celp_filters.c
Go to the documentation of this file.
1 /*
2  * various filters for ACELP-based codecs
3  *
4  * Copyright (c) 2008 Vladimir Voroshilov
5  *
6  * This file is part of Libav.
7  *
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.
12  *
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.
17  *
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
21  */
22 
23 #include <inttypes.h>
24 
25 #include "avcodec.h"
26 #include "celp_filters.h"
27 
28 void ff_celp_convolve_circ(int16_t* fc_out, const int16_t* fc_in,
29  const int16_t* filter, int len)
30 {
31  int i, k;
32 
33  memset(fc_out, 0, len * sizeof(int16_t));
34 
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++) {
38  if (fc_in[i]) {
39  for (k = 0; k < i; k++)
40  fc_out[k] += (fc_in[i] * filter[len + k - i]) >> 15;
41 
42  for (k = i; k < len; k++)
43  fc_out[k] += (fc_in[i] * filter[ k - i]) >> 15;
44  }
45  }
46 }
47 
48 void ff_celp_circ_addf(float *out, const float *in,
49  const float *lagged, int lag, float fac, int n)
50 {
51  int k;
52  for (k = 0; k < lag; k++)
53  out[k] = in[k] + fac * lagged[n + k - lag];
54  for (; k < n; k++)
55  out[k] = in[k] + fac * lagged[ k - lag];
56 }
57 
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 rounder)
62 {
63  int i,n;
64 
65  for (n = 0; n < buffer_length; n++) {
66  int sum = rounder;
67  for (i = 1; i <= filter_length; i++)
68  sum -= filter_coeffs[i-1] * out[n-i];
69 
70  sum = (sum >> 12) + in[n];
71 
72  if (sum + 0x8000 > 0xFFFFU) {
73  if (stop_on_overflow)
74  return 1;
75  sum = (sum >> 31) ^ 32767;
76  }
77  out[n] = sum;
78  }
79 
80  return 0;
81 }
82 
83 void ff_celp_lp_synthesis_filterf(float *out, const float *filter_coeffs,
84  const float* in, int buffer_length,
85  int filter_length)
86 {
87  int i,n;
88 
89 #if 0 // Unoptimized code path for improved readability
90  for (n = 0; n < buffer_length; n++) {
91  out[n] = in[n];
92  for (i = 1; i <= filter_length; i++)
93  out[n] -= filter_coeffs[i-1] * out[n-i];
94  }
95 #else
96  float out0, out1, out2, out3;
97  float old_out0, old_out1, old_out2, old_out3;
98  float a,b,c;
99 
100  a = filter_coeffs[0];
101  b = filter_coeffs[1];
102  c = filter_coeffs[2];
103  b -= filter_coeffs[0] * filter_coeffs[0];
104  c -= filter_coeffs[1] * filter_coeffs[0];
105  c -= filter_coeffs[0] * b;
106 
107  old_out0 = out[-4];
108  old_out1 = out[-3];
109  old_out2 = out[-2];
110  old_out3 = out[-1];
111  for (n = 0; n <= buffer_length - 4; n+=4) {
112  float tmp0,tmp1,tmp2;
113  float val;
114 
115  out0 = in[0];
116  out1 = in[1];
117  out2 = in[2];
118  out3 = in[3];
119 
120  out0 -= filter_coeffs[2] * old_out1;
121  out1 -= filter_coeffs[2] * old_out2;
122  out2 -= filter_coeffs[2] * old_out3;
123 
124  out0 -= filter_coeffs[1] * old_out2;
125  out1 -= filter_coeffs[1] * old_out3;
126 
127  out0 -= filter_coeffs[0] * old_out3;
128 
129  val = filter_coeffs[3];
130 
131  out0 -= val * old_out0;
132  out1 -= val * old_out1;
133  out2 -= val * old_out2;
134  out3 -= val * old_out3;
135 
136  for (i = 5; i <= filter_length; i += 2) {
137  old_out3 = out[-i];
138  val = filter_coeffs[i-1];
139 
140  out0 -= val * old_out3;
141  out1 -= val * old_out0;
142  out2 -= val * old_out1;
143  out3 -= val * old_out2;
144 
145  old_out2 = out[-i-1];
146 
147  val = filter_coeffs[i];
148 
149  out0 -= val * old_out2;
150  out1 -= val * old_out3;
151  out2 -= val * old_out0;
152  out3 -= val * old_out1;
153 
154  FFSWAP(float, old_out0, old_out2);
155  old_out1 = old_out3;
156  }
157 
158  tmp0 = out0;
159  tmp1 = out1;
160  tmp2 = out2;
161 
162  out3 -= a * tmp2;
163  out2 -= a * tmp1;
164  out1 -= a * tmp0;
165 
166  out3 -= b * tmp1;
167  out2 -= b * tmp0;
168 
169  out3 -= c * tmp0;
170 
171 
172  out[0] = out0;
173  out[1] = out1;
174  out[2] = out2;
175  out[3] = out3;
176 
177  old_out0 = out0;
178  old_out1 = out1;
179  old_out2 = out2;
180  old_out3 = out3;
181 
182  out += 4;
183  in += 4;
184  }
185 
186  out -= n;
187  in -= n;
188  for (; n < buffer_length; n++) {
189  out[n] = in[n];
190  for (i = 1; i <= filter_length; i++)
191  out[n] -= filter_coeffs[i-1] * out[n-i];
192  }
193 #endif
194 }
195 
196 void ff_celp_lp_zero_synthesis_filterf(float *out, const float *filter_coeffs,
197  const float *in, int buffer_length,
198  int filter_length)
199 {
200  int i,n;
201 
202  for (n = 0; n < buffer_length; n++) {
203  out[n] = in[n];
204  for (i = 1; i <= filter_length; i++)
205  out[n] += filter_coeffs[i-1] * in[n-i];
206  }
207 }