P20 crossover - convert the crossover module to only use the sink/source api by piotrhoppeintel · Pull Request #10999 · thesofproject/sof

piotrhoppeintel · 2026-07-15T07:27:35Z

Rework the crossover module to only use the sink/source api to
prepare the SOF for the full transition to pipeline 2.0.

Copilot

Pull request overview

Reworks the Crossover component’s processing path to use the SOF sof_source / sof_sink APIs (instead of stream buffers) as preparation for the pipeline 2.0 transition, and adds dedicated CMocka unit tests for LR4 filter behavior and split topology routing.

Changes:

Switch crossover processing entrypoint and processing function pointers to source/sink-based APIs.
Implement source/sink circular-buffer processing in crossover_generic.c for s16/s24/s32 formats plus passthrough copy via source_to_sink_copy().
Add CMocka unit tests and build integration for LR4 filter properties and 2/3/4-way split topology functions.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
test/cmocka/src/audio/crossover/crossover_test_mocks.c	Adds link-time stubs for source/sink APIs needed by unit tests that compile `crossover_generic.c`.
test/cmocka/src/audio/crossover/crossover_split_test.c	Adds split topology unit tests validating routing/energy behavior for 2/3/4-way splits.
test/cmocka/src/audio/crossover/crossover_lr4_test.c	Adds LR4 filter property tests (DC/HF behavior, split symmetry at fc, selectivity, denominator consistency).
test/cmocka/src/audio/crossover/CMakeLists.txt	Adds CMocka targets for crossover LR4 and split tests, wiring required sources.
test/cmocka/src/audio/CMakeLists.txt	Enables the new crossover test subdirectory behind `CONFIG_COMP_CROSSOVER`.
src/audio/crossover/crossover.h	Updates processing function typedefs to return `int` and take `sof_source`/`sof_sink` handles.
src/audio/crossover/crossover.c	Converts module `.process` implementation to source/sink APIs and updates sink assignment logic accordingly.
src/audio/crossover/crossover_generic.c	Reworks per-format processing/passthrough to use source/sink APIs and circular buffer helpers.
src/arch/xtensa/configs/unit_test_defconfig	Enables crossover component config for unit-test builds.

lyakh · 2026-07-16T07:05:24Z

 endif()
+if(CONFIG_COMP_CROSSOVER)
+	add_subdirectory(crossover)
+endif()


we've had a few PRs recently moving from over cmocka to Zephyr native tasting. Shouldn't this be done as a ztest too?

Done. Tests moved to the Ztest framework.

abonislawski · 2026-07-16T07:22:08Z

Please fix PR and commit title

btw: #10110

piotrhoppeintel · 2026-09-02T10:44:05Z

All comments have been addressed. @lyakh please review.

intel-sofci · 2026-09-02T12:07:24Z

PR 10999: test results

Run date: 2026-09-03 11:54 UTC

Tested commit: 9ee144db0f9b5256776660864b4a844e5fe0e6de

lyakh · 2026-09-03T07:24:04Z

+	rms_out_hf = rms_q31(out + WARMUP_LEN, SIGNAL_LEN - WARMUP_LEN);
+	printf("HP HF:  f=%.0fHz  input rms=%.4f  output rms=%.4f\n",
+	       f_hf, rms_in, rms_out_hf);
+	zassert_true(fabs(rms_out_hf - rms_in) / rms_in < 0.05,


do we have any science behind these numbers? Is 5% enough or do we need 0.05%? Similar for 0.01 in line 351 and supposedly other occurrences too.

zassert_true(fabs(rms_out_hf - rms_in) / rms_in < 0.05,

0.05 represents the maximum allowed relative RMS error of 5%. The test verifies that the LR4 high-pass filter passes a high-frequency signal at 0.45 * FS = 21.6 kHz with approximately unity gain:

[ \frac{|RMS_{out} - RMS_{in}|}{RMS_{in}} < 0.05 ]

Since rms_in ≈ 0.3536 , the permitted absolute difference is approximately 0.0177, giving an accepted output range of roughly 0.3359–0.3712 . This corresponds to a gain between 0.95 and 1.05, or approximately −0.45 dB to +0.42 dB.

The tolerance accounts for Q2.30 coefficient quantization, fixed-point rounding, and RMS measurement over a finite window that does not contain an integer number of signal periods. It is a test acceptance threshold, not a filter parameter.

right, but why 5%? why not 1% or 10% or 0.1%?

