liba52/srfftp.h

   1
   2 /*
   3  *  srfftp.h
   4  *
   5  *  Copyright (C) Yuqing Deng <Yuqing_Deng@brown.edu> - April 2000
   6  *
   7  *  64 and 128 point split radix fft for ac3dec
   8  *
   9  *  The algorithm is desribed in the book:
  10  *  "Computational Frameworks of the Fast Fourier Transform".
  11  *
  12  *  The ideas and the the organization of code borrowed from djbfft written by
  13  *  D. J. Bernstein <djb@cr.py.to>.  djbff can be found at
  14  *  http://cr.yp.to/djbfft.html.
  15  *
  16  *  srfftp.h is free software; you can redistribute it and/or modify
  17  *  it under the terms of the GNU General Public License as published by
  18  *  the Free Software Foundation; either version 2, or (at your option)
  19  *  any later version.
  20  *
  21  *  srfftp.h is distributed in the hope that it will be useful,
  22  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  23  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  24  *  GNU General Public License for more details.
  25  *
  26  *  You should have received a copy of the GNU General Public License
  27  *  along with GNU Make; see the file COPYING.  If not, write to
  28  *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  29  *
  30  */
  31
  32 #ifndef SRFFTP_H__
  33 #define SRFFTP_H__
  34
  35 static complex_t delta16[4] __attribute__((aligned(16))) =
  36  { {1.00000000000000,  0.00000000000000},
  37    {0.92387953251129, -0.38268343236509},
  38    {0.70710678118655, -0.70710678118655},
  39    {0.38268343236509, -0.92387953251129}};
  40
  41 static complex_t delta16_3[4] __attribute__((aligned(16))) =
  42  { {1.00000000000000,  0.00000000000000},
  43    {0.38268343236509, -0.92387953251129},
  44    {-0.70710678118655, -0.70710678118655},
  45    {-0.92387953251129, 0.38268343236509}};
  46
  47 static complex_t delta32[8] __attribute__((aligned(16))) =
  48  { {1.00000000000000,  0.00000000000000},
  49    {0.98078528040323, -0.19509032201613},
  50    {0.92387953251129, -0.38268343236509},
  51    {0.83146961230255, -0.55557023301960},
  52    {0.70710678118655, -0.70710678118655},
  53    {0.55557023301960, -0.83146961230255},
  54    {0.38268343236509, -0.92387953251129},
  55    {0.19509032201613, -0.98078528040323}};
  56
  57 static complex_t delta32_3[8] __attribute__((aligned(16))) =
  58  { {1.00000000000000,  0.00000000000000},
  59    {0.83146961230255, -0.55557023301960},
  60    {0.38268343236509, -0.92387953251129},
  61    {-0.19509032201613, -0.98078528040323},
  62    {-0.70710678118655, -0.70710678118655},
  63    {-0.98078528040323, -0.19509032201613},
  64    {-0.92387953251129, 0.38268343236509},
  65    {-0.55557023301960, 0.83146961230255}};
  66
  67 static complex_t delta64[16] __attribute__((aligned(16))) =
  68  { {1.00000000000000,  0.00000000000000},
  69    {0.99518472667220, -0.09801714032956},
  70    {0.98078528040323, -0.19509032201613},
  71    {0.95694033573221, -0.29028467725446},
  72    {0.92387953251129, -0.38268343236509},
  73    {0.88192126434836, -0.47139673682600},
  74    {0.83146961230255, -0.55557023301960},
  75    {0.77301045336274, -0.63439328416365},
  76    {0.70710678118655, -0.70710678118655},
  77    {0.63439328416365, -0.77301045336274},
  78    {0.55557023301960, -0.83146961230255},
  79    {0.47139673682600, -0.88192126434835},
  80    {0.38268343236509, -0.92387953251129},
  81    {0.29028467725446, -0.95694033573221},
  82    {0.19509032201613, -0.98078528040323},
  83    {0.09801714032956, -0.99518472667220}};
