FFmpeg: libavcodec/faandct.c Source File

00001 /*
00002  * Floating point AAN DCT
00003  * this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c)
00004  *
00005  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
00006  * Copyright (c) 2003 Roman Shaposhnik
00007  *
00008  * Permission to use, copy, modify, and/or distribute this software for any
00009  * purpose with or without fee is hereby granted, provided that the above
00010  * copyright notice and this permission notice appear in all copies.
00011  *
00012  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
00013  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
00014  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
00015  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
00016  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
00017  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
00018  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
00019  */
00020 
00028 #include "dsputil.h"
00029 #include "faandct.h"
00030 
00031 #define FLOAT float
00032 
00033 //numbers generated by simple c code (not as accurate as they could be)
00034 /*
00035 for(i=0; i<8; i++){
00036     printf("#define B%d %1.20llf\n", i, (long double)1.0/(cosl(i*acosl(-1.0)/(long double)16.0)*sqrtl(2)));
00037 }
00038 */
00039 #define B0 1.00000000000000000000
00040 #define B1 0.72095982200694791383 // (cos(pi*1/16)sqrt(2))^-1
00041 #define B2 0.76536686473017954350 // (cos(pi*2/16)sqrt(2))^-1
00042 #define B3 0.85043009476725644878 // (cos(pi*3/16)sqrt(2))^-1
00043 #define B4 1.00000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
00044 #define B5 1.27275858057283393842 // (cos(pi*5/16)sqrt(2))^-1
00045 #define B6 1.84775906502257351242 // (cos(pi*6/16)sqrt(2))^-1
00046 #define B7 3.62450978541155137218 // (cos(pi*7/16)sqrt(2))^-1
00047 
00048 
00049 #define A1 0.70710678118654752438 // cos(pi*4/16)
00050 #define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
00051 #define A5 0.38268343236508977170 // cos(pi*6/16)
00052 #define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)
00053 
00054 static const FLOAT postscale[64]={
00055 B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
00056 B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
00057 B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
00058 B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
00059 B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
00060 B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
00061 B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
00062 B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
00063 };
00064 
00065 static av_always_inline void row_fdct(FLOAT temp[64], DCTELEM * data)
00066 {
00067     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
00068     FLOAT tmp10, tmp11, tmp12, tmp13;
00069     FLOAT z2, z4, z11, z13;
00070     FLOAT av_unused z5;
00071     int i;
00072 
00073     for (i=0; i<8*8; i+=8) {
00074         tmp0= data[0 + i] + data[7 + i];
00075         tmp7= data[0 + i] - data[7 + i];
00076         tmp1= data[1 + i] + data[6 + i];
00077         tmp6= data[1 + i] - data[6 + i];
00078         tmp2= data[2 + i] + data[5 + i];
00079         tmp5= data[2 + i] - data[5 + i];
00080         tmp3= data[3 + i] + data[4 + i];
00081         tmp4= data[3 + i] - data[4 + i];
00082 
00083         tmp10= tmp0 + tmp3;
00084         tmp13= tmp0 - tmp3;
00085         tmp11= tmp1 + tmp2;
00086         tmp12= tmp1 - tmp2;
00087 
00088         temp[0 + i]= tmp10 + tmp11;
00089         temp[4 + i]= tmp10 - tmp11;
00090 
00091         tmp12 += tmp13;
00092         tmp12 *= A1;
00093         temp[2 + i]= tmp13 + tmp12;
00094         temp[6 + i]= tmp13 - tmp12;
00095 
00096         tmp4 += tmp5;
00097         tmp5 += tmp6;
00098         tmp6 += tmp7;
00099 
00100 #if 0
00101         z5= (tmp4 - tmp6) * A5;
00102         z2= tmp4*A2 + z5;
00103         z4= tmp6*A4 + z5;
00104 #else
00105         z2= tmp4*(A2+A5) - tmp6*A5;
00106         z4= tmp6*(A4-A5) + tmp4*A5;
00107 #endif
00108         tmp5*=A1;
00109 
