3gpplte_turbo_decoder_sse_16bit.c 57.2 KB
Newer Older
1 2 3
/* file: 3gpplte_turbo_decoder_sse_16bit.c
   purpose: Routines for implementing max-logmap decoding of Turbo-coded (DLSCH) transport channels from 36-212, V8.6 2009-03
   authors: raymond.knopp@eurecom.fr, Laurent Thomas (Alcatel-Lucent)
4
   date: 21.10.2009
5 6 7

   Note: This version of the routine currently requires SSE2,SSSE3 and SSE4.1 equipped computers.  It uses 16-bit inputs for
         LLRS and uses 16-bit arithmetic for the internal computations!
8

9 10 11 12 13 14 15 16 17 18
   Changelog: 17.11.2009 FK SSE4.1 not required anymore
   Aug. 2012 new parallelization options for higher speed (8-way parallelization)
   Jan. 2013 8-bit LLR support with 16-way parallelization
   Feb. 2013 New interleaving and hard-decision optimizations (L. Thomas)
   May 2013 Extracted 16bit code
*/

///
///

knopp's avatar
knopp committed
19
#undef __AVX2__
20
#include "PHY/sse_intrin.h"
21

22
#ifndef TEST_DEBUG
23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38
#include "PHY/defs.h"
#include "PHY/CODING/defs.h"
#include "PHY/CODING/lte_interleaver_inline.h"
#include "extern_3GPPinterleaver.h"
#else

#include "defs.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#endif

#ifdef MEX
#include "mex.h"
#endif

39 40 41 42 43
//#define DEBUG_LOGMAP

#ifdef DEBUG_LOGMAP
#define print_shorts(s,x) fprintf(fdsse4,"%s %d,%d,%d,%d,%d,%d,%d,%d\n",s,(x)[0],(x)[1],(x)[2],(x)[3],(x)[4],(x)[5],(x)[6],(x)[7])
#endif
44

45
#undef __AVX2__
46

47 48 49
#ifdef DEBUG_LOGMAP
FILE *fdsse4;
#endif
50

51 52 53 54
typedef int16_t llr_t; // internal decoder LLR data is 16-bit fixed
typedef int16_t channel_t;
#define MAX 256

55 56
void log_map16(llr_t* systematic,channel_t* y_parity, llr_t* m11, llr_t* m10, llr_t *alpha, llr_t *beta, llr_t* ext,unsigned short frame_length,unsigned char term_flag,unsigned char F,
               int offset8_flag,time_stats_t *alpha_stats,time_stats_t *beta_stats,time_stats_t *gamma_stats,time_stats_t *ext_stats);
57 58 59 60 61 62 63
void compute_gamma16(llr_t* m11,llr_t* m10,llr_t* systematic, channel_t* y_parity, unsigned short frame_length,unsigned char term_flag);
void compute_alpha16(llr_t*alpha,llr_t *beta, llr_t* m11,llr_t* m10, unsigned short frame_length,unsigned char F);
void compute_beta16(llr_t*alpha, llr_t* beta,llr_t* m11,llr_t* m10, unsigned short frame_length,unsigned char F,int offset8_flag);
void compute_ext16(llr_t* alpha,llr_t* beta,llr_t* m11,llr_t* m10,llr_t* extrinsic, llr_t* ap, unsigned short frame_length);


void log_map16(llr_t* systematic,
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79
               channel_t* y_parity,
               llr_t* m11,
               llr_t* m10,
               llr_t *alpha,
               llr_t *beta,
               llr_t* ext,
               unsigned short frame_length,
               unsigned char term_flag,
               unsigned char F,
               int offset8_flag,
               time_stats_t *alpha_stats,
               time_stats_t *beta_stats,
               time_stats_t *gamma_stats,
               time_stats_t *ext_stats)
{

80
#ifdef DEBUG_LOGMAP
81
  fprintf(fdsse4,"log_map, frame_length %d\n",frame_length);
82 83
#endif

84 85 86 87 88 89 90 91 92 93 94 95
  start_meas(gamma_stats) ;
  compute_gamma16(m11,m10,systematic,y_parity,frame_length,term_flag) ;
  stop_meas(gamma_stats);
  start_meas(alpha_stats) ;
  compute_alpha16(alpha,beta,m11,m10,frame_length,F)                  ;
  stop_meas(alpha_stats);
  start_meas(beta_stats)  ;
  compute_beta16(alpha,beta,m11,m10,frame_length,F,offset8_flag)      ;
  stop_meas(beta_stats);
  start_meas(ext_stats)   ;
  compute_ext16(alpha,beta,m11,m10,ext,systematic,frame_length)       ;
  stop_meas(ext_stats);
96 97 98 99 100


}

void compute_gamma16(llr_t* m11,llr_t* m10,llr_t* systematic,channel_t* y_parity,
101 102
                     unsigned short frame_length,unsigned char term_flag)
{
103
  int k,K1;
104
#if defined(__x86_64__)||defined(__i386__)
105 106 107 108
  __m128i *systematic128 = (__m128i *)systematic;
  __m128i *y_parity128   = (__m128i *)y_parity;
  __m128i *m10_128        = (__m128i *)m10;
  __m128i *m11_128        = (__m128i *)m11;
109 110 111 112 113 114
#elif defined(__arm__)
  int16x8_t *systematic128  = (int16x8_t *)systematic;
  int16x8_t *y_parity128    = (int16x8_t *)y_parity;
  int16x8_t *m10_128        = (int16x8_t *)m10;
  int16x8_t *m11_128        = (int16x8_t *)m11;
#endif
115 116

#ifdef DEBUG_LOGMAP
117
  fprintf(fdsse4,"compute_gamma (sse_16bit), %p,%p,%p,%p,framelength %d\n",m11,m10,systematic,y_parity,frame_length);
118 119
#endif

120
#ifndef __AVX2__
121
  K1=frame_length>>3;
122 123 124 125 126 127
#else
  if ((frame_length&15) > 0)
    K1=(frame_length+1)>>4;
  else
    K1=frame_length>>4;
#endif
128 129

  for (k=0; k<K1; k++) {
130
#if defined(__x86_64__) || defined(__i386__)
131
#ifndef __AVX2__
132 133
    m11_128[k] = _mm_srai_epi16(_mm_adds_epi16(systematic128[k],y_parity128[k]),1);
    m10_128[k] = _mm_srai_epi16(_mm_subs_epi16(systematic128[k],y_parity128[k]),1);
134 135 136 137 138
#else
    ((__m256i*)m11_128)[k] = _mm256_srai_epi16(_mm256_adds_epi16(((__m256i*)systematic128)[k],((__m256i*)y_parity128)[k]),1);
    //    ((__m256i*)m10_128)[k] = _mm256_srai_epi16(_mm256_subs_epi16(((__m256i*)y_parity128)[k],((__m256i*)systematic128)[k]),1);
    ((__m256i*)m10_128)[k] = _mm256_srai_epi16(_mm256_subs_epi16(((__m256i*)systematic128)[k],((__m256i*)y_parity128)[k]),1);
#endif
139 140 141 142
#elif defined(__arm__)
    m11_128[k] = vhaddq_s16(systematic128[k],y_parity128[k]);
    m10_128[k] = vhsubq_s16(systematic128[k],y_parity128[k]);
#endif
143

144
#ifdef DEBUG_LOGMAP
145
    fprintf(fdsse4,"Loop index k %d\n", k);
146 147 148 149 150
    print_shorts("sys",(int16_t*)&systematic128[k]);
    print_shorts("yp",(int16_t*)&y_parity128[k]);
    print_shorts("m11",(int16_t*)&m11_128[k]);
    print_shorts("m10",(int16_t*)&m10_128[k]);
#endif
151
  }
152

153
  k=frame_length>>3;
154
  // Termination
155
#if defined(__x86_64__) || defined(__i386__)
156
  m11_128[k] = _mm_srai_epi16(_mm_adds_epi16(systematic128[k+term_flag],y_parity128[k]),1);
157 158
  //#ifndef __AVX2__
#if 1
159
  m10_128[k] = _mm_srai_epi16(_mm_subs_epi16(systematic128[k+term_flag],y_parity128[k]),1);
160 161 162
#else
  m10_128[k] = _mm_srai_epi16(_mm_subs_epi16(y_parity128[k],systematic128[k+term_flag]),1);
#endif
163 164
#elif defined(__arm__)
  m11_128[k] = vhaddq_s16(systematic128[k+term_flag],y_parity128[k]);
165
  m10_128[k] = vhsubq_s16(systematic128[k+term_flag],y_parity128[k]);
166 167 168 169 170 171 172 173
#endif

#ifdef DEBUG_LOGMAP
fprintf(fdsse4,"Loop index k %d (term flag %d)\n", k,term_flag);
print_shorts("sys",(int16_t*)&systematic128[k]);
    print_shorts("yp",(int16_t*)&y_parity128[k]);
    print_shorts("m11",(int16_t*)&m11_128[k]);
    print_shorts("m10",(int16_t*)&m10_128[k]);
174
#endif
175 176 177 178
}

#define L 40

179 180
void compute_alpha16(llr_t* alpha,llr_t* beta,llr_t* m_11,llr_t* m_10,unsigned short frame_length,unsigned char F)
{
181
  int k,l,l2,K1,rerun_flag=0;
182
#if defined(__x86_64__) || defined(__i386__)
183 184 185 186
  __m128i *alpha128=(__m128i *)alpha,*alpha_ptr,*m11p,*m10p;
  //#ifndef __AVX2__
#if 1
  __m128i a0,a1,a2,a3,a4,a5,a6,a7;
187 188 189
  __m128i m_b0,m_b1,m_b2,m_b3,m_b4,m_b5,m_b6,m_b7;
  __m128i new0,new1,new2,new3,new4,new5,new6,new7;
  __m128i alpha_max;
190 191 192 193 194 195 196 197
#else
  __m256i *alpha256=(__m256i *)alpha,*alpha_ptr256,m11,m10;
  __m256i a01,a23,a45,a67,a02,a13,a64,a75;
  __m256i m_b01,m_b23,m_b45,m_b67,new01,new23,new45,new67;
  __m256i m11m10_256;
  __m256i alpha_max;
#endif

