Part 2 of compressed offload support for IPC4 (no topology changes) by ujfalusi · Pull Request #10534 · thesofproject/sof

ujfalusi · 2026-02-10T07:18:20Z

Finalizing the code part of the compr offload support for IPC4:

fix vorbis support
Add sof_info and within sof_info the codec_info into fw_config. The codec_info gives the needed information to host on what codecs are supported by the booted firmware in cadence module.
Stop printing 'no bytes to copy' in FAST_MODE as by nature it is a burst mode and a copy is not going to happen on all ticks. FAST_MODE is going to be used for compr streams to facilitate the non audio data.
Add support for EOS handling in module_adaptor. Decoders can buffer the uncompressed data and they will spill out audio way after when no new data is consumed by them
Add new module notification for cadence module to notify host about events, the first such notification is the the EOS (or DRAIN) completion which is sent when the EOS is expected and the decoder produces 0 bytes (end of stream)
Update toml files for mtl+ to enable cadence module support
Add compr_overlay.conf as an easy way to enable all currently supported codecs via cadence module and provide a standardized location for the libraries.

Copilot

Pull request overview

Adds compressed offload support plumbing for IPC4 by enabling Cadence codec modules, exposing supported codec info via fw_config, and improving EOS handling/notifications for compressed streams.

Changes:

Enable Cadence module inclusion in multiple rimage TOML configs and add a ready-to-use compr_overlay.conf.
Add IPC4 fw_config SOF-specific tuple(s) to report supported codec capabilities to the host.
Implement EOS handling and a module notification path for compressed streams; reduce noisy “no bytes to copy” logging in FAST_MODE.

Reviewed changes

Copilot reviewed 14 out of 14 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tools/rimage/config/wcl.toml.h	Conditionally includes Cadence module TOML for this platform config.
tools/rimage/config/ptl.toml.h	Conditionally includes Cadence module TOML for this platform config.
tools/rimage/config/nvl.toml.h	Conditionally includes Cadence module TOML for this platform config.
tools/rimage/config/mtl.toml.h	Conditionally includes Cadence module TOML for this platform config.
tools/rimage/config/lnl.toml.h	Conditionally includes Cadence module TOML for this platform config.
src/include/sof/audio/module_adapter/module/generic.h	Adds EOS state fields to module processing data.
src/include/ipc4/header.h	Adds a new magic value for compressed module event notifications.
src/include/ipc4/base_fw.h	Introduces `IPC4_FW_SOF_INFO` and SOF sub-parameter IDs.
src/audio/module_adapter/module/cadence_ipc4.c	Adds Vorbis config case + EOS notification emission and sink EOS propagation.
src/audio/module_adapter/module/cadence.c	Initializes and updates EOS state based on decoder output and expected EOS.
src/audio/module_adapter/CMakeLists.txt	Imports Vorbis decoder library when enabled.
src/audio/host-zephyr.c	Suppresses “no bytes to copy” logs in FAST_MODE and improves warning formatting.
src/audio/base_fw.c	Builds and publishes codec capability TLVs under fw_config.
app/compr_overlay.conf	Adds an overlay to enable Cadence codecs and point to libraries.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-02-10T07:19:43Z

src/audio/module_adapter/module/cadence_ipc4.c

 	case CADENCE_CODEC_AAC_DEC_ID:
 		return cadence_configure_aac_dec_params(mod);
+	case CADENCE_CODEC_VORBIS_DEC_ID:
+		/* No configuration needed fro Vorbis */


Correct typo in comment: 'fro' → 'for'.

Suggested change

/* No configuration needed fro Vorbis */

/* No configuration needed for Vorbis */

Copilot · 2026-02-10T07:19:44Z

app/compr/all_codecs.conf

+CONFIG_CADENCE_CODEC=y
+CONFIG_CADENCE_CODEC_MP3_DEC=y
+CONFIG_CADENCE_CODEC_MP3_DEC_LIB="../cadence_libs/xa_mp3_dec.a"
+CONFIG_CADENCE_CODEC_MP3_ENC=y
+CONFIG_CADENCE_CODEC_MP3_ENC_LIB="../cadence_libs/xa_mp3_enc.a"
+CONFIG_CADENCE_CODEC_AAC_DEC=y
+CONFIG_CADENCE_CODEC_AAC_DEC_LIB="../cadence_libs/xa_aac_dec.a"
+CONFIG_CADENCE_CODEC_VORBIS_DEC=y
+CONFIG_CADENCE_CODEC_VORBIS_DEC_LIB="../cadence_libs/xa_vorbis_dec.a"


This overlay relies on relative library paths (e.g., ../cadence_libs/...) but doesn’t document the expected directory layout or where the paths are resolved from during builds. Add brief comments at the top of the file explaining the expected location of cadence_libs/ relative to the build/app directory, and how users should adjust these paths for their environment.

