \return true if \a exp is constant, false otherwise.
*/
#ifdef __GNUC__
# define is_constant(exp) __builtin_constant_p(exp)
#else
# define is_constant(exp) (0)
#endif
//! @}
/*! \name Bit-Field Handling
*/
#include "bit_handling/clz_ctz.h"
//! @{
/*! \brief Reads the bits of a value specified by a given bit-mask.
*
* \param value Value to read bits from.
* \param mask Bit-mask indicating bits to read.
*
* \return Read bits.
*/
#define Rd_bits( value, mask) ((value)&(mask))
/*! \brief Writes the bits of a C lvalue specified by a given bit-mask.
*
* \param lvalue C lvalue to write bits to.
* \param mask Bit-mask indicating bits to write.
* \param bits Bits to write.
*
* \return Resulting value with written bits.
*/
#define Wr_bits(lvalue, mask, bits) ((lvalue) = ((lvalue) & ~(mask)) |\
((bits ) & (mask)))
/*! \brief Tests the bits of a value specified by a given bit-mask.
*
* \param value Value of which to test bits.
* \param mask Bit-mask indicating bits to test.
*
* \return \c 1 if at least one of the tested bits is set, else \c 0.
*/
#define Tst_bits( value, mask) (Rd_bits(value, mask) != 0)
/*! \brief Clears the bits of a C lvalue specified by a given bit-mask.
*
* \param lvalue C lvalue of which to clear bits.
* \param mask Bit-mask indicating bits to clear.
*
* \return Resulting value with cleared bits.
*/
#define Clr_bits(lvalue, mask) ((lvalue) &= ~(mask))
/*! \brief Sets the bits of a C lvalue specified by a given bit-mask.
*
* \param lvalue C lvalue of which to set bits.
* \param mask Bit-mask indicating bits to set.
*
* \return Resulting value with set bits.
*/
#define Set_bits(lvalue, mask) ((lvalue) |= (mask))
/*! \brief Toggles the bits of a C lvalue specified by a given bit-mask.
*
* \param lvalue C lvalue of which to toggle bits.
* \param mask Bit-mask indicating bits to toggle.
*
* \return Resulting value with toggled bits.
*/
#define Tgl_bits(lvalue, mask) ((lvalue) ^= (mask))
/*! \brief Reads the bit-field of a value specified by a given bit-mask.
*
* \param value Value to read a bit-field from.
* \param mask Bit-mask indicating the bit-field to read.
*
* \return Read bit-field.
*/
#define Rd_bitfield( value,mask) (Rd_bits( value, (uint32_t)mask) >> ctz(mask))
/*! \brief Writes the bit-field of a C lvalue specified by a given bit-mask.
*
* \param lvalue C lvalue to write a bit-field to.
* \param mask Bit-mask indicating the bit-field to write.
* \param bitfield Bit-field to write.
*
* \return Resulting value with written bit-field.
*/
#define Wr_bitfield(lvalue, mask, bitfield) (Wr_bits(lvalue, mask, (uint32_t)(bitfield) << ctz(mask)))
//! @}
/*! \brief This macro is used to test fatal errors.
*
* The macro tests if the expression is false. If it is, a fatal error is
* detected and the application hangs up. If TEST_SUITE_DEFINE_ASSERT_MACRO
* is defined, a unit test version of the macro is used, to allow execution
* of further tests after a false expression.
*
* \param expr Expression to evaluate and supposed to be nonzero.
*/
#if defined(_ASSERT_ENABLE_)
# if defined(TEST_SUITE_DEFINE_ASSERT_MACRO)
// Assert() is defined in unit_test/suite.h
# include "unit_test/suite.h"
# else
# define Assert(expr) \
{\
if (!(expr)) while (true);\
}
# endif
#else
# define Assert(expr) ((void) 0)
#endif
/*! \name Bit Reversing
*/
//! @{
/*! \brief Reverses the bits of \a u8.
*
* \param u8 U8 of which to reverse the bits.
*
* \return Value resulting from \a u8 with reversed bits.
*/
#define bit_reverse8(u8) ((U8)(bit_reverse32((U8)(u8)) >> 24))
/*! \brief Reverses the bits of \a u16.
*
* \param u16 U16 of which to reverse the bits.
