1 /* 2 * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 #include "webrtc/modules/audio_coding/codecs/isac/fix/source/settings.h" 12 #include "webrtc/typedefs.h" 13 14 // Filter ar_g_Q0[] and ar_f_Q0[] through an AR filter with coefficients 15 // cth_Q15[] and sth_Q15[]. 16 void WebRtcIsacfix_FilterArLoop(int16_t* ar_g_Q0, // Input samples 17 int16_t* ar_f_Q0, // Input samples 18 int16_t* cth_Q15, // Filter coefficients 19 int16_t* sth_Q15, // Filter coefficients 20 int16_t order_coef) { // order of the filter 21 int n = 0; 22 23 for (n = 0; n < HALF_SUBFRAMELEN - 1; n++) { 24 int count = order_coef - 1; 25 int offset; 26 #if !defined(MIPS_DSP_R1_LE) 27 int16_t* tmp_cth; 28 int16_t* tmp_sth; 29 int16_t* tmp_arg; 30 int32_t max_q16 = 0x7fff; 31 int32_t min_q16 = 0xffff8000; 32 #endif 33 // Declare variables used as temporary registers. 34 int32_t r0, r1, r2, t0, t1, t2, t_ar; 35 36 __asm __volatile ( 37 ".set push \n\t" 38 ".set noreorder \n\t" 39 "bltz %[count], 2f \n\t" 40 " lh %[t_ar], 0(%[tmp]) \n\t" 41 // Inner loop 42 "1: \n\t" 43 "sll %[offset], %[count], 1 \n\t" 44 #if defined(MIPS_DSP_R1_LE) 45 "lhx %[r0], %[offset](%[cth_Q15]) \n\t" 46 "lhx %[r1], %[offset](%[sth_Q15]) \n\t" 47 "lhx %[r2], %[offset](%[ar_g_Q0]) \n\t" 48 #else 49 "addu %[tmp_cth], %[cth_Q15], %[offset] \n\t" 50 "addu %[tmp_sth], %[sth_Q15], %[offset] \n\t" 51 "addu %[tmp_arg], %[ar_g_Q0], %[offset] \n\t" 52 "lh %[r0], 0(%[tmp_cth]) \n\t" 53 "lh %[r1], 0(%[tmp_sth]) \n\t" 54 "lh %[r2], 0(%[tmp_arg]) \n\t" 55 #endif 56 "mul %[t0], %[r0], %[t_ar] \n\t" 57 "mul %[t1], %[r1], %[t_ar] \n\t" 58 "mul %[t2], %[r1], %[r2] \n\t" 59 "mul %[r0], %[r0], %[r2] \n\t" 60 "subu %[t0], %[t0], %[t2] \n\t" 61 "addu %[t1], %[t1], %[r0] \n\t" 62 #if defined(MIPS_DSP_R1_LE) 63 "shra_r.w %[t1], %[t1], 15 \n\t" 64 "shra_r.w %[t0], %[t0], 15 \n\t" 65 #else 66 "addiu %[t1], %[t1], 0x4000 \n\t" 67 "sra %[t1], %[t1], 15 \n\t" 68 "addiu %[t0], %[t0], 0x4000 \n\t" 69 "sra %[t0], %[t0], 15 \n\t" 70 #endif 71 "addiu %[offset], %[offset], 2 \n\t" 72 #if defined(MIPS_DSP_R1_LE) 73 "shll_s.w %[t1], %[t1], 16 \n\t" 74 "shll_s.w %[t_ar], %[t0], 16 \n\t" 75 #else 76 "slt %[r0], %[t1], %[max_q16] \n\t" 77 "slt %[r1], %[t0], %[max_q16] \n\t" 78 "movz %[t1], %[max_q16], %[r0] \n\t" 79 "movz %[t0], %[max_q16], %[r1] \n\t" 80 #endif 81 "addu %[offset], %[offset], %[ar_g_Q0] \n\t" 82 #if defined(MIPS_DSP_R1_LE) 83 "sra %[t1], %[t1], 16 \n\t" 84 "sra %[t_ar], %[t_ar], 16 \n\t" 85 #else 86 "slt %[r0], %[t1], %[min_q16] \n\t" 87 "slt %[r1], %[t0], %[min_q16] \n\t" 88 "movn %[t1], %[min_q16], %[r0] \n\t" 89 "movn %[t0], %[min_q16], %[r1] \n\t" 90 "addu %[t_ar], $zero, %[t0] \n\t" 91 #endif 92 "sh %[t1], 0(%[offset]) \n\t" 93 "bgtz %[count], 1b \n\t" 94 " addiu %[count], %[count], -1 \n\t" 95 "2: \n\t" 96 "sh %[t_ar], 0(%[tmp]) \n\t" 97 "sh %[t_ar], 0(%[ar_g_Q0]) \n\t" 98 ".set pop \n\t" 99 : [t_ar] "=&r" (t_ar), [count] "+r" (count), [offset] "=&r" (offset), 100 [r0] "=&r" (r0), [r1] "=&r" (r1), [r2] "=&r" (r2), [t0] "=&r" (t0), 101 #if !defined(MIPS_DSP_R1_LE) 102 [tmp_cth] "=&r" (tmp_cth), [tmp_sth] "=&r" (tmp_sth), 103 [tmp_arg] "=&r" (tmp_arg), 104 #endif 105 [t1] "=&r" (t1), [t2] "=&r" (t2) 106 : [tmp] "r" (&ar_f_Q0[n+1]), [cth_Q15] "r" (cth_Q15), 107 #if !defined(MIPS_DSP_R1_LE) 108 [max_q16] "r" (max_q16), [min_q16] "r" (min_q16), 109 #endif 110 [sth_Q15] "r" (sth_Q15), [ar_g_Q0] "r" (ar_g_Q0) 111 : "memory", "hi", "lo" 112 ); 113 } 114 } 115 116 // MIPS optimization of the inner loop used for function 117 // WebRtcIsacfix_NormLatticeFilterMa(). It does: 118 // 119 // for 0 <= n < HALF_SUBFRAMELEN - 1: 120 // *ptr2 = input2 * (*ptr2) + input0 * (*ptr0)); 121 // *ptr1 = input1 * (*ptr0) + input0 * (*ptr2); 122 // 123 // Note, function WebRtcIsacfix_FilterMaLoopMIPS and WebRtcIsacfix_FilterMaLoopC 124 // are not bit-exact. The accuracy of the MIPS function is same or better. 125 void WebRtcIsacfix_FilterMaLoopMIPS(int16_t input0, // Filter coefficient 126 int16_t input1, // Filter coefficient 127 int32_t input2, // Inverse coeff (1/input1) 128 int32_t* ptr0, // Sample buffer 129 int32_t* ptr1, // Sample buffer 130 int32_t* ptr2) { // Sample buffer 131 #if defined(MIPS_DSP_R2_LE) 132 // MIPS DSPR2 version. 4 available accumulators allows loop unrolling 4 times. 133 // This variant is not bit-exact with WebRtcIsacfix_FilterMaLoopC, since we 134 // are exploiting 64-bit accumulators. The accuracy of the MIPS DSPR2 function 135 // is same or better. 136 int n = (HALF_SUBFRAMELEN - 1) >> 2; 137 int m = (HALF_SUBFRAMELEN - 1) & 3; 138 139 int r0, r1, r2, r3; 140 int t0, t1, t2, t3; 141 int s0, s1, s2, s3; 142 143 __asm __volatile ( 144 ".set push \n\t" 145 ".set noreorder \n\t" 146 "1: \n\t" 147 "lw %[r0], 0(%[ptr0]) \n\t" 148 "lw %[r1], 4(%[ptr0]) \n\t" 149 "lw %[r2], 8(%[ptr0]) \n\t" 150 "lw %[r3], 12(%[ptr0]) \n\t" 151 "mult $ac0, %[r0], %[input0] \n\t" 152 "mult $ac1, %[r1], %[input0] \n\t" 153 "mult $ac2, %[r2], %[input0] \n\t" 154 "mult $ac3, %[r3], %[input0] \n\t" 155 "lw %[t0], 0(%[ptr2]) \n\t" 156 "extr_rs.w %[s0], $ac0, 15 \n\t" 157 "extr_rs.w %[s1], $ac1, 15 \n\t" 158 "extr_rs.w %[s2], $ac2, 15 \n\t" 159 "extr_rs.w %[s3], $ac3, 15 \n\t" 160 "lw %[t1], 4(%[ptr2]) \n\t" 161 "lw %[t2], 8(%[ptr2]) \n\t" 162 "lw %[t3], 12(%[ptr2]) \n\t" 163 "addu %[t0], %[t0], %[s0] \n\t" 164 "addu %[t1], %[t1], %[s1] \n\t" 165 "addu %[t2], %[t2], %[s2] \n\t" 166 "addu %[t3], %[t3], %[s3] \n\t" 167 "mult $ac0, %[t0], %[input2] \n\t" 168 "mult $ac1, %[t1], %[input2] \n\t" 169 "mult $ac2, %[t2], %[input2] \n\t" 