Subversion Repositories AFRtranscoder

/******************************************************************************

 * @file     arm_math.h

 * @brief    Public header file for CMSIS DSP LibraryU

 * @version  V1.5.3

 * @date     10. January 2018

 ******************************************************************************/

/*

 * Copyright (c) 2010-2018 Arm Limited or its affiliates. All rights reserved.

 *

 * SPDX-License-Identifier: Apache-2.0

 *

 * Licensed under the Apache License, Version 2.0 (the License); you may

 * not use this file except in compliance with the License.

 * You may obtain a copy of the License at

 *

 * www.apache.org/licenses/LICENSE-2.0

 *

 * Unless required by applicable law or agreed to in writing, software

 * distributed under the License is distributed on an AS IS BASIS, WITHOUT

 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

 * See the License for the specific language governing permissions and

 * limitations under the License.

 */

/**

   \mainpage CMSIS DSP Software Library

   *

   * Introduction

   * ------------

   *

   * This user manual describes the CMSIS DSP software library,

   * a suite of common signal processing functions for use on Cortex-M processor based devices.

   *

   * The library is divided into a number of functions each covering a specific category:

   * - Basic math functions

   * - Fast math functions

   * - Complex math functions

   * - Filters

   * - Matrix functions

   * - Transforms

   * - Motor control functions

   * - Statistical functions

   * - Support functions

   * - Interpolation functions

   *

   * The library has separate functions for operating on 8-bit integers, 16-bit integers,

   * 32-bit integer and 32-bit floating-point values.

   *

   * Using the Library

   * ------------

   *

   * The library installer contains prebuilt versions of the libraries in the <code>Lib</code> folder.

   * - arm_cortexM7lfdp_math.lib (Cortex-M7, Little endian, Double Precision Floating Point Unit)

   * - arm_cortexM7bfdp_math.lib (Cortex-M7, Big endian, Double Precision Floating Point Unit)

   * - arm_cortexM7lfsp_math.lib (Cortex-M7, Little endian, Single Precision Floating Point Unit)

   * - arm_cortexM7bfsp_math.lib (Cortex-M7, Big endian and Single Precision Floating Point Unit on)

   * - arm_cortexM7l_math.lib (Cortex-M7, Little endian)

   * - arm_cortexM7b_math.lib (Cortex-M7, Big endian)

   * - arm_cortexM4lf_math.lib (Cortex-M4, Little endian, Floating Point Unit)

   * - arm_cortexM4bf_math.lib (Cortex-M4, Big endian, Floating Point Unit)

   * - arm_cortexM4l_math.lib (Cortex-M4, Little endian)

   * - arm_cortexM4b_math.lib (Cortex-M4, Big endian)

   * - arm_cortexM3l_math.lib (Cortex-M3, Little endian)

   * - arm_cortexM3b_math.lib (Cortex-M3, Big endian)

   * - arm_cortexM0l_math.lib (Cortex-M0 / Cortex-M0+, Little endian)

   * - arm_cortexM0b_math.lib (Cortex-M0 / Cortex-M0+, Big endian)

   * - arm_ARMv8MBLl_math.lib (Armv8-M Baseline, Little endian)

   * - arm_ARMv8MMLl_math.lib (Armv8-M Mainline, Little endian)

   * - arm_ARMv8MMLlfsp_math.lib (Armv8-M Mainline, Little endian, Single Precision Floating Point Unit)

   * - arm_ARMv8MMLld_math.lib (Armv8-M Mainline, Little endian, DSP instructions)

   * - arm_ARMv8MMLldfsp_math.lib (Armv8-M Mainline, Little endian, DSP instructions, Single Precision Floating Point Unit)

   *

   * The library functions are declared in the public file <code>arm_math.h</code> which is placed in the <code>Include</code> folder.

   * Simply include this file and link the appropriate library in the application and begin calling the library functions. The Library supports single

   * public header file <code> arm_math.h</code> for Cortex-M cores with little endian and big endian. Same header file will be used for floating point unit(FPU) variants.

