libavcodec/faandct.c

   1 /*
   2  * Floating point AAN DCT
   3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c)
   6  *
   7  * This file is part of FFmpeg.
   8  *
   9  * FFmpeg is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * FFmpeg is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with FFmpeg; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  *
  23  * The AAN DCT in this file except ff_faandct248() can also be used under the
  24  * new (3 clause) BSD license.
  25  */
  26
  27 /**
  28  * @file faandct.c
  29  * @brief
  30  *     Floating point AAN DCT
  31  * @author Michael Niedermayer <michaelni@gmx.at>
  32  */
  33
  34 #include "dsputil.h"
  35 #include "faandct.h"
  36
  37 #define FLOAT float
  38 #ifdef FAAN_POSTSCALE
  39 #    define SCALE(x) postscale[x]
  40 #else
  41 #    define SCALE(x) 1
  42 #endif
  43
  44 //numbers generated by simple c code (not as accurate as they could be)
  45 /*
  46 for(i=0; i<8; i++){
  47     printf("#define B%d %1.20llf\n", i, (long double)1.0/(cosl(i*acosl(-1.0)/(long double)16.0)*sqrtl(2)));
  48 }
  49 */
  50 #define B0 1.00000000000000000000
  51 #define B1 0.72095982200694791383 // (cos(pi*1/16)sqrt(2))^-1
  52 #define B2 0.76536686473017954350 // (cos(pi*2/16)sqrt(2))^-1
  53 #define B3 0.85043009476725644878 // (cos(pi*3/16)sqrt(2))^-1
  54 #define B4 1.00000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
  55 #define B5 1.27275858057283393842 // (cos(pi*5/16)sqrt(2))^-1
  56 #define B6 1.84775906502257351242 // (cos(pi*6/16)sqrt(2))^-1
  57 #define B7 3.62450978541155137218 // (cos(pi*7/16)sqrt(2))^-1
  58
  59
  60 #define A1 0.70710678118654752438 // cos(pi*4/16)
  61 #define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
  62 #define A5 0.38268343236508977170 // cos(pi*6/16)
  63 #define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)
  64
  65 static const FLOAT postscale[64]={
  66 B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
  67 B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
  68 B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
  69 B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
  70 B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
  71 B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
  72 B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
  73 B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
  74 };
  75
  76 static av_always_inline void row_fdct(FLOAT temp[64], DCTELEM * data)
  77 {
  78     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  79     FLOAT tmp10, tmp11, tmp12, tmp13;
  80     FLOAT z2, z4, z11, z13;
  81     FLOAT av_unused z5;
  82     int i;
  83
  84     for (i=0; i<8*8; i+=8) {
  85         tmp0= data[0 + i] + data[7 + i];
  86         tmp7= data[0 + i] - data[7 + i];
  87         tmp1= data[1 + i] + data[6 + i];
  88         tmp6= data[1 + i] - data[6 + i];
  89         tmp2= data[2 + i] + data[5 + i];
  90         tmp5= data[2 + i] - data[5 + i];
  91         tmp3= data[3 + i] + data[4 + i];
  92         tmp4= data[3 + i] - data[4 + i];
  93
  94         tmp10= tmp0 + tmp3;
  95         tmp13= tmp0 - tmp3;
  96         tmp11= tmp1 + tmp2;
  97         tmp12= tmp1 - tmp2;
  98
  99         temp[0 + i]= tmp10 + tmp11;
 100         temp[4 + i]= tmp10 - tmp11;
 101
 102         tmp12 += tmp13;
 103         tmp12 *= A1;
 104         temp[2 + i]= tmp13 + tmp12;
 105         temp[6 + i]= tmp13 - tmp12;
 106
 107         tmp4 += tmp5;
 108         tmp5 += tmp6;
 109         tmp6 += tmp7;
 110
 111 #if 0
 112         z5= (tmp4 - tmp6) * A5;
 113         z2= tmp4*A2 + z5;
 114         z4= tmp6*A4 + z5;
 115 #else
 116         z2= tmp4*(A2+A5) - tmp6*A5;
 117         z4= tmp6*(A4-A5) + tmp4*A5;
 118 #endif
 119         tmp5*=A1;
 120
 121         z11= tmp7 + tmp5;
 122         z13= tmp7 - tmp5;
 123
 124         temp[5 + i]= z13 + z2;
