1 /***********************************************************************
2 Copyright (c) 2006-2011, Skype Limited. All rights reserved.
3 Redistribution and use in source and binary forms, with or without
4 modification, are permitted provided that the following conditions
5 are met:
6 - Redistributions of source code must retain the above copyright notice,
7 this list of conditions and the following disclaimer.
8 - Redistributions in binary form must reproduce the above copyright
9 notice, this list of conditions and the following disclaimer in the
10 documentation and/or other materials provided with the distribution.
11 - Neither the name of Internet Society, IETF or IETF Trust, nor the
12 names of specific contributors, may be used to endorse or promote
13 products derived from this software without specific prior written
14 permission.
15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
16 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
19 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25 POSSIBILITY OF SUCH DAMAGE.
26 ***********************************************************************/
27
28 #ifdef HAVE_CONFIG_H
29 #include "config.h"
30 #endif
31
32 #include "main_FIX.h"
33 #include "tuning_parameters.h"
34
35 /* Head room for correlations */
36 #define LTP_CORRS_HEAD_ROOM 2
37
38 void silk_fit_LTP(
39 opus_int32 LTP_coefs_Q16[ LTP_ORDER ],
40 opus_int16 LTP_coefs_Q14[ LTP_ORDER ]
41 );
42
silk_find_LTP_FIX(opus_int16 b_Q14[MAX_NB_SUBFR * LTP_ORDER],opus_int32 WLTP[MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER],opus_int * LTPredCodGain_Q7,const opus_int16 r_lpc[],const opus_int lag[MAX_NB_SUBFR],const opus_int32 Wght_Q15[MAX_NB_SUBFR],const opus_int subfr_length,const opus_int nb_subfr,const opus_int mem_offset,opus_int corr_rshifts[MAX_NB_SUBFR],int arch)43 void silk_find_LTP_FIX(
44 opus_int16 b_Q14[ MAX_NB_SUBFR * LTP_ORDER ], /* O LTP coefs */
45 opus_int32 WLTP[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* O Weight for LTP quantization */
46 opus_int *LTPredCodGain_Q7, /* O LTP coding gain */
47 const opus_int16 r_lpc[], /* I residual signal after LPC signal + state for first 10 ms */
48 const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */
49 const opus_int32 Wght_Q15[ MAX_NB_SUBFR ], /* I weights */
50 const opus_int subfr_length, /* I subframe length */
51 const opus_int nb_subfr, /* I number of subframes */
52 const opus_int mem_offset, /* I number of samples in LTP memory */
53 opus_int corr_rshifts[ MAX_NB_SUBFR ], /* O right shifts applied to correlations */
54 int arch /* I Run-time architecture */
55 )
56 {
57 opus_int i, k, lshift;
58 const opus_int16 *r_ptr, *lag_ptr;
59 opus_int16 *b_Q14_ptr;
60
61 opus_int32 regu;
62 opus_int32 *WLTP_ptr;
63 opus_int32 b_Q16[ LTP_ORDER ], delta_b_Q14[ LTP_ORDER ], d_Q14[ MAX_NB_SUBFR ], nrg[ MAX_NB_SUBFR ], g_Q26;
