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

full version of the optimized 5NR LDPC decoder

Browse Source
This commit is contained in:
Sy
2020-06-03 19:04:20 +02:00
parent a6af0d99ce
commit 301317eca6
334 changed files with 44474 additions and 128094 deletions
Download Patch File Download Diff File Expand all files Collapse all files
openair1/PHY/CODING/nrLDPC_decoder/nrLDPC_mPass.h
+26 -20
View File
@@ -33,7 +33,7 @@
#include <string.h>
#include "nrLDPCdecoder_defs.h"
//#include <omp.h>
#include <omp.h>
/**
\brief Circular memcpy
|<- rem->|<- circular shift ->|
@@ -1018,9 +1018,9 @@ static inline void nrLDPC_bn2cnProcBuf_BG1(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf
// =====================================================================
// CN group with 3 BNs
bitOffsetInGroup = lut_numCnInCnGroups_BG1_R13[0]*NR_LDPC_ZMAX;
#pragma omp simd
for (j=0;j<2; j++)
{
p_cnProcBuf = &cnProcBuf[lut_startAddrCnGroups[0] + j*bitOffsetInGroup];
@@ -1031,27 +1031,28 @@ static inline void nrLDPC_bn2cnProcBuf_BG1(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf
// CN group with 4 BNs
bitOffsetInGroup = lut_numCnInCnGroups_BG1_R13[1]*NR_LDPC_ZMAX;
// #pragma omp simd
for (j=0; j<3; j++)
{
p_cnProcBuf = &cnProcBuf[lut_startAddrCnGroups[1] + j*bitOffsetInGroup];
for (i=0; i<lut_numCnInCnGroups[1]; i++)
#pragma omp simd
for (i=0; i<lut_numCnInCnGroups[1]; i++)
{
idxBn = lut_startAddrBnProcBuf_CNG4[j][i] + lut_bnPosBnProcBuf_CNG4[j][i]*Z;
nrLDPC_circ_memcpy(p_cnProcBuf, &bnProcBufRes[idxBn], Z, lut_circShift_CNG4[j][i]);
p_cnProcBuf += Z;
}
}
}
// =====================================================================
// CN group with 5 BNs
bitOffsetInGroup = lut_numCnInCnGroups_BG1_R13[2]*NR_LDPC_ZMAX;
// #pragma omp simd
for (j=0; j<4; j++)
{
p_cnProcBuf = &cnProcBuf[lut_startAddrCnGroups[2] + j*bitOffsetInGroup];
for (i=0; i<lut_numCnInCnGroups[2]; i++)
#pragma omp simd
for (i=0; i<lut_numCnInCnGroups[2]; i++)
{
idxBn = lut_startAddrBnProcBuf_CNG5[j][i] + lut_bnPosBnProcBuf_CNG5[j][i]*Z;
nrLDPC_circ_memcpy(p_cnProcBuf, &bnProcBufRes[idxBn], Z, lut_circShift_CNG5[j][i]);
@@ -1063,11 +1064,12 @@ static inline void nrLDPC_bn2cnProcBuf_BG1(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf
// CN group with 6 BNs
bitOffsetInGroup = lut_numCnInCnGroups_BG1_R13[3]*NR_LDPC_ZMAX;
//#pragma omp simd
for (j=0; j<5; j++)
{
p_cnProcBuf = &cnProcBuf[lut_startAddrCnGroups[3] + j*bitOffsetInGroup];
for (i=0; i<lut_numCnInCnGroups[3]; i++)
#pragma omp simd
for (i=0; i<lut_numCnInCnGroups[3]; i++)
{
idxBn = lut_startAddrBnProcBuf_CNG6[j][i] + lut_bnPosBnProcBuf_CNG6[j][i]*Z;
nrLDPC_circ_memcpy(p_cnProcBuf, &bnProcBufRes[idxBn], Z, lut_circShift_CNG6[j][i]);
