3gpplte_turbo_decoder_sse_16bit.c 49 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
*/

///
///

47
#include "PHY/sse_intrin.h"
48

49
#ifndef TEST_DEBUG
50
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
#define print_shorts(s,x) printf("%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])

69
70
//#define DEBUG_LOGMAP

71

72
73
74
75
typedef int16_t llr_t; // internal decoder LLR data is 16-bit fixed
typedef int16_t channel_t;
#define MAX 256

76
77
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);
78
79
80
81
82
83
84
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,
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
               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)
{

101
102
103
104
#ifdef DEBUG_LOGMAP
  msg("log_map, frame_length %d\n",frame_length);
#endif

105
106
107
108
109
110
111
112
113
114
115
116
  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);
117
118
119
120
121


}

void compute_gamma16(llr_t* m11,llr_t* m10,llr_t* systematic,channel_t* y_parity,
122
123
                     unsigned short frame_length,unsigned char term_flag)
{
124
  int k,K1;
125
#if defined(__x86_64__)||defined(__i386__)
126
127
128
129
  __m128i *systematic128 = (__m128i *)systematic;
  __m128i *y_parity128   = (__m128i *)y_parity;
  __m128i *m10_128        = (__m128i *)m10;
  __m128i *m11_128        = (__m128i *)m11;
130
131
132
133
134
135
#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
136
137
138
139
140
141

#ifdef DEBUG_LOGMAP
  msg("compute_gamma, %p,%p,%p,%p,framelength %d\n",m11,m10,systematic,y_parity,frame_length);
#endif

  K1=frame_length>>3;
142
143

  for (k=0; k<K1; k++) {
144
#if defined(__x86_64__) || defined(__i386__)
145
146
    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);
147
148
149
150
#elif defined(__arm__)
    m11_128[k] = vhaddq_s16(systematic128[k],y_parity128[k]);
    m10_128[k] = vhsubq_s16(systematic128[k],y_parity128[k]);
#endif
151

152
153
154
155
156
157
158
#ifdef DEBUG_LOGMAP
    printf("Loop index k, m11,m10\n");
    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
159
  }
160

161
  // Termination
162
#if defined(__x86_64__) || defined(__i386__)
163
164
  m11_128[k] = _mm_srai_epi16(_mm_adds_epi16(systematic128[k+term_flag],y_parity128[k]),1);
  m10_128[k] = _mm_srai_epi16(_mm_subs_epi16(systematic128[k+term_flag],y_parity128[k]),1);
165
166
167
168
#elif defined(__arm__)
  m11_128[k] = vhaddq_s16(systematic128[k+term_flag],y_parity128[k]);
  m10_128[k] = vhsubq_s16(systematic128[k+term_flag],y_parity128[k]);
#endif
169
170
171
172
}

#define L 40

173
174
void compute_alpha16(llr_t* alpha,llr_t* beta,llr_t* m_11,llr_t* m_10,unsigned short frame_length,unsigned char F)
{
175
  int k,l,l2,K1,rerun_flag=0;
176
#if defined(__x86_64__) || defined(__i386__)
177
178
179
180
181
  __m128i *alpha128=(__m128i *)alpha,*alpha_ptr;
  __m128i a0,a1,a2,a3,a4,a5,a6,a7,*m11p,*m10p;
  __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;
182
183
184
185
186
187
188
#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
189
190
191
  l2 = L>>3;
  K1 = (frame_length>>3);

192
  for (l=K1;; l=l2,rerun_flag=1) {
193
#if defined(__x86_64__) || defined(__i386__)
194
    alpha128 = (__m128i *)alpha;
195
196
197
#elif defined(__arm__)
    alpha128 = (int16x8_t *)alpha;
#endif
198

199
    if (rerun_flag == 0) {
200
#if defined(__x86_64__) || defined(__i386__)
201
202
203
204
205
206
207
208
      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);
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
#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
      printf("Initial alpha\n");
      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
231
    } else {
232
      //set initial alpha in columns 1-7 from final alpha from last run in columns 0-6
233
#if defined(__x86_64__) || defined(__i386__)
234
235
236
237
238
239
240
241
      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);
242
243
244
245
246
247
248
249
250
251
#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
252
253
254
255
256
257
258
259
      // 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;
260
261
262
263
264
265
266
267
268
269
270
271
#ifdef DEBUG_LOGMAP
      printf("Second run\n");
      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

