1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
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
368
369
370
371
372
373
374
375
376
377
378
379
380
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
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
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
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
/*
* Licensed to the OpenAirInterface (OAI) Software Alliance under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The OpenAirInterface Software Alliance licenses this file to You under
* the OAI Public License, Version 1.1 (the "License"); you may not use this file
* except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.openairinterface.org/?page_id=698
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*-------------------------------------------------------------------------------
* For more information about the OpenAirInterface (OAI) Software Alliance:
* contact@openairinterface.org
*/
/*!\file PHY/CODING/nrPolar_tools/nr_polar_decoding_tools.c
* \brief
* \author Turker Yilmaz
* \date 2018
* \version 0.1
* \company EURECOM
* \email turker.yilmaz@eurecom.fr
* \note
* \warning
*/
#include "PHY/CODING/nrPolar_tools/nr_polar_defs.h"
#include "PHY/sse_intrin.h"
#include "PHY/impl_defs_top.h"
//#define DEBUG_NEW_IMPL 1
void updateLLR(double ***llr,
uint8_t **llrU,
uint8_t ***bit,
uint8_t **bitU,
uint8_t listSize,
uint16_t row,
uint16_t col,
uint16_t xlen,
uint8_t ylen,
uint8_t approximation)
{
uint16_t offset = (xlen/(pow(2,(ylen-col-1))));
for (uint8_t i=0; i<listSize; i++) {
if (( (row) % (2*offset) ) >= offset ) {
if(bitU[row-offset][col]==0) updateBit(bit, bitU, listSize, (row-offset), col, xlen, ylen);
if(llrU[row-offset][col+1]==0) updateLLR(llr, llrU, bit, bitU, listSize, (row-offset), (col+1), xlen, ylen, approximation);
if(llrU[row][col+1]==0) updateLLR(llr, llrU, bit, bitU, listSize, row, (col+1), xlen, ylen, approximation);
llr[row][col][i] = (pow((-1),bit[row-offset][col][i])*llr[row-offset][col+1][i]) + llr[row][col+1][i];
} else {
if(llrU[row][col+1]==0) updateLLR(llr, llrU, bit, bitU, listSize, row, (col+1), xlen, ylen, approximation);
if(llrU[row+offset][col+1]==0) updateLLR(llr, llrU, bit, bitU, listSize, (row+offset), (col+1), xlen, ylen, approximation);
computeLLR(llr, row, col, i, offset, approximation);
}
}
llrU[row][col]=1;
// printf("LLR (a %f, b %f): llr[%d][%d] %f\n",32*a,32*b,col,row,32*llr[col][row]);
}
void updateBit(uint8_t ***bit,
uint8_t **bitU,
uint8_t listSize,
uint16_t row,
uint16_t col,
uint16_t xlen,
uint8_t ylen)
{
uint16_t offset = ( xlen/(pow(2,(ylen-col))) );
for (uint8_t i=0; i<listSize; i++) {
if (( (row) % (2*offset) ) >= offset ) {
if (bitU[row][col-1]==0) updateBit(bit, bitU, listSize, row, (col-1), xlen, ylen);
bit[row][col][i] = bit[row][col-1][i];
} else {
if (bitU[row][col-1]==0) updateBit(bit, bitU, listSize, row, (col-1), xlen, ylen);
if (bitU[row+offset][col-1]==0) updateBit(bit, bitU, listSize, (row+offset), (col-1), xlen, ylen);
bit[row][col][i] = ( (bit[row][col-1][i]+bit[row+offset][col-1][i]) % 2);