It is a subjective choice and is related to the description above so I added a safety margin that seemed reasonable for this test.

lyakh · 2026-09-03T07:32:23Z

+		     int32_t delay[LR4_NDELAY],
+		     const int32_t biquad_coef[BIQUAD_NCOEF])
+{
+	memcpy(coef,              biquad_coef, BIQUAD_NCOEF * sizeof(int32_t));


these spaces look strange. There's a similar location in crossover_split_ztest.c

I've removed spaces.

lyakh · 2026-09-03T07:38:39Z

+	double lp_50  = rms_buf(out[0] + WARMUP_LEN, SIGNAL_LEN - WARMUP_LEN);
+	double hp_50  = rms_buf(out[1] + WARMUP_LEN, SIGNAL_LEN - WARMUP_LEN);
+
+	/* 20 kHz -- should be dominated by HP (band 1) */


interesting, we split low / high at 1kHz, right? Should we also test frequencies like 950Hz / 1050Hz?

The main goal of these tests was to verify the migration to the new API. Please don't treat these tests as strictly as the crossover module tests. The test cases were chosen subjectively.

If you see any additional cases or frequencies that should be covered, please let me know. I'll try to add them.

lyakh · 2026-09-03T07:48:21Z

+		 */
+		for (i = 0; i < CROSSOVER_MAX_LR4; i++) {
+			double fc = (i == 0) ? fc_lo :
+				    (i == 1) ? ((fc_lo + fc_hi) / 2.0) : fc_hi;


ok, splitting hairs here, but how about

double fc[CROSSOVER_MAX_LR4] = {fc_lo, (fc_lo + fc_hi) / 2.0, fc_hi}; for (i = 0;...) { ... compute_biquad_lp(FS, fc[i], bq_lp);

I've implemented your suggestion.

Add seven LR4 filter tests and ten split topology tests as dedicated Ztest suites running on native_sim. Add the Twister test package and shared crossover mocks. Signed-off-by: Piotr Hoppe <piotr.hoppe@intel.com>

Rework the crossover module to only use the sink/source api to prepare the SOF for the full transition to pipeline 2.0. Signed-off-by: Piotr Hoppe <piotr.hoppe@intel.com>

Copilot AI review requested due to automatic review settings July 15, 2026 07:27

piotrhoppeintel requested review from a team, dbaluta, kv2019i, lbetlej, lgirdwood, mmaka1 and plbossart as code owners July 15, 2026 07:27

Copilot started reviewing on behalf of piotrhoppeintel July 15, 2026 07:35 View session

Copilot AI reviewed Jul 15, 2026

View reviewed changes

Comment thread src/audio/crossover/crossover_generic.c Outdated

Comment thread src/audio/crossover/crossover_generic.c Outdated

Comment thread src/audio/crossover/crossover_generic.c

lyakh reviewed Jul 16, 2026

View reviewed changes

piotrhoppeintel changed the title ~~P20 crossover~~ P20 crossover - convert the crossover module to only use the sink/source api Aug 11, 2026

piotrhoppeintel force-pushed the p20-crossover branch 3 times, most recently from 9fc9a6b to 2915070 Compare September 2, 2026 10:37

lyakh reviewed Sep 3, 2026

View reviewed changes

piotrhoppeintel added 2 commits September 3, 2026 13:14

test: crossover: add ztest unit tests for LR4 filter and split topology

cb6b7c7

Add seven LR4 filter tests and ten split topology tests as dedicated Ztest suites running on native_sim. Add the Twister test package and shared crossover mocks. Signed-off-by: Piotr Hoppe <piotr.hoppe@intel.com>

audio: crossover: rework module to use sink/source api

9ee144d

Rework the crossover module to only use the sink/source api to prepare the SOF for the full transition to pipeline 2.0. Signed-off-by: Piotr Hoppe <piotr.hoppe@intel.com>

piotrhoppeintel force-pushed the p20-crossover branch from 2915070 to 9ee144d Compare September 3, 2026 11:14

Uh oh!

Conversation

piotrhoppeintel commented Jul 15, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

abonislawski commented Jul 16, 2026

Uh oh!

piotrhoppeintel commented Sep 2, 2026

Uh oh!

intel-sofci commented Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR 10999: test results

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

piotrhoppeintel Sep 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

piotrhoppeintel Sep 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

intel-sofci commented Sep 2, 2026 •

edited

Loading

piotrhoppeintel Sep 3, 2026 •

edited

Loading

piotrhoppeintel Sep 3, 2026 •

edited

Loading