198 199 200 201 202 203 204
#elif defined(__arm__)
  int16x8_t *alpha128=(int16x8_t *)alpha,*alpha_ptr;
  int16x8_t a0,a1,a2,a3,a4,a5,a6,a7,*m11p,*m10p;
  int16x8_t m_b0,m_b1,m_b2,m_b3,m_b4,m_b5,m_b6,m_b7;
  int16x8_t new0,new1,new2,new3,new4,new5,new6,new7;
  int16x8_t alpha_max;
#endif
205 206
  l2 = L>>3;
  K1 = (frame_length>>3);
207 208 209
#ifdef DEBUG_LOGMAP
  fprintf(fdsse4,"compute_alpha (sse_16bit)\n");
#endif
210
  for (l=K1;; l=l2,rerun_flag=1) {
211
#if defined(__x86_64__) || defined(__i386__)
212
    alpha128 = (__m128i *)alpha;
213 214 215 216
    //#ifdef __AVX2__
#if 0
    alpha256 = (__m256i *)alpha;
#endif
217 218 219
#elif defined(__arm__)
    alpha128 = (int16x8_t *)alpha;
#endif
220

221
    if (rerun_flag == 0) {
222
#if defined(__x86_64__) || defined(__i386__)
223 224 225 226 227 228 229 230
      alpha128[0] = _mm_set_epi16(-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,0);
      alpha128[1] = _mm_set_epi16(-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2);
      alpha128[2] = _mm_set_epi16(-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2);
      alpha128[3] = _mm_set_epi16(-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2);
      alpha128[4] = _mm_set_epi16(-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2);
      alpha128[5] = _mm_set_epi16(-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2);
      alpha128[6] = _mm_set_epi16(-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2);
      alpha128[7] = _mm_set_epi16(-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2,-MAX/2);
231 232 233 234 235 236 237 238 239 240 241 242
#elif defined(__arm__)
      alpha128[0] = vdupq_n_s16(-MAX/2);
      alpha128[0] = vsetq_lane_s16(0,alpha128[0],0);
      alpha128[1] = vdupq_n_s16(-MAX/2);
      alpha128[2] = vdupq_n_s16(-MAX/2);
      alpha128[3] = vdupq_n_s16(-MAX/2);
      alpha128[4] = vdupq_n_s16(-MAX/2);
      alpha128[5] = vdupq_n_s16(-MAX/2);
      alpha128[6] = vdupq_n_s16(-MAX/2);
      alpha128[7] = vdupq_n_s16(-MAX/2);
#endif
#ifdef DEBUG_LOGMAP
243
      fprintf(fdsse4,"Initial alpha\n");
244 245 246 247 248 249 250 251 252
      print_shorts("a0",(int16_t*)&alpha128[0]);
      print_shorts("a1",(int16_t*)&alpha128[1]);
      print_shorts("a2",(int16_t*)&alpha128[2]);
      print_shorts("a3",(int16_t*)&alpha128[3]);
      print_shorts("a4",(int16_t*)&alpha128[4]);
      print_shorts("a5",(int16_t*)&alpha128[5]);
      print_shorts("a6",(int16_t*)&alpha128[6]);
      print_shorts("a7",(int16_t*)&alpha128[7]);
#endif
253
    } else {
254
      //set initial alpha in columns 1-7 from final alpha from last run in columns 0-6
255
#if defined(__x86_64__) || defined(__i386__)
256 257 258 259 260 261 262 263
      alpha128[0] = _mm_slli_si128(alpha128[frame_length],2);
      alpha128[1] = _mm_slli_si128(alpha128[1+frame_length],2);
      alpha128[2] = _mm_slli_si128(alpha128[2+frame_length],2);
      alpha128[3] = _mm_slli_si128(alpha128[3+frame_length],2);
      alpha128[4] = _mm_slli_si128(alpha128[4+frame_length],2);
      alpha128[5] = _mm_slli_si128(alpha128[5+frame_length],2);
      alpha128[6] = _mm_slli_si128(alpha128[6+frame_length],2);
      alpha128[7] = _mm_slli_si128(alpha128[7+frame_length],2);
264 265 266 267 268 269 270 271 272 273
#elif defined(__arm__)
      alpha128[0] = (int16x8_t)vshlq_n_s64((int64x2_t)alpha128[frame_length],16);   alpha128[0] = vsetq_lane_s16(alpha[8],alpha128[0],3);
      alpha128[1] = (int16x8_t)vshlq_n_s64((int64x2_t)alpha128[1+frame_length],16); alpha128[1] = vsetq_lane_s16(alpha[24],alpha128[0],3);
      alpha128[2] = (int16x8_t)vshlq_n_s64((int64x2_t)alpha128[2+frame_length],16); alpha128[2] = vsetq_lane_s16(alpha[40],alpha128[0],3);
      alpha128[3] = (int16x8_t)vshlq_n_s64((int64x2_t)alpha128[3+frame_length],16); alpha128[3] = vsetq_lane_s16(alpha[56],alpha128[0],3);
      alpha128[4] = (int16x8_t)vshlq_n_s64((int64x2_t)alpha128[4+frame_length],16); alpha128[4] = vsetq_lane_s16(alpha[72],alpha128[0],3);
      alpha128[5] = (int16x8_t)vshlq_n_s64((int64x2_t)alpha128[5+frame_length],16); alpha128[5] = vsetq_lane_s16(alpha[88],alpha128[0],3);
      alpha128[6] = (int16x8_t)vshlq_n_s64((int64x2_t)alpha128[6+frame_length],16); alpha128[6] = vsetq_lane_s16(alpha[104],alpha128[0],3);
      alpha128[7] = (int16x8_t)vshlq_n_s64((int64x2_t)alpha128[7+frame_length],16); alpha128[7] = vsetq_lane_s16(alpha[120],alpha128[0],3);
#endif
274 275 276 277 278 279 280 281
      // set initial alpha in column 0 to (0,-MAX/2,...,-MAX/2)
      alpha[8] = -MAX/2;
      alpha[16] = -MAX/2;
      alpha[24] = -MAX/2;
      alpha[32] = -MAX/2;
      alpha[40] = -MAX/2;
      alpha[48] = -MAX/2;
      alpha[56] = -MAX/2;
282
#ifdef DEBUG_LOGMAP
283
      fprintf(fdsse4,"Second run\n");
284 285 286 287 288 289 290 291 292 293
      print_shorts("a0",(int16_t*)&alpha128[0]);
      print_shorts("a1",(int16_t*)&alpha128[1]);
      print_shorts("a2",(int16_t*)&alpha128[2]);
      print_shorts("a3",(int16_t*)&alpha128[3]);
      print_shorts("a4",(int16_t*)&alpha128[4]);
      print_shorts("a5",(int16_t*)&alpha128[5]);
      print_shorts("a6",(int16_t*)&alpha128[6]);
      print_shorts("a7",(int16_t*)&alpha128[7]);
#endif

294
    }
295

296
    alpha_ptr = &alpha128[0];
297 298 299 300 301
    //#ifdef __AVX2__
#if 0
    alpha_ptr256 = &alpha256[0];
#endif

302
#if defined(__x86_64__) || defined(__i386__)
303 304
    m11p = (__m128i*)m_11;
    m10p = (__m128i*)m_10;
305 306 307 308
#elif defined(__arm__)
    m11p = (int16x8_t*)m_11;
    m10p = (int16x8_t*)m_10;
#endif
309
    for (k=0;
310 311
         k<l;
         k++) {
312

313
#if defined(__x86_64__) || defined(__i386__)
314 315
      //#ifndef __AVX2__
#if 1
316 317 318 319 320 321
      a1=_mm_load_si128(&alpha_ptr[1]);
      a3=_mm_load_si128(&alpha_ptr[3]);
      a5=_mm_load_si128(&alpha_ptr[5]);
      a7=_mm_load_si128(&alpha_ptr[7]);

      m_b0 = _mm_adds_epi16(a1,*m11p);  // m11
322
      m_b4 = _mm_subs_epi16(a1,*m11p);  // m00=-m11
323 324 325 326 327 328 329 330 331 332 333
      m_b1 = _mm_subs_epi16(a3,*m10p);  // m01=-m10
      m_b5 = _mm_adds_epi16(a3,*m10p);  // m10
      m_b2 = _mm_adds_epi16(a5,*m10p);  // m10
      m_b6 = _mm_subs_epi16(a5,*m10p);  // m01=-m10
      m_b3 = _mm_subs_epi16(a7,*m11p);  // m00=-m11
      m_b7 = _mm_adds_epi16(a7,*m11p);  // m11

      a0=_mm_load_si128(&alpha_ptr[0]);
      a2=_mm_load_si128(&alpha_ptr[2]);
      a4=_mm_load_si128(&alpha_ptr[4]);
      a6=_mm_load_si128(&alpha_ptr[6]);
334

335 336 337 338 339 340 341 342
      new0 = _mm_subs_epi16(a0,*m11p);  // m00=-m11
      new4 = _mm_adds_epi16(a0,*m11p);  // m11
      new1 = _mm_adds_epi16(a2,*m10p);  // m10
      new5 = _mm_subs_epi16(a2,*m10p);  // m01=-m10
      new2 = _mm_subs_epi16(a4,*m10p);  // m01=-m10
      new6 = _mm_adds_epi16(a4,*m10p);  // m10
      new3 = _mm_adds_epi16(a6,*m11p);  // m11
      new7 = _mm_subs_epi16(a6,*m11p);  // m00=-m11
343

