Commit 0f207da2 authored by Raymond Knopp's avatar Raymond Knopp

initial changes for multi-threaded TX for IF5

parent 6bb14205
......@@ -275,15 +275,15 @@ int connect_rau(RU_t *ru) {
// southbound IF5 fronthaul for 16-bit OAI format
void fh_if5_south_out(RU_t *ru, int frame, int slot, uint64_t timestamp) {
if (ru == RC.ru[0]) VCD_SIGNAL_DUMPER_DUMP_VARIABLE_BY_NAME( VCD_SIGNAL_DUMPER_VARIABLES_TRX_TST, ru->proc.timestamp_tx&0xffffffff );
void *buffs[ru->nb_tx];
int offset = ru->nr_frame_parms->get_samples_slot_timestamp(slot,ru->nr_frame_parms,0);
for (int aid=0;aid<ru->nb_tx;aid++) buffs[aid]=&ru->common.txdata[aid][offset];
start_meas(&ru->tx_fhaul);
ru->ifdevice.trx_write_func2(&ru->ifdevice,
timestamp,
buffs,
ru->nr_frame_parms->get_samples_per_slot(slot,ru->nr_frame_parms),
0);
for (int aid=0;aid<ru->nb_tx;aid++)
ru->ifdevice.trx_write_func2(&ru->ifdevice,
timestamp,
(void*)&ru->common.txdata[aid][offset],
aid,
ru->nr_frame_parms->get_samples_per_slot(slot,ru->nr_frame_parms),
0);
stop_meas(&ru->tx_fhaul);
}
......@@ -488,7 +488,7 @@ void fh_if5_north_asynch_in(RU_t *ru,int *frame,int *slot) {
RU_proc_t *proc = &ru->proc;
int tti_tx,frame_tx;
openair0_timestamp timestamp_tx;
recv_IF5(ru, &timestamp_tx, *slot, IF5_RRH_GW_DL,1);
AssertFatal(1==0,"Shouldn't get here\n");
// printf("Received subframe %d (TS %llu) from RCC\n",tti_tx,timestamp_tx);
frame_tx = (timestamp_tx / (fp->samples_per_subframe*10))&1023;
uint32_t idx_sf = timestamp_tx / fp->samples_per_subframe;
......@@ -565,11 +565,9 @@ void fh_if4p5_north_asynch_in(RU_t *ru,int *frame,int *slot) {
}
void fh_if5_north_out(RU_t *ru) {
RU_proc_t *proc=&ru->proc;
uint8_t seqno=0;
/// **** send_IF5 of rxdata to BBU **** ///
VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME( VCD_SIGNAL_DUMPER_FUNCTIONS_SEND_IF5, 1 );
send_IF5(ru, proc->timestamp_rx, proc->tti_rx, &seqno, IF5_RRH_GW_UL);
AssertFatal(1==0,"Shouldn't get here\n");
VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME( VCD_SIGNAL_DUMPER_FUNCTIONS_SEND_IF5, 0 );
}
......@@ -1181,9 +1179,15 @@ void *ru_stats_thread(void *param) {
print_meas(&ru->precoding_stats,"feptx_prec",NULL,NULL);
print_meas(&ru->txdataF_copy_stats,"txdataF_copy",NULL,NULL);
print_meas(&ru->ofdm_mod_stats,"feptx_ofdm",NULL,NULL);
print_meas(&ru->ofdm_total_stats,"feptx_total",NULL,NULL);
if (ru->txfh_in_fep) {
print_meas(&ru->ofdm_mod_stats,"feptx_ofdm (with txfh)",NULL,NULL);
print_meas(&ru->ofdm_total_stats,"feptx_total (with txfh)",NULL,NULL);
}
else {
print_meas(&ru->ofdm_mod_stats,"feptx_ofdm",NULL,NULL);
print_meas(&ru->ofdm_total_stats,"feptx_total",NULL,NULL);
}
}
print_meas(&ru->rx_fhaul,"rx_fhaul",NULL,NULL);
print_meas(&ru->tx_fhaul,"tx_fhaul",NULL,NULL);
......@@ -1851,11 +1855,12 @@ void set_function_spec_param(RU_t *ru) {
case REMOTE_IF5: // the remote unit is IF5 RRU
ru->do_prach = 0;
// ru->txfh_in_fep = 1;
ru->feprx = (get_thread_worker_conf() == WORKER_ENABLE) ? nr_fep_full_2thread : nr_fep_full; // this is frequency-shift + DFTs