OK, I'll add a comment to the compr_overlay.conf file as well.

kv2019i

Only minor things, looks good and easy to review code.

kv2019i · 2026-02-10T07:58:03Z

src/audio/module_adapter/module/cadence_ipc4.c

 	case CADENCE_CODEC_AAC_DEC_ID:
 		return cadence_configure_aac_dec_params(mod);
+	case CADENCE_CODEC_VORBIS_DEC_ID:
+		/* No configuration needed fro Vorbis */


kv2019i · 2026-02-10T07:59:12Z

src/audio/base_fw.c


+struct sof_ipc4_codec_info_data {
+	uint32_t count;
+	uint32_t items[32];


Why 32? Future-proofing or aligning to some existing spec/convention?

kv2019i · 2026-02-10T08:06:35Z

src/audio/module_adapter/module/cadence_ipc4.c

+			msg_module_data->event_id = SOF_IPC4_NOTIFY_MODULE_EVENTID_COMPR_MAGIC_VAL;
+			msg_module_data->event_data_size = 0;
+
+			comp_err(dev, "[peter] Sending notification");


I think this should have one peter less.

...and shouldn't be an "error." Debug would be enough?

[peter] usually is in warn level, I will drop this.

lyakh · 2026-02-10T09:43:25Z

src/audio/base_fw.c

+	uint32_t items[32];
+} __packed __aligned(4);
+
+#define SET_CODEC_INFO_ITEM(codec, dir)	(((codec) & 0xff) | (((dir) & 0xff) << 16))


internally 16 bits would be enough, I presume 32 bits are used for compatibility with external standards?

The codec ID (lower 16bit) is actually defined in compress_params.h to be stored as __u32, max is 0x11 atm, so 8bit should be fine with bit7 as direction, but if I recall the fw_config itself have 32bit alignment requirement, all other flags are also stored as u32, so we follow this through
Is this acceptable answer @lyakh, @kv2019i ?

any reason we are trying to save here, why not just go for 31bits ID and 1 bit direction to start with ?

Yes, that would make more sense I guess, need ot align the kernel PR as well.

compress direction have 3 values:

enum snd_compr_direction { SND_COMPRESS_PLAYBACK = 0, SND_COMPRESS_CAPTURE, SND_COMPRESS_ACCEL };

I guess we could use bit 30-31, that would allow one more value for future proofing.
I'll go with something which is bit more compact:
BIT 0-7 (8 bits): codec_id
BIT 8-11 (4 bits): dir
and we leave BIT 12-31 for future use if needed.

lyakh · 2026-02-10T09:56:15Z

src/audio/module_adapter/module/cadence_ipc4.c

+			msg_module_data->event_id = SOF_IPC4_NOTIFY_MODULE_EVENTID_COMPR_MAGIC_VAL;
+			msg_module_data->event_data_size = 0;
+
+			comp_err(dev, "[peter] Sending notification");


...and shouldn't be an "error." Debug would be enough?

ujfalusi · 2026-02-10T12:25:19Z

Changes since v1:

typo corrected
drop the debug print with [peter]
add comment in compr_overlay.conf

lgirdwood

LGTM. No new opens.

lgirdwood · 2026-02-10T12:52:27Z

src/audio/base_fw.c

+	uint32_t items[32];
+} __packed __aligned(4);
+
+#define SET_CODEC_INFO_ITEM(codec, dir)	(((codec) & 0xff) | (((dir) & 0xff) << 16))


any reason we are trying to save here, why not just go for 31bits ID and 1 bit direction to start with ?

src/include/ipc4/header.h

ujfalusi · 2026-02-10T15:02:22Z

src/include/ipc4/base_fw.h

 	/* Minimum size of host buffer in ms */
 	IPC4_FW_MIN_HOST_BUFFER_PERIODS = 33,
+	/* decoder/encoder codec information */
+	IPC4_FW_SOF_INFO = 35,


If this ID is good than we need to reserve it...

