apps/codecs/libfaad/fixed.h

   1 /*
   2 ** FAAD2 - Freeware Advanced Audio (AAC) Decoder including SBR decoding
   3 ** Copyright (C) 2003-2004 M. Bakker, Ahead Software AG, http://www.nero.com
   4 **
   5 ** This program is free software; you can redistribute it and/or modify
   6 ** it under the terms of the GNU General Public License as published by
   7 ** the Free Software Foundation; either version 2 of the License, or
   8 ** (at your option) any later version.
   9 **
  10 ** This program is distributed in the hope that it will be useful,
  11 ** but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 ** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 ** GNU General Public License for more details.
  14 **
  15 ** You should have received a copy of the GNU General Public License
  16 ** along with this program; if not, write to the Free Software
  17 ** Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18 **
  19 ** Any non-GPL usage of this software or parts of this software is strictly
  20 ** forbidden.
  21 **
  22 ** Commercial non-GPL licensing of this software is possible.
  23 ** For more info contact Ahead Software through Mpeg4AAClicense@nero.com.
  24 **
  25 ** $Id$
  26 **/
  27
  28 #ifndef __FIXED_H__
  29 #define __FIXED_H__
  30
  31 #ifdef __cplusplus
  32 extern "C" {
  33 #endif
  34
  35 #if defined(_WIN32_WCE) && defined(_ARM_)
  36 #include <cmnintrin.h>
  37 #endif
  38
  39 #define COEF_BITS 28
  40 #define COEF_PRECISION (1 << COEF_BITS)
  41 #define REAL_BITS 14 // MAXIMUM OF 14 FOR FIXED POINT SBR
  42 #define REAL_PRECISION (1 << REAL_BITS)
  43
  44 /* FRAC is the fractional only part of the fixed point number [0.0..1.0) */
  45 #define FRAC_SIZE 32 /* frac is a 32 bit integer */
  46 #define FRAC_BITS 31
  47 #define FRAC_PRECISION ((uint32_t)(1 << FRAC_BITS))
  48 #define FRAC_MAX 0x7FFFFFFF
  49
  50 typedef int32_t real_t;
  51
  52
  53 #define REAL_CONST(A) (((A) >= 0) ? ((real_t)((A)*(REAL_PRECISION)+0.5)) : ((real_t)((A)*(REAL_PRECISION)-0.5)))
  54 #define COEF_CONST(A) (((A) >= 0) ? ((real_t)((A)*(COEF_PRECISION)+0.5)) : ((real_t)((A)*(COEF_PRECISION)-0.5)))
  55 #define FRAC_CONST(A) (((A) == 1.00) ? ((real_t)FRAC_MAX) : (((A) >= 0) ? ((real_t)((A)*(FRAC_PRECISION)+0.5)) : ((real_t)((A)*(FRAC_PRECISION)-0.5))))
  56 //#define FRAC_CONST(A) (((A) >= 0) ? ((real_t)((A)*(FRAC_PRECISION)+0.5)) : ((real_t)((A)*(FRAC_PRECISION)-0.5)))
  57
  58 #define Q2_BITS 22
  59 #define Q2_PRECISION (1 << Q2_BITS)
  60 #define Q2_CONST(A) (((A) >= 0) ? ((real_t)((A)*(Q2_PRECISION)+0.5)) : ((real_t)((A)*(Q2_PRECISION)-0.5)))
  61
  62 #if defined(CPU_COLDFIRE)
  63
  64 static INLINE real_t MUL_F(real_t A, real_t B)
  65 {
  66     asm volatile (
  67         "mac.l %[A], %[B], %%acc0\n\t"
  68         "movclr.l %%acc0, %[A]"
  69         : [A] "+&r" (A) : [B] "r" (B)
  70     );
  71     return A;
  72 }
  73
  74 static INLINE real_t MUL_C(real_t A, real_t B)
