Imported Debian version 2.5.3~trusty1
[deb_ffmpeg.git] / ffmpeg / libavcodec / faandct.c
CommitLineData
2ba45a60
DM
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/*
36for(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
55static const FLOAT postscale[64]={
56B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
57B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
58B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
59B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
60B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
61B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
62B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
63B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
64};
65
66static 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 int i;
72
73 for (i=0; i<8*8; i+=8) {
74 tmp0= data[0 + i] + data[7 + i];
75 tmp7= data[0 + i] - data[7 + i];
76 tmp1= data[1 + i] + data[6 + i];
77 tmp6= data[1 + i] - data[6 + i];
78 tmp2= data[2 + i] + data[5 + i];
79 tmp5= data[2 + i] - data[5 + i];
80 tmp3= data[3 + i] + data[4 + i];
81 tmp4= data[3 + i] - data[4 + i];
82
83 tmp10= tmp0 + tmp3;
84 tmp13= tmp0 - tmp3;
85 tmp11= tmp1 + tmp2;
86 tmp12= tmp1 - tmp2;
87
88 temp[0 + i]= tmp10 + tmp11;
89 temp[4 + i]= tmp10 - tmp11;
90
91 tmp12 += tmp13;
92 tmp12 *= A1;
93 temp[2 + i]= tmp13 + tmp12;
94 temp[6 + i]= tmp13 - tmp12;
95
96 tmp4 += tmp5;
97 tmp5 += tmp6;
98 tmp6 += tmp7;
99
100#if 0
101 {
102 FLOAT z5;
103 z5 = (tmp4 - tmp6) * A5;
104 z2 = tmp4 * A2 + z5;
105 z4 = tmp6 * A4 + z5;
106 }
107#else
108 z2= tmp4*(A2+A5) - tmp6*A5;
109 z4= tmp6*(A4-A5) + tmp4*A5;
110#endif
111 tmp5*=A1;
112
113 z11= tmp7 + tmp5;
114 z13= tmp7 - tmp5;
115
116 temp[5 + i]= z13 + z2;
117 temp[3 + i]= z13 - z2;
118 temp[1 + i]= z11 + z4;
119 temp[7 + i]= z11 - z4;
120 }
121}
122
123void ff_faandct(int16_t *data)
124{
125 FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
126 FLOAT tmp10, tmp11, tmp12, tmp13;
127 FLOAT z2, z4, z11, z13;
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 {
164 FLOAT z5;
165 z5 = (tmp4 - tmp6) * A5;
166 z2 = tmp4 * A2 + z5;
167 z4 = tmp6 * A4 + z5;
168 }
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(postscale[8*5 + i] * (z13 + z2));
179 data[8*3 + i]= lrintf(postscale[8*3 + i] * (z13 - z2));
180 data[8*1 + i]= lrintf(postscale[8*1 + i] * (z11 + z4));
181 data[8*7 + i]= lrintf(postscale[8*7 + i] * (z11 - z4));
182 }
183}
184
185void ff_faandct248(int16_t *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(postscale[8*0 + i] * (tmp10 + tmp11));
212 data[8*4 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
213
214 tmp12 += tmp13;
215 tmp12 *= A1;
216 data[8*2 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
217 data[8*6 + i] = lrintf(postscale[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(postscale[8*0 + i] * (tmp10 + tmp11));
225 data[8*5 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
226
227 tmp12 += tmp13;
228 tmp12 *= A1;
229 data[8*3 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
230 data[8*7 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
231 }
232}