Without +crypto, GCC does not define __ARM_FEATURE_AES, so SIMDe's
simde_mm_clmulepi64_si128 falls back to a scalar carry-less multiply
(16 integer multiplies + many AND/XOR ops per call) instead of emitting
vmull_p64. The NXP LS2160A A72 has full hardware crypto (aes, pmull,
sha1, sha2) but -march=armv8-a+simd does not enable it.
Split the A72 / A53 march cases:
A72 (0xd08): -march=armv8-a+simd+crypto (hardware crypto confirmed)
A53 (0xd03): -march=armv8-a+simd (crypto optional on A53)
Result on A72 @ 2 GHz (nr_dlsim -n100 -e25 -s25 -R273 -b273 -P -x2 -y4 -z4 -p55):
DLSCH Outer CRC: ~250 us -> 22.9 us (~11x)
DLSCH segmentation: ~250 us -> 37.5 us (~6.7x)
Total DLSCH encoding: ~1500 us -> ~1057 us
Assisted by Claude Sonnet 4.6
Replace the generic j-loop (7 serial ORRs, 28-cycle critical path, j-loop
overhead) with a fully-unrolled ns=8 specialization for Cortex-A72:
- Pre-extract 8 segment base pointers into registers so GCC holds them
across the 528-iteration inner loop instead of reloading from input[].
- CONTRIB macro: vcombine(vld1_dup(p+bi), vld1_dup(p+bi+1)) + VTST +
VAND to produce each segment's bit-mask contribution in one expression;
avoids ANDNOT+CMPEQ+AND (saves one NEON op per segment per iteration).
- asm volatile barrier 1: "+w" on all 8 contributions forces GCC to
materialize r0..r7 as physical NEON registers before any ORR is issued.
Without it GCC left-folds each ORR greedily as soon as one input is
ready (28-cycle serial chain). With it the four level-1 ORRs are
data-independent and A72's 2-wide NEON pipe issues two per cycle.
- asm volatile barrier 2: "+w" on r01/r23/r45/r67 forces level-1 results
before level-2. GCC's instruction scheduler then fills the 4-cycle
level-2 ORR latency gap with the loop's scalar work (add i_byte, cmp),
placing those instructions between level-2 and level-3/4 ORRs. This
lets A72's out-of-order engine start the next iteration's address chain
(lsr bi = i_byte>>3) ~4 cycles earlier, hiding load latency for free.
- ns != 8 fallback: original j-loop preserved for 1..7 segments.
Measured on NXP LS2160A (Cortex-A72, 2.0 GHz), BG1 Zc=384 (-n100 -S8 -s3.0):
Before: ~13 µs → After: ~9-10 µs (~25% reduction)
Assisted by Claude Sonnet 4.6
Input interleaving (ldpc_encoder_optim8segmulti.c): replace the SIMDE
shuffle_epi8 path (-> NEON TBL, 5-cycle latency) with direct NEON
vcombine(vld1_dup_u8, vld1_dup_u8) to broadcast each input byte, and
use VTST+AND instead of ANDNOT+CMPEQ+AND (one fewer op per segment per
iteration). On NXP LS2160A (Cortex-A72, 2 GHz) this reduces the input
interleaving phase from 20.2 us to ~2.0 us (10x speedup), and total
encoder latency from ~126 us to ~107 us for 8 segments.
Parity-check VLA fix (ldpc_encode_parity_check.c): under USE_ALIGNR
(aarch64, BG1, Zc=384) the simd_size memcpy copies are skipped, so
allocate only one copy (vla_simd=1) instead of simd_size=32, reducing
the stack VLA from 528 KB to 16,896 bytes.
CMakeLists.txt: add Cortex-A72 (CPUPART 0xd08) detection with
-mcpu=cortex-a72 -march=armv8-a+simd (ARMv8.0-A only, no ARMv8.2-A).
ldpctest.c: wire per-phase timers (tinput, tparity, toutput,
tinput_memcpy) into the encoder_implemparams so each phase can be
measured independently.
ldpc_encoder.c / ldpc384_byte_128.c: add code generator and generated
output for a 4-wide 128-bit parity-check variant (ldpc_BG1_Zc384_byte_128_x4)
that unrolls 4 i2 iterations per outer loop; not yet wired into the
call path.
Assisted by Claude Sonnet 4.6
- make internal links relative where applicable
- delete link to the wiki, as the documentation is in the main repo, not
the wiki
- remove some "example in oai code" as the examples either don't exist,
or are not in that place, and we can reasonably expect people to grep
Signed-off-by: Robert Schmidt <robert.schmidt@openairinterface.org>
Add GitHub Actions workflow with 5 jobs:
- detect-changes: uses dorny/paths-filter to check whether files under
ci-scripts/ or .github/ were modified
- welcome-first-time-contributor: getting started and useful links for
first time contributors
- pr-retrigger-ci-instructions: instructions on how to retrigger the CI
without any PR changes
- require-maintainer-approval: requires manual approval via the
ci-approval GitHub environment when protected files changed,
preventing untrusted changes from triggering Jenkins automatically
- trigger-jenkins: triggers Jenkins by forwarding the GitHub event
payload via curl, skipping the approval gate when no protected files
changed
Add support for retriggering CI by adding a label to a PR. The label is
removed automatically after Jenkins is triggered, regardless of success
or failure, so it can be re-added at any time for subsequent retriggers.
