Add Nvidia L1 configuration file in sa_gnb_aerial and sa_gnb_aerial_ul

This commit is contained in:
Reem Bahsoun
2026-01-15 12:01:05 +01:00
parent a5e8e10acf
commit bf78ef08e1
2 changed files with 486 additions and 0 deletions

View File

@@ -0,0 +1,243 @@
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (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.apache.org/licenses/LICENSE-2.0
#
# 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.
aerial_sdk_version: 25-3-cubb
l2adapter_filename: l2_adapter_config_P5G_GH.yaml
aerial_metrics_backend_address: 127.0.0.1:8081
# CPU core shared by all low-priority threads
low_priority_core: 10
enable_ptp_svc_monitoring: 0
ptp_rms_threshold: 20 #ns
nic_tput_alert_threshold_mbps: 85000
cuphydriver_config:
standalone: 0
validation: 0
num_slots: 8
log_level: DBG
profiler_sec: 0
dpdk_thread: 10
dpdk_verbose_logs: 0
accu_tx_sched_res_ns: 500
accu_tx_sched_disable: 0 #Flag applicable only for CX6-DX : BF3/CX7 HW and beyond support accurate send scheduling directly on timestamp, for CX6-DX this is implemented via software emulation
fh_stats_dump_cpu_core: -1
pdump_client_thread: -1
use_green_contexts: 0
use_batched_memcpy: 0 # Experimental feature; requires >= CUDA 12.8
mps_sm_pusch: 100
mps_sm_pucch: 2
mps_sm_prach: 2
mps_sm_ul_order: 20
mps_sm_pdsch: 102
mps_sm_pdcch: 10
mps_sm_pbch: 2
mps_sm_gpu_comms: 16
mps_sm_srs: 16
pdsch_fallback: 0
dpdk_file_prefix: cuphycontroller
nics:
- nic: '0000:01:00.0'
mtu: 9216
cpu_mbufs: 196608
uplane_tx_handles: 64
txq_size: 8192
rxq_size: 16384
gpu: 0
cus_port_failover: 0
gpus:
- 0
# Set GPUID to the GPU sharing the PCIe switch as NIC
# run nvidia-smi topo -m to find out which GPU
workers_ul: [4,5]
workers_dl: [6,7,8]
debug_worker: -1
workers_sched_priority: 95
prometheus_thread: -1
start_section_id_srs: 3072
start_section_id_prach: 2048
enable_ul_cuphy_graphs: 1
enable_dl_cuphy_graphs: 1
ul_order_kernel_mode: 0 # 0: Ping-Pong, 1: dual CTA
ul_order_timeout_cpu_ns: 8000000
ul_order_timeout_log_interval_ns: 1000000000
ul_order_timeout_gpu_ns: 3000000
ul_order_timeout_gpu_srs_ns: 5200000
ul_srs_aggr3_task_launch_offset_ns: 500000
ul_order_timeout_gpu_log_enable: 0
ul_order_max_rx_pkts: 512
ul_order_rx_pkts_timeout_ns: 100000
cplane_disable: 0
gpu_init_comms_dl: 1
cell_group: 1
cell_group_num: 1
fix_beta_dl: 0
bfw_beta_prescaler: 16384
pusch_workCancelMode: 2 # set cuphyPuschStatPrms_t workCancelMode field to PUSCH_DEVICE_GRAPHS; this only sets the mode and does not cancel work;
# work cancellation (i.e., not launching some of the device graphs) can only occur after an explicit internal
# memory write is timely observed by a PUSCH graph.
pusch_sinr: 2
pusch_rssi: 1
pusch_tdi: 1
pusch_cfo: 1
pusch_dftsofdm: 0
pusch_to: 1
pusch_select_eqcoeffalgo: 3
pusch_select_chestalgo: 1
pusch_enable_perprgchest: 0
pusch_tbsizecheck: 1
pusch_deviceGraphLaunchEn: 1
pusch_waitTimeOutPreEarlyHarqUs: 3000
pusch_waitTimeOutPostEarlyHarqUs: 3000
puxch_polarDcdrListSz: 8
enable_cpu_task_tracing: 0
enable_prepare_tracing: 0
disable_empw: 0 #1=>Disables Multi packet WQE feature
cqe_tracer_config:
enable_dl_cqe_tracing: 0
cqe_trace_cell_mask: 1 #[64 bit mask: Bit0->Cell0, Bit1->Cell1...]