   * Define the appropriate preprocessor macro ARM_MATH_CM7 or ARM_MATH_CM4 or ARM_MATH_CM3 or

   * ARM_MATH_CM0 or ARM_MATH_CM0PLUS depending on the target processor in the application.

   * For Armv8-M cores define preprocessor macro ARM_MATH_ARMV8MBL or ARM_MATH_ARMV8MML.

   * Set preprocessor macro __DSP_PRESENT if Armv8-M Mainline core supports DSP instructions.

   *

   *

   * Examples

   * --------

   *

   * The library ships with a number of examples which demonstrate how to use the library functions.

   *

   * Toolchain Support

   * ------------

   *

   * The library has been developed and tested with MDK version 5.14.0.0

   * The library is being tested in GCC and IAR toolchains and updates on this activity will be made available shortly.

   *

   * Building the Library

   * ------------

   *

   * The library installer contains a project file to rebuild libraries on MDK toolchain in the <code>CMSIS\\DSP_Lib\\Source\\ARM</code> folder.

   * - arm_cortexM_math.uvprojx

   *

   *

   * The libraries can be built by opening the arm_cortexM_math.uvprojx project in MDK-ARM, selecting a specific target, and defining the optional preprocessor macros detailed above.

   *

   * Preprocessor Macros

   * ------------

   *

   * Each library project have different preprocessor macros.

   *

   * - UNALIGNED_SUPPORT_DISABLE:

   *

   * Define macro UNALIGNED_SUPPORT_DISABLE, If the silicon does not support unaligned memory access

   *

   * - ARM_MATH_BIG_ENDIAN:

   *

   * Define macro ARM_MATH_BIG_ENDIAN to build the library for big endian targets. By default library builds for little endian targets.

   *

   * - ARM_MATH_MATRIX_CHECK:

   *

   * Define macro ARM_MATH_MATRIX_CHECK for checking on the input and output sizes of matrices

   *

   * - ARM_MATH_ROUNDING:

   *

   * Define macro ARM_MATH_ROUNDING for rounding on support functions

   *

   * - ARM_MATH_CMx:

   *

   * Define macro ARM_MATH_CM4 for building the library on Cortex-M4 target, ARM_MATH_CM3 for building library on Cortex-M3 target

   * and ARM_MATH_CM0 for building library on Cortex-M0 target, ARM_MATH_CM0PLUS for building library on Cortex-M0+ target, and

   * ARM_MATH_CM7 for building the library on cortex-M7.

   *

   * - ARM_MATH_ARMV8MxL:

   *

   * Define macro ARM_MATH_ARMV8MBL for building the library on Armv8-M Baseline target, ARM_MATH_ARMV8MML for building library

   * on Armv8-M Mainline target.

   *

   * - __FPU_PRESENT:

   *

   * Initialize macro __FPU_PRESENT = 1 when building on FPU supported Targets. Enable this macro for floating point libraries.

   *

   * - __DSP_PRESENT:

   *

   * Initialize macro __DSP_PRESENT = 1 when Armv8-M Mainline core supports DSP instructions.

   *

   * <hr>

   * CMSIS-DSP in ARM::CMSIS Pack

   * -----------------------------

   *

   * The following files relevant to CMSIS-DSP are present in the <b>ARM::CMSIS</b> Pack directories:

   * |File/Folder                   |Content                                                                 |

   * |------------------------------|------------------------------------------------------------------------|

   * |\b CMSIS\\Documentation\\DSP  | This documentation                                                     |

   * |\b CMSIS\\DSP_Lib             | Software license agreement (license.txt)                               |

   * |\b CMSIS\\DSP_Lib\\Examples   | Example projects demonstrating the usage of the library functions      |

   * |\b CMSIS\\DSP_Lib\\Source     | Source files for rebuilding the library                                |

   *

   * <hr>

   * Revision History of CMSIS-DSP

   * ------------

   * Please refer to \ref ChangeLog_pg.

   *

   * Copyright Notice

   * ------------

   *

   * Copyright (C) 2010-2015 Arm Limited. All rights reserved.