Assisted-By: Claude:claude-sonnet-4-6
Co-authored-by: Sagar Arora <sagar.arora@openairinterface.org>
Co-authored-by: Shubhika Garg <shubhika.garg@openairinterface.org>
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Make the image tag unique per scheduled run and consistent with the branch
naming convention used in the other pipelines by suffixing the commit hash:
testBranch = "${sourceBranch}-${commitID}"
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Replace the manual fetch-and-merge approach with pre-built PR merge
refs to simplify checkout and eliminate explicit remote management:
- Add Job init stage to set build name and description for traceability
- Add Checkout stage that fetches refs/pull/<N>/merge for PR builds and
checks out targetBranch directly for push builds
- Remove sourceRepo parameter and "Add source repo & fetch branches" stage
- Remove "Merge target into source" stage (merge is handled by the GitHub
PR merge ref at checkout time)
Co-authored-by: Shubhika Garg <shubhika.garg@openairinterface.org>
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Replace GitLab-specific CI infrastructure with GitHub-compatible
equivalents in the main OAI RAN pipeline:
- Rename Jenkinsfile-GitLab-Container to Jenkinsfile-GitHub-Container
- Replace GitLab CI env vars with GitHub PR equivalents provided by
GitHub Integration plugin and GitHub Plugin
- Replace addGitLabMRComment / gitlabCommitStatus with githubPRComment
and githubNotify for PR comments and commit status reporting
- label validation within Jenkinsfile, remove the now-unused
checkGitLabMergeRequestLabels.sh helper script
- Remove git fetch from the pre-ci-check script as we now use
GITHUB_PR_HEAD_SHA. Fetching is redundant and may not retrieve
the PR head commit for fork-based contributions.
Move githubNotify for downstream jobs from triggerDownstreamJob into
finalizeDownstreamJob so each status check links directly to the HTML
test report artifact rather than the generic build page:
- Add downstreamResults map to pass job results (SUCCESS/FAILURE)
from trigger functions to finalizeDownstreamJob without changing the
existing always/failure post-block structure
- Map any non-success result (UNSTABLE, ABORTED) to FAILURE for GitHub
- Add targetUrl: BUILD_URL to the pipeline-level success/failure
githubNotify calls so the top-level check also links to the build
Consolidate GitHub PR comments to reduce noise:
- Buffer mrValidationWarning instead of posting it as a separate comment,
fold it into the final success or failure comment
- Remove the inline Local-Repo-Push-Dev failure comment, append to
failingStages so it appears in the single final failure comment
- Drop the success PR comment for clean builds - post only when there is
a validation warning to report
Assisted-By: Claude:claude-sonnet-4-6
Co-authored-by: Shubhika Garg <shubhika.garg@openairinterface.org>
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Integration: `2026.w22`
* !4140 Fix evaluation of PMI and CQI report when there is no RI report
* !4125 fix(phy): prevent division by zero in nr_csi_rs_pmi_estimation
* !4134 Add SRS LS and RSRP measurement logging for indoor positioning research
* !4086 vrtsim: Refactor peer antenna configuration and taps_client
* !4108 XNAP: Add encode/decode+unit tests for Xn HO Ack, Prep Failure, SN Status Transfer, UE Context Release
* !4096 O-RU fronthaul library
* !4147 1-port CQI report
* !4133 fix(config): fix memory safety, memory leaks, and add unit test coverage
* !4084 gold_cache: replace linear search with open-addressing hash table
* !4150 fix a few issues with unit tests
* !3827 Antenna port indexing of txdataF using FAPI spatial stream indices
* !4138 \[FHI72\] \[M-plane\] Implement dynamic compression
* !4144 fix(fhi72): Fix use-after-free bug in fhi_72 lib
* !3890 Fixes for multiplexing pattern 3 operations in FR2
* !4145 fix(vrtsim): handle CLOCK_REALTIME jump after PTP sync
* !4151 minor: fix nr rlc test make-based compilation
* !4141 wrong usage of extern declaration in C files, header cleanup
* !4153 doc: update 7.2 tutorial hardware and os configuration and add a doc to list supported hardware and OS
* !4066 Separate nfapi_p5.c into LTE and NR implementations
* !4098 \[CI OAI-FlexRIC\] Improve CI pipeline
* !3723 Support for aperiodic SRS at gNB
* !4076 Fix imscope and tracy build
* Fix the undefined `res` parameter if not defined K_RELEASE
* CI: Adjust timing thresholds in RAN-gNB-N300-Timing-Phytest-LDPC
* !3553 Add automated FHI72 4x4 nightly testing
* !4142 CI: Cleanup of Jenkinsfiles and parameters
* Adjust do_SRS after merge of !3723Closes#1086, #1097, #1022, and #1076
See merge request oai/openairinterface5g!4146
CI: Cleanup of Jenkinsfiles and parameters
- pass pre-computed branch from orchestrator to downstream jobs
- pass repository from orchestrator to downstream jobs, repository is now
configurable via ciRepositoryURL parameter of RAN-Container-Parent
- remove eNB_/ran-prefixed Jenkinsfile parameters
- address !4129 (comment 221898) and !4129 (comment 221900) and
!4129 (comment 221927)
Reviewed-By: Robert Schmidt <robert.schmidt@openairinterface.org>