  75 {
  76     asm volatile (
  77         "mac.l %[A], %[B], %%acc0\n\t"
  78         "movclr.l %%acc0, %[A]\n\t"
  79         : [A] "+&d" (A) : [B] "r" (B)
  80     );
  81     return A << 3;
  82 }
  83
  84 /* MUL_R needs too many shifts for us to just operate on the top 32 bits the
  85    emac unit gives as usual, so we do a full 64 bit mul here. */
  86 static INLINE real_t MUL_R(real_t x, real_t y)
  87 {
  88     real_t t1, t2;
  89     asm volatile (
  90         "mac.l   %[x],%[y],%%acc0\n" /* multiply */
  91         "mulu.l  %[y],%[x]   \n"     /* get lower half, avoid emac stall */
  92         "movclr.l %%acc0,%[t1]   \n" /* get higher half */
  93         "moveq.l #17,%[t2]   \n"
  94         "asl.l   %[t2],%[t1] \n"     /* hi <<= 17, plus one free */
  95         "moveq.l #14,%[t2]   \n"
  96         "lsr.l   %[t2],%[x]  \n"     /* (unsigned)lo >>= 14 */
  97         "or.l    %[x],%[t1]  \n"     /* combine result */
  98         : /* outputs */
  99         [t1]"=&d"(t1),
 100         [t2]"=&d"(t2),
 101         [x] "+d" (x)
 102         : /* inputs */
 103         [y] "d"  (y)
 104     );
 105     return t1;
 106 }
 107
 108 static INLINE void ComplexMult(real_t *y1, real_t *y2,
 109     real_t x1, real_t x2, real_t c1, real_t c2)
 110 {
 111     asm volatile(
 112         "mac.l %[x1], %[c1], %%acc0\n\t"
 113         "mac.l %[x2], %[c2], %%acc0\n\t"
 114         "mac.l %[x2], %[c1], %%acc1\n\t"
 115         "msac.l %[x1], %[c2], %%acc1\n\t"
 116         "movclr.l %%acc0, %[x1]\n\t"
 117         "move.l %[x1], (%[y1])\n\t"
 118         "movclr.l %%acc1, %[x1]\n\t"
 119         "move.l %[x1], (%[y2])"
 120         : [x1] "+&r" (x1)
 121         : [x2] "r" (x2), [y1] "a" (y1), [y2] "a" (y2),
 122           [c1] "r" (c1), [c2] "r" (c2)
 123         : "memory"
 124     );
 125 }
 126
 127   /* the following see little or no use, so just ignore them for now */
 128   #define MUL_Q2(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (Q2_BITS-1))) >> Q2_BITS)
 129   #define MUL_SHIFT6(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (6-1))) >> 6)
 130   #define MUL_SHIFT23(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (23-1))) >> 23)
 131
 132 #elif defined(__GNUC__) && defined (__arm__)
 133
 134 /* taken from MAD */
 135 #define arm_mul(x, y, SCALEBITS) \
 136 ({ \
 137     uint32_t __hi; \
 138     uint32_t __lo; \
 139     uint32_t __result; \
 140     asm("smull  %0, %1, %3, %4\n\t" \
 141         "movs   %0, %0, lsr %5\n\t" \
 142         "adc    %2, %0, %1, lsl %6" \
 143         : "=&r" (__lo), "=&r" (__hi), "=r" (__result) \
 144         : "%r" (x), "r" (y), \
 145         "M" (SCALEBITS), "M" (32 - (SCALEBITS)) \
 146         : "cc"); \
 147         __result; \
 148 })
 149
 150 static INLINE real_t MUL_R(real_t A, real_t B)
 151 {
 152     return arm_mul(A, B, REAL_BITS);
 153 }
 154
 155 static INLINE real_t MUL_C(real_t A, real_t B)
 156 {