64 opus_int32 w[ MAX_NB_SUBFR ], WLTP_max, max_abs_d_Q14, max_w_bits;
65
66 opus_int32 temp32, denom32;
67 opus_int extra_shifts;
68 opus_int rr_shifts, maxRshifts, maxRshifts_wxtra, LZs;
69 opus_int32 LPC_res_nrg, LPC_LTP_res_nrg, div_Q16;
70 opus_int32 Rr[ LTP_ORDER ], rr[ MAX_NB_SUBFR ];
71 opus_int32 wd, m_Q12;
72
73 b_Q14_ptr = b_Q14;
74 WLTP_ptr = WLTP;
75 r_ptr = &r_lpc[ mem_offset ];
76 for( k = 0; k < nb_subfr; k++ ) {
77 lag_ptr = r_ptr - ( lag[ k ] + LTP_ORDER / 2 );
78
79 silk_sum_sqr_shift( &rr[ k ], &rr_shifts, r_ptr, subfr_length ); /* rr[ k ] in Q( -rr_shifts ) */
80
81 /* Assure headroom */
82 LZs = silk_CLZ32( rr[k] );
83 if( LZs < LTP_CORRS_HEAD_ROOM ) {
84 rr[ k ] = silk_RSHIFT_ROUND( rr[ k ], LTP_CORRS_HEAD_ROOM - LZs );
85 rr_shifts += ( LTP_CORRS_HEAD_ROOM - LZs );
86 }
87 corr_rshifts[ k ] = rr_shifts;
88 silk_corrMatrix_FIX( lag_ptr, subfr_length, LTP_ORDER, LTP_CORRS_HEAD_ROOM, WLTP_ptr, &corr_rshifts[ k ], arch ); /* WLTP_fix_ptr in Q( -corr_rshifts[ k ] ) */
89
90 /* The correlation vector always has lower max abs value than rr and/or RR so head room is assured */
91 silk_corrVector_FIX( lag_ptr, r_ptr, subfr_length, LTP_ORDER, Rr, corr_rshifts[ k ], arch ); /* Rr_fix_ptr in Q( -corr_rshifts[ k ] ) */
92 if( corr_rshifts[ k ] > rr_shifts ) {
93 rr[ k ] = silk_RSHIFT( rr[ k ], corr_rshifts[ k ] - rr_shifts ); /* rr[ k ] in Q( -corr_rshifts[ k ] ) */
94 }
95 silk_assert( rr[ k ] >= 0 );
96
97 regu = 1;
98 regu = silk_SMLAWB( regu, rr[ k ], SILK_FIX_CONST( LTP_DAMPING/3, 16 ) );
99 regu = silk_SMLAWB( regu, matrix_ptr( WLTP_ptr, 0, 0, LTP_ORDER ), SILK_FIX_CONST( LTP_DAMPING/3, 16 ) );
100 regu = silk_SMLAWB( regu, matrix_ptr( WLTP_ptr, LTP_ORDER-1, LTP_ORDER-1, LTP_ORDER ), SILK_FIX_CONST( LTP_DAMPING/3, 16 ) );
101 silk_regularize_correlations_FIX( WLTP_ptr, &rr[k], regu, LTP_ORDER );
102
103 silk_solve_LDL_FIX( WLTP_ptr, LTP_ORDER, Rr, b_Q16 ); /* WLTP_fix_ptr and Rr_fix_ptr both in Q(-corr_rshifts[k]) */
104
105 /* Limit and store in Q14 */
106 silk_fit_LTP( b_Q16, b_Q14_ptr );
107
108 /* Calculate residual energy */
109 nrg[ k ] = silk_residual_energy16_covar_FIX( b_Q14_ptr, WLTP_ptr, Rr, rr[ k ], LTP_ORDER, 14 ); /* nrg_fix in Q( -corr_rshifts[ k ] ) */
110
111 /* temp = Wght[ k ] / ( nrg[ k ] * Wght[ k ] + 0.01f * subfr_length ); */
112 extra_shifts = silk_min_int( corr_rshifts[ k ], LTP_CORRS_HEAD_ROOM );
113 denom32 = silk_LSHIFT_SAT32( silk_SMULWB( nrg[ k ], Wght_Q15[ k ] ), 1 + extra_shifts ) + /* Q( -corr_rshifts[ k ] + extra_shifts ) */
114 silk_RSHIFT( silk_SMULWB( (opus_int32)subfr_length, 655 ), corr_rshifts[ k ] - extra_shifts ); /* Q( -corr_rshifts[ k ] + extra_shifts ) */
115 denom32 = silk_max( denom32, 1 );
116 silk_assert( ((opus_int64)Wght_Q15[ k ] << 16 ) < silk_int32_MAX ); /* Wght always < 0.5 in Q0 */