}
}
bitU[row][col]=1;
}
void updatePathMetric(double *pathMetric,
double ***llr,
uint8_t listSize,
uint8_t bitValue,
uint16_t row,
uint8_t approximation)
{
if (approximation) { //eq. (12)
for (uint8_t i=0; i<listSize; i++) {
if ((2*bitValue) != ( 1 - copysign(1.0,llr[row][0][i]) )) pathMetric[i] += fabs(llr[row][0][i]);
}
} else { //eq. (11b)
int8_t multiplier = (2*bitValue) - 1;
for (uint8_t i=0; i<listSize; i++) pathMetric[i] += log ( 1 + exp(multiplier*llr[row][0][i]) ) ;
}
}
void updatePathMetric2(double *pathMetric,
double ***llr,
uint8_t listSize,
uint16_t row,
uint8_t appr)
{
double *tempPM = malloc(sizeof(double) * listSize);
for (int i=0; i < listSize; i++) tempPM[i]=pathMetric[i];
uint8_t bitValue = 0;
if (appr) { //eq. (12)
for (uint8_t i = 0; i < listSize; i++) {
if ((2 * bitValue) != (1 - copysign(1.0, llr[row][0][i]))) pathMetric[i] += fabs(llr[row][0][i]);
}
} else { //eq. (11b)
int8_t multiplier = (2 * bitValue) - 1;
for (uint8_t i = 0; i < listSize; i++) pathMetric[i] += log(1 + exp(multiplier * llr[row][0][i]));
}
bitValue = 1;
if (appr) { //eq. (12)
for (uint8_t i = listSize; i < 2*listSize; i++) {
if ((2 * bitValue) != (1 - copysign(1.0, llr[row][0][(i-listSize)]))) pathMetric[i] = tempPM[(i-listSize)] + fabs(llr[row][0][(i-listSize)]);
}
} else { //eq. (11b)
int8_t multiplier = (2 * bitValue) - 1;
for (uint8_t i = listSize; i < 2*listSize; i++) pathMetric[i] = tempPM[(i-listSize)] + log(1 + exp(multiplier * llr[row][0][(i-listSize)]));
}
free(tempPM);
}
void computeLLR(double ***llr,
uint16_t row,
uint16_t col,
uint8_t i,
uint16_t offset,
uint8_t approximation)
{
double a = llr[row][col + 1][i];
double absA = fabs(a);
double b = llr[row + offset][col + 1][i];
double absB = fabs(b);
if (approximation || isinf(absA) || isinf(absB)) { //eq. (9)
llr[row][col][i] = copysign(1.0, a) * copysign(1.0, b) * fmin(absA, absB);
} else { //eq. (8a)
llr[row][col][i] = log((exp(a + b) + 1) / (exp(a) + exp(b)));
}
}
void updateCrcChecksum(uint8_t **crcChecksum,
uint8_t **crcGen,
uint8_t listSize,
uint32_t i2,
uint8_t len)
{
for (uint8_t i = 0; i < listSize; i++) {
for (uint8_t j = 0; j < len; j++) {
crcChecksum[j][i] = ( (crcChecksum[j][i] + crcGen[i2][j]) % 2 );
}
}
}
void updateCrcChecksum2(uint8_t **crcChecksum,
uint8_t **crcGen,
uint8_t listSize,
uint32_t i2,
uint8_t len)
{
for (uint8_t i = 0; i < listSize; i++) {
for (uint8_t j = 0; j < len; j++) {
crcChecksum[j][i+listSize] = ( (crcChecksum[j][i] + crcGen[i2][j]) % 2 );
}
}
}
decoder_node_t *new_decoder_node(int first_leaf_index,int level) {
decoder_node_t *node=(decoder_node_t *)malloc(sizeof(decoder_node_t));
node->first_leaf_index=first_leaf_index;
node->level=level;
node->Nv = 1<<level;
node->leaf = 0;
node->left=(decoder_node_t *)NULL;
node->right=(decoder_node_t *)NULL;
node->all_frozen=0;
node->alpha = (int16_t*)malloc16(node->Nv*sizeof(int16_t));
node->beta = (int16_t*)malloc16(node->Nv*sizeof(int16_t));
memset((void*)node->beta,-1,node->Nv*sizeof(int16_t));
return(node);
}
decoder_node_t *add_nodes(int level,int first_leaf_index,t_nrPolar_params *pp) {
int all_frozen_below=1;
