1/*
2 * various filters for ACELP-based codecs
3 *
4 * Copyright (c) 2008 Vladimir Voroshilov
5 *
6 * This file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23#include <inttypes.h>
24
25#include "avcodec.h"
26#include "celp_filters.h"
27#include "libavutil/avassert.h"
28#include "libavutil/common.h"
29
30void ff_celp_convolve_circ(int16_t* fc_out, const int16_t* fc_in,
31                           const int16_t* filter, int len)
32{
33    int i, k;
34
35    memset(fc_out, 0, len * sizeof(int16_t));
36
37    /* Since there are few pulses over an entire subframe (i.e. almost
38       all fc_in[i] are zero) it is faster to loop over fc_in first. */
39    for (i = 0; i < len; i++) {
40        if (fc_in[i]) {
41            for (k = 0; k < i; k++)
42                fc_out[k] += (fc_in[i] * filter[len + k - i]) >> 15;
43
44            for (k = i; k < len; k++)
45                fc_out[k] += (fc_in[i] * filter[      k - i]) >> 15;
46        }
47    }
48}
49
50void ff_celp_circ_addf(float *out, const float *in,
51                       const float *lagged, int lag, float fac, int n)
52{
53    int k;
54    for (k = 0; k < lag; k++)
55        out[k] = in[k] + fac * lagged[n + k - lag];
56    for (; k < n; k++)
57        out[k] = in[k] + fac * lagged[    k - lag];
58}
59
60int ff_celp_lp_synthesis_filter(int16_t *out, const int16_t *filter_coeffs,
61                                const int16_t *in, int buffer_length,
62                                int filter_length, int stop_on_overflow,
63                                int shift, int rounder)
64{
65    int i,n;
66
67    for (n = 0; n < buffer_length; n++) {
68        int sum = -rounder, sum1;
69        for (i = 1; i <= filter_length; i++)
70            sum += filter_coeffs[i-1] * out[n-i];
71
72        sum1 = ((-sum >> 12) + in[n]) >> shift;
73        sum  = av_clip_int16(sum1);
74
75        if (stop_on_overflow && sum != sum1)
76            return 1;
77
78        out[n] = sum;
79    }
80
81    return 0;
82}
83
84void ff_celp_lp_synthesis_filterf(float *out, const float *filter_coeffs,
85                                  const float* in, int buffer_length,
86                                  int filter_length)
87{
88    int i,n;
89
90#if 0 // Unoptimized code path for improved readability
91    for (n = 0; n < buffer_length; n++) {
92        out[n] = in[n];
93        for (i = 1; i <= filter_length; i++)
94            out[n] -= filter_coeffs[i-1] * out[n-i];
95    }
96#else
97    float out0, out1, out2, out3;
98    float old_out0, old_out1, old_out2, old_out3;
99    float a,b,c;
100
101    a = filter_coeffs[0];
102    b = filter_coeffs[1];
103    c = filter_coeffs[2];
104    b -= filter_coeffs[0] * filter_coeffs[0];
105    c -= filter_coeffs[1] * filter_coeffs[0];
106    c -= filter_coeffs[0] * b;
107
108    av_assert2((filter_length&1)==0 && filter_length>=4);
109
110    old_out0 = out[-4];
111    old_out1 = out[-3];
112    old_out2 = out[-2];
113    old_out3 = out[-1];
114    for (n = 0; n <= buffer_length - 4; n+=4) {
115        float tmp0,tmp1,tmp2;
116        float val;
117
118        out0 = in[0];
119        out1 = in[1];
120        out2 = in[2];
121        out3 = in[3];
122
123        out0 -= filter_coeffs[2] * old_out1;
124        out1 -= filter_coeffs[2] * old_out2;
125        out2 -= filter_coeffs[2] * old_out3;
126
127        out0 -= filter_coeffs[1] * old_out2;
128        out1 -= filter_coeffs[1] * old_out3;
129
130        out0 -= filter_coeffs[0] * old_out3;
131
132        val = filter_coeffs[3];
133
134        out0 -= val * old_out0;
135        out1 -= val * old_out1;
136        out2 -= val * old_out2;
137        out3 -= val * old_out3;
138
139        for (i = 5; i < filter_length; i += 2) {
140            old_out3 = out[-i];
141            val = filter_coeffs[i-1];
142
143            out0 -= val * old_out3;
144            out1 -= val * old_out0;
145            out2 -= val * old_out1;
146            out3 -= val * old_out2;
147
148            old_out2 = out[-i-1];
149
150            val = filter_coeffs[i];
151
152            out0 -= val * old_out2;
153            out1 -= val * old_out3;
154            out2 -= val * old_out0;
155            out3 -= val * old_out1;
156
157            FFSWAP(float, old_out0, old_out2);
158            old_out1 = old_out3;
159        }
160
161        tmp0 = out0;
162        tmp1 = out1;
163        tmp2 = out2;
164
165        out3 -= a * tmp2;
166        out2 -= a * tmp1;
167        out1 -= a * tmp0;
168
169        out3 -= b * tmp1;
170        out2 -= b * tmp0;
171
172        out3 -= c * tmp0;
173
174
175        out[0] = out0;
176        out[1] = out1;
177        out[2] = out2;
178        out[3] = out3;
179
180        old_out0 = out0;
181        old_out1 = out1;
182        old_out2 = out2;
183        old_out3 = out3;
184
185        out += 4;
186        in  += 4;
187    }
188
189    out -= n;
190    in -= n;
191    for (; n < buffer_length; n++) {
192        out[n] = in[n];
193        for (i = 1; i <= filter_length; i++)
194            out[n] -= filter_coeffs[i-1] * out[n-i];
195    }
196#endif
197}
198
199void ff_celp_lp_zero_synthesis_filterf(float *out, const float *filter_coeffs,
200                                       const float *in, int buffer_length,
201                                       int filter_length)
202{
203    int i,n;
204
205    for (n = 0; n < buffer_length; n++) {
206        out[n] = in[n];
207        for (i = 1; i <= filter_length; i++)
208            out[n] += filter_coeffs[i-1] * in[n-i];
209    }
210}
211
212void ff_celp_filter_init(CELPFContext *c)
213{
214    c->celp_lp_synthesis_filterf        = ff_celp_lp_synthesis_filterf;
215    c->celp_lp_zero_synthesis_filterf   = ff_celp_lp_zero_synthesis_filterf;
216
217    if(HAVE_MIPSFPU)
218        ff_celp_filter_init_mips(c);
219}
220