Chris@69
|
1 /***********************************************************************
|
Chris@69
|
2 Copyright (c) 2006-2011, Skype Limited. All rights reserved.
|
Chris@69
|
3 Redistribution and use in source and binary forms, with or without
|
Chris@69
|
4 modification, are permitted provided that the following conditions
|
Chris@69
|
5 are met:
|
Chris@69
|
6 - Redistributions of source code must retain the above copyright notice,
|
Chris@69
|
7 this list of conditions and the following disclaimer.
|
Chris@69
|
8 - Redistributions in binary form must reproduce the above copyright
|
Chris@69
|
9 notice, this list of conditions and the following disclaimer in the
|
Chris@69
|
10 documentation and/or other materials provided with the distribution.
|
Chris@69
|
11 - Neither the name of Internet Society, IETF or IETF Trust, nor the
|
Chris@69
|
12 names of specific contributors, may be used to endorse or promote
|
Chris@69
|
13 products derived from this software without specific prior written
|
Chris@69
|
14 permission.
|
Chris@69
|
15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
|
Chris@69
|
16 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
Chris@69
|
17 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
Chris@69
|
18 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
|
Chris@69
|
19 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
|
Chris@69
|
20 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
|
Chris@69
|
21 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
|
Chris@69
|
22 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
|
Chris@69
|
23 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
|
Chris@69
|
24 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
Chris@69
|
25 POSSIBILITY OF SUCH DAMAGE.
|
Chris@69
|
26 ***********************************************************************/
|
Chris@69
|
27
|
Chris@69
|
28 /*! \file silk_Inlines.h
|
Chris@69
|
29 * \brief silk_Inlines.h defines OPUS_INLINE signal processing functions.
|
Chris@69
|
30 */
|
Chris@69
|
31
|
Chris@69
|
32 #ifndef SILK_FIX_INLINES_H
|
Chris@69
|
33 #define SILK_FIX_INLINES_H
|
Chris@69
|
34
|
Chris@69
|
35 #ifdef __cplusplus
|
Chris@69
|
36 extern "C"
|
Chris@69
|
37 {
|
Chris@69
|
38 #endif
|
Chris@69
|
39
|
Chris@69
|
40 /* count leading zeros of opus_int64 */
|
Chris@69
|
41 static OPUS_INLINE opus_int32 silk_CLZ64( opus_int64 in )
|
Chris@69
|
42 {
|
Chris@69
|
43 opus_int32 in_upper;
|
Chris@69
|
44
|
Chris@69
|
45 in_upper = (opus_int32)silk_RSHIFT64(in, 32);
|
Chris@69
|
46 if (in_upper == 0) {
|
Chris@69
|
47 /* Search in the lower 32 bits */
|
Chris@69
|
48 return 32 + silk_CLZ32( (opus_int32) in );
|
Chris@69
|
49 } else {
|
Chris@69
|
50 /* Search in the upper 32 bits */
|
Chris@69
|
51 return silk_CLZ32( in_upper );
|
Chris@69
|
52 }
|
Chris@69
|
53 }
|
Chris@69
|
54
|
Chris@69
|
55 /* get number of leading zeros and fractional part (the bits right after the leading one */
|
Chris@69
|
