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

Merge remote-tracking branch 'origin/improve-nr_ul_channel_estimation' into integration_2022_wk35b

Browse Source
This commit is contained in:
Robert Schmidt
2022-09-04 21:46:17 +02:00
parent 203ac17833 4c74cc6686
commit 23459a487d
22 changed files with 767 additions and 1556 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/utils/LOG/log.c
+4
View File
@@ -48,6 +48,10 @@
// main log variables // main log variables
// Fixme: a better place to be shure it is called
void read_cpu_hardware (void) __attribute__ ((constructor));
void read_cpu_hardware (void) {__builtin_cpu_init(); }
log_mem_cnt_t log_mem_d[2]; log_mem_cnt_t log_mem_d[2];
int log_mem_flag=0; int log_mem_flag=0;
int log_mem_multi=1; int log_mem_multi=1;
common/utils/LOG/log.h
+1
View File
@@ -366,6 +366,7 @@ typedef struct {
#define MATLAB_CSHORT_BRACKET3 15 #define MATLAB_CSHORT_BRACKET3 15
int32_t write_file_matlab(const char *fname, const char *vname, void *data, int length, int dec, unsigned int format, int multiVec); int32_t write_file_matlab(const char *fname, const char *vname, void *data, int length, int dec, unsigned int format, int multiVec);
#define write_output(a, b, c, d, e, f) write_file_matlab(a, b, c, d, e, f, 0)
/*----------------macro definitions for reading log configuration from the config module */ /*----------------macro definitions for reading log configuration from the config module */
#define CONFIG_STRING_LOG_PREFIX "log_config" #define CONFIG_STRING_LOG_PREFIX "log_config"
common/utils/time_meas.c
+8 -8
View File
@@ -39,14 +39,14 @@ notifiedFIFO_t measur_fifo;
double get_cpu_freq_GHz(void) double get_cpu_freq_GHz(void)
{ {
if (cpu_freq_GHz <1 ) { if (cpu_freq_GHz <1 ) {
time_stats_t ts = {0}; time_stats_t ts = {0};
reset_meas(&ts); reset_meas(&ts);
ts.trials++; ts.trials++;
ts.in = rdtsc_oai(); ts.in = rdtsc_oai();
sleep(1); sleep(1);
ts.diff = (rdtsc_oai()-ts.in); ts.diff = (rdtsc_oai()-ts.in);
cpu_freq_GHz = (double)ts.diff/1000000000; cpu_freq_GHz = (double)ts.diff/1000000000;
printf("CPU Freq is %f \n", cpu_freq_GHz); printf("CPU Freq is %f \n", cpu_freq_GHz);
} }
return cpu_freq_GHz; return cpu_freq_GHz;
} }
openair1/PHY/INIT/lte_init.c
-7
View File
@@ -475,9 +475,7 @@ int phy_init_lte_eNB(PHY_VARS_eNB *eNB,
srs_vars[srs_id].srs = (int32_t *) malloc16_clear (2 * fp->ofdm_symbol_size * sizeof (int32_t)); srs_vars[srs_id].srs = (int32_t *) malloc16_clear (2 * fp->ofdm_symbol_size * sizeof (int32_t));
} }
LOG_I(PHY,"PRACH allocation\n");
// PRACH // PRACH
prach_vars->prachF = (int16_t *) malloc16_clear (1024 * 2 * sizeof (int16_t));
// assume maximum of 64 RX antennas for PRACH receiver // assume maximum of 64 RX antennas for PRACH receiver
prach_vars->prach_ifft[0] = (int32_t **) malloc16_clear (64 * sizeof (int32_t *)); prach_vars->prach_ifft[0] = (int32_t **) malloc16_clear (64 * sizeof (int32_t *));
@@ -485,8 +483,6 @@ int phy_init_lte_eNB(PHY_VARS_eNB *eNB,
prach_vars->prach_ifft[0][i] = (int32_t *) malloc16_clear (1024 * 2 * sizeof (int32_t)); prach_vars->prach_ifft[0][i] = (int32_t *) malloc16_clear (1024 * 2 * sizeof (int32_t));
prach_vars->rxsigF[0] = (int16_t **) malloc16_clear (64 * sizeof (int16_t *)); prach_vars->rxsigF[0] = (int16_t **) malloc16_clear (64 * sizeof (int16_t *));
// PRACH BR
prach_vars_br->prachF = (int16_t *)malloc16_clear( 1024*2*sizeof(int32_t) );
// assume maximum of 64 RX antennas for PRACH receiver // assume maximum of 64 RX antennas for PRACH receiver
for (int ce_level = 0; ce_level < 4; ce_level++) { for (int ce_level = 0; ce_level < 4; ce_level++) {
@@ -576,8 +572,6 @@ void phy_free_lte_eNB(PHY_VARS_eNB *eNB) {
for (UE_id=0; UE_id<NUMBER_OF_SRS_MAX; UE_id++) free_and_zero(srs_vars[UE_id].srs); for (UE_id=0; UE_id<NUMBER_OF_SRS_MAX; UE_id++) free_and_zero(srs_vars[UE_id].srs);
free_and_zero(prach_vars->prachF);
for (i = 0; i < 64; i++) free_and_zero(prach_vars->prach_ifft[0][i]); for (i = 0; i < 64; i++) free_and_zero(prach_vars->prach_ifft[0][i]);
free_and_zero(prach_vars->prach_ifft[0]); free_and_zero(prach_vars->prach_ifft[0]);
@@ -589,7 +583,6 @@ void phy_free_lte_eNB(PHY_VARS_eNB *eNB) {
free_and_zero(prach_vars->rxsigF[ce_level]); free_and_zero(prach_vars->rxsigF[ce_level]);
} }
free_and_zero(prach_vars_br->prachF);
free_and_zero(prach_vars->rxsigF[0]); free_and_zero(prach_vars->rxsigF[0]);
for (int ULSCH_id=0; ULSCH_id<NUMBER_OF_ULSCH_MAX; ULSCH_id++) { for (int ULSCH_id=0; ULSCH_id<NUMBER_OF_ULSCH_MAX; ULSCH_id++) {
openair1/PHY/LTE_ESTIMATION/lte_ul_channel_estimation.c
+5 -5
View File
@@ -106,9 +106,9 @@ int32_t lte_ul_channel_estimation(LTE_DL_FRAME_PARMS *frame_parms,
#ifdef DEBUG_CH #ifdef DEBUG_CH
if (Ns==0) if (Ns==0)
LOG_M("drs_seq0.m","drsseq0",ul_ref_sigs_rx[u][v][Msc_RS_idx],2*Msc_RS,2,1); LOG_M("drs_seq0.m","drsseq0",ul_ref_sigs_rx[u][v][Msc_RS_idx],2*Msc_RS,2,0);
else else
LOG_M("drs_seq1.m","drsseq1",ul_ref_sigs_rx[u][v][Msc_RS_idx],2*Msc_RS,2,1); LOG_M("drs_seq1.m","drsseq1",ul_ref_sigs_rx[u][v][Msc_RS_idx],2*Msc_RS,2,0);
#endif #endif
@@ -311,7 +311,7 @@ int32_t lte_ul_channel_estimation(LTE_DL_FRAME_PARMS *frame_parms,
// alpha = (int16_t) (((int32_t) SCALE * (int32_t) (pilot_pos2-k))/(pilot_pos2-pilot_pos1)); // alpha = (int16_t) (((int32_t) SCALE * (int32_t) (pilot_pos2-k))/(pilot_pos2-pilot_pos1));
// beta = (int16_t) (((int32_t) SCALE * (int32_t) (k-pilot_pos1))/(pilot_pos2-pilot_pos1)); // beta = (int16_t) (((int32_t) SCALE * (int32_t) (k-pilot_pos1))/(pilot_pos2-pilot_pos1));
#ifdef DEBUG_CH #ifdef DEBUG_CH
LOG_D(PHY,"lte_ul_channel_estimation: k=%d, alpha = %d, beta = %d\n",k,alpha,beta); LOG_D(PHY,"lte_ul_channel_estimation: k=%d\n",k);
#endif #endif
//symbol_offset_subframe = frame_parms->N_RB_UL*12*k; //symbol_offset_subframe = frame_parms->N_RB_UL*12*k;
@@ -436,9 +436,9 @@ int32_t lte_ul_channel_estimation_RRU(LTE_DL_FRAME_PARMS *frame_parms,
#ifdef DEBUG_CH #ifdef DEBUG_CH
if (l==pilot_pos1) if (l==pilot_pos1)
write_output("drs_seq0.m","drsseq0",ul_ref_sigs_rx[u][v][Msc_RS_idx],Msc_RS,1,1); write_output("drs_seq0.m","drsseq0",ul_ref_sigs_rx[u][v][Msc_RS_idx],Msc_RS,1,0);
else else
write_output("drs_seq1.m","drsseq1",ul_ref_sigs_rx[u][v][Msc_RS_idx],Msc_RS,1,1); write_output("drs_seq1.m","drsseq1",ul_ref_sigs_rx[u][v][Msc_RS_idx],Msc_RS,1,0);
#endif #endif
symbol_offset = frame_parms->N_RB_UL*12*l; symbol_offset = frame_parms->N_RB_UL*12*l;
openair1/PHY/LTE_TRANSPORT/phich.c
+10 -9
View File
@@ -74,7 +74,8 @@ void generate_phich(LTE_DL_FRAME_PARMS *frame_parms,