ru->feptx_prec = (get_thread_worker_conf() == WORKER_ENABLE) ? NULL : nr_feptx_prec; // need to do transmit Precoding + IDFTs
ru->feptx_ofdm = (get_thread_worker_conf() == WORKER_ENABLE) ? nr_feptx_ofdm_2thread : nr_feptx_ofdm; // need to do transmit Precoding + IDFTs
ru->fh_south_in = fh_if5_south_in; // synchronous IF5 reception
ru->fh_south_out = fh_if5_south_out; // synchronous IF5 transmission
ru->fh_south_out = (ru->txfh_in_fep) ? NULL : fh_if5_south_out; // synchronous IF5 transmission
ru->fh_south_asynch_in = NULL; // no asynchronous UL
ru->start_rf = ru->eth_params.transp_preference == ETH_UDP_IF5_ECPRI_MODE ? start_streaming : NULL;
ru->stop_rf = NULL;
......
......@@ -507,6 +507,8 @@ typedef struct RU_t_s {
int sf_ahead;
/// TX processing advance in slots (for NR)
int sl_ahead;
/// flag to indicated TX FH is embedded in TX FEP
int txfh_in_fep;
/// FAPI confiuration
nfapi_nr_config_request_scf_t config;
/// Frame parameters
......
......@@ -56,7 +56,7 @@ void nr_feptx0(RU_t *ru,int tti_tx,int first_symbol, int num_symbols, int aa) {
NR_DL_FRAME_PARMS *fp = ru->nr_frame_parms;
unsigned int slot_offset,slot_offsetF;
unsigned int slot_offset,slot_offsetF,initial_slot_offset,num_samples;;
int slot = tti_tx;
......@@ -71,17 +71,19 @@ void nr_feptx0(RU_t *ru,int tti_tx,int first_symbol, int num_symbols, int aa) {
slot_offset += (idx_sym%(0x7<<fp->numerology_index)) ? fp->nb_prefix_samples : fp->nb_prefix_samples0;
slot_offset += fp->ofdm_symbol_size*first_symbol;
initial_slot_offset = slot_offset;
LOG_D(PHY,"SFN/SF:RU:TX:%d/%d Generating slot %d (first_symbol %d num_symbols %d)\n",ru->proc.frame_tx, ru->proc.tti_tx,slot,first_symbol,num_symbols);
if (fp->Ncp == 1)
if (fp->Ncp == 1) {
PHY_ofdm_mod(&ru->common.txdataF_BF[aa][slot_offsetF],
(int*)&ru->common.txdata[aa][slot_offset],
fp->ofdm_symbol_size,
num_symbols,
fp->nb_prefix_samples,
CYCLIC_PREFIX);
else {
num_samples=num_symbols*(fp->nb_prefix_samples+fp->ofdm_symbol_size);
} else {
if (fp->numerology_index != 0) {
if (!(slot%(fp->slots_per_subframe/2))&&(first_symbol==0)) { // case where first symbol in slot has longer prefix
......@@ -98,6 +100,8 @@ void nr_feptx0(RU_t *ru,int tti_tx,int first_symbol, int num_symbols, int aa) {
num_symbols-1,
fp->nb_prefix_samples,
CYCLIC_PREFIX);
num_samples=(num_symbols-1)*(fp->nb_prefix_samples+fp->ofdm_symbol_size)+
fp->nb_prefix_samples0+fp->ofdm_symbol_size;
}
else { // all symbols in slot have shorter prefix
PHY_ofdm_mod(&ru->common.txdataF_BF[aa][slot_offsetF],
......@@ -106,6 +110,7 @@ void nr_feptx0(RU_t *ru,int tti_tx,int first_symbol, int num_symbols, int aa) {
num_symbols,
fp->nb_prefix_samples,
CYCLIC_PREFIX);
num_samples=num_symbols*(fp->nb_prefix_samples+fp->ofdm_symbol_size);
}
} // numerology_index!=0
else { //numerology_index == 0
......@@ -130,10 +135,20 @@ void nr_feptx0(RU_t *ru,int tti_tx,int first_symbol, int num_symbols, int aa) {
slot_offset += fp->nb_prefix_samples0+fp->ofdm_symbol_size;
slot_offsetF += fp->ofdm_symbol_size;
}
}
}
} // for(idx_symbol..