272
    }
273

274
    alpha_ptr = &alpha128[0];
275
#if defined(__x86_64__) || defined(__i386__)
276
277
    m11p = (__m128i*)m_11;
    m10p = (__m128i*)m_10;
278
279
280
281
#elif defined(__arm__)
    m11p = (int16x8_t*)m_11;
    m10p = (int16x8_t*)m_10;
#endif
282
    for (k=0;
283
284
         k<l;
         k++) {
285

286
#if defined(__x86_64__) || defined(__i386__)
287
288
289
290
291
292
      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
293
      m_b4 = _mm_subs_epi16(a1,*m11p);  // m00=-m11
294
295
296
297
298
299
300
301
302
303
304
      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]);
305

306
307
308
309
310
311
312
313
      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
314

315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
      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);
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
#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
368
369
370

      alpha_ptr+=8;
      m11p++;
371
      m10p++;
372
#if defined(__x86_64__) || defined(__i386__)
373
374
375
376
377
378
379
380
      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);
381
382
383
384
385
386
387
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
419
420
421
422
423
424
425
426
427
428
429
430
431
432
#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
      printf("Loop index %d, mb\n",k);
      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);

      printf("Loop index %d, new\n",k);
      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);

      printf("Loop index %d, after max\n",k);
      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);

      printf("Loop index %d\n",k);
      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
433
434
435
436
437

    }

    if (rerun_flag==1)
      break;
438
  }
439
440
441
}


442
443
444
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)
{

445
  int k,rerun_flag=0;
446
#if defined(__x86_64__) || defined(__i386__)
447
448
449
450
451
  __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;
452
  __m128i beta_max;
453
454
455
456
457
458
459
460
461
#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

462
  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;
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
  llr_t beta0,beta1;

#ifdef DEBUG_LOGMAP
  msg("compute_beta, %p,%p,%p,%p,framelength %d,F %d\n",
      beta,m_11,m_10,alpha,frame_length,F);
#endif


  // termination for beta initialization

  //  printf("beta init: offset8 %d\n",offset8_flag);
  m11=(int16_t)m_11[2+frame_length];
  m10=(int16_t)m_10[2+frame_length];

  //  printf("m11,m10 %d,%d\n",m11,m10);

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

  //  printf("m11,m10 %d,%d\n",m11,m10);

  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];
  //  printf("m11,m10 %d,%d (%p)\n",m11,m10,m_11+frame_length);

  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;
502

503
504
505
506
507
508
509
510

  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;
511

512
513
514
515
516
517
518
519
520
521

  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;

522
  for (rerun_flag=0;; rerun_flag=1) {
523
#if defined(__x86_64__) || defined(__i386__)
524
525
    beta_ptr   = (__m128i*)&beta[frame_length<<3];
    alpha128   = (__m128i*)&alpha[0];
526
527
528
529
#elif defined(__arm__)
    beta_ptr   = (int16x8_t*)&beta[frame_length<<3];
    alpha128   = (int16x8_t*)&alpha[0];
#endif
530
531
532
533
534
535
536
537
538
    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)];
539
    } else {
540
#if defined(__x86_64__) || defined(__i386__)
541
542
543
544
545
546
547
548
549
      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);
550
551
552
553
554
555
556
557
558
559
560
561
#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);
#endif
562
563
564
    }


565
#if defined(__x86_64__) || defined(__i386__)
566
567
568
569
570
571
572
573
    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);
574
575
576
577
578
579
580
581
582
583
#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
584

585
    int loopval=((rerun_flag==0)?0:((frame_length-L)>>3));
586
587

    for (k=(frame_length>>3)-1; k>=loopval; k--) {
588
#if defined(__x86_64__) || defined(__i386__)
589
      m11_128=((__m128i*)m_11)[k];
590
591
592
593
594
595
      m10_128=((__m128i*)m_10)[k];

      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
596
      m_b4 = _mm_adds_epi16(beta_ptr[6],m10_128);  //m10
597
598
599
      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
600

601
602
603
604
605
606
607
608
      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
609

610
      beta_ptr-=8;
611

612
613
614
615
616
617
618
619
      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);
620