56 static OPUS_INLINE void silk_CLZ_FRAC(
|
Chris@69
|
57 opus_int32 in, /* I input */
|
Chris@69
|
58 opus_int32 *lz, /* O number of leading zeros */
|
Chris@69
|
59 opus_int32 *frac_Q7 /* O the 7 bits right after the leading one */
|
Chris@69
|
60 )
|
Chris@69
|
61 {
|
Chris@69
|
62 opus_int32 lzeros = silk_CLZ32(in);
|
Chris@69
|
63
|
Chris@69
|
64 * lz = lzeros;
|
Chris@69
|
65 * frac_Q7 = silk_ROR32(in, 24 - lzeros) & 0x7f;
|
Chris@69
|
66 }
|
Chris@69
|
67
|
Chris@69
|
68 /* Approximation of square root */
|
Chris@69
|
69 /* Accuracy: < +/- 10% for output values > 15 */
|
Chris@69
|
70 /* < +/- 2.5% for output values > 120 */
|
Chris@69
|
71 static OPUS_INLINE opus_int32 silk_SQRT_APPROX( opus_int32 x )
|
Chris@69
|
72 {
|
Chris@69
|
73 opus_int32 y, lz, frac_Q7;
|
Chris@69
|
74
|
Chris@69
|
75 if( x <= 0 ) {
|
Chris@69
|
76 return 0;
|
Chris@69
|
77 }
|
Chris@69
|
78
|
Chris@69
|
79 silk_CLZ_FRAC(x, &lz, &frac_Q7);
|
Chris@69
|
80
|
Chris@69
|
81 if( lz & 1 ) {
|
Chris@69
|
82 y = 32768;
|
Chris@69
|
83 } else {
|
Chris@69
|
84 y = 46214; /* 46214 = sqrt(2) * 32768 */
|
Chris@69
|
85 }
|
Chris@69
|
86
|
Chris@69
|
87 /* get scaling right */
|
Chris@69
|
88 y >>= silk_RSHIFT(lz, 1);
|
Chris@69
|
89
|
Chris@69
|
90 /* increment using fractional part of input */
|
Chris@69
|
91 y = silk_SMLAWB(y, y, silk_SMULBB(213, frac_Q7));
|
Chris@69
|
92
|
Chris@69
|
93 return y;
|
Chris@69
|
94 }
|
Chris@69
|
95
|
Chris@69
|
96 /* Divide two int32 values and return result as int32 in a given Q-domain */
|
Chris@69
|
97 static OPUS_INLINE opus_int32 silk_DIV32_varQ( /* O returns a good approximation of "(a32 << Qres) / b32" */
|
Chris@69
|
98 const opus_int32 a32, /* I numerator (Q0) */
|
Chris@69
|
99 const opus_int32 b32, /* I denominator (Q0) */
|
Chris@69
|
100 const opus_int Qres /* I Q-domain of result (>= 0) */
|
Chris@69
|
101 )
|
Chris@69
|
102 {
|
Chris@69
|
103 opus_int a_headrm, b_headrm, lshift;
|
Chris@69
|
104 opus_int32 b32_inv, a32_nrm, b32_nrm, result;
|
Chris@69
|
105
|
Chris@69
|
106 silk_assert( b32 != 0 );
|
Chris@69
|
107 silk_assert( Qres >= 0 );
|
Chris@69
|
108
|
Chris@69
|
109 /* Compute number of bits head room and normalize inputs */
|
Chris@69
|
110 a_headrm = silk_CLZ32( silk_abs(a32) ) - 1;
|
Chris@69
|
111 a32_nrm = silk_LSHIFT(a32, a_headrm); /* Q: a_headrm */
|
Chris@69
|
112 b_headrm = silk_CLZ32( silk_abs(b32) ) - 1;
|
Chris@69
|
113 b32_nrm = silk_LSHIFT(b32, b_headrm); /* Q: b_headrm */
|
Chris@69
|
114
|
Chris@69
|
115 /* Inverse of b32, with 14 bits of precision */
|
Chris@69
|
116 b32_inv = silk_DIV32_16( silk_int32_MAX >> 2, silk_RSHIFT(b32_nrm, 16) ); /* Q: 29 + 16 - b_headrm */
|
Chris@69
|
117
|
Chris@69
|
118 /* First approximation */
|
Chris@69
|
119 result = silk_SMULWB(a32_nrm, b32_inv); /* Q: 29 + a_headrm - b_headrm */
|
Chris@69
|
120
|
Chris@69
|
121 /* Compute residual by subtracting product of denominator and first approximation */
|
Chris@69
|
122 /* It's OK to overflow because the final value of a32_nrm should always be small */