*
* \return Value resulting from \a u16 with reversed bits.
*/
#define bit_reverse16(u16) ((U16)(bit_reverse32((U16)(u16)) >> 16))
/*! \brief Reverses the bits of \a u32.
*
* \param u32 U32 of which to reverse the bits.
*
* \return Value resulting from \a u32 with reversed bits.
*/
#if (defined __GNUC__)
#define bit_reverse32(u32) \
(\
{\
unsigned int __value = (U32)(u32);\
__asm__ ("brev\t%0" : "+r" (__value) : : "cc");\
(U32)__value;\
}\
)
#elif (defined __ICCAVR__)
#define bit_reverse32(u32) ((U32)__bit_reverse((U32)(u32)))
#endif
/*! \brief Reverses the bits of \a u64.
*
* \param u64 U64 of which to reverse the bits.
*
* \return Value resulting from \a u64 with reversed bits.
*/
#define bit_reverse64(u64) ((U64)(((U64)bit_reverse32((U64)(u64) >> 32)) |\
((U64)bit_reverse32((U64)(u64)) << 32)))
//! @}
//! \name Logarithmic functions
//! @{
/**
* \internal
* Undefined function. Will cause a link failure if ilog2() is called
* with an invalid constant value.
*/
int_fast8_t ilog2_undefined(void);
/**
* \brief Calculate the base-2 logarithm of a number rounded down to
* the nearest integer.
*
* \param x A 32-bit value
* \return The base-2 logarithm of \a x, or -1 if \a x is 0.
*/
static inline int_fast8_t ilog2(uint32_t x)
{
if (is_constant(x))
return ((x) & (1ULL << 31) ? 31 :
(x) & (1ULL << 30) ? 30 :
(x) & (1ULL << 29) ? 29 :
(x) & (1ULL << 28) ? 28 :
(x) & (1ULL << 27) ? 27 :
(x) & (1ULL << 26) ? 26 :
(x) & (1ULL << 25) ? 25 :
(x) & (1ULL << 24) ? 24 :
(x) & (1ULL << 23) ? 23 :
(x) & (1ULL << 22) ? 22 :
(x) & (1ULL << 21) ? 21 :
(x) & (1ULL << 20) ? 20 :
(x) & (1ULL << 19) ? 19 :
(x) & (1ULL << 18) ? 18 :
(x) & (1ULL << 17) ? 17 :
(x) & (1ULL << 16) ? 16 :
(x) & (1ULL << 15) ? 15 :
(x) & (1ULL << 14) ? 14 :
(x) & (1ULL << 13) ? 13 :
(x) & (1ULL << 12) ? 12 :
(x) & (1ULL << 11) ? 11 :
(x) & (1ULL << 10) ? 10 :
(x) & (1ULL << 9) ? 9 :
(x) & (1ULL << 8) ? 8 :
(x) & (1ULL << 7) ? 7 :
(x) & (1ULL << 6) ? 6 :
(x) & (1ULL << 5) ? 5 :
(x) & (1ULL << 4) ? 4 :
(x) & (1ULL << 3) ? 3 :
(x) & (1ULL << 2) ? 2 :
(x) & (1ULL << 1) ? 1 :
(x) & (1ULL << 0) ? 0 :
ilog2_undefined());
return 31 - clz(x);
}
//! @}
/*! \name Alignment
*/
//! @{
/*! \brief Tests alignment of the number \a val with the \a n boundary.
*
* \param val Input value.
* \param n Boundary.
*
* \return \c 1 if the number \a val is aligned with the \a n boundary, else \c 0.
*/
#define Test_align(val, n ) (!Tst_bits( val, (n) - 1 ) )
/*! \brief Gets alignment of the number \a val with respect to the \a n boundary.
*
* \param val Input value.
* \param n Boundary.
*
* \return Alignment of the number \a val with respect to the \a n boundary.
*/
#define Get_align( val, n ) ( Rd_bits( val, (n) - 1 ) )
/*! \brief Sets alignment of the lvalue number \a lval to \a alg with respect to the \a n boundary.
*
* \param lval Input/output lvalue.
* \param n Boundary.
* \param alg Alignment.
*
* \return New value of \a lval resulting from its alignment set to \a alg with respect to the \a n boundary.