621
622
623
624
625
626
627
628
629
630
631
632
633
634
      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);
635
      beta_ptr[6] = _mm_subs_epi16(beta_ptr[6],beta_max);
636
      beta_ptr[7] = _mm_subs_epi16(beta_ptr[7],beta_max);
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
#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;
659

660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
      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
686
687
688
689
690
691
692
693
694


    }

    if (rerun_flag==1)
      break;
  }
}

695
696
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)
{
697
#if defined(__x86_64__) || defined(__i386__)
698
699
700
701
702
703
704
705
  __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;
706
707
708
709
710
711
712
713
714
715
716
#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

717
718
719
720
721
722
723
724
725
726
727
728
729
730
  int k;

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

#ifdef DEBUG_LOGMAP
  msg("compute_ext, %p, %p, %p, %p, %p, %p ,framelength %d\n",alpha,beta,m_11,m_10,ext,systematic,frame_length);
#endif

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


731
  for (k=0; k<(frame_length>>3); k++) {
732

733
#if defined(__x86_64__) || defined(__i386__)
734
735
736
    m11_128        = (__m128i*)&m_11[k<<3];
    m10_128        = (__m128i*)&m_10[k<<3];
    ext_128        = (__m128i*)&ext[k<<3];
737

738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
    /*
      printf("EXT %03d\n",k);
      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]);
756
    */
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
    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;
    /*
      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);
805

806
807
808
809
810
811
812
813
814
815
816
817
818
    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);
819

820
821
822
823
824
825
826
827
    /*
      print_shorts("ext:",ext_128);
      print_shorts("m11:",m11_128);
      print_shorts("m10:",m10_128);
      print_shorts("m10_1:",&m10_1);
      print_shorts("m01_1:",&m01_1);
      print_shorts("syst:",systematic_128);
    */
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
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
#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
876
877
878
879
880
881
882
883
884
885
    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];

886
887
888
889
890
891
892
893
894
895
void free_td16(void)
{
  int ind;

  for (ind=0; ind<188; ind++) {
    free(pi2tab16[ind]);
    free(pi5tab16[ind]);
    free(pi4tab16[ind]);
    free(pi6tab16[ind]);
  }
896
897
}

898
899
void init_td16()
{
900
901
902
903

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

904
  for (ind=0; ind<188; ind++) {
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920

    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

921
    for (i=i2=0; i2<8; i2++) {
922
      j=i2;
923
924
925
926
927
928
929
930

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

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

        pi2tab16[ind][i]  = j;
        //    printf("pi2[%d] = %d\n",i,j);
931
932
      }
    }
933
934

    for (i=0; i<n; i++) {
935
936
937
938
939
      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];
940
    }
941
942
943
944
945

  }
}

unsigned char phy_threegpplte_turbo_decoder16(short *y,
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
    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)
{

962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
  /*  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 */


  llr_t systematic0[n+16] __attribute__ ((aligned(16)));
  llr_t systematic1[n+16] __attribute__ ((aligned(16)));
  llr_t systematic2[n+16] __attribute__ ((aligned(16)));
  llr_t yparity1[n+16] __attribute__ ((aligned(16)));
  llr_t yparity2[n+16] __attribute__ ((aligned(16)));

  llr_t ext[n+128] __attribute__((aligned(16)));
  llr_t ext2[n+128] __attribute__((aligned(16)));

  llr_t alpha[(n+16)*8] __attribute__ ((aligned(16)));
  llr_t beta[(n+16)*8] __attribute__ ((aligned(16)));
  llr_t m11[n+16] __attribute__ ((aligned(16)));
  llr_t m10[n+16] __attribute__ ((aligned(16)));

981

982
983
984
985
986
  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;
gauthier's avatar
gauthier committed
987
  uint8_t temp;
988

989
990
#if defined(__x86_64__) || defined(__i386__)
  __m128i *yp128;
991
992
  __m128i tmp, zeros=_mm_setzero_si128();
  register __m128i tmpe;
993
994
995
996
997
998
999
1000
#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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
  int offset8_flag=0;

  if (crc_type > 3) {
    msg("Illegal crc length!\n");
    return 255;
  }


  start_meas(init_stats);



