From 9ddb8c0b9ffbf011ef89ac67e5051445cc922e6c Mon Sep 17 00:00:00 2001 From: Laurent THOMAS Date: Thu, 14 Aug 2025 12:16:20 +0200 Subject: [PATCH] Remove prach_ifft from premanent allocation Allocate on the stack when necessary. --- openair1/PHY/INIT/nr_init.c | 7 +- openair1/PHY/NR_TRANSPORT/nr_prach.c | 220 +++++++++++---------------- openair1/PHY/defs_gNB.h | 2 - 3 files changed, 87 insertions(+), 142 deletions(-) diff --git a/openair1/PHY/INIT/nr_init.c b/openair1/PHY/INIT/nr_init.c index 778903be91..23334c8626 100644 --- a/openair1/PHY/INIT/nr_init.c +++ b/openair1/PHY/INIT/nr_init.c @@ -190,8 +190,7 @@ void phy_init_nr_gNB(PHY_VARS_gNB *gNB) common_vars->debugBuff_sample_offset = 0; // PRACH - prach_vars->rxsigF = (int16_t **)malloc16_clear(Prx*sizeof(int16_t*)); - prach_vars->prach_ifft = (int32_t *)malloc16_clear(1024*2*sizeof(int32_t)); + prach_vars->rxsigF = (int16_t **)malloc16_clear(Prx * sizeof(int16_t *)); init_prach_list(gNB); @@ -266,10 +265,6 @@ void phy_free_nr_gNB(PHY_VARS_gNB *gNB) free_and_zero(common_vars->debugBuff); - NR_gNB_PRACH* prach_vars = &gNB->prach_vars; - free_and_zero(prach_vars->rxsigF); - free_and_zero(prach_vars->prach_ifft); - for (int ULSCH_id = 0; ULSCH_id < gNB->max_nb_pusch; ULSCH_id++) { NR_gNB_PUSCH *pusch_vars = &gNB->pusch_vars[ULSCH_id]; for (int i = 0; i < n_buf; i++) { diff --git a/openair1/PHY/NR_TRANSPORT/nr_prach.c b/openair1/PHY/NR_TRANSPORT/nr_prach.c index 76d9b1185f..f278d4c845 100644 --- a/openair1/PHY/NR_TRANSPORT/nr_prach.c +++ b/openair1/PHY/NR_TRANSPORT/nr_prach.c @@ -452,81 +452,67 @@ void rx_nr_prach(PHY_VARS_gNB *gNB, { AssertFatal(gNB != NULL, "Can only be called from gNB\n"); nfapi_nr_prach_config_t *cfg = &gNB->gNB_config.prach_config; - NR_DL_FRAME_PARMS *fp; + c16_t **rxsigF = (c16_t **)gNB->prach_vars.rxsigF; + uint16_t preamble_index0 = 0; + uint16_t numshift = 0; + int first_nonzero_root_idx = 0; + bool new_dft = false; + int log2_ifft_size = 10; - uint16_t rootSequenceIndex; - int numrootSequenceIndex; - uint8_t restricted_set; - uint8_t n_ra_prb=0xFF; - int nb_rx; + NR_DL_FRAME_PARMS *fp = &gNB->frame_parms; - int16_t **rxsigF = gNB->prach_vars.rxsigF; + const int nb_rx = gNB->gNB_config.carrier_config.num_rx_ant.value; + const int rootSequenceIndex = cfg->num_prach_fd_occasions_list[prach_pdu->num_ra].prach_root_sequence_index.value; + const int numrootSequenceIndex = cfg->num_prach_fd_occasions_list[prach_pdu->num_ra].num_root_sequences.value; + const int NCS = prach_pdu->num_cs; // cfg->num_prach_fd_occasions_list[0].prach_zero_corr_conf.value; + const int prach_sequence_length = cfg->prach_sequence_length.value; + const int msg1_frequencystart = cfg->num_prach_fd_occasions_list[prach_pdu->num_ra].k1.value; + const int restricted_set = cfg->restricted_set_config.value; + const int prach_fmt = prach_pdu->prach_format; + const int N_ZC = prach_sequence_length == 0 ? 