  84
  85 static complex_t delta64_3[16] __attribute__((aligned(16))) =
  86  { {1.00000000000000,  0.00000000000000},
  87    {0.95694033573221, -0.29028467725446},
  88    {0.83146961230255, -0.55557023301960},
  89    {0.63439328416365, -0.77301045336274},
  90    {0.38268343236509, -0.92387953251129},
  91    {0.09801714032956, -0.99518472667220},
  92    {-0.19509032201613, -0.98078528040323},
  93    {-0.47139673682600, -0.88192126434836},
  94    {-0.70710678118655, -0.70710678118655},
  95    {-0.88192126434835, -0.47139673682600},
  96    {-0.98078528040323, -0.19509032201613},
  97    {-0.99518472667220, 0.09801714032956},
  98    {-0.92387953251129, 0.38268343236509},
  99    {-0.77301045336274, 0.63439328416365},
 100    {-0.55557023301960, 0.83146961230255},
 101    {-0.29028467725446, 0.95694033573221}};
 102
 103 static complex_t delta128[32] __attribute__((aligned(16))) =
 104  { {1.00000000000000,  0.00000000000000},
 105    {0.99879545620517, -0.04906767432742},
 106    {0.99518472667220, -0.09801714032956},
 107    {0.98917650996478, -0.14673047445536},
 108    {0.98078528040323, -0.19509032201613},
 109    {0.97003125319454, -0.24298017990326},
 110    {0.95694033573221, -0.29028467725446},
 111    {0.94154406518302, -0.33688985339222},
 112    {0.92387953251129, -0.38268343236509},
 113    {0.90398929312344, -0.42755509343028},
 114    {0.88192126434836, -0.47139673682600},
 115    {0.85772861000027, -0.51410274419322},
 116    {0.83146961230255, -0.55557023301960},
 117    {0.80320753148064, -0.59569930449243},
 118    {0.77301045336274, -0.63439328416365},
 119    {0.74095112535496, -0.67155895484702},
 120    {0.70710678118655, -0.70710678118655},
 121    {0.67155895484702, -0.74095112535496},
 122    {0.63439328416365, -0.77301045336274},
 123    {0.59569930449243, -0.80320753148064},
 124    {0.55557023301960, -0.83146961230255},
 125    {0.51410274419322, -0.85772861000027},
 126    {0.47139673682600, -0.88192126434835},
 127    {0.42755509343028, -0.90398929312344},
 128    {0.38268343236509, -0.92387953251129},
 129    {0.33688985339222, -0.94154406518302},
 130    {0.29028467725446, -0.95694033573221},
 131    {0.24298017990326, -0.97003125319454},
 132    {0.19509032201613, -0.98078528040323},
 133    {0.14673047445536, -0.98917650996478},
 134    {0.09801714032956, -0.99518472667220},
 135    {0.04906767432742, -0.99879545620517}};
 136
 137 static complex_t delta128_3[32] __attribute__((aligned(16))) =
 138  { {1.00000000000000,  0.00000000000000},
 139    {0.98917650996478, -0.14673047445536},
 140    {0.95694033573221, -0.29028467725446},
 141    {0.90398929312344, -0.42755509343028},
 142    {0.83146961230255, -0.55557023301960},
 143    {0.74095112535496, -0.67155895484702},
 144    {0.63439328416365, -0.77301045336274},
 145    {0.51410274419322, -0.85772861000027},
 146    {0.38268343236509, -0.92387953251129},
 147    {0.24298017990326, -0.97003125319454},
 148    {0.09801714032956, -0.99518472667220},
 149    {-0.04906767432742, -0.99879545620517},
 150    {-0.19509032201613, -0.98078528040323},
 151    {-0.33688985339222, -0.94154406518302},
 152    {-0.47139673682600, -0.88192126434836},
 153    {-0.59569930449243, -0.80320753148065},
 154    {-0.70710678118655, -0.70710678118655},
 155    {-0.80320753148065, -0.59569930449243},
 156    {-0.88192126434835, -0.47139673682600},
 157    {-0.94154406518302, -0.33688985339222},
 158    {-0.98078528040323, -0.19509032201613},
 159    {-0.99879545620517, -0.04906767432742},
