| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* -*- c++ -*- */ | ||
| 2 | /* | ||
| 3 | * Copyright 2014 Free Software Foundation, Inc. | ||
| 4 | * | ||
| 5 | * This file is part of VOLK | ||
| 6 | * | ||
| 7 | * SPDX-License-Identifier: LGPL-3.0-or-later | ||
| 8 | */ | ||
| 9 | |||
| 10 | #ifndef INCLUDED_volk_8u_conv_k7_r2puppet_8u_H | ||
| 11 | #define INCLUDED_volk_8u_conv_k7_r2puppet_8u_H | ||
| 12 | |||
| 13 | #include <string.h> | ||
| 14 | #include <volk/volk.h> | ||
| 15 | #include <volk/volk_8u_x4_conv_k7_r2_8u.h> | ||
| 16 | |||
| 17 | typedef union { | ||
| 18 | // decision_t is a BIT vector | ||
| 19 | unsigned char* t; | ||
| 20 | unsigned int* w; | ||
| 21 | } p_decision_t; | ||
| 22 | |||
| 23 | 256 | static inline int parity(int x, unsigned char* Partab) | |
| 24 | { | ||
| 25 | 256 | x ^= (x >> 16); | |
| 26 | 256 | x ^= (x >> 8); | |
| 27 | 256 | return Partab[x]; | |
| 28 | } | ||
| 29 | |||
| 30 | 4 | static inline int chainback_viterbi(unsigned char* data, | |
| 31 | unsigned int nbits, | ||
| 32 | unsigned int endstate, | ||
| 33 | unsigned int tailsize, | ||
| 34 | unsigned char* decisions) | ||
| 35 | { | ||
| 36 | unsigned char* d; | ||
| 37 | 4 | int d_ADDSHIFT = 0; | |
| 38 | 4 | int d_numstates = (1 << 6); | |
| 39 | 4 | int d_decision_t_size = d_numstates / 8; | |
| 40 | 4 | unsigned int d_k = 7; | |
| 41 | 4 | int d_framebits = nbits; | |
| 42 | /* ADDSHIFT and SUBSHIFT make sure that the thing returned is a byte. */ | ||
| 43 | 4 | d = decisions; | |
| 44 | /* Make room beyond the end of the encoder register so we can | ||
| 45 | * accumulate a full byte of decoded data | ||
| 46 | */ | ||
| 47 | |||
| 48 | 4 | endstate = (endstate % d_numstates) << d_ADDSHIFT; | |
| 49 | |||
| 50 | /* The store into data[] only needs to be done every 8 bits. | ||
| 51 | * But this avoids a conditional branch, and the writes will | ||
| 52 | * combine in the cache anyway | ||
| 53 | */ | ||
| 54 | |||
| 55 | 4 | d += tailsize * d_decision_t_size; /* Look past tail */ | |
| 56 | int retval; | ||
| 57 | 4 | int dif = tailsize - (d_k - 1); | |
| 58 | // printf("break, %d, %d\n", dif, (nbits+dif)%d_framebits); | ||
| 59 | p_decision_t dec; | ||
| 60 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 4 times.
|
28 | while (nbits-- > d_framebits - (d_k - 1)) { |
| 61 | int k; | ||
| 62 | 24 | dec.t = &d[nbits * d_decision_t_size]; | |
| 63 | 24 | k = (dec.w[(endstate >> d_ADDSHIFT) / 32] >> ((endstate >> d_ADDSHIFT) % 32)) & 1; | |
| 64 | |||
| 65 | 24 | endstate = (endstate >> 1) | (k << (d_k - 2 + d_ADDSHIFT)); | |
| 66 | // data[((nbits+dif)%nbits)>>3] = endstate>>d_SUBSHIFT; | ||
| 67 | // printf("%d, %d\n", k, (nbits+dif)%d_framebits); | ||
| 68 | 24 | data[((nbits + dif) % d_framebits)] = k; | |
| 69 | |||
| 70 | 24 | retval = endstate; | |
| 71 | } | ||
| 72 | 4 | nbits += 1; | |
| 73 | |||
| 74 |
2/2✓ Branch 0 taken 262092 times.
