rockbox/apps/codecs/libwma/wmafixed.c

//#include "asf.h"
#include "wmadec.h"
#include "wmafixed.h"
#include <codecs.h>

fixed64 IntTo64(int x){
    fixed64 res = 0;
    unsigned char *p = (unsigned char *)&res;

#ifdef ROCKBOX_BIG_ENDIAN
    p[5] = x & 0xff;
    p[4] = (x & 0xff00)>>8;
    p[3] = (x & 0xff0000)>>16;
    p[2] = (x & 0xff000000)>>24;
#else
    p[2] = x & 0xff;
    p[3] = (x & 0xff00)>>8;
    p[4] = (x & 0xff0000)>>16;
    p[5] = (x & 0xff000000)>>24;
#endif
    return res;
}

int IntFrom64(fixed64 x)
{
    int res = 0;
    unsigned char *p = (unsigned char *)&x;

#ifdef ROCKBOX_BIG_ENDIAN
    res = p[5] | (p[4]<<8) | (p[3]<<16) | (p[2]<<24);
#else
    res = p[2] | (p[3]<<8) | (p[4]<<16) | (p[5]<<24);
#endif
    return res;
}

fixed32 Fixed32From64(fixed64 x)
{
  return x & 0xFFFFFFFF;
}

fixed64 Fixed32To64(fixed32 x)
{
  return (fixed64)x;
}


/*
	Fixed precision multiply code.

*/

/*Sign-15.16 format */
#ifdef CPU_ARM
/* these are defines in wmafixed.h*/
#elif defined(CPU_COLDFIRE)

#else

fixed32 fixmul32(fixed32 x, fixed32 y)
{
    fixed64 temp;
    temp = x;
    temp *= y;

    temp >>= PRECISION;

    return (fixed32)temp;
}

#endif
/*
	Special fixmul32 that does a 16.16 x 1.31 multiply that returns a 16.16 value.
	this is needed because the fft constants are all normalized to be less then 1
	and can't fit into a 16 bit number without excessive rounding


*/
#ifndef CPU_ARM
fixed32 fixmul32b(fixed32 x, fixed32 y)
{
    fixed64 temp;

    temp = x;
    temp *= y;

    temp >>= 31;        //16+31-16 = 31 bits

    return (fixed32)temp;
}
#endif



/*
	Not performance senstitive code here

*/


fixed64 fixmul64byfixed(fixed64 x, fixed32 y)
{

    //return x * y;
     return (x * y);
 // return (fixed64) fixmul32(Fixed32From64(x),y);
}


fixed32 fixdiv32(fixed32 x, fixed32 y)
{
    fixed64 temp;

    if(x == 0)
        return 0;
    if(y == 0)
        return 0x7fffffff;
    temp = x;
    temp <<= PRECISION;
    return (fixed32)(temp / y);
}

fixed64 fixdiv64(fixed64 x, fixed64 y)
{
    fixed64 temp;

    if(x == 0)
        return 0;
    if(y == 0)
        return 0x07ffffffffffffffLL;
    temp = x;
    temp <<= PRECISION64;
    return (fixed64)(temp / y);
}

 fixed32 fixsqrt32(fixed32 x)
{

    unsigned long r = 0, s, v = (unsigned long)x;

#define STEP(k) s = r + (1 << k * 2); r >>= 1; \
    if (s <= v) { v -= s; r |= (1 << k * 2); }

    STEP(15);
    STEP(14);
    STEP(13);
    STEP(12);
    STEP(11);
    STEP(10);
    STEP(9);
    STEP(8);
    STEP(7);
    STEP(6);
    STEP(5);
    STEP(4);
    STEP(3);
    STEP(2);
    STEP(1);
    STEP(0);

    return (fixed32)(r << (PRECISION / 2));
}



/* Inverse gain of circular cordic rotation in s0.31 format. */
static const long cordic_circular_gain = 0xb2458939; /* 0.607252929 */