int Nv = 1<<level;
decoder_node_t *new_node = new_decoder_node(first_leaf_index,level);
#ifdef DEBUG_NEW_IMPL
printf("New node %d order %d, level %d\n",pp->tree.num_nodes,Nv,level);
#endif
pp->tree.num_nodes++;
if (level==0) {
#ifdef DEBUG_NEW_IMPL
printf("leaf %d (%s)\n",first_leaf_index,pp->information_bit_pattern[first_leaf_index]==1 ? "information or crc" : "frozen");
#endif
new_node->leaf=1;
if (pp->information_bit_pattern[first_leaf_index]==0) new_node->all_frozen = 1;
else {
new_node->all_frozen = 0;
int i=0;
for (;i<pp->K;i++)
if (pp->Q_I_N[i] == first_leaf_index) break;
AssertFatal(i<pp->K,"Couldn't find a valid bit position for leaf index %d\n",first_leaf_index);
new_node->bit_index = i;
}
return new_node; // this is a leaf node
}
for (int i=0;i<Nv;i++) {
if (pp->information_bit_pattern[i+first_leaf_index]>0) all_frozen_below=0;
}
if (all_frozen_below==0) new_node->left=add_nodes(level-1,first_leaf_index,pp);
else {
#ifdef DEBUG_NEW_IMPL
printf("aggregating frozen bits %d ... %d at level %d (%s)\n",first_leaf_index,first_leaf_index+Nv-1,level,((first_leaf_index/Nv)&1)==0?"left":"right");
#endif
new_node->leaf=1;
new_node->all_frozen=1;
}
if (all_frozen_below==0) new_node->right=add_nodes(level-1,first_leaf_index+(Nv/2),pp);
#ifdef DEBUG_NEW_IMPL
printf("new_node (%d): first_leaf_index %d, left %p, right %p\n",Nv,first_leaf_index,new_node->left,new_node->right);
#endif
return(new_node);
}
void build_decoder_tree(t_nrPolar_params *pp) {
pp->tree.num_nodes=0;
pp->tree.root = add_nodes(pp->n,0,pp);
#ifdef DEBUG_NEW_IMPL
printf("root : left %p, right %p\n",pp->tree.root->left,pp->tree.root->right);
#endif
}
inline void update_min(t_nrPolar_params *pp,int bi,int i,int u,int16_t abs_alpha) __attribute__((always_inline));
inline void update_min(t_nrPolar_params *pp,int bi,int i,int u,int16_t abs_alpha) {
int newPM = pp->tree.PM[i]-abs_alpha;
int minPM=0;
int minentry=pp->tree.minentry;
if (newPM > pp->tree.minPM) {
pp->tree.PM[minentry] = newPM;
pp->tree.decoderout[minentry][bi>>6] &= ~(((uint64_t)1)<<(bi&63));
pp->tree.decoderout[minentry][bi>>6] |= (((uint64_t)u)<<(bi&63));
}
for (int j=0;j<pp->tree.numentries;j++)
if (pp->tree.PM[i] < minPM) {
pp->tree.PM[i] = minPM;
pp->tree.minentry = i;
}
}
#if defined(__arm__) || defined(__aarch64__)
// translate 1-1 SIMD functions from SSE to NEON
#define __m128i int16x8_t
#define __m64 int8x8_t
#define _mm_abs_epi16(a) vabsq_s16(a)
#define _mm_min_epi16(a,b) vminq_s16(a,b)
#define _mm_subs_epi16(a,b) vsubq_s16(a,b)
#define _mm_abs_pi16(a) vabs_s16(a)
#define _mm_min_pi16(a,b) vmin_s16(a,b)
#define _mm_subs_pi16(a,b) vsub_s16(a,b)
#endif
void applyFtoleft(t_nrPolar_params *pp,decoder_node_t *node) {
int16_t *alpha_v=node->alpha;
int16_t *alpha_l=node->left->alpha;
int16_t *betal = node->left->beta;
int16_t a,b,absa,absb,maska,maskb,minabs;
#ifdef DEBUG_NEW_IMPL
printf("applyFtoleft %d, Nv %d (level %d,node->left (leaf %d, AF %d))\n",node->first_leaf_index,node->Nv,node->level,node->left->leaf,node->left->all_frozen);
