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.h"
33
34 /* Entropy constrained matrix-weighted VQ, hard-coded to 5-element vectors, for a single input data vector */
silk_VQ_WMat_EC_c(opus_int8 * ind,opus_int32 * res_nrg_Q15,opus_int32 * rate_dist_Q8,opus_int * gain_Q7,const opus_int32 * XX_Q17,const opus_int32 * xX_Q17,const opus_int8 * cb_Q7,const opus_uint8 * cb_gain_Q7,const opus_uint8 * cl_Q5,const opus_int subfr_len,const opus_int32 max_gain_Q7,const opus_int L)35 void silk_VQ_WMat_EC_c(
36 opus_int8 *ind, /* O index of best codebook vector */
37 opus_int32 *res_nrg_Q15, /* O best residual energy */
38 opus_int32 *rate_dist_Q8, /* O best total bitrate */
39 opus_int *gain_Q7, /* O sum of absolute LTP coefficients */
40 const opus_int32 *XX_Q17, /* I correlation matrix */
41 const opus_int32 *xX_Q17, /* I correlation vector */
42 const opus_int8 *cb_Q7, /* I codebook */
43 const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */
44 const opus_uint8 *cl_Q5, /* I code length for each codebook vector */
45 const opus_int subfr_len, /* I number of samples per subframe */
46 const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */
47 const opus_int L /* I number of vectors in codebook */
48 )
49 {
50 opus_int k, gain_tmp_Q7;
51 const opus_int8 *cb_row_Q7;
52 opus_int32 neg_xX_Q24[ 5 ];
53 opus_int32 sum1_Q15, sum2_Q24;
54 opus_int32 bits_res_Q8, bits_tot_Q8;
55
56 /* Negate and convert to new Q domain */
57 neg_xX_Q24[ 0 ] = -silk_LSHIFT32( xX_Q17[ 0 ], 7 );
58 neg_xX_Q24[ 1 ] = -silk_LSHIFT32( xX_Q17[ 1 ], 7 );
59 neg_xX_Q24[ 2 ] = -silk_LSHIFT32( xX_Q17[ 2 ], 7 );
60 neg_xX_Q24[ 3 ] = -silk_LSHIFT32( xX_Q17[ 3 ], 7 );
61 neg_xX_Q24[ 4 ] = -silk_LSHIFT32( xX_Q17[ 4 ], 7 );
62
63 /* Loop over codebook */
64 *rate_dist_Q8 = silk_int32_MAX;
65 *res_nrg_Q15 = silk_int32_MAX;
66 cb_row_Q7 = cb_Q7;
67 /* If things go really bad, at least *ind is set to something safe. */
68 *ind = 0;
69 for( k = 0; k < L; k++ ) {
70 opus_int32 penalty;
71 gain_tmp_Q7 = cb_gain_Q7[k];
72 /* Weighted rate */
73 /* Quantization error: 1 - 2 * xX * cb + cb' * XX * cb */
74 sum1_Q15 = SILK_FIX_CONST( 1.001, 15 );
75
76 /* Penalty for too large gain */
77 penalty = silk_LSHIFT32( silk_max( silk_SUB32( gain_tmp_Q7, max_gain_Q7 ), 0 ), 11 );
78
79 /* first row of XX_Q17 */
80 sum2_Q24 = silk_MLA( neg_xX_Q24[ 0 ], XX_Q17[ 1 ], cb_row_Q7[ 1 ] );
81 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 2 ], cb_row_Q7[ 2 ] );
82 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 3 ], cb_row_Q7[ 3 ] );
83 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 4 ], cb_row_Q7[ 4 ] );
84 sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 );
85 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 0 ], cb_row_Q7[ 0 ] );
86 sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 0 ] );
87
88 /* second row of XX_Q17 */
89 sum2_Q24 = silk_MLA( neg_xX_Q24[ 1 ], XX_Q17[ 7 ], cb_row_Q7[ 2 ] );
90 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 8 ], cb_row_Q7[ 3 ] );
91 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 9 ], cb_row_Q7[ 4 ] );
92 sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 );
93 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 6 ], cb_row_Q7[ 1 ] );
94 sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 1 ] );
95
96 /* third row of XX_Q17 */
97 sum2_Q24 = silk_MLA( neg_xX_Q24[ 2 ], XX_Q17[ 13 ], cb_row_Q7[ 3 ] );
98 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 14 ], cb_row_Q7[ 4 ] );
99 sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 );
100 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 12 ], cb_row_Q7[ 2 ] );
101 sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 2 ] );
102
103 /* fourth row of XX_Q17 */
104 sum2_Q24 = silk_MLA( neg_xX_Q24[ 3 ], XX_Q17[ 19 ], cb_row_Q7[ 4 ] );
105 sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 );
106 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 18 ], cb_row_Q7[ 3 ] );
107 sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 3 ] );
108
109 /* last row of XX_Q17 */
110 sum2_Q24 = silk_LSHIFT32( neg_xX_Q24[ 4 ], 1 );
111 sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 24 ], cb_row_Q7[ 4 ] );
112 sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 4 ] );
113
114 /* find best */
115 if( sum1_Q15 >= 0 ) {
116 /* Translate residual energy to bits using high-rate assumption (6 dB ==> 1 bit/sample) */
117 bits_res_Q8 = silk_SMULBB( subfr_len, silk_lin2log( sum1_Q15 + penalty) - (15 << 7) );
118 /* In the following line we reduce the codelength component by half ("-1"); seems to slightly improve quality */
119 bits_tot_Q8 = silk_ADD_LSHIFT32( bits_res_Q8, cl_Q5[ k ], 3-1 );
120 if( bits_tot_Q8 <= *rate_dist_Q8 ) {
121 *rate_dist_Q8 = bits_tot_Q8;
122 *res_nrg_Q15 = sum1_Q15 + penalty;
123 *ind = (opus_int8)k;
124 *gain_Q7 = gain_tmp_Q7;
125 }
126 }
127
128 /* Go to next cbk vector */
129 cb_row_Q7 += LTP_ORDER;
130 }
131 }
132