✓ Branch 1 taken 4 times.
|
262096 | while (nbits-- != 0) { |
| 75 | int k; | ||
| 76 | |||
| 77 | 262092 | dec.t = &d[nbits * d_decision_t_size]; | |
| 78 | |||
| 79 | 262092 | k = (dec.w[(endstate >> d_ADDSHIFT) / 32] >> ((endstate >> d_ADDSHIFT) % 32)) & 1; | |
| 80 | |||
| 81 | 262092 | endstate = (endstate >> 1) | (k << (d_k - 2 + d_ADDSHIFT)); | |
| 82 | 262092 | data[((nbits + dif) % d_framebits)] = k; | |
| 83 | } | ||
| 84 | // printf("%d, %d, %d, %d, %d, %d, %d, %d\n", | ||
| 85 | // data[4095],data[4094],data[4093],data[4092],data[4091],data[4090],data[4089],data[4088]); | ||
| 86 | |||
| 87 | |||
| 88 | 4 | return retval >> d_ADDSHIFT; | |
| 89 | } | ||
| 90 | |||
| 91 | |||
| 92 | #if LV_HAVE_SSE3 | ||
| 93 | |||
| 94 | #include <emmintrin.h> | ||
| 95 | #include <mmintrin.h> | ||
| 96 | #include <pmmintrin.h> | ||
| 97 | #include <stdio.h> | ||
| 98 | #include <xmmintrin.h> | ||
| 99 | |||
| 100 | 2 | static inline void volk_8u_conv_k7_r2puppet_8u_spiral(unsigned char* syms, | |
| 101 | unsigned char* dec, | ||
| 102 | unsigned int framebits) | ||
| 103 | { | ||
| 104 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 | if (framebits < 12) { |
| 105 | ✗ | return; | |
| 106 | } | ||
| 107 | |||
| 108 | static int once = 1; | ||
| 109 | 2 | int d_numstates = (1 << 6); | |
| 110 | 2 | int rate = 2; | |
| 111 | static unsigned char* D; | ||
| 112 | static unsigned char* Y; | ||
| 113 | static unsigned char* X; | ||
| 114 | static unsigned int excess = 6; | ||
| 115 | static unsigned char* Branchtab; | ||
| 116 | static unsigned char Partab[256]; | ||
| 117 | |||
| 118 | 2 | int d_polys[2] = { 79, 109 }; | |
| 119 | |||
| 120 | |||
| 121 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (once) { |
| 122 | |||
| 123 | 2 | X = (unsigned char*)volk_malloc(2 * d_numstates, volk_get_alignment()); | |
| 124 | 2 | Y = X + d_numstates; | |
| 125 | 2 | Branchtab = | |
| 126 | 2 | (unsigned char*)volk_malloc(d_numstates / 2 * rate, volk_get_alignment()); | |
| 127 | 2 | D = (unsigned char*)volk_malloc((d_numstates / 8) * (framebits + 6), | |
| 128 | volk_get_alignment()); | ||
| 129 | int state, i; | ||
| 130 | int cnt, ti; | ||
| 131 | |||
| 132 | /* Initialize parity lookup table */ | ||
| 133 |
2/2✓ Branch 0 taken 512 times.
✓ Branch 1 taken 2 times.
|
514 | for (i = 0; i < 256; i++) { |
| 134 | 512 | cnt = 0; | |
| 135 | 512 | ti = i; | |
| 136 |
2/2✓ Branch 0 taken 3586 times.
✓ Branch 1 taken 512 times.
|
4098 | while (ti) { |
| 137 |
2/2✓ Branch 0 taken 2048 times.
✓ Branch 1 taken 1538 times.
|
3586 | if (ti & 1) |
| 138 | 2048 | cnt++; | |
| 139 | 3586 | ti >>= 1; | |
| 140 | } | ||
| 141 | 512 | Partab[i] = cnt & 1; | |
| 142 | } | ||
| 143 | /* Initialize the branch table */ | ||
| 144 |
2/2✓ Branch 0 taken 64 times.
✓ Branch 1 taken 2 times.
|
66 | for (state = 0; state < d_numstates / 2; state++) { |
| 145 |
2/2✓ Branch 0 taken 128 times.
✓ Branch 1 taken 64 times.
|
192 | for (i = 0; i < rate; i++) { |
| 146 |
2/2✓ Branch 0 taken 64 times.
✓ Branch 1 taken 64 times.
|
128 | Branchtab[i * d_numstates / 2 + state] = |
| 147 | 128 | parity((2 * state) & d_polys[i], Partab) ? 255 : 0; | |
| 148 | } | ||
| 149 | } | ||
| 150 | |||
| 151 | 2 | once = 0; | |
| 152 | } | ||
| 153 | |||
| 154 | // unbias the old_metrics | ||
| 155 | 2 | memset(X, 31, d_numstates); | |
| 156 | |||
| 157 | // initialize decisions | ||
| 158 | 2 | memset(D, 0, (d_numstates / 8) * (framebits + 6)); | |
| 159 | |||
| 160 | 2 | volk_8u_x4_conv_k7_r2_8u_spiral( | |
| 161 | 2 | Y, X, syms, D, framebits / 2 - excess, excess, Branchtab); | |
| 162 | |||
| 163 | 2 | unsigned int min = X[0]; | |
| 164 | 2 | int i = 0, state = 0; | |
| 165 |
2/2✓ Branch 0 taken 128 times.
