maki/openairinterface5g
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(nrLDPC_coding_segment_encoder): address review

Browse Source 
1. nrLDPC_coding_segment_encoding: Clarifying encoder parameters
   The way the encoder implementation params was used could be made clearer:
   * `macro_num` was always used in order to calculate segment indexes.
     It is clearer to pass directly the first segment index.
   * The initialisation of a common implementation parameters structure can be made clearer.

2. nrLDPC_coding_segment_encoder: Remove multi segment dimension in encoder output
   The output of `LDPCencoder` was a 2D unsigned char array.
   But all the segments were actually packed in the first segment, 8 bits of different segments per byte of the first segment.
   So the first dimension was useless. This commit removes this dimension.

3. nrLDPC_coding_segment_encoder: Rename `nrLDPC_prepare_TB_encoding` into `nrLDPC_launch_TB_encoding`

4. nrLDPC_coding_segment_encoder: Remove the useless second rate matching for E2, one rate matching with max(E, E2) does the job.

5. nr_interleaving_ldpc: cleanup
   Rewrite code to use different SIMD sizes in a linear fashion.
   This means cascading preprocessor directives are used to start from a high SIMD size an decrease it progressively.

6. ldpc_encoder_optim8segmulti: linear usage of SIMD
   Use cascading preprocessor directives to include different SIMD sizes and types depending on which are available on the system.
   The different loops use a common index that is progressively incremented.
   Also add AVX512 version of bits distribution from 8 segments into one byte.

