FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
faandct.c
Go to the documentation of this file.
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 "faandct.h"
29 #include "libavutil/internal.h"
30 #include "libavutil/libm.h"
31 
32 #define FLOAT float
33 
34 //numbers generated by simple c code (not as accurate as they could be)
35 /*
36 for(i=0; i<8; i++){
37  printf("#define B%d %1.20llf\n", i, (long double)1.0/(cosl(i*acosl(-1.0)/(long double)16.0)*sqrtl(2)));
38 }
39 */
40 #define B0 1.00000000000000000000
41 #define B1 0.72095982200694791383 // (cos(pi*1/16)sqrt(2))^-1
42 #define B2 0.76536686473017954350 // (cos(pi*2/16)sqrt(2))^-1
43 #define B3 0.85043009476725644878 // (cos(pi*3/16)sqrt(2))^-1
44 #define B4 1.00000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
45 #define B5 1.27275858057283393842 // (cos(pi*5/16)sqrt(2))^-1
46 #define B6 1.84775906502257351242 // (cos(pi*6/16)sqrt(2))^-1
47 #define B7 3.62450978541155137218 // (cos(pi*7/16)sqrt(2))^-1
48 
49 
50 #define A1 0.70710678118654752438 // cos(pi*4/16)
51 #define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
52 #define A5 0.38268343236508977170 // cos(pi*6/16)
53 #define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)
54 
55 static const FLOAT postscale[64]={
56 B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
57 B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
58 B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
59 B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
60 B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
61 B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
62 B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
63 B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
64 };
65 
66 static av_always_inline void row_fdct(FLOAT temp[64], int16_t *data)
67 {
68  FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
69  FLOAT tmp10, tmp11, tmp12, tmp13;
70  FLOAT z2, z4, z11, z13;
71  FLOAT av_unused z5;
72  int i;
73 
74  for (i=0; i<8*8; i+=8) {
75  tmp0= data[0 + i] + data[7 + i];
76  tmp7= data[0 + i] - data[7 + i];
77  tmp1= data[1 + i] + data[6 + i];
78  tmp6= data[1 + i] - data[6 + i];
79  tmp2= data[2 + i] + data[5 + i];
80  tmp5= data[2 + i] - data[5 + i];
81  tmp3= data[3 + i] + data[4 + i];
82  tmp4= data[3 + i] - data[4 + i];
83 
84  tmp10= tmp0 + tmp3;
85  tmp13= tmp0 - tmp3;
86  tmp11= tmp1 + tmp2;
87  tmp12= tmp1 - tmp2;
88 
89  temp[0 + i]= tmp10 + tmp11;
90  temp[4 + i]= tmp10 - tmp11;
91 
92  tmp12 += tmp13;
93  tmp12 *= A1;
94  temp[2 + i]= tmp13 + tmp12;
95  temp[6 + i]= tmp13 - tmp12;
96 
97  tmp4 += tmp5;
98  tmp5 += tmp6;
99  tmp6 += tmp7;
100 
101 #if 0
102  z5= (tmp4 - tmp6) * A5;
103  z2= tmp4*A2 + z5;
104  z4= tmp6*A4 + z5;
105 #else
106  z2= tmp4*(A2+A5) - tmp6*A5;
107  z4= tmp6*(A4-A5) + tmp4*A5;
108 #endif
109  tmp5*=A1;
110 
111  z11= tmp7 + tmp5;
112  z13= tmp7 - tmp5;
113 
114  temp[5 + i]= z13 + z2;
115  temp[3 + i]= z13 - z2;
116  temp[1 + i]= z11 + z4;
117  temp[7 + i]= z11 - z4;
118  }
119 }
120 
121 void ff_faandct(int16_t *data)
122 {
123  FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
124  FLOAT tmp10, tmp11, tmp12, tmp13;
125  FLOAT z2, z4, z11, z13;
126  FLOAT av_unused z5;
127  FLOAT temp[64];
128  int i;
129 
130  emms_c();
131 
132  row_fdct(temp, data);
133 
134  for (i=0; i<8; i++) {
135  tmp0= temp[8*0 + i] + temp[8*7 + i];
136  tmp7= temp[8*0 + i] - temp[8*7 + i];
137  tmp1= temp[8*1 + i] + temp[8*6 + i];
138  tmp6= temp[8*1 + i] - temp[8*6 + i];
139  tmp2= temp[8*2 + i] + temp[8*5 + i];
140  tmp5= temp[8*2 + i] - temp[8*5 + i];
141  tmp3= temp[8*3 + i] + temp[8*4 + i];
142  tmp4= temp[8*3 + i] - temp[8*4 + i];
143 
144  tmp10= tmp0 + tmp3;
145  tmp13= tmp0 - tmp3;
146  tmp11= tmp1 + tmp2;
147  tmp12= tmp1 - tmp2;
148 
149  data[8*0 + i]= lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
150  data[8*4 + i]= lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
151 
152  tmp12 += tmp13;
153  tmp12 *= A1;
154  data[8*2 + i]= lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
155  data[8*6 + i]= lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
156 
157  tmp4 += tmp5;
158  tmp5 += tmp6;
159  tmp6 += tmp7;
160 
161 #if 0
162  z5= (tmp4 - tmp6) * A5;
163  z2= tmp4*A2 + z5;
164  z4= tmp6*A4 + z5;
165 #else
166  z2= tmp4*(A2+A5) - tmp6*A5;
167  z4= tmp6*(A4-A5) + tmp4*A5;
168 #endif
169  tmp5*=A1;
170 
171  z11= tmp7 + tmp5;
172  z13= tmp7 - tmp5;
173 
174  data[8*5 + i]= lrintf(postscale[8*5 + i] * (z13 + z2));
175  data[8*3 + i]= lrintf(postscale[8*3 + i] * (z13 - z2));
176  data[8*1 + i]= lrintf(postscale[8*1 + i] * (z11 + z4));
177  data[8*7 + i]= lrintf(postscale[8*7 + i] * (z11 - z4));
178  }
179 }
180 
181 void ff_faandct248(int16_t *data)
182 {
183  FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
184  FLOAT tmp10, tmp11, tmp12, tmp13;
185  FLOAT temp[64];
186  int i;
187 
188  emms_c();
189 
190  row_fdct(temp, data);
191 
192  for (i=0; i<8; i++) {
193  tmp0 = temp[8*0 + i] + temp[8*1 + i];
194  tmp1 = temp[8*2 + i] + temp[8*3 + i];
195  tmp2 = temp[8*4 + i] + temp[8*5 + i];
196  tmp3 = temp[8*6 + i] + temp[8*7 + i];
197  tmp4 = temp[8*0 + i] - temp[8*1 + i];
198  tmp5 = temp[8*2 + i] - temp[8*3 + i];
199  tmp6 = temp[8*4 + i] - temp[8*5 + i];
200  tmp7 = temp[8*6 + i] - temp[8*7 + i];
201 
202  tmp10 = tmp0 + tmp3;
203  tmp11 = tmp1 + tmp2;
204  tmp12 = tmp1 - tmp2;
205  tmp13 = tmp0 - tmp3;
206 
207  data[8*0 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
208  data[8*4 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
209 
210  tmp12 += tmp13;
211  tmp12 *= A1;
212  data[8*2 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
213  data[8*6 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
214 
215  tmp10 = tmp4 + tmp7;
216  tmp11 = tmp5 + tmp6;
217  tmp12 = tmp5 - tmp6;
218  tmp13 = tmp4 - tmp7;
219 
220  data[8*1 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
221  data[8*5 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
222 
223  tmp12 += tmp13;
224  tmp12 *= A1;
225  data[8*3 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
226  data[8*7 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
227  }
228 }