Repository navigation
topology: validator: check IPC4 pin format compatibility - #11443
Conversation
There was a problem hiding this comment.
🟡 Changes recommended
Removing the sole build target for sof-dmic-4ch-wov.tplg would stop production builds from generating the shared PTL/WCL WoV feature topology.
4 open issues
What changed in this PR
This PR extends SOF’s topology validator to check declared format compatibility between connected IPC4 pins, without predicting runtime format selection. It also updates topology definitions and adds regression coverage.
Changes:
- Add pin-aware compatibility checks and sample-type/bit-depth validation.
- Adjust HDA, SoundWire, and WebRTC topology formats and rates.
- Document validation behavior and add unit, CLI, and integration tests.
| File | Description |
|---|---|
tools/topology/validator/tests/test_rules.py |
Add format-validation and CLI regression cases. |
tools/topology/validator/tests/test_integration.py |
Check SoundWire echo-reference pin compatibility. |
tools/topology/validator/README.md |
Explain format checks, severities, and limitations. |
tools/topology/validator/checks/tokens.py |
Add T008 sample-type and bit-depth validation. |
tools/topology/validator/checks/paths.py |
Update declared-format normalization. |
tools/topology/validator/checks/ipc4.py |
Add F003 connected-pin compatibility checks. |
tools/topology/topology2/README.md |
Document HDA benchmark rates and filter behavior. |
tools/topology/topology2/production/tplg-targets-ace3.cmake |
Remove the DMIC WoV production registration. |
tools/topology/topology2/platform/intel/sdw-amp-xover.conf |
Adjust integer sample-type declarations. |
tools/topology/topology2/include/pipelines/cavs/mixout-gain-alh-dai-copier-playback.conf |
Use LSB integer input format. |
tools/topology/topology2/include/components/webrtc-vad.conf |
Make VAD sample rate configurable. |
tools/topology/topology2/development/cavs-nocodec-webrtc-ns.conf |
Align VAD rate and capture output depth. |
tools/topology/topology2/cavs-sdw-src-gain-mixin.conf |
Use LSB integer output format. |
tools/topology/topology2/cavs-mixin-mixout-hda.conf |
Align component formats and rate-specific filtering. |
tools/topology/topology2/cavs-benchmark-hda.conf |
Gate default DAI pipelines and align MFCC rates. |
Review effort: Balanced
| if not selected and output and pin > 0 and widget.type_name in PROCESS_TYPES: | ||
| selected = [f for f in fmts if not f.get(index, 0)] | ||
| # Incomplete formats are diagnosed by T006; do not infer incompatibility. | ||
| if not selected or any(not f.get(rate) or not f.get(cfg) for f in selected): |
| bits = (cfg >> 8) & 0xff | ||
| stype = (cfg >> 16) & 0xff if bits in (8, 32) else TYPE_LSB_INTEGER | ||
| stype = (cfg >> 16) & 0xff | ||
| if bits == 32 and fmt.get(depth_tok, 0) == 32 and stype == 0: |
| problem = f"valid bit depth {bits} exceeds container depth {depth}" | ||
| elif stype == 4 and (depth, bits) != (32, 32): | ||
| problem = f"float requires 32/32 bits, got {depth}/{bits}" | ||
| elif stype in (5, 6) and (depth, bits) != (8, 8): |
PR 11443: test resultsRun date: 2026-11-13 13:55 UTC Tested commit: 18727ad1226517c6a7b85f51806a37db92472b6e |
01872cb to
18727ad
Compare
|
Changes since v1:
|
|
@ujfalusi can you check CI thanks ! |
CI tests have unrelated fails, the GH tests fail to install deb packages and unrelated errors. |
|
@singalsu pls check for wov and webrtc |
18727ad to
9764f18
Compare
Remove the standalone DMIC WoV target. Its KPB secondary output declares 16 kHz while the connected host copier only declares 48 kHz input formats, which F003 reports as incompatible. Signed-off-by: Peter Ujfalusi <[email protected]>
Include the sample-type bits in the selector input and playback copier input formats. These internal connections carry 32-bit LSB integer samples, matching the DRC and crossover FIR outputs. Leave the external SoundWire output format unchanged. Signed-off-by: Peter Ujfalusi <[email protected]>
…gateways The generic benchmark provides its own playback and capture DAI pipelines, but also instantiates the default component-benchmark DAI gateways. The overlapping pipeline IDs and duplicate widget names cause the wrong format declarations and pipeline ownership to be used. Instantiate the default gateways only with HDA_CONFIG=none. Keep the custom gateways for HDA_CONFIG=benchmark and document the distinction. Verify analog playback and capture at 48 kHz in S16_LE, S24_LE and S32_LE on a TGL HDA device with kernel 7.2.14. Both directions now run instead of playback timing out and capture failing at pipeline start. Signed-off-by: Peter Ujfalusi <[email protected]>
