Thread-pool support for TX symbol processing.
This MR adds thread-pool support for TX symbol processing. It allows modulation
/mapping/layer-precoding to run in parallel and offers a speedup of around 3
compared to single-thread execution. This is particularly important for large
bandwidths and 4 or more TX antenna ports where the precoding operation is quite
computationally-intensive.
Please see the description of !3489 for the performance comparison results and
the nr_dlsim timing measurements.
* Add option `--L1s.L1_num_tx_sym_per_thread` for the softmodems and `-Y` for `nr_dlsim`
to provide the number of symbols processed per thread.
It defaults to 0 which makes that every symbols are processed in one thread.
* The last symbol processing task is processed in the L1 TX thread.
Further unused arguments cleanup
More fixes for unused function arguments (see #1057),
add_compile_options(-Wunused-parameter) added in folders:
- nfapi
- openair2
- openair3
- USRP
- rfsimulator
* Global timers were started and stopped
in the TX symbol processing tasks
which is not thread safe
* gNBs nrL1_stats.log now shows
PDSCH generation time rather than
layer mapping and precoding times
There was an indexing error in the calculation of the symbol offset `re_beginning_of_symbol`.
The symbol was tested to hold PTRS or DMRS based on the index of the
first symbol processed in the task and not based on the index of the
symbol as it should be.
Then PDSCH generation was not working properly for more than one symbol
per task.
Cleanup common_lib.h
This change reduces the number of dependecies of common_lib.h and simplifies
the task of implementing an external OAI radio library.
Remove a lot of unused code, defines, and config options
Remove unused header files and defines.
Slightly clean up gnb_config.c by removing all _IDX variables. Remove
these unused parameters:
- MACRLCs.[0].num_cc
- MACRLCs.[0].local_n_portc
- MACRLCs.[0].remove_n_portc
- MACRLCs.[0].remote_s_portc
- MACRLCs.[0].remote_s_portd
- L1s.[0].num_cc
- L1s.[0].local_n_portc
- remove test_multipath and test_noise from ChanelSim tests of
RAN-Channel-Simulation pipeline - will be executed together with the
other CUDA-enabled unit tests
- add benchmark_channel_pipeline to CTest
- add test_channel_pipeline to CTest
- add `cuda` label to CUDA-dependent tests for easier execution and
filtering
This change removes the cudaDevAttrIntegrated check and only requires
pageable memory access. This change allows to run GPU accelerated
channel convolution on GH machines.
Update doc: split RUNMODEM.md, clarify USRP-specific workarounds
This reworks the documentation:
- split up RUNMODEM.md into separate documents for gNB, UE, NTN
- explains USRP workaround and patches
- makes a standalone UE doc file that explains UE-specific configuration
& modes
- better explains physical simulators
- adds a document on tracy
- cleanup
More compilation improvements
Mostly to decouple NR files from LTE headers.
At least in one instance, ulsch_input_buffer_array, using an LTE constant in
NR code was quite dangerous because the buffer in NR could easily be larger
than what the LTE constant foresee.
UE: Fix CSI RS Measurement
1. Fix regression introduced in !3834.
2. ZMQ has no interference or artificial noise so the interference plus noise
measured is always 0. nr_csi_rs_pmi_estimation() returns SINR as 0 if noise
is 0 which reports CQI as 0. This fix computes SINR if noise is 0 instead of
returning.
This allows to compile physical simulators by compiling only tests with
"ninja tests". It fixes a regression.
Fixes: ec8efe1b7e ("Unify physim test definition")
Added two versions of channel convolution and noise generation:
- accelerated via threadpool
- accelerated using CUDA
The main difference between this and previous versions of channel
convolution implementations is that these functions take a real-world
approach to input/output where both could be split unevenly over a set of
buffers, e.g. ring buffers used in vrtsim.
