2 * various filters for ACELP-based codecs
4 * Copyright (c) 2008 Vladimir Voroshilov
6 * This file is part of Libav.
8 * Libav 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.
13 * Libav 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.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 #include "celp_filters.h"
28 void ff_celp_convolve_circ(int16_t* fc_out, const int16_t* fc_in,
29 const int16_t* filter, int len)
33 memset(fc_out, 0, len * sizeof(int16_t));
35 /* Since there are few pulses over an entire subframe (i.e. almost
36 all fc_in[i] are zero) it is faster to loop over fc_in first. */
37 for (i = 0; i < len; i++) {
39 for (k = 0; k < i; k++)
40 fc_out[k] += (fc_in[i] * filter[len + k - i]) >> 15;
42 for (k = i; k < len; k++)
43 fc_out[k] += (fc_in[i] * filter[ k - i]) >> 15;
48 void ff_celp_circ_addf(float *out, const float *in,
49 const float *lagged, int lag, float fac, int n)
52 for (k = 0; k < lag; k++)
53 out[k] = in[k] + fac * lagged[n + k - lag];
55 out[k] = in[k] + fac * lagged[ k - lag];
58 int ff_celp_lp_synthesis_filter(int16_t *out, const int16_t *filter_coeffs,
59 const int16_t *in, int buffer_length,
60 int filter_length, int stop_on_overflow,
61 int shift, int rounder)
65 for (n = 0; n < buffer_length; n++) {
66 int sum = -rounder, sum1;
67 for (i = 1; i <= filter_length; i++)
68 sum += filter_coeffs[i-1] * out[n-i];
70 sum1 = ((-sum >> 12) + in[n]) >> shift;
71 sum = av_clip_int16(sum1);
73 if (stop_on_overflow && sum != sum1)
82 void ff_celp_lp_synthesis_filterf(float *out, const float *filter_coeffs,
83 const float* in, int buffer_length,
88 #if 0 // Unoptimized code path for improved readability
89 for (n = 0; n < buffer_length; n++) {
91 for (i = 1; i <= filter_length; i++)
92 out[n] -= filter_coeffs[i-1] * out[n-i];
95 float out0, out1, out2, out3;
96 float old_out0, old_out1, old_out2, old_out3;
100 b = filter_coeffs[1];
101 c = filter_coeffs[2];
102 b -= filter_coeffs[0] * filter_coeffs[0];
103 c -= filter_coeffs[1] * filter_coeffs[0];
104 c -= filter_coeffs[0] * b;
110 for (n = 0; n <= buffer_length - 4; n+=4) {
111 float tmp0,tmp1,tmp2;
119 out0 -= filter_coeffs[2] * old_out1;
120 out1 -= filter_coeffs[2] * old_out2;
121 out2 -= filter_coeffs[2] * old_out3;
123 out0 -= filter_coeffs[1] * old_out2;
124 out1 -= filter_coeffs[1] * old_out3;
126 out0 -= filter_coeffs[0] * old_out3;
128 val = filter_coeffs[3];
130 out0 -= val * old_out0;
131 out1 -= val * old_out1;
132 out2 -= val * old_out2;
133 out3 -= val * old_out3;
135 for (i = 5; i <= filter_length; i += 2) {
137 val = filter_coeffs[i-1];
139 out0 -= val * old_out3;
140 out1 -= val * old_out0;
141 out2 -= val * old_out1;
142 out3 -= val * old_out2;
144 old_out2 = out[-i-1];
146 val = filter_coeffs[i];
148 out0 -= val * old_out2;
149 out1 -= val * old_out3;
150 out2 -= val * old_out0;
151 out3 -= val * old_out1;
153 FFSWAP(float, old_out0, old_out2);
187 for (; n < buffer_length; n++) {
189 for (i = 1; i <= filter_length; i++)
190 out[n] -= filter_coeffs[i-1] * out[n-i];
195 void ff_celp_lp_zero_synthesis_filterf(float *out, const float *filter_coeffs,
196 const float *in, int buffer_length,
201 for (n = 0; n < buffer_length; n++) {
203 for (i = 1; i <= filter_length; i++)
204 out[n] += filter_coeffs[i-1] * in[n-i];