uint32_t subframe_offset=((frame_parms->Ncp==0)?14:12)*frame_parms->ofdm_symbol_size*subframe; uint32_t subframe_offset=((frame_parms->Ncp==0)?14:12)*frame_parms->ofdm_symbol_size*subframe;
memset(d,0,24*sizeof(int16_t)); memset(d,0,24*sizeof(int16_t));
const int nushift=frame_parms->nushift;
if (frame_parms->nb_antenna_ports_eNB==1) if (frame_parms->nb_antenna_ports_eNB==1)
gain_lin_QPSK = (int16_t)(((int32_t)amp*ONE_OVER_SQRT2_Q15)>>15); gain_lin_QPSK = (int16_t)(((int32_t)amp*ONE_OVER_SQRT2_Q15)>>15);
else else
@@ -245,7 +246,7 @@ void generate_phich(LTE_DL_FRAME_PARMS *frame_parms,
y1_16[7] = -y0_16[5]; y1_16[7] = -y0_16[5];
for (i=0,j=0,m=0; i<6; i++,j+=2) { for (i=0,j=0,m=0; i<6; i++,j+=2) {
if ((i!=(frame_parms->nushift))&&(i!=(frame_parms->nushift+3))) { if ((i!=(nushift))&&(i!=((nushift+3)%6))) {
y0[j] += y0_16[m]; y0[j] += y0_16[m];
y1[j] += y1_16[m++]; y1[j] += y1_16[m++];
y0[j+1] += y0_16[m]; y0[j+1] += y0_16[m];
@@ -287,7 +288,7 @@ void generate_phich(LTE_DL_FRAME_PARMS *frame_parms,
y1_16[7] = -y0_16[5]; y1_16[7] = -y0_16[5];
for (i=0,j=0,m=0; i<6; i++,j+=2) { for (i=0,j=0,m=0; i<6; i++,j+=2) {
if ((i!=(frame_parms->nushift))&&(i!=(frame_parms->nushift+3))) { if ((i!=(nushift))&&(i!=((nushift+3)%6))) {
y0[j] += y0_16[m]; y0[j] += y0_16[m];
y1[j] += y1_16[m++]; y1[j] += y1_16[m++];
y0[j+1] += y0_16[m]; y0[j+1] += y0_16[m];
@@ -329,7 +330,7 @@ void generate_phich(LTE_DL_FRAME_PARMS *frame_parms,
y1_16[7] = -y0_16[5]; y1_16[7] = -y0_16[5];
for (i=0,j=0,m=0; i<6; i++,j+=2) { for (i=0,j=0,m=0; i<6; i++,j+=2) {
if ((i!=(frame_parms->nushift))&&(i!=(frame_parms->nushift+3))) { if ((i!=(nushift))&&(i!=((nushift+3)%6))) {
y0[j] += y0_16[m]; y0[j] += y0_16[m];
y1[j] += y1_16[m++]; y1[j] += y1_16[m++];
y0[j+1] += y0_16[m]; y0[j+1] += y0_16[m];
@@ -360,7 +361,7 @@ void generate_phich(LTE_DL_FRAME_PARMS *frame_parms,
y0_16[7] = d[7]*gain_lin_QPSK; y0_16[7] = d[7]*gain_lin_QPSK;
for (i=0,j=0,m=0; i<6; i++,j+=2) { for (i=0,j=0,m=0; i<6; i++,j+=2) {
if ((i!=(frame_parms->nushift))&&(i!=(frame_parms->nushift+3))) { if ((i!=(nushift))&&(i!=((nushift+3)%6))) {
y0[j] += y0_16[m++]; y0[j] += y0_16[m++];
y0[j+1] += y0_16[m++]; y0[j+1] += y0_16[m++];
} }
@@ -385,7 +386,7 @@ void generate_phich(LTE_DL_FRAME_PARMS *frame_parms,
y0_16[7] = d[15]*gain_lin_QPSK; y0_16[7] = d[15]*gain_lin_QPSK;
for (i=0,j=0,m=0; i<6; i++,j+=2) { for (i=0,j=0,m=0; i<6; i++,j+=2) {
if ((i!=(frame_parms->nushift))&&(i!=(frame_parms->nushift+3))) { if ((i!=(nushift))&&(i!=((nushift+3)%6))) {
y0[j] += y0_16[m++]; y0[j] += y0_16[m++];
y0[j+1] += y0_16[m++]; y0[j+1] += y0_16[m++];
} }
@@ -410,7 +411,7 @@ void generate_phich(LTE_DL_FRAME_PARMS *frame_parms,
y0_16[7] = d[23]*gain_lin_QPSK; y0_16[7] = d[23]*gain_lin_QPSK;
for (i=0,j=0,m=0; i<6; i++,j+=2) { for (i=0,j=0,m=0; i<6; i++,j+=2) {
if ((i!=(frame_parms->nushift))&&(i!=(frame_parms->nushift+3))) { if ((i!=(nushift))&&(i!=((nushift+3)%6))) {
y0[j] += y0_16[m++]; y0[j] += y0_16[m++];
y0[j+1] += y0_16[m++]; y0[j+1] += y0_16[m++];
} }
@@ -533,7 +534,7 @@ void generate_phich(LTE_DL_FRAME_PARMS *frame_parms,
y1_16[7] = -y0_16[5]; y1_16[7] = -y0_16[5];
for (i=0,j=0,m=0; i<6; i++,j+=2) { for (i=0,j=0,m=0; i<6; i++,j+=2) {
if ((i!=(frame_parms->nushift))&&(i!=(frame_parms->nushift+3))) { if ((i!=(nushift))&&(i!=((nushift+3)%6))) {
y0[j] += y0_16[m]; y0[j] += y0_16[m];
y1[j] += y1_16[m++]; y1[j] += y1_16[m++];
y0[j+1] += y0_16[m]; y0[j+1] += y0_16[m];
@@ -644,7 +645,7 @@ void generate_phich(LTE_DL_FRAME_PARMS *frame_parms,
y0_16[7] = d[7]*gain_lin_QPSK; y0_16[7] = d[7]*gain_lin_QPSK;
for (i=0,j=0,m=0; i<6; i++,j+=2) { for (i=0,j=0,m=0; i<6; i++,j+=2) {
if ((i!=(frame_parms->nushift))&&(i!=(frame_parms->nushift+3))) { if ((i!=(nushift))&&(i!=((nushift+3)%6))) {
y0[j] += y0_16[m++]; y0[j] += y0_16[m++];
y0[j+1] += y0_16[m++]; y0[j+1] += y0_16[m++];
} }
openair1/PHY/LTE_TRANSPORT/prach.c
+29 -37
View File
@@ -62,9 +62,7 @@ void rx_prach0(PHY_VARS_eNB *eNB,
uint8_t Ncs_config; uint8_t Ncs_config;
uint8_t restricted_set; uint8_t restricted_set;
uint8_t n_ra_prb; uint8_t n_ra_prb;
int16_t *prachF=NULL;
int16_t **rxsigF=NULL; int16_t **rxsigF=NULL;
int16_t *prach2;
uint8_t preamble_index; uint8_t preamble_index;
uint16_t NCS,NCS2; uint16_t NCS,NCS2;
uint16_t preamble_offset=0,preamble_offset_old; uint16_t preamble_offset=0,preamble_offset_old;
@@ -77,12 +75,10 @@ void rx_prach0(PHY_VARS_eNB *eNB,
uint8_t not_found; uint8_t not_found;
int k=0; int k=0;
uint16_t u; uint16_t u;
int16_t *Xu=0; c16_t *Xu=0;
uint16_t offset;
int16_t Ncp; int16_t Ncp;
uint16_t first_nonzero_root_idx=0; uint16_t first_nonzero_root_idx=0;
uint8_t new_dft=0; uint8_t new_dft=0;
uint8_t aa;
int32_t lev; int32_t lev;
int16_t levdB; int16_t levdB;
int fft_size=0,log2_ifft_size; int fft_size=0,log2_ifft_size;
@@ -130,14 +126,11 @@ void rx_prach0(PHY_VARS_eNB *eNB,
tdd_mapindex,Nf); tdd_mapindex,Nf);
} }
int16_t *prach[nb_rx];
uint8_t prach_fmt = get_prach_fmt(prach_ConfigIndex,frame_type); uint8_t prach_fmt = get_prach_fmt(prach_ConfigIndex,frame_type);
uint16_t N_ZC = (prach_fmt <4)?839:139; uint16_t N_ZC = (prach_fmt <4)?839:139;
if (eNB) { if (eNB) {
if (br_flag == 1) { if (br_flag == 1) {
prach_ifftp = eNB->prach_vars_br.prach_ifft[ce_level]; prach_ifftp = eNB->prach_vars_br.prach_ifft[ce_level];
prachF = eNB->prach_vars_br.prachF;
rxsigF = eNB->prach_vars_br.rxsigF[ce_level]; rxsigF = eNB->prach_vars_br.rxsigF[ce_level];
if (LOG_DEBUGFLAG(PRACH)) { if (LOG_DEBUGFLAG(PRACH)) {
@@ -151,7 +144,6 @@ void rx_prach0(PHY_VARS_eNB *eNB,
} }
} else { } else {
prach_ifftp = eNB->prach_vars.prach_ifft[0]; prach_ifftp = eNB->prach_vars.prach_ifft[0];
prachF = eNB->prach_vars.prachF;
rxsigF = eNB->prach_vars.rxsigF[0]; rxsigF = eNB->prach_vars.rxsigF[0];
//if (LOG_DEBUGFLAG(PRACH)) { //if (LOG_DEBUGFLAG(PRACH)) {
@@ -179,7 +171,8 @@ void rx_prach0(PHY_VARS_eNB *eNB,
AssertFatal(ru!=NULL,"ru is null\n"); AssertFatal(ru!=NULL,"ru is null\n");
int8_t dBEn0=0; int8_t dBEn0=0;
for (aa=0; aa<nb_rx; aa++) { int16_t *prach[nb_rx];
for (int aa=0; aa<nb_rx; aa++) {
if (ru->if_south == LOCAL_RF || ru->function == NGFI_RAU_IF5) { // set the time-domain signal if we have to use it in this node if (ru->if_south == LOCAL_RF || ru->function == NGFI_RAU_IF5) { // set the time-domain signal if we have to use it in this node
// DJP - indexing below in subframe zero takes us off the beginning of the array??? // DJP - indexing below in subframe zero takes us off the beginning of the array???