*/
#define Set_align(lval, n, alg) ( Wr_bits(lval, (n) - 1, alg) )
/*! \brief Aligns the number \a val with the upper \a n boundary.
*
* \param val Input value.
* \param n Boundary.
*
* \return Value resulting from the number \a val aligned with the upper \a n boundary.
*/
#define Align_up( val, n ) (((val) + ((n) - 1)) & ~((n) - 1))
/*! \brief Aligns the number \a val with the lower \a n boundary.
*
* \param val Input value.
* \param n Boundary.
*
* \return Value resulting from the number \a val aligned with the lower \a n boundary.
*/
#define Align_down(val, n ) ( (val) & ~((n) - 1))
//! @}
/*! \name Mathematics
*
* Compiler optimization for non-constant expressions, only for abs under WinAVR
*/
//! @{
/*! \brief Takes the absolute value of \a a.
*
* \param a Input value.
*
* \return Absolute value of \a a.
*
* \note More optimized if only used with values known at compile time.
*/
#define Abs(a) (((a) < 0 ) ? -(a) : (a))
#ifndef abs
#define abs(a) Abs(a)
#endif
/*! \brief Takes the minimal value of \a a and \a b.
*
* \param a Input value.
* \param b Input value.
*
* \return Minimal value of \a a and \a b.
*
* \note More optimized if only used with values known at compile time.
*/
#define Min(a, b) (((a) < (b)) ? (a) : (b))
#define min(a, b) Min(a, b)
/*! \brief Takes the maximal value of \a a and \a b.
*
* \param a Input value.
* \param b Input value.
*
* \return Maximal value of \a a and \a b.
*
* \note More optimized if only used with values known at compile time.
*/
#define Max(a, b) (((a) > (b)) ? (a) : (b))
#define max(a, b) Max(a, b)
//! @}
/*! \brief Calls the routine at address \a addr.
*
* It generates a long call opcode.
*
* For example, `Long_call(0x80000000)' generates a software reset on a UC3 if
* it is invoked from the CPU supervisor mode.
*
* \param addr Address of the routine to call.
*
* \note It may be used as a long jump opcode in some special cases.
*/
#define Long_call(addr) ((*(void (*)(void))(addr))())
/*! \name System Register Access
*/
//! @{
/*! \brief Gets the value of the \a sysreg system register.
*
* \param sysreg Address of the system register of which to get the value.
*
* \return Value of the \a sysreg system register.
*/
#if (defined __GNUC__)
#define Get_system_register(sysreg) __builtin_mfsr(sysreg)
#elif (defined __ICCAVR__)
#define Get_system_register(sysreg) __get_system_register(sysreg)
#endif
/*! \brief Sets the value of the \a sysreg system register to \a value.
*
* \param sysreg Address of the system register of which to set the value.
* \param value Value to set the \a sysreg system register to.
*/
#if (defined __GNUC__)
#define Set_system_register(sysreg, value) __builtin_mtsr(sysreg, value)
#elif (defined __ICCAVR__)
#define Set_system_register(sysreg, value) __set_system_register(sysreg, value)
#endif
//! @}
/*! \name Debug Register Access
*/
//! @{
/*! \brief Gets the value of the \a dbgreg debug register.
*
* \param dbgreg Address of the debug register of which to get the value.
*
* \return Value of the \a dbgreg debug register.
*/
#if (defined __GNUC__)
#define Get_debug_register(dbgreg) __builtin_mfdr(dbgreg)
#elif (defined __ICCAVR__)
#define Get_debug_register(dbgreg) __get_debug_register(dbgreg)
#endif
/*! \brief Sets the value of the \a dbgreg debug register to \a value.
*
* \param dbgreg Address of the debug register of which to set the value.
* \param value Value to set the \a dbgreg debug register to.
*/
#if (defined __GNUC__)
#define Set_debug_register(dbgreg, value) __builtin_mtdr(dbgreg, value)
#elif (defined __ICCAVR__)
#define Set_debug_register(dbgreg, value) __set_debug_register(dbgreg, value)
#endif
//! @}
/*! \name MCU Endianism Handling
* xmega is a MCU little endianism.