117 temp32 = silk_DIV32( silk_LSHIFT( (opus_int32)Wght_Q15[ k ], 16 ), denom32 ); /* Q( 15 + 16 + corr_rshifts[k] - extra_shifts ) */
118 temp32 = silk_RSHIFT( temp32, 31 + corr_rshifts[ k ] - extra_shifts - 26 ); /* Q26 */
119
120 /* Limit temp such that the below scaling never wraps around */
121 WLTP_max = 0;
122 for( i = 0; i < LTP_ORDER * LTP_ORDER; i++ ) {
123 WLTP_max = silk_max( WLTP_ptr[ i ], WLTP_max );
124 }
125 lshift = silk_CLZ32( WLTP_max ) - 1 - 3; /* keep 3 bits free for vq_nearest_neighbor_fix */
126 silk_assert( 26 - 18 + lshift >= 0 );
127 if( 26 - 18 + lshift < 31 ) {
128 temp32 = silk_min_32( temp32, silk_LSHIFT( (opus_int32)1, 26 - 18 + lshift ) );
129 }
130
131 silk_scale_vector32_Q26_lshift_18( WLTP_ptr, temp32, LTP_ORDER * LTP_ORDER ); /* WLTP_ptr in Q( 18 - corr_rshifts[ k ] ) */
132
133 w[ k ] = matrix_ptr( WLTP_ptr, LTP_ORDER/2, LTP_ORDER/2, LTP_ORDER ); /* w in Q( 18 - corr_rshifts[ k ] ) */
134 silk_assert( w[k] >= 0 );
135
136 r_ptr += subfr_length;
137 b_Q14_ptr += LTP_ORDER;
138 WLTP_ptr += LTP_ORDER * LTP_ORDER;
139 }
140
141 maxRshifts = 0;
142 for( k = 0; k < nb_subfr; k++ ) {
143 maxRshifts = silk_max_int( corr_rshifts[ k ], maxRshifts );
144 }
145
146 /* Compute LTP coding gain */
147 if( LTPredCodGain_Q7 != NULL ) {
148 LPC_LTP_res_nrg = 0;
149 LPC_res_nrg = 0;
150 silk_assert( LTP_CORRS_HEAD_ROOM >= 2 ); /* Check that no overflow will happen when adding */
151 for( k = 0; k < nb_subfr; k++ ) {
152 LPC_res_nrg = silk_ADD32( LPC_res_nrg, silk_RSHIFT( silk_ADD32( silk_SMULWB( rr[ k ], Wght_Q15[ k ] ), 1 ), 1 + ( maxRshifts - corr_rshifts[ k ] ) ) ); /* Q( -maxRshifts ) */
153 LPC_LTP_res_nrg = silk_ADD32( LPC_LTP_res_nrg, silk_RSHIFT( silk_ADD32( silk_SMULWB( nrg[ k ], Wght_Q15[ k ] ), 1 ), 1 + ( maxRshifts - corr_rshifts[ k ] ) ) ); /* Q( -maxRshifts ) */
154 }
155 LPC_LTP_res_nrg = silk_max( LPC_LTP_res_nrg, 1 ); /* avoid division by zero */
156
157 div_Q16 = silk_DIV32_varQ( LPC_res_nrg, LPC_LTP_res_nrg, 16 );
158 *LTPredCodGain_Q7 = ( opus_int )silk_SMULBB( 3, silk_lin2log( div_Q16 ) - ( 16 << 7 ) );
159
160 silk_assert( *LTPredCodGain_Q7 == ( opus_int )silk_SAT16( silk_MUL( 3, silk_lin2log( div_Q16 ) - ( 16 << 7 ) ) ) );
161 }
162
163 /* smoothing */
164 /* d = sum( B, 1 ); */
165 b_Q14_ptr = b_Q14;
166 for( k = 0; k < nb_subfr; k++ ) {
167 d_Q14[ k ] = 0;
168 for( i = 0; i < LTP_ORDER; i++ ) {
169 d_Q14[ k ] += b_Q14_ptr[ i ];
170 }
171 b_Q14_ptr += LTP_ORDER;
172 }
173
174 /* m = ( w * d' ) / ( sum( w ) + 1e-3 ); */
175
176 /* Find maximum absolute value of d_Q14 and the bits used by w in Q0 */
177 max_abs_d_Q14 = 0;
178 max_w_bits = 0;
179 for( k = 0; k < nb_subfr; k++ ) {
180 max_abs_d_Q14 = silk_max_32( max_abs_d_Q14, silk_abs( d_Q14[ k ] ) );
181 /* w[ k ] is in Q( 18 - corr_rshifts[ k ] ) */