839 : 139; - uint8_t preamble_index; - uint16_t NCS=99,NCS2; - uint16_t preamble_offset=0,preamble_offset_old; - int16_t preamble_shift=0; - uint32_t preamble_shift2; - uint16_t preamble_index0=0,n_shift_ra=0,n_shift_ra_bar; - uint16_t d_start=0; - uint16_t numshift=0; - const uint16_t *prach_root_sequence_map; - uint8_t not_found; - uint16_t u; - int16_t *Xu=0; - uint16_t first_nonzero_root_idx=0; - uint8_t new_dft=0; - int32_t lev; - int16_t levdB; - int log2_ifft_size=10; - int16_t prach_ifft_tmp[2048*2] __attribute__((aligned(32))); - int32_t *prach_ifft=(int32_t*)NULL; - - fp = &gNB->frame_parms; + LOG_D(PHY, + "L1 PRACH RX: rooSequenceIndex %d, numRootSeqeuences %d, NCS %d, N_ZC %d, format %d \n", + rootSequenceIndex, + numrootSequenceIndex, + NCS, + N_ZC, + prach_fmt); - nb_rx = gNB->gNB_config.carrier_config.num_rx_ant.value; - rootSequenceIndex = cfg->num_prach_fd_occasions_list[prach_pdu->num_ra].prach_root_sequence_index.value; - numrootSequenceIndex = cfg->num_prach_fd_occasions_list[prach_pdu->num_ra].num_root_sequences.value; - NCS = prach_pdu->num_cs;//cfg->num_prach_fd_occasions_list[0].prach_zero_corr_conf.value; - int prach_sequence_length = cfg->prach_sequence_length.value; - int msg1_frequencystart = cfg->num_prach_fd_occasions_list[prach_pdu->num_ra].k1.value; - // int num_unused_root_sequences = cfg->num_prach_fd_occasions_list[0].num_unused_root_sequences.value; - // cfg->num_prach_fd_occasions_list[0].unused_root_sequences_list - - restricted_set = cfg->restricted_set_config.value; - - uint8_t prach_fmt = prach_pdu->prach_format; - uint16_t N_ZC = (prach_sequence_length==0)?839:139; - - LOG_D(PHY,"L1 PRACH RX: rooSequenceIndex %d, numRootSeqeuences %d, NCS %d, N_ZC %d, format %d \n",rootSequenceIndex,numrootSequenceIndex,NCS,N_ZC,prach_fmt); - - prach_ifft = gNB->prach_vars.prach_ifft; if (LOG_DEBUGFLAG(DEBUG_PRACH)) { - if ((frame&1023) < 20) LOG_D(PHY,"PRACH (gNB) : running rx_prach for slot %d, msg1_frequencystart %d, rootSequenceIndex %d\n", slot, msg1_frequencystart, rootSequenceIndex); + if ((frame & 1023) < 20) + LOG_D(PHY, + "PRACH (gNB) : running rx_prach for slot %d, msg1_frequencystart %d, rootSequenceIndex %d\n", + slot, + msg1_frequencystart, + rootSequenceIndex); } start_meas(&gNB->rx_prach); - prach_root_sequence_map = (cfg->prach_sequence_length.value==0) ? prach_root_sequence_map_0_3 : prach_root_sequence_map_abc; + const uint16_t *prach_root_sequence_map = + cfg->prach_sequence_length.value == 0 ? prach_root_sequence_map_0_3 : prach_root_sequence_map_abc; // PDP is oversampled, e.g. 1024 sample instead of 839 // Adapt the NCS (zero-correlation zones) with oversampling factor e.g. 1024/839 - NCS2 = (N_ZC==839) ? ((NCS<<10)/839) : ((NCS<<8)/139); + int NCS2 = N_ZC == 839 ? (NCS << 10) / 839 : (NCS << 8) / 139; - if (NCS2==0) NCS2 = N_ZC; + if (NCS2 == 0) + NCS2 = N_ZC; + int preamble_offset = 0, preamble_offset_old = 99; - preamble_offset_old = 99; - - *max_preamble_energy = 0; *max_preamble_delay = 0; *max_preamble = 0; - int16_t prachF[2 * 1024]; - for (preamble_index=0 ; preamble_index<64 ; preamble_index++) { + int16_t preamble_shift = 0; + const int dft_sz = N_ZC == 839 ? 