…arks MFCC and MFCC-mel benchmarks configure the host copiers and processing modules for 16 kHz, but their HDA DAI copiers inherit 48 kHz formats. F003 reports incompatible formats on both connections to the DAIs. Use the configured HDA playback and capture rates for both sides of the DAI copiers, and select 16 kHz for all six MFCC benchmark variants. Keep 48 kHz defaults for other component benchmarks. The copiers convert sample formats, not sample rates, so the HDA link must also use 16 kHz. Document the benchmark rate and feature-data output. Verify three-second playback and capture streams at 16 kHz for each variant in its native S16_LE, S24_LE or S32_LE format on a TGL HDA device with kernel 7.2.14. This checks stream operation, not MFCC algorithm correctness. Representative unaffected benchmark binaries remain byte-identical. Signed-off-by: Peter Ujfalusi <[email protected]>
The standalone SRC-gain-mixin topology declares MSB integer samples on internal connections whose peers declare LSB integer samples. F003 reports mismatches at both the playback and capture copiers. Use LSB integer samples for the shared playback copier input and the standalone capture copier output. Remove their explicit fmt_cfg overrides so the base format classes encode the selected sample type. Keep the SoundWire-facing formats as MSB integer. These connections carry 32 valid bits in 32-bit containers, so sample placement is unchanged. This corrects the format declarations rather than an observed runtime audio failure. Signed-off-by: Peter Ujfalusi <[email protected]>
…cted rate The HDA analog PCM advertises only 48 kHz in its rate list even when its minimum and maximum rates are set to 96 kHz. The resulting empty rate constraints prevent the stream from opening. The playback gain, mixers and effects also retain 48 kHz defaults when 96 kHz is selected. Move the pre-mixer gain formats out of the mixer control and configure the internal playback chain at the selected rate. Allow the PCM rate list to include the configured rate while retaining the fixed minimum and maximum constraints. Keep the capture high-pass cutoff at 40 Hz for 96 kHz by using the 20 Hz / 48 kHz coefficients, which have the same normalized cutoff. All nine 96 kHz development targets pass F003 with the validator in this series. Playback and capture were tested on a TGL HDA device with S16_LE, S24_LE and S32_LE at 96 kHz and with the production topology at 48 kHz. The updated high-pass coefficients have not been tested on hardware. Signed-off-by: Peter Ujfalusi <[email protected]>
The nocodec WebRTC test topology declares a 16-bit capture DAI output feeding a copier configured for 32-bit input. Its VAD also declares 48 kHz formats while the following noise suppressor uses 16 kHz. Configure the DAI output as 32-bit, matching the first module copier. Make the VAD format rate selectable, defaulting to 48 kHz, and set it to 16 kHz in this test topology. This preserves the existing sample format and channel choices while aligning the connected formats. The rebuilt topology passes F003. Hardware testing requires an SSP nocodec loopback setup and has not been performed. Signed-off-by: Peter Ujfalusi <[email protected]>
Resolve route endpoints to their bound input and output pins and report an error when declared format sets have no compatible format. Do not infer queue allocation for unbound multi-pin endpoints or constrain undeclared secondary copier outputs. Compare mixer outputs using the shared format set selected by the kernel for all pins. Check sample representations independently of compatibility normalization and reject invalid sample types and bit depths. Treat full-width 32-bit MSB and LSB integers as equivalent, and include container depth when evaluating declared sample types. These checks validate declared format compatibility, not the behavior of a particular kernel version or runtime sample alignment. Matching declarations do not simulate runtime selection, and module-specific configuration can require further review. Cover pin bindings, sample types, base topologies, and siggen echo reference with regression tests. Signed-off-by: Peter Ujfalusi <[email protected]>
Check that each declared copier output can be selected at the input rate using fixed output parameters or preserved input parameters. Require the resulting combination to be declared. Report undeclared secondary outputs and incomplete formats as unresolved. Document the selection contract and add focused tests. Signed-off-by: Peter Ujfalusi <[email protected]>
|
Changes since v2:
|
|
@ujfalusi not expecting CI to fail for this, can you confirm we are good. Thanks |
failed to connect to the docker API at unix:///var/run/docker.sock; check if the path is correct and if the daemon is running: dial unix /var/run/docker.sock: connect: no such file or directory I don't think it is caused by this PR. |


Validate format compatibility between connected IPC4 pins using explicit pin bindings, and report incompatible declared format sets. Add checks for invalid sample types and bit depths, with regression coverage. The validator checks declared format compatibility; it does not predict behavior across kernel versions or runtime format selection.
The fixed topologies all have incompatible formats on connections, like:
sof-hda-benchmark-generic.tplg: warning: F003 route
'dai-copier.HDA.Analog.capture' -> 'eqiir.5.1':
output pin 0 [48000Hz/2ch/24bit/type1] has no compatible format with
input pin 0 [48000Hz/2ch/32bit/type1]