for (int i=0;i<node->Nv;i++) printf("i%d (frozen %d): alpha_v[i] = %d\n",i,1-pp->information_bit_pattern[node->first_leaf_index+i],alpha_v[i]);
#endif
if (node->left->all_frozen == 0) {
#if defined(__AVX2__)
int avx2mod = (node->Nv/2)&15;
if (avx2mod == 0) {
__m256i a256,b256,absa256,absb256,minabs256;
int avx2len = node->Nv/2/16;
// printf("avx2len %d\n",avx2len);
for (int i=0;i<avx2len;i++) {
a256 =((__m256i*)alpha_v)[i];
b256 =((__m256i*)alpha_v)[i+avx2len];
absa256 =_mm256_abs_epi16(a256);
absb256 =_mm256_abs_epi16(b256);
minabs256 =_mm256_min_epi16(absa256,absb256);
((__m256i*)alpha_l)[i] =_mm256_sign_epi16(minabs256,_mm256_sign_epi16(a256,b256));
}
}
else if (avx2mod == 8) {
__m128i a128,b128,absa128,absb128,minabs128;
a128 =*((__m128i*)alpha_v);
b128 =((__m128i*)alpha_v)[1];
absa128 =_mm_abs_epi16(a128);
absb128 =_mm_abs_epi16(b128);
minabs128 =_mm_min_epi16(absa128,absb128);
*((__m128i*)alpha_l) =_mm_sign_epi16(minabs128,_mm_sign_epi16(a128,b128));
}
else if (avx2mod == 4) {
__m64 a64,b64,absa64,absb64,minabs64;
a64 =*((__m64*)alpha_v);
b64 =((__m64*)alpha_v)[1];
absa64 =_mm_abs_pi16(a64);
absb64 =_mm_abs_pi16(b64);
minabs64 =_mm_min_pi16(absa64,absb64);
*((__m64*)alpha_l) =_mm_sign_pi16(minabs64,_mm_sign_pi16(a64,b64));
}
else
#else
int sse4mod = (node->Nv/2)&7;
int sse4len = node->Nv/2/8;
#if defined(__arm__) || defined(__aarch64__)
int16x8_t signatimesb,comp1,comp2,negminabs128;
int16x8_t zero=vdupq_n_s16(0);
#endif
if (sse4mod == 0) {
for (int i=0;i<sse4len;i++) {
__m128i a128,b128,absa128,absb128,minabs128;
int sse4len = node->Nv/2/8;
a128 =*((__m128i*)alpha_v);
b128 =((__m128i*)alpha_v)[1];
absa128 =_mm_abs_epi16(a128);
absb128 =_mm_abs_epi16(b128);
minabs128 =_mm_min_epi16(absa128,absb128);
#if defined(__arm__) || defined(__aarch64__)
// unfortunately no direct equivalent to _mm_sign_epi16
signatimesb=vxorrq_s16(a128,b128);
comp1=vcltq_s16(signatimesb,zero);
comp2=vcgeq_s16(signatimesb,zero);
negminabs128=vnegq_s16(minabs128);
*((__m128i*)alpha_l) =vorrq_s16(vandq_s16(minabs128,comp0),vandq_s16(negminabs128,comp1));
#else
*((__m128i*)alpha_l) =_mm_sign_epi16(minabs128,_mm_sign_epi16(a128,b128));
#endif
}
}
else if (sse4mod == 4) {
__m64 a64,b64,absa64,absb64,minabs64;
a64 =*((__m64*)alpha_v);
b64 =((__m64*)alpha_v)[1];
absa64 =_mm_abs_pi16(a64);
absb64 =_mm_abs_pi16(b64);
minabs64 =_mm_min_pi16(absa64,absb64);
#if defined(__arm__) || defined(__aarch64__)
AssertFatal(1==0,"Need to do this still for ARM\n");
#else
*((__m64*)alpha_l) =_mm_sign_pi16(minabs64,_mm_sign_epi16(a64,b64));
#endif
}
else
#endif
{ // equvalent scalar code to above, activated only on non x86/ARM architectures
for (int i=0;i<node->Nv/2;i++) {
a=alpha_v[i];
b=alpha_v[i+(node->Nv/2)];
maska=a>>15;
maskb=b>>15;
absa=(a+maska)^maska;
absb=(b+maskb)^maskb;
minabs = absa<absb ? absa : absb;
alpha_l[i] = (maska^maskb)==0 ? minabs : -minabs;
// printf("alphal[%d] %d (%d,%d)\n",i,alpha_l[i],a,b);
}
}
if (node->Nv == 2) { // apply hard decision on left node
betal[0] = (alpha_l[0]>0) ? -1 : 1;
#ifdef DEBUG_NEW_IMPL
printf("betal[0] %d (%p)\n",betal[0],&betal[0]);
#endif
// pp->nr_polar_U[node->first_leaf_index] = (1+betal[0])>>1;