1024 : 256; + int32_t prach_ifft[dft_sz] __attribute__((aligned(32))); + for (int preamble_index = 0; preamble_index < 64; preamble_index++) { if (LOG_DEBUGFLAG(DEBUG_PRACH)) { - int en = dB_fixed(signal_energy((int32_t*)&rxsigF[0][0],(N_ZC==839) ? 840: 140)); - if (en>60) LOG_D(PHY,"frame %d, slot %d : Trying preamble %d \n",frame,slot,preamble_index); + int en = dB_fixed(signal_energy((int32_t *)rxsigF[0], N_ZC == 839 ? 840 : 140)); + if (en > 60) + LOG_D(PHY, "frame %d, slot %d : Trying preamble %d \n", frame, slot, preamble_index); } if (restricted_set == 0) { // This is the relative offset in the root sequence table (5.7.2-4 from 36.211) for the given preamble index @@ -534,48 +520,43 @@ void rx_nr_prach(PHY_VARS_gNB *gNB, if (preamble_offset != preamble_offset_old) { preamble_offset_old = preamble_offset; - new_dft = 1; + new_dft = true; // This is the \nu corresponding to the preamble index preamble_shift = 0; - } - - else { - preamble_shift -= NCS; + } else { + preamble_shift -= NCS; if (preamble_shift < 0) - preamble_shift+=N_ZC; + preamble_shift += N_ZC; } } else { // This is the high-speed case - new_dft = 0; + new_dft = false; uint16_t nr_du[NR_PRACH_SEQ_LEN_L - 1]; nr_fill_du(N_ZC, prach_root_sequence_map, nr_du); // set preamble_offset to initial rootSequenceIndex and look if we need more root sequences for this // preamble index and find the corresponding cyclic shift // Check if all shifts for that root have been processed + int n_shift_ra = 0, n_shift_ra_bar, d_start = 0; if (preamble_index0 == numshift) { - not_found = 1; - new_dft = 1; + bool not_found = true; + new_dft = true; preamble_index0 -= numshift; - //(preamble_offset==0 && numshift==0) ? (preamble_offset) : (preamble_offset++); - - while (not_found == 1) { + while (not_found) { // current root depending on rootSequenceIndex int index = (rootSequenceIndex + preamble_offset) % N_ZC; - - u = prach_root_sequence_map[index]; - + int u = nr_du[prach_root_sequence_map[index]]; uint16_t n_group_ra = 0; - - if ((nr_du[u] < (N_ZC / 3)) && (nr_du[u] >= NCS) ) { - n_shift_ra = nr_du[u] / NCS; - d_start = (nr_du[u] << 1) + (n_shift_ra * NCS); + + if (u < (N_ZC / 3) && u >= NCS) { + n_shift_ra = u / NCS; + d_start = (u << 1) + (n_shift_ra * NCS); n_group_ra = N_ZC / d_start; - n_shift_ra_bar = max(0, (N_ZC-(nr_du[u] << 1) - (n_group_ra * d_start)) / N_ZC); - } else if ((nr_du[u] >= (N_ZC / 3)) && (nr_du[u] <= ((N_ZC - NCS) >> 1))) { - n_shift_ra = (N_ZC - (nr_du[u] << 1)) / NCS; - d_start = N_ZC - (nr_du[u] << 1) + (n_shift_ra * NCS); - n_group_ra = nr_du[u] / d_start; - n_shift_ra_bar = min(n_shift_ra, max(0, (nr_du[u]- (n_group_ra * d_start)) / NCS)); + n_shift_ra_bar = max(0, (N_ZC - (u << 1) - (n_group_ra * d_start)) / N_ZC); + } else if (u >= (N_ZC / 3) && u <= ((N_ZC - NCS) >> 1)) { + n_shift_ra = (N_ZC - (u << 1)) / NCS; + d_start = N_ZC - (u << 1) + (n_shift_ra * NCS); + n_group_ra = u / d_start; + n_shift_ra_bar = min(n_shift_ra, max(0, (u - (n_group_ra * d_start)) / NCS)); } else { n_shift_ra = 0; n_shift_ra_bar = 0; @@ -584,12 +565,13 @@ void rx_nr_prach(PHY_VARS_gNB *gNB, // This is the number of cyclic shifts for the current root u numshift = (n_shift_ra * n_group_ra) + n_shift_ra_bar; // skip to next root and recompute parameters if numshift==0 - (numshift>0) ? (not_found = 0) : (preamble_offset++); + numshift > 0 ? not_found = false : preamble_offset++; } } if (n_shift_ra>0) - preamble_shift = -((d_start * (preamble_index0/n_shift_ra)) + ((preamble_index0%n_shift_ra)*NCS)); // minus because the channel is h(t -\tau + Cv) + preamble_shift = -(d_start * (preamble_index0 / n_shift_ra) + + (preamble_index0 % n_shift_ra) * NCS); // minus because the channel is h(t -\tau + Cv) else preamble_shift = 0; @@ -604,7 +586,7 @@ void rx_nr_prach(PHY_VARS_gNB *gNB, // Compute DFT of RX signal (conjugate input, results in conjugate output) for each new rootSequenceIndex if (LOG_DEBUGFLAG(DEBUG_PRACH)) { - int en = dB_fixed(signal_energy((int32_t*)&rxsigF[0][0],840)); + int en = dB_fixed(signal_energy((int32_t *)rxsigF[0], 840)); if (en>60) LOG_D(PHY, "frame %d, slot %d : preamble index %d, NCS %d, N_ZC/NCS %d: offset %d, preamble shift %d , en %d)\n", @@ -620,42 +602,30 @@ void rx_nr_prach(PHY_VARS_gNB *gNB, LOG_D(PHY,"PRACH RX preamble_index %d, preamble_offset %d\n",preamble_index,preamble_offset); + if (new_dft) { + new_dft = false; - if (new_dft == 1) { - new_dft = 0; - - Xu = (int16_t*)gNB->X_u[preamble_offset-first_nonzero_root_idx]; - + c16_t *Xu = (c16_t *)gNB->X_u[preamble_offset - first_nonzero_root_idx]; LOG_D(PHY,"PRACH RX new dft preamble_offset-first_nonzero_root_idx %d\n",preamble_offset-first_nonzero_root_idx); - memset(prach_ifft,0,((N_ZC==839) ? 2048 : 256)*sizeof(int32_t)); - memset(prachF, 0, sizeof(int16_t) * 2 * 1024); + memset(prach_ifft, 0, sizeof(prach_ifft)); if (LOG_DUMPFLAG(DEBUG_PRACH)) { LOG_M("prach_rxF0.m","prach_rxF0",rxsigF[0],N_ZC,1,1); LOG_M("prach_rxF1.m","prach_rxF1",rxsigF[1],6144,1,1); } - + c16_t prachF[dft_sz] __attribute__((aligned(32))); for (int aa = 0; aa < nb_rx; aa++) { - // Do componentwise product with Xu* on each antenna - - for (int offset = 0; offset < (N_ZC << 1); offset += 2) { - prachF[offset] = (int16_t)(((int32_t)Xu[offset]*rxsigF[aa][offset] + (int32_t)Xu[offset+1]*rxsigF[aa][offset+1])>>15); - prachF[offset+1] = (int16_t)(((int32_t)Xu[offset]*rxsigF[aa][offset+1] - (int32_t)Xu[offset+1]*rxsigF[aa][offset])>>15); + // Do componentwise