 125         temp[3 + i]= z13 - z2;
 126         temp[1 + i]= z11 + z4;
 127         temp[7 + i]= z11 - z4;
 128     }
 129 }
 130
 131 void ff_faandct(DCTELEM * data)
 132 {
 133     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
 134     FLOAT tmp10, tmp11, tmp12, tmp13;
 135     FLOAT z2, z4, z11, z13;
 136     FLOAT av_unused z5;
 137     FLOAT temp[64];
 138     int i;
 139
 140     emms_c();
 141
 142     row_fdct(temp, data);
 143
 144     for (i=0; i<8; i++) {
 145         tmp0= temp[8*0 + i] + temp[8*7 + i];
 146         tmp7= temp[8*0 + i] - temp[8*7 + i];
 147         tmp1= temp[8*1 + i] + temp[8*6 + i];
 148         tmp6= temp[8*1 + i] - temp[8*6 + i];
 149         tmp2= temp[8*2 + i] + temp[8*5 + i];
 150         tmp5= temp[8*2 + i] - temp[8*5 + i];
 151         tmp3= temp[8*3 + i] + temp[8*4 + i];
 152         tmp4= temp[8*3 + i] - temp[8*4 + i];
 153
 154         tmp10= tmp0 + tmp3;
 155         tmp13= tmp0 - tmp3;
 156         tmp11= tmp1 + tmp2;
 157         tmp12= tmp1 - tmp2;
 158
 159         data[8*0 + i]= lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
 160         data[8*4 + i]= lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));
 161
 162         tmp12 += tmp13;
 163         tmp12 *= A1;
 164         data[8*2 + i]= lrintf(SCALE(8*2 + i) * (tmp13 + tmp12));
 165         data[8*6 + i]= lrintf(SCALE(8*6 + i) * (tmp13 - tmp12));
 166
 167         tmp4 += tmp5;
 168         tmp5 += tmp6;
 169         tmp6 += tmp7;
 170
 171 #if 0
 172         z5= (tmp4 - tmp6) * A5;
 173         z2= tmp4*A2 + z5;
 174         z4= tmp6*A4 + z5;
 175 #else
 176         z2= tmp4*(A2+A5) - tmp6*A5;
 177         z4= tmp6*(A4-A5) + tmp4*A5;
 178 #endif
 179         tmp5*=A1;
 180
 181         z11= tmp7 + tmp5;
 182         z13= tmp7 - tmp5;
 183
 184         data[8*5 + i]= lrintf(SCALE(8*5 + i) * (z13 + z2));
 185         data[8*3 + i]= lrintf(SCALE(8*3 + i) * (z13 - z2));
 186         data[8*1 + i]= lrintf(SCALE(8*1 + i) * (z11 + z4));
 187         data[8*7 + i]= lrintf(SCALE(8*7 + i) * (z11 - z4));
 188     }
 189 }
 190
 191 void ff_faandct248(DCTELEM * data)
 192 {
 193     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
 194     FLOAT tmp10, tmp11, tmp12, tmp13;
 195     FLOAT temp[64];
 196     int i;
 197
 198     emms_c();
 199
 200     row_fdct(temp, data);
 201
 202     for (i=0; i<8; i++) {
 203         tmp0 = temp[8*0 + i] + temp[8*1 + i];
 204         tmp1 = temp[8*2 + i] + temp[8*3 + i];
 205         tmp2 = temp[8*4 + i] + temp[8*5 + i];
 206         tmp3 = temp[8*6 + i] + temp[8*7 + i];
 207         tmp4 = temp[8*0 + i] - temp[8*1 + i];
 208         tmp5 = temp[8*2 + i] - temp[8*3 + i];
 209         tmp6 = temp[8*4 + i] - temp[8*5 + i];
 210         tmp7 = temp[8*6 + i] - temp[8*7 + i];
 211
 212         tmp10 = tmp0 + tmp3;
 213         tmp11 = tmp1 + tmp2;
 214         tmp12 = tmp1 - tmp2;
 215         tmp13 = tmp0 - tmp3;
 216
 217         data[8*0 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
 218         data[8*4 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));
 219
 220         tmp12 += tmp13;
 221         tmp12 *= A1;
 222         data[8*2 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + tmp12));
 223         data[8*6 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - tmp12));
 224
 225         tmp10 = tmp4 + tmp7;
 226         tmp11 = tmp5 + tmp6;
 227         tmp12 = tmp5 - tmp6;
 228         tmp13 = tmp4 - tmp7;
 229
 230         data[8*1 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
 231         data[8*5 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));
 232
 233         tmp12 += tmp13;
 234         tmp12 *= A1;
 235         data[8*3 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + tmp12));
 236         data[8*7 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - tmp12));
 237     }
 238 }