libavcodec/faandct.c

   1 /*
   2  * Floating point AAN DCT
   3  * this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c)
   4  *
   5  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   6  * Copyright (c) 2003 Roman Shaposhnik
   7  *
   8  * Permission to use, copy, modify, and/or distribute this software for any
   9  * purpose with or without fee is hereby granted, provided that the above
  10  * copyright notice and this permission notice appear in all copies.
  11  *
  12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
  13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
  14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
  15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
  16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
  17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
  19  */
  20
  21 /**
  22  * @file
  23  * @brief
  24  *     Floating point AAN DCT
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #include "dsputil.h"
  29 #include "faandct.h"
  30 #include "libavutil/internal.h"
  31 #include "libavutil/libm.h"
  32
  33 #define FLOAT float
  34
  35 //numbers generated by simple c code (not as accurate as they could be)
  36 /*
  37 for(i=0; i<8; i++){
  38     printf("#define B%d %1.20llf\n", i, (long double)1.0/(cosl(i*acosl(-1.0)/(long double)16.0)*sqrtl(2)));
  39 }
  40 */
  41 #define B0 1.00000000000000000000
  42 #define B1 0.72095982200694791383 // (cos(pi*1/16)sqrt(2))^-1
  43 #define B2 0.76536686473017954350 // (cos(pi*2/16)sqrt(2))^-1
  44 #define B3 0.85043009476725644878 // (cos(pi*3/16)sqrt(2))^-1
  45 #define B4 1.00000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
  46 #define B5 1.27275858057283393842 // (cos(pi*5/16)sqrt(2))^-1
  47 #define B6 1.84775906502257351242 // (cos(pi*6/16)sqrt(2))^-1
  48 #define B7 3.62450978541155137218 // (cos(pi*7/16)sqrt(2))^-1
  49
  50
  51 #define A1 0.70710678118654752438 // cos(pi*4/16)
  52 #define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
  53 #define A5 0.38268343236508977170 // cos(pi*6/16)
  54 #define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)
  55
  56 static const FLOAT postscale[64]={
  57 B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
  58 B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
  59 B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
  60 B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
  61 B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
  62 B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
  63 B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
  64 B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
  65 };
  66
  67 static av_always_inline void row_fdct(FLOAT temp[64], int16_t *data)
  68 {
  69     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  70     FLOAT tmp10, tmp11, tmp12, tmp13;
  71     FLOAT z2, z4, z11, z13;
  72     FLOAT av_unused z5;
  73     int i;
  74
  75     for (i=0; i<8*8; i+=8) {
  76         tmp0= data[0 + i] + data[7 + i];
  77         tmp7= data[0 + i] - data[7 + i];
  78         tmp1= data[1 + i] + data[6 + i];
  79         tmp6= data[1 + i] - data[6 + i];
  80         tmp2= data[2 + i] + data[5 + i];
  81         tmp5= data[2 + i] - data[5 + i];
  82         tmp3= data[3 + i] + data[4 + i];
  83         tmp4= data[3 + i] - data[4 + i];
  84
  85         tmp10= tmp0 + tmp3;
  86         tmp13= tmp0 - tmp3;
  87         tmp11= tmp1 + tmp2;
  88         tmp12= tmp1 - tmp2;
  89
  90         temp[0 + i]= tmp10 + tmp11;
  91         temp[4 + i]= tmp10 - tmp11;
  92
  93         tmp12 += tmp13;
  94         tmp12 *= A1;
  95         temp[2 + i]= tmp13 + tmp12;
  96         temp[6 + i]= tmp13 - tmp12;
  97
  98         tmp4 += tmp5;
  99         tmp5 += tmp6;
 100         tmp6 += tmp7;
 101
 102 #if 0
 103         z5= (tmp4 - tmp6) * A5;
 104         z2= tmp4*A2 + z5;
 105         z4= tmp6*A4 + z5;
 106 #else
 107         z2= tmp4*(A2+A5) - tmp6*A5;
 108         z4= tmp6*(A4-A5) + tmp4*A5;