✓ Branch 1 taken 2 times.
|
130 | for (i = 0; i < (d_numstates); ++i) { |
| 166 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 128 times.
|
128 | if (X[i] < min) { |
| 167 | ✗ | min = X[i]; | |
| 168 | ✗ | state = i; | |
| 169 | } | ||
| 170 | } | ||
| 171 | |||
| 172 | 2 | chainback_viterbi(dec, framebits / 2 - excess, state, excess, D); | |
| 173 | |||
| 174 | 2 | return; | |
| 175 | } | ||
| 176 | |||
| 177 | #endif /*LV_HAVE_SSE3*/ | ||
| 178 | |||
| 179 | |||
| 180 | #if LV_HAVE_NEON | ||
| 181 | |||
| 182 | #include "volk/sse2neon.h" | ||
| 183 | |||
| 184 | static inline void volk_8u_conv_k7_r2puppet_8u_neonspiral(unsigned char* syms, | ||
| 185 | unsigned char* dec, | ||
| 186 | unsigned int framebits) | ||
| 187 | { | ||
| 188 | if (framebits < 12) { | ||
| 189 | return; | ||
| 190 | } | ||
| 191 | |||
| 192 | static int once = 1; | ||
| 193 | int d_numstates = (1 << 6); | ||
| 194 | int rate = 2; | ||
| 195 | static unsigned char* D; | ||
| 196 | static unsigned char* Y; | ||
| 197 | static unsigned char* X; | ||
| 198 | static unsigned int excess = 6; | ||
| 199 | static unsigned char* Branchtab; | ||
| 200 | static unsigned char Partab[256]; | ||
| 201 | |||
| 202 | int d_polys[2] = { 79, 109 }; | ||
| 203 | |||
| 204 | |||
| 205 | if (once) { | ||
| 206 | |||
| 207 | X = (unsigned char*)volk_malloc(2 * d_numstates, volk_get_alignment()); | ||
| 208 | Y = X + d_numstates; | ||
| 209 | Branchtab = | ||
| 210 | (unsigned char*)volk_malloc(d_numstates / 2 * rate, volk_get_alignment()); | ||
| 211 | D = (unsigned char*)volk_malloc((d_numstates / 8) * (framebits + 6), | ||
| 212 | volk_get_alignment()); | ||
| 213 | int state, i; | ||
| 214 | int cnt, ti; | ||
| 215 | |||
| 216 | /* Initialize parity lookup table */ | ||
| 217 | for (i = 0; i < 256; i++) { | ||
| 218 | cnt = 0; | ||
| 219 | ti = i; | ||
| 220 | while (ti) { | ||
| 221 | if (ti & 1) | ||
| 222 | cnt++; | ||
| 223 | ti >>= 1; | ||
| 224 | } | ||
| 225 | Partab[i] = cnt & 1; | ||
| 226 | } | ||
| 227 | /* Initialize the branch table */ | ||
| 228 | for (state = 0; state < d_numstates / 2; state++) { | ||
| 229 | for (i = 0; i < rate; i++) { | ||
| 230 | Branchtab[i * d_numstates / 2 + state] = | ||
| 231 | parity((2 * state) & d_polys[i], Partab) ? 255 : 0; | ||
| 232 | } | ||
| 233 | } | ||
| 234 | |||
| 235 | once = 0; | ||
| 236 | } | ||
| 237 | |||
| 238 | // unbias the old_metrics | ||
| 239 | memset(X, 31, d_numstates); | ||
| 240 | |||
| 241 | // initialize decisions | ||
| 242 | memset(D, 0, (d_numstates / 8) * (framebits + 6)); | ||
| 243 | |||
| 244 | volk_8u_x4_conv_k7_r2_8u_neonspiral( | ||
| 245 | Y, X, syms, D, framebits / 2 - excess, excess, Branchtab); | ||
| 246 | |||
| 247 | unsigned int min = X[0]; | ||
| 248 | int i = 0, state = 0; | ||
| 249 | for (i = 0; i < (d_numstates); ++i) { | ||
| 250 | if (X[i] < min) { | ||
| 251 | min = X[i]; | ||
| 252 | state = i; | ||
| 253 | } | ||
| 254 | } | ||
| 255 | |||
| 256 | chainback_viterbi(dec, framebits / 2 - excess, state, excess, D); | ||
| 257 | |||
| 258 | return; | ||
| 259 | } | ||
| 260 | |||
| 261 | #endif /*LV_HAVE_NEON*/ | ||
| 262 | |||
| 263 | |||
| 264 | //#if LV_HAVE_AVX2 | ||
