Cinelerra: hvirtual/quicktime/ffmpeg/libavcodec/faandct.c Source File

00001 /*
00002  * Floating point AAN DCT
00003  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
00004  *
00005  * This library is free software; you can redistribute it and/or
00006  * modify it under the terms of the GNU Lesser General Public
00007  * License as published by the Free Software Foundation; either
00008  * version 2 of the License, or (at your option) any later version.
00009  *
00010  * This library is distributed in the hope that it will be useful,
00011  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00012  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00013  * Lesser General Public License for more details.
00014  *
00015  * You should have received a copy of the GNU Lesser General Public
00016  * License along with this library; if not, write to the Free Software
00017  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
00018  *
00019  * this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c)
00020  */
00021 
00029 #include "dsputil.h"
00030 #include "faandct.h"
00031 
00032 #define FLOAT float
00033 #ifdef FAAN_POSTSCALE
00034 #    define SCALE(x) postscale[x]
00035 #else
00036 #    define SCALE(x) 1
00037 #endif
00038 
00039 //numbers generated by simple c code (not as accurate as they could be)
00040 /*
00041 for(i=0; i<8; i++){
00042     printf("#define B%d %1.20llf\n", i, (long double)1.0/(cosl(i*acosl(-1.0)/(long double)16.0)*sqrtl(2)));
00043 }
00044 */
00045 #define B0 1.00000000000000000000
00046 #define B1 0.72095982200694791383 // (cos(pi*1/16)sqrt(2))^-1
00047 #define B2 0.76536686473017954350 // (cos(pi*2/16)sqrt(2))^-1
00048 #define B3 0.85043009476725644878 // (cos(pi*3/16)sqrt(2))^-1
00049 #define B4 1.00000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
00050 #define B5 1.27275858057283393842 // (cos(pi*5/16)sqrt(2))^-1
00051 #define B6 1.84775906502257351242 // (cos(pi*6/16)sqrt(2))^-1
00052 #define B7 3.62450978541155137218 // (cos(pi*7/16)sqrt(2))^-1
00053 
00054 
00055 #define A1 0.70710678118654752438 // cos(pi*4/16)
00056 #define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
00057 #define A5 0.38268343236508977170 // cos(pi*6/16)
00058 #define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)
00059 
00060 static FLOAT postscale[64]={
00061 B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
00062 B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
00063 B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
00064 B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
00065 B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
00066 B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
00067 B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
00068 B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
00069 };
00070 
00071 static always_inline void row_fdct(FLOAT temp[64], DCTELEM * data)
00072 {
00073     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
00074     FLOAT tmp10, tmp11, tmp12, tmp13;
00075     FLOAT z1, z2, z3, z4, z5, z11, z13;
00076     int i;
00077 
00078     for (i=0; i<8*8; i+=8) {
00079         tmp0= data[0 + i] + data[7 + i];
00080         tmp7= data[0 + i] - data[7 + i];
00081         tmp1= data[1 + i] + data[6 + i];
00082         tmp6= data[1 + i] - data[6 + i];
00083         tmp2= data[2 + i] + data[5 + i];
00084         tmp5= data[2 + i] - data[5 + i];
00085         tmp3= data[3 + i] + data[4 + i];
00086         tmp4= data[3 + i] - data[4 + i];
00087         
00088         tmp10= tmp0 + tmp3;
00089         tmp13= tmp0 - tmp3;
00090         tmp11= tmp1 + tmp2;
00091         tmp12= tmp1 - tmp2;
00092         
00093         temp[0 + i]= tmp10 + tmp11;
00094         temp[4 + i]= tmp10 - tmp11;
00095         
00096         z1= (tmp12 + tmp13)*A1;
00097         temp[2 + i]= tmp13 + z1;
00098         temp[6 + i]= tmp13 - z1;
00099         
00100         tmp10= tmp4 + tmp5;
00101         tmp11= tmp5 + tmp6;
00102         tmp12= tmp6 + tmp7;
00103 
00104         z5= (tmp10 - tmp12) * A5;
00105         z2= tmp10*A2 + z5;
