yading@10: /*
yading@10:  * various filters for ACELP-based codecs
yading@10:  *
yading@10:  * Copyright (c) 2008 Vladimir Voroshilov
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include <inttypes.h>
yading@10: 
yading@10: #include "avcodec.h"
yading@10: #include "celp_filters.h"
yading@10: #include "libavutil/avassert.h"
yading@10: #include "libavutil/common.h"
yading@10: 
yading@10: void ff_celp_convolve_circ(int16_t* fc_out, const int16_t* fc_in,
yading@10:                            const int16_t* filter, int len)
yading@10: {
yading@10:     int i, k;
yading@10: 
yading@10:     memset(fc_out, 0, len * sizeof(int16_t));
yading@10: 
yading@10:     /* Since there are few pulses over an entire subframe (i.e. almost
yading@10:        all fc_in[i] are zero) it is faster to loop over fc_in first. */
yading@10:     for (i = 0; i < len; i++) {
yading@10:         if (fc_in[i]) {
yading@10:             for (k = 0; k < i; k++)
yading@10:                 fc_out[k] += (fc_in[i] * filter[len + k - i]) >> 15;
yading@10: 
yading@10:             for (k = i; k < len; k++)
yading@10:                 fc_out[k] += (fc_in[i] * filter[      k - i]) >> 15;
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_celp_circ_addf(float *out, const float *in,
yading@10:                        const float *lagged, int lag, float fac, int n)
yading@10: {
yading@10:     int k;
yading@10:     for (k = 0; k < lag; k++)
yading@10:         out[k] = in[k] + fac * lagged[n + k - lag];
yading@10:     for (; k < n; k++)
yading@10:         out[k] = in[k] + fac * lagged[    k - lag];
yading@10: }
yading@10: 
yading@10: int ff_celp_lp_synthesis_filter(int16_t *out, const int16_t *filter_coeffs,
yading@10:                                 const int16_t *in, int buffer_length,
yading@10:                                 int filter_length, int stop_on_overflow,
yading@10:                                 int shift, int rounder)
yading@10: {
yading@10:     int i,n;
yading@10: 
yading@10:     for (n = 0; n < buffer_length; n++) {
yading@10:         int sum = -rounder, sum1;
yading@10:         for (i = 1; i <= filter_length; i++)
yading@10:             sum += filter_coeffs[i-1] * out[n-i];
yading@10: 
yading@10:         sum1 = ((-sum >> 12) + in[n]) >> shift;
yading@10:         sum  = av_clip_int16(sum1);
yading@10: 
yading@10:         if (stop_on_overflow && sum != sum1)
yading@10:             return 1;
yading@10: 
yading@10:         out[n] = sum;
yading@10:     }
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: void ff_celp_lp_synthesis_filterf(float *out, const float *filter_coeffs,
yading@10:                                   const float* in, int buffer_length,
yading@10:                                   int filter_length)
yading@10: {
yading@10:     int i,n;
yading@10: 
yading@10: #if 0 // Unoptimized code path for improved readability
yading@10:     for (n = 0; n < buffer_length; n++) {
yading@10:         out[n] = in[n];
yading@10:         for (i = 1; i <= filter_length; i++)
yading@10:             out[n] -= filter_coeffs[i-1] * out[n-i];
yading@10:     }
yading@10: #else
yading@10:     float out0, out1, out2, out3;
yading@10:     float old_out0, old_out1, old_out2, old_out3;
yading@10:     float a,b,c;
yading@10: 
yading@10:     a = filter_coeffs[0];
yading@10:     b = filter_coeffs[1];
yading@10:     c = filter_coeffs[2];
yading@10:     b -= filter_coeffs[0] * filter_coeffs[0];
yading@10:     c -= filter_coeffs[1] * filter_coeffs[0];
yading@10:     c -= filter_coeffs[0] * b;
yading@10: 
yading@10:     av_assert2((filter_length&1)==0 && filter_length>=4);
yading@10: 
yading@10:     old_out0 = out[-4];
yading@10:     old_out1 = out[-3];
yading@10:     old_out2 = out[-2];