   */

/**

 * @defgroup groupMath Basic Math Functions

 */

/**

 * @defgroup groupFastMath Fast Math Functions

 * This set of functions provides a fast approximation to sine, cosine, and square root.

 * As compared to most of the other functions in the CMSIS math library, the fast math functions

 * operate on individual values and not arrays.

 * There are separate functions for Q15, Q31, and floating-point data.

 *

 */

/**

 * @defgroup groupCmplxMath Complex Math Functions

 * This set of functions operates on complex data vectors.

 * The data in the complex arrays is stored in an interleaved fashion

 * (real, imag, real, imag, ...).

 * In the API functions, the number of samples in a complex array refers

 * to the number of complex values; the array contains twice this number of

 * real values.

 */

/**

 * @defgroup groupFilters Filtering Functions

 */

/**

 * @defgroup groupMatrix Matrix Functions

 *

 * This set of functions provides basic matrix math operations.

 * The functions operate on matrix data structures.  For example,

 * the type

 * definition for the floating-point matrix structure is shown

 * below:

 * <pre>

 *     typedef struct

 *     {

 *       uint16_t numRows;     // number of rows of the matrix.

 *       uint16_t numCols;     // number of columns of the matrix.

 *       float32_t *pData;     // points to the data of the matrix.

 *     } arm_matrix_instance_f32;

 * </pre>

 * There are similar definitions for Q15 and Q31 data types.

 *

 * The structure specifies the size of the matrix and then points to

 * an array of data.  The array is of size <code>numRows X numCols</code>

 * and the values are arranged in row order.  That is, the

 * matrix element (i, j) is stored at:

 * <pre>

 *     pData[i*numCols + j]

 * </pre>

 *

 * \par Init Functions

 * There is an associated initialization function for each type of matrix

 * data structure.

 * The initialization function sets the values of the internal structure fields.

 * Refer to the function <code>arm_mat_init_f32()</code>, <code>arm_mat_init_q31()</code>

 * and <code>arm_mat_init_q15()</code> for floating-point, Q31 and Q15 types,  respectively.

 *

 * \par

 * Use of the initialization function is optional. However, if initialization function is used

 * then the instance structure cannot be placed into a const data section.

 * To place the instance structure in a const data

 * section, manually initialize the data structure.  For example:

 * <pre>

 * <code>arm_matrix_instance_f32 S = {nRows, nColumns, pData};</code>

 * <code>arm_matrix_instance_q31 S = {nRows, nColumns, pData};</code>

 * <code>arm_matrix_instance_q15 S = {nRows, nColumns, pData};</code>

 * </pre>

 * where <code>nRows</code> specifies the number of rows, <code>nColumns</code>

 * specifies the number of columns, and <code>pData</code> points to the

 * data array.

 *

 * \par Size Checking

 * By default all of the matrix functions perform size checking on the input and

 * output matrices. For example, the matrix addition function verifies that the

 * two input matrices and the output matrix all have the same number of rows and

 * columns. If the size check fails the functions return:

 * <pre>

 *     ARM_MATH_SIZE_MISMATCH

 * </pre>

 * Otherwise the functions return

 * <pre>

 *     ARM_MATH_SUCCESS

 * </pre>

 * There is some overhead associated with this matrix size checking.

 * The matrix size checking is enabled via the \#define

 * <pre>

 *     ARM_MATH_MATRIX_CHECK

 * </pre>

 * within the library project settings.  By default this macro is defined

 * and size checking is enabled. By changing the project settings and

 * undefining this macro size checking is eliminated and the functions

 * run a bit faster. With size checking disabled the functions always

 * return <code>ARM_MATH_SUCCESS</code>.

 */

/**

 * @defgroup groupTransforms Transform Functions

 */

/**

 * @defgroup groupController Controller Functions

 */

/**

 * @defgroup groupStats Statistics Functions

 */

/**

 * @defgroup groupSupport Support Functions

 */

/**

 * @defgroup groupInterpolation Interpolation Functions

 * These functions perform 1- and 2-dimensional interpolation of data.

