Alexandre Lision | 744f742 | 2013-09-25 11:39:37 -0400 | [diff] [blame] | 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 | /*****************************/ |
| 36 | /* Internal function headers */ |
| 37 | /*****************************/ |
| 38 | |
| 39 | typedef struct { |
| 40 | opus_int32 Q36_part; |
| 41 | opus_int32 Q48_part; |
| 42 | } inv_D_t; |
| 43 | |
| 44 | /* Factorize square matrix A into LDL form */ |
| 45 | static inline void silk_LDL_factorize_FIX( |
| 46 | opus_int32 *A, /* I/O Pointer to Symetric Square Matrix */ |
| 47 | opus_int M, /* I Size of Matrix */ |
| 48 | opus_int32 *L_Q16, /* I/O Pointer to Square Upper triangular Matrix */ |
| 49 | inv_D_t *inv_D /* I/O Pointer to vector holding inverted diagonal elements of D */ |
| 50 | ); |
| 51 | |
| 52 | /* Solve Lx = b, when L is lower triangular and has ones on the diagonal */ |
| 53 | static inline void silk_LS_SolveFirst_FIX( |
| 54 | const opus_int32 *L_Q16, /* I Pointer to Lower Triangular Matrix */ |
| 55 | opus_int M, /* I Dim of Matrix equation */ |
| 56 | const opus_int32 *b, /* I b Vector */ |
| 57 | opus_int32 *x_Q16 /* O x Vector */ |
| 58 | ); |
| 59 | |
| 60 | /* Solve L^t*x = b, where L is lower triangular with ones on the diagonal */ |
| 61 | static inline void silk_LS_SolveLast_FIX( |
| 62 | const opus_int32 *L_Q16, /* I Pointer to Lower Triangular Matrix */ |
| 63 | const opus_int M, /* I Dim of Matrix equation */ |
| 64 | const opus_int32 *b, /* I b Vector */ |
| 65 | opus_int32 *x_Q16 /* O x Vector */ |
| 66 | ); |
| 67 | |
| 68 | static inline void silk_LS_divide_Q16_FIX( |
| 69 | opus_int32 T[], /* I/O Numenator vector */ |
| 70 | inv_D_t *inv_D, /* I 1 / D vector */ |
| 71 | opus_int M /* I dimension */ |
| 72 | ); |
| 73 | |
| 74 | /* Solves Ax = b, assuming A is symmetric */ |
| 75 | void silk_solve_LDL_FIX( |
| 76 | opus_int32 *A, /* I Pointer to symetric square matrix A */ |
| 77 | opus_int M, /* I Size of matrix */ |
| 78 | const opus_int32 *b, /* I Pointer to b vector */ |
| 79 | opus_int32 *x_Q16 /* O Pointer to x solution vector */ |
| 80 | ) |
| 81 | { |
| 82 | opus_int32 L_Q16[ MAX_MATRIX_SIZE * MAX_MATRIX_SIZE ]; |
| 83 | opus_int32 Y[ MAX_MATRIX_SIZE ]; |
| 84 | inv_D_t inv_D[ MAX_MATRIX_SIZE ]; |
| 85 | |
| 86 | silk_assert( M <= MAX_MATRIX_SIZE ); |
| 87 | |
| 88 | /*************************************************** |
| 89 | Factorize A by LDL such that A = L*D*L', |
| 90 | where L is lower triangular with ones on diagonal |
| 91 | ****************************************************/ |
| 92 | silk_LDL_factorize_FIX( A, M, L_Q16, inv_D ); |
| 93 | |
| 94 | /**************************************************** |
| 95 | * substitute D*L'*x = Y. ie: |
| 96 | L*D*L'*x = b => L*Y = b <=> Y = inv(L)*b |
| 97 | ******************************************************/ |
| 98 | silk_LS_SolveFirst_FIX( L_Q16, M, b, Y ); |
| 99 | |
| 100 | /**************************************************** |