344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359
      a0 = _mm_max_epi16(m_b0,new0);
      a1 = _mm_max_epi16(m_b1,new1);
      a2 = _mm_max_epi16(m_b2,new2);
      a3 = _mm_max_epi16(m_b3,new3);
      a4 = _mm_max_epi16(m_b4,new4);
      a5 = _mm_max_epi16(m_b5,new5);
      a6 = _mm_max_epi16(m_b6,new6);
      a7 = _mm_max_epi16(m_b7,new7);

      alpha_max = _mm_max_epi16(a0,a1);
      alpha_max = _mm_max_epi16(alpha_max,a2);
      alpha_max = _mm_max_epi16(alpha_max,a3);
      alpha_max = _mm_max_epi16(alpha_max,a4);
      alpha_max = _mm_max_epi16(alpha_max,a5);
      alpha_max = _mm_max_epi16(alpha_max,a6);
      alpha_max = _mm_max_epi16(alpha_max,a7);
360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390
#else
      a02=_mm256_load_si256(&alpha_ptr256[0]);
      a13=_mm256_load_si256(&alpha_ptr256[1]);
      a64=_mm256_load_si256(&alpha_ptr256[2]);
      a75=_mm256_load_si256(&alpha_ptr256[3]);
      m11m10_256 = _mm256_insertf128_si256(m11m10_256,*m11p,0);
      m11m10_256 = _mm256_insertf128_si256(m11m10_256,*m10p,1);


      m_b01 = _mm256_adds_epi16(a13,m11m10_256); //negative m10
      m_b23 = _mm256_subs_epi16(a75,m11m10_256); //negative m10
      m_b45 = _mm256_subs_epi16(a13,m11m10_256); //negative m10
      m_b67 = _mm256_adds_epi16(a75,m11m10_256); //negative m10

      new01 = _mm256_subs_epi16(a02,m11m10_256);  //negative m10
      new23 = _mm256_adds_epi16(a64,m11m10_256);  //negative m10
      new45 = _mm256_adds_epi16(a02,m11m10_256);  //negative m10
      new67 = _mm256_subs_epi16(a64,m11m10_256);  //negative m10

      a01   = _mm256_max_epi16(m_b01,new01);
      a23   = _mm256_max_epi16(m_b23,new23);
      a45   = _mm256_max_epi16(m_b45,new45);
      a67   = _mm256_max_epi16(m_b67,new67);

      alpha_max = _mm256_max_epi16(a01,a23);
      alpha_max = _mm256_max_epi16(alpha_max,a45);
      alpha_max = _mm256_max_epi16(alpha_max,a67);
      alpha_max = _mm256_max_epi16(alpha_max,_mm256_permutevar8x32_epi32(alpha_max,_mm256_set_epi32(3,2,1,0,7,6,5,4)));
      
      
#endif
391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427
#elif defined(__arm__)
      m_b0 = vqaddq_s16(alpha_ptr[1],*m11p);  // m11
      m_b4 = vqsubq_s16(alpha_ptr[1],*m11p);  // m00=-m11
      m_b1 = vqsubq_s16(alpha_ptr[3],*m10p);  // m01=-m10
      m_b5 = vqaddq_s16(alpha_ptr[3],*m10p);  // m10
      m_b2 = vqaddq_s16(alpha_ptr[5],*m10p);  // m10
      m_b6 = vqsubq_s16(alpha_ptr[5],*m10p);  // m01=-m10
      m_b3 = vqsubq_s16(alpha_ptr[7],*m11p);  // m00=-m11
      m_b7 = vqaddq_s16(alpha_ptr[7],*m11p);  // m11

      new0 = vqsubq_s16(alpha_ptr[0],*m11p);  // m00=-m11
      new4 = vqaddq_s16(alpha_ptr[0],*m11p);  // m11
      new1 = vqaddq_s16(alpha_ptr[2],*m10p);  // m10
      new5 = vqsubq_s16(alpha_ptr[2],*m10p);  // m01=-m10
      new2 = vqsubq_s16(alpha_ptr[4],*m10p);  // m01=-m10
      new6 = vqaddq_s16(alpha_ptr[4],*m10p);  // m10
      new3 = vqaddq_s16(alpha_ptr[6],*m11p);  // m11
      new7 = vqsubq_s16(alpha_ptr[6],*m11p);  // m00=-m11
      a0 = vmaxq_s16(m_b0,new0);
      a1 = vmaxq_s16(m_b1,new1);
      a2 = vmaxq_s16(m_b2,new2);
      a3 = vmaxq_s16(m_b3,new3);
      a4 = vmaxq_s16(m_b4,new4);
      a5 = vmaxq_s16(m_b5,new5);
      a6 = vmaxq_s16(m_b6,new6);
      a7 = vmaxq_s16(m_b7,new7);

      // compute and subtract maxima
      alpha_max = vmaxq_s16(a0,a1);
      alpha_max = vmaxq_s16(alpha_max,a2);
      alpha_max = vmaxq_s16(alpha_max,a3);
      alpha_max = vmaxq_s16(alpha_max,a4);
      alpha_max = vmaxq_s16(alpha_max,a5);
      alpha_max = vmaxq_s16(alpha_max,a6);
      alpha_max = vmaxq_s16(alpha_max,a7);

#endif
428 429

      alpha_ptr+=8;
430 431 432 433
      //#ifdef __AVX2__
#if 0
      alpha_ptr256+=4;
#endif
434
      m11p++;
435
      m10p++;
436
#if defined(__x86_64__) || defined(__i386__)
437 438
      //#ifndef __AVX2__
#if 1
439 440 441 442 443 444 445 446
      alpha_ptr[0] = _mm_subs_epi16(a0,alpha_max);
      alpha_ptr[1] = _mm_subs_epi16(a1,alpha_max);
      alpha_ptr[2] = _mm_subs_epi16(a2,alpha_max);
      alpha_ptr[3] = _mm_subs_epi16(a3,alpha_max);
      alpha_ptr[4] = _mm_subs_epi16(a4,alpha_max);
      alpha_ptr[5] = _mm_subs_epi16(a5,alpha_max);
      alpha_ptr[6] = _mm_subs_epi16(a6,alpha_max);
      alpha_ptr[7] = _mm_subs_epi16(a7,alpha_max);
447 448 449 450 451 452 453 454 455 456 457 458
#else

      a01   = _mm256_subs_epi16(a01,alpha_max);
      a23   = _mm256_subs_epi16(a23,alpha_max);
      a45   = _mm256_subs_epi16(a45,alpha_max);
      a67   = _mm256_subs_epi16(a67,alpha_max);

      alpha_ptr256[0] = _mm256_permute2x128_si256(a01,a23,0x20);  //a02
      alpha_ptr256[1] = _mm256_permute2x128_si256(a01,a23,0x13);  //a13
      alpha_ptr256[2] = _mm256_permute2x128_si256(a45,a67,0x02);  //a64
      alpha_ptr256[3] = _mm256_permute2x128_si256(a45,a67,0x31);  //a75
#endif
459 460 461 462 463 464 465 466 467 468 469 470
#elif defined(__arm__)
      alpha_ptr[0] = vqsubq_s16(a0,alpha_max);
      alpha_ptr[1] = vqsubq_s16(a1,alpha_max);
      alpha_ptr[2] = vqsubq_s16(a2,alpha_max);
      alpha_ptr[3] = vqsubq_s16(a3,alpha_max);
      alpha_ptr[4] = vqsubq_s16(a4,alpha_max);
      alpha_ptr[5] = vqsubq_s16(a5,alpha_max);
      alpha_ptr[6] = vqsubq_s16(a6,alpha_max);
      alpha_ptr[7] = vqsubq_s16(a7,alpha_max);
#endif

#ifdef DEBUG_LOGMAP
471
      fprintf(fdsse4,"Loop index %d\n",k);
472 473 474 475 476 477 478 479 480
      print_shorts("mb0",(int16_t*)&m_b0);
      print_shorts("mb1",(int16_t*)&m_b1);
      print_shorts("mb2",(int16_t*)&m_b2);
      print_shorts("mb3",(int16_t*)&m_b3);
      print_shorts("mb4",(int16_t*)&m_b4);
      print_shorts("mb5",(int16_t*)&m_b5);
      print_shorts("mb6",(int16_t*)&m_b6);
      print_shorts("mb7",(int16_t*)&m_b7);

481
      fprintf(fdsse4,"Loop index %d, new\n",k);
482 483 484 485 486 487 488 489 490
      print_shorts("new0",(int16_t*)&new0);
      print_shorts("new1",(int16_t*)&new1);
      print_shorts("new2",(int16_t*)&new2);
      print_shorts("new3",(int16_t*)&new3);
      print_shorts("new4",(int16_t*)&new4);
      print_shorts("new5",(int16_t*)&new5);
      print_shorts("new6",(int16_t*)&new6);
      print_shorts("new7",(int16_t*)&new7);

491
      fprintf(fdsse4,"Loop index %d, after max\n",k);
492 493 494 495 496 497 498 499 500
      print_shorts("a0",(int16_t*)&a0);
      print_shorts("a1",(int16_t*)&a1);
      print_shorts("a2",(int16_t*)&a2);
      print_shorts("a3",(int16_t*)&a3);
      print_shorts("a4",(int16_t*)&a4);
      print_shorts("a5",(int16_t*)&a5);
      print_shorts("a6",(int16_t*)&a6);
      print_shorts("a7",(int16_t*)&a7);

501
      fprintf(fdsse4,"Loop index %d\n",k);
502 503 504 505 506 507 508 509 510
      print_shorts("a0",(int16_t*)&alpha_ptr[0]);
      print_shorts("a1",(int16_t*)&alpha_ptr[1]);
      print_shorts("a2",(int16_t*)&alpha_ptr[2]);
      print_shorts("a3",(int16_t*)&alpha_ptr[3]);
      print_shorts("a4",(int16_t*)&alpha_ptr[4]);
      print_shorts("a5",(int16_t*)&alpha_ptr[5]);
      print_shorts("a6",(int16_t*)&alpha_ptr[6]);
      print_shorts("a7",(int16_t*)&alpha_ptr[7]);
#endif
511 512 513 514 515

    }

    if (rerun_flag==1)
      break;
516
  }
517 518 519
}


520 521 522
void compute_beta16(llr_t* alpha,llr_t* beta,llr_t *m_11,llr_t* m_10,unsigned short frame_length,unsigned char F,int offset8_flag)
{