Co-authored-by: Laurent Thomas <laurent.thomas@open-cells.com>
This commit is contained in:
Romain Beurdouche
2025-05-13 14:08:33 +00:00
co-authored by Laurent Thomas
parent 1b31ac417b
commit 06c81030dc
6 changed files with 515 additions and 552 deletions
Download Patch File Download Diff File Expand all files Collapse all files
openair1/PHY/CODING/nrLDPC_encoder/ldpc_encoder_optim8segmulti.c
+73 -69
View File
@@ -43,7 +43,7 @@
#include "ldpc_encode_parity_check.c"
#include "ldpc_generate_coefficient.c"
int LDPCencoder(uint8_t **input, uint8_t **output, encoder_implemparams_t *impp)
int LDPCencoder(uint8_t **input, uint8_t *output, encoder_implemparams_t *impp)
{
//set_log(PHY, 4);
@@ -61,41 +61,10 @@ int LDPCencoder(uint8_t **input, uint8_t **output, encoder_implemparams_t *impp)
unsigned int macro_segment, macro_segment_end;
macro_segment = 8*impp->macro_num;
macro_segment_end = (impp->n_segments > 8*(impp->macro_num+1)) ? 8*(impp->macro_num+1) : impp->n_segments;
macro_segment = impp->first_seg;
macro_segment_end = (impp->n_segments > impp->first_seg + 8) ? impp->first_seg + 8 : impp->n_segments;
///printf("macro_segment: %d\n", macro_segment);
///printf("macro_segment_end: %d\n", macro_segment_end );
#ifdef __aarch64__
simde__m128i shufmask = simde_mm_set_epi64x(0x0101010101010101, 0x0000000000000000);
simde__m128i andmask = simde_mm_set1_epi64x(0x0102040810204080); // every 8 bits -> 8 bytes, pattern repeats.
simde__m128i zero128 = simde_mm_setzero_si128();
simde__m128i masks[8];
register simde__m128i c128;
masks[0] = simde_mm_set1_epi8(0x1);
masks[1] = simde_mm_set1_epi8(0x2);
masks[2] = simde_mm_set1_epi8(0x4);
masks[3] = simde_mm_set1_epi8(0x8);
masks[4] = simde_mm_set1_epi8(0x10);
masks[5] = simde_mm_set1_epi8(0x20);
masks[6] = simde_mm_set1_epi8(0x40);
masks[7] = simde_mm_set1_epi8(0x80);
#else
simde__m256i shufmask = simde_mm256_set_epi64x(0x0303030303030303, 0x0202020202020202,0x0101010101010101, 0x0000000000000000);
simde__m256i andmask = simde_mm256_set1_epi64x(0x0102040810204080); // every 8 bits -> 8 bytes, pattern repeats.
simde__m256i zero256 = simde_mm256_setzero_si256();
simde__m256i masks[8];
register simde__m256i c256;
masks[0] = simde_mm256_set1_epi8(0x1);
masks[1] = simde_mm256_set1_epi8(0x2);
masks[2] = simde_mm256_set1_epi8(0x4);
masks[3] = simde_mm256_set1_epi8(0x8);
masks[4] = simde_mm256_set1_epi8(0x10);
masks[5] = simde_mm256_set1_epi8(0x20);
masks[6] = simde_mm256_set1_epi8(0x40);
masks[7] = simde_mm256_set1_epi8(0x80);
#endif
//determine number of bits in codeword
if (BG==1)
@@ -136,55 +105,90 @@ int LDPCencoder(uint8_t **input, uint8_t **output, encoder_implemparams_t *impp)
if(impp->tinput != NULL) start_meas(impp->tinput);
//interleave up to 8 transport-block segements at a time
#if 0
for (i=0; i<block_length; i++) {
//for (j=0; j<n_segments; j++) {
for (j=macro_segment; j < macro_segment_end; j++) {
temp = (input[j][i/8]&(1<<(i&7)))>>(i&7);
//printf("c(%d,%d)=%d\n",j,i,temp);
c[i] |= (temp << (j-macro_segment));
}
}
#elif defined(__aarch64__)
for (int i=0; i<block_length>>4; i++) {
c128 = simde_mm_and_si128(simde_mm_cmpeq_epi8(simde_mm_andnot_si128(simde_mm_shuffle_epi8(simde_mm_set1_epi16(((uint16_t*)input[macro_segment])[i]), shufmask),andmask),zero128),masks[0]);
//for (j=1; j<n_segments; j++) {
for (int j=macro_segment+1; j < macro_segment_end; j++) {
c128 = simde_mm_or_si128(simde_mm_and_si128(simde_mm_cmpeq_epi8(simde_mm_andnot_si128(simde_mm_shuffle_epi8(simde_mm_set1_epi32(((uint16_t*)input[j])[i]), shufmask),andmask),zero128),masks[j-macro_segment]),c128);
}
((simde__m128i *)cc)[i] = c128;
}
unsigned int i_byte = 0;
for (int i=(block_length>>4)<<4;i<block_length;i++) {
//for (j=0; j<n_segments; j++) {
for (int j=macro_segment; j < macro_segment_end; j++) {
#if defined(__AVX512F__) && defined(__AVX512BW__) && defined(__AVX512VBMI__)
const __m512i masks5[8] = { _mm512_set1_epi8(0x1), _mm512_set1_epi8(0x2),
_mm512_set1_epi8(0x4), _mm512_set1_epi8(0x8),
_mm512_set1_epi8(0x10), _mm512_set1_epi8(0x20),
_mm512_set1_epi8(0x40), _mm512_set1_epi8(0x80)};
const __m512i zero512 = _mm512_setzero_si512();
const uint8_t perm[64]__attribute__((aligned(64))) = {7,6,5,4,3,2,1,0, 15,14,13,12,11,10,9,8,