num_samples=slot_offset-initial_slot_offset;
} // numerology 0
}
if (ru->txfh_in_fep) {
ru->ifdevice.trx_write_func2(&ru->ifdevice,
ru->proc.timestamp_tx,
(void*)&ru->common.txdata[aa][initial_slot_offset],
aa,
num_samples,
0);
}
//VCD_SIGNAL_DUMPER_DUMP_FUNCTION_BY_NAME(VCD_SIGNAL_DUMPER_FUNCTIONS_PHY_PROCEDURES_RU_FEPTX_OFDM+(first_symbol!=0?1:0), 0);
}
......
......@@ -608,25 +608,10 @@ bool allocate_dl_retransmission(module_id_t module_id,
/* the retransmission will use a different time domain allocation, check
* that we have enough resources */
NR_pdsch_semi_static_t temp_ps = *ps;
<<<<<<< HEAD
const uint16_t slbitmap = SL_to_bitmap(ps->startSymbolIndex, ps->nrOfSymbols);
nr_set_pdsch_semi_static(sib1,scc, cg, sched_ctrl->active_bwp, bwpd,tda, ps->nrOfLayers, sched_ctrl, &temp_ps);
while (rbStart < bwpSize &&
!(rballoc_mask[rbStart]&SL_to_bitmap(temp_ps.startSymbolIndex, temp_ps.nrOfSymbols)))
=======
nr_set_pdsch_semi_static(sib1,
scc,
cg,
sched_ctrl->active_bwp,
bwpd,
tda,
ps->nrOfLayers,
sched_ctrl,
&temp_ps);
const uint16_t slbitmap = SL_to_bitmap(temp_ps.startSymbolIndex, temp_ps.nrOfSymbols);
while (rbStart < bwpSize && (rballoc_mask[rbStart] & slbitmap) != slbitmap)
>>>>>>> origin/develop
rbStart++;
while (rbStart + rbSize < bwpSize && (rballoc_mask[rbStart + rbSize] & slbitmap) == slbitmap)
......
......@@ -423,7 +423,7 @@ struct openair0_device_t {
@param antenna_id index of the antenna if the device has multiple anteannas
@param flags flags must be set to TRUE if timestamp parameter needs to be applied
*/
int (*trx_write_func2)(openair0_device *device, openair0_timestamp timestamp, void **buff, int nsamps, int flags);
int (*trx_write_func2)(openair0_device *device, openair0_timestamp timestamp, void *buff, int aid, int nsamps, int flags);
/*! \brief Receive samples from hardware.
* Read \ref nsamps samples from each channel to buffers. buff[0] is the array for
......
