liba52/srfftp_3dnow.h

   1 /*
   2  *  srfftp.h
   3  *
   4  *  Copyright (C) Yuqing Deng <Yuqing_Deng@brown.edu> - April 2000
   5  *
   6  *  64 and 128 point split radix fft for ac3dec
   7  *
   8  *  The algorithm is desribed in the book:
   9  *  "Computational Frameworks of the Fast Fourier Transform".
  10  *
  11  *  The ideas and the the organization of code borrowed from djbfft written by
  12  *  D. J. Bernstein <djb@cr.py.to>.  djbff can be found at
  13  *  http://cr.yp.to/djbfft.html.
  14  *
  15  *  srfftp.h is free software; you can redistribute it and/or modify
  16  *  it under the terms of the GNU General Public License as published by
  17  *  the Free Software Foundation; either version 2, or (at your option)
  18  *  any later version.
  19  *
  20  *  srfftp.h is distributed in the hope that it will be useful,
  21  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  23  *  GNU General Public License for more details.
  24  *
  25  *  You should have received a copy of the GNU General Public License
  26  *  along with GNU Make; see the file COPYING.  If not, write to
  27  *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  28  *
  29  *  Modified for using AMD's 3DNow! - 3DNowEx(DSP)! SIMD operations
  30  *  by Nick Kurshev <nickols_k@mail.ru>
  31  */
  32
  33 #ifndef SRFFTP_3DNOW_H__
  34 #define SRFFTP_3DNOW_H__
  35
  36 typedef struct
  37 {
  38   unsigned long val[2];
  39 }i_cmplx_t;
  40
  41 #define TRANS_FILL_MM6_MM7_3DNOW()\
  42     __asm__ __volatile__(\
  43         "movq   %1, %%mm7\n\t"\
  44         "movq   %0, %%mm6\n\t"\
  45         ::"m"(x_plus_minus_3dnow),\
  46         "m"(x_minus_plus_3dnow)\
  47         :"memory");
  48
  49 #ifdef HAVE_3DNOWEX
  50 #define PSWAP_MM(mm_base,mm_hlp) "pswapd        "mm_base","mm_base"\n\t"
  51 #else
  52 #define PSWAP_MM(mm_base,mm_hlp)\
  53         "movq   "mm_base","mm_hlp"\n\t"\
  54         "psrlq $32, "mm_base"\n\t"\
  55         "punpckldq "mm_hlp","mm_base"\n\t"
  56 #endif
  57 #ifdef HAVE_3DNOWEX
  58 #define PFNACC_MM(mm_base,mm_hlp)       "pfnacc "mm_base","mm_base"\n\t"
  59 #else
  60 #define PFNACC_MM(mm_base,mm_hlp)\
  61         "movq   "mm_base","mm_hlp"\n\t"\
  62         "psrlq  $32,"mm_hlp"\n\t"\
  63         "punpckldq "mm_hlp","mm_hlp"\n\t"\
  64         "pfsub  "mm_hlp","mm_base"\n\t"
  65 #endif
  66
  67 #define TRANSZERO_3DNOW(A0,A4,A8,A12) \
  68 { \
  69     __asm__ __volatile__(\
  70         "movq   %4, %%mm0\n\t" /* mm0 = wTB[0]*/\
  71         "movq   %5, %%mm1\n\t" /* mm1 = wTB[k*2]*/ \
  72         "movq   %%mm0, %%mm5\n\t"/*u.re = wTB[0].re + wTB[k*2].re;*/\