 * Linear interpolation is used for 1-dimensional data and

 * bilinear interpolation is used for 2-dimensional data.

 */

/**

 * @defgroup groupExamples Examples

 */

#ifndef _ARM_MATH_H

#define _ARM_MATH_H

/* Compiler specific diagnostic adjustment */

#if   defined ( __CC_ARM )

#elif defined ( __ARMCC_VERSION ) && ( __ARMCC_VERSION >= 6010050 )

#elif defined ( __GNUC__ )

#pragma GCC diagnostic push

#pragma GCC diagnostic ignored "-Wsign-conversion"

#pragma GCC diagnostic ignored "-Wconversion"

#pragma GCC diagnostic ignored "-Wunused-parameter"

#elif defined ( __ICCARM__ )

#elif defined ( __TI_ARM__ )

#elif defined ( __CSMC__ )

#elif defined ( __TASKING__ )

#else

  #error Unknown compiler

#endif

#define __CMSIS_GENERIC         /* disable NVIC and Systick functions */

#if defined(ARM_MATH_CM7)

  #include "core_cm7.h"

  #define ARM_MATH_DSP

#elif defined (ARM_MATH_CM4)

  #include "core_cm4.h"

  #define ARM_MATH_DSP

#elif defined (ARM_MATH_CM3)

  #include "core_cm3.h"

#elif defined (ARM_MATH_CM0)

  #include "core_cm0.h"

  #define ARM_MATH_CM0_FAMILY

#elif defined (ARM_MATH_CM0PLUS)

  #include "core_cm0plus.h"

  #define ARM_MATH_CM0_FAMILY

#elif defined (ARM_MATH_ARMV8MBL)

  #include "core_armv8mbl.h"

  #define ARM_MATH_CM0_FAMILY

#elif defined (ARM_MATH_ARMV8MML)

  #include "core_armv8mml.h"

  #if (defined (__DSP_PRESENT) && (__DSP_PRESENT == 1))

    #define ARM_MATH_DSP

  #endif

#else

  #error "Define according the used Cortex core ARM_MATH_CM7, ARM_MATH_CM4, ARM_MATH_CM3, ARM_MATH_CM0PLUS, ARM_MATH_CM0, ARM_MATH_ARMV8MBL, ARM_MATH_ARMV8MML"

#endif

#undef  __CMSIS_GENERIC         /* enable NVIC and Systick functions */

#include "string.h"

#include "math.h"

#ifdef   __cplusplus

extern "C"

{

#endif

  /**

   * @brief Macros required for reciprocal calculation in Normalized LMS

   */

#define DELTA_Q31          (0x100)

#define DELTA_Q15          0x5

#define INDEX_MASK         0x0000003F

#ifndef PI

  #define PI               3.14159265358979f

#endif

  /**

   * @brief Macros required for SINE and COSINE Fast math approximations

   */

#define FAST_MATH_TABLE_SIZE  512

#define FAST_MATH_Q31_SHIFT   (32 - 10)

#define FAST_MATH_Q15_SHIFT   (16 - 10)

#define CONTROLLER_Q31_SHIFT  (32 - 9)

#define TABLE_SPACING_Q31     0x400000

#define TABLE_SPACING_Q15     0x80

  /**

   * @brief Macros required for SINE and COSINE Controller functions

   */

  /* 1.31(q31) Fixed value of 2/360 */

  /* -1 to +1 is divided into 360 values so total spacing is (2/360) */

#define INPUT_SPACING         0xB60B61

  /**

   * @brief Macro for Unaligned Support

   */

#ifndef UNALIGNED_SUPPORT_DISABLE

    #define ALIGN4

#else

  #if defined  (__GNUC__)

    #define ALIGN4 __attribute__((aligned(4)))

  #else

    #define ALIGN4 __align(4)

  #endif

#endif   /* #ifndef UNALIGNED_SUPPORT_DISABLE */

  /**

   * @brief Error status returned by some functions in the library.