|
Chris@69
|
123 a32_nrm = silk_SUB32_ovflw(a32_nrm, silk_LSHIFT_ovflw( silk_SMMUL(b32_nrm, result), 3 )); /* Q: a_headrm */
|
Chris@69
|
124
|
Chris@69
|
125 /* Refinement */
|
Chris@69
|
126 result = silk_SMLAWB(result, a32_nrm, b32_inv); /* Q: 29 + a_headrm - b_headrm */
|
Chris@69
|
127
|
Chris@69
|
128 /* Convert to Qres domain */
|
Chris@69
|
129 lshift = 29 + a_headrm - b_headrm - Qres;
|
Chris@69
|
130 if( lshift < 0 ) {
|
Chris@69
|
131 return silk_LSHIFT_SAT32(result, -lshift);
|
Chris@69
|
132 } else {
|
Chris@69
|
133 if( lshift < 32){
|
Chris@69
|
134 return silk_RSHIFT(result, lshift);
|
Chris@69
|
135 } else {
|
Chris@69
|
136 /* Avoid undefined result */
|
Chris@69
|
137 return 0;
|
Chris@69
|
138 }
|
Chris@69
|
139 }
|
Chris@69
|
140 }
|
Chris@69
|
141
|
Chris@69
|
142 /* Invert int32 value and return result as int32 in a given Q-domain */
|
Chris@69
|
143 static OPUS_INLINE opus_int32 silk_INVERSE32_varQ( /* O returns a good approximation of "(1 << Qres) / b32" */
|
Chris@69
|
144 const opus_int32 b32, /* I denominator (Q0) */
|
Chris@69
|
145 const opus_int Qres /* I Q-domain of result (> 0) */
|
Chris@69
|
146 )
|
Chris@69
|
147 {
|
Chris@69
|
148 opus_int b_headrm, lshift;
|
Chris@69
|
149 opus_int32 b32_inv, b32_nrm, err_Q32, result;
|
Chris@69
|
150
|
Chris@69
|
151 silk_assert( b32 != 0 );
|
Chris@69
|
152 silk_assert( Qres > 0 );
|
Chris@69
|
153
|
Chris@69
|
154 /* Compute number of bits head room and normalize input */
|
Chris@69
|
155 b_headrm = silk_CLZ32( silk_abs(b32) ) - 1;
|
Chris@69
|
156 b32_nrm = silk_LSHIFT(b32, b_headrm); /* Q: b_headrm */
|
Chris@69
|
157
|
Chris@69
|
158 /* Inverse of b32, with 14 bits of precision */
|
Chris@69
|
159 b32_inv = silk_DIV32_16( silk_int32_MAX >> 2, silk_RSHIFT(b32_nrm, 16) ); /* Q: 29 + 16 - b_headrm */
|
Chris@69
|
160
|
Chris@69
|
161 /* First approximation */
|
Chris@69
|
162 result = silk_LSHIFT(b32_inv, 16); /* Q: 61 - b_headrm */
|
Chris@69
|
163
|
Chris@69
|
164 /* Compute residual by subtracting product of denominator and first approximation from one */
|
Chris@69
|
165 err_Q32 = silk_LSHIFT( ((opus_int32)1<<29) - silk_SMULWB(b32_nrm, b32_inv), 3 ); /* Q32 */
|
Chris@69
|
166
|
Chris@69
|
167 /* Refinement */
|
Chris@69
|
168 result = silk_SMLAWW(result, err_Q32, b32_inv); /* Q: 61 - b_headrm */
|
Chris@69
|
169
|
Chris@69
|
170 /* Convert to Qres domain */
|
Chris@69
|
171 lshift = 61 - b_headrm - Qres;
|
Chris@69
|
172 if( lshift <= 0 ) {
|
Chris@69
|
173 return silk_LSHIFT_SAT32(result, -lshift);
|
Chris@69
|
174 } else {
|
Chris@69
|
175 if( lshift < 32){
|
Chris@69
|
176 return silk_RSHIFT(result, lshift);
|
Chris@69
|
177 }else{
|
Chris@69
|
178 /* Avoid undefined result */
|
Chris@69
|
179 return 0;
|
Chris@69
|
180 }
|
Chris@69
|
181 }
|
Chris@69
|
182 }
|
Chris@69
|
183
|
Chris@69
|
184 #ifdef __cplusplus
|
Chris@69
|
185 }
|
Chris@69
|
186 #endif
|
Chris@69
|
187
|
Chris@69
|
188 #endif /* SILK_FIX_INLINES_H */
|