libavcodec/acelp_filters.c

   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 "acelp_filters.h"
  27 #define FRAC_BITS 13
  28 #include "mathops.h"
  29
  30 const int16_t ff_acelp_interp_filter[61] =
  31 { /* (0.15) */
  32   29443, 28346, 25207, 20449, 14701,  8693,
  33    3143, -1352, -4402, -5865, -5850, -4673,
  34   -2783,  -672,  1211,  2536,  3130,  2991,
  35    2259,  1170,     0, -1001, -1652, -1868,
  36   -1666, -1147,  -464,   218,   756,  1060,
  37    1099,   904,   550,   135,  -245,  -514,
  38    -634,  -602,  -451,  -231,     0,   191,
  39     308,   340,   296,   198,    78,   -36,
  40    -120,  -163,  -165,  -132,   -79,   -19,
  41      34,    73,    91,    89,    70,    38,
  42       0,
  43 };
  44
  45 void ff_acelp_interpolate(
  46         int16_t* out,
  47         const int16_t* in,
  48         const int16_t* filter_coeffs,
  49         int precision,
  50         int pitch_delay_frac,
  51         int filter_length,
  52         int length)
  53 {
  54     int n, i;
  55
  56     assert(pitch_delay_frac >= 0 && pitch_delay_frac < precision);
  57
  58     for(n=0; n<length; n++)
  59     {
  60         int idx = 0;
  61         int v = 0x4000;
  62
  63         for(i=0; i<filter_length;)
  64         {
  65
  66             /* The reference G.729 and AMR fixed point code performs clipping after
  67                each of the two following accumulations.
  68                Since clipping affects only the synthetic OVERFLOW test without
  69                causing an int type overflow, it was moved outside the loop. */
  70
  71             /*  R(x):=ac_v[-k+x]
  72                 v += R(n-i)*ff_acelp_interp_filter(t+6i)
  73                 v += R(n+i+1)*ff_acelp_interp_filter(6-t+6i) */
  74
  75             v += in[n + i] * filter_coeffs[idx + pitch_delay_frac];
  76             idx += precision;
  77             i++;
  78             v += in[n - i] * filter_coeffs[idx - pitch_delay_frac];
  79         }
  80         out[n] = av_clip_int16(v >> 15);
  81     }
  82 }
  83
  84 void ff_acelp_convolve_circ(
  85         int16_t* fc_out,
  86         const int16_t* fc_in,
  87         const int16_t* filter,
  88         int subframe_size)
  89 {
  90     int i, k;
  91
  92     memset(fc_out, 0, subframe_size * sizeof(int16_t));
  93
  94     /* Since there are few pulses over an entire subframe (i.e. almost
  95        all fc_in[i] are zero) it is faster to swap two loops and process
  96        non-zero samples only. In the case of G.729D the buffer contains
  97        two non-zero samples before the call to ff_acelp_enhance_harmonics
  98        and, due to pitch_delay being bounded by [20; 143], a maximum
  99        of four non-zero samples for a total of 40 after the call. */
 100     for(i=0; i<subframe_size; i++)
 101     {
 102         if(fc_in[i])
 103         {
 104             for(k=0; k<i; k++)
 105                 fc_out[k] += (fc_in[i] * filter[subframe_size + k - i]) >> 15;
 106
 107             for(k=i; k<subframe_size; k++)
 108                 fc_out[k] += (fc_in[i] * filter[k - i]) >> 15;
 109         }
 110     }
 111 }
 112
 113 int ff_acelp_lp_synthesis_filter(
 114         int16_t *out,
 115         const int16_t* filter_coeffs,
 116         const int16_t* in,
 117         int buffer_length,
 118         int filter_length,
 119         int stop_on_overflow)
 120 {
 121     int i,n;
 122
 123     for(n=0; n<buffer_length; n++)
 124     {
 125         int sum = 0x800;
 126         for(i=1; i<filter_length; i++)
 127             sum -= filter_coeffs[i] * out[n-i];
 128
 129         sum = (sum >> 12) + in[n];
 130
 131         /* Check for overflow */
 132         if(sum + 0x8000 > 0xFFFFU)
 133         {
 134             if(stop_on_overflow)
 135                 return 1;
 136             sum = (sum >> 31) ^ 32767;
 137         }
 138         out[n] = sum;
 139     }
 140
 141     return 0;
 142 }
 143
 144 void ff_acelp_weighted_filter(
 145         int16_t *out,
 146         const int16_t* in,
 147         const int16_t *weight_pow,
 148         int filter_length)
 149 {
 150     int n;
 151     for(n=0; n<filter_length; n++)
 152         out[n] = (in[n] * weight_pow[n] + 0x4000) >> 15; /* (3.12) = (0.15) * (3.12) with rounding */
 153 }
 154
 155 void ff_acelp_high_pass_filter(
 156         int16_t* out,
 157         int hpf_f[2],
 158         const int16_t* in,
 159         int length)
 160 {
 161     int i;
 162     int tmp;
 163
 164     for(i=0; i<length; i++)
 165     {
 166         tmp =  MULL(hpf_f[0], 15836);                     /* (14.13) = (13.13) * (1.13) */
 167         tmp += MULL(hpf_f[1], -7667);                     /* (13.13) = (13.13) * (0.13) */
 168         tmp += 7699 * (in[i] - 2*in[i-1] + in[i-2]); /* (14.13) =  (0.13) * (14.0) */
 169
 170         /* Multiplication by 2 with rounding can cause short type
 171            overflow, thus clipping is required. */
 172
 173         out[i] = av_clip_int16((tmp + 0x800) >> 12);      /* (15.0) = 2 * (13.13) = (14.13) */
 174
 175         hpf_f[1] = hpf_f[0];
 176         hpf_f[0] = tmp;
 177     }
 178 }