3gpplte_turbo_decoder_sse_16bit.c 58.5 KB
Newer Older
1
/*******************************************************************************
2
    OpenAirInterface
ghaddab's avatar
ghaddab committed
3
    Copyright(c) 1999 - 2014 Eurecom
4

ghaddab's avatar
ghaddab committed
5 6 7 8
    OpenAirInterface is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation, either version 3 of the License, or
    (at your option) any later version.
9 10


ghaddab's avatar
ghaddab committed
11 12 13 14
    OpenAirInterface is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.
15

ghaddab's avatar
ghaddab committed
16
    You should have received a copy of the GNU General Public License
17 18
    along with OpenAirInterface.The full GNU General Public License is
   included in this distribution in the file called "COPYING". If not,
ghaddab's avatar
ghaddab committed
19
   see <http://www.gnu.org/licenses/>.
20 21

  Contact Information
ghaddab's avatar
ghaddab committed
22 23
  OpenAirInterface Admin: openair_admin@eurecom.fr
  OpenAirInterface Tech : openair_tech@eurecom.fr
24
  OpenAirInterface Dev  : openair4g-devel@lists.eurecom.fr
25

ghaddab's avatar
ghaddab committed
26
  Address      : Eurecom, Campus SophiaTech, 450 Route des Chappes, CS 50193 - 06904 Biot Sophia Antipolis cedex, FRANCE
27

ghaddab's avatar
ghaddab committed
28
 *******************************************************************************/
29 30 31
/* 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)
32
   date: 21.10.2009
33 34 35

   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!
36

37 38 39 40 41 42 43 44 45 46
   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
47
#undef __AVX2__
48
#include "PHY/sse_intrin.h"
49

50
#ifndef TEST_DEBUG
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
#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

67 68 69 70 71
//#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
72

73

74

75 76 77
#ifdef DEBUG_LOGMAP
FILE *fdsse4;
#endif
78

79 80 81 82
typedef int16_t llr_t; // internal decoder LLR data is 16-bit fixed
typedef int16_t channel_t;
#define MAX 256

83 84
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);
85 86 87 88 89 90 91
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,
92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107
               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)
{

108
#ifdef DEBUG_LOGMAP
109
  fprintf(fdsse4,"log_map, frame_length %d\n",frame_length);
110 111
#endif

112 113 114 115 116 117 118 119 120 121 122 123
  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);
124 125 126 127 128


}

void compute_gamma16(llr_t* m11,llr_t* m10,llr_t* systematic,channel_t* y_parity,
129 130
                     unsigned short frame_length,unsigned char term_flag)
{
131
  int k,K1;
132
#if defined(__x86_64__)||defined(__i386__)
133 134 135 136
  __m128i *systematic128 = (__m128i *)systematic;
  __m128i *y_parity128   = (__m128i *)y_parity;
  __m128i *m10_128        = (__m128i *)m10;
  __m128i *m11_128        = (__m128i *)m11;
137 138 139 140 141 142
#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
143 144

#ifdef DEBUG_LOGMAP
145
  fprintf(fdsse4,"compute_gamma (sse_16bit), %p,%p,%p,%p,framelength %d\n",m11,m10,systematic,y_parity,frame_length);
146 147
#endif

148
#ifndef __AVX2__
149
  K1=frame_length>>3;
150 151 152 153 154 155
#else
  if ((frame_length&15) > 0)
    K1=(frame_length+1)>>4;
  else
    K1=frame_length>>4;
#endif
156 157

  for (k=0; k<K1; k++) {
158
#if defined(__x86_64__) || defined(__i386__)
159
#ifndef __AVX2__
160 161
    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);
162 163 164 165 166
#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
167 168 169 170
#elif defined(__arm__)
    m11_128[k] = vhaddq_s16(systematic128[k],y_parity128[k]);
    m10_128[k] = vhsubq_s16(systematic128[k],y_parity128[k]);
#endif
171

172
#ifdef DEBUG_LOGMAP
173
    fprintf(fdsse4,"Loop index k %d\n", k);
174 175 176 177 178
    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
179
  }
180

181
  k=frame_length>>3;
182
  // Termination
183
#if defined(__x86_64__) || defined(__i386__)
184
  m11_128[k] = _mm_srai_epi16(_mm_adds_epi16(systematic128[k+term_flag],y_parity128[k]),1);
185 186
  //#ifndef __AVX2__
#if 1
187
  m10_128[k] = _mm_srai_epi16(_mm_subs_epi16(systematic128[k+term_flag],y_parity128[k]),1);
188 189 190
#else
  m10_128[k] = _mm_srai_epi16(_mm_subs_epi16(y_parity128[k],systematic128[k+term_flag]),1);
#endif
191 192
#elif defined(__arm__)
  m11_128[k] = vhaddq_s16(systematic128[k+term_flag],y_parity128[k]);
193
  m10_128[k] = vhsubq_s16(systematic128[k+term_flag],y_parity128[k]);
194 195 196 197 198 199 200 201
#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]);
202
#endif
203 204 205 206
}