/* Table of values of atan(2^-i) in 0.32 format fractions of pi where pi = 0xffffffff / 2 */
static const unsigned long atan_table[] = {
    0x1fffffff, /* +0.785398163 (or pi/4) */
    0x12e4051d, /* +0.463647609 */
    0x09fb385b, /* +0.244978663 */
    0x051111d4, /* +0.124354995 */
    0x028b0d43, /* +0.062418810 */
    0x0145d7e1, /* +0.031239833 */
    0x00a2f61e, /* +0.015623729 */
    0x00517c55, /* +0.007812341 */
    0x0028be53, /* +0.003906230 */
    0x00145f2e, /* +0.001953123 */
    0x000a2f98, /* +0.000976562 */
    0x000517cc, /* +0.000488281 */
    0x00028be6, /* +0.000244141 */
    0x000145f3, /* +0.000122070 */
    0x0000a2f9, /* +0.000061035 */
    0x0000517c, /* +0.000030518 */
    0x000028be, /* +0.000015259 */
    0x0000145f, /* +0.000007629 */
    0x00000a2f, /* +0.000003815 */
    0x00000517, /* +0.000001907 */
    0x0000028b, /* +0.000000954 */
    0x00000145, /* +0.000000477 */
    0x000000a2, /* +0.000000238 */
    0x00000051, /* +0.000000119 */
    0x00000028, /* +0.000000060 */
    0x00000014, /* +0.000000030 */
    0x0000000a, /* +0.000000015 */
    0x00000005, /* +0.000000007 */
    0x00000002, /* +0.000000004 */
    0x00000001, /* +0.000000002 */
    0x00000000, /* +0.000000001 */
    0x00000000, /* +0.000000000 */
};


/*

	Below here functions do not use standard fixed precision!
*/


/**
 * Implements sin and cos using CORDIC rotation.
 *
 * @param phase has range from 0 to 0xffffffff, representing 0 and
 *        2*pi respectively.
 * @param cos return address for cos
 * @return sin of phase, value is a signed value from LONG_MIN to LONG_MAX,
 *         representing -1 and 1 respectively.
 *
 *        Gives at least 24 bits precision (last 2-8 bits or so are probably off)
 */
long fsincos(unsigned long phase, fixed32 *cos)
{
    int32_t x, x1, y, y1;
    unsigned long z, z1;
    int i;

    /* Setup initial vector */
    x = cordic_circular_gain;
    y = 0;
    z = phase;

    /* The phase has to be somewhere between 0..pi for this to work right */
    if (z < 0xffffffff / 4) {
        /* z in first quadrant, z += pi/2 to correct */
        x = -x;
        z += 0xffffffff / 4;
    } else if (z < 3 * (0xffffffff / 4)) {
        /* z in third quadrant, z -= pi/2 to correct */
        z -= 0xffffffff / 4;
    } else {
        /* z in fourth quadrant, z -= 3pi/2 to correct */
        x = -x;
        z -= 3 * (0xffffffff / 4);
    }

    /* Each iteration adds roughly 1-bit of extra precision */
    for (i = 0; i < 31; i++) {
        x1 = x >> i;
        y1 = y >> i;
        z1 = atan_table[i];

        /* Decided which direction to rotate vector. Pivot point is pi/2 */
        if (z >= 0xffffffff / 4) {
            x -= y1;
            y += x1;
            z -= z1;
        } else {
            x += y1;
            y -= x1;
            z += z1;
        }
    }

    if (cos)
        *cos = x;

    return y;
}


/*
	Old trig functions.  Still used in 1 place each.

*/

#if 0
fixed32 fixsin32(fixed32 x)
{

    fixed64 x2, temp;
    int     sign = 1;

    if(x < 0)
    {
        sign = -1;
        x = -x;
    }
    while (x > 0x19220)
    {
        x -= M_PI_F;
        sign = -sign;
    }
    if (x > 0x19220)
    {
        x = M_PI_F - x;
    }
    x2 = (fixed64)x * x;
    x2 >>= PRECISION;
    if(sign != 1)
    {
        x = -x;
    }
    /**
    temp = ftofix32(-.0000000239f) * x2;
    temp >>= PRECISION;
    **/
    temp = 0; // PJJ
    //temp = (temp + 0x0) * x2;    //MGG:  this can't possibly do anything?
    //temp >>= PRECISION;
    temp = (temp - 0xd) * x2;
    temp >>= PRECISION;
    temp = (temp + 0x222) * x2;
    temp >>= PRECISION;
    temp = (temp - 0x2aab) * x2;
    temp >>= PRECISION;
    temp += 0x10000;
    temp = temp * x;
    temp >>= PRECISION;

    return  (fixed32)(temp);
}

fixed32 fixcos32(fixed32 x)
{
    return fixsin32(x - (M_PI_F>>1))*-1;
}
#endif