int bi = node->left->bit_index;
int16_t abs_alpha = ((alpha_l[0]>0) ? alpha_l[0] : -alpha_l[0]);
int ne2=pp->tree.numentries<<1;
if (ne2 > pp->tree.list_size) ne2 = pp->tree.list_size;
if (pp->tree.numentries < pp->tree.list_size) {
// update list for SC hard-decision
for (int i=0;i<pp->tree.numentries;i++) pp->tree.decoderout[i][bi>>6] |= (((1+betal[0])>>1)<<(bi&63));
// update list for flipped SC hard-decision
for (int i=pp->tree.numentries;i<ne2;i++) {
pp->tree.decoderout[i][bi>>6] = pp->tree.decoderout[i-pp->tree.numentries][bi>>6] ^ (((uint64_t)1)<<(bi&63));
pp->tree.PM[i] = pp->tree.PM[i-pp->tree.numentries] - abs_alpha;
if (pp->tree.PM[i] < pp->tree.minPM) {
pp->tree.minPM = pp->tree.PM[i];
pp->tree.minentry = i;
}
}
}
for (int i=0;i<(pp->tree.numentries<<1)-ne2;i++) update_min(pp,bi,i,(1-betal[0])>>1,abs_alpha);
#ifdef DEBUG_NEW_IMPL
int j;
printf("Setting bit %d (%d) to %d (LLR %d)\n",node->first_leaf_index,bit_index,(betal[0]+1)>>1,alpha_l[0]);
#endif
}
}
}
void applyGtoright(t_nrPolar_params *pp,decoder_node_t *node) {
int16_t *alpha_v=node->alpha;
int16_t *alpha_r=node->right->alpha;
int16_t *betal = node->left->beta;
int16_t *betar = node->right->beta;
#ifdef DEBUG_NEW_IMPL
printf("applyGtoright %d, Nv %d (level %d), (leaf %d, AF %d)\n",node->first_leaf_index,node->Nv,node->level,node->right->leaf,node->right->all_frozen);
#endif
if (node->right->all_frozen == 0) {
#if defined(__AVX2__)
int avx2mod = (node->Nv/2)&15;
if (avx2mod == 0) {
int avx2len = node->Nv/2/16;
for (int i=0;i<avx2len;i++) {
((__m256i *)alpha_r)[i] =
_mm256_subs_epi16(((__m256i *)alpha_v)[i+avx2len],
_mm256_sign_epi16(((__m256i *)alpha_v)[i],
((__m256i *)betal)[i]));
}
}
else if (avx2mod == 8) {
((__m128i *)alpha_r)[0] = _mm_subs_epi16(((__m128i *)alpha_v)[1],_mm_sign_epi16(((__m128i *)alpha_v)[0],((__m128i *)betal)[0]));
}
else if (avx2mod == 4) {
((__m64 *)alpha_r)[0] = _mm_subs_pi16(((__m64 *)alpha_v)[1],_mm_sign_pi16(((__m64 *)alpha_v)[0],((__m64 *)betal)[0]));
}
else
#else
int sse4mod = (node->Nv/2)&7;
if (sse4mod == 0) {
int sse4len = node->Nv/2/8;
for (int i=0;i<sse4len;i++) {
#if defined(__arm__) || defined(__aarch64__)
((int16x8_t *)alpha_r)[0] = vsubq_s16(((int16x8_t *)alpha_v)[1],vmulq_epi16(((int16x8_t *)alpha_v)[0],((int16x8_t *)betal)[0]));
#else
((__m128i *)alpha_r)[0] = _mm_subs_epi16(((__m128i *)alpha_v)[1],_mm_sign_epi16(((__m128i *)alpha_v)[0],((__m128i *)betal)[0]));
#endif
}
}
else if (sse4mod == 4) {
#if defined(__arm__) || defined(__aarch64__)
((int16x4_t *)alpha_r)[0] = vsub_s16(((int16x4_t *)alpha_v)[1],vmul_epi16(((int16x4_t *)alpha_v)[0],((int16x4_t *)betal)[0]));
#else
((__m64 *)alpha_r)[0] = _mm_subs_pi16(((__m64 *)alpha_v)[1],_mm_sign_pi16(((__64 *)alpha_v)[0],((__m64 *)betal)[0]));
#endif
}
else
#endif
{// equvalent scalar code to above, activated only on non x86/ARM architectures or Nv=1,2
for (int i=0;i<node->Nv/2;i++) {
alpha_r[i] = alpha_v[i+(node->Nv/2)] - (betal[i]*alpha_v[i]);
}
}
if (node->Nv == 2) { // apply hard decision on right node
betar[0] = (alpha_r[0]>0) ? -1 : 1;
// pp->nr_polar_U[node->first_leaf_index+1] = (1+betar[0])>>1;