cqe_trace_slot_mask: 196800 #[20 bit mask: Bit0->Slot0, Bit1->Slot1...] 196800=>All *6,*7 DL Slots
ul_rx_pkt_tracing_level: 0
ul_rx_pkt_tracing_level_srs: 0
pmu_metrics: 0 #Enable and select PMU metrics, 0=disabled, 1=General Counters (platform agnostic), 2=Topdown Metrics (Grace only), 3=Cache Metrics (Grace only)
enable_h2d_copy_thread: 1
h2d_copy_thread_cpu_affinity : 11
h2d_copy_thread_sched_priority : 95 #0->SCHED_OTHER, >0->Actual
split_ul_cuda_streams: 0 # 1=Put UL slot 4 and slot 5 on different streams for DDDSUUDDDD pattern
serialize_pucch_pusch: 0 # 1=Force serialization of PUSCH/PUCCH
# Note: for Early Harq (EH) order is PUSCH EH -> PUCCH -> PUSCH non EH
# for non-Early Harq order is PUCCH -> all PUSCH processing
aggr_obj_non_avail_th: 5 # Threshold for consecutive non-availability of Aggregated objects(UL/DL) or DL/UL buffers
dl_wait_th_ns:
- 500000 #H2D copy wait threshold
- 4000000 #cuPHY DL channel wait threshold
sendCPlane_timing_error_th_ns : 0
sendCPlane_ulbfw_backoff_th_ns: 300000 # ULBFW must be completed by cplane send time - this is the backoff time
sendCPlane_dlbfw_backoff_th_ns: 300000 # DLBFW must be completed by cplane send time - this is the backoff time
pusch_forcedNumCsi2Bits: 0
mMIMO_enable: 0
enable_srs: 1
ue_mode: 0
mCh_segment_proc_enable: 0
pusch_aggr_per_ctx: 3
prach_aggr_per_ctx: 2
pucch_aggr_per_ctx: 4
srs_aggr_per_ctx: 3
ul_input_buffer_per_cell: 10
ul_input_buffer_per_cell_srs: 6
max_harq_pools: 384
max_harq_tx_count_bundled: 10
max_harq_tx_count_non_bundled: 4
ok_testbench_config:
enable_ok_tb: 0
num_ok_tb_slot: 0 #Presently not used
max_ru_unhealthy_ul_slots: 100
srs_chest_algo_type: 0
srs_chest_tol2_normalization_algo_type: 1
srs_chest_tol2_constant_scaler: 32768.0
bfw_power_normalization_alg_selector: 1
total_num_srs_chest_buffers: 6144
send_static_bfw_wt_all_cplane: 1
ul_pcap_capture_enable: 0
ul_pcap_capture_thread_cpu_affinity: 10
ul_pcap_capture_thread_sched_priority: 95
pcap_logger_ul_cplane_enable: 0
pcap_logger_dl_cplane_enable: 0
pcap_logger_thread_cpu_affinity: 10
pcap_logger_thread_sched_prio: 95
pcap_logger_file_save_dir: ./
dlc_bfw_enable_divide_per_cell: 0 # Enable/disable dividing DL BFW C-Plane transmission window among multiple cells in a slot
ulc_bfw_enable_divide_per_cell: 0 # Enable/disable dividing UL BFW C-Plane transmission window among multiple cells in a slot
dlc_alloc_cplane_bfw_txq: 0 # Allocates a second separate queue for BFW DL C-Plane transmission
ulc_alloc_cplane_bfw_txq: 1 # Allocates a second separate queue for BFW UL C-Plane transmission
#static beam id range
static_beam_id_start: 1
static_beam_id_end: 16527
#dynamic beam id range