170 "mult $ac3, %[t3], %[input2] \n\t" 171 "addiu %[ptr0], %[ptr0], 16 \n\t" 172 "extr_rs.w %[t0], $ac0, 16 \n\t" 173 "extr_rs.w %[t1], $ac1, 16 \n\t" 174 "extr_rs.w %[t2], $ac2, 16 \n\t" 175 "extr_rs.w %[t3], $ac3, 16 \n\t" 176 "addiu %[n], %[n], -1 \n\t" 177 "mult $ac0, %[r0], %[input1] \n\t" 178 "mult $ac1, %[r1], %[input1] \n\t" 179 "mult $ac2, %[r2], %[input1] \n\t" 180 "mult $ac3, %[r3], %[input1] \n\t" 181 "sw %[t0], 0(%[ptr2]) \n\t" 182 "extr_rs.w %[s0], $ac0, 15 \n\t" 183 "extr_rs.w %[s1], $ac1, 15 \n\t" 184 "extr_rs.w %[s2], $ac2, 15 \n\t" 185 "extr_rs.w %[s3], $ac3, 15 \n\t" 186 "sw %[t1], 4(%[ptr2]) \n\t" 187 "sw %[t2], 8(%[ptr2]) \n\t" 188 "sw %[t3], 12(%[ptr2]) \n\t" 189 "mult $ac0, %[t0], %[input0] \n\t" 190 "mult $ac1, %[t1], %[input0] \n\t" 191 "mult $ac2, %[t2], %[input0] \n\t" 192 "mult $ac3, %[t3], %[input0] \n\t" 193 "addiu %[ptr2], %[ptr2], 16 \n\t" 194 "extr_rs.w %[t0], $ac0, 15 \n\t" 195 "extr_rs.w %[t1], $ac1, 15 \n\t" 196 "extr_rs.w %[t2], $ac2, 15 \n\t" 197 "extr_rs.w %[t3], $ac3, 15 \n\t" 198 "addu %[t0], %[t0], %[s0] \n\t" 199 "addu %[t1], %[t1], %[s1] \n\t" 200 "addu %[t2], %[t2], %[s2] \n\t" 201 "addu %[t3], %[t3], %[s3] \n\t" 202 "sw %[t0], 0(%[ptr1]) \n\t" 203 "sw %[t1], 4(%[ptr1]) \n\t" 204 "sw %[t2], 8(%[ptr1]) \n\t" 205 "sw %[t3], 12(%[ptr1]) \n\t" 206 "bgtz %[n], 1b \n\t" 207 " addiu %[ptr1], %[ptr1], 16 \n\t" 208 "beq %[m], %0, 3f \n\t" 209 " nop \n\t" 210 "2: \n\t" 211 "lw %[r0], 0(%[ptr0]) \n\t" 212 "lw %[t0], 0(%[ptr2]) \n\t" 213 "addiu %[ptr0], %[ptr0], 4 \n\t" 214 "mult $ac0, %[r0], %[input0] \n\t" 215 "mult $ac1, %[r0], %[input1] \n\t" 216 "extr_rs.w %[r1], $ac0, 15 \n\t" 217 "extr_rs.w %[t1], $ac1, 15 \n\t" 218 "addu %[t0], %[t0], %[r1] \n\t" 219 "mult $ac0, %[t0], %[input2] \n\t" 220 "extr_rs.w %[t0], $ac0, 16 \n\t" 221 "sw %[t0], 0(%[ptr2]) \n\t" 222 "mult $ac0, %[t0], %[input0] \n\t" 223 "addiu %[ptr2], %[ptr2], 4 \n\t" 224 "addiu %[m], %[m], -1 \n\t" 225 "extr_rs.w %[t0], $ac0, 15 \n\t" 226 "addu %[t0], %[t0], %[t1] \n\t" 227 "sw %[t0], 0(%[ptr1]) \n\t" 228 "bgtz %[m], 2b \n\t" 229 " addiu %[ptr1], %[ptr1], 4 \n\t" 230 "3: \n\t" 231 ".set pop \n\t" 232 : [r0] "=&r" (r0), [r1] "=&r" (r1), [r2] "=&r" (r2), 233 [r3] "=&r" (r3), [t0] "=&r" (t0), [t1] "=&r" (t1), 234 [t2] "=&r" (t2), [t3] "=&r" (t3), [s0] "=&r" (s0), 235 [s1] "=&r" (s1), [s2] "=&r" (s2), [s3] "=&r" (s3), 236 [ptr0] "+r" (ptr0), [ptr1] "+r" (ptr1), [m] "+r" (m), 237 [ptr2] "+r" (ptr2), [n] "+r" (n) 238 : [input0] "r" (input0), [input1] "r" (input1), 239 [input2] "r" (input2) 240 : "memory", "hi", "lo", "$ac1hi", "$ac1lo", "$ac2hi", 241 "$ac2lo", "$ac3hi", "$ac3lo" 242 ); 243 #else 244 // Non-DSPR2 version of the function. Avoiding the accumulator usage due to 245 // large latencies. This