*/
//! @{
#define MSB(u16) (((uint8_t* )&u16)[1])
#define LSB(u16) (((uint8_t* )&u16)[0])
#define MSW(u32) (((uint16_t*)&u32)[1])
#define LSW(u32) (((uint16_t*)&u32)[0])
#define MSB0W(u32) (((uint8_t*)&(u32))[3]) //!< Most significant byte of 1st rank of \a u32.
#define MSB1W(u32) (((uint8_t*)&(u32))[2]) //!< Most significant byte of 2nd rank of \a u32.
#define MSB2W(u32) (((uint8_t*)&(u32))[1]) //!< Most significant byte of 3rd rank of \a u32.
#define MSB3W(u32) (((uint8_t*)&(u32))[0]) //!< Most significant byte of 4th rank of \a u32.
#define LSB3W(u32) MSB0W(u32) //!< Least significant byte of 4th rank of \a u32.
#define LSB2W(u32) MSB1W(u32) //!< Least significant byte of 3rd rank of \a u32.
#define LSB1W(u32) MSB2W(u32) //!< Least significant byte of 2nd rank of \a u32.
#define LSB0W(u32) MSB3W(u32) //!< Least significant byte of 1st rank of \a u32.
#define MSB0(u32) (((uint8_t*)&u32)[3])
#define MSB1(u32) (((uint8_t*)&u32)[2])
#define MSB2(u32) (((uint8_t*)&u32)[1])
#define MSB3(u32) (((uint8_t*)&u32)[0])
#define LSB0(u32) MSB3(u32)
#define LSB1(u32) MSB2(u32)
#define LSB2(u32) MSB1(u32)
#define LSB3(u32) MSB0(u32)
#define LE16(x) (x)
#define le16_to_cpu(x) (x)
#define cpu_to_le16(x) (x)
#define LE16_TO_CPU(x) (x)
#define CPU_TO_LE16(x) (x)
#define BE16(x) Swap16(x)
#define be16_to_cpu(x) swap16(x)
#define cpu_to_be16(x) swap16(x)
#define BE16_TO_CPU(x) Swap16(x)
#define CPU_TO_BE16(x) Swap16(x)
#define LE32(x) (x)
#define le32_to_cpu(x) (x)
#define cpu_to_le32(x) (x)
#define LE32_TO_CPU(x) (x)
#define CPU_TO_LE32(x) (x)
#define BE32(x) Swap32(x)
#define be32_to_cpu(x) swap32(x)
#define cpu_to_be32(x) swap32(x)
#define BE32_TO_CPU(x) Swap32(x)
#define CPU_TO_BE32(x) Swap32(x)
//! @}
/*! \name Endianism Conversion
*
* The same considerations as for clz and ctz apply here but AVR32-GCC's
* __builtin_bswap_16 and __builtin_bswap_32 do not behave like macros when
* applied to constant expressions, so two sets of macros are defined here:
* - Swap16, Swap32 and Swap64 to apply to constant expressions (values known
* at compile time);
* - swap16, swap32 and swap64 to apply to non-constant expressions (values
* unknown at compile time).
*/
//! @{
/*! \brief Toggles the endianism of \a u16 (by swapping its bytes).
*
* \param u16 U16 of which to toggle the endianism.
*
* \return Value resulting from \a u16 with toggled endianism.
*
* \note More optimized if only used with values known at compile time.
*/
#define Swap16(u16) ((U16)(((U16)(u16) >> 8) |\
((U16)(u16) << 8)))
/*! \brief Toggles the endianism of \a u32 (by swapping its bytes).
*
* \param u32 U32 of which to toggle the endianism.
*
* \return Value resulting from \a u32 with toggled endianism.
*
* \note More optimized if only used with values known at compile time.
*/
#define Swap32(u32) ((U32)(((U32)Swap16((U32)(u32) >> 16)) |\
((U32)Swap16((U32)(u32)) << 16)))
/*! \brief Toggles the endianism of \a u64 (by swapping its bytes).
*
* \param u64 U64 of which to toggle the endianism.
*
* \return Value resulting from \a u64 with toggled endianism.
*
* \note More optimized if only used with values known at compile time.
*/
#define Swap64(u64) ((U64)(((U64)Swap32((U64)(u64) >> 32)) |\
((U64)Swap32((U64)(u64)) << 32)))
/*! \brief Toggles the endianism of \a u16 (by swapping its bytes).