| 265 | // | ||
| 266 | //#include <immintrin.h> | ||
| 267 | //#include <stdio.h> | ||
| 268 | // | ||
| 269 | // static inline void volk_8u_conv_k7_r2puppet_8u_avx2(unsigned char* syms, | ||
| 270 | // unsigned char* dec, | ||
| 271 | // unsigned int framebits) | ||
| 272 | //{ | ||
| 273 | // if (framebits < 12) { | ||
| 274 | // return; | ||
| 275 | // } | ||
| 276 | // | ||
| 277 | // static int once = 1; | ||
| 278 | // int d_numstates = (1 << 6); | ||
| 279 | // int rate = 2; | ||
| 280 | // static unsigned char* D; | ||
| 281 | // static unsigned char* Y; | ||
| 282 | // static unsigned char* X; | ||
| 283 | // static unsigned int excess = 6; | ||
| 284 | // static unsigned char* Branchtab; | ||
| 285 | // static unsigned char Partab[256]; | ||
| 286 | // | ||
| 287 | // int d_polys[2] = { 79, 109 }; | ||
| 288 | // | ||
| 289 | // | ||
| 290 | // if (once) { | ||
| 291 | // | ||
| 292 | // X = (unsigned char*)volk_malloc(2 * d_numstates, volk_get_alignment()); | ||
| 293 | // Y = X + d_numstates; | ||
| 294 | // Branchtab = | ||
| 295 | // (unsigned char*)volk_malloc(d_numstates / 2 * rate, volk_get_alignment()); | ||
| 296 | // D = (unsigned char*)volk_malloc((d_numstates / 8) * (framebits + 6), | ||
| 297 | // volk_get_alignment()); | ||
| 298 | // int state, i; | ||
| 299 | // int cnt, ti; | ||
| 300 | // | ||
| 301 | // /* Initialize parity lookup table */ | ||
| 302 | // for (i = 0; i < 256; i++) { | ||
| 303 | // cnt = 0; | ||
| 304 | // ti = i; | ||
| 305 | // while (ti) { | ||
| 306 | // if (ti & 1) | ||
| 307 | // cnt++; | ||
| 308 | // ti >>= 1; | ||
| 309 | // } | ||
| 310 | // Partab[i] = cnt & 1; | ||
| 311 | // } | ||
| 312 | // /* Initialize the branch table */ | ||
| 313 | // for (state = 0; state < d_numstates / 2; state++) { | ||
| 314 | // for (i = 0; i < rate; i++) { | ||
| 315 | // Branchtab[i * d_numstates / 2 + state] = | ||
| 316 | // parity((2 * state) & d_polys[i], Partab) ? 255 : 0; | ||
| 317 | // } | ||
| 318 | // } | ||
| 319 | // | ||
| 320 | // once = 0; | ||
| 321 | // } | ||
| 322 | // | ||
| 323 | // // unbias the old_metrics | ||
| 324 | // memset(X, 31, d_numstates); | ||
| 325 | // | ||
| 326 | // // initialize decisions | ||
| 327 | // memset(D, 0, (d_numstates / 8) * (framebits + 6)); | ||
| 328 | // | ||
| 329 | // volk_8u_x4_conv_k7_r2_8u_avx2( | ||
| 330 | // Y, X, syms, D, framebits / 2 - excess, excess, Branchtab); | ||
| 331 | // | ||
| 332 | // unsigned int min = X[0]; | ||
| 333 | // int i = 0, state = 0; | ||
| 334 | // for (i = 0; i < (d_numstates); ++i) { | ||
| 335 | // if (X[i] < min) { | ||
| 336 | // min = X[i]; | ||
| 337 | // state = i; | ||
| 338 | // } | ||
| 339 | // } | ||
| 340 | // | ||
| 341 | // chainback_viterbi(dec, framebits / 2 - excess, state, excess, D); | ||
| 342 | // | ||
| 343 | // return; | ||
| 344 | //} | ||
| 345 | // | ||
| 346 | //#endif /*LV_HAVE_AVX2*/ | ||
| 347 | |||
| 348 | |||
| 349 | #if LV_HAVE_GENERIC | ||
| 350 | |||
| 351 | |||
| 352 | 2 | static inline void volk_8u_conv_k7_r2puppet_8u_generic(unsigned char* syms, | |
| 353 | unsigned char* dec, | ||
| 354 | unsigned int framebits) | ||
| 355 | { | ||
| 356 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 | if (framebits < 12) { |
| 357 | ✗ | return; | |
| 358 | } | ||
| 359 | |||
| 360 | static int once = 1; | ||
| 361 | 2 | int d_numstates = (1 << 6); | |
| 362 | 2 | int rate = 2; | |
| 363 | static unsigned char* Y; | ||
| 364 | static unsigned char* X; | ||
| 365 | static unsigned char* D; | ||
| 366 | static unsigned int excess = 6; | ||
| 367 | static unsigned char* Branchtab; | ||
| 368 | static unsigned char Partab[256]; | ||
| 369 | |||
| 370 | 2 | int d_polys[2] = { 79, 109 }; | |
| 371 | |||
| 372 | |||
| 373 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (once) { |
| 374 | |||
| 375 | 2 | X = (unsigned char*)volk_malloc(2 * d_numstates, volk_get_alignment()); | |
| 376 | 2 | Y = X + d_numstates; | |
| 377 | 2 | Branchtab = | |
| 378 | 2 | (unsigned char*)volk_malloc(d_numstates / 2 * rate, volk_get_alignment()); | |
| 379 | 2 | D = (unsigned char*)volk_malloc((d_numstates / 8) * (framebits + 6), | |
| 380 | volk_get_alignment()); | ||
| 381 | |||
| 382 | int state, i; | ||
| 383 | int cnt, ti; | ||
| 384 | |||
| 385 | /* Initialize parity lookup table */ | ||
| 386 |
2/2✓ Branch 0 taken 512 times.
✓ Branch 1 taken 2 times.
|
514 | for (i = 0; i < 256; i++) { |
| 387 | 512 | cnt = 0; | |
| 388 | 512 | ti = i; | |
| 389 |
2/2✓ Branch 0 taken 3586 times.
✓ Branch 1 taken 512 times.
|
4098 | while (ti) { |
| 390 |
2/2✓ Branch 0 taken 2048 times.
✓ Branch 1 taken 1538 times.
|
3586 | if (ti & 1) |
| 391 | 2048 | cnt++; | |
| 392 | 3586 | ti >>= 1; | |
| 393 | } | ||
| 394 | 512 | Partab[i] = cnt & 1; | |
| 395 | } | ||
| 396 | /* Initialize the branch table */ | ||
| 397 |
2/2✓ Branch 0 taken 64 times.
✓ Branch 1 taken 2 times.
|
66 | for (state = 0; state < d_numstates / 2; state++) { |
| 398 |
2/2✓ Branch 0 taken 128 times.
✓ Branch 1 taken 64 times.
|
192 | for (i = 0; i < rate; i++) { |
| 399 |
2/2✓ Branch 0 taken 64 times.
✓ Branch 1 taken 64 times.
|
128 | Branchtab[i * d_numstates / 2 + state] = |
| 400 | 128 | parity((2 * state) & d_polys[i], Partab) ? 255 : 0; | |
| 401 | } | ||
| 402 | } | ||
| 403 | |||
| 404 | 2 | once = 0; | |
| 405 | } | ||
| 406 | |||
| 407 | // unbias the old_metrics | ||
| 408 | 2 | memset(X, 31, d_numstates); | |
| 409 | |||
| 410 | // initialize decisions | ||
| 411 | 2 | memset(D, 0, (d_numstates / 8) * (framebits + 6)); | |
| 412 | |||
| 413 | 2 | volk_8u_x4_conv_k7_r2_8u_generic( | |
| 414 | 2 | Y, X, syms, D, framebits / 2 - excess, excess, Branchtab); | |
| 415 | |||
| 416 | 2 | unsigned int min = X[0]; | |
| 417 | 2 | int i = 0, state = 0; | |
| 418 |
2/2✓ Branch 0 taken 128 times.
✓ Branch 1 taken 2 times.
|
130 | for (i = 0; i < (d_numstates); ++i) { |
| 419 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 128 times.
|
128 | if (X[i] < min) { |
| 420 | ✗ | min = X[i]; | |
| 421 | ✗ | state = i; | |
| 422 | } | ||
| 423 | } | ||
| 424 | |||
| 425 | 2 | chainback_viterbi(dec, framebits / 2 - excess, state, excess, D); | |
| 426 | |||
| 427 | 2 | return; | |
| 428 | } | ||
| 429 | |||
| 430 | #endif /* LV_HAVE_GENERIC */ | ||
| 431 | |||
| 432 | #endif /*INCLUDED_volk_8u_conv_k7_r2puppet_8u_H*/ | ||
| 433 |