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 typedef float FLOAT;
33 
34 /* numbers generated by arbitrary precision arithmetic followed by truncation
35 to 36 fractional digits (enough for a 128-bit IEEE quad, see /usr/include/math.h
36 for this approach). Unfortunately, long double is not always available correctly,
37 e.g ppc has issues.
38 TODO: 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 
54 static const FLOAT postscale[64]={
55 B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
56 B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
57 B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
58 B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
59 B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
60 B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
61 B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
62 B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
63 };
64 
65 static 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 #if 0
100  {
101  FLOAT z5;
102  z5 = (tmp4 - tmp6) * A5;
103  z2 = tmp4 * A2 + z5;
104  z4 = tmp6 * A4 + z5;
105  }
106 #else
107  z2= tmp4*(A2+A5) - tmp6*A5;
108  z4= tmp6*(A4-A5) + tmp4*A5;
109 #endif
110  tmp5*=A1;
111 
112  z11= tmp7 + tmp5;
113  z13= tmp7 - tmp5;
114 
115  temp[5 + i]= z13 + z2;
116  temp[3 + i]= z13 - z2;
117  temp[1 + i]= z11 + z4;
118  temp[7 + i]= z11 - z4;
119  }
120 }
121 
122 void ff_faandct(int16_t *data)
123 {
124  FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
125  FLOAT tmp10, tmp11, tmp12, tmp13;
126  FLOAT z2, z4, z11, z13;
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  {
163  FLOAT z5;
164  z5 = (tmp4 - tmp6) * A5;
165  z2 = tmp4 * A2 + z5;
166  z4 = tmp6 * A4 + z5;
167  }
168 #else
169  z2= tmp4*(A2+A5) - tmp6*A5;
170  z4= tmp6*(A4-A5) + tmp4*A5;
171 #endif
172  tmp5*=A1;
173 
174  z11= tmp7 + tmp5;
175  z13= tmp7 - tmp5;
176 
177  data[8*5 + i]= lrintf(postscale[8*5 + i] * (z13 + z2));
178  data[8*3 + i]= lrintf(postscale[8*3 + i] * (z13 - z2));
179  data[8*1 + i]= lrintf(postscale[8*1 + i] * (z11 + z4));
180  data[8*7 + i]= lrintf(postscale[8*7 + i] * (z11 - z4));
181  }
182 }
183 
184 void ff_faandct248(int16_t *data)
185 {
186  FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
187  FLOAT tmp10, tmp11, tmp12, tmp13;
188  FLOAT temp[64];
189  int i;
190 
191  emms_c();
192 
193  row_fdct(temp, data);
194 
195  for (i=0; i<8; i++) {
196  tmp0 = temp[8*0 + i] + temp[8*1 + i];
197  tmp1 = temp[8*2 + i] + temp[8*3 + i];
198  tmp2 = temp[8*4 + i] + temp[8*5 + i];
199  tmp3 = temp[8*6 + i] + temp[8*7 + i];
200  tmp4 = temp[8*0 + i] - temp[8*1 + i];
201  tmp5 = temp[8*2 + i] - temp[8*3 + i];
202  tmp6 = temp[8*4 + i] - temp[8*5 + i];
203  tmp7 = temp[8*6 + i] - temp[8*7 + i];
204 
205  tmp10 = tmp0 + tmp3;
206  tmp11 = tmp1 + tmp2;
207  tmp12 = tmp1 - tmp2;
208  tmp13 = tmp0 - tmp3;
209 
210  data[8*0 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
211  data[8*4 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
212 
213  tmp12 += tmp13;
214  tmp12 *= A1;
215  data[8*2 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
216  data[8*6 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
217 
218  tmp10 = tmp4 + tmp7;
219  tmp11 = tmp5 + tmp6;
220  tmp12 = tmp5 - tmp6;
221  tmp13 = tmp4 - tmp7;
222 
223  data[8*1 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
224  data[8*5 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));
225 
226  tmp12 += tmp13;
227  tmp12 *= A1;
228  data[8*3 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
229  data[8*7 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
230  }
231 }
ptrdiff_t const GLvoid * data
Definition: opengl_enc.c:101
else temp
Definition: vf_mcdeint.c:259
void ff_faandct248(int16_t *data)
Definition: faandct.c:184
#define B7
Definition: faandct.c:47
#define B1
Definition: faandct.c:41
#define B4
Definition: faandct.c:44
#define lrintf(x)
Definition: libm_mips.h:70
#define B2
Definition: faandct.c:42
#define A4
Definition: faandct.c:52
static const FLOAT postscale[64]
Definition: faandct.c:54
#define B5
Definition: faandct.c:45
common internal API header
uint32_t i
Definition: intfloat.h:28
float FLOAT
Definition: faandct.c:32
void ff_faandct(int16_t *data)
Definition: faandct.c:122
#define B6
Definition: faandct.c:46
Replacements for frequently missing libm functions.
static av_always_inline void row_fdct(FLOAT temp[64], int16_t *data)
Definition: faandct.c:65
#define A2
Definition: faandct.c:50
#define A5
Definition: faandct.c:51
#define B3
Definition: faandct.c:43
#define av_always_inline
Definition: attributes.h:39
Floating point AAN DCT
#define A1
Definition: faandct.c:49
#define B0
Definition: faandct.c:40