23,22,21,20,19,18,17,16, 31,30,29,28,27,26,25,24,
39,38,37,36,35,34,33,32, 47,46,45,44,43,42,41,40,
55,54,53,52,51,50,49,48, 63,62,61,60,59,58,57,56};
register __m512i c512;
temp = (input[j][i/8]&(128>>(i&7)))>>(7-(i&7));
//printf("c(%d,%d)=%d\n",j,i,temp);
cc[i] |= (temp << (j-macro_segment));
for (; i_byte < ((block_length >> 6) << 6); i_byte += 64) {
unsigned int i = i_byte >> 6;
c512 = _mm512_mask_blend_epi8(((uint64_t *)&input[macro_segment][0])[i], zero512, masks5[0]);
for (int j = macro_segment + 1; j < macro_segment_end; j++) {
c512 = _mm512_or_si512(c512, _mm512_mask_blend_epi8(((uint64_t *)&input[j][0])[i], zero512, masks5[j - macro_segment]));
}
c512 = _mm512_permutexvar_epi8(*(__m512i*)perm, c512);
((__m512i *)cc)[i] = c512;
}
#else
for (int i=0; i<block_length>>5; i++) {
#endif
#ifndef __aarch64__
simde__m256i shufmask = simde_mm256_set_epi64x(0x0303030303030303, 0x0202020202020202,0x0101010101010101, 0x0000000000000000);
simde__m256i andmask = simde_mm256_set1_epi64x(0x0102040810204080); // every 8 bits -> 8 bytes, pattern repeats.
simde__m256i zero256 = simde_mm256_setzero_si256();
simde__m256i masks[8];
register simde__m256i c256;
masks[0] = simde_mm256_set1_epi8(0x1);
masks[1] = simde_mm256_set1_epi8(0x2);
masks[2] = simde_mm256_set1_epi8(0x4);
masks[3] = simde_mm256_set1_epi8(0x8);
masks[4] = simde_mm256_set1_epi8(0x10);
masks[5] = simde_mm256_set1_epi8(0x20);
masks[6] = simde_mm256_set1_epi8(0x40);
masks[7] = simde_mm256_set1_epi8(0x80);
for (; i_byte < ((block_length >> 5 ) << 5); i_byte += 32) {
unsigned int i = i_byte >> 5;
c256 = simde_mm256_and_si256(simde_mm256_cmpeq_epi8(simde_mm256_andnot_si256(simde_mm256_shuffle_epi8(simde_mm256_set1_epi32(((uint32_t*)input[macro_segment])[i]), shufmask),andmask),zero256),masks[0]);
//for (j=1; j<n_segments; j++) {
for (int j=macro_segment+1; j < macro_segment_end; j++) {
c256 = simde_mm256_or_si256(simde_mm256_and_si256(simde_mm256_cmpeq_epi8(simde_mm256_andnot_si256(simde_mm256_shuffle_epi8(simde_mm256_set1_epi32(((uint32_t*)input[j])[i]), shufmask),andmask),zero256),masks[j-macro_segment]),c256);
}
((simde__m256i *)cc)[i] = c256;
}
#endif
for (int i=(block_length>>5)<<5;i<block_length;i++) {
//for (j=0; j<n_segments; j++) {
for (int j=macro_segment; j < macro_segment_end; j++) {
#ifdef __aarch64__
simde__m128i shufmask = simde_mm_set_epi64x(0x0101010101010101, 0x0000000000000000);
simde__m128i andmask = simde_mm_set1_epi64x(0x0102040810204080); // every 8 bits -> 8 bytes, pattern repeats.
simde__m128i zero128 = simde_mm_setzero_si128();
simde__m128i masks[8];
register simde__m128i c128;
masks[0] = simde_mm_set1_epi8(0x1);
masks[1] = simde_mm_set1_epi8(0x2);
masks[2] = simde_mm_set1_epi8(0x4);
masks[3] = simde_mm_set1_epi8(0x8);
masks[4] = simde_mm_set1_epi8(0x10);
masks[5] = simde_mm_set1_epi8(0x20);
masks[6] = simde_mm_set1_epi8(0x40);
masks[7] = simde_mm_set1_epi8(0x80);
temp = (input[j][i/8]&(128>>(i&7)))>>(7-(i&7));
//printf("c(%d,%d)=%d\n",j,i,temp);
for (; i_byte < ((block_length >> 4 ) << 4); i_byte += 16) {
unsigned int i = i_byte >> 4;
c128 = simde_mm_and_si128(simde_mm_cmpeq_epi8(simde_mm_andnot_si128(simde_mm_shuffle_epi8(simde_mm_set1_epi16(((uint16_t*)input[macro_segment])[i]), shufmask),andmask),zero128),masks[0]);
for (int j=macro_segment+1; j < macro_segment_end; j++) {
c128 = simde_mm_or_si128(simde_mm_and_si128(simde_mm_cmpeq_epi8(simde_mm_andnot_si128(simde_mm_shuffle_epi8(simde_mm_set1_epi32(((uint16_t*)input[j])[i]), shufmask),andmask),zero128),masks[j-macro_segment]),c128);
}
((simde__m128i *)cc)[i] = c128;
}
#endif
for (; i_byte < block_length; i_byte++) {
unsigned int i = i_byte;
for (int j = macro_segment; j < macro_segment_end; j++) {
temp = (input[j][i/8]&(128>>(i&7)))>>(7-(i&7));
cc[i] |= (temp << (j-macro_segment));
}
}
#endif
if(impp->tinput != NULL) stop_meas(impp->tinput);
@@ -203,8 +207,8 @@ int LDPCencoder(uint8_t **input, uint8_t **output, encoder_implemparams_t *impp)
return(-1);
}
}
memcpy(output[0],&cc[2*Zc],(block_length-(2*Zc)));
memcpy(output[0]+block_length-(2*Zc),dd,((nrows-no_punctured_columns) * Zc-removed_bit));
memcpy(output,&cc[2*Zc],(block_length-(2*Zc)));
memcpy(output+block_length-(2*Zc),dd,((nrows-no_punctured_columns) * Zc-removed_bit));
return 0;
}