1/***********************************************************************
2Copyright (c) 2006-2011, Skype Limited. All rights reserved.
3Redistribution and use in source and binary forms, with or without
4modification, are permitted provided that the following conditions
5are met:
6- Redistributions of source code must retain the above copyright notice,
7this list of conditions and the following disclaimer.
8- Redistributions in binary form must reproduce the above copyright
9notice, this list of conditions and the following disclaimer in the
10documentation and/or other materials provided with the distribution.
11- Neither the name of Internet Society, IETF or IETF Trust, nor the
12names of specific contributors, may be used to endorse or promote
13products derived from this software without specific prior written
14permission.
15THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
16AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
19LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25POSSIBILITY OF SUCH DAMAGE.
26***********************************************************************/
27
28/*! \file silk_Inlines.h
29 *  \brief silk_Inlines.h defines OPUS_INLINE signal processing functions.
30 */
31
32#ifndef SILK_FIX_INLINES_H
33#define SILK_FIX_INLINES_H
34
35#ifdef  __cplusplus
36extern "C"
37{
38#endif
39
40/* count leading zeros of opus_int64 */
41static OPUS_INLINE opus_int32 silk_CLZ64( opus_int64 in )
42{
43    opus_int32 in_upper;
44
45    in_upper = (opus_int32)silk_RSHIFT64(in, 32);
46    if (in_upper == 0) {
47        /* Search in the lower 32 bits */
48        return 32 + silk_CLZ32( (opus_int32) in );
49    } else {
50        /* Search in the upper 32 bits */
51        return silk_CLZ32( in_upper );
52    }
53}
54
55/* get number of leading zeros and fractional part (the bits right after the leading one */
56static OPUS_INLINE void silk_CLZ_FRAC(
57    opus_int32 in,            /* I  input                               */
58    opus_int32 *lz,           /* O  number of leading zeros             */
59    opus_int32 *frac_Q7       /* O  the 7 bits right after the leading one */
60)
61{
62    opus_int32 lzeros = silk_CLZ32(in);
63
64    * lz = lzeros;
65    * frac_Q7 = silk_ROR32(in, 24 - lzeros) & 0x7f;
66}
67
68/* Approximation of square root                                          */
69/* Accuracy: < +/- 10%  for output values > 15                           */
70/*           < +/- 2.5% for output values > 120                          */
71static OPUS_INLINE opus_int32 silk_SQRT_APPROX( opus_int32 x )
72{
73    opus_int32 y, lz, frac_Q7;
74
75    if( x <= 0 ) {
76        return 0;
77    }
78
79    silk_CLZ_FRAC(x, &lz, &frac_Q7);
80
81    if( lz & 1 ) {
82        y = 32768;
83    } else {
84        y = 46214;        /* 46214 = sqrt(2) * 32768 */
85    }
86
87    /* get scaling right */
88    y >>= silk_RSHIFT(lz, 1);
89
90    /* increment using fractional part of input */
91    y = silk_SMLAWB(y, y, silk_SMULBB(213, frac_Q7));
92
93    return y;
94}
95
96/* Divide two int32 values and return result as int32 in a given Q-domain */
97static OPUS_INLINE opus_int32 silk_DIV32_varQ(   /* O    returns a good approximation of "(a32 << Qres) / b32" */
98    const opus_int32     a32,               /* I    numerator (Q0)                  */
99    const opus_int32     b32,               /* I    denominator (Q0)                */
100    const opus_int       Qres               /* I    Q-domain of result (>= 0)       */
101)
102{
103    opus_int   a_headrm, b_headrm, lshift;
104    opus_int32 b32_inv, a32_nrm, b32_nrm, result;
105
106    silk_assert( b32 != 0 );
107    silk_assert( Qres >= 0 );
108
109    /* Compute number of bits head room and normalize inputs */
110    a_headrm = silk_CLZ32( silk_abs(a32) ) - 1;
111    a32_nrm = silk_LSHIFT(a32, a_headrm);                                       /* Q: a_headrm                  */
112    b_headrm = silk_CLZ32( silk_abs(b32) ) - 1;
113    b32_nrm = silk_LSHIFT(b32, b_headrm);                                       /* Q: b_headrm                  */
114
115    /* Inverse of b32, with 14 bits of precision */
116    b32_inv = silk_DIV32_16( silk_int32_MAX >> 2, silk_RSHIFT(b32_nrm, 16) );   /* Q: 29 + 16 - b_headrm        */
117
118    /* First approximation */
119    result = silk_SMULWB(a32_nrm, b32_inv);                                     /* Q: 29 + a_headrm - b_headrm  */
120
121    /* Compute residual by subtracting product of denominator and first approximation */
122    /* It's OK to overflow because the final value of a32_nrm should always be small */
123    a32_nrm = silk_SUB32_ovflw(a32_nrm, silk_LSHIFT_ovflw( silk_SMMUL(b32_nrm, result), 3 ));  /* Q: a_headrm   */
124
125    /* Refinement */
126    result = silk_SMLAWB(result, a32_nrm, b32_inv);                             /* Q: 29 + a_headrm - b_headrm  */
127
128    /* Convert to Qres domain */
129    lshift = 29 + a_headrm - b_headrm - Qres;
130    if( lshift < 0 ) {
131        return silk_LSHIFT_SAT32(result, -lshift);
132    } else {
133        if( lshift < 32){
134            return silk_RSHIFT(result, lshift);
135        } else {
136            /* Avoid undefined result */
137            return 0;
138        }
139    }
140}
141
142/* Invert int32 value and return result as int32 in a given Q-domain */
143static OPUS_INLINE opus_int32 silk_INVERSE32_varQ(   /* O    returns a good approximation of "(1 << Qres) / b32" */
144    const opus_int32     b32,                   /* I    denominator (Q0)                */
145    const opus_int       Qres                   /* I    Q-domain of result (> 0)        */
146)
147{
148    opus_int   b_headrm, lshift;
149    opus_int32 b32_inv, b32_nrm, err_Q32, result;
150
151    silk_assert( b32 != 0 );
152    silk_assert( Qres > 0 );
153
154    /* Compute number of bits head room and normalize input */
155    b_headrm = silk_CLZ32( silk_abs(b32) ) - 1;
156    b32_nrm = silk_LSHIFT(b32, b_headrm);                                       /* Q: b_headrm                */
157
158    /* Inverse of b32, with 14 bits of precision */
159    b32_inv = silk_DIV32_16( silk_int32_MAX >> 2, silk_RSHIFT(b32_nrm, 16) );   /* Q: 29 + 16 - b_headrm    */
160
161    /* First approximation */
162    result = silk_LSHIFT(b32_inv, 16);                                          /* Q: 61 - b_headrm            */
163
164    /* Compute residual by subtracting product of denominator and first approximation from one */
165    err_Q32 = silk_LSHIFT( ((opus_int32)1<<29) - silk_SMULWB(b32_nrm, b32_inv), 3 );        /* Q32                        */
166
167    /* Refinement */
168    result = silk_SMLAWW(result, err_Q32, b32_inv);                             /* Q: 61 - b_headrm            */
169
170    /* Convert to Qres domain */
171    lshift = 61 - b_headrm - Qres;
172    if( lshift <= 0 ) {
173        return silk_LSHIFT_SAT32(result, -lshift);
174    } else {
175        if( lshift < 32){
176            return silk_RSHIFT(result, lshift);
177        }else{
178            /* Avoid undefined result */
179            return 0;
180        }
181    }
182}
183
184#ifdef  __cplusplus
185}
186#endif
187
188#endif /* SILK_FIX_INLINES_H */
189