#define L 40

207 208
void compute_alpha16(llr_t* alpha,llr_t* beta,llr_t* m_11,llr_t* m_10,unsigned short frame_length,unsigned char F)
{
209
  int k,l,l2,K1,rerun_flag=0;
210
#if defined(__x86_64__) || defined(__i386__)
211 212 213 214
  __m128i *alpha128=(__m128i *)alpha,*alpha_ptr,*m11p,*m10p;
  //#ifndef __AVX2__
#if 1
  __m128i a0,a1,a2,a3,a4,a5,a6,a7;
215 216 217
  __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;
218 219 220 221 222 223 224 225
#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

226 227 228 229 230 231 232
#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
233 234
  l2 = L>>3;
  K1 = (frame_length>>3);
235 236 237
#ifdef DEBUG_LOGMAP
  fprintf(fdsse4,"compute_alpha (sse_16bit)\n");
#endif
238
  for (l=K1;; l=l2,rerun_flag=1) {
239
#if defined(__x86_64__) || defined(__i386__)
240
    alpha128 = (__m128i *)alpha;
241 242 243 244
    //#ifdef __AVX2__
#if 0
    alpha256 = (__m256i *)alpha;
#endif
245 246 247
#elif defined(__arm__)
    alpha128 = (int16x8_t *)alpha;
#endif
248

249
    if (rerun_flag == 0) {
250
#if defined(__x86_64__) || defined(__i386__)
251 252 253 254 255 256 257 258
      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);
259 260 261 262 263 264 265 266 267 268 269 270
#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
271
      fprintf(fdsse4,"Initial alpha\n");
272 273 274 275 276 277 278 279 280
      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
281
    } else {
282
      //set initial alpha in columns 1-7 from final alpha from last run in columns 0-6
283
#if defined(__x86_64__) || defined(__i386__)
284 285 286 287 288 289 290 291
      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);
292 293 294 295 296 297 298 299 300 301
#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
302 303 304 305 306 307 308 309
      // 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;
310
#ifdef DEBUG_LOGMAP
311
      fprintf(fdsse4,"Second run\n");
312 313 314 315 316 317 318 319 320 321
      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

322
    }
323

324
    alpha_ptr = &alpha128[0];
325 326 327 328 329
    //#ifdef __AVX2__
#if 0
    alpha_ptr256 = &alpha256[0];
#endif

330
#if defined(__x86_64__) || defined(__i386__)
331 332
    m11p = (__m128i*)m_11;
    m10p = (__m128i*)m_10;
333 334 335 336
#elif defined(__arm__)
    m11p = (int16x8_t*)m_11;
    m10p = (int16x8_t*)m_10;
#endif
337
    for (k=0;
338 339
         k<l;
         k++) {
340

341
#if defined(__x86_64__) || defined(__i386__)
342 343
      //#ifndef __AVX2__
#if 1
344 345 346 347 348 349
      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
350
      m_b4 = _mm_subs_epi16(a1,*m11p);  // m00=-m11
351 352 353 354 355 356 357 358 359 360 361
      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]);
362

363 364 365 366 367 368 369 370
      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
371

372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387
      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);
388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418
#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
419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455
#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
456 457

      alpha_ptr+=8;
458 459 460 461
      //#ifdef __AVX2__
#if 0
      alpha_ptr256+=4;
#endif
462
      m11p++;
463
      m10p++;
464
#if defined(__x86_64__) || defined(__i386__)
465 466
      //#ifndef __AVX2__
#if 1
467 468 469 470 471 472 473 474
      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);
475 476 477 478 479 480 481 482 483 484 485 486
#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
487 488 489 490 491 492 493 494 495 496 497 498
#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
499
      fprintf(fdsse4,"Loop index %d\n",k);
500 501 502 503 504 505 506 507 508
      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);

509
      fprintf(fdsse4,"Loop index %d, new\n",k);
510 511 512 513 514 515 516 517 518
      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);

519
      fprintf(fdsse4,"Loop index %d, after max\n",k);
520 521 522 523 524 525 526 527 528
      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);

529
      fprintf(fdsse4,"Loop index %d\n",k);
530 531 532 533 534 535 536 537 538
      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
539 540 541 542 543

    }

    if (rerun_flag==1)
      break;
544
  }
545 546 547
}


548 549 550
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)
{

551
  int k,rerun_flag=0;
552
#if defined(__x86_64__) || defined(__i386__)
553 554 555 556 557
  __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;
558
  __m128i beta_max;
559 560 561 562 563 564 565 566 567
#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

568
  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;
569 570 571
  llr_t beta0,beta1;

#ifdef DEBUG_LOGMAP
572 573
  fprintf(fdsse4,"compute_beta, %p,%p,%p,%p,framelength %d,F %d\n",
	  beta,m_11,m_10,alpha,frame_length,F);
574 575 576 577 578
#endif