   */

  typedef enum

  {

    ARM_MATH_SUCCESS = 0,                /**< No error */

    ARM_MATH_ARGUMENT_ERROR = -1,        /**< One or more arguments are incorrect */

    ARM_MATH_LENGTH_ERROR = -2,          /**< Length of data buffer is incorrect */

    ARM_MATH_SIZE_MISMATCH = -3,         /**< Size of matrices is not compatible with the operation. */

    ARM_MATH_NANINF = -4,                /**< Not-a-number (NaN) or infinity is generated */

    ARM_MATH_SINGULAR = -5,              /**< Generated by matrix inversion if the input matrix is singular and cannot be inverted. */

    ARM_MATH_TEST_FAILURE = -6           /**< Test Failed  */

  } arm_status;

  /**

   * @brief 8-bit fractional data type in 1.7 format.

   */

  typedef int8_t q7_t;

  /**

   * @brief 16-bit fractional data type in 1.15 format.

   */

  typedef int16_t q15_t;

  /**

   * @brief 32-bit fractional data type in 1.31 format.

   */

  typedef int32_t q31_t;

  /**

   * @brief 64-bit fractional data type in 1.63 format.

   */

  typedef int64_t q63_t;

  /**

   * @brief 32-bit floating-point type definition.

   */

  typedef float float32_t;

  /**

   * @brief 64-bit floating-point type definition.

   */

  typedef double float64_t;

  /**

   * @brief definition to read/write two 16 bit values.

   */

#if   defined ( __CC_ARM )

  #define __SIMD32_TYPE int32_t __packed

  #define CMSIS_UNUSED __attribute__((unused))

  #define CMSIS_INLINE __attribute__((always_inline))

#elif defined ( __ARMCC_VERSION ) && ( __ARMCC_VERSION >= 6010050 )

  #define __SIMD32_TYPE int32_t

  #define CMSIS_UNUSED __attribute__((unused))

  #define CMSIS_INLINE __attribute__((always_inline))

#elif defined ( __GNUC__ )

  #define __SIMD32_TYPE int32_t

  #define CMSIS_UNUSED __attribute__((unused))

  #define CMSIS_INLINE __attribute__((always_inline))

#elif defined ( __ICCARM__ )

  #define __SIMD32_TYPE int32_t __packed

  #define CMSIS_UNUSED

  #define CMSIS_INLINE

#elif defined ( __TI_ARM__ )

  #define __SIMD32_TYPE int32_t

  #define CMSIS_UNUSED __attribute__((unused))

  #define CMSIS_INLINE

#elif defined ( __CSMC__ )

  #define __SIMD32_TYPE int32_t

  #define CMSIS_UNUSED

  #define CMSIS_INLINE

#elif defined ( __TASKING__ )

  #define __SIMD32_TYPE __unaligned int32_t

  #define CMSIS_UNUSED

  #define CMSIS_INLINE

#else

  #error Unknown compiler

#endif

#define __SIMD32(addr)        (*(__SIMD32_TYPE **) & (addr))

#define __SIMD32_CONST(addr)  ((__SIMD32_TYPE *)(addr))

#define _SIMD32_OFFSET(addr)  (*(__SIMD32_TYPE *)  (addr))

#define __SIMD64(addr)        (*(int64_t **) & (addr))

#if !defined (ARM_MATH_DSP)

  /**

   * @brief definition to pack two 16 bit values.

   */

#define __PKHBT(ARG1, ARG2, ARG3) ( (((int32_t)(ARG1) <<    0) & (int32_t)0x0000FFFF) | \

                                    (((int32_t)(ARG2) << ARG3) & (int32_t)0xFFFF0000)  )

#define __PKHTB(ARG1, ARG2, ARG3) ( (((int32_t)(ARG1) <<    0) & (int32_t)0xFFFF0000) | \

                                    (((int32_t)(ARG2) >> ARG3) & (int32_t)0x0000FFFF)  )

#endif /* !defined (ARM_MATH_DSP) */

   /**

   * @brief definition to pack four 8 bit values.