yading@10:     old_out3 = out[-1];
yading@10:     for (n = 0; n <= buffer_length - 4; n+=4) {
yading@10:         float tmp0,tmp1,tmp2;
yading@10:         float val;
yading@10: 
yading@10:         out0 = in[0];
yading@10:         out1 = in[1];
yading@10:         out2 = in[2];
yading@10:         out3 = in[3];
yading@10: 
yading@10:         out0 -= filter_coeffs[2] * old_out1;
yading@10:         out1 -= filter_coeffs[2] * old_out2;
yading@10:         out2 -= filter_coeffs[2] * old_out3;
yading@10: 
yading@10:         out0 -= filter_coeffs[1] * old_out2;
yading@10:         out1 -= filter_coeffs[1] * old_out3;
yading@10: 
yading@10:         out0 -= filter_coeffs[0] * old_out3;
yading@10: 
yading@10:         val = filter_coeffs[3];
yading@10: 
yading@10:         out0 -= val * old_out0;
yading@10:         out1 -= val * old_out1;
yading@10:         out2 -= val * old_out2;
yading@10:         out3 -= val * old_out3;
yading@10: 
yading@10:         for (i = 5; i < filter_length; i += 2) {
yading@10:             old_out3 = out[-i];
yading@10:             val = filter_coeffs[i-1];
yading@10: 
yading@10:             out0 -= val * old_out3;
yading@10:             out1 -= val * old_out0;
yading@10:             out2 -= val * old_out1;
yading@10:             out3 -= val * old_out2;
yading@10: 
yading@10:             old_out2 = out[-i-1];
yading@10: 
yading@10:             val = filter_coeffs[i];
yading@10: 
yading@10:             out0 -= val * old_out2;
yading@10:             out1 -= val * old_out3;
yading@10:             out2 -= val * old_out0;
yading@10:             out3 -= val * old_out1;
yading@10: 
yading@10:             FFSWAP(float, old_out0, old_out2);
yading@10:             old_out1 = old_out3;
yading@10:         }
yading@10: 
yading@10:         tmp0 = out0;
yading@10:         tmp1 = out1;
yading@10:         tmp2 = out2;
yading@10: 
yading@10:         out3 -= a * tmp2;
yading@10:         out2 -= a * tmp1;
yading@10:         out1 -= a * tmp0;
yading@10: 
yading@10:         out3 -= b * tmp1;
yading@10:         out2 -= b * tmp0;
yading@10: 
yading@10:         out3 -= c * tmp0;
yading@10: 
yading@10: 
yading@10:         out[0] = out0;
yading@10:         out[1] = out1;
yading@10:         out[2] = out2;
yading@10:         out[3] = out3;
yading@10: 
yading@10:         old_out0 = out0;
yading@10:         old_out1 = out1;
yading@10:         old_out2 = out2;
yading@10:         old_out3 = out3;
yading@10: 
yading@10:         out += 4;
yading@10:         in  += 4;
yading@10:     }
yading@10: 
yading@10:     out -= n;
yading@10:     in -= n;
yading@10:     for (; n < buffer_length; n++) {
yading@10:         out[n] = in[n];
yading@10:         for (i = 1; i <= filter_length; i++)
yading@10:             out[n] -= filter_coeffs[i-1] * out[n-i];
yading@10:     }
yading@10: #endif
yading@10: }
yading@10: 
yading@10: void ff_celp_lp_zero_synthesis_filterf(float *out, const float *filter_coeffs,
yading@10:                                        const float *in, int buffer_length,
yading@10:                                        int filter_length)
yading@10: {
yading@10:     int i,n;
yading@10: 
yading@10:     for (n = 0; n < buffer_length; n++) {
yading@10:         out[n] = in[n];
yading@10:         for (i = 1; i <= filter_length; i++)
yading@10:             out[n] += filter_coeffs[i-1] * in[n-i];
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_celp_filter_init(CELPFContext *c)
yading@10: {
yading@10:     c->celp_lp_synthesis_filterf        = ff_celp_lp_synthesis_filterf;
yading@10:     c->celp_lp_zero_synthesis_filterf   = ff_celp_lp_zero_synthesis_filterf;
yading@10: 
yading@10:     if(HAVE_MIPSFPU)
yading@10:         ff_celp_filter_init_mips(c);
yading@10: }