1/* Copyright (c) 2008-2011 Octasic Inc.
2   Written by Jean-Marc Valin */
3/*
4   Redistribution and use in source and binary forms, with or without
5   modification, are permitted provided that the following conditions
6   are met:
7
8   - Redistributions of source code must retain the above copyright
9   notice, this list of conditions and the following disclaimer.
10
11   - Redistributions in binary form must reproduce the above copyright
12   notice, this list of conditions and the following disclaimer in the
13   documentation and/or other materials provided with the distribution.
14
15   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
18   A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
19   CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
20   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
21   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
22   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
23   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
24   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
25   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26*/
27
28#ifdef HAVE_CONFIG_H
29#include "config.h"
30#endif
31
32#include "opus_types.h"
33#include "opus_defines.h"
34
35#include <math.h>
36#include "mlp.h"
37#include "arch.h"
38#include "tansig_table.h"
39#define MAX_NEURONS 100
40
41#if 0
42static OPUS_INLINE opus_val16 tansig_approx(opus_val32 _x) /* Q19 */
43{
44    int i;
45    opus_val16 xx; /* Q11 */
46    /*double x, y;*/
47    opus_val16 dy, yy; /* Q14 */
48    /*x = 1.9073e-06*_x;*/
49    if (_x>=QCONST32(8,19))
50        return QCONST32(1.,14);
51    if (_x<=-QCONST32(8,19))
52        return -QCONST32(1.,14);
53    xx = EXTRACT16(SHR32(_x, 8));
54    /*i = lrint(25*x);*/
55    i = SHR32(ADD32(1024,MULT16_16(25, xx)),11);
56    /*x -= .04*i;*/
57    xx -= EXTRACT16(SHR32(MULT16_16(20972,i),8));
58    /*x = xx*(1./2048);*/
59    /*y = tansig_table[250+i];*/
60    yy = tansig_table[250+i];
61    /*y = yy*(1./16384);*/
62    dy = 16384-MULT16_16_Q14(yy,yy);
63    yy = yy + MULT16_16_Q14(MULT16_16_Q11(xx,dy),(16384 - MULT16_16_Q11(yy,xx)));
64    return yy;
65}
66#else
67/*extern const float tansig_table[501];*/
68static OPUS_INLINE float tansig_approx(float x)
69{
70    int i;
71    float y, dy;
72    float sign=1;
73    /* Tests are reversed to catch NaNs */
74    if (!(x<8))
75        return 1;
76    if (!(x>-8))
77        return -1;
78#ifndef FIXED_POINT
79    /* Another check in case of -ffast-math */
80    if (celt_isnan(x))
81       return 0;
82#endif
83    if (x<0)
84    {
85       x=-x;
86       sign=-1;
87    }
88    i = (int)floor(.5f+25*x);
89    x -= .04f*i;
90    y = tansig_table[i];
91    dy = 1-y*y;
92    y = y + x*dy*(1 - y*x);
93    return sign*y;
94}
95#endif
96
97#if 0
98void mlp_process(const MLP *m, const opus_val16 *in, opus_val16 *out)
99{
100    int j;
101    opus_val16 hidden[MAX_NEURONS];
102    const opus_val16 *W = m->weights;
103    /* Copy to tmp_in */
104    for (j=0;j<m->topo[1];j++)
105    {
106        int k;
107        opus_val32 sum = SHL32(EXTEND32(*W++),8);
108        for (k=0;k<m->topo[0];k++)
109            sum = MAC16_16(sum, in[k],*W++);
110        hidden[j] = tansig_approx(sum);
111    }
112    for (j=0;j<m->topo[2];j++)
113    {
114        int k;
115        opus_val32 sum = SHL32(EXTEND32(*W++),14);
116        for (k=0;k<m->topo[1];k++)
117            sum = MAC16_16(sum, hidden[k], *W++);
118        out[j] = tansig_approx(EXTRACT16(PSHR32(sum,17)));
119    }
120}
121#else
122void mlp_process(const MLP *m, const float *in, float *out)
123{
124    int j;
125    float hidden[MAX_NEURONS];
126    const float *W = m->weights;
127    /* Copy to tmp_in */
128    for (j=0;j<m->topo[1];j++)
129    {
130        int k;
131        float sum = *W++;
132        for (k=0;k<m->topo[0];k++)
133            sum = sum + in[k]**W++;
134        hidden[j] = tansig_approx(sum);
135    }
136    for (j=0;j<m->topo[2];j++)
137    {
138        int k;
139        float sum = *W++;
140        for (k=0;k<m->topo[1];k++)
141            sum = sum + hidden[k]**W++;
142        out[j] = tansig_approx(sum);
143    }
144}
145#endif
146