   */

#ifndef ARM_MATH_BIG_ENDIAN

#define __PACKq7(v0,v1,v2,v3) ( (((int32_t)(v0) <<  0) & (int32_t)0x000000FF) | \

                                (((int32_t)(v1) <<  8) & (int32_t)0x0000FF00) | \

                                (((int32_t)(v2) << 16) & (int32_t)0x00FF0000) | \

                                (((int32_t)(v3) << 24) & (int32_t)0xFF000000)  )

#else

#define __PACKq7(v0,v1,v2,v3) ( (((int32_t)(v3) <<  0) & (int32_t)0x000000FF) | \

                                (((int32_t)(v2) <<  8) & (int32_t)0x0000FF00) | \

                                (((int32_t)(v1) << 16) & (int32_t)0x00FF0000) | \

                                (((int32_t)(v0) << 24) & (int32_t)0xFF000000)  )

#endif

  /**

   * @brief Clips Q63 to Q31 values.

   */

  CMSIS_INLINE __STATIC_INLINE q31_t clip_q63_to_q31(

  q63_t x)

  {

    return ((q31_t) (x >> 32) != ((q31_t) x >> 31)) ?

      ((0x7FFFFFFF ^ ((q31_t) (x >> 63)))) : (q31_t) x;

  }

  /**

   * @brief Clips Q63 to Q15 values.

   */

  CMSIS_INLINE __STATIC_INLINE q15_t clip_q63_to_q15(

  q63_t x)

  {

    return ((q31_t) (x >> 32) != ((q31_t) x >> 31)) ?

      ((0x7FFF ^ ((q15_t) (x >> 63)))) : (q15_t) (x >> 15);

  }

  /**

   * @brief Clips Q31 to Q7 values.

   */

  CMSIS_INLINE __STATIC_INLINE q7_t clip_q31_to_q7(

  q31_t x)

  {

    return ((q31_t) (x >> 24) != ((q31_t) x >> 23)) ?

      ((0x7F ^ ((q7_t) (x >> 31)))) : (q7_t) x;

  }

  /**

   * @brief Clips Q31 to Q15 values.

   */

  CMSIS_INLINE __STATIC_INLINE q15_t clip_q31_to_q15(

  q31_t x)

  {

    return ((q31_t) (x >> 16) != ((q31_t) x >> 15)) ?

      ((0x7FFF ^ ((q15_t) (x >> 31)))) : (q15_t) x;

  }

  /**

   * @brief Multiplies 32 X 64 and returns 32 bit result in 2.30 format.

   */

  CMSIS_INLINE __STATIC_INLINE q63_t mult32x64(

  q63_t x,

  q31_t y)

  {

    return ((((q63_t) (x & 0x00000000FFFFFFFF) * y) >> 32) +

            (((q63_t) (x >> 32) * y)));

  }

  /**

   * @brief Function to Calculates 1/in (reciprocal) value of Q31 Data type.

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t arm_recip_q31(

  q31_t in,

  q31_t * dst,

  q31_t * pRecipTable)

  {

    q31_t out;

    uint32_t tempVal;

    uint32_t index, i;

    uint32_t signBits;

    if (in > 0)

    {

      signBits = ((uint32_t) (__CLZ( in) - 1));

    }

    else

    {

      signBits = ((uint32_t) (__CLZ(-in) - 1));

    }

    /* Convert input sample to 1.31 format */

    in = (in << signBits);

    /* calculation of index for initial approximated Val */

    index = (uint32_t)(in >> 24);

    index = (index & INDEX_MASK);

    /* 1.31 with exp 1 */

    out = pRecipTable[index];

    /* calculation of reciprocal value */

    /* running approximation for two iterations */

    for (i = 0U; i < 2U; i++)

    {

      tempVal = (uint32_t) (((q63_t) in * out) >> 31);

      tempVal = 0x7FFFFFFFu - tempVal;

      /*      1.31 with exp 1 */

      /* out = (q31_t) (((q63_t) out * tempVal) >> 30); */

      out = clip_q63_to_q31(((q63_t) out * tempVal) >> 30);

    }

    /* write output */

    *dst = out;

    /* return num of signbits of out = 1/in value */

    return (signBits + 1U);

  }

  /**

   * @brief Function to Calculates 1/in (reciprocal) value of Q15 Data type.