 160    {-0.99518472667220, 0.09801714032956},
 161    {-0.97003125319454, 0.24298017990326},
 162    {-0.92387953251129, 0.38268343236509},
 163    {-0.85772861000027, 0.51410274419322},
 164    {-0.77301045336274, 0.63439328416365},
 165    {-0.67155895484702, 0.74095112535496},
 166    {-0.55557023301960, 0.83146961230255},
 167    {-0.42755509343028, 0.90398929312344},
 168    {-0.29028467725446, 0.95694033573221},
 169    {-0.14673047445536, 0.98917650996478}};
 170
 171 #define HSQRT2 0.707106781188;
 172
 173 #define TRANSZERO(A0,A4,A8,A12) { \
 174   u_r = wTB[0].real; \
 175   v_i = u_r - wTB[k*2].real; \
 176   u_r += wTB[k*2].real; \
 177   u_i = wTB[0].imag; \
 178   v_r = wTB[k*2].imag - u_i; \
 179   u_i += wTB[k*2].imag; \
 180   a_r = A0.real; \
 181   a_i = A0.imag; \
 182   a1_r = a_r; \
 183   a1_r += u_r; \
 184   A0.real = a1_r; \
 185   a_r -= u_r; \
 186   A8.real = a_r; \
 187   a1_i = a_i; \
 188   a1_i += u_i; \
 189   A0.imag = a1_i; \
 190   a_i -= u_i; \
 191   A8.imag = a_i; \
 192   a1_r = A4.real; \
 193   a1_i = A4.imag; \
 194   a_r = a1_r; \
 195   a_r -= v_r; \
 196   A4.real = a_r; \
 197   a1_r += v_r; \
 198   A12.real = a1_r; \
 199   a_i = a1_i; \
 200   a_i -= v_i; \
 201   A4.imag = a_i; \
 202   a1_i += v_i; \
 203   A12.imag = a1_i; \
 204   }
 205
 206 #define TRANSHALF_16(A2,A6,A10,A14) {\
 207   u_r = wTB[2].real; \
 208   a_r = u_r; \
 209   u_i = wTB[2].imag; \
 210   u_r += u_i; \
 211   u_i -= a_r; \
 212   a_r = wTB[6].real; \
 213   a1_r = a_r; \
 214   a_i = wTB[6].imag; \
 215   a_r = a_i - a_r; \
 216   a_i += a1_r; \
 217   v_i = u_r - a_r; \
 218   u_r += a_r; \
 219   v_r = u_i + a_i; \
 220   u_i -= a_i; \
 221   v_i *= HSQRT2; \
 222   v_r *= HSQRT2; \
 223   u_r *= HSQRT2; \
 224   u_i *= HSQRT2; \
 225   a_r = A2.real; \
 226   a_i = A2.imag; \
 227   a1_r = a_r; \
 228   a1_r += u_r; \
 229   A2.real = a1_r; \
 230   a_r -= u_r; \
 231   A10.real = a_r; \
 232   a1_i = a_i; \
 233   a1_i += u_i; \
 234   A2.imag = a1_i; \
 235   a_i -= u_i; \
 236   A10.imag = a_i; \
 237   a1_r = A6.real; \
 238   a1_i = A6.imag;  \
 239   a_r = a1_r; \
 240   a1_r += v_r; \
 241   A6.real = a1_r; \
 242   a_r -= v_r; \
 243   A14.real = a_r; \
 244   a_i = a1_i; \
 245   a1_i -= v_i; \
 246   A6.imag = a1_i; \
 247   a_i += v_i; \
 248   A14.imag = a_i; \
 249   }
 250
 251 #define TRANS(A1,A5,A9,A13,WT,WB,D,D3) { \
 252   u_r = WT.real; \
 253   a_r = u_r; \
 254   a_r *= D.imag; \
 255   u_r *= D.real; \
 256   a_i = WT.imag; \
 257   a1_i = a_i; \
 258   a1_i *= D.real; \
 259   a_i *= D.imag; \
 260   u_r -= a_i; \
 261   u_i = a_r; \
 262   u_i += a1_i; \
 263   a_r = WB.real; \
 264   a1_r = a_r; \
 265   a1_r *= D3.real; \
 266   a_r *= D3.imag; \
 267   a_i = WB.imag; \
 268   a1_i = a_i; \
 269   a_i *= D3.real; \
 270   a1_i *= D3.imag; \
 271   a1_r -= a1_i; \
 272   a_r += a_i; \
 273   v_i = u_r - a1_r; \
 274   u_r += a1_r; \
 275   v_r = a_r - u_i; \
 276   u_i += a_r; \
 277   a_r = A1.real; \
 278   a_i = A1.imag; \
 279   a1_r = a_r; \
 280   a1_r += u_r; \
 281   A1.real = a1_r; \
 282   a_r -= u_r; \
 283   A9.real = a_r; \
 284   a1_i = a_i; \
 285   a1_i += u_i; \
 286   A1.imag = a1_i; \
 287   a_i -= u_i; \
 288   A9.imag = a_i; \
 289   a1_r = A5.real; \
 290   a1_i = A5.imag;  \
 291   a_r = a1_r; \
 292   a1_r -= v_r; \
 293   A5.real = a1_r; \
 294   a_r += v_r; \
 295   A13.real = a_r; \
 296   a_i = a1_i; \
 297   a1_i -= v_i; \
 298   A5.imag = a1_i; \
 299   a_i += v_i; \
 300   A13.imag = a_i; \
 301   }
 302
 303 #endif