dynamic_beam_id_start: 16528 #Dynamic beam ID start should be after static beam ID end
dynamic_beam_id_end: 32767 #15bit max value
bfw_c_plane_chaining_mode: 0 # 0: Default memcpy of BFW weights, 1: CPU mbuf chaining of BFW weights, 2: GPU mbuf chaining
enable_l1_param_sanity_check: 0
pusch_nMaxLdpcHetConfigs: 32
notify_ul_harq_buffer_release: 0
# data_config:
# datalake_core: 12 # This must be set for Data Lake and/or E3 Agent to be used
# datalake_db_write_enable: 1 # Enable/disable ClickHouse writes based on datalake_data_types
# datalake_samples: 1000000
# datalake_address: localhost
# datalake_engine: "MergeTree() PRIMARY KEY (TsTaiNs)"
# # datalake_drop_tables: 0 # Set to 1 to drop Clickhouse tables at startup
# datalake_data_types: [fh, pusch, hest] # Data types to write to ClickHouse DB (all types always collected in memory for E3 Agent):
# fh = Fronthaul IQ samples,pusch = PUSCH/FAPI data, hest = channel estimates
# datalake_store_failed_pdu: 0 # Set to 1 to store PDU data even when CRC fails
# e3_agent_enable: 0 # Enable/disable E3 Agent and dApp capabilities
# num_rows_fh: 120
# num_rows_pusch: 200
# num_rows_hest: 140
# e3_pub_port: 5555 # E3 publisher port for indication messages
# e3_rep_port: 5556 # E3 reply port for E3AP requests
# e3_sub_port: 5560 # E3 subscriber port for Manager messages
cells:
- name: O-RU 0
cell_id: 1
ru_type: 1
# set to 00:00:00:00:00:00 to use the MAC address of the NIC port to use
src_mac_addr: 00:00:00:00:00:00
dst_mac_addr: e8:c7:cf:ad:12:0b #WNC n78
nic: '0000:01:00.0'
vlan: 500
pcp: 7
txq_count_uplane: 1
eAxC_id_ssb_pbch: [0, 1, 2, 3]
eAxC_id_pdcch: [0, 1, 2, 3]
eAxC_id_pdsch: [0, 1, 2, 3]
eAxC_id_csirs: [0, 1, 2, 3]
eAxC_id_pusch: [0, 1, 2, 3]
eAxC_id_pucch: [0, 1, 2, 3]
eAxC_id_srs: [8, 9, 10, 11]
eAxC_id_prach: [4, 5, 6, 7]
dl_iq_data_fmt: {comp_meth: 1, bit_width: 9}
ul_iq_data_fmt: {comp_meth: 1, bit_width: 9}
section_3_time_offset: 484
fs_offset_dl: 7
exponent_dl: 4
ref_dl: 0
fs_offset_ul: -5
exponent_ul: 4
max_amp_ul: 65504
mu: 1
T1a_max_up_ns: 280000
T1a_max_cp_ul_ns: 405000
Ta4_min_ns: 50000
Ta4_max_ns: 331000
Tcp_adv_dl_ns: 125000
ul_u_plane_tx_offset_ns: 280000
fh_len_range: 0
pusch_prb_stride: 273
prach_prb_stride: 12
srs_prb_stride: 273
pusch_ldpc_max_num_itr_algo_type: 1
pusch_fixed_max_num_ldpc_itrs: 10
pusch_ldpc_early_termination: 0
pusch_ldpc_algo_index: 0
pusch_ldpc_flags: 2
pusch_ldpc_use_half: 1
pusch_nMaxPrb: 273
pusch_nMaxRx: 4
ul_gain_calibration: 78.68
lower_guard_bw: 845
tv_pusch: cuPhyChEstCoeffs.h5

View File

@@ -0,0 +1,243 @@
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (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.apache.org/licenses/LICENSE-2.0