| 101 | D*L'*x = Y <=> L'*x = inv(D)*Y, because D is |
| 102 | diagonal just multiply with 1/d_i |
| 103 | ****************************************************/ |
| 104 | silk_LS_divide_Q16_FIX( Y, inv_D, M ); |
| 105 | |
| 106 | /**************************************************** |
| 107 | x = inv(L') * inv(D) * Y |
| 108 | *****************************************************/ |
| 109 | silk_LS_SolveLast_FIX( L_Q16, M, Y, x_Q16 ); |
| 110 | } |
| 111 | |
| 112 | static inline void silk_LDL_factorize_FIX( |
| 113 | opus_int32 *A, /* I/O Pointer to Symetric Square Matrix */ |
| 114 | opus_int M, /* I Size of Matrix */ |
| 115 | opus_int32 *L_Q16, /* I/O Pointer to Square Upper triangular Matrix */ |
| 116 | inv_D_t *inv_D /* I/O Pointer to vector holding inverted diagonal elements of D */ |
| 117 | ) |
| 118 | { |
| 119 | opus_int i, j, k, status, loop_count; |
| 120 | const opus_int32 *ptr1, *ptr2; |
| 121 | opus_int32 diag_min_value, tmp_32, err; |
| 122 | opus_int32 v_Q0[ MAX_MATRIX_SIZE ], D_Q0[ MAX_MATRIX_SIZE ]; |
| 123 | opus_int32 one_div_diag_Q36, one_div_diag_Q40, one_div_diag_Q48; |
| 124 | |
| 125 | silk_assert( M <= MAX_MATRIX_SIZE ); |
| 126 | |
| 127 | status = 1; |
| 128 | diag_min_value = silk_max_32( silk_SMMUL( silk_ADD_SAT32( A[ 0 ], A[ silk_SMULBB( M, M ) - 1 ] ), SILK_FIX_CONST( FIND_LTP_COND_FAC, 31 ) ), 1 << 9 ); |
| 129 | for( loop_count = 0; loop_count < M && status == 1; loop_count++ ) { |
| 130 | status = 0; |
| 131 | for( j = 0; j < M; j++ ) { |
| 132 | ptr1 = matrix_adr( L_Q16, j, 0, M ); |
| 133 | tmp_32 = 0; |
| 134 | for( i = 0; i < j; i++ ) { |
| 135 | v_Q0[ i ] = silk_SMULWW( D_Q0[ i ], ptr1[ i ] ); /* Q0 */ |
| 136 | tmp_32 = silk_SMLAWW( tmp_32, v_Q0[ i ], ptr1[ i ] ); /* Q0 */ |
| 137 | } |
| 138 | tmp_32 = silk_SUB32( matrix_ptr( A, j, j, M ), tmp_32 ); |
| 139 | |
| 140 | if( tmp_32 < diag_min_value ) { |
| 141 | tmp_32 = silk_SUB32( silk_SMULBB( loop_count + 1, diag_min_value ), tmp_32 ); |
| 142 | /* Matrix not positive semi-definite, or ill conditioned */ |
| 143 | for( i = 0; i < M; i++ ) { |
| 144 | matrix_ptr( A, i, i, M ) = silk_ADD32( matrix_ptr( A, i, i, M ), tmp_32 ); |
| 145 | } |
| 146 | status = 1; |
| 147 | break; |
| 148 | } |
| 149 | D_Q0[ j ] = tmp_32; /* always < max(Correlation) */ |
| 150 | |
| 151 | /* two-step division */ |
| 152 | one_div_diag_Q36 = silk_INVERSE32_varQ( tmp_32, 36 ); /* Q36 */ |
| 153 | one_div_diag_Q40 = silk_LSHIFT( one_div_diag_Q36, 4 ); /* Q40 */ |
| 154 | err = silk_SUB32( (opus_int32)1 << 24, silk_SMULWW( tmp_32, one_div_diag_Q40 ) ); /* Q24 */ |
| 155 | one_div_diag_Q48 = silk_SMULWW( err, one_div_diag_Q40 ); /* Q48 */ |
| 156 | |
| 157 | /* Save 1/Ds */ |
| 158 | inv_D[ j ].Q36_part = one_div_diag_Q36; |
| 159 | inv_D[ j ].Q48_part = one_div_diag_Q48; |
| 160 | |
| 161 | matrix_ptr( L_Q16, j, j, M ) = 65536; /* 1.0 in Q16 */ |
| 162 | ptr1 = matrix_adr( A, j, 0, M ); |
| 163 | ptr2 = matrix_adr( L_Q16, j + 1, 0, M ); |