prach[aa] = (int16_t *)&ru->common.rxdata[aa][(subframe*fp->samples_per_tti)-ru->N_TA_offset]; prach[aa] = (int16_t *)&ru->common.rxdata[aa][(subframe*fp->samples_per_tti)-ru->N_TA_offset];
@@ -291,9 +284,9 @@ void rx_prach0(PHY_VARS_eNB *eNB,
LOG_D(PHY,"rx_prach: Doing FFT for N_RB_UL %d nb_rx:%d Ncp:%d\n",fp->N_RB_UL, nb_rx, Ncp); LOG_D(PHY,"rx_prach: Doing FFT for N_RB_UL %d nb_rx:%d Ncp:%d\n",fp->N_RB_UL, nb_rx, Ncp);
} }
for (aa=0; aa<nb_rx; aa++) { for (int aa=0; aa<nb_rx; aa++) {
AssertFatal(prach[aa]!=NULL,"prach[%d] is null\n",aa); AssertFatal(prach[aa]!=NULL,"prach[%d] is null\n",aa);
prach2 = prach[aa] + (Ncp<<1); int16_t *prach2 = prach[aa] + (Ncp<<1); // <<1 because type int16 but input is c16
// do DFT // do DFT
switch (fp->N_RB_UL) { switch (fp->N_RB_UL) {
@@ -562,37 +555,38 @@ void rx_prach0(PHY_VARS_eNB *eNB,
if (new_dft == 1) { if (new_dft == 1) {
new_dft = 0; new_dft = 0;
if (br_flag == 1) { if (br_flag == 1) {
Xu=(int16_t *)eNB->X_u_br[ce_level][preamble_offset-first_nonzero_root_idx]; Xu=(c16_t *)eNB->X_u_br[ce_level][preamble_offset-first_nonzero_root_idx];
prach_ifft = prach_ifftp[prach_ifft_cnt++]; prach_ifft = prach_ifftp[prach_ifft_cnt++];
if (eNB->prach_vars_br.repetition_number[ce_level]==1) memset(prach_ifft,0,((N_ZC==839)?2048:256)*sizeof(int32_t)); if (eNB->prach_vars_br.repetition_number[ce_level]==1) memset(prach_ifft,0,((N_ZC==839)?2048:256)*sizeof(int32_t));
} else { } else {
Xu=(int16_t *)eNB->X_u[preamble_offset-first_nonzero_root_idx]; Xu=(c16_t *)eNB->X_u[preamble_offset-first_nonzero_root_idx];
prach_ifft = prach_ifftp[0]; prach_ifft = prach_ifftp[0];
memset(prach_ifft,0,((N_ZC==839) ? 2048 : 256)*sizeof(int32_t)); memset(prach_ifft,0,((N_ZC==839) ? 2048 : 256)*sizeof(int32_t));
} }
c16_t prachF[1024] __attribute__((aligned(32)))={0};
memset(prachF, 0, sizeof(int16_t)*2*1024 );
if (LOG_DUMPFLAG(PRACH))
if (LOG_DUMPFLAG(PRACH)) { for (int z=0; z<nb_rx; z++)
if (prach[0]!= NULL) LOG_M("prach_rx0.m","prach_rx0",prach[0],6144+792,1,1); if( prach[z] ) {
char tmp[128];
LOG_M("prach_rx1.m","prach_rx1",prach[1],6144+792,1,1); sprintf(tmp,"prach_rx%d.m", z);
LOG_M("prach_rxF0.m","prach_rxF0",rxsigF[0],12288,1,1); LOG_M(tmp,tmp,prach[z],6144+792,1,1);
LOG_M("prach_rxF1.m","prach_rxF1",rxsigF[1],12288,1,1); sprintf(tmp,"prach_rxF%d.m", z);
} LOG_M(tmp,tmp,rxsigF[z],12288,1,1);
}
for (aa=0; aa<nb_rx; aa++) {
for (int aa=0; aa<nb_rx; aa++) {
// Do componentwise product with Xu* on each antenna // Do componentwise product with Xu* on each antenna
k=0; k=0;
for (offset=0; offset<(N_ZC<<1); offset+=2) { for (int offset=0; offset<N_ZC; offset++) {
prachF[offset] = (int16_t)(((int32_t)Xu[offset]*rxsigF[aa][k] + (int32_t)Xu[offset+1]*rxsigF[aa][k+1])>>15); prachF[offset].r = (int16_t)(((int32_t)Xu[offset].r*rxsigF[aa][k] + (int32_t)Xu[offset].i*rxsigF[aa][k+1])>>15);
prachF[offset+1] = (int16_t)(((int32_t)Xu[offset]*rxsigF[aa][k+1] - (int32_t)Xu[offset+1]*rxsigF[aa][k])>>15); prachF[offset].i = (int16_t)(((int32_t)Xu[offset].r*rxsigF[aa][k+1] - (int32_t)Xu[offset].i*rxsigF[aa][k])>>15);
k+=2; k+=2;
if (k==(12*2*fp->ofdm_symbol_size)) if (k==(12*2*fp->ofdm_symbol_size))
k=0; k=0;
} }
@@ -600,13 +594,13 @@ void rx_prach0(PHY_VARS_eNB *eNB,
// Now do IFFT of size 1024 (N_ZC=839) or 256 (N_ZC=139) // Now do IFFT of size 1024 (N_ZC=839) or 256 (N_ZC=139)
if (N_ZC == 839) { if (N_ZC == 839) {
log2_ifft_size = 10; log2_ifft_size = 10;
idft(IDFT_1024,prachF,prach_ifft_tmp,1); idft(IDFT_1024,(int16_t*)prachF,prach_ifft_tmp,1);
// compute energy and accumulate over receive antennas and repetitions for BR // compute energy and accumulate over receive antennas and repetitions for BR
for (i=0; i<2048; i++) for (i=0; i<2048; i++)
prach_ifft[i] += (prach_ifft_tmp[i<<1]*prach_ifft_tmp[i<<1] + prach_ifft_tmp[1+(i<<1)]*prach_ifft_tmp[1+(i<<1)])>>9; prach_ifft[i] += (prach_ifft_tmp[i<<1]*prach_ifft_tmp[i<<1] + prach_ifft_tmp[1+(i<<1)]*prach_ifft_tmp[1+(i<<1)])>>9;
} else { } else {
idft(IDFT_256,prachF,prach_ifft_tmp,1); idft(IDFT_256,(int16_t*)prachF,prach_ifft_tmp,1);
log2_ifft_size = 8; log2_ifft_size = 8;
// compute energy and accumulate over receive antennas and repetitions for BR // compute energy and accumulate over receive antennas and repetitions for BR
@@ -674,14 +668,12 @@ void rx_prach0(PHY_VARS_eNB *eNB,
if (br_flag == 0) { if (br_flag == 0) {
LOG_M("rxsigF.m","prach_rxF",&rxsigF[0][0],12288,1,1); LOG_M("rxsigF.m","prach_rxF",&rxsigF[0][0],12288,1,1);
LOG_M("prach_rxF_comp0.m","prach_rxF_comp0",prachF,1024,1,1);
LOG_M("Xu.m","xu",Xu,N_ZC,1,1); LOG_M("Xu.m","xu",Xu,N_ZC,1,1);
LOG_M("prach_ifft0.m","prach_t0",prach_ifft,1024,1,1); LOG_M("prach_ifft0.m","prach_t0",prach_ifft,1024,1,1);
LOG_M("SF2_3.m","sf2_3",&ru->common.rxdata[0][2*fp->samples_per_tti],2*fp->samples_per_tti,1,1); LOG_M("SF2_3.m","sf2_3",&ru->common.rxdata[0][2*fp->samples_per_tti],2*fp->samples_per_tti,1,1);
} else { } else {
LOG_E(PHY,"Dumping prach (br_flag %d), k = %d (n_ra_prb %d)\n",br_flag,k,n_ra_prb); LOG_E(PHY,"Dumping prach (br_flag %d), k = %d (n_ra_prb %d)\n",br_flag,k,n_ra_prb);
LOG_M("rxsigF_br.m","prach_rxF_br",&rxsigF[0][0],12288,1,1); LOG_M("rxsigF_br.m","prach_rxF_br",&rxsigF[0][0],12288,1,1);
LOG_M("prach_rxF_comp0_br.m","prach_rxF_comp0_br",prachF,1024,1,1);
LOG_M("Xu_br.m","xu_br",Xu,N_ZC,1,1); LOG_M("Xu_br.m","xu_br",Xu,N_ZC,1,1);
LOG_M("prach_ifft0_br.m","prach_t0_br",prach_ifft,1024,1,1); LOG_M("prach_ifft0_br.m","prach_t0_br",prach_ifft,1024,1,1);
} }
openair1/PHY/NR_ESTIMATION/nr_measurements_gNB.c
+1 -1
View File
@@ -67,7 +67,7 @@ int nr_est_timing_advance_pusch(PHY_VARS_gNB* gNB, int UE_id)
return max_pos - sync_pos; return max_pos - sync_pos;
} }
int nr_est_timing_advance_srs(const NR_DL_FRAME_PARMS *frame_parms, int nr_est_timing_advance_srs(const NR_DL_FRAME_PARMS *frame_parms,
const int32_t srs_estimated_channel_time[][frame_parms->ofdm_symbol_size]) { const int32_t srs_estimated_channel_time[][frame_parms->ofdm_symbol_size]) {
int timing_advance = 0; int timing_advance = 0;
int max_val = 0; int max_val = 0;
openair1/PHY/NR_ESTIMATION/nr_ul_channel_estimation.c
+348 -588
View File
File diff suppressed because it is too large Load Diff
openair1/PHY/NR_TRANSPORT/srs_rx.c
+1 -1
View File
@@ -183,4 +183,4 @@ int nr_get_srs_signal(PHY_VARS_gNB *gNB,
} else { } else {
return 0; return 0;
} }
} }
openair1/PHY/NR_UE_TRANSPORT/nr_dlsch_demodulation.c
+37 -37
View File
@@ -480,45 +480,45 @@ int nr_rx_pdsch(PHY_VARS_NR_UE *ue,
nbSymb = dlsch1_harq->nb_symbols; nbSymb = dlsch1_harq->nb_symbols;
pduBitmap = dlsch1_harq->pduBitmap; pduBitmap = dlsch1_harq->pduBitmap;
} }
/* Check for PTRS bitmap and process it respectively */ /* Check for PTRS bitmap and process it respectively */
if((pduBitmap & 0x1) && (type == PDSCH)) { if((pduBitmap & 0x1) && (type == PDSCH)) {
nr_pdsch_ptrs_processing(ue, nr_pdsch_ptrs_processing(ue,
pdsch_vars, pdsch_vars,
frame_parms, frame_parms,
dlsch0_harq, dlsch0_harq,
dlsch1_harq, dlsch1_harq,
gNB_id, gNB_id,
nr_slot_rx, nr_slot_rx,
symbol, symbol,
(nb_rb_pdsch*12), (nb_rb_pdsch*12),
dlsch[0]->rnti,rx_type); dlsch[0]->rnti,rx_type);
pdsch_vars[gNB_id]->dl_valid_re[symbol-1] -= pdsch_vars[gNB_id]->ptrs_re_per_slot[0][symbol]; pdsch_vars[gNB_id]->dl_valid_re[symbol-1] -= pdsch_vars[gNB_id]->ptrs_re_per_slot[0][symbol];
} }
