/* * This file is subject to the terms of the GFX License. If a copy of * the license was not distributed with this file, you can obtain one at: * * http://ugfx.org/license.html */ /** * @file src/gmisc/sys_defs.h * @brief GMISC - Miscellaneous Routines header file. * * @addtogroup GMISC * * @brief Module which contains different features such as array conversions * * @{ */ #ifndef _GMISC_H #define _GMISC_H #include "gfx.h" /*===========================================================================*/ /* Type definitions */ /*===========================================================================*/ /** * @brief Sample data formats * @note These are defined regardless of whether you use the GMISC module * or not as they are used in lots of places. */ typedef enum ArrayDataFormat_e { ARRAY_DATA_4BITUNSIGNED = 4, ARRAY_DATA_4BITSIGNED = 5, ARRAY_DATA_8BITUNSIGNED = 8, ARRAY_DATA_8BITSIGNED = 9, ARRAY_DATA_10BITUNSIGNED = 10, ARRAY_DATA_10BITSIGNED = 11, ARRAY_DATA_12BITUNSIGNED = 12, ARRAY_DATA_12BITSIGNED = 13, ARRAY_DATA_14BITUNSIGNED = 14, ARRAY_DATA_14BITSIGNED = 15, ARRAY_DATA_16BITUNSIGNED = 16, ARRAY_DATA_16BITSIGNED = 17, } ArrayDataFormat; /** * @brief Is the sample data format a "signed" data format? */ #define gfxSampleFormatIsSigned(fmt) ((fmt) & 1) /** * @brief How many bits are in the sample data format */ #define gfxSampleFormatBits(fmt) ((fmt) & ~1) /** * @brief The type for a fixed point type. * @details The top 16 bits are the integer component, the bottom 16 bits are the real component. */ typedef int32_t fixed; /** * @brief Macros to convert to and from a fixed point. * @{ */ #define FIXED(x) ((fixed)(x)<<16) /* @< integer to fixed */ #define NONFIXED(x) ((x)>>16) /* @< fixed to integer */ #define FIXED0_5 32768 /* @< 0.5 as a fixed (used for rounding) */ #define FP2FIXED(x) ((fixed)((x)*65536.0)) /* @< floating point to fixed */ #define FIXED2FP(x) ((double)(x)/65536.0) /* @< fixed to floating point */ /** @} */ /** * @brief The famous number pi */ #define PI 3.1415926535897932384626433832795028841971693993751 /** * @brief pi as a fixed point */ #define FIXED_PI FP2FIXED(PI) /*===========================================================================*/ /* External declarations. */ /*===========================================================================*/ #if GFX_USE_GMISC || defined(__DOXYGEN__) #ifdef __cplusplus extern "C" { #endif #if GMISC_NEED_ARRAYOPS || defined(__DOXYGEN__) /** * @brief Convert from one array format to another array format. * * @param[in] srcfmt The format of the source array * @param[in] src The source array * @param[in] dstfmt The format of the destination array * @param[in] dst The dstination array * @param[in] cnt The number of array elements to convert * * @note Assumes the destination buffer is large enough for the resultant data. * @note This routine is optimised to perform as fast as possible. * @note No type checking is performed on the source format. It is assumed to * have only valid values eg. ARRAY_DATA_4BITSIGNED will have values * 0000 -> 0111 for positive numbers and 1111 -> 1000 for negative numbers * Bits 5 -> 8 in the storage byte are treated in an undefined manner. * @note If srcfmt or dstfmt is an unknown format, this routine does nothing * with no warning that something is wrong * * @api */ void gmiscArrayConvert(ArrayDataFormat srcfmt, void *src, ArrayDataFormat dstfmt, void *dst, size_t cnt); #if 0 void gmiscArrayTranslate(ArrayDataFormat fmt, void *src, void *dst, size_t cnt, int trans); void gmiscArrayMultiply(ArrayDataFormat fmt, void *src, void *dst, size_t cnt, int mult); void gmiscArrayDivide(ArrayDataFormat fmt, void *src, void *dst, size_t cnt, int mdiv); void gmiscArrayMultDiv(ArrayDataFormat fmt, void *src, void *dst, size_t cnt, int mult, int div); void gmiscArrayAdd(ArrayDataFormat fmt, void *src1, void *src2, void *dst, size_t cnt); void gmiscArrayAddNoOverflow(ArrayDataFormat fmt, void *src1, void *src2, void *dst, size_t cnt); #endif #endif #if GMISC_NEED_FASTTRIG || defined(__DOXYGEN__) extern const double sintabledouble[]; /** * @brief Fast Table Based Trig functions * @return A double in the range -1.0 .. 0.0 .. 1.0 * * @param[in] degrees The angle in degrees (not radians) * * @note These functions use degrees rather than radians to describe the angle. * * @api * @{ */ double fsin(int degrees); double fcos(int degrees); /** @} * * @brief Fast Table Based Trig functions * @return A double in the range -1.0 .. 0.0 .. 1.0 * * @param[in] degrees The angle in degrees 0 .. 359 * * @note These functions use degrees rather than radians to describe the angle. * @note These functions are super fast but require the parameter to be in range. * Use the lowercase functions if the parameter may not be in range or if a * required trig function is not supported in this form. * * @api * @{ */ #define FSIN(degrees) sintabledouble[degrees]; /** @} */ #endif #if GMISC_NEED_FIXEDTRIG || defined(__DOXYGEN__) extern const fixed sintablefixed[]; /** * @brief Fast Table Based Trig functions * @return A fixed point in the range -1.0 .. 0.0 .. 1.0 * * @param[in] degrees The angle in degrees (not radians) * * @note These functions use degrees rather than radians to describe the angle. * * @api * @{ */ fixed ffsin(int degrees); fixed ffcos(int degrees); /** @} * * @brief Fast Table Based Trig functions * @return A fixed point in the range -1.0 .. 0.0 .. 1.0 * * @param[in] degrees The angle in degrees 0 .. 359 * * @note These functions use degrees rather than radians to describe the angle. * @note These functions are super fast but require the parameter to be in range. * Use the lowercase functions if the parameter may not be in range or if a * required trig function is not supported in this form. * * @api * @{ */ #define FFSIN(degrees) sintablefixed[degrees]; /** @} */ #endif #if GMISC_NEED_INVSQRT /** * @brief Fast inverse square root function (x^-1/2) * @return The approximate inverse square root * * @param[in] n The number to find the inverse square root of * * @note This function generates an approximate result. Higher accuracy (at the expense * of speed) can be obtained by modifying the source code (the necessary line * is already there - just commented out). * @note This function relies on the internal machine format of a float and a long. * If your machine architecture is very unusual this function may not work. * * @api */ float invsqrt(float n); #endif #ifdef __cplusplus } #endif #endif /* GFX_USE_MISC */ #endif /* _GMISC_H */ /** @} */