......@@ -289,101 +289,98 @@ int trx_eth_write_udp_IF4p5(openair0_device *device, openair0_timestamp timestam
return (bytes_sent);
}
int trx_eth_write_udp(openair0_device *device, openair0_timestamp timestamp, void **buff, int nsamps, int flags) {
int trx_eth_write_udp(openair0_device *device, openair0_timestamp timestamp, void *buff, int aid, int nsamps, int flags) {
int bytes_sent=0;
eth_state_t *eth = (eth_state_t*)device->priv;
int sendto_flag =0;
fhstate_t *fhstate = &device->fhstate;
AssertFatal((nsamps&255) == 0,"transport size should be a multiple of 256 samples, have %d\n",nsamps);
//sendto_flag|=flags;
eth->tx_nsamps=nsamps;
int ntx = device->openair0_cfg->tx_num_channels;
void *buff2[ntx];
void *buff2;
#if defined(__x86_64) || defined(__i386__)
#ifdef __AVX2__
int nsamps2 = 256>>3;
__m256i buff_tx[ntx][nsamps2+1];
for (int aid=0;aid<ntx;aid++) buff2[aid]=(void*)&buff_tx[aid][1] - APP_HEADER_SIZE_BYTES;
__m256i buff_tx[nsamps2+1];
buff2=(void*)&buff_tx[1] - APP_HEADER_SIZE_BYTES;
#else
int nsamps2 = 256>>2;
__m128i buff_tx[ntx][nsamps2+2];
for (int aid=0;aid<ntx;aid++) buff2[aid]=(void*)&buff_tx[aid][2] - APP_HEADER_SIZE_BYTES;
__m128i buff_tx[nsamps2+2];
buff=(void*)&buff_tx[2] - APP_HEADER_SIZE_BYTES;
#endif
#elif defined(__arm__) || defined(__aarch64__)
int nsamps2 = 256>>2;
int16x8_t buff_tx[ntx][nsamps2+2];
for (int aid=0;aid<ntx;aid++) buff2[aid]=(void*)&buff_tx[aid][2] - APP_HEADER_SIZE_BYTES;
int16x8_t buff_tx[nsamps2+2];
buff2=(void*)&buff_tx[2] - APP_HEADER_SIZE_BYTES;
#else
#error Unsupported CPU architecture, ethernet device cannot be built
#endif
for (int aid=0;aid<ntx;aid++) {
/* construct application header */
// ECPRI Protocol revision + reserved bits (1 byte)
*(uint8_t *)buff2[aid] = ECPRIREV;
// ECPRI Message type (1 byte)
*(uint8_t *)(buff2[aid] + 1) = 64;
// ECPRI Payload Size (2 bytes)
*(uint8_t *)(buff2[aid] + 2) = 4;// 1024>>6;
*(uint8_t *)(buff2[aid] + 3) = 0;// 1024&0xff;
// ECPRI PC_ID (2 bytes)
*(uint16_t *)(buff2[aid] + 4) = aid;
/* construct application header */
// ECPRI Protocol revision + reserved bits (1 byte)
*(uint8_t *)buff2 = ECPRIREV;
// ECPRI Message type (1 byte)
*(uint8_t *)(buff2 + 1) = 64;
// ECPRI PC_ID (2 bytes)
*(uint16_t *)(buff2 + 4) = aid;
}
openair0_timestamp TS = timestamp + fhstate->TS0;
TS = 6*device->sampling_rate_ratio_d*(TS/device->sampling_rate_ratio_n);
TS -= device->txrx_offset;
int TSinc = 6*256*device->sampling_rate_ratio_d/device->sampling_rate_ratio_n;
int len=256;
for (int offset=0;offset<nsamps;offset+=256,TS+=TSinc) {
// OAI modified SEQ_ID (4 bytes)
for (int aid=0;aid<ntx;aid++) {
*(uint64_t *)(buff2[aid] + 6) = TS;
*(uint64_t *)(buff2 + 6) = TS;
if ((offset + 256) <= nsamps) len=1024;
else len = (nsamps-offset)<<2;
// ECPRI Payload Size (2 bytes)
*(uint8_t *)(buff2 + 2) = len>>8;
*(uint8_t *)(buff2 + 3) = len&0xff;
//LOG_I(PHY,"TS %llu aa %d : offset %d, len %d\n",(unsigned long long)TS,aid,offset,len);
// bring TX data into 12 MSBs
#if defined(__x86_64__) || defined(__i386__)
#ifdef __AVX2__