35 seems free. We need to reserve it.

Yes, can we initiate the reservation process?
This will reduce the collision surface for future in both ways.

softwarecki

My two cents. Nice to see the EOS functionality starting to get actual use :)

softwarecki · 2026-02-10T16:32:58Z

src/audio/module_adapter/module/cadence_ipc4.c

+		primary->r.type = SOF_IPC4_GLB_NOTIFICATION;
+		primary->r.rsp = SOF_IPC4_MESSAGE_DIR_MSG_REQUEST;
+		primary->r.msg_tgt = SOF_IPC4_MESSAGE_TARGET_FW_GEN_MSG;
+		msg = ipc_msg_w_ext_init(msg_proto.header, msg_proto.extension,


Looks like a good place to use ipc_notification_pool_get

Would a followup patch do?
All module messages are using this mode and it will be better to update them with a single PR.
We can most likely also add helper to src/ipc/ipc4/notification.c for these at the same round?

softwarecki · 2026-02-10T16:39:22Z

src/include/ipc4/base_fw.h

 	/* Minimum size of host buffer in ms */
 	IPC4_FW_MIN_HOST_BUFFER_PERIODS = 33,
+	/* decoder/encoder codec information */
+	IPC4_FW_SOF_INFO = 35,


35 seems free. We need to reserve it.

The zephyr_library_import() is needed to compile the vorbis decoder support Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

setup_cfg can be in theory undefined when no module_data is provided, initialize it to NULL and check for NULL pointer before accessing to it's member. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

The decoder for Vorbis does not need any configuration. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

…yload the sof_info (param id 35) contains SOF specific information in a tuple based array. The first such info block is the codec_info (sub ID: 0) to hold information about the codecs supported fro compress offload. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

In FAST_MODE the host DMA is not paced with the period size and can copy more data once to do no copying for a other periods. In other words: in FAST_MODE the host is not running in stream mode, but in batch mode. Printing warnings in this does not make sense and just fills up the firmware log. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

…ruct To facilitate the End Of Stream handling in processing modules add two flags: eos_reached: when set, the processing module reached EOS by processing all data. EOS can happen when the pipeline have expect_eos flag set, indicating that EOS is going to happen eos_notification_sent: if the module uses notification to host, this flag can be used make sure to not flood with IPCs Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

0xC0C0 magic value is to be used by compr module as module notification of drain completion. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

…host When the pipeline has the expect_eos flag set and the cadence module produces 0 bytes during process is an indication that the EOS has been completed and the module drained all data from input to output. Send a notification about this to host and set EOS state on downstream module instances. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

Convert TGL platform to modular toml configuration to align with more modern architectures. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

Convert TGL-H platform to modular toml configuration to align with more modern architectures. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

…module if enabled Include the cadence.toml if CONFIG_CADENCE_CODEC is enabled. Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com> Co-developed-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com> Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

The needed libraries should be placed in sof/../cadence_libs/ directory if the overlay is used. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

ujfalusi · 2026-02-11T11:54:22Z

Changes since v2:

the codec_info u32 has been re-partitioned to 8bit codec_id and 4bit direction
noticed a bug in cadence_ipc4 which can dereference a random pointer during init, fix included
converting tgl and tgl-h to dynamic toml since a base cadence module will be needed to load topologies with compr enabled.
toml update for all Intel platforms when cadence is enabled
add set of overlay files to allow codec selection and to be able to include only the base cadence module w/o codecs

lyakh · 2026-02-11T11:56:30Z

src/audio/base_fw.c

+{
+	char sof_config_data[SOF_CONFIG_SIZE_MAX] = { 0 };
+	struct sof_tlv *sof_config_tuple = (struct sof_tlv *)sof_config_data;
+	uint32_t sof_config_size;


size_t resp. ssize_t?