523
  int k,rerun_flag=0;
524
#if defined(__x86_64__) || defined(__i386__)
525 526 527 528 529
  __m128i m11_128,m10_128;
  __m128i m_b0,m_b1,m_b2,m_b3,m_b4,m_b5,m_b6,m_b7;
  __m128i new0,new1,new2,new3,new4,new5,new6,new7;

  __m128i *beta128,*alpha128,*beta_ptr;
530
  __m128i beta_max;
531 532 533 534 535 536 537 538 539
#elif defined(__arm__)
  int16x8_t m11_128,m10_128;
  int16x8_t m_b0,m_b1,m_b2,m_b3,m_b4,m_b5,m_b6,m_b7;
  int16x8_t new0,new1,new2,new3,new4,new5,new6,new7;

  int16x8_t *beta128,*alpha128,*beta_ptr;
  int16x8_t beta_max;
#endif

540
  int16_t m11,m10,beta0_16,beta1_16,beta2_16,beta3_16,beta4_16,beta5_16,beta6_16,beta7_16,beta0_2,beta1_2,beta2_2,beta3_2,beta_m;
541 542 543
  llr_t beta0,beta1;

#ifdef DEBUG_LOGMAP
544 545
  fprintf(fdsse4,"compute_beta, %p,%p,%p,%p,framelength %d,F %d\n",
	  beta,m_11,m_10,alpha,frame_length,F);
546 547 548 549 550
#endif


  // termination for beta initialization

551
  //  fprintf(fdsse4,"beta init: offset8 %d\n",offset8_flag);
552
  m11=(int16_t)m_11[2+frame_length];
553 554
  //#ifndef __AVX2__
#if 1
555
  m10=(int16_t)m_10[2+frame_length];
556 557 558
#else
  m10=-(int16_t)m_10[2+frame_length];
#endif
559 560 561
#ifdef DEBUG_LOGMAP
  fprintf(fdsse4,"m11,m10 %d,%d\n",m11,m10);
#endif
562 563 564 565 566 567

  beta0 = -m11;//M0T_TERM;
  beta1 = m11;//M1T_TERM;
  m11=(int16_t)m_11[1+frame_length];
  m10=(int16_t)m_10[1+frame_length];

568 569 570
#ifdef DEBUG_LOGMAP
  fprintf(fdsse4,"m11,m10 %d,%d\n",m11,m10);
#endif
571 572 573 574 575 576 577

  beta0_2 = beta0-m11;//+M0T_TERM;
  beta1_2 = beta0+m11;//+M1T_TERM;
  beta2_2 = beta1+m10;//M2T_TERM;
  beta3_2 = beta1-m10;//+M3T_TERM;
  m11=(int16_t)m_11[frame_length];
  m10=(int16_t)m_10[frame_length];
578 579 580
#ifdef DEBUG_LOGMAP
  fprintf(fdsse4,"m11,m10 %d,%d\n",m11,m10);
#endif
581 582 583 584 585 586 587 588
  beta0_16 = beta0_2-m11;//+M0T_TERM;
  beta1_16 = beta0_2+m11;//+M1T_TERM;
  beta2_16 = beta1_2+m10;//+M2T_TERM;
  beta3_16 = beta1_2-m10;//+M3T_TERM;
  beta4_16 = beta2_2-m10;//+M4T_TERM;
  beta5_16 = beta2_2+m10;//+M5T_TERM;
  beta6_16 = beta3_2+m11;//+M6T_TERM;
  beta7_16 = beta3_2-m11;//+M7T_TERM;
589

590 591 592 593 594 595 596 597

  beta_m = (beta0_16>beta1_16) ? beta0_16 : beta1_16;
  beta_m = (beta_m>beta2_16) ? beta_m : beta2_16;
  beta_m = (beta_m>beta3_16) ? beta_m : beta3_16;
  beta_m = (beta_m>beta4_16) ? beta_m : beta4_16;
  beta_m = (beta_m>beta5_16) ? beta_m : beta5_16;
  beta_m = (beta_m>beta6_16) ? beta_m : beta6_16;
  beta_m = (beta_m>beta7_16) ? beta_m : beta7_16;
598

599 600 601 602 603 604 605 606 607 608

  beta0_16=beta0_16-beta_m;
  beta1_16=beta1_16-beta_m;
  beta2_16=beta2_16-beta_m;
  beta3_16=beta3_16-beta_m;
  beta4_16=beta4_16-beta_m;
  beta5_16=beta5_16-beta_m;
  beta6_16=beta6_16-beta_m;
  beta7_16=beta7_16-beta_m;

609
  for (rerun_flag=0;; rerun_flag=1) {
610
#if defined(__x86_64__) || defined(__i386__)
611 612
    beta_ptr   = (__m128i*)&beta[frame_length<<3];
    alpha128   = (__m128i*)&alpha[0];
613 614 615 616
#elif defined(__arm__)
    beta_ptr   = (int16x8_t*)&beta[frame_length<<3];
    alpha128   = (int16x8_t*)&alpha[0];
#endif
617 618 619 620 621 622 623 624 625
    if (rerun_flag == 0) {
      beta_ptr[0] = alpha128[(frame_length)];
      beta_ptr[1] = alpha128[1+(frame_length)];
      beta_ptr[2] = alpha128[2+(frame_length)];
      beta_ptr[3] = alpha128[3+(frame_length)];
      beta_ptr[4] = alpha128[4+(frame_length)];
      beta_ptr[5] = alpha128[5+(frame_length)];
      beta_ptr[6] = alpha128[6+(frame_length)];
      beta_ptr[7] = alpha128[7+(frame_length)];
626 627 628 629 630 631 632 633 634 635 636
#ifdef DEBUG_LOGMAP
      fprintf(fdsse4,"beta init \n");
      print_shorts("b0",(int16_t*)&beta_ptr[0]);
      print_shorts("b1",(int16_t*)&beta_ptr[1]);
      print_shorts("b2",(int16_t*)&beta_ptr[2]);
      print_shorts("b3",(int16_t*)&beta_ptr[3]);
      print_shorts("b4",(int16_t*)&beta_ptr[4]);
      print_shorts("b5",(int16_t*)&beta_ptr[5]);
      print_shorts("b6",(int16_t*)&beta_ptr[6]);
      print_shorts("b7",(int16_t*)&beta_ptr[7]);
#endif
637
    } else {
638
#if defined(__x86_64__) || defined(__i386__)
639 640 641 642 643 644 645 646 647
      beta128 = (__m128i*)&beta[0];
      beta_ptr[0] = _mm_srli_si128(beta128[0],2);
      beta_ptr[1] = _mm_srli_si128(beta128[1],2);
      beta_ptr[2] = _mm_srli_si128(beta128[2],2);
      beta_ptr[3] = _mm_srli_si128(beta128[3],2);
      beta_ptr[4] = _mm_srli_si128(beta128[4],2);
      beta_ptr[5] = _mm_srli_si128(beta128[5],2);
      beta_ptr[6] = _mm_srli_si128(beta128[6],2);
      beta_ptr[7] = _mm_srli_si128(beta128[7],2);
648 649 650 651 652 653 654 655 656 657 658
#elif defined(__arm__)
      beta128 = (int16x8_t*)&beta[0];
      beta_ptr   = (int16x8_t*)&beta[frame_length<<3];
      beta_ptr[0] = (int16x8_t)vshrq_n_s64((int64x2_t)beta128[0],16);   beta_ptr[0] = vsetq_lane_s16(beta[3],beta_ptr[0],4);
      beta_ptr[1] = (int16x8_t)vshrq_n_s64((int64x2_t)beta128[1],16);   beta_ptr[1] = vsetq_lane_s16(beta[11],beta_ptr[1],4);
      beta_ptr[2] = (int16x8_t)vshrq_n_s64((int64x2_t)beta128[2],16);   beta_ptr[2] = vsetq_lane_s16(beta[19],beta_ptr[2],4);
      beta_ptr[3] = (int16x8_t)vshrq_n_s64((int64x2_t)beta128[3],16);   beta_ptr[3] = vsetq_lane_s16(beta[27],beta_ptr[3],4);
      beta_ptr[4] = (int16x8_t)vshrq_n_s64((int64x2_t)beta128[4],16);   beta_ptr[4] = vsetq_lane_s16(beta[35],beta_ptr[4],4);
      beta_ptr[5] = (int16x8_t)vshrq_n_s64((int64x2_t)beta128[5],16);   beta_ptr[5] = vsetq_lane_s16(beta[43],beta_ptr[5],4);
      beta_ptr[6] = (int16x8_t)vshrq_n_s64((int64x2_t)beta128[6],16);   beta_ptr[6] = vsetq_lane_s16(beta[51],beta_ptr[6],4);
      beta_ptr[7] = (int16x8_t)vshrq_n_s64((int64x2_t)beta128[7],16);   beta_ptr[7] = vsetq_lane_s16(beta[59],beta_ptr[7],4);
659 660 661 662 663 664 665 666 667 668 669
#endif
#ifdef DEBUG_LOGMAP
      fprintf(fdsse4,"beta init (second run) \n");
      print_shorts("b0",(int16_t*)&beta_ptr[0]);
      print_shorts("b1",(int16_t*)&beta_ptr[1]);
      print_shorts("b2",(int16_t*)&beta_ptr[2]);
      print_shorts("b3",(int16_t*)&beta_ptr[3]);
      print_shorts("b4",(int16_t*)&beta_ptr[4]);
      print_shorts("b5",(int16_t*)&beta_ptr[5]);
      print_shorts("b6",(int16_t*)&beta_ptr[6]);
      print_shorts("b7",(int16_t*)&beta_ptr[7]);
670
#endif
671 672 673
    }