__m256i *buff256 = (__m256i *)&((uint32_t*)buff[aid])[offset];
for (int j=0; j<32; j+=8) {
buff_tx[aid][1+j] = _mm256_slli_epi16(buff256[j],4);
buff_tx[aid][2+j] = _mm256_slli_epi16(buff256[j+1],4);
buff_tx[aid][3+j] = _mm256_slli_epi16(buff256[j+2],4);
buff_tx[aid][4+j] = _mm256_slli_epi16(buff256[j+3],4);
buff_tx[aid][5+j] = _mm256_slli_epi16(buff256[j+4],4);
buff_tx[aid][6+j] = _mm256_slli_epi16(buff256[j+5],4);
buff_tx[aid][7+j] = _mm256_slli_epi16(buff256[j+6],4);
buff_tx[aid][8+j] = _mm256_slli_epi16(buff256[j+7],4);
}
__m256i *buff256 = (__m256i *)&((uint32_t*)buff)[offset];
for (int j=0; j<32; j+=8) {
buff_tx[1+j] = _mm256_slli_epi16(buff256[j],4);
buff_tx[2+j] = _mm256_slli_epi16(buff256[j+1],4);
buff_tx[3+j] = _mm256_slli_epi16(buff256[j+2],4);
buff_tx[4+j] = _mm256_slli_epi16(buff256[j+3],4);
buff_tx[5+j] = _mm256_slli_epi16(buff256[j+4],4);
buff_tx[6+j] = _mm256_slli_epi16(buff256[j+5],4);
buff_tx[7+j] = _mm256_slli_epi16(buff256[j+6],4);
buff_tx[8+j] = _mm256_slli_epi16(buff256[j+7],4);
}
#else
__m128i *buff128 = (__m128i *)&buff[aid][offset];
for (int j=0; j<64; j++) buff_tx[aid][2+j] = _mm_slli_epi16(buff128[j],4);
__m128i *buff128 = (__m128i *)&buff[offset];
for (int j=0; j<64; j++) buff_tx[2+j] = _mm_slli_epi16(buff128[j],4);
#endif
#elif defined(__arm__)
int16x8_t *buff128 = (__int16x8_t*)&buff[aid][offset];
for (int j=0; j<64; j++) buff_tx[aid][2+j] = vshlq_n_s16(((int16x8_t *)buff128)[j],4);
int16x8_t *buff128 = (__int16x8_t*)&buff[offset];
for (int j=0; j<64; j++) buff_tx[2+j] = vshlq_n_s16(((int16x8_t *)buff128)[j],4);
#endif
/* Send packet */
bytes_sent = sendto(eth->sockfdd[aid%eth->num_fd],
buff2[aid],
UDP_PACKET_SIZE_BYTES(256),
sendto_flag,
(struct sockaddr*)&eth->dest_addrd,
eth->addr_len);
bytes_sent = sendto(eth->sockfdd[aid%eth->num_fd],
buff2,
UDP_PACKET_SIZE_BYTES(len>>2),
sendto_flag,
(struct sockaddr*)&eth->dest_addrd,
eth->addr_len);
if ( bytes_sent == -1) {
if ( bytes_sent == -1) {
eth->num_tx_errors++;
perror("ETHERNET WRITE: ");
exit(-1);
} else {
} else {
eth->tx_actual_nsamps=bytes_sent>>2;
eth->tx_count++;
}
}
}
}
return (bytes_sent-APP_HEADER_SIZE_BYTES)>>2;
}
......@@ -426,12 +423,11 @@ void *udp_read_thread(void *arg) {
}
sleep(1);
}
return(0);
}
int trx_eth_read_udp(openair0_device *device, openair0_timestamp *timestamp, uint32_t **buff, int nsamps) {
eth_state_t *eth = (eth_state_t*)device->priv;
fhstate_t *fhstate = &device->fhstate;
openair0_timestamp prev_read_TS= fhstate->TS_read, min_TS;
// block until FH is ready
......
......@@ -241,7 +241,7 @@ int ethernet_tune(openair0_device *device, unsigned int option, int value);
* @ingroup _oai
*/
int eth_socket_init_udp(openair0_device *device);
int trx_eth_write_udp(openair0_device *device, openair0_timestamp timestamp, void **buff, int nsamps, int flags);
int trx_eth_write_udp(openair0_device *device, openair0_timestamp timestamp, void *buf, int aid, int nsamps, int flags);
int trx_eth_read_udp(openair0_device *device, openair0_timestamp *timestamp, uint32_t **buff, int nsamps);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment