git.gag.com Git - debian/gnuradio/blob - gcell/include/gcell/spu/fft_1d.h

   1 /* --------------------------------------------------------------  */
   2 /* (C)Copyright 2001,2007,                                         */
   3 /* International Business Machines Corporation,                    */
   4 /* Sony Computer Entertainment, Incorporated,                      */
   5 /* Toshiba Corporation,                                            */
   6 /*                                                                 */
   7 /* All Rights Reserved.                                            */
   8 /*                                                                 */
   9 /* Redistribution and use in source and binary forms, with or      */
  10 /* without modification, are permitted provided that the           */
  11 /* following conditions are met:                                   */
  12 /*                                                                 */
  13 /* - Redistributions of source code must retain the above copyright*/
  14 /*   notice, this list of conditions and the following disclaimer. */
  15 /*                                                                 */
  16 /* - Redistributions in binary form must reproduce the above       */
  17 /*   copyright notice, this list of conditions and the following   */
  18 /*   disclaimer in the documentation and/or other materials        */
  19 /*   provided with the distribution.                               */
  20 /*                                                                 */
  21 /* - Neither the name of IBM Corporation nor the names of its      */
  22 /*   contributors may be used to endorse or promote products       */
  23 /*   derived from this software without specific prior written     */
  24 /*   permission.                                                   */
  25 /*                                                                 */
  26 /* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND          */
  27 /* CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,     */
  28 /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF        */
  29 /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE        */
  30 /* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR            */
  31 /* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,    */
  32 /* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT    */
  33 /* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;    */
  34 /* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)        */
  35 /* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN       */
  36 /* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR    */
  37 /* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,  */
  38 /* EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.              */
  39 /* --------------------------------------------------------------  */
  40 /* PROLOG END TAG zYx                                              */
  41 #ifndef _FFT_1D_H_
  42 #define _FFT_1D_H_      1
  43
  44 #include <spu_intrinsics.h>
  45
  46 /* BIT_SWAP - swaps up to 16 bits of the integer _i according to the
  47  *            pattern specified by _pat.
  48  */
  49 #define BIT_SWAP(_i, _pat)        spu_extract(spu_gather(spu_shuffle(spu_maskb(_i), _pat, _pat)), 0)
  50
  51
  52 #ifndef MAX_FFT_1D_SIZE
  53 #define MAX_FFT_1D_SIZE 8192
  54 #endif
  55
  56 #ifndef INV_SQRT_2
  57 #define INV_SQRT_2      0.7071067811865
  58 #endif
  59
  60
  61 /* The following macro, FFT_1D_BUTTERFLY, performs a 4 way SIMD basic butterfly
  62  * operation. The inputs are in parallel arrays (seperate real and imaginary
  63  * vectors).
  64  *
  65  *          p --------------------------> P = p + q*Wi
  66  *                        \      /
  67  *                         \    /
  68  *                          \  /
  69  *                           \/
  70  *                           /\
  71  *                          /  \
  72  *                         /    \
  73  *               ____     /      \
  74  *          q --| Wi |-----------------> Q = p - q*Wi
  75  *               ----
  76  */
  77
  78 #define FFT_1D_BUTTERFLY(_P_re, _P_im, _Q_re, _Q_im, _p_re, _p_im, _q_re, _q_im, _W_re, _W_im) {        \
  79   vector float _qw_re, _qw_im;                                                                          \
  80                                                                                                         \
  81   _qw_re = spu_msub(_q_re, _W_re, spu_mul(_q_im, _W_im));                                               \
  82   _qw_im = spu_madd(_q_re, _W_im, spu_mul(_q_im, _W_re));                                               \
  83   _P_re  = spu_add(_p_re, _qw_re);                                                                      \
  84   _P_im  = spu_add(_p_im, _qw_im);                                                                      \
  85   _Q_re  = spu_sub(_p_re, _qw_re);                                                                      \
  86   _Q_im  = spu_sub(_p_im, _qw_im);                                                                      \
  87 }
  88
  89
  90 /* FFT_1D_BUTTERFLY_HI is equivalent to FFT_1D_BUTTERFLY with twiddle factors (W_im, -W_re)
  91  */
  92 #define FFT_1D_BUTTERFLY_HI(_P_re, _P_im, _Q_re, _Q_im, _p_re, _p_im, _q_re, _q_im, _W_re, _W_im) {     \
  93   vector float _qw_re, _qw_im;                                                                          \
  94                                                                                                         \
  95   _qw_re = spu_madd(_q_re, _W_im, spu_mul(_q_im, _W_re));                                               \
  96   _qw_im = spu_msub(_q_im, _W_im, spu_mul(_q_re, _W_re));                                               \
  97   _P_re  = spu_add(_p_re, _qw_re);                                                                      \
  98   _P_im  = spu_add(_p_im, _qw_im);                                                                      \
  99   _Q_re  = spu_sub(_p_re, _qw_re);                                                                      \
 100   _Q_im  = spu_sub(_p_im, _qw_im);                                                                      \
 101 }
 102
 103 #endif /* _FFT_1D_H_ */