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t arm_recip_q15(

  q15_t in,

  q15_t * dst,

  q15_t * pRecipTable)

  {

    q15_t out = 0;

    uint32_t tempVal = 0;

    uint32_t index = 0, i = 0;

    uint32_t signBits = 0;

    if (in > 0)

    {

      signBits = ((uint32_t)(__CLZ( in) - 17));

    }

    else

    {

      signBits = ((uint32_t)(__CLZ(-in) - 17));

    }

    /* Convert input sample to 1.15 format */

    in = (in << signBits);

    /* calculation of index for initial approximated Val */

    index = (uint32_t)(in >>  8);

    index = (index & INDEX_MASK);

    /*      1.15 with exp 1  */

    out = pRecipTable[index];

    /* calculation of reciprocal value */

    /* running approximation for two iterations */

    for (i = 0U; i < 2U; i++)

    {

      tempVal = (uint32_t) (((q31_t) in * out) >> 15);

      tempVal = 0x7FFFu - tempVal;

      /*      1.15 with exp 1 */

      out = (q15_t) (((q31_t) out * tempVal) >> 14);

      /* out = clip_q31_to_q15(((q31_t) out * tempVal) >> 14); */

    }

    /* write output */

    *dst = out;

    /* return num of signbits of out = 1/in value */

    return (signBits + 1);

  }

/*

 * @brief C custom defined intrinsic function for M3 and M0 processors

 */

#if !defined (ARM_MATH_DSP)

  /*

   * @brief C custom defined QADD8 for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __QADD8(

  uint32_t x,

  uint32_t y)

  {

    q31_t r, s, t, u;

    r = __SSAT(((((q31_t)x << 24) >> 24) + (((q31_t)y << 24) >> 24)), 8) & (int32_t)0x000000FF;

    s = __SSAT(((((q31_t)x << 16) >> 24) + (((q31_t)y << 16) >> 24)), 8) & (int32_t)0x000000FF;

    t = __SSAT(((((q31_t)x <<  8) >> 24) + (((q31_t)y <<  8) >> 24)), 8) & (int32_t)0x000000FF;

    u = __SSAT(((((q31_t)x      ) >> 24) + (((q31_t)y      ) >> 24)), 8) & (int32_t)0x000000FF;

    return ((uint32_t)((u << 24) | (t << 16) | (s <<  8) | (r      )));

  }

  /*

   * @brief C custom defined QSUB8 for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __QSUB8(

  uint32_t x,

  uint32_t y)

  {

    q31_t r, s, t, u;

    r = __SSAT(((((q31_t)x << 24) >> 24) - (((q31_t)y << 24) >> 24)), 8) & (int32_t)0x000000FF;

    s = __SSAT(((((q31_t)x << 16) >> 24) - (((q31_t)y << 16) >> 24)), 8) & (int32_t)0x000000FF;

    t = __SSAT(((((q31_t)x <<  8) >> 24) - (((q31_t)y <<  8) >> 24)), 8) & (int32_t)0x000000FF;

    u = __SSAT(((((q31_t)x      ) >> 24) - (((q31_t)y      ) >> 24)), 8) & (int32_t)0x000000FF;

    return ((uint32_t)((u << 24) | (t << 16) | (s <<  8) | (r      )));

  }

  /*

   * @brief C custom defined QADD16 for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __QADD16(

  uint32_t x,

  uint32_t y)

  {

/*  q31_t r,     s;  without initialisation 'arm_offset_q15 test' fails  but 'intrinsic' tests pass! for armCC */

    q31_t r = 0, s = 0;

    r = __SSAT(((((q31_t)x << 16) >> 16) + (((q31_t)y << 16) >> 16)), 16) & (int32_t)0x0000FFFF;

    s = __SSAT(((((q31_t)x      ) >> 16) + (((q31_t)y      ) >> 16)), 16) & (int32_t)0x0000FFFF;

    return ((uint32_t)((s << 16) | (r      )));

