Libav 0.7.1
|
00001 /* 00002 * various filters for ACELP-based codecs 00003 * 00004 * Copyright (c) 2008 Vladimir Voroshilov 00005 * 00006 * This file is part of Libav. 00007 * 00008 * Libav is free software; you can redistribute it and/or 00009 * modify it under the terms of the GNU Lesser General Public 00010 * License as published by the Free Software Foundation; either 00011 * version 2.1 of the License, or (at your option) any later version. 00012 * 00013 * Libav is distributed in the hope that it will be useful, 00014 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00015 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00016 * Lesser General Public License for more details. 00017 * 00018 * You should have received a copy of the GNU Lesser General Public 00019 * License along with Libav; if not, write to the Free Software 00020 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00021 */ 00022 00023 #include <inttypes.h> 00024 00025 #include "avcodec.h" 00026 #include "celp_filters.h" 00027 00028 void ff_celp_convolve_circ(int16_t* fc_out, const int16_t* fc_in, 00029 const int16_t* filter, int len) 00030 { 00031 int i, k; 00032 00033 memset(fc_out, 0, len * sizeof(int16_t)); 00034 00035 /* Since there are few pulses over an entire subframe (i.e. almost 00036 all fc_in[i] are zero) it is faster to loop over fc_in first. */ 00037 for (i = 0; i < len; i++) { 00038 if (fc_in[i]) { 00039 for (k = 0; k < i; k++) 00040 fc_out[k] += (fc_in[i] * filter[len + k - i]) >> 15; 00041 00042 for (k = i; k < len; k++) 00043 fc_out[k] += (fc_in[i] * filter[ k - i]) >> 15; 00044 } 00045 } 00046 } 00047 00048 void ff_celp_circ_addf(float *out, const float *in, 00049 const float *lagged, int lag, float fac, int n) 00050 { 00051 int k; 00052 for (k = 0; k < lag; k++) 00053 out[k] = in[k] + fac * lagged[n + k - lag]; 00054 for (; k < n; k++) 00055 out[k] = in[k] + fac * lagged[ k - lag]; 00056 } 00057 00058 int ff_celp_lp_synthesis_filter(int16_t *out, const int16_t *filter_coeffs, 00059 const int16_t *in, int buffer_length, 00060 int filter_length, int stop_on_overflow, 00061 int rounder) 00062 { 00063 int i,n; 00064 00065 for (n = 0; n < buffer_length; n++) { 00066 int sum = rounder; 00067 for (i = 1; i <= filter_length; i++) 00068 sum -= filter_coeffs[i-1] * out[n-i]; 00069 00070 sum = (sum >> 12) + in[n]; 00071 00072 if (sum + 0x8000 > 0xFFFFU) { 00073 if (stop_on_overflow) 00074 return 1; 00075 sum = (sum >> 31) ^ 32767; 00076 } 00077 out[n] = sum; 00078 } 00079 00080 return 0; 00081 } 00082 00083 void ff_celp_lp_synthesis_filterf(float *out, const float *filter_coeffs, 00084 const float* in, int buffer_length, 00085 int filter_length) 00086 { 00087 int i,n; 00088 00089 #if 0 // Unoptimized code path for improved readability 00090 for (n = 0; n < buffer_length; n++) { 00091 out[n] = in[n]; 00092 for (i = 1; i <= filter_length; i++) 00093 out[n] -= filter_coeffs[i-1] * out[n-i]; 00094 } 00095 #else 00096 float out0, out1, out2, out3; 00097 float old_out0, old_out1, old_out2, old_out3; 00098 float a,b,c; 00099 00100 a = filter_coeffs[0]; 00101 b = filter_coeffs[1]; 00102 c = filter_coeffs[2]; 00103 b -= filter_coeffs[0] * filter_coeffs[0]; 00104 c -= filter_coeffs[1] * filter_coeffs[0]; 00105 c -= filter_coeffs[0] * b; 00106 00107 old_out0 = out[-4]; 00108 old_out1 = out[-3]; 00109 old_out2 = out[-2]; 00110 old_out3 = out[-1]; 00111 for (n = 0; n <= buffer_length - 4; n+=4) { 00112 float tmp0,tmp1,tmp2; 00113 float val; 00114 00115 out0 = in[0]; 00116 out1 = in[1]; 00117 out2 = in[2]; 00118 out3 = in[3]; 00119 00120 out0 -= filter_coeffs[2] * old_out1; 00121 out1 -= filter_coeffs[2] * old_out2; 00122 out2 -= filter_coeffs[2] * old_out3; 00123 00124 out0 -= filter_coeffs[1] * old_out2; 00125 out1 -= filter_coeffs[1] * old_out3; 00126 00127 out0 -= filter_coeffs[0] * old_out3; 00128 00129 val = filter_coeffs[3]; 00130 00131 out0 -= val * old_out0; 00132 out1 -= val * old_out1; 00133 out2 -= val * old_out2; 00134 out3 -= val * old_out3; 00135 00136 old_out3 = out[-5]; 00137 00138 for (i = 5; i <= filter_length; i += 2) { 00139 val = filter_coeffs[i-1]; 00140 00141 out0 -= val * old_out3; 00142 out1 -= val * old_out0; 00143 out2 -= val * old_out1; 00144 out3 -= val * old_out2; 00145 00146 old_out2 = out[-i-1]; 00147 00148 val = filter_coeffs[i]; 00149 00150 out0 -= val * old_out2; 00151 out1 -= val * old_out3; 00152 out2 -= val * old_out0; 00153 out3 -= val * old_out1; 00154 00155 FFSWAP(float, old_out0, old_out2); 00156 old_out1 = old_out3; 00157 old_out3 = out[-i-2]; 00158 } 00159 00160 tmp0 = out0; 00161 tmp1 = out1; 00162 tmp2 = out2; 00163 00164 out3 -= a * tmp2; 00165 out2 -= a * tmp1; 00166 out1 -= a * tmp0; 00167 00168 out3 -= b * tmp1; 00169 out2 -= b * tmp0; 00170 00171 out3 -= c * tmp0; 00172 00173 00174 out[0] = out0; 00175 out[1] = out1; 00176 out[2] = out2; 00177 out[3] = out3; 00178 00179 old_out0 = out0; 00180 old_out1 = out1; 00181 old_out2 = out2; 00182 old_out3 = out3; 00183 00184 out += 4; 00185 in += 4; 00186 } 00187 00188 out -= n; 00189 in -= n; 00190 for (; n < buffer_length; n++) { 00191 out[n] = in[n]; 00192 for (i = 1; i <= filter_length; i++) 00193 out[n] -= filter_coeffs[i-1] * out[n-i]; 00194 } 00195 #endif 00196 } 00197 00198 void ff_celp_lp_zero_synthesis_filterf(float *out, const float *filter_coeffs, 00199 const float *in, int buffer_length, 00200 int filter_length) 00201 { 00202 int i,n; 00203 00204 for (n = 0; n < buffer_length; n++) { 00205 out[n] = in[n]; 00206 for (i = 1; i <= filter_length; i++) 00207 out[n] += filter_coeffs[i-1] * in[n-i]; 00208 } 00209 }