  // termination for beta initialization

579
  //  fprintf(fdsse4,"beta init: offset8 %d\n",offset8_flag);
580
  m11=(int16_t)m_11[2+frame_length];
581 582
  //#ifndef __AVX2__
#if 1
583
  m10=(int16_t)m_10[2+frame_length];
584 585 586
#else
  m10=-(int16_t)m_10[2+frame_length];
#endif
587 588 589
#ifdef DEBUG_LOGMAP
  fprintf(fdsse4,"m11,m10 %d,%d\n",m11,m10);
#endif
590 591 592 593 594 595

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

596 597 598
#ifdef DEBUG_LOGMAP
  fprintf(fdsse4,"m11,m10 %d,%d\n",m11,m10);
#endif
599 600 601 602 603 604 605

  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];
606 607 608
#ifdef DEBUG_LOGMAP
  fprintf(fdsse4,"m11,m10 %d,%d\n",m11,m10);
#endif
609 610 611 612 613 614 615 616
  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;
617

618 619 620 621 622 623 624 625

  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;
626

627 628 629 630 631 632 633 634 635 636

  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;

637
  for (rerun_flag=0;; rerun_flag=1) {
638
#if defined(__x86_64__) || defined(__i386__)
639 640
    beta_ptr   = (__m128i*)&beta[frame_length<<3];
    alpha128   = (__m128i*)&alpha[0];
641 642 643 644
#elif defined(__arm__)
    beta_ptr   = (int16x8_t*)&beta[frame_length<<3];
    alpha128   = (int16x8_t*)&alpha[0];
#endif
645 646 647 648 649 650 651 652 653
    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)];
654 655 656 657 658 659 660 661 662 663 664
#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
665
    } else {
666
#if defined(__x86_64__) || defined(__i386__)
667 668 669 670 671 672 673 674 675
      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);
676 677 678 679 680 681 682 683 684 685 686
#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);
687 688 689 690 691 692 693 694 695 696 697
#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]);
698
#endif
699 700 701
    }


702
#if defined(__x86_64__) || defined(__i386__)
703 704 705 706 707 708 709 710
    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);
711 712 713 714 715 716 717 718 719 720
#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
721

722 723 724 725 726 727 728 729 730 731 732
#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
733
    int loopval=((rerun_flag==0)?0:((frame_length-L)>>3));
734 735

    for (k=(frame_length>>3)-1; k>=loopval; k--) {
736
#if defined(__x86_64__) || defined(__i386__)
737
      m11_128=((__m128i*)m_11)[k];
738 739
      m10_128=((__m128i*)m_10)[k];

740 741 742

      //#ifndef __AVX2__
#if 1
743 744 745 746
      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
747
      m_b4 = _mm_adds_epi16(beta_ptr[6],m10_128);  //m10
748 749 750
      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
751

752

753 754 755 756 757 758 759 760
      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
761

762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780
#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

781
      beta_ptr-=8;
782

783 784
      //#ifndef __AVX2__
#if 1
785 786 787 788 789 790 791 792
      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);
793

794 795 796 797 798 799 800 801 802 803 804 805 806 807
      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);
808
      beta_ptr[6] = _mm_subs_epi16(beta_ptr[6],beta_max);
809
      beta_ptr[7] = _mm_subs_epi16(beta_ptr[7],beta_max);
810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831
#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

832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853
#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;
854

855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880
      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
881

882 883 884 885 886 887 888 889 890 891 892 893
#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
894 895 896 897 898 899 900 901

    }

    if (rerun_flag==1)
      break;
  }
}

902 903
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)
{
904
#if defined(__x86_64__) || defined(__i386__)
905 906 907 908 909 910 911 912
  __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;
913 914 915 916 917 918 919 920 921 922 923
#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

924 925 926 927 928 929 930
  int k;

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

#ifdef DEBUG_LOGMAP
931
  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);
932 933 934 935 936 937
#endif

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


938
  for (k=0; k<(frame_length>>3); k++) {
939

940
#if defined(__x86_64__) || defined(__i386__)
941 942 943
    m11_128        = (__m128i*)&m_11[k<<3];
    m10_128        = (__m128i*)&m_10[k<<3];
    ext_128        = (__m128i*)&ext[k<<3];
944

945
    /*
946
      fprintf(fdsse4,"EXT %03d\n",k);
947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962
      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]);
963
    */
964 965 966

    //#ifndef __AVX2__
#if 1
967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982
    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;
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
#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

1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040
    /*
      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);
1041

1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054
    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);
1055 1056 1057 1058 1059 1060 1061 1062
#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
1063

1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111
#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
1112 1113 1114 1115 1116 1117 1118 1119 1120 1121
    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];

1122 1123 1124 1125 1126 1127 1128 1129 1130 1131
void free_td16(void)
{
  int ind;

  for (ind=0; ind<188; ind++) {
    free(pi2tab16[ind]);
    free(pi5tab16[ind]);
    free(pi4tab16[ind]);
    free(pi6tab16[ind]);
  }
1132 1133
}

1134 1135
void init_td16()
{
1136 1137 1138 1139

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

1140
  for (ind=0; ind<188; ind++) {
1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155