int bi = node->right->bit_index;
int16_t abs_alpha = ((alpha_r[0]>0) ? alpha_r[0] : -alpha_r[0]);
int ne2=pp->tree.numentries<<1;
if (ne2 > pp->tree.list_size) ne2 = pp->tree.list_size;
if (pp->tree.numentries < pp->tree.list_size) {
// update list for SC hard-decision
for (int i=0;i<pp->tree.numentries;i++) pp->tree.decoderout[i][bi>>6] |= (((1+betar[0])>>1)<<(bi&63));
// update list for flipped SC hard-decision
for (int i=pp->tree.numentries;i<ne2;i++) {
pp->tree.decoderout[i][bi>>6] = pp->tree.decoderout[i-pp->tree.numentries][bi>>6] ^ (((uint64_t)1)<<(bi&63));
pp->tree.PM[i] = pp->tree.PM[i-pp->tree.numentries] - abs_alpha;
if (pp->tree.PM[i] < pp->tree.minPM) {
pp->tree.minPM = pp->tree.PM[i];
pp->tree.minentry = i;
}
}
}
for (int i=0;i<(pp->tree.numentries<<1)-ne2;i++) update_min(pp,bi,i,(1-betar[0])>>1,abs_alpha);
#ifdef DEBUG_NEW_IMPL
// int j;
// for (j=0;j<pp->K;j++) if (pp->Q_I_N[j] == (node->first_leaf_index+1)) break;
printf("Setting bit %d (%d) to %d (LLR %d)\n",node->first_leaf_index+1,j,(betar[0]+1)>>1,alpha_r[0]);
#endif
}
}
}
int16_t all1[16] = {1,1,1,1,
1,1,1,1,
1,1,1,1,
1,1,1,1};
void computeBeta(t_nrPolar_params *pp,decoder_node_t *node) {
int16_t *betav = node->beta;
int16_t *betal = node->left->beta;
int16_t *betar = node->right->beta;
#ifdef DEBUG_NEW_IMPL
printf("Computing beta @ level %d first_leaf_index %d (all_frozen %d)\n",node->level,node->first_leaf_index,node->left->all_frozen);
#endif
if (node->left->all_frozen==0) { // if left node is not aggregation of frozen bits
#if defined(__AVX2__)
int avx2mod = (node->Nv/2)&15;
register __m256i allones=*((__m256i*)all1);
if (avx2mod == 0) {
int avx2len = node->Nv/2/16;
for (int i=0;i<avx2len;i++) {
((__m256i*)betav)[i] = _mm256_or_si256(_mm256_cmpeq_epi16(((__m256i*)betar)[i],
((__m256i*)betal)[i]),allones);
}
}
else if (avx2mod == 8) {
((__m128i*)betav)[0] = _mm_or_si128(_mm_cmpeq_epi16(((__m128i*)betar)[0],
((__m128i*)betal)[0]),*((__m128i*)all1));
}
else if (avx2mod == 4) {
((__m64*)betav)[0] = _mm_or_si64(_mm_cmpeq_pi16(((__m64*)betar)[0],
((__m64*)betal)[0]),*((__m64*)all1));
}
else
#else
int sse4mod = (node->Nv/2)&7;
if (ssr4mod == 0) {
int ssr4len = node->Nv/2/8;
register __m128i allones=*((__m128i*)all1);
for (int i=0;i<sse4len;i++) {
((__m128i*)betav)[i] = _mm_or_si128(_mm_cmpeq_epi16(((__m128i*)betar)[i],
((__m128i*)betal)[i]),allones);
}
}
else if (sse4mod == 4) {
((__m64*)betav)[0] = _mm_or_si64(_mm_cmpeq_pi16(((__m64*)betar)[0],
((__m64*)betal)[0]),*((__m64*)all1));
}
else
#endif
{
for (int i=0;i<node->Nv/2;i++) {
betav[i] = (betal[i] != betar[i]) ? 1 : -1;
}
}
}
else memcpy((void*)&betav[0],betar,(node->Nv/2)*sizeof(int16_t));
memcpy((void*)&betav[node->Nv/2],betar,(node->Nv/2)*sizeof(int16_t));
}
void generic_polar_decoder(t_nrPolar_params *pp,decoder_node_t *node) {
// Apply F to left
applyFtoleft(pp,node);
// if left is not a leaf recurse down to the left
if (node->left->leaf==0) generic_polar_decoder(pp,node->left);
applyGtoright(pp,node);
if (node->right->leaf==0) generic_polar_decoder(pp,node->right);
computeBeta(pp,node);
}