Add automated FHI72 4x4 nightly testing
Add configuration files (conf, YAML, XML) for a new pipeline targeting
nightly high-throughput testing with FHI7.2 RUs. This pipeline is
designed to perform daily performance and stability testing using
either:
- the integration branch (to detect regressions in high-throughput
scenarios), or
- the develop branch (to collect up-to-date performance
metrics and observe long-term stability trends)
Details:
- testing with F1 split: DU runs on stonechat, CU runs on OC
- 4x4 MIMO, 4 DL layers layers with VVDN, LiteON, Metanoia, Bennetel550
and Benetel650 FHI7.2 RUs
- T2 card for LDPC processing
- TDD: DDDSU (special: DDDDDDDDGGGGUU) / DDDSU (special: DDDDDDGGGGUUUU)
for Benetel configrations
Reviewed-By: Robert Schmidt <robert.schmidt@openairinterface.org>
!3827 improved feprx processing time in 2x2 setup with 60MHz BW. Update
the timing thresholds accordingly.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Avoid the following error in the always post condition:
Error when executing always post condition:
Also: org.jenkinsci.plugins.workflow.actions.ErrorAction$ErrorId: 1540dd3e-f4ca-410c-8c37-8977ada45325
hudson.AbortException: Failed to copy artifacts from RAN-ARM-Cross-Compile-Builder with filter: test_results*.html
Add "optional: true" to copyArtifacts so that a missing HTML report
is handled by the existing placeholder fallback instead of aborting
the post condition.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Since MR !4116 removed support for building the L2sim proxy, there is no
longer a need to publish the proxy image to Docker Hub.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Branch names containing '/' are invalid in Docker image tags. Compute a
sanitized testBranch variable once (replacing '/' with '-') in the
Verify Guidelines stage and pass it to both triggerSlaveJob (as the new
testBranch parameter) and triggerCN5GSlaveJob (as fullRanTag).
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
This change enables configuring the repository and branch to test
directly from Jenkins.
The default checkout performed by the testing pipeline is now skipped,
since it does not support parameterized git URLs. Instead, the SCM
checkout is executed during the Build Init step, allowing dynamic
repository selection.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Jenkinsfile-GitLab-Container now constructs CI testing branch as
{sourceBranch}-{sourceCommit} and passes it explicitly to all downstream
jobs. Both Jenkinsfile and Jenkinsfile-push-local-repo consume it via
params.branch instead of independently reconstructing the name, ensuring
that the branch pushed by RAN-Local-Repo-Push and the branch checked out
by test jobs are always identical.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
UP3 development UE is used since no dedicated CI UE is available for outdoor
testing. CN with PLMN 20899 can be deployed locally on demand, however for
now always-on CN is used (AMF IP address 172.21.6.5).
The gNB is currently configured with pdsch_AntennaPorts_N1 = 2 instead
of 4, as the proper 8x8 antenna configuration triggers an assertion in
polar procedures (see issue 1067). This will be corrected once the assertion
is resolved.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Introduce a dedicated Jenkinsfile for nightly 4x4 runs. Increase the timeout
per test step to 150 minutes (vs. standard 60 minutes). Support dynamic branch
selection - if eNB_CommitID or eNB_Branch is set to "latest", the pipeline
resolves the latest develop or integration branch using get-latest-ref.sh script.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Test description:
- FHI7.2 test with Metanoia RU and Quectel module
- 4 DL layers, 2 UL layers
- stonechat as DU host, CU running on OC
- LDPC processing on the T2 card
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Test description:
- FHI7.2 test with LiteON RU and Quectel module
- 4 DL layers, 2 UL layers
- stonechat as DU host, CU running on OC
- LDPC processing on the T2 card
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
This commit is intended primarily for comparison with CU/DU split deployments,
enabling data collection and performance evaluation across both configurations.
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
Test description:
- FHI7.2 test with VVDN RU and Quectel module
- 4 DL layers, 2 UL layers
- stonechat as DU host, CU running on OC
- LDPC processing on the T2 card
Enable automatic retry of DU deployment to recover in case of following failure
during T2 initialization:
EAL: Unable to reset device! Error: 11 (Resource temporarily unavailable)
EAL: 0000:01:00.0 setup device failed
EAL: Driver cannot attach the device (01:00.0)
EAL: Failed to attach device on primary process
[PHY] T2 card 01:00.0 not found
Assertion (itf->nrLDPC_coding_init() == 0) failed!
In load_nrLDPC_coding_interface() /oai-ran/openair1/PHY/CODING/nrLDPC_coding/nrLDPC_coding_interface_load.c:66
error starting LDPC library ldpc (null)
Signed-off-by: Jaroslava Fiedlerova <jaroslava.fiedlerova@openairinterface.org>