product with Xu* on each antenna + for (int offset = 0; offset < N_ZC; offset++) { + prachF[offset] = c16MulConjShift(Xu[offset], rxsigF[aa][offset], 15); } - + memset(prachF + N_ZC, 0, sizeof(*prachF) * (dft_sz - N_ZC)); // Now do IFFT of size 1024 (N_ZC=839) or 256 (N_ZC=139) - if (N_ZC == 839) { - idft(IDFT_1024, prachF, prach_ifft_tmp, 1); - // compute energy and accumulate over receive antennas - for (int i = 0; i < 1024; i++) - prach_ifft[i] += (int32_t)prach_ifft_tmp[i<<1]*(int32_t)prach_ifft_tmp[i<<1] + (int32_t)prach_ifft_tmp[1+(i<<1)]*(int32_t)prach_ifft_tmp[1+(i<<1)]; - } else { - idft(IDFT_256, prachF, prach_ifft_tmp, 1); - log2_ifft_size = 8; - // compute energy and accumulate over receive antennas and repetitions for BR - for (int i = 0; i < 256; i++) - prach_ifft[i] += (int32_t)prach_ifft_tmp[i<<1]*(int32_t)prach_ifft_tmp[(i<<1)] + (int32_t)prach_ifft_tmp[1+(i<<1)]*(int32_t)prach_ifft_tmp[1+(i<<1)]; - } + c16_t prach_ifft_tmp[dft_sz] __attribute__((aligned(32))); + idft(get_idft(dft_sz), (int16_t *)prachF, (int16_t *)prach_ifft_tmp, 1); + // compute energy and accumulate over receive antennas + for (int i = 0; i < dft_sz; i++) + prach_ifft[i] += squaredMod(prach_ifft_tmp[i]); if (LOG_DUMPFLAG(DEBUG_PRACH)) { if (aa == 0) @@ -678,14 +648,14 @@ void rx_nr_prach(PHY_VARS_gNB *gNB, } } // new dft - - // check energy in nth time shift, for - preamble_shift2 = ((preamble_shift==0) ? 0 : ((preamble_shift<*max_preamble_energy) { LOG_D(PHY,"preamble_index %d, delay %d en %d dB > %d dB\n",preamble_index,i,levdB,*max_preamble_energy); *max_preamble_energy = levdB; @@ -693,8 +663,7 @@ void rx_nr_prach(PHY_VARS_gNB *gNB, *max_preamble = preamble_index; } } - }// preamble_index - + } // preamble_index // The conversion from *max_preamble_delay from TA value is done here. // It is normalized to the 30.72 Ms/s, considering the numerology, N_RB and the sampling rate @@ -721,23 +690,6 @@ void rx_nr_prach(PHY_VARS_gNB *gNB, } else *TA = *TA/2; - if (LOG_DUMPFLAG(DEBUG_PRACH)) { - //int en = dB_fixed(signal_energy((int32_t*)&rxsigF[0][0],840)); - // if (en>60) { - int k = (12 * n_ra_prb) - 6 * fp->N_RB_UL; - if (k < 0) - k += fp->ofdm_symbol_size; - - k*=12; - k+=13; - k*=2; - - 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("prach_ifft0.m","prach_t0", prach_ifft, 1024, 1, 1); - // } - } stop_meas(&gNB->rx_prach); } diff --git a/openair1/PHY/defs_gNB.h b/openair1/PHY/defs_gNB.h index ebf459402e..9575e33c7b 100644 --- a/openair1/PHY/defs_gNB.h +++ b/openair1/PHY/defs_gNB.h @@ -151,8 +151,6 @@ typedef struct { /// second index: rx antenna [0..63] (hard coded) \note Hard coded array size indexed by \c nb_antennas_rx. /// third index: frequency-domain sample [0..ofdm_symbol_size*12[ int16_t **rxsigF; - /// \brief local buffer to compute prach_ifft - int32_t *prach_ifft; gNB_PRACH_list_t list[NUMBER_OF_NR_PRACH_MAX]; } NR_gNB_PRACH;