00106         z4= tmp12*A4 + z5;
00107         z3= tmp11*A1;
00108 
00109         z11= tmp7 + z3;
00110         z13= tmp7 - z3;
00111 
00112         temp[5 + i]= z13 + z2;
00113         temp[3 + i]= z13 - z2;
00114         temp[1 + i]= z11 + z4;
00115         temp[7 + i]= z11 - z4;
00116     }    
00117 }
00118 
00119 void ff_faandct(DCTELEM * data)
00120 {
00121     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
00122     FLOAT tmp10, tmp11, tmp12, tmp13;
00123     FLOAT z1, z2, z3, z4, z5, z11, z13;
00124     FLOAT temp[64];
00125     int i;
00126 
00127     emms_c();
00128 
00129     row_fdct(temp, data);
00130 
00131     for (i=0; i<8; i++) {
00132         tmp0= temp[8*0 + i] + temp[8*7 + i];
00133         tmp7= temp[8*0 + i] - temp[8*7 + i];
00134         tmp1= temp[8*1 + i] + temp[8*6 + i];
00135         tmp6= temp[8*1 + i] - temp[8*6 + i];
00136         tmp2= temp[8*2 + i] + temp[8*5 + i];
00137         tmp5= temp[8*2 + i] - temp[8*5 + i];
00138         tmp3= temp[8*3 + i] + temp[8*4 + i];
00139         tmp4= temp[8*3 + i] - temp[8*4 + i];
00140         
00141         tmp10= tmp0 + tmp3;
00142         tmp13= tmp0 - tmp3;
00143         tmp11= tmp1 + tmp2;
00144         tmp12= tmp1 - tmp2;
00145         
00146         data[8*0 + i]= lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
00147         data[8*4 + i]= lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));
00148         
00149         z1= (tmp12 + tmp13)* A1;
00150         data[8*2 + i]= lrintf(SCALE(8*2 + i) * (tmp13 + z1));
00151         data[8*6 + i]= lrintf(SCALE(8*6 + i) * (tmp13 - z1));
00152         
00153         tmp10= tmp4 + tmp5;
00154         tmp11= tmp5 + tmp6;
00155         tmp12= tmp6 + tmp7;
00156 
00157         z5= (tmp10 - tmp12) * A5;
00158         z2= tmp10*A2 + z5;
00159         z4= tmp12*A4 + z5;
00160         z3= tmp11*A1;
00161 
00162         z11= tmp7 + z3;
00163         z13= tmp7 - z3;
00164 
00165         data[8*5 + i]= lrintf(SCALE(8*5 + i) * (z13 + z2));
00166         data[8*3 + i]= lrintf(SCALE(8*3 + i) * (z13 - z2));
00167         data[8*1 + i]= lrintf(SCALE(8*1 + i) * (z11 + z4));
00168         data[8*7 + i]= lrintf(SCALE(8*7 + i) * (z11 - z4));
00169     }
00170 }
00171 
00172 void ff_faandct248(DCTELEM * data)
00173 {
00174     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
00175     FLOAT tmp10, tmp11, tmp12, tmp13;
00176     FLOAT z1;
00177     FLOAT temp[64];
00178     int i;
00179 
00180     emms_c();
00181 
00182     row_fdct(temp, data);
00183 
00184     for (i=0; i<8; i++) {
00185         tmp0 = temp[8*0 + i] + temp[8*1 + i];
00186         tmp1 = temp[8*2 + i] + temp[8*3 + i];
00187         tmp2 = temp[8*4 + i] + temp[8*5 + i];
00188         tmp3 = temp[8*6 + i] + temp[8*7 + i];
00189         tmp4 = temp[8*0 + i] - temp[8*1 + i];
00190         tmp5 = temp[8*2 + i] - temp[8*3 + i];
00191         tmp6 = temp[8*4 + i] - temp[8*5 + i];
00192         tmp7 = temp[8*6 + i] - temp[8*7 + i];
00193         
00194         tmp10 = tmp0 + tmp3;
00195         tmp11 = tmp1 + tmp2;
00196         tmp12 = tmp1 - tmp2;
00197         tmp13 = tmp0 - tmp3;
00198         
00199         data[8*0 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
00200         data[8*4 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));
00201         
00202         z1 = (tmp12 + tmp13)* A1;
00203         data[8*2 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + z1));
00204         data[8*6 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - z1));
00205         
00206         tmp10 = tmp4 + tmp7;
00207         tmp11 = tmp5 + tmp6;
00208         tmp12 = tmp5 - tmp6;
00209         tmp13 = tmp4 - tmp7;
00210 
00211         data[8*1 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
00212         data[8*5 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));
00213 
00214         z1 = (tmp12 + tmp13)* A1;
00215         data[8*3 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + z1));
00216         data[8*7 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - z1));
00217     }
00218 }