 157     return arm_mul(A, B, COEF_BITS);
 158 }
 159
 160 static INLINE real_t MUL_Q2(real_t A, real_t B)
 161 {
 162     return arm_mul(A, B, Q2_BITS);
 163 }
 164
 165 static INLINE real_t MUL_SHIFT6(real_t A, real_t B)
 166 {
 167     return arm_mul(A, B, 6);
 168 }
 169
 170 static INLINE real_t MUL_SHIFT23(real_t A, real_t B)
 171 {
 172     return arm_mul(A, B, 23);
 173 }
 174
 175 static INLINE real_t _MulHigh(real_t x, real_t y)
 176 {
 177     uint32_t __lo;
 178     uint32_t __hi;
 179     asm("smull\t%0, %1, %2, %3"
 180         : "=&r"(__lo),"=&r"(__hi)
 181         : "%r"(x),"r"(y)
 182         : "cc");
 183     return __hi;
 184 }
 185
 186 static INLINE real_t MUL_F(real_t A, real_t B)
 187 {
 188     return _MulHigh(A, B) << (FRAC_SIZE-FRAC_BITS);
 189 }
 190
 191 /* Complex multiplication */
 192 static INLINE void ComplexMult(real_t *y1, real_t *y2,
 193     real_t x1, real_t x2, real_t c1, real_t c2)
 194 {
 195     int32_t tmp, yt1, yt2;
 196     asm("smull %0, %1, %4, %6\n\t"
 197         "smlal %0, %1, %5, %7\n\t"
 198         "rsb   %3, %4, #0\n\t"
 199         "smull %0, %2, %5, %6\n\t"
 200         "smlal %0, %2, %3, %7"
 201         : "=&r" (tmp), "=&r" (yt1), "=&r" (yt2), "=r" (x1)
 202         : "3" (x1), "r" (x2), "r" (c1), "r" (c2)
 203         : "cc" );
 204     *y1 = yt1 << (FRAC_SIZE-FRAC_BITS);
 205     *y2 = yt2 << (FRAC_SIZE-FRAC_BITS);
 206 }
 207
 208 #else
 209
 210   /* multiply with real shift */
 211   #define MUL_R(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (REAL_BITS-1))) >> REAL_BITS)
 212   /* multiply with coef shift */
 213   #define MUL_C(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (COEF_BITS-1))) >> COEF_BITS)
 214   /* multiply with fractional shift */
 215 #if defined(_WIN32_WCE) && defined(_ARM_)
 216   /* eVC for PocketPC has an intrinsic function that returns only the high 32 bits of a 32x32 bit multiply */
 217   static INLINE real_t MUL_F(real_t A, real_t B)
 218   {
 219       return _MulHigh(A,B) << (32-FRAC_BITS);
 220   }
 221 #else
 222   #define _MulHigh(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (FRAC_SIZE-1))) >> FRAC_SIZE)
 223   #define MUL_F(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (FRAC_BITS-1))) >> FRAC_BITS)
 224 #endif
 225   #define MUL_Q2(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (Q2_BITS-1))) >> Q2_BITS)
 226   #define MUL_SHIFT6(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (6-1))) >> 6)
 227   #define MUL_SHIFT23(A,B) (real_t)(((int64_t)(A)*(int64_t)(B)+(1 << (23-1))) >> 23)
 228
 229 /* Complex multiplication */
 230 static INLINE void ComplexMult(real_t *y1, real_t *y2,
 231     real_t x1, real_t x2, real_t c1, real_t c2)
 232 {
 233     *y1 = (_MulHigh(x1, c1) + _MulHigh(x2, c2))<<(FRAC_SIZE-FRAC_BITS);
 234     *y2 = (_MulHigh(x2, c1) - _MulHigh(x1, c2))<<(FRAC_SIZE-FRAC_BITS);
 235 }
 236
 237 #endif
 238
 239
 240
 241 #ifdef __cplusplus
 242 }
 243 #endif
 244 #endif