674
#if defined(__x86_64__) || defined(__i386__)
675 676 677 678 679 680 681 682
    beta_ptr[0] = _mm_insert_epi16(beta_ptr[0],beta0_16,7);
    beta_ptr[1] = _mm_insert_epi16(beta_ptr[1],beta1_16,7);
    beta_ptr[2] = _mm_insert_epi16(beta_ptr[2],beta2_16,7);
    beta_ptr[3] = _mm_insert_epi16(beta_ptr[3],beta3_16,7);
    beta_ptr[4] = _mm_insert_epi16(beta_ptr[4],beta4_16,7);
    beta_ptr[5] = _mm_insert_epi16(beta_ptr[5],beta5_16,7);
    beta_ptr[6] = _mm_insert_epi16(beta_ptr[6],beta6_16,7);
    beta_ptr[7] = _mm_insert_epi16(beta_ptr[7],beta7_16,7);
683 684 685 686 687 688 689 690 691 692
#elif defined(__arm__)
    beta_ptr[0] = vsetq_lane_s16(beta0_16,beta_ptr[0],7);
    beta_ptr[1] = vsetq_lane_s16(beta1_16,beta_ptr[1],7);
    beta_ptr[2] = vsetq_lane_s16(beta2_16,beta_ptr[2],7);
    beta_ptr[3] = vsetq_lane_s16(beta3_16,beta_ptr[3],7);
    beta_ptr[4] = vsetq_lane_s16(beta4_16,beta_ptr[4],7);
    beta_ptr[5] = vsetq_lane_s16(beta5_16,beta_ptr[5],7);
    beta_ptr[6] = vsetq_lane_s16(beta6_16,beta_ptr[6],7);
    beta_ptr[7] = vsetq_lane_s16(beta7_16,beta_ptr[7],7);
#endif
693

694 695 696 697 698 699 700 701 702 703 704
#ifdef DEBUG_LOGMAP
      fprintf(fdsse4,"beta init (after insert) \n");
      print_shorts("b0",(int16_t*)&beta_ptr[0]);
      print_shorts("b1",(int16_t*)&beta_ptr[1]);
      print_shorts("b2",(int16_t*)&beta_ptr[2]);
      print_shorts("b3",(int16_t*)&beta_ptr[3]);
      print_shorts("b4",(int16_t*)&beta_ptr[4]);
      print_shorts("b5",(int16_t*)&beta_ptr[5]);
      print_shorts("b6",(int16_t*)&beta_ptr[6]);
      print_shorts("b7",(int16_t*)&beta_ptr[7]);
#endif
705
    int loopval=((rerun_flag==0)?0:((frame_length-L)>>3));
706 707

    for (k=(frame_length>>3)-1; k>=loopval; k--) {
708
#if defined(__x86_64__) || defined(__i386__)
709
      m11_128=((__m128i*)m_11)[k];
710 711
      m10_128=((__m128i*)m_10)[k];

712 713 714

      //#ifndef __AVX2__
#if 1
715 716 717 718
      m_b0 = _mm_adds_epi16(beta_ptr[4],m11_128);  //m11
      m_b1 = _mm_subs_epi16(beta_ptr[4],m11_128);  //m00
      m_b2 = _mm_subs_epi16(beta_ptr[5],m10_128);  //m01
      m_b3 = _mm_adds_epi16(beta_ptr[5],m10_128);  //m10
719
      m_b4 = _mm_adds_epi16(beta_ptr[6],m10_128);  //m10
720 721 722
      m_b5 = _mm_subs_epi16(beta_ptr[6],m10_128);  //m01
      m_b6 = _mm_subs_epi16(beta_ptr[7],m11_128);  //m00
      m_b7 = _mm_adds_epi16(beta_ptr[7],m11_128);  //m11
723

724

725 726 727 728 729 730 731 732
      new0 = _mm_subs_epi16(beta_ptr[0],m11_128);  //m00
      new1 = _mm_adds_epi16(beta_ptr[0],m11_128);  //m11
      new2 = _mm_adds_epi16(beta_ptr[1],m10_128);  //m10
      new3 = _mm_subs_epi16(beta_ptr[1],m10_128);  //m01
      new4 = _mm_subs_epi16(beta_ptr[2],m10_128);  //m01
      new5 = _mm_adds_epi16(beta_ptr[2],m10_128);  //m10
      new6 = _mm_adds_epi16(beta_ptr[3],m11_128);  //m11
      new7 = _mm_subs_epi16(beta_ptr[3],m11_128);  //m00
733

734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752
#else
      b01=_mm256_load_si256(&((_m256i*)beta_ptr)[0]);
      b23=_mm256_load_si256(&((_m256i*)beta_ptr)[1]);
      b45=_mm256_load_si256(&((_m256i*)beta_ptr)[2]);
      b67=_mm256_load_si256(&((_m256i*)beta_ptr)[3]);
      m11m10_256 = _mm256_insertf128_si256(m11m10_256,m11_128,0);
      m11m10_256 = _mm256_insertf128_si256(m11m10_256,m10_128,1);


      m_b02 = _mm256_adds_epi16(b45,m11m10_256); //negative m10
      m_b13 = _mm256_subs_epi16(b45,m11m10_256); //negative m10
      m_b64 = _mm256_subs_epi16(b67,m11m10_256); //negative m10
      m_b75 = _mm256_adds_epi16(b67,m11m10_256); //negative m10
      new02 = _mm256_subs_epi16(b01,m11m10_256);  //negative m10
      new13 = _mm256_adds_epi16(b01,m11m10_256);  //negative m10
      new64 = _mm256_adds_epi16(b23,m11m10_256);  //negative m10
      new75 = _mm256_subs_epi16(b24,m11m10_256);  //negative m10
#endif

753
      beta_ptr-=8;
754

755 756
      //#ifndef __AVX2__
#if 1
757 758 759 760 761 762 763 764
      beta_ptr[0] = _mm_max_epi16(m_b0,new0);
      beta_ptr[1] = _mm_max_epi16(m_b1,new1);
      beta_ptr[2] = _mm_max_epi16(m_b2,new2);
      beta_ptr[3] = _mm_max_epi16(m_b3,new3);
      beta_ptr[4] = _mm_max_epi16(m_b4,new4);
      beta_ptr[5] = _mm_max_epi16(m_b5,new5);
      beta_ptr[6] = _mm_max_epi16(m_b6,new6);
      beta_ptr[7] = _mm_max_epi16(m_b7,new7);
765

766 767 768 769 770 771 772 773 774 775 776 777 778 779
      beta_max = _mm_max_epi16(beta_ptr[0],beta_ptr[1]);
      beta_max = _mm_max_epi16(beta_max   ,beta_ptr[2]);
      beta_max = _mm_max_epi16(beta_max   ,beta_ptr[3]);
      beta_max = _mm_max_epi16(beta_max   ,beta_ptr[4]);
      beta_max = _mm_max_epi16(beta_max   ,beta_ptr[5]);
      beta_max = _mm_max_epi16(beta_max   ,beta_ptr[6]);
      beta_max = _mm_max_epi16(beta_max   ,beta_ptr[7]);

      beta_ptr[0] = _mm_subs_epi16(beta_ptr[0],beta_max);
      beta_ptr[1] = _mm_subs_epi16(beta_ptr[1],beta_max);
      beta_ptr[2] = _mm_subs_epi16(beta_ptr[2],beta_max);
      beta_ptr[3] = _mm_subs_epi16(beta_ptr[3],beta_max);
      beta_ptr[4] = _mm_subs_epi16(beta_ptr[4],beta_max);
      beta_ptr[5] = _mm_subs_epi16(beta_ptr[5],beta_max);
780
      beta_ptr[6] = _mm_subs_epi16(beta_ptr[6],beta_max);
781
      beta_ptr[7] = _mm_subs_epi16(beta_ptr[7],beta_max);
782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803
#else
      b02   = _mm256_max_epi16(m_b02,new02);
      b13   = _mm256_max_epi16(m_b13,new13);
      b64   = _mm256_max_epi16(m_b64,new64);
      b75   = _mm256_max_epi16(m_b75,new75);
      
      beta_max = _mm256_max_epi16(b02,b13);
      beta_max = _mm256_max_epi16(beta_max,b64);
      beta_max = _mm256_max_epi16(beta_max,b75);
      beta_max = _mm256_max_epi16(beta_max,_mm256_permutevar8x32_epi32(betaa_max,_mm256_set_epi32(3,2,1,0,7,6,5,4)));

      b02   = _mm256_subs_epi16(b02,beta_max);
      b13   = _mm256_subs_epi16(b13,beta_max);
      b64   = _mm256_subs_epi16(b64,beta_max);
      b75   = _mm256_subs_epi16(b75,beta_max);

      ((_m256i*)beta_ptr)[0]) = _mm256_permute2x128_si256(b02,b13,0x02);  //b01
      ((_m256i*)beta_ptr)[1]) = _mm256_permute2x128_si256(b02,b13,0x31);  //b23
      ((_m256i*)beta_ptr)[2]) = _mm256_permute2x128_si256(b64,b75,0x13);  //b45
      ((_m256i*)beta_ptr)[3]) = _mm256_permute2x128_si256(b64,b75,0x20);  //b67
#endif

804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825
#elif defined(__arm__)
      m11_128=((int16x8_t*)m_11)[k];
      m10_128=((int16x8_t*)m_10)[k];
      m_b0 = vqaddq_s16(beta_ptr[4],m11_128);  //m11
      m_b1 = vqsubq_s16(beta_ptr[4],m11_128);  //m00
      m_b2 = vqsubq_s16(beta_ptr[5],m10_128);  //m01
      m_b3 = vqaddq_s16(beta_ptr[5],m10_128);  //m10
      m_b4 = vqaddq_s16(beta_ptr[6],m10_128);  //m10
      m_b5 = vqsubq_s16(beta_ptr[6],m10_128);  //m01
      m_b6 = vqsubq_s16(beta_ptr[7],m11_128);  //m00
      m_b7 = vqaddq_s16(beta_ptr[7],m11_128);  //m11

      new0 = vqsubq_s16(beta_ptr[0],m11_128);  //m00
      new1 = vqaddq_s16(beta_ptr[0],m11_128);  //m11
      new2 = vqaddq_s16(beta_ptr[1],m10_128);  //m10
      new3 = vqsubq_s16(beta_ptr[1],m10_128);  //m01
      new4 = vqsubq_s16(beta_ptr[2],m10_128);  //m01
      new5 = vqaddq_s16(beta_ptr[2],m10_128);  //m10
      new6 = vqaddq_s16(beta_ptr[3],m11_128);  //m11
      new7 = vqsubq_s16(beta_ptr[3],m11_128);  //m00

      beta_ptr-=8;
826

827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852
      beta_ptr[0] = vmaxq_s16(m_b0,new0);
      beta_ptr[1] = vmaxq_s16(m_b1,new1);
      beta_ptr[2] = vmaxq_s16(m_b2,new2);
      beta_ptr[3] = vmaxq_s16(m_b3,new3);
      beta_ptr[4] = vmaxq_s16(m_b4,new4);
      beta_ptr[5] = vmaxq_s16(m_b5,new5);
      beta_ptr[6] = vmaxq_s16(m_b6,new6);
      beta_ptr[7] = vmaxq_s16(m_b7,new7);

      beta_max = vmaxq_s16(beta_ptr[0],beta_ptr[1]);
      beta_max = vmaxq_s16(beta_max   ,beta_ptr[2]);
      beta_max = vmaxq_s16(beta_max   ,beta_ptr[3]);
      beta_max = vmaxq_s16(beta_max   ,beta_ptr[4]);
      beta_max = vmaxq_s16(beta_max   ,beta_ptr[5]);
      beta_max = vmaxq_s16(beta_max   ,beta_ptr[6]);
      beta_max = vmaxq_s16(beta_max   ,beta_ptr[7]);

      beta_ptr[0] = vqsubq_s16(beta_ptr[0],beta_max);
      beta_ptr[1] = vqsubq_s16(beta_ptr[1],beta_max);
      beta_ptr[2] = vqsubq_s16(beta_ptr[2],beta_max);
      beta_ptr[3] = vqsubq_s16(beta_ptr[3],beta_max);
      beta_ptr[4] = vqsubq_s16(beta_ptr[4],beta_max);
      beta_ptr[5] = vqsubq_s16(beta_ptr[5],beta_max);
      beta_ptr[6] = vqsubq_s16(beta_ptr[6],beta_max);
      beta_ptr[7] = vqsubq_s16(beta_ptr[7],beta_max);
#endif
853

854 855 856 857 858 859 860 861 862 863 864 865
#ifdef DEBUG_LOGMAP
      fprintf(fdsse4,"Loop index %d, mb\n",k);
      fprintf(fdsse4,"beta init (after max)\n");
      print_shorts("b0",(int16_t*)&beta_ptr[0]);
      print_shorts("b1",(int16_t*)&beta_ptr[1]);
      print_shorts("b2",(int16_t*)&beta_ptr[2]);
      print_shorts("b3",(int16_t*)&beta_ptr[3]);
      print_shorts("b4",(int16_t*)&beta_ptr[4]);
      print_shorts("b5",(int16_t*)&beta_ptr[5]);
      print_shorts("b6",(int16_t*)&beta_ptr[6]);
      print_shorts("b7",(int16_t*)&beta_ptr[7]);
#endif
866 867 868 869 870 871 872 873

    }

    if (rerun_flag==1)
      break;
  }
}

874 875
void compute_ext16(llr_t* alpha,llr_t* beta,llr_t* m_11,llr_t* m_10,llr_t* ext, llr_t* systematic,unsigned short frame_length)
{
876
#if defined(__x86_64__) || defined(__i386__)
877 878 879 880 881 882 883 884
  __m128i *alpha128=(__m128i *)alpha;
  __m128i *beta128=(__m128i *)beta;
  __m128i *m11_128,*m10_128,*ext_128;
  __m128i *alpha_ptr,*beta_ptr;
  __m128i m00_1,m00_2,m00_3,m00_4;
  __m128i m01_1,m01_2,m01_3,m01_4;
  __m128i m10_1,m10_2,m10_3,m10_4;
  __m128i m11_1,m11_2,m11_3,m11_4;
885 886 887 888 889 890 891 892 893 894 895
#elif defined(__arm__)
  int16x8_t *alpha128=(int16x8_t *)alpha;
  int16x8_t *beta128=(int16x8_t *)beta;
  int16x8_t *m11_128,*m10_128,*ext_128;
  int16x8_t *alpha_ptr,*beta_ptr;
  int16x8_t m00_1,m00_2,m00_3,m00_4;
  int16x8_t m01_1,m01_2,m01_3,m01_4;
  int16x8_t m10_1,m10_2,m10_3,m10_4;
  int16x8_t m11_1,m11_2,m11_3,m11_4;
#endif

896 897 898 899 900 901 902
  int k;

  //
  // LLR computation, 8 consequtive bits per loop
  //

#ifdef DEBUG_LOGMAP
903
  fprintf(fdsse4,"compute_ext (sse_16bit), %p, %p, %p, %p, %p, %p ,framelength %d\n",alpha,beta,m_11,m_10,ext,systematic,frame_length);
904 905 906 907 908 909
#endif

  alpha_ptr = alpha128;
  beta_ptr = &beta128[8];


910
  for (k=0; k<(frame_length>>3); k++) {
911

912
#if defined(__x86_64__) || defined(__i386__)
913 914 915
    m11_128        = (__m128i*)&m_11[k<<3];
    m10_128        = (__m128i*)&m_10[k<<3];
    ext_128        = (__m128i*)&ext[k<<3];
916

917
    /*
918
      fprintf(fdsse4,"EXT %03d\n",k);
919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934
      print_shorts("a0:",&alpha_ptr[0]);
      print_shorts("a1:",&alpha_ptr[1]);
      print_shorts("a2:",&alpha_ptr[2]);
      print_shorts("a3:",&alpha_ptr[3]);
      print_shorts("a4:",&alpha_ptr[4]);
      print_shorts("a5:",&alpha_ptr[5]);
      print_shorts("a6:",&alpha_ptr[6]);
      print_shorts("a7:",&alpha_ptr[7]);
      print_shorts("b0:",&beta_ptr[0]);
      print_shorts("b1:",&beta_ptr[1]);
      print_shorts("b2:",&beta_ptr[2]);
      print_shorts("b3:",&beta_ptr[3]);
      print_shorts("b4:",&beta_ptr[4]);
      print_shorts("b5:",&beta_ptr[5]);
      print_shorts("b6:",&beta_ptr[6]);
      print_shorts("b7:",&beta_ptr[7]);
935
    */
936 937 938

    //#ifndef __AVX2__
#if 1
939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954
    m00_4 = _mm_adds_epi16(alpha_ptr[7],beta_ptr[3]); //ALPHA_BETA_4m00;
    m11_4 = _mm_adds_epi16(alpha_ptr[7],beta_ptr[7]); //ALPHA_BETA_4m11;
    m00_3 = _mm_adds_epi16(alpha_ptr[6],beta_ptr[7]); //ALPHA_BETA_3m00;
    m11_3 = _mm_adds_epi16(alpha_ptr[6],beta_ptr[3]); //ALPHA_BETA_3m11;
    m00_2 = _mm_adds_epi16(alpha_ptr[1],beta_ptr[4]); //ALPHA_BETA_2m00;
    m11_2 = _mm_adds_epi16(alpha_ptr[1],beta_ptr[0]); //ALPHA_BETA_2m11;
    m11_1 = _mm_adds_epi16(alpha_ptr[0],beta_ptr[4]); //ALPHA_BETA_1m11;
    m00_1 = _mm_adds_epi16(alpha_ptr[0],beta_ptr[0]); //ALPHA_BETA_1m00;
    m01_4 = _mm_adds_epi16(alpha_ptr[5],beta_ptr[6]); //ALPHA_BETA_4m01;
    m10_4 = _mm_adds_epi16(alpha_ptr[5],beta_ptr[2]); //ALPHA_BETA_4m10;
    m01_3 = _mm_adds_epi16(alpha_ptr[4],beta_ptr[2]); //ALPHA_BETA_3m01;
    m10_3 = _mm_adds_epi16(alpha_ptr[4],beta_ptr[6]); //ALPHA_BETA_3m10;
    m01_2 = _mm_adds_epi16(alpha_ptr[3],beta_ptr[1]); //ALPHA_BETA_2m01;
    m10_2 = _mm_adds_epi16(alpha_ptr[3],beta_ptr[5]); //ALPHA_BETA_2m10;
    m10_1 = _mm_adds_epi16(alpha_ptr[2],beta_ptr[1]); //ALPHA_BETA_1m10;
    m01_1 = _mm_adds_epi16(alpha_ptr[2],beta_ptr[5]); //ALPHA_BETA_1m01;
955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980
#else


    m00_1 = _mm_adds_epi16(alpha_ptr[0],beta_ptr[0]); //ALPHA_BETA_1m00;
    m10_1 = _mm_adds_epi16(alpha_ptr[2],beta_ptr[1]); //ALPHA_BETA_1m10;
    m11_1 = _mm_adds_epi16(alpha_ptr[0],beta_ptr[4]); //ALPHA_BETA_1m11;
    m01_1 = _mm_adds_epi16(alpha_ptr[2],beta_ptr[5]); //ALPHA_BETA_1m01;

    m11_2 = _mm_adds_epi16(alpha_ptr[1],beta_ptr[0]); //ALPHA_BETA_2m11;
    m01_2 = _mm_adds_epi16(alpha_ptr[3],beta_ptr[1]); //ALPHA_BETA_2m01;
    m00_2 = _mm_adds_epi16(alpha_ptr[1],beta_ptr[4]); //ALPHA_BETA_2m00;
    m10_2 = _mm_adds_epi16(alpha_ptr[3],beta_ptr[5]); //ALPHA_BETA_2m10;

    m11_3 = _mm_adds_epi16(alpha_ptr[6],beta_ptr[3]); //ALPHA_BETA_3m11;
    m01_3 = _mm_adds_epi16(alpha_ptr[4],beta_ptr[2]); //ALPHA_BETA_3m01;
    m00_3 = _mm_adds_epi16(alpha_ptr[6],beta_ptr[7]); //ALPHA_BETA_3m00;
    m10_3 = _mm_adds_epi16(alpha_ptr[4],beta_ptr[6]); //ALPHA_BETA_3m10;

    m00_4 = _mm_adds_epi16(alpha_ptr[7],beta_ptr[3]); //ALPHA_BETA_4m00;
    m10_4 = _mm_adds_epi16(alpha_ptr[5],beta_ptr[2]); //ALPHA_BETA_4m10;
    m11_4 = _mm_adds_epi16(alpha_ptr[7],beta_ptr[7]); //ALPHA_BETA_4m11;
    m01_4 = _mm_adds_epi16(alpha_ptr[5],beta_ptr[6]); //ALPHA_BETA_4m01;


