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 libavcodec/faandct.c
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
31#define FLOAT float
32#ifdef FAAN_POSTSCALE
33#    define SCALE(x) postscale[x]
34#else
35#    define SCALE(x) 1
36#endif
37
38//numbers generated by simple c code (not as accurate as they could be)
39/*
40for(i=0; i<8; i++){
41    printf("#define B%d %1.20llf\n", i, (long double)1.0/(cosl(i*acosl(-1.0)/(long double)16.0)*sqrtl(2)));
42}
43*/
44#define B0 1.00000000000000000000
45#define B1 0.72095982200694791383 // (cos(pi*1/16)sqrt(2))^-1
46#define B2 0.76536686473017954350 // (cos(pi*2/16)sqrt(2))^-1
47#define B3 0.85043009476725644878 // (cos(pi*3/16)sqrt(2))^-1
48#define B4 1.00000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
49#define B5 1.27275858057283393842 // (cos(pi*5/16)sqrt(2))^-1
50#define B6 1.84775906502257351242 // (cos(pi*6/16)sqrt(2))^-1
51#define B7 3.62450978541155137218 // (cos(pi*7/16)sqrt(2))^-1
52
53
54#define A1 0.70710678118654752438 // cos(pi*4/16)
55#define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
56#define A5 0.38268343236508977170 // cos(pi*6/16)
57#define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)
58
59static const FLOAT postscale[64]={
60B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
61B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
62B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
63B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
64B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
65B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
66B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
67B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
68};
69
70static av_always_inline void row_fdct(FLOAT temp[64], DCTELEM * data)
71{
72    FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
73    FLOAT tmp10, tmp11, tmp12, tmp13;
74    FLOAT z2, z4, z11, z13;
75    FLOAT av_unused z5;
76    int i;
77
78    for (i=0; i<8*8; i+=8) {
79        tmp0= data[0 + i] + data[7 + i];
80        tmp7= data[0 + i] - data[7 + i];
81        tmp1= data[1 + i] + data[6 + i];
82        tmp6= data[1 + i] - data[6 + i];
83        tmp2= data[2 + i] + data[5 + i];
84        tmp5= data[2 + i] - data[5 + i];
85        tmp3= data[3 + i] + data[4 + i];
86        tmp4= data[3 + i] - data[4 + i];
87
88        tmp10= tmp0 + tmp3;
89        tmp13= tmp0 - tmp3;
90        tmp11= tmp1 + tmp2;
91        tmp12= tmp1 - tmp2;
92
93        temp[0 + i]= tmp10 + tmp11;
94        temp[4 + i]= tmp10 - tmp11;
95
96        tmp12 += tmp13;
97        tmp12 *= A1;
98        temp[2 + i]= tmp13 + tmp12;
99        temp[6 + i]= tmp13 - tmp12;
100
101        tmp4 += tmp5;
102        tmp5 += tmp6;
103        tmp6 += tmp7;
104
105#if 0
106        z5= (tmp4 - tmp6) * A5;
107        z2= tmp4*A2 + z5;
108        z4= tmp6*A4 + z5;
109#else
110        z2= tmp4*(A2+A5) - tmp6*A5;
111        z4= tmp6*(A4-A5) + tmp4*A5;
112#endif
113        tmp5*=A1;
114
115        z11= tmp7 + tmp5;
116        z13= tmp7 - tmp5;
117
118        temp[5 + i]= z13 + z2;
119        temp[3 + i]= z13 - z2;
120        temp[1 + i]= z11 + z4;
121        temp[7 + i]= z11 - z4;
122    }
123}
124
125void ff_faandct(DCTELEM * data)
126{
127    FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
128    FLOAT tmp10, tmp11, tmp12, tmp13;