  73         "pfadd  %%mm1, %%mm5\n\t"/*u.im = wTB[0].im + wTB[k*2].im; mm5 = u*/\
  74         "pxor   %%mm6, %%mm0\n\t"/*mm0 = wTB[0].re | -wTB[0].im */\
  75         "pxor   %%mm7, %%mm1\n\t"/*mm1 = -wTB[k*2].re | wTB[k*2].im */\
  76         "pfadd  %%mm1, %%mm0\n\t"/*v.im = wTB[0].re - wTB[k*2].re;*/\
  77         "movq   %%mm0, %%mm4\n\t"/*v.re =-wTB[0].im + wTB[k*2].im;*/\
  78         PSWAP_MM("%%mm4","%%mm2")/* mm4 = v*/\
  79         "movq   %6, %%mm0\n\t" /* a1 = A0;*/\
  80         "movq   %7, %%mm2\n\t" /* a1 = A4;*/\
  81         "movq   %%mm0, %%mm1\n\t"\
  82         "movq   %%mm2, %%mm3\n\t"\
  83         "pfadd  %%mm5, %%mm0\n\t" /*A0 = a1 + u;*/\
  84         "pfadd  %%mm4, %%mm2\n\t" /*A12 = a1 + v;*/\
  85         "movq   %%mm0, %0\n\t"\
  86         "pfsub  %%mm5, %%mm1\n\t" /*A1 = a1 - u;*/\
  87         "movq   %%mm2, %3\n\t"\
  88         "pfsub  %%mm4, %%mm3\n\t" /*A4  = a1 - v;*/\
  89         "movq   %%mm1, %1\n\t"\
  90         "movq   %%mm3, %2"\
  91         :"=m"(A0), "=m"(A8), "=m"(A4), "=m"(A12)\
  92         :"m"(wTB[0]), "m"(wTB[k*2]), "m"(A0), "m"(A4)\
  93         :"memory");\
  94 }
  95
  96 #define TRANSHALF_16_3DNOW(A2,A6,A10,A14)\
  97 {\
  98     __asm__ __volatile__(\
  99         "movq   %4, %%mm0\n\t"/*u.re = wTB[2].im + wTB[2].re;*/\
 100         "movq   %%mm0, %%mm1\n\t"\
 101         "pxor   %%mm7, %%mm1\n\t"\
 102         "pfacc  %%mm1, %%mm0\n\t"/*u.im = wTB[2].im - wTB[2].re; mm0 = u*/\
 103         "movq   %5, %%mm1\n\t"  /*a.re = wTB[6].im - wTB[6].re; */\
 104         "movq   %%mm1, %%mm2\n\t"\
 105         "pxor   %%mm7, %%mm1\n\t"\
 106         "pfacc  %%mm2, %%mm1\n\t"/*a.im = wTB[6].im + wTB[6].re;  mm1 = a*/\
 107         "movq   %%mm1, %%mm2\n\t"\
 108         "pxor   %%mm7, %%mm2\n\t"/*v.im = u.re - a.re;*/\
 109         "movq   %%mm0, %%mm3\n\t"/*v.re = u.im + a.im;*/\
 110         "pfadd  %%mm2, %%mm3\n\t"\
 111         PSWAP_MM("%%mm3","%%mm2")/*mm3 = v*/\
 112         "pxor   %%mm6, %%mm1\n\t"/*u.re = u.re + a.re;*/\
 113         "pfadd  %%mm1, %%mm0\n\t"/*u.im = u.im - a.im; mm0 = u*/\
 114         "movq   %8, %%mm2\n\t"\
 115         "pfmul  %%mm2, %%mm3\n\t" /* v *= HSQRT2_3DNOW; */\
 116         "pfmul  %%mm2, %%mm0\n\t" /* u *= HSQRT2_3DNOW; */\
 117         "movq   %6, %%mm1\n\t" /* a1 = A2;*/\
 118         "movq   %7, %%mm5\n\t" /* a1 = A6;*/\
 119         "movq   %%mm1, %%mm2\n\t"\
 120         "movq   %%mm3, %%mm4\n\t"\
 121         "pfadd  %%mm0, %%mm1\n\t" /*A2 = a1 + u;*/\
 122         "pxor   %%mm6, %%mm4\n\t"/*A6.re  = a1.re + v.re;*/\
 123         "pfsub  %%mm0, %%mm2\n\t" /*A2 = a1 - u;*/\