/* at last symbol in a slot calculate LLR's for whole slot */ /* at last symbol in a slot calculate LLR's for whole slot */
if(symbol == (startSymbIdx + nbSymb -1)) { if(symbol == (startSymbIdx + nbSymb -1)) {
for(uint8_t i =startSymbIdx; i < (startSymbIdx+nbSymb);i++) { for(uint8_t i =startSymbIdx; i < (startSymbIdx+nbSymb);i++) {
/* re evaluating the first symbol flag as LLR's are done in symbol loop */ /* re evaluating the first symbol flag as LLR's are done in symbol loop */
if(i == startSymbIdx && i < 3) { if(i == startSymbIdx && i < 3) {
first_symbol_flag =1; first_symbol_flag =1;
} }
else { else {
first_symbol_flag=0; first_symbol_flag=0;
} }
/* Calculate LLR's for each symbol */ /* Calculate LLR's for each symbol */
nr_dlsch_llr(pdsch_vars, frame_parms, nr_dlsch_llr(pdsch_vars, frame_parms,
rxdataF_comp_ptr, dl_ch_mag_ptr, rxdataF_comp_ptr, dl_ch_mag_ptr,
dlsch0_harq, dlsch1_harq, dlsch0_harq, dlsch1_harq,
rx_type, harq_pid, rx_type, harq_pid,
gNB_id, gNB_id_i, gNB_id, gNB_id_i,
first_symbol_flag, first_symbol_flag,
i, nb_rb_pdsch, i, nb_rb_pdsch,
codeword_TB0, codeword_TB1, codeword_TB0, codeword_TB1,
pdsch_vars[gNB_id]->dl_valid_re[i-1], pdsch_vars[gNB_id]->dl_valid_re[i-1],
nr_slot_rx, beamforming_mode); nr_slot_rx, beamforming_mode);
} }
int dmrs_type = dlsch[0]->harq_processes[harq_pid]->dmrsConfigType; int dmrs_type = dlsch[0]->harq_processes[harq_pid]->dmrsConfigType;
uint8_t nb_re_dmrs; uint8_t nb_re_dmrs;
uint16_t dmrs_len = get_num_dmrs(dlsch[0]->harq_processes[harq_pid]->dlDmrsSymbPos); uint16_t dmrs_len = get_num_dmrs(dlsch[0]->harq_processes[harq_pid]->dlDmrsSymbPos);
@@ -528,29 +528,29 @@ int nr_rx_pdsch(PHY_VARS_NR_UE *ue,
nb_re_dmrs = 4*dlsch[0]->harq_processes[harq_pid]->n_dmrs_cdm_groups; nb_re_dmrs = 4*dlsch[0]->harq_processes[harq_pid]->n_dmrs_cdm_groups;
} }
dlsch[0]->harq_processes[harq_pid]->G = nr_get_G(dlsch[0]->harq_processes[harq_pid]->nb_rb, dlsch[0]->harq_processes[harq_pid]->G = nr_get_G(dlsch[0]->harq_processes[harq_pid]->nb_rb,
dlsch[0]->harq_processes[harq_pid]->nb_symbols, dlsch[0]->harq_processes[harq_pid]->nb_symbols,
nb_re_dmrs, nb_re_dmrs,
dmrs_len, dmrs_len,
dlsch[0]->harq_processes[harq_pid]->Qm, dlsch[0]->harq_processes[harq_pid]->Qm,
dlsch[0]->harq_processes[harq_pid]->Nl); dlsch[0]->harq_processes[harq_pid]->Nl);
nr_dlsch_layer_demapping(pdsch_vars[gNB_id]->llr, nr_dlsch_layer_demapping(pdsch_vars[gNB_id]->llr,
dlsch[0]->harq_processes[harq_pid]->Nl, dlsch[0]->harq_processes[harq_pid]->Nl,
dlsch[0]->harq_processes[harq_pid]->Qm, dlsch[0]->harq_processes[harq_pid]->Qm,
dlsch[0]->harq_processes[harq_pid]->G, dlsch[0]->harq_processes[harq_pid]->G,
codeword_TB0, codeword_TB0,
codeword_TB1, codeword_TB1,
pdsch_vars[gNB_id]->layer_llr); pdsch_vars[gNB_id]->layer_llr);
} }
stop_meas(&ue->generic_stat_bis[proc->thread_id][slot]); stop_meas(&ue->generic_stat_bis[proc->thread_id][slot]);
if (cpumeas(CPUMEAS_GETSTATE)) if (cpumeas(CPUMEAS_GETSTATE))
LOG_D(PHY, "[AbsSFN %u.%d] Slot%d Symbol %d: LLR Computation %5.2f \n",frame,nr_slot_rx,slot,symbol,ue->generic_stat_bis[proc->thread_id][slot].p_time/(cpuf*1000.0)); LOG_D(PHY, "[AbsSFN %u.%d] Slot%d Symbol %d: LLR Computation %5.2f \n",frame,nr_slot_rx,slot,symbol,ue->generic_stat_bis[proc->thread_id][slot].p_time/(cpuf*1000.0));
// Please keep it: useful for debugging // Please keep it: useful for debugging
#ifdef DEBUG_PDSCH_RX #ifdef DEBUG_PDSCH_RX
char filename[50]; char filename[50];
uint8_t aa = 0; uint8_t aa = 0;
snprintf(filename, 50, "rxdataF0_symb_%d_nr_slot_rx_%d.m", symbol, nr_slot_rx); snprintf(filename, 50, "rxdataF0_symb_%d_nr_slot_rx_%d.m", symbol, nr_slot_rx);
write_output(filename, "rxdataF0", &common_vars->common_vars_rx_data_per_thread[proc->thread_id].rxdataF[0][0], NR_SYMBOLS_PER_SLOT*frame_parms->ofdm_symbol_size, 1, 1); write_output(filename, "rxdataF0", &common_vars->common_vars_rx_data_per_thread[proc->thread_id].rxdataF[0][0], NR_SYMBOLS_PER_SLOT*frame_parms->ofdm_symbol_size, 1, 1);
openair1/PHY/TOOLS/8bit_rxdemux.c
-62
View File
@@ -1,62 +0,0 @@
/*
* 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
*/
#include "PHY/types.h"
#include "PHY/defs.h"
#include "PHY/extern.h"
void bit8_rxdemux(int length,int offset)
{
int i;
short *rx1_ptr = (short *)&PHY_vars->rx_vars[1].RX_DMA_BUFFER[offset];
short *rx0_ptr = (short *)&PHY_vars->rx_vars[0].RX_DMA_BUFFER[offset];
char *rx0_ptr2 = (char *)(&PHY_vars->rx_vars[0].RX_DMA_BUFFER[offset]);
// short tmp;
short r0,i0,r1,i1;
// printf("demuxing: %d,%d\n",length,offset);
// printf("%x %x\n",PHY_vars->chsch_data[0].CHSCH_f_sync[0], PHY_vars->chsch_data[0].CHSCH_f_sync[1]);
for (i=0; i<length; i++) {
r0= (short)(rx0_ptr2[i<<2]); // Re 0
i0= (short)(rx0_ptr2[(i<<2)+1]); // Im 0
r1= (short)(rx0_ptr2[(i<<2)+2]); // Re 1
i1= (short)(rx0_ptr2[(i<<2)+3]); // Im 1
rx0_ptr[(i<<1)] = r0;
rx0_ptr[(i<<1)+1] =i0;
rx1_ptr[i<<1] =r1;
rx1_ptr[(i<<1)+1] =i1;
}
// printf("%x %x\n",PHY_vars->chsch_data[0].CHSCH_f_sync[0], PHY_vars->chsch_data[0].CHSCH_f_sync[1]);
}
openair1/PHY/TOOLS/8bit_txmux.c
-48
View File
@@ -1,48 +0,0 @@
/*
* 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
*/
#include "PHY/types.h"
#include "PHY/defs.h"
#include "PHY/extern.h"
void bit8_txmux(int length,int offset)
{
int i;
short *dest,*dest2;
for (i=0; i<length; i++) {
dest = (short *)&PHY_vars->tx_vars[0].TX_DMA_BUFFER[i+offset];
dest2 = (short *)&PHY_vars->tx_vars[1].TX_DMA_BUFFER[i+offset];
((char *)dest)[0] = (char)(dest[0]>>BIT8_TX_SHIFT);
((char *)dest)[1] = (char)(dest[1]>>BIT8_TX_SHIFT);
((char *)dest)[2] = (char)(dest2[0]>>BIT8_TX_SHIFT);
((char *)dest)[3] = (char)(dest2[1]>>BIT8_TX_SHIFT);
}
}
openair1/PHY/TOOLS/cadd_sv.c
-163
View File
@@ -1,163 +0,0 @@
/*
* 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
*/
#include "defs.h"
static __m128i alpha_128 __attribute__ ((aligned(16)));
static __m128i shift __attribute__ ((aligned(16)));
int add_cpx_vector(short *x,
short *alpha,
short *y,
unsigned int N)
{
unsigned int i; // loop counter
__m128i *x_128;
__m128i *y_128;
x_128 = (__m128i *)&x[0];
y_128 = (__m128i *)&y[0];
alpha_128 = _mm_set1_epi32(*((int*)alpha));
// we compute 4 cpx multiply for each loop
for(i=0; i<(N>>3); i++) {
y_128[0] = _mm_adds_epi16(alpha_128, x_128[0]);
y_128[1] = _mm_adds_epi16(alpha_128, x_128[1]);
y_128[2] = _mm_adds_epi16(alpha_128, x_128[2]);
y_128[3] = _mm_adds_epi16(alpha_128, x_128[3]);
x_128+=4;
y_128 +=4;
}
return (0);
}
int add_vector32_scalar(short *x,
int alpha,
short *y,
unsigned int N)
{
unsigned int i; // loop counter
__m128i *x_128;
__m128i *y_128;
x_128 = (__m128i *)&x[0];
y_128 = (__m128i *)&y[0];
alpha_128 = _mm_setr_epi32(alpha,0,alpha,0);
// we compute 4 cpx multiply for each loop
for(i=0; i<(N>>3); i++) {
y_128[0] = _mm_add_epi32(alpha_128, x_128[0]);
y_128[1] = _mm_add_epi32(alpha_128, x_128[1]);
y_128[2] = _mm_add_epi32(alpha_128, x_128[2]);
y_128[3] = _mm_add_epi32(alpha_128, x_128[3]);
x_128+=4;
y_128 +=4;
}
return (0);
}
int add_real_vector64_scalar(short *x,
long long int a,
short *y,
unsigned int N)
{
unsigned int i; // loop counter
__m128i *x_128;
__m128i *y_128;
x_128 = (__m128i *)&x[0];
y_128 = (__m128i *)&y[0];
alpha_128 = _mm_set1_epi64((__m64) a);
// we compute 4 cpx multiply for each loop
for(i=0; i<(N>>3); i++) {
y_128[0] = _mm_add_epi64(alpha_128, x_128[0]);
y_128[1] = _mm_add_epi64(alpha_128, x_128[1]);