The CUDA-accelerated version only works on systems with unified memory, e.g. NVidia
DGX Spark or GH
- all RAN code, CI code, configuration files, dockerfiles, in CSSL v1.0
- all deployment code (openshift, charts, ancillary files like shell
scripts), in MIT
- documentation in CC-BY-4.0
- exceptions might apply and are listed in NOTICE
- there is a new LICENSES folder with all licenses
- CONTRIBUTIONS.md has been updated accordingly
For automated changes based on OAI PL v1.1:
perl -i~ -0pe 's/\/\*.*Licensed to the OpenAirInterface.*openairinterface.org\n#?/\/*\n * SPDX-License-Identifier: LicenseRef-CSSL-1.0\n/s' **/*.{c,h,cpp}
perl -i~ -0pe 's/\/\*.*Licensed to the OpenAirInterface.*openairinterface.org\n#?/\/*\n * SPDX-License-Identifier: LicenseRef-CSSL-1.0\n/s' **/*.ts
perl -i~ -0pe 's/<!--.*Licensed to the OpenAirInterface.*openairinterface.org\n.*-->/<!-- SPDX-License-Identifier: LicenseRef-CSSL-1.0 -->/s' **/*.xml
The rest (cmake, files with missing license, cmake) manually.
Remove the \file directive, as it is always superfluous because the
current file is implicit [1]:
> If the file name is omitted (i.e. the line after \file is left blank)
> then the documentation block that contains the \file command will belong
> to the file it is located in.
Author names and e-mails are not relevant for us: it can always be
inferred from git blame, and is often outdated.
Eurecom code has been contributed and was under OAI PL v1.0/v1.1.
For the cpack package contact: put generic email address that is
independent of an individual and that will remain reachable.
[1] https://www.doxygen.nl/manual/commands.html#cmdfile
Fix Delta MCS mode, reimplement UL power control
This change set fixes the "delta-MCS" mode, and reimplements the UL power
control loops at gNB, i.e., for PUSCH and PUCCH. It further adds new T traces
to log PUSCH/PUCCH power infrmation, and cleans up code.
The "delta-MCS" mode in OAI refers to configuring a UE with deltaMCS in the
PUSCH-PowerControl IE (see TS 38.331). In this mode, the UE adjusts the PUSCH
power for transmissions to take into account for MCS (see TS 38.213 Sec. 7.1).
In other words, the target SNR can be set to a somewhat low value, and the UE
will still use the right power for UL transmissions. Note that the spec only
foresees this adjustment of UL power when using one layer. In contrast, in the
"normal" mode, the gNB sends TPC commands to keep a UE at a fixed target SNR
(default: 20dB), regardless of the MCS or number of layers. Future work is
planned to completely remove this fixed target SNR, which is not done in this
MR yet.
Further, there is a new power control loop implementation. Prior to this MR, on
each UL transmissions, the gNB uses the SNR to decide about TPC sent to a UE in
the next transmissions. The new implementation relies on averaging the SNR, and
a "TPC in flight" average to account for the slow reaction of the average SNR.
The sum of both quantities is the current SNR that is used by the UE. See the
commit messages or documentation for more information.
Further cleanup to group data (e.g., power control configuration, refactoring of
code to save space, ...) is done. Two new T traces GNB_MAC_PUSCH_POWER_CONTROL
and GNB_MAC_PUCCH_POWER_CONTROL are added; documentation exists to explain how
to graphically plot corresponding graphs.
PUCCH Format 1 decoder implementation and validation with nr_pucchsim
This is the implementation of the PUCCH Format 1 decoder, validated with
pucchsim in the current branch.
Build:
./build_oai --nrUE --gNB --ninja --phy_simulators
Run tests:
ctest -L nr_pucchsim -j 8
This commit completes the PUCCH Format 1 decoder implementation using
maximum-likelihood detection with RE averaging across symbols/RB (and
hop-wise averaging for intra-slot hopping). It also updates the PUCCH
simulation/test flow to align with the decoder changes, including
payload and logging consistency fixes needed for correct validation.
Remove old CUDA LDPC implementation
A new implementation is under way in !3955. To simplify the MR, remove the old
LDPC CUDA implementation in a first step.
* Option to build ldpc_cuda was remaining in the build script
* Build script option to build ldpc_cuda was still used in CI
* Mentions of ldpc_cuda were remaining in the documentation
Note: In the documentation, the example of ldpctest usage that was
loading ldpc_cuda was replaced by an example in which it is loading
ldpc_orig even though this is pointless as it makes ldpctest compare
ldpc_orig with ldpc_orig. On the other hand, it is the only library
other than the default ldpc that uses the former interface.
replace static allocation of max antennas in a dynamic structure for gnb rach
NB_ANTENNAS_RX was used in 5G by mistake
there was also a error in O-RU implementation, with no consequences
(index not in right order)
Jaroslava Fiedlerova