*
* \param u16 U16 of which to toggle the endianism.
*
* \return Value resulting from \a u16 with toggled endianism.
*
* \note More optimized if only used with values unknown at compile time.
*/
#define swap16(u16) Swap16(u16)
/*! \brief Toggles the endianism of \a u32 (by swapping its bytes).
*
* \param u32 U32 of which to toggle the endianism.
*
* \return Value resulting from \a u32 with toggled endianism.
*
* \note More optimized if only used with values unknown at compile time.
*/
#define swap32(u32) Swap32(u32)
/*! \brief Toggles the endianism of \a u64 (by swapping its bytes).
*
* \param u64 U64 of which to toggle the endianism.
*
* \return Value resulting from \a u64 with toggled endianism.
*
* \note More optimized if only used with values unknown at compile time.
*/
#define swap64(u64) ((U64)(((U64)swap32((U64)(u64) >> 32)) |\
((U64)swap32((U64)(u64)) << 32)))
//! @}
/*! \name Target Abstraction
*/
//! @{
#define _GLOBEXT_ extern //!< extern storage-class specifier.
#define _CONST_TYPE_ const //!< const type qualifier.
#define _MEM_TYPE_SLOW_ //!< Slow memory type.
#define _MEM_TYPE_MEDFAST_ //!< Fairly fast memory type.
#define _MEM_TYPE_FAST_ //!< Fast memory type.
typedef U8 Byte; //!< 8-bit unsigned integer.
#define memcmp_ram2ram memcmp //!< Target-specific memcmp of RAM to RAM.
#define memcmp_code2ram memcmp //!< Target-specific memcmp of RAM to NVRAM.
#define memcpy_ram2ram memcpy //!< Target-specific memcpy from RAM to RAM.
#define memcpy_code2ram memcpy //!< Target-specific memcpy from NVRAM to RAM.
//! @}
/**
* \brief Calculate \f$ \left\lceil \frac{a}{b} \right\rceil \f$ using
* integer arithmetic.
*
* \param a An integer
* \param b Another integer
*
* \return (\a a / \a b) rounded up to the nearest integer.
*/
#define div_ceil(a, b) (((a) + (b) - 1) / (b))
#include "preprocessor.h"
#include "progmem.h"
#include "interrupt.h"
#if (defined __GNUC__)
#define SHORTENUM __attribute__ ((packed))
#elif (defined __ICCAVR__)
#define SHORTENUM /**/
#endif
#if (defined __GNUC__)
#define FUNC_PTR void *
#elif (defined __ICCAVR__)
#if (FLASHEND > 0x1FFFF) // Required for program code larger than 128K
#define FUNC_PTR void __farflash *
#else
#define FUNC_PTR void *
#endif /* ENABLE_FAR_FLASH */
#endif
#if (defined __GNUC__)
#define FLASH_DECLARE(x) const x __attribute__((__progmem__))
#elif (defined __ICCAVR__)
#define FLASH_DECLARE(x) const __flash x
#endif
#if (defined __GNUC__)
#define FLASH_EXTERN(x) extern const x
#elif (defined __ICCAVR__)
#define FLASH_EXTERN(x) extern const __flash x
#endif
/*Defines the Flash Storage for the request and response of MAC*/
#define CMD_ID_OCTET (0)
/* Converting of values from CPU endian to little endian. */
#define CPU_ENDIAN_TO_LE16(x) (x)
#define CPU_ENDIAN_TO_LE32(x) (x)
#define CPU_ENDIAN_TO_LE64(x) (x)
/* Converting of values from little endian to CPU endian. */
#define LE16_TO_CPU_ENDIAN(x) (x)
#define LE32_TO_CPU_ENDIAN(x) (x)
#define LE64_TO_CPU_ENDIAN(x) (x)
/* Converting of constants from little endian to CPU endian. */
#define CLE16_TO_CPU_ENDIAN(x) (x)
#define CLE32_TO_CPU_ENDIAN(x) (x)
#define CLE64_TO_CPU_ENDIAN(x) (x)
/* Converting of constants from CPU endian to little endian. */