00110         z11= tmp7 + tmp5;
00111         z13= tmp7 - tmp5;
00112 
00113         temp[5 + i]= z13 + z2;
00114         temp[3 + i]= z13 - z2;
00115         temp[1 + i]= z11 + z4;
00116         temp[7 + i]= z11 - z4;
00117     }
00118 }
00119 
00120 void ff_faandct(DCTELEM * data)
00121 {
00122     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
00123     FLOAT tmp10, tmp11, tmp12, tmp13;
00124     FLOAT z2, z4, z11, z13;
00125     FLOAT av_unused z5;
00126     FLOAT temp[64];
00127     int i;
00128 
00129     emms_c();
00130 
00131     row_fdct(temp, data);
00132 
00133     for (i=0; i<8; i++) {
00134         tmp0= temp[8*0 + i] + temp[8*7 + i];
00135         tmp7= temp[8*0 + i] - temp[8*7 + i];
00136         tmp1= temp[8*1 + i] + temp[8*6 + i];
00137         tmp6= temp[8*1 + i] - temp[8*6 + i];
00138         tmp2= temp[8*2 + i] + temp[8*5 + i];
00139         tmp5= temp[8*2 + i] - temp[8*5 + i];
00140         tmp3= temp[8*3 + i] + temp[8*4 + i];
00141         tmp4= temp[8*3 + i] - temp[8*4 + i];
00142 
00143         tmp10= tmp0 + tmp3;
00144         tmp13= tmp0 - tmp3;
00145         tmp11= tmp1 + tmp2;
00146         tmp12= tmp1 - tmp2;
00147 
00148         data[8*0 + i]= lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
00149         data[8*4 + i]= lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
00150 
00151         tmp12 += tmp13;
00152         tmp12 *= A1;
00153         data[8*2 + i]= lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
00154         data[8*6 + i]= lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
00155 
00156         tmp4 += tmp5;
00157         tmp5 += tmp6;
00158         tmp6 += tmp7;
00159 
00160 #if 0
00161         z5= (tmp4 - tmp6) * A5;
00162         z2= tmp4*A2 + z5;
00163         z4= tmp6*A4 + z5;
00164 #else
00165         z2= tmp4*(A2+A5) - tmp6*A5;
00166         z4= tmp6*(A4-A5) + tmp4*A5;
00167 #endif
00168         tmp5*=A1;
00169 
00170         z11= tmp7 + tmp5;
00171         z13= tmp7 - tmp5;
00172 
00173         data[8*5 + i]= lrintf(postscale[8*5 + i] * (z13 + z2));
00174         data[8*3 + i]= lrintf(postscale[8*3 + i] * (z13 - z2));
00175         data[8*1 + i]= lrintf(postscale[8*1 + i] * (z11 + z4));
00176         data[8*7 + i]= lrintf(postscale[8*7 + i] * (z11 - z4));
00177     }
00178 }
00179 
00180 void ff_faandct248(DCTELEM * data)
00181 {
00182     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
00183     FLOAT tmp10, tmp11, tmp12, tmp13;
00184     FLOAT temp[64];
00185     int i;
00186 
00187     emms_c();
00188 
00189     row_fdct(temp, data);
00190 
00191     for (i=0; i<8; i++) {
00192         tmp0 = temp[8*0 + i] + temp[8*1 + i];
00193         tmp1 = temp[8*2 + i] + temp[8*3 + i];
00194         tmp2 = temp[8*4 + i] + temp[8*5 + i];
00195         tmp3 = temp[8*6 + i] + temp[8*7 + i];
00196         tmp4 = temp[8*0 + i] - temp[8*1 + i];
00197         tmp5 = temp[8*2 + i] - temp[8*3 + i];
00198         tmp6 = temp[8*4 + i] - temp[8*5 + i];
00199         tmp7 = temp[8*6 + i] - temp[8*7 + i];
00200 
00201         tmp10 = tmp0 + tmp3;
00202         tmp11 = tmp1 + tmp2;
00203         tmp12 = tmp1 - tmp2;
00204         tmp13 = tmp0 - tmp3;
00205 
00206         data[8*0 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
00207         data[8*4 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
00208 
00209         tmp12 += tmp13;
00210         tmp12 *= A1;
00211         data[8*2 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
00212         data[8*6 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
00213 
00214         tmp10 = tmp4 + tmp7;
00215         tmp11 = tmp5 + tmp6;
00216         tmp12 = tmp5 - tmp6;
00217         tmp13 = tmp4 - tmp7;
00218 
00219         data[8*1 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
00220         data[8*5 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
00221 
00222         tmp12 += tmp13;
00223         tmp12 *= A1;
00224         data[8*3 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
00225         data[8*7 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
00226     }
00227 }