1013
  for (iind=0; iind < 188 && f1f2mat[iind].nb_bits != n; iind++);
1014

1015
1016
1017
1018
  if ( iind == 188 ) {
    msg("Illegal frame length!\n");
    return 255;
  }
1019

1020
1021
1022
1023
1024
  switch (crc_type) {
  case CRC24_A:
  case CRC24_B:
    crc_len=3;
    break;
1025

1026
1027
1028
  case CRC16:
    crc_len=2;
    break;
1029

1030
1031
1032
1033
  case CRC8:
    crc_len=1;
    break;

1034
1035
1036
1037
  default:
    crc_len=3;
  }

1038
#if defined(__x86_64__) || defined(__i386__)
1039
  yp128 = (__m128i*)y;
1040
1041
1042
#elif defined(__arm__)
  yp128 = (int16x8_t*)y;
#endif
1043
1044
1045



1046
1047

  s = systematic0;
1048
1049
1050
1051
1052
1053
1054
  s1 = systematic1;
  s2 = systematic2;
  yp1 = yparity1;
  yp2 = yparity2;



1055
  for (i=0; i<n; i+=8) {
1056
    pi2_p = &pi2tab16[iind][i];
1057

1058
1059
    j=pi2_p[0];

1060
#if defined(__x86_64__) || defined(__i386__)
1061
1062
    tmpe = _mm_load_si128(yp128);

1063
1064
    s[j]   = _mm_extract_epi16(tmpe,0);
    yp1[j] = _mm_extract_epi16(tmpe,1);
1065
1066
    yp2[j] = _mm_extract_epi16(tmpe,2);
    //        printf("init: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
1067

1068
1069
    j=pi2_p[1];

1070
1071
    s[j]   = _mm_extract_epi16(tmpe,3);
    yp1[j] = _mm_extract_epi16(tmpe,4);
1072
1073
1074
1075
1076
    yp2[j] = _mm_extract_epi16(tmpe,5);
    //    printf("init: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);

    j=pi2_p[2];

1077
1078
    s[j]   = _mm_extract_epi16(tmpe,6);
    yp1[j] = _mm_extract_epi16(tmpe,7);
1079
1080
1081
1082
1083
1084
    tmpe = _mm_load_si128(&yp128[1]);
    yp2[j] = _mm_extract_epi16(tmpe,0);
    //    printf("init: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);

    j=pi2_p[3];

1085
1086
    s[j]   = _mm_extract_epi16(tmpe,1);
    yp1[j] = _mm_extract_epi16(tmpe,2);
1087
1088
1089
1090
1091
    yp2[j] = _mm_extract_epi16(tmpe,3);
    //    printf("init: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);

    j=pi2_p[4];

1092
1093
    s[j]   = _mm_extract_epi16(tmpe,4);
    yp1[j] = _mm_extract_epi16(tmpe,5);
1094
1095
    yp2[j] = _mm_extract_epi16(tmpe,6);
    //    printf("init: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
1096

1097
1098
    j=pi2_p[5];

1099
    s[j]   = _mm_extract_epi16(tmpe,7);
1100
    tmpe = _mm_load_si128(&yp128[2]);
1101
    yp1[j] = _mm_extract_epi16(tmpe,0);
1102
1103
    yp2[j] = _mm_extract_epi16(tmpe,1);
    //    printf("init: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);
1104

1105
1106
    j=pi2_p[6];

1107
1108
    s[j]   = _mm_extract_epi16(tmpe,2);
    yp1[j] = _mm_extract_epi16(tmpe,3);
1109
1110
1111
1112
1113
    yp2[j] = _mm_extract_epi16(tmpe,4);
    //    printf("init: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);

    j=pi2_p[7];

1114
1115
    s[j]   = _mm_extract_epi16(tmpe,5);
    yp1[j] = _mm_extract_epi16(tmpe,6);
1116
1117
1118
    yp2[j] = _mm_extract_epi16(tmpe,7);
    //    printf("init: j %d, s[j] %d yp1[j] %d yp2[j] %d\n",j,s[j],yp1[j],yp2[j]);

1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
#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
1159
1160
1161
1162
1163
1164
1165
1166
    yp128+=3;

  }

  yp=(llr_t*)yp128;