| 164 | for( i = j + 1; i < M; i++ ) { |
| 165 | tmp_32 = 0; |
| 166 | for( k = 0; k < j; k++ ) { |
| 167 | tmp_32 = silk_SMLAWW( tmp_32, v_Q0[ k ], ptr2[ k ] ); /* Q0 */ |
| 168 | } |
| 169 | tmp_32 = silk_SUB32( ptr1[ i ], tmp_32 ); /* always < max(Correlation) */ |
| 170 | |
| 171 | /* tmp_32 / D_Q0[j] : Divide to Q16 */ |
| 172 | matrix_ptr( L_Q16, i, j, M ) = silk_ADD32( silk_SMMUL( tmp_32, one_div_diag_Q48 ), |
| 173 | silk_RSHIFT( silk_SMULWW( tmp_32, one_div_diag_Q36 ), 4 ) ); |
| 174 | |
| 175 | /* go to next column */ |
| 176 | ptr2 += M; |
| 177 | } |
| 178 | } |
| 179 | } |
| 180 | |
| 181 | silk_assert( status == 0 ); |
| 182 | } |
| 183 | |
| 184 | static inline void silk_LS_divide_Q16_FIX( |
| 185 | opus_int32 T[], /* I/O Numenator vector */ |
| 186 | inv_D_t *inv_D, /* I 1 / D vector */ |
| 187 | opus_int M /* I dimension */ |
| 188 | ) |
| 189 | { |
| 190 | opus_int i; |
| 191 | opus_int32 tmp_32; |
| 192 | opus_int32 one_div_diag_Q36, one_div_diag_Q48; |
| 193 | |
| 194 | for( i = 0; i < M; i++ ) { |
| 195 | one_div_diag_Q36 = inv_D[ i ].Q36_part; |
| 196 | one_div_diag_Q48 = inv_D[ i ].Q48_part; |
| 197 | |
| 198 | tmp_32 = T[ i ]; |
| 199 | T[ i ] = silk_ADD32( silk_SMMUL( tmp_32, one_div_diag_Q48 ), silk_RSHIFT( silk_SMULWW( tmp_32, one_div_diag_Q36 ), 4 ) ); |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | /* Solve Lx = b, when L is lower triangular and has ones on the diagonal */ |
| 204 | static inline void silk_LS_SolveFirst_FIX( |
| 205 | const opus_int32 *L_Q16, /* I Pointer to Lower Triangular Matrix */ |
| 206 | opus_int M, /* I Dim of Matrix equation */ |
| 207 | const opus_int32 *b, /* I b Vector */ |
| 208 | opus_int32 *x_Q16 /* O x Vector */ |
| 209 | ) |
| 210 | { |
| 211 | opus_int i, j; |
| 212 | const opus_int32 *ptr32; |
| 213 | opus_int32 tmp_32; |
| 214 | |
| 215 | for( i = 0; i < M; i++ ) { |
| 216 | ptr32 = matrix_adr( L_Q16, i, 0, M ); |
| 217 | tmp_32 = 0; |
| 218 | for( j = 0; j < i; j++ ) { |
| 219 | tmp_32 = silk_SMLAWW( tmp_32, ptr32[ j ], x_Q16[ j ] ); |
| 220 | } |
| 221 | x_Q16[ i ] = silk_SUB32( b[ i ], tmp_32 ); |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | /* Solve L^t*x = b, where L is lower triangular with ones on the diagonal */ |
| 226 | static inline void silk_LS_SolveLast_FIX( |
| 227 | const opus_int32 *L_Q16, /* I Pointer to Lower Triangular Matrix */ |
| 228 | const opus_int M, /* I Dim of Matrix equation */ |
| 229 | const opus_int32 *b, /* I b Vector */ |
| 230 | opus_int32 *x_Q16 /* O x Vector */ |
| 231 | ) |
| 232 | { |
| 233 | opus_int i, j; |
| 234 | const opus_int32 *ptr32; |
| 235 | opus_int32 tmp_32; |
| 236 | |
| 237 | for( i = M - 1; i >= 0; i-- ) { |
| 238 | ptr32 = matrix_adr( L_Q16, 0, i, M ); |
| 239 | tmp_32 = 0; |
| 240 | for( j = M - 1; j > i; j-- ) { |
| 241 | tmp_32 = silk_SMLAWW( tmp_32, ptr32[ silk_SMULBB( j, M ) ], x_Q16[ j ] ); |
| 242 | } |
| 243 | x_Q16[ i ] = silk_SUB32( b[ i ], tmp_32 ); |
| 244 | } |
| 245 | } |