y_128[2] = _mm_add_epi64(alpha_128, x_128[2]);
y_128[3] = _mm_add_epi64(alpha_128, x_128[3]);
x_128+=4;
y_128+=4;
}
return(0);
}
#ifdef MAIN
#include <stdio.h>
main ()
{
short input[256] __attribute__((aligned(16)));
short output[256] __attribute__((aligned(16)));
int i;
c16_t alpha;
Zero_Buffer(output,256*2);
input[0] = 100;
input[1] = 200;
input[2] = 100;
input[3] = 200;
input[4] = 1234;
input[5] = -1234;
input[6] = 1234;
input[7] = -1234;
input[8] = 100;
input[9] = 200;
input[10] = 100;
input[11] = 200;
input[12] = 1000;
input[13] = 2000;
input[14] = 1000;
input[15] = 2000;
alpha.r = 10;
alpha.i = -10;
add_cpx_vector(input,(short*) &alpha,input,8);
for (i=0; i<16; i+=2)
printf("output[%d] = %d + %d i\n",i,input[i],input[i+1]);
}
#endif //MAIN
openair1/PHY/TOOLS/cadd_vv.c
+1 -215
View File
@@ -23,153 +23,6 @@
#include "tools_defs.h" #include "tools_defs.h"
int add_vector16(short *x,
short *y,
short *z,
unsigned int N)
{
unsigned int i; // loop counter
__m128i *x_128;
__m128i *y_128;
__m128i *z_128;
x_128 = (__m128i *)&x[0];
y_128 = (__m128i *)&y[0];
z_128 = (__m128i *)&z[0];
for(i=0; i<(N>>5); i++) {
/*
printf("i = %d\n",i);
print_shorts(x_128[0],"x[0]=");
print_shorts(y_128[0],"y[0]=");
*/
z_128[0] = _mm_adds_epi16(x_128[0],y_128[0]);
/*
print_shorts(z_128[0],"z[0]=");
print_shorts(x_128[1],"x[1]=");
print_shorts(y_128[1],"y[1]=");
*/
z_128[1] = _mm_adds_epi16(x_128[1],y_128[1]);
/*
print_shorts(z_128[1],"z[1]=");
print_shorts(x_128[2],"x[2]=");
print_shorts(y_128[2],"y[2]=");
*/
z_128[2] = _mm_adds_epi16(x_128[2],y_128[2]);
/*
print_shorts(z_128[2],"z[2]=");
print_shorts(x_128[3],"x[3]=");
print_shorts(y_128[3],"y[3]=");
*/
z_128[3] = _mm_adds_epi16(x_128[3],y_128[3]);
/*
print_shorts(z_128[3],"z[3]=");
*/
x_128 +=4;
y_128 +=4;
z_128 +=4;
}
_mm_empty();
_m_empty();
return(0);
}
/*
print_shorts64(__m64 x,char *s) {
printf("%s ",s);
printf("%d %d %d %d\n",((short *)&x)[0],((short *)&x)[1],((short *)&x)[2],((short *)&x)[3]);
}
*/
int add_vector16_64(short *x,
short *y,
short *z,
unsigned int N)
{
unsigned int i; // loop counter
__m64 *x_64;
__m64 *y_64;
__m64 *z_64;
x_64 = (__m64 *)&x[0];
y_64 = (__m64 *)&y[0];
z_64 = (__m64 *)&z[0];
for(i=0; i<(N>>2); i++) {
/*
printf("i = %d\n",i);
print_shorts64(x_64[i],"x[i]=");
print_shorts64(y_64[i],"y[i]=");
*/
z_64[i] = _mm_adds_pi16(x_64[i],y_64[i]);
/*
print_shorts64(z_64[i],"z[i]=");
*/
}
_mm_empty();
_m_empty();
return(0);
}
int add_cpx_vector32(short *x,
short *y,
short *z,
unsigned int N)
{
unsigned int i; // loop counter
__m128i *x_128;
__m128i *y_128;
__m128i *z_128;
x_128 = (__m128i *)&x[0];
y_128 = (__m128i *)&y[0];
z_128 = (__m128i *)&z[0];
for(i=0; i<(N>>3); i++) {
z_128[0] = _mm_add_epi32(x_128[0],y_128[0]);
z_128[1] = _mm_add_epi32(x_128[1],y_128[1]);
z_128[2] = _mm_add_epi32(x_128[2],y_128[2]);
z_128[3] = _mm_add_epi32(x_128[3],y_128[3]);
x_128+=4;
y_128 +=4;
z_128 +=4;
}
_mm_empty();
_m_empty();
return(0);
}
int32_t sub_cpx_vector16(int16_t *x, int32_t sub_cpx_vector16(int16_t *x,
int16_t *y, int16_t *y,
int16_t *z, int16_t *z,
@@ -201,73 +54,6 @@ int32_t sub_cpx_vector16(int16_t *x,
int add_real_vector64(short *x,
short* y,
short *z,
unsigned int N)
{
unsigned int i; // loop counter
__m128i *x_128;
__m128i *y_128;
__m128i *z_128;
x_128 = (__m128i *)&x[0];
y_128 = (__m128i *)&y[0];
z_128 = (__m128i *)&z[0];
for(i=0; i<(N>>3); i++) {
z_128[0] = _mm_add_epi64(x_128[0], y_128[0]);
z_128[1] = _mm_add_epi64(x_128[1], y_128[1]);
z_128[2] = _mm_add_epi64(x_128[2], y_128[2]);
z_128[3] = _mm_add_epi64(x_128[3], y_128[3]);
x_128+=4;
y_128+=4;
z_128+=4;
}
_mm_empty();
_m_empty();
return(0);
}
int sub_real_vector64(short *x,
short* y,
short *z,
unsigned int N)
{
unsigned int i; // loop counter
__m128i *x_128;
__m128i *y_128;
__m128i *z_128;
x_128 = (__m128i *)&x[0];
y_128 = (__m128i *)&y[0];
z_128 = (__m128i *)&z[0];
for(i=0; i<(N>>3); i++) {
z_128[0] = _mm_sub_epi64(x_128[0], y_128[0]);
z_128[1] = _mm_sub_epi64(x_128[1], y_128[1]);
z_128[2] = _mm_sub_epi64(x_128[2], y_128[2]);
z_128[3] = _mm_sub_epi64(x_128[3], y_128[3]);
x_128+=4;
y_128+=4;
z_128+=4;
}
_mm_empty();
_m_empty();
return(0);
}
#ifdef MAIN #ifdef MAIN
#include <stdio.h> #include <stdio.h>
@@ -280,7 +66,7 @@ main ()
int i; int i;
c16_t alpha; c16_t alpha;
Zero_Buffer(output,256*2); memset(output,0,256*2);
input[0] = 100; input[0] = 100;
input[1] = 200; input[1] = 200;
openair1/PHY/TOOLS/cmult_sv.c
+114 -215
View File
@@ -22,28 +22,6 @@
#include "PHY/sse_intrin.h" #include "PHY/sse_intrin.h"
#include "tools_defs.h" #include "tools_defs.h"
#if defined(__x86_64__) || defined(__i386__)
#define simd_q15_t __m128i
#define simdshort_q15_t __m64
#define shiftright_int16(a,shift) _mm_srai_epi16(a,shift)
#define set1_int16(a) _mm_set1_epi16(a)
#define mulhi_int16(a,b) _mm_slli_epi16(_mm_mulhi_epi16(a,b),1)
#define mulhi_s1_int16(a,b) _mm_slli_epi16(_mm_mulhi_epi16(a,b),2)
#define adds_int16(a,b) _mm_adds_epi16(a,b)
#define mullo_int16(a,b) _mm_mullo_epi16(a,b)
#elif defined(__arm__)
#define simd_q15_t int16x8_t
#define simdshort_q15_t int16x4_t
#define shiftright_int16(a,shift) vshrq_n_s16(a,shift)
#define set1_int16(a) vdupq_n_s16(a)
#define mulhi_int16(a,b) vqdmulhq_s16(a,b)
#define mulhi_s1_int16(a,b) vshlq_n_s16(vqdmulhq_s16(a,b),1)
#define adds_int16(a,b) vqaddq_s16(a,b)
#define mullo_int16(a,b) vmulq_s16(a,b)
#define _mm_empty()
#define _m_empty()
#endif
void multadd_complex_vector_real_scalar(int16_t *x, void multadd_complex_vector_real_scalar(int16_t *x,
int16_t alpha, int16_t alpha,
@@ -111,216 +89,137 @@ void multadd_real_vector_complex_scalar(int16_t *x,
} }
} }
void multadd_real_four_symbols_vector_complex_scalar(int16_t *x,
int16_t *alpha,
int16_t *y)
{
// do 8 multiplications at a time
simd_q15_t alpha_r_128,alpha_i_128,yr,yi,*x_128=(simd_q15_t*)x;
simd_q15_t y_128;
y_128 = _mm_loadu_si128((simd_q15_t*)y);
alpha_r_128 = set1_int16(alpha[0]);
alpha_i_128 = set1_int16(alpha[1]);
yr = mulhi_s1_int16(alpha_r_128,x_128[0]);
yi = mulhi_s1_int16(alpha_i_128,x_128[0]);
y_128 = _mm_adds_epi16(y_128,_mm_unpacklo_epi16(yr,yi));
y_128 = _mm_adds_epi16(y_128,_mm_unpackhi_epi16(yr,yi));
_mm_storeu_si128((simd_q15_t*)y, y_128);
_mm_empty();
_m_empty();
}
#ifdef __AVX2__
void rotate_cpx_vector(c16_t *x, void rotate_cpx_vector(c16_t *x,
c16_t *alpha, c16_t *alpha,
c16_t *y, c16_t *y,
uint32_t N, uint32_t N,
uint16_t output_shift) uint16_t output_shift)
{ {
// Multiply elementwise two complex vectors of N elements // multiply a complex vector with a complex value (alpha)
// x - input 1 in the format |Re0 Im0 |,......,|Re(N-1) Im(N-1)| // stores result in y
// We assume x1 with a dynamic of 15 bit maximum // N is the number of complex numbers
// // output_shift reduces the result of the multiplication by this number of bits
// alpha - input 2 in the format |Re0 Im0| //AssertFatal(N%8==0, "To be developped");
// We assume x2 with a dynamic of 15 bit maximum #ifdef __AVX2__
// if ( (intptr_t)x%32 == 0 && !(intptr_t)y%32 == 0 && __builtin_cpu_supports("avx2")) {
// y - output in the format |Re0 Im0|,......,|Re(N-1) Im(N-1)| // output is 32 bytes aligned, but not the input
//
// N - the size f the vectors (this function does N cpx mpy. WARNING: N>=4; const c16_t for_re={alpha->r, -alpha->i};
// __m256i const alpha_for_real = _mm256_set1_epi32(*(uint32_t*)&for_re);