 109 #endif
 110         tmp5*=A1;
 111
 112         z11= tmp7 + tmp5;
 113         z13= tmp7 - tmp5;
 114
 115         temp[5 + i]= z13 + z2;
 116         temp[3 + i]= z13 - z2;
 117         temp[1 + i]= z11 + z4;
 118         temp[7 + i]= z11 - z4;
 119     }
 120 }
 121
 122 void ff_faandct(int16_t *data)
 123 {
 124     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
 125     FLOAT tmp10, tmp11, tmp12, tmp13;
 126     FLOAT z2, z4, z11, z13;
 127     FLOAT av_unused z5;
 128     FLOAT temp[64];
 129     int i;
 130
 131     emms_c();
 132
 133     row_fdct(temp, data);
 134
 135     for (i=0; i<8; i++) {
 136         tmp0= temp[8*0 + i] + temp[8*7 + i];
 137         tmp7= temp[8*0 + i] - temp[8*7 + i];
 138         tmp1= temp[8*1 + i] + temp[8*6 + i];
 139         tmp6= temp[8*1 + i] - temp[8*6 + i];
 140         tmp2= temp[8*2 + i] + temp[8*5 + i];
 141         tmp5= temp[8*2 + i] - temp[8*5 + i];
 142         tmp3= temp[8*3 + i] + temp[8*4 + i];
 143         tmp4= temp[8*3 + i] - temp[8*4 + i];
 144
 145         tmp10= tmp0 + tmp3;
 146         tmp13= tmp0 - tmp3;
 147         tmp11= tmp1 + tmp2;
 148         tmp12= tmp1 - tmp2;
 149
 150         data[8*0 + i]= lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
 151         data[8*4 + i]= lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
 152
 153         tmp12 += tmp13;
 154         tmp12 *= A1;
 155         data[8*2 + i]= lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
 156         data[8*6 + i]= lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
 157
 158         tmp4 += tmp5;
 159         tmp5 += tmp6;
 160         tmp6 += tmp7;
 161
 162 #if 0
 163         z5= (tmp4 - tmp6) * A5;
 164         z2= tmp4*A2 + z5;
 165         z4= tmp6*A4 + z5;
 166 #else
 167         z2= tmp4*(A2+A5) - tmp6*A5;
 168         z4= tmp6*(A4-A5) + tmp4*A5;
 169 #endif
 170         tmp5*=A1;
 171
 172         z11= tmp7 + tmp5;
 173         z13= tmp7 - tmp5;
 174
 175         data[8*5 + i]= lrintf(postscale[8*5 + i] * (z13 + z2));
 176         data[8*3 + i]= lrintf(postscale[8*3 + i] * (z13 - z2));
 177         data[8*1 + i]= lrintf(postscale[8*1 + i] * (z11 + z4));
 178         data[8*7 + i]= lrintf(postscale[8*7 + i] * (z11 - z4));
 179     }
 180 }
 181
 182 void ff_faandct248(int16_t *data)
 183 {
 184     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
 185     FLOAT tmp10, tmp11, tmp12, tmp13;
 186     FLOAT temp[64];
 187     int i;
 188
 189     emms_c();
 190
 191     row_fdct(temp, data);
 192
 193     for (i=0; i<8; i++) {
 194         tmp0 = temp[8*0 + i] + temp[8*1 + i];
 195         tmp1 = temp[8*2 + i] + temp[8*3 + i];
 196         tmp2 = temp[8*4 + i] + temp[8*5 + i];
 197         tmp3 = temp[8*6 + i] + temp[8*7 + i];
 198         tmp4 = temp[8*0 + i] - temp[8*1 + i];
 199         tmp5 = temp[8*2 + i] - temp[8*3 + i];
 200         tmp6 = temp[8*4 + i] - temp[8*5 + i];
 201         tmp7 = temp[8*6 + i] - temp[8*7 + i];
 202
 203         tmp10 = tmp0 + tmp3;
 204         tmp11 = tmp1 + tmp2;
 205         tmp12 = tmp1 - tmp2;
 206         tmp13 = tmp0 - tmp3;
 207
 208         data[8*0 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
 209         data[8*4 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
 210
 211         tmp12 += tmp13;
 212         tmp12 *= A1;
 213         data[8*2 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
 214         data[8*6 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
 215
 216         tmp10 = tmp4 + tmp7;
 217         tmp11 = tmp5 + tmp6;
 218         tmp12 = tmp5 - tmp6;
 219         tmp13 = tmp4 - tmp7;
 220
 221         data[8*1 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
 222         data[8*5 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
 223
 224         tmp12 += tmp13;
 225         tmp12 *= A1;
 226         data[8*3 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
 227         data[8*7 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
 228     }
 229 }