#
# 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.
aerial_sdk_version: 25-3-cubb
l2adapter_filename: l2_adapter_config_P5G_GH.yaml
aerial_metrics_backend_address: 127.0.0.1:8081
# CPU core shared by all low-priority threads
low_priority_core: 10
enable_ptp_svc_monitoring: 0
ptp_rms_threshold: 20 #ns
nic_tput_alert_threshold_mbps: 85000
cuphydriver_config:
standalone: 0
validation: 0
num_slots: 8
log_level: DBG
profiler_sec: 0
dpdk_thread: 10
dpdk_verbose_logs: 0
accu_tx_sched_res_ns: 500
accu_tx_sched_disable: 0 #Flag applicable only for CX6-DX : BF3/CX7 HW and beyond support accurate send scheduling directly on timestamp, for CX6-DX this is implemented via software emulation
fh_stats_dump_cpu_core: -1
pdump_client_thread: -1
use_green_contexts: 0
use_batched_memcpy: 0 # Experimental feature; requires >= CUDA 12.8
mps_sm_pusch: 100
mps_sm_pucch: 2
mps_sm_prach: 2
mps_sm_ul_order: 20
mps_sm_pdsch: 102
mps_sm_pdcch: 10
mps_sm_pbch: 2
mps_sm_gpu_comms: 16
mps_sm_srs: 16
pdsch_fallback: 0
dpdk_file_prefix: cuphycontroller
nics:
- nic: '0000:01:00.0'
mtu: 9216
cpu_mbufs: 196608
uplane_tx_handles: 64
txq_size: 8192
rxq_size: 16384
gpu: 0
cus_port_failover: 0
gpus:
- 0
# Set GPUID to the GPU sharing the PCIe switch as NIC
# run nvidia-smi topo -m to find out which GPU
workers_ul: [4,5]
workers_dl: [6,7,8]
debug_worker: -1
workers_sched_priority: 95
prometheus_thread: -1
start_section_id_srs: 3072
start_section_id_prach: 2048
enable_ul_cuphy_graphs: 1
enable_dl_cuphy_graphs: 1
ul_order_kernel_mode: 0 # 0: Ping-Pong, 1: dual CTA
ul_order_timeout_cpu_ns: 8000000
ul_order_timeout_log_interval_ns: 1000000000
ul_order_timeout_gpu_ns: 3000000
ul_order_timeout_gpu_srs_ns: 5200000
ul_srs_aggr3_task_launch_offset_ns: 500000
ul_order_timeout_gpu_log_enable: 0
ul_order_max_rx_pkts: 512
ul_order_rx_pkts_timeout_ns: 100000
cplane_disable: 0
gpu_init_comms_dl: 1
cell_group: 1
cell_group_num: 1
fix_beta_dl: 0
bfw_beta_prescaler: 16384
pusch_workCancelMode: 2 # set cuphyPuschStatPrms_t workCancelMode field to PUSCH_DEVICE_GRAPHS; this only sets the mode and does not cancel work;
# work cancellation (i.e., not launching some of the device graphs) can only occur after an explicit internal
# memory write is timely observed by a PUSCH graph.
pusch_sinr: 2
pusch_rssi: 1
pusch_tdi: 1
pusch_cfo: 1
pusch_dftsofdm: 0
pusch_to: 1
pusch_select_eqcoeffalgo: 3
pusch_select_chestalgo: 1
pusch_enable_perprgchest: 0
pusch_tbsizecheck: 1
pusch_deviceGraphLaunchEn: 1
pusch_waitTimeOutPreEarlyHarqUs: 3000
pusch_waitTimeOutPostEarlyHarqUs: 3000
puxch_polarDcdrListSz: 8
enable_cpu_task_tracing: 0
enable_prepare_tracing: 0
disable_empw: 0 #1=>Disables Multi packet WQE feature
cqe_tracer_config:
enable_dl_cqe_tracing: 0
cqe_trace_cell_mask: 1 #[64 bit mask: Bit0->Cell0, Bit1->Cell1...]