variant is bit-exact with C code. 246 int n = HALF_SUBFRAMELEN - 1; 247 int32_t t16a, t16b; 248 int32_t r0, r1, r2, r3, r4; 249 250 __asm __volatile ( 251 ".set push \n\t" 252 ".set noreorder \n\t" 253 "sra %[t16a], %[input2], 16 \n\t" 254 "andi %[t16b], %[input2], 0xFFFF \n\t" 255 #if defined(MIPS32R2_LE) 256 "seh %[t16b], %[t16b] \n\t" 257 "seh %[input0], %[input0] \n\t" 258 "seh %[input1], %[input1] \n\t" 259 #else 260 "sll %[t16b], %[t16b], 16 \n\t" 261 "sra %[t16b], %[t16b], 16 \n\t" 262 "sll %[input0], %[input0], 16 \n\t" 263 "sra %[input0], %[input0], 16 \n\t" 264 "sll %[input1], %[input1], 16 \n\t" 265 "sra %[input1], %[input1], 16 \n\t" 266 #endif 267 "addiu %[r0], %[t16a], 1 \n\t" 268 "slt %[r1], %[t16b], $zero \n\t" 269 "movn %[t16a], %[r0], %[r1] \n\t" 270 "1: \n\t" 271 "lw %[r0], 0(%[ptr0]) \n\t" 272 "lw %[r1], 0(%[ptr2]) \n\t" 273 "addiu %[ptr0], %[ptr0], 4 \n\t" 274 "sra %[r2], %[r0], 16 \n\t" 275 "andi %[r0], %[r0], 0xFFFF \n\t" 276 "mul %[r3], %[r2], %[input0] \n\t" 277 "mul %[r4], %[r0], %[input0] \n\t" 278 "mul %[r2], %[r2], %[input1] \n\t" 279 "mul %[r0], %[r0], %[input1] \n\t" 280 "addiu %[ptr2], %[ptr2], 4 \n\t" 281 "sll %[r3], %[r3], 1 \n\t" 282 "sra %[r4], %[r4], 1 \n\t" 283 "addiu %[r4], %[r4], 0x2000 \n\t" 284 "sra %[r4], %[r4], 14 \n\t" 285 "addu %[r3], %[r3], %[r4] \n\t" 286 "addu %[r1], %[r1], %[r3] \n\t" 287 "sra %[r3], %[r1], 16 \n\t" 288 "andi %[r4], %[r1], 0xFFFF \n\t" 289 "sra %[r4], %[r4], 1 \n\t" 290 "mul %[r1], %[r1], %[t16a] \n\t" 291 "mul %[r3], %[r3], %[t16b] \n\t" 292 "mul %[r4], %[r4], %[t16b] \n\t" 293 "sll %[r2], %[r2], 1 \n\t" 294 "sra %[r0], %[r0], 1 \n\t" 295 "addiu %[r0], %[r0], 0x2000 \n\t" 296 "sra %[r0], %[r0], 14 \n\t" 297 "addu %[r0], %[r0], %[r2] \n\t" 298 "addiu %[n], %[n], -1 \n\t" 299 "addu %[r1], %[r1], %[r3] \n\t" 300 "addiu %[r4], %[r4], 0x4000 \n\t" 301 "sra %[r4], %[r4], 15 \n\t" 302 "addu %[r1], %[r1], %[r4] \n\t" 303 "sra %[r2], %[r1], 16 \n\t" 304 "andi %[r3], %[r1], 0xFFFF \n\t" 305 "mul %[r3], %[r3], %[input0] \n\t" 306 "mul %[r2], %[r2], %[input0] \n\t" 307 "sw %[r1], -4(%[ptr2]) \n\t" 308 "sra %[r3], %[r3], 1 \n\t" 309 "addiu %[r3], %[r3], 0x2000 \n\t" 310 "sra %[r3], %[r3], 14 \n\t" 311 "addu %[r0], %[r0], %[r3] \n\t" 312 "sll %[r2], %[r2], 1 \n\t" 313 "addu %[r0], %[r0], %[r2] \n\t" 314 "sw %[r0], 0(%[ptr1]) \n\t" 315 "bgtz %[n], 1b \n\t" 316 " addiu %[ptr1], %[ptr1], 4 \n\t" 317 ".set pop \n\t" 318 : [t16a] "=&r" (t16a), [t16b] "=&r" (t16b), [r0] "=&r" (r0), 319 [r1] "=&r" (r1), [r2] "=&r" (r2), [r3] "=&r" (r3), 320 [r4] "=&r" (r4), [ptr0] "+r" (ptr0), [ptr1] "+r" (ptr1), 321 [ptr2] "+r" (ptr2), [n] "+r" (n) 322 : [input0] "r" (input0), [input1] "r" (input1), 323 [input2] "r" (input2) 324 : "hi", "lo", "memory" 325 ); 326 #endif 327 } 328