 124         "pxor   %%mm7, %%mm3\n\t"/*A14.re = a1.re - v.re;*/\
 125         "movq   %%mm1, %0\n\t"\
 126         "movq   %%mm2, %1\n\t"\
 127         "movq   %%mm5, %%mm2\n\t"\
 128         "pfadd  %%mm4, %%mm5\n\t"/*A6.im  = a1.im - v.im;*/\
 129         "pfadd  %%mm3, %%mm2\n\t"/*A14.im = a1.im + v.im;*/\
 130         "movq   %%mm5, %2\n\t"\
 131         "movq   %%mm2, %3"\
 132         :"=m"(A2), "=m"(A10), "=m"(A6), "=m"(A14)\
 133         :"m"(wTB[2]), "m"(wTB[6]), "m"(A2), "m"(A6), "m"(HSQRT2_3DNOW)\
 134         :"memory");\
 135 }
 136
 137 #define TRANS_3DNOW(A1,A5,A9,A13,WT,WB,D,D3)\
 138 { \
 139     __asm__ __volatile__(\
 140         "movq   %1,     %%mm4\n\t"\
 141         "movq   %%mm4,  %%mm5\n\t"\
 142         "punpckldq %%mm4, %%mm4\n\t"/*mm4 = D.re | D.re */\
 143         "punpckhdq %%mm5, %%mm5\n\t"/*mm5 = D.im | D.im */\
 144         "movq   %0,     %%mm0\n\t"\
 145         "pfmul  %%mm0,  %%mm4\n\t"/* mm4 =u.re | u.im */\
 146         "pfmul  %%mm0,  %%mm5\n\t"/* mm5 = a.re | a.im */\
 147         PSWAP_MM("%%mm5","%%mm3")\
 148         "pxor   %%mm7,  %%mm5\n\t"\
 149         "pfadd  %%mm5,  %%mm4\n\t"/* mm4 = u*/\
 150         "movq   %3,     %%mm1\n\t"\
 151         "movq   %2,     %%mm0\n\t"\
 152         PSWAP_MM("%%mm1","%%mm3")\
 153         "movq   %%mm0,  %%mm2\n\t"\
 154         "pfmul  %%mm1,  %%mm0\n\t"/* mm0 = a*/\
 155         "pfmul  %3,     %%mm2\n\t"/* mm2 = v*/\
 156         PFNACC_MM("%%mm2","%%mm3")\
 157         "pfacc  %%mm0,  %%mm0\n\t"\
 158         "movq   %%mm4,  %%mm5\n\t"\
 159         "punpckldq %%mm0,%%mm2\n\t"/*mm2 = v.re | a.re*/\
 160         "pxor   %%mm6,  %%mm5\n\t"\
 161         "movq   %%mm2,  %%mm3\n\t"\
 162         "pxor   %%mm7,  %%mm3\n\t"\
 163         "pfadd  %%mm3,  %%mm5\n\t"\
 164         PSWAP_MM("%%mm5","%%mm3")/* mm5 = v*/\
 165         "pfadd  %%mm2,  %%mm4\n\t"\
 166         :\
 167         :"m"(WT), "m"(D), "m"(WB), "m"(D3)\
 168         :"memory");\
 169     __asm__ __volatile__(\
 170         "movq   %4, %%mm0\n\t"/* a1 = A1*/\
 171         "movq   %5, %%mm2\n\t"/* a1 = A5*/\
 172         "movq   %%mm0, %%mm1\n\t"\
 173         "movq   %%mm2, %%mm3\n\t"\
 174         "pfadd  %%mm4, %%mm0\n\t"/*A1 = a1 + u*/\
 175         "pfsub  %%mm5, %%mm2\n\t"/*A5 = a1 - v*/\
 176         "movq   %%mm0, %0\n\t"\
 177         "pfsub  %%mm4, %%mm1\n\t"/*A9 = a1 - u*/\
 178         "movq   %%mm2, %2\n\t"\
 179         "pfadd  %%mm5, %%mm3\n\t"/*A9 = a1 + v*/\
 180         "movq   %%mm1, %1\n\t"\
 181         "movq   %%mm3, %3"\
 182         :"=m"(A1), "=m"(A9), "=m"(A5), "=m"(A13)\
 183         :"m"(A1), "m"(A5)\
 184         :"memory");\
 185 }
 186
 187 #endif