  // Termination
1167
1168
1169
1170
1171
1172
1173
  for (i=n; i<n+3; i++) {
    s[i]= *yp;
    s1[i] = s[i] ;
    s2[i] = s[i];
    yp++;
    yp1[i] = *yp;
    yp++;
1174
1175
1176
1177
1178
#ifdef DEBUG_LOGMAP
    msg("Term 1 (%d): %d %d\n",i,s[i],yp1[i]);
#endif //DEBUG_LOGMAP
  }

1179
1180
1181
1182
1183
1184
1185
  for (i=n+8; i<n+11; i++) {
    s[i]= *yp;
    s1[i] = s[i] ;
    s2[i] = s[i];
    yp++;
    yp2[i-8] = *yp;
    yp++;
1186
#ifdef DEBUG_LOGMAP
1187
    msg("Term 2 (%d): %d %d\n",i-3,s[i],yp2[i-8]);
1188
1189
1190
1191
1192
1193
#endif //DEBUG_LOGMAP
  }

#ifdef DEBUG_LOGMAP
  msg("\n");
#endif //DEBUG_LOGMAP
1194

1195
1196
1197
1198
1199
1200
1201
  stop_meas(init_stats);

  // do log_map from first parity bit

  log_map16(systematic0,yparity1,m11,m10,alpha,beta,ext,n,0,F,offset8_flag,alpha_stats,beta_stats,gamma_stats,ext_stats);

  while (iteration_cnt++ < max_iterations) {
1202

1203
1204
1205
#ifdef DEBUG_LOGMAP
    printf("\n*******************ITERATION %d (n %d), ext %p\n\n",iteration_cnt,n,ext);
#endif //DEBUG_LOGMAP
1206

1207
    start_meas(intl1_stats);
1208

1209
    pi4_p=pi4tab16[iind];
1210
1211

    for (i=0; i<(n>>3); i++) { // steady-state portion
1212
#if defined(__x86_64__) || defined(__i386__)
1213
1214
1215
1216
1217
1218
1219
1220
      ((__m128i *)systematic2)[i]=_mm_insert_epi16(((__m128i *)systematic2)[i],((llr_t*)ext)[*pi4_p++],0);
      ((__m128i *)systematic2)[i]=_mm_insert_epi16(((__m128i *)systematic2)[i],((llr_t*)ext)[*pi4_p++],1);
      ((__m128i *)systematic2)[i]=_mm_insert_epi16(((__m128i *)systematic2)[i],((llr_t*)ext)[*pi4_p++],2);
      ((__m128i *)systematic2)[i]=_mm_insert_epi16(((__m128i *)systematic2)[i],((llr_t*)ext)[*pi4_p++],3);
      ((__m128i *)systematic2)[i]=_mm_insert_epi16(((__m128i *)systematic2)[i],((llr_t*)ext)[*pi4_p++],4);
      ((__m128i *)systematic2)[i]=_mm_insert_epi16(((__m128i *)systematic2)[i],((llr_t*)ext)[*pi4_p++],5);
      ((__m128i *)systematic2)[i]=_mm_insert_epi16(((__m128i *)systematic2)[i],((llr_t*)ext)[*pi4_p++],6);
      ((__m128i *)systematic2)[i]=_mm_insert_epi16(((__m128i *)systematic2)[i],((llr_t*)ext)[*pi4_p++],7);
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
#elif defined(__arm__)
      ((int16x8_t*)systematic2)[i]=vsetq_lane_s16(((llr_t*)ext)[*pi4_p++],((int16x8_t*)systematic2)[i],0);
      ((int16x8_t*)systematic2)[i]=vsetq_lane_s16(((llr_t*)ext)[*pi4_p++],((int16x8_t*)systematic2)[i],1);
      ((int16x8_t*)systematic2)[i]=vsetq_lane_s16(((llr_t*)ext)[*pi4_p++],((int16x8_t*)systematic2)[i],2);
      ((int16x8_t*)systematic2)[i]=vsetq_lane_s16(((llr_t*)ext)[*pi4_p++],((int16x8_t*)systematic2)[i],3);
      ((int16x8_t*)systematic2)[i]=vsetq_lane_s16(((llr_t*)ext)[*pi4_p++],((int16x8_t*)systematic2)[i],4);
      ((int16x8_t*)systematic2)[i]=vsetq_lane_s16(((llr_t*)ext)[*pi4_p++],((int16x8_t*)systematic2)[i],5);
      ((int16x8_t*)systematic2)[i]=vsetq_lane_s16(((llr_t*)ext)[*pi4_p++],((int16x8_t*)systematic2)[i],6);
      ((int16x8_t*)systematic2)[i]=vsetq_lane_s16(((llr_t*)ext)[*pi4_p++],((int16x8_t*)systematic2)[i],7);
#endif
1231
1232
    }