#endif

981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012
    /*
      print_shorts("m11_1:",&m11_1);
      print_shorts("m11_2:",&m11_2);
      print_shorts("m11_3:",&m11_3);
      print_shorts("m11_4:",&m11_4);
      print_shorts("m00_1:",&m00_1);
      print_shorts("m00_2:",&m00_2);
      print_shorts("m00_3:",&m00_3);
      print_shorts("m00_4:",&m00_4);
      print_shorts("m10_1:",&m10_1);
      print_shorts("m10_2:",&m10_2);
      print_shorts("m10_3:",&m10_3);
      print_shorts("m10_4:",&m10_4);
      print_shorts("m01_1:",&m01_1);
      print_shorts("m01_2:",&m01_2);
      print_shorts("m01_3:",&m01_3);
      print_shorts("m01_4:",&m01_4);
    */
    m01_1 = _mm_max_epi16(m01_1,m01_2);
    m01_1 = _mm_max_epi16(m01_1,m01_3);
    m01_1 = _mm_max_epi16(m01_1,m01_4);
    m00_1 = _mm_max_epi16(m00_1,m00_2);
    m00_1 = _mm_max_epi16(m00_1,m00_3);
    m00_1 = _mm_max_epi16(m00_1,m00_4);
    m10_1 = _mm_max_epi16(m10_1,m10_2);
    m10_1 = _mm_max_epi16(m10_1,m10_3);
    m10_1 = _mm_max_epi16(m10_1,m10_4);
    m11_1 = _mm_max_epi16(m11_1,m11_2);
    m11_1 = _mm_max_epi16(m11_1,m11_3);
    m11_1 = _mm_max_epi16(m11_1,m11_4);

    //      print_shorts("m11_1:",&m11_1);
1013

1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026
    m01_1 = _mm_subs_epi16(m01_1,*m10_128);
    m00_1 = _mm_subs_epi16(m00_1,*m11_128);
    m10_1 = _mm_adds_epi16(m10_1,*m10_128);
    m11_1 = _mm_adds_epi16(m11_1,*m11_128);

    //      print_shorts("m10_1:",&m10_1);
    //      print_shorts("m11_1:",&m11_1);
    m01_1 = _mm_max_epi16(m01_1,m00_1);
    m10_1 = _mm_max_epi16(m10_1,m11_1);
    //      print_shorts("m01_1:",&m01_1);
    //      print_shorts("m10_1:",&m10_1);

    *ext_128 = _mm_subs_epi16(m10_1,m01_1);
1027 1028 1029 1030 1031 1032 1033 1034
#ifdef DEBUG_LOGMAP
    fprintf(fdsse4,"ext %p\n",ext_128);
    print_shorts("ext:",(int16_t*)ext_128);
    print_shorts("m11:",(int16_t*)m11_128);
    print_shorts("m10:",(int16_t*)m10_128);
    print_shorts("m10_1:",(int16_t*)&m10_1);
    print_shorts("m01_1:",(int16_t*)&m01_1);
#endif
1035

1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
#elif defined(__arm__)
    m11_128        = (int16x8_t*)&m_11[k<<3];
    m10_128        = (int16x8_t*)&m_10[k<<3];
    ext_128        = (int16x8_t*)&ext[k<<3];

    m00_4 = vqaddq_s16(alpha_ptr[7],beta_ptr[3]); //ALPHA_BETA_4m00;
    m11_4 = vqaddq_s16(alpha_ptr[7],beta_ptr[7]); //ALPHA_BETA_4m11;
    m00_3 = vqaddq_s16(alpha_ptr[6],beta_ptr[7]); //ALPHA_BETA_3m00;
    m11_3 = vqaddq_s16(alpha_ptr[6],beta_ptr[3]); //ALPHA_BETA_3m11;
    m00_2 = vqaddq_s16(alpha_ptr[1],beta_ptr[4]); //ALPHA_BETA_2m00;
    m11_2 = vqaddq_s16(alpha_ptr[1],beta_ptr[0]); //ALPHA_BETA_2m11;
    m11_1 = vqaddq_s16(alpha_ptr[0],beta_ptr[4]); //ALPHA_BETA_1m11;
    m00_1 = vqaddq_s16(alpha_ptr[0],beta_ptr[0]); //ALPHA_BETA_1m00;
    m01_4 = vqaddq_s16(alpha_ptr[5],beta_ptr[6]); //ALPHA_BETA_4m01;
    m10_4 = vqaddq_s16(alpha_ptr[5],beta_ptr[2]); //ALPHA_BETA_4m10;
    m01_3 = vqaddq_s16(alpha_ptr[4],beta_ptr[2]); //ALPHA_BETA_3m01;
    m10_3 = vqaddq_s16(alpha_ptr[4],beta_ptr[6]); //ALPHA_BETA_3m10;
    m01_2 = vqaddq_s16(alpha_ptr[3],beta_ptr[1]); //ALPHA_BETA_2m01;
    m10_2 = vqaddq_s16(alpha_ptr[3],beta_ptr[5]); //ALPHA_BETA_2m10;
    m10_1 = vqaddq_s16(alpha_ptr[2],beta_ptr[1]); //ALPHA_BETA_1m10;
    m01_1 = vqaddq_s16(alpha_ptr[2],beta_ptr[5]); //ALPHA_BETA_1m01;

    m01_1 = vmaxq_s16(m01_1,m01_2);
    m01_1 = vmaxq_s16(m01_1,m01_3);
    m01_1 = vmaxq_s16(m01_1,m01_4);
    m00_1 = vmaxq_s16(m00_1,m00_2);
    m00_1 = vmaxq_s16(m00_1,m00_3);
    m00_1 = vmaxq_s16(m00_1,m00_4);
    m10_1 = vmaxq_s16(m10_1,m10_2);
    m10_1 = vmaxq_s16(m10_1,m10_3);
    m10_1 = vmaxq_s16(m10_1,m10_4);
    m11_1 = vmaxq_s16(m11_1,m11_2);
    m11_1 = vmaxq_s16(m11_1,m11_3);
    m11_1 = vmaxq_s16(m11_1,m11_4);


    m01_1 = vqsubq_s16(m01_1,*m10_128);
    m00_1 = vqsubq_s16(m00_1,*m11_128);
    m10_1 = vqaddq_s16(m10_1,*m10_128);
    m11_1 = vqaddq_s16(m11_1,*m11_128);


    m01_1 = vmaxq_s16(m01_1,m00_1);
    m10_1 = vmaxq_s16(m10_1,m11_1);


    *ext_128 = vqsubq_s16(m10_1,m01_1);
#endif
1084 1085 1086 1087 1088 1089 1090 1091 1092 1093
    alpha_ptr+=8;
    beta_ptr+=8;
  }
}



//int pi2[n],pi3[n+8],pi5[n+8],pi4[n+8],pi6[n+8],
int *pi2tab16[188],*pi5tab16[188],*pi4tab16[188],*pi6tab16[188];

1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
void free_td16(void)
{
  int ind;

  for (ind=0; ind<188; ind++) {
    free(pi2tab16[ind]);
    free(pi5tab16[ind]);
    free(pi4tab16[ind]);
    free(pi6tab16[ind]);
  }
1104 1105
}

1106 1107
void init_td16()
{
1108 1109 1110 1111

  int ind,i,i2,i3,j,n,pi,pi3;
  short * base_interleaver;

1112
  for (ind=0; ind<188; ind++) {
1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128

    n = f1f2mat[ind].nb_bits;
    base_interleaver=il_tb+f1f2mat[ind].beg_index;
#ifdef MEX
    // This is needed for the Mex implementation to make the memory persistent
    pi2tab16[ind] = mxMalloc((n+8)*sizeof(int));
    pi5tab16[ind] = mxMalloc((n+8)*sizeof(int));
    pi4tab16[ind] = mxMalloc((n+8)*sizeof(int));
    pi6tab16[ind] = mxMalloc((n+8)*sizeof(int));
#else
    pi2tab16[ind] = malloc((n+8)*sizeof(int));
    pi5tab16[ind] = malloc((n+8)*sizeof(int));
    pi4tab16[ind] = malloc((n+8)*sizeof(int));
    pi6tab16[ind] = malloc((n+8)*sizeof(int));
#endif

1129
    for (i=i2=0; i2<8; i2++) {
1130
      j=i2;
1131 1132 1133 1134 1135 1136 1137

      for (i3=0; i3<(n>>3); i3++,i++,j+=8) {

        //    if (j>=n)
        //      j-=(n-1);

        pi2tab16[ind][i]  = j;
1138
        //    fprintf(fdsse4,"pi2[%d] = %d\n",i,j);
1139 1140
      }
    }
1141 1142

    for (i=0; i<n; i++) {
1143 1144 1145 1146 1147
      pi = base_interleaver[i];//(unsigned int)threegpplte_interleaver(f1,f2,n);
      pi3 = pi2tab16[ind][pi];
      pi4tab16[ind][pi2tab16[ind][i]] = pi3;
      pi5tab16[ind][pi3] = pi2tab16[ind][i];
      pi6tab16[ind][pi] = pi2tab16[ind][i];
1148
    }
1149 1150 1151 1152 1153