#define CCPU_ENDIAN_TO_LE16(x) (x)
#define CCPU_ENDIAN_TO_LE32(x) (x)
#define CCPU_ENDIAN_TO_LE64(x) (x)
#if (defined __GNUC__)
#define ADDR_COPY_DST_SRC_16(dst, src) memcpy((&(dst)), (&(src)), sizeof(uint16_t))
#define ADDR_COPY_DST_SRC_64(dst, src) memcpy((&(dst)), (&(src)), sizeof(uint64_t))
/* Converts a 2 Byte array into a 16-Bit value */
#define convert_byte_array_to_16_bit(data) \
(*(uint16_t *)(data))
/* Converts a 4 Byte array into a 32-Bit value */
#define convert_byte_array_to_32_bit(data) \
(*(uint32_t *)(data))
/* Converts a 8 Byte array into a 64-Bit value */
#define convert_byte_array_to_64_bit(data) \
(*(uint64_t *)(data))
/* Converts a 16-Bit value into a 2 Byte array */
#define convert_16_bit_to_byte_array(value, data) \
((*(uint16_t *)(data)) = (uint16_t)(value))
/* Converts spec 16-Bit value into a 2 Byte array */
#define convert_spec_16_bit_to_byte_array(value, data) \
((*(uint16_t *)(data)) = (uint16_t)(value))
/* Converts spec 16-Bit value into a 2 Byte array */
#define convert_16_bit_to_byte_address(value, data) \
((*(uint16_t *)(data)) = (uint16_t)(value))
/* Converts a 32-Bit value into a 4 Byte array */
#define convert_32_bit_to_byte_array(value, data) \
((*(uint32_t *)(data)) = (uint32_t)(value))
/* Converts a 64-Bit value into a 8 Byte array */
/* Here memcpy requires much less footprint */
#define convert_64_bit_to_byte_array(value, data) \
memcpy((data), (&(value)), sizeof(uint64_t))
#elif (defined __ICCAVR__)
#define ADDR_COPY_DST_SRC_16(dst, src) ((dst) = (src))
#define ADDR_COPY_DST_SRC_64(dst, src) ((dst) = (src))
/* Converts a 2 Byte array into a 16-Bit value */
#define convert_byte_array_to_16_bit(data) \
(*(uint16_t *)(data))
/* Converts a 4 Byte array into a 32-Bit value */
#define convert_byte_array_to_32_bit(data) \
(*(uint32_t *)(data))
/* Converts a 8 Byte array into a 64-Bit value */
#define convert_byte_array_to_64_bit(data) \
(*(uint64_t *)(data))
/* Converts a 16-Bit value into a 2 Byte array */
#define convert_16_bit_to_byte_array(value, data) \
((*(uint16_t *)(data)) = (uint16_t)(value))
/* Converts spec 16-Bit value into a 2 Byte array */
#define convert_spec_16_bit_to_byte_array(value, data) \
((*(uint16_t *)(data)) = (uint16_t)(value))
/* Converts spec 16-Bit value into a 2 Byte array */
#define convert_16_bit_to_byte_address(value, data) \
((*(uint16_t *)(data)) = (uint16_t)(value))
/* Converts a 32-Bit value into a 4 Byte array */
#define convert_32_bit_to_byte_array(value, data) \
((*(uint32_t *)(data)) = (uint32_t)(value))
/* Converts a 64-Bit value into a 8 Byte array */
#define convert_64_bit_to_byte_array(value, data) \
((*(uint64_t *)(data)) = (uint64_t)(value))
#endif
#define MEMCPY_ENDIAN memcpy
#define PGM_READ_BLOCK(dst, src, len) memcpy_P((dst), (src), (len))
#if (defined __GNUC__)
#define PGM_READ_BYTE(x) pgm_read_byte(x)
#define PGM_READ_WORD(x) pgm_read_word(x)
#elif (defined __ICCAVR__)
#define PGM_READ_BYTE(x) *(x)
#define PGM_READ_WORD(x) *(x)
#endif
#if (defined __GNUC__)
#define nop() do { __asm__ __volatile__ ("nop"); } while (0)
#elif (defined __ICCAVR__)
#define nop() __no_operation()
#endif
/**
* \}
*/
#endif // UTILS_COMPILER_H