cqe_trace_slot_mask: 196800 #[20 bit mask: Bit0->Slot0, Bit1->Slot1...] 196800=>All *6,*7 DL Slots
ul_rx_pkt_tracing_level: 0
ul_rx_pkt_tracing_level_srs: 0
pmu_metrics: 0 #Enable and select PMU metrics, 0=disabled, 1=General Counters (platform agnostic), 2=Topdown Metrics (Grace only), 3=Cache Metrics (Grace only)
enable_h2d_copy_thread: 1
h2d_copy_thread_cpu_affinity : 11
h2d_copy_thread_sched_priority : 95 #0->SCHED_OTHER, >0->Actual
split_ul_cuda_streams: 0 # 1=Put UL slot 4 and slot 5 on different streams for DDDSUUDDDD pattern
serialize_pucch_pusch: 0 # 1=Force serialization of PUSCH/PUCCH
# Note: for Early Harq (EH) order is PUSCH EH -> PUCCH -> PUSCH non EH
# for non-Early Harq order is PUCCH -> all PUSCH processing
aggr_obj_non_avail_th: 5 # Threshold for consecutive non-availability of Aggregated objects(UL/DL) or DL/UL buffers
dl_wait_th_ns:
- 500000 #H2D copy wait threshold
- 4000000 #cuPHY DL channel wait threshold
sendCPlane_timing_error_th_ns : 0
sendCPlane_ulbfw_backoff_th_ns: 300000 # ULBFW must be completed by cplane send time - this is the backoff time
sendCPlane_dlbfw_backoff_th_ns: 300000 # DLBFW must be completed by cplane send time - this is the backoff time
pusch_forcedNumCsi2Bits: 0
mMIMO_enable: 0
enable_srs: 1
ue_mode: 0
mCh_segment_proc_enable: 0
pusch_aggr_per_ctx: 9 #3
prach_aggr_per_ctx: 2
pucch_aggr_per_ctx: 4
srs_aggr_per_ctx: 3
ul_input_buffer_per_cell: 10
ul_input_buffer_per_cell_srs: 6
max_harq_pools: 384
max_harq_tx_count_bundled: 10
max_harq_tx_count_non_bundled: 4
ok_testbench_config:
enable_ok_tb: 0
num_ok_tb_slot: 0 #Presently not used
max_ru_unhealthy_ul_slots: 100
srs_chest_algo_type: 0
srs_chest_tol2_normalization_algo_type: 1
srs_chest_tol2_constant_scaler: 32768.0
bfw_power_normalization_alg_selector: 1
total_num_srs_chest_buffers: 6144
send_static_bfw_wt_all_cplane: 1
ul_pcap_capture_enable: 0
ul_pcap_capture_thread_cpu_affinity: 10
ul_pcap_capture_thread_sched_priority: 95
pcap_logger_ul_cplane_enable: 0
pcap_logger_dl_cplane_enable: 0
pcap_logger_thread_cpu_affinity: 10
pcap_logger_thread_sched_prio: 95
pcap_logger_file_save_dir: ./
dlc_bfw_enable_divide_per_cell: 0 # Enable/disable dividing DL BFW C-Plane transmission window among multiple cells in a slot
ulc_bfw_enable_divide_per_cell: 0 # Enable/disable dividing UL BFW C-Plane transmission window among multiple cells in a slot
dlc_alloc_cplane_bfw_txq: 0 # Allocates a second separate queue for BFW DL C-Plane transmission
ulc_alloc_cplane_bfw_txq: 1 # Allocates a second separate queue for BFW UL C-Plane transmission
#static beam id range
