annotate celp_math.h @ 8442:60be1c7d8345 libavcodec

Make h264 parser 50% faster.
author michael
date Tue, 23 Dec 2008 02:34:58 +0000
parents cf4d575b1982
children c8743c33eeef
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
1 /*
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
2 * Various fixed-point math operations
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
3 *
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
4 * Copyright (c) 2008 Vladimir Voroshilov
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
5 *
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
6 * This file is part of FFmpeg.
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
7 *
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
8 * FFmpeg is free software; you can redistribute it and/or
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
9 * modify it under the terms of the GNU Lesser General Public
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
10 * License as published by the Free Software Foundation; either
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
11 * version 2.1 of the License, or (at your option) any later version.
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
12 *
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
13 * FFmpeg is distributed in the hope that it will be useful,
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
16 * Lesser General Public License for more details.
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
17 *
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
18 * You should have received a copy of the GNU Lesser General Public
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
19 * License along with FFmpeg; if not, write to the Free Software
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
21 */
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
22
8048
ecb1962c12f3 Rename acelp_math.[ch] to celp_math.[ch] to prepare for QCELP decoder merge.
diego
parents: 7760
diff changeset
23 #ifndef AVCODEC_CELP_MATH_H
ecb1962c12f3 Rename acelp_math.[ch] to celp_math.[ch] to prepare for QCELP decoder merge.
diego
parents: 7760
diff changeset
24 #define AVCODEC_CELP_MATH_H
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
25
6735
459fe563a4dc add necessary #includes in headers
mru
parents: 6706
diff changeset
26 #include <stdint.h>
459fe563a4dc add necessary #includes in headers
mru
parents: 6706
diff changeset
27
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
28 /**
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
29 * fixed-point implementation of cosine in [0; PI) domain.
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
30 * @param arg fixed-point cosine argument, 0 <= arg < 0x4000
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
31 *
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
32 * @return value of (1<<15) * cos(arg * PI / (1<<14)), -0x8000 <= result <= 0x7fff
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
33 */
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
34 int16_t ff_cos(uint16_t arg);
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
35
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
36 /**
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
37 * fixed-point implementation of exp2(x) in [0; 1] domain.
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
38 * @param power argument to exp2, 0 <= power <= 0x7fff
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
39 *
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
40 * @return value of (1<<20) * exp2(power / (1<<15))
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
41 * 0x8000c <= result <= 0xfffea
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
42 */
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
43 int ff_exp2(uint16_t power);
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
44
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
45 /**
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
46 * Calculates log2(x).
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
47 * @param value function argument, 0 < value <= 7fff ffff
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
48 *
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
49 * @return value of (1<<15) * log2(value)
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
50 */
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
51 int ff_log2(uint32_t value);
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
52
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
53 /**
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
54 * returns the dot product.
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
55 * @param a input data array
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
56 * @param b input data array
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
57 * @param length number of elements
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
58 * @param shift right shift by this value will be done after multiplication
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
59 *
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
60 * @return dot product = sum of elementwise products
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
61 */
7186
ff5ef7becd94 make sum_of_squares() more generic
michael
parents: 6777
diff changeset
62 static int dot_product(const int16_t* a, const int16_t* b, int length, int shift)
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
63 {
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
64 int sum = 0;
7186
ff5ef7becd94 make sum_of_squares() more generic
michael
parents: 6777
diff changeset
65 int i;
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
66
7186
ff5ef7becd94 make sum_of_squares() more generic
michael
parents: 6777
diff changeset
67 for(i=0; i<length; i++)
ff5ef7becd94 make sum_of_squares() more generic
michael
parents: 6777
diff changeset
68 sum += (a[i] * b[i]) >> shift;
6682
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
69
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
70 return sum;
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
71 }
1064be7fea22 Fixed-point implementation of ff_cos, ff_log2, ff_exp2.
voroshil
parents:
diff changeset
72
6777
6f402a181803 Implement bidirectional (positive offset - left, negative - right)
voroshil
parents: 6739
diff changeset
73 /**
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
74 * Shift value left or right depending on sign of offset parameter.
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
75 * @param value value to shift
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
76 * @param offset shift offset
6777
6f402a181803 Implement bidirectional (positive offset - left, negative - right)
voroshil
parents: 6739
diff changeset
77 *
7655
f91a738da9c3 Doxygen syntax consistency.
michael
parents: 7188
diff changeset
78 * @return value << offset, if offset>=0; value >> -offset - otherwise
6777
6f402a181803 Implement bidirectional (positive offset - left, negative - right)
voroshil
parents: 6739
diff changeset
79 */
6f402a181803 Implement bidirectional (positive offset - left, negative - right)
voroshil
parents: 6739
diff changeset
80 static inline int bidir_sal(int value, int offset)
6f402a181803 Implement bidirectional (positive offset - left, negative - right)
voroshil
parents: 6739
diff changeset
81 {
6f402a181803 Implement bidirectional (positive offset - left, negative - right)
voroshil
parents: 6739
diff changeset
82 if(offset < 0) return value >> -offset;
6f402a181803 Implement bidirectional (positive offset - left, negative - right)
voroshil
parents: 6739
diff changeset
83 else return value << offset;
6f402a181803 Implement bidirectional (positive offset - left, negative - right)
voroshil
parents: 6739
diff changeset
84 }
6f402a181803 Implement bidirectional (positive offset - left, negative - right)
voroshil
parents: 6739
diff changeset
85
8090
1c07635d7334 Add ff_dot_productf() to celp_math.{c,h}
vitor
parents: 8048
diff changeset
86 /**
1c07635d7334 Add ff_dot_productf() to celp_math.{c,h}
vitor
parents: 8048
diff changeset
87 * returns the dot product.
1c07635d7334 Add ff_dot_productf() to celp_math.{c,h}
vitor
parents: 8048
diff changeset
88 * @param a input data array
1c07635d7334 Add ff_dot_productf() to celp_math.{c,h}
vitor
parents: 8048
diff changeset
89 * @param b input data array
1c07635d7334 Add ff_dot_productf() to celp_math.{c,h}
vitor
parents: 8048
diff changeset
90 * @param length number of elements
1c07635d7334 Add ff_dot_productf() to celp_math.{c,h}
vitor
parents: 8048
diff changeset
91 *
1c07635d7334 Add ff_dot_productf() to celp_math.{c,h}
vitor
parents: 8048
diff changeset
92 * @return dot product = sum of elementwise products
1c07635d7334 Add ff_dot_productf() to celp_math.{c,h}
vitor
parents: 8048
diff changeset
93 */
8250
cf4d575b1982 Delete unnecessary 'extern' keywords.
diego
parents: 8090
diff changeset
94 float ff_dot_productf(const float* a, const float* b, int length);
8090
1c07635d7334 Add ff_dot_productf() to celp_math.{c,h}
vitor
parents: 8048
diff changeset
95
8048
ecb1962c12f3 Rename acelp_math.[ch] to celp_math.[ch] to prepare for QCELP decoder merge.
diego
parents: 7760
diff changeset
96 #endif /* AVCODEC_CELP_MATH_H */