  }
}

unsigned char phy_threegpplte_turbo_decoder16(short *y,
1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169
    unsigned char *decoded_bytes,
    unsigned short n,
    unsigned short f1,
    unsigned short f2,
    unsigned char max_iterations,
    unsigned char crc_type,
    unsigned char F,
    time_stats_t *init_stats,
    time_stats_t *alpha_stats,
    time_stats_t *beta_stats,
    time_stats_t *gamma_stats,
    time_stats_t *ext_stats,
    time_stats_t *intl1_stats,
    time_stats_t *intl2_stats)
{

1170 1171 1172 1173 1174
  /*  y is a pointer to the input
      decoded_bytes is a pointer to the decoded output
      n is the size in bits of the coded block, with the tail */


1175 1176 1177 1178 1179
  llr_t systematic0[n+16] __attribute__ ((aligned(32)));
  llr_t systematic1[n+16] __attribute__ ((aligned(32)));
  llr_t systematic2[n+16] __attribute__ ((aligned(32)));
  llr_t yparity1[n+16] __attribute__ ((aligned(32)));
  llr_t yparity2[n+16] __attribute__ ((aligned(32)));
1180

1181 1182
  llr_t ext[n+128] __attribute__((aligned(32)));
  llr_t ext2[n+128] __attribute__((aligned(32)));
1183

1184 1185 1186 1187
  llr_t alpha[(n+16)*8] __attribute__ ((aligned(32)));
  llr_t beta[(n+16)*8] __attribute__ ((aligned(32)));
  llr_t m11[n+32] __attribute__ ((aligned(32)));
  llr_t m10[n+32] __attribute__ ((aligned(32)));
1188

1189

1190 1191 1192 1193 1194
  int *pi2_p,*pi4_p,*pi5_p,*pi6_p;
  llr_t *s,*s1,*s2,*yp1,*yp2,*yp;
  unsigned int i,j,iind;//,pi;
  unsigned char iteration_cnt=0;
  unsigned int crc,oldcrc,crc_len;
1195
  uint8_t temp;
1196

1197 1198
#if defined(__x86_64__) || defined(__i386__)
  __m128i *yp128;
1199
  __m128i tmp, zeros=_mm_setzero_si128();
1200
  __m128i tmpe;
1201 1202 1203 1204 1205 1206 1207 1208
#elif defined(__arm__)
  int16x8_t *yp128;
//  int16x8_t tmp128[(n+8)>>3];
  int16x8_t tmp, zeros=vdupq_n_s16(0);
  const uint16_t __attribute__ ((aligned (16))) _Powers[8]= 
    { 1, 2, 4, 8, 16, 32, 64, 128};
  uint16x8_t Powers= vld1q_u16(_Powers);
#endif
1209 1210
  int offset8_flag=0;

1211 1212 1213 1214 1215 1216 1217
#ifdef DEBUG_LOGMAP
  fdsse4 = fopen("dump_sse4.txt","w");


  printf("tc sse4_16 (y) %p\n",y);
#endif

1218
  if (crc_type > 3) {
1219
    printf("Illegal crc length!\n");
1220 1221 1222 1223
    return 255;
  }


1224

1225 1226 1227 1228
  start_meas(init_stats);



1229
  for (iind=0; iind < 188 && f1f2mat[iind].nb_bits != n; iind++);
1230

1231
  if ( iind == 188 ) {
1232
    printf("Illegal frame length!\n");
1233 1234
    return 255;
  }
1235

1236 1237 1238 1239 1240
  switch (crc_type) {
  case CRC24_A:
  case CRC24_B:
    crc_len=3;
    break;
1241

1242 1243 1244
  case CRC16:
    crc_len=2;
    break;
1245

1246 1247 1248 1249
  case CRC8:
    crc_len=1;
    break;

1250 1251 1252 1253
  default:
    crc_len=3;
  }

1254
#if defined(__x86_64__) || defined(__i386__)
1255
  yp128 = (__m128i*)y;
1256 1257 1258
#elif defined(__arm__)
  yp128 = (int16x8_t*)y;
#endif
1259 1260 1261



1262 1263

  s = systematic0;
1264 1265 1266 1267 1268 1269 1270
  s1 = systematic1;
  s2 = systematic2;
  yp1 = yparity1;
  yp2 = yparity2;



1271
  for (i=0; i<n; i+=8) {
1272
    pi2_p = &pi2tab16[iind][i];
1273

1274 1275
    j=pi2_p[0];

1276
#if defined(__x86_64__) || defined(__i386__)
1277
    tmpe = _mm_load_si128(yp128);
1278 1279
    //    fprintf(fdsse4,"yp128 %p\n",yp128);
    //    print_shorts("tmpe",(int16_t *)&tmpe);
1280

1281 1282
    s[j]   = _mm_extract_epi16(tmpe,0);
    yp1[j] = _mm_extract_epi16(tmpe,1);
1283
    yp2[j] = _mm_extract_epi16(tmpe,2);
1284 1285 1286
#ifdef DEBUG_LOGMAP
    fprintf(fdsse4,"init0: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
#endif
1287 1288
    j=pi2_p[1];

1289 1290
    s[j]   = _mm_extract_epi16(tmpe,3);
    yp1[j] = _mm_extract_epi16(tmpe,4);
1291
    yp2[j] = _mm_extract_epi16(tmpe,5);
1292 1293 1294
#ifdef DEBUG_LOGMAP
    fprintf(fdsse4,"init1: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
#endif
1295 1296
    j=pi2_p[2];

1297 1298
    s[j]   = _mm_extract_epi16(tmpe,6);
    yp1[j] = _mm_extract_epi16(tmpe,7);
1299 1300
    tmpe = _mm_load_si128(&yp128[1]);
    yp2[j] = _mm_extract_epi16(tmpe,0);
1301 1302 1303
#ifdef DEBUG_LOGMAP
    fprintf(fdsse4,"init2: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
#endif
1304 1305
    j=pi2_p[3];

1306 1307
    s[j]   = _mm_extract_epi16(tmpe,1);
    yp1[j] = _mm_extract_epi16(tmpe,2);
1308
    yp2[j] = _mm_extract_epi16(tmpe,3);
1309 1310 1311
#ifdef DEBUG_LOGMAP
    fprintf(fdsse4,"init3: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
#endif
1312 1313
    j=pi2_p[4];

1314 1315
    s[j]   = _mm_extract_epi16(tmpe,4);
    yp1[j] = _mm_extract_epi16(tmpe,5);
1316
    yp2[j] = _mm_extract_epi16(tmpe,6);
1317 1318 1319
#ifdef DEBUG_LOGMAP
    fprintf(fdsse4,"init4: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
#endif
1320 1321
    j=pi2_p[5];

1322
    s[j]   = _mm_extract_epi16(tmpe,7);
1323
    tmpe = _mm_load_si128(&yp128[2]);
1324
    yp1[j] = _mm_extract_epi16(tmpe,0);
1325
    yp2[j] = _mm_extract_epi16(tmpe,1);
1326 1327 1328
#ifdef DEBUG_LOGMAP
    fprintf(fdsse4,"init5: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
#endif
1329

1330 1331
    j=pi2_p[6];

1332 1333
    s[j]   = _mm_extract_epi16(tmpe,2);
    yp1[j] = _mm_extract_epi16(tmpe,3);
1334
    yp2[j] = _mm_extract_epi16(tmpe,4);
1335 1336 1337
#ifdef DEBUG_LOGMAP
    fprintf(fdsse4,"init6: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
#endif
1338 1339
    j=pi2_p[7];

1340 1341
    s[j]   = _mm_extract_epi16(tmpe,5);
    yp1[j] = _mm_extract_epi16(tmpe,6);
1342
    yp2[j] = _mm_extract_epi16(tmpe,7);
1343 1344 1345
#ifdef DEBUG_LOGMAP
    fprintf(fdsse4,"init7: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
#endif
1346

1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386
#elif defined(__arm__)
    s[j]   = vgetq_lane_s16(yp128[0],0);
    yp1[j] = vgetq_lane_s16(yp128[0],1);
    yp2[j] = vgetq_lane_s16(yp128[0],2);

    j=pi2_p[1];
    s[j]   = vgetq_lane_s16(yp128[0],3);
    yp1[j] = vgetq_lane_s16(yp128[0],4);
    yp2[j] = vgetq_lane_s16(yp128[0],5);

    j=pi2_p[2];
    s[j]   = vgetq_lane_s16(yp128[0],6);
    yp1[j] = vgetq_lane_s16(yp128[0],7);
    yp2[j] = vgetq_lane_s16(yp128[1],0);

    j=pi2_p[3];
    s[j]   = vgetq_lane_s16(yp128[1],1);
    yp1[j] = vgetq_lane_s16(yp128[1],2);
    yp2[j] = vgetq_lane_s16(yp128[1],3);

    j=pi2_p[4];
    s[j]   = vgetq_lane_s16(yp128[1],4);
    yp1[j] = vgetq_lane_s16(yp128[1],5);
    yp2[j] = vgetq_lane_s16(yp128[1],6);

    j=pi2_p[5];
    s[j]   = vgetq_lane_s16(yp128[1],7);
    yp1[j] = vgetq_lane_s16(yp128[2],0);
    yp2[j] = vgetq_lane_s16(yp128[2],1);

    j=pi2_p[6];
    s[j]   = vgetq_lane_s16(yp128[2],2);
    yp1[j] = vgetq_lane_s16(yp128[2],3);
    yp2[j] = vgetq_lane_s16(yp128[2],4);

    j=pi2_p[7];
    s[j]   = vgetq_lane_s16(yp128[2],5);
    yp1[j] = vgetq_lane_s16(yp128[2],6);
    yp2[j] = vgetq_lane_s16(yp128[2],7);
#endif
1387 1388 1389 1390 1391 1392 1393 1394
    yp128+=3;

  }

  yp=(llr_t*)yp128;


  // Termination
1395 1396 1397 1398 1399 1400 1401
  for (i=n; i<n+3; i++) {
    s[i]= *yp;
    s1[i] = s[i] ;
    s2[i] = s[i];
    yp++;
    yp1[i] = *yp;
    yp++;
1402
#ifdef DEBUG_LOGMAP
1403
    fprintf(fdsse4