// log2_amp - increase the output amplitude by a factor 2^log2_amp (default is 0) const c16_t for_im={alpha->i, alpha->r};
// WARNING: log2_amp>0 can cause overflow!! __m256i const alpha_for_im= _mm256_set1_epi32(*(uint32_t*)&for_im);
__m256i const perm_mask =
uint32_t i; // loop counter _mm256_set_epi8(31,30,23,22,29,28,21,20,27,26,19,18,25,24,17,16,
15,14,7,6,13,12,5,4,11,10,3,2,9,8,1,0);
__m256i* xd= (__m256i*)x;
simd_q15_t *y_128,alpha_128; const __m256i *end=xd+N/8;
int32_t *xd=(int32_t *)x; for( __m256i* yd = (__m256i *)y; xd<end ; yd++, xd++) {
const __m256i xre = _mm256_srai_epi32(_mm256_madd_epi16(*xd,alpha_for_real),
#if defined(__x86_64__) || defined(__i386__) output_shift);
__m128i shift = _mm_cvtsi32_si128(output_shift); const __m256i xim = _mm256_srai_epi32(_mm256_madd_epi16(*xd,alpha_for_im),
register simd_q15_t m0,m1,m2,m3; output_shift);
// a bit faster than unpacklo+unpackhi+packs
((int16_t *)&alpha_128)[0] = alpha->r; const __m256i tmp=_mm256_packs_epi32(xre,xim);
((int16_t *)&alpha_128)[1] = -alpha->i; *yd=_mm256_shuffle_epi8(tmp,perm_mask);
((int16_t *)&alpha_128)[2] = alpha->i; }
((int16_t *)&alpha_128)[3] = alpha->r; c16_t* alpha16=(c16_t*) alpha, *yLast;
((int16_t *)&alpha_128)[4] = alpha->r; yLast=((c16_t*)y)+(N/8)*8;
((int16_t *)&alpha_128)[5] = -alpha->i; for (c16_t* xTail=(c16_t*) end;
((int16_t *)&alpha_128)[6] = alpha->i; xTail<((c16_t*) x)+N;
((int16_t *)&alpha_128)[7] = alpha->r; xTail++, yLast++) {
#elif defined(__arm__) *yLast=c16mulShift(*xTail,*alpha16,output_shift);
int32x4_t shift; }
int32x4_t ab_re0,ab_re1,ab_im0,ab_im1,re32,im32; } else {
int16_t reflip[8] __attribute__((aligned(16))) = {1,-1,1,-1,1,-1,1,-1};
int32x4x2_t xtmp;
((int16_t *)&alpha_128)[0] = alpha->r;
((int16_t *)&alpha_128)[1] = alpha->i;
((int16_t *)&alpha_128)[2] = alpha->r;
((int16_t *)&alpha_128)[3] = alpha->i;
((int16_t *)&alpha_128)[4] = alpha->r;
((int16_t *)&alpha_128)[5] = alpha->i;
((int16_t *)&alpha_128)[6] = alpha->r;
((int16_t *)&alpha_128)[7] = alpha->i;
int16x8_t bflip = vrev32q_s16(alpha_128);
int16x8_t bconj = vmulq_s16(alpha_128,*(int16x8_t *)reflip);
shift = vdupq_n_s32(-output_shift);
#endif #endif
y_128 = (simd_q15_t *) y; // Multiply elementwise two complex vectors of N elements
// x - input 1 in the format |Re0 Im0 |,......,|Re(N-1) Im(N-1)|
// We assume x1 with a dynamic of 15 bit maximum
//
// alpha - input 2 in the format |Re0 Im0|
// We assume x2 with a dynamic of 15 bit maximum
//
// y - output in the format |Re0 Im0|,......,|Re(N-1) Im(N-1)|
//
// N - the size f the vectors (this function does N cpx mpy. WARNING: N>=4;
//
// log2_amp - increase the output amplitude by a factor 2^log2_amp (default is 0)
// WARNING: log2_amp>0 can cause overflow!!
uint32_t i; // loop counter
for(i=0; i<N>>2; i++) { simd_q15_t *y_128,alpha_128;
int32_t *xd=(int32_t *)x;
#if defined(__x86_64__) || defined(__i386__) #if defined(__x86_64__) || defined(__i386__)
m0 = _mm_setr_epi32(xd[0],xd[0],xd[1],xd[1]); __m128i shift = _mm_cvtsi32_si128(output_shift);
m1 = _mm_setr_epi32(xd[2],xd[2],xd[3],xd[3]); register simd_q15_t m0,m1,m2,m3;
m2 = _mm_madd_epi16(m0,alpha_128); //complex multiply. result is 32bit [Re Im Re Im]
m3 = _mm_madd_epi16(m1,alpha_128); //complex multiply. result is 32bit [Re Im Re Im]
m2 = _mm_sra_epi32(m2,shift); // shift right by shift in order to compensate for the input amplitude
m3 = _mm_sra_epi32(m3,shift); // shift right by shift in order to compensate for the input amplitude
y_128[0] = _mm_packs_epi32(m2,m3); // pack in 16bit integers with saturation [re im re im re im re im] ((int16_t *)&alpha_128)[0] = alpha->r;
//print_ints("y_128[0]=", &y_128[0]); ((int16_t *)&alpha_128)[1] = -alpha->i;
((int16_t *)&alpha_128)[2] = alpha->i;
((int16_t *)&alpha_128)[3] = alpha->r;
((int16_t *)&alpha_128)[4] = alpha->r;
((int16_t *)&alpha_128)[5] = -alpha->i;
((int16_t *)&alpha_128)[6] = alpha->i;
((int16_t *)&alpha_128)[7] = alpha->r;
#elif defined(__arm__)
int32x4_t shift;
int32x4_t ab_re0,ab_re1,ab_im0,ab_im1,re32,im32;
int16_t reflip[8] __attribute__((aligned(16))) = {1,-1,1,-1,1,-1,1,-1};
int32x4x2_t xtmp;
((int16_t *)&alpha_128)[0] = alpha->r;
((int16_t *)&alpha_128)[1] = alpha->i;
((int16_t *)&alpha_128)[2] = alpha->r;
((int16_t *)&alpha_128)[3] = alpha->i;
((int16_t *)&alpha_128)[4] = alpha->r;
((int16_t *)&alpha_128)[5] = alpha->i;
((int16_t *)&alpha_128)[6] = alpha->r;
((int16_t *)&alpha_128)[7] = alpha->i;
int16x8_t bflip = vrev32q_s16(alpha_128);
int16x8_t bconj = vmulq_s16(alpha_128,*(int16x8_t *)reflip);
shift = vdupq_n_s32(-output_shift);
#endif
y_128 = (simd_q15_t *) y;
for(i=0; i<N>>2; i++) {
#if defined(__x86_64__) || defined(__i386__)
m0 = _mm_setr_epi32(xd[0],xd[0],xd[1],xd[1]);
m1 = _mm_setr_epi32(xd[2],xd[2],xd[3],xd[3]);
m2 = _mm_madd_epi16(m0,alpha_128); //complex multiply. result is 32bit [Re Im Re Im]
m3 = _mm_madd_epi16(m1,alpha_128); //complex multiply. result is 32bit [Re Im Re Im]
m2 = _mm_sra_epi32(m2,shift); // shift right by shift in order to compensate for the input amplitude
m3 = _mm_sra_epi32(m3,shift); // shift right by shift in order to compensate for the input amplitude
y_128[0] = _mm_packs_epi32(m2,m3); // pack in 16bit integers with saturation [re im re im re im re im]
//print_ints("y_128[0]=", &y_128[0]);
#elif defined(__arm__) #elif defined(__arm__)
ab_re0 = vmull_s16(((int16x4_t*)xd)[0],((int16x4_t*)&bconj)[0]); ab_re0 = vmull_s16(((int16x4_t*)xd)[0],((int16x4_t*)&bconj)[0]);
ab_re1 = vmull_s16(((int16x4_t*)xd)[1],((int16x4_t*)&bconj)[1]); ab_re1 = vmull_s16(((int16x4_t*)xd)[1],((int16x4_t*)&bconj)[1]);
ab_im0 = vmull_s16(((int16x4_t*)xd)[0],((int16x4_t*)&bflip)[0]); ab_im0 = vmull_s16(((int16x4_t*)xd)[0],((int16x4_t*)&bflip)[0]);
ab_im1 = vmull_s16(((int16x4_t*)xd)[1],((int16x4_t*)&bflip)[1]); ab_im1 = vmull_s16(((int16x4_t*)xd)[1],((int16x4_t*)&bflip)[1]);
re32 = vshlq_s32(vcombine_s32(vpadd_s32(((int32x2_t*)&ab_re0)[0],((int32x2_t*)&ab_re0)[1]), re32 = vshlq_s32(vcombine_s32(vpadd_s32(((int32x2_t*)&ab_re0)[0],((int32x2_t*)&ab_re0)[1]),
vpadd_s32(((int32x2_t*)&ab_re1)[0],((int32x2_t*)&ab_re1)[1])), vpadd_s32(((int32x2_t*)&ab_re1)[0],((int32x2_t*)&ab_re1)[1])),
shift); shift);
im32 = vshlq_s32(vcombine_s32(vpadd_s32(((int32x2_t*)&ab_im0)[0],((int32x2_t*)&ab_im0)[1]), im32 = vshlq_s32(vcombine_s32(vpadd_s32(((int32x2_t*)&ab_im0)[0],((int32x2_t*)&ab_im0)[1]),
vpadd_s32(((int32x2_t*)&ab_im1)[0],((int32x2_t*)&ab_im1)[1])), vpadd_s32(((int32x2_t*)&ab_im1)[0],((int32x2_t*)&ab_im1)[1])),
shift); shift);
xtmp = vzipq_s32(re32,im32); xtmp = vzipq_s32(re32,im32);
y_128[0] = vcombine_s16(vmovn_s32(xtmp.val[0]),vmovn_s32(xtmp.val[1])); y_128[0] = vcombine_s16(vmovn_s32(xtmp.val[0]),vmovn_s32(xtmp.val[1]));
#endif #endif
xd+=4; xd+=4;
y_128+=1; y_128+=1;
}
} }
_mm_empty();
_m_empty();
return;
} }
#endif
/*
int mult_vector32_scalar(int16_t *x1,
int x2,
int16_t *y,
uint32_t N)
{
// Multiply elementwise two real vectors of N elements
// x1 - input 1 in the format |Re(0) xxx Re(1) xxx|,......,|Re(N-2) xxx Re(N-1) xxx|
// We assume x1 with a dinamic of 31 bit maximum
//
// x1 - input 2
//
// y - output in the format |Re0 (64bit) |,......,|Re(N-1) (64bit)|
//
// N - the size f the vectors (this function does N cpx mpy. WARNING: N>=4;
//
uint32_t i; // loop counter
__m128i *x1_128;
__m128i x2_128;
__m128i *y_128;
x1_128 = (__m128i *)&x1[0];