182 /* Find bits needed in Q( 18 - maxRshifts ) */
183 max_w_bits = silk_max_32( max_w_bits, 32 - silk_CLZ32( w[ k ] ) + corr_rshifts[ k ] - maxRshifts );
184 }
185
186 /* max_abs_d_Q14 = (5 << 15); worst case, i.e. LTP_ORDER * -silk_int16_MIN */
187 silk_assert( max_abs_d_Q14 <= ( 5 << 15 ) );
188
189 /* How many bits is needed for w*d' in Q( 18 - maxRshifts ) in the worst case, of all d_Q14's being equal to max_abs_d_Q14 */
190 extra_shifts = max_w_bits + 32 - silk_CLZ32( max_abs_d_Q14 ) - 14;
191
192 /* Subtract what we got available; bits in output var plus maxRshifts */
193 extra_shifts -= ( 32 - 1 - 2 + maxRshifts ); /* Keep sign bit free as well as 2 bits for accumulation */
194 extra_shifts = silk_max_int( extra_shifts, 0 );
195
196 maxRshifts_wxtra = maxRshifts + extra_shifts;
197
198 temp32 = silk_RSHIFT( 262, maxRshifts + extra_shifts ) + 1; /* 1e-3f in Q( 18 - (maxRshifts + extra_shifts) ) */
199 wd = 0;
200 for( k = 0; k < nb_subfr; k++ ) {
201 /* w has at least 2 bits of headroom so no overflow should happen */
202 temp32 = silk_ADD32( temp32, silk_RSHIFT( w[ k ], maxRshifts_wxtra - corr_rshifts[ k ] ) ); /* Q( 18 - maxRshifts_wxtra ) */
203 wd = silk_ADD32( wd, silk_LSHIFT( silk_SMULWW( silk_RSHIFT( w[ k ], maxRshifts_wxtra - corr_rshifts[ k ] ), d_Q14[ k ] ), 2 ) ); /* Q( 18 - maxRshifts_wxtra ) */
204 }
205 m_Q12 = silk_DIV32_varQ( wd, temp32, 12 );
206
207 b_Q14_ptr = b_Q14;
208 for( k = 0; k < nb_subfr; k++ ) {
209 /* w_fix[ k ] from Q( 18 - corr_rshifts[ k ] ) to Q( 16 ) */
210 if( 2 - corr_rshifts[k] > 0 ) {
211 temp32 = silk_RSHIFT( w[ k ], 2 - corr_rshifts[ k ] );
212 } else {
213 temp32 = silk_LSHIFT_SAT32( w[ k ], corr_rshifts[ k ] - 2 );
214 }
215
216 g_Q26 = silk_MUL(
217 silk_DIV32(
218 SILK_FIX_CONST( LTP_SMOOTHING, 26 ),
219 silk_RSHIFT( SILK_FIX_CONST( LTP_SMOOTHING, 26 ), 10 ) + temp32 ), /* Q10 */
220 silk_LSHIFT_SAT32( silk_SUB_SAT32( (opus_int32)m_Q12, silk_RSHIFT( d_Q14[ k ], 2 ) ), 4 ) ); /* Q16 */
221
222 temp32 = 0;
223 for( i = 0; i < LTP_ORDER; i++ ) {
224 delta_b_Q14[ i ] = silk_max_16( b_Q14_ptr[ i ], 1638 ); /* 1638_Q14 = 0.1_Q0 */
225 temp32 += delta_b_Q14[ i ]; /* Q14 */
226 }
227 temp32 = silk_DIV32( g_Q26, temp32 ); /* Q14 -> Q12 */
228 for( i = 0; i < LTP_ORDER; i++ ) {
229 b_Q14_ptr[ i ] = silk_LIMIT_32( (opus_int32)b_Q14_ptr[ i ] + silk_SMULWB( silk_LSHIFT_SAT32( temp32, 4 ), delta_b_Q14[ i ] ), -16000, 28000 );
230 }
231 b_Q14_ptr += LTP_ORDER;
232 }
233 }
234
silk_fit_LTP(opus_int32 LTP_coefs_Q16[LTP_ORDER],opus_int16 LTP_coefs_Q14[LTP_ORDER])235 void silk_fit_LTP(
236 opus_int32 LTP_coefs_Q16[ LTP_ORDER ],
237 opus_int16 LTP_coefs_Q14[ LTP_ORDER ]
238 )
239 {
240 opus_int i;
241
242 for( i = 0; i < LTP_ORDER; i++ ) {
243 LTP_coefs_Q14[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( LTP_coefs_Q16[ i ], 2 ) );
244 }
245 }
246