129    FLOAT z2, z4, z11, z13;
130    FLOAT av_unused z5;
131    FLOAT temp[64];
132    int i;
133
134    emms_c();
135
136    row_fdct(temp, data);
137
138    for (i=0; i<8; i++) {
139        tmp0= temp[8*0 + i] + temp[8*7 + i];
140        tmp7= temp[8*0 + i] - temp[8*7 + i];
141        tmp1= temp[8*1 + i] + temp[8*6 + i];
142        tmp6= temp[8*1 + i] - temp[8*6 + i];
143        tmp2= temp[8*2 + i] + temp[8*5 + i];
144        tmp5= temp[8*2 + i] - temp[8*5 + i];
145        tmp3= temp[8*3 + i] + temp[8*4 + i];
146        tmp4= temp[8*3 + i] - temp[8*4 + i];
147
148        tmp10= tmp0 + tmp3;
149        tmp13= tmp0 - tmp3;
150        tmp11= tmp1 + tmp2;
151        tmp12= tmp1 - tmp2;
152
153        data[8*0 + i]= lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
154        data[8*4 + i]= lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));
155
156        tmp12 += tmp13;
157        tmp12 *= A1;
158        data[8*2 + i]= lrintf(SCALE(8*2 + i) * (tmp13 + tmp12));
159        data[8*6 + i]= lrintf(SCALE(8*6 + i) * (tmp13 - tmp12));
160
161        tmp4 += tmp5;
162        tmp5 += tmp6;
163        tmp6 += tmp7;
164
165#if 0
166        z5= (tmp4 - tmp6) * A5;
167        z2= tmp4*A2 + z5;
168        z4= tmp6*A4 + z5;
169#else
170        z2= tmp4*(A2+A5) - tmp6*A5;
171        z4= tmp6*(A4-A5) + tmp4*A5;
172#endif
173        tmp5*=A1;
174
175        z11= tmp7 + tmp5;
176        z13= tmp7 - tmp5;
177
178        data[8*5 + i]= lrintf(SCALE(8*5 + i) * (z13 + z2));
179        data[8*3 + i]= lrintf(SCALE(8*3 + i) * (z13 - z2));
180        data[8*1 + i]= lrintf(SCALE(8*1 + i) * (z11 + z4));
181        data[8*7 + i]= lrintf(SCALE(8*7 + i) * (z11 - z4));
182    }
183}
184
185void ff_faandct248(DCTELEM * data)
186{
187    FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
188    FLOAT tmp10, tmp11, tmp12, tmp13;
189    FLOAT temp[64];
190    int i;
191
192    emms_c();
193
194    row_fdct(temp, data);
195
196    for (i=0; i<8; i++) {
197        tmp0 = temp[8*0 + i] + temp[8*1 + i];
198        tmp1 = temp[8*2 + i] + temp[8*3 + i];
199        tmp2 = temp[8*4 + i] + temp[8*5 + i];
200        tmp3 = temp[8*6 + i] + temp[8*7 + i];
201        tmp4 = temp[8*0 + i] - temp[8*1 + i];
202        tmp5 = temp[8*2 + i] - temp[8*3 + i];
203        tmp6 = temp[8*4 + i] - temp[8*5 + i];
204        tmp7 = temp[8*6 + i] - temp[8*7 + i];
205
206        tmp10 = tmp0 + tmp3;
207        tmp11 = tmp1 + tmp2;
208        tmp12 = tmp1 - tmp2;
209        tmp13 = tmp0 - tmp3;
210
211        data[8*0 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
212        data[8*4 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));
213
214        tmp12 += tmp13;
215        tmp12 *= A1;
216        data[8*2 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + tmp12));
217        data[8*6 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - tmp12));
218
219        tmp10 = tmp4 + tmp7;
220        tmp11 = tmp5 + tmp6;
221        tmp12 = tmp5 - tmp6;
222        tmp13 = tmp4 - tmp7;
223
224        data[8*1 + i] = lrintf(SCALE(8*0 + i) * (tmp10 + tmp11));
225        data[8*5 + i] = lrintf(SCALE(8*4 + i) * (tmp10 - tmp11));
226
227        tmp12 += tmp13;
228        tmp12 *= A1;
229        data[8*3 + i] = lrintf(SCALE(8*2 + i) * (tmp13 + tmp12));
230        data[8*7 + i] = lrintf(SCALE(8*6 + i) * (tmp13 - tmp12));
231    }
232}
233