x2_128 = _mm_setr_epi32(x2,0,x2,0);
y_128 = (__m128i *)&y[0];
// we compute 4 cpx multiply for each loop
for(i=0; i<(N>>3); i++) {
y_128[0] = _mm_mul_epu32(x1_128[0],x2_128);
y_128[1] = _mm_mul_epu32(x1_128[1],x2_128);
y_128[2] = _mm_mul_epu32(x1_128[2],x2_128);
y_128[3] = _mm_mul_epu32(x1_128[3],x2_128);
x1_128+=4;
y_128 +=4;
}
_mm_empty();
_m_empty();
return(0);
}
*/
int complex_conjugate(int16_t *x1,
int16_t *y,
uint32_t N)
{
uint32_t i; // loop counter
simd_q15_t *x1_128;
simd_q15_t *y_128;
int16_t x2[8] __attribute__((aligned(16))) = {1,-1,1,-1,1,-1,1,-1};
simd_q15_t *x2_128 = (simd_q15_t*)&x2[0];
x1_128 = (simd_q15_t *)&x1[0];
y_128 = (simd_q15_t *)&y[0];
// we compute 4 cpx multiply for each loop
for(i=0; i<(N>>3); i++) {
y_128[0] = mullo_int16(x1_128[0],*x2_128);
y_128[1] = mullo_int16(x1_128[1],*x2_128);
y_128[2] = mullo_int16(x1_128[2],*x2_128);
y_128[3] = mullo_int16(x1_128[3],*x2_128);
x1_128+=4;
y_128 +=4;
}
_mm_empty();
_m_empty();
return(0);
}
#ifdef MAIN #ifdef MAIN
#define L 8 #define L 8
openair1/PHY/TOOLS/memory_routines.c
-67
View File
@@ -1,67 +0,0 @@
/*
* 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
*/
#include "PHY/sse_intrin.h"
void Zero_Buffer(void *buf,unsigned int length)
{
// zeroes the mmx_t buffer 'buf' starting from buf[0] to buf[length-1] in bytes
int i;
register __m64 mm0;
__m64 *mbuf = (__m64 *)buf;
// length>>=3; // put length in quadwords
mm0 = _m_pxor(mm0,mm0); // clear the register
for(i=0; i<length>>3; i++) // for each i
mbuf[i] = mm0; // put 0 in buf[i]
_mm_empty();
}
void mmxcopy(void *dest,void *src,int size)
{
// copy size bytes from src to dest
register int i;
register __m64 mm0;
__m64 *mmsrc = (__m64 *)src, *mmdest= (__m64 *)dest;
for (i=0; i<size>>3; i++) {
mm0 = mmsrc[i];
mmdest[i] = mm0;
}
_mm_empty();
}
void Zero_Buffer_nommx(void *buf,unsigned int length)
{
int i;
for (i=0; i<length>>2; i++)
((int *)buf)[i] = 0;
}
openair1/PHY/TOOLS/oai_dfts.c
-1
View File
@@ -2575,7 +2575,6 @@ const static int16_t tw64c[96] __attribute__((aligned(32))) = {
#define simd_q15_t __m128i #define simd_q15_t __m128i
#define simdshort_q15_t __m64 #define simdshort_q15_t __m64
#define shiftright_int16(a,shift) _mm_srai_epi16(a,shift) #define shiftright_int16(a,shift) _mm_srai_epi16(a,shift)
#define set1_int16(a) _mm_set1_epi16(a);
#define mulhi_int16(a,b) _mm_mulhrs_epi16 (a,b) #define mulhi_int16(a,b) _mm_mulhrs_epi16 (a,b)
#ifdef __AVX2__ #ifdef __AVX2__
#define simd256_q15_t __m256i #define simd256_q15_t __m256i
openair1/PHY/TOOLS/tools_defs.h
+206 -87
View File
@@ -29,46 +29,199 @@
*/ */
#ifdef __cplusplus
extern "C" {
#endif
#include <stdio.h> #include <stdio.h>
#include <stdint.h> #include <stdint.h>
#include <assert.h> #include <assert.h>
#include "PHY/sse_intrin.h" #include "PHY/sse_intrin.h"
#include "common/utils/assertions.h" #include "common/utils/assertions.h"
#if defined(__x86_64__) || defined(__i386__)
#define simd_q15_t __m128i
#define simdshort_q15_t __m64
#define shiftright_int16(a,shift) _mm_srai_epi16(a,shift)
#define set1_int16(a) _mm_set1_epi16(a)
#define mulhi_int16(a,b) _mm_mulhrs_epi16 (a,b)
#define mulhi_s1_int16(a,b) _mm_slli_epi16(_mm_mulhi_epi16(a,b),2)
#define adds_int16(a,b) _mm_adds_epi16(a,b)
#define mullo_int16(a,b) _mm_mullo_epi16(a,b)
#elif defined(__arm__)
#define simd_q15_t int16x8_t
#define simdshort_q15_t int16x4_t
#define shiftright_int16(a,shift) vshrq_n_s16(a,shift)
#define set1_int16(a) vdupq_n_s16(a)
#define mulhi_int16(a,b) vqdmulhq_s16(a,b)
#define mulhi_s1_int16(a,b) vshlq_n_s16(vqdmulhq_s16(a,b),1)
#define adds_int16(a,b) vqaddq_s16(a,b)
#define mullo_int16(a,b) vmulq_s16(a,b)
#define _mm_empty()
#define _m_empty()
#endif
#ifdef __cplusplus
extern "C" {
#endif
#define CEILIDIV(a,b) ((a+b-1)/b) #define CEILIDIV(a,b) ((a+b-1)/b)
#define ROUNDIDIV(a,b) (((a<<1)+b)/(b<<1)) #define ROUNDIDIV(a,b) (((a<<1)+b)/(b<<1))
typedef struct complexd { typedef struct complexd {
double r; double r;
double i; double i;
} cd_t; } cd_t;
typedef struct complexf { typedef struct complexf {
float r; float r;
float i; float i;
} cf_t; } cf_t;
typedef struct complex16 { typedef struct complex16 {
int16_t r; int16_t r;
int16_t i; int16_t i;
} c16_t; } c16_t;
typedef struct complex32 { typedef struct complex32 {
int32_t r; int32_t r;
int32_t i; int32_t i;
} c32_t; } c32_t;
typedef struct complex64 {
int64_t r;
int64_t i;
} c64_t;
#define squaredMod(a) ((a).r*(a).r + (a).i*(a).i)
#define csum(res, i1, i2) (res).r = (i1).r + (i2).r ; (res).i = (i1).i + (i2).i
__attribute__((always_inline)) inline c16_t c16mulShift(const c16_t a, const c16_t b, const int Shift) {
return (c16_t) {
.r = (int16_t)((a.r * b.r - a.i * b.i) >> Shift),
.i = (int16_t)((a.r * b.i + a.i * b.r) >> Shift)
};
}
typedef struct complex64 { __attribute__((always_inline)) inline c16_t c16divShift(const c16_t a, const c16_t b, const int Shift) {
int64_t r; return (c16_t) {
int64_t i; .r = (int16_t)((a.r * b.r + a.i * b.i) >> Shift),
} c64_t; .i = (int16_t)((a.r * b.i - a.i * b.r) >> Shift)
};
}
__attribute__((always_inline)) inline c16_t c16maddShift(const c16_t a, const c16_t b, c16_t c, const int Shift) {
return (c16_t) {
.r = (int16_t)(((a.r * b.r - a.i * b.i ) >> Shift) + c.r),
.i = (int16_t)(((a.r * b.i + a.i * b.r ) >> Shift) + c.i)
};
}
#define squaredMod(a) ((a).r*(a).r+(a).i*(a).i) __attribute__((always_inline)) inline c32_t c32x16mulShift(const c16_t a, const c16_t b, const int Shift) {
#define csum(res, i1, i2) (res).r=(i1).r+(i2).r ; (res).i=(i1).i+(i2).i return (c32_t) {
.r = (a.r * b.r - a.i * b.i) >> Shift,
.i = (a.r * b.i + a.i * b.r) >> Shift
};
}
__attribute__((always_inline)) inline c32_t c32x16maddShift(const c16_t a, const c16_t b, const c32_t c, const int Shift) {
return (c32_t) {
.r = ((a.r * b.r - a.i * b.i) >> Shift) + c.r,
.i = ((a.r * b.i + a.i * b.r) >> Shift) + c.i
};
}
__attribute__((always_inline)) inline c16_t c16x32div(const c32_t a, const int div) {
return (c16_t) {
.r = (int16_t)(a.r / div),
.i = (int16_t)(a.i / div)
};
}
// On N complex numbers
// y.r += (x * alpha.r) >> 14
// y.i += (x * alpha.i) >> 14
// See regular C implementation at the end
__attribute__((always_inline)) inline void c16multaddVectRealComplex(const int16_t *x,
const c16_t *alpha,
c16_t *y,
const int N) {
#ifdef __AVX2__
const int8_t makePairs[32] __attribute__((aligned(32)))={
0,1,0+16,1+16,
2,3,2+16,3+16,
4,5,4+16,5+16,
6,7,6+16,7+16,
8,9,8+16,9+16,
10,11,10+16,11+16,
12,13,12+16,13+16,
14,15,14+16,15+16};
__m256i alpha256= _mm256_set1_epi32(*(int32_t *)alpha);
__m128i *x128=(__m128i *)x;
__m128i *y128=(__m128i *)y;
AssertFatal(N%8==0,"Not implemented\n");
for (int i=0; i<N/8; i++) {
const __m256i xduplicate=_mm256_broadcastsi128_si256(*x128);
const __m256i x_duplicate_ordered=_mm256_shuffle_epi8(xduplicate,*(__m256i*)makePairs);
const __m256i x_mul_alpha_shift15 =_mm256_mulhrs_epi16(alpha256, x_duplicate_ordered);
// Existing multiplication normalization is weird, constant table in alpha need to be doubled
const __m256i x_mul_alpha_x2= _mm256_adds_epi16(x_mul_alpha_shift15,x_mul_alpha_shift15);
*y128= _mm_adds_epi16(_mm256_extracti128_si256(x_mul_alpha_x2,0),*y128);
y128++;
*y128= _mm_adds_epi16(_mm256_extracti128_si256(x_mul_alpha_x2,1),*y128);
y128++;
x128++;
}
#elif defined(__x86_64__) || defined(__i386__) || defined(__arm__)