@@ -1079,12 +1081,13 @@ static inline void nrLDPC_bn2cnProcBuf_BG1(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf
// CN group with 7 BNs
bitOffsetInGroup = lut_numCnInCnGroups_BG1_R13[4]*NR_LDPC_ZMAX;
//#pragma omp simd
for (j=0; j<6; j++)
{
p_cnProcBuf = &cnProcBuf[lut_startAddrCnGroups[4] + j*bitOffsetInGroup];
for (i=0; i<lut_numCnInCnGroups[4]; i++)
#pragma omp simd
for (i=0; i<lut_numCnInCnGroups[4]; i++)
{
idxBn = lut_startAddrBnProcBuf_CNG7[j][i] + lut_bnPosBnProcBuf_CNG7[j][i]*Z;
nrLDPC_circ_memcpy(p_cnProcBuf, &bnProcBufRes[idxBn], Z, lut_circShift_CNG7[j][i]);
@@ -1096,7 +1099,7 @@ static inline void nrLDPC_bn2cnProcBuf_BG1(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf
// CN group with 8 BNs
bitOffsetInGroup = lut_numCnInCnGroups_BG1_R13[5]*NR_LDPC_ZMAX;
#pragma omp simd
for (j=0; j<7; j++)
{
p_cnProcBuf = &cnProcBuf[lut_startAddrCnGroups[5] + j*bitOffsetInGroup];
@@ -1112,11 +1115,12 @@ static inline void nrLDPC_bn2cnProcBuf_BG1(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf
// CN group with 9 BNs
bitOffsetInGroup = lut_numCnInCnGroups_BG1_R13[6]*NR_LDPC_ZMAX;
// #pragma omp simd
for (j=0; j<8; j++)
{
p_cnProcBuf = &cnProcBuf[lut_startAddrCnGroups[6] + j*bitOffsetInGroup];
for (i=0; i<lut_numCnInCnGroups[6]; i++)
#pragma omp simd
for (i=0; i<lut_numCnInCnGroups[6]; i++)
{
idxBn = lut_startAddrBnProcBuf_CNG9[j][i] + lut_bnPosBnProcBuf_CNG9[j][i]*Z;
nrLDPC_circ_memcpy(p_cnProcBuf, &bnProcBufRes[idxBn], Z, lut_circShift_CNG9[j][i]);
@@ -1128,11 +1132,12 @@ static inline void nrLDPC_bn2cnProcBuf_BG1(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf
// CN group with 10 BNs
bitOffsetInGroup = lut_numCnInCnGroups_BG1_R13[7]*NR_LDPC_ZMAX;
//#pragma omp simd
for (j=0; j<9; j++)
{
p_cnProcBuf = &cnProcBuf[lut_startAddrCnGroups[7] + j*bitOffsetInGroup];
for (i=0; i<lut_numCnInCnGroups[7]; i++)
#pragma omp simd
for (i=0; i<lut_numCnInCnGroups[7]; i++)
{
idxBn = lut_startAddrBnProcBuf_CNG10[j][i] + lut_bnPosBnProcBuf_CNG10[j][i]*Z;
nrLDPC_circ_memcpy(p_cnProcBuf, &bnProcBufRes[idxBn], Z, lut_circShift_CNG10[j][i]);
@@ -1144,11 +1149,12 @@ static inline void nrLDPC_bn2cnProcBuf_BG1(t_nrLDPC_lut* p_lut, t_nrLDPC_procBuf
// CN group with 19 BNs
bitOffsetInGroup = lut_numCnInCnGroups_BG1_R13[8]*NR_LDPC_ZMAX;
//#pragma omp simd
for (j=0; j<19; j++)
{
p_cnProcBuf = &cnProcBuf[lut_startAddrCnGroups[8] + j*bitOffsetInGroup];
for (i=0; i<lut_numCnInCnGroups[8]; i++)
#pragma omp simd
for (i=0; i<lut_numCnInCnGroups[8]; i++)
{
idxBn = lut_startAddrBnProcBuf_CNG19[j][i] + lut_bnPosBnProcBuf_CNG19[j][i]*Z;
nrLDPC_circ_memcpy(p_cnProcBuf, &bnProcBufRes[idxBn], Z, lut_circShift_CNG19[j][i]);