1233
    stop_meas(intl1_stats);
1234

1235
    // do log_map from second parity bit
1236
1237
1238
1239
1240
1241

    log_map16(systematic2,yparity2,m11,m10,alpha,beta,ext2,n,1,F,offset8_flag,alpha_stats,beta_stats,gamma_stats,ext_stats);



    pi5_p=pi5tab16[iind];
1242
1243

    for (i=0; i<(n>>3); i++) {
1244
#if defined(__x86_64__) || defined(__i386__)
1245
1246
1247
1248
1249
1250
1251
1252
1253
      tmp=_mm_insert_epi16(tmp,ext2[*pi5_p++],0);
      tmp=_mm_insert_epi16(tmp,ext2[*pi5_p++],1);
      tmp=_mm_insert_epi16(tmp,ext2[*pi5_p++],2);
      tmp=_mm_insert_epi16(tmp,ext2[*pi5_p++],3);
      tmp=_mm_insert_epi16(tmp,ext2[*pi5_p++],4);
      tmp=_mm_insert_epi16(tmp,ext2[*pi5_p++],5);
      tmp=_mm_insert_epi16(tmp,ext2[*pi5_p++],6);
      tmp=_mm_insert_epi16(tmp,ext2[*pi5_p++],7);
      ((__m128i *)systematic1)[i] = _mm_adds_epi16(_mm_subs_epi16(tmp,((__m128i*)ext)[i]),((__m128i *)systematic0)[i]);
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
#elif defined(__arm__)
        tmp=vsetq_lane_s16(ext2[*pi5_p++],tmp,0);
        tmp=vsetq_lane_s16(ext2[*pi5_p++],tmp,1);
        tmp=vsetq_lane_s16(ext2[*pi5_p++],tmp,2);
        tmp=vsetq_lane_s16(ext2[*pi5_p++],tmp,3);
        tmp=vsetq_lane_s16(ext2[*pi5_p++],tmp,4);
        tmp=vsetq_lane_s16(ext2[*pi5_p++],tmp,5);
        tmp=vsetq_lane_s16(ext2[*pi5_p++],tmp,6);
        tmp=vsetq_lane_s16(ext2[*pi5_p++],tmp,7);
	((int16x8_t *)systematic1)[i] = vqaddq_s16(vqsubq_s16(tmp,((int16x8_t*)ext)[i]),((int16x8_t *)systematic0)[i]);
#endif
1265
    }
1266

1267
1268
1269
    if (iteration_cnt>1) {
      start_meas(intl2_stats);
      pi6_p=pi6tab16[iind];
1270
1271

