xref: /third_party/ffmpeg/libavcodec/faandct.c (revision cabdff1a) 
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
32typedef float FLOAT;
33
34/* numbers generated by arbitrary precision arithmetic followed by truncation
35to 36 fractional digits (enough for a 128-bit IEEE quad, see /usr/include/math.h
36for this approach). Unfortunately, long double is not always available correctly,
37e.g ppc has issues.
38TODO: add L suffixes when ppc and toolchains sort out their stuff.
39*/
40#define B0 1.000000000000000000000000000000000000
41#define B1 0.720959822006947913789091890943021267 // (cos(pi*1/16)sqrt(2))^-1
42#define B2 0.765366864730179543456919968060797734 // (cos(pi*2/16)sqrt(2))^-1
43#define B3 0.850430094767256448766702844371412325 // (cos(pi*3/16)sqrt(2))^-1
44#define B4 1.000000000000000000000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
45#define B5 1.272758580572833938461007018281767032 // (cos(pi*5/16)sqrt(2))^-1
46#define B6 1.847759065022573512256366378793576574 // (cos(pi*6/16)sqrt(2))^-1
47#define B7 3.624509785411551372409941227504289587 // (cos(pi*7/16)sqrt(2))^-1
48
49#define A1 M_SQRT1_2              // cos(pi*4/16)
50#define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
51#define A5 0.38268343236508977170 // cos(pi*6/16)
52#define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)
53
54static const FLOAT postscale[64]={
55B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
56B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
57B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
58B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
59B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
60B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
61B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
62B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
63};
64
65static av_always_inline void row_fdct(FLOAT temp[64], int16_t *data)
66{
67    FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
68    FLOAT tmp10, tmp11, tmp12, tmp13;
69    FLOAT z2, z4, z11, z13;
70    int i;
71
72    for (i=0; i<8*8; i+=8) {
73        tmp0= data[0 + i] + data[7 + i];
74        tmp7= data[0 + i] - data[7 + i];
75        tmp1= data[1 + i] + data[6 + i];
76        tmp6= data[1 + i] - data[6 + i];
77        tmp2= data[2 + i] + data[5 + i];
78        tmp5= data[2 + i] - data[5 + i];
79        tmp3= data[3 + i] + data[4 + i];
80        tmp4= data[3 + i] - data[4 + i];
81
82        tmp10= tmp0 + tmp3;
83        tmp13= tmp0 - tmp3;
84        tmp11= tmp1 + tmp2;
85        tmp12= tmp1 - tmp2;
86
87        temp[0 + i]= tmp10 + tmp11;
88        temp[4 + i]= tmp10 - tmp11;
89
90        tmp12 += tmp13;
91        tmp12 *= A1;
92        temp[2 + i]= tmp13 + tmp12;
93        temp[6 + i]= tmp13 - tmp12;
94
95        tmp4 += tmp5;
96        tmp5 += tmp6;
97        tmp6 += tmp7;
98
99        z2= tmp4*(A2+A5) - tmp6*A5;
100        z4= tmp6*(A4-A5) + tmp4*A5;
101
102        tmp5*=A1;
103
104        z11= tmp7 + tmp5;
105        z13= tmp7 - tmp5;
106
107        temp[5 + i]= z13 + z2;
108        temp[3 + i]= z13 - z2;
109        temp[1 + i]= z11 + z4;
110        temp[7 + i]= z11 - z4;
111    }
112}
113
114void ff_faandct(int16_t *data)
115{
116    FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
117    FLOAT tmp10, tmp11, tmp12, tmp13;
118    FLOAT z2, z4, z11, z13;
119    FLOAT temp[64];
120    int i;
121
122    emms_c();
123
124    row_fdct(temp, data);
125
126    for (i=0; i<8; i++) {
127        tmp0= temp[8*0 + i] + temp[8*7 + i];
128        tmp7= temp[8*0 + i] - temp[8*7 + i];
129        tmp1= temp[8*1 + i] + temp[8*6 + i];
130        tmp6= temp[8*1 + i] - temp[8*6 + i];
131        tmp2= temp[8*2 + i] + temp[8*5 + i];
132        tmp5= temp[8*2 + i] - temp[8*5 + i];
133        tmp3= temp[8*3 + i] + temp[8*4 + i];
134        tmp4= temp[8*3 + i] - temp[8*4 + i];
135
136        tmp10= tmp0 + tmp3;
137        tmp13= tmp0 - tmp3;
138        tmp11= tmp1 + tmp2;
139        tmp12= tmp1 - tmp2;
140
141        data[8*0 + i]= lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
142        data[8*4 + i]= lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
143
144        tmp12 += tmp13;
145        tmp12 *= A1;
146        data[8*2 + i]= lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
147        data[8*6 + i]= lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
148
149        tmp4 += tmp5;
150        tmp5 += tmp6;
151        tmp6 += tmp7;
152
153        z2= tmp4*(A2+A5) - tmp6*A5;
154        z4= tmp6*(A4-A5) + tmp4*A5;
155
156        tmp5*=A1;
157
158        z11= tmp7 + tmp5;
159        z13= tmp7 - tmp5;
160
161        data[8*5 + i]= lrintf(postscale[8*5 + i] * (z13 + z2));
162        data[8*3 + i]= lrintf(postscale[8*3 + i] * (z13 - z2));
163        data[8*1 + i]= lrintf(postscale[8*1 + i] * (z11 + z4));
164        data[8*7 + i]= lrintf(postscale[8*7 + i] * (z11 - z4));
165    }
166}
167
168void ff_faandct248(int16_t *data)
169{
170    FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
171    FLOAT tmp10, tmp11, tmp12, tmp13;
172    FLOAT temp[64];
173    int i;
174
175    emms_c();
176
177    row_fdct(temp, data);
178
179    for (i=0; i<8; i++) {
180        tmp0 = temp[8*0 + i] + temp[8*1 + i];
181        tmp1 = temp[8*2 + i] + temp[8*3 + i];
182        tmp2 = temp[8*4 + i] + temp[8*5 + i];
183        tmp3 = temp[8*6 + i] + temp[8*7 + i];
184        tmp4 = temp[8*0 + i] - temp[8*1 + i];
185        tmp5 = temp[8*2 + i] - temp[8*3 + i];
186        tmp6 = temp[8*4 + i] - temp[8*5 + i];
187        tmp7 = temp[8*6 + i] - temp[8*7 + i];
188
189        tmp10 = tmp0 + tmp3;
190        tmp11 = tmp1 + tmp2;
191        tmp12 = tmp1 - tmp2;
192        tmp13 = tmp0 - tmp3;
193
194        data[8*0 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
195        data[8*4 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
196
197        tmp12 += tmp13;
198        tmp12 *= A1;
199        data[8*2 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
200        data[8*6 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
201
202        tmp10 = tmp4 + tmp7;
203        tmp11 = tmp5 + tmp6;
204        tmp12 = tmp5 - tmp6;
205        tmp13 = tmp4 - tmp7;
206
207        data[8*1 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
208        data[8*5 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
209
210        tmp12 += tmp13;
211        tmp12 *= A1;
212        data[8*3 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
213        data[8*7 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
214    }
215}
216