  }

  /*

   * @brief C custom defined SHADD16 for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __SHADD16(

  uint32_t x,

  uint32_t y)

  {

    q31_t r, s;

    r = (((((q31_t)x << 16) >> 16) + (((q31_t)y << 16) >> 16)) >> 1) & (int32_t)0x0000FFFF;

    s = (((((q31_t)x      ) >> 16) + (((q31_t)y      ) >> 16)) >> 1) & (int32_t)0x0000FFFF;

    return ((uint32_t)((s << 16) | (r      )));

  }

  /*

   * @brief C custom defined QSUB16 for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __QSUB16(

  uint32_t x,

  uint32_t y)

  {

    q31_t r, s;

    r = __SSAT(((((q31_t)x << 16) >> 16) - (((q31_t)y << 16) >> 16)), 16) & (int32_t)0x0000FFFF;

    s = __SSAT(((((q31_t)x      ) >> 16) - (((q31_t)y      ) >> 16)), 16) & (int32_t)0x0000FFFF;

    return ((uint32_t)((s << 16) | (r      )));

  }

  /*

   * @brief C custom defined SHSUB16 for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __SHSUB16(

  uint32_t x,

  uint32_t y)

  {

    q31_t r, s;

    r = (((((q31_t)x << 16) >> 16) - (((q31_t)y << 16) >> 16)) >> 1) & (int32_t)0x0000FFFF;

    s = (((((q31_t)x      ) >> 16) - (((q31_t)y      ) >> 16)) >> 1) & (int32_t)0x0000FFFF;

    return ((uint32_t)((s << 16) | (r      )));

  }

  /*

   * @brief C custom defined QASX for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __QASX(

  uint32_t x,

  uint32_t y)

  {

    q31_t r, s;

    r = __SSAT(((((q31_t)x << 16) >> 16) - (((q31_t)y      ) >> 16)), 16) & (int32_t)0x0000FFFF;

    s = __SSAT(((((q31_t)x      ) >> 16) + (((q31_t)y << 16) >> 16)), 16) & (int32_t)0x0000FFFF;

    return ((uint32_t)((s << 16) | (r      )));

  }

  /*

   * @brief C custom defined SHASX for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __SHASX(

  uint32_t x,

  uint32_t y)

  {

    q31_t r, s;

    r = (((((q31_t)x << 16) >> 16) - (((q31_t)y      ) >> 16)) >> 1) & (int32_t)0x0000FFFF;

    s = (((((q31_t)x      ) >> 16) + (((q31_t)y << 16) >> 16)) >> 1) & (int32_t)0x0000FFFF;

    return ((uint32_t)((s << 16) | (r      )));

  }

  /*

   * @brief C custom defined QSAX for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __QSAX(

  uint32_t x,

  uint32_t y)

  {

    q31_t r, s;

    r = __SSAT(((((q31_t)x << 16) >> 16) + (((q31_t)y      ) >> 16)), 16) & (int32_t)0x0000FFFF;

    s = __SSAT(((((q31_t)x      ) >> 16) - (((q31_t)y << 16) >> 16)), 16) & (int32_t)0x0000FFFF;

    return ((uint32_t)((s << 16) | (r      )));

  }

  /*

   * @brief C custom defined SHSAX for M3 and M0 processors

   */

  CMSIS_INLINE __STATIC_INLINE uint32_t __SHSAX(

  uint32_t x,

  uint32_t y)

  {

    q31_t r, s;

    r = (((((q31_t)x << 16) >> 16) + (((q31_t)y      ) >> 16)) >> 1) & (int32_t)0x0000FFFF;

    s = (((((q31_t)x      ) >> 16) - (((q31_t)y << 16) >> 16)) >> 1) & (int32_t)0x0000FFFF;