      for (i=0; i<(n>>3); i++) {
1272
#if defined(__x86_64__) || defined(__i386__)
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
        tmp=_mm_insert_epi16(tmp, ((llr_t*)ext2)[*pi6_p++],7);
        tmp=_mm_insert_epi16(tmp, ((llr_t*)ext2)[*pi6_p++],6);
        tmp=_mm_insert_epi16(tmp, ((llr_t*)ext2)[*pi6_p++],5);
        tmp=_mm_insert_epi16(tmp, ((llr_t*)ext2)[*pi6_p++],4);
        tmp=_mm_insert_epi16(tmp, ((llr_t*)ext2)[*pi6_p++],3);
        tmp=_mm_insert_epi16(tmp, ((llr_t*)ext2)[*pi6_p++],2);
        tmp=_mm_insert_epi16(tmp, ((llr_t*)ext2)[*pi6_p++],1);
        tmp=_mm_insert_epi16(tmp, ((llr_t*)ext2)[*pi6_p++],0);
        tmp=_mm_cmpgt_epi8(_mm_packs_epi16(tmp,zeros),zeros);
        decoded_bytes[i]=(unsigned char)_mm_movemask_epi8(tmp);
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
#elif defined(__arm__)
        tmp=vsetq_lane_s16(ext2[*pi6_p++],tmp,7);
        tmp=vsetq_lane_s16(ext2[*pi6_p++],tmp,6);
        tmp=vsetq_lane_s16(ext2[*pi6_p++],tmp,5);
        tmp=vsetq_lane_s16(ext2[*pi6_p++],tmp,4);
        tmp=vsetq_lane_s16(ext2[*pi6_p++],tmp,3);
        tmp=vsetq_lane_s16(ext2[*pi6_p++],tmp,2);
        tmp=vsetq_lane_s16(ext2[*pi6_p++],tmp,1);
        tmp=vsetq_lane_s16(ext2[*pi6_p++],tmp,0);
// This does:
// [1 2 4 8 16 32 64 128] .* I(ext_i > 0) = 2.^[b0 b1 b2 b3 b4 b5 b6 b7], where bi =I(ext_i > 0)
// [2^b0 + 2^b1 2^b2 + 2^b3 2^b4 + 2^b5 2^b6 + 2^b7]
// [2^b0 + 2^b1 + 2^b2 + 2^b3   2^b4 + 2^b5 + 2^b6 + 2^b7] 
// Mask64 = 2^b0 + 2^b1 + 2^b2 + 2^b3 + 2^b4 + 2^b5 + 2^b6 + 2^b7
	uint64x2_t Mask   = vpaddlq_u32(vpaddlq_u16(vandq_u16(vcgtq_s16(tmp,zeros), Powers)));
        uint64x1_t Mask64 = vget_high_u64(Mask)+vget_low_u64(Mask);
        decoded_bytes[i] = (uint8_t)Mask64;
#endif
1301
1302
      }
    }
1303

1304
1305
1306
    // check status on output
    if (iteration_cnt>1) {
      oldcrc= *((unsigned int *)(&decoded_bytes[(n>>3)-crc_len]));
1307

1308
      switch (crc_type) {
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318

      case CRC24_A:
        oldcrc&=0x00ffffff;
        crc = crc24a(&decoded_bytes[F>>3],
                     n-24-F)>>8;
        temp=((uint8_t *)&crc)[2];
        ((uint8_t *)&crc)[2] = ((uint8_t *)&crc)[0];
        ((uint8_t *)&crc)[0] = temp;
        break;

1319
      case CRC24_B:
1320
1321
1322
1323
1324
1325
1326
1327
        oldcrc&=0x00ffffff;
        crc = crc24b(decoded_bytes,
                     n-24)>>8;
        temp=((uint8_t *)&crc)[2];
        ((uint8_t *)&crc)[2] = ((uint8_t *)&crc)[0];
        ((uint8_t *)&crc)[0] = temp;
        break;

1328
      case CRC16:
1329
1330
1331
1332
1333
        oldcrc&=0x0000ffff;
        crc = crc16(decoded_bytes,
                    n-16)>>16;
        break;

1334
      case CRC8:
1335
1336
1337
1338
1339
        oldcrc&=0x000000ff;
        crc = crc8(decoded_bytes,
                   n-8)>>24;
        break;

1340
      default:
1341
1342
1343
        printf("FATAL: 3gpplte_turbo_decoder_sse.c: Unknown CRC\n");
        return(255);
        break;
1344
      }
1345

1346
      stop_meas(intl2_stats);
1347

1348
      if ((crc == oldcrc) && (crc!=0)) {
1349
        return(iteration_cnt);
1350
1351
      }
    }
1352

1353
1354
1355
    // do log_map from first parity bit
    if (iteration_cnt < max_iterations) {
      log_map16(systematic1,yparity1,m11,m10,alpha,beta,ext,n,0,F,offset8_flag,alpha_stats,beta_stats,gamma_stats,ext_stats);
1356
#if defined(__x86_64__) || defined(__i386__)
1357
1358
1359
      __m128i* ext_128=(__m128i*) ext;
      __m128i* s1_128=(__m128i*) systematic1;
      __m128i* s0_128=(__m128i*) systematic0;
1360
1361
1362
1363
1364
#elif defined(__arm__)
      int16x8_t* ext_128=(int16x8_t*) ext;
      int16x8_t* s1_128=(int16x8_t*) systematic1;
      int16x8_t* s0_128=(int16x8_t*) systematic0;
#endif
Cedric Roux's avatar