uint32_t i;
// do 8 multiplications at a time
simd_q15_t alpha_r_128,alpha_i_128,yr,yi,*x_128=(simd_q15_t*)x,*y_128=(simd_q15_t*)y;
int j;
// printf("alpha = %d,%d\n",alpha[0],alpha[1]);
alpha_r_128 = set1_int16(alpha->r);
alpha_i_128 = set1_int16(alpha->i);
j=0;
for (i=0; i<N>>3; i++) {
yr = mulhi_s1_int16(alpha_r_128,x_128[i]);
yi = mulhi_s1_int16(alpha_i_128,x_128[i]);
#if defined(__x86_64__) || defined(__i386__)
y_128[j] = _mm_adds_epi16(y_128[j],_mm_unpacklo_epi16(yr,yi));
j++;
y_128[j] = _mm_adds_epi16(y_128[j],_mm_unpackhi_epi16(yr,yi));
j++;
#elif defined(__arm__)
int16x8x2_t yint;
yint = vzipq_s16(yr,yi);
y_128[j] = adds_int16(y_128[j],yint.val[0]);
j++;
y_128[j] = adds_int16(y_128[j],yint.val[1]);
j++;
#endif
}
#else
for (int i=0; i<N; i++) {
int tmpr=y[i].r+((x[i]*alpha->r)>>14);
if (tmpr>INT16_MAX)
tmpr=INT16_MAX;
if (tmpr<INT16_MIN)
tmpr=INT16_MIN;
int tmpi=y[i].i+((x[i]*alpha->i)>>14);
if (tmpi>INT16_MAX)
tmpi=INT16_MAX;
if (tmpi<INT16_MIN)
tmpi=INT16_MIN;
y[i].r=(int16_t)tmpr;
y[i].i=(int16_t)tmpi;
}
#endif
}
//cmult_sv.h //cmult_sv.h
/*!\fn void multadd_real_vector_complex_scalar(int16_t *x,int16_t *alpha,int16_t *y,uint32_t N) /*!\fn void multadd_real_vector_complex_scalar(int16_t *x,int16_t *alpha,int16_t *y,uint32_t N)
@@ -85,10 +238,29 @@ void multadd_real_vector_complex_scalar(int16_t *x,
int16_t *y, int16_t *y,
uint32_t N); uint32_t N);
void multadd_real_four_symbols_vector_complex_scalar(int16_t *x, __attribute__((always_inline)) inline void multadd_real_four_symbols_vector_complex_scalar(int16_t *x,
int16_t *alpha, c16_t *alpha,
int16_t *y); c16_t *y)
{
// do 8 multiplications at a time
simd_q15_t alpha_r_128,alpha_i_128,yr,yi,*x_128=(simd_q15_t*)x;
simd_q15_t y_128;
y_128 = _mm_loadu_si128((simd_q15_t*)y);
alpha_r_128 = set1_int16(alpha->r);
alpha_i_128 = set1_int16(alpha->i);
yr = mulhi_s1_int16(alpha_r_128,x_128[0]);
yi = mulhi_s1_int16(alpha_i_128,x_128[0]);
y_128 = _mm_adds_epi16(y_128,_mm_unpacklo_epi16(yr,yi));
y_128 = _mm_adds_epi16(y_128,_mm_unpackhi_epi16(yr,yi));
_mm_storeu_si128((simd_q15_t*)y, y_128);
}
/*!\fn void multadd_complex_vector_real_scalar(int16_t *x,int16_t alpha,int16_t *y,uint8_t zero_flag,uint32_t N) /*!\fn void multadd_complex_vector_real_scalar(int16_t *x,int16_t alpha,int16_t *y,uint8_t zero_flag,uint32_t N)
This function performs componentwise multiplication and accumulation of a real scalar and a complex vector. This function performs componentwise multiplication and accumulation of a real scalar and a complex vector.
@param x Vector input (Q1.15) in the format |Re0 Im0|Re1 Im 1| ... @param x Vector input (Q1.15) in the format |Re0 Im0|Re1 Im 1| ...
@@ -114,7 +286,7 @@ void multadd_complex_vector_real_scalar(int16_t *x,
//cmult_vv.c //cmult_vv.c
/*! /*!
Multiply elementwise the complex conjugate of x1 with x2. Multiply elementwise the complex conjugate of x1 with x2.
@param x1 - input 1 in the format |Re0 Im0 Re1 Im1|,......,|Re(N-2) Im(N-2) Re(N-1) Im(N-1)| @param x1 - input 1 in the format |Re0 Im0 Re1 Im1|,......,|Re(N-2) Im(N-2) Re(N-1) Im(N-1)|
We assume x1 with a dinamic of 15 bit maximum We assume x1 with a dinamic of 15 bit maximum
@param x2 - input 2 in the format |Re0 Im0 Re1 Im1|,......,|Re(N-2) Im(N-2) Re(N-1) Im(N-1)| @param x2 - input 2 in the format |Re0 Im0 Re1 Im1|,......,|Re(N-2) Im(N-2) Re(N-1) Im(N-1)|
@@ -364,8 +536,8 @@ dft_size_idx_t get_dft(int ofdm_symbol_size)
printf("function get_dft : unsupported ofdm symbol size \n"); printf("function get_dft : unsupported ofdm symbol size \n");
assert(0); assert(0);
break; break;
} }
return DFT_SIZE_IDXTABLESIZE; // never reached and will trigger assertion in idft function; return DFT_SIZE_IDXTABLESIZE; // never reached and will trigger assertion in idft function;
} }
typedef enum idft_size_idx { typedef enum idft_size_idx {
@@ -448,8 +620,8 @@ idft_size_idx_t get_idft(int ofdm_symbol_size)
printf("function get_idft : unsupported ofdm symbol size \n"); printf("function get_idft : unsupported ofdm symbol size \n");
assert(0); assert(0);
break; break;
} }
return IDFT_SIZE_IDXTABLESIZE; // never reached and will trigger assertion in idft function return IDFT_SIZE_IDXTABLESIZE; // never reached and will trigger assertion in idft function
} }
@@ -491,58 +663,6 @@ int32_t sub_cpx_vector16(int16_t *x,
int16_t *z, int16_t *z,
uint32_t N); uint32_t N);
int32_t add_cpx_vector32(int16_t *x,
int16_t *y,
int16_t *z,
uint32_t N);
int32_t add_real_vector64(int16_t *x,
int16_t *y,
int16_t *z,
uint32_t N);
int32_t sub_real_vector64(int16_t *x,
int16_t *y,
int16_t *z,
uint32_t N);
int32_t add_real_vector64_scalar(int16_t *x,
long long int a,
int16_t *y,
uint32_t N);
/*!\fn int32_t add_vector16(int16_t *x,int16_t *y,int16_t *z,uint32_t N)
This function performs componentwise addition of two vectors with Q1.15 components.
@param x Vector input (Q1.15)
@param y Scalar input (Q1.15)
@param z Scalar output (Q1.15)
@param N Length of x WARNING: N must be a multiple of 32
The function implemented is : \f$\mathbf{z} = \mathbf{x} + \mathbf{y}\f$
*/
int32_t add_vector16(int16_t *x,
int16_t *y,
int16_t *z,
uint32_t N);
int32_t add_vector16_64(int16_t *x,
int16_t *y,
int16_t *z,
uint32_t N);
int32_t complex_conjugate(int16_t *x1,
int16_t *y,
uint32_t N);
void bit8_txmux(int32_t length,int32_t offset);
void bit8_rxdemux(int32_t length,int32_t offset);
void Zero_Buffer(void *,uint32_t);
void Zero_Buffer_nommx(void *buf,uint32_t length);
void mmxcopy(void *dest,void *src,int size);
/*!\fn int32_t signal_energy(int *,uint32_t); /*!\fn int32_t signal_energy(int *,uint32_t);
\brief Computes the signal energy per subcarrier \brief Computes the signal energy per subcarrier
*/ */
@@ -631,7 +751,6 @@ int64_t dot_product64(int16_t *x,
double interp(double x, double *xs, double *ys, int count); double interp(double x, double *xs, double *ys, int count);
int write_output(const char *fname,const char *vname,void *data,int length,int dec,char format);
#ifdef __cplusplus #ifdef __cplusplus
} }
openair1/PHY/defs_eNB.h
-3
View File
@@ -181,9 +181,6 @@ typedef struct {
typedef struct { typedef struct {
/// \brief ?.
/// first index: ? [0..1023] (hard coded)
int16_t *prachF;
/// \brief ?. /// \brief ?.
/// first index: ce_level [0..3] /// first index: ce_level [0..3]
/// second index: rx antenna [0..63] (hard coded) \note Hard coded array size indexed by \c nb_antennas_rx. /// second index: rx antenna [0..63] (hard coded) \note Hard coded array size indexed by \c nb_antennas_rx.
openair1/SCHED/prach_procedures.c
+1 -1
View File
@@ -49,7 +49,7 @@ extern int oai_nfapi_rach_ind(nfapi_rach_indication_t *rach_ind);
void prach_procedures(PHY_VARS_eNB *eNB, void prach_procedures(PHY_VARS_eNB *eNB,
int br_flag ) { int br_flag ) {
uint16_t max_preamble[4],max_preamble_energy[4],max_preamble_delay[4],avg_preamble_energy[4]; uint16_t max_preamble[4]={0},max_preamble_energy[4]={0},max_preamble_delay[4]={0},avg_preamble_energy[4]={0};
uint16_t i; uint16_t i;
int frame,subframe; int frame,subframe;
openair3/ocp-gtpu/gtp_itf.cpp
+1 -1
View File
@@ -145,7 +145,7 @@ class gtpEndPoints {
~gtpEndPoints() { ~gtpEndPoints() {
// automatically close all sockets on quit // automatically close all sockets on quit
for (const auto p : instances) for (const auto &p : instances)
close(p.first); close(p.first);
} }
}; };