lyakh · 2026-02-11T11:57:01Z

src/audio/base_fw.c

+	uint32_t sof_config_size;
+
+	get_codec_info(&sof_config_tuple);
+	sof_config_size = (uint32_t)((char *)sof_config_tuple - sof_config_data);


you could just type-cast sof_config_tuple to uintptr_t to avoid double-casting

lyakh · 2026-02-11T11:57:24Z

src/audio/base_fw.c

+
+	get_codec_info(&sof_config_tuple);
+	sof_config_size = (uint32_t)((char *)sof_config_tuple - sof_config_data);
+	if (sof_config_size == 0)


can it turn negative?

ranj063

looks great @ujfalusi

lgirdwood · 2026-02-12T10:10:32Z

Lets deal with any of the minor requests in the next update as this one is passing CI and there is an update fro FLAC coming.

Copilot AI review requested due to automatic review settings February 10, 2026 07:18

ujfalusi requested review from RanderWang, dbaluta, jxstelter, kv2019i, lbetlej, lgirdwood, marcinszkudlinski, mmaka1, pblaszko, plbossart and ranj063 as code owners February 10, 2026 07:18

Copilot AI reviewed Feb 10, 2026

View reviewed changes

kv2019i reviewed Feb 10, 2026

View reviewed changes

lyakh reviewed Feb 10, 2026

View reviewed changes

ujfalusi force-pushed the peter/pr/ipc4_compr_part2 branch from 22b3ab7 to c28c64c Compare February 10, 2026 12:24

lgirdwood reviewed Feb 10, 2026

View reviewed changes

ujfalusi commented Feb 10, 2026

View reviewed changes

src/include/ipc4/header.h Show resolved Hide resolved

ujfalusi commented Feb 10, 2026

View reviewed changes

softwarecki reviewed Feb 10, 2026

View reviewed changes

ujfalusi mentioned this pull request Feb 11, 2026

ASoC: SOF: Initial support for IPC4 compress offload thesofproject/linux#5647

Open

ujfalusi added 8 commits February 11, 2026 13:46

audio: module_adapter: Add zephyr_library_import() for Vorbis decoder

7966ba4

The zephyr_library_import() is needed to compile the vorbis decoder support Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

module: cadence_ipc4: Fix dereferencing of uninitialized pointer

e79442a

setup_cfg can be in theory undefined when no module_data is provided, initialize it to NULL and check for NULL pointer before accessing to it's member. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

module: cadence_ipc4: Enable support for VORBIS decoder

c357597

The decoder for Vorbis does not need any configuration. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

ipc4: header: Add new event ID for COMPR in module notification

02d8b40

0xC0C0 magic value is to be used by compr module as module notification of drain completion. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

ujfalusi and others added 4 commits February 11, 2026 13:47

rimage: config: toml: Convert tgl to modular configuration

7a5dae5

Convert TGL platform to modular toml configuration to align with more modern architectures. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

rimage: config: toml: Convert tgl-h to modular configuration

3e38b64

Convert TGL-H platform to modular toml configuration to align with more modern architectures. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

app: Add overlays to enable compr offload support (mp3, aac and vorbis)

f3f018b

The needed libraries should be placed in sof/../cadence_libs/ directory if the overlay is used. Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>

ujfalusi force-pushed the peter/pr/ipc4_compr_part2 branch from c28c64c to f3f018b Compare February 11, 2026 11:47

lyakh reviewed Feb 11, 2026

View reviewed changes

ranj063 reviewed Feb 11, 2026

View reviewed changes

ranj063 approved these changes Feb 11, 2026

View reviewed changes

lgirdwood approved these changes Feb 12, 2026

View reviewed changes

lgirdwood merged commit 2925c4b into thesofproject:main Feb 12, 2026
48 of 54 checks passed

	/* No configuration needed fro Vorbis */
	/* No configuration needed for Vorbis */

Conversation

ujfalusi commented Feb 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Feb 10, 2026

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!

Copilot AI Feb 10, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kv2019i left a comment

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!

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!

ujfalusi commented Feb 10, 2026

Uh oh!

lgirdwood left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

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!

softwarecki left a comment

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!

ujfalusi commented Feb 11, 2026

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!

ranj063 left a comment

Choose a reason for hiding this comment

Uh oh!

ujfalusi commented Feb 10, 2026 •

edited

Loading