static_beam_id_start: 1
static_beam_id_end: 16527
#dynamic beam id range
dynamic_beam_id_start: 16528 #Dynamic beam ID start should be after static beam ID end
dynamic_beam_id_end: 32767 #15bit max value
bfw_c_plane_chaining_mode: 0 # 0: Default memcpy of BFW weights, 1: CPU mbuf chaining of BFW weights, 2: GPU mbuf chaining
enable_l1_param_sanity_check: 0
pusch_nMaxLdpcHetConfigs: 32
notify_ul_harq_buffer_release: 0
# data_config:
# datalake_core: 12 # This must be set for Data Lake and/or E3 Agent to be used
# datalake_db_write_enable: 1 # Enable/disable ClickHouse writes based on datalake_data_types
# datalake_samples: 1000000
# datalake_address: localhost
# datalake_engine: "MergeTree() PRIMARY KEY (TsTaiNs)"
# # datalake_drop_tables: 0 # Set to 1 to drop Clickhouse tables at startup
# datalake_data_types: [fh, pusch, hest] # Data types to write to ClickHouse DB (all types always collected in memory for E3 Agent):
# fh = Fronthaul IQ samples,pusch = PUSCH/FAPI data, hest = channel estimates
# datalake_store_failed_pdu: 0 # Set to 1 to store PDU data even when CRC fails
# e3_agent_enable: 0 # Enable/disable E3 Agent and dApp capabilities
# num_rows_fh: 120
# num_rows_pusch: 200
# num_rows_hest: 140
# e3_pub_port: 5555 # E3 publisher port for indication messages
# e3_rep_port: 5556 # E3 reply port for E3AP requests
# e3_sub_port: 5560 # E3 subscriber port for Manager messages
cells:
- name: O-RU 0
cell_id: 1
ru_type: 1
# set to 00:00:00:00:00:00 to use the MAC address of the NIC port to use
src_mac_addr: 00:00:00:00:00:00
dst_mac_addr: e8:c7:cf:ad:12:0b #WNC n78
nic: '0000:01:00.0'
vlan: 500
pcp: 7
txq_count_uplane: 1
eAxC_id_ssb_pbch: [0, 1, 2, 3]
eAxC_id_pdcch: [0, 1, 2, 3]
eAxC_id_pdsch: [0, 1, 2, 3]
eAxC_id_csirs: [0, 1, 2, 3]
eAxC_id_pusch: [0, 1, 2, 3]
eAxC_id_pucch: [0, 1, 2, 3]
eAxC_id_srs: [8, 9, 10, 11]
eAxC_id_prach: [4, 5, 6, 7]
dl_iq_data_fmt: {comp_meth: 1, bit_width: 9}
ul_iq_data_fmt: {comp_meth: 1, bit_width: 9}
section_3_time_offset: 484
fs_offset_dl: 7
exponent_dl: 4
ref_dl: 0
fs_offset_ul: -5
exponent_ul: 4
max_amp_ul: 65504
mu: 1
T1a_max_up_ns: 280000
T1a_max_cp_ul_ns: 405000
Ta4_min_ns: 50000
Ta4_max_ns: 331000
Tcp_adv_dl_ns: 125000
ul_u_plane_tx_offset_ns: 280000
fh_len_range: 0
pusch_prb_stride: 273
prach_prb_stride: 12
srs_prb_stride: 273
pusch_ldpc_max_num_itr_algo_type: 1
pusch_fixed_max_num_ldpc_itrs: 10
pusch_ldpc_early_termination: 0
pusch_ldpc_algo_index: 0
pusch_ldpc_flags: 2
pusch_ldpc_use_half: 1
pusch_nMaxPrb: 273
pusch_nMaxRx: 4
ul_gain_calibration: 78.68
lower_guard_bw: 845
tv_pusch: cuPhyChEstCoeffs.h5